[PATCH v7 06/19] smb: client: don't give up writeback on the first replayable error

[Paulo Alcantara <pc@manguebit.org>]

Read and write retries follow the hard and retrans mount options, and
mounts are soft with no retrans by default.  For writeback this means
that a single replayable error, e.g. a reconnect while sending, fails
the subrequest and netfs ends writeback with its folios clean: the data
is gone, possibly long after write(2) succeeded, with only fsync(2) left
to notice.  NFS soft mounts, by comparison, wait minutes for a server to
answer before failing a request.

Keep retrying writeback on replayable errors, with exponential backoff:
forever on hard mounts, and on soft mounts for max(3, retrans) echo
intervals from the first failure - 3 minutes by default, the time after
which a silent server is considered unresponsive.  The window applies to
each write sent to the server, not to the whole writeback.  umount -f or
a shutdown still stops it right away.  Direct and unbuffered writes,
whose callers get the error, keep following the mount options.

Fixes: 2c1238a7477a ("cifs: make retry logic in read/write path consistent with other paths")
Signed-off-by: Paulo Alcantara <pc@manguebit.org>
Cc: stable@vger.kernel.org
---
 fs/smb/client/cifsglob.h  |  1 +
 fs/smb/client/file.c      | 21 +++++++++++++++++++++
 fs/smb/client/smb2ops.c   |  9 +--------
 fs/smb/client/smb2proto.h |  9 +++++++++
 4 files changed, 32 insertions(+), 8 deletions(-)

diff --git a/fs/smb/client/cifsglob.h b/fs/smb/client/cifsglob.h
index fb732c178e76..b6a993cc0245 100644
--- a/fs/smb/client/cifsglob.h
+++ b/fs/smb/client/cifsglob.h
@@ -1499,6 +1499,7 @@ struct cifs_io_subrequest {
 	bool				replay;
 	unsigned int			retries;	/* number of retries so far */
 	unsigned int			cur_sleep;	/* time to sleep before replay */
+	unsigned long			retry_deadline;
 	struct kvec			iov[2];
 	struct TCP_Server_Info		*server;
 #ifdef CONFIG_CIFS_SMB_DIRECT
diff --git a/fs/smb/client/file.c b/fs/smb/client/file.c
index 270858e526f9..e8557c670dc3 100644
--- a/fs/smb/client/file.c
+++ b/fs/smb/client/file.c
@@ -53,12 +53,33 @@ bool cifs_io_stopped(struct cifs_io_subrequest *subreq)
 	return aborts && rwsem_is_contended(&sb->s_umount);
 }
 
+static bool cifs_writeback_retry_expired(struct cifs_io_subrequest *subreq,
+					 struct cifs_tcon *tcon)
+{
+	struct TCP_Server_Info *server = tcon->ses->server;
+	unsigned long window;
+
+	if (tcon->retry)
+		return false;
+	if (!subreq->retry_deadline) {
+		window = max(3U, server->retrans) * server->echo_interval;
+		subreq->retry_deadline = (jiffies + window) | 1;
+	}
+	return time_after(jiffies, subreq->retry_deadline);
+}
+
 bool cifs_io_should_replay(struct cifs_io_subrequest *subreq)
 {
 	struct cifs_tcon *tcon = tlink_tcon(subreq->req->cfile->tlink);
 
 	if (cifs_io_stopped(subreq))
 		return false;
+	if (subreq->rreq->origin == NETFS_WRITEBACK) {
+		if (cifs_writeback_retry_expired(subreq, tcon))
+			return false;
+		smb2_replay_backoff(&subreq->cur_sleep);
+		return true;
+	}
 	return smb2_should_replay(tcon, &subreq->retries, &subreq->cur_sleep);
 }
 
diff --git a/fs/smb/client/smb2ops.c b/fs/smb/client/smb2ops.c
index dc55cd512616..457f18ce3245 100644
--- a/fs/smb/client/smb2ops.c
+++ b/fs/smb/client/smb2ops.c
@@ -2934,14 +2934,7 @@ bool smb2_should_replay(struct cifs_tcon *tcon,
 		return false;
 
 	if (tcon->retry || (*pretries)++ < tcon->ses->server->retrans) {
-		/* Update sleep time for exponential backoff */
-		if (!(*pcur_sleep))
-			(*pcur_sleep) = 1;
-		else {
-			(*pcur_sleep) = ((*pcur_sleep) << 1);
-			if ((*pcur_sleep) > CIFS_MAX_SLEEP)
-				(*pcur_sleep) = CIFS_MAX_SLEEP;
-		}
+		smb2_replay_backoff(pcur_sleep);
 		return true;
 	}
 
diff --git a/fs/smb/client/smb2proto.h b/fs/smb/client/smb2proto.h
index c79e5cb2ce26..b2acff8938f3 100644
--- a/fs/smb/client/smb2proto.h
+++ b/fs/smb/client/smb2proto.h
@@ -121,6 +121,15 @@ unsigned long smb_rqst_len(struct TCP_Server_Info *server,
 void smb2_set_next_command(struct cifs_tcon *tcon, struct smb_rqst *rqst);
 void smb2_set_related(struct smb_rqst *rqst);
 void smb2_set_replay(struct TCP_Server_Info *server, struct smb_rqst *rqst);
+
+static inline void smb2_replay_backoff(int *pcur_sleep)
+{
+	if (!*pcur_sleep)
+		*pcur_sleep = 1;
+	else
+		*pcur_sleep = min(*pcur_sleep << 1, CIFS_MAX_SLEEP);
+}
+
 bool smb2_should_replay(struct cifs_tcon *tcon, int *pretries,
 			int *pcur_sleep);
 
-- 
2.56.0