race between __sync_single_inode() and iput()/bdev_clear_inode()
Hi,
EIP is at filemap_fdatawait+0xe/0x80
eax: e7461ad8 ebx: 00000000 ecx: 00000001 edx: 00000000
esi: e5334c40 edi: 6b6b6b6b ebp: de239e88 esp: de239e70
ds: 007b es: 007b ss: 0068
Process fsstress (pid: 31048, threadinfo=de239000 task=e60cd020)
Stack: e7d34358 de239e88 c01834f7 00000000 e5334c40 e7461ad8 de239eb0 c01836bf
e7461ad8 00000001 00000000 00000001 e7f0eac8 e5334c40 e7f0eac8 e7d2ddf4
de239f0c c0183777 e5334c40 de239f4c e5334c40 e7f0eac8 e7d2ddf4 de239f0c
Call Trace:
[show_stack+127/160] show_stack+0x7f/0xa0
[show_registers+351/464] show_registers+0x15f/0x1d0
[die+244/384] die+0xf4/0x180
[do_page_fault+886/1749] do_page_fault+0x376/0x6d5
[error_code+43/48] error_code+0x2b/0x30
[__sync_single_inode+447/512] __sync_single_inode+0x1bf/0x200
[__writeback_single_inode+119/352] __writeback_single_inode+0x77/0x160
[sync_sb_inodes+452/736] sync_sb_inodes+0x1c4/0x2e0
[sync_inodes_sb+126/144] sync_inodes_sb+0x7e/0x90
[sync_inodes+140/176] sync_inodes+0x8c/0xb0
[do_sync+90/144] do_sync+0x5a/0x90
[sys_sync+18/32] sys_sync+0x12/0x20
[syscall_call+7/11] syscall_call+0x7/0xb
Code: ab 89 1c 24 8b 75 e4 8b 45 ec 89 74 24 08 89 44 24 04 e8 16 fd ff ff eb 93 8d 74 26 00 55 89 e5 57 56 53 83 ec 0c 8b 45 08 8b 38 <8b> 97 5c 01 00 00 8d b6 00 00 00 00 8d bf 00 00 00 00 89 d0 f0Dump of assembler code for function filemap_fdatawait:
0xc013e290 <filemap_fdatawait+0>: push %ebp 0xc013e291 <filemap_fdatawait+1>: mov %esp,%ebp 0xc013e293 <filemap_fdatawait+3>: push %edi 0xc013e294 <filemap_fdatawait+4>: push %esi 0xc013e295 <filemap_fdatawait+5>: push %ebx 0xc013e296 <filemap_fdatawait+6>: sub $0xc,%esp 0xc013e299 <filemap_fdatawait+9>: mov 0x8(%ebp),%eax <- %eax is mapping 0xc013e29c <filemap_fdatawait+12>: mov (%eax),%edi <- %edi is mapping->host 0xc013e29e <filemap_fdatawait+14>: mov 0x15c(%edi),%edx <- Oops here 0xc013e2a4 <filemap_fdatawait+20>: lea 0x0(%esi),%esi
I got the above Oops while doing fs stress test.
The cause of this was race condition between __sync_single_inode() and iput()/bdev_clear_inode().
This race seems following condition.
cpu0 (fs's inode) cpu1 (bdev's inode)
------------------------------------------------------------------------
close("/dev/hda2")
[...]__sync_single_inode()
/* copy the bdev's ->i_mapping */
mapping = inode->i_mapping;
generic_forget_inode()
bdev_clear_inode()
/* restre the fs's ->i_mapping */
inode->i_mapping = &inode->i_data;
/* bdev's inode was freed */
destroy_inode(inode); if (wait) {
/* dereference a freed bdev's mapping->host */
filemap_fdatawait(mapping); /* Oops */I wrote the attached patch for making sure fs's inode is not in __sync_single_inode().
What do you think of this?
Thanks.
Signature · 118 lines
OGAWA Hirofumi <[email protected]> Signed-off-by: OGAWA Hirofumi <[email protected]> --- fs/block_dev.c | 27 ++++++++++++++++++++++++++- fs/fs-writeback.c | 34 ++++++++++++++++++++-------------- 2 files changed, 46 insertions(+), 15 deletions(-) diff -puN fs/block_dev.c~bdev-inode-sync fs/block_dev.c --- linux-2.6.12-rc1/fs/block_dev.c~bdev-inode-sync 2005-03-20 02:59:24.000000000 +0900 +++ linux-2.6.12-rc1-hirofumi/fs/block_dev.c 2005-03-20 03:59:54.000000000 +0900 @@ -23,6 +23,7 @@ #include <linux/mount.h> #include <linux/uio.h> #include <linux/namei.h> +#include <linux/writeback.h> #include <asm/uaccess.h> struct bdev_inode { @@ -282,11 +283,35 @@ static inline void __bd_forget(struct in static void bdev_clear_inode(struct inode *inode) { + extern void wait_inode_ilock(struct inode *inode); struct block_device *bdev = &BDEV_I(inode)->bdev; struct list_head *p; + struct inode *i; + spin_lock(&bdev_lock); while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { - __bd_forget(list_entry(p, struct inode, i_devices)); + inode = list_entry(p, struct inode, i_devices); + i = igrab(inode); + spin_unlock(&bdev_lock); + /* + * Preparation for changeing the ->i_mapping. Make + * sure this inode is not in __sync_single_inode(). + */ + if (i) { + spin_lock(&inode_lock); + wait_inode_ilock(i); + inode->i_state |= I_LOCK; + spin_unlock(&inode_lock); + } + spin_lock(&bdev_lock); + spin_lock(&inode_lock); + if (i) { + inode->i_state &= ~I_LOCK; + wake_up_inode(i); + iput(i); + } + __bd_forget(inode); + spin_unlock(&inode_lock); } list_del_init(&bdev->bd_list); spin_unlock(&bdev_lock); diff -puN fs/fs-writeback.c~bdev-inode-sync fs/fs-writeback.c --- linux-2.6.12-rc1/fs/fs-writeback.c~bdev-inode-sync 2005-03-20 02:59:24.000000000 +0900 +++ linux-2.6.12-rc1-hirofumi/fs/fs-writeback.c 2005-03-20 04:00:15.000000000 +0900 @@ -140,6 +140,25 @@ static int write_inode(struct inode *ino return 0; } +/* Called under inode_lock. */ +void wait_inode_ilock(struct inode *inode) +{ + wait_queue_head_t *wqh; + DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); + + if (!(inode->i_state & I_LOCK)) + return; + + wqh = bit_waitqueue(&inode->i_state, __I_LOCK); + do { + __iget(inode); + spin_unlock(&inode_lock); + __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); + iput(inode); + spin_lock(&inode_lock); + } while (inode->i_state & I_LOCK); +} + /* * Write a single inode's dirty pages and inode data out to disk. * If `wait' is set, wait on the writeout. @@ -244,8 +263,6 @@ static int __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) { - wait_queue_head_t *wqh; - if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_LOCK)) { list_move(&inode->i_list, &inode->i_sb->s_dirty); return 0; @@ -254,19 +271,8 @@ __writeback_single_inode(struct inode *i /* * It's a data-integrity sync. We must wait. */ - if (inode->i_state & I_LOCK) { - DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); + wait_inode_ilock(inode); - wqh = bit_waitqueue(&inode->i_state, __I_LOCK); - do { - __iget(inode); - spin_unlock(&inode_lock); - __wait_on_bit(wqh, &wq, inode_wait, - TASK_UNINTERRUPTIBLE); - iput(inode); - spin_lock(&inode_lock); - } while (inode->i_state & I_LOCK); - } return __sync_single_inode(inode, wbc); } _
Quoted 24 lines · OGAWA Hirofumi
...
I got the above Oops while doing fs stress test.The cause of this was race condition between
__sync_single_inode()and iput()/bdev_clear_inode().This race seems following condition.
cpu0 (fs's inode) cpu1 (bdev's inode)
------------------------------------------------------------------------close("/dev/hda2") [...]
__sync_single_inode()
/* copy the bdev's ->i_mapping */
mapping =inode->i_mapping;generic_forget_inode() bdev_clear_inode() /* restre the fs's ->i_mapping */ inode->i_mapping = &inode->i_data; /* bdev's inode was freed */ destroy_inode(inode);
Quoted 4 lines
if (wait) {
/* dereference a freed bdev's mapping->host */
filemap_fdatawait(mapping); /* Oops */The __sync_single_inode() caller takes a ref on the inode to prevent things like this from happening.
What was the call path on the other process? The one running destroy_inode()? unmount?
Quoted 2 lines
I wrote the attached patch for making sure fs's inode is not in
__sync_single_inode().
It would be nicer to honour the extra inode ref in the unmount path, if poss. Only swizzle i_mapping when the inode is really not in use.
Quoted 8 lines
(quoted diff) +8
-0+/* Called under inode_lock. */ +void wait_inode_ilock(struct inode *inode) +{ + wait_queue_head_t *wqh; + DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); + + if (!(inode->i_state & I_LOCK)) + return;
Review
Quoted 10 lines
(quoted diff) +10
-0+ + wqh = bit_waitqueue(&inode->i_state, __I_LOCK); + do { + __iget(inode); + spin_unlock(&inode_lock); + __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); + iput(inode); + spin_lock(&inode_lock); + } while (inode->i_state & I_LOCK); +}
Does this differ from wait_on_inode()?
Quoted 2 lines · Andrew Morton
The
__sync_single_inode()caller takes a ref on the inode to prevent things like this from happening.
Yes. But, in this case, that inode->i_mapping is pointing the bdev's ->i_mapping by open("/dev/hda2").
open("/dev/hda2") -> blkdev_open() -> bd_acquire()
inode->i_bdev = bdev; inode->i_mapping = bdev->bd_inode->i_mapping; list_add(&inode->i_devices, &bdev->bd_inodes);
In this race case, the inode is not freeing, but ->i_mapping is freed.
Quoted 2 lines
What was the call path on the other process? The one running
destroy_inode()? unmount?
close("/dev/hda2").
The _bdev's_ inode is freed by close() path after restoreing the inode->i_mapping of filesytem in bdev_clear_inode().
close("/dev/hda2")
-> blkdev_close()
-> [...]
-> iput()
-> generic_delete_inode()
-> bdev_clear_inode()
-> __bd_forget()
/* inode is filesystem's inode, not bdev. */
list_del_init(&inode->i_devices);
inode->i_bdev = NULL;
inode->i_mapping = &inode->i_data;
-> destroy_inode() <- is freeing the bdev's inode.And __sync_single_inode() side is updating the inode->i_atime on filesystem.
Quoted 12 lines
Quoted 12 lines
(quoted diff) +12
-0+/* Called under inode_lock. */ +void wait_inode_ilock(struct inode *inode) +{ + wait_queue_head_t *wqh; + DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); + + if (!(inode->i_state & I_LOCK)) + return; + + wqh = bit_waitqueue(&inode->i_state, __I_LOCK); + do { + __iget(inode);
Quoted 8 lines
Review
Quoted 6 lines
(quoted diff) +6
-0+ spin_unlock(&inode_lock); + __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); + iput(inode); + spin_lock(&inode_lock); + } while (inode->i_state & I_LOCK); +}Does this differ from
wait_on_inode()?
This checks I_LOCK after taking the inode_lock. So, caller can set the I_LOCK after this.
Signature · 1 line
OGAWA Hirofumi <[email protected]>
OGAWA Hirofumi <[email protected]> writes:
Grr.. that patch was calling iput() under inode_lock. Fixed patch is below.
Signature · 118 lines
OGAWA Hirofumi <[email protected]> Signed-off-by: OGAWA Hirofumi <[email protected]> --- fs/block_dev.c | 27 ++++++++++++++++++++++++++- fs/fs-writeback.c | 34 ++++++++++++++++++++-------------- 2 files changed, 46 insertions(+), 15 deletions(-) diff -puN fs/block_dev.c~bdev-inode-sync fs/block_dev.c --- linux-2.6.12-rc1/fs/block_dev.c~bdev-inode-sync 2005-03-20 04:49:31.000000000 +0900 +++ linux-2.6.12-rc1-hirofumi/fs/block_dev.c 2005-03-20 07:14:43.000000000 +0900 @@ -23,6 +23,7 @@ #include <linux/mount.h> #include <linux/uio.h> #include <linux/namei.h> +#include <linux/writeback.h> #include <asm/uaccess.h> struct bdev_inode { @@ -282,11 +283,35 @@ static inline void __bd_forget(struct in static void bdev_clear_inode(struct inode *inode) { + extern void wait_inode_ilock(struct inode *inode); struct block_device *bdev = &BDEV_I(inode)->bdev; struct list_head *p; + struct inode *i; + spin_lock(&bdev_lock); while ( (p = bdev->bd_inodes.next) != &bdev->bd_inodes ) { - __bd_forget(list_entry(p, struct inode, i_devices)); + inode = list_entry(p, struct inode, i_devices); + i = igrab(inode); + spin_unlock(&bdev_lock); + /* + * Preparation for changeing the ->i_mapping. Make + * sure this inode is not in __sync_single_inode(). + */ + if (i) { + spin_lock(&inode_lock); + wait_inode_ilock(i); + inode->i_state |= I_LOCK; + spin_unlock(&inode_lock); + } + spin_lock(&bdev_lock); + spin_lock(&inode_lock); + __bd_forget(inode); + if (i) { + inode->i_state &= ~I_LOCK; + wake_up_inode(i); + } + spin_unlock(&inode_lock); + iput(i); } list_del_init(&bdev->bd_list); spin_unlock(&bdev_lock); diff -puN fs/fs-writeback.c~bdev-inode-sync fs/fs-writeback.c --- linux-2.6.12-rc1/fs/fs-writeback.c~bdev-inode-sync 2005-03-20 04:49:31.000000000 +0900 +++ linux-2.6.12-rc1-hirofumi/fs/fs-writeback.c 2005-03-20 04:49:31.000000000 +0900 @@ -140,6 +140,25 @@ static int write_inode(struct inode *ino return 0; } +/* Called under inode_lock. */ +void wait_inode_ilock(struct inode *inode) +{ + wait_queue_head_t *wqh; + DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); + + if (!(inode->i_state & I_LOCK)) + return; + + wqh = bit_waitqueue(&inode->i_state, __I_LOCK); + do { + __iget(inode); + spin_unlock(&inode_lock); + __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); + iput(inode); + spin_lock(&inode_lock); + } while (inode->i_state & I_LOCK); +} + /* * Write a single inode's dirty pages and inode data out to disk. * If `wait' is set, wait on the writeout. @@ -244,8 +263,6 @@ static int __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) { - wait_queue_head_t *wqh; - if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_LOCK)) { list_move(&inode->i_list, &inode->i_sb->s_dirty); return 0; @@ -254,19 +271,8 @@ __writeback_single_inode(struct inode *i /* * It's a data-integrity sync. We must wait. */ - if (inode->i_state & I_LOCK) { - DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); + wait_inode_ilock(inode); - wqh = bit_waitqueue(&inode->i_state, __I_LOCK); - do { - __iget(inode); - spin_unlock(&inode_lock); - __wait_on_bit(wqh, &wq, inode_wait, - TASK_UNINTERRUPTIBLE); - iput(inode); - spin_lock(&inode_lock); - } while (inode->i_state & I_LOCK); - } return __sync_single_inode(inode, wbc); } _