Ext4 has a filesystem wide lock protecting ext4_writepages() calls to
avoid races with switching of journalled data flag or inode format. This
lock can however cause a deadlock like:
CPU0 CPU1
ext4_writepages()
percpu_down_read(sbi->s_writepages_rwsem);
ext4_change_inode_journal_flag()
percpu_down_write(sbi->s_writepages_rwsem);
- blocks, all readers block from now on
ext4_do_writepages()
ext4_init_io_end()
kmem_cache_zalloc(io_end_cachep, GFP_KERNEL)
fs_reclaim frees dentry...
dentry_unlink_inode()
iput() - last ref =>
iput_final() - inode dirty =>
write_inode_now()...
ext4_writepages() tries to acquire sbi->s_writepages_rwsem
and blocks forever
Make sure we cannot recurse into filesystem reclaim from writeback code
to avoid the deadlock.
Reported-by: [email protected]
Link: https://lore.kernel.org/all/[email protected]
Fixes: c8585c6fcaf2 ("ext4: fix races between changing inode journal mode and ext4_writepages")
CC: [email protected]
Signed-off-by: Jan Kara <[email protected]>
---
fs/ext4/ext4.h | 24 ++++++++++++++++++++++++
fs/ext4/inode.c | 18 ++++++++++--------
fs/ext4/migrate.c | 11 ++++++-----
3 files changed, 40 insertions(+), 13 deletions(-)
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 18cb2680dc39..92df861227dd 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1684,6 +1684,30 @@ static inline struct ext4_inode_info *EXT4_I(struct inode *inode)
return container_of(inode, struct ext4_inode_info, vfs_inode);
}
+static inline int ext4_writepages_down_read(struct super_block *sb)
+{
+ percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
+ return memalloc_nofs_save();
+}
+
+static inline void ext4_writepages_up_read(struct super_block *sb, int ctx)
+{
+ memalloc_nofs_restore(ctx);
+ percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
+}
+
+static inline int ext4_writepages_down_write(struct super_block *sb)
+{
+ percpu_down_write(&EXT4_SB(sb)->s_writepages_rwsem);
+ return memalloc_nofs_save();
+}
+
+static inline void ext4_writepages_up_write(struct super_block *sb, int ctx)
+{
+ memalloc_nofs_restore(ctx);
+ percpu_up_write(&EXT4_SB(sb)->s_writepages_rwsem);
+}
+
static inline int ext4_valid_inum(struct super_block *sb, unsigned long ino)
{
return ino == EXT4_ROOT_INO ||
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index ffbbd9626bd8..b6dc795b61ad 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2783,11 +2783,12 @@ static int ext4_writepages(struct address_space *mapping,
.can_map = 1,
};
int ret;
+ int alloc_ctx;
if (unlikely(ext4_forced_shutdown(EXT4_SB(sb))))
return -EIO;
- percpu_down_read(&EXT4_SB(sb)->s_writepages_rwsem);
+ alloc_ctx = ext4_writepages_down_read(sb);
ret = ext4_do_writepages(&mpd);
/*
* For data=journal writeback we could have come across pages marked
@@ -2796,7 +2797,7 @@ static int ext4_writepages(struct address_space *mapping,
*/
if (!ret && mpd.journalled_more_data)
ret = ext4_do_writepages(&mpd);
- percpu_up_read(&EXT4_SB(sb)->s_writepages_rwsem);
+ ext4_writepages_up_read(sb, alloc_ctx);
return ret;
}
@@ -2824,17 +2825,18 @@ static int ext4_dax_writepages(struct address_space *mapping,
long nr_to_write = wbc->nr_to_write;
struct inode *inode = mapping->host;
struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
+ int alloc_ctx;
if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
return -EIO;
- percpu_down_read(&sbi->s_writepages_rwsem);
+ alloc_ctx = ext4_writepages_down_read(inode->i_sb);
trace_ext4_writepages(inode, wbc);
ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
trace_ext4_writepages_result(inode, wbc, ret,
nr_to_write - wbc->nr_to_write);
- percpu_up_read(&sbi->s_writepages_rwsem);
+ ext4_writepages_up_read(inode->i_sb, alloc_ctx);
return ret;
}
@@ -5925,7 +5927,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
journal_t *journal;
handle_t *handle;
int err;
- struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
+ int alloc_ctx;
/*
* We have to be very careful here: changing a data block's
@@ -5963,7 +5965,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
}
}
- percpu_down_write(&sbi->s_writepages_rwsem);
+ alloc_ctx = ext4_writepages_down_write(inode->i_sb);
jbd2_journal_lock_updates(journal);
/*
@@ -5980,7 +5982,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
err = jbd2_journal_flush(journal, 0);
if (err < 0) {
jbd2_journal_unlock_updates(journal);
- percpu_up_write(&sbi->s_writepages_rwsem);
+ ext4_writepages_up_write(inode->i_sb, alloc_ctx);
return err;
}
ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
@@ -5988,7 +5990,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
ext4_set_aops(inode);
jbd2_journal_unlock_updates(journal);
- percpu_up_write(&sbi->s_writepages_rwsem);
+ ext4_writepages_up_write(inode->i_sb, alloc_ctx);
if (val)
filemap_invalidate_unlock(inode->i_mapping);
diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
index a19a9661646e..d98ac2af8199 100644
--- a/fs/ext4/migrate.c
+++ b/fs/ext4/migrate.c
@@ -408,7 +408,6 @@ static int free_ext_block(handle_t *handle, struct inode *inode)
int ext4_ext_migrate(struct inode *inode)
{
- struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
handle_t *handle;
int retval = 0, i;
__le32 *i_data;
@@ -418,6 +417,7 @@ int ext4_ext_migrate(struct inode *inode)
unsigned long max_entries;
__u32 goal, tmp_csum_seed;
uid_t owner[2];
+ int alloc_ctx;
/*
* If the filesystem does not support extents, or the inode
@@ -434,7 +434,7 @@ int ext4_ext_migrate(struct inode *inode)
*/
return retval;
- percpu_down_write(&sbi->s_writepages_rwsem);
+ alloc_ctx = ext4_writepages_down_write(inode->i_sb);
/*
* Worst case we can touch the allocation bitmaps and a block
@@ -586,7 +586,7 @@ int ext4_ext_migrate(struct inode *inode)
unlock_new_inode(tmp_inode);
iput(tmp_inode);
out_unlock:
- percpu_up_write(&sbi->s_writepages_rwsem);
+ ext4_writepages_up_write(inode->i_sb, alloc_ctx);
return retval;
}
@@ -605,6 +605,7 @@ int ext4_ind_migrate(struct inode *inode)
ext4_fsblk_t blk;
handle_t *handle;
int ret, ret2 = 0;
+ int alloc_ctx;
if (!ext4_has_feature_extents(inode->i_sb) ||
(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
@@ -621,7 +622,7 @@ int ext4_ind_migrate(struct inode *inode)
if (test_opt(inode->i_sb, DELALLOC))
ext4_alloc_da_blocks(inode);
- percpu_down_write(&sbi->s_writepages_rwsem);
+ alloc_ctx = ext4_writepages_down_write(inode->i_sb);
handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
if (IS_ERR(handle)) {
@@ -665,6 +666,6 @@ int ext4_ind_migrate(struct inode *inode)
ext4_journal_stop(handle);
up_write(&EXT4_I(inode)->i_data_sem);
out_unlock:
- percpu_up_write(&sbi->s_writepages_rwsem);
+ ext4_writepages_up_write(inode->i_sb, alloc_ctx);
return ret;
}
--
2.35.3
On Thu, 04 May 2023 14:47:23 +0200, Jan Kara wrote:
> Ext4 has a filesystem wide lock protecting ext4_writepages() calls to
> avoid races with switching of journalled data flag or inode format. This
> lock can however cause a deadlock like:
>
> CPU0 CPU1
>
> ext4_writepages()
> percpu_down_read(sbi->s_writepages_rwsem);
> ext4_change_inode_journal_flag()
> percpu_down_write(sbi->s_writepages_rwsem);
> - blocks, all readers block from now on
> ext4_do_writepages()
> ext4_init_io_end()
> kmem_cache_zalloc(io_end_cachep, GFP_KERNEL)
> fs_reclaim frees dentry...
> dentry_unlink_inode()
> iput() - last ref =>
> iput_final() - inode dirty =>
> write_inode_now()...
> ext4_writepages() tries to acquire sbi->s_writepages_rwsem
> and blocks forever
>
> [...]
Applied, thanks!
[1/1] ext4: Avoid deadlock in fs reclaim with page writeback
commit: 568e5b263e8bf81ffb575686f980bd18fdb7428f
Best regards,
--
Theodore Ts'o <[email protected]>