LU-3680 ptlrpc: Fix assertion failure of null_alloc_rs()

lustre_get_emerg_rs() set the size of the reply buffer to zero
by mistake, which will cause LBUG in null_alloc_rs() when memory
pressure is high. This patch fix this problem and adds a size
check to avoid the problem of insufficient buffer size.

Signed-off-by: Li Xi <lixi@ddn.com>
Signed-off-by: Patrick Farrell <paf@cray.com>
Change-Id: I9fbd4f14e8e1263de2af564c4f2e420f5f2b43bc
Reviewed-on: http://review.whamcloud.com/8200
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: John L. Hammond <john.hammond@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>

diff --git a/lustre/ptlrpc/pack_generic.c b/lustre/ptlrpc/pack_generic.c
index 2d6d7bd..c33a6da 100644 (file)
--- a/lustre/ptlrpc/pack_generic.c
+++ b/lustre/ptlrpc/pack_generic.c
@@ -311,6 +311,7 @@ lustre_get_emerg_rs(struct ptlrpc_service_part *svcpt)
        spin_unlock(&svcpt->scp_rep_lock);
 
        memset(rs, 0, svcpt->scp_service->srv_max_reply_size);
        spin_unlock(&svcpt->scp_rep_lock);
 
        memset(rs, 0, svcpt->scp_service->srv_max_reply_size);

+       rs->rs_size = svcpt->scp_service->srv_max_reply_size;

        rs->rs_svcpt = svcpt;
        rs->rs_prealloc = 1;
 out:
        rs->rs_svcpt = svcpt;
        rs->rs_prealloc = 1;
 out:

diff --git a/lustre/ptlrpc/sec.c b/lustre/ptlrpc/sec.c
index 1c2e30e..fbea227 100644 (file)
--- a/lustre/ptlrpc/sec.c
+++ b/lustre/ptlrpc/sec.c
@@ -2111,8 +2111,18 @@ int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen)
 
         rc = policy->sp_sops->alloc_rs(req, msglen);
         if (unlikely(rc == -ENOMEM)) {
 
         rc = policy->sp_sops->alloc_rs(req, msglen);
         if (unlikely(rc == -ENOMEM)) {

+               struct ptlrpc_service_part *svcpt = req->rq_rqbd->rqbd_svcpt;
+               if (svcpt->scp_service->srv_max_reply_size <
+                  msglen + sizeof(struct ptlrpc_reply_state)) {
+                       /* Just return failure if the size is too big */
+                       CERROR("size of message is too big (%zd), %d allowed",
+                               msglen + sizeof(struct ptlrpc_reply_state),
+                               svcpt->scp_service->srv_max_reply_size);
+                       RETURN(-ENOMEM);
+               }
+

                 /* failed alloc, try emergency pool */
                 /* failed alloc, try emergency pool */

-               rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_svcpt);
+               rs = lustre_get_emerg_rs(svcpt);

                 if (rs == NULL)
                         RETURN(-ENOMEM);
 
                 if (rs == NULL)
                         RETURN(-ENOMEM);