f2fs: Fix deadlock between f2fs_quota_sync and block_operation
Subsystems (2)
PATCH 1/1 f2fs: Fix deadlock between f2fs_quota_sync and block_operation
This deadlock is hitting Android users (Pixel 3/3a/4) with Magisk, due to frequent umount/mount operations that trigger quota_sync, hitting the race. See https://github.com/topjohnwu/Magisk/issues/3171 for additional impact discussion.
In commit db6ec53b7e03, we added a semaphore to protect quota flags. As part of this commit, we changed f2fs_quota_sync to call f2fs_lock_op, in an attempt to prevent an AB/BA type deadlock with quota_sem locking in block_operation. However, rwsem in Linux is not recursive. Therefore, the following deadlock can occur:
f2fs_quota_syncdown_read(cp_rwsem) // f2fs_lock_opfilemap_fdatawritef2fs_write_data_pages
...
block_opertaion down_write(cp_rwsem) - marks rwsem as
"writer pending"down_read_trylock(cp_rwsem) - fails as there is
a writer pending.
Code keeps on trying, live-locking the filesystem.
We solve this by creating a new rwsem, used specifically to synchronize this case, instead of attempting to reuse an existing lock.
--- fs/f2fs/f2fs.h | 3 +++ fs/f2fs/super.c | 13 +++++++++++-- 2 files changed, 14 insertions(+), 2 deletions(-)
fs/f2fs/f2fs.h +3 -0
@@ -1448,6 +1448,7 @@ struct f2fs_sb_info {struct inode *meta_inode;/* cache meta blocks */ struct mutex cp_mutex;/* checkpoint procedure lock */ struct rw_semaphore cp_rwsem;/* blocking FS operations */ +struct rw_semaphore cp_quota_rwsem;/* blocking quota sync operations */ struct rw_semaphore node_write;/* locking node writes */ struct rw_semaphore node_change;/* locking node change */ wait_queue_head_t cp_wait;
@@ -1961,12 +1962,14 @@ static inline void f2fs_unlock_op(struct f2fs_sb_info *sbi)static inline void f2fs_lock_all (struct f2fs_sb_info *sbi) { +down_write (&sbi->cp_quota_rwsem);down_write (&sbi->cp_rwsem); }static inline void f2fs_unlock_all (struct f2fs_sb_info *sbi) {up_write (&sbi->cp_rwsem); +up_write (&sbi->cp_quota_rwsem); }static inline int __get_cp_reason (struct f2fs_sb_info *sbi)
fs/f2fs/super.c +11 -2
@@ -2209,8 +2209,16 @@ int f2fs_quota_sync(struct super_block *sb, int type)* f2fs_dquot_commit * block_operation * down_read(quota_sem) +* +* However, we cannot use the cp_rwsem to prevent this +* deadlock, as the cp_rwsem is taken for read inside the +* f2fs_dquot_commit code, and rwsem is not recursive. +* +* We therefore use a special lock to synchronize +* f2fs_quota_sync with block_operations, as this is the only +* place where such recursion occurs. */ -f2fs_lock_op (sbi); +down_read (&sbi->cp_quota_rwsem);down_read (&sbi->quota_sem); ret =dquot_writeback_dquots (sb, type);
@@ -2251,7 +2259,7 @@ int f2fs_quota_sync(struct super_block *sb, int type)if (ret)set_sbi_flag (F2FS_SB (sb),SBI_QUOTA_NEED_REPAIR );up_read (&sbi->quota_sem); -f2fs_unlock_op (sbi); +up_read (&sbi->cp_quota_rwsem);return ret; }
@@ -3599,6 +3607,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)init_rwsem (&sbi->cp_rwsem);init_rwsem (&sbi->quota_sem); +init_rwsem (&sbi->cp_quota_rwsem);init_waitqueue_head (&sbi->cp_wait);init_sb_info (sbi);
Signature · 8 lines
2.29.2 _______________________________________________ Linux-f2fs-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
Quoted 11 lines · Shachar Raindel
This deadlock is hitting Android users (Pixel 3/3a/4) with Magisk, due to frequent umount/mount operations that trigger
quota_sync, hitting the race. See https://github.com/topjohnwu/Magisk/issues/3171 for additional impact discussion.In commit
db6ec53b7e03, we added a semaphore to protect quota flags. As part of this commit, we changedf2fs_quota_syncto callf2fs_lock_op, in an attempt to prevent an AB/BA type deadlock withquota_semlocking inblock_operation. However, rwsem in Linux is not recursive. Therefore, the following deadlock can occur:
Quoted 12 lines
f2fs_quota_syncdown_read(cp_rwsem)//f2fs_lock_opfilemap_fdatawritef2fs_write_data_pages
...
block_opertaion
down_write(cp_rwsem)- marks rwsem as
"writer pending"down_read_trylock(cp_rwsem)- fails as there is
a writer pending.
Code keeps on trying, live-locking the filesystem.
f2fs_write_single_data_page() will not grab read lock of cp_rwsem now, could you please check f2fs code in mainline?
/* Dentry/quota blocks are controlled by checkpoint */
if (S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) {
/*
* We need to wait for node_write to avoid block allocation during
* checkpoint. This can only happen to quota writes which can cause
* the below discard race condition.
*/
if (IS_NOQUOTA(inode))
down_read(&sbi->node_write);
fio.need_lock = LOCK_DONE;
err = f2fs_do_write_data_page(&fio);
if (IS_NOQUOTA(inode))
up_read(&sbi->node_write);
goto done;
}Thanks,
Quoted 73 lines
We solve this by creating a new rwsem, used specifically to synchronize this case, instead of attempting to reuse an existing lock.
Signed-off-by Shachar Raindel <[email protected]>Fixes db6ec53b7e03 f2fs: add a rw_sem to cover quota flag changes--- fs/f2fs/f2fs.h | 3 +++ fs/f2fs/super.c | 13 +++++++++++-- 2 files changed, 14 insertions(+), 2 deletions(-)fs/f2fs/f2fs.h +3
-0@@ -1448,6 +1448,7 @@ struct f2fs_sb_info {struct inode *meta_inode;/* cache meta blocks */ struct mutex cp_mutex;/* checkpoint procedure lock */ struct rw_semaphore cp_rwsem;/* blocking FS operations */ +struct rw_semaphore cp_quota_rwsem;/* blocking quota sync operations */ struct rw_semaphore node_write;/* locking node writes */ struct rw_semaphore node_change;/* locking node change */ wait_queue_head_t cp_wait;@@ -1961,12 +1962,14 @@ static inline void f2fs_unlock_op(struct f2fs_sb_info *sbi)static inline void f2fs_lock_all (struct f2fs_sb_info *sbi) { +down_write (&sbi->cp_quota_rwsem);down_write (&sbi->cp_rwsem); }static inline void f2fs_unlock_all (struct f2fs_sb_info *sbi) {up_write (&sbi->cp_rwsem); +up_write (&sbi->cp_quota_rwsem); }static inline int __get_cp_reason (struct f2fs_sb_info *sbi)fs/f2fs/super.c +10
-2@@ -2209,8 +2209,16 @@ int f2fs_quota_sync(struct super_block *sb, int type)* f2fs_dquot_commit * block_operation * down_read(quota_sem) +* +* However, we cannot use the cp_rwsem to prevent this +* deadlock, as the cp_rwsem is taken for read inside the +* f2fs_dquot_commit code, and rwsem is not recursive. +* +* We therefore use a special lock to synchronize +* f2fs_quota_sync with block_operations, as this is the only +* place where such recursion occurs. */ -f2fs_lock_op (sbi); +down_read (&sbi->cp_quota_rwsem);down_read (&sbi->quota_sem); ret =dquot_writeback_dquots (sb, type);@@ -2251,7 +2259,7 @@ int f2fs_quota_sync(struct super_block *sb, int type)if (ret)set_sbi_flag (F2FS_SB (sb),SBI_QUOTA_NEED_REPAIR );up_read (&sbi->quota_sem); -f2fs_unlock_op (sbi); +up_read (&sbi->cp_quota_rwsem);return ret; }
fs/f2fs/super.c static int f2fs_fill_super(struct super_block *sb, void *data, int silent) · line 3613
Quoted 7 lines
fs/f2fs/super.c +1
-0@@ -3599,6 +3607,7 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)init_rwsem (&sbi->cp_rwsem);init_rwsem (&sbi->quota_sem); +init_rwsem (&sbi->cp_quota_rwsem);init_waitqueue_head (&sbi->cp_wait);init_sb_info (sbi);
_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel