[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <20211216144558.63931-1-zhangjiachen.jaycee@bytedance.com>
Date: Thu, 16 Dec 2021 22:45:58 +0800
From: Jiachen Zhang <zhangjiachen.jaycee@...edance.com>
To: miklos@...redi.hu
Cc: linux-fsdevel@...r.kernel.org, linux-kernel@...r.kernel.org,
xieyongji@...edance.com,
Jiachen Zhang <zhangjiachen.jaycee@...edance.com>
Subject: [PATCH] fuse: fix deadlock between O_TRUNC open() and non-O_TRUNC open()
fuse_finish_open() will be called with FUSE_NOWRITE set in case of atomic
O_TRUNC open(), so commit 76224355db75 ("fuse: truncate pagecache on
atomic_o_trunc") replaced invalidate_inode_pages2() by truncate_pagecache()
in such a case to avoid the A-A deadlock. However, we found another A-B-B-A
deadlock related to the case above, which will cause the xfstests
generic/464 testcase hung in our virtio-fs test environment.
Consider two processes concurrently open one same file, one with O_TRUNC
and another without O_TRUNC. The deadlock case is described below, if
open(O_TRUNC) is already set_nowrite(acquired A), and is trying to lock
a page (acquiring B), open() could have held the page lock (acquired B),
and waiting on the page writeback (acquiring A). This would lead to
deadlocks.
This commit tries to fix it by locking inode in fuse_open_common() even
if O_TRUNC is not set. This introduces a lock C to protect the area with
A-B-B-A the deadlock rick.
open(O_TRUNC)
----------------------------------------------------------------
fuse_open_common
lock_inode [C acquire]
fuse_set_nowrite [A acquire]
fuse_finish_open
truncate_pagecache
lock_page [B acquire]
truncate_inode_page
unlock_page [B release]
fuse_release_nowrite [A release]
unlock_inode [C release]
----------------------------------------------------------------
open()
----------------------------------------------------------------
fuse_open_common
(lock_inode) [C acquire, introduced by this commit]
fuse_finish_open
invalidate_inode_pages2
lock_page [B acquire]
fuse_launder_page
fuse_wait_on_page_writeback [A acquire]
unlock_page [B release]
(unlock_inode) [C release, introduced by this commit]
----------------------------------------------------------------
Signed-off-by: Jiachen Zhang <zhangjiachen.jaycee@...edance.com>
---
fs/fuse/file.c | 16 +++++++++-------
1 file changed, 9 insertions(+), 7 deletions(-)
diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 829094451774..73fec30e0a37 100644
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -229,8 +229,10 @@ int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
bool is_wb_truncate = (file->f_flags & O_TRUNC) &&
fc->atomic_o_trunc &&
fc->writeback_cache;
- bool dax_truncate = (file->f_flags & O_TRUNC) &&
+ bool is_dax_truncate = (file->f_flags & O_TRUNC) &&
fc->atomic_o_trunc && FUSE_IS_DAX(inode);
+ bool may_truncate = fc->atomic_o_trunc &&
+ (fc->writeback_cache || FUSE_IS_DAX(inode));
if (fuse_is_bad(inode))
return -EIO;
@@ -239,12 +241,12 @@ int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
if (err)
return err;
- if (is_wb_truncate || dax_truncate) {
+ if (may_truncate)
inode_lock(inode);
+ if (is_wb_truncate || is_dax_truncate)
fuse_set_nowrite(inode);
- }
- if (dax_truncate) {
+ if (is_dax_truncate) {
filemap_invalidate_lock(inode->i_mapping);
err = fuse_dax_break_layouts(inode, 0, 0);
if (err)
@@ -256,13 +258,13 @@ int fuse_open_common(struct inode *inode, struct file *file, bool isdir)
fuse_finish_open(inode, file);
out:
- if (dax_truncate)
+ if (is_dax_truncate)
filemap_invalidate_unlock(inode->i_mapping);
- if (is_wb_truncate | dax_truncate) {
+ if (is_wb_truncate | is_dax_truncate)
fuse_release_nowrite(inode);
+ if (may_truncate)
inode_unlock(inode);
- }
return err;
}
--
2.20.1
Powered by blists - more mailing lists