Sat, Feb 9 2019, 10:56AM, Sasha Levin sashal@kernel.org wrote:
From: Pavel Shilovsky pshilov@microsoft.com
[ Upstream commit ee258d79159afed52ca9372aeb9c1a51e89b32ee ]
Currently we account for credits in the thread initiating a request and waiting for a response. The demultiplex thread receives the response, wakes up the thread and the latter collects credits from the response buffer and add them to the server structure on the client. This approach is not accurate, because it may race with reconnect events in the demultiplex thread which resets the number of credits.
Fix this by moving credit processing to new mid callbacks that collect credits granted by the server from the response in the demultiplex thread.
Signed-off-by: Pavel Shilovsky pshilov@microsoft.com Signed-off-by: Steve French stfrench@microsoft.com Signed-off-by: Sasha Levin sashal@kernel.org
fs/cifs/transport.c | 51 ++++++++++++++++++++++++++++++--------------- 1 file changed, 34 insertions(+), 17 deletions(-)
diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c index 4dbf62bb51b2..0dab276eced8 100644 --- a/fs/cifs/transport.c +++ b/fs/cifs/transport.c @@ -781,12 +781,7 @@ cifs_setup_request(struct cifs_ses *ses, struct smb_rqst *rqst) }
static void -cifs_noop_callback(struct mid_q_entry *mid) -{ -}
-static void -cifs_cancelled_callback(struct mid_q_entry *mid) +cifs_compound_callback(struct mid_q_entry *mid) { struct TCP_Server_Info *server = mid->server; unsigned int optype = mid->optype; @@ -799,10 +794,23 @@ cifs_cancelled_callback(struct mid_q_entry *mid) cifs_dbg(FYI, "Bad state for cancelled MID\n"); }
DeleteMidQEntry(mid); add_credits(server, credits_received, optype);}
+static void +cifs_compound_last_callback(struct mid_q_entry *mid) +{
cifs_compound_callback(mid);cifs_wake_up_task(mid);+}
+static void +cifs_cancelled_callback(struct mid_q_entry *mid) +{
cifs_compound_callback(mid);DeleteMidQEntry(mid);+}
int compound_send_recv(const unsigned int xid, struct cifs_ses *ses, const int flags, const int num_rqst, struct smb_rqst *rqst, @@ -880,11 +888,14 @@ compound_send_recv(const unsigned int xid, struct cifs_ses *ses, midQ[i]->mid_state = MID_REQUEST_SUBMITTED; midQ[i]->optype = optype; /*
* We don't invoke the callback compounds unless it is the last* request.
* Invoke callback for every part of the compound chain* to calculate credits properly. Wake up this thread only when* the last element is received. */ if (i < num_rqst - 1)
midQ[i]->callback = cifs_noop_callback;
midQ[i]->callback = cifs_compound_callback;elsemidQ[i]->callback = cifs_compound_last_callback; } cifs_in_send_inc(ses->server); rc = smb_send_rqst(ses->server, num_rqst, rqst, flags);@@ -898,8 +909,20 @@ compound_send_recv(const unsigned int xid, struct cifs_ses *ses,
mutex_unlock(&ses->server->srv_mutex);
if (rc < 0)
if (rc < 0) {/* Sending failed for some reason - return credits back */for (i = 0; i < num_rqst; i++)add_credits(ses->server, credits[i], optype); goto out;}/** At this point the request is passed to the network stack - we assume* that any credits taken from the server structure on the client have* been spent and we can't return them back. Once we receive responses* we will collect credits granted by the server in the mid callbacks* and add those credits to the server structure.*/ /* * Compounding is never used during session establish.@@ -932,11 +955,6 @@ compound_send_recv(const unsigned int xid, struct cifs_ses *ses, } }
for (i = 0; i < num_rqst; i++)if (!cancelled_mid[i] && midQ[i]->resp_buf&& (midQ[i]->mid_state == MID_RESPONSE_RECEIVED))credits[i] = ses->server->ops->get_credits(midQ[i]);for (i = 0; i < num_rqst; i++) { if (rc < 0) goto out;@@ -995,7 +1013,6 @@ out: for (i = 0; i < num_rqst; i++) { if (!cancelled_mid[i]) cifs_delete_mid(midQ[i]);
add_credits(ses->server, credits[i], optype); } return rc;-- 2.19.1
Please apply the following two patches too:
https://patchwork.ozlabs.org/patch/1030180/ https://patchwork.ozlabs.org/patch/1030181/
The 1st fixes some minor regressions introduces by the proposing patch and the 2nd adds the same logic to processing of async responses as the proposing patch is doing for sync ones.
-- Best regards, Pavel Shilovsky