mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Siddharth Chintamaneni <sidchintamaneni@gmail.com>
To: Paulo Alcantara <pc@manguebit.org>
Cc: Ronnie Sahlberg <ronniesahlberg@gmail.com>,
	Shyam Prasad N <sprasad@microsoft.com>,
	Tom Talpey <tom@talpey.com>, Bharath SM <bharathsm@microsoft.com>,
	Allen Pais <apais@microsoft.com>,
	Rachel Menge <rachelmenge@linux.microsoft.com>,
	Hardik Garg <hargar@linux.microsoft.com>,
	Adam Kim <adamkim@microsoft.com>,
	Siddharth Chintamaneni <sidchintamaneni@gmail.com>,
	linux-cifs@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: [PATCH v1 1/2] smb: client: fix busy dentry and inode on unmount from late oplock break
Date: Tue, 17 Nov 2026 00:17:32 +0000	[thread overview]
Message-ID: <20261117001733.192399-2-sidchintamaneni@gmail.com> (raw)
In-Reply-To: <20261117001733.192399-1-sidchintamaneni@gmail.com>

During unmount, cifs_kill_sb() flushes cifsoplockd_wq and deferredclose_wq
before waiting on cifs_sb->outstanding_rreq for in-flight netfs I/O
requests to finish their cleanup_work in cifs_free_request().

While cifs_kill_sb() is waiting on outstanding_rreq, any cifsFileInfo
referenced by an in-flight request (req->cfile) still has cfile->count >= 1
and remains on tcon->openFileList. If the SMB server sends a lease or
oplock break for that file during this window, the demultiplex thread
(cifsd) finds cfile on tcon->openFileList and calls
cifs_queue_oplock_break(cfile), which takes a new reference on cfile and
queues cfile->oplock_break onto cifsoplockd_wq after cifsoplockd_wq was
already flushed:

  Thread 1 (umount)      Thread 2 (cifsd)       Thread 3 (workers)
  -----------------      ----------------       ------------------
  cifs_kill_sb()
  flush(cifsoplockd_wq)
  flush(deferredclose_wq)
  wait(outstanding_rreq)
    [blocks: rreq == 1]
                         SMB2 Lease Break
                         finds cfile on list
                         cifs_queue_oplock_break()
                           queue(cifsoplockd_wq)
                           cifsFileInfo_get(cfile)
                           [cfile->count: 1 -> 2]
                                                cifs_oplock_break()
                                                  ihold(inode)
                                                cifs_free_request()
                                                  cifsFileInfo_put()
                                                  [cfile->count: 2 -> 1]
                                                  dec(outstanding_rreq)
    [wakes: rreq == 0]
  flush(serverclose_wq)
  flush(fileinfo_put_wq)
  kill_anon_super(sb)
    -> generic_shutdown_super()
       BUG: Dentry still in use (1)
       VFS: Busy inodes after unmount!

Furthermore, cifs_oplock_break() itself holds the inode via ihold() after
dropping cfile and may initiate writeback via filemap_fdatawrite(),
spawning new netfs I/O requests.

Fix this by counting queued oplock breaks in cifs_sb->outstanding_rreq
from when cifs_queue_oplock_break() queues the work until
cifs_oplock_break() finishes, so wait_var_event() in cifs_kill_sb() waits
for both in-flight I/O requests and any late oplock breaks to complete.

Fixes: 340cea84f691 ("cifs: open files should not hold ref on superblock")
Reported-by: Adam Kim <adamkim@microsoft.com>
Signed-off-by: Siddharth Chintamaneni <sidchintamaneni@gmail.com>
---
 fs/smb/client/cifsfs.c | 10 +++++-----
 fs/smb/client/file.c   |  2 ++
 fs/smb/client/misc.c   |  6 +++++-
 3 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/fs/smb/client/cifsfs.c b/fs/smb/client/cifsfs.c
index 8cff7857b4509..9faba158f8b6a 100644
--- a/fs/smb/client/cifsfs.c
+++ b/fs/smb/client/cifsfs.c
@@ -418,11 +418,11 @@ static void cifs_kill_sb(struct super_block *sb)
 		flush_workqueue(deferredclose_wq);
 
 		/*
-		 * Wait for all in-flight netfs I/O requests to finish their
-		 * cleanup_work so that any cifsFileInfo final puts they queue
-		 * to fileinfo_put_wq/serverclose_wq have been queued, then
-		 * drain the workqueue so the cfile dentry refs are dropped to
-		 * avoid the busy dentry warning.
+		 * Wait for all in-flight netfs I/O requests and oplock breaks
+		 * to finish so that any cifsFileInfo final puts they queue to
+		 * fileinfo_put_wq/serverclose_wq have been queued, then drain
+		 * the workqueue so the cfile dentry refs are dropped to avoid
+		 * the busy dentry warning.
 		 */
 		wait_var_event(&cifs_sb->outstanding_rreq,
 			       !atomic_read(&cifs_sb->outstanding_rreq));
diff --git a/fs/smb/client/file.c b/fs/smb/client/file.c
index f51653cb6432a..146ff2adfa244 100644
--- a/fs/smb/client/file.c
+++ b/fs/smb/client/file.c
@@ -4023,6 +4023,8 @@ void cifs_oplock_break(struct work_struct *work)
 out:
 	cifs_done_oplock_break(cinode);
 	iput(inode);
+	if (atomic_dec_and_test(&cifs_sb->outstanding_rreq))
+		wake_up_var(&cifs_sb->outstanding_rreq);
 }
 
 #ifdef CONFIG_SWAP
diff --git a/fs/smb/client/misc.c b/fs/smb/client/misc.c
index 4d0df73d15d7a..72a986bfaefc4 100644
--- a/fs/smb/client/misc.c
+++ b/fs/smb/client/misc.c
@@ -371,6 +371,8 @@ void cifs_put_writer(struct cifsInodeInfo *cinode)
  */
 void cifs_queue_oplock_break(struct cifsFileInfo *cfile)
 {
+	struct cifs_sb_info *cifs_sb = CIFS_SB(d_inode(cfile->dentry)->i_sb);
+
 	/*
 	 * Bump the handle refcount now while we hold the
 	 * open_file_lock to enforce the validity of it for the oplock
@@ -379,8 +381,10 @@ void cifs_queue_oplock_break(struct cifsFileInfo *cfile)
 	 *
 	 * Only take a reference if the work is actually queued.
 	 */
-	if (queue_work(cifsoplockd_wq, &cfile->oplock_break))
+	if (queue_work(cifsoplockd_wq, &cfile->oplock_break)) {
 		cifsFileInfo_get(cfile);
+		atomic_inc(&cifs_sb->outstanding_rreq);
+	}
 }
 
 void cifs_done_oplock_break(struct cifsInodeInfo *cinode)
-- 
2.43.0


  reply	other threads:[~2026-11-17  0:17 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-17  0:17 [PATCH v1 0/2] smb: client: fix unmount busy dentry/inode races on openFileList Siddharth Chintamaneni
2026-11-17  0:17 ` Siddharth Chintamaneni [this message]
2026-11-17  0:17 ` [PATCH v1 2/2] smb: client: fix busy dentry on unmount from persistent handle reopen Siddharth Chintamaneni

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261117001733.192399-2-sidchintamaneni@gmail.com \
    --to=sidchintamaneni@gmail.com \
    --cc=adamkim@microsoft.com \
    --cc=apais@microsoft.com \
    --cc=bharathsm@microsoft.com \
    --cc=hargar@linux.microsoft.com \
    --cc=linux-cifs@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=pc@manguebit.org \
    --cc=rachelmenge@linux.microsoft.com \
    --cc=ronniesahlberg@gmail.com \
    --cc=sprasad@microsoft.com \
    --cc=tom@talpey.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®