Re: [RFC PATCH] f2fs: don't set SBI_QUOTA_NEED_REPAIR flag if receive SIGKILL

From: Chao Yu
Date: Mon Sep 02 2024 - 03:32:44 EST


On 2024/8/27 14:22, wangzijie wrote:
Thread A
-dquot_initialize
-dqget
-f2fs_dquot_acquire
-v2_read_dquot
-qtree_read_dquot
-find_tree_dqentry
-f2fs_quota_read
-read_cache_page_gfp
-do_read_cache_folio
-fiemap_read_folio
-folio_wait_locked_killable
-receive SIGKILL : return -EINTR
-set SBI_QUOTA_NEED_REPAIR
-set SBI_QUOTA_NEED_REPAIR

When calling read_cache_page_gfp in quota read, thread may receive SIGKILL and
set SBI_QUOTA_NEED_REPAIR, should we set SBI_QUOTA_NEED_REPAIR in this error path?

f2fs_quota_read() can be called in a lot of contexts, can we just ignore -EINTR
for f2fs_dquot_initialize() case?

Thanks,


Signed-off-by: wangzijie <wangzijie1@xxxxxxxxx>
---
fs/f2fs/inode.c | 3 ++-
fs/f2fs/super.c | 6 +++---
2 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/fs/f2fs/inode.c b/fs/f2fs/inode.c
index ed629dabb..2af98e2b7 100644
--- a/fs/f2fs/inode.c
+++ b/fs/f2fs/inode.c
@@ -837,8 +837,9 @@ void f2fs_evict_inode(struct inode *inode)
err = f2fs_dquot_initialize(inode);
if (err) {
+ if (err != -EINTR)
+ set_sbi_flag(sbi, SBI_QUOTA_NEED_REPAIR);
err = 0;
- set_sbi_flag(sbi, SBI_QUOTA_NEED_REPAIR);
}
f2fs_remove_ino_entry(sbi, inode->i_ino, APPEND_INO);
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 1f1b3647a..f99a36ff3 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -2650,8 +2650,8 @@ static ssize_t f2fs_quota_read(struct super_block *sb, int type, char *data,
if (PTR_ERR(page) == -ENOMEM) {
memalloc_retry_wait(GFP_NOFS);
goto repeat;
- }
- set_sbi_flag(F2FS_SB(sb), SBI_QUOTA_NEED_REPAIR);
+ } else if (PTR_ERR(page) != -EINTR)
+ set_sbi_flag(F2FS_SB(sb), SBI_QUOTA_NEED_REPAIR);
return PTR_ERR(page);
}
@@ -3070,7 +3070,7 @@ static int f2fs_dquot_acquire(struct dquot *dquot)
f2fs_down_read(&sbi->quota_sem);
ret = dquot_acquire(dquot);
- if (ret < 0)
+ if (ret < 0 && ret != -EINTR)
set_sbi_flag(sbi, SBI_QUOTA_NEED_REPAIR);
f2fs_up_read(&sbi->quota_sem);
return ret;