Linux CIFS filesystem development
 help / color / mirror / Atom feed
From: Paulo Alcantara <pc@manguebit.org>
To: linux-cifs@vger.kernel.org, netfs@lists.linux.dev,
	David Howells <dhowells@redhat.com>
Cc: Namjae Jeon <linkinjeon@kernel.org>,
	Ronnie Sahlberg <ronniesahlberg@gmail.com>,
	Shyam Prasad N <sprasad@microsoft.com>,
	Tom Talpey <tom@talpey.com>, Bharath SM <bharathsm@microsoft.com>,
	Christian Brauner <brauner@kernel.org>,
	stable@vger.kernel.org
Subject: [PATCH v7 01/19] smb: client: don't write back through O_DIRECT handles
Date: Fri,  6 Nov 2026 20:18:09 -0300	[thread overview]
Message-ID: <20261106231827.2604956-2-pc@manguebit.org> (raw)
In-Reply-To: <20261106231827.2604956-1-pc@manguebit.org>

Writeback uses the first writable handle of the inode, which may have
been opened with O_DIRECT.  Such handles are opened with
FILE_NO_INTERMEDIATE_BUFFERING, for which Windows rejects any unaligned
I/O with STATUS_INVALID_PARAMETER.  Writeback of partly dirty pages and
of the unaligned tail of a file through one of them fails, so netfs ends
it with the folios clean and the data never reaches the server.

xfstests generic/051, where fsstress mixes both kinds of opens, loses
about 1% of its writes this way:

  smb3_write_err: ... offset=0xde90d7 len=0xbd0b rc=-22

All the rejected writes in a network capture were unaligned writes on a
no-buffering handle; none of the aligned ones, and none of the unaligned
ones on regular handles, were.

Prefer a handle that was not opened with O_DIRECT, and use one that was
only if there is no other.

Fixes: 1013e760d10e ("SMB3: Don't ignore O_SYNC/O_DSYNC and O_DIRECT flags")
Signed-off-by: Paulo Alcantara <pc@manguebit.org>
Cc: stable@vger.kernel.org
---
 fs/smb/client/cifsglob.h |  1 +
 fs/smb/client/file.c     | 15 ++++++++++++---
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/fs/smb/client/cifsglob.h b/fs/smb/client/cifsglob.h
index d3ca688b6aba..dbf980fcf05b 100644
--- a/fs/smb/client/cifsglob.h
+++ b/fs/smb/client/cifsglob.h
@@ -1891,6 +1891,7 @@ enum cifs_find_flags {
 	FIND_WITH_DELETE	= (1U << 1),
 	FIND_NO_PENDING_DELETE	= (1U << 2),
 	FIND_OPEN_FLAGS		= (1U << 3),
+	FIND_NO_DIRECT		= (1U << 4),
 };
 
 #define   MID_FREE 0
diff --git a/fs/smb/client/file.c b/fs/smb/client/file.c
index 0767eb5f4687..aa2d09dac44c 100644
--- a/fs/smb/client/file.c
+++ b/fs/smb/client/file.c
@@ -255,9 +255,15 @@ static void cifs_begin_writeback(struct netfs_io_request *wreq)
 	struct cifs_io_request *req = container_of(wreq, struct cifs_io_request, rreq);
 	int ret;
 
-	ret = cifs_get_writable_file(CIFS_I(wreq->inode), FIND_ANY, &req->cfile);
+	ret = cifs_get_writable_file(CIFS_I(wreq->inode), FIND_NO_DIRECT,
+				     &req->cfile);
 	if (ret) {
-		cifs_dbg(VFS, "No writable handle in writepages ret=%d\n", ret);
+		ret = cifs_get_writable_file(CIFS_I(wreq->inode), FIND_ANY,
+					     &req->cfile);
+	}
+	if (ret) {
+		cifs_dbg(VFS, "%s: couldn't find writable handle: %d\n",
+			 __func__, ret);
 		return;
 	}
 
@@ -2722,7 +2728,7 @@ int __cifs_get_writable_file(struct cifsInodeInfo *cifs_inode,
 	}
 
 	if (test_bit(CIFS_INO_TMPFILE, &cifs_inode->flags))
-		find_flags = FIND_ANY;
+		find_flags &= FIND_NO_DIRECT;
 
 	cifs_sb = CIFS_SB(cifs_inode);
 
@@ -2752,6 +2758,9 @@ int __cifs_get_writable_file(struct cifsInodeInfo *cifs_inode,
 		    !open_flags_match(cifs_inode, open_flags,
 				      open_file->f_flags))
 			continue;
+		if ((find_flags & FIND_NO_DIRECT) &&
+		    (open_file->f_flags & O_DIRECT))
+			continue;
 		if (OPEN_FMODE(open_file->f_flags) & FMODE_WRITE) {
 			if (!open_file->invalidHandle) {
 				/* found a good writable file */
-- 
2.56.0


  reply	other threads:[~2026-11-06 23:18 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-06 23:18 [PATCH v7 00/19] smb: client: fix data loss in writeback Paulo Alcantara
2026-11-06 23:18 ` Paulo Alcantara [this message]
2026-11-06 23:18 ` [PATCH v7 02/19] netfs: preserve prepare_write errno on reissue Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 03/19] netfs: keep the prepare_read errno when abandoning read retries Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 04/19] netfs: don't hang a read whose retry is abandoned Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 05/19] smb: client: let umount -f abort retried reads and writes Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 06/19] smb: client: don't give up writeback on the first replayable error Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 07/19] smb: client: don't retry failed writes forever regardless of mount options Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 08/19] smb: client: don't spend read retries on reads cut short by a reconnect Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 09/19] smb: client: don't block dying tasks waiting for writeback on close Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 10/19] smb: client: issue writes from a worker if the task is dying Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 11/19] smb: client: don't fail writeback preparation " Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 12/19] smb: client: retry writeback on interrupts and allocation failures Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 13/19] smb: client: don't block dying tasks waiting for writeback on fsync Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 14/19] smb: client: retry writes that couldn't be issued due to a reconnect Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 15/19] smb: client: retry writeback after losing the connection, share or handle Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 16/19] smb: client: let umount -f stop waiting for reconnects on hard mounts Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 17/19] smb: client: don't replay handle reopens inside SMB2_open() Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 18/19] smb: client: fail writeback sends and reopens fast when the server is gone Paulo Alcantara
2026-11-06 23:18 ` [PATCH v7 19/19] smb: client: stop reconnecting and replaying during a forced unmount Paulo Alcantara

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=20261106231827.2604956-2-pc@manguebit.org \
    --to=pc@manguebit.org \
    --cc=bharathsm@microsoft.com \
    --cc=brauner@kernel.org \
    --cc=dhowells@redhat.com \
    --cc=linkinjeon@kernel.org \
    --cc=linux-cifs@vger.kernel.org \
    --cc=netfs@lists.linux.dev \
    --cc=ronniesahlberg@gmail.com \
    --cc=sprasad@microsoft.com \
    --cc=stable@vger.kernel.org \
    --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