staging/lustre/ptlrpc: unlink request buffer correctly
authorAlexey Lyashkov <alexey_lyashkov@xyratex.com>
Mon, 23 Jun 2014 01:32:18 +0000 (21:32 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 27 Jun 2014 00:45:06 +0000 (20:45 -0400)
outgoning buffer may be hold by lnet and don't unlinked fast,
it's break unloading a lustre modules as request hold a
reference to the export/obd

Signed-off-by: Alexey Lyashkov <alexey_lyashkov@xyratex.com>
Xyratex-bug-id: MRP-1848
Reviewed-on: http://review.whamcloud.com/10353
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-5073
Reviewed-by: Mike Pershin <mike.pershin@intel.com>
Reviewed-by: Liang Zhen <liang.zhen@intel.com>
Reviewed-by: Isaac Huang <he.huang@intel.com>
Signed-off-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/include/lustre_net.h
drivers/staging/lustre/lustre/ptlrpc/client.c
drivers/staging/lustre/lustre/ptlrpc/events.c
drivers/staging/lustre/lustre/ptlrpc/niobuf.c

index f6b7d10cb78cf64c46652370cf803c4af0e1b6a5..b837d34f0b4d0d545009aa646920a0542a4f28d6 100644 (file)
@@ -1591,7 +1591,8 @@ struct ptlrpc_request {
                rq_replay:1,
                rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
                rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
-               rq_early:1, rq_must_unlink:1,
+               rq_early:1,
+               rq_req_unlink:1, rq_reply_unlink:1,
                rq_memalloc:1,      /* req originated from "kswapd" */
                /* server-side flags */
                rq_packed_final:1,  /* packed final reply */
@@ -3039,7 +3040,8 @@ ptlrpc_client_recv_or_unlink(struct ptlrpc_request *req)
                spin_unlock(&req->rq_lock);
                return 1;
        }
-       rc = req->rq_receiving_reply || req->rq_must_unlink;
+       rc = req->rq_receiving_reply;
+       rc = rc || req->rq_req_unlink || req->rq_reply_unlink;
        spin_unlock(&req->rq_lock);
        return rc;
 }
index 189048287946ec191b1ceac7dd7056bbb40857f8..0e0ea5cd5f3695063f2851d6bd269fd4f755e2cd 100644 (file)
@@ -1202,7 +1202,7 @@ static int after_reply(struct ptlrpc_request *req)
 
        LASSERT(obd != NULL);
        /* repbuf must be unlinked */
-       LASSERT(!req->rq_receiving_reply && !req->rq_must_unlink);
+       LASSERT(!req->rq_receiving_reply && !req->rq_reply_unlink);
 
        if (req->rq_reply_truncate) {
                if (ptlrpc_no_resend(req)) {
@@ -2406,9 +2406,10 @@ int ptlrpc_unregister_reply(struct ptlrpc_request *request, int async)
                }
 
                LASSERT(rc == -ETIMEDOUT);
-               DEBUG_REQ(D_WARNING, request, "Unexpectedly long timeout "
-                         "rvcng=%d unlnk=%d", request->rq_receiving_reply,
-                         request->rq_must_unlink);
+               DEBUG_REQ(D_WARNING, request,
+                         "Unexpectedly long timeout rvcng=%d unlnk=%d/%d",
+                         request->rq_receiving_reply,
+                         request->rq_req_unlink, request->rq_reply_unlink);
        }
        return 0;
 }
@@ -3081,7 +3082,7 @@ void *ptlrpcd_alloc_work(struct obd_import *imp,
        req->rq_interpret_reply = work_interpreter;
        /* don't want reply */
        req->rq_receiving_reply = 0;
-       req->rq_must_unlink = 0;
+       req->rq_req_unlink = req->rq_reply_unlink = 0;
        req->rq_no_delay = req->rq_no_resend = 1;
        req->rq_pill.rc_fmt = (void *)&worker_format;
 
index 9f9b8d1f835eeb6b459c4134bfd2b13be5fa9cef..209fcc14429825de4117cb757972b91f335adf6a 100644 (file)
@@ -63,19 +63,20 @@ void request_out_callback(lnet_event_t *ev)
        DEBUG_REQ(D_NET, req, "type %d, status %d", ev->type, ev->status);
 
        sptlrpc_request_out_callback(req);
+       spin_lock(&req->rq_lock);
        req->rq_real_sent = cfs_time_current_sec();
+       if (ev->unlinked)
+               req->rq_req_unlink = 0;
 
        if (ev->type == LNET_EVENT_UNLINK || ev->status != 0) {
 
                /* Failed send: make it seem like the reply timed out, just
                 * like failing sends in client.c does currently...  */
 
-               spin_lock(&req->rq_lock);
                req->rq_net_err = 1;
-               spin_unlock(&req->rq_lock);
-
                ptlrpc_client_wake_req(req);
        }
+       spin_unlock(&req->rq_lock);
 
        ptlrpc_req_finished(req);
 }
@@ -102,7 +103,7 @@ void reply_in_callback(lnet_event_t *ev)
        req->rq_receiving_reply = 0;
        req->rq_early = 0;
        if (ev->unlinked)
-               req->rq_must_unlink = 0;
+               req->rq_reply_unlink = 0;
 
        if (ev->status)
                goto out_wake;
index f760504e0696acc4c3398e01cb4f8e999f7636ff..3f0ca23f7b9133790bfa9be9b8096f6a77c41fb0 100644 (file)
@@ -580,8 +580,9 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
        spin_lock(&request->rq_lock);
        /* If the MD attach succeeds, there _will_ be a reply_in callback */
        request->rq_receiving_reply = !noreply;
+       request->rq_req_unlink = 1;
        /* We are responsible for unlinking the reply buffer */
-       request->rq_must_unlink = !noreply;
+       request->rq_reply_unlink = !noreply;
        /* Clear any flags that may be present from previous sends. */
        request->rq_replied = 0;
        request->rq_err = 0;
@@ -604,7 +605,7 @@ int ptl_send_rpc(struct ptlrpc_request *request, int noreply)
                reply_md.user_ptr  = &request->rq_reply_cbid;
                reply_md.eq_handle = ptlrpc_eq_h;
 
-               /* We must see the unlink callback to unset rq_must_unlink,
+               /* We must see the unlink callback to unset rq_reply_unlink,
                   so we can't auto-unlink */
                rc = LNetMDAttach(reply_me_h, reply_md, LNET_RETAIN,
                                  &request->rq_reply_md_h);