From: Patrick Farrell <p...@cray.com> 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 <l...@ddn.com> Signed-off-by: Patrick Farrell <p...@cray.com> Reviewed-on: http://review.whamcloud.com/8200 Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-3680 Reviewed-by: John L. Hammond <john.hamm...@intel.com> Signed-off-by: Oleg Drokin <oleg.dro...@intel.com> --- drivers/staging/lustre/lustre/ptlrpc/pack_generic.c | 1 + drivers/staging/lustre/lustre/ptlrpc/sec.c | 12 +++++++++++- 2 files changed, 12 insertions(+), 1 deletion(-) diff --git a/drivers/staging/lustre/lustre/ptlrpc/pack_generic.c b/drivers/staging/lustre/lustre/ptlrpc/pack_generic.c index 45c0b84..cddeeb6 100644 --- a/drivers/staging/lustre/lustre/ptlrpc/pack_generic.c +++ b/drivers/staging/lustre/lustre/ptlrpc/pack_generic.c @@ -300,6 +300,7 @@ 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); + rs->rs_size = svcpt->scp_service->srv_max_reply_size; rs->rs_svcpt = svcpt; rs->rs_prealloc = 1; out: diff --git a/drivers/staging/lustre/lustre/ptlrpc/sec.c b/drivers/staging/lustre/lustre/ptlrpc/sec.c index b0a1c5a..5e75392 100644 --- a/drivers/staging/lustre/lustre/ptlrpc/sec.c +++ b/drivers/staging/lustre/lustre/ptlrpc/sec.c @@ -2086,8 +2086,18 @@ int sptlrpc_svc_alloc_rs(struct ptlrpc_request *req, int msglen) 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 */ - rs = lustre_get_emerg_rs(req->rq_rqbd->rqbd_svcpt); + rs = lustre_get_emerg_rs(svcpt); if (rs == NULL) return -ENOMEM; -- 1.8.5.3 _______________________________________________ devel mailing list de...@linuxdriverproject.org http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel