* [PATCH v6 0/4] dm-replicator: introduce new remote replication target
@ 2009-12-18 15:44 heinzm
2009-12-18 15:44 ` [PATCH v6 1/4] dm-replicator: documentation and module registry heinzm
0 siblings, 1 reply; 9+ messages in thread
From: heinzm @ 2009-12-18 15:44 UTC (permalink / raw)
To: dm-devel; +Cc: Heinz Mauelshagen
From: Heinz Mauelshagen <heinzm@redhat.com>
* 6th version of patch series (initial version dated Oct 23 2009) *
Rebased to 2.6.33-rc1 (dm-dirty-log interface change).
Fixed some sparse warnings spotted by Alasdair (make C=1).
This is a series of 4 patches introducing the device-mapper remote
data replication target "dm-replicator" to kernel 2.6.
Userspace support for remote data replication will be in
a future LVM2 version.
The target supports disaster recovery by replicating groups of active
mapped devices (ie. receiving io from applications) to one or more
remote sites to paired groups of equally sized passive block devices
(ie. no application access). Synchronous, asynchronous replication
(with fallbehind settings) and temporary downtime of transports
are supported.
It utilizes a replication log to ensure write ordering fidelity for
the whole group of replicated devices, hence allowing for consistent
recovery after failover of arbitrary applications
(eg. DBMS utilizing N > 1 devices).
In case the replication log runs full, it is capable to fall back
to dirty logging utilizing the existing dm-log module, hence keeping
track of regions of devices wich need resynchronization after access
to the transport returned.
Access logic of the replication log and the site links are implemented
as loadable modules, hence allowing for future implementations with
different capabilities in terms of additional plugins.
A "ringbuffer" replication log module implements a circular ring buffer
store for all writes being processed. Other replication log handlers
may follow this one as plugins too.
A "blockdev" site link module implements block devices access to all remote
devices, ie. all devices exposed via the Linux block device layer
(eg. iSCSI, FC).
Again, other eg. network type transport site link handlers may
follow as plugins.
Please review for upstream inclusion.
Heinz
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCH v6 1/4] dm-replicator: documentation and module registry
2009-12-18 15:44 [PATCH v6 0/4] dm-replicator: introduce new remote replication target heinzm
@ 2009-12-18 15:44 ` heinzm
2009-12-18 15:44 ` [PATCH v6 2/4] dm-replicator: replication log and site link handler interfaces and main replicator module heinzm
2010-01-07 10:18 ` [PATCH v6 1/4] dm-replicator: documentation and module registry 张宇
0 siblings, 2 replies; 9+ messages in thread
From: heinzm @ 2009-12-18 15:44 UTC (permalink / raw)
To: dm-devel; +Cc: Heinz Mauelshagen
From: Heinz Mauelshagen <heinzm@redhat.com>
The dm-registry module is a general purpose registry for modules.
The remote replicator utilizes it to register its ringbuffer log and
site link handlers in order to avoid duplicating registry code and logic.
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Reviewed-by: Jon Brassow <jbrassow@redhat.com>
Tested-by: Jon Brassow <jbrassow@redhat.com>
---
Documentation/device-mapper/replicator.txt | 203 +++++++++++++++++++++++++
drivers/md/Kconfig | 8 +
drivers/md/Makefile | 1 +
drivers/md/dm-registry.c | 224 ++++++++++++++++++++++++++++
drivers/md/dm-registry.h | 38 +++++
5 files changed, 474 insertions(+), 0 deletions(-)
create mode 100644 Documentation/device-mapper/replicator.txt
create mode 100644 drivers/md/dm-registry.c
create mode 100644 drivers/md/dm-registry.h
diff --git 2.6.33-rc1.orig/Documentation/device-mapper/replicator.txt 2.6.33-rc1/Documentation/device-mapper/replicator.txt
new file mode 100644
index 0000000..1d408a6
--- /dev/null
+++ 2.6.33-rc1/Documentation/device-mapper/replicator.txt
@@ -0,0 +1,203 @@
+dm-replicator
+=============
+
+Device-mapper replicator is designed to enable redundant copies of
+storage devices to be made - preferentially, to remote locations.
+RAID1 (aka mirroring) is often used to maintain redundant copies of
+storage for fault tolerance purposes. Unlike RAID1, which often
+assumes similar device characteristics, dm-replicator is designed to
+handle devices with different latency and bandwidth characteristics
+which are often the result of the geograhic disparity of multi-site
+architectures. Simply put, you might choose RAID1 to protect from
+a single device failure, but you would choose remote replication
+via dm-replicator for protection against a site failure.
+
+dm-replicator works by first sending write requests to the "replicator
+log". Not to be confused with the device-mapper dirty log, this
+replicator log behaves similarly to that of a journal. Write requests
+go to this log first and then are copied to all the replicate devices
+at their various locations. Requests are cleared from the log once all
+replicate devices confirm the data is received/copied. This architecture
+allows dm-replicator to be flexible in terms of device characteristics.
+If one device should fall behind the others - perhaps due to high latency -
+the slack is picked up by the log. The user has a great deal of
+flexibility in specifying to what degree a particular site is allowed to
+fall behind - if at all.
+
+Device-Mapper's dm-replicator has two targets, "replicator" and
+"replicator-dev". The "replicator" target is used to setup the
+aforementioned log and allow the specification of site link properties.
+Through the "replicator" target, the user might specify that writes
+that are copied to the local site must happen synchronously (i.e the
+writes are complete only after they have passed through the log device
+and have landed on the local site's disk). They may also specify that
+a remote link should asynchronously complete writes, but that the remote
+link should never fall more than 100MB behind in terms of processing.
+Again, the "replicator" target is used to define the replicator log and
+the characteristics of each site link.
+
+The "replicator-dev" target is used to define the devices used and
+associate them with a particular replicator log. You might think of
+this stage in a similar way to setting up RAID1 (mirroring). You
+define a set of devices which will be copies of each other, but
+access the device through the mirror virtual device which takes care
+of the copying. The user accessible replicator device is analogous
+to the mirror virtual device, while the set of devices being copied
+to are analogous to the mirror images (sometimes called 'legs').
+When creating a replicator device via the "replicator-dev" target,
+it must be associated with the replicator log (created with the
+aforementioned "replicator" target). When each redundant device
+is specified as part of the replicator device, it is associated with
+a site link whose properties were defined when the "replicator"
+target was created.
+
+The user can go farther than simply replicating one device. They
+can continue to add replicator devices - associating them with a
+particular replicator log. Writes that go through the replicator
+log are guarenteed to have their write ordering preserved. So, if
+you associate more than one replicator device to a particular
+replicator log, you are preserving write ordering across multiple
+devices. This might be useful if you had a database that spanned
+multiple disks and write ordering must be preserved or any transaction
+accounting scheme would be foiled. (You can imagine this like
+preserving write ordering across a number of mirrored devices, where
+each mirror has images/legs in different geographic locations.)
+
+dm-replicator has a modular architecture. Future implementations for
+the replicator log and site link modules are allowed. The current
+replication log is ringbuffer - utilized to store all writes being
+subject to replication and enforce write ordering. The current site
+link code is based on accessing block devices (iSCSI, FC, etc) and
+does device recovery including (initial) resynchronization.
+
+
+Picture of a 2 site configuration with 3 local devices (LDs) in a
+primary site being resycnhronied to 3 remotes sites with 3 remote
+devices (RDs) each via site links (SLINK) 1-2 with site link 0
+as a special case to handle the local devices:
+
+ |
+ Local (primary) site | Remote sites
+ -------------------- | ------------
+ |
+ D1 D2 Dn |
+ | | | |
+ +---+- ... -+ |
+ | |
+ REPLOG-----------------+- SLINK1 ------------+
+ | | | |
+ SLINK0 (special case) | | |
+ | | | |
+ +-----+ ... + | | +----+- ... -+
+ | | | | | | | |
+ LD1 LD2 LDn | | RD1 RD2 RDn
+ | |
+ +-- SLINK2------------+
+ | | |
+ | | +----+- ... -+
+ | | | | |
+ | | RD1 RD2 RDn
+ | |
+ | |
+ | |
+ +- SLINKm ------------+
+ | |
+ | +----+- ... -+
+ | | | |
+ | RD1 RD2 RDn
+
+
+
+
+The following are descriptions of the device-mapper tables used to
+construct the "replicator" and "replicator-dev" targets.
+
+"replicator" target parameters:
+-------------------------------
+<start> <length> replicator \
+ <replog_type> <#replog_params> <replog_params> \
+ [<slink_type_0> <#slink_params_0> <slink_params_0>]{1..N}
+
+<replog_type> = "ringbuffer" is currently the only available type
+<#replog_params> = # of args following this one intended for the replog (2 or 4)
+<replog_params> = <dev_path> <dev_start> [auto/create/open <size>]
+ <dev_path> = device path of replication log (REPLOG) backing store
+ <dev_start> = offset to REPLOG header
+ create = The replication log will be initialized if not active
+ and sized to "size". (If already active, the create
+ will fail.) Size is always in sectors.
+ open = The replication log must be initialized and valid or
+ the constructor will fail.
+ auto = If a valid replication log header is found on the
+ replication device, this will behave like 'open'.
+ Otherwise, this option behaves like 'create'.
+
+<slink_type> = "blockdev" is currently the only available type
+<#slink_params> = 1/2/4
+<slink_params> = <slink_nr> [<slink_policy> [<fall_behind> <N>]]
+ <slink_nr> = This is a unique number that is used to identify a
+ particular site/location. '0' is always used to
+ identify the local site, while increasing integers
+ are used to identify remote sites.
+ <slink_policy> = The policy can be either 'sync' or 'async'.
+ 'sync' means write requests will not return until
+ the data is on the storage device. 'async' allows
+ a device to "fall behind"; that is, outstanding
+ write requests are waiting in the replication log
+ to be processed for this site, but it is not delaying
+ the writes of other sites.
+ <fall_behind> = This field is used to specify how far the user is
+ willing to allow write requests to this specific site
+ to "fall behind" in processing before switching to
+ a 'sync' policy. This "fall behind" threshhold can
+ be specified in three ways: ios, size, or timeout.
+ 'ios' is the number of pending I/Os allowed (e.g.
+ "ios 10000"). 'size' is the amount of pending data
+ allowed (e.g. "size 200m"). Size labels include:
+ s (sectors), k, m, g, t, p, and e. 'timeout' is
+ the amount of time allowed for writes to be
+ outstanding. Time labels include: s, m, h, and d.
+
+
+"replicator-dev" target parameters:
+-----------------------------------
+start> <length> replicator-dev
+ <replicator_device> <dev_nr> \
+ [<slink_nr> <#dev_params> <dev_params>
+ <dlog_type> <#dlog_params> <dlog_params>]{1..N}
+
+<replicator_device> = device previously constructed via "replication" target
+<dev_nr> = An integer that is used to 'tag' write requests as
+ belonging to a particular set of devices - specifically,
+ the devices that follow this argument (i.e. the site
+ link devices).
+<slink_nr> = This number identifies the site/location where the next
+ device to be specified comes from. It is exactly the
+ same number used to identify the site/location (and its
+ policies) in the "replicator" target. Interestingly,
+ while one might normally expect a "dev_type" argument
+ here, it can be deduced from the site link number and
+ the 'slink_type' given in the "replication" target.
+<#dev_params> = '1' (The number of allowed parameters actually depends
+ on the 'slink_type' given in the "replication" target.
+ Since our only option there is "blockdev", the only
+ allowable number here is currently '1'.)
+<dev_params> = 'dev_path' (Again, since "blockdev" is the only
+ 'slink_type' available, the only allowable argument here
+ is the path to the device.)
+<dlog_type> = Not to be confused with the "replicator log", this is
+ the type of dirty log associated with this particular
+ device. Dirty logs are used for synchronization, during
+ initialization or fall behind conditions, to bring devices
+ into a coherent state with its peers - analogous to
+ rebuilding a RAID1 (mirror) device. Available dirty
+ log types include: 'nolog', 'core', and 'disk'
+<#dlog_params> = The number of arguments required for a particular log
+ type - 'nolog' = 0, 'core' = 1/2, 'disk' = 2/3.
+<dlog_params> = 'nolog' => ~no arguments~
+ 'core' => <region_size> [sync | nosync]
+ 'disk' => <dlog_dev_path> <region_size> [sync | nosync]
+ <region_size> = This sets the granularity at which the dirty log
+ tracks what areas of the device is in-sync.
+ [sync | nosync] = Optionally specify whether the sync should be forced
+ or avoided initially.
diff --git 2.6.33-rc1.orig/drivers/md/Kconfig 2.6.33-rc1/drivers/md/Kconfig
index acb3a4e..62c9766 100644
--- 2.6.33-rc1.orig/drivers/md/Kconfig
+++ 2.6.33-rc1/drivers/md/Kconfig
@@ -313,6 +313,14 @@ config DM_DELAY
If unsure, say N.
+config DM_REPLICATOR
+ tristate "Replication target (EXPERIMENTAL)"
+ depends on BLK_DEV_DM && EXPERIMENTAL
+ ---help---
+ A target that supports replication of local devices to remote sites.
+
+ If unsure, say N.
+
config DM_UEVENT
bool "DM uevents (EXPERIMENTAL)"
depends on BLK_DEV_DM && EXPERIMENTAL
diff --git 2.6.33-rc1.orig/drivers/md/Makefile 2.6.33-rc1/drivers/md/Makefile
index e355e7f..be05b39 100644
--- 2.6.33-rc1.orig/drivers/md/Makefile
+++ 2.6.33-rc1/drivers/md/Makefile
@@ -44,6 +44,7 @@ obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o dm-region-hash.o
obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
obj-$(CONFIG_DM_ZERO) += dm-zero.o
+obj-$(CONFIG_DM_REPLICATOR) += dm-log.o dm-registry.o
quiet_cmd_unroll = UNROLL $@
cmd_unroll = $(AWK) -f$(srctree)/$(src)/unroll.awk -vN=$(UNROLL) \
diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.c 2.6.33-rc1/drivers/md/dm-registry.c
new file mode 100644
index 0000000..fb8abbf
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-registry.c
@@ -0,0 +1,224 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * Generic registry for arbitrary structures
+ * (needs dm_registry_type structure upfront each registered structure).
+ *
+ * This file is released under the GPL.
+ *
+ * FIXME: use as registry for e.g. dirty log types as well.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+
+#include "dm-registry.h"
+
+#define DM_MSG_PREFIX "dm-registry"
+
+static const char *version = "0.001";
+
+/* Sizable class registry. */
+static unsigned num_classes;
+static struct list_head *_classes;
+static rwlock_t *_locks;
+
+void *
+dm_get_type(const char *type_name, enum dm_registry_class class)
+{
+ struct dm_registry_type *t;
+
+ read_lock(_locks + class);
+ list_for_each_entry(t, _classes + class, list) {
+ if (!strcmp(type_name, t->name)) {
+ if (!t->use_count && !try_module_get(t->module)) {
+ read_unlock(_locks + class);
+ return ERR_PTR(-ENOMEM);
+ }
+
+ t->use_count++;
+ read_unlock(_locks + class);
+ return t;
+ }
+ }
+
+ read_unlock(_locks + class);
+ return ERR_PTR(-ENOENT);
+}
+EXPORT_SYMBOL(dm_get_type);
+
+void
+dm_put_type(void *type, enum dm_registry_class class)
+{
+ struct dm_registry_type *t = type;
+
+ read_lock(_locks + class);
+ if (!--t->use_count)
+ module_put(t->module);
+
+ read_unlock(_locks + class);
+}
+EXPORT_SYMBOL(dm_put_type);
+
+/* Add a type to the registry. */
+int
+dm_register_type(void *type, enum dm_registry_class class)
+{
+ struct dm_registry_type *t = type, *tt;
+
+ if (unlikely(class >= num_classes))
+ return -EINVAL;
+
+ tt = dm_get_type(t->name, class);
+ if (unlikely(!IS_ERR(tt))) {
+ dm_put_type(t, class);
+ return -EEXIST;
+ }
+
+ write_lock(_locks + class);
+ t->use_count = 0;
+ list_add(&t->list, _classes + class);
+ write_unlock(_locks + class);
+
+ return 0;
+}
+EXPORT_SYMBOL(dm_register_type);
+
+/* Remove a type from the registry. */
+int
+dm_unregister_type(void *type, enum dm_registry_class class)
+{
+ struct dm_registry_type *t = type;
+
+ if (unlikely(class >= num_classes)) {
+ DMERR("Attempt to unregister invalid class");
+ return -EINVAL;
+ }
+
+ write_lock(_locks + class);
+
+ if (unlikely(t->use_count)) {
+ write_unlock(_locks + class);
+ DMWARN("Attempt to unregister a type that is still in use");
+ return -ETXTBSY;
+ } else
+ list_del(&t->list);
+
+ write_unlock(_locks + class);
+ return 0;
+}
+EXPORT_SYMBOL(dm_unregister_type);
+
+/*
+ * Return kmalloc'ed NULL terminated pointer
+ * array of all type names of the given class.
+ *
+ * Caller has to kfree the array!.
+ */
+const char **dm_types_list(enum dm_registry_class class)
+{
+ unsigned i = 0, count = 0;
+ const char **r;
+ struct dm_registry_type *t;
+
+ /* First count the registered types in the class. */
+ read_lock(_locks + class);
+ list_for_each_entry(t, _classes + class, list)
+ count++;
+ read_unlock(_locks + class);
+
+ /* None registered in this class. */
+ if (!count)
+ return NULL;
+
+ /* One member more for array NULL termination. */
+ r = kzalloc((count + 1) * sizeof(*r), GFP_KERNEL);
+ if (!r)
+ return ERR_PTR(-ENOMEM);
+
+ /*
+ * Go with the counted ones.
+ * Any new added ones after we counted will be ignored!
+ */
+ read_lock(_locks + class);
+ list_for_each_entry(t, _classes + class, list) {
+ r[i++] = t->name;
+ if (!--count)
+ break;
+ }
+ read_unlock(_locks + class);
+
+ return r;
+}
+EXPORT_SYMBOL(dm_types_list);
+
+int __init
+dm_registry_init(void)
+{
+ unsigned n;
+
+ BUG_ON(_classes);
+ BUG_ON(_locks);
+
+ /* Module parameter given ? */
+ if (!num_classes)
+ num_classes = DM_REGISTRY_CLASS_END;
+
+ n = num_classes;
+ _classes = kmalloc(n * sizeof(*_classes), GFP_KERNEL);
+ if (!_classes) {
+ DMERR("Failed to allocate classes registry");
+ return -ENOMEM;
+ }
+
+ _locks = kmalloc(n * sizeof(*_locks), GFP_KERNEL);
+ if (!_locks) {
+ DMERR("Failed to allocate classes locks");
+ kfree(_classes);
+ _classes = NULL;
+ return -ENOMEM;
+ }
+
+ while (n--) {
+ INIT_LIST_HEAD(_classes + n);
+ rwlock_init(_locks + n);
+ }
+
+ DMINFO("initialized %s for max %u classes", version, num_classes);
+ return 0;
+}
+
+void __exit
+dm_registry_exit(void)
+{
+ BUG_ON(!_classes);
+ BUG_ON(!_locks);
+
+ kfree(_classes);
+ _classes = NULL;
+ kfree(_locks);
+ _locks = NULL;
+ DMINFO("exit %s", version);
+}
+
+/* Module hooks */
+module_init(dm_registry_init);
+module_exit(dm_registry_exit);
+module_param(num_classes, uint, 0);
+MODULE_PARM_DESC(num_classes, "Maximum number of classes");
+MODULE_DESCRIPTION(DM_NAME "device-mapper registry");
+MODULE_AUTHOR("Heinz Mauelshagen <heinzm@redhat.com>");
+MODULE_LICENSE("GPL");
+
+#ifndef MODULE
+static int __init num_classes_setup(char *str)
+{
+ num_classes = simple_strtol(str, NULL, 0);
+ return num_classes ? 1 : 0;
+}
+
+__setup("num_classes=", num_classes_setup);
+#endif
diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.h 2.6.33-rc1/drivers/md/dm-registry.h
new file mode 100644
index 0000000..1cb0ce8
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-registry.h
@@ -0,0 +1,38 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * Generic registry for arbitrary structures.
+ * (needs dm_registry_type structure upfront each registered structure).
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm.h"
+
+#ifndef DM_REGISTRY_H
+#define DM_REGISTRY_H
+
+enum dm_registry_class {
+ DM_REPLOG = 0,
+ DM_SLINK,
+ DM_LOG,
+ DM_REGION_HASH,
+ DM_REGISTRY_CLASS_END,
+};
+
+struct dm_registry_type {
+ struct list_head list; /* Linked list of types in this class. */
+ const char *name;
+ struct module *module;
+ unsigned int use_count;
+};
+
+void *dm_get_type(const char *type_name, enum dm_registry_class class);
+void dm_put_type(void *type, enum dm_registry_class class);
+int dm_register_type(void *type, enum dm_registry_class class);
+int dm_unregister_type(void *type, enum dm_registry_class class);
+const char **dm_types_list(enum dm_registry_class class);
+
+#endif
--
1.6.2.5
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCH v6 2/4] dm-replicator: replication log and site link handler interfaces and main replicator module
2009-12-18 15:44 ` [PATCH v6 1/4] dm-replicator: documentation and module registry heinzm
@ 2009-12-18 15:44 ` heinzm
2009-12-18 15:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler heinzm
2010-01-07 10:18 ` [PATCH v6 1/4] dm-replicator: documentation and module registry 张宇
1 sibling, 1 reply; 9+ messages in thread
From: heinzm @ 2009-12-18 15:44 UTC (permalink / raw)
To: dm-devel; +Cc: Heinz Mauelshagen
From: Heinz Mauelshagen <heinzm@redhat.com>
These are the interface definitions for the replication log and the site link
handler plus the main replicator module plugging into the dm core interface
to construct/destruct/... replication control ("replicator" target)
and data devices ("replicator-dev" target).
The "replicator" control target handles the replication log and
the site link properties (eg. log size or asynchronous replication),
while the "replicator-dev" target handles all local and remote device
properties (eg. their paths and dirty log parameters).
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Reviewed-by: Jon Brassow <jbrassow@redhat.com>
Tested-by: Jon Brassow <jbrassow@redhat.com>
---
drivers/md/Makefile | 4 +-
drivers/md/dm-repl-log.h | 120 +++
drivers/md/dm-repl-slink.h | 313 +++++++
drivers/md/dm-repl.c | 2004 ++++++++++++++++++++++++++++++++++++++++++++
drivers/md/dm-repl.h | 127 +++
drivers/md/dm.c | 1 +
6 files changed, 2568 insertions(+), 1 deletions(-)
create mode 100644 drivers/md/dm-repl-log.h
create mode 100644 drivers/md/dm-repl-slink.h
create mode 100644 drivers/md/dm-repl.c
create mode 100644 drivers/md/dm-repl.h
diff --git 2.6.33-rc1.orig/drivers/md/Makefile 2.6.33-rc1/drivers/md/Makefile
index be05b39..832d547 100644
--- 2.6.33-rc1.orig/drivers/md/Makefile
+++ 2.6.33-rc1/drivers/md/Makefile
@@ -8,6 +8,7 @@ dm-multipath-y += dm-path-selector.o dm-mpath.o
dm-snapshot-y += dm-snap.o dm-exception-store.o dm-snap-transient.o \
dm-snap-persistent.o
dm-mirror-y += dm-raid1.o
+dm-replicator-y += dm-repl.o
dm-log-userspace-y \
+= dm-log-userspace-base.o dm-log-userspace-transfer.o
md-mod-y += md.o bitmap.o
@@ -44,7 +45,8 @@ obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o dm-region-hash.o
obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
obj-$(CONFIG_DM_ZERO) += dm-zero.o
-obj-$(CONFIG_DM_REPLICATOR) += dm-log.o dm-registry.o
+obj-$(CONFIG_DM_REPLICATOR) += dm-replicator.o \
+ dm-log.o dm-registry.o
quiet_cmd_unroll = UNROLL $@
cmd_unroll = $(AWK) -f$(srctree)/$(src)/unroll.awk -vN=$(UNROLL) \
diff --git 2.6.33-rc1.orig/drivers/md/dm-repl-log.h 2.6.33-rc1/drivers/md/dm-repl-log.h
new file mode 100644
index 0000000..cff74ed
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-repl-log.h
@@ -0,0 +1,120 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * This file is released under the GPL.
+ */
+
+/*
+ * API calling convention to create a replication mapping:
+ *
+ * 1. get a replicator log handle, hence creating a new persistent
+ * log or accessing an existing one
+ * 2. get an slink handle, hence creating a new transient
+ * slink or accessing an existing one
+ * 2(cont). repeat the previous step for multiple slinks (eg. one for
+ * local and one for remote device access)
+ * 3. bind a (remote) device to a particlar slink created in a previous step
+ * 3(cont). repeat the device binding for any additional devices on that slink
+ * 4. bind the created slink which has device(s) bound to it to the replog
+ * 4(cont). repeat the slink binding to the replog for all created slinks
+ * 5. call the replog io function for each IO.
+ *
+ * Reverse steps 1-4 to tear a replication mapping down, hence freeing all
+ * transient resources allocated to it.
+ */
+
+#ifndef _DM_REPL_LOG_H
+#define _DM_REPL_LOG_H
+
+#include "dm-repl.h"
+#include "dm-registry.h"
+#include "dm-repl-slink.h"
+
+/* Handle to access a replicator log. */
+struct dm_repl_log {
+ struct dm_repl_log_type *ops;
+ void *context;
+};
+
+/* List of site links hanging off of each replicator log. */
+struct dm_repl_log_slink_list {
+ rwlock_t lock;
+ struct list_head list; /* List of site links hanging of off this log. */
+ void *context; /* Caller context. */
+};
+
+struct dm_repl_log_type {
+ struct dm_registry_type type;
+
+ /* Construct/destruct a replicator log. */
+ int (*ctr)(struct dm_repl_log *, struct dm_target *,
+ unsigned argc, char **argv);
+ void (*dtr)(struct dm_repl_log *, struct dm_target *);
+
+ /*
+ * There are times when we want the log to be quiet.
+ * Ie. no entries of the log will be copied accross site links.
+ */
+ int (*postsuspend)(struct dm_repl_log *log, int dev_number);
+ int (*resume)(struct dm_repl_log *log, int dev_number);
+
+ /* Flush the current log contents. This function may block. */
+ int (*flush)(struct dm_repl_log *log);
+
+ /*
+ * Read a bio either from a replicator logs backing store
+ * (if supported) or from the replicated device if no buffer entry.
+ * - or-
+ * write a bio to a replicator logs backing store buffer.
+ *
+ * This includes buffer allocation in case of a write and
+ * inititation of copies accross an/multiple site link(s).
+ *
+ * In case of a read with (partial) writes in the buffer,
+ * the replog may postpone the read until the buffer content has
+ * been copied accross the local site link *or* optimize by reading
+ * (parts of) the bio off the buffer.
+ *
+ * Tag us a unique tag identifying a data set.
+ */
+ int (*io)(struct dm_repl_log *, struct bio *, unsigned long long tag);
+
+ /* Endio function to call from dm_repl core on bio endio processing. */
+ int (*endio) (struct dm_repl_log *, struct bio *bio, int error,
+ union map_info *map_context);
+
+ /* Set global I/O completion notification function and context- */
+ void (*io_notify_fn_set)(struct dm_repl_log *,
+ dm_repl_notify_fn, void *context);
+
+ /*
+ * Add (tie) a site link to a replication
+ * log for site link copy processing.
+ */
+ int (*slink_add)(struct dm_repl_log *, struct dm_repl_slink *);
+
+ /* Remove (untie) a site link from a replication log. */
+ int (*slink_del)(struct dm_repl_log *, struct dm_repl_slink *);
+
+ /*
+ * Return list of site links added to a replication log.
+ *
+ * This method eases slink handler coding to
+ * keep such replication log site link list.
+ */
+ struct dm_repl_log_slink_list *(*slinks)(struct dm_repl_log *);
+
+ /* Return maximum number of supported site links. */
+ int (*slink_max)(struct dm_repl_log *);
+
+ /* REPLOG messages. */
+ int (*message)(struct dm_repl_log *, unsigned argc, char **argv);
+
+ /* Support function for replicator log status requests. */
+ int (*status)(struct dm_repl_log *, int dev_number, status_type_t,
+ char *result, unsigned maxlen);
+};
+
+#endif /* #ifndef _DM_REPL_LOG_H */
diff --git 2.6.33-rc1.orig/drivers/md/dm-repl-slink.h 2.6.33-rc1/drivers/md/dm-repl-slink.h
new file mode 100644
index 0000000..ddf4ef7
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-repl-slink.h
@@ -0,0 +1,313 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * This file is released under the GPL.
+ */
+
+/*
+ * API calling convention to create a replication mapping:
+ *
+ * 1. get a replicator log handle, hence creating a new persistent
+ * log or accessing an existing one
+ * 2. get an slink handle, hence creating a new transient
+ * slink or accessing an existing one
+ * 2(cont). repeat the previous step for multiple slinks (eg. one for
+ * local and one for remote device access)
+ * 3. bind a (remote) device to a particlar slink created in a previous step
+ * 3(cont). repeat the device binding for any additional devices on that slink
+ * 4. bind the created slink which has device(s) bound to it to the replog
+ * 4(cont). repeat the slink binding to the replog for all created slinks
+ * 5. call the replog write function for each write IO and the replog hit
+ * function for each read IO..
+ *
+ * Reverse steps 1-4 to tear a replication mapping down, hence freeing all
+ * transient resources allocated to it.
+ */
+
+#ifndef _DM_REPL_SLINK_IO_H
+#define _DM_REPL_SLINK_IO_H
+
+#include "dm.h"
+#include "dm-repl.h"
+#include "dm-registry.h"
+
+#include <linux/dm-io.h>
+
+/* Handle to access a site link. */
+struct dm_repl_slink {
+ struct dm_repl_slink_type *ops;
+ void *context; /* Private slink (callee) context. */
+ void *caller; /* Caller context to (optionally) tie to slink. */
+};
+
+/*
+ * Start copy function parameters.
+ */
+/* Copy device address union content type. */
+enum dm_repl_slink_dev_type {
+ DM_REPL_SLINK_BLOCK_DEVICE, /* Copying from/to block_device. */
+ DM_REPL_SLINK_DEV_NUMBER, /* Copying from/to device number. */
+};
+
+/* Copy device address. */
+struct dm_repl_slink_copy_addr {
+ /* Union content type. */
+ enum dm_repl_slink_dev_type type;
+
+ /* Either address is block_device or slink/device # pair. */
+ union {
+ struct block_device *bdev;
+ struct {
+ unsigned slink;
+ unsigned dev;
+ } number;
+ } dev;
+
+ /* Sector offset on device to copy to/from. */
+ sector_t sector;
+};
+
+/* Copy notification callback parameters. */
+struct dm_repl_slink_notify_ctx {
+ dm_repl_notify_fn fn;
+ void *context;
+};
+
+/* Copy function structure to pass in from caller. */
+struct dm_repl_slink_copy {
+ struct dm_repl_slink_copy_addr src; /* Source address of copy. */
+ struct dm_repl_slink_copy_addr dst; /* Destination address of copy. */
+ unsigned size; /* Size of copy [bytes]. */
+
+ /* Notification callback for data transfered to (remote) RAM. */
+ struct dm_repl_slink_notify_ctx ram;
+ /* Notification callback for data transfered to (remote) disk. */
+ struct dm_repl_slink_notify_ctx disk;
+};
+/*
+ * End copy function parameters.
+ */
+
+/* SLINK policies */
+enum dm_repl_slink_policy_type {
+ DM_REPL_SLINK_ASYNC,
+ DM_REPL_SLINK_SYNC,
+ DM_REPL_SLINK_STALL,
+};
+
+/* SLINK states */
+enum dm_repl_slink_state_type {
+ DM_REPL_SLINK_DOWN,
+ DM_REPL_SLINK_READ_ERROR,
+ DM_REPL_SLINK_WRITE_ERROR,
+};
+
+/* SLINK fallbehind information. */
+/* Definition of fall behind values. */
+enum dm_repl_slink_fallbehind_type {
+ DM_REPL_SLINK_FB_IOS, /* Number of IOs. */
+ DM_REPL_SLINK_FB_SIZE, /* In sectors unless unit. */
+ DM_REPL_SLINK_FB_TIMEOUT, /* In ms unless unit. */
+};
+struct dm_repl_slink_fallbehind {
+ enum dm_repl_slink_fallbehind_type type;
+ sector_t value;
+ sector_t multiplier;
+ char unit;
+};
+
+struct dm_repl_log;
+
+/* SLINK handler interface type. */
+struct dm_repl_slink_type {
+ /* Must be first to allow for registry abstraction! */
+ struct dm_registry_type type;
+
+ /* Construct/destruct a site link. */
+ int (*ctr)(struct dm_repl_slink *, struct dm_repl_log *,
+ unsigned argc, char **argv);
+ void (*dtr)(struct dm_repl_slink *);
+
+ /*
+ * There are times when we want the slink to be quiet.
+ * Ie. no checks will run on slinks and no initial
+ * resynchronization will be performed.
+ */
+ int (*postsuspend)(struct dm_repl_slink *slink, int dev_number);
+ int (*resume)(struct dm_repl_slink *slink, int dev_number);
+
+ /* Add a device to a site link. */
+ int (*dev_add)(struct dm_repl_slink *, int dev_number,
+ struct dm_target *ti, unsigned argc, char **argv);
+
+ /* Delete a device from a site link. */
+ int (*dev_del)(struct dm_repl_slink *, int dev_number);
+
+ /*
+ * Initiate data copy across a site link.
+ *
+ * This function may be used to copy a buffer entry *or*
+ * for resynchronizing regions initially or when an SLINK
+ * has fallen back to dirty log (bitmap) mode.
+ *
+ * The dm_repl_slink_copy can be allocated on the stack,
+ * because copies of its members are taken before the function returns.
+ *
+ * The function will call 2 callbacks, one to report data in (remote)
+ * RAM and another one to report data on (remote) disk
+ * (see dm_repl_slink_copy structure for details).
+ *
+ * Tag is a unique tag to identify a data set.
+ *
+ *
+ * The return codes are defined as follows:
+ *
+ * o -EAGAIN in case of prohibiting I/O because
+ * of device inaccessibility/suspension
+ * or device I/O errors
+ * (i.e. link temporarilly down) ->
+ * caller is allowed to retry the I/O later once
+ * he'll have received a callback.
+ *
+ * o -EACCES in case a region is being resynchronized
+ * and the source region is being read to copy data
+ * accross to the same region of the replica (RD) ->
+ * caller is allowed to retry the I/O later once
+ * he'll have received a callback.
+ *
+ * o -ENODEV in case a device is not configured
+ * caller must drop the I/O to the device/slink pair.
+ *
+ * o -EPERM in case a region is out of sync ->
+ * caller must drop the I/O to the device/slink pair.
+ */
+ int (*copy)(struct dm_repl_slink *, struct dm_repl_slink_copy *,
+ unsigned long long tag);
+
+ /* Submit bio to underlying transport. */
+ int (*io)(struct dm_repl_slink *, struct bio *,
+ unsigned long long tag);
+
+ /* Endio function to call from dm_repl core on bio endio processing. */
+ int (*endio) (struct dm_repl_slink *, struct bio *bio, int error,
+ union map_info *map_context);
+
+ /* Unplug request queues on all devices on slink. */
+ int (*unplug)(struct dm_repl_slink *);
+
+ /* Set global recovery notification function and context- */
+ void (*recover_notify_fn_set)(struct dm_repl_slink *,
+ dm_repl_notify_fn, void *context);
+
+ /* Set/clear sync status of sector. */
+ int (*set_sync)(struct dm_repl_slink *, int dev_number,
+ sector_t sector, int in_sync);
+
+ /* Flush any dirty logs on slink. */
+ int (*flush_sync)(struct dm_repl_slink *);
+
+ /* Trigger resynchronization of devices on slink. */
+ int (*resync)(struct dm_repl_slink *slink, int resync);
+
+ /* Return > 0 if region is in sync on all slinks. */
+ int (*in_sync)(struct dm_repl_slink *slink, int dev_number,
+ sector_t region);
+
+ /* Site link policies. */
+ enum dm_repl_slink_policy_type (*policy)(struct dm_repl_slink *);
+
+ /* Site link state. */
+ enum dm_repl_slink_state_type (*state)(struct dm_repl_slink *);
+
+ /* Return reference to fallbehind information. */
+ struct dm_repl_slink_fallbehind *(*fallbehind)(struct dm_repl_slink *);
+
+ /* Return device number for block_device on slink if any. */
+ int (*dev_number)(struct dm_repl_slink *, struct block_device *);
+
+ /* Return # of the SLINK. */
+ int (*slink_number)(struct dm_repl_slink *);
+
+ /* Return SLINK by number. */
+ struct dm_repl_slink *(*slink)(struct dm_repl_log *,
+ unsigned slink_number);
+
+ /* SLINK status requests. */
+ int (*status)(struct dm_repl_slink *, int dev_number,
+ status_type_t, char *result, unsigned int maxlen);
+
+ /* SLINK messages (eg. change policy). */
+ int (*message)(struct dm_repl_slink *, unsigned argc, char **argv);
+};
+
+/* Policy and state access inlines. */
+/* Policy synchronous. */
+static inline int
+slink_policy_synchronous(enum dm_repl_slink_policy_type policy)
+{
+ return test_bit(DM_REPL_SLINK_SYNC, (unsigned long *) &policy);
+}
+
+/* Slink synchronous. */
+static inline int
+slink_synchronous(struct dm_repl_slink *slink)
+{
+ return slink_policy_synchronous(slink->ops->policy(slink));
+}
+
+/* Policy asynchronous. */
+static inline int
+slink_policy_asynchronous(enum dm_repl_slink_policy_type policy)
+{
+ return test_bit(DM_REPL_SLINK_ASYNC, (unsigned long *) &policy);
+}
+
+/* Slink asynchronous. */
+static inline int
+slink_asynchronous(struct dm_repl_slink *slink)
+{
+ return slink_policy_asynchronous(slink->ops->policy(slink));
+}
+
+/* Policy stall. */
+static inline int
+slink_policy_stall(enum dm_repl_slink_policy_type policy)
+{
+ return test_bit(DM_REPL_SLINK_STALL, (unsigned long *) &policy);
+}
+
+/* Slink stall. */
+static inline int
+slink_stall(struct dm_repl_slink *slink)
+{
+ return slink_policy_stall(slink->ops->policy(slink));
+}
+
+/* State down.*/
+static inline int
+slink_state_down(enum dm_repl_slink_state_type state)
+{
+ return test_bit(DM_REPL_SLINK_DOWN, (unsigned long *) &state);
+}
+
+/* Slink down.*/
+static inline int
+slink_down(struct dm_repl_slink *slink)
+{
+ return slink_state_down(slink->ops->state(slink));
+}
+
+/* Setup of site links. */
+/* Create/destroy a transient replicator site link */
+struct dm_repl_slink *
+dm_repl_slink_get(char *name, struct dm_repl_log *,
+ unsigned argc, char **argv);
+void dm_repl_slink_put(struct dm_repl_slink *);
+
+/* init/exit functions. */
+int dm_repl_slink_init(void);
+void dm_repl_slink_exit(void);
+
+#endif
diff --git 2.6.33-rc1.orig/drivers/md/dm-repl.c 2.6.33-rc1/drivers/md/dm-repl.c
new file mode 100644
index 0000000..86d1b48
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-repl.c
@@ -0,0 +1,2004 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen <HeinzM@redhat.com>
+ *
+ * This file is released under the GPL.
+ *
+ * Remote Replication target.
+ *
+ * Features:
+ * o Logs writes to circular buffer keeping persistent state metadata.
+ * o Writes data from log synchronuously or asynchronuously
+ * to multiple (1-N) remote replicas.
+ * o stores CRCs with metadata for integrity checks
+ * o stores versions with metadata to support future metadata migration
+ *
+ *
+ * For disk layout of backing store see dm-repl-log implementation.
+ *
+ *
+ * This file is the control module of the replication target, which
+ * controls the construction/destruction and mapping of replication
+ * mappings interfacing into seperate log and site link (transport)
+ * handler modules.
+ *
+ * That architecture allows the control module to be log *and* transport
+ * implementation agnostic.
+ */
+
+static const char version[] = "v0.028";
+
+#include "dm.h"
+#include "dm-repl.h"
+#include "dm-repl-log.h"
+#include "dm-repl-slink.h"
+
+#include <stdarg.h>
+#include <linux/dm-dirty-log.h>
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/crc32.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/namei.h>
+#include <linux/types.h>
+#include <linux/vmalloc.h>
+#include <linux/workqueue.h>
+
+#define DM_MSG_PREFIX "dm-repl"
+#define DAEMON DM_MSG_PREFIX "d"
+
+/* Default local device read ahead pages. */
+#define LD_RA_PAGES_DEFAULT 8
+
+/* Factor out to dm.[ch] */
+/* Return type for name. */
+int
+dm_descr_type(const struct dm_str_descr *descr, unsigned len, const char *name)
+{
+ while (len--) {
+ if (!strncmp(STR_LEN(name, descr[len].name)))
+ return descr[len].type;
+ }
+
+ return -ENOENT;
+}
+EXPORT_SYMBOL_GPL(dm_descr_type);
+
+/* Return name for type. */
+const char *
+dm_descr_name(const struct dm_str_descr *descr, unsigned len, const int type)
+{
+ while (len--) {
+ if (type == descr[len].type)
+ return descr[len].name;
+ }
+
+ return NULL;
+}
+EXPORT_SYMBOL_GPL(dm_descr_name);
+/* END Factor out to dm.[ch] */
+
+/* Global list of replication log contexts for ctr/dtr and lock. */
+static LIST_HEAD(replog_c_list);
+static struct mutex replog_c_list_mutex;
+
+/* Statistics. */
+struct stats {
+ atomic_t io[2];
+ atomic_t submitted_io[2];
+ atomic_t congested_fn[2];
+};
+
+/* Reset statistics variables. */
+static void
+stats_reset(struct stats *stats)
+{
+ int i = 2;
+
+ while (i--) {
+ atomic_set(stats->io + i, 0);
+ atomic_set(stats->submitted_io + i, 0);
+ atomic_set(stats->congested_fn + i, 0);
+ }
+}
+
+/* Per site link context. */
+struct slink_c {
+ struct {
+ struct list_head slink_c;
+ struct list_head dc; /* List of replication device contexts. */
+ } lists;
+
+ /* Reference count (ie. number of devices on this site link) */
+ struct kref ref;
+
+ /* Slink handle. */
+ struct dm_repl_slink *slink;
+
+ /* Replog context. */
+ struct replog_c *replog_c;
+};
+
+/* Global context kept with replicator log. */
+enum replog_c_flags {
+ REPLOG_C_BLOCKED,
+ REPLOG_C_DEVEL_STATS,
+ REPLOG_C_IO_INFLIGHT
+};
+struct replog_c {
+ struct {
+ struct list_head replog_c;/* To add to global replog_c list. */
+ struct list_head slink_c; /* Site link context elements. */
+ } lists;
+
+ struct dm_target *ti;
+
+ /* Reference count (ie. # of slinks * # of devices on this replog) */
+ struct kref ref;
+
+ /* Back pointer to replication log. */
+ struct dm_repl_log *replog;
+ dev_t dev; /* Replicator control device major:minor. */
+
+ /* Global io housekeeping on site link 0. */
+ struct repl_io {
+ unsigned long flags; /* I/O state flags. */
+
+ struct bio_list in; /* Pending bios (central input list).*/
+ spinlock_t in_lock; /* Protects central input list.*/
+ atomic_t in_flight; /* In flight io counter. */
+
+ /* IO workqueue. */
+ struct workqueue_struct *wq;
+ struct work_struct ws;
+
+ /* Statistics. */
+ struct stats stats;
+
+ /* slink+I/O teardown synchronization. */
+ wait_queue_head_t waiters;
+ } io;
+};
+DM_BITOPS(ReplBlocked, replog_c, REPLOG_C_BLOCKED);
+DM_BITOPS(ReplDevelStats, replog_c, REPLOG_C_DEVEL_STATS);
+DM_BITOPS(ReplIoInflight, replog_c, REPLOG_C_IO_INFLIGHT);
+
+/*
+ * Per device replication context kept with any mapped device and
+ * any associated remote device, which doesn't have a local mapping.
+ */
+struct device_c {
+ struct list_head list; /* To add to slink_c rc list. */
+
+ /* Local device ti (i.e. head). */
+ struct dm_target *ti;
+
+ /* replicator control device reference. */
+ struct dm_dev *replicator_dev;
+
+ /* SLINK handle. */
+ struct slink_c *slink_c;
+
+ /* This device's number. */
+ int number;
+};
+
+/* IO in flight wait qeue handling during suspension. */
+static void
+replog_c_io_get(struct replog_c *replog_c)
+{
+ SetReplIoInflight(replog_c);
+ atomic_inc(&replog_c->io.in_flight);
+}
+
+/* Drop io in flight reference. */
+static void
+replog_c_io_put(struct replog_c *replog_c)
+{
+ if (atomic_dec_and_test(&replog_c->io.in_flight)) {
+ ClearReplIoInflight(replog_c);
+ wake_up(&replog_c->io.waiters);
+ }
+}
+
+/* Get a handle on a replicator log. */
+static struct dm_repl_log *
+repl_log_ctr(const char *name, struct dm_target *ti,
+ unsigned int argc, char **argv)
+{
+ int r;
+ struct dm_repl_log_type *type;
+ struct dm_repl_log *log;
+
+ log = kzalloc(sizeof(*log), GFP_KERNEL);
+ if (unlikely(!log))
+ return ERR_PTR(-ENOMEM);
+
+ /* Load requested replication log module. */
+ r = request_module("dm-repl-log-%s", name);
+ if (r < 0) {
+ DMERR("replication log module for \"%s\" not found", name);
+ kfree(log);
+ return ERR_PTR(-ENOENT);
+ }
+
+ type = dm_get_type(name, DM_REPLOG);
+ if (unlikely(IS_ERR(type))) {
+ DMERR("replication log registry type not found");
+ kfree(log);
+ return (struct dm_repl_log *) type;
+ }
+
+ log->ops = type;
+ r = type->ctr(log, ti, argc, argv);
+ if (unlikely(r < 0)) {
+ DMERR("%s: constructor failed", __func__);
+ dm_put_type(type, DM_REPLOG);
+ kfree(log);
+ log = ERR_PTR(r);
+ }
+
+ return log;
+}
+
+/* Put a handle on a replicator log. */
+static void
+repl_log_dtr(struct dm_repl_log *log, struct dm_target *ti)
+{
+ /* Frees log on last drop. */
+ log->ops->dtr(log, ti);
+ dm_put_type(log->ops, DM_REPLOG);
+ kfree(log);
+}
+
+/*
+ * Create/destroy a transient replicator site link on initial get/last out.
+ */
+static struct dm_repl_slink *
+repl_slink_ctr(char *name, struct dm_repl_log *replog,
+ unsigned argc, char **argv)
+{
+ int r;
+ struct dm_repl_slink_type *type;
+ struct dm_repl_slink *slink;
+
+ slink = kzalloc(sizeof(*slink), GFP_KERNEL);
+ if (unlikely(!slink))
+ return ERR_PTR(-ENOMEM);
+
+ /* Load requested replication site link module. */
+ r = request_module("dm-repl-slink-%s", name);
+ if (r < 0) {
+ DMERR("replication slink module for \"%s\" not found", name);
+ kfree(slink);
+ return ERR_PTR(-ENOENT);
+ }
+
+ type = dm_get_type(name, DM_SLINK);
+ if (unlikely(IS_ERR(type))) {
+ DMERR("replication slink registry type not found");
+ kfree(slink);
+ return (struct dm_repl_slink *) type;
+ }
+
+ r = type->ctr(slink, replog, argc, argv);
+ if (unlikely(r < 0)) {
+ DMERR("%s: constructor failed", __func__);
+ dm_put_type(type, DM_SLINK);
+ kfree(slink);
+ return ERR_PTR(r);
+ }
+
+ slink->ops = type;
+ return slink;
+}
+
+static void
+slink_destroy(struct dm_repl_slink *slink)
+{
+ /* Frees slink on last reference drop. */
+ slink->ops->dtr(slink);
+ dm_put_type(slink->ops, DM_SLINK);
+ kfree(slink);
+}
+
+
+/* Wake worker. */
+static void do_repl(struct work_struct *ws);
+static void
+wake_do_repl(struct replog_c *replog_c)
+{
+ queue_work(replog_c->io.wq, &replog_c->io.ws);
+}
+
+/* Called from the replog in case we can queue more bios. */
+static void
+io_callback(int read_err, int write_err, void *context)
+{
+ struct replog_c *replog_c = context;
+
+ DMDEBUG_LIMIT("%s", __func__);
+ _BUG_ON_PTR(replog_c);
+ ClearReplBlocked(replog_c);
+ wake_do_repl(replog_c);
+}
+
+/* Get a reference on a replog_c by replog reference. */
+static struct replog_c *
+replog_c_get(struct replog_c *replog_c)
+{
+ kref_get(&replog_c->ref);
+ return replog_c;
+}
+
+/* Destroy replog_c object. */
+static int slink_c_put(struct slink_c *slink_c);
+static void
+replog_c_release(struct kref *ref)
+{
+ struct replog_c *replog_c = container_of(ref, struct replog_c, ref);
+
+ BUG_ON(!list_empty(&replog_c->lists.replog_c));
+ BUG_ON(!list_empty(&replog_c->lists.slink_c));
+ kfree(replog_c);
+}
+
+/* Release reference on replog_c, releasing resources on last drop. */
+static int
+replog_c_put(struct replog_c *replog_c)
+{
+ _BUG_ON_PTR(replog_c);
+ return kref_put(&replog_c->ref, replog_c_release);
+}
+
+/*
+ * Find a replog_c by replog reference in the global replog context list.
+ *
+ * Call with replog_c_list_mutex held.
+ */
+static struct replog_c *
+replog_c_get_by_dev(dev_t dev)
+{
+ struct replog_c *replog_c;
+
+ list_for_each_entry(replog_c, &replog_c_list, lists.replog_c) {
+ if (dev == replog_c->dev)
+ return replog_c_get(replog_c);
+ }
+
+ return ERR_PTR(-ENOENT);
+}
+
+/* Get replicator control device major:minor. */
+static dev_t
+get_ctrl_dev(struct dm_target *ti)
+{
+ dev_t dev;
+ struct mapped_device *md = dm_table_get_md(ti->table);
+ struct block_device *bdev = bdget_disk(dm_disk(md), 0);
+
+ dev = bdev->bd_dev;
+ bdput(bdev);
+ dm_put(md);
+ return dev;
+}
+
+/* Allocate a replication control context. */
+static struct replog_c *
+replog_c_alloc(void)
+{
+ struct replog_c *replog_c = kzalloc(sizeof(*replog_c), GFP_KERNEL);
+ struct repl_io *io;
+
+ if (unlikely(!replog_c))
+ return ERR_PTR(-ENOMEM);
+
+ io = &replog_c->io;
+
+ /* Create singlethread workqueue for this replog's io. */
+ io->wq = create_singlethread_workqueue(DAEMON);
+ if (unlikely(!io->wq)) {
+ kfree(replog_c);
+ return ERR_PTR(-ENOMEM);
+ }
+
+ kref_init(&replog_c->ref);
+ INIT_LIST_HEAD(&replog_c->lists.slink_c);
+ ClearReplDevelStats(replog_c);
+ ClearReplBlocked(replog_c);
+ spin_lock_init(&io->in_lock);
+ bio_list_init(&io->in);
+ atomic_set(&io->in_flight, 0);
+ INIT_WORK(&io->ws, do_repl);
+ stats_reset(&io->stats);
+ init_waitqueue_head(&io->waiters);
+ return replog_c;
+}
+
+/* Create replog_c context. */
+static struct replog_c *
+replog_c_create(struct dm_target *ti, struct dm_repl_log *replog)
+{
+ dev_t replicator_dev;
+ struct replog_c *replog_c, *replog_c_tmp;
+
+ /* Get replicator control device major:minor. */
+ replicator_dev = get_ctrl_dev(ti);
+
+ /* Allcate and init replog_c object. */
+ replog_c = replog_c_alloc();
+ if (IS_ERR(replog_c))
+ return replog_c;
+
+ /* Add to global replog_c list. */
+ mutex_lock(&replog_c_list_mutex);
+ replog_c_tmp = replog_c_get_by_dev(replicator_dev);
+ if (likely(IS_ERR(replog_c_tmp))) {
+ /* We won any potential race. */
+ /* Set replog global I/O callback and context. */
+ replog->ops->io_notify_fn_set(replog, io_callback,
+ replog_c);
+ replog_c->dev = replicator_dev;
+ replog_c->ti = ti;
+ replog_c->replog = replog;
+ list_add_tail(&replog_c->lists.replog_c,
+ &replog_c_list);
+ mutex_unlock(&replog_c_list_mutex);
+ } else {
+ /* Lost a potential race. */
+ mutex_unlock(&replog_c_list_mutex);
+
+ destroy_workqueue(replog_c->io.wq);
+ kfree(replog_c);
+ replog_c = replog_c_tmp;
+ }
+
+ return replog_c;
+}
+
+/* Find dc on slink_c list by dev_nr. */
+static struct device_c *
+device_c_find(struct slink_c *slink_c, unsigned dev_nr)
+{
+ struct device_c *dc;
+
+ list_for_each_entry(dc, &slink_c->lists.dc, list) {
+ if (dev_nr == dc->number)
+ return dc;
+ }
+
+ return ERR_PTR(-ENOENT);
+}
+
+/* Get a reference on an slink_c by slink reference. */
+static struct slink_c *
+slink_c_get(struct slink_c *slink_c)
+{
+ kref_get(&slink_c->ref);
+ return slink_c;
+}
+
+/* Find an slink_c by slink number on the replog slink list. */
+static struct slink_c *
+slink_c_get_by_number(struct replog_c *replog_c, int slink_nr)
+{
+ struct slink_c *slink_c;
+
+ list_for_each_entry(slink_c, &replog_c->lists.slink_c, lists.slink_c) {
+ int slink_nr_tmp =
+ slink_c->slink->ops->slink_number(slink_c->slink);
+
+ if (slink_nr == slink_nr_tmp)
+ return slink_c_get(slink_c);
+ }
+
+ return ERR_PTR(-ENOENT);
+}
+
+/* Site link constructor helper to create a slink_c object. */
+static struct slink_c *
+slink_c_create(struct replog_c *replog_c, struct dm_repl_slink *slink)
+{
+ int r, slink_nr = slink->ops->slink_number(slink);
+ struct slink_c *slink_c, *slink_c_tmp;
+ struct dm_repl_log *replog = replog_c->replog;
+
+ BUG_ON(slink_nr < 0);
+ DMDEBUG("%s creating slink_c for site link=%d", __func__, slink_nr);
+
+ slink_c = kzalloc(sizeof(*slink_c), GFP_KERNEL);
+ if (unlikely(!slink_c))
+ return ERR_PTR(-ENOMEM);
+
+ r = replog->ops->slink_add(replog, slink);
+ if (unlikely(r < 0)) {
+ kfree(slink_c);
+ return ERR_PTR(r);
+ }
+
+ DMDEBUG("%s added site link=%d", __func__,
+ slink->ops->slink_number(slink));
+
+ kref_init(&slink_c->ref);
+ INIT_LIST_HEAD(&slink_c->lists.dc);
+ slink_c->replog_c = replog_c;
+ slink_c->slink = slink;
+
+ /* Check creation race and add to per replog_c slink_c list. */
+ mutex_lock(&replog_c_list_mutex);
+ slink_c_tmp = slink_c_get_by_number(replog_c, slink_nr);
+ if (likely(IS_ERR(slink_c_tmp)))
+ list_add_tail(&slink_c->lists.slink_c,
+ &replog_c->lists.slink_c);
+ else {
+ kfree(slink_c);
+ slink_c = slink_c_tmp;
+ }
+
+ mutex_unlock(&replog_c_list_mutex);
+ return slink_c;
+}
+
+/*
+ * Release reference on slink_c, removing dc from
+ * it and releasing resources on last drop.
+ */
+static void
+slink_c_release(struct kref *ref)
+{
+ struct slink_c *slink_c = container_of(ref, struct slink_c, ref);
+
+ BUG_ON(!list_empty(&slink_c->lists.dc));
+ kfree(slink_c);
+}
+
+/*
+ * Release reference on slink_c, removing dc from
+ * it and releasing resources on last drop.
+ */
+static int
+slink_c_put(struct slink_c *slink_c)
+{
+ return kref_put(&slink_c->ref, slink_c_release);
+}
+
+/* Either set ti->error or call DMERR() depending on ctr call type. */
+enum ctr_call_type { CTR_CALL, MESSAGE_CALL };
+static void
+ti_or_dmerr(enum ctr_call_type call_type, struct dm_target *ti, char *msg)
+{
+ if (call_type == CTR_CALL)
+ ti->error = msg;
+ else
+ DMERR("%s", msg);
+}
+
+/*
+ * Check, if @str is listed on variable (const char *) list of strings.
+ *
+ * Returns 1 for found on list and 0 for failure.
+ */
+static int
+str_listed(const char *str, ...)
+{
+ int r = 0;
+ const char *s;
+ va_list str_list;
+
+ va_start(str_list, str);
+
+ while ((s = va_arg(str_list, const char *))) {
+ if (!strncmp(str, s, strlen(str))) {
+ r = 1;
+ break;
+ }
+ }
+
+ va_end(str_list);
+ return r;
+}
+
+/*
+ * Worker thread.
+ *
+ * o work on all new queued bios io'ing them to the REPLOG
+ * o break out if replog reports -EWOULDBLOCK until called back
+ */
+static void
+do_repl(struct work_struct *ws)
+{
+ struct replog_c *replog_c = container_of(ws, struct replog_c, io.ws);
+ struct dm_repl_log *replog = replog_c->replog;
+ struct bio *bio;
+ struct bio_list ios;
+
+ _BUG_ON_PTR(replog);
+
+ if (ReplBlocked(replog_c))
+ return;
+
+ bio_list_init(&ios);
+
+ /* Quickly grab all (new) input bios queued. */
+ spin_lock(&replog_c->io.in_lock);
+ bio_list_merge(&ios, &replog_c->io.in);
+ bio_list_init(&replog_c->io.in);
+ spin_unlock(&replog_c->io.in_lock);
+
+ /* Work all deferred or new bios on work list. */
+ while ((bio = bio_list_pop(&ios))) {
+ int r = replog->ops->io(replog, bio, 0);
+
+ if (r == -EWOULDBLOCK) {
+ SetReplBlocked(replog_c);
+ DMDEBUG_LIMIT("%s SetReplBlocked", __func__);
+
+ /* Push non-processed bio back to the work list. */
+ bio_list_push(&ios, bio);
+
+ /*
+ * Merge non-processed bios
+ * back to the input list head.
+ */
+ spin_lock(&replog_c->io.in_lock);
+ bio_list_merge_head(&replog_c->io.in, &ios);
+ spin_unlock(&replog_c->io.in_lock);
+
+ break;
+ } else
+ BUG_ON(r);
+ }
+}
+
+/* Replication congested function. */
+static int
+repl_congested(void *congested_data, int bdi_bits)
+{
+ int r;
+ struct device_c *dc = congested_data;
+ struct replog_c *replog_c;
+
+ _BUG_ON_PTR(dc);
+ _BUG_ON_PTR(dc->slink_c);
+ replog_c = dc->slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+ r = !!ReplBlocked(replog_c);
+ atomic_inc(&replog_c->io.stats.congested_fn[r]);
+ return r;
+}
+
+/* Set backing device congested function of a local replicated device. */
+static void
+dc_set_bdi(struct device_c *dc)
+{
+ struct mapped_device *md = dm_table_get_md(dc->ti->table);
+ struct backing_dev_info *bdi = &dm_disk(md)->queue->backing_dev_info;
+
+ /* Set congested function and data. */
+ bdi->congested_fn = repl_congested;
+ bdi->congested_data = dc;
+ dm_put(md);
+}
+
+/* Get device on slink and unlink it from the list of devices. */
+static struct device_c *
+dev_get_del(struct device_c *dc, int slink_nr, struct list_head *dc_list)
+{
+ int dev_nr;
+ struct slink_c *slink_c;
+ struct dm_repl_slink *slink;
+ struct dm_repl_log *replog;
+ struct replog_c *replog_c;
+
+ _BUG_ON_PTR(dc);
+ dev_nr = dc->number;
+ BUG_ON(dev_nr < 0);
+ slink_c = dc->slink_c;
+ _BUG_ON_PTR(slink_c);
+ slink = slink_c->slink;
+ _BUG_ON_PTR(slink);
+ replog_c = slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+ replog = replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ /* Get the slink by number. */
+ slink = slink->ops->slink(replog, slink_nr);
+ if (IS_ERR(slink))
+ return (struct device_c *) slink;
+
+ slink_c = slink_c_get_by_number(replog_c, slink_nr);
+ if (IS_ERR(slink_c))
+ return (struct device_c *) slink_c;
+
+ dc = device_c_find(slink_c, dev_nr);
+ if (IS_ERR(dc))
+ DMERR("No device %d on slink %d", dev_nr, slink_nr);
+ else
+ list_move(&dc->list, dc_list);
+
+ BUG_ON(slink_c_put(slink_c));
+ return dc;
+}
+
+/* Free device and put references. */
+static int
+dev_free_put(struct device_c *dc, int slink_nr)
+{
+ int r;
+ struct slink_c *slink_c;
+ struct dm_repl_slink *slink;
+
+ _BUG_ON_PTR(dc);
+ BUG_ON(dc->number < 0);
+ BUG_ON(slink_nr < 0);
+ slink_c = dc->slink_c;
+ _BUG_ON_PTR(slink_c);
+ slink = slink_c->slink;
+ _BUG_ON_PTR(slink);
+
+ /* Delete device from slink. */
+ r = slink->ops->dev_del(slink, dc->number);
+ if (r < 0) {
+ DMERR("Error %d deleting device %d from "
+ "site link %d", r, dc->number, slink_nr);
+ } else
+ /* Drop reference on replicator control device. */
+ dm_put_device(dc->ti, dc->replicator_dev);
+
+ kfree(dc);
+
+ if (!r)
+ /* Drop reference on slink_c, freeing it on last one. */
+ BUG_ON(slink_c_put(slink_c));
+
+ return r;
+}
+
+/*
+ * Replication device "replicator-dev" destructor method.
+ *
+ * Either on slink0 in case slink_nr == 0 for mapped devices;
+ * the whole chain of LD + its RDs will be deleted
+ * -or-
+ * on slink > 0 in case of message interface calls (just one RD)
+ */
+static int
+_replicator_dev_dtr(struct dm_target *ti, int slink_nr)
+{
+ int r;
+ struct device_c *dc = ti->private, *dc_tmp, *dc_n;
+ struct slink_c *slink_c, *slink_c_n;
+ struct replog_c *replog_c;
+ struct dm_repl_slink *slink;
+ struct list_head dc_list;
+
+ BUG_ON(slink_nr < 0);
+ _BUG_ON_PTR(dc);
+ INIT_LIST_HEAD(&dc_list);
+ slink_c = dc->slink_c;
+ _BUG_ON_PTR(slink_c);
+ replog_c = slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+
+ /* First pull device out on all slinks holding lock. */
+ mutex_lock(&replog_c_list_mutex);
+ /* Call from message interface wih slink_nr > 0. */
+ if (slink_nr)
+ dev_get_del(dc, slink_nr, &dc_list);
+ else {
+ /* slink number 0 -> delete LD and any RDs. */
+ list_for_each_entry_safe(slink_c, slink_c_n,
+ &replog_c->lists.slink_c,
+ lists.slink_c) {
+ slink = slink_c->slink;
+ _BUG_ON_PTR(slink);
+ slink_nr = slink->ops->slink_number(slink);
+ BUG_ON(slink_nr < 0);
+ dev_get_del(dc, slink_nr, &dc_list);
+ }
+ }
+
+ mutex_unlock(&replog_c_list_mutex);
+
+ r = !list_empty(&dc_list);
+
+ /* Now delete devices on pulled out list. */
+ list_for_each_entry_safe(dc_tmp, dc_n, &dc_list, list) {
+ slink = dc_tmp->slink_c->slink;
+ dev_free_put(dc_tmp, slink->ops->slink_number(slink));
+ }
+
+ ti->private = NULL;
+ return r;
+}
+
+/* Replicator device destructor. Autodestructs devices on slink > 0. */
+static void
+replicator_dev_dtr(struct dm_target *ti)
+{
+ _replicator_dev_dtr(ti, 0); /* Slink 0 device destruction. */
+}
+
+/* Construct a local/remote device. */
+/*
+ * slink_nr dev_nr dev_path dirty_log_params
+ *
+ * [0 1 /dev/mapper/local_device \ # local device being replicated
+ * nolog 0]{1..N} # no dirty log with local devices
+ */
+#define MIN_DEV_ARGS 5
+static int
+device_ctr(enum ctr_call_type call_type, struct dm_target *ti,
+ struct replog_c *replog_c,
+ const char *replicator_path, unsigned dev_nr,
+ unsigned argc, char **argv, int *args_used)
+{
+ int dev_params, dirtylog_params, params, r, slink_nr;
+ struct dm_repl_slink *slink; /* Site link handle. */
+ struct slink_c *slink_c; /* Site link context. */
+ struct device_c *dc; /* Replication device context. */
+
+ SHOW_ARGV;
+
+ if (argc < MIN_DEV_ARGS) {
+ ti_or_dmerr(call_type, ti, "Not enough device arguments");
+ return -EINVAL;
+ }
+
+ /* Get slink number. */
+ params = 0;
+ if (unlikely(sscanf(argv[params], "%d", &slink_nr) != 1 ||
+ slink_nr < 0)) {
+ ti_or_dmerr(call_type, ti,
+ "Invalid site link number argument");
+ return -EINVAL;
+ }
+
+ /* Get #dev_params. */
+ params++;
+ if (unlikely(sscanf(argv[params], "%d", &dev_params) != 1 ||
+ dev_params < 0 ||
+ dev_params + 4 > argc)) {
+ ti_or_dmerr(call_type, ti,
+ "Invalid device parameter number argument");
+ return -EINVAL;
+ }
+
+ /* Get #dirtylog_params. */
+ params += dev_params + 2;
+ if (unlikely(sscanf(argv[params], "%d", &dirtylog_params) != 1 ||
+ dirtylog_params < 0 ||
+ params + dirtylog_params + 1 > argc)) {
+ ti_or_dmerr(call_type, ti,
+ "Invalid dirtylog parameter number argument");
+ return -EINVAL;
+ }
+
+ /* Check that all parameters are sane. */
+ params = dev_params + dirtylog_params + 3;
+ if (params > argc) {
+ ti_or_dmerr(call_type, ti,
+ "Invalid device/dirtylog argument count");
+ return -EINVAL;
+ }
+
+ /* Get SLINK handle. */
+ mutex_lock(&replog_c_list_mutex);
+ slink_c = slink_c_get_by_number(replog_c, slink_nr);
+ mutex_unlock(&replog_c_list_mutex);
+
+ if (unlikely(IS_ERR(slink_c))) {
+ ti_or_dmerr(call_type, ti, "Cannot find site link context");
+ return -ENOENT;
+ }
+
+ slink = slink_c->slink;
+ _BUG_ON_PTR(slink);
+
+ /* Allocate replication context for new device. */
+ dc = kzalloc(sizeof(*dc), GFP_KERNEL);
+ if (unlikely(!dc)) {
+ ti_or_dmerr(call_type, ti, "Cannot allocate device context");
+ BUG_ON(slink_c_put(slink_c));
+ return -ENOMEM;
+ }
+
+ INIT_LIST_HEAD(&dc->list);
+ dc->slink_c = slink_c;
+ dc->ti = ti;
+
+ /*
+ * Get reference on replicator control device.
+ *
+ * Dummy start/size sufficient here.
+ */
+ r = dm_get_device(ti, replicator_path, 0, 1,
+ FMODE_WRITE, &dc->replicator_dev);
+ if (unlikely(r < 0)) {
+ ti_or_dmerr(call_type, ti,
+ "Can't access replicator control device");
+ goto err_slink_put;
+ }
+
+ /* Add device to slink. */
+ /*
+ * ti->split_io for all local devices must be set
+ * to the unique region_size of the remote devices.
+ */
+ r = slink->ops->dev_add(slink, dev_nr, ti, params, argv + 1);
+ if (unlikely(r < 0)) {
+ ti_or_dmerr(call_type, ti, r == -EEXIST ?
+ "device already in use on site link" :
+ "Failed to add device to site link");
+ goto err_device_put;
+ }
+
+ dc->number = r;
+
+ /* Only set bdi properties on local devices. */
+ if (!slink_nr) {
+ /* Preset, will be set to region size in the slink code. */
+ ti->split_io = DM_REPL_MIN_SPLIT_IO;
+
+ /*
+ * Init ti reference on slink0 devices only,
+ * because they only have a local mapping!
+ */
+ ti->private = dc;
+ dc_set_bdi(dc);
+ }
+
+ /* Add rc to slink_c list. */
+ mutex_lock(&replog_c_list_mutex);
+ list_add_tail(&dc->list, &slink_c->lists.dc);
+ mutex_unlock(&replog_c_list_mutex);
+
+ *args_used = dev_params + dirtylog_params + 4;
+ DMDEBUG("%s added device=%d to site link=%u", __func__,
+ r, slink->ops->slink_number(slink));
+ return 0;
+
+err_device_put:
+ dm_put_device(ti, dc->replicator_dev);
+err_slink_put:
+ BUG_ON(slink_c_put(slink_c));
+ kfree(dc);
+ return r;
+}
+
+/*
+ * Replication device "replicator-dev" constructor method.
+ *
+ * <start> <length> replicator-dev
+ * <replicator_device> <dev_nr> \
+ * [<slink_nr> <#dev_params> <dev_params>
+ * <dlog_type> <#dlog_params> <dlog_params>]{1..N}
+ *
+ * <replicator_device> = device previously constructed via "replication" target
+ * <dev_nr> = An integer that is used to 'tag' write requests as
+ * belonging to a particular set of devices - specifically,
+ * the devices that follow this argument (i.e. the site
+ * link devices).
+ * <slink_nr> = This number identifies the site/location where the next
+ * device to be specified comes from. It is exactly the
+ * same number used to identify the site/location (and its
+ * policies) in the "replicator" target. Interestingly,
+ * while one might normally expect a "dev_type" argument
+ * here, it can be deduced from the site link number and
+ * the 'slink_type' given in the "replication" target.
+ * <#dev_params> = '1' (The number of allowed parameters actually depends
+ * on the 'slink_type' given in the "replication" target.
+ * Since our only option there is "blockdev", the only
+ * allowable number here is currently '1'.)
+ * <dev_params> = 'dev_path' (Again, since "blockdev" is the only
+ * 'slink_type' available, the only allowable argument here
+ * is the path to the device.)
+ * <dlog_type> = Not to be confused with the "replicator log", this is
+ * the type of dirty log associated with this particular
+ * device. Dirty logs are used for synchronization, during
+ * initialization or fall behind conditions, to bring devices
+ * into a coherent state with its peers - analogous to
+ * rebuilding a RAID1 (mirror) device. Available dirty
+ * log types include: 'nolog', 'core', and 'disk'
+ * <#dlog_params> = The number of arguments required for a particular log
+ * type - 'nolog' = 0, 'core' = 1/2, 'disk' = 2/3.
+ * <dlog_params> = 'nolog' => ~no arguments~
+ * 'core' => <region_size> [sync | nosync]
+ * 'disk' => <dlog_dev_path> <region_size> [sync | nosync]
+ * <region_size> = This sets the granularity at which the dirty log
+ * tracks what areas of the device is in-sync.
+ * [sync | nosync] = Optionally specify whether the sync should be forced
+ * or avoided initially.
+ */
+#define LOG_ARGS 2
+#define DEV_MIN_ARGS 5
+static int
+_replicator_dev_ctr(enum ctr_call_type call_type, struct dm_target *ti,
+ unsigned argc, char **argv)
+{
+ int args_used, r, tmp;
+ unsigned dev_nr;
+ char *replicator_path = argv[0];
+ struct dm_dev *ctrl_dev;
+ struct replog_c *replog_c;
+
+ SHOW_ARGV;
+
+ if (argc < LOG_ARGS + DEV_MIN_ARGS)
+ goto err_args;
+
+ /*
+ * Get reference on replicator control device.
+ *
+ * Dummy start/size sufficient here.
+ */
+ r = dm_get_device(ti, replicator_path, 0, 1, FMODE_WRITE, &ctrl_dev);
+ if (unlikely(r < 0)) {
+ ti_or_dmerr(CTR_CALL, ti,
+ "Can't access replicator control device");
+ return r;
+ }
+
+ if (sscanf(argv[1], "%d", &tmp) != 1 ||
+ tmp < 0) {
+ dm_put_device(ti, ctrl_dev);
+ ti_or_dmerr(call_type, ti, "Invalid device number argument");
+ return -EINVAL;
+ }
+
+ dev_nr = tmp;
+
+ /* Find precreated replog context by device, taking out a reference. */
+ mutex_lock(&replog_c_list_mutex);
+ replog_c = replog_c_get_by_dev(ctrl_dev->bdev->bd_dev);
+ mutex_unlock(&replog_c_list_mutex);
+
+ if (unlikely(IS_ERR(replog_c))) {
+ dm_put_device(ti, ctrl_dev);
+ ti_or_dmerr(call_type, ti, "Failed to find replication log");
+ return PTR_ERR(replog_c);
+ }
+
+ _BUG_ON_PTR(replog_c->replog);
+ argc -= LOG_ARGS;
+ argv += LOG_ARGS;
+
+ /*
+ * Iterate all slinks/rds if multiple device/dirty
+ * log tuples present on mapping table line.
+ */
+ while (argc >= DEV_MIN_ARGS) {
+ /* Create slink+device context. */
+ r = device_ctr(call_type, ti, replog_c, replicator_path,
+ dev_nr, argc, argv, &args_used);
+ if (unlikely(r))
+ goto device_ctr_err;
+
+ BUG_ON(args_used > argc);
+ argc -= args_used;
+ argv += args_used;
+ }
+
+ /* All arguments consumed? */
+ if (argc) {
+ r = -EINVAL;
+ goto invalid_args;
+ }
+
+ /* Drop initially taken replog reference. */
+ BUG_ON(replog_c_put(replog_c));
+ dm_put_device(ti, ctrl_dev);
+ return 0;
+
+invalid_args:
+ ti_or_dmerr(call_type, ti, "Invalid device arguments");
+device_ctr_err:
+ /* Drop the initially taken replog reference. */
+ BUG_ON(replog_c_put(replog_c));
+ dm_put_device(ti, ctrl_dev);
+
+ /* If we get an error in ctr -> tear down. */
+ if (call_type == CTR_CALL)
+ replicator_dev_dtr(ti);
+
+ return r;
+
+err_args:
+ ti_or_dmerr(call_type, ti, "Not enough device arguments");
+ return -EINVAL;
+}
+
+/* Constructor method. */
+static int
+replicator_dev_ctr(struct dm_target *ti, unsigned argc, char **argv)
+{
+ return _replicator_dev_ctr(CTR_CALL, ti, argc, argv);
+}
+
+/* Device flush method. */
+static void
+replicator_dev_flush(struct dm_target *ti)
+{
+ struct device_c *dc = ti->private;
+ struct dm_repl_log *replog;
+
+ _BUG_ON_PTR(dc);
+ _BUG_ON_PTR(dc->slink_c);
+ _BUG_ON_PTR(dc->slink_c->replog_c);
+ replog = dc->slink_c->replog_c->replog;
+ _BUG_ON_PTR(replog);
+ BUG_ON(!replog->ops->flush);
+ replog->ops->flush(replog);
+}
+
+/* Queues bios to the cache and wakes up worker thread. */
+static inline void
+queue_bio(struct device_c *dc, struct bio *bio)
+{
+ struct replog_c *replog_c = dc->slink_c->replog_c;
+
+ atomic_inc(replog_c->io.stats.io + !!(bio_data_dir(bio) == WRITE));
+
+ spin_lock(&replog_c->io.in_lock);
+ bio_list_add(&replog_c->io.in, bio);
+ replog_c_io_get(replog_c);
+ spin_unlock(&replog_c->io.in_lock);
+
+ /* Wakeup worker to deal with bio input list. */
+ wake_do_repl(replog_c);
+}
+
+/*
+ * Map a replicated device io by handling it in the worker
+ * thread in order to avoid delays in the fast path.
+ */
+static int
+replicator_dev_map(struct dm_target *ti, struct bio *bio,
+ union map_info *map_context)
+{
+ map_context->ptr = bio->bi_private;
+ bio->bi_sector -= ti->begin; /* Remap sector to target begin. */
+ queue_bio(ti->private, bio); /* Queue bio to the worker. */
+ return DM_MAPIO_SUBMITTED; /* Handle later. */
+}
+
+
+/* Replication device suspend/resume helper. */
+enum suspend_resume_type { POSTSUSPEND, RESUME };
+static void
+_replicator_dev_suspend_resume(struct dm_target *ti,
+ enum suspend_resume_type type)
+{
+ struct device_c *dc = ti->private;
+ struct replog_c *replog_c;
+ struct slink_c *slink_c, *n;
+ int dev_nr = dc->number, slinks = 0;
+
+ DMDEBUG("%s %s", __func__, type == RESUME ? "resume" : "postsusend");
+ _BUG_ON_PTR(dc);
+ _BUG_ON_PTR(dc->slink_c);
+ replog_c = dc->slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+ BUG_ON(dev_nr < 0);
+
+ /* Suspend/resume device on all slinks. */
+ list_for_each_entry_safe(slink_c, n, &replog_c->lists.slink_c,
+ lists.slink_c) {
+ int r;
+ struct dm_repl_slink *slink = slink_c->slink;
+
+ _BUG_ON_PTR(slink);
+
+ r = type == RESUME ?
+ slink->ops->resume(slink, dev_nr) :
+ slink->ops->postsuspend(slink, dev_nr);
+ if (r < 0)
+ DMERR("Error %d %s device=%d on site link %u",
+ r, type == RESUME ?
+ "resuming" : "postsuspending",
+ dev_nr, slink->ops->slink_number(slink));
+ else
+ slinks++;
+ }
+
+ if (type == RESUME && slinks)
+ wake_do_repl(replog_c);
+}
+
+/* Replication device post suspend method. */
+static void
+replicator_dev_postsuspend(struct dm_target *ti)
+{
+ _replicator_dev_suspend_resume(ti, POSTSUSPEND);
+}
+
+/* Replicatin device resume method. */
+static void
+replicator_dev_resume(struct dm_target *ti)
+{
+ _replicator_dev_suspend_resume(ti, RESUME);
+}
+
+/* Pass endio calls down to the replicator log if requested. */
+static int
+replicator_dev_endio(struct dm_target *ti, struct bio *bio,
+ int error, union map_info *map_context)
+{
+ int rr, rs;
+ struct device_c *dc = ti->private;
+ struct replog_c *replog_c;
+ struct dm_repl_log *replog;
+ struct dm_repl_slink *slink;
+
+ _BUG_ON_PTR(dc);
+ _BUG_ON_PTR(dc->slink_c);
+ slink = dc->slink_c->slink;
+ replog_c = dc->slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+ replog = dc->slink_c->replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ rr = replog->ops->endio ?
+ replog->ops->endio(replog, bio, error, map_context) : 0;
+ rs = slink->ops->endio ?
+ slink->ops->endio(slink, bio, error, map_context) : 0;
+ replog_c_io_put(replog_c);
+ return rs < 0 ? rs : rr;
+}
+
+/*
+ * Replication device message method.
+ *
+ * Arguments:
+ * device add/del \
+ * 63:4 0 \ # replication log on 63:4 and device number '0'
+ * [0 1 /dev/mapper/local_device \ # local device being replicated
+ * nolog 0]{1..N} # no dirty log with local devices
+ *
+ * start/resume all/device # Resume whole replicator/
+ * # a single device
+ */
+static int
+replicator_dev_message(struct dm_target *ti, unsigned argc, char **argv)
+{
+ int slink_nr;
+ struct device_c *dc = ti->private;
+ struct replog_c *replog_c;
+ struct dm_repl_log *replog;
+
+ SHOW_ARGV;
+
+ _BUG_ON_PTR(dc);
+ _BUG_ON_PTR(dc->slink_c);
+ replog_c = dc->slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+ replog = dc->slink_c->replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ /* Check minimum arguments. */
+ if (unlikely(argc < 1))
+ goto err_args;
+
+ /* Add/delete a device to/from a site link. */
+ if (str_listed(argv[0], "device", NULL)) {
+ if (argc < 2)
+ goto err_args;
+
+ /* We've got the target index of an SLINK0 device here. */
+ if (str_listed(argv[1], "add", NULL))
+ return _replicator_dev_ctr(MESSAGE_CALL, ti,
+ argc - 2, argv + 2);
+ else if (str_listed(argv[1], "del", NULL)) {
+ if (argc < 3)
+ goto err_args;
+
+ if (sscanf(argv[2], "%d", &slink_nr) != 1 ||
+ slink_nr < 1)
+ DM_EINVAL("invalid site link number "
+ "argument; must be > 0");
+
+ return _replicator_dev_dtr(ti, slink_nr);
+ } else
+ DM_EINVAL("invalid device command argument");
+
+ /* Start replication on single device on all slinks. */
+ } else if (str_listed(argv[0], "start", "resume", NULL))
+ replicator_dev_resume(ti);
+
+ /* Stop replication for single device on all slinks. */
+ else if (str_listed(argv[0], "stop", "suspend", "postsuspend", NULL))
+ replicator_dev_postsuspend(ti);
+ else
+ DM_EINVAL("invalid message command");
+
+ return 0;
+
+err_args:
+ DM_EINVAL("too few message arguments");
+}
+
+/* Replication device status output method. */
+static int
+replicator_dev_status(struct dm_target *ti, status_type_t type,
+ char *result, unsigned maxlen)
+{
+ ssize_t sz = 0;
+ static char buffer[2048];
+ struct device_c *dc = ti->private;
+ struct replog_c *replog_c;
+ struct dm_repl_slink *slink;
+
+ mutex_lock(&replog_c_list_mutex);
+ _BUG_ON_PTR(dc);
+ _BUG_ON_PTR(dc->slink_c);
+ slink = dc->slink_c->slink;
+ _BUG_ON_PTR(slink);
+ replog_c = dc->slink_c->replog_c;
+ _BUG_ON_PTR(replog_c);
+
+ DMEMIT("%s %d ", format_dev_t(buffer, replog_c->dev), dc->number);
+ mutex_unlock(&replog_c_list_mutex);
+ slink->ops->status(slink, dc->number, type, buffer, sizeof(buffer));
+ DMEMIT("%s", buffer);
+ return 0;
+}
+
+/* Replicator device interface. */
+static struct target_type replicator_dev_target = {
+ .name = "replicator-dev",
+ .version = {1, 0, 0},
+ .module = THIS_MODULE,
+ .ctr = replicator_dev_ctr,
+ .dtr = replicator_dev_dtr,
+ .flush = replicator_dev_flush,
+ .map = replicator_dev_map,
+ .postsuspend = replicator_dev_postsuspend,
+ .resume = replicator_dev_resume,
+ .end_io = replicator_dev_endio,
+ .message = replicator_dev_message,
+ .status = replicator_dev_status,
+};
+
+
+/*
+ * Replication log destructor.
+ */
+static void
+replicator_dtr(struct dm_target *ti)
+{
+ int r, slink_nr;
+ struct replog_c *replog_c = ti->private;
+ struct dm_repl_log *replog;
+ struct slink_c *slink_c, *n;
+ struct dm_repl_slink *slink;
+
+ _BUG_ON_PTR(replog_c);
+ replog = replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ /* Pull out replog_c to process destruction cleanly. */
+ mutex_lock(&replog_c_list_mutex);
+ list_del_init(&replog_c->lists.replog_c);
+ mutex_unlock(&replog_c_list_mutex);
+
+ /* Put all replog's slink contexts. */
+ list_for_each_entry_safe(slink_c, n, &replog_c->lists.slink_c,
+ lists.slink_c) {
+ list_del_init(&slink_c->lists.slink_c);
+ slink = slink_c->slink;
+ _BUG_ON_PTR(slink);
+ slink_nr = slink->ops->slink_number(slink);
+ r = replog->ops->slink_del(replog, slink);
+ BUG_ON(r < 0);
+ slink_destroy(slink);
+ BUG_ON(replog_c_put(replog_c));
+ BUG_ON(!slink_c_put(slink_c));
+ }
+
+ /* Drop work queue. */
+ destroy_workqueue(replog_c->io.wq);
+
+ /* Drop reference on replog. */
+ repl_log_dtr(replog_c->replog, replog_c->ti);
+
+ BUG_ON(!replog_c_put(replog_c));
+}
+
+/*
+ * Replication constructor helpers.
+ */
+/* Create a site link tying it to the replication log. */
+/*
+ * E.g.: "local 4 1 async ios 10000"
+ */
+#define MIN_SLINK_ARGS 3
+static int
+_replicator_slink_ctr(enum ctr_call_type call_type, struct dm_target *ti,
+ struct replog_c *replog_c,
+ unsigned argc, char **argv, unsigned *args_used)
+{
+ int first_slink, slink_nr, slink_params;
+ struct dm_repl_slink *slink; /* Site link handle. */
+ struct slink_c *slink_c; /* Site link context. */
+
+ SHOW_ARGV;
+
+ if (argc < MIN_SLINK_ARGS)
+ return -EINVAL;
+
+ /* Get #slink_params. */
+ if (unlikely(sscanf(argv[1], "%d", &slink_params) != 1 ||
+ slink_params < 0 ||
+ slink_params + 2 > argc)) {
+ ti_or_dmerr(call_type, ti,
+ "Invalid site link parameter number argument");
+ return -EINVAL;
+ }
+
+ /* Get slink #. */
+ if (unlikely(sscanf(argv[2], "%d", &slink_nr) != 1 ||
+ slink_nr < 0)) {
+ ti_or_dmerr(call_type, ti,
+ "Invalid site link number argument");
+ return -EINVAL;
+ }
+
+ /* Check first slink is slink 0. */
+ mutex_lock(&replog_c_list_mutex);
+ first_slink = !list_first_entry(&replog_c->lists.slink_c,
+ struct slink_c, lists.slink_c);
+ if (first_slink && slink_nr) {
+ mutex_unlock(&replog_c_list_mutex);
+ ti_or_dmerr(call_type, ti, "First site link must be 0");
+ return -EINVAL;
+ }
+
+ slink_c = slink_c_get_by_number(replog_c, slink_nr);
+ mutex_unlock(&replog_c_list_mutex);
+
+ if (!IS_ERR(slink_c)) {
+ ti_or_dmerr(call_type, ti, "slink already existing");
+ BUG_ON(slink_c_put(slink_c));
+ return -EPERM;
+ }
+
+ /* Get SLINK handle. */
+ slink = repl_slink_ctr(argv[0], replog_c->replog,
+ slink_params + 1, argv + 1);
+ if (unlikely(IS_ERR(slink))) {
+ ti_or_dmerr(call_type, ti, "Cannot create site link context");
+ return PTR_ERR(slink);
+ }
+
+ slink_c = slink_c_create(replog_c, slink);
+ if (unlikely(IS_ERR(slink_c))) {
+ ti_or_dmerr(call_type, ti, "Cannot allocate site link context");
+ slink_destroy(slink);
+ return PTR_ERR(slink_c);
+ }
+
+ *args_used = slink_params + 2;
+ DMDEBUG("%s added site link=%d", __func__, slink_nr);
+ return 0;
+}
+
+/*
+ * Construct a replicator mapping to log writes of one or more local mapped
+ * devices in a local replicator log (REPLOG) in order to replicate them to
+ * one or multiple site links (SLINKs) while ensuring write order fidelity.
+ *
+ *******************************
+ *
+ * "replicator" constructor table:
+ *
+ * <start> <length> replicator \
+ * <replog_type> <#replog_params> <replog_params> \
+ * [<slink_type_0> <#slink_params_0> <slink_params_0>]{1..N}
+ *
+ * <replog_type> = "ringbuffer" is currently the only available type
+ * <#replog_params> = # of args intended for the replog (2 or 4)
+ * <replog_params> = <dev_path> <dev_start> [auto/create/open <size>]
+ * <dev_path> = device path of replication log (REPLOG) backing store
+ * <dev_start> = offset to REPLOG header
+ * create = The replication log will be initialized if not active
+ * and sized to "size". (If already active, the create
+ * will fail.) Size is always in sectors.
+ * open = The replication log must be initialized and valid or
+ * the constructor will fail.
+ * auto = If a valid replication log header is found on the
+ * replication device, this will behave like 'open'.
+ * Otherwise, this option behaves like 'create'.
+ *
+ * <slink_type> = "blockdev" is currently the only available type
+ * <#slink_params> = 1/2/4
+ * <slink_params> = <slink_nr> [<slink_policy> [<fall_behind> <N>]]
+ * <slink_nr> = This is a unique number that is used to identify a
+ * particular site/location. '0' is always used to
+ * identify the local site, while increasing integers
+ * are used to identify remote sites.
+ * <slink_policy> = The policy can be either 'sync' or 'async'.
+ * 'sync' means write requests will not return until
+ * the data is on the storage device. 'async' allows
+ * a device to "fall behind"; that is, outstanding
+ * write requests are waiting in the replication log
+ * to be processed for this site, but it is not delaying
+ * the writes of other sites.
+ * <fall_behind> = This field is used to specify how far the user is
+ * willing to allow write requests to this specific site
+ * to "fall behind" in processing before switching to
+ * a 'sync' policy. This "fall behind" threshhold can
+ * be specified in three ways: ios, size, or timeout.
+ * 'ios' is the number of pending I/Os allowed (e.g.
+ * "ios 10000"). 'size' is the amount of pending data
+ * allowed (e.g. "size 200m"). Size labels include:
+ * s (sectors), k, m, g, t, p, and e. 'timeout' is
+ * the amount of time allowed for writes to be
+ * outstanding. Time labels include: s, m, h, and d.
+ */
+#define MIN_CONTROL_ARGS 3
+static int
+replicator_ctr(struct dm_target *ti, unsigned argc, char **argv)
+{
+ int args_used = 0, params, r;
+ struct dm_dev *backing_dev;
+ struct dm_repl_log *replog; /* Replicator log handle. */
+ struct replog_c *replog_c; /* Replication log context. */
+
+ SHOW_ARGV;
+
+ if (unlikely(argc < MIN_CONTROL_ARGS)) {
+ ti->error = "Invalid argument count";
+ return -EINVAL;
+ }
+
+ /* Get # of replog params. */
+ if (unlikely(sscanf(argv[1], "%d", ¶ms) != 1 ||
+ params < 2 ||
+ params + 3 > argc)) {
+ ti->error = "Invalid replicator log parameter number";
+ return -EINVAL;
+ }
+
+ /* Check for site link 0 parameter count. */
+ if (params + 4 > argc) {
+ ti->error = "Invalid replicator site link parameter number";
+ return -EINVAL;
+ }
+
+ /*
+ * Get reference on replicator control device.
+ *
+ * Dummy start/size sufficient here.
+ */
+ r = dm_get_device(ti, argv[2], 0, 1, FMODE_WRITE, &backing_dev);
+ if (unlikely(r < 0)) {
+ ti->error = "Can't access replicator control device";
+ return r;
+ }
+
+
+ /* Lookup replog_c by dev_t. */
+ mutex_lock(&replog_c_list_mutex);
+ replog_c = replog_c_get_by_dev(backing_dev->bdev->bd_dev);
+ mutex_unlock(&replog_c_list_mutex);
+
+ if (unlikely(!IS_ERR(replog_c))) {
+ BUG_ON(replog_c_put(replog_c));
+ dm_put_device(ti, backing_dev);
+ ti->error = "Recreating replication log prohibited";
+ return -EPERM;
+ }
+
+ /* Get a reference on the replication log. */
+ replog = repl_log_ctr(argv[0], ti, params, argv + 1);
+ dm_put_device(ti, backing_dev);
+ if (unlikely(IS_ERR(replog))) {
+ ti->error = "Cannot create replication log context";
+ return PTR_ERR(replog);
+ }
+
+ _BUG_ON_PTR(replog->ops->postsuspend);
+ _BUG_ON_PTR(replog->ops->resume);
+
+ /* Create global replication control context. */
+ replog_c = replog_c_create(ti, replog);
+ if (unlikely(IS_ERR(replog_c))) {
+ ti->error = "Cannot allocate replication device context";
+ return PTR_ERR(replog_c);
+ } else
+ ti->private = replog_c;
+
+ /* Work any slink parameter tupels. */
+ params += 2;
+ BUG_ON(argc < params);
+ argc -= params;
+ argv += params;
+ r = 0;
+
+ while (argc > 0) {
+ r = _replicator_slink_ctr(CTR_CALL, ti, replog_c,
+ argc, argv, &args_used);
+ if (r)
+ break;
+
+ /* Take per site link reference out. */
+ replog_c_get(replog_c);
+
+ BUG_ON(argc < args_used);
+ argc -= args_used;
+ argv += args_used;
+ }
+
+ return r;
+}
+
+/*
+ * Replication log map function.
+ *
+ * No io to replication log device allowed: ignore it
+ * by returning zeroes on read and ignoring writes silently.
+ */
+static int
+replicator_map(struct dm_target *ti, struct bio *bio,
+ union map_info *map_context)
+{
+ /* Readahead of null bytes only wastes buffer cache. */
+ if (bio_rw(bio) == READA)
+ return -EIO;
+ else if (bio_rw(bio) == READ)
+ zero_fill_bio(bio);
+
+ bio_endio(bio, 0);
+ return DM_MAPIO_SUBMITTED; /* Accepted bio, don't make new request. */
+}
+
+/* Replication log suspend/resume helper. */
+static void
+_replicator_suspend_resume(struct replog_c *replog_c,
+ enum suspend_resume_type type)
+{
+ struct dm_repl_log *replog;
+
+ DMDEBUG("%s %s", __func__, type == RESUME ? "resume" : "postsusend");
+ _BUG_ON_PTR(replog_c);
+ replog = replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ /* FIXME: device number not utilized yet. */
+ switch (type) {
+ case POSTSUSPEND:
+ ClearReplBlocked(replog_c);
+ flush_workqueue(replog_c->io.wq);
+ wait_event(replog_c->io.waiters, !ReplIoInflight(replog_c));
+ replog->ops->postsuspend(replog, -1);
+ break;
+ case RESUME:
+ replog->ops->resume(replog, -1);
+ ClearReplBlocked(replog_c);
+ wake_do_repl(replog_c);
+ break;
+ default:
+ BUG();
+ }
+}
+
+
+/* Suspend/Resume all. */
+static void
+_replicator_suspend_resume_all(struct replog_c *replog_c,
+ enum suspend_resume_type type)
+{
+ struct device_c *dc;
+ struct slink_c *slink_c0;
+
+ _BUG_ON_PTR(replog_c);
+
+ /* First entry on replog_c slink_c list is slink0. */
+ slink_c0 = list_first_entry(&replog_c->lists.slink_c,
+ struct slink_c, lists.slink_c);
+ _BUG_ON_PTR(slink_c0);
+
+ /* Walk all slink device_c dc and resume slinks. */
+ if (type == RESUME)
+ list_for_each_entry(dc, &slink_c0->lists.dc, list)
+ _replicator_dev_suspend_resume(dc->ti, type);
+
+ _replicator_suspend_resume(replog_c, type);
+
+ /* Walk all slink device_c dc and resume slinks. */
+ if (type == POSTSUSPEND)
+ list_for_each_entry(dc, &slink_c0->lists.dc, list)
+ _replicator_dev_suspend_resume(dc->ti, type);
+}
+
+/* Replication control post suspend method. */
+static void
+replicator_postsuspend(struct dm_target *ti)
+{
+ _replicator_suspend_resume(ti->private, POSTSUSPEND);
+}
+
+/* Replication control resume method. */
+static void
+replicator_resume(struct dm_target *ti)
+{
+ _replicator_suspend_resume(ti->private, RESUME);
+}
+
+/*
+ * Replication log message method.
+ *
+ * Arguments: start/resume/stop/suspend/statistics/replog
+ */
+static int
+_replicator_slink_message(struct dm_target *ti, int argc, char **argv)
+{
+ int args_used, r, tmp;
+ unsigned slink_nr;
+ struct replog_c *replog_c = ti->private;
+ struct dm_repl_slink *slink;
+ struct slink_c *slink_c;
+
+ if (sscanf(argv[2], "%d", &tmp) != 1 || tmp < 1)
+ DM_EINVAL("site link number invalid");
+
+ slink_nr = tmp;
+
+ if (str_listed(argv[1], "add", "del", NULL) &&
+ !slink_nr)
+ DM_EPERM("Can't add/delete site link 0 via message");
+
+ mutex_lock(&replog_c_list_mutex);
+ slink_c = slink_c_get_by_number(replog_c, slink_nr);
+ mutex_unlock(&replog_c_list_mutex);
+
+ if (str_listed(argv[1], "add", NULL)) {
+ if (IS_ERR(slink_c)) {
+ r = _replicator_slink_ctr(MESSAGE_CALL, ti,
+ replog_c,
+ argc - 2, argv + 2,
+ &args_used);
+ if (r)
+ DMERR("Error creating site link");
+
+ return r;
+ } else {
+ BUG_ON(slink_c_put(slink_c));
+ DM_EPERM("site link already exists");
+ }
+ } else if (str_listed(argv[1], "del", NULL)) {
+ if (IS_ERR(slink_c))
+ DM_EPERM("site link doesn't exist");
+ else {
+ if (!list_empty(&slink_c->lists.dc)) {
+ slink_c_put(slink_c);
+ DM_EPERM("site link still has devices");
+ }
+
+ slink_c_put(slink_c);
+ r = slink_c_put(slink_c);
+ if (!r)
+ DMERR("site link still exists (race)!");
+
+ return r;
+ }
+ } else if (str_listed(argv[1], "message", NULL)) {
+ slink = slink_c->slink;
+ _BUG_ON_PTR(slink);
+
+ if (slink->ops->message)
+ return slink->ops->message(slink,
+ argc - 2, argv + 2);
+ else
+ DM_EPERM("no site link message interface");
+ }
+
+ return r;
+}
+
+static int
+replicator_message(struct dm_target *ti, unsigned argc, char **argv)
+{
+ int r, resume, suspend;
+ struct replog_c *replog_c = ti->private;
+ struct dm_repl_log *replog;
+
+ SHOW_ARGV;
+ _BUG_ON_PTR(replog_c);
+ replog = replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ /* Check minimum arguments. */
+ if (unlikely(argc < 1))
+ goto err_args;
+
+ resume = str_listed(argv[0], "resume", "start", NULL);
+ /* Hrm, bogus: need a NULL end arg to make it work!? */
+ suspend = !resume &&
+ str_listed(argv[0], "suspend", "postsuspend", "stop", NULL);
+
+ /*
+ * Start/resume replicaton log or
+ * start/resume it and all slinks+devices.
+ */
+ if (suspend || resume) {
+ int all;
+
+ if (!range_ok(argc, 1, 2)) {
+ DMERR("Invalid suspend/resume argument count");
+ return -EINVAL;
+ }
+
+ all = (argc == 2 && str_listed(argv[1], "all", NULL));
+
+ if (resume) {
+ if (all)
+ _replicator_suspend_resume_all(replog_c,
+ RESUME);
+ else
+ _replicator_suspend_resume(replog_c,
+ RESUME);
+
+ /* Stop replication log. */
+ } else {
+ if (all) {
+ _replicator_suspend_resume_all(replog_c,
+ POSTSUSPEND);
+ } else
+ _replicator_suspend_resume(replog_c,
+ POSTSUSPEND);
+ }
+
+ /* Site link message. */
+ } else if (str_listed(argv[0], "slink", NULL)) {
+ /* E.g.: "local 4 1 async ios 10000" */
+ /* Check minimum arguments. */
+ if (unlikely(argc < 3))
+ goto err_args;
+
+ r = _replicator_slink_message(ti, argc, argv);
+ if (r)
+ return r;
+ /* Statistics. */
+ } else if (str_listed(argv[0], "statistics", NULL)) {
+ if (argc != 2)
+ DM_EINVAL("too many message arguments");
+
+ _BUG_ON_PTR(replog_c);
+ if (str_listed(argv[1], "on", NULL))
+ SetReplDevelStats(replog_c);
+ else if (str_listed(argv[1], "off", NULL))
+ ClearReplDevelStats(replog_c);
+ else if (str_listed(argv[1], "reset", NULL))
+ stats_reset(&replog_c->io.stats);
+
+ /* Replication log message. */
+ } else if (str_listed(argv[0], "replog", NULL)) {
+ if (argc < 2)
+ goto err_args;
+
+ if (replog->ops->message)
+ return replog->ops->message(replog, argc - 1, argv + 1);
+ else
+ DM_EPERM("no replication log message interface");
+ } else
+ DM_EINVAL("invalid message received");
+
+ return 0;
+
+err_args:
+ DM_EINVAL("too few message arguments");
+}
+
+/* Replication log status output method. */
+static int
+replicator_status(struct dm_target *ti, status_type_t type,
+ char *result, unsigned maxlen)
+{
+ unsigned dev_nr = 0;
+ ssize_t sz = 0;
+ static char buffer[2048];
+ struct replog_c *replog_c = ti->private;
+ struct dm_repl_log *replog;
+ struct slink_c *slink_c0;
+ struct dm_repl_slink *slink;
+
+ mutex_lock(&replog_c_list_mutex);
+ _BUG_ON_PTR(replog_c);
+ replog = replog_c->replog;
+ _BUG_ON_PTR(replog);
+
+ if (type == STATUSTYPE_INFO) {
+ if (ReplDevelStats(replog_c)) {
+ struct stats *s = &replog_c->io.stats;
+
+ DMEMIT("v=%s r=%u w=%u rs=%u "
+ "ws=%u nc=%u c=%u ",
+ version,
+ atomic_read(s->io), atomic_read(s->io + 1),
+ atomic_read(s->submitted_io),
+ atomic_read(s->submitted_io + 1),
+ atomic_read(s->congested_fn),
+ atomic_read(s->congested_fn + 1));
+ }
+ }
+
+ mutex_unlock(&replog_c_list_mutex);
+
+ /* Get status from replog. */
+ /* FIXME: dev_nr superfluous? */
+ replog->ops->status(replog, dev_nr, type, buffer, sizeof(buffer));
+ DMEMIT("%s", buffer);
+
+ slink_c0 = list_first_entry(&replog_c->lists.slink_c,
+ struct slink_c, lists.slink_c);
+ slink = slink_c0->slink;
+ _BUG_ON_PTR(slink);
+ /* Get status from slink. */
+ *buffer = 0;
+ slink->ops->status(slink, -1, type, buffer, sizeof(buffer));
+ DMEMIT(" %s", buffer);
+ return 0;
+}
+
+/* Replicator control interface. */
+static struct target_type replicator_target = {
+ .name = "replicator",
+ .version = {1, 0, 0},
+ .module = THIS_MODULE,
+ .ctr = replicator_ctr,
+ .dtr = replicator_dtr,
+ .map = replicator_map,
+ .postsuspend = replicator_postsuspend,
+ .resume = replicator_resume,
+ .message = replicator_message,
+ .status = replicator_status,
+};
+
+static int __init dm_repl_init(void)
+{
+ int r;
+
+ INIT_LIST_HEAD(&replog_c_list);
+ mutex_init(&replog_c_list_mutex);
+
+ r = dm_register_target(&replicator_target);
+ if (r < 0)
+ DMERR("failed to register %s %s [%d]",
+ replicator_target.name, version, r);
+ else {
+ DMINFO("registered %s target %s",
+ replicator_target.name, version);
+ r = dm_register_target(&replicator_dev_target);
+ if (r < 0) {
+ DMERR("Failed to register %s %s [%d]",
+ replicator_dev_target.name, version, r);
+ dm_unregister_target(&replicator_target);
+ } else
+ DMINFO("registered %s target %s",
+ replicator_dev_target.name, version);
+ }
+
+ return r;
+}
+
+static void __exit
+dm_repl_exit(void)
+{
+ dm_unregister_target(&replicator_dev_target);
+ DMINFO("unregistered target %s %s",
+ replicator_dev_target.name, version);
+ dm_unregister_target(&replicator_target);
+ DMINFO("unregistered target %s %s", replicator_target.name, version);
+}
+
+/* Module hooks */
+module_init(dm_repl_init);
+module_exit(dm_repl_exit);
+
+MODULE_DESCRIPTION(DM_NAME " remote replication target");
+MODULE_AUTHOR("Heinz Mauelshagen <heinzm@redhat.com>");
+MODULE_LICENSE("GPL");
diff --git 2.6.33-rc1.orig/drivers/md/dm-repl.h 2.6.33-rc1/drivers/md/dm-repl.h
new file mode 100644
index 0000000..20d0c99
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-repl.h
@@ -0,0 +1,127 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * This file is released under the GPL.
+ */
+
+/*
+ * API calling convention to create a replication mapping:
+ *
+ * 1. get a replicator log handle, hence creating a new persistent
+ * log or accessing an existing one
+ * 2. get an slink handle, hence creating a new transient
+ * slink or accessing an existing one
+ * 2(cont). repeat the previous step for multiple slinks (eg. one for
+ * local and one for remote device access)
+ * 3. bind a (remote) device to a particlar slink created in a previous step
+ * 3(cont). repeat the device binding for any additional devices on that slink
+ * 4. bind the created slink which has device(s) bound to it to the replog
+ * 4(cont). repeat the slink binding to the replog for all created slinks
+ * 5. call the replog io function for each IO.
+ *
+ * Reverse steps 1-4 to tear a replication mapping down, hence freeing all
+ * transient resources allocated to it.
+ */
+
+#ifndef _DM_REPL_H
+#define _DM_REPL_H
+
+#include <linux/device-mapper.h>
+
+/* FIXME: factor these macros out to dm.h */
+#define STR_LEN(ptr, str) ptr, str, strlen(ptr)
+#define ARRAY_END(a) ((a) + ARRAY_SIZE(a))
+#define range_ok(i, min, max) (i >= min && i <= max)
+
+#define TI_ERR_RET(str, ret) \
+do { \
+ ti->error = DM_MSG_PREFIX ": " str; \
+ return ret; } \
+while (0)
+#define TI_ERR(str) TI_ERR_RET(str, -EINVAL)
+
+#define DM_ERR_RET(ret, x...) do { DMERR(x); return ret; } while (0)
+#define DM_EINVAL(x...) DM_ERR_RET(-EINVAL, x)
+#define DM_EPERM(x...) DM_ERR_RET(-EPERM, x)
+
+/*
+ * Minimum split_io of target to preset for local devices in repl_ctr().
+ * Will be adjusted while constructing (a) remote device(s).
+ */
+#define DM_REPL_MIN_SPLIT_IO BIO_MAX_SECTORS
+
+/* REMOVEME: devel testing. */
+#if 0
+#define SHOW_ARGV \
+ do { \
+ int i; \
+\
+ DMINFO("%s: called with the following args:", __func__); \
+ for (i = 0; i < argc; i++) \
+ DMINFO("%d: %s", i, argv[i]); \
+ } while (0)
+#else
+#define SHOW_ARGV
+#endif
+
+
+/* Factor out to dm-bio-list.h */
+static inline void
+bio_list_push(struct bio_list *bl, struct bio *bio)
+{
+ bio->bi_next = bl->head;
+ bl->head = bio;
+
+ if (!bl->tail)
+ bl->tail = bio;
+}
+
+/* REMOVEME: development */
+#define _BUG_ON_PTR(ptr) \
+ do { \
+ BUG_ON(!ptr); \
+ BUG_ON(IS_ERR(ptr)); \
+ } while (0)
+
+/* Callback function. */
+typedef void
+(*dm_repl_notify_fn)(int read_err, int write_err, void *context);
+
+/*
+ * Macros to access bitfields in the structures io.flags member.
+ * Mixed case naming examples are in the page cache as well.
+ */
+#define DM_BITOPS(name, var, flag) \
+static inline int \
+TestClear ## name(struct var *v) \
+{ return test_and_clear_bit(flag, &v->io.flags); } \
+static inline int \
+TestSet ## name(struct var *v) \
+{ return test_and_set_bit(flag, &v->io.flags); } \
+static inline void \
+Clear ## name(struct var *v) \
+{ clear_bit(flag, &v->io.flags); smp_mb(); } \
+static inline void \
+Set ## name(struct var *v) \
+{ set_bit(flag, &v->io.flags); smp_mb(); } \
+static inline int \
+name(struct var *v) \
+{ return test_bit(flag, &v->io.flags); }
+
+/* FIXME: move to dm core. */
+/* Search routines for descriptor arrays. */
+struct dm_str_descr {
+ const int type;
+ const char *name;
+};
+
+/* Return type for name. */
+extern int
+dm_descr_type(const struct dm_str_descr *descr, unsigned len, const char *name);
+/* Return name for type. */
+extern const char *
+dm_descr_name(const struct dm_str_descr *descr, unsigned len, const int type);
+
+#endif
diff --git 2.6.33-rc1.orig/drivers/md/dm.c 2.6.33-rc1/drivers/md/dm.c
index 3167480..0048958 100644
--- 2.6.33-rc1.orig/drivers/md/dm.c
+++ 2.6.33-rc1/drivers/md/dm.c
@@ -2653,6 +2653,7 @@ struct gendisk *dm_disk(struct mapped_device *md)
{
return md->disk;
}
+EXPORT_SYMBOL_GPL(dm_disk);
struct kobject *dm_kobject(struct mapped_device *md)
{
--
1.6.2.5
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler
2009-12-18 15:44 ` [PATCH v6 2/4] dm-replicator: replication log and site link handler interfaces and main replicator module heinzm
@ 2009-12-18 15:44 ` heinzm
2009-12-18 15:44 ` [PATCH v6 4/4] dm-replicator: blockdev site link handler heinzm
2011-07-18 9:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler Busby
0 siblings, 2 replies; 9+ messages in thread
From: heinzm @ 2009-12-18 15:44 UTC (permalink / raw)
To: dm-devel; +Cc: Heinz Mauelshagen
From: Heinz Mauelshagen <heinzm@redhat.com>
This is the "ringbuffer" type replication log handler module
plugging into the main replicator module.
It abstracts the handling of the log from the main module
allowing it to be log type agnostic. It uses the abstracted
device access logic of the site link module, hence allowing
it to be transport type agnostic.
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Reviewed-by: Jon Brassow <jbrassow@redhat.com>
Tested-by: Jon Brassow <jbrassow@redhat.com>
---
drivers/md/Makefile | 1 +
drivers/md/dm-repl-log-ringbuffer.c | 5000 +++++++++++++++++++++++++++++++++++
2 files changed, 5001 insertions(+), 0 deletions(-)
create mode 100644 drivers/md/dm-repl-log-ringbuffer.c
diff --git 2.6.33-rc1.orig/drivers/md/Makefile 2.6.33-rc1/drivers/md/Makefile
index 832d547..dcb1f69 100644
--- 2.6.33-rc1.orig/drivers/md/Makefile
+++ 2.6.33-rc1/drivers/md/Makefile
@@ -46,6 +46,7 @@ obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o dm-region-hash.o
obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
obj-$(CONFIG_DM_ZERO) += dm-zero.o
obj-$(CONFIG_DM_REPLICATOR) += dm-replicator.o \
+ dm-repl-log-ringbuffer.o \
dm-log.o dm-registry.o
quiet_cmd_unroll = UNROLL $@
diff --git 2.6.33-rc1.orig/drivers/md/dm-repl-log-ringbuffer.c 2.6.33-rc1/drivers/md/dm-repl-log-ringbuffer.c
new file mode 100644
index 0000000..e07ffaa
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-repl-log-ringbuffer.c
@@ -0,0 +1,5000 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Authors: Jeff Moyer (jmoyer@redhat.com)
+ * Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * This file is released under the GPL.
+ *
+ * "default" device-mapper replication log type implementing a ring buffer
+ * for write IOs, which will be copied accross site links to devices.
+ *
+ * A log like this allows for write coalescing enhancements in order
+ * to reduce network traffic at the cost of larger fallbehind windows.
+ */
+
+/*
+ * Locking:
+ * l->io.lock for io (de)queueing / slink manipulation
+ * l->lists.lock for copy contexts moved around lists
+ *
+ * The ringbuffer lock does not need to be held in order to take the io.lock,
+ * but if they are both acquired, the ordering must be as indicated above.
+ */
+
+#include "dm-repl.h"
+#include "dm-registry.h"
+#include "dm-repl-log.h"
+#include "dm-repl-slink.h"
+
+#include <linux/crc32.h>
+#include <linux/dm-io.h>
+#include <linux/kernel.h>
+#include <linux/version.h>
+
+static const char version[] = "v0.028";
+static struct dm_repl_log_type ringbuffer_type;
+
+static struct mutex list_mutex;
+
+#define DM_MSG_PREFIX "dm-repl-log-ringbuffer"
+#define DAEMON DM_MSG_PREFIX "d"
+
+/* Maximum number of site links supported. */
+#define MAX_DEFAULT_SLINKS 2048
+
+#define DEFAULT_BIOS 16 /* Default number of max bios -> ring buffer */
+
+#define LOG_SIZE_MIN (2 * BIO_MAX_SECTORS)
+#define REGIONS_MAX 32768
+
+/* Later kernels have this macro in bitops.h */
+#ifndef for_each_bit
+#define for_each_bit(bit, addr, size) \
+ for ((bit) = find_first_bit((void *)(addr), (size)); \
+ (bit) < (size); \
+ (bit) = find_next_bit((void *)(addr), (size), (bit) + 1))
+#endif
+
+#define _BUG_ON_SLINK_NR(l, slink_nr) \
+ do { \
+ BUG_ON(slink_nr < 0); \
+ } while (0);
+
+/* Replicator log metadata version. */
+struct repl_log_version {
+ unsigned major;
+ unsigned minor;
+ unsigned subminor;
+};
+
+/*
+ * Each version of the log code may get a separate source module, so
+ * we store the version information in the .c file.
+ */
+#define DM_REPL_LOG_MAJOR 0
+#define DM_REPL_LOG_MINOR 0
+#define DM_REPL_LOG_MICRO 1
+
+#define DM_REPL_LOG_VERSION \
+ { DM_REPL_LOG_MAJOR, \
+ DM_REPL_LOG_MINOR, \
+ DM_REPL_LOG_MICRO, }
+
+static struct version {
+ uint16_t major;
+ uint16_t minor;
+ uint16_t subminor;
+} my_version = DM_REPL_LOG_VERSION;
+
+/* 1 */
+/* Shall be 16 bytes long */
+static const char log_header_magic[] = "dm-replicatorHJM";
+#define MAGIC_LEN (sizeof(log_header_magic) - 1)
+#define HANDLER_LEN MAGIC_LEN
+
+/* Header format on disk */
+struct log_header_disk {
+ uint8_t magic[MAGIC_LEN];
+ uint32_t crc;
+ struct version version;
+ uint64_t size;
+ uint64_t buffer_header; /* sector of first
+ * buffer_header_disk */
+ uint8_t handler_name[HANDLER_LEN];
+ /* Free space. */
+} __attribute__((__packed__));
+
+/* Macros for bitmap access. */
+#define BITMAP_SIZE(l) ((l)->slink.bitmap_size)
+#define BITMAP_ELEMS(l) ((l)->slink.bitmap_elems)
+#define BITMAP_ELEMS_MAX 32
+
+/* Header format in core (only one of these per log device). */
+struct log_header {
+ struct repl_log_version version;
+ sector_t size;
+ sector_t buffer_header;
+
+ /* Bitarray of configured slinks to copy accross and those to I/O to. */
+ struct {
+ uint64_t slinks[BITMAP_ELEMS_MAX];
+ uint64_t ios[BITMAP_ELEMS_MAX];
+ uint64_t set_accessible[BITMAP_ELEMS_MAX];
+ uint64_t inaccessible[BITMAP_ELEMS_MAX];
+ } slink_bits;
+};
+#define LOG_SLINKS(l) ((void *) (l)->header.log->slink_bits.slinks)
+#define LOG_SLINKS_IO(l) ((void *) (l)->header.log->slink_bits.ios)
+#define LOG_SLINKS_INACCESSIBLE(l) \
+ ((void *)(l)->header.log->slink_bits.inaccessible)
+#define LOG_SLINKS_SET_ACCESSIBLE(l) \
+ ((void *)(l)->header.log->slink_bits.set_accessible)
+
+static void
+log_header_to_disk(unsigned slinks, void *d_ptr, void *c_ptr)
+{
+ struct log_header_disk *d = d_ptr;
+ struct log_header *c = c_ptr;
+
+ strncpy((char *) d->magic, log_header_magic, MAGIC_LEN);
+ strncpy((char *) d->handler_name,
+ ringbuffer_type.type.name, HANDLER_LEN);
+ d->version.major = cpu_to_le16(c->version.major);
+ d->version.minor = cpu_to_le16(c->version.minor);
+ d->version.subminor = cpu_to_le16(c->version.subminor);
+ d->size = cpu_to_le64(c->size);
+ d->buffer_header = cpu_to_le64(c->buffer_header);
+ d->crc = 0;
+ d->crc = crc32(~0, d, sizeof(d));
+}
+
+static int
+log_header_to_core(unsigned slinks, void *c_ptr, void *d_ptr)
+{
+ int r;
+ uint32_t crc;
+ struct log_header *c = c_ptr;
+ struct log_header_disk *d = d_ptr;
+
+ r = strncmp((char *) d->magic, log_header_magic, MAGIC_LEN);
+ if (r)
+ return -EINVAL;
+
+ /* Check, if acceptible to this replication log handler. */
+ r = strncmp((char *) d->handler_name, ringbuffer_type.type.name,
+ HANDLER_LEN);
+ if (r)
+ return -EINVAL;
+
+ c->version.major = le16_to_cpu(d->version.major);
+ c->version.minor = le16_to_cpu(d->version.minor);
+ c->version.subminor = le16_to_cpu(d->version.subminor);
+ c->size = le64_to_cpu(d->size);
+ c->buffer_header = le64_to_cpu(d->buffer_header);
+ crc = d->crc;
+ d->crc = 0;
+ return (crc == crc32(~0, d, sizeof(d))) ? 0 : -EINVAL;
+}
+
+/* 1a */
+static const char *buffer_header_magic = "dm-replbufferHJM";
+
+/*
+ * meta-data for the ring buffer, one per replog:
+ *
+ * start: location on disk
+ * head: ring buffer head, first data item to be replicated
+ * tail: points to one after the last data item to be replicated
+ *
+ * The ring buffer is full of data_header(_disk) entries.
+ */
+struct buffer_header_disk {
+ uint8_t magic[MAGIC_LEN];
+ uint32_t crc;
+ struct buffer_disk {
+ uint64_t start;
+ uint64_t head;
+ uint64_t tail;
+ } buffer;
+
+ uint64_t flags;
+ /* Free space. */
+} __attribute__((__packed__));
+
+/*
+ * In-core format of the buffer_header_disk structure
+ *
+ * start, head, and tail are as described above for buffer_header_disk.
+ *
+ * next_avail points to the next available sector for placing a log entry.
+ * It is important to distinguish this from tail, as we can issue I/O to
+ * multiple log entries at a time.
+ *
+ * end is the end sector of the log device
+ *
+ * len is the total length of the log device, handy to keep around for maths
+ * free represents the amount of free space in the log. This number
+ * reflects the free space in the log given the current outstanding I/O's.
+ * In other words, it is the distance between next_avail and head.
+ */
+/*
+ * My guess is that this should be subsumed by the repl_log structure, as
+ * much of the data is copied from there, anyway. The question is just
+ * how to organize it in a readable and efficient way.
+ */
+/* Ring state flag(s). */
+enum ring_status_type {
+ RING_BLOCKED,
+ RING_BUFFER_ERROR,
+ RING_BUFFER_DATA_ERROR,
+ RING_BUFFER_HEADER_ERROR,
+ RING_BUFFER_HEAD_ERROR,
+ RING_BUFFER_TAIL_ERROR,
+ RING_BUFFER_FULL,
+ RING_BUFFER_IO_QUEUED,
+ RING_SUSPENDED
+};
+
+/*
+ * Pools types for:
+ * o ring buffer entries
+ * o data headers.
+ * o disk data headers.
+ * o slink copy contexts
+ */
+enum ring_pool_type {
+ ENTRY, /* Ring buffer entries. */
+ DATA_HEADER, /* Ring buffer data headers. */
+ DATA_HEADER_DISK, /* Ring buffer ondisk data headers. */
+ COPY_CONTEXT, /* Context for any single slink copy. */
+ NR_RING_POOLS,
+};
+
+struct sector_range {
+ sector_t start;
+ sector_t end;
+};
+
+struct ringbuffer {
+ sector_t start; /* Start sector of the log space on disk. */
+ sector_t head; /* Sector of the first log entry. */
+ sector_t tail; /* Sector of the last valid log entry. */
+
+ struct mutex mutex; /* Mutex hold on member updates below. */
+
+ /* The following fields are useful to keep track of in-core state. */
+ sector_t next_avail; /* In-memory tail of the log. */
+ sector_t end; /* 1st sector past end of log device. */
+ sector_t free; /* Free space left in the log. */
+ sector_t pending; /* sectors queued but not allocated */
+
+ struct {
+ unsigned long flags; /* Buffer state flags. */
+ } io;
+
+ /* Dirty sectors for slink0. */
+ struct sector_hash {
+ struct list_head *hash;
+ unsigned buckets;
+ unsigned mask;
+ } busy_sectors;
+
+ /* Waiting for all I/O to be flushed. */
+ wait_queue_head_t flushq;
+ mempool_t *pools[NR_RING_POOLS];
+};
+
+DM_BITOPS(RingBlocked, ringbuffer, RING_BLOCKED)
+DM_BITOPS(RingBufferError, ringbuffer, RING_BUFFER_ERROR)
+DM_BITOPS(RingBufferDataError, ringbuffer, RING_BUFFER_DATA_ERROR)
+DM_BITOPS(RingBufferHeaderError, ringbuffer, RING_BUFFER_HEADER_ERROR)
+DM_BITOPS(RingBufferHeadError, ringbuffer, RING_BUFFER_HEAD_ERROR)
+DM_BITOPS(RingBufferTailError, ringbuffer, RING_BUFFER_TAIL_ERROR)
+DM_BITOPS(RingBufferFull, ringbuffer, RING_BUFFER_FULL)
+DM_BITOPS(RingBufferIOQueued, ringbuffer, RING_BUFFER_IO_QUEUED)
+DM_BITOPS(RingSuspended, ringbuffer, RING_SUSPENDED)
+
+#define CC_POOL_MIN 4
+#define HEADER_POOL_MIN 32
+#define ENTRY_POOL_MIN 32
+
+static void
+buffer_header_to_disk(unsigned slinks, void *d_ptr, void *c_ptr)
+{
+ struct buffer_header_disk *d = d_ptr;
+ struct ringbuffer *c = c_ptr;
+
+ strncpy((char *) d->magic, buffer_header_magic, MAGIC_LEN);
+ d->buffer.start = cpu_to_le64(to_bytes(c->start));
+ d->buffer.head = cpu_to_le64(to_bytes(c->head));
+ d->buffer.tail = cpu_to_le64(to_bytes(c->tail));
+ d->flags = cpu_to_le64(c->io.flags);
+ d->crc = 0;
+ d->crc = crc32(~0, d, sizeof(d));
+}
+
+static int
+buffer_header_to_core(unsigned slinks, void *c_ptr, void *d_ptr)
+{
+ int r;
+ uint32_t crc;
+ struct ringbuffer *c = c_ptr;
+ struct buffer_header_disk *d = d_ptr;
+
+ r = strncmp((char *) d->magic, buffer_header_magic, MAGIC_LEN);
+ if (r)
+ return -EINVAL;
+
+ c->start = to_sector(le64_to_cpu(d->buffer.start));
+ c->head = to_sector(le64_to_cpu(d->buffer.head));
+ c->tail = to_sector(le64_to_cpu(d->buffer.tail));
+ c->io.flags = le64_to_cpu(d->flags);
+ crc = d->crc;
+ d->crc = 0;
+ return likely(crc == crc32(~0, d, sizeof(d))) ? 0 : -EINVAL;
+}
+
+/* 3 */
+/* The requirement is to support devices with 4k sectors. */
+#define HEADER_SECTORS to_sector(4096)
+
+static const char *data_header_magic = "dm-replicdataHJM";
+
+/* FIXME: XXX adjust for larger sector size! */
+#define DATA_HEADER_DISK_SIZE 512
+enum entry_wrap_type { WRAP_NONE, WRAP_DATA, WRAP_NEXT };
+struct data_header_disk {
+ uint8_t magic[MAGIC_LEN];
+ uint32_t crc;
+ uint32_t filler;
+
+ struct {
+ /* Internal namespace to get rid of major/minor. -HJM */
+ uint64_t dev;
+ uint64_t offset;
+ uint64_t size;
+ } region;
+
+ /* Position of header and data on disk in bytes. */
+ struct {
+ uint64_t header; /* Offset of this header */
+ uint64_t data; /* Offset of data (ie. the bio). */
+ } pos;
+
+ uint8_t valid; /* FIXME: XXX this needs to be in memory copy, too */
+ uint8_t wrap; /* Above enum entry_wrap_type. */
+ uint8_t barrier;/* Be prepared for write barrier support. */
+
+ /*
+ * Free space: fill up to offset 256.
+ */
+ uint8_t filler1[189];
+
+ /* Offset 256! */
+ /* Bitmap, bit position set to 0 for uptodate slink */
+ uint64_t slink_bits[BITMAP_ELEMS_MAX];
+
+ /* Free space. */
+} __attribute__((__packed__));
+
+struct data_header {
+ struct list_head list;
+
+ /* Bitmap, bit position set to 0 for uptodate slink. */
+ uint64_t slink_bits[BITMAP_ELEMS_MAX];
+
+ /*
+ * Reference count indicating the number of endios
+ * expected while writing the header and bitmap.
+ */
+ atomic_t cnt;
+
+ struct data_header_region {
+ /* dev, sector, and size are taken from the initial bio. */
+ unsigned long dev;
+ sector_t sector;
+ unsigned size;
+ } region;
+
+ /* Position of header and data on disk and size in sectors. */
+ struct {
+ sector_t header; /* sector of this header on disk */
+ sector_t data; /* Offset of data (ie. the bio). */
+ unsigned data_sectors; /* Useful for sector calculation. */
+ } pos;
+
+ /* Next data or complete entry wraps. */
+ enum entry_wrap_type wrap;
+};
+
+/* Round size in bytes up to multiples of HEADER_SECTORS. */
+enum distance_type { FULL_SECTORS, DATA_SECTORS };
+static inline sector_t
+_roundup_sectors(unsigned sectors, enum distance_type type)
+{
+ return HEADER_SECTORS *
+ (!!(type == FULL_SECTORS) + dm_div_up(sectors, HEADER_SECTORS));
+}
+
+/* Header + data. */
+static inline sector_t
+roundup_sectors(unsigned sectors)
+{
+ return _roundup_sectors(sectors, FULL_SECTORS);
+}
+
+/* Data only. */
+static inline sector_t
+roundup_data_sectors(unsigned sectors)
+{
+ return _roundup_sectors(sectors, DATA_SECTORS);
+}
+
+static void
+data_header_to_disk(unsigned bitmap_elems, void *d_ptr, void *c_ptr)
+{
+ unsigned i = bitmap_elems;
+ struct data_header_disk *d = d_ptr;
+ struct data_header *c = c_ptr;
+
+ BUG_ON(!i);
+
+ strncpy((char *) d->magic, data_header_magic, MAGIC_LEN);
+ d->region.dev = cpu_to_le64(c->region.dev);
+ d->region.offset = cpu_to_le64(to_bytes(c->region.sector));
+ d->region.size = cpu_to_le64(c->region.size);
+
+ /* Xfer bitmap. */
+ while (i--)
+ d->slink_bits[i] = cpu_to_le64(c->slink_bits[i]);
+
+ d->valid = 1;
+ d->wrap = c->wrap;
+ d->pos.header = cpu_to_le64(to_bytes(c->pos.header));
+ d->pos.data = cpu_to_le64(to_bytes(c->pos.data));
+ d->crc = 0;
+ d->crc = crc32(~0, d, sizeof(d));
+}
+
+static int
+data_header_to_core(unsigned bitmap_elems, void *c_ptr, void *d_ptr)
+{
+ int r;
+ unsigned i = bitmap_elems;
+ uint32_t crc;
+ struct data_header *c = c_ptr;
+ struct data_header_disk *d = d_ptr;
+
+ BUG_ON(!i);
+
+ r = strncmp((char *) d->magic, data_header_magic, MAGIC_LEN);
+ if (r)
+ return -EINVAL;
+
+ c->region.dev = le64_to_cpu(d->region.dev);
+ c->region.sector = to_sector(le64_to_cpu(d->region.offset));
+ c->region.size = le64_to_cpu(d->region.size);
+
+ /* Xfer bitmap. */
+ while (i--)
+ c->slink_bits[i] = le64_to_cpu(d->slink_bits[i]);
+
+ c->pos.header = to_sector(le64_to_cpu(d->pos.header));
+ c->pos.data = to_sector(le64_to_cpu(d->pos.data));
+ c->pos.data_sectors = roundup_data_sectors(to_sector(c->region.size));
+ c->wrap = d->wrap;
+
+ if (unlikely(!d->valid) ||
+ !c->region.size)
+ return -EINVAL;
+
+ crc = d->crc;
+ d->crc = 0;
+ return likely(crc == crc32(~0, d, sizeof(d))) ? 0 : -EINVAL;
+}
+
+static inline void
+slink_set_bit(int bit, uint64_t *ptr)
+{
+ set_bit(bit, (unsigned long *)ptr);
+ smp_mb();
+}
+
+static inline void
+slink_clear_bit(int bit, uint64_t *ptr)
+{
+ clear_bit(bit, (unsigned long *)ptr);
+ smp_mb();
+}
+
+static inline int
+slink_test_bit(int bit, uint64_t *ptr)
+{
+ return test_bit(bit, (unsigned long *)ptr);
+}
+
+
+/* entry list types and access macros. */
+enum entry_list_type {
+ E_BUSY_HASH, /* Busys entries hash. */
+ E_COPY_CONTEXT, /* Copyies accross slinks in progress for entry. */
+ E_ORDERED, /* Ordered for advancing the ring buffer head. */
+ E_WRITE_OR_COPY,/* Add to l->lists.l[L_ENTRY_RING_WRITE/L_SLINK_COPY] */
+ E_NR_LISTS,
+};
+#define E_BUSY_HASH_LIST(entry) (entry->lists.l + E_BUSY_HASH)
+#define E_COPY_CONTEXT_LIST(entry) (entry->lists.l + E_COPY_CONTEXT)
+#define E_ORDERED_LIST(entry) (entry->lists.l + E_ORDERED)
+#define E_WRITE_OR_COPY_LIST(entry) (entry->lists.l + E_WRITE_OR_COPY)
+
+/*
+ * Container for the data_header and the associated data pages.
+ */
+struct ringbuffer_entry {
+ struct {
+ struct list_head l[E_NR_LISTS];
+ } lists;
+
+ struct ringbuffer *ring; /* Back pointer. */
+
+ /* Reference count. */
+ atomic_t ref;
+
+ /*
+ * Reference count indicating the number of endios expected
+ * while writing its header and data to the ring buffer log
+ * -or- future use:
+ * how many copies accross site links are active and how many
+ * reads are being sattisfied from the entry.
+ */
+ atomic_t endios;
+
+ struct entry_data {
+ struct data_header *header;
+ struct data_header_disk *disk_header;
+ struct {
+ unsigned long data;
+ unsigned long header;
+ } error;
+ } data;
+
+ struct {
+ struct bio *read; /* bio to read. */
+ struct bio *write; /* Original bio to write. */
+ } bios;
+
+ struct {
+ /* Bitmask of slinks the entry has active copies accross. */
+ uint64_t ios[BITMAP_ELEMS_MAX];
+ /* Bitmask of synchronuous slinks for endio. */
+ /* FIXME: drop in favour of slink inquiry of sync state ? */
+ uint64_t sync[BITMAP_ELEMS_MAX];
+ /* Bitmask of slinks with errors. */
+ uint64_t error[BITMAP_ELEMS_MAX];
+ } slink_bits;
+};
+#define ENTRY_SLINKS(l) ((void *) (entry)->data.header->slink_bits)
+#define ENTRY_IOS(entry) ((void *) (entry)->slink_bits.ios)
+#define ENTRY_SYNC(entry) ((entry)->slink_bits.sync)
+#define ENTRY_ERROR(entry) ((entry)->slink_bits.error)
+
+/* FIXME: XXX
+ * For now, the copy context has a backpointer to the ring buffer entry.
+ * This means that a ring buffer entry has to remain in memory until all
+ * of the slink copies have finished. Heinz, you mentioned that this was
+ * not a good idea. I'm open to suggestions on how better to organize this.
+ */
+enum error_type { ERR_DISK, ERR_RAM, NR_ERR_TYPES };
+struct slink_copy_error {
+ int read;
+ int write;
+};
+
+struct slink_copy_context {
+ /*
+ * List first points to the copy context list in the ring buffer
+ * entry. Then, upon completion it gets moved to the slink endios
+ * list.
+ */
+ struct list_head list;
+ atomic_t cnt;
+ struct ringbuffer_entry *entry;
+ struct dm_repl_slink *slink;
+ struct slink_copy_error error[NR_ERR_TYPES];
+ unsigned long start_jiffies;
+};
+
+/* Development statistics. */
+struct stats {
+ atomic_t io[2];
+ atomic_t writes_pending;
+ atomic_t hash_elem;
+
+ unsigned copy[2];
+ unsigned wrap;
+ unsigned hash_insert;
+ unsigned hash_insert_max;
+ unsigned stall;
+};
+
+/* Per site link measure/state. */
+enum slink_status_type {
+ SS_SYNC, /* slink fell behind an I/O threshold. */
+ SS_TEARDOWN, /* Flag site link teardown. */
+};
+struct slink_state {
+ unsigned slink_nr;
+ struct repl_log *l;
+
+ struct {
+
+ /*
+ * Difference of time (measured in jiffies) between the
+ * first outstanding copy for this slink and the last
+ * outstanding copy.
+ */
+ unsigned long head_jiffies;
+
+ /* Number of ios/sectors currently copy() to this slink. */
+ struct {
+ sector_t sectors;
+ uint64_t ios;
+ } outstanding;
+ } fb;
+
+ struct {
+ unsigned long flags; /* slink_state flags._*/
+
+ /* slink+I/O teardown synchronization. */
+ wait_queue_head_t waiters;
+ atomic_t in_flight;
+ } io;
+};
+DM_BITOPS(SsSync, slink_state, SS_SYNC)
+DM_BITOPS(SsTeardown, slink_state, SS_TEARDOWN)
+
+enum open_type { OT_AUTO, OT_OPEN, OT_CREATE };
+enum replog_status_type {
+ LOG_DEVEL_STATS, /* Turn on development stats. */
+ LOG_INITIALIZED, /* Log initialization finished. */
+ LOG_RESIZE, /* Log resize requested. */
+};
+
+/* repl_log list types and access macros. */
+enum replog_list_type {
+ L_REPLOG, /* Linked list of replogs. */
+ L_SLINK_COPY, /* Entries to copy accross slinks. */
+ L_SLINK_ENDIO, /* Entries to endio process. */
+ L_ENTRY_RING_WRITE, /* Entries to write to ring buffer */
+ L_ENTRY_ORDERED, /* Ordered list of entries (write fidelity). */
+ L_NR_LISTS,
+};
+#define L_REPLOG_LIST(l) (l->lists.l + L_REPLOG)
+#define L_SLINK_COPY_LIST(l) (l->lists.l + L_SLINK_COPY)
+#define L_SLINK_ENDIO_LIST(l) (l->lists.l + L_SLINK_ENDIO)
+#define L_ENTRY_RING_WRITE_LIST(l) (l->lists.l + L_ENTRY_RING_WRITE)
+#define L_ENTRY_ORDERED_LIST(l) (l->lists.l + L_ENTRY_ORDERED)
+
+/* The replication log in core. */
+struct repl_log {
+ struct dm_repl_log *log;
+
+ struct kref ref; /* Pin count. */
+
+ struct dm_repl_log *replog;
+ struct dm_repl_slink *slink0;
+
+ struct stats stats; /* Development statistics. */
+
+ struct repl_params {
+ enum open_type open_type;
+ unsigned count;
+ struct repl_dev {
+ struct dm_dev *dm_dev;
+ sector_t start;
+ sector_t size;
+ } dev;
+ } params;
+
+ struct {
+ spinlock_t lock; /* Lock on pending list below. */
+ struct bio_list in; /* pending list of bios */
+ struct dm_io_client *io_client;
+ struct workqueue_struct *wq;
+ struct work_struct ws;
+ unsigned long flags; /* State flags. */
+ /* Preallocated header. We only need one at a time.*/
+ struct buffer_header_disk *buffer_header_disk;
+ } io;
+
+ struct ringbuffer ringbuffer;
+
+ /* Useful for runtime performance on bitmap accesses. */
+ struct {
+ int count; /* Actual # of slinks in this replog. */
+ unsigned max; /* Actual maximum added site link #. */
+ unsigned bitmap_elems; /* Actual used elements in bitmaps. */
+ unsigned bitmap_size; /* Actual bitmap size (for memcpy). */
+ } slink;
+
+ struct {
+ struct log_header *log;
+ } header;
+
+ struct {
+ /* List of site links. */
+ struct dm_repl_log_slink_list slinks;
+
+ /*
+ * A single lock for all of these lists should be sufficient
+ * given that each list is processed in-turn (see do_log()).
+ *
+ * The lock has to protect the L_SLINK_ENDIO list
+ * and the entry ring write lists below.
+ *
+ * We got to streamline these lists vs. the lock. -HJM
+ * The others are accessed by one thread only. -HJM
+ */
+ rwlock_t lock;
+
+ /*
+ * Lists for entry slink copies, entry endios,
+ * ring buffer writes and ordered entries.
+ */
+ struct list_head l[L_NR_LISTS];
+ } lists;
+
+ /* Caller callback function and context. */
+ struct replog_notify {
+ dm_repl_notify_fn fn;
+ void *context;
+ } notify;
+};
+
+#define _SET_AND_BUG_ON_L(l, log) \
+ do { \
+ _BUG_ON_PTR(log); \
+ (l) = (log)->context; \
+ _BUG_ON_PTR(l); \
+ } while (0);
+
+/* Define log bitops. */
+DM_BITOPS(LogDevelStats, repl_log, LOG_DEVEL_STATS);
+DM_BITOPS(LogInitialized, repl_log, LOG_INITIALIZED);
+DM_BITOPS(LogResize, repl_log, LOG_RESIZE);
+
+static inline struct repl_log *
+ringbuffer_repl_log(struct ringbuffer *ring)
+{
+ return container_of(ring, struct repl_log, ringbuffer);
+}
+
+static inline struct block_device *
+repl_log_bdev(struct repl_log *l)
+{
+ return l->params.dev.dm_dev->bdev;
+}
+
+static inline struct block_device *
+ringbuffer_bdev(struct ringbuffer *ring)
+{
+ return repl_log_bdev(ringbuffer_repl_log(ring));
+}
+
+/* Check MAX_SLINKS bit array for busy bits. */
+static inline int
+entry_busy(struct repl_log *l, void *bits)
+{
+ return find_first_bit(bits, l->slink.max) < l->slink.max;
+}
+
+static inline int
+entry_endios_pending(struct ringbuffer_entry *entry)
+{
+ return entry_busy(ringbuffer_repl_log(entry->ring), ENTRY_IOS(entry));
+}
+
+static inline int
+ss_io(struct slink_state *ss)
+{
+ _BUG_ON_PTR(ss);
+ return atomic_read(&ss->io.in_flight);
+}
+
+static void
+ss_io_get(struct slink_state *ss)
+{
+ BUG_ON(!ss || IS_ERR(ss));
+ atomic_inc(&ss->io.in_flight);
+}
+
+static void
+ss_io_put(struct slink_state *ss)
+{
+ _BUG_ON_PTR(ss);
+ if (atomic_dec_and_test((&ss->io.in_flight)))
+ wake_up(&ss->io.waiters);
+ else
+ BUG_ON(ss_io(ss) < 0);
+}
+
+static void
+ss_wait_on_io(struct slink_state *ss)
+{
+ _BUG_ON_PTR(ss);
+ while (ss_io(ss)) {
+ flush_workqueue(ss->l->io.wq);
+ wait_event(ss->io.waiters, !ss_io(ss));
+ }
+}
+
+/* Wait for I/O to finish on all site links. */
+static inline void
+ss_all_wait_on_ios(struct repl_log *l)
+{
+ unsigned long slink_nr;
+
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max) {
+ struct dm_repl_slink *slink =
+ l->slink0->ops->slink(l->replog, slink_nr);
+ struct slink_state *ss;
+
+ if (IS_ERR(slink)) {
+ DMERR_LIMIT("%s slink error", __func__);
+ continue;
+ }
+
+ ss = slink->caller;
+ _BUG_ON_PTR(ss);
+ ss_wait_on_io(ss);
+ }
+}
+
+static inline struct dm_io_client *
+replog_io_client(struct repl_log *l)
+{
+ return l->io.io_client;
+}
+
+static inline struct repl_log *
+dev_repl_log(struct repl_dev *dev)
+{
+ return container_of(dev, struct repl_log, params.dev);
+}
+
+/* Define mempool_{alloc,free}() functions for the ring buffer pools. */
+#define ALLOC_FREE_ELEM(name, type) \
+static void *\
+alloc_ ## name(struct ringbuffer *ring) \
+{ \
+ return mempool_alloc(ring->pools[(type)], GFP_KERNEL); \
+} \
+\
+static inline void \
+free_ ## name(void *ptr, struct ringbuffer *ring) \
+{ \
+ _BUG_ON_PTR(ptr); \
+ mempool_free(ptr, ring->pools[(type)]); \
+}
+
+ALLOC_FREE_ELEM(entry, ENTRY)
+ALLOC_FREE_ELEM(header, DATA_HEADER)
+ALLOC_FREE_ELEM(data_header_disk, DATA_HEADER_DISK)
+ALLOC_FREE_ELEM(copy_context, COPY_CONTEXT)
+#undef ALLOC_FREE_ELEM
+
+/* Additional alloc/free functions for header_io() abstraction. */
+/* No need to allocate bitmaps, because they are transient. */
+static void *
+alloc_log_header_disk(struct ringbuffer *ring)
+{
+ return kmalloc(to_bytes(1), GFP_KERNEL);
+}
+
+static void
+free_log_header_disk(void *ptr, struct ringbuffer *ring)
+{
+ kfree(ptr);
+}
+
+/* Dummies to allow for abstraction. */
+static void *
+alloc_buffer_header_disk(struct ringbuffer *ring)
+{
+ return ringbuffer_repl_log(ring)->io.buffer_header_disk;
+}
+
+static void
+free_buffer_header_disk(void *ptr, struct ringbuffer *ring)
+{
+}
+
+/*********************************************************************
+ * Busys entries hash.
+ */
+/* Initialize/destroy sector hash. */
+static int
+sector_hash_init(struct sector_hash *hash, sector_t size)
+{
+ unsigned buckets = roundup_pow_of_two(size / BIO_MAX_SECTORS);
+
+ if (buckets > 4) {
+ if (buckets > REGIONS_MAX)
+ buckets = REGIONS_MAX;
+
+ buckets /= 4;
+ }
+
+ /* Allocate stripe hash. */
+ hash->hash = vmalloc(buckets * sizeof(*hash->hash));
+ if (!hash->hash)
+ return -ENOMEM;
+
+ hash->buckets = hash->mask = buckets;
+ hash->mask--;
+
+ /* Initialize buckets. */
+ while (buckets--)
+ INIT_LIST_HEAD(hash->hash + buckets);
+
+ return 0;
+}
+
+static void
+sector_hash_exit(struct sector_hash *hash)
+{
+ if (hash->hash) {
+ vfree(hash->hash);
+ hash->hash = NULL;
+ }
+}
+
+/* Hash function. */
+static inline struct list_head *
+hash_bucket(struct sector_hash *hash, sector_t sector)
+{
+ sector_div(sector, BIO_MAX_SECTORS);
+ return hash->hash + (unsigned) (sector & hash->mask);
+}
+
+/* Insert an entry into a sector hash. */
+static inline void
+sector_hash_elem_insert(struct sector_hash *hash,
+ struct ringbuffer_entry *entry)
+{
+ struct repl_log *l;
+ struct stats *s;
+ struct list_head *bucket =
+ hash_bucket(hash, entry->data.header->region.sector);
+
+ BUG_ON(!bucket);
+ _BUG_ON_PTR(entry->ring);
+ l = ringbuffer_repl_log(entry->ring);
+ s = &l->stats;
+
+ BUG_ON(!list_empty(E_BUSY_HASH_LIST(entry)));
+ list_add_tail(E_BUSY_HASH_LIST(entry), bucket);
+
+ atomic_inc(&s->hash_elem);
+ if (++s->hash_insert > s->hash_insert_max)
+ s->hash_insert_max = s->hash_insert;
+}
+
+/* Return first sector # of bio. */
+static inline sector_t
+bio_begin(struct bio *bio)
+{
+ return bio->bi_sector;
+}
+
+/* Return last sector # of bio. */
+static inline sector_t bio_end(struct bio *bio)
+{
+ return bio_begin(bio) + bio_sectors(bio);
+}
+
+/* Roundup size to sectors. */
+static inline sector_t round_up_to_sector(unsigned size)
+{
+ return to_sector(dm_round_up(size, to_bytes(1)));
+}
+
+/* Check if a bio and a range overlap. */
+static inline int
+_ranges_overlap(struct sector_range *r1, struct sector_range *r2)
+{
+ return r1->start >= r2->start &&
+ r1->start < r2->end;
+}
+
+static inline int
+ranges_overlap(struct sector_range *elem_range, struct sector_range *bio_range)
+{
+ return _ranges_overlap(elem_range, bio_range) ||
+ _ranges_overlap(bio_range, elem_range);
+}
+
+/* Take an entry ref reference out. */
+static inline void
+entry_get(struct ringbuffer_entry *entry)
+{
+ atomic_inc(&entry->ref);
+}
+
+/*
+ * Check if bio's address range has writes pending.
+ *
+ * Must be called with the read hash lock held.
+ */
+static int
+ringbuffer_writes_pending(struct sector_hash *hash, struct bio *bio,
+ struct list_head *buckets[2])
+{
+ int r = 0;
+ unsigned end, i;
+ struct ringbuffer_entry *entry;
+ /* Setup a range for the bio. */
+ struct sector_range bio_range = {
+ .start = bio_begin(bio),
+ .end = bio_end(bio),
+ }, entry_range;
+
+ buckets[0] = hash_bucket(hash, bio_range.start);
+ buckets[1] = hash_bucket(hash, bio_range.end);
+ if (buckets[0] == buckets[1]) {
+ end = 1;
+ buckets[1] = NULL;
+ } else
+ end = 2;
+
+ for (i = 0; i < end; i++) {
+ /* Walk the entries checking for overlaps. */
+ list_for_each_entry_reverse(entry, buckets[i],
+ lists.l[E_BUSY_HASH]) {
+ entry_range.start = entry->data.header->region.sector;
+ entry_range.end = entry_range.start +
+ round_up_to_sector(entry->data.header->region.size);
+
+ if (ranges_overlap(&entry_range, &bio_range))
+ return atomic_read(&entry->endios) ? -EBUSY : 1;
+ }
+ }
+
+ return r;
+}
+
+/* Clear a sector range busy. */
+static void
+entry_put(struct ringbuffer_entry *entry)
+{
+ _BUG_ON_PTR(entry);
+
+ if (atomic_dec_and_test(&entry->ref)) {
+ struct ringbuffer *ring = entry->ring;
+ struct stats *s;
+ struct repl_log *l;
+
+ _BUG_ON_PTR(ring);
+ l = ringbuffer_repl_log(ring);
+ s = &l->stats;
+
+ /*
+ * We don't need locking here because the last
+ * put is carried out in daemon context.
+ */
+ BUG_ON(list_empty(E_BUSY_HASH_LIST(entry)));
+ list_del_init(E_BUSY_HASH_LIST(entry));
+
+ atomic_dec(&s->hash_elem);
+ s->hash_insert--;
+ } else
+ BUG_ON(atomic_read(&entry->ref) < 0);
+}
+
+static inline void
+sector_range_clear_busy(struct ringbuffer_entry *entry)
+{
+ entry_put(entry);
+}
+
+/*
+ * Mark a sector range start and length busy.
+ *
+ * Caller has to serialize calls.
+ */
+static void
+sector_range_mark_busy(struct ringbuffer_entry *entry)
+{
+ _BUG_ON_PTR(entry);
+ entry_get(entry);
+
+ /* Insert new element into hash. */
+ sector_hash_elem_insert(&entry->ring->busy_sectors, entry);
+}
+
+static void
+stats_init(struct repl_log *l)
+{
+ unsigned i = 2;
+ struct stats *s = &l->stats;
+
+ memset(s, 0, sizeof(*s));
+
+ while (i--)
+ atomic_set(s->io + i, 0);
+
+ atomic_set(&s->writes_pending, 0);
+ atomic_set(&s->hash_elem, 0);
+}
+
+/* Global replicator log list. */
+static LIST_HEAD(replog_list);
+
+/* Wake worker. */
+static void
+wake_do_log(struct repl_log *l)
+{
+ queue_work(l->io.wq, &l->io.ws);
+}
+
+struct dm_repl_slink *
+slink_find(struct repl_log *l, int slink_nr)
+{
+ struct dm_repl_slink *slink0 = l->slink0;
+
+ if (!slink0)
+ return ERR_PTR(-ENOENT);
+
+ _BUG_ON_SLINK_NR(l, slink_nr);
+ return slink_nr ? slink0->ops->slink(l->replog, slink_nr) : slink0;
+}
+
+/*
+ * If an slink is asynchronous, check to see if it needs to fall
+ * back to synchronous mode due to falling too far behind.
+ *
+ * Declare a bunch of fallbehind specific small functions in order
+ * to avoid conditions in the fast path by accessing them via
+ * function pointers.
+ */
+/* True if slink exceeds fallbehind threshold. */
+static int
+slink_fallbehind_exceeded(struct repl_log *l, struct slink_state *ss,
+ struct dm_repl_slink_fallbehind *fallbehind,
+ unsigned amount)
+{
+ sector_t *sectors;
+ uint64_t *ios;
+ unsigned long *head_jiffies;
+
+ _BUG_ON_PTR(l);
+ _BUG_ON_PTR(ss);
+ _BUG_ON_PTR(fallbehind);
+ ios = &ss->fb.outstanding.ios;
+ sectors = &ss->fb.outstanding.sectors;
+
+ spin_lock(&l->io.lock);
+ (*ios)++;
+ (*sectors) += amount;
+ spin_unlock(&l->io.lock);
+
+ if (!fallbehind->value)
+ return 0;
+
+ switch (fallbehind->type) {
+ case DM_REPL_SLINK_FB_IOS:
+ return *ios > fallbehind->value;
+
+ case DM_REPL_SLINK_FB_SIZE:
+ return *sectors > fallbehind->value;
+
+ case DM_REPL_SLINK_FB_TIMEOUT:
+ head_jiffies = &ss->fb.head_jiffies;
+ if (unlikely(!*head_jiffies))
+ *head_jiffies = jiffies;
+
+ return time_after(jiffies, *head_jiffies +
+ msecs_to_jiffies(fallbehind->value));
+
+ default:
+ BUG();
+ }
+
+ return 0;
+}
+
+/*
+ * True if slink falls below fallbehind threshold.
+ *
+ * Can be called from interrupt context.
+ */
+static int
+slink_fallbehind_recovered(struct repl_log *l, struct slink_state *ss,
+ struct dm_repl_slink_fallbehind *fallbehind,
+ unsigned amount)
+{
+ sector_t *sectors;
+ uint64_t *ios;
+
+ _BUG_ON_PTR(ss);
+ _BUG_ON_PTR(fallbehind);
+ ios = &ss->fb.outstanding.ios;
+ sectors = &ss->fb.outstanding.sectors;
+
+ /* Need the non-irq versions here, because IRQs are already disabled. */
+ spin_lock(&l->io.lock);
+ (*ios)--;
+ (*sectors) -= amount;
+ spin_unlock(&l->io.lock);
+
+ if (!fallbehind->value)
+ return 0;
+
+ switch (fallbehind->type) {
+ case DM_REPL_SLINK_FB_IOS:
+ return *ios <= fallbehind->value;
+
+ case DM_REPL_SLINK_FB_SIZE:
+ return *sectors <= fallbehind->value;
+
+ case DM_REPL_SLINK_FB_TIMEOUT:
+ return time_before(jiffies, ss->fb.head_jiffies +
+ msecs_to_jiffies(fallbehind->value));
+ default:
+ BUG();
+ }
+
+ return 0;
+}
+
+/*
+ * Update fallbehind account.
+ *
+ * Has to be called with rw lock held.
+ */
+/* FIXME: account for resynchronization. */
+enum fb_update_type { UPD_INC, UPD_DEC };
+static void
+slink_fallbehind_update(enum fb_update_type type,
+ struct dm_repl_slink *slink,
+ struct ringbuffer_entry *entry)
+{
+ int slink_nr, sync;
+ struct repl_log *l;
+ struct slink_state *ss;
+ struct data_header_region *region;
+ struct dm_repl_slink_fallbehind *fallbehind;
+ struct ringbuffer_entry *pos;
+
+ _BUG_ON_PTR(slink);
+ fallbehind = slink->ops->fallbehind(slink);
+ _BUG_ON_PTR(fallbehind);
+ _BUG_ON_PTR(entry);
+ l = ringbuffer_repl_log(entry->ring);
+ _BUG_ON_PTR(l);
+ slink_nr = slink->ops->slink_number(slink);
+ _BUG_ON_SLINK_NR(l, slink_nr);
+ region = &entry->data.header->region;
+ _BUG_ON_PTR(region);
+
+ /*
+ * We can access ss w/o a lock, because it's referenced by
+ * inflight I/Os and by the running worker which processes
+ * this function.
+ */
+ ss = slink->caller;
+ if (!ss)
+ return;
+
+ _BUG_ON_PTR(ss);
+ sync = SsSync(ss);
+
+ switch (type) {
+ case UPD_INC:
+ if (slink_fallbehind_exceeded(l, ss, fallbehind,
+ region->size) &&
+ !TestSetSsSync(ss) &&
+ !sync)
+ DMINFO("enforcing fallbehind sync on slink=%d at %u",
+ slink_nr, jiffies_to_msecs(jiffies));
+ break;
+
+ case UPD_DEC:
+ /*
+ * Walk the list of outstanding copy I/Os and update the
+ * start_jiffies value with the first entry found.
+ */
+ list_for_each_entry(pos, L_SLINK_COPY_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ struct slink_copy_context *cc;
+
+ list_for_each_entry(cc, E_COPY_CONTEXT_LIST(pos),
+ list) {
+ if (cc->slink->ops->slink_number(cc->slink) ==
+ slink_nr) {
+ ss->fb.head_jiffies = cc->start_jiffies;
+ break;
+ }
+ }
+ }
+
+ if (slink_fallbehind_recovered(l, ss, fallbehind,
+ region->size)) {
+ ss->fb.head_jiffies = 0;
+
+ if (TestClearSsSync(ss) && sync) {
+ DMINFO("releasing fallbehind sync on slink=%d"
+ " at %u",
+ slink_nr, jiffies_to_msecs(jiffies));
+ wake_do_log(l);
+ }
+ }
+
+ break;
+
+ default:
+ BUG();
+ }
+}
+
+static inline void
+slink_fallbehind_inc(struct dm_repl_slink *slink,
+ struct ringbuffer_entry *entry)
+{
+ slink_fallbehind_update(UPD_INC, slink, entry);
+}
+
+static inline void
+slink_fallbehind_dec(struct dm_repl_slink *slink,
+ struct ringbuffer_entry *entry)
+{
+ slink_fallbehind_update(UPD_DEC, slink, entry);
+}
+
+/* Caller properties definition for dev_io(). */
+struct dev_io_params {
+ struct repl_dev *dev;
+ sector_t sector;
+ unsigned size;
+ struct dm_io_memory mem;
+ struct dm_io_notify notify;
+ unsigned long flags;
+};
+
+/*
+ * Read/write device items.
+ *
+ * In case of dio->fn, an asynchronous dm_io()
+ * call will be performed, else synchronous.
+ */
+static int
+dev_io(int rw, struct ringbuffer *ring, struct dev_io_params *dio)
+{
+ BUG_ON(dio->size > BIO_MAX_SIZE);
+ DMDEBUG_LIMIT("%s: rw: %d, %u sectors at sector %llu, dev %p",
+ __func__, rw, dio->size,
+ (unsigned long long) dio->sector,
+ dio->dev->dm_dev->bdev);
+
+ /* Flag IO queued on asynchronous calls. */
+ if (dio->notify.fn)
+ SetRingBufferIOQueued(ring);
+
+ return dm_io(
+ &(struct dm_io_request) {
+ .bi_rw = rw,
+ .mem = dio->mem,
+ .notify = dio->notify,
+ .client = replog_io_client(dev_repl_log(dio->dev))
+ }, 1 /* 1 region following */,
+ &(struct dm_io_region) {
+ .bdev = dio->dev->dm_dev->bdev,
+ .sector = dio->sector,
+ .count = round_up_to_sector(dio->size),
+ },
+ NULL
+ );
+}
+
+/* Definition of properties/helper functions for header IO. */
+struct header_io_spec {
+ const char *name; /* Header identifier (eg. 'data'). */
+ unsigned size; /* Size of ondisk structure. */
+ /* Disk structure allocation helper. */
+ void *(*alloc_disk)(struct ringbuffer *);
+ /* Disk structure deallocation helper. */
+ void (*free_disk)(void *, struct ringbuffer *);
+ /* Disk structure to core structure xfer helper. */
+ int (*to_core_fn)(unsigned bitmap_elems, void *, void *);
+ /* Core structure to disk structure xfer helper. */
+ void (*to_disk_fn)(unsigned bitmap_elems, void *, void *);
+};
+/* Macro to initialize type specific header_io_spec structure. */
+#define IO_SPEC(header) \
+ { .name = # header, \
+ .size = sizeof(struct header ## _header_disk), \
+ .alloc_disk = alloc_ ## header ## _header_disk, \
+ .free_disk = free_ ## header ## _header_disk, \
+ .to_core_fn = header ## _header_to_core, \
+ .to_disk_fn = header ## _header_to_disk }
+
+enum header_type { IO_LOG, IO_BUFFER, IO_DATA };
+struct header_io_params {
+ enum header_type type;
+ struct repl_log *l;
+ void *core_header;
+ sector_t sector;
+ void (*disk_header_fn)(void *);
+};
+
+/* Read /write a {log,buffer,data} header to disk. */
+static int
+header_io(int rw, struct header_io_params *hio)
+{
+ int r;
+ struct repl_log *l = hio->l;
+ struct ringbuffer *ring = &l->ringbuffer;
+ /* Specs of all log headers. Must be in 'enum header_type' order! */
+ static const struct header_io_spec io_specs[] = {
+ IO_SPEC(log),
+ IO_SPEC(buffer),
+ IO_SPEC(data),
+ };
+ const struct header_io_spec *io = io_specs + hio->type;
+ void *disk_header = io->alloc_disk(ring);
+ struct dev_io_params dio = {
+ &l->params.dev, hio->sector, io->size,
+ .mem = { DM_IO_KMEM, { .addr = disk_header }, 0 },
+ .notify = { NULL, NULL}
+ };
+
+ BUG_ON(io < io_specs || io >= ARRAY_END(io_specs));
+ BUG_ON(!hio->core_header);
+ BUG_ON(!disk_header);
+ memset(disk_header, 0, io->size);
+
+ if (rw == WRITE) {
+ io->to_disk_fn(BITMAP_ELEMS(l), disk_header, hio->core_header);
+
+ /* If disk header needs special handling before write. */
+ if (hio->disk_header_fn)
+ hio->disk_header_fn(disk_header);
+ }
+
+ r = dev_io(rw, ring, &dio);
+ if (unlikely(r)) {
+ SetRingBufferError(ring);
+ DMERR("Failed to %s %s header!",
+ rw == WRITE ? "write" : "read", io->name);
+ } else if (rw == READ) {
+ r = io->to_core_fn(BITMAP_ELEMS(l), hio->core_header,
+ disk_header);
+ if (unlikely(r))
+ DMERR("invalid %s header/sector=%llu",
+ io->name, (unsigned long long) hio->sector);
+ }
+
+ io->free_disk(disk_header, ring);
+ return r;
+}
+
+/* Read/write the log header synchronously. */
+static inline int
+log_header_io(int rw, struct repl_log *l)
+{
+ return header_io(rw, &(struct header_io_params) {
+ IO_LOG, l, l->header.log, l->params.dev.start, NULL });
+}
+
+/* Read/write the ring buffer header synchronously. */
+static inline int
+buffer_header_io(int rw, struct repl_log *l)
+{
+ return header_io(rw, &(struct header_io_params) {
+ IO_BUFFER, l, &l->ringbuffer,
+ l->header.log->buffer_header, NULL });
+}
+
+/* Read/write a data header to/from the ring buffer synchronously. */
+static inline int
+data_header_io(int rw, struct repl_log *l,
+ struct data_header *header, sector_t sector)
+{
+ return header_io(rw, &(struct header_io_params) {
+ IO_DATA, l, header, sector, NULL });
+}
+
+/* Notify dm-repl.c to submit more IO. */
+static void
+notify_caller(struct repl_log *l, int rw, int error)
+{
+ struct replog_notify notify;
+
+ _BUG_ON_PTR(l);
+
+ spin_lock(&l->io.lock);
+ notify = l->notify;
+ spin_unlock(&l->io.lock);
+
+ if (likely(notify.fn)) {
+ if (rw == READ)
+ notify.fn(error, 0, notify.context);
+ else
+ notify.fn(0, error, notify.context);
+ }
+}
+
+/*
+ * Ring buffer routines.
+ *
+ * The ring buffer needs to keep track of arbitrarily-sized data items.
+ * HEAD points to the first data header that needs to be replicated. This
+ * can mean it has been partially replicated or not replicated at all.
+ * The ring buffer is empty if HEAD == TAIL.
+ * The ring buffer is full if HEAD == TAIL + len(TAIL) modulo device size.
+ *
+ * An entry in the buffer is not valid until both the data header and the
+ * associated data items are on disk. Multiple data headers and data items
+ * may be written in parallel. This means that, in addition to the
+ * traditional HEAD and TAIL pointers, we need to keep track of an in-core
+ * variable reflecting the next area in the log that is unallocated. We also
+ * need to keep an ordered list of pending and completd buffer entry writes.
+ */
+/*
+ * Check and wrap a ring buffer offset around ring buffer end.
+ *
+ * There are three cases to distinguish here:
+ * 1. header and data fit before ring->end
+ * 2. header fits before ring->end, data doesn't -> remap data to ring->start
+ * 3. header doesn't fit before ring->end -> remap both to ring->start
+ *
+ * Function returns the next rounded offset *after* any
+ * conditional remapping of the actual header.
+ *
+ */
+static sector_t
+sectors_unused(struct ringbuffer *ring, sector_t first_free)
+{
+ return (ring->end < first_free) ? 0 : ring->end - first_free;
+}
+
+/*
+ * Return the first sector past the end of the header
+ * (i.e. the first data sector).
+ */
+static inline sector_t
+data_start(struct data_header *header)
+{
+ return header->pos.header + HEADER_SECTORS;
+}
+
+/*
+ * Return the first sector past the end of the entry.
+ * (i.e.(the first unused sector).
+ */
+static inline sector_t
+next_start(struct data_header *header)
+{
+ return header->pos.data + header->pos.data_sectors;
+}
+
+static inline sector_t
+next_start_adjust(struct ringbuffer *ring, struct data_header *header)
+{
+ sector_t next_sector = next_start(header);
+
+ return likely(sectors_unused(ring, next_sector) < HEADER_SECTORS) ?
+ ring->start : next_sector;
+}
+
+/* True if entry doesn't wrap. */
+static inline int
+not_wrapped(struct data_header *header)
+{
+ return header->wrap == WRAP_NONE;
+}
+
+/* True if header at ring end and data wrapped to ring start. */
+static inline int
+data_wrapped(struct data_header *header)
+{
+ return header->wrap == WRAP_DATA;
+}
+
+/* True if next entry wraps to ring start. */
+static inline int
+next_entry_wraps(struct data_header *header)
+{
+ return header->wrap == WRAP_NEXT;
+}
+
+/* Return amount of skipped sectors in case of wrapping. */
+static unsigned
+sectors_skipped(struct ringbuffer *ring, struct data_header *header)
+{
+ if (likely(not_wrapped(header)))
+ /* noop */ ;
+ else if (data_wrapped(header))
+ return sectors_unused(ring, data_start(header));
+ else if (next_entry_wraps(header))
+ return sectors_unused(ring, next_start(header));
+
+ return 0;
+}
+
+/* Emmit only once log error messages. */
+static void
+ringbuffer_error(enum ring_status_type type,
+ struct ringbuffer *ring, int error)
+{
+ struct error {
+ enum ring_status_type type;
+ int (*f)(struct ringbuffer *);
+ const char *msg;
+ };
+ static const struct error errors[] = {
+ { RING_BUFFER_DATA_ERROR, TestSetRingBufferDataError, "data" },
+ { RING_BUFFER_HEAD_ERROR, TestSetRingBufferHeadError, "head" },
+ { RING_BUFFER_HEADER_ERROR, TestSetRingBufferHeaderError,
+ "header" },
+ { RING_BUFFER_TAIL_ERROR, TestSetRingBufferTailError, "tail" },
+ };
+ const struct error *e = ARRAY_END(errors);
+
+ while (e-- > errors) {
+ if (type == e->type) {
+ if (!e->f(ring))
+ DMERR("ring buffer %s I/O error %d",
+ e->msg, error);
+
+ return SetRingBufferError(ring);
+ }
+ }
+
+ BUG();
+}
+
+/*
+ * Allocate space for a data item in the ring buffer.
+ *
+ * header->pos is filled in with the sectors for the header and data in
+ * the ring buffer. The free space in the ring buffer is decremented to
+ * account for this entry. The return value is the sector address for the
+ * next data_header_disk.
+ */
+
+/* Increment buffer offset past actual header, optionaly wrapping data. */
+static sector_t
+ringbuffer_inc(struct ringbuffer *ring, struct data_header *header)
+{
+ sector_t sectors;
+
+ /* Initialize the header with the common case */
+ header->pos.header = ring->next_avail;
+ header->pos.data = data_start(header);
+
+ /*
+ * Header doesn't fit before ring->end.
+ *
+ * This can only happen when we are started with an empty ring
+ * buffer that has its tail near the end of the device.
+ */
+ if (unlikely(data_start(header) > ring->end)) {
+ /*
+ * Wrap an entire entry (header + data) to the beginning of
+ * the log device. This will update the ring free sector
+ * count to account for the unused sectors at the end
+ * of the device.
+ */
+ header->pos.header = ring->start;
+ header->pos.data = data_start(header);
+ /* Data doesn't fit before ring->end. */
+ } else if (unlikely(next_start(header) > ring->end)) {
+ /*
+ * Wrap the data portion of a ring buffer entry to the
+ * beginning of the log device. This will update the ring
+ * free sector count to account for the unused sectors at
+ * the end of the device.
+ */
+ header->pos.data = ring->start;
+ header->wrap = WRAP_DATA;
+
+ ringbuffer_repl_log(ring)->stats.wrap++;
+ } else
+ header->wrap = WRAP_NONE;
+
+ sectors = roundup_sectors(header->pos.data_sectors);
+ BUG_ON(sectors > ring->pending);
+ ring->pending -= sectors;
+
+ sectors = next_start_adjust(ring, header);
+ if (sectors == ring->start) {
+ header->wrap = WRAP_NEXT;
+
+ ringbuffer_repl_log(ring)->stats.wrap++;
+ }
+
+ return sectors;
+}
+
+/* Slab and mempool definition. */
+struct cache_defs {
+ const enum ring_pool_type type;
+ const int min;
+ const size_t size;
+ struct kmem_cache *slab_pool;
+ const char *slab_name;
+ const size_t align;
+};
+
+/* Slab and mempool declarations. */
+static struct cache_defs cache_defs[] = {
+ { ENTRY, ENTRY_POOL_MIN, sizeof(struct ringbuffer_entry),
+ NULL, "dm_repl_log_entry", 0 },
+ { DATA_HEADER, HEADER_POOL_MIN, sizeof(struct data_header),
+ NULL, "dm_repl_log_header", 0 },
+ { DATA_HEADER_DISK, HEADER_POOL_MIN, DATA_HEADER_DISK_SIZE,
+ NULL, "dm_repl_log_disk_header", DATA_HEADER_DISK_SIZE },
+ { COPY_CONTEXT, CC_POOL_MIN, sizeof(struct slink_copy_context),
+ NULL, "dm_repl_log_copy", 0 },
+};
+
+/* Destroy all memory pools for a ring buffer. */
+static void
+ringbuffer_exit(struct ringbuffer *ring)
+{
+ mempool_t **pool = ARRAY_END(ring->pools);
+
+ sector_hash_exit(&ring->busy_sectors);
+
+ while (pool-- > ring->pools) {
+ if (likely(*pool)) {
+ mempool_destroy(*pool);
+ *pool = NULL;
+ }
+ }
+}
+
+/* Create all mempools for a ring buffer. */
+static int
+ringbuffer_init(struct ringbuffer *ring)
+{
+ int r;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+ struct cache_defs *pd = ARRAY_END(cache_defs);
+
+
+ mutex_init(&l->ringbuffer.mutex);
+ init_waitqueue_head(&ring->flushq);
+
+ /* Create slab pools. */
+ while (pd-- > cache_defs) {
+ /* Bitmap is not a slab pool. */
+ if (!pd->size)
+ continue;
+
+ ring->pools[pd->type] =
+ mempool_create_slab_pool(pd->min, pd->slab_pool);
+
+ if (unlikely(!ring->pools[pd->type])) {
+ DMERR("Error creating mempool %s", pd->slab_name);
+ goto bad;
+ }
+ }
+
+ /* Initialize busy sector hash. */
+ r = sector_hash_init(&ring->busy_sectors, l->params.dev.size);
+ if (r < 0) {
+ DMERR("Failed to allocate sector busy hash!");
+ goto bad;
+ }
+
+ return 0;
+
+bad:
+ ringbuffer_exit(ring);
+ return -ENOMEM;
+}
+
+/*
+ * Reserve space in the ring buffer for the
+ * given bio data and associated header.
+ *
+ * Correct ring->free by any skipped sectors at the end of the ring buffer.
+ */
+static int
+ringbuffer_reserve_space(struct ringbuffer *ring, struct bio *bio)
+{
+ unsigned nsectors = roundup_sectors(bio_sectors(bio));
+ sector_t end_space, start_sector;
+
+ if (!nsectors)
+ return -EPERM;
+
+ BUG_ON(!mutex_is_locked(&ring->mutex));
+
+ if (unlikely(ring->free < nsectors)) {
+ SetRingBufferFull(ring);
+ return -EBUSY;
+ }
+
+ /*
+ * Account for the sectors that are queued for do_log()
+ * but have not been accounted for on the disk. We need this
+ * calculation to see if any sectors will be lost from our
+ * free pool at the end of ring buffer.
+ */
+ start_sector = ring->next_avail + ring->pending;
+ end_space = sectors_unused(ring, start_sector);
+
+ /* if the whole I/O won't fit before the end of the disk. */
+ if (unlikely(end_space && end_space < nsectors)) {
+ sector_t skipped = end_space >= HEADER_SECTORS ?
+ sectors_unused(ring, start_sector + HEADER_SECTORS) :
+ end_space;
+
+ /* Don't subtract skipped sectors in case the bio won't fit. */
+ if (ring->free - skipped < nsectors)
+ return -EBUSY;
+
+ /*
+ * We subtract the amount of skipped sectors
+ * from ring->free here..
+ *
+ * ringbuffer_advance_head() will add them back on.
+ */
+ ring->free -= skipped;
+ }
+
+ ring->free -= nsectors;
+ ring->pending += nsectors;
+ return 0;
+}
+
+static int
+ringbuffer_empty_nolock(struct ringbuffer *ring)
+{
+ return (ring->head == ring->tail) && !RingBufferFull(ring);
+}
+
+static int
+ringbuffer_empty(struct ringbuffer *ring)
+{
+ int r;
+
+ mutex_lock(&ring->mutex);
+ r = ringbuffer_empty_nolock(ring);
+ mutex_unlock(&ring->mutex);
+
+ return r;
+}
+
+static void
+set_sync_mask(struct repl_log *l, struct ringbuffer_entry *entry)
+{
+ unsigned long slink_nr;
+
+ /* Bitmask of slinks with synchronous I/O completion policy. */
+ for_each_bit(slink_nr, ENTRY_SLINKS(entry), l->slink.max) {
+ struct dm_repl_slink *slink = slink_find(l, slink_nr);
+
+ /* Slink not configured. */
+ if (unlikely(IS_ERR(slink)))
+ continue;
+
+ /* If an slink has fallen behind an I/O threshold, it
+ * must be marked for synchronous I/O completion. */
+ if (slink_synchronous(slink) ||
+ SsSync(slink->caller))
+ slink_set_bit(slink_nr, ENTRY_SYNC(entry));
+ }
+}
+
+/*
+ * Always returns an initialized write entry,
+ * unless fatal memory allocation happens.
+ */
+static struct ringbuffer_entry *
+ringbuffer_alloc_entry(struct ringbuffer *ring, struct bio *bio)
+{
+ int dev_number, i;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+ struct ringbuffer_entry *entry = alloc_entry(ring);
+ struct data_header *header = alloc_header(ring);
+ struct data_header_region *region;
+
+ BUG_ON(!entry);
+ BUG_ON(!header);
+ memset(entry, 0, sizeof(*entry));
+ memset(header, 0, sizeof(*header));
+
+ /* Now setup the ringbuffer_entry. */
+ atomic_set(&entry->endios, 0);
+ atomic_set(&entry->ref, 0);
+ entry->ring = ring;
+ entry->data.header = header;
+ header->wrap = WRAP_NONE;
+
+ i = ARRAY_SIZE(entry->lists.l);
+ while (i--)
+ INIT_LIST_HEAD(entry->lists.l + i);
+
+ /*
+ * In case we're called with a bio, we're creating a new entry
+ * or we're allocating it for reading the header in during init.
+ */
+ if (bio) {
+ struct dm_repl_slink *slink0 = slink_find(l, 0);
+
+ _BUG_ON_PTR(slink0);
+
+ /* Setup the header region. */
+ dev_number = slink0->ops->dev_number(slink0, bio->bi_bdev);
+ BUG_ON(dev_number < 0);
+ region = &header->region;
+ region->dev = dev_number;
+ region->sector = bio_begin(bio);
+ region->size = bio->bi_size;
+ BUG_ON(!region->size);
+ header->pos.data_sectors =
+ roundup_data_sectors(bio_sectors(bio));
+
+ entry->bios.write = bio;
+ sector_range_mark_busy(entry);
+
+ /*
+ * Successfully allocated space in the ring buffer
+ * for this entry. Advance our in-memory tail pointer.
+ * Round up to HEADER_SECTORS boundary for supporting
+ * up to 4k sector sizes.
+ */
+ mutex_lock(&ring->mutex);
+ ring->next_avail = ringbuffer_inc(ring, header);
+ mutex_unlock(&ring->mutex);
+
+ /* Bitmask of slinks to initiate copies accross. */
+ memcpy(ENTRY_SLINKS(entry), LOG_SLINKS(l), BITMAP_SIZE(l));
+
+ /* Set synchronous I/O policy mask. */
+ set_sync_mask(l, entry);
+ }
+
+ /* Add header to the ordered list of headers. */
+ list_add_tail(E_ORDERED_LIST(entry), L_ENTRY_ORDERED_LIST(l));
+
+ DMDEBUG_LIMIT("%s header->pos.header=%llu header->pos.data=%llu "
+ "advancing ring->next_avail=%llu", __func__,
+ (unsigned long long) header->pos.header,
+ (unsigned long long) header->pos.data,
+ (unsigned long long) ring->next_avail);
+ return entry;
+}
+
+/* Free a ring buffer entry and the data header hanging off it. */
+static void
+ringbuffer_free_entry(struct ringbuffer_entry *entry)
+{
+ struct ringbuffer *ring;
+
+ _BUG_ON_PTR(entry);
+ _BUG_ON_PTR(entry->data.header);
+
+ ring = entry->ring;
+ _BUG_ON_PTR(ring);
+
+ /*
+ * Will need to change once ringbuffer_entry is
+ * not kept around as long as the data header.
+ */
+ if (!list_empty(E_BUSY_HASH_LIST(entry))) {
+ DMERR("%s E_BUSY_HAS_LIST not empty!", __func__);
+ BUG();
+ }
+
+ if (!list_empty(E_COPY_CONTEXT_LIST(entry))) {
+ DMERR("%s E_COPY_CONTEXT_LIST not empty!", __func__);
+ BUG();
+ }
+
+ if (!list_empty(E_ORDERED_LIST(entry)))
+ list_del(E_ORDERED_LIST(entry));
+
+ if (!list_empty(E_WRITE_OR_COPY_LIST(entry)))
+ list_del(E_WRITE_OR_COPY_LIST(entry));
+
+ free_header(entry->data.header, ring);
+ free_entry(entry, ring);
+}
+
+/* Mark a ring buffer entry invalid on the backing store device. */
+static void
+disk_header_set_invalid(void *ptr)
+{
+ ((struct data_header_disk *) ptr)->valid = 0;
+}
+
+static int
+ringbuffer_mark_entry_invalid(struct ringbuffer *ring,
+ struct ringbuffer_entry *entry)
+{
+ struct data_header *header = entry->data.header;
+
+ return header_io(WRITE, &(struct header_io_params) {
+ DATA_HEADER, ringbuffer_repl_log(ring),
+ header, header->pos.header, disk_header_set_invalid });
+}
+
+enum endio_type { HEADER_ENDIO = 0, DATA_ENDIO, NR_ENDIOS };
+static void
+endio(struct ringbuffer_entry *entry,
+ enum endio_type type, unsigned long error)
+{
+ *(type == DATA_ENDIO ? &entry->data.error.data :
+ &entry->data.error.header) = error;
+
+ if (atomic_dec_and_test(&entry->endios))
+ /*
+ * Endio processing requires disk writes to advance the log
+ * tail pointer. So, we need to defer this to process context.
+ * The endios are processed from the l->lists.entry.io list,
+ * and the entry is already on that list.
+ */
+ wake_do_log(ringbuffer_repl_log(entry->ring));
+ else
+ BUG_ON(atomic_read(&entry->endios) < 0);
+}
+
+/* Endio routine for data header io. */
+static void
+header_endio(unsigned long error, void *context)
+{
+ endio(context, HEADER_ENDIO, error);
+}
+
+/* Endio routine for data io (ie. the bio data written for an entry). */
+static void
+data_endio(unsigned long error, void *context)
+{
+ endio(context, DATA_ENDIO, error);
+}
+
+/*
+ * Place the data contained in bio asynchronously
+ * into the replog's ring buffer.
+ *
+ * This can be void, because any allocation failure is fatal and any
+ * IO errors will be reported asynchronously via dm_io() callbacks.
+ */
+static void
+ringbuffer_write_entry(struct repl_log *l, struct bio *bio)
+{
+ int i;
+ struct ringbuffer *ring = &l->ringbuffer;
+ /*
+ * ringbuffer_alloc_entry returns an entry,
+ * including an initialized data_header.
+ */
+ struct ringbuffer_entry *entry = ringbuffer_alloc_entry(ring, bio);
+ struct data_header_disk *disk_header = alloc_data_header_disk(ring);
+ struct data_header *header = entry->data.header;
+ struct dev_io_params dio[] = {
+ { /* Data IO specs. */
+ &l->params.dev, header->pos.data, bio->bi_size,
+ .mem = { DM_IO_BVEC, { .bvec = bio_iovec(bio) },
+ bio_offset(bio) },
+ .notify = { data_endio, entry }
+ },
+ { /* Header IO specs. */
+ &l->params.dev, header->pos.header, DATA_HEADER_DISK_SIZE,
+ .mem = { DM_IO_KMEM, { .addr = disk_header }, 0 },
+ .notify = { header_endio, entry }
+ },
+ };
+
+ DMDEBUG_LIMIT("in %s %u", __func__, jiffies_to_msecs(jiffies));
+ BUG_ON(!disk_header);
+ entry->data.disk_header = disk_header;
+ data_header_to_disk(BITMAP_ELEMS(l), disk_header, header);
+
+ /* Take ringbuffer IO reference out vs. slink0. */
+ ss_io_get(l->slink0->caller);
+
+ /* Add to ordered list of active entries. */
+ list_add_tail(E_WRITE_OR_COPY_LIST(entry), L_ENTRY_RING_WRITE_LIST(l));
+
+ DMDEBUG_LIMIT("%s writing header to offset=%llu and bio for "
+ "sector=%llu to sector=%llu/size=%llu", __func__,
+ (unsigned long long) entry->data.header->pos.header,
+ (unsigned long long) bio_begin(bio),
+ (unsigned long long) entry->data.header->pos.data,
+ (unsigned long long) to_sector(dio[1].size));
+
+ /*
+ * Submit the writes.
+ *
+ * 1 I/O count for header + 1 for data
+ */
+ i = ARRAY_SIZE(dio);
+ atomic_set(&entry->endios, i);
+ while (i--)
+ BUG_ON(dev_io(WRITE, ring, dio + i));
+
+ DMDEBUG_LIMIT("out %s %u", __func__, jiffies_to_msecs(jiffies));
+}
+
+/* Endio routine for bio data reads of off the ring buffer. */
+static void
+read_bio_vec_endio(unsigned long error, void *context)
+{
+ struct ringbuffer_entry *entry = context;
+ struct ringbuffer *ring = entry->ring;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+
+ atomic_dec(&entry->endios);
+ BUG_ON(!entry->bios.read);
+ bio_endio(entry->bios.read, error ? -EIO : 0);
+ entry->bios.read = NULL;
+ entry_put(entry);
+ wake_do_log(l);
+
+ /* Release IO reference on slink0. */
+ ss_io_put(l->slink0->caller);
+}
+
+/* Read bio data of off the ring buffer. */
+static void
+ringbuffer_read_bio_vec(struct repl_log *l,
+ struct ringbuffer_entry *entry, sector_t offset,
+ struct bio *bio)
+{
+ /* Data IO specs. */
+ struct dev_io_params dio = {
+ &l->params.dev,
+ entry->data.header->pos.data + offset, bio->bi_size,
+ .mem = { DM_IO_BVEC, { .bvec = bio_iovec(bio) },
+ bio_offset(bio) },
+ .notify = { read_bio_vec_endio, entry }
+ };
+
+ DMDEBUG_LIMIT("in %s %u", __func__, jiffies_to_msecs(jiffies));
+ _BUG_ON_PTR(entry);
+ entry_get(entry);
+ atomic_inc(&entry->endios);
+
+ /* Take IO reference out vs. slink0. */
+ ss_io_get(l->slink0->caller);
+
+ DMDEBUG("%s reading bio data bio for sector=%llu/size=%llu",
+ __func__, (unsigned long long) bio_begin(bio),
+ (unsigned long long) to_sector(dio.size));
+
+ /*
+ * Submit the read.
+ */
+ BUG_ON(dev_io(READ, &l->ringbuffer, &dio));
+ DMDEBUG_LIMIT("out %s %u", __func__, jiffies_to_msecs(jiffies));
+}
+
+/*
+ * Advances the ring buffer head pointer, updating the in-core data
+ * and writing it to the backing store device, but only if there are
+ * inactive entries (ie. those with copies to all slinks) at the head.
+ *
+ * Returns -ve errno on failure, otherwise the number of entries freed.
+ */
+static int
+ringbuffer_advance_head(const char *caller, struct ringbuffer *ring)
+{
+ int r;
+ unsigned entries_freed = 0;
+ sector_t sectors_freed = 0;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+ struct ringbuffer_entry *entry, *entry_last = NULL, *n;
+
+ /* Count any freeable entries and remeber last one. */
+ list_for_each_entry(entry, L_ENTRY_ORDERED_LIST(l),
+ lists.l[E_ORDERED]) {
+ /* Can't advance past dirty entry. */
+ if (entry_busy(l, ENTRY_SLINKS(entry)) ||
+ atomic_read(&entry->endios))
+ break;
+
+ BUG_ON(entry_endios_pending(entry));
+ entry_last = entry;
+ entries_freed++;
+ }
+
+ /* No entries to free. */
+ if (!entries_freed)
+ return 0;
+
+ BUG_ON(!entry_last);
+
+ /* Need safe version, because ringbuffer_free_entry removes entry. */
+ list_for_each_entry_safe(entry, n, L_ENTRY_ORDERED_LIST(l),
+ lists.l[E_ORDERED]) {
+ struct data_header *header = entry->data.header;
+
+ BUG_ON(entry_busy(l, ENTRY_SLINKS(entry)) ||
+ entry_endios_pending(entry) ||
+ atomic_read(&entry->endios));
+
+ /*
+ * If the entry wrapped around between the header and
+ * the data or if the next entry wraps, free the
+ * unused sectors at the end of the device.
+ */
+ mutex_lock(&ring->mutex);
+ sectors_freed += roundup_sectors(header->pos.data_sectors)
+ + sectors_skipped(ring, header);
+ if (likely(ring->head != ring->tail))
+ ring->head = next_start_adjust(ring, header);
+ BUG_ON(ring->head >= ring->end);
+ mutex_unlock(&ring->mutex);
+
+ /* Don't access entry after this call! */
+ ringbuffer_free_entry(entry);
+
+ if (entry == entry_last)
+ break;
+ }
+
+ DMDEBUG_LIMIT("%s (%s) advancing ring buffer head for %u "
+ "entries to %llu",
+ __func__, caller, entries_freed,
+ (unsigned long long) ring->head);
+
+ /* Update ring buffer pointers in buffer header. */
+ r = buffer_header_io(WRITE, l);
+ if (likely(!r)) {
+ /* Buffer header written... */
+ mutex_lock(&ring->mutex);
+ ring->free += sectors_freed;
+ mutex_unlock(&ring->mutex);
+ }
+
+ /* Inform caller, that we're willing to receive more I/Os. */
+ ClearRingBlocked(ring);
+ ClearRingBufferFull(ring);
+ notify_caller(l, WRITE, 0);
+ if (unlikely(r < 0))
+ ringbuffer_error(RING_BUFFER_HEAD_ERROR, ring, r);
+
+ return r ? r : entries_freed;
+}
+
+/*
+ * Advances the tail pointer after a successful
+ * write of an entry to the log.
+ */
+static int
+ringbuffer_advance_tail(struct ringbuffer_entry *entry)
+{
+ int r;
+ sector_t new_tail, old_tail;
+ struct ringbuffer *ring = entry->ring;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+ struct data_header *header = entry->data.header;
+
+/*
+ if (unlikely(ring->tail != header->pos.header)) {
+ DMERR("ring->tail %llu header->pos.header %llu",
+ (unsigned long long) ring->tail,
+ (unsigned long long) header->pos.header);
+ BUG();
+ }
+*/
+
+ mutex_lock(&ring->mutex);
+ old_tail = ring->tail;
+ /* Should we let this get out of sync? */
+ new_tail = ring->tail = next_start_adjust(ring, header);
+ BUG_ON(ring->tail >= ring->end);
+ mutex_unlock(&ring->mutex);
+
+ DMDEBUG_LIMIT("%s header->pos.header=%llu header->pos.data=%llu "
+ "ring->tail=%llu; "
+ "advancing ring tail pointer to %llu",
+ __func__,
+ (unsigned long long) header->pos.header,
+ (unsigned long long) header->pos.data,
+ (unsigned long long) ring->tail,
+ (unsigned long long) ring->tail);
+
+ r = buffer_header_io(WRITE, l);
+ if (unlikely(r < 0)) {
+ /* Return the I/O size to ring->free. */
+ mutex_lock(&ring->mutex);
+ /* Make sure it wasn't changed. */
+ BUG_ON(ring->tail != new_tail);
+ ring->tail = old_tail;
+ mutex_unlock(&ring->mutex);
+
+ ringbuffer_error(RING_BUFFER_TAIL_ERROR, ring, r);
+ }
+
+ return r;
+}
+
+/* Open type <-> name mapping. */
+static const struct dm_str_descr open_types[] = {
+ { OT_AUTO, "auto" },
+ { OT_OPEN, "open" },
+ { OT_CREATE, "create" },
+};
+
+/* Get slink policy flags. */
+static inline int
+_open_type(const char *name)
+{
+ return dm_descr_type(open_types, ARRAY_SIZE(open_types), name);
+}
+
+/* Get slink policy name. */
+static inline const char *
+_open_str(const int type)
+{
+ return dm_descr_name(open_types, ARRAY_SIZE(open_types), type);
+}
+
+/*
+ * Amount of free sectors in ring buffer. This function does not take
+ * into account unused sectors at the end of the log device.
+ */
+static sector_t
+ring_free(struct ringbuffer *ring)
+{
+ if (unlikely(ring->head == ring->next_avail))
+ return ring->end - ring->start;
+ else
+ return ring->head > ring->tail ?
+ ring->head - ring->tail :
+ (ring->head - ring->start) + (ring->end - ring->tail);
+}
+
+static struct log_header *
+alloc_log_header(struct repl_log *l)
+{
+ struct log_header *log_header =
+ kzalloc(sizeof(*log_header), GFP_KERNEL);
+
+ if (log_header)
+ l->header.log = log_header;
+
+ return log_header;
+}
+
+static void free_log_header(struct log_header *log_header,
+ struct ringbuffer *ring)
+{
+ kfree(log_header);
+}
+
+/* Create a new dirty log. */
+static int
+log_create(struct repl_log *l)
+{
+ int r;
+ struct log_header *log_header = l->header.log;
+ struct repl_dev *dev = &l->params.dev;
+ struct repl_params *params = &l->params;
+ struct ringbuffer *ring = &l->ringbuffer;
+
+ DMINFO("%s: creating new log", __func__);
+ _BUG_ON_PTR(log_header);
+
+ /* First, create the in-memory representation */
+ log_header->version.major = DM_REPL_LOG_MAJOR;
+ log_header->version.minor = DM_REPL_LOG_MINOR;
+ log_header->version.subminor = DM_REPL_LOG_MICRO;
+ log_header->size = params->dev.size;
+ log_header->buffer_header = dev->start + HEADER_SECTORS;
+
+ /* Write log header to device. */
+ r = log_header_io(WRITE, l);
+ if (unlikely(r < 0)) {
+ free_log_header(log_header, ring);
+ l->header.log = NULL;
+ return r;
+ }
+
+ /*
+ * Initialize the ring buffer.
+ *
+ * Start is behind the buffer header which follows the log header.
+ */
+ ring->start = params->dev.start;
+ ring->end = ring->start + params->dev.size;
+ ring->start += 2 * HEADER_SECTORS;
+ ring->head = ring->tail = ring->next_avail = ring->start;
+ ring->free = ring_free(ring);
+
+ DMDEBUG("%s start=%llu end=%llu free=%llu", __func__,
+ (unsigned long long) ring->start,
+ (unsigned long long) ring->end,
+ (unsigned long long) ring->free);
+
+ r = buffer_header_io(WRITE, l);
+ if (unlikely(r < 0)) {
+ free_log_header(log_header, ring);
+ l->header.log = NULL;
+ return r;
+ }
+
+ return 0;
+}
+
+/* Allocate a log_header and read header in from disk. */
+static int
+log_read(struct repl_log *l)
+{
+ int r;
+ struct log_header *log_header = l->header.log;
+ struct repl_dev *dev;
+ struct ringbuffer *ring;
+ char buf[BDEVNAME_SIZE];
+
+ _BUG_ON_PTR(log_header);
+ r = log_header_io(READ, l);
+ if (unlikely(r < 0))
+ return r;
+
+ format_dev_t(buf, l->params.dev.dm_dev->bdev->bd_dev);
+
+ /* Make sure that we can handle this version of the log. */
+ if (memcmp(&log_header->version, &my_version, sizeof(my_version)))
+ DMINFO("Found valid log header on %s", buf);
+ else
+ DM_EINVAL("On-disk version (%d.%d.%d) is "
+ "not supported by this module.",
+ log_header->version.major, log_header->version.minor,
+ log_header->version.subminor);
+
+ /*
+ * Read in the buffer_header_disk
+ */
+ r = buffer_header_io(READ, l);
+ if (unlikely(r < 0))
+ return r;
+
+ dev = &l->params.dev;
+ ring = &l->ringbuffer;
+
+ /*
+ * We'll go with the size in the log header and
+ * adjust it in the worker thread when possible.
+ */
+ ring->end = dev->start + log_header->size;
+ ring->next_avail = ring->tail;
+
+ /*
+ * The following call to ring_free is incorrect as the free
+ * space in the ring has to take into account the potential
+ * for unused sectors at the end of the device. However, once
+ * do_log_init is called, any discrepencies are fixed there.
+ */
+ ring->free = ring_free(ring);
+ return 0;
+}
+
+/*
+ * Open and read/initialize a replicator log backing store device.
+ *
+ * Must be called with dm_io client set up, because we dm_io to the device.
+ */
+/* Try to read an existing log or create a new one. */
+static int
+log_init(struct repl_log *l)
+{
+ int r;
+ struct repl_params *p = &l->params;
+ struct log_header *log_header = alloc_log_header(l);
+
+ BUG_ON(!log_header);
+
+ /* Read the log header in from disk. */
+ r = log_read(l);
+ switch (r) {
+ case 0:
+ /* Sucessfully read in the log. */
+ if (p->open_type == OT_CREATE)
+ DMERR("OT_CREATE requested: "
+ "initializing existing log!");
+ else
+ p->dev.size = l->header.log->size;
+
+ break;
+ case -EINVAL:
+ /*
+ * Most likely this is the initial create of the log.
+ * But, if this is an open, return failure.
+ */
+ if (p->open_type == OT_OPEN)
+ DMWARN("Can't create new replog on open!");
+ else
+ /* Try to create a new log. */
+ r = log_create(l);
+
+ break;
+ case -EIO:
+ DMERR("log_read IO error!");
+ break;
+ default:
+ DMERR("log_read failed with %d?", r);
+ }
+
+ return r;
+}
+
+/* Find a replog on the global list checking for bdev and start offset. */
+static struct repl_log *
+replog_find(dev_t dev, sector_t dev_start)
+{
+ struct repl_log *replog;
+
+ list_for_each_entry(replog, &replog_list, lists.l[L_REPLOG]) {
+ if (replog->params.dev.dm_dev->bdev->bd_dev == dev)
+ return likely(replog->params.dev.start == dev_start) ?
+ replog : ERR_PTR(-EINVAL);
+ }
+
+ return ERR_PTR(-ENOENT);
+}
+
+/* Clear all allocated slab objects in case of busys teardown. */
+static void
+ringbuffer_free_entries(struct ringbuffer *ring)
+{
+ struct ringbuffer_entry *entry, *n;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+
+ list_for_each_entry_safe(entry, n, L_ENTRY_ORDERED_LIST(l),
+ lists.l[E_ORDERED]) {
+ if (atomic_read(&entry->ref))
+ sector_range_clear_busy(entry);
+
+ ringbuffer_free_entry(entry);
+ }
+}
+
+static void
+replog_release(struct kref *ref)
+{
+ struct repl_log *l = container_of(ref, struct repl_log, ref);
+
+ BUG_ON(!list_empty(L_REPLOG_LIST(l)));
+ kfree(l);
+}
+
+/* Destroy replication log. */
+static void
+replog_destroy(struct repl_log *l)
+{
+ _BUG_ON_PTR(l);
+
+ if (l->io.wq)
+ destroy_workqueue(l->io.wq);
+
+ free_log_header(l->header.log, &l->ringbuffer);
+ ringbuffer_free_entries(&l->ringbuffer);
+ ringbuffer_exit(&l->ringbuffer);
+ kfree(l->io.buffer_header_disk);
+
+ if (l->io.io_client)
+ dm_io_client_destroy(l->io.io_client);
+}
+
+/* Release a reference on a replog freeing its resources on last drop. */
+static int
+replog_put(struct dm_repl_log *log, struct dm_target *ti)
+{
+ struct repl_log *l;
+
+ _SET_AND_BUG_ON_L(l, log);
+ dm_put_device(ti, l->params.dev.dm_dev);
+ return kref_put(&l->ref, replog_release);
+}
+
+/* Return ringbuffer log device size. */
+static sector_t
+replog_dev_size(struct dm_dev *dm_dev, sector_t size_wanted)
+{
+ sector_t dev_size = i_size_read(dm_dev->bdev->bd_inode) >> SECTOR_SHIFT;
+
+ return (!dev_size || size_wanted > dev_size) ? 0 : dev_size;
+}
+
+/* Get a reference on a replicator log. */
+static void do_log(struct work_struct *ws);
+static struct repl_log *
+replog_get(struct dm_repl_log *log, struct dm_target *ti,
+ const char *path, struct repl_params *params)
+{
+ int i, r;
+ dev_t dev;
+ sector_t dev_size;
+ struct dm_dev *dm_dev;
+ char buf[BDEVNAME_SIZE];
+ struct repl_log *l;
+ struct dm_io_client *io_client;
+
+ /* Get device with major:minor or device path. */
+ r = dm_get_device(ti, path, params->dev.start, params->dev.size,
+ FMODE_WRITE, &dm_dev);
+ if (r) {
+ DMERR("Failed to open replicator log device \"%s\" [%d]",
+ path, r);
+ return ERR_PTR(r);
+ }
+
+ dev = dm_dev->bdev->bd_dev;
+ dev_size = replog_dev_size(dm_dev, params->dev.size);
+ if (!dev_size)
+ return ERR_PTR(-EINVAL);
+
+ /* Check if we already have a handle to this device. */
+ mutex_lock(&list_mutex);
+ l = replog_find(dev, params->dev.start);
+ if (IS_ERR(l)) {
+ mutex_unlock(&list_mutex);
+
+ if (unlikely(l == ERR_PTR(-EINVAL))) {
+ DMERR("Device open with different start offset!");
+ dm_put_device(ti, dm_dev);
+ return l;
+ }
+ } else {
+ /* Cannot create if there is an open reference. */
+ if (params->open_type == OT_CREATE) {
+ mutex_unlock(&list_mutex);
+ DMERR("OT_CREATE requested, but existing log found!");
+ dm_put_device(ti, dm_dev);
+ return ERR_PTR(-EPERM);
+ }
+
+ /* Take reference on replication log out. */
+ kref_get(&l->ref);
+ mutex_unlock(&list_mutex);
+
+ DMINFO("Found existing replog=%s", format_dev_t(buf, dev));
+
+ /* Found one, return it. */
+ log->context = l;
+ return l;
+ }
+
+ /*
+ * There is no open log, so time to look for one on disk.
+ */
+ l = kzalloc(sizeof(*l), GFP_KERNEL);
+ if (unlikely(!l)) {
+ DMERR("failed to allocate replicator log context");
+ dm_put_device(ti, dm_dev);
+ return ERR_PTR(-ENOMEM);
+ }
+
+ /* Preserve constructor parameters. */
+ l->params = *params;
+ l->params.dev.dm_dev = dm_dev;
+
+ log->context = l;
+ l->replog = log;
+
+ /* Init basic members. */
+ rwlock_init(&l->lists.lock);
+ rwlock_init(&l->lists.slinks.lock);
+ INIT_LIST_HEAD(&l->lists.slinks.list);
+
+ i = L_NR_LISTS;
+ while (i--)
+ INIT_LIST_HEAD(l->lists.l + i);
+
+ spin_lock_init(&l->io.lock);
+ bio_list_init(&l->io.in);
+
+ /* Take first reference out. */
+ kref_init(&l->ref);
+
+ /* Initialize ring buffer. */
+ r = ringbuffer_init(&l->ringbuffer);
+ if (unlikely(r < 0)) {
+ DMERR("failed to initialize ring buffer %d", r);
+ goto bad;
+ }
+
+ /* Preallocate to avoid stalling on OOM. */
+ l->io.buffer_header_disk =
+ kzalloc(dm_round_up(sizeof(l->io.buffer_header_disk),
+ to_bytes(1)), GFP_KERNEL);
+ if (unlikely(!l->io.buffer_header_disk)) {
+ DMERR("failed to allocate ring buffer disk header");
+ r = -ENOMEM;
+ goto bad;
+ }
+
+ /*
+ * ringbuffer_io will only be called with I/O sizes of ti->split_io
+ * or fewer bytes, which are boundary checked too.
+ *
+ * The io_client needs to be setup before we can call log_init below.
+ */
+ io_client = dm_io_client_create(DEFAULT_BIOS * (1 + BIO_MAX_PAGES));
+ if (unlikely(IS_ERR(io_client))) {
+ DMERR("dm_io_client_create failed!");
+ r = PTR_ERR(io_client);
+ goto bad;
+ } else
+ l->io.io_client = io_client;
+
+ /* Create one worker per replog. */
+ l->io.wq = create_singlethread_workqueue(DAEMON);
+ if (unlikely(!l->io.wq)) {
+ DMERR("failed to create workqueue");
+ r = -ENOMEM;
+ goto bad;
+ } else
+ INIT_WORK(&l->io.ws, do_log);
+
+ /* Try to read an existing log or create a new one. */
+ r = log_init(l);
+ if (unlikely(r < 0))
+ goto bad;
+
+ stats_init(l);
+ ClearLogDevelStats(l);
+
+ /* Start out suspended, dm core will resume us. */
+ SetRingSuspended(&l->ringbuffer);
+ SetRingBlocked(&l->ringbuffer);
+
+ /* Link the new replog into the global list */
+ mutex_lock(&list_mutex);
+ list_add_tail(L_REPLOG_LIST(l), &replog_list);
+ mutex_unlock(&list_mutex);
+
+ return l;
+
+bad:
+ replog_destroy(l);
+ return ERR_PTR(r);
+}
+
+/* Account and entry for fallbehind and put on copy list. */
+static void
+entry_account_and_copy(struct ringbuffer_entry *entry)
+{
+ unsigned long slink_nr;
+ struct repl_log *l;
+
+ _BUG_ON_PTR(entry);
+ l = ringbuffer_repl_log(entry->ring);
+ _BUG_ON_PTR(l);
+
+ /* If there's no outstanding copies for this entry -> bail out. */
+ if (!entry_busy(l, ENTRY_SLINKS(entry)))
+ return;
+
+ _BUG_ON_PTR(entry->ring);
+
+ /* Account for fallbehind. */
+ for_each_bit(slink_nr, ENTRY_SLINKS(entry), l->slink.max) {
+ struct dm_repl_slink *slink = slink_find(l, slink_nr);
+
+ if (!IS_ERR(slink))
+ slink_fallbehind_inc(slink, entry);
+ }
+
+ /*
+ * Initiate copies across all SLINKS by moving to
+ * copy list in order. Because we are already processing
+ * do_log before do_slink_ios(), we need not call wake_do_log.
+ */
+ list_move_tail(E_WRITE_OR_COPY_LIST(entry), L_SLINK_COPY_LIST(l));
+}
+
+/* Adjust the log size. */
+static void
+do_log_resize(struct repl_log *l)
+{
+ int r = 0;
+ sector_t size_cur = l->header.log->size,
+ size_dev = l->params.dev.size;
+
+ /* If size change requested, adjust when possible. */
+ if (size_cur != size_dev) {
+ int write = 0;
+ int grow = size_dev > size_cur;
+ struct ringbuffer *ring = &l->ringbuffer;
+
+ mutex_lock(&ring->mutex);
+
+ /* Ringbuffer empty easy case. */
+ r = ringbuffer_empty_nolock(ring);
+ if (r) {
+ ring->head = ring->tail = \
+ ring->next_avail = ring->start;
+ write = true;
+ /* Ringbuffer grow easy case. */
+ /* FIXME: check for device size valid! */
+ } else if (grow) {
+ write = true;
+ /* Ringbuffer shrink case. */
+ } else if (ring->head < ring->tail &&
+ max(ring->tail, ring->next_avail) < size_dev)
+ write = true;
+
+ if (write) {
+ ring->end = l->header.log->size = size_dev;
+ ring->free = ring_free(ring);
+ mutex_unlock(&ring->mutex);
+
+ r = log_header_io(WRITE, l);
+ if (r)
+ DMERR("failed to write log header "
+ "while resizing!");
+ else
+ DMINFO("%sing ringbuffer to %llu sectors",
+ grow ? "grow" : "shrink",
+ (unsigned long long) size_dev);
+ } else {
+ mutex_unlock(&ring->mutex);
+ r = 0;
+ }
+
+ ClearLogResize(l);
+ }
+}
+
+/*
+ * Initialize logs incore metadata.
+ */
+static void
+do_log_init(struct repl_log *l)
+{
+ int entries = 0, r;
+ sector_t sector;
+ struct ringbuffer *ring = &l->ringbuffer;
+ struct ringbuffer_entry *entry;
+
+ /* NOOP in case we're initialized already. */
+ if (TestSetLogInitialized(l))
+ return;
+
+ DMDEBUG("%s ring->head=%llu ring->tail=%llu",
+ __func__,
+ (unsigned long long) ring->head,
+ (unsigned long long) ring->tail);
+
+ /* Nothing to do if the log is empty */
+ if (ringbuffer_empty(ring))
+ goto out;
+
+ /*
+ * Start at head and walk to tail, queuing I/O to slinks.
+ */
+ for (sector = ring->head; sector != ring->tail;) {
+ struct data_header *header;
+
+ entry = ringbuffer_alloc_entry(ring, NULL); /* No bio alloc. */
+ header = entry->data.header;
+ r = data_header_io(READ, l, header, sector);
+ if (unlikely(r < 0)) {
+ /*
+ * FIXME: as written, this is not recoverable.
+ * All ios have to be errored because
+ * of RingBufferError().
+ */
+ ringbuffer_error(RING_BUFFER_HEADER_ERROR, ring,
+ PTR_ERR(entry));
+ ringbuffer_free_entry(entry);
+ break;
+ } else {
+ /* Set synchronous I/O policy mask. */
+ set_sync_mask(l, entry);
+
+ /* Adjust ring->free for any skipped sectors. */
+ ring->free -= sectors_skipped(ring, header);
+
+ /*
+ * Mark sector range busy in case the
+ * entry hasn't been copied to slink0 yet.
+ */
+ if (slink_test_bit(0, ENTRY_SLINKS(entry)))
+ sector_range_mark_busy(entry);
+
+ /*
+ * Account entry for fallbehind and
+ * put on slink copy list if needed.
+ */
+ entry_account_and_copy(entry);
+
+ /* Advance past this entry. */
+ sector = unlikely(next_entry_wraps(header)) ?
+ ring->start : next_start(header);
+ entries++;
+ }
+ }
+
+ DMINFO("found %d entries in the log", entries);
+
+ /* Advance head past any already copied entries. */
+ r = ringbuffer_advance_head(__func__, ring);
+ if (r >= 0)
+ DMINFO("%d entries freed", r);
+ else
+ DMERR_LIMIT("Error %d advancing ring buffer head!", r);
+
+out:
+ ClearRingBlocked(ring);
+ notify_caller(l, READ, 0);
+}
+
+/*
+ * Conditionally endio a bio, when no copies on sync slinks are pending.
+ *
+ * In case an error on site link 0 occured, the bio will be errored!
+ */
+/*
+ * FIXME: in case of no synchronous site links, the entry hasn't hit
+ * the local device yet, so a potential io error on it ain't
+ * available while endio processing the bio.
+ */
+static void
+entry_nosync_endio(struct ringbuffer_entry *entry)
+{
+ struct bio *bio = entry->bios.write;
+
+ /* If all sync slinks processed (if any). */
+ if (bio && !entry_busy(ringbuffer_repl_log(entry->ring),
+ ENTRY_SYNC(entry))) {
+ DMDEBUG_LIMIT("Calling bio_endio with %u, bi_endio %p",
+ entry->data.header->region.size, bio->bi_end_io);
+
+ /* Only error in case of site link 0 errors. */
+ bio_endio(bio,
+ slink_test_bit(0, ENTRY_ERROR(entry)) ? -EIO : 0);
+ entry->bios.write = NULL;
+ }
+}
+
+/*
+ * Error endio the entries bio, mark the ring
+ * buffer entry invalid and advance the tail.
+ */
+static void
+entry_endio_invalid(struct repl_log *l, struct ringbuffer_entry *entry)
+{
+ int r;
+
+ DMDEBUG_LIMIT("entry %p header_err %lu, data_err %lu", entry,
+ entry->data.error.header, entry->data.error.data);
+ BUG_ON(!entry->bios.write);
+ bio_endio(entry->bios.write, -EIO);
+
+ /* Mark the header as invalid so it is not queued for slink copies. */
+ r = ringbuffer_mark_entry_invalid(&l->ringbuffer, entry);
+ if (unlikely(r < 0)) {
+ /* FIXME: XXX
+ * Take the device offline?
+ */
+ DMERR("%s: I/O to sector %llu of log device "
+ "failed, and failed to mark header "
+ "invalid. Taking device off-line.",
+ __func__,
+ (unsigned long long)
+ entry->data.header->region.sector);
+ }
+
+ ringbuffer_free_entry(entry);
+}
+
+static inline int
+cc_error_read(struct slink_copy_context *cc)
+{
+ return cc->error[ERR_DISK].read ||
+ cc->error[ERR_RAM].read;
+}
+
+static inline int
+cc_error_write(struct slink_copy_context *cc)
+{
+ return cc->error[ERR_DISK].write ||
+ cc->error[ERR_RAM].write;
+}
+
+static inline int
+cc_error(struct slink_copy_context *cc)
+{
+ return cc_error_read(cc) ||
+ cc_error_write(cc);
+}
+
+/*
+ * Set state of slink_copy_context to completion.
+ *
+ * slink_copy_conmtext is the object describing a *single* copy
+ * of a particular ringbuffer entry to *one* site link.
+ *
+ * Called with list lock held.
+ */
+static void
+slink_copy_complete(struct slink_copy_context *cc)
+{
+ int slink_nr;
+ struct dm_repl_slink *slink = cc->slink;
+ struct ringbuffer_entry *entry = cc->entry;
+ struct repl_log *l = ringbuffer_repl_log(entry->ring);
+
+ _BUG_ON_PTR(slink);
+ _BUG_ON_PTR(slink->caller);
+ _BUG_ON_PTR(entry);
+ _BUG_ON_PTR(l);
+ slink_nr = slink->ops->slink_number(slink);
+ _BUG_ON_SLINK_NR(l, slink_nr);
+
+ /* The entry is no longer under I/O accross this slink. */
+ slink_clear_bit(slink_nr, ENTRY_IOS(entry));
+
+ /* The slink is no longer under I/O. */
+ slink_clear_bit(slink_nr, LOG_SLINKS_IO(l));
+
+ /* Update the I/O threshold counters */
+ slink_fallbehind_dec(slink, entry);
+
+ DMDEBUG_LIMIT("processing I/O completion for slink%d", slink_nr);
+
+ if (unlikely(cc_error(cc)) &&
+ slink_test_bit(slink_nr, LOG_SLINKS(l))) {
+ slink_set_bit(slink_nr, ENTRY_ERROR(entry));
+ DMERR_LIMIT("copy on slink%d failed", slink_nr);
+ } else {
+ /* Flag entry copied to slink_nr. */
+ slink_clear_bit(slink_nr, ENTRY_SLINKS(entry));
+
+ /* Reset any sync copy request on entry to slink_nr. */
+ slink_clear_bit(slink_nr, ENTRY_SYNC(entry));
+ }
+
+ free_copy_context(cc, entry->ring);
+
+ /* Release slink state reference after completion. */
+ ss_io_put(slink->caller);
+}
+
+/* Check for entry with endios pending at ring buffer head. */
+static int
+ringbuffer_head_busy(struct repl_log *l)
+{
+ int r;
+ struct ringbuffer_entry *entry;
+
+ mutex_lock(&l->ringbuffer.mutex);
+
+ /*
+ * This shouldn't happen. Presumably this function is called
+ * when the ring buffer is overflowing, so you would expect
+ * at least one entry on the list!
+ */
+ if (unlikely(list_empty(L_ENTRY_ORDERED_LIST(l))))
+ goto out_unlock;
+
+ /* The first entry on this list is the ring head. */
+ entry = list_first_entry(L_ENTRY_ORDERED_LIST(l),
+ struct ringbuffer_entry,
+ lists.l[E_ORDERED]);
+ r = entry_endios_pending(entry);
+ mutex_unlock(&l->ringbuffer.mutex);
+ return r;
+
+out_unlock:
+ mutex_unlock(&l->ringbuffer.mutex);
+ DMERR_LIMIT("%s called with an empty ring!", __func__);
+ return 0;
+}
+
+/*
+ * Find the first ring buffer entry with outstanding copies
+ * and record each slink that hasn't completed the copy I/O.
+ */
+static int
+find_slow_slinks(struct repl_log *l, uint64_t *slow_slinks)
+{
+ int r = 0;
+ struct ringbuffer_entry *entry;
+
+ DMDEBUG("%s", __func__);
+ /* Needed for E_COPY_CONTEXT_LIST() access. */
+ list_for_each_entry(entry, L_SLINK_COPY_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ int slink_nr;
+ struct slink_copy_context *cc;
+
+ /*
+ * There may or may not be slink copy contexts hanging
+ * off of the entry. If there aren't any, it means the
+ * copy has already completed.
+ */
+ list_for_each_entry(cc, E_COPY_CONTEXT_LIST(entry), list) {
+ struct dm_repl_slink *slink = cc->slink;
+
+ slink_nr = slink->ops->slink_number(slink);
+ _BUG_ON_SLINK_NR(l, slink_nr);
+ slink_set_bit(slink_nr, slow_slinks);
+ r = 1;
+ break;
+ }
+
+ }
+
+ if (r) {
+ /*
+ * Check to see if all slinks are slow! slink0 should
+ * not be slow, one would hope! But, we need to deal
+ * with that case.
+ */
+ if (slink_test_bit(0, slow_slinks)) {
+ struct slink_state *ss;
+
+ _BUG_ON_PTR(l->slink0);
+ ss = l->slink0->caller;
+ _BUG_ON_PTR(ss);
+
+ /*
+ * If slink0 is slow, there is
+ * obviously some other problem!
+ */
+ DMWARN("%s: slink0 copy taking a long time "
+ "(%u ms)", __func__,
+ jiffies_to_msecs(jiffies) -
+ jiffies_to_msecs(ss->fb.head_jiffies));
+ r = 0;
+ } else if (!memcmp(slow_slinks, LOG_SLINKS(l),
+ sizeof(LOG_SLINKS(l))))
+ r = 0;
+
+ if (!r)
+ memset(slow_slinks, 0, BITMAP_SIZE(l));
+ }
+
+ return r;
+}
+
+/* Check if entry has ios scheduled on slow slinks. */
+static int
+entry_is_slow(struct ringbuffer_entry *entry, uint64_t *slow_slinks)
+{
+ unsigned long slink_nr;
+
+ for_each_bit(slink_nr, ENTRY_IOS(entry),
+ ringbuffer_repl_log(entry->ring)->slink.max) {
+ if (test_bit(slink_nr, (void *) slow_slinks))
+ return 1;
+ }
+
+ return 0;
+}
+
+/*
+ * Cancel slink_copies to the slinks specified in the slow_slinks bitmask.
+ *
+ * This function starts at the beginning of the ordered slink copy list
+ * and frees up ring buffer entries which are waiting only for the slow
+ * slinks. This is accomplished by marking the regions under I/O as
+ * dirty in the slink dirty logs and advancing the ring head pointer.
+ * Once a ring buffer entry is encountered that is waiting for more
+ * than just the slinks specified, the function terminates.
+ */
+static void
+repl_log_cancel_copies(struct repl_log *l, uint64_t *slow_slinks)
+{
+ int r;
+ unsigned long slink_nr;
+ struct ringbuffer *ring = &l->ringbuffer;
+ struct ringbuffer_entry *entry;
+ struct dm_repl_slink *slink;
+ struct data_header_region *region;
+ struct slink_copy_context *cc, *n;
+ static uint64_t flush_slinks[BITMAP_ELEMS_MAX],
+ flush_error[BITMAP_ELEMS_MAX],
+ stall_slinks[BITMAP_ELEMS_MAX];
+
+ DMDEBUG("%s", __func__);
+ memset(flush_slinks, 0, BITMAP_SIZE(l));
+ memset(flush_error, 0, BITMAP_SIZE(l));
+ memset(stall_slinks, 0, BITMAP_SIZE(l));
+
+ /* First walk the entry list setting region nosync state. */
+ list_for_each_entry(entry, L_SLINK_COPY_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ if (!entry_is_slow(entry, slow_slinks) ||
+ entry_endios_pending(entry))
+ break;
+
+ region = &entry->data.header->region;
+
+ /* Needed for E_COPY_CONTEXT_LIST() access. */
+ read_lock_irq(&l->lists.lock);
+
+ /* Walk the copy context list. */
+ list_for_each_entry_safe(cc, n, E_COPY_CONTEXT_LIST(entry),
+ list) {
+ slink = cc->slink;
+ _BUG_ON_PTR(slink);
+ slink_nr = slink->ops->slink_number(slink);
+ _BUG_ON_SLINK_NR(l, slink_nr);
+
+ /* Stall IO policy set. */
+ if (slink_stall(slink)) {
+ DMINFO_LIMIT("slink=%lu stall", slink_nr);
+ /*
+ * Keep stall policy in bitarray
+ * to avoid policy change race.
+ */
+ slink_set_bit(slink_nr, stall_slinks);
+ l->stats.stall++;
+ continue;
+ }
+
+ r = slink->ops->in_sync(slink,
+ region->dev, region->sector);
+ if (r)
+ slink_set_bit(slink_nr, flush_slinks);
+
+ r = slink->ops->set_sync(slink, region->dev,
+ region->sector, 0);
+ BUG_ON(r);
+ }
+
+ read_unlock_irq(&l->lists.lock);
+ }
+
+ /*
+ * The dirty logs of all devices on this slink must be flushed in
+ * this second step for performance reasons before advancing the
+ * ring head.
+ */
+ for_each_bit(slink_nr, (void *) flush_slinks, l->slink.max) {
+ slink = slink_find(l, slink_nr);
+ r = slink->ops->flush_sync(slink);
+
+ if (unlikely(r)) {
+ /*
+ * What happens when the region is
+ * marked but not flushed? Will we
+ * still get an endio?
+ * This code assumes not. -JEM
+ *
+ * If a region is marked sync, the slink
+ * code won't select it for resync,
+ * Hence we got to keep the buffer entries,
+ * because we can't assume resync is
+ * ever going to happen. -HJM
+ */
+ DMERR_LIMIT("error flushing dirty logs "
+ "on slink=%d",
+ slink->ops->slink_number(slink));
+ slink_set_bit(slink_nr, flush_error);
+ } else {
+ /* Trigger resynchronization on slink. */
+ r = slink->ops->resync(slink, 1);
+ BUG_ON(r);
+ }
+ }
+
+ /* Now release copy contexts, declaring copy completion. */
+ list_for_each_entry(entry, L_SLINK_COPY_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ if (!entry_is_slow(entry, slow_slinks) ||
+ entry_endios_pending(entry))
+ break;
+
+ /* Needed for E_COPY_CONTEXT_LIST() access. */
+ write_lock_irq(&l->lists.lock);
+
+ /* Walk the copy context list. */
+ list_for_each_entry_safe(cc, n, E_COPY_CONTEXT_LIST(entry),
+ list) {
+ slink = cc->slink;
+ slink_nr = slink->ops->slink_number(slink);
+
+ /* Stall IO policy set. */
+ if (slink_test_bit(slink_nr, stall_slinks))
+ continue;
+
+ /* Error flushing dirty log, keep entry. */
+ if (unlikely(slink_test_bit(slink_nr, flush_error)))
+ continue;
+
+ BUG_ON(list_empty(&cc->list));
+ list_del_init(&cc->list);
+
+ /* Do not reference cc after this call. */
+ slink_copy_complete(cc);
+ }
+
+ write_unlock_irq(&l->lists.lock);
+ }
+
+ /*
+ * Now advance the head pointer to free up room in the ring buffer.
+ * In case we fail here, we've got both entries in the ring buffer
+ * *and* nosync regions to recover.
+ */
+ ringbuffer_advance_head(__func__, ring);
+}
+
+/*
+ * This function is called to free up some ring buffer space when a
+ * full condition is encountered. The basic idea is to walk through
+ * the list of outstanding copies and see which slinks are slow to
+ * respond. Then, we free up as many of the entries as possible and
+ * advance the ring head.
+ */
+static void
+ring_check_fallback(struct ringbuffer *ring)
+{
+ int r;
+ struct repl_log *l = ringbuffer_repl_log(ring);
+ static uint64_t slow_slinks[BITMAP_ELEMS_MAX];
+
+ DMDEBUG("%s", __func__);
+ /*
+ * First, check to see if we can simply
+ * free entries at the head of the ring.
+ */
+ r = ringbuffer_advance_head(__func__, ring);
+ if (r > 0) {
+ DMINFO_LIMIT("%s: able to advance head", __func__);
+ return;
+ }
+
+ /*
+ * Check to see if any entries at the head of the ring buffer
+ * are currently queued for completion. If they are, then
+ * don't do anything here; simply allow the I/O completion to
+ * proceed.
+ */
+ r = ringbuffer_head_busy(l);
+ if (r) {
+ DMINFO_LIMIT("%s: endios pending.", __func__);
+ return;
+ }
+
+ /*
+ * Take a look at the first entry in the copy list with outstanding
+ * I/O and figure out which slinks are holding up progress.
+ */
+ memset(slow_slinks, 0, BITMAP_SIZE(l));
+
+ r = find_slow_slinks(l, slow_slinks);
+ if (r) {
+ DMINFO_LIMIT("%s: slow slinks found.", __func__);
+ /*
+ * Now, walk the copy list from the beginning and free
+ * any entry which is awaiting copy completion from the
+ * slow slinks. Once we hit an entry which is awaiting
+ * completion from an slink other than the slow ones, we stop.
+ */
+ repl_log_cancel_copies(l, slow_slinks);
+ } else
+ DMINFO_LIMIT("%s: no slow slinks found.", __func__);
+}
+
+static int
+entry_error(struct ringbuffer_entry *entry)
+{
+ struct entry_data *data = &entry->data;
+
+ if (unlikely(data->error.header ||
+ data->error.data)) {
+ if (data->error.header)
+ ringbuffer_error(RING_BUFFER_HEADER_ERROR,
+ entry->ring, -EIO);
+
+ if (data->error.data)
+ ringbuffer_error(RING_BUFFER_DATA_ERROR,
+ entry->ring, -EIO);
+
+ return -EIO;
+ }
+
+ return 0;
+}
+
+/*
+ * Ring buffer endio processing. The ring buffer tail cannot be
+ * advanced until both the data and data_header portions are written
+ * to the log, AND all of the buffer I/O's preceding this one are in
+ * the log have completed.
+ */
+#define MIN_ENTRIES_INACTIVE 128
+static void
+do_ringbuffer_endios(struct repl_log *l)
+{
+ int r;
+ unsigned count = 0;
+ struct ringbuffer *ring = &l->ringbuffer;
+ struct ringbuffer_entry *entry, *entry_last = NULL, *n;
+
+ DMDEBUG_LIMIT("%s", __func__);
+
+ /*
+ * The l->lists.entry.io list is sorted by on-disk order. The first
+ * entry in the list will correspond to the current ring buffer tail
+ * plus the size of the last valid entry. We process endios in
+ * order so that the tail is not advanced past unfinished entries.
+ */
+
+ list_for_each_entry(entry, L_ENTRY_RING_WRITE_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ if (atomic_read(&entry->endios))
+ break;
+
+ count++;
+ entry_last = entry;
+ }
+
+ /* No inactive entries on list -> bail out. */
+ if (!count)
+ return;
+
+ BUG_ON(!entry_last);
+
+ /* Update the tail pointer once for a list of entries. */
+ DMDEBUG_LIMIT("%s advancing ring buffer tail %u entries",
+ __func__, count);
+ r = ringbuffer_advance_tail(entry_last);
+
+ /* Now check for any errored entries. */
+ list_for_each_entry_safe(entry, n, L_ENTRY_RING_WRITE_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ struct entry_data *data = &entry->data;
+
+ _BUG_ON_PTR(data->disk_header);
+ free_data_header_disk(data->disk_header, ring);
+ data->disk_header = NULL;
+
+ ss_io_put(l->slink0->caller);
+
+ /*
+ * Tail update error before or header/data
+ * ring buffer write error -> error bio.
+ */
+ if (unlikely(r || entry_error(entry)))
+ entry_endio_invalid(l, entry);
+ else {
+ /*
+ * Handle the slink policy for sync vs. async here.
+ *
+ * Synchronous link means, that endio needs to be
+ * reported *after* the slink copy of the entry
+ * succeeded and *not* after the entry got stored
+ * in the ring buffer. -HJM
+ */
+ /* Endio bio in case of no sync slinks. */
+ entry_nosync_endio(entry);
+
+ /*
+ * Account entry for fallbehind
+ * and put on slink copy list.
+ *
+ * WARNING: removes entry from write list!
+ */
+ entry_account_and_copy(entry);
+ }
+
+ if (entry == entry_last)
+ break;
+ }
+
+ /* On ring full, check if we need to fall back to bitmap mode. */
+ if (RingBufferFull(ring))
+ ring_check_fallback(ring);
+
+ /* Wake up any waiters. */
+ wake_up(&ring->flushq);
+}
+
+/*
+ * Work all site link endios (i.e. all slink_copy contexts).
+ */
+static struct slink_copy_context *
+cc_pop(struct repl_log *l)
+{
+ struct slink_copy_context *cc;
+
+ /* Pop copy_context from copy contexts list. */
+ if (list_empty(L_SLINK_ENDIO_LIST(l)))
+ cc = NULL;
+ else {
+ cc = list_first_entry(L_SLINK_ENDIO_LIST(l),
+ struct slink_copy_context, list);
+ list_del(&cc->list);
+ }
+
+ return cc;
+}
+
+static void
+do_slink_endios(struct repl_log *l)
+{
+ int r;
+ LIST_HEAD(slink_endios);
+ struct ringbuffer *ring = &l->ringbuffer;
+ struct ringbuffer_entry *entry = NULL;
+ struct data_header *header;
+
+ DMDEBUG_LIMIT("%s", __func__);
+
+ while (1) {
+ int slink_nr;
+ struct slink_copy_context *cc;
+ struct dm_repl_slink *slink;
+
+ /* Pop copy_context from copy contexts list. */
+ write_lock_irq(&l->lists.lock);
+ cc = cc_pop(l);
+ if (!cc) {
+ write_unlock_irq(&l->lists.lock);
+ break;
+ }
+
+ /* No active copy on endios list! */
+ BUG_ON(atomic_read(&cc->cnt));
+
+ slink = cc->slink;
+ entry = cc->entry;
+
+ /* Do not reference cc after this call. */
+ slink_copy_complete(cc);
+
+ write_unlock_irq(&l->lists.lock);
+
+ _BUG_ON_PTR(slink);
+ _BUG_ON_PTR(slink->ops);
+ _BUG_ON_PTR(entry);
+
+ /*
+ * All reads are serviced from slink0 (for now), so mark
+ * sectors as no longer under I/O once the copy to slink0
+ * is complete.
+ */
+ slink_nr = slink->ops->slink_number(slink);
+ _BUG_ON_SLINK_NR(l, slink_nr);
+ if (!slink_nr)
+ sector_range_clear_busy(entry);
+
+ /* If all synchronous site links processed, endio here. */
+ entry_nosync_endio(entry);
+
+ /*
+ * Update data header on disk to reflect the ENTRY_SLINK
+ * change so that we don't pick up a copy which has
+ * finished again on restart.
+ *
+ * FIXME: this throttles throughput on fast site links.
+ */
+ header = entry->data.header;
+ _BUG_ON_PTR(header);
+ r = data_header_io(WRITE, l, header, header->pos.header);
+ if (unlikely(r < 0)) {
+ DMERR_LIMIT("Writing data header at %llu",
+ (unsigned long long) header->pos.header);
+
+ /* Flag error on all slinks because we can't recover. */
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max)
+ slink_set_bit(slink_nr, ENTRY_ERROR(entry));
+ }
+ }
+
+ /*
+ * If all slinks are up-to-date, then we can advance
+ * the ring buffer head pointer and remove the entry
+ * from the slink copy list.
+ */
+ r = ringbuffer_advance_head(__func__, ring);
+ if (r < 0)
+ DMERR_LIMIT("Error %d advancing ring buffer head!", r);
+}
+
+/*
+ * Read a bio (partially) of off log:
+ *
+ * o check if bio's data is completely in the log
+ * -> redirect N reads to the log
+ * (N = 1 for simple cases to N > 1)
+ * o check if bio's data is split between log and LD
+ * -> redirect N parts to the log
+ * -> redirect 1 part to the LD
+ * o if bio'data is on the LD
+ */
+#define DO_INFO1 \
+DMDEBUG_LIMIT("%s overlap for bio_range.start=%llu bio_range.end=%llu " \
+ "entry_range.start=%llu entry_range.end=%llu", __func__, \
+ (unsigned long long) bio_range.start, \
+ (unsigned long long) bio_range.end, \
+ (unsigned long long) entry_range.start, \
+ (unsigned long long) entry_range.end);
+#define DO_INFO2 \
+DMDEBUG_LIMIT("%s NO overlap for bio_range.start=%llu bio_range.end=%llu " \
+ "entry_range.start=%llu entry_range.end=%llu", __func__, \
+ (unsigned long long) bio_range.start, \
+ (unsigned long long) bio_range.end, \
+ (unsigned long long) entry_range.start, \
+ (unsigned long long) entry_range.end);
+static int
+bio_read(struct repl_log *l, struct bio *bio, struct list_head *buckets[2])
+{
+ int r;
+ unsigned i;
+ struct ringbuffer_entry *entry;
+ struct sector_range bio_range = {
+ .start = bio_begin(bio),
+ .end = bio_end(bio),
+ }, entry_range;
+
+ /* Figure overlapping areas. */
+ r = 0;
+ for (i = 0; buckets[i] && i < 2; i++) {
+ /* Find entry from end of bucket. */
+ list_for_each_entry_reverse(entry, buckets[i],
+ lists.l[E_BUSY_HASH]) {
+ entry_range.start = entry->data.header->region.sector;
+ entry_range.end = entry_range.start +
+ round_up_to_sector(entry->data.header->region.size);
+
+ if (ranges_overlap(&bio_range, &entry_range)) {
+ if (bio_range.start >= entry_range.start &&
+ bio_range.end <= entry_range.end) {
+ sector_t off;
+
+ entry->bios.read = bio;
+ DO_INFO1;
+ off = bio_range.start -
+ entry_range.start;
+ ringbuffer_read_bio_vec(l, entry,
+ off, bio);
+ return 0;
+ } else
+ DO_INFO2;
+ } else
+ goto out;
+ }
+ }
+
+ /*
+ * slink->ops->io() will check if region is in sync
+ * and return -EAGAIN in case the I/O needs
+ * to be delayed. Returning -ENODEV etc. is fatal.
+ *
+ * WARNING: bio->bi_bdev changed after return!
+ */
+ /*
+ * Reading of off log:
+ * o check if bio's data is completely in the log
+ * -> redirect N reads to the log
+ * (N = 1 for simple cases to N > 1)
+ * o check if bio's data is split between log and LD
+ * -> redirect N parts to the log
+ * -> redirect 1 part to the LD
+ * o if bio'data is on the LD
+ */
+out:
+ return -EAGAIN;
+}
+#undef DO_INFO1
+#undef DO_INFO2
+
+static int
+ringbuffer_read_bio(struct repl_log *l, struct bio *bio)
+{
+ int r;
+ struct ringbuffer *ring = &l->ringbuffer;
+ struct dm_repl_slink *slink0 = slink_find(l, 0);
+ struct list_head *buckets[2];
+
+ if (IS_ERR(slink0))
+ return PTR_ERR(slink0);
+
+ /*
+ * Check if there's writes pending to the area the bio intends
+ * to read and if so, satisfy request from ring buffer.
+ */
+ /* We've got writes in the log for this bio. */
+ r = ringbuffer_writes_pending(&ring->busy_sectors, bio, buckets);
+ if (r) {
+ atomic_inc(&l->stats.writes_pending);
+ r = bio_read(l, bio, buckets);
+ /* Simple case: no writes in the log for this bio. */
+ } else {
+ /*
+ * slink->ops->io() will check if region is in sync
+ * and return -EAGAIN in case the I/O needs
+ * to be delayed. Returning -ENODEV etc. is fatal.
+ *
+ * WARNING: bio->bi_bdev changed after return!
+ */
+ r = slink0->ops->io(slink0, bio, 0);
+ if (r < 0)
+ /* No retry possibility is fatal. */
+ BUG_ON(unlikely(r != -EAGAIN));
+ }
+
+ return r;
+}
+
+/* Work on any IOS queued into the ring buffer. */
+static void
+do_ringbuffer_ios(struct repl_log *l)
+{
+ int r;
+ struct bio *bio;
+ struct bio_list ios_in;
+
+ DMDEBUG_LIMIT("%s %u start", __func__, jiffies_to_msecs(jiffies));
+
+ bio_list_init(&ios_in);
+
+ /* Quickly grab the bio input list. */
+ spin_lock(&l->io.lock);
+ bio_list_merge(&ios_in, &l->io.in);
+ bio_list_init(&l->io.in);
+ spin_unlock(&l->io.lock);
+
+ while ((bio = bio_list_pop(&ios_in))) {
+ /* FATAL: ring buffer I/O error ocurred! */
+ if (unlikely(RingBufferError(&l->ringbuffer)))
+ bio_endio(bio, -EIO);
+ else if (bio_data_dir(bio) == READ) {
+ r = ringbuffer_read_bio(l, bio);
+ /* We have to wait. */
+ if (r < 0) {
+ bio_list_push(&ios_in, bio);
+ break;
+ }
+ } else
+ /* Insert new write bio into ring buffer. */
+ ringbuffer_write_entry(l, bio);
+ }
+
+ DMDEBUG_LIMIT("%s %u end ", __func__, jiffies_to_msecs(jiffies));
+
+ if (!bio_list_empty(&ios_in)) {
+ spin_lock(&l->io.lock);
+ bio_list_merge_head(&l->io.in, &ios_in);
+ spin_unlock(&l->io.lock);
+ }
+}
+
+/*
+ * Set any slinks requested by the recovery callback to accessible.
+ *
+ * Needs doing in the main worker thread in order to avoid
+ * a race between do_slink_ios() and slink_recover_callback(),
+ * which is being called asynchrnously from the slink module.
+ */
+static void
+do_slinks_accessible(struct repl_log *l)
+{
+ unsigned long slink_nr;
+
+ /* Reset any requested inaccessible bits. */
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max) {
+ if (slink_test_bit(slink_nr, LOG_SLINKS_SET_ACCESSIBLE(l))) {
+ slink_clear_bit(slink_nr, LOG_SLINKS_INACCESSIBLE(l));
+ slink_clear_bit(slink_nr, LOG_SLINKS_SET_ACCESSIBLE(l));
+ }
+ }
+}
+
+/* Drop reference on a copy context and put on endio list on last drop. */
+static void
+slink_copy_context_put(struct slink_copy_context *cc)
+{
+ DMDEBUG_LIMIT("%s", __func__);
+
+ if (atomic_dec_and_test(&cc->cnt)) {
+ int slink_nr;
+ unsigned long flags;
+ struct repl_log *l = ringbuffer_repl_log(cc->entry->ring);
+ struct dm_repl_slink *slink = cc->slink;
+
+ /* last put, schedule completion */
+ DMDEBUG_LIMIT("last put, scheduling do_log");
+
+ _BUG_ON_PTR(l);
+ _BUG_ON_PTR(slink);
+ slink_nr = slink->ops->slink_number(slink);
+ _BUG_ON_SLINK_NR(l, slink_nr);
+
+ write_lock_irqsave(&l->lists.lock, flags);
+ BUG_ON(list_empty(&cc->list));
+ list_move_tail(&cc->list, L_SLINK_ENDIO_LIST(l));
+ write_unlock_irqrestore(&l->lists.lock, flags);
+
+ wake_do_log(l);
+ } else
+ BUG_ON(atomic_read(&cc->cnt) < 0);
+}
+
+enum slink_endio_type { SLINK_ENDIO_RAM, SLINK_ENDIO_DISK };
+static void
+slink_copy_endio(enum slink_endio_type type, int read_err, int write_err,
+ void *context)
+{
+ struct slink_copy_context *cc = context;
+ struct slink_copy_error *error;
+
+ DMDEBUG_LIMIT("%s", __func__);
+ _BUG_ON_PTR(cc);
+ error = cc->error;
+
+ if (type == SLINK_ENDIO_RAM) {
+ /* On RAM endio error, no disk callback will be performed. */
+ if (unlikely(read_err || write_err))
+ atomic_dec(&cc->cnt);
+
+ error += ERR_RAM;
+ } else
+ error += ERR_DISK;
+
+ error->read = read_err;
+ error->write = write_err;
+ slink_copy_context_put(cc);
+}
+
+/* Callback for copy in RAM. */
+static void
+slink_copy_ram_endio(int read_err, int write_err, void *context)
+{
+ slink_copy_endio(SLINK_ENDIO_RAM, read_err, write_err, context);
+}
+
+/* Callback for copy on disk. */
+static void
+slink_copy_disk_endio(int read_err, int write_err, void *context)
+{
+ slink_copy_endio(SLINK_ENDIO_DISK, read_err, write_err, context);
+}
+
+/*
+ * Called back when:
+ *
+ * o site link recovered from failure
+ * o site link recovered a region.
+ */
+static void
+slink_recover_callback(int read_err, int write_err, void *context)
+{
+ unsigned slink_nr;
+ struct repl_log *l;
+ struct slink_state *ss = context;
+
+ _BUG_ON_PTR(ss);
+ l = ss->l;
+ _BUG_ON_PTR(l);
+ slink_nr = ss->slink_nr;
+ _BUG_ON_SLINK_NR(l, slink_nr);
+
+ DMDEBUG_LIMIT("%s slink=%d", __func__, slink_nr);
+
+ if (!read_err && !write_err)
+ slink_set_bit(slink_nr, LOG_SLINKS_SET_ACCESSIBLE(l));
+
+ /* Inform caller, that we're willing to receive more I/Os. */
+ notify_caller(l, WRITE, 0);
+
+ /* Wakeup worker to allow for further IO. */
+ wake_do_log(l);
+}
+
+/* Initialize slink_copy global properties independent of entry. */
+static void
+slink_copy_init(struct dm_repl_slink_copy *slink_copy, struct repl_log *l)
+{
+ /*
+ * The source block device (ie. the ring buffer device)
+ * is the same for all I/Os.
+ */
+ slink_copy->src.type = DM_REPL_SLINK_BLOCK_DEVICE;
+ slink_copy->src.dev.bdev = repl_log_bdev(l);
+
+ /* The destination is identified by slink and device number. */
+ slink_copy->dst.type = DM_REPL_SLINK_DEV_NUMBER;
+
+ /* RAM, disk, slink recovery callbacks. */
+ slink_copy->ram.fn = slink_copy_ram_endio;
+ slink_copy->disk.fn = slink_copy_disk_endio;
+}
+
+/* Initialize slink_copy global properties dependent of entry. */
+static void
+slink_copy_addr(struct dm_repl_slink_copy *slink_copy,
+ struct ringbuffer_entry *entry)
+{
+ struct data_header *header = entry->data.header;
+ struct data_header_region *region;
+
+ _BUG_ON_PTR(header);
+ region = &header->region;
+ _BUG_ON_PTR(region);
+
+ /* The offset/size to copy from is given by the entry. */
+ slink_copy->src.sector = header->pos.data;
+
+ /* Most of the destination is the same across slinks. */
+ slink_copy->dst.dev.number.dev = region->dev;
+ slink_copy->dst.sector = region->sector;
+ slink_copy->size = region->size;
+}
+
+/* Allocate and initialize and slink_copy_context structure. */
+static inline struct slink_copy_context *
+slink_copy_context_alloc(struct ringbuffer_entry *entry,
+ struct dm_repl_slink *slink)
+{
+ struct slink_copy_context *cc = alloc_copy_context(entry->ring);
+
+ BUG_ON(!cc);
+ memset(cc, 0, sizeof(*cc));
+
+ /* NR_ENDIOS # of endios callbacks per copy (RAM and disk). */
+ atomic_set(&cc->cnt, NR_ENDIOS);
+ cc->entry = entry;
+ cc->slink = slink;
+ cc->start_jiffies = jiffies;
+ return cc;
+}
+
+/* Trigger/prohibit resynchronization on all site links. */
+enum resync_switch { RESYNC_OFF = 0, RESYNC_ON };
+static void
+resync_on_off(struct repl_log *l, enum resync_switch resync)
+{
+ unsigned long slink_nr;
+ struct dm_repl_slink *slink;
+
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max) {
+ slink = slink_find(l, slink_nr);
+ if (!IS_ERR(slink))
+ slink->ops->resync(slink, resync);
+ }
+}
+
+/* Return true if all slinks processed (either active or inaccessible). */
+static int
+all_slinks_processed(struct repl_log *l)
+{
+ unsigned slinks = 0;
+ unsigned long slink_nr;
+
+ for_each_bit(slink_nr, LOG_SLINKS_IO(l), l->slink.max)
+ slinks++;
+
+ for_each_bit(slink_nr, LOG_SLINKS_INACCESSIBLE(l), l->slink.max)
+ slinks++;
+
+ return slinks >= l->slink.count;
+}
+
+/*
+ * Work all site link copy orders.
+ */
+static void
+do_slink_ios(struct repl_log *l)
+{
+ unsigned long slink_nr;
+ struct ringbuffer_entry *entry;
+ struct dm_repl_slink *slink;
+ static struct dm_repl_slink_copy slink_copy;
+
+ /* If there's no entries on the copy list, allow resync. */
+ if (list_empty(L_SLINK_COPY_LIST(l)))
+ return resync_on_off(l, RESYNC_ON);
+
+ /*
+ * ...else prohibit resync.
+ *
+ * We'll deal with any active resynchronization based
+ * on the return code of slink->ops->copy() below.
+ */
+ resync_on_off(l, RESYNC_OFF);
+
+ /*
+ * This list is ordered, how do we keep it so that endio processing
+ * is ordered? We need this so that head pointer advances in order.
+ *
+ * We do that by changing ringbuffer_advance_head() to check
+ * for entry_busy(l, ENTRY_SLINKS(entry))) and stop processing. -HJM
+ */
+
+ /* Initialize global properties, which are independent of the entry. */
+ slink_copy_init(&slink_copy, l);
+
+ /* Walk all entries on the slink copy list. */
+ list_for_each_entry(entry, L_SLINK_COPY_LIST(l),
+ lists.l[E_WRITE_OR_COPY]) {
+ int r;
+ unsigned copies = 0;
+
+ /* Check, if all slinks processed now. */
+ r = all_slinks_processed(l);
+ if (r)
+ break;
+
+ /* Set common parts independent of slink up. */
+ slink_copy_addr(&slink_copy, entry);
+
+ /* Walk all slinks, which still need this entry. */
+ for_each_bit(slink_nr, ENTRY_SLINKS(entry), l->slink.max) {
+ int teardown;
+ struct slink_copy_context *cc;
+ struct slink_state *ss;
+
+ /*
+ * One maximum write pending to slink already
+ * -or-
+ * slink is recovering this region.
+ */
+ if (slink_test_bit(slink_nr, LOG_SLINKS_IO(l)) ||
+ slink_test_bit(slink_nr,
+ LOG_SLINKS_INACCESSIBLE(l)))
+ continue;
+
+ /*
+ * Check for deleted or being torn down site link.
+ */
+ slink = slink_find(l, slink_nr);
+ if (unlikely(IS_ERR(slink))) {
+ DMERR_LIMIT("%s no slink!", __func__);
+ ss = NULL;
+ teardown = 0;
+ } else {
+ ss = slink->caller;
+ _BUG_ON_PTR(ss);
+ teardown = SsTeardown(ss);
+ }
+
+ if (unlikely(IS_ERR(slink) ||
+ teardown ||
+ !slink_test_bit(slink_nr,
+ LOG_SLINKS(l)))) {
+drop_copy:
+ if (IS_ERR(slink))
+ DMERR_LIMIT("%s: slink %lu not "
+ "configured!",
+ __func__, slink_nr);
+ else
+ /* Correct fallbehind account. */
+ slink_fallbehind_dec(slink, entry);
+
+ /* Flag entry copied to slink_nr. */
+ slink_clear_bit(slink_nr, ENTRY_SLINKS(entry));
+
+ /* Reset any sync copy request to slink_nr. */
+ slink_clear_bit(slink_nr, ENTRY_SYNC(entry));
+
+ if (!slink_nr)
+ sector_range_clear_busy(entry);
+
+ continue;
+ }
+
+ /* Take slink reference out. */
+ ss_io_get(ss);
+
+ /* Flag active copy to slink+entry, */
+ slink_set_bit(slink_nr, LOG_SLINKS_IO(l));
+ slink_set_bit(slink_nr, ENTRY_IOS(entry));
+
+ /* Fill in the destination slink number. */
+ slink_copy.dst.dev.number.slink = slink_nr;
+
+ /* Setup the callback data. */
+ cc = slink_copy_context_alloc(entry, slink);
+ BUG_ON(!cc);
+ slink_copy.ram.context = slink_copy.disk.context = cc;
+
+ /*
+ * Add to entrys copy list of active copies in
+ * order to avoid race with ->copy() endio function
+ * accessing cc->list.
+ */
+ write_lock_irq(&l->lists.lock);
+ list_add_tail(&cc->list, E_COPY_CONTEXT_LIST(entry));
+ write_unlock_irq(&l->lists.lock);
+
+ DMDEBUG("slink0->ops->copy() from log, sector=%llu, "
+ "size=%u to dev_number=%d, sector=%llu "
+ "on slink=%u",
+ (unsigned long long) slink_copy.src.sector,
+ slink_copy.size,
+ slink_copy.dst.dev.number.dev,
+ (unsigned long long) slink_copy.dst.sector,
+ slink_copy.dst.dev.number.slink);
+
+
+ /*
+ * slink->ops->copy() may return:
+ *
+ * o -EAGAIN in case of prohibiting I/O because
+ * of device inaccessibility/suspension
+ * or device I/O errors
+ * (i.e. link temporarilly down) ->
+ * caller is allowed to retry the I/O later once
+ * he'll have received a callback.
+ *
+ * o -EACCES in case a region is being resynchronized
+ * and the source region is being read to copy data
+ * accross to the same region of the replica (RD) ->
+ * caller is allowed to retry the I/O later once
+ * he'll have received a callback.
+ *
+ * o -ENODEV in case a device is not configured
+ * caller must drop the I/O to the device/slink pair.
+ *
+ * o -EPERM in case a region is out of sync ->
+ * caller must drop the I/O to the device/slink pair.
+ */
+ r = slink->ops->copy(slink, &slink_copy, 0);
+ if (unlikely(r < 0)) {
+ DMDEBUG_LIMIT("Copy to slink%d/dev%d/"
+ "sector=%llu failed with %d.",
+ slink_copy.dst.dev.number.slink,
+ slink_copy.dst.dev.number.dev,
+ (unsigned long long)
+ slink_copy.dst.sector, r);
+
+ /*
+ * Failed -> take off entrys copies list
+ * and free copy contrext.
+ */
+ write_lock_irq(&l->lists.lock);
+ list_del_init(&cc->list);
+ write_unlock_irq(&l->lists.lock);
+
+ free_copy_context(cc, entry->ring);
+
+ /* Reset active I/O on slink+entry. */
+ slink_clear_bit(slink_nr, LOG_SLINKS_IO(l));
+ slink_clear_bit(slink_nr, ENTRY_IOS(entry));
+
+ /* Release slink reference. */
+ ss_io_put(ss);
+
+ /*
+ * Source region is being read for recovery
+ * or device is temporarilly inaccessible ->
+ * retry later once accessible again.
+ */
+ if (r == -EACCES ||
+ r == -EAGAIN) {
+ slink_set_bit(slink_nr,
+ LOG_SLINKS_INACCESSIBLE(l));
+
+ /*
+ * Device not on slink
+ * -or-
+ * region not in sync -> avoid copy.
+ */
+ } else if (r == -ENODEV ||
+ r == -EPERM)
+ goto drop_copy;
+ else
+ BUG();
+ } else
+ copies++;
+ }
+
+ if (copies)
+ l->stats.copy[copies > 1]++;
+ }
+}
+
+/* Unplug device queues with entries on all site links. */
+static void
+do_unplug(struct repl_log *l)
+{
+ struct dm_repl_slink *slink;
+ unsigned long slink_nr;
+
+ /* Conditionally unplug ring buffer. */
+ if (TestClearRingBufferIOQueued(&l->ringbuffer))
+ blk_unplug(bdev_get_queue(ringbuffer_bdev(&l->ringbuffer)));
+
+ /* Unplug any devices with queued IO on site links. */
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max) {
+ slink = slink_find(l, slink_nr);
+ if (!IS_ERR(slink))
+ slink->ops->unplug(slink);
+ }
+}
+
+/* Take out/drop slink state references to synchronize with slink delition. */
+enum reference_type { REF_GET, REF_PUT };
+static inline void
+ss_ref(enum reference_type type, struct repl_log *l)
+{
+ unsigned long slink_nr;
+ void (*f)(struct slink_state *) =
+ type == REF_GET ? ss_io_get : ss_io_put;
+
+ if (!l->slink0)
+ return;
+
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max) {
+ struct dm_repl_slink *slink =
+ l->slink0->ops->slink(l->replog, slink_nr);
+
+ _BUG_ON_PTR(slink);
+ f(slink->caller);
+ }
+}
+
+/*
+ * Worker thread.
+ *
+ * Belabour any:
+ * o replicator log ring buffer initialization
+ * o endios on the ring buffer
+ * o endios on any site links
+ * o I/O on site links (copies of buffer entries via site links to [LR]Ds
+ * o I/O to the ring buffer
+ *
+ */
+static void
+do_log(struct work_struct *ws)
+{
+ struct repl_log *l = container_of(ws, struct repl_log, io.ws);
+
+ /* Take out references vs. removal races. */
+ spin_lock(&l->io.lock);
+ ss_ref(REF_GET, l);
+ spin_unlock(&l->io.lock);
+
+ if (!RingSuspended(&l->ringbuffer)) {
+ do_log_init(l);
+ do_log_resize(l);
+ }
+
+ /* Allow for endios at any time, even while suspended. */
+ do_ringbuffer_endios(l); /* Must be called before do_slink_ios. */
+
+ /* Don't allow for new I/Os while suspended. */
+ if (!RingSuspended(&l->ringbuffer)) {
+ int r;
+
+ do_slink_endios(l);
+ do_ringbuffer_ios(l);
+
+ /*
+ * Set any slinks requested to accessible
+ * before checking all_slinks_processed().
+ */
+ do_slinks_accessible(l);
+
+ /* Only initiate slink copies if not all slinks active. */
+ r = all_slinks_processed(l);
+ if (!r)
+ do_slink_ios(l);
+
+ do_unplug(l);
+ }
+
+ ss_ref(REF_PUT, l);
+}
+
+/*
+ * Start methods of "default" type
+ */
+/* Destroy a replicator log context. */
+static void
+ringbuffer_dtr(struct dm_repl_log *log, struct dm_target *ti)
+{
+ struct repl_log *l;
+
+ DMDEBUG("%s: log %p", __func__, log);
+ _SET_AND_BUG_ON_L(l, log);
+
+ /* Remove from the global list of replogs. */
+ mutex_lock(&list_mutex);
+ list_del_init(L_REPLOG_LIST(l));
+ mutex_unlock(&list_mutex);
+
+ replog_destroy(l);
+ BUG_ON(!replog_put(log, ti));
+}
+
+/*
+ * Construct a replicator log context.
+ *
+ * Arguments:
+ * #replog_params dev_path dev_start [auto/create/open [size]
+ *
+ * dev_path = device path of replication log (REPLOG) backing store
+ * dev_start = offset in sectors to REPLOG header
+ *
+ * auto = causes open of an REPLOG with a valid header or
+ * creation of a new REPLOG in case the header's invalid.
+ * <#replog_params> = 2 or (3 and "open")
+ * -> the cache device must be initialized or the constructor will fail.
+ * <#replog_params> = 4 and "auto"
+ * -> if not already initialized, the log device will get initialized
+ * and sized to "size", otherwise it'll be opened.
+ * <#replog_params> = 4 and 'create'
+ * -> the log device will get initialized if not active and sized to
+ * "size"; if the REPLOG is active 'create' will fail.
+ *
+ * The above roughly translates to:
+ * argv[0] == #params
+ * argv[1] == dev_name
+ * argv[2] == dev_start
+ * argv[3] == OT_OPEN|OT_CREATE|OT_AUTO
+ * argv[4] == size in sectors
+ */
+#define MIN_ARGS 3
+static int
+ringbuffer_ctr(struct dm_repl_log *log, struct dm_target *ti,
+ unsigned argc, char **argv)
+{
+ int open_type, params;
+ unsigned long long tmp;
+ struct repl_log *l;
+ struct repl_params p;
+
+ SHOW_ARGV;
+
+ if (unlikely(argc < MIN_ARGS))
+ DM_EINVAL("%s: at least 3 args required, only got %d\n",
+ __func__, argc);
+
+ memset(&p, 0, sizeof(p));
+
+ /* Get # of parameters. */
+ if (unlikely(sscanf(argv[0], "%d", ¶ms) != 1 ||
+ params < 2 ||
+ params > 5)) {
+ DM_EINVAL("invalid replicator log device start");
+ } else
+ p.count = params;
+
+ if (params == 2)
+ open_type = OT_OPEN;
+ else {
+ open_type = _open_type(argv[3]);
+ if (unlikely(open_type < 0))
+ return -EINVAL;
+ else if (unlikely(open_type == OT_OPEN && params > 3))
+ DM_EINVAL("3 arguments required for open, %d given.",
+ params);
+ }
+
+ p.open_type = open_type;
+
+ if (params > 3) {
+ /* Get device size argument. */
+ if (unlikely(sscanf(argv[4], "%llu", &tmp) != 1 ||
+ tmp < LOG_SIZE_MIN)) {
+ DM_EINVAL("invalid replicator log device size");
+ } else
+ p.dev.size = tmp;
+
+ } else
+ p.dev.size = LOG_SIZE_MIN;
+
+ if (unlikely((open_type == OT_AUTO || open_type == OT_CREATE) &&
+ params < 4))
+ DM_EINVAL("4 arguments required for auto and create");
+
+ /* Get device start argument. */
+ if (unlikely(sscanf(argv[2], "%llu", &tmp) != 1))
+ DM_EINVAL("invalid replicator log device start");
+ else
+ p.dev.start = tmp;
+
+ /* Get a reference on the replog. */
+ l = replog_get(log, ti, argv[1], &p);
+ if (unlikely(IS_ERR(l)))
+ return PTR_ERR(l);
+
+ return 0;
+}
+
+/* Flush the current log contents. This function may block. */
+static int
+ringbuffer_flush(struct dm_repl_log *log)
+{
+ struct repl_log *l;
+ struct ringbuffer *ring;
+
+ DMDEBUG("%s", __func__);
+ _SET_AND_BUG_ON_L(l, log);
+ ring = &l->ringbuffer;
+
+ wake_do_log(l);
+ wait_event(ring->flushq, ringbuffer_empty(ring));
+ return 0;
+}
+
+/* Suspend method. */
+/*
+ * FIXME: we're suspending/resuming the whole ring buffer,
+ * not just the device requested. Avoiding this complete
+ * suspension would afford knowledge on the reason for the suspension.
+ * E.g. in case of device removal, we could avoid suspending completely.
+ * Don't know how we can optimize this w/o a bitmap
+ * for the devices, hence limiting dev_numbers. -HJM
+ */
+static int
+ringbuffer_postsuspend(struct dm_repl_log *log, int dev_number)
+{
+ struct repl_log *l;
+
+ _SET_AND_BUG_ON_L(l, log);
+ flush_workqueue(l->io.wq);
+
+ if (TestSetRingSuspended(&l->ringbuffer))
+ DMWARN("%s ring buffer already suspended", __func__);
+
+ flush_workqueue(l->io.wq);
+ SetRingBlocked(&l->ringbuffer);
+ ss_all_wait_on_ios(l);
+ return 0;
+}
+
+/* Resume method. */
+static int
+ringbuffer_resume(struct dm_repl_log *log, int dev_number)
+{
+ struct repl_log *l;
+ struct ringbuffer *ring;
+
+ _SET_AND_BUG_ON_L(l, log);
+
+ ring = &l->ringbuffer;
+ if (!TestClearRingSuspended(ring))
+ DMWARN("%s ring buffer already resumed", __func__);
+
+ ClearRingBlocked(ring);
+ notify_caller(l, WRITE, 0);
+ wake_do_log(l);
+ return 0;
+}
+
+/*
+ * Queue a bio to the worker thread ensuring, that
+ * there's enough space for writes in the ring buffer.
+ */
+static inline int
+queue_bio(struct repl_log *l, struct bio *bio)
+{
+ int rw = bio_data_dir(bio);
+ struct ringbuffer *ring = &l->ringbuffer;
+
+ /*
+ * Try reserving space for the bio in the
+ * buffer and mark the sector range busy.
+ */
+ if (rw == WRITE) {
+ int r;
+
+ mutex_lock(&ring->mutex);
+ r = ringbuffer_reserve_space(ring, bio);
+ mutex_unlock(&ring->mutex);
+
+ /* Ring buffer full. */
+ if (r < 0)
+ return r;
+ }
+
+ spin_lock(&l->io.lock);
+ bio_list_add(&l->io.in, bio);
+ spin_unlock(&l->io.lock);
+
+ atomic_inc(l->stats.io + !!rw);
+ wake_do_log(l);
+ return 0;
+}
+
+/*
+ * Read a bio either from a replicator log's ring buffer
+ * or from the replicated device if no buffer entry.
+ * - or-
+ * write a bio to a replicator log's ring
+ * buffer (increments buffer tail).
+ *
+ * This includes buffer allocation in case of a write and
+ * inititation of copies accross an/multiple SLINK(s).
+ *
+ * In case of a read with (partial) writes in the buffer,
+ * the replog may postpone the read until the buffer content has
+ * been copied accross the local SLINK *or* optimize by reading
+ * (parts of) the bio off the buffer.
+ */
+/*
+ * Returns 0 on success, -EWOULDBLOCK if this is a WRITE request
+ * and buffer space could not be allocated. Returns -EWOULDBLOCK if
+ * this is a READ request and the call would block due to the
+ * requested region being currently under WRITE I/O.
+ */
+static int
+ringbuffer_io(struct dm_repl_log *log, struct bio *bio, unsigned long long tag)
+{
+ int r = 0;
+ struct repl_log *l;
+ struct ringbuffer *ring;
+
+ _SET_AND_BUG_ON_L(l, log);
+ ring = &l->ringbuffer;
+
+ if (RingBlocked(ring) ||
+ !LogInitialized(l))
+ goto out_blocked;
+
+ if (unlikely(RingSuspended(ring)))
+ goto set_blocked;
+
+ /*
+ * Queue writes to the daemon in order to avoid sleeping
+ * on allocations. queue_bio() checks to see if there is
+ * enough space in the log for this bio and all of the
+ * other bios currently queued for the daemon.
+ */
+ r = queue_bio(l, bio);
+ if (!r)
+ return r;
+
+set_blocked:
+ SetRingBlocked(ring);
+out_blocked:
+ DMDEBUG_LIMIT("%s Ring blocked", __func__);
+ return -EWOULDBLOCK;
+}
+
+/* Set maximum slink # for bitarray access optimization. */
+static void replog_set_slink_max(struct repl_log *l)
+{
+ unsigned long bit_nr;
+
+ l->slink.max = 0;
+ for_each_bit(bit_nr, LOG_SLINKS(l), MAX_DEFAULT_SLINKS)
+ l->slink.max = bit_nr;
+
+ l->slink.max++;
+ BITMAP_ELEMS(l) = dm_div_up(dm_div_up(l->slink.max, BITS_PER_BYTE),
+ sizeof(uint64_t));
+ BITMAP_SIZE(l) = BITMAP_ELEMS(l) * sizeof(uint64_t);
+}
+
+/* Set replog global I/O notification function and context. */
+static void
+ringbuffer_io_notify_fn_set(struct dm_repl_log *log,
+ dm_repl_notify_fn fn, void *notify_context)
+{
+ struct repl_log *l;
+
+ _SET_AND_BUG_ON_L(l, log);
+
+ spin_lock(&l->io.lock);
+ l->notify.fn = fn;
+ l->notify.context = notify_context;
+ spin_unlock(&l->io.lock);
+}
+
+/* Add (tie) a site link to a replication log for SLINK copy processing. */
+static int
+ringbuffer_slink_add(struct dm_repl_log *log, struct dm_repl_slink *slink)
+{
+ int slink_nr;
+ struct repl_log *l;
+ struct slink_state *ss;
+
+ /* FIXME: XXX lock the repl_log */
+ DMDEBUG("ringbuffer_slink_add");
+ _BUG_ON_PTR(slink);
+ _SET_AND_BUG_ON_L(l, log);
+
+ /* See if slink was already added. */
+ slink_nr = slink->ops->slink_number(slink);
+ if (slink_nr >= MAX_DEFAULT_SLINKS)
+ DM_EINVAL("slink number larger than maximum "
+ "for 'default' replication log.");
+
+ DMDEBUG("%s: attempting to add slink%d", __func__, slink_nr);
+
+ /* No entry -> add a new one. */
+ ss = kzalloc(sizeof(*ss), GFP_KERNEL);
+ if (unlikely(!ss))
+ return -ENOMEM;
+
+ ss->slink_nr = slink_nr;
+ ss->l = l;
+ atomic_set(&ss->io.in_flight, 0);
+ init_waitqueue_head(&ss->io.waiters);
+
+ spin_lock(&l->io.lock);
+
+ if (unlikely(slink->caller)) {
+ spin_unlock(&l->io.lock);
+ kfree(ss);
+ DMERR("slink already exists.");
+ return -EEXIST;
+ }
+
+ ClearSsTeardown(ss);
+
+ /* Keep slink state reference. */
+ slink->caller = ss;
+
+ if (!slink_nr)
+ l->slink0 = slink;
+
+ l->slink.count++;
+
+ /* Set site link recovery notification. */
+ slink->ops->recover_notify_fn_set(slink, slink_recover_callback, ss);
+
+ /* Update log_header->slinks bit mask before setting max slink #! */
+ slink_set_bit(slink_nr, LOG_SLINKS(l));
+
+ /* Set maximum slink # for bitarray access optimization. */
+ replog_set_slink_max(l);
+
+ spin_unlock(&l->io.lock);
+ return 0;
+}
+
+/* Remove (untie) a site link from a replication log. */
+/*
+ * How do we tell if this is a configuration change or just a shutdown?
+ * After _repl_ctr, the RDs on the site link are either there or not.
+ */
+static int
+ringbuffer_slink_del(struct dm_repl_log *log, struct dm_repl_slink *slink)
+{
+ int r, slink_nr;
+ struct repl_log *l;
+ struct ringbuffer *ring;
+ struct slink_state *ss;
+
+ DMDEBUG("%s", __func__);
+ _BUG_ON_PTR(slink);
+ _SET_AND_BUG_ON_L(l, log);
+ ring = &l->ringbuffer;
+
+ /* Find entry to be deleted. */
+ slink_nr = slink->ops->slink_number(slink);
+ DMDEBUG("%s slink_nr=%d", __func__, slink_nr);
+
+ spin_lock(&l->io.lock);
+ ss = slink->caller;
+ if (likely(ss)) {
+ BUG_ON(atomic_read(&ss->io.in_flight));
+
+ /* No new I/Os on this slink and no duplicate deletion calls. */
+ if (TestSetSsTeardown(ss)) {
+ spin_unlock(&l->io.lock);
+ return -EPERM;
+ }
+
+ /* Wait on worker and any async I/O to finish on site link. */
+ do {
+ spin_unlock(&l->io.lock);
+ ss_wait_on_io(ss);
+ spin_lock(&l->io.lock);
+
+ if (!ss_io(ss)) {
+ slink_clear_bit(slink_nr, LOG_SLINKS(l));
+ slink->caller = NULL;
+ slink->ops->recover_notify_fn_set(slink,
+ NULL, NULL);
+ if (!slink_nr)
+ l->slink0 = NULL;
+
+ l->slink.count--;
+ replog_set_slink_max(l); /* Set l->slink.max. */
+ }
+ } while (slink->caller);
+
+ spin_unlock(&l->io.lock);
+
+ BUG_ON(l->slink.count < 0);
+ kfree(ss);
+ DMDEBUG("%s removed slink=%u", __func__, slink_nr);
+ r = 0;
+ } else {
+ spin_unlock(&l->io.lock);
+ r = -EINVAL;
+ }
+
+ wake_do_log(l);
+ return r;
+}
+
+/* Return head of the list of site links for this replicator log. */
+static struct dm_repl_log_slink_list
+*ringbuffer_slinks(struct dm_repl_log *log)
+{
+ struct repl_log *l;
+
+ _SET_AND_BUG_ON_L(l, log);
+ return &l->lists.slinks;
+}
+
+/* Return maximum number of supported site links. */
+static int
+ringbuffer_slink_max(struct dm_repl_log *log)
+{
+ return MAX_DEFAULT_SLINKS;
+}
+
+/*
+ * Message interface
+ *
+ * 'sta[tistics] {on,of[f],r[eset]}' # e.g. 'stat of'
+ */
+static int
+ringbuffer_message(struct dm_repl_log *log, unsigned argc, char **argv)
+{
+ static const char stat[] = "statistics";
+ static const char resize[] = "resize";
+ struct repl_log *l;
+
+ _SET_AND_BUG_ON_L(l, log);
+
+ if (argc != 2)
+ DM_EINVAL("Invalid number of arguments.");
+
+ if (!strnicmp(STR_LEN(argv[0], stat))) {
+ if (!strnicmp(STR_LEN(argv[1], "on")))
+ set_bit(LOG_DEVEL_STATS, &l->io.flags);
+ else if (!strnicmp(STR_LEN(argv[1], "off")))
+ clear_bit(LOG_DEVEL_STATS, &l->io.flags);
+ else if (!strnicmp(STR_LEN(argv[1], "reset")))
+ stats_init(l);
+ else
+ DM_EINVAL("Invalid '%s' arguments.", stat);
+ } else if (!strnicmp(STR_LEN(argv[0], resize))) {
+ if (TestSetLogResize(l))
+ DM_EPERM("Log resize already in progress");
+ else {
+ unsigned long long tmp;
+ sector_t dev_size;
+
+ if (unlikely(sscanf(argv[1], "%llu", &tmp) != 1) ||
+ tmp < LOG_SIZE_MIN)
+ DM_EINVAL("Invalid log %s argument.", resize);
+
+ dev_size = replog_dev_size(l->params.dev.dm_dev, tmp);
+ if (!dev_size)
+ DM_EINVAL("Invalid log size requested.");
+
+ l->params.dev.size = tmp;
+ wake_do_log(l); /* Let the worker do the resize. */
+ }
+ } else
+ DM_EINVAL("Invalid argument.");
+
+ return 0;
+}
+
+/* Support function for replicator log status requests. */
+static int
+ringbuffer_status(struct dm_repl_log *log, int dev_number,
+ status_type_t type, char *result, unsigned int maxlen)
+{
+ unsigned long slink_nr;
+ size_t sz = 0;
+ sector_t ios, sectors;
+ char buf[BDEVNAME_SIZE];
+ struct repl_log *l;
+ struct stats *s;
+ struct ringbuffer *ring;
+ struct repl_params *p;
+
+ _SET_AND_BUG_ON_L(l, log);
+ s = &l->stats;
+ ring = &l->ringbuffer;
+ p = &l->params;
+
+ switch (type) {
+ case STATUSTYPE_INFO:
+ ios = sectors = 0;
+
+ /* Output ios/sectors stats. */
+ spin_lock(&l->io.lock);
+ for_each_bit(slink_nr, LOG_SLINKS(l), l->slink.max) {
+ struct dm_repl_slink *slink = slink_find(l, slink_nr);
+ struct slink_state *ss;
+
+ _BUG_ON_PTR(slink);
+ ss = slink->caller;
+ _BUG_ON_PTR(ss);
+
+ DMEMIT(" %s,%llu,%llu",
+ SsSync(ss) ? "S" : "A",
+ (unsigned long long) ss->fb.outstanding.ios,
+ (unsigned long long) ss->fb.outstanding.sectors);
+ ios += ss->fb.outstanding.ios;
+ sectors += ss->fb.outstanding.sectors;
+ }
+
+ DMEMIT(" %llu/%llu/%llu",
+ (unsigned long long) ios,
+ (unsigned long long) sectors,
+ (unsigned long long) l->params.dev.size);
+
+ spin_unlock(&l->io.lock);
+
+ if (LogDevelStats(l))
+ DMEMIT(" ring->start=%llu "
+ "ring->head=%llu ring->tail=%llu "
+ "ring->next_avail=%llu ring->end=%llu "
+ "ring_free=%llu wrap=%d r=%d w=%d wp=%d he=%d "
+ "hash_insert=%u hash_insert_max=%u "
+ "single=%u multi=%u stall=%u",
+ (unsigned long long) ring->start,
+ (unsigned long long) ring->head,
+ (unsigned long long) ring->tail,
+ (unsigned long long) ring->next_avail,
+ (unsigned long long) ring->end,
+ (unsigned long long) ring_free(ring),
+ s->wrap,
+ atomic_read(s->io + 0), atomic_read(s->io + 1),
+ atomic_read(&s->writes_pending),
+ atomic_read(&s->hash_elem),
+ s->hash_insert, s->hash_insert_max,
+ s->copy[0], s->copy[1],
+ s->stall);
+
+ break;
+
+ case STATUSTYPE_TABLE:
+ DMEMIT("%s %d %s %llu", ringbuffer_type.type.name, p->count,
+ format_dev_t(buf, p->dev.dm_dev->bdev->bd_dev),
+ (unsigned long long) p->dev.start);
+
+ if (p->count > 2) {
+ DMEMIT(" %s", _open_str(p->open_type));
+
+ if (p->count > 3)
+ DMEMIT(" %llu",
+ (unsigned long long) p->dev.size);
+ }
+ }
+
+ return 0;
+}
+
+/*
+ * End methods of "ring-buffer" type
+ */
+
+/* "ring-buffer" replication log type. */
+static struct dm_repl_log_type ringbuffer_type = {
+ .type.name = "ringbuffer",
+ .type.module = THIS_MODULE,
+
+ .ctr = ringbuffer_ctr,
+ .dtr = ringbuffer_dtr,
+
+ .postsuspend = ringbuffer_postsuspend,
+ .resume = ringbuffer_resume,
+ .flush = ringbuffer_flush,
+ .io = ringbuffer_io,
+ .io_notify_fn_set = ringbuffer_io_notify_fn_set,
+
+ .slink_add = ringbuffer_slink_add,
+ .slink_del = ringbuffer_slink_del,
+ .slinks = ringbuffer_slinks,
+ .slink_max = ringbuffer_slink_max,
+
+ .message = ringbuffer_message,
+ .status = ringbuffer_status,
+};
+
+/* Destroy kmem caches on module unload. */
+static int
+replog_kmem_caches_exit(void)
+{
+ struct cache_defs *pd = ARRAY_END(cache_defs);
+
+ while (pd-- > cache_defs) {
+ if (unlikely(!pd->slab_pool))
+ continue;
+
+ DMDEBUG("Destroying kmem_cache %p", pd->slab_pool);
+ kmem_cache_destroy(pd->slab_pool);
+ pd->slab_pool = NULL;
+ }
+
+ return 0;
+}
+
+/* Create kmem caches on module load. */
+static int
+replog_kmem_caches_init(void)
+{
+ int r = 0;
+ struct cache_defs *pd = ARRAY_END(cache_defs);
+
+ while (pd-- > cache_defs) {
+ BUG_ON(pd->slab_pool);
+
+ /* No slab pool. */
+ if (!pd->size)
+ continue;
+
+ pd->slab_pool = kmem_cache_create(pd->slab_name, pd->size,
+ pd->align, 0, NULL);
+ if (likely(pd->slab_pool))
+ DMDEBUG("Created kmem_cache %p", pd->slab_pool);
+ else {
+ DMERR("failed to create slab %s for replication log "
+ " handler %s %s",
+ pd->slab_name, ringbuffer_type.type.name,
+ version);
+ replog_kmem_caches_exit();
+ r = -ENOMEM;
+ break;
+ }
+ }
+
+ return r;
+}
+
+int __init
+dm_repl_log_init(void)
+{
+ int r;
+
+ if (sizeof(struct data_header_disk) != DATA_HEADER_DISK_SIZE)
+ DM_EINVAL("invalid size of 'struct data_header_disk' for %s %s",
+ ringbuffer_type.type.name, version);
+
+ mutex_init(&list_mutex);
+
+ r = replog_kmem_caches_init();
+ if (r < 0) {
+ DMERR("failed to init %s kmem caches %s",
+ ringbuffer_type.type.name, version);
+ return r;
+ }
+
+ r = dm_register_type(&ringbuffer_type, DM_REPLOG);
+ if (r < 0) {
+ DMERR("failed to register replication log %s handler %s [%d]",
+ ringbuffer_type.type.name, version, r);
+ replog_kmem_caches_exit();
+ } else
+ DMINFO("registered replication log %s handler %s",
+ ringbuffer_type.type.name, version);
+
+ return r;
+}
+
+void __exit
+dm_repl_log_exit(void)
+{
+ int r = dm_unregister_type(&ringbuffer_type, DM_REPLOG);
+
+ replog_kmem_caches_exit();
+
+ if (r)
+ DMERR("failed to unregister replication log %s handler %s [%d]",
+ ringbuffer_type.type.name, version, r);
+ else
+ DMINFO("unregistered replication log %s handler %s",
+ ringbuffer_type.type.name, version);
+}
+
+/* Module hooks */
+module_init(dm_repl_log_init);
+module_exit(dm_repl_log_exit);
+
+MODULE_DESCRIPTION(DM_NAME " remote replication target \"ringbuffer\" "
+ "log handler");
+MODULE_AUTHOR("Jeff Moyer <jmoyer@redhat.com>, "
+ "Heinz Mauelshagen <heinzm@redhat.com");
+MODULE_LICENSE("GPL");
--
1.6.2.5
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCH v6 4/4] dm-replicator: blockdev site link handler
2009-12-18 15:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler heinzm
@ 2009-12-18 15:44 ` heinzm
2011-07-18 9:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler Busby
1 sibling, 0 replies; 9+ messages in thread
From: heinzm @ 2009-12-18 15:44 UTC (permalink / raw)
To: dm-devel; +Cc: Heinz Mauelshagen
From: Heinz Mauelshagen <heinzm@redhat.com>
This is the "blockdev" type site link handler module plugging
into the main replication module.
It abstracts the transport access logic used, allowing the
replication log to be agnostic about it (block device transport
in this handlers case).
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Reviewed-by: Jon Brassow <jbrassow@redhat.com>
Tested-by: Jon Brassow <jbrassow@redhat.com>
---
drivers/md/Makefile | 1 +
drivers/md/dm-repl-slink-blockdev.c | 3054 +++++++++++++++++++++++++++++++++++
2 files changed, 3055 insertions(+), 0 deletions(-)
create mode 100644 drivers/md/dm-repl-slink-blockdev.c
diff --git 2.6.33-rc1.orig/drivers/md/Makefile 2.6.33-rc1/drivers/md/Makefile
index dcb1f69..a5e38dd 100644
--- 2.6.33-rc1.orig/drivers/md/Makefile
+++ 2.6.33-rc1/drivers/md/Makefile
@@ -47,6 +47,7 @@ obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
obj-$(CONFIG_DM_ZERO) += dm-zero.o
obj-$(CONFIG_DM_REPLICATOR) += dm-replicator.o \
dm-repl-log-ringbuffer.o \
+ dm-repl-slink-blockdev.o \
dm-log.o dm-registry.o
quiet_cmd_unroll = UNROLL $@
diff --git 2.6.33-rc1.orig/drivers/md/dm-repl-slink-blockdev.c 2.6.33-rc1/drivers/md/dm-repl-slink-blockdev.c
new file mode 100644
index 0000000..951b260
--- /dev/null
+++ 2.6.33-rc1/drivers/md/dm-repl-slink-blockdev.c
@@ -0,0 +1,3054 @@
+/*
+ * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
+ *
+ * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
+ *
+ * This file is released under the GPL.
+ *
+ *
+ * "blockdev" site link handler for the replicator target supporting
+ * devices on block transports with device node access abstracting
+ * the nature of the access to the caller.
+ *
+ * It handles the fallbehind thresholds, temporary transport failures,
+ * their recovery and initial/partial device resynchronization.
+ *
+ * Locking Hierarchy:
+ * 1) repl_slinks->lock
+ * 2) sl->lock
+ *
+ */
+
+static const char version[] = "v0.022";
+
+#include "dm.h"
+#include "dm-repl.h"
+#include "dm-registry.h"
+#include "dm-repl-log.h"
+#include "dm-repl-slink.h"
+
+#include <linux/dm-dirty-log.h>
+#include <linux/dm-kcopyd.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+
+#define DM_MSG_PREFIX "dm-repl-slink-blockdev"
+#define DAEMON DM_MSG_PREFIX "d"
+#define COPY_PAGES BIO_MAX_PAGES
+#define RESYNC_PAGES (BIO_MAX_PAGES / 2)
+#define RESYNC_SIZE (to_sector(PAGE_SIZE) * RESYNC_PAGES)
+#define MIN_IOS 1
+
+/* Jiffies to wait before retrying a device. */
+#define SLINK_TEST_JIFFIES (15 * HZ)
+#define SLINK_TEST_SIZE 4096
+
+#define _SET_AND_BUG_ON_SL(sl, slink) \
+ do { \
+ _BUG_ON_PTR(slink); \
+ (sl) = slink_check(slink); \
+ _BUG_ON_PTR(sl); \
+ } while (0);
+
+/* An slink device. */
+enum sdev_list_type {
+ SDEV_SLINK, /* To put on slink's device list. */
+ SDEV_RESYNC, /* To put on slink's resynchronization list. */
+ SDEV_TEST, /* To put on so_slink_test()'s test list. */
+ NR_SDEV_LISTS,
+};
+struct test_buffer;
+struct sdev {
+ struct kref ref; /* Reference counter. */
+
+ /* Lists to hang off slink, resync and flush lists */
+ struct list_head lists[NR_SDEV_LISTS];
+ struct dm_target *ti;
+ struct slink *sl; /* Backpointer for callback. */
+
+ struct {
+ struct sdev_resync {
+ sector_t region; /* Region being resynchronized. */
+ sector_t writer_region; /* region being written to. */
+ sector_t start; /* Start of actual resync copy. */
+ sector_t end; /* End of actual resync copy. */
+ unsigned len; /* Length of actual copy. */
+ /* Source device pointer for resync callback. */
+ struct sdev *from;
+ } resync;
+
+ struct {
+ unsigned long time;
+ struct test_buffer *buffer;
+ } test;
+
+
+ /* kcopyd resynchronization client. */
+ struct dm_kcopyd_client *kcopyd_client;
+
+ /* Teardown synchronization. */
+ wait_queue_head_t waiters;
+
+ sector_t split_io;
+
+ unsigned long flags;
+ } io;
+
+ /* Device properties. */
+ struct sdev_dev {
+ struct {
+ unsigned count; /* Ctr parameters count. */
+ const char *path; /* Device path/major:minor. */
+ } params;
+
+ struct dm_dev *dm_dev;
+ struct dm_dirty_log *dl;
+ unsigned number;
+ } dev;
+};
+
+/* Macros to access sdev lists. */
+#define SDEV_SLINK_LIST(sl) (sl->lists + SDEV_SLINK)
+#define SDEV_RESYNC_LIST(sl) (sl->lists + SDEV_RESYNC)
+
+/* Status flags for device. */
+enum dev_flags {
+ DEV_ERROR_READ, /* Read error on device. */
+ DEV_ERROR_WRITE, /* Write error on device. */
+ DEV_IO_QUEUED, /* Request(s) to device queued. */
+ DEV_IO_UNPLUG, /* Unplug device queue. */
+ DEV_OPEN, /* Device got opend during ctr. */
+ DEV_RESYNC, /* Device may resync. */
+ DEV_RESYNC_END, /* Flag device resynchronization end. */
+ DEV_RESYNCING, /* Device has active resync. */
+ DEV_SUSPENDED, /* Device suspended. */
+ DEV_TEARDOWN, /* Device is being deleted. */
+};
+
+/* Create slink bitfield (io.flags) access inline definitions. */
+DM_BITOPS(DevErrorRead, sdev, DEV_ERROR_READ)
+DM_BITOPS(DevErrorWrite, sdev, DEV_ERROR_WRITE)
+DM_BITOPS(DevIOQueued, sdev, DEV_IO_QUEUED)
+DM_BITOPS(DevIOUnplug, sdev, DEV_IO_UNPLUG)
+DM_BITOPS(DevOpen, sdev, DEV_OPEN)
+DM_BITOPS(DevResync, sdev, DEV_RESYNC)
+DM_BITOPS(DevResyncEnd, sdev, DEV_RESYNC_END)
+DM_BITOPS(DevResyncing, sdev, DEV_RESYNCING)
+DM_BITOPS(DevSuspended, sdev, DEV_SUSPENDED)
+DM_BITOPS(DevTeardown, sdev, DEV_TEARDOWN)
+
+/* Internal site link representation. */
+enum slink_list_type { SLINK_DEVS, SLINK_REPLOG, SLINK_RESYNC, NR_SLINK_LISTS };
+enum cache_type { COPY_CACHE, TEST_CACHE, NR_CACHES };
+struct slink {
+ struct kref ref; /* Reference count. */
+ /*
+ * Protect slink lists.
+ *
+ * Has to be spinlock, because the global replog lock
+ * needs to be one to be used from interrupt context
+ * and they are both taken in some places.
+ */
+ rwlock_t lock; /* Protect slink lists. */
+
+ /* Devices on this slink, on replog list and on resync list. */
+ struct list_head lists[NR_SLINK_LISTS];
+
+ /* List of all slinks for a replog. */
+ struct dm_repl_log_slink_list *repl_slinks;
+
+ unsigned number; /* slink number. */
+
+ struct slink_params {
+ unsigned count;
+ struct dm_repl_slink_fallbehind fallbehind;
+ enum dm_repl_slink_policy_type policy;
+ } params;
+
+ struct dm_repl_slink *slink;
+
+ struct slink_io {
+ unsigned long flags;
+ struct dm_kcopyd_client *kcopyd_client;
+ struct dm_io_client *dm_io_client;
+
+ /* Copy context and test buffer mempools. */
+ mempool_t *pool[NR_CACHES];
+
+ /* io work. */
+ struct workqueue_struct *wq;
+ struct delayed_work dws;
+
+ struct sdev *dev_test;
+ } io;
+
+ /* Callback for slink recovered. */
+ struct dm_repl_slink_notify_ctx recover;
+};
+
+/* Macros to access slink lists. */
+#define SLINK_DEVS_LIST(sl) (sl->lists + SLINK_DEVS)
+#define SLINK_REPLOG_LIST(sl) (sl->lists + SLINK_REPLOG)
+#define SLINK_RESYNC_LIST(sl) (sl->lists + SLINK_RESYNC)
+
+/* Status flags for slink. */
+enum slink_flags {
+ SLINK_ERROR_READ, /* Read error on site link. */
+ SLINK_ERROR_WRITE, /* Write error on site link. */
+ SLINK_IMMEDIATE_WORK, /* Flag immediate worker run. */
+ SLINK_RESYNC_PROCESSING,/* Resync is being processed on slink. */
+ SLINK_TEST_ACTIVE, /* Device test active on slink. */
+ SLINK_WRITER, /* Slink is being written to. */
+};
+
+/* Create slink bitfield (io.flags) access inline definitions. */
+DM_BITOPS(SlinkErrorRead, slink, SLINK_ERROR_READ)
+DM_BITOPS(SlinkErrorWrite, slink, SLINK_ERROR_WRITE)
+DM_BITOPS(SlinkImmediateWork, slink, SLINK_IMMEDIATE_WORK)
+DM_BITOPS(SlinkResyncProcessing, slink, SLINK_RESYNC_PROCESSING)
+DM_BITOPS(SlinkTestActive, slink, SLINK_TEST_ACTIVE)
+DM_BITOPS(SlinkWriter, slink, SLINK_WRITER)
+
+/* Copy context to carry from blockdev_copy() to copy_endio(). */
+struct copy_context {
+ struct sdev *dev_to; /* Device to copy to. */
+
+ /* Callback for data in RAM (noop for 'blockdev' type). */
+ struct dm_repl_slink_notify_ctx ram;
+
+ /* Callback for data on disk. */
+ struct dm_repl_slink_notify_ctx disk;
+};
+
+/* Allocate/free blockdev copy context. */
+static inline struct copy_context *alloc_copy_context(struct slink *sl)
+{
+ return mempool_alloc(sl->io.pool[COPY_CACHE], GFP_KERNEL);
+}
+
+static inline void free_copy_context(struct copy_context *cc, struct slink *sl)
+{
+ mempool_free(cc, sl->io.pool[COPY_CACHE]);
+}
+
+/* Allocate/free blockdev test io buffer. */
+static inline struct test_buffer *alloc_test_buffer(struct slink *sl)
+{
+ return mempool_alloc(sl->io.pool[TEST_CACHE], GFP_KERNEL);
+}
+
+static inline void free_test_buffer(struct test_buffer *tb, struct slink *sl)
+{
+ mempool_free(tb, sl->io.pool[TEST_CACHE]);
+}
+
+/* Destcriptor type <-> name mapping. */
+static const struct dm_str_descr policies[] = {
+ { DM_REPL_SLINK_ASYNC, "asynchronous" },
+ { DM_REPL_SLINK_SYNC, "synchronous" },
+ { DM_REPL_SLINK_STALL, "stall" },
+};
+
+/* Get slink policy flags. */
+static int _slink_policy_type(char *name)
+{
+ int r = dm_descr_type(policies, ARRAY_SIZE(policies), name);
+
+ if (r < 0)
+ DMERR("Invalid site link policy %s", name);
+
+ return r;
+}
+
+/* Get slink policy name. */
+static const char *
+_slink_policy_name(const int type)
+{
+ return dm_descr_name(policies, ARRAY_SIZE(policies), type);
+}
+
+#define SEPARATOR '+'
+static int
+get_slink_policy(char *arg)
+{
+ int policy = 0, r;
+ char *sep;
+
+ DMDEBUG_LIMIT("%s arg=%s", __func__, arg);
+
+ /*
+ * Check substrings of the compound policy
+ * string separated by SEPARATOR.
+ */
+ do {
+ sep = strchr(arg, SEPARATOR);
+ if (sep)
+ *sep = 0;
+ else
+ sep = arg;
+
+ r = _slink_policy_type(arg);
+ if (sep != arg) {
+ arg = sep + 1;
+ *sep = SEPARATOR;
+ }
+
+ if (r < 0)
+ return r;
+ else
+ set_bit(r, (unsigned long *) &policy);
+ } while (sep != arg);
+
+ smp_mb();
+ return policy;
+}
+
+/* String print policies. */
+static char *
+snprint_policies(enum dm_repl_slink_policy_type policies,
+ char *result, size_t maxlen)
+{
+ int bits = sizeof(policies) * 8, i;
+ size_t sz = 0;
+
+ *result = 0;
+ for (i = 0; i < bits; i++) {
+ if (test_bit(i, (unsigned long *) &policies)) {
+ const char *name = _slink_policy_name(i);
+
+ if (name) {
+ if (*result)
+ DMEMIT("%c", SEPARATOR);
+
+ DMEMIT("%s", name);
+ }
+ }
+ }
+
+ return result;
+}
+
+/* Fallbehind type <-> name mappings. */
+static const struct dm_str_descr fb_types[] = {
+ { DM_REPL_SLINK_FB_IOS, "ios" },
+ { DM_REPL_SLINK_FB_SIZE, "size" },
+ { DM_REPL_SLINK_FB_TIMEOUT, "timeout" },
+};
+
+/* Return name of fallbehind parameter by type. */
+static const char *
+fb_name(enum dm_repl_slink_fallbehind_type type)
+{
+ return dm_descr_name(fb_types, ARRAY_SIZE(fb_types), type);
+}
+
+/* String print fallbehind. */
+static char *
+snprint_fallbehind(struct dm_repl_slink_fallbehind *fallbehind,
+ char *result, size_t maxlen)
+{
+ size_t sz = 0;
+ sector_t value = fallbehind->value;
+
+ sector_div(value, fallbehind->multiplier);
+ DMEMIT("%s %llu%c", fb_name(fallbehind->type),
+ (unsigned long long) value, fallbehind->unit);
+ return result;
+}
+
+/*
+ * Check and get fallbehind value and type.
+ * Pay attention to unit qualifiers.
+ */
+static int
+_get_slink_fallbehind(int argc, char **argv,
+ enum dm_repl_slink_fallbehind_type fb_type,
+ struct dm_repl_slink_fallbehind *fb)
+{
+ int arg = 0, r;
+ unsigned multi = 1;
+ long long tmp;
+ char *unit;
+ const char *name = fb_name(fb_type);
+
+ fb->unit = 0;
+
+ /* Old syntax e.g. "ios=1000". */
+ r = sscanf(argv[arg] + strlen(name), "=%lld", &tmp) != 1 || tmp < 0;
+ if (r) {
+ if (argc < 2)
+ goto bad_value;
+
+ /* New syntax e.g. "ios 1000". */
+ r = sscanf(argv[++arg], "%lld", &tmp) != 1 || tmp < 0;
+ }
+
+ unit = argv[arg] + strlen(argv[arg]) - 1;
+ unit = (*unit < '0' || *unit > '9') ? unit : NULL;
+
+ if (r)
+ goto bad_value;
+
+ if (unit) {
+ const struct units {
+ const char *chars;
+ const sector_t multiplier[];
+ } *u = NULL;
+ static const struct units size = {
+ "sSkKmMgGtTpPeE",
+ #define TWO (sector_t) 2
+ /* sectors, KB,MB, GB, TB, PB, EB */
+ { 1, 2, TWO<<10, TWO<<20, TWO<<30, TWO<<40, TWO<<50 },
+ #undef TWO
+ }, timeout = {
+ "tTsSmMhHdD",
+ /*ms, sec, minute, hour, day */
+ { 1, 1000, 60*1000, 60*60*1000, 24*60*60*1000 },
+ };
+ const char *c;
+
+ switch (fb_type) {
+ case DM_REPL_SLINK_FB_SIZE:
+ u = &size;
+ goto retrieve;
+ case DM_REPL_SLINK_FB_TIMEOUT:
+ u = &timeout;
+retrieve:
+ /* Skip to unit identifier character. */
+ for (c = u->chars, multi = 0;
+ *c && *c != *unit;
+ c++, multi++)
+ ;
+
+ if (*c) {
+ fb->unit = *c;
+ multi = u->multiplier[(multi + 2) / 2];
+ } else
+ goto bad_unit;
+ case DM_REPL_SLINK_FB_IOS:
+ break;
+
+ default:
+ BUG();
+ }
+ }
+
+ fb->type = fb_type;
+ fb->multiplier = multi;
+ fb->value = tmp * multi;
+ return 0;
+
+bad_value:
+ DMERR("invalid fallbehind %s value", argv[0]);
+ return -EINVAL;
+
+bad_unit:
+ DMERR("invalid slink fallbehind unit");
+ return -EINVAL;
+}
+
+static int
+get_slink_fallbehind(int argc, char **argv, struct dm_repl_slink_fallbehind *fb)
+{
+ const struct dm_str_descr *f = ARRAY_END(fb_types);
+
+ while (f-- > fb_types) {
+ /* Check for fallbehind argument. */
+ if (!strnicmp(STR_LEN(fb_name(f->type), argv[0])))
+ return _get_slink_fallbehind(argc, argv, f->type, fb);
+ }
+
+ DMERR("invalid fallbehind type %s", argv[0]);
+ return -EINVAL;
+}
+
+/* Return region on device fro given sector. */
+static sector_t
+sector_to_region(struct sdev *dev, sector_t sector)
+{
+ sector_div(sector, dev->ti->split_io);
+ return sector;
+}
+
+/* Check dm_repl_slink and slink ok. */
+static struct slink *
+slink_check(struct dm_repl_slink *slink)
+{
+ struct slink *sl;
+
+ if (unlikely(!slink))
+ return ERR_PTR(-EINVAL);
+
+ if (unlikely(IS_ERR(slink)))
+ return (struct slink *) slink;
+
+ sl = slink->context;
+ return sl ? sl : ERR_PTR(-EINVAL);
+}
+
+struct cache_defs {
+ enum cache_type type;
+ const int min;
+ struct kmem_cache *cache;
+ const char *name;
+ const size_t size;
+};
+
+/* Slabs for the copy context structures and for device test I/O buffers. */
+static struct cache_defs cache_defs[] = {
+ { COPY_CACHE, MIN_IOS, NULL,
+ "dm_repl_slink_copy", sizeof(struct copy_context) },
+ { TEST_CACHE, MIN_IOS, NULL,
+ "dm_repl_slink_test", SLINK_TEST_SIZE },
+};
+
+/*
+ * Release resources when last reference dropped.
+ *
+ * Gets called with lock hold to atomically delete slink from list.
+ */
+static void
+slink_release(struct kref *ref)
+{
+ struct slink *sl = container_of(ref, struct slink, ref);
+
+ DMDEBUG("%s slink=%d released", __func__, sl->number);
+ kfree(sl);
+}
+
+/* Take out reference on slink. */
+static struct slink *
+slink_get(struct slink *sl)
+{
+ kref_get(&sl->ref);
+ return sl;
+}
+
+/* Drop reference on slink and destroy it on last release. */
+static int
+slink_put(struct slink *sl)
+{
+ return kref_put(&sl->ref, slink_release);
+}
+
+/* Find slink on global slink list by number. */
+static struct slink *
+slink_get_by_number(struct dm_repl_log_slink_list *repl_slinks,
+ unsigned slink_number)
+{
+ struct slink *sl;
+
+ BUG_ON(!repl_slinks);
+
+ list_for_each_entry(sl, &repl_slinks->list, lists[SLINK_REPLOG]) {
+ if (slink_number == sl->number)
+ return slink_get(sl);
+ }
+
+ return ERR_PTR(-ENOENT);
+}
+
+/* Destroy slink object. */
+static void
+slink_destroy(struct slink *sl)
+{
+ struct slink_io *io;
+ struct cache_defs *cd;
+
+ _BUG_ON_PTR(sl);
+ _BUG_ON_PTR(sl->repl_slinks);
+ io = &sl->io;
+
+ write_lock(&sl->repl_slinks->lock);
+ if (!list_empty(SLINK_REPLOG_LIST(sl)))
+ list_del(SLINK_REPLOG_LIST(sl));
+ write_unlock(&sl->repl_slinks->lock);
+
+ /* Destroy workqueue before freeing resources. */
+ if (io->wq)
+ destroy_workqueue(io->wq);
+
+ if (io->kcopyd_client)
+ dm_kcopyd_client_destroy(io->kcopyd_client);
+
+ if (io->dm_io_client)
+ dm_io_client_destroy(io->dm_io_client);
+
+ cd = ARRAY_END(cache_defs);
+ while (cd-- > cache_defs) {
+ if (io->pool[cd->type]) {
+ mempool_destroy(io->pool[cd->type]);
+ io->pool[cd->type] = NULL;
+ }
+ }
+}
+
+/*
+ * Get slink from global slink list by number or create
+ * new one and put it on list; take out reference.
+ */
+static void do_slink(struct work_struct *ws);
+static struct slink *
+slink_create(struct dm_repl_slink *slink,
+ struct dm_repl_log_slink_list *repl_slinks,
+ struct slink_params *params, unsigned slink_number)
+{
+ int i, r;
+ struct slink *sl;
+
+ DMDEBUG_LIMIT("%s %u", __func__, slink_number);
+
+ /* Make sure, slink0 exists when creating slink > 0. */
+ if (slink_number) {
+ struct slink *sl0;
+
+ read_lock(&repl_slinks->lock);
+ sl0 = slink_get_by_number(repl_slinks, 0);
+ read_unlock(&repl_slinks->lock);
+
+ if (IS_ERR(sl0)) {
+ DMERR("Can't create slink=%u w/o slink0.",
+ slink_number);
+ return ERR_PTR(-EPERM);
+ }
+
+ BUG_ON(slink_put(sl0));
+ }
+
+ read_lock(&repl_slinks->lock);
+ sl = slink_get_by_number(repl_slinks, slink_number);
+ read_unlock(&repl_slinks->lock);
+
+ if (IS_ERR(sl)) {
+ struct slink *sl_tmp;
+ struct slink_io *io;
+ struct cache_defs *cd;
+
+ if (!params)
+ return sl;
+
+ /* Preallocate internal slink struct. */
+ sl = kzalloc(sizeof(*sl), GFP_KERNEL);
+ if (unlikely(!sl))
+ return ERR_PTR(-ENOMEM);
+
+ rwlock_init(&sl->lock);
+ kref_init(&sl->ref);
+
+#ifdef CONFIG_LOCKDEP
+ {
+ static struct lock_class_key slink_number_lock;
+
+ lockdep_set_class_and_subclass(&sl->lock,
+ &slink_number_lock,
+ slink_number);
+ }
+#endif
+
+ i = ARRAY_SIZE(sl->lists);
+ while (i--)
+ INIT_LIST_HEAD(sl->lists + i);
+
+ /* Copy (parsed) fallbehind arguments accross. */
+ io = &sl->io;
+ sl->params = *params;
+ sl->number = slink_number;
+ sl->repl_slinks = repl_slinks;
+ sl->slink = slink;
+ slink->context = sl;
+
+ /* Create kcopyd client for data copies to slinks. */
+ r = dm_kcopyd_client_create(COPY_PAGES, &io->kcopyd_client);
+ if (unlikely(r < 0)) {
+ io->kcopyd_client = NULL;
+ goto bad;
+ }
+
+ /* Create dm-io client context for test I/Os on slinks. */
+ io->dm_io_client = dm_io_client_create(1);
+ if (unlikely(IS_ERR(io->dm_io_client))) {
+ r = PTR_ERR(io->dm_io_client);
+ io->dm_io_client = NULL;
+ goto bad;
+ }
+
+ r = -ENOMEM;
+
+ /* Create slab mempools for copy contexts and test buffers. */
+ cd = ARRAY_END(cache_defs);
+ while (cd-- > cache_defs) {
+ io->pool[cd->type] =
+ mempool_create_slab_pool(cd->min, cd->cache);
+ if (unlikely(!io->pool[cd->type])) {
+ DMERR("Failed to create mempool %p",
+ io->pool[cd->type]);
+ goto bad;
+ }
+ }
+
+ io->wq = create_singlethread_workqueue(DAEMON);
+ if (likely(io->wq))
+ INIT_DELAYED_WORK(&sl->io.dws, do_slink);
+ else
+ goto bad;
+
+ /* Add to replog list. */
+ write_lock(&repl_slinks->lock);
+ sl_tmp = slink_get_by_number(repl_slinks, slink_number);
+ if (likely(IS_ERR(sl_tmp))) {
+ /* We won the race -> add to list. */
+ list_add_tail(SLINK_REPLOG_LIST(sl),
+ &repl_slinks->list);
+ write_unlock(&repl_slinks->lock);
+ } else {
+ /* We lost the race, take the winner. */
+ write_unlock(&repl_slinks->lock);
+ /* Will release sl. */
+ slink_destroy(sl);
+ sl = sl_tmp;
+ }
+
+ return sl;
+ }
+
+ slink_put(sl);
+ return ERR_PTR(-EEXIST);
+
+bad:
+ slink_destroy(sl);
+ return ERR_PTR(r);
+}
+
+/* Return slink count. */
+static unsigned
+slink_count(struct slink *sl)
+{
+ unsigned count = 0;
+ struct slink *sl_cur;
+
+ _BUG_ON_PTR(sl);
+
+ list_for_each_entry(sl_cur, &sl->repl_slinks->list, lists[SLINK_REPLOG])
+ count++;
+
+ return count;
+}
+
+/* Return number of regions for device. */
+static inline sector_t
+region_count(struct sdev *dev)
+{
+ return dm_sector_div_up(dev->ti->len, dev->ti->split_io);
+}
+
+
+/*
+ * Site link worker.
+ */
+/* Queue (optionally delayed) io work. */
+static void
+wake_do_slink_delayed(struct slink *sl, unsigned long delay)
+{
+ struct delayed_work *dws = &sl->io.dws;
+
+ if (delay) {
+ /* Avoid delaying if immediate worker run already requested. */
+ if (SlinkImmediateWork(sl))
+ return;
+ } else
+ SetSlinkImmediateWork(sl);
+
+ if (delayed_work_pending(dws))
+ cancel_delayed_work(dws);
+
+ queue_delayed_work(sl->io.wq, dws, delay);
+}
+
+/* Queue io work immediately. */
+static void
+wake_do_slink(void *context)
+{
+ wake_do_slink_delayed(context, 0);
+}
+
+/* Set/get device test timeouts. */
+/* FIXME: algorithm to have flexible test timing? */
+static inline void
+set_dev_test_time(struct sdev *dev)
+{
+ unsigned long time = jiffies + SLINK_TEST_JIFFIES;
+
+ /* Check jiffies wrap. */
+ if (unlikely(time < jiffies))
+ time = SLINK_TEST_JIFFIES;
+
+ dev->io.test.time = time;
+}
+
+static inline unsigned long
+get_dev_test_time(struct sdev *dev)
+{
+ return dev->io.test.time;
+}
+
+/*
+ * Get device object reference count.
+ *
+ * A reference count > 1 indicates IO in flight on the device.
+ *
+ */
+static int dev_io(struct sdev *dev)
+{
+ return atomic_read(&dev->ref.refcount) > 1;
+}
+
+/* Take device object reference out. */
+static struct sdev *dev_get(struct sdev *dev)
+{
+ kref_get(&dev->ref);
+ return dev;
+}
+
+/* Release sdev object. */
+static void
+dev_release(struct kref *ref)
+{
+ struct sdev *dev = container_of(ref, struct sdev, ref);
+ struct slink *sl = dev->sl;
+
+ _BUG_ON_PTR(sl);
+
+ kfree(dev->dev.params.path);
+ DMDEBUG("%s dev=%d slink=%d released", __func__,
+ dev->dev.number, sl->number);
+ kfree(dev);
+}
+
+/* Drop device object reference. */
+static int dev_put(struct sdev *dev)
+{
+ int r = kref_put(&dev->ref, dev_release);
+
+ if (!r) {
+ if (!dev_io(dev))
+ wake_up(&dev->io.waiters);
+ }
+
+ return r;
+}
+
+/* Find device by device number. */
+static struct sdev *dev_get_by_number(struct slink *sl, int dev_number)
+{
+ struct sdev *dev;
+
+ _BUG_ON_PTR(sl);
+
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ if (dev_number == dev->dev.number)
+ return dev_get(dev);
+ }
+
+ return ERR_PTR(-ENODEV);
+}
+
+/* Find device by bdev. */
+static struct sdev *dev_get_by_bdev(struct slink *sl,
+ struct block_device *bdev)
+{
+ struct sdev *dev;
+
+ _BUG_ON_PTR(sl);
+
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ struct mapped_device *md = dm_table_get_md(dev->ti->table);
+ struct gendisk *gd = dm_disk(md);
+
+ dm_put(md);
+
+ if (bdev->bd_disk == gd)
+ return dev_get(dev);
+ }
+
+ return ERR_PTR(-ENODEV);
+}
+
+/* Find device by path. */
+static struct sdev *dev_get_by_path(struct slink *sl,
+ const char *path)
+{
+ struct sdev *dev;
+
+ _BUG_ON_PTR(sl);
+
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ if (!strcmp(dev->dev.params.path, path))
+ return dev_get(dev);
+ }
+
+ return ERR_PTR(-ENODEV);
+}
+
+static struct sdev *
+dev_get_on_any_slink(struct slink *sl, struct sdev *dev)
+{
+ struct slink *sl_cur = NULL;
+ struct sdev *dev_r;
+
+ list_for_each_entry(sl_cur, &sl->repl_slinks->list,
+ lists[SLINK_REPLOG]) {
+ /* Check by path if device already present. */
+ if (sl_cur != sl)
+ read_lock(&sl_cur->lock);
+
+ /* Check by bdev/number depending on device open or not. */
+ dev_r = DevOpen(dev) ?
+ dev_get_by_bdev(sl_cur, dev->dev.dm_dev->bdev) :
+ dev_get_by_path(sl_cur, dev->dev.params.path);
+
+ if (sl_cur != sl)
+ read_unlock(&sl_cur->lock);
+
+ if (unlikely(!IS_ERR(dev_r)))
+ return dev_r;
+ }
+
+ return ERR_PTR(-ENOENT);
+}
+
+/* Callback for site link accessibility tests. */
+static void
+dev_test_endio(unsigned long error, void *context)
+{
+ struct sdev *dev = context;
+ struct slink *sl;
+
+ _BUG_ON_PTR(dev);
+ sl = dev->sl;
+ _BUG_ON_PTR(sl);
+
+ if (error)
+ set_dev_test_time(dev);
+ else {
+ ClearDevErrorRead(dev);
+ ClearDevErrorWrite(dev);
+ }
+
+ /* Release test io buffer. */
+ free_test_buffer(dev->io.test.buffer, sl);
+
+ ClearSlinkTestActive(sl);
+ BUG_ON(dev_put(dev)); /* Release reference. */
+ wake_do_slink(sl);
+}
+
+/* Submit a read to sector 0 of a remote device to test access to it. */
+static void
+dev_test(struct slink *sl, struct sdev *dev)
+{
+ struct dm_io_region region = {
+ .bdev = dev->dev.dm_dev->bdev,
+ .sector = 0,
+ .count = 1,
+ };
+ struct dm_io_request req = {
+ .bi_rw = READ,
+ .mem.type = DM_IO_KMEM,
+ .mem.ptr.addr = alloc_test_buffer(sl),
+ .notify.fn = dev_test_endio,
+ .notify.context = dev,
+ .client = sl->io.dm_io_client,
+ };
+
+ /* FIXME: flush_workqueue should care for race. */
+ sl->io.dev_test = dev;
+ dev->io.test.buffer = req.mem.ptr.addr;
+ set_dev_test_time(dev);
+ BUG_ON(dm_io(&req, 1, ®ion, NULL));
+}
+
+
+/*
+ * Callback replog handler in case of a region
+ * resynchronized or a device recovered.
+ */
+static inline void
+recover_callback(struct slink *sl, int read_err, int write_err)
+{
+ struct dm_repl_slink_notify_ctx recover;
+
+ _BUG_ON_PTR(sl);
+ read_lock(&sl->lock);
+ recover = sl->recover;
+ read_unlock(&sl->lock);
+
+ /* Optionally call back site link recovery. */
+ if (likely(recover.fn))
+ recover.fn(read_err, write_err, recover.context);
+}
+
+/* Try to open device. */
+static int
+try_dev_open(struct sdev *dev)
+{
+ int r = 0;
+
+ if (!DevOpen(dev)) {
+ /* Try getting device with limit checks. */
+ r = dm_get_device(dev->ti, dev->dev.params.path,
+ 0, dev->ti->len,
+ dm_table_get_mode(dev->ti->table),
+ &dev->dev.dm_dev);
+ if (r) {
+ set_dev_test_time(dev);
+ SetDevErrorRead(dev);
+ } else {
+ SetDevOpen(dev);
+ ClearDevErrorRead(dev);
+ }
+ }
+
+ return r;
+}
+
+/* Check devices for error condition and initiate test io on those. */
+static void
+do_slink_test(struct slink *sl)
+{
+ int r;
+ unsigned error_count = 0;
+ /* FIXME: jiffies may differ on MP ? */
+ unsigned long delay = ~0, j = jiffies;
+ struct sdev *dev, *dev_t = NULL;
+
+ read_lock(&sl->lock);
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ if ((DevErrorRead(dev) || DevErrorWrite(dev))) {
+ error_count++;
+
+ if (!DevTeardown(dev) && !DevSuspended(dev)) {
+ unsigned long t = get_dev_test_time(dev);
+
+ /* Check we didn't reach the test time jet. */
+ if (time_before(j, t)) {
+ unsigned long d = t - j;
+
+ if (d < delay)
+ delay = d;
+ } else {
+ dev_t = dev;
+ slink_get(sl);
+ break;
+ }
+ }
+ }
+ }
+
+ read_unlock(&sl->lock);
+
+ if (!error_count) {
+ /*
+ * If all are ok -> reset site link error state.
+ *
+ * We can't allow submission of writes
+ * before all devices are accessible.
+ */
+ /*
+ * FIXME: I actually test the remote device only so
+ * I shouldn't update state on the local reader side!
+ *
+ * Question is, where to update this or leave it
+ * to the caller to fail fataly when it can't read data
+ * of off the log or the replicated device any more.
+ */
+ if (TestClearSlinkErrorRead(sl))
+ error_count++;
+
+ if (TestClearSlinkErrorWrite(sl))
+ error_count++;
+
+ if (error_count)
+ recover_callback(sl, 0, 0);
+
+ return;
+ }
+
+ j = jiffies;
+
+ /* Check for jiffies overrun. */
+ if (unlikely(j < SLINK_TEST_JIFFIES))
+ set_dev_test_time(dev);
+
+ if (!TestSetSlinkTestActive(sl)) {
+ dev_get(dev); /* Take out device reference. */
+
+ /* Check device is open or open it. */
+ r = try_dev_open(dev_t);
+ if (r) {
+ ClearSlinkTestActive(sl);
+ BUG_ON(dev_put(dev_t));
+ } else
+ /* If open, do test. */
+ dev_test(sl, dev_t);
+ }
+
+ if (!SlinkTestActive(sl) && delay < ~0)
+ wake_do_slink_delayed(sl, delay);
+
+ slink_put(sl);
+}
+
+/* Set device error state and throw a dm event. */
+static void
+resync_error_dev(struct sdev *dev)
+{
+ SetDevErrorRead(dev);
+ dm_table_event(dev->ti->table);
+}
+
+/* Resync copy callback. */
+static void
+resync_endio(int read_err, unsigned long write_err, void *context)
+{
+ struct sdev *dev_from, *dev_to = context;
+ struct sdev_resync *resync;
+
+ _BUG_ON_PTR(dev_to);
+ resync = &dev_to->io.resync;
+ dev_from = resync->from;
+ _BUG_ON_PTR(dev_from);
+
+ if (unlikely(read_err))
+ resync_error_dev(dev_from);
+
+ if (unlikely(write_err))
+ resync_error_dev(dev_to);
+
+ resync->start += resync->len;
+ ClearDevResyncing(dev_to);
+ dev_put(dev_from);
+ dev_put(dev_to);
+ wake_do_slink(dev_to->sl);
+}
+
+/* Unplug a block devices queue. */
+static inline void
+dev_unplug(struct block_device *bdev)
+{
+ blk_unplug(bdev_get_queue(bdev));
+}
+
+/* Resync copy function. */
+static void
+resync_copy(struct sdev *dev_to, struct sdev *dev_from,
+ struct sdev_resync *resync)
+{
+ sector_t max_len = dev_from->ti->len;
+ struct dm_io_region src = {
+ .bdev = dev_from->dev.dm_dev->bdev,
+ .sector = resync->start,
+ }, dst = {
+ .bdev = dev_to->dev.dm_dev->bdev,
+ .sector = resync->start,
+ };
+
+ src.count = dst.count = unlikely(src.sector + resync->len > max_len) ?
+ max_len - src.sector : resync->len;
+ BUG_ON(!src.count);
+ BUG_ON(src.sector + src.count > max_len);
+ dev_to->io.resync.from = dev_from;
+ SetDevResyncing(dev_to);
+ BUG_ON(dm_kcopyd_copy(dev_to->io.kcopyd_client, &src, 1, &dst, 0,
+ resync_endio, dev_to));
+ dev_unplug(src.bdev);
+ dev_unplug(dst.bdev);
+}
+
+/* Return length of segment to resynchronoze. */
+static inline sector_t
+resync_len(struct sdev *dev)
+{
+ sector_t r = RESYNC_SIZE, region_size = dev->ti->split_io;
+ struct sdev_resync *resync = &dev->io.resync;
+
+ if (unlikely(r > region_size))
+ r = region_size;
+
+ if (unlikely(resync->start + r > resync->end))
+ r = resync->end - resync->start;
+
+ return r;
+}
+
+/* Initiate recovery on all site links registered for. */
+static void
+do_slink_resync(struct slink *sl)
+{
+ struct slink *sl0;
+ struct sdev *dev_from, *dev_n, *dev_to;
+ struct list_head resync_list;
+
+ _BUG_ON_PTR(sl);
+
+ if (!sl->number)
+ return;
+
+ /*
+ * Protect the global site link list from
+ * changes while getting slink 0.
+ */
+ read_lock(&sl->repl_slinks->lock);
+ sl0 = slink_get_by_number(sl->repl_slinks, 0);
+ read_unlock(&sl->repl_slinks->lock);
+
+ _BUG_ON_PTR(sl0);
+
+ /*
+ * Quickly take out resync list for local unlocked processing
+ * and take references per device for suspend/delete race prevention
+ */
+ INIT_LIST_HEAD(&resync_list);
+ read_lock(&sl0->lock);
+ write_lock(&sl->lock);
+ SetSlinkResyncProcessing(sl);
+
+ list_splice(SLINK_RESYNC_LIST(sl), &resync_list);
+ INIT_LIST_HEAD(SLINK_RESYNC_LIST(sl));
+
+ list_for_each_entry(dev_to, &resync_list, lists[SDEV_RESYNC]) {
+ dev_from = dev_get_by_number(sl0, dev_to->dev.number);
+ _BUG_ON_PTR(dev_from);
+ dev_get(dev_to);
+ /* Memorize device to copy from. */
+ dev_to->io.resync.from = dev_from;
+ }
+
+ write_unlock(&sl->lock);
+ read_unlock(&sl0->lock);
+
+ /*
+ * Process all devvices needing
+ * resynchronization on the private list.
+ *
+ * "dev_to" is device to copy to.
+ */
+ list_for_each_entry(dev_to, &resync_list, lists[SDEV_RESYNC]) {
+ unsigned region_size;
+ struct sdev_resync *resync;
+ struct dm_dirty_log *dl = dev_to->dev.dl;
+
+ /* Can't resync w/o dirty log. */
+ _BUG_ON_PTR(dl);
+
+ /* Device closed/copy active/suspended/being torn down/error. */
+ if (!DevOpen(dev_to) ||
+ DevResyncing(dev_to) ||
+ DevSuspended(dev_to) ||
+ DevTeardown(dev_to) ||
+ DevErrorWrite(dev_to))
+ continue;
+
+ /* Device to copy from. */
+ resync = &dev_to->io.resync;
+ dev_from = resync->from;
+
+ /* slink0 device suspended/being torn down or I/O error. */
+ if (DevSuspended(dev_from) ||
+ DevTeardown(dev_from) ||
+ DevErrorRead(dev_from))
+ continue;
+
+ /* No copy active if resync->end == 0. */
+ if (!resync->end) {
+ int r;
+ sector_t region;
+
+ /* Ask dirty region log for another region to sync. */
+ r = dl->type->get_resync_work(dl, ®ion);
+ if (r) {
+ write_lock(&sl0->lock);
+
+ /* Region is being written to -> postpone. */
+ if (unlikely(SlinkWriter(sl0) &&
+ resync->writer_region == region)) {
+ write_unlock(&sl0->lock);
+ continue;
+ }
+
+ region_size = dev_to->ti->split_io;
+ resync->region = region;
+ resync->start = region * region_size;
+ resync->end = resync->start + region_size;
+ if (unlikely(resync->end > dev_to->ti->len))
+ resync->end = dev_to->ti->len;
+
+ write_unlock(&sl0->lock);
+ } else {
+ /* No more regions to recover. */
+ SetDevResyncEnd(dev_to);
+ continue;
+ }
+ }
+
+ /* More to copy for this region. */
+ if (resync->start < resync->end) {
+ resync->len = resync_len(dev_to);
+ BUG_ON(!resync->len);
+
+ /*
+ * Take out references in order
+ * to not race with deletion.
+ *
+ * resync_endio will release them.
+ */
+ dev_get(dev_from);
+ dev_get(dev_to);
+ resync_copy(dev_to, dev_from, resync);
+
+ /*
+ * Done with copying this region:
+ * mark in sync and flush dirty log.
+ */
+ } else {
+ dl->type->set_region_sync(dl, resync->region, 1);
+ dl->type->flush(dl);
+
+ /* Optionally call back site link recovery. */
+ recover_callback(sl, 0, 0);
+ resync->end = 0;
+
+ /* Another run to check for more resync work. */
+ wake_do_slink(sl);
+ }
+ }
+
+ /* Put race device references. */
+ read_lock(&sl0->lock);
+ write_lock(&sl->lock);
+ list_for_each_entry_safe(dev_to, dev_n, &resync_list,
+ lists[SDEV_RESYNC]) {
+ if (TestClearDevResyncEnd(dev_to))
+ list_del_init(SDEV_RESYNC_LIST(dev_to));
+
+ dev_from = dev_get_by_number(sl0, dev_to->dev.number);
+ /* 1 put just taken, 1 put for the one initially taken. */
+ _BUG_ON_PTR(dev_from);
+ BUG_ON(dev_put(dev_from));
+ BUG_ON(dev_put(dev_from));
+ BUG_ON(dev_put(dev_to));
+ }
+
+ list_splice(&resync_list, SLINK_RESYNC_LIST(sl));
+ ClearSlinkResyncProcessing(sl);
+
+ write_unlock(&sl->lock);
+ read_unlock(&sl0->lock);
+
+ BUG_ON(slink_put(sl0));
+}
+
+/* Main worker thread function. */
+static void
+do_slink(struct work_struct *ws)
+{
+ int must_resync;
+ struct slink *sl = container_of(ws, struct slink, io.dws.work);
+
+ if (!SlinkTestActive(sl))
+ do_slink_test(sl);
+
+ write_lock(&sl->lock);
+ must_resync = !list_empty(SLINK_RESYNC_LIST(sl));
+ write_unlock(&sl->lock);
+
+ if (must_resync)
+ do_slink_resync(sl);
+
+ ClearSlinkImmediateWork(sl);
+}
+
+/*
+ * End site link worker.
+ */
+
+/* Allocate/init an sdev structure and dm_get_device(). */
+static struct sdev *
+dev_create(struct slink *sl, struct dm_target *ti,
+ char *path, unsigned dev_number)
+{
+ int i, r;
+ struct sdev *dev;
+
+ _BUG_ON_PTR(sl);
+ _BUG_ON_PTR(ti);
+ _BUG_ON_PTR(path);
+
+ /* Preallocate site link device structure. */
+ dev = kzalloc(sizeof(*dev), GFP_KERNEL);
+ if (unlikely(!dev))
+ goto bad_dev_alloc;
+
+ dev->ti = ti;
+ dev->dev.number = dev_number;
+ init_waitqueue_head(&dev->io.waiters);
+ kref_init(&dev->ref);
+
+ i = ARRAY_SIZE(dev->lists);
+ while (i--)
+ INIT_LIST_HEAD(dev->lists + i);
+
+ dev->dev.params.path = kstrdup(path, GFP_KERNEL);
+ if (unlikely(!dev->dev.params.path))
+ goto bad_dev_path_alloc;
+
+ /* FIXME: closed device handling 29.09.2009
+ *
+ * Allow inaccessible devices to be created, hence
+ * opening them during transport failure discovery.
+ */
+ ClearDevOpen(dev);
+ try_dev_open(dev);
+
+ /*
+ * Create kcopyd client for resynchronization copies to slinks.
+ * Only needed for remote devices and hence slinks > 0.
+ */
+ if (sl->number) {
+ r = dm_kcopyd_client_create(RESYNC_PAGES,
+ &dev->io.kcopyd_client);
+ if (unlikely(r < 0))
+ goto bad_kcopyd_client;
+ }
+
+ dev->sl = sl;
+ SetDevSuspended(dev);
+ return dev;
+
+bad_dev_alloc:
+ DMERR("site link device allocation failed");
+ return ERR_PTR(-ENOMEM);
+
+bad_dev_path_alloc:
+ DMERR("site link device path allocation failed");
+ r = -ENOMEM;
+ goto bad;
+
+bad_kcopyd_client:
+ DMERR("site link device allocation failed");
+bad:
+ dev_release(&dev->ref);
+ return ERR_PTR(r);
+}
+
+/* Add an sdev to an slink. */
+static int
+dev_add(struct slink *sl, struct sdev *dev)
+{
+ int r;
+ struct slink *sl0;
+ struct sdev *dev_tmp;
+
+ _BUG_ON_PTR(sl);
+ _BUG_ON_PTR(dev);
+
+ /* Check by number if device got already added to this site link. */
+ dev_tmp = dev_get_by_number(sl, dev->dev.number);
+ if (unlikely(!IS_ERR(dev_tmp)))
+ goto bad_device;
+
+ /* Check by bdev/path if device got already added to any site link. */
+ dev_tmp = dev_get_on_any_slink(sl, dev);
+ if (unlikely(!IS_ERR(dev_tmp)))
+ goto bad_device;
+
+ /* Sibling device on local slink 0 registered yet ? */
+ if (sl->number) {
+ sl0 = slink_get_by_number(sl->repl_slinks, 0);
+ if (unlikely(IS_ERR(sl0)))
+ goto bad_slink0;
+
+ read_lock(&sl0->lock);
+ dev_tmp = dev_get_by_number(sl0, dev->dev.number);
+ read_unlock(&sl0->lock);
+
+ BUG_ON(slink_put(sl0));
+
+ if (unlikely(IS_ERR(dev_tmp)))
+ goto bad_sibling;
+
+ BUG_ON(dev_put(dev_tmp));
+ }
+
+ /* Add to slink's list of devices. */
+ list_add_tail(SDEV_SLINK_LIST(dev), SLINK_DEVS_LIST(sl));
+
+ /* All ok, add to list of remote devices to resync. */
+ if (sl->number) {
+ SetDevResync(dev);
+ list_add_tail(SDEV_RESYNC_LIST(dev), SLINK_RESYNC_LIST(sl));
+ }
+
+ return 0;
+
+bad_device:
+ DMERR("device already exists");
+ BUG_ON(dev_put(dev_tmp));
+ return -EBUSY;
+
+bad_slink0:
+ DMERR("SLINK0 doesn't exit!");
+ r = PTR_ERR(sl0);
+ return r;
+
+bad_sibling:
+ DMERR("Sibling device=%d on SLINK0 doesn't exist!", dev->dev.number);
+ r = PTR_ERR(dev_tmp);
+ return r;
+}
+
+/*
+ * Set up dirty log for new device.
+ *
+ * For local devices, no dirty log is allowed.
+ * For remote devices, a dirty log is mandatory.
+ *
+ * dirtylog_type = "nolog"/"core"/"disk",
+ * #dirtylog_params = 0-3 (1-2 for core dirty log type, 3 for
+ * disk dirty log only)
+ * dirtylog_params = [dirty_log_path] region_size [[no]sync])
+ */
+static int
+dirty_log_create(struct slink *sl, struct sdev *dev, unsigned argc, char **argv)
+{
+ struct dm_dirty_log *dl;
+
+ SHOW_ARGV;
+
+ _BUG_ON_PTR(sl);
+ _BUG_ON_PTR(dev);
+
+ if (unlikely(argc < 2))
+ goto bad_params;
+
+ /* Check for no dirty log with local devices. */
+ if (!strcmp(argv[0], "nolog") ||
+ !strcmp(argv[0], "-")) {
+ if (argc != 2 ||
+ strcmp(argv[1], "0"))
+ goto bad_params;
+
+ dev->dev.dl = NULL;
+ dev->io.split_io = DM_REPL_MIN_SPLIT_IO;
+
+ /* Mandatory dirty logs on SLINK > 0. */
+ if (sl->number)
+ goto bad_need_dl;
+
+ return 0;
+ }
+
+ /* No dirty logs on SLINK0. */
+ if (unlikely(!sl->number))
+ goto bad_site_link;
+
+ dl = dm_dirty_log_create(argv[0], dev->ti, NULL, argc - 2, argv + 2);
+ if (unlikely(!dl))
+ goto bad_dirty_log;
+
+ dev->dev.dl = dl;
+ dev->io.split_io = dl->type->get_region_size(dl);
+ if (dev->io.split_io < BIO_MAX_SECTORS)
+ DM_EINVAL("Invalid dirty log region size");
+
+ return 0;
+
+bad_params:
+ DMERR("invalid dirty log parameter count");
+ return -EINVAL;
+
+bad_need_dl:
+ DMERR("dirty log mandatory on SLINKs > 0");
+ return -EINVAL;
+
+bad_site_link:
+ DMERR("no dirty log allowed on SLINK0");
+ return -EINVAL;
+
+bad_dirty_log:
+ DMERR("failed to create dirty log");
+ return -ENXIO;
+}
+
+/*
+ * Check and adjust split_io on all replicated devices.
+ *
+ * Called with write repl_slinks->lock and write sl->lock hold.
+ *
+ * All remote devices must go by the same dirty log
+ * region size in order to keep the caller simple.
+ *
+ * @sl = slink > 0
+ * @dev = device to check and use to set ti->split_io
+ *
+ */
+static int
+set_split_io(struct slink *sl, struct sdev *dev)
+{
+ sector_t split_io_1st, split_io_ref = 0;
+ struct slink *sl_cur, *sl0;
+ struct sdev *dev_cur;
+
+ /* Nonsense to proceed on SLINK0. */
+ _BUG_ON_PTR(sl);
+ if (!sl->number)
+ return 0;
+
+ _BUG_ON_PTR(dev);
+
+ sl0 = slink_get_by_number(sl->repl_slinks, 0);
+ _BUG_ON_PTR(sl0);
+
+ /* Get split_io from any existing dev on this actual slink. */
+ if (list_empty(SLINK_DEVS_LIST(sl)))
+ split_io_1st = 0;
+ else {
+ dev_cur = list_first_entry(SLINK_DEVS_LIST(sl), struct sdev,
+ lists[SDEV_SLINK]);
+ split_io_1st = dev_cur->io.split_io;
+ }
+
+ /* Find any preset split_io on any (slink > 0 && slink != sl) device. */
+ list_for_each_entry(sl_cur, &sl0->repl_slinks->list,
+ lists[SLINK_REPLOG]) {
+ if (!sl_cur->number ||
+ sl_cur->number == sl->number)
+ continue;
+
+ if (!list_empty(SLINK_DEVS_LIST(sl_cur))) {
+ dev_cur = list_first_entry(SLINK_DEVS_LIST(sl_cur),
+ struct sdev,
+ lists[SDEV_SLINK]);
+ split_io_ref = dev_cur->io.split_io;
+ break;
+ }
+ }
+
+
+ /*
+ * The region size *must* be the same for all devices
+ * in order to simplify the related caller logic.
+ */
+ if ((split_io_ref && split_io_1st && split_io_ref != split_io_1st) ||
+ (split_io_1st && split_io_1st != dev->io.split_io) ||
+ (split_io_ref && split_io_ref != dev->io.split_io))
+ DM_EINVAL("region size argument must be the "
+ "same for all devices");
+
+ /* Lock sl0, because we ain't get here with sl == sl0. */
+ write_lock(&sl0->lock);
+ list_for_each_entry(dev_cur, SLINK_DEVS_LIST(sl0), lists[SDEV_SLINK])
+ dev_cur->ti->split_io = dev->io.split_io;
+
+ write_unlock(&sl0->lock);
+
+ BUG_ON(slink_put(sl0));
+ return 0;
+}
+
+/* Wait on device in flight device I/O before allowing device destroy. */
+static void
+slink_wait_on_io(struct sdev *dev)
+{
+ while (dev_io(dev)) {
+ flush_workqueue(dev->sl->io.wq);
+ wait_event(dev->io.waiters, !dev_io(dev));
+ }
+}
+
+/* Postsuspend method. */
+static int
+blockdev_postsuspend(struct dm_repl_slink *slink, int dev_number)
+{
+ struct slink *sl;
+ struct sdev *dev;
+
+ DMDEBUG_LIMIT("%s dev_number=%d", __func__, dev_number);
+ _SET_AND_BUG_ON_SL(sl, slink);
+
+ if (dev_number < 0)
+ return -EINVAL;
+
+ write_lock(&sl->lock);
+ dev = dev_get_by_number(sl, dev_number);
+ if (unlikely(IS_ERR(dev))) {
+ write_unlock(&sl->lock);
+ return PTR_ERR(dev);
+ }
+
+ /* Set device suspended. */
+ SetDevSuspended(dev);
+ write_unlock(&sl->lock);
+
+ dev_put(dev);
+
+ /* Wait for any device io to finish. */
+ slink_wait_on_io(dev);
+ return 0;
+}
+
+/* Resume method. */
+static int
+blockdev_resume(struct dm_repl_slink *slink, int dev_number)
+{
+ struct slink *sl;
+ struct sdev *dev;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+ DMDEBUG("%s sl_number=%d dev_number=%d", __func__,
+ sl->number, dev_number);
+
+ if (dev_number < 0)
+ return -EINVAL;
+
+ read_lock(&sl->lock);
+ dev = dev_get_by_number(sl, dev_number);
+ read_unlock(&sl->lock);
+
+ if (unlikely(IS_ERR(dev)))
+ return PTR_ERR(dev);
+
+ /* Clear device suspended. */
+ ClearDevSuspended(dev);
+ BUG_ON(dev_put(dev));
+ wake_do_slink(sl);
+ return 0;
+}
+
+/* Destroy device resources. */
+static void
+dev_destroy(struct sdev *dev)
+{
+ if (dev->dev.dl)
+ dm_dirty_log_destroy(dev->dev.dl);
+
+ if (dev->io.kcopyd_client)
+ dm_kcopyd_client_destroy(dev->io.kcopyd_client);
+
+ if (dev->dev.dm_dev)
+ dm_put_device(dev->ti, dev->dev.dm_dev);
+
+ BUG_ON(!dev_put(dev));
+}
+
+/*
+ * Method to add a device to a given site link
+ * and optionally create a dirty log for it.
+ *
+ * @dev_number = unsigned int stored in the REPLOG to associate to a dev_path
+ * @ti = dm_target ptr (needed for dm functions)
+ * @argc = 4...
+ * @argv = dev_params# dev_path dirty_log_args
+ */
+#define MIN_DEV_ARGS 4
+static int
+blockdev_dev_add(struct dm_repl_slink *slink, int dev_number,
+ struct dm_target *ti, unsigned argc, char **argv)
+{
+ int r;
+ unsigned dev_params, params, sl_count;
+ long long tmp;
+ struct slink *sl;
+ struct sdev *dev;
+
+ SHOW_ARGV;
+
+ if (dev_number < 0)
+ return -EINVAL;
+
+ /* Two more because of the following dirty log parameters. */
+ if (unlikely(argc < MIN_DEV_ARGS))
+ DM_EINVAL("invalid device parameters count");
+
+ /* Get #dev_params. */
+ if (unlikely(sscanf(argv[0], "%lld", &tmp) != 1 ||
+ tmp != 1)) {
+ DM_EINVAL("invalid device parameters argument");
+ } else
+ dev_params = tmp;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+ _BUG_ON_PTR(sl->repl_slinks);
+
+ dev = dev_create(sl, ti, argv[1], dev_number);
+ if (unlikely(IS_ERR(dev)))
+ return PTR_ERR(dev);
+
+ dev->dev.params.count = dev_params;
+
+ /* Work on dirty log paramaters. */
+ params = dev_params + 1;
+ r = dirty_log_create(sl, dev, argc - params, argv + params);
+ if (unlikely(r < 0))
+ goto bad;
+
+ /* Take out global and local lock to update the configuration. */
+ write_lock(&sl->repl_slinks->lock);
+ write_lock(&sl->lock);
+
+ /* Set split io value on all replicated devices. */
+ r = set_split_io(sl, dev);
+ if (unlikely(r < 0))
+ goto bad_unlock;
+
+ /*
+ * Now that dev is all set, add it to the slink.
+ *
+ * If callers are racing for the same device,
+ * dev_add() will catch that case too.
+ */
+ r = dev_add(sl, dev);
+ if (unlikely(r < 0))
+ goto bad_unlock;
+
+ write_unlock(&sl->lock);
+
+ sl_count = slink_count(sl);
+ write_unlock(&sl->repl_slinks->lock);
+
+ /* Ignore any resize problem and live with what we got. */
+ if (sl_count > 1)
+ dm_io_client_resize(sl_count, sl->io.dm_io_client);
+
+ DMDEBUG("%s added device=%u to slink=%u",
+ __func__, dev_number, sl->number);
+ return dev_number;
+
+bad_unlock:
+ write_unlock(&sl->lock);
+ write_unlock(&sl->repl_slinks->lock);
+bad:
+ BUG_ON(dev_io(dev));
+ dev_destroy(dev);
+ return r;
+}
+
+/* Method to delete a device from a given site link. */
+static int
+blockdev_dev_del(struct dm_repl_slink *slink, int dev_number)
+{
+ int i;
+ unsigned sl_count;
+ struct slink *sl;
+ struct sdev *dev;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+
+ if (dev_number < 0)
+ return -EINVAL;
+
+ /* Check if device is active! */
+ write_lock(&sl->lock);
+ dev = dev_get_by_number(sl, dev_number);
+ if (unlikely(IS_ERR(dev))) {
+ write_unlock(&sl->lock);
+ return PTR_ERR(dev);
+ }
+
+ SetDevTeardown(dev);
+ write_unlock(&sl->lock);
+
+ /* Release the new reference taken out via dev_get_by_number() .*/
+ BUG_ON(dev_put(dev));
+
+ /* Wait for any device I/O to finish. */
+ slink_wait_on_io(dev);
+ BUG_ON(dev_io(dev));
+
+ /* Take device off any lists. */
+ write_lock(&sl->lock);
+ i = ARRAY_SIZE(dev->lists);
+ while (i--) {
+ if (!list_empty(dev->lists + i))
+ list_del(dev->lists + i);
+ }
+
+ write_unlock(&sl->lock);
+
+ /* Destroy device. */
+ dev_destroy(dev);
+
+ /* Ignore any resize problem. */
+ sl_count = slink_count(sl);
+ dm_io_client_resize(sl_count ? sl_count : 1, sl->io.dm_io_client);
+ DMDEBUG("%s deleted device=%u from slink=%u",
+ __func__, dev_number, sl->number);
+ return 0;
+}
+
+/* Check slink properties for consistency. */
+static int slink_check_properties(struct slink_params *params)
+{
+ enum dm_repl_slink_policy_type policy = params->policy;
+
+ if (slink_policy_synchronous(policy) &&
+ slink_policy_asynchronous(policy))
+ DM_EINVAL("synchronous and asynchronous slink "
+ "policies are mutually exclusive!");
+
+ if (slink_policy_synchronous(policy) &&
+ params->fallbehind.value)
+ DM_EINVAL("synchronous slink policy and fallbehind "
+ "are mutually exclusive!");
+ return 0;
+}
+
+/*
+ * Start methods of "blockdev" slink type.
+ */
+/* Method to destruct a site link context. */
+static void
+blockdev_dtr(struct dm_repl_slink *slink)
+{
+ struct slink *sl;
+
+ _BUG_ON_PTR(slink);
+ sl = slink->context;
+ _BUG_ON_PTR(sl);
+
+ slink_destroy(sl);
+ BUG_ON(!slink_put(sl));
+}
+
+/*
+ * Method to construct a site link context.
+ *
+ * #slink_params = 1-4
+ * <slink_params> = slink# [slink_policy [fall_behind value]]
+ * slink# = used to tie the host+dev_path to a particular SLINK; 0 is used
+ * for the local site link and 1-M are for remote site links.
+ * slink_policy = policy to set on the slink (eg. async/sync)
+ * fall_behind = # of ios the SLINK can fall behind before switching to
+ * synchronous mode (ios N, data N[kmgtpe], timeout N[smhd])
+ */
+static int
+blockdev_ctr(struct dm_repl_slink *slink, struct dm_repl_log *replog,
+ unsigned argc, char **argv)
+{
+ int check, r;
+ long long tmp;
+ unsigned slink_number, slink_params;
+ struct slink *sl;
+ struct slink_params params;
+ /* Makes our task to keep a unique list of slinks per replog easier. */
+ struct dm_repl_log_slink_list *repl_slinks =
+ replog->ops->slinks(replog);
+
+ SHOW_ARGV;
+ _BUG_ON_PTR(repl_slinks);
+
+ memset(¶ms, 0, sizeof(params));
+ params.policy = DM_REPL_SLINK_ASYNC;
+ params.fallbehind.type = DM_REPL_SLINK_FB_IOS;
+ params.fallbehind.multiplier = 1;
+
+ if (unlikely(argc < 2))
+ DM_EINVAL("invalid number of slink arguments");
+
+ /* Get # of slink parameters. */
+ if (unlikely(sscanf(argv[0], "%lld", &tmp) != 1 ||
+ tmp < 1 || tmp > argc))
+ DM_EINVAL("invalid slink parameter argument");
+ else
+ params.count = slink_params = tmp;
+
+
+ /* Get slink#. */
+ if (unlikely(sscanf(argv[1], "%lld", &tmp) != 1 ||
+ tmp < 0 || tmp >= replog->ops->slink_max(replog))) {
+ DM_EINVAL("invalid slink number argument");
+ } else
+ slink_number = tmp;
+
+ if (slink_params > 1) {
+ /* Handle policy argument. */
+ r = get_slink_policy(argv[2]);
+ if (unlikely(r < 0))
+ return r;
+
+ params.policy = r;
+ check = 1;
+
+ /* Handle fallbehind argument. */
+ if (slink_params > 2) {
+ r = get_slink_fallbehind(slink_params,
+ argv + 3, ¶ms.fallbehind);
+ if (unlikely(r < 0))
+ return r;
+ }
+ } else
+ check = 0;
+
+ /* Check that policies make sense vs. fallbehind. */
+ if (check) {
+ r = slink_check_properties(¶ms);
+ if (r < 0)
+ return r;
+ }
+
+ /* Get/create an slink context. */
+ sl = slink_create(slink, repl_slinks, ¶ms, slink_number);
+ return unlikely(IS_ERR(sl)) ? PTR_ERR(sl) : 0;
+}
+
+/*
+ * Initiate data copy across a site link.
+ *
+ * This function may be used to copy a buffer entry *or*
+ * for resynchronizing regions initially or when an SLINK
+ * has fallen back to dirty log (bitmap) mode.
+ */
+/* Get sdev ptr from copy address. */
+static struct sdev *
+dev_get_by_addr(struct slink *sl, struct dm_repl_slink_copy_addr *addr)
+{
+ BUG_ON(addr->type != DM_REPL_SLINK_BLOCK_DEVICE &&
+ addr->type != DM_REPL_SLINK_DEV_NUMBER);
+ return addr->type == DM_REPL_SLINK_BLOCK_DEVICE ?
+ dev_get_by_bdev(sl, addr->dev.bdev) :
+ dev_get_by_number(sl, addr->dev.number.dev);
+}
+
+/*
+ * Needs to be called with sl->lock held.
+ *
+ * Return 0 in case io is allowed to the region the sector
+ * is in and take out an I/O reference on the device.
+ *
+ * If I/O isn't allowd, no I/O reference will be taken out
+ * and the follwoing return codes apply to caller actions:
+ *
+ * o -EAGAIN in case of prohibiting I/O because of device suspension
+ * or device I/O errors (i.e. link temporarilly down) ->
+ * caller is allowed to retry the I/O later once
+ * he'll have received a callback.
+ *
+ * o -EACCES in case a region is being resynchronized and the source
+ * region is being read to copy data accross to the same region
+ * of the replica (RD) ->
+ * caller is allowed to retry the I/O later once
+ * he'll have received a callback.
+ *
+ * o -ENODEV in case a device is not configured
+ * caller must drop the I/O to the device/slink pair.
+ *
+ * o -EPERM in case a region is out of sync ->
+ * caller must drop the I/O to the device/slink pair.
+ */
+static int
+may_io(int rw, struct slink *sl, struct sdev *dev,
+ sector_t sector, const char *action)
+{
+ int r;
+ sector_t region;
+ struct slink *sl_cur;
+ struct sdev *dev_tmp;
+ struct sdev_resync *resync;
+
+ _BUG_ON_PTR(sl);
+
+ if (IS_ERR(dev))
+ return PTR_ERR(dev);
+
+ region = sector_to_region(dev, sector);
+
+ /*
+ * It's a caller error to call for multiple copies per slink.
+ */
+ if (rw == WRITE &&
+ SlinkWriter(sl)) {
+ DMERR_LIMIT("%s %s to slink%d, dev%d, region=%llu "
+ "while write pending to region=%llu!",
+ __func__, action, sl->number, dev->dev.number,
+ (unsigned long long) region,
+ (unsigned long long) dev->io.resync.writer_region);
+ return -EPERM;
+ }
+
+ /*
+ * If the device is suspended, being torn down,
+ * closed or errored, retry again later.
+ */
+ if (!DevOpen(dev) ||
+ DevSuspended(dev) ||
+ DevTeardown(dev) ||
+ DevErrorRead(dev) ||
+ DevErrorWrite(dev))
+ return -EAGAIN;
+
+ /* slink > 0 may read and write in case region is in sync. */
+ if (sl->number) {
+ struct dm_dirty_log *dl = dev->dev.dl;
+
+ /*
+ * Ask dirty log for region in sync.
+ *
+ * In sync -> allow reads and writes.
+ * Out of sync -> prohibit them.
+ */
+ _BUG_ON_PTR(dl);
+ r = dl->type->in_sync(dl, region, 0); /* Don't block. */
+ r = r ? 0 : -EPERM;
+ } else {
+ /* slink0 may always read. */
+ if (rw == READ)
+ return 0;
+
+ read_lock(&sl->repl_slinks->lock);
+
+ /*
+ * Walk all slinks and check if anyone is syncing this region,
+ * in which case no write is allowed to it on slink0.
+ */
+ list_for_each_entry(sl_cur, &sl->repl_slinks->list,
+ lists[SLINK_REPLOG]) {
+ /* Avoid local devices. */
+ if (!sl_cur->number)
+ continue;
+
+ /*
+ * If device exists and the LD is
+ * being read off for resync.
+ */
+ read_lock(&sl_cur->lock);
+ dev_tmp = dev_get_by_number(sl_cur, dev->dev.number);
+ read_unlock(&sl_cur->lock);
+
+ if (!IS_ERR(dev_tmp)) {
+ resync = &dev_tmp->io.resync;
+ if (resync->end &&
+ resync->region == region) {
+ BUG_ON(dev_put(dev_tmp));
+ r = -EACCES;
+ goto out;
+ }
+
+ BUG_ON(dev_put(dev_tmp));
+ }
+ }
+
+ /* We're allowed to write to this LD -> indicate we do. */
+ r = 0;
+out:
+ read_unlock(&sl->repl_slinks->lock);
+ }
+
+ if (!r && rw == WRITE) {
+ write_lock(&sl->lock);
+ /*
+ * Memorize region being synchronized
+ * to check in do_slink_resync().
+ */
+ dev->io.resync.writer_region = region;
+ SetSlinkWriter(sl);
+ write_unlock(&sl->lock);
+ }
+
+ return r;
+}
+
+/* Set source/destination address of the copy. */
+static int
+copy_addr_init(struct slink *sl, struct dm_io_region *io,
+ struct dm_repl_slink_copy_addr *addr, unsigned size)
+{
+
+ if (addr->type == DM_REPL_SLINK_BLOCK_DEVICE) {
+ io->bdev = addr->dev.bdev;
+ } else if (addr->type == DM_REPL_SLINK_DEV_NUMBER) {
+ struct sdev *dev;
+ struct slink *sl_tmp;
+
+ /* Check that slink number is correct. */
+ read_lock(&sl->repl_slinks->lock);
+ sl_tmp = slink_get_by_number(sl->repl_slinks,
+ addr->dev.number.slink);
+ if (unlikely(IS_ERR(sl_tmp))) {
+ int r = PTR_ERR(sl_tmp);
+
+ read_unlock(&sl->repl_slinks->lock);
+ return r;
+ }
+
+ read_unlock(&sl->repl_slinks->lock);
+
+ if (unlikely(sl != sl_tmp))
+ return -EINVAL;
+
+ read_lock(&sl_tmp->lock);
+ dev = dev_get_by_number(sl_tmp, addr->dev.number.dev);
+ read_unlock(&sl_tmp->lock);
+
+ BUG_ON(slink_put(sl_tmp));
+
+ if (unlikely(IS_ERR(dev)))
+ return PTR_ERR(dev);
+
+ io->bdev = dev->dev.dm_dev->bdev;
+ BUG_ON(dev_put(dev));
+ } else
+ BUG();
+
+ io->sector = addr->sector;
+ io->count = dm_div_up(size, to_bytes(1));
+ return 0;
+}
+
+/*
+ * Copy endio function.
+ *
+ * For the "blockdev" type, both states (data in (remote) ram and data
+ * on (remote) disk) are reported here at once. For future transports
+ * those will be reported seperately.
+ */
+static void
+copy_endio(int read_err, unsigned long write_err, void *context)
+{
+ struct copy_context *ctx = context;
+ struct slink *sl;
+ struct sdev *dev_to;
+
+ _BUG_ON_PTR(ctx);
+ dev_to = ctx->dev_to;
+ _BUG_ON_PTR(dev_to);
+ sl = dev_to->sl;
+ _BUG_ON_PTR(sl);
+
+ /* Throw a table event in case of a site link device copy error. */
+ if (unlikely(read_err || write_err)) {
+ if (read_err) {
+ if (!TestSetDevErrorRead(dev_to)) {
+ SetSlinkErrorRead(sl);
+ dm_table_event(dev_to->ti->table);
+ }
+ } else if (!TestSetDevErrorWrite(dev_to)) {
+ SetSlinkErrorWrite(sl);
+ dm_table_event(dev_to->ti->table);
+ }
+
+ set_dev_test_time(dev_to);
+ }
+
+ /* Must clear before calling back. */
+ ClearSlinkWriter(sl);
+
+ /*
+ * FIXME: check if caller has set region to NOSYNC and
+ * and if so, avoid calling callbacks completely.
+ */
+ if (ctx->ram.fn)
+ ctx->ram.fn(read_err, write_err ? -EIO : 0, ctx->ram.context);
+
+ /* Only call when no error or when no ram callback defined. */
+ if (likely((!read_err && !write_err) || !ctx->ram.fn))
+ ctx->disk.fn(read_err, write_err, ctx->disk.context);
+
+ /* Copy done slinkX device. */
+ free_copy_context(ctx, sl);
+
+ /* Release device reference. */
+ BUG_ON(dev_put(dev_to));
+
+ /* Wake slink worker to reschedule any postponed resynchronization. */
+ wake_do_slink(sl);
+}
+
+/* Site link copy method. */
+static int
+blockdev_copy(struct dm_repl_slink *slink, struct dm_repl_slink_copy *copy,
+ unsigned long long tag)
+{
+ int r;
+ struct slink *sl;
+ struct sdev *dev_to;
+ struct copy_context *ctx = NULL;
+ static struct dm_io_region src, dst;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+ if (unlikely(SlinkErrorRead(sl) || SlinkErrorWrite(sl)))
+ return -EAGAIN;
+
+ /* Get device by address taking out reference. */
+ read_lock(&sl->lock);
+ dev_to = dev_get_by_addr(sl, ©->dst);
+ read_unlock(&sl->lock);
+
+ /* Check if io is allowed or a resync is active. */
+ r = may_io(WRITE, sl, dev_to, copy->dst.sector, "copy");
+ if (r < 0)
+ goto bad;
+
+ ctx = alloc_copy_context(sl);
+ BUG_ON(!ctx);
+
+ /* Device to copy to. */
+ ctx->dev_to = dev_to;
+
+ /* Save caller context. */
+ ctx->ram = copy->ram;
+ ctx->disk = copy->disk;
+
+ /* Setup copy source. */
+ r = copy_addr_init(sl, &src, ©->src, copy->size);
+ if (unlikely(r < 0))
+ goto bad;
+
+ /* Setup copy destination. */
+ r = copy_addr_init(sl, &dst, ©->dst, copy->size);
+ if (unlikely(r < 0))
+ goto bad;
+
+ /* FIXME: can we avoid reading per copy on multiple slinks ? */
+ r = dm_kcopyd_copy(sl->io.kcopyd_client, &src, 1, &dst, 0,
+ copy_endio, ctx);
+ BUG_ON(r); /* dm_kcopyd_copy() may never fail. */
+ SetDevIOQueued(dev_to); /* dev_unplug(src.bdev); */
+ return r;
+
+bad:
+ if (!IS_ERR(dev_to))
+ BUG_ON(dev_put(dev_to));
+
+ if (ctx)
+ free_copy_context(ctx, sl);
+
+ return r;
+}
+
+/* Method to get site link policy. */
+static enum dm_repl_slink_policy_type
+blockdev_policy(struct dm_repl_slink *slink)
+{
+ struct slink *sl = slink_check(slink);
+
+ return IS_ERR(sl) ? PTR_ERR(sl) : sl->params.policy;
+}
+
+/* Method to get site link State. */
+static enum dm_repl_slink_state_type
+blockdev_state(struct dm_repl_slink *slink)
+{
+ enum dm_repl_slink_state_type state = 0;
+ struct slink *sl = slink_check(slink);
+
+ if (unlikely(IS_ERR(sl)))
+ return PTR_ERR(sl);
+
+ if (SlinkErrorRead(sl))
+ set_bit(DM_REPL_SLINK_READ_ERROR, (unsigned long *) &state);
+
+ if (SlinkErrorWrite(sl))
+ set_bit(DM_REPL_SLINK_DOWN, (unsigned long *) &state);
+
+ return state;
+}
+
+/* Method to get reference to site link fallbehind parameters. */
+static struct dm_repl_slink_fallbehind *
+blockdev_fallbehind(struct dm_repl_slink *slink)
+{
+ struct slink *sl = slink_check(slink);
+
+ return IS_ERR(sl) ? ((struct dm_repl_slink_fallbehind *) sl) :
+ &sl->params.fallbehind;
+}
+
+/* Return # of the device. */
+static int
+blockdev_dev_number(struct dm_repl_slink *slink, struct block_device *bdev)
+{
+ struct slink *sl = slink_check(slink);
+ struct sdev *dev;
+ struct mapped_device *md;
+
+ if (unlikely(IS_ERR(sl)))
+ return PTR_ERR(sl);
+
+ if (unlikely(sl->number)) {
+ DMERR("Can't retrieve device number from slink > 0");
+ return -EINVAL;
+ }
+
+ read_lock(&sl->lock);
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ md = dm_table_get_md(dev->ti->table);
+ if (bdev->bd_disk == dm_disk(md)) {
+ read_unlock(&sl->lock);
+ dm_put(md);
+ return dev->dev.number;
+ }
+
+ dm_put(md);
+ }
+
+ read_unlock(&sl->lock);
+
+ /*
+ * The caller might have removed the device from SLINK0 but
+ * have an order to copy to the device in its metadata still,
+ * so he has to react accordingly (ie. remove device copy request).
+ */
+ return -ENOENT;
+}
+
+/* Method to remap bio to underlying device on slink0. */
+static int
+blockdev_io(struct dm_repl_slink *slink, struct bio *bio,
+ unsigned long long tag)
+{
+ int r, rw = bio_data_dir(bio);
+ struct slink *sl;
+ struct sdev *dev;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+
+ /*
+ * Prohibit slink > 0 I/O, because the resync
+ * code can't cope with it for now...
+ */
+ if (sl->number)
+ DM_EPERM("I/O to slink > 0 prohibited!");
+
+ if (rw == WRITE)
+ DM_EPERM("Writes to slink0 prohibited!");
+
+ read_lock(&sl->lock);
+ dev = dev_get_by_bdev(sl, bio->bi_bdev);
+ read_unlock(&sl->lock);
+
+ /* Check if io is allowed or a resync is active. */
+ r = may_io(rw, sl, dev, bio->bi_sector, "io");
+ if (likely(!r)) {
+ bio->bi_bdev = dev->dev.dm_dev->bdev;
+ generic_make_request(bio);
+ SetDevIOQueued(dev);
+ }
+
+ if (!IS_ERR(dev))
+ BUG_ON(dev_put(dev));
+
+ return r;
+}
+
+/* Method to unplug all device queues on a site link. */
+static int
+blockdev_unplug(struct dm_repl_slink *slink)
+{
+ struct slink *sl = slink_check(slink);
+ struct sdev *dev, *dev_n;
+
+ if (unlikely(IS_ERR(sl)))
+ return PTR_ERR(sl);
+
+ /* Take out device references for all devices with IO queued. */
+ read_lock(&sl->lock);
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ if (TestClearDevIOQueued(dev)) {
+ dev_get(dev);
+ SetDevIOUnplug(dev);
+ }
+ }
+
+ read_unlock(&sl->lock);
+
+ list_for_each_entry_safe(dev, dev_n,
+ SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ if (TestClearDevIOUnplug(dev)) {
+ if (DevOpen(dev) &&
+ !DevSuspended(dev) &&
+ !DevTeardown(dev))
+ dev_unplug(dev->dev.dm_dev->bdev);
+
+ BUG_ON(dev_put(dev));
+ }
+ }
+
+ return 0;
+}
+
+/* Method to set global recovery function and context. */
+static void
+blockdev_recover_notify_fn_set(struct dm_repl_slink *slink,
+ dm_repl_notify_fn fn, void *context)
+{
+ struct slink *sl;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+
+ write_lock(&sl->lock);
+ sl->recover.fn = fn;
+ sl->recover.context = context;
+ write_unlock(&sl->lock);
+}
+
+/* Method to return # of the SLINK. */
+static int
+blockdev_slink_number(struct dm_repl_slink *slink)
+{
+ struct slink *sl = slink_check(slink);
+
+ return unlikely(IS_ERR(sl)) ? PTR_ERR(sl) : sl->number;
+}
+
+/* Method to return SLINK by number. */
+static struct dm_repl_slink *
+blockdev_slink(struct dm_repl_log *replog, unsigned slink_number)
+{
+ struct slink *sl;
+ struct dm_repl_slink *slink;
+ struct dm_repl_log_slink_list *repl_slinks;
+
+ _BUG_ON_PTR(replog);
+ repl_slinks = replog->ops->slinks(replog);
+ _BUG_ON_PTR(repl_slinks);
+
+ read_lock(&repl_slinks->lock);
+ sl = slink_get_by_number(repl_slinks, slink_number);
+ if (IS_ERR(sl))
+ slink = (struct dm_repl_slink *) sl;
+ else {
+ slink = sl->slink;
+ BUG_ON(slink_put(sl));
+ }
+
+ read_unlock(&repl_slinks->lock);
+ return slink;
+}
+
+/* Method to set SYNC state of a region of a device. */
+static int
+blockdev_set_sync(struct dm_repl_slink *slink, int dev_number,
+ sector_t sector, int in_sync)
+{
+ struct sdev *dev;
+ struct sdev_dev *sd;
+ struct dm_dirty_log *dl;
+ struct slink *sl;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+
+ if (dev_number < 0)
+ return -EINVAL;
+
+ read_lock(&sl->lock);
+ dev = dev_get_by_number(sl, dev_number);
+ read_unlock(&sl->lock);
+
+ if (IS_ERR(dev))
+ return PTR_ERR(dev);
+
+ sd = &dev->dev;
+ dl = sd->dl;
+ if (dl)
+ dl->type->set_region_sync(dl, sector_to_region(dev, sector),
+ in_sync);
+ BUG_ON(dev_put(dev));
+ return 0;
+}
+
+/* Method to flush all dirty logs on slink's devices. */
+static int
+blockdev_flush_sync(struct dm_repl_slink *slink)
+{
+ int r = 0;
+ struct slink *sl;
+ struct sdev *dev;
+ struct list_head resync_list;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+ if (!sl->number)
+ return -EINVAL;
+
+ INIT_LIST_HEAD(&resync_list);
+
+ write_lock(&sl->lock);
+ if (SlinkResyncProcessing(sl)) {
+ write_unlock(&sl->lock);
+ return -EAGAIN;
+ }
+
+ /* Take out resync list in order to process flushs unlocked. */
+ list_splice(SLINK_RESYNC_LIST(sl), &resync_list);
+ INIT_LIST_HEAD(SLINK_RESYNC_LIST(sl));
+
+ /* Get race references on devices. */
+ list_for_each_entry(dev, &resync_list, lists[SDEV_RESYNC]) {
+ _BUG_ON_PTR(dev->dev.dl);
+ dev_get(dev);
+ }
+
+ write_unlock(&sl->lock);
+
+ list_for_each_entry(dev, &resync_list, lists[SDEV_RESYNC]) {
+ if (DevOpen(dev) &&
+ !(DevSuspended(dev) || DevTeardown(dev))) {
+ int rr;
+ struct dm_dirty_log *dl = dev->dev.dl;
+
+ rr = dl->type->flush(dl);
+ if (rr && !r)
+ r = rr;
+ }
+ }
+
+ /* Put race device references. */
+ write_lock(&sl->lock);
+ list_for_each_entry(dev, &resync_list, lists[SDEV_RESYNC])
+ BUG_ON(dev_put(dev));
+
+ list_splice(&resync_list, SLINK_RESYNC_LIST(sl));
+ write_unlock(&sl->lock);
+
+ return r;
+}
+
+/*
+ * Method to trigger/prohibit resynchronzation on all devices by
+ * adding to the slink resync list and waking up the worker.
+ *
+ * We may *not* remove from the slink resync list here,
+ * because we'd end up with partitally resynchronized
+ * regions in do_slink_resync() otherwise.
+ */
+static int
+blockdev_resync(struct dm_repl_slink *slink, int resync)
+{
+ struct slink *sl;
+ struct sdev *dev;
+
+ _SET_AND_BUG_ON_SL(sl, slink);
+
+ /* Don't proceed on site link 0. */
+ if (!sl->number)
+ return -EINVAL;
+
+ /* If resync processing, we need to postpone. */
+ write_lock(&sl->lock);
+ if (SlinkResyncProcessing(sl)) {
+ write_unlock(&sl->lock);
+ return -EAGAIN;
+ }
+
+ list_for_each_entry(dev, SLINK_DEVS_LIST(sl), lists[SDEV_SLINK]) {
+ BUG_ON(IS_ERR(dev));
+
+ if (resync) {
+ SetDevResync(dev);
+
+ /* Add to resync list if not yet on. */
+ if (list_empty(SDEV_RESYNC_LIST(dev))) {
+ list_add_tail(SDEV_RESYNC_LIST(dev),
+ SLINK_RESYNC_LIST(sl));
+ break;
+ }
+ } else {
+ ClearDevResync(dev);
+
+ /* emove from resync list if on. */
+ if (!list_empty(SDEV_RESYNC_LIST(dev)))
+ list_del_init(SDEV_RESYNC_LIST(dev));
+ }
+ }
+
+ write_unlock(&sl->lock);
+ wake_do_slink(sl);
+ return 0;
+}
+
+/*
+ * Method to check if a region is in sync
+ * by sector on all devices on all slinks.
+ */
+static int
+blockdev_in_sync(struct dm_repl_slink *slink, int dev_number, sector_t sector)
+{
+ int nosync = 0;
+ sector_t region = 0;
+ struct slink *sl = slink_check(slink);
+
+ if (IS_ERR(sl) ||
+ dev_number < 0)
+ return -EINVAL;
+
+ BUG_ON(!sl->repl_slinks);
+
+ read_lock(&sl->repl_slinks->lock);
+ list_for_each_entry(sl, &sl->repl_slinks->list, lists[SLINK_REPLOG]) {
+ int r;
+ struct dm_dirty_log *dl;
+ struct sdev *dev;
+
+ if (!sl->number)
+ continue;
+
+ read_lock(&sl->lock);
+ dev = dev_get_by_number(sl, dev_number);
+ read_unlock(&sl->lock);
+
+ if (IS_ERR(dev))
+ continue;
+
+ dl = dev->dev.dl;
+ _BUG_ON_PTR(dl);
+
+ /* Calculate region once for all devices on any slinks. */
+ if (!region)
+ region = sector_to_region(dev, sector);
+
+ r = dl->type->in_sync(dl, region, 0);
+ BUG_ON(dev_put(dev));
+ if (!r) {
+ nosync = 1;
+ break;
+ }
+ }
+
+ read_unlock(&sl->repl_slinks->lock);
+ return nosync;
+}
+
+/*
+ * Method for site link messages.
+ *
+ * fallbehind ios/size/timeout=X[unit]
+ * policy X
+ */
+static int
+blockdev_message(struct dm_repl_slink *slink, unsigned argc, char **argv)
+{
+ int r;
+ struct slink_params params;
+ struct slink *sl = slink_check(slink);
+
+ if (IS_ERR(sl))
+ return PTR_ERR(sl);
+
+ if (unlikely(argc < 2))
+ goto bad_arguments;
+
+ /* Preserve parameters. */
+ params = sl->params;
+
+ if (!strnicmp(STR_LEN(argv[0], "fallbehind"))) {
+ if (argc != 2)
+ DM_EINVAL("wrong fallbehind argument count");
+
+ r = get_slink_fallbehind(argc - 1, argv + 1,
+ ¶ms.fallbehind);
+ if (r < 0)
+ return r;
+ } else if (!strnicmp(STR_LEN(argv[0], "policy"))) {
+ if (argc != 2)
+ DM_EINVAL("wrong policy argument count");
+
+ r = get_slink_policy(argv[1]);
+ if (r < 0)
+ return r;
+
+ params.policy = r;
+ } else
+ DM_EINVAL("invalid message received");
+
+ /* Check properties' consistency. */
+ r = slink_check_properties(¶ms);
+ if (r < 0)
+ return r;
+
+ /* Set parameters. */
+ sl->params = params;
+ return 0;
+
+bad_arguments:
+ DM_EINVAL("too few message arguments");
+}
+
+/* String print site link error state. */
+static const char *
+snprint_slink_error(struct slink *sl, char *result, size_t maxlen)
+{
+ size_t sz = 0;
+
+ *result = 0;
+ if (SlinkErrorRead(sl))
+ DMEMIT("R");
+
+ if (SlinkErrorWrite(sl))
+ DMEMIT("W");
+
+ if (!*result)
+ DMEMIT("A");
+
+ return result;
+}
+
+/* String print device status. */
+static const char *
+snprint_device(struct slink *sl, struct sdev *dev,
+ status_type_t type, char *result, unsigned maxlen)
+{
+ size_t sz = 0;
+ static char buf[BDEVNAME_SIZE];
+ struct sdev_dev *sd;
+ struct dm_dirty_log *dl;
+
+ *result = 0;
+ if (IS_ERR(dev))
+ goto out;
+
+ sd = &dev->dev;
+ DMEMIT("%u %s ", sd->params.count,
+ sd->dm_dev ?
+ format_dev_t(buf, sd->dm_dev->bdev->bd_dev) :
+ sd->params.path);
+ dl = sd->dl;
+ if (dl)
+ dl->type->status(dl, type, result + sz, maxlen - sz);
+ else
+ DMEMIT("nolog 0");
+
+out:
+ return result;
+}
+
+/* String print device resynchronization state. */
+static const char *
+snprint_sync_count(struct slink *sl, struct sdev *dev,
+ char *result, unsigned maxlen)
+{
+ size_t sz = 0;
+ struct dm_dirty_log *dl;
+
+ if (IS_ERR(dev))
+ goto no_dev;
+
+ dl = dev->dev.dl;
+ if (dl) {
+ DMEMIT("%llu%s/%llu",
+ (unsigned long long) dl->type->get_sync_count(dl),
+ DevResyncing(dev) ? "+" : "",
+ (unsigned long long) region_count(dev));
+ } else {
+no_dev:
+ DMEMIT("-");
+ }
+
+ return result;
+}
+
+/* Method for site link status requests. */
+static struct dm_repl_slink_type blockdev_type;
+static int
+blockdev_status(struct dm_repl_slink *slink, int dev_number,
+ status_type_t type, char *result, unsigned int maxlen)
+{
+ size_t sz = 0;
+ static char buffer[256];
+ struct slink *sl_cur, *sl = slink_check(slink);
+ struct sdev *dev;
+ struct slink_params *p;
+ struct list_head *sl_list;
+
+ if (unlikely(IS_ERR(sl)))
+ return PTR_ERR(sl);
+
+ if (dev_number < -1)
+ return -EINVAL;
+
+ BUG_ON(!sl->repl_slinks);
+ sl_list = &sl->repl_slinks->list;
+
+ read_lock(&sl->repl_slinks->lock);
+
+ switch (type) {
+ case STATUSTYPE_INFO:
+ list_for_each_entry(sl_cur, sl_list, lists[SLINK_REPLOG]) {
+ read_lock(&sl_cur->lock);
+ dev = dev_get_by_number(sl_cur, dev_number);
+ read_unlock(&sl_cur->lock);
+
+ if (!IS_ERR(dev)) {
+ DMEMIT("%s,",
+ snprint_slink_error(sl_cur, buffer,
+ sizeof(buffer)));
+
+ DMEMIT("%s ",
+ snprint_sync_count(sl_cur, dev, buffer,
+ sizeof(buffer)));
+ BUG_ON(dev_put(dev));
+ }
+ }
+
+ break;
+
+ case STATUSTYPE_TABLE:
+ list_for_each_entry(sl_cur, sl_list, lists[SLINK_REPLOG]) {
+ read_lock(&sl_cur->lock);
+ if (dev_number < 0) {
+ p = &sl_cur->params;
+ DMEMIT("%s %u %u ",
+ blockdev_type.type.name,
+ p->count, sl_cur->number);
+
+ if (p->count > 1) {
+ snprint_policies(p->policy, buffer,
+ sizeof(buffer));
+ DMEMIT("%s ", buffer);
+ snprint_fallbehind(&p->fallbehind,
+ buffer,
+ sizeof(buffer));
+ if (p->count > 2)
+ DMEMIT("%s ", buffer);
+ }
+ } else {
+ dev = dev_get_by_number(sl_cur, dev_number);
+ if (!IS_ERR(dev)) {
+ DMEMIT("%u %s ", sl_cur->number,
+ snprint_device(sl_cur, dev, type,
+ buffer,
+ sizeof(buffer)));
+ BUG_ON(dev_put(dev));
+ }
+ }
+
+ read_unlock(&sl_cur->lock);
+ }
+ }
+
+ read_unlock(&sl->repl_slinks->lock);
+ return 0;
+}
+
+/*
+ * End methods of "blockdev" slink type.
+ */
+
+/* "blockdev" SLINK handler interface type. */
+static struct dm_repl_slink_type blockdev_type = {
+ .type.name = "blockdev",
+ .type.module = THIS_MODULE,
+
+ .ctr = blockdev_ctr,
+ .dtr = blockdev_dtr,
+
+ .postsuspend = blockdev_postsuspend,
+ .resume = blockdev_resume,
+
+ .dev_add = blockdev_dev_add,
+ .dev_del = blockdev_dev_del,
+
+ .copy = blockdev_copy,
+ .io = blockdev_io,
+ .unplug = blockdev_unplug,
+ .recover_notify_fn_set = blockdev_recover_notify_fn_set,
+ .set_sync = blockdev_set_sync,
+ .flush_sync = blockdev_flush_sync,
+ .resync = blockdev_resync,
+ .in_sync = blockdev_in_sync,
+
+ .policy = blockdev_policy,
+ .state = blockdev_state,
+ .fallbehind = blockdev_fallbehind,
+ .dev_number = blockdev_dev_number,
+ .slink_number = blockdev_slink_number,
+ .slink = blockdev_slink,
+
+ .message = blockdev_message,
+ .status = blockdev_status,
+};
+
+/* Destroy kmem caches on module unload. */
+static void
+slink_kmem_caches_exit(void)
+{
+ struct cache_defs *cd = ARRAY_END(cache_defs);
+
+ while (cd-- > cache_defs) {
+ if (cd->cache) {
+ kmem_cache_destroy(cd->cache);
+ cd->cache = NULL;
+ }
+ }
+}
+
+/* Create kmem caches on module load. */
+static int
+slink_kmem_caches_init(void)
+{
+ int r = 0;
+ struct cache_defs *cd = ARRAY_END(cache_defs);
+
+ while (cd-- > cache_defs) {
+ cd->cache = kmem_cache_create(cd->name, cd->size, 0, 0, NULL);
+
+ if (unlikely(!cd->cache)) {
+ DMERR("failed to create %s slab for site link "
+ "handler %s %s",
+ cd->name, blockdev_type.type.name, version);
+ slink_kmem_caches_exit();
+ r = -ENOMEM;
+ break;
+ }
+ }
+
+ return r;
+}
+
+int __init
+dm_repl_slink_init(void)
+{
+ int r;
+
+ /* Create slabs for the copy contexts and test buffers. */
+ r = slink_kmem_caches_init();
+ if (r) {
+ DMERR("failed to init %s kmem caches", blockdev_type.type.name);
+ return r;
+ }
+
+ r = dm_register_type(&blockdev_type, DM_SLINK);
+ if (unlikely(r < 0)) {
+ DMERR("failed to register replication site "
+ "link handler %s %s [%d]",
+ blockdev_type.type.name, version, r);
+ slink_kmem_caches_exit();
+ } else
+ DMINFO("registered replication site link handler %s %s",
+ blockdev_type.type.name, version);
+
+ return r;
+}
+
+void __exit
+dm_repl_slink_exit(void)
+{
+ int r = dm_unregister_type(&blockdev_type, DM_SLINK);
+
+ slink_kmem_caches_exit();
+
+ if (r)
+ DMERR("failed to unregister replication site "
+ "link handler %s %s [%d]",
+ blockdev_type.type.name, version, r);
+ else
+ DMINFO("unregistered replication site link handler %s %s",
+ blockdev_type.type.name, version);
+
+}
+
+/* Module hooks. */
+module_init(dm_repl_slink_init);
+module_exit(dm_repl_slink_exit);
+
+MODULE_DESCRIPTION(DM_NAME " remote replication target \"blockdev\" "
+ "site link (SLINK) handler");
+MODULE_AUTHOR("Heinz Mauelshagen <heinzm@redhat.com>");
+MODULE_LICENSE("GPL");
--
1.6.2.5
^ permalink raw reply related [flat|nested] 9+ messages in thread
* Re: [PATCH v6 1/4] dm-replicator: documentation and module registry
2009-12-18 15:44 ` [PATCH v6 1/4] dm-replicator: documentation and module registry heinzm
2009-12-18 15:44 ` [PATCH v6 2/4] dm-replicator: replication log and site link handler interfaces and main replicator module heinzm
@ 2010-01-07 10:18 ` 张宇
2010-01-08 19:44 ` Heinz Mauelshagen
1 sibling, 1 reply; 9+ messages in thread
From: 张宇 @ 2010-01-07 10:18 UTC (permalink / raw)
To: device-mapper development
[-- Attachment #1.1: Type: text/plain, Size: 23164 bytes --]
Is there any command-line example to explain how to use this patch?
I have compiled it and loaded the modules, the '<start><length>' target
parameters in both replicator ant replicator-dev targets means what?
how can I construct these target?
I haven't read the source code in detail till now, sorry.
2009/12/18 <heinzm@redhat.com>
> From: Heinz Mauelshagen <heinzm@redhat.com>
>
> The dm-registry module is a general purpose registry for modules.
>
> The remote replicator utilizes it to register its ringbuffer log and
> site link handlers in order to avoid duplicating registry code and logic.
>
>
> Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
> Reviewed-by: Jon Brassow <jbrassow@redhat.com>
> Tested-by: Jon Brassow <jbrassow@redhat.com>
> ---
> Documentation/device-mapper/replicator.txt | 203
> +++++++++++++++++++++++++
> drivers/md/Kconfig | 8 +
> drivers/md/Makefile | 1 +
> drivers/md/dm-registry.c | 224
> ++++++++++++++++++++++++++++
> drivers/md/dm-registry.h | 38 +++++
> 5 files changed, 474 insertions(+), 0 deletions(-)
> create mode 100644 Documentation/device-mapper/replicator.txt
> create mode 100644 drivers/md/dm-registry.c
> create mode 100644 drivers/md/dm-registry.h
>
> diff --git 2.6.33-rc1.orig/Documentation/device-mapper/replicator.txt
> 2.6.33-rc1/Documentation/device-mapper/replicator.txt
> new file mode 100644
> index 0000000..1d408a6
> --- /dev/null
> +++ 2.6.33-rc1/Documentation/device-mapper/replicator.txt
> @@ -0,0 +1,203 @@
> +dm-replicator
> +=============
> +
> +Device-mapper replicator is designed to enable redundant copies of
> +storage devices to be made - preferentially, to remote locations.
> +RAID1 (aka mirroring) is often used to maintain redundant copies of
> +storage for fault tolerance purposes. Unlike RAID1, which often
> +assumes similar device characteristics, dm-replicator is designed to
> +handle devices with different latency and bandwidth characteristics
> +which are often the result of the geograhic disparity of multi-site
> +architectures. Simply put, you might choose RAID1 to protect from
> +a single device failure, but you would choose remote replication
> +via dm-replicator for protection against a site failure.
> +
> +dm-replicator works by first sending write requests to the "replicator
> +log". Not to be confused with the device-mapper dirty log, this
> +replicator log behaves similarly to that of a journal. Write requests
> +go to this log first and then are copied to all the replicate devices
> +at their various locations. Requests are cleared from the log once all
> +replicate devices confirm the data is received/copied. This architecture
> +allows dm-replicator to be flexible in terms of device characteristics.
> +If one device should fall behind the others - perhaps due to high latency
> -
> +the slack is picked up by the log. The user has a great deal of
> +flexibility in specifying to what degree a particular site is allowed to
> +fall behind - if at all.
> +
> +Device-Mapper's dm-replicator has two targets, "replicator" and
> +"replicator-dev". The "replicator" target is used to setup the
> +aforementioned log and allow the specification of site link properties.
> +Through the "replicator" target, the user might specify that writes
> +that are copied to the local site must happen synchronously (i.e the
> +writes are complete only after they have passed through the log device
> +and have landed on the local site's disk). They may also specify that
> +a remote link should asynchronously complete writes, but that the remote
> +link should never fall more than 100MB behind in terms of processing.
> +Again, the "replicator" target is used to define the replicator log and
> +the characteristics of each site link.
> +
> +The "replicator-dev" target is used to define the devices used and
> +associate them with a particular replicator log. You might think of
> +this stage in a similar way to setting up RAID1 (mirroring). You
> +define a set of devices which will be copies of each other, but
> +access the device through the mirror virtual device which takes care
> +of the copying. The user accessible replicator device is analogous
> +to the mirror virtual device, while the set of devices being copied
> +to are analogous to the mirror images (sometimes called 'legs').
> +When creating a replicator device via the "replicator-dev" target,
> +it must be associated with the replicator log (created with the
> +aforementioned "replicator" target). When each redundant device
> +is specified as part of the replicator device, it is associated with
> +a site link whose properties were defined when the "replicator"
> +target was created.
> +
> +The user can go farther than simply replicating one device. They
> +can continue to add replicator devices - associating them with a
> +particular replicator log. Writes that go through the replicator
> +log are guarenteed to have their write ordering preserved. So, if
> +you associate more than one replicator device to a particular
> +replicator log, you are preserving write ordering across multiple
> +devices. This might be useful if you had a database that spanned
> +multiple disks and write ordering must be preserved or any transaction
> +accounting scheme would be foiled. (You can imagine this like
> +preserving write ordering across a number of mirrored devices, where
> +each mirror has images/legs in different geographic locations.)
> +
> +dm-replicator has a modular architecture. Future implementations for
> +the replicator log and site link modules are allowed. The current
> +replication log is ringbuffer - utilized to store all writes being
> +subject to replication and enforce write ordering. The current site
> +link code is based on accessing block devices (iSCSI, FC, etc) and
> +does device recovery including (initial) resynchronization.
> +
> +
> +Picture of a 2 site configuration with 3 local devices (LDs) in a
> +primary site being resycnhronied to 3 remotes sites with 3 remote
> +devices (RDs) each via site links (SLINK) 1-2 with site link 0
> +as a special case to handle the local devices:
> +
> + |
> + Local (primary) site | Remote sites
> + -------------------- | ------------
> + |
> + D1 D2 Dn |
> + | | | |
> + +---+- ... -+ |
> + | |
> + REPLOG-----------------+- SLINK1 ------------+
> + | | | |
> + SLINK0 (special case) | | |
> + | | | |
> + +-----+ ... + | | +----+- ... -+
> + | | | | | | | |
> + LD1 LD2 LDn | | RD1 RD2 RDn
> + | |
> + +-- SLINK2------------+
> + | | |
> + | | +----+- ... -+
> + | | | | |
> + | | RD1 RD2 RDn
> + | |
> + | |
> + | |
> + +- SLINKm ------------+
> + | |
> + | +----+- ... -+
> + | | | |
> + | RD1 RD2 RDn
> +
> +
> +
> +
> +The following are descriptions of the device-mapper tables used to
> +construct the "replicator" and "replicator-dev" targets.
> +
> +"replicator" target parameters:
> +-------------------------------
> +<start> <length> replicator \
> + <replog_type> <#replog_params> <replog_params> \
> + [<slink_type_0> <#slink_params_0> <slink_params_0>]{1..N}
> +
> +<replog_type> = "ringbuffer" is currently the only available type
> +<#replog_params> = # of args following this one intended for the replog (2
> or 4)
> +<replog_params> = <dev_path> <dev_start> [auto/create/open <size>]
> + <dev_path> = device path of replication log (REPLOG) backing store
> + <dev_start> = offset to REPLOG header
> + create = The replication log will be initialized if not active
> + and sized to "size". (If already active, the create
> + will fail.) Size is always in sectors.
> + open = The replication log must be initialized and valid or
> + the constructor will fail.
> + auto = If a valid replication log header is found on the
> + replication device, this will behave like 'open'.
> + Otherwise, this option behaves like 'create'.
> +
> +<slink_type> = "blockdev" is currently the only available type
> +<#slink_params> = 1/2/4
> +<slink_params> = <slink_nr> [<slink_policy> [<fall_behind> <N>]]
> + <slink_nr> = This is a unique number that is used to identify a
> + particular site/location. '0' is always used to
> + identify the local site, while increasing integers
> + are used to identify remote sites.
> + <slink_policy> = The policy can be either 'sync' or 'async'.
> + 'sync' means write requests will not return until
> + the data is on the storage device. 'async' allows
> + a device to "fall behind"; that is, outstanding
> + write requests are waiting in the replication log
> + to be processed for this site, but it is not
> delaying
> + the writes of other sites.
> + <fall_behind> = This field is used to specify how far the user is
> + willing to allow write requests to this specific
> site
> + to "fall behind" in processing before switching to
> + a 'sync' policy. This "fall behind" threshhold
> can
> + be specified in three ways: ios, size, or timeout.
> + 'ios' is the number of pending I/Os allowed (e.g.
> + "ios 10000"). 'size' is the amount of pending
> data
> + allowed (e.g. "size 200m"). Size labels include:
> + s (sectors), k, m, g, t, p, and e. 'timeout' is
> + the amount of time allowed for writes to be
> + outstanding. Time labels include: s, m, h, and d.
> +
> +
> +"replicator-dev" target parameters:
> +-----------------------------------
> +start> <length> replicator-dev
> + <replicator_device> <dev_nr> \
> + [<slink_nr> <#dev_params> <dev_params>
> + <dlog_type> <#dlog_params> <dlog_params>]{1..N}
> +
> +<replicator_device> = device previously constructed via "replication"
> target
> +<dev_nr> = An integer that is used to 'tag' write requests as
> + belonging to a particular set of devices -
> specifically,
> + the devices that follow this argument (i.e. the site
> + link devices).
> +<slink_nr> = This number identifies the site/location where the
> next
> + device to be specified comes from. It is exactly the
> + same number used to identify the site/location (and
> its
> + policies) in the "replicator" target. Interestingly,
> + while one might normally expect a "dev_type" argument
> + here, it can be deduced from the site link number and
> + the 'slink_type' given in the "replication" target.
> +<#dev_params> = '1' (The number of allowed parameters actually
> depends
> + on the 'slink_type' given in the "replication"
> target.
> + Since our only option there is "blockdev", the only
> + allowable number here is currently '1'.)
> +<dev_params> = 'dev_path' (Again, since "blockdev" is the only
> + 'slink_type' available, the only allowable argument
> here
> + is the path to the device.)
> +<dlog_type> = Not to be confused with the "replicator log", this is
> + the type of dirty log associated with this particular
> + device. Dirty logs are used for synchronization,
> during
> + initialization or fall behind conditions, to bring
> devices
> + into a coherent state with its peers - analogous to
> + rebuilding a RAID1 (mirror) device. Available dirty
> + log types include: 'nolog', 'core', and 'disk'
> +<#dlog_params> = The number of arguments required for a particular log
> + type - 'nolog' = 0, 'core' = 1/2, 'disk' = 2/3.
> +<dlog_params> = 'nolog' => ~no arguments~
> + 'core' => <region_size> [sync | nosync]
> + 'disk' => <dlog_dev_path> <region_size> [sync |
> nosync]
> + <region_size> = This sets the granularity at which the dirty log
> + tracks what areas of the device is in-sync.
> + [sync | nosync] = Optionally specify whether the sync should be
> forced
> + or avoided initially.
> diff --git 2.6.33-rc1.orig/drivers/md/Kconfig 2.6.33-rc1/drivers/md/Kconfig
> index acb3a4e..62c9766 100644
> --- 2.6.33-rc1.orig/drivers/md/Kconfig
> +++ 2.6.33-rc1/drivers/md/Kconfig
> @@ -313,6 +313,14 @@ config DM_DELAY
>
> If unsure, say N.
>
> +config DM_REPLICATOR
> + tristate "Replication target (EXPERIMENTAL)"
> + depends on BLK_DEV_DM && EXPERIMENTAL
> + ---help---
> + A target that supports replication of local devices to remote
> sites.
> +
> + If unsure, say N.
> +
> config DM_UEVENT
> bool "DM uevents (EXPERIMENTAL)"
> depends on BLK_DEV_DM && EXPERIMENTAL
> diff --git 2.6.33-rc1.orig/drivers/md/Makefile
> 2.6.33-rc1/drivers/md/Makefile
> index e355e7f..be05b39 100644
> --- 2.6.33-rc1.orig/drivers/md/Makefile
> +++ 2.6.33-rc1/drivers/md/Makefile
> @@ -44,6 +44,7 @@ obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
> obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o
> dm-region-hash.o
> obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
> obj-$(CONFIG_DM_ZERO) += dm-zero.o
> +obj-$(CONFIG_DM_REPLICATOR) += dm-log.o dm-registry.o
>
> quiet_cmd_unroll = UNROLL $@
> cmd_unroll = $(AWK) -f$(srctree)/$(src)/unroll.awk -vN=$(UNROLL) \
> diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.c
> 2.6.33-rc1/drivers/md/dm-registry.c
> new file mode 100644
> index 0000000..fb8abbf
> --- /dev/null
> +++ 2.6.33-rc1/drivers/md/dm-registry.c
> @@ -0,0 +1,224 @@
> +/*
> + * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
> + *
> + * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
> + *
> + * Generic registry for arbitrary structures
> + * (needs dm_registry_type structure upfront each registered structure).
> + *
> + * This file is released under the GPL.
> + *
> + * FIXME: use as registry for e.g. dirty log types as well.
> + */
> +
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/moduleparam.h>
> +
> +#include "dm-registry.h"
> +
> +#define DM_MSG_PREFIX "dm-registry"
> +
> +static const char *version = "0.001";
> +
> +/* Sizable class registry. */
> +static unsigned num_classes;
> +static struct list_head *_classes;
> +static rwlock_t *_locks;
> +
> +void *
> +dm_get_type(const char *type_name, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t;
> +
> + read_lock(_locks + class);
> + list_for_each_entry(t, _classes + class, list) {
> + if (!strcmp(type_name, t->name)) {
> + if (!t->use_count && !try_module_get(t->module)) {
> + read_unlock(_locks + class);
> + return ERR_PTR(-ENOMEM);
> + }
> +
> + t->use_count++;
> + read_unlock(_locks + class);
> + return t;
> + }
> + }
> +
> + read_unlock(_locks + class);
> + return ERR_PTR(-ENOENT);
> +}
> +EXPORT_SYMBOL(dm_get_type);
> +
> +void
> +dm_put_type(void *type, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t = type;
> +
> + read_lock(_locks + class);
> + if (!--t->use_count)
> + module_put(t->module);
> +
> + read_unlock(_locks + class);
> +}
> +EXPORT_SYMBOL(dm_put_type);
> +
> +/* Add a type to the registry. */
> +int
> +dm_register_type(void *type, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t = type, *tt;
> +
> + if (unlikely(class >= num_classes))
> + return -EINVAL;
> +
> + tt = dm_get_type(t->name, class);
> + if (unlikely(!IS_ERR(tt))) {
> + dm_put_type(t, class);
> + return -EEXIST;
> + }
> +
> + write_lock(_locks + class);
> + t->use_count = 0;
> + list_add(&t->list, _classes + class);
> + write_unlock(_locks + class);
> +
> + return 0;
> +}
> +EXPORT_SYMBOL(dm_register_type);
> +
> +/* Remove a type from the registry. */
> +int
> +dm_unregister_type(void *type, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t = type;
> +
> + if (unlikely(class >= num_classes)) {
> + DMERR("Attempt to unregister invalid class");
> + return -EINVAL;
> + }
> +
> + write_lock(_locks + class);
> +
> + if (unlikely(t->use_count)) {
> + write_unlock(_locks + class);
> + DMWARN("Attempt to unregister a type that is still in
> use");
> + return -ETXTBSY;
> + } else
> + list_del(&t->list);
> +
> + write_unlock(_locks + class);
> + return 0;
> +}
> +EXPORT_SYMBOL(dm_unregister_type);
> +
> +/*
> + * Return kmalloc'ed NULL terminated pointer
> + * array of all type names of the given class.
> + *
> + * Caller has to kfree the array!.
> + */
> +const char **dm_types_list(enum dm_registry_class class)
> +{
> + unsigned i = 0, count = 0;
> + const char **r;
> + struct dm_registry_type *t;
> +
> + /* First count the registered types in the class. */
> + read_lock(_locks + class);
> + list_for_each_entry(t, _classes + class, list)
> + count++;
> + read_unlock(_locks + class);
> +
> + /* None registered in this class. */
> + if (!count)
> + return NULL;
> +
> + /* One member more for array NULL termination. */
> + r = kzalloc((count + 1) * sizeof(*r), GFP_KERNEL);
> + if (!r)
> + return ERR_PTR(-ENOMEM);
> +
> + /*
> + * Go with the counted ones.
> + * Any new added ones after we counted will be ignored!
> + */
> + read_lock(_locks + class);
> + list_for_each_entry(t, _classes + class, list) {
> + r[i++] = t->name;
> + if (!--count)
> + break;
> + }
> + read_unlock(_locks + class);
> +
> + return r;
> +}
> +EXPORT_SYMBOL(dm_types_list);
> +
> +int __init
> +dm_registry_init(void)
> +{
> + unsigned n;
> +
> + BUG_ON(_classes);
> + BUG_ON(_locks);
> +
> + /* Module parameter given ? */
> + if (!num_classes)
> + num_classes = DM_REGISTRY_CLASS_END;
> +
> + n = num_classes;
> + _classes = kmalloc(n * sizeof(*_classes), GFP_KERNEL);
> + if (!_classes) {
> + DMERR("Failed to allocate classes registry");
> + return -ENOMEM;
> + }
> +
> + _locks = kmalloc(n * sizeof(*_locks), GFP_KERNEL);
> + if (!_locks) {
> + DMERR("Failed to allocate classes locks");
> + kfree(_classes);
> + _classes = NULL;
> + return -ENOMEM;
> + }
> +
> + while (n--) {
> + INIT_LIST_HEAD(_classes + n);
> + rwlock_init(_locks + n);
> + }
> +
> + DMINFO("initialized %s for max %u classes", version, num_classes);
> + return 0;
> +}
> +
> +void __exit
> +dm_registry_exit(void)
> +{
> + BUG_ON(!_classes);
> + BUG_ON(!_locks);
> +
> + kfree(_classes);
> + _classes = NULL;
> + kfree(_locks);
> + _locks = NULL;
> + DMINFO("exit %s", version);
> +}
> +
> +/* Module hooks */
> +module_init(dm_registry_init);
> +module_exit(dm_registry_exit);
> +module_param(num_classes, uint, 0);
> +MODULE_PARM_DESC(num_classes, "Maximum number of classes");
> +MODULE_DESCRIPTION(DM_NAME "device-mapper registry");
> +MODULE_AUTHOR("Heinz Mauelshagen <heinzm@redhat.com>");
> +MODULE_LICENSE("GPL");
> +
> +#ifndef MODULE
> +static int __init num_classes_setup(char *str)
> +{
> + num_classes = simple_strtol(str, NULL, 0);
> + return num_classes ? 1 : 0;
> +}
> +
> +__setup("num_classes=", num_classes_setup);
> +#endif
> diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.h
> 2.6.33-rc1/drivers/md/dm-registry.h
> new file mode 100644
> index 0000000..1cb0ce8
> --- /dev/null
> +++ 2.6.33-rc1/drivers/md/dm-registry.h
> @@ -0,0 +1,38 @@
> +/*
> + * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
> + *
> + * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
> + *
> + * Generic registry for arbitrary structures.
> + * (needs dm_registry_type structure upfront each registered structure).
> + *
> + * This file is released under the GPL.
> + */
> +
> +#include "dm.h"
> +
> +#ifndef DM_REGISTRY_H
> +#define DM_REGISTRY_H
> +
> +enum dm_registry_class {
> + DM_REPLOG = 0,
> + DM_SLINK,
> + DM_LOG,
> + DM_REGION_HASH,
> + DM_REGISTRY_CLASS_END,
> +};
> +
> +struct dm_registry_type {
> + struct list_head list; /* Linked list of types in this class. */
> + const char *name;
> + struct module *module;
> + unsigned int use_count;
> +};
> +
> +void *dm_get_type(const char *type_name, enum dm_registry_class class);
> +void dm_put_type(void *type, enum dm_registry_class class);
> +int dm_register_type(void *type, enum dm_registry_class class);
> +int dm_unregister_type(void *type, enum dm_registry_class class);
> +const char **dm_types_list(enum dm_registry_class class);
> +
> +#endif
> --
> 1.6.2.5
>
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel
>
[-- Attachment #1.2: Type: text/html, Size: 25722 bytes --]
[-- Attachment #2: Type: text/plain, Size: 0 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v6 1/4] dm-replicator: documentation and module registry
2010-01-07 10:18 ` [PATCH v6 1/4] dm-replicator: documentation and module registry 张宇
@ 2010-01-08 19:44 ` Heinz Mauelshagen
2010-02-09 1:48 ` Busby
0 siblings, 1 reply; 9+ messages in thread
From: Heinz Mauelshagen @ 2010-01-08 19:44 UTC (permalink / raw)
To: device-mapper development
On Thu, 2010-01-07 at 18:18 +0800, 张宇 wrote:
> Is there any command-line example to explain how to use this patch?
Look at Documentation/device-mapper/replicator.txt and at comments above
replicator_ctr() and _replicator_dev_ctr() in dm-repl.c for mapping
table syntax and examples.
> I have compiled it and loaded the modules, the '<start><length>'
> target parameters in both replicator ant replicator-dev targets means
> what?
replicator doesn't provide a direct mapping of its own, so <length> is
arbitrary and <start> shal be 0.
> how can I construct these target?
See documentation hints above.
> I haven't read the source code in detail till now, sorry.
You will now ;-)
Regards,
Heinz
>
>
> 2009/12/18 <heinzm@redhat.com>
> From: Heinz Mauelshagen <heinzm@redhat.com>
>
> The dm-registry module is a general purpose registry for
> modules.
>
> The remote replicator utilizes it to register its ringbuffer
> log and
> site link handlers in order to avoid duplicating registry code
> and logic.
>
>
> Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
> Reviewed-by: Jon Brassow <jbrassow@redhat.com>
> Tested-by: Jon Brassow <jbrassow@redhat.com>
> ---
> Documentation/device-mapper/replicator.txt | 203
> +++++++++++++++++++++++++
> drivers/md/Kconfig | 8 +
> drivers/md/Makefile | 1 +
> drivers/md/dm-registry.c | 224
> ++++++++++++++++++++++++++++
> drivers/md/dm-registry.h | 38 +++++
> 5 files changed, 474 insertions(+), 0 deletions(-)
> create mode 100644 Documentation/device-mapper/replicator.txt
> create mode 100644 drivers/md/dm-registry.c
> create mode 100644 drivers/md/dm-registry.h
>
> diff --git
> 2.6.33-rc1.orig/Documentation/device-mapper/replicator.txt
> 2.6.33-rc1/Documentation/device-mapper/replicator.txt
> new file mode 100644
> index 0000000..1d408a6
> --- /dev/null
> +++ 2.6.33-rc1/Documentation/device-mapper/replicator.txt
> @@ -0,0 +1,203 @@
> +dm-replicator
> +=============
> +
> +Device-mapper replicator is designed to enable redundant
> copies of
> +storage devices to be made - preferentially, to remote
> locations.
> +RAID1 (aka mirroring) is often used to maintain redundant
> copies of
> +storage for fault tolerance purposes. Unlike RAID1, which
> often
> +assumes similar device characteristics, dm-replicator is
> designed to
> +handle devices with different latency and bandwidth
> characteristics
> +which are often the result of the geograhic disparity of
> multi-site
> +architectures. Simply put, you might choose RAID1 to protect
> from
> +a single device failure, but you would choose remote
> replication
> +via dm-replicator for protection against a site failure.
> +
> +dm-replicator works by first sending write requests to the
> "replicator
> +log". Not to be confused with the device-mapper dirty log,
> this
> +replicator log behaves similarly to that of a journal. Write
> requests
> +go to this log first and then are copied to all the replicate
> devices
> +at their various locations. Requests are cleared from the
> log once all
> +replicate devices confirm the data is received/copied. This
> architecture
> +allows dm-replicator to be flexible in terms of device
> characteristics.
> +If one device should fall behind the others - perhaps due to
> high latency -
> +the slack is picked up by the log. The user has a great deal
> of
> +flexibility in specifying to what degree a particular site is
> allowed to
> +fall behind - if at all.
> +
> +Device-Mapper's dm-replicator has two targets, "replicator"
> and
> +"replicator-dev". The "replicator" target is used to setup
> the
> +aforementioned log and allow the specification of site link
> properties.
> +Through the "replicator" target, the user might specify that
> writes
> +that are copied to the local site must happen synchronously
> (i.e the
> +writes are complete only after they have passed through the
> log device
> +and have landed on the local site's disk). They may also
> specify that
> +a remote link should asynchronously complete writes, but that
> the remote
> +link should never fall more than 100MB behind in terms of
> processing.
> +Again, the "replicator" target is used to define the
> replicator log and
> +the characteristics of each site link.
> +
> +The "replicator-dev" target is used to define the devices
> used and
> +associate them with a particular replicator log. You might
> think of
> +this stage in a similar way to setting up RAID1 (mirroring).
> You
> +define a set of devices which will be copies of each other,
> but
> +access the device through the mirror virtual device which
> takes care
> +of the copying. The user accessible replicator device is
> analogous
> +to the mirror virtual device, while the set of devices being
> copied
> +to are analogous to the mirror images (sometimes called
> 'legs').
> +When creating a replicator device via the "replicator-dev"
> target,
> +it must be associated with the replicator log (created with
> the
> +aforementioned "replicator" target). When each redundant
> device
> +is specified as part of the replicator device, it is
> associated with
> +a site link whose properties were defined when the
> "replicator"
> +target was created.
> +
> +The user can go farther than simply replicating one device.
> They
> +can continue to add replicator devices - associating them
> with a
> +particular replicator log. Writes that go through the
> replicator
> +log are guarenteed to have their write ordering preserved.
> So, if
> +you associate more than one replicator device to a particular
> +replicator log, you are preserving write ordering across
> multiple
> +devices. This might be useful if you had a database that
> spanned
> +multiple disks and write ordering must be preserved or any
> transaction
> +accounting scheme would be foiled. (You can imagine this
> like
> +preserving write ordering across a number of mirrored
> devices, where
> +each mirror has images/legs in different geographic
> locations.)
> +
> +dm-replicator has a modular architecture. Future
> implementations for
> +the replicator log and site link modules are allowed. The
> current
> +replication log is ringbuffer - utilized to store all writes
> being
> +subject to replication and enforce write ordering. The
> current site
> +link code is based on accessing block devices (iSCSI, FC,
> etc) and
> +does device recovery including (initial) resynchronization.
> +
> +
> +Picture of a 2 site configuration with 3 local devices (LDs)
> in a
> +primary site being resycnhronied to 3 remotes sites with 3
> remote
> +devices (RDs) each via site links (SLINK) 1-2 with site link
> 0
> +as a special case to handle the local devices:
> +
> + |
> + Local (primary) site | Remote
> sites
> + -------------------- |
> ------------
> + |
> + D1 D2 Dn |
> + | | | |
> + +---+- ... -+ |
> + | |
> + REPLOG-----------------+- SLINK1 ------------+
> + | | | |
> + SLINK0 (special case) | | |
> + | | | |
> + +-----+ ... + | | +----+- ... -+
> + | | | | | | | |
> + LD1 LD2 LDn | | RD1 RD2
> RDn
> + | |
> + +-- SLINK2------------+
> + | | |
> + | | +----+- ... -+
> + | | | | |
> + | | RD1 RD2
> RDn
> + | |
> + | |
> + | |
> + +- SLINKm ------------+
> + | |
> + | +----+- ... -+
> + | | | |
> + | RD1 RD2
> RDn
> +
> +
> +
> +
> +The following are descriptions of the device-mapper tables
> used to
> +construct the "replicator" and "replicator-dev" targets.
> +
> +"replicator" target parameters:
> +-------------------------------
> +<start> <length> replicator \
> + <replog_type> <#replog_params> <replog_params> \
> + [<slink_type_0> <#slink_params_0>
> <slink_params_0>]{1..N}
> +
> +<replog_type> = "ringbuffer" is currently the only
> available type
> +<#replog_params> = # of args following this one intended for
> the replog (2 or 4)
> +<replog_params> = <dev_path> <dev_start> [auto/create/open
> <size>]
> + <dev_path> = device path of replication log (REPLOG)
> backing store
> + <dev_start> = offset to REPLOG header
> + create = The replication log will be initialized
> if not active
> + and sized to "size". (If already
> active, the create
> + will fail.) Size is always in sectors.
> + open = The replication log must be initialized
> and valid or
> + the constructor will fail.
> + auto = If a valid replication log header is
> found on the
> + replication device, this will behave
> like 'open'.
> + Otherwise, this option behaves like
> 'create'.
> +
> +<slink_type> = "blockdev" is currently the only available
> type
> +<#slink_params> = 1/2/4
> +<slink_params> = <slink_nr> [<slink_policy> [<fall_behind>
> <N>]]
> + <slink_nr> = This is a unique number that is used
> to identify a
> + particular site/location. '0' is
> always used to
> + identify the local site, while
> increasing integers
> + are used to identify remote sites.
> + <slink_policy> = The policy can be either 'sync' or
> 'async'.
> + 'sync' means write requests will not
> return until
> + the data is on the storage device.
> 'async' allows
> + a device to "fall behind"; that is,
> outstanding
> + write requests are waiting in the
> replication log
> + to be processed for this site, but it
> is not delaying
> + the writes of other sites.
> + <fall_behind> = This field is used to specify how far
> the user is
> + willing to allow write requests to
> this specific site
> + to "fall behind" in processing before
> switching to
> + a 'sync' policy. This "fall behind"
> threshhold can
> + be specified in three ways: ios,
> size, or timeout.
> + 'ios' is the number of pending I/Os
> allowed (e.g.
> + "ios 10000"). 'size' is the amount
> of pending data
> + allowed (e.g. "size 200m"). Size
> labels include:
> + s (sectors), k, m, g, t, p, and e.
> 'timeout' is
> + the amount of time allowed for writes
> to be
> + outstanding. Time labels include: s,
> m, h, and d.
> +
> +
> +"replicator-dev" target parameters:
> +-----------------------------------
> +start> <length> replicator-dev
> + <replicator_device> <dev_nr> \
> + [<slink_nr> <#dev_params> <dev_params>
> + <dlog_type> <#dlog_params> <dlog_params>]{1..N}
> +
> +<replicator_device> = device previously constructed via
> "replication" target
> +<dev_nr> = An integer that is used to 'tag' write
> requests as
> + belonging to a particular set of devices
> - specifically,
> + the devices that follow this argument
> (i.e. the site
> + link devices).
> +<slink_nr> = This number identifies the site/location
> where the next
> + device to be specified comes from. It
> is exactly the
> + same number used to identify the
> site/location (and its
> + policies) in the "replicator" target.
> Interestingly,
> + while one might normally expect a
> "dev_type" argument
> + here, it can be deduced from the site
> link number and
> + the 'slink_type' given in the
> "replication" target.
> +<#dev_params> = '1' (The number of allowed parameters
> actually depends
> + on the 'slink_type' given in the
> "replication" target.
> + Since our only option there is
> "blockdev", the only
> + allowable number here is currently '1'.)
> +<dev_params> = 'dev_path' (Again, since "blockdev" is
> the only
> + 'slink_type' available, the only
> allowable argument here
> + is the path to the device.)
> +<dlog_type> = Not to be confused with the "replicator
> log", this is
> + the type of dirty log associated with
> this particular
> + device. Dirty logs are used for
> synchronization, during
> + initialization or fall behind
> conditions, to bring devices
> + into a coherent state with its peers -
> analogous to
> + rebuilding a RAID1 (mirror) device.
> Available dirty
> + log types include: 'nolog', 'core', and
> 'disk'
> +<#dlog_params> = The number of arguments required for a
> particular log
> + type - 'nolog' = 0, 'core' = 1/2, 'disk'
> = 2/3.
> +<dlog_params> = 'nolog' => ~no arguments~
> + 'core' => <region_size> [sync | nosync]
> + 'disk' => <dlog_dev_path> <region_size>
> [sync | nosync]
> + <region_size> = This sets the granularity at which
> the dirty log
> + tracks what areas of the device is
> in-sync.
> + [sync | nosync] = Optionally specify whether the sync
> should be forced
> + or avoided initially.
> diff --git 2.6.33-rc1.orig/drivers/md/Kconfig
> 2.6.33-rc1/drivers/md/Kconfig
> index acb3a4e..62c9766 100644
> --- 2.6.33-rc1.orig/drivers/md/Kconfig
> +++ 2.6.33-rc1/drivers/md/Kconfig
> @@ -313,6 +313,14 @@ config DM_DELAY
>
> If unsure, say N.
>
> +config DM_REPLICATOR
> + tristate "Replication target (EXPERIMENTAL)"
> + depends on BLK_DEV_DM && EXPERIMENTAL
> + ---help---
> + A target that supports replication of local devices to
> remote sites.
> +
> + If unsure, say N.
> +
> config DM_UEVENT
> bool "DM uevents (EXPERIMENTAL)"
> depends on BLK_DEV_DM && EXPERIMENTAL
> diff --git 2.6.33-rc1.orig/drivers/md/Makefile
> 2.6.33-rc1/drivers/md/Makefile
> index e355e7f..be05b39 100644
> --- 2.6.33-rc1.orig/drivers/md/Makefile
> +++ 2.6.33-rc1/drivers/md/Makefile
> @@ -44,6 +44,7 @@ obj-$(CONFIG_DM_SNAPSHOT) +=
> dm-snapshot.o
> obj-$(CONFIG_DM_MIRROR) += dm-mirror.o
> dm-log.o dm-region-hash.o
> obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
> obj-$(CONFIG_DM_ZERO) += dm-zero.o
> +obj-$(CONFIG_DM_REPLICATOR) += dm-log.o dm-registry.o
>
> quiet_cmd_unroll = UNROLL $@
> cmd_unroll = $(AWK) -f$(srctree)/$(src)/unroll.awk -vN=
> $(UNROLL) \
> diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.c
> 2.6.33-rc1/drivers/md/dm-registry.c
> new file mode 100644
> index 0000000..fb8abbf
> --- /dev/null
> +++ 2.6.33-rc1/drivers/md/dm-registry.c
> @@ -0,0 +1,224 @@
> +/*
> + * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
> + *
> + * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
> + *
> + * Generic registry for arbitrary structures
> + * (needs dm_registry_type structure upfront each registered
> structure).
> + *
> + * This file is released under the GPL.
> + *
> + * FIXME: use as registry for e.g. dirty log types as well.
> + */
> +
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/moduleparam.h>
> +
> +#include "dm-registry.h"
> +
> +#define DM_MSG_PREFIX "dm-registry"
> +
> +static const char *version = "0.001";
> +
> +/* Sizable class registry. */
> +static unsigned num_classes;
> +static struct list_head *_classes;
> +static rwlock_t *_locks;
> +
> +void *
> +dm_get_type(const char *type_name, enum dm_registry_class
> class)
> +{
> + struct dm_registry_type *t;
> +
> + read_lock(_locks + class);
> + list_for_each_entry(t, _classes + class, list) {
> + if (!strcmp(type_name, t->name)) {
> + if (!t->use_count && !
> try_module_get(t->module)) {
> + read_unlock(_locks + class);
> + return ERR_PTR(-ENOMEM);
> + }
> +
> + t->use_count++;
> + read_unlock(_locks + class);
> + return t;
> + }
> + }
> +
> + read_unlock(_locks + class);
> + return ERR_PTR(-ENOENT);
> +}
> +EXPORT_SYMBOL(dm_get_type);
> +
> +void
> +dm_put_type(void *type, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t = type;
> +
> + read_lock(_locks + class);
> + if (!--t->use_count)
> + module_put(t->module);
> +
> + read_unlock(_locks + class);
> +}
> +EXPORT_SYMBOL(dm_put_type);
> +
> +/* Add a type to the registry. */
> +int
> +dm_register_type(void *type, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t = type, *tt;
> +
> + if (unlikely(class >= num_classes))
> + return -EINVAL;
> +
> + tt = dm_get_type(t->name, class);
> + if (unlikely(!IS_ERR(tt))) {
> + dm_put_type(t, class);
> + return -EEXIST;
> + }
> +
> + write_lock(_locks + class);
> + t->use_count = 0;
> + list_add(&t->list, _classes + class);
> + write_unlock(_locks + class);
> +
> + return 0;
> +}
> +EXPORT_SYMBOL(dm_register_type);
> +
> +/* Remove a type from the registry. */
> +int
> +dm_unregister_type(void *type, enum dm_registry_class class)
> +{
> + struct dm_registry_type *t = type;
> +
> + if (unlikely(class >= num_classes)) {
> + DMERR("Attempt to unregister invalid class");
> + return -EINVAL;
> + }
> +
> + write_lock(_locks + class);
> +
> + if (unlikely(t->use_count)) {
> + write_unlock(_locks + class);
> + DMWARN("Attempt to unregister a type that is
> still in use");
> + return -ETXTBSY;
> + } else
> + list_del(&t->list);
> +
> + write_unlock(_locks + class);
> + return 0;
> +}
> +EXPORT_SYMBOL(dm_unregister_type);
> +
> +/*
> + * Return kmalloc'ed NULL terminated pointer
> + * array of all type names of the given class.
> + *
> + * Caller has to kfree the array!.
> + */
> +const char **dm_types_list(enum dm_registry_class class)
> +{
> + unsigned i = 0, count = 0;
> + const char **r;
> + struct dm_registry_type *t;
> +
> + /* First count the registered types in the class. */
> + read_lock(_locks + class);
> + list_for_each_entry(t, _classes + class, list)
> + count++;
> + read_unlock(_locks + class);
> +
> + /* None registered in this class. */
> + if (!count)
> + return NULL;
> +
> + /* One member more for array NULL termination. */
> + r = kzalloc((count + 1) * sizeof(*r), GFP_KERNEL);
> + if (!r)
> + return ERR_PTR(-ENOMEM);
> +
> + /*
> + * Go with the counted ones.
> + * Any new added ones after we counted will be
> ignored!
> + */
> + read_lock(_locks + class);
> + list_for_each_entry(t, _classes + class, list) {
> + r[i++] = t->name;
> + if (!--count)
> + break;
> + }
> + read_unlock(_locks + class);
> +
> + return r;
> +}
> +EXPORT_SYMBOL(dm_types_list);
> +
> +int __init
> +dm_registry_init(void)
> +{
> + unsigned n;
> +
> + BUG_ON(_classes);
> + BUG_ON(_locks);
> +
> + /* Module parameter given ? */
> + if (!num_classes)
> + num_classes = DM_REGISTRY_CLASS_END;
> +
> + n = num_classes;
> + _classes = kmalloc(n * sizeof(*_classes), GFP_KERNEL);
> + if (!_classes) {
> + DMERR("Failed to allocate classes registry");
> + return -ENOMEM;
> + }
> +
> + _locks = kmalloc(n * sizeof(*_locks), GFP_KERNEL);
> + if (!_locks) {
> + DMERR("Failed to allocate classes locks");
> + kfree(_classes);
> + _classes = NULL;
> + return -ENOMEM;
> + }
> +
> + while (n--) {
> + INIT_LIST_HEAD(_classes + n);
> + rwlock_init(_locks + n);
> + }
> +
> + DMINFO("initialized %s for max %u classes", version,
> num_classes);
> + return 0;
> +}
> +
> +void __exit
> +dm_registry_exit(void)
> +{
> + BUG_ON(!_classes);
> + BUG_ON(!_locks);
> +
> + kfree(_classes);
> + _classes = NULL;
> + kfree(_locks);
> + _locks = NULL;
> + DMINFO("exit %s", version);
> +}
> +
> +/* Module hooks */
> +module_init(dm_registry_init);
> +module_exit(dm_registry_exit);
> +module_param(num_classes, uint, 0);
> +MODULE_PARM_DESC(num_classes, "Maximum number of classes");
> +MODULE_DESCRIPTION(DM_NAME "device-mapper registry");
> +MODULE_AUTHOR("Heinz Mauelshagen <heinzm@redhat.com>");
> +MODULE_LICENSE("GPL");
> +
> +#ifndef MODULE
> +static int __init num_classes_setup(char *str)
> +{
> + num_classes = simple_strtol(str, NULL, 0);
> + return num_classes ? 1 : 0;
> +}
> +
> +__setup("num_classes=", num_classes_setup);
> +#endif
> diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.h
> 2.6.33-rc1/drivers/md/dm-registry.h
> new file mode 100644
> index 0000000..1cb0ce8
> --- /dev/null
> +++ 2.6.33-rc1/drivers/md/dm-registry.h
> @@ -0,0 +1,38 @@
> +/*
> + * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
> + *
> + * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
> + *
> + * Generic registry for arbitrary structures.
> + * (needs dm_registry_type structure upfront each registered
> structure).
> + *
> + * This file is released under the GPL.
> + */
> +
> +#include "dm.h"
> +
> +#ifndef DM_REGISTRY_H
> +#define DM_REGISTRY_H
> +
> +enum dm_registry_class {
> + DM_REPLOG = 0,
> + DM_SLINK,
> + DM_LOG,
> + DM_REGION_HASH,
> + DM_REGISTRY_CLASS_END,
> +};
> +
> +struct dm_registry_type {
> + struct list_head list; /* Linked list of types in
> this class. */
> + const char *name;
> + struct module *module;
> + unsigned int use_count;
> +};
> +
> +void *dm_get_type(const char *type_name, enum
> dm_registry_class class);
> +void dm_put_type(void *type, enum dm_registry_class class);
> +int dm_register_type(void *type, enum dm_registry_class
> class);
> +int dm_unregister_type(void *type, enum dm_registry_class
> class);
> +const char **dm_types_list(enum dm_registry_class class);
> +
> +#endif
> --
> 1.6.2.5
>
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel
>
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel
--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v6 1/4] dm-replicator: documentation and module registry
2010-01-08 19:44 ` Heinz Mauelshagen
@ 2010-02-09 1:48 ` Busby
0 siblings, 0 replies; 9+ messages in thread
From: Busby @ 2010-02-09 1:48 UTC (permalink / raw)
To: heinzm, device-mapper development
[-- Attachment #1.1: Type: text/plain, Size: 33098 bytes --]
Hi ,
I have some questions about this replicator module after deployed on my
systems:
1.when the connection to the RD failed at the moment reading or writing
data to the D_link, the D_link will fail? can't D_link & LD be well in this
situation? (one LD and one RD, yes, RD can be RDs such as RAID1, however,
all of the RDs disconnected, the D_link also will fail?it will affect on the
user's applications, in gernal speaking, RD can be fail, but the LD (D_link
for user )should be ok at this situation. )
2.Any deals with the data initiation of the LD and RD in this repl
module, not using 'dd' cmds. is there any interface to do it in the
architecture design of this system (roadmap)?
3.While doing the repl, sometime, the RD or LD failed, is there any
function to resync th RD and LD when the LD or RD system come back and then
continue to do the repl?
4.Any design when LD fails, the LD come back, the RD to restore the LD?
not use 'dd' cmd. (as 3rd question said, but this situation just take
attention to the restore repl from the RD.)
I am very interest in this repl module, sorry for asking so many
questions.
DRBD of HA feild is a good tool for HA, but not for repl. Is there
other idea of replication design in Linux system?
Thank you very much.
Regards,
Busby
2010/1/9 Heinz Mauelshagen <heinzm@redhat.com>
> On Thu, 2010-01-07 at 18:18 +0800, 张宇 wrote:
> > Is there any command-line example to explain how to use this patch?
>
> Look at Documentation/device-mapper/replicator.txt and at comments above
> replicator_ctr() and _replicator_dev_ctr() in dm-repl.c for mapping
> table syntax and examples.
>
> > I have compiled it and loaded the modules, the '<start><length>'
> > target parameters in both replicator ant replicator-dev targets means
> > what?
>
> replicator doesn't provide a direct mapping of its own, so <length> is
> arbitrary and <start> shal be 0.
>
> > how can I construct these target?
>
> See documentation hints above.
>
> > I haven't read the source code in detail till now, sorry.
>
> You will now ;-)
>
> Regards,
> Heinz
>
> >
> >
> > 2009/12/18 <heinzm@redhat.com>
> > From: Heinz Mauelshagen <heinzm@redhat.com>
> >
> > The dm-registry module is a general purpose registry for
> > modules.
> >
> > The remote replicator utilizes it to register its ringbuffer
> > log and
> > site link handlers in order to avoid duplicating registry code
> > and logic.
> >
> >
> > Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
> > Reviewed-by: Jon Brassow <jbrassow@redhat.com>
> > Tested-by: Jon Brassow <jbrassow@redhat.com>
> > ---
> > Documentation/device-mapper/replicator.txt | 203
> > +++++++++++++++++++++++++
> > drivers/md/Kconfig | 8 +
> > drivers/md/Makefile | 1 +
> > drivers/md/dm-registry.c | 224
> > ++++++++++++++++++++++++++++
> > drivers/md/dm-registry.h | 38 +++++
> > 5 files changed, 474 insertions(+), 0 deletions(-)
> > create mode 100644 Documentation/device-mapper/replicator.txt
> > create mode 100644 drivers/md/dm-registry.c
> > create mode 100644 drivers/md/dm-registry.h
> >
> > diff --git
> > 2.6.33-rc1.orig/Documentation/device-mapper/replicator.txt
> > 2.6.33-rc1/Documentation/device-mapper/replicator.txt
> > new file mode 100644
> > index 0000000..1d408a6
> > --- /dev/null
> > +++ 2.6.33-rc1/Documentation/device-mapper/replicator.txt
> > @@ -0,0 +1,203 @@
> > +dm-replicator
> > +=============
> > +
> > +Device-mapper replicator is designed to enable redundant
> > copies of
> > +storage devices to be made - preferentially, to remote
> > locations.
> > +RAID1 (aka mirroring) is often used to maintain redundant
> > copies of
> > +storage for fault tolerance purposes. Unlike RAID1, which
> > often
> > +assumes similar device characteristics, dm-replicator is
> > designed to
> > +handle devices with different latency and bandwidth
> > characteristics
> > +which are often the result of the geograhic disparity of
> > multi-site
> > +architectures. Simply put, you might choose RAID1 to protect
> > from
> > +a single device failure, but you would choose remote
> > replication
> > +via dm-replicator for protection against a site failure.
> > +
> > +dm-replicator works by first sending write requests to the
> > "replicator
> > +log". Not to be confused with the device-mapper dirty log,
> > this
> > +replicator log behaves similarly to that of a journal. Write
> > requests
> > +go to this log first and then are copied to all the replicate
> > devices
> > +at their various locations. Requests are cleared from the
> > log once all
> > +replicate devices confirm the data is received/copied. This
> > architecture
> > +allows dm-replicator to be flexible in terms of device
> > characteristics.
> > +If one device should fall behind the others - perhaps due to
> > high latency -
> > +the slack is picked up by the log. The user has a great deal
> > of
> > +flexibility in specifying to what degree a particular site is
> > allowed to
> > +fall behind - if at all.
> > +
> > +Device-Mapper's dm-replicator has two targets, "replicator"
> > and
> > +"replicator-dev". The "replicator" target is used to setup
> > the
> > +aforementioned log and allow the specification of site link
> > properties.
> > +Through the "replicator" target, the user might specify that
> > writes
> > +that are copied to the local site must happen synchronously
> > (i.e the
> > +writes are complete only after they have passed through the
> > log device
> > +and have landed on the local site's disk). They may also
> > specify that
> > +a remote link should asynchronously complete writes, but that
> > the remote
> > +link should never fall more than 100MB behind in terms of
> > processing.
> > +Again, the "replicator" target is used to define the
> > replicator log and
> > +the characteristics of each site link.
> > +
> > +The "replicator-dev" target is used to define the devices
> > used and
> > +associate them with a particular replicator log. You might
> > think of
> > +this stage in a similar way to setting up RAID1 (mirroring).
> > You
> > +define a set of devices which will be copies of each other,
> > but
> > +access the device through the mirror virtual device which
> > takes care
> > +of the copying. The user accessible replicator device is
> > analogous
> > +to the mirror virtual device, while the set of devices being
> > copied
> > +to are analogous to the mirror images (sometimes called
> > 'legs').
> > +When creating a replicator device via the "replicator-dev"
> > target,
> > +it must be associated with the replicator log (created with
> > the
> > +aforementioned "replicator" target). When each redundant
> > device
> > +is specified as part of the replicator device, it is
> > associated with
> > +a site link whose properties were defined when the
> > "replicator"
> > +target was created.
> > +
> > +The user can go farther than simply replicating one device.
> > They
> > +can continue to add replicator devices - associating them
> > with a
> > +particular replicator log. Writes that go through the
> > replicator
> > +log are guarenteed to have their write ordering preserved.
> > So, if
> > +you associate more than one replicator device to a particular
> > +replicator log, you are preserving write ordering across
> > multiple
> > +devices. This might be useful if you had a database that
> > spanned
> > +multiple disks and write ordering must be preserved or any
> > transaction
> > +accounting scheme would be foiled. (You can imagine this
> > like
> > +preserving write ordering across a number of mirrored
> > devices, where
> > +each mirror has images/legs in different geographic
> > locations.)
> > +
> > +dm-replicator has a modular architecture. Future
> > implementations for
> > +the replicator log and site link modules are allowed. The
> > current
> > +replication log is ringbuffer - utilized to store all writes
> > being
> > +subject to replication and enforce write ordering. The
> > current site
> > +link code is based on accessing block devices (iSCSI, FC,
> > etc) and
> > +does device recovery including (initial) resynchronization.
> > +
> > +
> > +Picture of a 2 site configuration with 3 local devices (LDs)
> > in a
> > +primary site being resycnhronied to 3 remotes sites with 3
> > remote
> > +devices (RDs) each via site links (SLINK) 1-2 with site link
> > 0
> > +as a special case to handle the local devices:
> > +
> > + |
> > + Local (primary) site | Remote
> > sites
> > + -------------------- |
> > ------------
> > + |
> > + D1 D2 Dn |
> > + | | | |
> > + +---+- ... -+ |
> > + | |
> > + REPLOG-----------------+- SLINK1 ------------+
> > + | | | |
> > + SLINK0 (special case) | | |
> > + | | | |
> > + +-----+ ... + | | +----+- ... -+
> > + | | | | | | | |
> > + LD1 LD2 LDn | | RD1 RD2
> > RDn
> > + | |
> > + +-- SLINK2------------+
> > + | | |
> > + | | +----+- ... -+
> > + | | | | |
> > + | | RD1 RD2
> > RDn
> > + | |
> > + | |
> > + | |
> > + +- SLINKm ------------+
> > + | |
> > + | +----+- ... -+
> > + | | | |
> > + | RD1 RD2
> > RDn
> > +
> > +
> > +
> > +
> > +The following are descriptions of the device-mapper tables
> > used to
> > +construct the "replicator" and "replicator-dev" targets.
> > +
> > +"replicator" target parameters:
> > +-------------------------------
> > +<start> <length> replicator \
> > + <replog_type> <#replog_params> <replog_params> \
> > + [<slink_type_0> <#slink_params_0>
> > <slink_params_0>]{1..N}
> > +
> > +<replog_type> = "ringbuffer" is currently the only
> > available type
> > +<#replog_params> = # of args following this one intended for
> > the replog (2 or 4)
> > +<replog_params> = <dev_path> <dev_start> [auto/create/open
> > <size>]
> > + <dev_path> = device path of replication log (REPLOG)
> > backing store
> > + <dev_start> = offset to REPLOG header
> > + create = The replication log will be initialized
> > if not active
> > + and sized to "size". (If already
> > active, the create
> > + will fail.) Size is always in sectors.
> > + open = The replication log must be initialized
> > and valid or
> > + the constructor will fail.
> > + auto = If a valid replication log header is
> > found on the
> > + replication device, this will behave
> > like 'open'.
> > + Otherwise, this option behaves like
> > 'create'.
> > +
> > +<slink_type> = "blockdev" is currently the only available
> > type
> > +<#slink_params> = 1/2/4
> > +<slink_params> = <slink_nr> [<slink_policy> [<fall_behind>
> > <N>]]
> > + <slink_nr> = This is a unique number that is used
> > to identify a
> > + particular site/location. '0' is
> > always used to
> > + identify the local site, while
> > increasing integers
> > + are used to identify remote sites.
> > + <slink_policy> = The policy can be either 'sync' or
> > 'async'.
> > + 'sync' means write requests will not
> > return until
> > + the data is on the storage device.
> > 'async' allows
> > + a device to "fall behind"; that is,
> > outstanding
> > + write requests are waiting in the
> > replication log
> > + to be processed for this site, but it
> > is not delaying
> > + the writes of other sites.
> > + <fall_behind> = This field is used to specify how far
> > the user is
> > + willing to allow write requests to
> > this specific site
> > + to "fall behind" in processing before
> > switching to
> > + a 'sync' policy. This "fall behind"
> > threshhold can
> > + be specified in three ways: ios,
> > size, or timeout.
> > + 'ios' is the number of pending I/Os
> > allowed (e.g.
> > + "ios 10000"). 'size' is the amount
> > of pending data
> > + allowed (e.g. "size 200m"). Size
> > labels include:
> > + s (sectors), k, m, g, t, p, and e.
> > 'timeout' is
> > + the amount of time allowed for writes
> > to be
> > + outstanding. Time labels include: s,
> > m, h, and d.
> > +
> > +
> > +"replicator-dev" target parameters:
> > +-----------------------------------
> > +start> <length> replicator-dev
> > + <replicator_device> <dev_nr> \
> > + [<slink_nr> <#dev_params> <dev_params>
> > + <dlog_type> <#dlog_params> <dlog_params>]{1..N}
> > +
> > +<replicator_device> = device previously constructed via
> > "replication" target
> > +<dev_nr> = An integer that is used to 'tag' write
> > requests as
> > + belonging to a particular set of devices
> > - specifically,
> > + the devices that follow this argument
> > (i.e. the site
> > + link devices).
> > +<slink_nr> = This number identifies the site/location
> > where the next
> > + device to be specified comes from. It
> > is exactly the
> > + same number used to identify the
> > site/location (and its
> > + policies) in the "replicator" target.
> > Interestingly,
> > + while one might normally expect a
> > "dev_type" argument
> > + here, it can be deduced from the site
> > link number and
> > + the 'slink_type' given in the
> > "replication" target.
> > +<#dev_params> = '1' (The number of allowed parameters
> > actually depends
> > + on the 'slink_type' given in the
> > "replication" target.
> > + Since our only option there is
> > "blockdev", the only
> > + allowable number here is currently '1'.)
> > +<dev_params> = 'dev_path' (Again, since "blockdev" is
> > the only
> > + 'slink_type' available, the only
> > allowable argument here
> > + is the path to the device.)
> > +<dlog_type> = Not to be confused with the "replicator
> > log", this is
> > + the type of dirty log associated with
> > this particular
> > + device. Dirty logs are used for
> > synchronization, during
> > + initialization or fall behind
> > conditions, to bring devices
> > + into a coherent state with its peers -
> > analogous to
> > + rebuilding a RAID1 (mirror) device.
> > Available dirty
> > + log types include: 'nolog', 'core', and
> > 'disk'
> > +<#dlog_params> = The number of arguments required for a
> > particular log
> > + type - 'nolog' = 0, 'core' = 1/2, 'disk'
> > = 2/3.
> > +<dlog_params> = 'nolog' => ~no arguments~
> > + 'core' => <region_size> [sync | nosync]
> > + 'disk' => <dlog_dev_path> <region_size>
> > [sync | nosync]
> > + <region_size> = This sets the granularity at which
> > the dirty log
> > + tracks what areas of the device is
> > in-sync.
> > + [sync | nosync] = Optionally specify whether the sync
> > should be forced
> > + or avoided initially.
> > diff --git 2.6.33-rc1.orig/drivers/md/Kconfig
> > 2.6.33-rc1/drivers/md/Kconfig
> > index acb3a4e..62c9766 100644
> > --- 2.6.33-rc1.orig/drivers/md/Kconfig
> > +++ 2.6.33-rc1/drivers/md/Kconfig
> > @@ -313,6 +313,14 @@ config DM_DELAY
> >
> > If unsure, say N.
> >
> > +config DM_REPLICATOR
> > + tristate "Replication target (EXPERIMENTAL)"
> > + depends on BLK_DEV_DM && EXPERIMENTAL
> > + ---help---
> > + A target that supports replication of local devices to
> > remote sites.
> > +
> > + If unsure, say N.
> > +
> > config DM_UEVENT
> > bool "DM uevents (EXPERIMENTAL)"
> > depends on BLK_DEV_DM && EXPERIMENTAL
> > diff --git 2.6.33-rc1.orig/drivers/md/Makefile
> > 2.6.33-rc1/drivers/md/Makefile
> > index e355e7f..be05b39 100644
> > --- 2.6.33-rc1.orig/drivers/md/Makefile
> > +++ 2.6.33-rc1/drivers/md/Makefile
> > @@ -44,6 +44,7 @@ obj-$(CONFIG_DM_SNAPSHOT) +=
> > dm-snapshot.o
> > obj-$(CONFIG_DM_MIRROR) += dm-mirror.o
> > dm-log.o dm-region-hash.o
> > obj-$(CONFIG_DM_LOG_USERSPACE) += dm-log-userspace.o
> > obj-$(CONFIG_DM_ZERO) += dm-zero.o
> > +obj-$(CONFIG_DM_REPLICATOR) += dm-log.o dm-registry.o
> >
> > quiet_cmd_unroll = UNROLL $@
> > cmd_unroll = $(AWK) -f$(srctree)/$(src)/unroll.awk -vN=
> > $(UNROLL) \
> > diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.c
> > 2.6.33-rc1/drivers/md/dm-registry.c
> > new file mode 100644
> > index 0000000..fb8abbf
> > --- /dev/null
> > +++ 2.6.33-rc1/drivers/md/dm-registry.c
> > @@ -0,0 +1,224 @@
> > +/*
> > + * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
> > + *
> > + * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
> > + *
> > + * Generic registry for arbitrary structures
> > + * (needs dm_registry_type structure upfront each registered
> > structure).
> > + *
> > + * This file is released under the GPL.
> > + *
> > + * FIXME: use as registry for e.g. dirty log types as well.
> > + */
> > +
> > +#include <linux/init.h>
> > +#include <linux/module.h>
> > +#include <linux/moduleparam.h>
> > +
> > +#include "dm-registry.h"
> > +
> > +#define DM_MSG_PREFIX "dm-registry"
> > +
> > +static const char *version = "0.001";
> > +
> > +/* Sizable class registry. */
> > +static unsigned num_classes;
> > +static struct list_head *_classes;
> > +static rwlock_t *_locks;
> > +
> > +void *
> > +dm_get_type(const char *type_name, enum dm_registry_class
> > class)
> > +{
> > + struct dm_registry_type *t;
> > +
> > + read_lock(_locks + class);
> > + list_for_each_entry(t, _classes + class, list) {
> > + if (!strcmp(type_name, t->name)) {
> > + if (!t->use_count && !
> > try_module_get(t->module)) {
> > + read_unlock(_locks + class);
> > + return ERR_PTR(-ENOMEM);
> > + }
> > +
> > + t->use_count++;
> > + read_unlock(_locks + class);
> > + return t;
> > + }
> > + }
> > +
> > + read_unlock(_locks + class);
> > + return ERR_PTR(-ENOENT);
> > +}
> > +EXPORT_SYMBOL(dm_get_type);
> > +
> > +void
> > +dm_put_type(void *type, enum dm_registry_class class)
> > +{
> > + struct dm_registry_type *t = type;
> > +
> > + read_lock(_locks + class);
> > + if (!--t->use_count)
> > + module_put(t->module);
> > +
> > + read_unlock(_locks + class);
> > +}
> > +EXPORT_SYMBOL(dm_put_type);
> > +
> > +/* Add a type to the registry. */
> > +int
> > +dm_register_type(void *type, enum dm_registry_class class)
> > +{
> > + struct dm_registry_type *t = type, *tt;
> > +
> > + if (unlikely(class >= num_classes))
> > + return -EINVAL;
> > +
> > + tt = dm_get_type(t->name, class);
> > + if (unlikely(!IS_ERR(tt))) {
> > + dm_put_type(t, class);
> > + return -EEXIST;
> > + }
> > +
> > + write_lock(_locks + class);
> > + t->use_count = 0;
> > + list_add(&t->list, _classes + class);
> > + write_unlock(_locks + class);
> > +
> > + return 0;
> > +}
> > +EXPORT_SYMBOL(dm_register_type);
> > +
> > +/* Remove a type from the registry. */
> > +int
> > +dm_unregister_type(void *type, enum dm_registry_class class)
> > +{
> > + struct dm_registry_type *t = type;
> > +
> > + if (unlikely(class >= num_classes)) {
> > + DMERR("Attempt to unregister invalid class");
> > + return -EINVAL;
> > + }
> > +
> > + write_lock(_locks + class);
> > +
> > + if (unlikely(t->use_count)) {
> > + write_unlock(_locks + class);
> > + DMWARN("Attempt to unregister a type that is
> > still in use");
> > + return -ETXTBSY;
> > + } else
> > + list_del(&t->list);
> > +
> > + write_unlock(_locks + class);
> > + return 0;
> > +}
> > +EXPORT_SYMBOL(dm_unregister_type);
> > +
> > +/*
> > + * Return kmalloc'ed NULL terminated pointer
> > + * array of all type names of the given class.
> > + *
> > + * Caller has to kfree the array!.
> > + */
> > +const char **dm_types_list(enum dm_registry_class class)
> > +{
> > + unsigned i = 0, count = 0;
> > + const char **r;
> > + struct dm_registry_type *t;
> > +
> > + /* First count the registered types in the class. */
> > + read_lock(_locks + class);
> > + list_for_each_entry(t, _classes + class, list)
> > + count++;
> > + read_unlock(_locks + class);
> > +
> > + /* None registered in this class. */
> > + if (!count)
> > + return NULL;
> > +
> > + /* One member more for array NULL termination. */
> > + r = kzalloc((count + 1) * sizeof(*r), GFP_KERNEL);
> > + if (!r)
> > + return ERR_PTR(-ENOMEM);
> > +
> > + /*
> > + * Go with the counted ones.
> > + * Any new added ones after we counted will be
> > ignored!
> > + */
> > + read_lock(_locks + class);
> > + list_for_each_entry(t, _classes + class, list) {
> > + r[i++] = t->name;
> > + if (!--count)
> > + break;
> > + }
> > + read_unlock(_locks + class);
> > +
> > + return r;
> > +}
> > +EXPORT_SYMBOL(dm_types_list);
> > +
> > +int __init
> > +dm_registry_init(void)
> > +{
> > + unsigned n;
> > +
> > + BUG_ON(_classes);
> > + BUG_ON(_locks);
> > +
> > + /* Module parameter given ? */
> > + if (!num_classes)
> > + num_classes = DM_REGISTRY_CLASS_END;
> > +
> > + n = num_classes;
> > + _classes = kmalloc(n * sizeof(*_classes), GFP_KERNEL);
> > + if (!_classes) {
> > + DMERR("Failed to allocate classes registry");
> > + return -ENOMEM;
> > + }
> > +
> > + _locks = kmalloc(n * sizeof(*_locks), GFP_KERNEL);
> > + if (!_locks) {
> > + DMERR("Failed to allocate classes locks");
> > + kfree(_classes);
> > + _classes = NULL;
> > + return -ENOMEM;
> > + }
> > +
> > + while (n--) {
> > + INIT_LIST_HEAD(_classes + n);
> > + rwlock_init(_locks + n);
> > + }
> > +
> > + DMINFO("initialized %s for max %u classes", version,
> > num_classes);
> > + return 0;
> > +}
> > +
> > +void __exit
> > +dm_registry_exit(void)
> > +{
> > + BUG_ON(!_classes);
> > + BUG_ON(!_locks);
> > +
> > + kfree(_classes);
> > + _classes = NULL;
> > + kfree(_locks);
> > + _locks = NULL;
> > + DMINFO("exit %s", version);
> > +}
> > +
> > +/* Module hooks */
> > +module_init(dm_registry_init);
> > +module_exit(dm_registry_exit);
> > +module_param(num_classes, uint, 0);
> > +MODULE_PARM_DESC(num_classes, "Maximum number of classes");
> > +MODULE_DESCRIPTION(DM_NAME "device-mapper registry");
> > +MODULE_AUTHOR("Heinz Mauelshagen <heinzm@redhat.com>");
> > +MODULE_LICENSE("GPL");
> > +
> > +#ifndef MODULE
> > +static int __init num_classes_setup(char *str)
> > +{
> > + num_classes = simple_strtol(str, NULL, 0);
> > + return num_classes ? 1 : 0;
> > +}
> > +
> > +__setup("num_classes=", num_classes_setup);
> > +#endif
> > diff --git 2.6.33-rc1.orig/drivers/md/dm-registry.h
> > 2.6.33-rc1/drivers/md/dm-registry.h
> > new file mode 100644
> > index 0000000..1cb0ce8
> > --- /dev/null
> > +++ 2.6.33-rc1/drivers/md/dm-registry.h
> > @@ -0,0 +1,38 @@
> > +/*
> > + * Copyright (C) 2009 Red Hat, Inc. All rights reserved.
> > + *
> > + * Module Author: Heinz Mauelshagen (heinzm@redhat.com)
> > + *
> > + * Generic registry for arbitrary structures.
> > + * (needs dm_registry_type structure upfront each registered
> > structure).
> > + *
> > + * This file is released under the GPL.
> > + */
> > +
> > +#include "dm.h"
> > +
> > +#ifndef DM_REGISTRY_H
> > +#define DM_REGISTRY_H
> > +
> > +enum dm_registry_class {
> > + DM_REPLOG = 0,
> > + DM_SLINK,
> > + DM_LOG,
> > + DM_REGION_HASH,
> > + DM_REGISTRY_CLASS_END,
> > +};
> > +
> > +struct dm_registry_type {
> > + struct list_head list; /* Linked list of types in
> > this class. */
> > + const char *name;
> > + struct module *module;
> > + unsigned int use_count;
> > +};
> > +
> > +void *dm_get_type(const char *type_name, enum
> > dm_registry_class class);
> > +void dm_put_type(void *type, enum dm_registry_class class);
> > +int dm_register_type(void *type, enum dm_registry_class
> > class);
> > +int dm_unregister_type(void *type, enum dm_registry_class
> > class);
> > +const char **dm_types_list(enum dm_registry_class class);
> > +
> > +#endif
> > --
> > 1.6.2.5
> >
> > --
> > dm-devel mailing list
> > dm-devel@redhat.com
> > https://www.redhat.com/mailman/listinfo/dm-devel
> >
> > --
> > dm-devel mailing list
> > dm-devel@redhat.com
> > https://www.redhat.com/mailman/listinfo/dm-devel
>
> --
> dm-devel mailing list
> dm-devel@redhat.com
> https://www.redhat.com/mailman/listinfo/dm-devel
>
[-- Attachment #1.2: Type: text/html, Size: 61247 bytes --]
[-- Attachment #2: Type: text/plain, Size: 0 bytes --]
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler
2009-12-18 15:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler heinzm
2009-12-18 15:44 ` [PATCH v6 4/4] dm-replicator: blockdev site link handler heinzm
@ 2011-07-18 9:44 ` Busby
1 sibling, 0 replies; 9+ messages in thread
From: Busby @ 2011-07-18 9:44 UTC (permalink / raw)
To: device-mapper development; +Cc: Heinz Mauelshagen
is there any bug in slink_fallbehind_update function?
I use the dm-repl module, while the oops came as:
localhost kernel: Oops: 0010 [#2] SMP
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: last sysfs file: /sys/module/scsi_transport_iscsi/version
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: Stack:
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: ffff88005f5019e0 0000000000000000 ffff8800574d58c0
0000000000000000
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: 0000000000000000 ffff8800be3e3800 ffffffffa0320d50
0000000000000000
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: Call Trace:
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffffa031ddb7>] ?
slink_fallbehind_update+0x1f7/0x30e [dm_repl_log_ringbuffer]
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffffa0320d50>] ? do_log+0xf07/0x12ab
[dm_repl_log_ringbuffer]
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffffa031c4aa>] ? data_endio+0x0/0x10
[dm_repl_log_ringbuffer]
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffffa031c4ba>] ? header_endio+0x0/0xd
[dm_repl_log_ringbuffer]
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff8020c9ee>] ? apic_timer_interrupt+0xe/0x20
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff8022e37c>] ? update_curr+0x6f/0xaa
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff80236661>] ? dequeue_task_fair+0x93/0x105
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff804aab5d>] ? thread_return+0x3d/0xb0
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffffa031fe49>] ? do_log+0x0/0x12ab
[dm_repl_log_ringbuffer]
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff8024802c>] ? run_workqueue+0x7a/0x102
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff8024894c>] ? worker_thread+0xd5/0xe0
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff8024b2a6>] ? autoremove_wake_function+0x0/0x2e
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff80248877>] ? worker_thread+0x0/0xe0
Message from syslogd@ at Thu Jul 21 06:42:27 2011 ...
localhost kernel: [<ffffffff8024b176>] ? kthread+0x47/0x75
it seems there is something wrong in this section:
list_for_each_entry(pos, L_SLINK_COPY_LIST(l),
lists.l[E_WRITE_OR_COPY]) {
struct slink_copy_context *cc;
list_for_each_entry(cc, E_COPY_CONTEXT_LIST(pos),
list) {
if (cc->slink->ops->slink_number(cc->slink) ==
slink_nr) {
ss->fb.head_jiffies = cc->start_jiffies;
break;
}
}
}
maybe "cc->slink->ops->slink_number(cc->slink)" return an err?
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2011-07-18 9:44 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2009-12-18 15:44 [PATCH v6 0/4] dm-replicator: introduce new remote replication target heinzm
2009-12-18 15:44 ` [PATCH v6 1/4] dm-replicator: documentation and module registry heinzm
2009-12-18 15:44 ` [PATCH v6 2/4] dm-replicator: replication log and site link handler interfaces and main replicator module heinzm
2009-12-18 15:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler heinzm
2009-12-18 15:44 ` [PATCH v6 4/4] dm-replicator: blockdev site link handler heinzm
2011-07-18 9:44 ` [PATCH v6 3/4] dm-replicator: ringbuffer replication log handler Busby
2010-01-07 10:18 ` [PATCH v6 1/4] dm-replicator: documentation and module registry 张宇
2010-01-08 19:44 ` Heinz Mauelshagen
2010-02-09 1:48 ` Busby
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.