LU-12616 obclass: fix MDS start/stop race

[fs/lustre-release.git] / lustre / osp / osp_trans.c

diff --git a/lustre/osp/osp_trans.c b/lustre/osp/osp_trans.c
index ebecd8a..a69a9cb 100644 (file)
--- a/lustre/osp/osp_trans.c
+++ b/lustre/osp/osp_trans.c
@@ -20,7 +20,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2014, 2016, Intel Corporation.
+ * Copyright (c) 2014, 2017, Intel Corporation.
  */
 /*
  * lustre/osp/osp_trans.c
@@ -97,20 +97,6 @@ struct osp_update_callback {
        osp_update_interpreter_t        ouc_interpreter;
 };
 
-static struct object_update_request *object_update_request_alloc(size_t size)
-{
-       struct object_update_request *ourq;
-
-       OBD_ALLOC_LARGE(ourq, size);
-       if (ourq == NULL)
-               return ERR_PTR(-ENOMEM);
-
-       ourq->ourq_magic = UPDATE_REQUEST_MAGIC;
-       ourq->ourq_count = 0;
-
-       return ourq;
-}
-
 /**
  * Allocate new update request
  *
@@ -126,21 +112,28 @@ int osp_object_update_request_create(struct osp_update_request *our,
                                     size_t size)
 {
        struct osp_update_request_sub *ours;
+       struct object_update_request *ourq;
 
        OBD_ALLOC_PTR(ours);
        if (ours == NULL)
                return -ENOMEM;
 
-       if (size < OUT_UPDATE_INIT_BUFFER_SIZE)
-               size = OUT_UPDATE_INIT_BUFFER_SIZE;
-
-       ours->ours_req = object_update_request_alloc(size);
-
-       if (IS_ERR(ours->ours_req)) {
+       /* The object update request will be added to an SG list for
+        * bulk transfer. Some IB HW cannot handle partial pages in SG
+        * lists (since they create gaps in memory regions) so we
+        * round the size up to the next multiple of PAGE_SIZE. See
+        * LU-9983. */
+       LASSERT(size > 0);
+       size = round_up(size, PAGE_SIZE);
+       OBD_ALLOC_LARGE(ourq, size);
+       if (ourq == NULL) {
                OBD_FREE_PTR(ours);
                return -ENOMEM;
        }
 
+       ourq->ourq_magic = UPDATE_REQUEST_MAGIC;
+       ourq->ourq_count = 0;
+       ours->ours_req = ourq;
        ours->ours_req_size = size;
        INIT_LIST_HEAD(&ours->ours_list);
        list_add_tail(&ours->ours_list, &our->our_req_list);
@@ -199,7 +192,7 @@ struct osp_update_request *osp_update_request_create(struct dt_device *dt)
        INIT_LIST_HEAD(&our->our_invalidate_cb_list);
        spin_lock_init(&our->our_list_lock);
 
-       rc = osp_object_update_request_create(our, OUT_UPDATE_INIT_BUFFER_SIZE);
+       rc = osp_object_update_request_create(our, PAGE_SIZE);
        if (rc != 0) {
                OBD_FREE_PTR(our);
                return ERR_PTR(rc);
@@ -240,7 +233,7 @@ void osp_update_request_destroy(const struct lu_env *env,
                        list_del_init(&obj->opo_invalidate_cb_list);
                        spin_unlock(&obj->opo_lock);
 
-                       lu_object_put(env, &obj->opo_obj.do_lu);
+                       dt_object_put(env, &obj->opo_obj);
                }
 
                if (env == &lenv)
@@ -383,8 +376,8 @@ int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
                buf_count++;
        }
        repsize += sizeof(*reply);
-       repsize = (repsize + OUT_UPDATE_REPLY_SIZE - 1) &
-                       ~(OUT_UPDATE_REPLY_SIZE - 1);
+       if (repsize < OUT_UPDATE_REPLY_SIZE)
+               repsize = OUT_UPDATE_REPLY_SIZE;
        LASSERT(buf_count > 0);
 
        req = ptlrpc_request_alloc(imp, &RQF_OUT_UPDATE);
@@ -527,7 +520,7 @@ static void osp_thandle_invalidate_object(const struct lu_env *env,
                list_del_init(&obj->opo_invalidate_cb_list);
                spin_unlock(&obj->opo_lock);
 
-               lu_object_put(env, &obj->opo_obj.do_lu);
+               dt_object_put(env, &obj->opo_obj);
        }
 }
 
@@ -609,17 +602,17 @@ static void osp_update_callback_fini(const struct lu_env *env,
  * \retval             negative error number on failure
  */
 static int osp_update_interpret(const struct lu_env *env,
-                               struct ptlrpc_request *req, void *arg, int rc)
+                               struct ptlrpc_request *req, void *args, int rc)
 {
-       struct object_update_reply      *reply  = NULL;
-       struct osp_update_args          *oaua   = arg;
-       struct osp_update_request       *our = oaua->oaua_update;
-       struct osp_thandle              *oth;
-       struct osp_update_callback      *ouc;
-       struct osp_update_callback      *next;
-       int                              count  = 0;
-       int                              index  = 0;
-       int                              rc1    = 0;
+       struct object_update_reply *reply = NULL;
+       struct osp_update_args *oaua = args;
+       struct osp_update_request *our = oaua->oaua_update;
+       struct osp_thandle *oth;
+       struct osp_update_callback *ouc;
+       struct osp_update_callback *next;
+       int count = 0;
+       int index = 0;
+       int rc1 = 0;
 
        ENTRY;
 
@@ -973,7 +966,6 @@ struct thandle *osp_trans_create(const struct lu_env *env, struct dt_device *d)
        oth->ot_magic = OSP_THANDLE_MAGIC;
        th = &oth->ot_super;
        th->th_dev = d;
-       th->th_tags = LCT_TX_HANDLE;
 
        atomic_set(&oth->ot_refcount, 1);
        INIT_LIST_HEAD(&oth->ot_commit_dcb_list);
@@ -1151,7 +1143,7 @@ static int osp_send_update_req(const struct lu_env *env,
                if (!osp->opd_connect_mdt) {
                        down_read(&osp->opd_async_updates_rwsem);
                        args->oaua_count = &osp->opd_async_updates_count;
-                       args->oaua_waitq = &osp->opd_syn_barrier_waitq;
+                       args->oaua_waitq = &osp->opd_sync_barrier_waitq;
                        up_read(&osp->opd_async_updates_rwsem);
                        atomic_inc(args->oaua_count);
                }
@@ -1181,7 +1173,10 @@ static int osp_send_update_req(const struct lu_env *env,
                rc = ptlrpc_queue_wait(req);
                if (osp->opd_connect_mdt)
                        osp_put_rpc_lock(osp);
-               if ((rc == -ENOMEM && req->rq_set == NULL) ||
+
+               /* We use rq_queued_time to distinguish between local
+                * and remote -ENOMEM. */
+               if ((rc == -ENOMEM && req->rq_queued_time == 0) ||
                    (req->rq_transno == 0 && !req->rq_committed)) {
                        if (args->oaua_update != NULL) {
                                /* If osp_update_interpret is not being called,
@@ -1208,7 +1203,7 @@ out:
  * Get local thandle for osp_thandle
  *
  * Get the local OSD thandle from the OSP thandle. Currently, there
- * are a few OSP API (osp_object_create() and osp_sync_add()) needs
+ * are a few OSP API (osp_create() and osp_sync_add()) needs
  * to update the object on local OSD device.
  *
  * If the osp_thandle comes from normal stack (MDD->LOD->OSP), then