* [linux-lvm] PATCH: volume drivers in ll_rw_block
@ 1999-08-09 19:52 David Teigland
1999-08-10 10:09 ` Klaus Strebel
0 siblings, 1 reply; 3+ messages in thread
From: David Teigland @ 1999-08-09 19:52 UTC (permalink / raw)
To: linux-raid; +Cc: linux-lvm, gfs-devel
I believe a change is needed in the way ll_rw_block() handles volume
drivers. Currently, each volume driver requires a separate chunk of code
to handle its own mapping function. As more volume drivers are used (MD,
LVM and GFS' Pool), ll_rw_block gets long and messy. (See the combination
of MD and LVM in 2.2.10-ac12). This can be simplified by having all
drivers use a single function pointer to call a previously assigned
mapping or make_request function.
This patch adds the function pointers map_fn and makerq_fn to the blk_dev
structure and uses them like request_fn. If a block driver sets either of
these pointers in its init routine, the specific map or makerq function
will be called from ll_rw_block.
This eliminates repetitious driver-specific code and provides a clean way
to make MD, LVM and Pool modular. MD devices can currently be used within
LVM, but this patch, in a general sense, imposes no limit on the number or
order in which volumes are built on one another.
Unaddressed by this method is more than one stacked volume driver using
makerq_fn. If stacking volume drivers, only the bottom one can use a
make_request function.
The patch below to 2.2.10+raid0145 implements the generic mapping. Two
other patches can be found at ftp://globalfilesystem.org/pub/GFS/patches/
- modular_md-2.2.10 adds module support for MD using this method.
- lvm_map-0.7 allows LVM to be used with generic_map.
David Teigland
--------------------Cut Here---------------- generic_map-2.2.10-raid
diff -urN linux-raid-1/drivers/block/ll_rw_blk.c linux-raid-2/drivers/block/ll_rw_blk.c
--- linux-raid-1/drivers/block/ll_rw_blk.c Mon Aug 16 14:42:42 1999
+++ linux-raid-2/drivers/block/ll_rw_blk.c Mon Aug 16 14:46:08 1999
@@ -578,7 +578,7 @@
{
unsigned int major;
int correct_size;
- struct blk_dev_struct * dev;
+ struct blk_dev_struct * dev, * tdev = NULL;
int i;
/* Make sure that the first block contains something reasonable */
@@ -608,7 +608,10 @@
/* Verify requested block sizes. */
for (i = 0; i < nr; i++) {
- if (bh[i] && bh[i]->b_size != correct_size) {
+ if (!bh[i])
+ continue;
+
+ if (bh[i]->b_size != correct_size) {
printk(KERN_NOTICE "ll_rw_block: device %s: "
"only %d-char blocks implemented (%lu)\n",
kdevname(bh[0]->b_dev),
@@ -619,15 +622,17 @@
/* Md remaps blocks now */
bh[i]->b_rdev = bh[i]->b_dev;
bh[i]->b_rsector=bh[i]->b_blocknr*(bh[i]->b_size >> 9);
-#ifdef CONFIG_BLK_DEV_MD
- if (major==MD_MAJOR &&
- md_map (bh[i]->b_dev, &bh[i]->b_rdev,
- &bh[i]->b_rsector, bh[i]->b_size >> 9)) {
- printk (KERN_ERR
- "Bad md_map in ll_rw_block\n");
- goto sorry;
- }
-#endif
+
+ tdev = dev;
+ while (tdev->map_fn) {
+ if (tdev->map_fn (bh[i]->b_rdev, &bh[i]->b_rdev,
+ &bh[i]->b_rsector,
+ bh[i]->b_size >> 9)) {
+ printk (KERN_ERR "Bad map in ll_rw_block\n");
+ goto sorry;
+ }
+ tdev = blk_dev + MAJOR(bh[i]->b_rdev);
+ }
}
if ((rw == WRITE || rw == WRITEA) && is_read_only(bh[0]->b_dev)) {
@@ -639,12 +644,10 @@
for (i = 0; i < nr; i++) {
if (bh[i]) {
set_bit(BH_Req, &bh[i]->b_state);
-#ifdef CONFIG_BLK_DEV_MD
- if (MAJOR(bh[i]->b_dev) == MD_MAJOR) {
- md_make_request(bh[i], rw);
- continue;
- }
-#endif
+ if (tdev->makerq_fn) {
+ tdev->makerq_fn (bh[i], rw);
+ continue;
+ }
make_request(MAJOR(bh[i]->b_rdev), rw, bh[i]);
}
}
@@ -724,6 +727,8 @@
for (dev = blk_dev + MAX_BLKDEV; dev-- != blk_dev;) {
dev->request_fn = NULL;
+ dev->map_fn = NULL;
+ dev->makerq_fn = NULL;
dev->queue = NULL;
dev->current_request = NULL;
dev->plug.rq_status = RQ_INACTIVE;
diff -urN linux-raid-1/drivers/block/md.c linux-raid-2/drivers/block/md.c
--- linux-raid-1/drivers/block/md.c Mon Aug 16 14:42:42 1999
+++ linux-raid-2/drivers/block/md.c Mon Aug 16 14:43:15 1999
@@ -3894,6 +3894,8 @@
}
blk_dev[MD_MAJOR].request_fn = DEVICE_REQUEST;
+ blk_dev[MD_MAJOR].makerq_fn=md_make_request;
+ blk_dev[MD_MAJOR].map_fn=md_map;
blk_dev[MD_MAJOR].current_request = NULL;
read_ahead[MD_MAJOR] = INT_MAX;
md_gendisk.next = gendisk_head;
diff -urN linux-raid-1/include/linux/blkdev.h linux-raid-2/include/linux/blkdev.h
--- linux-raid-1/include/linux/blkdev.h Mon Aug 16 14:42:42 1999
+++ linux-raid-2/include/linux/blkdev.h Mon Aug 16 14:46:53 1999
@@ -34,10 +34,14 @@
};
typedef void (request_fn_proc) (void);
+typedef int (makerq_fn_proc) (struct buffer_head *, int rw);
+typedef int (map_fn_proc) (kdev_t, kdev_t *, unsigned long *, unsigned long);
typedef struct request ** (queue_proc) (kdev_t dev);
struct blk_dev_struct {
request_fn_proc *request_fn;
+ makerq_fn_proc *makerq_fn;
+ map_fn_proc *map_fn;
/*
* queue_proc has to be atomic
*/
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [linux-lvm] PATCH: volume drivers in ll_rw_block
1999-08-09 19:52 [linux-lvm] PATCH: volume drivers in ll_rw_block David Teigland
@ 1999-08-10 10:09 ` Klaus Strebel
1999-08-10 19:39 ` [linux-lvm] " David Teigland
0 siblings, 1 reply; 3+ messages in thread
From: Klaus Strebel @ 1999-08-10 10:09 UTC (permalink / raw)
To: David Teigland; +Cc: linux-raid, linux-lvm, gfs-devel
David Teigland wrote:
> This eliminates repetitious driver-specific code and provides a clean way
> to make MD, LVM and Pool modular. MD devices can currently be used within
> LVM, but this patch, in a general sense, imposes no limit on the number or
> order in which volumes are built on one another.
>
> Unaddressed by this method is more than one stacked volume driver using
> makerq_fn. If stacking volume drivers, only the bottom one can use a
> make_request function.
>
> The patch below to 2.2.10+raid0145 implements the generic mapping. Two
> other patches can be found at ftp://globalfilesystem.org/pub/GFS/patches/
>
> - modular_md-2.2.10 adds module support for MD using this method.
> - lvm_map-0.7 allows LVM to be used with generic_map.
Hi all,
the lvm_map-0.7 lacks one thing in ll_rw_blk.c, i show below in
generic_map-2.2.10-raid what and where.
> --------------------Cut Here---------------- generic_map-2.2.10-raid
>
> diff -urN linux-raid-1/drivers/block/ll_rw_blk.c linux-raid-2/drivers/block/ll_rw_blk.c
> --- linux-raid-1/drivers/block/ll_rw_blk.c Mon Aug 16 14:42:42 1999
> +++ linux-raid-2/drivers/block/ll_rw_blk.c Mon Aug 16 14:46:08 1999
...
> @@ -619,15 +622,17 @@
> /* Md remaps blocks now */
> bh[i]->b_rdev = bh[i]->b_dev;
> bh[i]->b_rsector=bh[i]->b_blocknr*(bh[i]->b_size >> 9);
-#if defined CONFIG_BLK_DEV_LVM || defined CONFIG_BLK_DEV_LVM_MODULE
- major = MAJOR(bh[i]->b_dev);
- if ( major == LVM_BLK_MAJOR) {
- int ret;
-
- if ( lvm_map_ptr == NULL) {
- printk ( KERN_ERR
- "Bad lvm_map_ptr in ll_rw_block\n");
- goto sorry;
- }
- if ( ( ret = ( lvm_map_ptr) ( MINOR ( bh[i]->b_dev),
- &bh[i]->b_rdev,
- &bh[i]->b_rsector,
- bh[i]->b_size >> 9,
- rw)) != 0) {
- printk ( KERN_ERR
- "Bad lvm_map in ll_rw_block\n");
- goto sorry;
- }
- /* remap major too ... */
- major = MAJOR(bh[i]->b_rdev);
- }
-#endif
> -#ifdef CONFIG_BLK_DEV_MD
> - if (major==MD_MAJOR &&
> - md_map (bh[i]->b_dev, &bh[i]->b_rdev,
> - &bh[i]->b_rsector, bh[i]->b_size >> 9)) {
> - printk (KERN_ERR
> - "Bad md_map in ll_rw_block\n");
> - goto sorry;
> - }
> -#endif
> +
> + tdev = dev;
> + while (tdev->map_fn) {
> + if (tdev->map_fn (bh[i]->b_rdev, &bh[i]->b_rdev,
> + &bh[i]->b_rsector,
> + bh[i]->b_size >> 9)) {
> + printk (KERN_ERR "Bad map in ll_rw_block\n");
> + goto sorry;
> + }
> + tdev = blk_dev + MAJOR(bh[i]->b_rdev);
> + }
> }
>
> if ((rw == WRITE || rw == WRITEA) && is_read_only(bh[0]->b_dev)) {
or : in my 1st effort i just did this :
-#if defined CONFIG_BLK_DEV_LVM || defined CONFIG_BLK_DEV_LVM_MODULE
+#if defined CONFIG_BLK_DEV_LVM_NEVER || defined
CONFIG_BLK_DEV_LVM_MODULE_NEVER
at that place in ll_rw_block. Whithout it, you�ll get the "Bad lvm_map
in ll_rw_block" when fsck tries to read the superblock!
Ciao
Klaus
BTW: I use lvm since release 0.5alpha and had (almost ;-)) never any
problems with it!
--
Klaus Strebel
stb@ep-ag.com
EIGNER + PARTNER AG - The Engineering Warehouse Company -
<http://www.ep-ag.com>
-----------------------------------------------------------------------
^ permalink raw reply [flat|nested] 3+ messages in thread
* [linux-lvm] Re: PATCH: volume drivers in ll_rw_block
1999-08-10 10:09 ` Klaus Strebel
@ 1999-08-10 19:39 ` David Teigland
0 siblings, 0 replies; 3+ messages in thread
From: David Teigland @ 1999-08-10 19:39 UTC (permalink / raw)
To: Klaus Strebel; +Cc: linux-raid, linux-lvm, gfs-devel
> the lvm_map-0.7 lacks one thing in ll_rw_blk.c, i show below in
> generic_map-2.2.10-raid what and where.
> -#if defined CONFIG_BLK_DEV_LVM || defined CONFIG_BLK_DEV_LVM_MODULE
> - major = MAJOR(bh[i]->b_dev);
> - if ( major == LVM_BLK_MAJOR) {
> - int ret;
> -
> - if ( lvm_map_ptr == NULL) {
> - printk ( KERN_ERR
> - "Bad lvm_map_ptr in ll_rw_block\n");
> - goto sorry;
> - }
> - if ( ( ret = ( lvm_map_ptr) ( MINOR ( bh[i]->b_dev),
> - &bh[i]->b_rdev,
> - &bh[i]->b_rsector,
> - bh[i]->b_size >> 9,
> - rw)) != 0) {
> - printk ( KERN_ERR
> - "Bad lvm_map in ll_rw_block\n");
> - goto sorry;
> - }
> - /* remap major too ... */
> - major = MAJOR(bh[i]->b_rdev);
> - }
> -#endif
Correct. This chunk is no longer needed. In the lvm_map-0.7 patch I
wasn't trying to fix everything from patch-2.2.10-LVM, only lvm.c.
To adhere with the generic mapping, patch-2.2.10-LVM should change in the
following ways:
- remove all modifications to the function ll_rw_block()
- remove all references to lvm_map from ll_rw_blk.c
- lvm.c should include the changes in lvm_map-0.7
In other words, when using the generic mapping, the LVM stuff shouldn't
affect ll_rw_blk.c at all except the insertion of lvm_init at the very
end. This simplifies the LVM patch.
Thanks for pointing that out. I'm not sure if this addresses your fsck
problem or not.
David Teigland
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~1999-08-10 19:39 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
1999-08-09 19:52 [linux-lvm] PATCH: volume drivers in ll_rw_block David Teigland
1999-08-10 10:09 ` Klaus Strebel
1999-08-10 19:39 ` [linux-lvm] " David Teigland
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).