* [PATCHv3 0/5] Dynamic growing data area support
@ 2017-03-20 9:09 lixiubo
2017-03-20 9:09 ` [PATCHv3 1/5] tcmu: Add dynamic growing data area feature support lixiubo
` (5 more replies)
0 siblings, 6 replies; 9+ messages in thread
From: lixiubo @ 2017-03-20 9:09 UTC (permalink / raw)
To: agrover, nab, mchristi
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly, Xiubo Li
From: Xiubo Li <lixiubo@cmss.chinamobile.com>
Changed for V3:
- [PATCHv2 2/5] fix double usage of blocks and possible page fault
call trace.
- [PATCHv2 5/5] fix a mistake.
Changed for V2:
- [PATCHv2 1/5] just fixes some small spelling and other mistakes.
And as the initial patch, here sets cmd area to 8M and data area to
1G(1M fixed and 1023M growing)
- [PATCHv2 2/5] is a new one, adding global data block pool support.
The max total size of the pool is 2G and all the targets will get
growing blocks from here.
Test this using multi-targets at the same time.
- [PATCHv2 3/5] changed nothing, respin it to avoid the conflict.
- [PATCHv2 4/5] and [PATCHv2 5/5] are new ones.
Xiubo Li (5):
tcmu: Add dynamic growing data area feature support
tcmu: Add global data block pool support
target/user: Fix possible overwrite of t_data_sg's last iov[]
target/user: Fix wrongly calculating of the base_command_size
target/user: Clean up tcmu_queue_cmd_ring
drivers/target/target_core_user.c | 621 +++++++++++++++++++++++++++++++-------
1 file changed, 514 insertions(+), 107 deletions(-)
--
1.8.3.1
^ permalink raw reply [flat|nested] 9+ messages in thread
* [PATCHv3 1/5] tcmu: Add dynamic growing data area feature support
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
@ 2017-03-20 9:09 ` lixiubo
2017-03-20 9:09 ` [PATCHv3 2/5] tcmu: Add global data block pool support lixiubo
` (4 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: lixiubo @ 2017-03-20 9:09 UTC (permalink / raw)
To: agrover, nab, mchristi
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly, Xiubo Li
From: Xiubo Li <lixiubo@cmss.chinamobile.com>
Currently for the TCMU, the ring buffer size is fixed to 64K cmd
area + 1M data area, and this will be bottlenecks for high iops.
The struct tcmu_cmd_entry {} size is fixed about 112 bytes with
iovec[N] & N <= 4, and the size of struct iovec is about 16 bytes.
If N == 0, the ratio will be sizeof(cmd entry) : sizeof(datas) ==
112Bytes : (N * 4096)Bytes = 28 : 0, no data area is need.
If 0 < N <=4, the ratio will be sizeof(cmd entry) : sizeof(datas)
== 112Bytes : (N * 4096)Bytes = 28 : (N * 1024), so the max will
be 28 : 1024.
If N > 4, the sizeof(cmd entry) will be [(N - 4) *16 + 112] bytes,
and its corresponding data size will be [N * 4096], so the ratio
of sizeof(cmd entry) : sizeof(datas) == [(N - 4) * 16 + 112)Bytes
: (N * 4096)Bytes == 4/1024 - 12/(N * 1024), so the max is about
4 : 1024.
When N is bigger, the ratio will be smaller.
As the initial patch, we will set the cmd area size to 2M, and
the cmd area size to 32M. The TCMU will dynamically grows the data
area from 0 to max 32M size as needed.
The cmd area memory will be allocated through vmalloc(), and the
data area's blocks will be allocated individually later when needed.
The allocated data area block memory will be managed via radix tree.
For now the bitmap still be the most efficient way to search and
manage the block index, this could be update later.
Signed-off-by: Xiubo Li <lixiubo@cmss.chinamobile.com>
Signed-off-by: Jianfei Hu <hujianfei@cmss.chinamobile.com>
---
drivers/target/target_core_user.c | 286 ++++++++++++++++++++++++++++----------
1 file changed, 215 insertions(+), 71 deletions(-)
diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
index c6874c3..fc9f237 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -2,6 +2,7 @@
* Copyright (C) 2013 Shaohua Li <shli@kernel.org>
* Copyright (C) 2014 Red Hat, Inc.
* Copyright (C) 2015 Arrikto, Inc.
+ * Copyright (C) 2017 Chinamobile, Inc.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
@@ -25,6 +26,7 @@
#include <linux/parser.h>
#include <linux/vmalloc.h>
#include <linux/uio_driver.h>
+#include <linux/radix-tree.h>
#include <linux/stringify.h>
#include <linux/bitops.h>
#include <linux/highmem.h>
@@ -63,15 +65,17 @@
* this may have a 'UAM' comment.
*/
-
#define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
-#define DATA_BLOCK_BITS 256
-#define DATA_BLOCK_SIZE 4096
+/* For cmd area, the size is fixed 2M */
+#define CMDR_SIZE (2 * 1024 * 1024)
-#define CMDR_SIZE (16 * 4096)
+/* For data area, the size is fixed 32M */
+#define DATA_BLOCK_BITS (8 * 1024)
+#define DATA_BLOCK_SIZE 4096
#define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
+/* The ring buffer size is 34M */
#define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
static struct device *tcmu_root_device;
@@ -103,12 +107,14 @@ struct tcmu_dev {
size_t data_off;
size_t data_size;
- DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
-
wait_queue_head_t wait_cmdr;
/* TODO should this be a mutex? */
spinlock_t cmdr_lock;
+ uint32_t dbi_cur;
+ DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
+ struct radix_tree_root data_blocks;
+
struct idr commands;
spinlock_t commands_lock;
@@ -130,7 +136,9 @@ struct tcmu_cmd {
/* Can't use se_cmd when cleaning up expired cmds, because if
cmd has been completed then accessing se_cmd is off limits */
- DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
+ uint32_t dbi_len;
+ uint32_t dbi_cur;
+ uint32_t *dbi;
unsigned long deadline;
@@ -161,10 +169,80 @@ enum tcmu_multicast_groups {
.netnsok = true,
};
+static int tcmu_get_empty_block(struct tcmu_dev *udev, void **addr)
+{
+ void *p;
+ uint32_t dbi;
+ int ret;
+
+ dbi = find_first_zero_bit(udev->data_bitmap, DATA_BLOCK_BITS);
+ if (dbi > udev->dbi_cur)
+ udev->dbi_cur = dbi;
+
+ set_bit(dbi, udev->data_bitmap);
+
+ p = radix_tree_lookup(&udev->data_blocks, dbi);
+ if (!p) {
+ p = kzalloc(DATA_BLOCK_SIZE, GFP_ATOMIC);
+ if (!p) {
+ clear_bit(dbi, udev->data_bitmap);
+ return -ENOMEM;
+ }
+
+ ret = radix_tree_insert(&udev->data_blocks, dbi, p);
+ if (ret) {
+ kfree(p);
+ clear_bit(dbi, udev->data_bitmap);
+ return ret;
+ }
+ }
+
+ *addr = p;
+ return dbi;
+}
+
+static void *tcmu_get_block_addr(struct tcmu_dev *udev, uint32_t dbi)
+{
+ return radix_tree_lookup(&udev->data_blocks, dbi);
+}
+
+#define tcmu_cmd_reset_dbi_cur(cmd) ((cmd)->dbi_cur = 0)
+#define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
+#define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
+
+static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd)
+{
+ struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
+ uint32_t bi;
+
+ for (bi = 0; bi < tcmu_cmd->dbi_len; bi++)
+ clear_bit(tcmu_cmd->dbi[bi], udev->data_bitmap);
+}
+
+static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
+{
+ kfree(tcmu_cmd->dbi);
+ kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
+}
+
+static inline uint32_t tcmu_cmd_get_dbi_len(struct se_cmd *se_cmd)
+{
+ size_t data_length = se_cmd->data_length;
+ uint32_t dbi_len;
+
+ if (se_cmd->se_cmd_flags & SCF_BIDI)
+ data_length += se_cmd->t_bidi_data_sg->length;
+
+ dbi_len = (data_length + DATA_BLOCK_SIZE - 1) / DATA_BLOCK_SIZE;
+
+ return dbi_len;
+}
+
static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
{
struct se_device *se_dev = se_cmd->se_dev;
struct tcmu_dev *udev = TCMU_DEV(se_dev);
+ uint32_t dbi_len = tcmu_cmd_get_dbi_len(se_cmd);
struct tcmu_cmd *tcmu_cmd;
int cmd_id;
@@ -178,6 +256,14 @@ static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
tcmu_cmd->deadline = jiffies +
msecs_to_jiffies(udev->cmd_time_out);
+ tcmu_cmd_reset_dbi_cur(tcmu_cmd);
+ tcmu_cmd->dbi_len = dbi_len;
+ tcmu_cmd->dbi = kcalloc(dbi_len, sizeof(uint32_t), GFP_KERNEL);
+ if (!tcmu_cmd->dbi) {
+ kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
+ return NULL;
+ }
+
idr_preload(GFP_KERNEL);
spin_lock_irq(&udev->commands_lock);
cmd_id = idr_alloc(&udev->commands, tcmu_cmd, 0,
@@ -186,7 +272,7 @@ static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
idr_preload_end();
if (cmd_id < 0) {
- kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
+ tcmu_free_cmd(tcmu_cmd);
return NULL;
}
tcmu_cmd->cmd_id = cmd_id;
@@ -248,10 +334,10 @@ static inline void new_iov(struct iovec **iov, int *iov_cnt,
#define UPDATE_HEAD(head, used, size) smp_store_release(&head, ((head % size) + used) % size)
/* offset is relative to mb_addr */
-static inline size_t get_block_offset(struct tcmu_dev *dev,
- int block, int remaining)
+static inline size_t get_block_offset_user(struct tcmu_dev *dev,
+ int dbi, int remaining)
{
- return dev->data_off + block * DATA_BLOCK_SIZE +
+ return dev->data_off + dbi * DATA_BLOCK_SIZE +
DATA_BLOCK_SIZE - remaining;
}
@@ -260,14 +346,15 @@ static inline size_t iov_tail(struct tcmu_dev *udev, struct iovec *iov)
return (size_t)iov->iov_base + iov->iov_len;
}
-static void alloc_and_scatter_data_area(struct tcmu_dev *udev,
- struct scatterlist *data_sg, unsigned int data_nents,
- struct iovec **iov, int *iov_cnt, bool copy_data)
+static int alloc_and_scatter_data_area(struct tcmu_dev *udev,
+ struct tcmu_cmd *tcmu_cmd, struct scatterlist *data_sg,
+ unsigned int data_nents, struct iovec **iov,
+ int *iov_cnt, bool copy_data)
{
- int i, block;
+ int i, dbi;
int block_remaining = 0;
- void *from, *to;
- size_t copy_bytes, to_offset;
+ void *from, *to = NULL;
+ size_t copy_bytes, to_offset, offset;
struct scatterlist *sg;
for_each_sg(data_sg, sg, data_nents, i) {
@@ -275,22 +362,26 @@ static void alloc_and_scatter_data_area(struct tcmu_dev *udev,
from = kmap_atomic(sg_page(sg)) + sg->offset;
while (sg_remaining > 0) {
if (block_remaining == 0) {
- block = find_first_zero_bit(udev->data_bitmap,
- DATA_BLOCK_BITS);
block_remaining = DATA_BLOCK_SIZE;
- set_bit(block, udev->data_bitmap);
+ dbi = tcmu_get_empty_block(udev, &to);
+ if (dbi < 0)
+ return dbi;
+ tcmu_cmd_set_dbi(tcmu_cmd, dbi);
}
+
copy_bytes = min_t(size_t, sg_remaining,
block_remaining);
- to_offset = get_block_offset(udev, block,
+ to_offset = get_block_offset_user(udev, dbi,
block_remaining);
- to = (void *)udev->mb_addr + to_offset;
+ offset = DATA_BLOCK_SIZE - block_remaining;
+ to = (void *)(unsigned long)to + offset;
+
if (*iov_cnt != 0 &&
to_offset == iov_tail(udev, *iov)) {
(*iov)->iov_len += copy_bytes;
} else {
new_iov(iov, iov_cnt, udev);
- (*iov)->iov_base = (void __user *) to_offset;
+ (*iov)->iov_base = (void __user *)to_offset;
(*iov)->iov_len = copy_bytes;
}
if (copy_data) {
@@ -303,21 +394,17 @@ static void alloc_and_scatter_data_area(struct tcmu_dev *udev,
}
kunmap_atomic(from - sg->offset);
}
-}
-static void free_data_area(struct tcmu_dev *udev, struct tcmu_cmd *cmd)
-{
- bitmap_xor(udev->data_bitmap, udev->data_bitmap, cmd->data_bitmap,
- DATA_BLOCK_BITS);
+ return 0;
}
-static void gather_data_area(struct tcmu_dev *udev, unsigned long *cmd_bitmap,
+static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
struct scatterlist *data_sg, unsigned int data_nents)
{
- int i, block;
+ int i, dbi;
int block_remaining = 0;
void *from, *to;
- size_t copy_bytes, from_offset;
+ size_t copy_bytes, offset;
struct scatterlist *sg;
for_each_sg(data_sg, sg, data_nents, i) {
@@ -325,16 +412,14 @@ static void gather_data_area(struct tcmu_dev *udev, unsigned long *cmd_bitmap,
to = kmap_atomic(sg_page(sg)) + sg->offset;
while (sg_remaining > 0) {
if (block_remaining == 0) {
- block = find_first_bit(cmd_bitmap,
- DATA_BLOCK_BITS);
block_remaining = DATA_BLOCK_SIZE;
- clear_bit(block, cmd_bitmap);
+ dbi = tcmu_cmd_get_dbi(tcmu_cmd);
+ from = tcmu_get_block_addr(udev, dbi);
}
copy_bytes = min_t(size_t, sg_remaining,
block_remaining);
- from_offset = get_block_offset(udev, block,
- block_remaining);
- from = (void *) udev->mb_addr + from_offset;
+ offset = DATA_BLOCK_SIZE - block_remaining;
+ from = (void *)(unsigned long)from + offset;
tcmu_flush_dcache_range(from, copy_bytes);
memcpy(to + sg->length - sg_remaining, from,
copy_bytes);
@@ -403,12 +488,11 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t d
struct tcmu_mailbox *mb;
struct tcmu_cmd_entry *entry;
struct iovec *iov;
- int iov_cnt;
+ int iov_cnt, ret;
uint32_t cmd_head;
uint64_t cdb_off;
bool copy_to_data_area;
size_t data_length;
- DECLARE_BITMAP(old_bitmap, DATA_BLOCK_BITS);
if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
@@ -498,28 +582,31 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t d
entry->hdr.kflags = 0;
entry->hdr.uflags = 0;
- bitmap_copy(old_bitmap, udev->data_bitmap, DATA_BLOCK_BITS);
-
/* Handle allocating space from the data area */
iov = &entry->req.iov[0];
iov_cnt = 0;
copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
|| se_cmd->se_cmd_flags & SCF_BIDI);
- alloc_and_scatter_data_area(udev, se_cmd->t_data_sg,
+ ret = alloc_and_scatter_data_area(udev, tcmu_cmd, se_cmd->t_data_sg,
se_cmd->t_data_nents, &iov, &iov_cnt, copy_to_data_area);
+ if (ret) {
+ pr_err("tcmu: alloc and scatter data failed\n");
+ return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+ }
entry->req.iov_cnt = iov_cnt;
entry->req.iov_dif_cnt = 0;
/* Handle BIDI commands */
iov_cnt = 0;
- alloc_and_scatter_data_area(udev, se_cmd->t_bidi_data_sg,
- se_cmd->t_bidi_data_nents, &iov, &iov_cnt, false);
+ ret = alloc_and_scatter_data_area(udev, tcmu_cmd,
+ se_cmd->t_bidi_data_sg, se_cmd->t_bidi_data_nents,
+ &iov, &iov_cnt, false);
+ if (ret) {
+ pr_err("tcmu: alloc and scatter bidi data failed\n");
+ return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+ }
entry->req.iov_bidi_cnt = iov_cnt;
- /* cmd's data_bitmap is what changed in process */
- bitmap_xor(tcmu_cmd->data_bitmap, old_bitmap, udev->data_bitmap,
- DATA_BLOCK_BITS);
-
/* All offsets relative to mb_addr, not start of entry! */
cdb_off = CMDR_OFF + cmd_head + base_command_size;
memcpy((void *) mb + cdb_off, se_cmd->t_task_cdb, scsi_command_size(se_cmd->t_task_cdb));
@@ -560,7 +647,7 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t d
idr_remove(&udev->commands, tcmu_cmd->cmd_id);
spin_unlock_irq(&udev->commands_lock);
- kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
+ tcmu_free_cmd(tcmu_cmd);
}
return ret;
@@ -582,41 +669,34 @@ static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *
return;
}
+ tcmu_cmd_reset_dbi_cur(cmd);
+
if (entry->hdr.uflags & TCMU_UFLAG_UNKNOWN_OP) {
- free_data_area(udev, cmd);
pr_warn("TCMU: Userspace set UNKNOWN_OP flag on se_cmd %p\n",
cmd->se_cmd);
entry->rsp.scsi_status = SAM_STAT_CHECK_CONDITION;
} else if (entry->rsp.scsi_status == SAM_STAT_CHECK_CONDITION) {
memcpy(se_cmd->sense_buffer, entry->rsp.sense_buffer,
se_cmd->scsi_sense_length);
- free_data_area(udev, cmd);
} else if (se_cmd->se_cmd_flags & SCF_BIDI) {
- DECLARE_BITMAP(bitmap, DATA_BLOCK_BITS);
-
/* Get Data-In buffer before clean up */
- bitmap_copy(bitmap, cmd->data_bitmap, DATA_BLOCK_BITS);
- gather_data_area(udev, bitmap,
+ gather_data_area(udev, cmd,
se_cmd->t_bidi_data_sg, se_cmd->t_bidi_data_nents);
- free_data_area(udev, cmd);
} else if (se_cmd->data_direction == DMA_FROM_DEVICE) {
- DECLARE_BITMAP(bitmap, DATA_BLOCK_BITS);
-
- bitmap_copy(bitmap, cmd->data_bitmap, DATA_BLOCK_BITS);
- gather_data_area(udev, bitmap,
+ gather_data_area(udev, cmd,
se_cmd->t_data_sg, se_cmd->t_data_nents);
- free_data_area(udev, cmd);
} else if (se_cmd->data_direction == DMA_TO_DEVICE) {
- free_data_area(udev, cmd);
+ /* TODO: */
} else if (se_cmd->data_direction != DMA_NONE) {
pr_warn("TCMU: data direction was %d!\n",
se_cmd->data_direction);
}
target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
- cmd->se_cmd = NULL;
- kmem_cache_free(tcmu_cmd_cache, cmd);
+ cmd->se_cmd = NULL;
+ tcmu_cmd_free_data(cmd);
+ tcmu_free_cmd(cmd);
}
static unsigned int tcmu_handle_completions(struct tcmu_dev *udev)
@@ -774,6 +854,54 @@ static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
return 0;
}
+static void tcmu_blocks_release(struct tcmu_dev *udev, bool release_pending)
+{
+ uint32_t dbi, end;
+ void *addr;
+
+ spin_lock_irq(&udev->cmdr_lock);
+
+ end = udev->dbi_cur + 1;
+
+ /* try to release all unused blocks */
+ dbi = find_first_zero_bit(udev->data_bitmap, end);
+ if (dbi >= end) {
+ spin_unlock_irq(&udev->cmdr_lock);
+ return;
+ }
+ do {
+ addr = radix_tree_delete(&udev->data_blocks, dbi);
+ kfree(addr);
+
+ dbi = find_next_zero_bit(udev->data_bitmap, end, dbi + 1);
+ } while (dbi < end);
+
+ if (!release_pending)
+ return;
+
+ /* try to release all pending blocks */
+ dbi = find_first_bit(udev->data_bitmap, end);
+ if (dbi >= end) {
+ spin_unlock_irq(&udev->cmdr_lock);
+ return;
+ }
+ do {
+ addr = radix_tree_delete(&udev->data_blocks, dbi);
+ kfree(addr);
+
+ dbi = find_next_bit(udev->data_bitmap, end, dbi + 1);
+ } while (dbi < end);
+
+ spin_unlock_irq(&udev->cmdr_lock);
+}
+
+static void tcmu_vma_close(struct vm_area_struct *vma)
+{
+ struct tcmu_dev *udev = vma->vm_private_data;
+
+ tcmu_blocks_release(udev, false);
+}
+
/*
* mmap code from uio.c. Copied here because we want to hook mmap()
* and this stuff must come along.
@@ -809,17 +937,28 @@ static int tcmu_vma_fault(struct vm_fault *vmf)
*/
offset = (vmf->pgoff - mi) << PAGE_SHIFT;
- addr = (void *)(unsigned long)info->mem[mi].addr + offset;
- if (info->mem[mi].memtype == UIO_MEM_LOGICAL)
- page = virt_to_page(addr);
- else
+ if (offset < udev->data_off) {
+ /* For the vmalloc()ed cmd area pages */
+ addr = (void *)(unsigned long)info->mem[mi].addr + offset;
page = vmalloc_to_page(addr);
+ } else {
+ /* For the dynamically growing data area pages */
+ uint32_t dbi;
+
+ dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
+ addr = tcmu_get_block_addr(udev, dbi);
+ if (!addr)
+ return VM_FAULT_NOPAGE;
+ page = virt_to_page(addr);
+ }
+
get_page(page);
vmf->page = page;
return 0;
}
static const struct vm_operations_struct tcmu_vm_ops = {
+ .close = tcmu_vma_close,
.fault = tcmu_vma_fault,
};
@@ -927,7 +1066,7 @@ static int tcmu_configure_device(struct se_device *dev)
info->name = str;
- udev->mb_addr = vzalloc(TCMU_RING_SIZE);
+ udev->mb_addr = vzalloc(CMDR_SIZE);
if (!udev->mb_addr) {
ret = -ENOMEM;
goto err_vzalloc;
@@ -936,8 +1075,9 @@ static int tcmu_configure_device(struct se_device *dev)
/* mailbox fits in first part of CMDR space */
udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
udev->data_off = CMDR_SIZE;
- udev->data_size = TCMU_RING_SIZE - CMDR_SIZE;
+ udev->data_size = DATA_SIZE;
+ /* Initialise the mailbox of the ring buffer */
mb = udev->mb_addr;
mb->version = TCMU_MAILBOX_VERSION;
mb->flags = TCMU_MAILBOX_FLAG_CAP_OOOC;
@@ -948,12 +1088,14 @@ static int tcmu_configure_device(struct se_device *dev)
WARN_ON(udev->data_size % PAGE_SIZE);
WARN_ON(udev->data_size % DATA_BLOCK_SIZE);
+ INIT_RADIX_TREE(&udev->data_blocks, GFP_ATOMIC);
+
info->version = __stringify(TCMU_MAILBOX_VERSION);
info->mem[0].name = "tcm-user command & data buffer";
info->mem[0].addr = (phys_addr_t)(uintptr_t)udev->mb_addr;
info->mem[0].size = TCMU_RING_SIZE;
- info->mem[0].memtype = UIO_MEM_VIRTUAL;
+ info->mem[0].memtype = UIO_MEM_NONE;
info->irqcontrol = tcmu_irqcontrol;
info->irq = UIO_IRQ_CUSTOM;
@@ -1034,6 +1176,8 @@ static void tcmu_free_device(struct se_device *dev)
spin_unlock_irq(&udev->commands_lock);
WARN_ON(!all_expired);
+ tcmu_blocks_release(udev, true);
+
if (tcmu_dev_configured(udev)) {
tcmu_netlink_event(TCMU_CMD_REMOVED_DEVICE, udev->uio_info.name,
udev->uio_info.uio_dev->minor);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCHv3 2/5] tcmu: Add global data block pool support
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
2017-03-20 9:09 ` [PATCHv3 1/5] tcmu: Add dynamic growing data area feature support lixiubo
@ 2017-03-20 9:09 ` lixiubo
2017-03-20 9:09 ` [PATCHv3 3/5] target/user: Fix possible overwrite of t_data_sg's last iov[] lixiubo
` (3 subsequent siblings)
5 siblings, 0 replies; 9+ messages in thread
From: lixiubo @ 2017-03-20 9:09 UTC (permalink / raw)
To: agrover, nab, mchristi
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly, Xiubo Li
From: Xiubo Li <lixiubo@cmss.chinamobile.com>
For each target there will be one ring, when the target number
grows larger and larger, it could eventually runs out of the
system memories.
In this patch for each target ring, for the cmd area the size
will be limited to 8MB and for the data area the size will be
limited to 256K * PAGE_SIZE.
For all the targets' data areas, they will get empty blocks
from the "global data block pool", which has limited to 512K *
PAGE_SIZE for now.
When the "global data block pool" has been used up, then any
target could trigger the unmapping thread routine to shrink the
targets' rings. And for the idle targets the unmapping routine
will reserve 256 blocks at least.
When user space has touched the data blocks out of the iov[N],
the tcmu_page_fault() will return one zeroed blocks.
Signed-off-by: Xiubo Li <lixiubo@cmss.chinamobile.com>
Signed-off-by: Jianfei Hu <hujianfei@cmss.chinamobile.com>
---
drivers/target/target_core_user.c | 426 ++++++++++++++++++++++++++++++--------
1 file changed, 339 insertions(+), 87 deletions(-)
diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
index fc9f237..8cbe196 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -31,6 +31,8 @@
#include <linux/bitops.h>
#include <linux/highmem.h>
#include <linux/configfs.h>
+#include <linux/mutex.h>
+#include <linux/kthread.h>
#include <net/genetlink.h>
#include <scsi/scsi_common.h>
#include <scsi/scsi_proto.h>
@@ -67,17 +69,24 @@
#define TCMU_TIME_OUT (30 * MSEC_PER_SEC)
-/* For cmd area, the size is fixed 2M */
-#define CMDR_SIZE (2 * 1024 * 1024)
+/* For cmd area, the size is fixed 8MB */
+#define CMDR_SIZE (8 * 1024 * 1024)
-/* For data area, the size is fixed 32M */
-#define DATA_BLOCK_BITS (8 * 1024)
-#define DATA_BLOCK_SIZE 4096
+/*
+ * For data area, the block size is PAGE_SIZE and
+ * the total size is 256K * PAGE_SIZE.
+ */
+#define DATA_BLOCK_SIZE PAGE_SIZE
+#define DATA_BLOCK_BITS (256 * 1024)
#define DATA_SIZE (DATA_BLOCK_BITS * DATA_BLOCK_SIZE)
+#define DATA_BLOCK_RES_BITS 256
-/* The ring buffer size is 34M */
+/* The total size of the ring is 8M + 256K * PAGE_SIZE */
#define TCMU_RING_SIZE (CMDR_SIZE + DATA_SIZE)
+/* Default maximum of the global data blocks(512K * PAGE_SIZE) */
+#define TCMU_GLOBAL_MAX_BLOCKS (512 * 1024)
+
static struct device *tcmu_root_device;
struct tcmu_hba {
@@ -87,6 +96,8 @@ struct tcmu_hba {
#define TCMU_CONFIG_LEN 256
struct tcmu_dev {
+ struct list_head node;
+
struct se_device se_dev;
char *name;
@@ -98,6 +109,16 @@ struct tcmu_dev {
struct uio_info uio_info;
+ struct inode *inode;
+
+ struct mutex unmap_mutex;
+ bool unmapping;
+ bool waiting_global;
+ uint32_t dbi_cur;
+ uint32_t dbi_thresh;
+ DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
+ struct radix_tree_root data_blocks;
+
struct tcmu_mailbox *mb_addr;
size_t dev_size;
u32 cmdr_size;
@@ -111,10 +132,6 @@ struct tcmu_dev {
/* TODO should this be a mutex? */
spinlock_t cmdr_lock;
- uint32_t dbi_cur;
- DECLARE_BITMAP(data_bitmap, DATA_BLOCK_BITS);
- struct radix_tree_root data_blocks;
-
struct idr commands;
spinlock_t commands_lock;
@@ -146,6 +163,14 @@ struct tcmu_cmd {
unsigned long flags;
};
+static struct task_struct *unmap_thread;
+static wait_queue_head_t unmap_wait;
+static DEFINE_MUTEX(udev_mutex);
+static LIST_HEAD(root_udev);
+
+static spinlock_t db_count_lock;
+static unsigned long global_db_count;
+
static struct kmem_cache *tcmu_cmd_cache;
/* multicast group */
@@ -169,54 +194,90 @@ enum tcmu_multicast_groups {
.netnsok = true,
};
-static int tcmu_get_empty_block(struct tcmu_dev *udev, void **addr)
+#define tcmu_cmd_reset_dbi_cur(cmd) ((cmd)->dbi_cur = 0)
+#define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
+#define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
+
+static inline void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd, uint32_t len)
{
- void *p;
- uint32_t dbi;
- int ret;
+ struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
+ uint32_t i;
- dbi = find_first_zero_bit(udev->data_bitmap, DATA_BLOCK_BITS);
- if (dbi > udev->dbi_cur)
- udev->dbi_cur = dbi;
+ for (i = 0; i < len; i++)
+ clear_bit(tcmu_cmd->dbi[i], udev->data_bitmap);
+}
- set_bit(dbi, udev->data_bitmap);
+static inline bool tcmu_get_empty_block(struct tcmu_dev *udev,
+ struct tcmu_cmd *tcmu_cmd)
+{
+ struct page *page;
+ int ret, dbi;
- p = radix_tree_lookup(&udev->data_blocks, dbi);
- if (!p) {
- p = kzalloc(DATA_BLOCK_SIZE, GFP_ATOMIC);
- if (!p) {
- clear_bit(dbi, udev->data_bitmap);
- return -ENOMEM;
+ dbi = find_first_zero_bit(udev->data_bitmap, udev->dbi_thresh);
+ if (dbi == udev->dbi_thresh)
+ return false;
+
+ page = radix_tree_lookup(&udev->data_blocks, dbi);
+ if (!page) {
+ /* try to get new page from the mm */
+ spin_lock_irq(&db_count_lock);
+ if (global_db_count >= TCMU_GLOBAL_MAX_BLOCKS) {
+ spin_unlock_irq(&db_count_lock);
+ wake_up(&unmap_wait);
+ return false;
}
+ global_db_count++;
+ spin_unlock_irq(&db_count_lock);
+
+ page = alloc_page(GFP_ATOMIC);
+ if (!page)
+ return false;
- ret = radix_tree_insert(&udev->data_blocks, dbi, p);
+ ret = radix_tree_insert(&udev->data_blocks, dbi, page);
if (ret) {
- kfree(p);
- clear_bit(dbi, udev->data_bitmap);
- return ret;
+ __free_page(page);
+ return false;
}
}
- *addr = p;
- return dbi;
+ if (dbi > udev->dbi_cur)
+ udev->dbi_cur = dbi;
+
+ set_bit(dbi, udev->data_bitmap);
+ tcmu_cmd_set_dbi(tcmu_cmd, dbi);
+
+ return true;
}
-static void *tcmu_get_block_addr(struct tcmu_dev *udev, uint32_t dbi)
+static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
+ struct tcmu_cmd *tcmu_cmd)
{
- return radix_tree_lookup(&udev->data_blocks, dbi);
-}
+ int i;
-#define tcmu_cmd_reset_dbi_cur(cmd) ((cmd)->dbi_cur = 0)
-#define tcmu_cmd_set_dbi(cmd, index) ((cmd)->dbi[(cmd)->dbi_cur++] = (index))
-#define tcmu_cmd_get_dbi(cmd) ((cmd)->dbi[(cmd)->dbi_cur++])
+ tcmu_cmd_reset_dbi_cur(tcmu_cmd);
+ for (i = 0; i < tcmu_cmd->dbi_len; i++) {
+ if (!tcmu_get_empty_block(udev, tcmu_cmd))
+ goto err;
+ }
+ return true;
+
+err:
+ pr_debug("no blocks: only %u blocks available, but ask for %u\n",
+ tcmu_cmd->dbi_len, tcmu_cmd->dbi_cur);
+ tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
+ udev->waiting_global = true;
+ return false;
+}
-static void tcmu_cmd_free_data(struct tcmu_cmd *tcmu_cmd)
+static struct page *tcmu_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
{
- struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
- uint32_t bi;
+ struct page *page;
+
+ page = radix_tree_lookup(&udev->data_blocks, dbi);
+ if (!page)
+ return NULL;
- for (bi = 0; bi < tcmu_cmd->dbi_len; bi++)
- clear_bit(tcmu_cmd->dbi[bi], udev->data_bitmap);
+ return page;
}
static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
@@ -356,17 +417,20 @@ static int alloc_and_scatter_data_area(struct tcmu_dev *udev,
void *from, *to = NULL;
size_t copy_bytes, to_offset, offset;
struct scatterlist *sg;
+ struct page *page;
for_each_sg(data_sg, sg, data_nents, i) {
int sg_remaining = sg->length;
from = kmap_atomic(sg_page(sg)) + sg->offset;
while (sg_remaining > 0) {
if (block_remaining == 0) {
+ if (to)
+ kunmap_atomic(to);
+
block_remaining = DATA_BLOCK_SIZE;
- dbi = tcmu_get_empty_block(udev, &to);
- if (dbi < 0)
- return dbi;
- tcmu_cmd_set_dbi(tcmu_cmd, dbi);
+ dbi = tcmu_cmd_get_dbi(tcmu_cmd);
+ page = tcmu_get_block_page(udev, dbi);
+ to = kmap_atomic(page);
}
copy_bytes = min_t(size_t, sg_remaining,
@@ -394,6 +458,8 @@ static int alloc_and_scatter_data_area(struct tcmu_dev *udev,
}
kunmap_atomic(from - sg->offset);
}
+ if (to)
+ kunmap_atomic(to);
return 0;
}
@@ -403,18 +469,23 @@ static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
{
int i, dbi;
int block_remaining = 0;
- void *from, *to;
+ void *from = NULL, *to;
size_t copy_bytes, offset;
struct scatterlist *sg;
+ struct page *page;
for_each_sg(data_sg, sg, data_nents, i) {
int sg_remaining = sg->length;
to = kmap_atomic(sg_page(sg)) + sg->offset;
while (sg_remaining > 0) {
if (block_remaining == 0) {
+ if (from)
+ kunmap_atomic(from);
+
block_remaining = DATA_BLOCK_SIZE;
dbi = tcmu_cmd_get_dbi(tcmu_cmd);
- from = tcmu_get_block_addr(udev, dbi);
+ page = tcmu_get_block_page(udev, dbi);
+ from = kmap_atomic(page);
}
copy_bytes = min_t(size_t, sg_remaining,
block_remaining);
@@ -429,12 +500,13 @@ static void gather_data_area(struct tcmu_dev *udev, struct tcmu_cmd *tcmu_cmd,
}
kunmap_atomic(to - sg->offset);
}
+ if (from)
+ kunmap_atomic(from);
}
-static inline size_t spc_bitmap_free(unsigned long *bitmap)
+static inline size_t spc_bitmap_free(unsigned long *bitmap, uint32_t thresh)
{
- return DATA_BLOCK_SIZE * (DATA_BLOCK_BITS -
- bitmap_weight(bitmap, DATA_BLOCK_BITS));
+ return DATA_BLOCK_SIZE * (thresh - bitmap_weight(bitmap, thresh));
}
/*
@@ -443,12 +515,14 @@ static inline size_t spc_bitmap_free(unsigned long *bitmap)
*
* Called with ring lock held.
*/
-static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t data_needed)
+static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
+ size_t cmd_size, size_t data_needed)
{
struct tcmu_mailbox *mb = udev->mb_addr;
size_t space, cmd_needed;
u32 cmd_head;
+ udev->waiting_global = false;
tcmu_flush_dcache_range(mb, sizeof(*mb));
cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
@@ -469,13 +543,31 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t d
return false;
}
- space = spc_bitmap_free(udev->data_bitmap);
+ /* try to check and get the data blocks as needed */
+ space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
if (space < data_needed) {
- pr_debug("no data space: only %zu available, but ask for %zu\n",
- space, data_needed);
- return false;
+ uint32_t dbi_cnt;
+
+ if (udev->unmapping) {
+ pr_debug("the unmap thread is running, try it later");
+ return false;
+ }
+
+ dbi_cnt = (data_needed + DATA_BLOCK_SIZE - 1) /
+ DATA_BLOCK_BITS;
+ udev->dbi_thresh += max(udev->dbi_thresh / 2, dbi_cnt);
+ udev->dbi_thresh = min((int)udev->dbi_thresh, DATA_BLOCK_BITS);
+ space = spc_bitmap_free(udev->data_bitmap, udev->dbi_thresh);
+ if (space < data_needed) {
+ pr_debug("no data space: only %zu available, but ask for %zu\n",
+ space, data_needed);
+ return false;
+ }
}
+ if (!tcmu_get_empty_blocks(udev, cmd))
+ return false;
+
return true;
}
@@ -531,7 +623,7 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t d
return TCM_INVALID_CDB_FIELD;
}
- while (!is_ring_space_avail(udev, command_size, data_length)) {
+ while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
int ret;
DEFINE_WAIT(__wait);
@@ -583,6 +675,7 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, size_t cmd_size, size_t d
entry->hdr.uflags = 0;
/* Handle allocating space from the data area */
+ tcmu_cmd_reset_dbi_cur(tcmu_cmd);
iov = &entry->req.iov[0];
iov_cnt = 0;
copy_to_data_area = (se_cmd->data_direction == DMA_TO_DEVICE
@@ -663,10 +756,7 @@ static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *
* cmd has been completed already from timeout, just reclaim
* data area space and free cmd
*/
- free_data_area(udev, cmd);
-
- kmem_cache_free(tcmu_cmd_cache, cmd);
- return;
+ goto out;
}
tcmu_cmd_reset_dbi_cur(cmd);
@@ -694,8 +784,9 @@ static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *
target_complete_cmd(cmd->se_cmd, entry->rsp.scsi_status);
+out:
cmd->se_cmd = NULL;
- tcmu_cmd_free_data(cmd);
+ tcmu_cmd_free_data(cmd, cmd->dbi_len);
tcmu_free_cmd(cmd);
}
@@ -830,6 +921,8 @@ static struct se_device *tcmu_alloc_device(struct se_hba *hba, const char *name)
return NULL;
}
+ mutex_init(&udev->unmap_mutex);
+
udev->hba = hba;
udev->cmd_time_out = TCMU_TIME_OUT;
@@ -856,41 +949,53 @@ static int tcmu_irqcontrol(struct uio_info *info, s32 irq_on)
static void tcmu_blocks_release(struct tcmu_dev *udev, bool release_pending)
{
- uint32_t dbi, end;
- void *addr;
+ int dbi = -1, end;
+ struct page *page;
spin_lock_irq(&udev->cmdr_lock);
-
end = udev->dbi_cur + 1;
- /* try to release all unused blocks */
- dbi = find_first_zero_bit(udev->data_bitmap, end);
- if (dbi >= end) {
- spin_unlock_irq(&udev->cmdr_lock);
- return;
- }
+ /* Try to release all empty blocks */
do {
- addr = radix_tree_delete(&udev->data_blocks, dbi);
- kfree(addr);
-
dbi = find_next_zero_bit(udev->data_bitmap, end, dbi + 1);
- } while (dbi < end);
+ if (dbi == end)
+ break;
- if (!release_pending)
- return;
+ page = radix_tree_delete(&udev->data_blocks, dbi);
+ if (page) {
+ __free_page(page);
+ spin_lock_irq(&db_count_lock);
+ global_db_count--;
+ spin_unlock_irq(&db_count_lock);
+ }
+ } while (1);
- /* try to release all pending blocks */
- dbi = find_first_bit(udev->data_bitmap, end);
- if (dbi >= end) {
+ if (!release_pending) {
spin_unlock_irq(&udev->cmdr_lock);
return;
}
- do {
- addr = radix_tree_delete(&udev->data_blocks, dbi);
- kfree(addr);
+ /* Try to release all pending blocks */
+ dbi = -1;
+ do {
dbi = find_next_bit(udev->data_bitmap, end, dbi + 1);
- } while (dbi < end);
+ if (dbi == end)
+ break;
+
+ clear_bit(dbi, udev->data_bitmap);
+
+ page = radix_tree_delete(&udev->data_blocks, dbi);
+ if (page) {
+ __free_page(page);
+ spin_lock_irq(&db_count_lock);
+ global_db_count--;
+ spin_unlock_irq(&db_count_lock);
+ } else {
+ pr_err("block page not found, ring is broken\n");
+ set_bit(TCMU_DEV_BIT_BROKEN, &udev->flags);
+ break;
+ }
+ } while (1);
spin_unlock_irq(&udev->cmdr_lock);
}
@@ -919,6 +1024,53 @@ static int tcmu_find_mem_index(struct vm_area_struct *vma)
return -1;
}
+/*
+ * Normally it shouldn't be here. This is just for avoid
+ * the page fault call trace, and will return zeroed page.
+ */
+static struct page *tcmu_try_get_block_page(struct tcmu_dev *udev, uint32_t dbi)
+{
+ struct page *page;
+ int ret;
+
+ spin_lock_irq(&udev->cmdr_lock);
+ page = tcmu_get_block_page(udev, dbi);
+ if (page) {
+ spin_unlock_irq(&udev->cmdr_lock);
+ return page;
+ }
+
+ page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
+ if (!page) {
+ spin_unlock_irq(&udev->cmdr_lock);
+ return NULL;
+ }
+
+ if (dbi >= udev->dbi_thresh) {
+ /*
+ * Since this case is rare in page fault routine, here we
+ * will allow the global_db_count >= TCMU_GLOBAL_MAX_BLOCKS
+ * to reduce possible page fault call trace.
+ */
+ spin_lock_irq(&db_count_lock);
+ global_db_count++;
+ spin_unlock_irq(&db_count_lock);
+
+ udev->dbi_thresh = dbi;
+ udev->dbi_cur = dbi;
+ }
+
+ ret = radix_tree_insert(&udev->data_blocks, dbi, page);
+ if (ret) {
+ spin_unlock_irq(&udev->cmdr_lock);
+ __free_page(page);
+ return NULL;
+ }
+ spin_unlock_irq(&udev->cmdr_lock);
+
+ return page;
+}
+
static int tcmu_vma_fault(struct vm_fault *vmf)
{
struct tcmu_dev *udev = vmf->vma->vm_private_data;
@@ -942,14 +1094,15 @@ static int tcmu_vma_fault(struct vm_fault *vmf)
addr = (void *)(unsigned long)info->mem[mi].addr + offset;
page = vmalloc_to_page(addr);
} else {
- /* For the dynamically growing data area pages */
uint32_t dbi;
+ /* For the dynamically growing data area pages */
dbi = (offset - udev->data_off) / DATA_BLOCK_SIZE;
- addr = tcmu_get_block_addr(udev, dbi);
- if (!addr)
+ mutex_lock(&udev->unmap_mutex);
+ page = tcmu_try_get_block_page(udev, dbi);
+ mutex_unlock(&udev->unmap_mutex);
+ if (!page)
return VM_FAULT_NOPAGE;
- page = virt_to_page(addr);
}
get_page(page);
@@ -986,6 +1139,8 @@ static int tcmu_open(struct uio_info *info, struct inode *inode)
if (test_and_set_bit(TCMU_DEV_BIT_OPEN, &udev->flags))
return -EBUSY;
+ udev->inode = inode;
+
pr_debug("open\n");
return 0;
@@ -1076,6 +1231,8 @@ static int tcmu_configure_device(struct se_device *dev)
udev->cmdr_size = CMDR_SIZE - CMDR_OFF;
udev->data_off = CMDR_SIZE;
udev->data_size = DATA_SIZE;
+ udev->dbi_thresh = DATA_BLOCK_BITS;
+ udev->unmapping = false;
/* Initialise the mailbox of the ring buffer */
mb = udev->mb_addr;
@@ -1121,6 +1278,10 @@ static int tcmu_configure_device(struct se_device *dev)
if (ret)
goto err_netlink;
+ mutex_lock(&udev_mutex);
+ list_add(&udev->node, &root_udev);
+ mutex_unlock(&udev_mutex);
+
return 0;
err_netlink:
@@ -1164,6 +1325,10 @@ static void tcmu_free_device(struct se_device *dev)
del_timer_sync(&udev->timeout);
+ mutex_lock(&udev_mutex);
+ list_del(&udev->node);
+ mutex_unlock(&udev_mutex);
+
vfree(udev->mb_addr);
/* Upper layer should drain all requests before calling this */
@@ -1369,12 +1534,87 @@ static ssize_t tcmu_cmd_time_out_store(struct config_item *item, const char *pag
.tb_dev_attrib_attrs = NULL,
};
+static int unmap_thread_fn(void *data)
+{
+ struct tcmu_dev *udev;
+ loff_t off;
+ uint32_t start, end, dbi;
+ struct page *page;
+ int i;
+
+ while (1) {
+ DEFINE_WAIT(__wait);
+
+ prepare_to_wait(&unmap_wait, &__wait, TASK_INTERRUPTIBLE);
+ schedule();
+ finish_wait(&unmap_wait, &__wait);
+
+ mutex_lock(&udev_mutex);
+ list_for_each_entry(udev, &root_udev, node) {
+ mutex_lock(&udev->unmap_mutex);
+ spin_lock_irq(&udev->cmdr_lock);
+ end = udev->dbi_cur + 1;
+ dbi = find_last_bit(udev->data_bitmap, end);
+ if (dbi == end) {
+ /*
+ * Reserving about DATA_BLOCK_RES_BITS
+ * blocks for idle udev
+ */
+ dbi = DATA_BLOCK_RES_BITS - 1;
+ udev->dbi_cur = 0;
+ } else {
+ udev->dbi_cur = dbi;
+
+ if (dbi < DATA_BLOCK_RES_BITS - 1)
+ dbi = DATA_BLOCK_RES_BITS - 1;
+ }
+
+ udev->dbi_thresh = start = dbi + 1;
+ if (start >= end) {
+ spin_unlock_irq(&udev->cmdr_lock);
+ mutex_unlock(&udev->unmap_mutex);
+ continue;
+ }
+ udev->unmapping = true;
+ spin_unlock_irq(&udev->cmdr_lock);
+
+ /* Here will truncate the ring from offset */
+ off = udev->data_off + start * DATA_BLOCK_SIZE;
+ unmap_mapping_range(udev->inode->i_mapping, off, 0, 1);
+
+ spin_lock_irq(&udev->cmdr_lock);
+ for (i = start; i < end; i++) {
+ page = radix_tree_delete(&udev->data_blocks, i);
+ if (page) {
+ __free_page(page);
+ spin_lock_irq(&db_count_lock);
+ global_db_count--;
+ spin_unlock_irq(&db_count_lock);
+ }
+ }
+ udev->unmapping = false;
+ spin_unlock_irq(&udev->cmdr_lock);
+ mutex_unlock(&udev->unmap_mutex);
+ }
+
+ list_for_each_entry(udev, &root_udev, node) {
+ if (udev->waiting_global)
+ wake_up(&udev->wait_cmdr);
+ }
+ mutex_unlock(&udev_mutex);
+ }
+
+ return 0;
+}
+
static int __init tcmu_module_init(void)
{
int ret, i, len = 0;
BUILD_BUG_ON((sizeof(struct tcmu_cmd_entry) % TCMU_OP_ALIGN_SIZE) != 0);
+ spin_lock_init(&db_count_lock);
+
tcmu_cmd_cache = kmem_cache_create("tcmu_cmd_cache",
sizeof(struct tcmu_cmd),
__alignof__(struct tcmu_cmd),
@@ -1414,8 +1654,17 @@ static int __init tcmu_module_init(void)
if (ret)
goto out_attrs;
+ init_waitqueue_head(&unmap_wait);
+ unmap_thread = kthread_run(unmap_thread_fn, NULL, "tcmu_unmap");
+ if (IS_ERR(unmap_thread)) {
+ unmap_thread = NULL;
+ goto out_unreg_transport;
+ }
+
return 0;
+out_unreg_transport:
+ target_backend_unregister(&tcmu_ops);
out_attrs:
kfree(tcmu_attrs);
out_unreg_genl:
@@ -1430,6 +1679,9 @@ static int __init tcmu_module_init(void)
static void __exit tcmu_module_exit(void)
{
+ if (unmap_thread)
+ kthread_stop(unmap_thread);
+
target_backend_unregister(&tcmu_ops);
kfree(tcmu_attrs);
genl_unregister_family(&tcmu_genl_family);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCHv3 3/5] target/user: Fix possible overwrite of t_data_sg's last iov[]
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
2017-03-20 9:09 ` [PATCHv3 1/5] tcmu: Add dynamic growing data area feature support lixiubo
2017-03-20 9:09 ` [PATCHv3 2/5] tcmu: Add global data block pool support lixiubo
@ 2017-03-20 9:09 ` lixiubo
2017-03-21 19:10 ` Mike Christie
2017-03-20 9:09 ` [PATCHv3 4/5] target/user: Fix wrongly calculating of the base_command_size lixiubo
` (2 subsequent siblings)
5 siblings, 1 reply; 9+ messages in thread
From: lixiubo @ 2017-03-20 9:09 UTC (permalink / raw)
To: agrover, nab, mchristi
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly, Xiubo Li
From: Xiubo Li <lixiubo@cmss.chinamobile.com>
If there has BIDI data, its first iov[] will overwrite the last
iov[] for se_cmd->t_data_sg.
To fix this, we can just increase the iov pointer, but this may
introuduce a new memory leakage bug: If the se_cmd->data_length
and se_cmd->t_bidi_data_sg->length are all not aligned up to the
DATA_BLOCK_SIZE, the actual length needed maybe larger than just
sum of them.
So, this could be avoided by rounding all the data lengthes up
to DATA_BLOCK_SIZE.
Signed-off-by: Xiubo Li <lixiubo@cmss.chinamobile.com>
Signed-off-by: Bryant G. Ly <bryantly@linux.vnet.ibm.com>
---
drivers/target/target_core_user.c | 32 +++++++++++++++++++-------------
1 file changed, 19 insertions(+), 13 deletions(-)
diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
index 8cbe196..780c30f 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -288,13 +288,14 @@ static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
static inline uint32_t tcmu_cmd_get_dbi_len(struct se_cmd *se_cmd)
{
- size_t data_length = se_cmd->data_length;
+ size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
uint32_t dbi_len;
if (se_cmd->se_cmd_flags & SCF_BIDI)
- data_length += se_cmd->t_bidi_data_sg->length;
+ data_length += round_up(se_cmd->t_bidi_data_sg->length,
+ DATA_BLOCK_SIZE);
- dbi_len = (data_length + DATA_BLOCK_SIZE - 1) / DATA_BLOCK_SIZE;
+ dbi_len = data_length / DATA_BLOCK_SIZE;
return dbi_len;
}
@@ -609,10 +610,11 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
mb = udev->mb_addr;
cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
- data_length = se_cmd->data_length;
+ data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
if (se_cmd->se_cmd_flags & SCF_BIDI) {
BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
- data_length += se_cmd->t_bidi_data_sg->length;
+ data_length += round_up(se_cmd->t_bidi_data_sg->length,
+ DATA_BLOCK_SIZE);
}
if ((command_size > (udev->cmdr_size / 2)) ||
data_length > udev->data_size) {
@@ -690,15 +692,19 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
entry->req.iov_dif_cnt = 0;
/* Handle BIDI commands */
- iov_cnt = 0;
- ret = alloc_and_scatter_data_area(udev, tcmu_cmd,
- se_cmd->t_bidi_data_sg, se_cmd->t_bidi_data_nents,
- &iov, &iov_cnt, false);
- if (ret) {
- pr_err("tcmu: alloc and scatter bidi data failed\n");
- return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+ if (se_cmd->se_cmd_flags & SCF_BIDI) {
+ iov_cnt = 0;
+ iov++;
+ ret = alloc_and_scatter_data_area(udev, tcmu_cmd,
+ se_cmd->t_bidi_data_sg,
+ se_cmd->t_bidi_data_nents,
+ &iov, &iov_cnt, false);
+ if (ret) {
+ pr_err("tcmu: alloc and scatter bidi data failed\n");
+ return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+ }
+ entry->req.iov_bidi_cnt = iov_cnt;
}
- entry->req.iov_bidi_cnt = iov_cnt;
/* All offsets relative to mb_addr, not start of entry! */
cdb_off = CMDR_OFF + cmd_head + base_command_size;
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCHv3 4/5] target/user: Fix wrongly calculating of the base_command_size
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
` (2 preceding siblings ...)
2017-03-20 9:09 ` [PATCHv3 3/5] target/user: Fix possible overwrite of t_data_sg's last iov[] lixiubo
@ 2017-03-20 9:09 ` lixiubo
2017-03-20 9:09 ` [PATCHv3 5/5] target/user: Clean up tcmu_queue_cmd_ring lixiubo
2017-03-21 5:24 ` [PATCHv3 0/5] Dynamic growing data area support Nicholas A. Bellinger
5 siblings, 0 replies; 9+ messages in thread
From: lixiubo @ 2017-03-20 9:09 UTC (permalink / raw)
To: agrover, nab, mchristi
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly, Xiubo Li
From: Xiubo Li <lixiubo@cmss.chinamobile.com>
The t_data_nents and t_bidi_data_nents are all the numbers of the
segments, and we couldn't be sure the size of the data area block
will equal to size of the segment.
Use the actually block number needed intead of the sum of segments.
Signed-off-by: Xiubo Li <lixiubo@cmss.chinamobile.com>
---
drivers/target/target_core_user.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
index 780c30f..e3daf15 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -598,8 +598,7 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
* expensive to tell how many regions are freed in the bitmap
*/
base_command_size = max(offsetof(struct tcmu_cmd_entry,
- req.iov[se_cmd->t_bidi_data_nents +
- se_cmd->t_data_nents]),
+ req.iov[tcmu_cmd->dbi_len]),
sizeof(struct tcmu_cmd_entry));
command_size = base_command_size
+ round_up(scsi_command_size(se_cmd->t_task_cdb), TCMU_OP_ALIGN_SIZE);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread
* [PATCHv3 5/5] target/user: Clean up tcmu_queue_cmd_ring
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
` (3 preceding siblings ...)
2017-03-20 9:09 ` [PATCHv3 4/5] target/user: Fix wrongly calculating of the base_command_size lixiubo
@ 2017-03-20 9:09 ` lixiubo
2017-03-21 5:24 ` [PATCHv3 0/5] Dynamic growing data area support Nicholas A. Bellinger
5 siblings, 0 replies; 9+ messages in thread
From: lixiubo @ 2017-03-20 9:09 UTC (permalink / raw)
To: agrover, nab, mchristi
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly, Xiubo Li
From: Xiubo Li <lixiubo@cmss.chinamobile.com>
Add two helpers to simplify the code, and move some code out of
the cmdr spin lock to reduce the size of critical region.
Signed-off-by: Xiubo Li <lixiubo@cmss.chinamobile.com>
---
drivers/target/target_core_user.c | 68 +++++++++++++++++++++------------------
1 file changed, 37 insertions(+), 31 deletions(-)
diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
index e3daf15..52fa988 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -153,7 +153,7 @@ struct tcmu_cmd {
/* Can't use se_cmd when cleaning up expired cmds, because if
cmd has been completed then accessing se_cmd is off limits */
- uint32_t dbi_len;
+ uint32_t dbi_cnt;
uint32_t dbi_cur;
uint32_t *dbi;
@@ -255,7 +255,7 @@ static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
int i;
tcmu_cmd_reset_dbi_cur(tcmu_cmd);
- for (i = 0; i < tcmu_cmd->dbi_len; i++) {
+ for (i = 0; i < tcmu_cmd->dbi_cnt; i++) {
if (!tcmu_get_empty_block(udev, tcmu_cmd))
goto err;
}
@@ -263,7 +263,7 @@ static bool tcmu_get_empty_blocks(struct tcmu_dev *udev,
err:
pr_debug("no blocks: only %u blocks available, but ask for %u\n",
- tcmu_cmd->dbi_len, tcmu_cmd->dbi_cur);
+ tcmu_cmd->dbi_cnt, tcmu_cmd->dbi_cur);
tcmu_cmd_free_data(tcmu_cmd, tcmu_cmd->dbi_cur);
udev->waiting_global = true;
return false;
@@ -286,25 +286,29 @@ static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
}
-static inline uint32_t tcmu_cmd_get_dbi_len(struct se_cmd *se_cmd)
+static inline uint32_t tcmu_cmd_get_data_length(struct se_cmd *se_cmd)
{
size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
- uint32_t dbi_len;
if (se_cmd->se_cmd_flags & SCF_BIDI)
data_length += round_up(se_cmd->t_bidi_data_sg->length,
DATA_BLOCK_SIZE);
- dbi_len = data_length / DATA_BLOCK_SIZE;
+ return data_length;
+}
+
+static inline uint32_t tcmu_cmd_get_dbi_cnt(struct se_cmd *se_cmd)
+{
+ size_t data_length = tcmu_cmd_get_data_length(se_cmd);
- return dbi_len;
+ return data_length / DATA_BLOCK_SIZE;
}
static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
{
struct se_device *se_dev = se_cmd->se_dev;
struct tcmu_dev *udev = TCMU_DEV(se_dev);
- uint32_t dbi_len = tcmu_cmd_get_dbi_len(se_cmd);
+ uint32_t dbi_cnt = tcmu_cmd_get_dbi_cnt(se_cmd);
struct tcmu_cmd *tcmu_cmd;
int cmd_id;
@@ -319,8 +323,8 @@ static struct tcmu_cmd *tcmu_alloc_cmd(struct se_cmd *se_cmd)
msecs_to_jiffies(udev->cmd_time_out);
tcmu_cmd_reset_dbi_cur(tcmu_cmd);
- tcmu_cmd->dbi_len = dbi_len;
- tcmu_cmd->dbi = kcalloc(dbi_len, sizeof(uint32_t), GFP_KERNEL);
+ tcmu_cmd->dbi_cnt = dbi_cnt;
+ tcmu_cmd->dbi = kcalloc(dbi_cnt, sizeof(uint32_t), GFP_KERNEL);
if (!tcmu_cmd->dbi) {
kmem_cache_free(tcmu_cmd_cache, tcmu_cmd);
return NULL;
@@ -572,13 +576,27 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
return true;
}
+static inline void tcmu_cmd_get_cmd_size(struct tcmu_cmd *tcmu_cmd,
+ size_t *base_size, size_t *cmd_size)
+{
+ struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
+
+ *base_size = max(offsetof(struct tcmu_cmd_entry,
+ req.iov[tcmu_cmd->dbi_cnt]),
+ sizeof(struct tcmu_cmd_entry));
+
+ *cmd_size = round_up(scsi_command_size(se_cmd->t_task_cdb),
+ TCMU_OP_ALIGN_SIZE) + *base_size;
+ WARN_ON(*cmd_size & (TCMU_OP_ALIGN_SIZE-1));
+}
+
static sense_reason_t
tcmu_queue_cmd_ring(struct tcmu_cmd *tcmu_cmd)
{
struct tcmu_dev *udev = tcmu_cmd->tcmu_dev;
struct se_cmd *se_cmd = tcmu_cmd->se_cmd;
size_t base_command_size, command_size;
- struct tcmu_mailbox *mb;
+ struct tcmu_mailbox *mb = udev->mb_addr;
struct tcmu_cmd_entry *entry;
struct iovec *iov;
int iov_cnt, ret;
@@ -590,6 +608,8 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
if (test_bit(TCMU_DEV_BIT_BROKEN, &udev->flags))
return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
+ if (se_cmd->se_cmd_flags & SCF_BIDI)
+ BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
/*
* Must be a certain minimum size for response sense info, but
* also may be larger if the iov array is large.
@@ -597,33 +617,19 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
* We prepare way too many iovs for potential uses here, because it's
* expensive to tell how many regions are freed in the bitmap
*/
- base_command_size = max(offsetof(struct tcmu_cmd_entry,
- req.iov[tcmu_cmd->dbi_len]),
- sizeof(struct tcmu_cmd_entry));
- command_size = base_command_size
- + round_up(scsi_command_size(se_cmd->t_task_cdb), TCMU_OP_ALIGN_SIZE);
-
- WARN_ON(command_size & (TCMU_OP_ALIGN_SIZE-1));
-
- spin_lock_irq(&udev->cmdr_lock);
-
- mb = udev->mb_addr;
- cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
- data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
- if (se_cmd->se_cmd_flags & SCF_BIDI) {
- BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
- data_length += round_up(se_cmd->t_bidi_data_sg->length,
- DATA_BLOCK_SIZE);
- }
+ tcmu_cmd_get_cmd_size(tcmu_cmd, &base_command_size, &command_size);
+ data_length = tcmu_cmd_get_data_length(se_cmd);
if ((command_size > (udev->cmdr_size / 2)) ||
data_length > udev->data_size) {
pr_warn("TCMU: Request of size %zu/%zu is too big for %u/%zu "
"cmd ring/data area\n", command_size, data_length,
udev->cmdr_size, udev->data_size);
- spin_unlock_irq(&udev->cmdr_lock);
return TCM_INVALID_CDB_FIELD;
}
+ spin_lock_irq(&udev->cmdr_lock);
+
+ cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
while (!is_ring_space_avail(udev, tcmu_cmd, command_size, data_length)) {
int ret;
DEFINE_WAIT(__wait);
@@ -791,7 +797,7 @@ static void tcmu_handle_completion(struct tcmu_cmd *cmd, struct tcmu_cmd_entry *
out:
cmd->se_cmd = NULL;
- tcmu_cmd_free_data(cmd, cmd->dbi_len);
+ tcmu_cmd_free_data(cmd, cmd->dbi_cnt);
tcmu_free_cmd(cmd);
}
--
1.8.3.1
^ permalink raw reply related [flat|nested] 9+ messages in thread
* Re: [PATCHv3 0/5] Dynamic growing data area support
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
` (4 preceding siblings ...)
2017-03-20 9:09 ` [PATCHv3 5/5] target/user: Clean up tcmu_queue_cmd_ring lixiubo
@ 2017-03-21 5:24 ` Nicholas A. Bellinger
2017-03-21 5:34 ` Xiubo Li
5 siblings, 1 reply; 9+ messages in thread
From: Nicholas A. Bellinger @ 2017-03-21 5:24 UTC (permalink / raw)
To: lixiubo
Cc: agrover, mchristi, shli, sheng, linux-scsi, target-devel,
namei.unix, bryantly
Hi Xiubo !
On Mon, 2017-03-20 at 17:09 +0800, lixiubo@cmss.chinamobile.com wrote:
> From: Xiubo Li <lixiubo@cmss.chinamobile.com>
>
> Changed for V3:
> - [PATCHv2 2/5] fix double usage of blocks and possible page fault
> call trace.
> - [PATCHv2 5/5] fix a mistake.
>
> Changed for V2:
> - [PATCHv2 1/5] just fixes some small spelling and other mistakes.
> And as the initial patch, here sets cmd area to 8M and data area to
> 1G(1M fixed and 1023M growing)
> - [PATCHv2 2/5] is a new one, adding global data block pool support.
> The max total size of the pool is 2G and all the targets will get
> growing blocks from here.
> Test this using multi-targets at the same time.
> - [PATCHv2 3/5] changed nothing, respin it to avoid the conflict.
> - [PATCHv2 4/5] and [PATCHv2 5/5] are new ones.
>
>
> Xiubo Li (5):
> tcmu: Add dynamic growing data area feature support
> tcmu: Add global data block pool support
> target/user: Fix possible overwrite of t_data_sg's last iov[]
> target/user: Fix wrongly calculating of the base_command_size
> target/user: Clean up tcmu_queue_cmd_ring
>
> drivers/target/target_core_user.c | 621 +++++++++++++++++++++++++++++++-------
> 1 file changed, 514 insertions(+), 107 deletions(-)
>
I've not been following the details of your TCMU efforts, so will have
defer to MNC + AGrover for Acked-by + Reviewed-bys here. ;)
One comment on the series ordering though..
Patches #1 + #2 introduce new features, while patches #3 + #4 are
bug-fixes to (existing..?) code. AFAICT, patches #3 + #4 are
stand-alone that don't depend on the new features. Is that correct..?
If so, I'd prefer to apply #3 + #4 to target-pending/master for
v4.11-rcX (eg: bug-fixes only), and include new features in #1 + #2 and
cleanup in #5 to target-pending/for-next. (eg: next merge window for
v4.12-rc1).
Usually the preferred way when submitting patches is to always put
bug-fixes first in the series, followed by new features, further
cleanups, etc.
That way it's easy for a maintainer to split out / cherry-pick bug-fixes
from the series as necessary, without needing to worry about
dependencies in the earlier patches.
That said, if patch #3 + #4 are stand-alone bug-fixes, would you be so
kind to re-order them at the head of the series, and re-submit..?
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCHv3 0/5] Dynamic growing data area support
2017-03-21 5:24 ` [PATCHv3 0/5] Dynamic growing data area support Nicholas A. Bellinger
@ 2017-03-21 5:34 ` Xiubo Li
0 siblings, 0 replies; 9+ messages in thread
From: Xiubo Li @ 2017-03-21 5:34 UTC (permalink / raw)
To: Nicholas A. Bellinger
Cc: agrover, mchristi, shli, sheng, linux-scsi, target-devel,
namei.unix, bryantly
On 2017年03月21日 13:24, Nicholas A. Bellinger wrote:
> Hi Xiubo !
>
> On Mon, 2017-03-20 at 17:09 +0800, lixiubo@cmss.chinamobile.com wrote:
>> From: Xiubo Li <lixiubo@cmss.chinamobile.com>
>>
>> Changed for V3:
>> - [PATCHv2 2/5] fix double usage of blocks and possible page fault
>> call trace.
>> - [PATCHv2 5/5] fix a mistake.
>>
>> Changed for V2:
>> - [PATCHv2 1/5] just fixes some small spelling and other mistakes.
>> And as the initial patch, here sets cmd area to 8M and data area to
>> 1G(1M fixed and 1023M growing)
>> - [PATCHv2 2/5] is a new one, adding global data block pool support.
>> The max total size of the pool is 2G and all the targets will get
>> growing blocks from here.
>> Test this using multi-targets at the same time.
>> - [PATCHv2 3/5] changed nothing, respin it to avoid the conflict.
>> - [PATCHv2 4/5] and [PATCHv2 5/5] are new ones.
>>
>>
>> Xiubo Li (5):
>> tcmu: Add dynamic growing data area feature support
>> tcmu: Add global data block pool support
>> target/user: Fix possible overwrite of t_data_sg's last iov[]
>> target/user: Fix wrongly calculating of the base_command_size
>> target/user: Clean up tcmu_queue_cmd_ring
>>
>> drivers/target/target_core_user.c | 621 +++++++++++++++++++++++++++++++-------
>> 1 file changed, 514 insertions(+), 107 deletions(-)
>>
> I've not been following the details of your TCMU efforts, so will have
> defer to MNC + AGrover for Acked-by + Reviewed-bys here. ;)
>
> One comment on the series ordering though..
>
> Patches #1 + #2 introduce new features, while patches #3 + #4 are
> bug-fixes to (existing..?) code. AFAICT, patches #3 + #4 are
> stand-alone that don't depend on the new features. Is that correct..?
Yes, that's right.
> If so, I'd prefer to apply #3 + #4 to target-pending/master for
> v4.11-rcX (eg: bug-fixes only), and include new features in #1 + #2 and
> cleanup in #5 to target-pending/for-next. (eg: next merge window for
> v4.12-rc1).
>
> Usually the preferred way when submitting patches is to always put
> bug-fixes first in the series, followed by new features, further
> cleanups, etc.
>
> That way it's easy for a maintainer to split out / cherry-pick bug-fixes
> from the series as necessary, without needing to worry about
> dependencies in the earlier patches.
>
> That said, if patch #3 + #4 are stand-alone bug-fixes, would you be so
> kind to re-order them at the head of the series, and re-submit..?
>
Sure.
I will split the #3, #4 separately.
Thanks,
BRs
Xiubo
^ permalink raw reply [flat|nested] 9+ messages in thread
* Re: [PATCHv3 3/5] target/user: Fix possible overwrite of t_data_sg's last iov[]
2017-03-20 9:09 ` [PATCHv3 3/5] target/user: Fix possible overwrite of t_data_sg's last iov[] lixiubo
@ 2017-03-21 19:10 ` Mike Christie
0 siblings, 0 replies; 9+ messages in thread
From: Mike Christie @ 2017-03-21 19:10 UTC (permalink / raw)
To: lixiubo, agrover, nab
Cc: shli, sheng, linux-scsi, target-devel, namei.unix, bryantly
On 03/20/2017 05:09 AM, lixiubo@cmss.chinamobile.com wrote:
> From: Xiubo Li <lixiubo@cmss.chinamobile.com>
>
> If there has BIDI data, its first iov[] will overwrite the last
> iov[] for se_cmd->t_data_sg.
>
> To fix this, we can just increase the iov pointer, but this may
> introuduce a new memory leakage bug: If the se_cmd->data_length
> and se_cmd->t_bidi_data_sg->length are all not aligned up to the
> DATA_BLOCK_SIZE, the actual length needed maybe larger than just
> sum of them.
>
> So, this could be avoided by rounding all the data lengthes up
> to DATA_BLOCK_SIZE.
>
> Signed-off-by: Xiubo Li <lixiubo@cmss.chinamobile.com>
> Signed-off-by: Bryant G. Ly <bryantly@linux.vnet.ibm.com>
> ---
> drivers/target/target_core_user.c | 32 +++++++++++++++++++-------------
> 1 file changed, 19 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/target/target_core_user.c b/drivers/target/target_core_user.c
> index 8cbe196..780c30f 100644
> --- a/drivers/target/target_core_user.c
> +++ b/drivers/target/target_core_user.c
> @@ -288,13 +288,14 @@ static inline void tcmu_free_cmd(struct tcmu_cmd *tcmu_cmd)
>
> static inline uint32_t tcmu_cmd_get_dbi_len(struct se_cmd *se_cmd)
> {
> - size_t data_length = se_cmd->data_length;
> + size_t data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
> uint32_t dbi_len;
>
> if (se_cmd->se_cmd_flags & SCF_BIDI)
> - data_length += se_cmd->t_bidi_data_sg->length;
> + data_length += round_up(se_cmd->t_bidi_data_sg->length,
> + DATA_BLOCK_SIZE);
>
> - dbi_len = (data_length + DATA_BLOCK_SIZE - 1) / DATA_BLOCK_SIZE;
> + dbi_len = data_length / DATA_BLOCK_SIZE;
>
> return dbi_len;
> }
> @@ -609,10 +610,11 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
>
> mb = udev->mb_addr;
> cmd_head = mb->cmd_head % udev->cmdr_size; /* UAM */
> - data_length = se_cmd->data_length;
> + data_length = round_up(se_cmd->data_length, DATA_BLOCK_SIZE);
> if (se_cmd->se_cmd_flags & SCF_BIDI) {
> BUG_ON(!(se_cmd->t_bidi_data_sg && se_cmd->t_bidi_data_nents));
> - data_length += se_cmd->t_bidi_data_sg->length;
> + data_length += round_up(se_cmd->t_bidi_data_sg->length,
> + DATA_BLOCK_SIZE);
> }
> if ((command_size > (udev->cmdr_size / 2)) ||
> data_length > udev->data_size) {
> @@ -690,15 +692,19 @@ static bool is_ring_space_avail(struct tcmu_dev *udev, struct tcmu_cmd *cmd,
> entry->req.iov_dif_cnt = 0;
>
> /* Handle BIDI commands */
> - iov_cnt = 0;
> - ret = alloc_and_scatter_data_area(udev, tcmu_cmd,
> - se_cmd->t_bidi_data_sg, se_cmd->t_bidi_data_nents,
> - &iov, &iov_cnt, false);
> - if (ret) {
> - pr_err("tcmu: alloc and scatter bidi data failed\n");
> - return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
> + if (se_cmd->se_cmd_flags & SCF_BIDI) {
> + iov_cnt = 0;
> + iov++;
> + ret = alloc_and_scatter_data_area(udev, tcmu_cmd,
> + se_cmd->t_bidi_data_sg,
> + se_cmd->t_bidi_data_nents,
> + &iov, &iov_cnt, false);
> + if (ret) {
> + pr_err("tcmu: alloc and scatter bidi data failed\n");
> + return TCM_LOGICAL_UNIT_COMMUNICATION_FAILURE;
> + }
> + entry->req.iov_bidi_cnt = iov_cnt;
> }
> - entry->req.iov_bidi_cnt = iov_cnt;
>
> /* All offsets relative to mb_addr, not start of entry! */
> cdb_off = CMDR_OFF + cmd_head + base_command_size;
>
Looks ok to me. Thanks.
Reviewed-by: Mike Christie <mchristi@redhat.com>
^ permalink raw reply [flat|nested] 9+ messages in thread
end of thread, other threads:[~2017-03-21 19:10 UTC | newest]
Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-03-20 9:09 [PATCHv3 0/5] Dynamic growing data area support lixiubo
2017-03-20 9:09 ` [PATCHv3 1/5] tcmu: Add dynamic growing data area feature support lixiubo
2017-03-20 9:09 ` [PATCHv3 2/5] tcmu: Add global data block pool support lixiubo
2017-03-20 9:09 ` [PATCHv3 3/5] target/user: Fix possible overwrite of t_data_sg's last iov[] lixiubo
2017-03-21 19:10 ` Mike Christie
2017-03-20 9:09 ` [PATCHv3 4/5] target/user: Fix wrongly calculating of the base_command_size lixiubo
2017-03-20 9:09 ` [PATCHv3 5/5] target/user: Clean up tcmu_queue_cmd_ring lixiubo
2017-03-21 5:24 ` [PATCHv3 0/5] Dynamic growing data area support Nicholas A. Bellinger
2017-03-21 5:34 ` Xiubo Li
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.