Whamcloud - gitweb
LU-12651 osc: always call update_next_shrink
[fs/lustre-release.git] / lustre / osc / osc_request.c
index 5624214..a33641d 100644 (file)
@@ -231,8 +231,7 @@ int osc_setattr_async(struct obd_export *exp, struct obdo *oa,
        } else {
                req->rq_interpret_reply = osc_setattr_interpret;
 
-               CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
-               sa = ptlrpc_req_async_args(req);
+               sa = ptlrpc_req_async_args(sa, req);
                sa->sa_oa = oa;
                sa->sa_upcall = upcall;
                sa->sa_cookie = cookie;
@@ -320,8 +319,7 @@ int osc_ladvise_base(struct obd_export *exp, struct obdo *oa,
        }
 
        req->rq_interpret_reply = osc_ladvise_interpret;
-       CLASSERT(sizeof(*la) <= sizeof(req->rq_async_args));
-       la = ptlrpc_req_async_args(req);
+       la = ptlrpc_req_async_args(la, req);
        la->la_oa = oa;
        la->la_upcall = upcall;
        la->la_cookie = cookie;
@@ -417,8 +415,7 @@ int osc_punch_send(struct obd_export *exp, struct obdo *oa,
        ptlrpc_request_set_replen(req);
 
        req->rq_interpret_reply = osc_setattr_interpret;
-       CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
-       sa = ptlrpc_req_async_args(req);
+       sa = ptlrpc_req_async_args(sa, req);
        sa->sa_oa = oa;
        sa->sa_upcall = upcall;
        sa->sa_cookie = cookie;
@@ -496,8 +493,7 @@ int osc_sync_base(struct osc_object *obj, struct obdo *oa,
        ptlrpc_request_set_replen(req);
        req->rq_interpret_reply = osc_sync_interpret;
 
-       CLASSERT(sizeof(*fa) <= sizeof(req->rq_async_args));
-       fa = ptlrpc_req_async_args(req);
+       fa = ptlrpc_req_async_args(fa, req);
        fa->fa_obj = obj;
        fa->fa_oa = oa;
        fa->fa_upcall = upcall;
@@ -581,7 +577,7 @@ static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
         struct client_obd     *cli = &exp->exp_obd->u.cli;
         struct ptlrpc_request *req;
         struct ost_body       *body;
-       struct list_head       cancels = LIST_HEAD_INIT(cancels);
+       LIST_HEAD(cancels);
         int rc, count;
         ENTRY;
 
@@ -649,21 +645,18 @@ static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
                oa->o_dirty = cli->cl_dirty_grant;
        else
                oa->o_dirty = cli->cl_dirty_pages << PAGE_SHIFT;
-       if (unlikely(cli->cl_dirty_pages - cli->cl_dirty_transit >
-                    cli->cl_dirty_max_pages)) {
-               CERROR("dirty %lu - %lu > dirty_max %lu\n",
-                      cli->cl_dirty_pages, cli->cl_dirty_transit,
+       if (unlikely(cli->cl_dirty_pages > cli->cl_dirty_max_pages)) {
+               CERROR("dirty %lu > dirty_max %lu\n",
+                      cli->cl_dirty_pages,
                       cli->cl_dirty_max_pages);
                oa->o_undirty = 0;
-       } else if (unlikely(atomic_long_read(&obd_dirty_pages) -
-                           atomic_long_read(&obd_dirty_transit_pages) >
+       } else if (unlikely(atomic_long_read(&obd_dirty_pages) >
                            (long)(obd_max_dirty_pages + 1))) {
                /* The atomic_read() allowing the atomic_inc() are
                 * not covered by a lock thus they may safely race and trip
                 * this CERROR() unless we add in a small fudge factor (+1). */
-               CERROR("%s: dirty %ld - %ld > system dirty_max %ld\n",
+               CERROR("%s: dirty %ld > system dirty_max %ld\n",
                       cli_name(cli), atomic_long_read(&obd_dirty_pages),
-                      atomic_long_read(&obd_dirty_transit_pages),
                       obd_max_dirty_pages);
                oa->o_undirty = 0;
        } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
@@ -755,6 +748,7 @@ static int osc_shrink_grant_interpret(const struct lu_env *env,
        osc_update_grant(cli, body);
 out:
        OBD_SLAB_FREE_PTR(aa->aa_oa, osc_obdo_kmem);
+       aa->aa_oa = NULL;
 
        return rc;
 }
@@ -848,9 +842,11 @@ static int osc_should_shrink_grant(struct client_obd *client)
        if (client->cl_import == NULL)
                return 0;
 
-        if ((client->cl_import->imp_connect_data.ocd_connect_flags &
-             OBD_CONNECT_GRANT_SHRINK) == 0)
-                return 0;
+       if (!OCD_HAS_FLAG(&client->cl_import->imp_connect_data, GRANT_SHRINK) ||
+           client->cl_import->imp_grant_shrink_disabled) {
+               osc_update_next_shrink(client);
+               return 0;
+       }
 
        if (ktime_get_seconds() >= next_shrink - 5) {
                /* Get the current RPC size directly, instead of going via:
@@ -902,12 +898,13 @@ static void osc_grant_work_handler(struct work_struct *data)
        if (client_gtd.gtd_stopped == 1)
                return;
 
-       if (next_shrink > ktime_get_seconds())
-               schedule_delayed_work(&work, msecs_to_jiffies(
-                                       (next_shrink - ktime_get_seconds()) *
-                                       MSEC_PER_SEC));
-       else
+       if (next_shrink > ktime_get_seconds()) {
+               time64_t delay = next_shrink - ktime_get_seconds();
+
+               schedule_delayed_work(&work, cfs_time_seconds(delay));
+       } else {
                schedule_work(&work.work);
+       }
 }
 
 void osc_schedule_grant_work(void)
@@ -1548,8 +1545,7 @@ no_bulk:
        }
        ptlrpc_request_set_replen(req);
 
-       CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-       aa = ptlrpc_req_async_args(req);
+       aa = ptlrpc_req_async_args(aa, req);
        aa->aa_oa = oa;
        aa->aa_requested_nob = requested_nob;
        aa->aa_nio_count = niocount;
@@ -1627,11 +1623,10 @@ static void dump_all_bulk_pages(struct obdo *oa, __u32 page_count,
                kunmap(pga[i]->pg);
        }
 
-       rc = ll_vfs_fsync_range(filp, 0, LLONG_MAX, 1);
+       rc = vfs_fsync_range(filp, 0, LLONG_MAX, 1);
        if (rc)
                CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
        filp_close(filp, NULL);
-       return;
 }
 
 static int
@@ -1731,19 +1726,20 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
                &req->rq_import->imp_connection->c_peer;
        struct ost_body *body;
        u32 client_cksum = 0;
-        ENTRY;
 
-        if (rc < 0 && rc != -EDQUOT) {
-                DEBUG_REQ(D_INFO, req, "Failed request with rc = %d\n", rc);
-                RETURN(rc);
-        }
+       ENTRY;
 
-        LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
-        body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        if (body == NULL) {
-                DEBUG_REQ(D_INFO, req, "Can't unpack body\n");
-                RETURN(-EPROTO);
-        }
+       if (rc < 0 && rc != -EDQUOT) {
+               DEBUG_REQ(D_INFO, req, "Failed request: rc = %d", rc);
+               RETURN(rc);
+       }
+
+       LASSERTF(req->rq_repmsg != NULL, "rc = %d\n", rc);
+       body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+       if (body == NULL) {
+               DEBUG_REQ(D_INFO, req, "cannot unpack body");
+               RETURN(-EPROTO);
+       }
 
        /* set/clear over quota flag for a uid/gid/projid */
        if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE &&
@@ -1751,42 +1747,45 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
                unsigned qid[LL_MAXQUOTAS] = {
                                         body->oa.o_uid, body->oa.o_gid,
                                         body->oa.o_projid };
-               CDEBUG(D_QUOTA, "setdq for [%u %u %u] with valid %#llx, flags %x\n",
+               CDEBUG(D_QUOTA,
+                      "setdq for [%u %u %u] with valid %#llx, flags %x\n",
                       body->oa.o_uid, body->oa.o_gid, body->oa.o_projid,
                       body->oa.o_valid, body->oa.o_flags);
                       osc_quota_setdq(cli, req->rq_xid, qid, body->oa.o_valid,
                                       body->oa.o_flags);
-        }
+       }
 
-        osc_update_grant(cli, body);
+       osc_update_grant(cli, body);
 
-        if (rc < 0)
-                RETURN(rc);
+       if (rc < 0)
+               RETURN(rc);
 
-        if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
-                client_cksum = aa->aa_oa->o_cksum; /* save for later */
+       if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM)
+               client_cksum = aa->aa_oa->o_cksum; /* save for later */
 
-        if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
-                if (rc > 0) {
-                        CERROR("Unexpected +ve rc %d\n", rc);
-                        RETURN(-EPROTO);
-                }
+       if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
+               if (rc > 0) {
+                       CERROR("%s: unexpected positive size %d\n",
+                              obd_name, rc);
+                       RETURN(-EPROTO);
+               }
 
                if (req->rq_bulk != NULL &&
                    sptlrpc_cli_unwrap_bulk_write(req, req->rq_bulk))
-                        RETURN(-EAGAIN);
+                       RETURN(-EAGAIN);
 
-                if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
-                    check_write_checksum(&body->oa, peer, client_cksum,
+               if ((aa->aa_oa->o_valid & OBD_MD_FLCKSUM) && client_cksum &&
+                   check_write_checksum(&body->oa, peer, client_cksum,
                                         body->oa.o_cksum, aa))
-                        RETURN(-EAGAIN);
+                       RETURN(-EAGAIN);
 
-                rc = check_write_rcs(req, aa->aa_requested_nob,aa->aa_nio_count,
-                                     aa->aa_page_count, aa->aa_ppga);
-                GOTO(out, rc);
-        }
+               rc = check_write_rcs(req, aa->aa_requested_nob,
+                                    aa->aa_nio_count, aa->aa_page_count,
+                                    aa->aa_ppga);
+               GOTO(out, rc);
+       }
 
-        /* The rest of this function executes only for OST_READs */
+       /* The rest of this function executes only for OST_READs */
 
        if (req->rq_bulk == NULL) {
                rc = req_capsule_get_size(&req->rq_pill, &RMF_SHORT_IO,
@@ -1796,20 +1795,20 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
                /* if unwrap_bulk failed, return -EAGAIN to retry */
                rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk, rc);
        }
-        if (rc < 0)
-                GOTO(out, rc = -EAGAIN);
+       if (rc < 0)
+               GOTO(out, rc = -EAGAIN);
 
-        if (rc > aa->aa_requested_nob) {
-                CERROR("Unexpected rc %d (%d requested)\n", rc,
-                       aa->aa_requested_nob);
-                RETURN(-EPROTO);
-        }
+       if (rc > aa->aa_requested_nob) {
+               CERROR("%s: unexpected size %d, requested %d\n", obd_name,
+                      rc, aa->aa_requested_nob);
+               RETURN(-EPROTO);
+       }
 
        if (req->rq_bulk != NULL && rc != req->rq_bulk->bd_nob_transferred) {
-                CERROR ("Unexpected rc %d (%d transferred)\n",
-                        rc, req->rq_bulk->bd_nob_transferred);
-                return (-EPROTO);
-        }
+               CERROR("%s: unexpected size %d, transferred %d\n", obd_name,
+                      rc, req->rq_bulk->bd_nob_transferred);
+               RETURN(-EPROTO);
+       }
 
        if (req->rq_bulk == NULL) {
                /* short io */
@@ -1903,32 +1902,34 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc)
                        CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
                        rc = 0;
                }
-        } else if (unlikely(client_cksum)) {
-                static int cksum_missed;
-
-                cksum_missed++;
-                if ((cksum_missed & (-cksum_missed)) == cksum_missed)
-                        CERROR("Checksum %u requested from %s but not sent\n",
-                               cksum_missed, libcfs_nid2str(peer->nid));
-        } else {
-                rc = 0;
-        }
+       } else if (unlikely(client_cksum)) {
+               static int cksum_missed;
+
+               cksum_missed++;
+               if ((cksum_missed & (-cksum_missed)) == cksum_missed)
+                       CERROR("%s: checksum %u requested from %s but not sent\n",
+                              obd_name, cksum_missed,
+                              libcfs_nid2str(peer->nid));
+       } else {
+               rc = 0;
+       }
 out:
        if (rc >= 0)
                lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
                                     aa->aa_oa, &body->oa);
 
-        RETURN(rc);
+       RETURN(rc);
 }
 
 static int osc_brw_redo_request(struct ptlrpc_request *request,
                                struct osc_brw_async_args *aa, int rc)
 {
-        struct ptlrpc_request *new_req;
-        struct osc_brw_async_args *new_aa;
-        struct osc_async_page *oap;
-        ENTRY;
+       struct ptlrpc_request *new_req;
+       struct osc_brw_async_args *new_aa;
+       struct osc_async_page *oap;
+       ENTRY;
 
+       /* The below message is checked in replay-ost-single.sh test_8ae*/
        DEBUG_REQ(rc == -EINPROGRESS ? D_RPCTRACE : D_ERROR, request,
                  "redo for recoverable error %d", rc);
 
@@ -1967,7 +1968,7 @@ static int osc_brw_redo_request(struct ptlrpc_request *request,
         new_req->rq_generation_set = 1;
         new_req->rq_import_generation = request->rq_import_generation;
 
-        new_aa = ptlrpc_req_async_args(new_req);
+       new_aa = ptlrpc_req_async_args(new_aa, new_req);
 
        INIT_LIST_HEAD(&new_aa->aa_oaps);
        list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
@@ -2121,6 +2122,7 @@ static int brw_interpret(const struct lu_env *env,
                cl_object_attr_unlock(obj);
        }
        OBD_SLAB_FREE_PTR(aa->aa_oa, osc_obdo_kmem);
+       aa->aa_oa = NULL;
 
        if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
                osc_inc_unstable_pages(req);
@@ -2201,7 +2203,7 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
        int                             grant = 0;
        int                             rc;
        __u32                           layout_version = 0;
-       struct list_head                rpc_list = LIST_HEAD_INIT(rpc_list);
+       LIST_HEAD(rpc_list);
        struct ost_body                 *body;
        ENTRY;
        LASSERT(!list_empty(ext_list));
@@ -2312,8 +2314,7 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
        cl_req_attr_set(env, osc2cl(obj), crattr);
        lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
 
-       CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-       aa = ptlrpc_req_async_args(req);
+       aa = ptlrpc_req_async_args(aa, req);
        INIT_LIST_HEAD(&aa->aa_oaps);
        list_splice_init(&rpc_list, &aa->aa_oaps);
        INIT_LIST_HEAD(&aa->aa_exts);
@@ -2336,7 +2337,7 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
        }
        spin_unlock(&cli->cl_loi_list_lock);
 
-       DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %ur/%uw in flight",
+       DEBUG_REQ(D_INODE, req, "%d pages, aa %p, now %ur/%uw in flight",
                  page_count, aa, cli->cl_r_in_flight,
                  cli->cl_w_in_flight);
        OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_DELAY_IO, cfs_fail_val);
@@ -2488,9 +2489,8 @@ struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
  * release locks just after they are obtained. */
 int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
                     __u64 *flags, union ldlm_policy_data *policy,
-                    struct ost_lvb *lvb, int kms_valid,
-                    osc_enqueue_upcall_f upcall, void *cookie,
-                    struct ldlm_enqueue_info *einfo,
+                    struct ost_lvb *lvb, osc_enqueue_upcall_f upcall,
+                    void *cookie, struct ldlm_enqueue_info *einfo,
                     struct ptlrpc_request_set *rqset, int async,
                     bool speculative)
 {
@@ -2508,15 +2508,6 @@ int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
        policy->l_extent.start -= policy->l_extent.start & ~PAGE_MASK;
        policy->l_extent.end |= ~PAGE_MASK;
 
-       /*
-        * kms is not valid when either object is completely fresh (so that no
-        * locks are cached), or object was evicted. In the latter case cached
-        * lock cannot be used, because it would prime inode state with
-        * potentially stale LVB.
-        */
-       if (!kms_valid)
-               goto no_match;
-
         /* Next, search for already existing extent locks that will cover us */
         /* If we're trying to read, we also search for an existing PW lock.  The
          * VFS and page cache already protect us locally, so lots of readers/
@@ -2579,7 +2570,6 @@ int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
                }
        }
 
-no_match:
        if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
                RETURN(-ENOLCK);
 
@@ -2608,8 +2598,7 @@ no_match:
        if (async) {
                if (!rc) {
                        struct osc_enqueue_args *aa;
-                       CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-                       aa = ptlrpc_req_async_args(req);
+                       aa = ptlrpc_req_async_args(aa, req);
                        aa->oa_exp         = exp;
                        aa->oa_mode        = einfo->ei_mode;
                        aa->oa_type        = einfo->ei_type;
@@ -2647,9 +2636,10 @@ no_match:
        RETURN(rc);
 }
 
-int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
-                  enum ldlm_type type, union ldlm_policy_data *policy,
-                  enum ldlm_mode mode, __u64 *flags, void *data,
+int osc_match_base(const struct lu_env *env, struct obd_export *exp,
+                  struct ldlm_res_id *res_id, enum ldlm_type type,
+                  union ldlm_policy_data *policy, enum ldlm_mode mode,
+                  __u64 *flags, struct osc_object *obj,
                   struct lustre_handle *lockh, int unref)
 {
        struct obd_device *obd = exp->exp_obd;
@@ -2677,11 +2667,19 @@ int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
        if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
                RETURN(rc);
 
-       if (data != NULL) {
+       if (obj != NULL) {
                struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 
                LASSERT(lock != NULL);
-               if (!osc_set_lock_data(lock, data)) {
+               if (osc_set_lock_data(lock, obj)) {
+                       lock_res_and_lock(lock);
+                       if (!ldlm_is_lvb_cached(lock)) {
+                               LASSERT(lock->l_ast_data == obj);
+                               osc_lock_lvb_update(env, obj, lock, NULL);
+                               ldlm_set_lvb_cached(lock);
+                       }
+                       unlock_res_and_lock(lock);
+               } else {
                        ldlm_lock_decref(lockh, rc);
                        rc = 0;
                }
@@ -2775,8 +2773,7 @@ static int osc_statfs_async(struct obd_export *exp,
        }
 
        req->rq_interpret_reply = osc_statfs_interpret;
-       CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-       aa = ptlrpc_req_async_args(req);
+       aa = ptlrpc_req_async_args(aa, req);
        aa->aa_oi = oinfo;
 
        ptlrpc_set_add_req(rqset, req);
@@ -2960,8 +2957,7 @@ int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
                struct osc_grant_args *aa;
                struct obdo *oa;
 
-               CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-               aa = ptlrpc_req_async_args(req);
+               aa = ptlrpc_req_async_args(aa, req);
                OBD_SLAB_ALLOC_PTR_GFP(oa, osc_obdo_kmem, GFP_NOFS);
                if (!oa) {
                        ptlrpc_req_finished(req);
@@ -3350,7 +3346,7 @@ int osc_cleanup_common(struct obd_device *obd)
 }
 EXPORT_SYMBOL(osc_cleanup_common);
 
-static struct obd_ops osc_obd_ops = {
+static const struct obd_ops osc_obd_ops = {
         .o_owner                = THIS_MODULE,
         .o_setup                = osc_setup,
         .o_precleanup           = osc_precleanup,
@@ -3373,7 +3369,7 @@ static struct obd_ops osc_obd_ops = {
 };
 
 static struct shrinker *osc_cache_shrinker;
-struct list_head osc_shrink_list = LIST_HEAD_INIT(osc_shrink_list);
+LIST_HEAD(osc_shrink_list);
 DEFINE_SPINLOCK(osc_shrink_lock);
 
 #ifndef HAVE_SHRINKER_COUNT
@@ -3383,10 +3379,6 @@ static int osc_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
                .nr_to_scan = shrink_param(sc, nr_to_scan),
                .gfp_mask   = shrink_param(sc, gfp_mask)
        };
-#if !defined(HAVE_SHRINKER_WANT_SHRINK_PTR) && !defined(HAVE_SHRINK_CONTROL)
-       struct shrinker *shrinker = NULL;
-#endif
-
        (void)osc_cache_shrink_scan(shrinker, &scv);
 
        return osc_cache_shrink_count(shrinker, &scv);