lkml.org 
[lkml]   [2018]   [Aug]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [f2fs-dev] [PATCH] f2fs: avoid fi->i_gc_rwsem[WRITE] lock in f2fs_gc
From
Date
On 2018/8/4 10:31, Chao Yu wrote:
> How about keep lock order as:
>
> - inode_lock
> - i_mmap_sem
> - lock_all()
> - unlock_all()
> - i_gc_rwsem[WRITE]
> - lock_op()

I got below warning when testing last dev-test:

- f2fs_direct_IO current lock dependency
- i_gc_rwsem[WRITE]
- i_mmap_sem
- do_blockdev_direct_IO
- i_mmap_sem
- i_gc_rwsem[WRITE]

So I guess still we should grab i_gc_rwsem[WRITE] lock before i_mmap_sem, any idea?

run fstests generic/208 at 2018-08-12 18:10:39

======================================================
WARNING: possible circular locking dependency detected
4.18.0-rc2+ #39 Tainted: G O
------------------------------------------------------
aio-dio-invalid/20621 is trying to acquire lock:
e47a5a00 (&mm->mmap_sem){++++}, at: get_user_pages_unlocked+0x38/0x1d0

but task is already holding lock:
82073b2b (&fi->i_gc_rwsem[WRITE]){++++}, at: f2fs_direct_IO+0x16c/0x590 [f2fs]

which lock already depends on the new lock.


the existing dependency chain (in reverse order) is:

-> #2 (&fi->i_gc_rwsem[WRITE]){++++}:
lock_acquire+0xae/0x220
down_write+0x38/0x60
f2fs_setattr+0x187/0x5b0 [f2fs]
notify_change+0x22b/0x400
do_truncate+0x5c/0x90
path_openat+0xaf5/0x1370
do_filp_open+0x5c/0xb0
do_sys_open+0xf8/0x1d0
sys_open+0x22/0x30
do_fast_syscall_32+0xaa/0x22c
entry_SYSENTER_32+0x53/0x86

-> #1 (&fi->i_mmap_sem){++++}:
lock_acquire+0xae/0x220
down_read+0x38/0x60
f2fs_filemap_fault+0x21/0x40 [f2fs]
__do_fault+0x16/0x30
handle_mm_fault+0xa37/0x10c0
__do_page_fault+0x19f/0x530
do_page_fault+0x20/0x280
common_exception+0x89/0x8e

-> #0 (&mm->mmap_sem){++++}:
__lock_acquire+0xe89/0x10e0
lock_acquire+0xae/0x220
down_read+0x38/0x60
get_user_pages_unlocked+0x38/0x1d0
get_user_pages_fast+0x70/0xe1
iov_iter_get_pages+0x94/0x250
do_blockdev_direct_IO+0x2191/0x25a0
__blockdev_direct_IO+0x4a/0x50
f2fs_direct_IO+0x332/0x590 [f2fs]
generic_file_direct_write+0xe9/0x2c0
__generic_file_write_iter+0x9a/0x1f0
f2fs_file_write_iter+0xdd/0x3b0 [f2fs]
aio_write.isra.20+0xe0/0x190
sys_io_submit+0x464/0x650
do_int80_syscall_32+0x6c/0x190
restore_all+0x0/0x6a

other info that might help us debug this:

Chain exists of:
&mm->mmap_sem --> &fi->i_mmap_sem --> &fi->i_gc_rwsem[WRITE]

Possible unsafe locking scenario:

CPU0 CPU1
---- ----
lock(&fi->i_gc_rwsem[WRITE]);
lock(&fi->i_mmap_sem);
lock(&fi->i_gc_rwsem[WRITE]);
lock(&mm->mmap_sem);

*** DEADLOCK ***

2 locks held by aio-dio-invalid/20621:
#0: ca54a0ec (&sb->s_type->i_mutex_key#17){+.+.}, at:
f2fs_file_write_iter+0x6f/0x3b0 [f2fs]
#1: 82073b2b (&fi->i_gc_rwsem[WRITE]){++++}, at: f2fs_direct_IO+0x16c/0x590 [f2fs]

stack backtrace:
CPU: 1 PID: 20621 Comm: aio-dio-invalid Tainted: G O 4.18.0-rc2+ #39
Hardware name: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
Call Trace:
dump_stack+0x5f/0x86
print_circular_bug.isra.35+0x1b6/0x1c0
check_prev_add.constprop.44+0x67a/0x6a0
__lock_acquire+0xe89/0x10e0
lock_acquire+0xae/0x220
? get_user_pages_unlocked+0x38/0x1d0
down_read+0x38/0x60
? get_user_pages_unlocked+0x38/0x1d0
get_user_pages_unlocked+0x38/0x1d0
? mark_held_locks+0x5d/0x80
? get_user_pages_fast+0xb7/0xe1
? trace_hardirqs_on_caller+0xdd/0x1c0
get_user_pages_fast+0x70/0xe1
iov_iter_get_pages+0x94/0x250
? lockdep_init_map+0x12/0x20
? __raw_spin_lock_init+0x31/0x60
do_blockdev_direct_IO+0x2191/0x25a0
? __blockdev_direct_IO+0x4a/0x50
? __this_cpu_preempt_check+0xf/0x20
? free_unref_page_list+0x1c7/0x2a0
? trace_hardirqs_on_caller+0xdd/0x1c0
? __get_data_block+0xc0/0xc0 [f2fs]
__blockdev_direct_IO+0x4a/0x50
? __get_data_block+0xc0/0xc0 [f2fs]
f2fs_direct_IO+0x332/0x590 [f2fs]
? __get_data_block+0xc0/0xc0 [f2fs]
generic_file_direct_write+0xe9/0x2c0
__generic_file_write_iter+0x9a/0x1f0
f2fs_file_write_iter+0xdd/0x3b0 [f2fs]
aio_write.isra.20+0xe0/0x190
? sys_io_submit+0x1ab/0x650
sys_io_submit+0x464/0x650
? sys_io_submit+0x13a/0x650
do_int80_syscall_32+0x6c/0x190
entry_INT80_32+0x36/0x36


>
> Thanks,
>
>>
>> From f6341121ee0c07fa834960a7c86cb0ea3f824231 Mon Sep 17 00:00:00 2001
>> From: Jaegeuk Kim <jaegeuk@kernel.org>
>> Date: Wed, 25 Jul 2018 12:11:56 +0900
>> Subject: [PATCH] f2fs: avoid fi->i_gc_rwsem[WRITE] lock in f2fs_gc
>>
>> The f2fs_gc() called by f2fs_balance_fs() requires to be called outside of
>> fi->i_gc_rwsem[WRITE], since f2fs_gc() can try to grab it in a loop.
>>
>> If it hits the miximum retrials in GC, let's give a chance to release
>> gc_mutex for a short time in order not to go into live lock in the worst
>> case.
>>
>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
>> ---
>> v2:
>> - add rwsem on start_atomic_write
>>
>> fs/f2fs/f2fs.h | 1 +
>> fs/f2fs/file.c | 71 ++++++++++++++++++++++++-----------------------
>> fs/f2fs/gc.c | 22 +++++++++++----
>> fs/f2fs/segment.c | 5 +++-
>> fs/f2fs/segment.h | 2 +-
>> 5 files changed, 58 insertions(+), 43 deletions(-)
>>
>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>> index a9447c7d6570..50349780001b 100644
>> --- a/fs/f2fs/f2fs.h
>> +++ b/fs/f2fs/f2fs.h
>> @@ -1223,6 +1223,7 @@ struct f2fs_sb_info {
>> unsigned int gc_mode; /* current GC state */
>> /* for skip statistic */
>> unsigned long long skipped_atomic_files[2]; /* FG_GC and BG_GC */
>> + unsigned long long skipped_gc_rwsem; /* FG_GC only */
>>
>> /* threshold for gc trials on pinned files */
>> u64 gc_pin_file_threshold;
>> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
>> index 78c1bd6b8497..a960869bf60f 100644
>> --- a/fs/f2fs/file.c
>> +++ b/fs/f2fs/file.c
>> @@ -1179,10 +1179,12 @@ static int __exchange_data_block(struct inode *src_inode,
>> return ret;
>> }
>>
>> -static int f2fs_do_collapse(struct inode *inode, pgoff_t start, pgoff_t end)
>> +static int f2fs_do_collapse(struct inode *inode, loff_t offset, loff_t len)
>> {
>> struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
>> pgoff_t nrpages = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
>> + pgoff_t start = offset >> PAGE_SHIFT;
>> + pgoff_t end = (offset + len) >> PAGE_SHIFT;
>> int ret;
>>
>> f2fs_balance_fs(sbi, true);
>> @@ -1190,14 +1192,18 @@ static int f2fs_do_collapse(struct inode *inode, pgoff_t start, pgoff_t end)
>>
>> f2fs_drop_extent_tree(inode);
>>
>> + /* avoid gc operation during block exchange */
>> + down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> + truncate_pagecache(inode, offset);
>> ret = __exchange_data_block(inode, inode, end, start, nrpages - end, true);
>> + up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> +
>> f2fs_unlock_op(sbi);
>> return ret;
>> }
>>
>> static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len)
>> {
>> - pgoff_t pg_start, pg_end;
>> loff_t new_size;
>> int ret;
>>
>> @@ -1212,21 +1218,13 @@ static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len)
>> if (ret)
>> return ret;
>>
>> - pg_start = offset >> PAGE_SHIFT;
>> - pg_end = (offset + len) >> PAGE_SHIFT;
>> -
>> - /* avoid gc operation during block exchange */
>> - down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> -
>> down_write(&F2FS_I(inode)->i_mmap_sem);
>> /* write out all dirty pages from offset */
>> ret = filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
>> if (ret)
>> goto out_unlock;
>>
>> - truncate_pagecache(inode, offset);
>> -
>> - ret = f2fs_do_collapse(inode, pg_start, pg_end);
>> + ret = f2fs_do_collapse(inode, offset, len);
>> if (ret)
>> goto out_unlock;
>>
>> @@ -1242,7 +1240,6 @@ static int f2fs_collapse_range(struct inode *inode, loff_t offset, loff_t len)
>> f2fs_i_size_write(inode, new_size);
>> out_unlock:
>> up_write(&F2FS_I(inode)->i_mmap_sem);
>> - up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> return ret;
>> }
>>
>> @@ -1417,9 +1414,6 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
>>
>> f2fs_balance_fs(sbi, true);
>>
>> - /* avoid gc operation during block exchange */
>> - down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> -
>> down_write(&F2FS_I(inode)->i_mmap_sem);
>> ret = f2fs_truncate_blocks(inode, i_size_read(inode), true);
>> if (ret)
>> @@ -1430,13 +1424,15 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
>> if (ret)
>> goto out;
>>
>> - truncate_pagecache(inode, offset);
>> -
>> pg_start = offset >> PAGE_SHIFT;
>> pg_end = (offset + len) >> PAGE_SHIFT;
>> delta = pg_end - pg_start;
>> idx = (i_size_read(inode) + PAGE_SIZE - 1) / PAGE_SIZE;
>>
>> + /* avoid gc operation during block exchange */
>> + down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> + truncate_pagecache(inode, offset);
>> +
>> while (!ret && idx > pg_start) {
>> nr = idx - pg_start;
>> if (nr > delta)
>> @@ -1450,6 +1446,7 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
>> idx + delta, nr, false);
>> f2fs_unlock_op(sbi);
>> }
>> + up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>>
>> /* write out all moved pages, if possible */
>> filemap_write_and_wait_range(inode->i_mapping, offset, LLONG_MAX);
>> @@ -1459,7 +1456,6 @@ static int f2fs_insert_range(struct inode *inode, loff_t offset, loff_t len)
>> f2fs_i_size_write(inode, new_size);
>> out:
>> up_write(&F2FS_I(inode)->i_mmap_sem);
>> - up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> return ret;
>> }
>>
>> @@ -1706,8 +1702,6 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>>
>> inode_lock(inode);
>>
>> - down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> -
>> if (f2fs_is_atomic_file(inode)) {
>> if (is_inode_flag_set(inode, FI_ATOMIC_REVOKE_REQUEST))
>> ret = -EINVAL;
>> @@ -1718,6 +1712,8 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>> if (ret)
>> goto out;
>>
>> + down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> +
>> if (!get_dirty_pages(inode))
>> goto skip_flush;
>>
>> @@ -1725,18 +1721,20 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>> "Unexpected flush for atomic writes: ino=%lu, npages=%u",
>> inode->i_ino, get_dirty_pages(inode));
>> ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
>> - if (ret)
>> + if (ret) {
>> + up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> goto out;
>> + }
>> skip_flush:
>> set_inode_flag(inode, FI_ATOMIC_FILE);
>> clear_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
>> - f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
>> + up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>>
>> + f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
>> F2FS_I(inode)->inmem_task = current;
>> stat_inc_atomic_write(inode);
>> stat_update_max_atomic_write(inode);
>> out:
>> - up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> inode_unlock(inode);
>> mnt_drop_write_file(filp);
>> return ret;
>> @@ -1754,9 +1752,9 @@ static int f2fs_ioc_commit_atomic_write(struct file *filp)
>> if (ret)
>> return ret;
>>
>> - inode_lock(inode);
>> + f2fs_balance_fs(F2FS_I_SB(inode), true);
>>
>> - down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> + inode_lock(inode);
>>
>> if (f2fs_is_volatile_file(inode)) {
>> ret = -EINVAL;
>> @@ -1782,7 +1780,6 @@ static int f2fs_ioc_commit_atomic_write(struct file *filp)
>> clear_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
>> ret = -EINVAL;
>> }
>> - up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> inode_unlock(inode);
>> mnt_drop_write_file(filp);
>> return ret;
>> @@ -2378,15 +2375,10 @@ static int f2fs_move_file_range(struct file *file_in, loff_t pos_in,
>> }
>>
>> inode_lock(src);
>> - down_write(&F2FS_I(src)->i_gc_rwsem[WRITE]);
>> if (src != dst) {
>> ret = -EBUSY;
>> if (!inode_trylock(dst))
>> goto out;
>> - if (!down_write_trylock(&F2FS_I(dst)->i_gc_rwsem[WRITE])) {
>> - inode_unlock(dst);
>> - goto out;
>> - }
>> }
>>
>> ret = -EINVAL;
>> @@ -2432,6 +2424,14 @@ static int f2fs_move_file_range(struct file *file_in, loff_t pos_in,
>>
>> f2fs_balance_fs(sbi, true);
>> f2fs_lock_op(sbi);
>> +
>> + down_write(&F2FS_I(src)->i_gc_rwsem[WRITE]);
>> + if (src != dst) {
>> + ret = -EBUSY;
>> + if (!down_write_trylock(&F2FS_I(dst)->i_gc_rwsem[WRITE]))
>> + goto out_src;
>> + }
>> +
>> ret = __exchange_data_block(src, dst, pos_in >> F2FS_BLKSIZE_BITS,
>> pos_out >> F2FS_BLKSIZE_BITS,
>> len >> F2FS_BLKSIZE_BITS, false);
>> @@ -2442,14 +2442,15 @@ static int f2fs_move_file_range(struct file *file_in, loff_t pos_in,
>> else if (dst_osize != dst->i_size)
>> f2fs_i_size_write(dst, dst_osize);
>> }
>> + if (src != dst)
>> + up_write(&F2FS_I(dst)->i_gc_rwsem[WRITE]);
>> +out_src:
>> + up_write(&F2FS_I(src)->i_gc_rwsem[WRITE]);
>> f2fs_unlock_op(sbi);
>> out_unlock:
>> - if (src != dst) {
>> - up_write(&F2FS_I(dst)->i_gc_rwsem[WRITE]);
>> + if (src != dst)
>> inode_unlock(dst);
>> - }
>> out:
>> - up_write(&F2FS_I(src)->i_gc_rwsem[WRITE]);
>> inode_unlock(src);
>> return ret;
>> }
>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>> index e352fbd33848..cac317e37306 100644
>> --- a/fs/f2fs/gc.c
>> +++ b/fs/f2fs/gc.c
>> @@ -884,6 +884,7 @@ static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
>> if (!down_write_trylock(
>> &F2FS_I(inode)->i_gc_rwsem[WRITE])) {
>> iput(inode);
>> + sbi->skipped_gc_rwsem++;
>> continue;
>> }
>>
>> @@ -913,6 +914,7 @@ static void gc_data_segment(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
>> continue;
>> if (!down_write_trylock(
>> &fi->i_gc_rwsem[WRITE])) {
>> + sbi->skipped_gc_rwsem++;
>> up_write(&fi->i_gc_rwsem[READ]);
>> continue;
>> }
>> @@ -1062,6 +1064,7 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
>> prefree_segments(sbi));
>>
>> cpc.reason = __get_cp_reason(sbi);
>> + sbi->skipped_gc_rwsem = 0;
>> gc_more:
>> if (unlikely(!(sbi->sb->s_flags & SB_ACTIVE))) {
>> ret = -EINVAL;
>> @@ -1103,7 +1106,8 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
>> total_freed += seg_freed;
>>
>> if (gc_type == FG_GC) {
>> - if (sbi->skipped_atomic_files[FG_GC] > last_skipped)
>> + if (sbi->skipped_atomic_files[FG_GC] > last_skipped ||
>> + sbi->skipped_gc_rwsem)
>> skipped_round++;
>> last_skipped = sbi->skipped_atomic_files[FG_GC];
>> round++;
>> @@ -1112,15 +1116,21 @@ int f2fs_gc(struct f2fs_sb_info *sbi, bool sync,
>> if (gc_type == FG_GC)
>> sbi->cur_victim_sec = NULL_SEGNO;
>>
>> - if (!sync) {
>> - if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
>> - if (skipped_round > MAX_SKIP_ATOMIC_COUNT &&
>> - skipped_round * 2 >= round)
>> - f2fs_drop_inmem_pages_all(sbi, true);
>> + if (sync)
>> + goto stop;
>> +
>> + if (has_not_enough_free_secs(sbi, sec_freed, 0)) {
>> + if (skipped_round <= MAX_SKIP_GC_COUNT ||
>> + skipped_round * 2 < round) {
>> segno = NULL_SEGNO;
>> goto gc_more;
>> }
>>
>> + if (sbi->skipped_atomic_files[FG_GC] == last_skipped) {
>> + f2fs_drop_inmem_pages_all(sbi, true);
>> + segno = NULL_SEGNO;
>> + goto gc_more;
>> + }
>> if (gc_type == FG_GC)
>> ret = f2fs_write_checkpoint(sbi, &cpc);
>> }
>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>> index 3662e1f429b4..15b3b095fd58 100644
>> --- a/fs/f2fs/segment.c
>> +++ b/fs/f2fs/segment.c
>> @@ -444,10 +444,12 @@ int f2fs_commit_inmem_pages(struct inode *inode)
>> struct f2fs_inode_info *fi = F2FS_I(inode);
>> int err;
>>
>> - f2fs_balance_fs(sbi, true);
>> + f2fs_balance_fs(F2FS_I_SB(inode), true);
>> +
>> f2fs_lock_op(sbi);
>>
>> set_inode_flag(inode, FI_ATOMIC_COMMIT);
>> + down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>>
>> mutex_lock(&fi->inmem_lock);
>> err = __f2fs_commit_inmem_pages(inode);
>> @@ -458,6 +460,7 @@ int f2fs_commit_inmem_pages(struct inode *inode)
>> spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
>> mutex_unlock(&fi->inmem_lock);
>>
>> + up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>> clear_inode_flag(inode, FI_ATOMIC_COMMIT);
>>
>> f2fs_unlock_op(sbi);
>> diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
>> index 50495515f0a0..b3d9e317ff0c 100644
>> --- a/fs/f2fs/segment.h
>> +++ b/fs/f2fs/segment.h
>> @@ -215,7 +215,7 @@ struct segment_allocation {
>> #define IS_DUMMY_WRITTEN_PAGE(page) \
>> (page_private(page) == (unsigned long)DUMMY_WRITTEN_PAGE)
>>
>> -#define MAX_SKIP_ATOMIC_COUNT 16
>> +#define MAX_SKIP_GC_COUNT 16
>>
>> struct inmem_pages {
>> struct list_head list;
>>
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> Linux-f2fs-devel mailing list
> Linux-f2fs-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
>

\
 
 \ /
  Last update: 2018-08-12 12:24    [W:0.125 / U:0.328 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site