Whamcloud - gitweb
LU-12681 osc: wrong cache of LVB attrs
[fs/lustre-release.git] / lustre / osc / osc_request.c
index 7ba86bc..96a077b 100644 (file)
@@ -41,7 +41,6 @@
 #include <uapi/linux/lustre/lustre_ioctl.h>
 #include <lustre_net.h>
 #include <lustre_obdo.h>
-#include <uapi/linux/lustre/lustre_param.h>
 #include <obd.h>
 #include <obd_cksum.h>
 #include <obd_class.h>
@@ -180,24 +179,25 @@ out:
 }
 
 static int osc_setattr_interpret(const struct lu_env *env,
-                                 struct ptlrpc_request *req,
-                                 struct osc_setattr_args *sa, int rc)
+                                struct ptlrpc_request *req, void *args, int rc)
 {
-        struct ost_body *body;
-        ENTRY;
+       struct osc_setattr_args *sa = args;
+       struct ost_body *body;
 
-        if (rc != 0)
-                GOTO(out, rc);
+       ENTRY;
 
-        body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        if (body == NULL)
-                GOTO(out, rc = -EPROTO);
+       if (rc != 0)
+               GOTO(out, rc);
+
+       body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+       if (body == NULL)
+               GOTO(out, rc = -EPROTO);
 
        lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
                             &body->oa);
 out:
-        rc = sa->sa_upcall(sa->sa_cookie, rc);
-        RETURN(rc);
+       rc = sa->sa_upcall(sa->sa_cookie, rc);
+       RETURN(rc);
 }
 
 int osc_setattr_async(struct obd_export *exp, struct obdo *oa,
@@ -229,11 +229,9 @@ int osc_setattr_async(struct obd_export *exp, struct obdo *oa,
                /* Do not wait for response. */
                ptlrpcd_add_req(req);
        } else {
-               req->rq_interpret_reply =
-                       (ptlrpc_interpterer_t)osc_setattr_interpret;
+               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;
@@ -321,8 +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;
@@ -417,9 +414,8 @@ int osc_punch_send(struct obd_export *exp, struct obdo *oa,
 
        ptlrpc_request_set_replen(req);
 
-       req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_setattr_interpret;
-       CLASSERT(sizeof(*sa) <= sizeof(req->rq_async_args));
-       sa = ptlrpc_req_async_args(req);
+       req->rq_interpret_reply = osc_setattr_interpret;
+       sa = ptlrpc_req_async_args(sa, req);
        sa->sa_oa = oa;
        sa->sa_upcall = upcall;
        sa->sa_cookie = cookie;
@@ -431,14 +427,13 @@ int osc_punch_send(struct obd_export *exp, struct obdo *oa,
 EXPORT_SYMBOL(osc_punch_send);
 
 static int osc_sync_interpret(const struct lu_env *env,
-                              struct ptlrpc_request *req,
-                              void *arg, int rc)
+                             struct ptlrpc_request *req, void *args, int rc)
 {
-       struct osc_fsync_args   *fa = arg;
-       struct ost_body         *body;
-       struct cl_attr          *attr = &osc_env_info(env)->oti_attr;
-       unsigned long           valid = 0;
-       struct cl_object        *obj;
+       struct osc_fsync_args *fa = args;
+       struct ost_body *body;
+       struct cl_attr *attr = &osc_env_info(env)->oti_attr;
+       unsigned long valid = 0;
+       struct cl_object *obj;
        ENTRY;
 
        if (rc != 0)
@@ -498,8 +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;
@@ -549,13 +543,13 @@ static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
 }
 
 static int osc_destroy_interpret(const struct lu_env *env,
-                                struct ptlrpc_request *req, void *data,
-                                int rc)
+                                struct ptlrpc_request *req, void *args, int rc)
 {
        struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
 
        atomic_dec(&cli->cl_destroy_in_flight);
        wake_up(&cli->cl_destroy_waitq);
+
        return 0;
 }
 
@@ -694,8 +688,8 @@ static void osc_announce_cached(struct client_obd *cli, struct obdo *oa,
                /* Do not ask for more than OBD_MAX_GRANT - a margin for server
                 * to add extent tax, etc.
                 */
-               oa->o_undirty = min(undirty, OBD_MAX_GRANT -
-                                   (PTLRPC_MAX_BRW_PAGES << PAGE_SHIFT)*4UL);
+               oa->o_undirty = min(undirty, OBD_MAX_GRANT &
+                                   ~(PTLRPC_MAX_BRW_SIZE * 4UL));
         }
        oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
         oa->o_dropped = cli->cl_lost_grant;
@@ -740,24 +734,26 @@ struct grant_thread_data {
 static struct grant_thread_data client_gtd;
 
 static int osc_shrink_grant_interpret(const struct lu_env *env,
-                                      struct ptlrpc_request *req,
-                                      void *aa, int rc)
+                                     struct ptlrpc_request *req,
+                                     void *args, int rc)
 {
-        struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
-        struct obdo *oa = ((struct osc_grant_args *)aa)->aa_oa;
-        struct ost_body *body;
+       struct osc_grant_args *aa = args;
+       struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
+       struct ost_body *body;
 
-        if (rc != 0) {
-                __osc_update_grant(cli, oa->o_grant);
-                GOTO(out, rc);
-        }
+       if (rc != 0) {
+               __osc_update_grant(cli, aa->aa_oa->o_grant);
+               GOTO(out, rc);
+       }
 
-        body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        LASSERT(body);
-        osc_update_grant(cli, body);
+       body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+       LASSERT(body);
+       osc_update_grant(cli, body);
 out:
-        OBDO_FREE(oa);
-        return rc;
+       OBD_SLAB_FREE_PTR(aa->aa_oa, osc_obdo_kmem);
+       aa->aa_oa = NULL;
+
+       return rc;
 }
 
 static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
@@ -817,6 +813,11 @@ int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
        osc_announce_cached(cli, &body->oa, 0);
 
        spin_lock(&cli->cl_loi_list_lock);
+       if (target_bytes >= cli->cl_avail_grant) {
+               /* available grant has changed since target calculation */
+               spin_unlock(&cli->cl_loi_list_lock);
+               GOTO(out_free, rc = 0);
+       }
        body->oa.o_grant = cli->cl_avail_grant - target_bytes;
        cli->cl_avail_grant = target_bytes;
        spin_unlock(&cli->cl_loi_list_lock);
@@ -832,6 +833,7 @@ int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
                                 sizeof(*body), body, NULL);
         if (rc != 0)
                 __osc_update_grant(cli, body->oa.o_grant);
+out_free:
         OBD_FREE_PTR(body);
         RETURN(rc);
 }
@@ -877,9 +879,11 @@ static void osc_grant_work_handler(struct work_struct *data)
        mutex_lock(&client_gtd.gtd_mutex);
        list_for_each_entry(cli, &client_gtd.gtd_clients,
                            cl_grant_chain) {
-               if (++rpc_sent < GRANT_SHRINK_RPC_BATCH &&
-                   osc_should_shrink_grant(cli))
+               if (rpc_sent < GRANT_SHRINK_RPC_BATCH &&
+                   osc_should_shrink_grant(cli)) {
                        osc_shrink_grant(cli);
+                       rpc_sent++;
+               }
 
                if (!init_next_shrink) {
                        if (cli->cl_next_shrink_grant < next_shrink &&
@@ -895,12 +899,19 @@ 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)
+{
+       cancel_delayed_work_sync(&work);
+       schedule_work(&work.work);
 }
 
 /**
@@ -1055,8 +1066,11 @@ static int check_write_rcs(struct ptlrpc_request *req,
 
         /* return error if any niobuf was in error */
         for (i = 0; i < niocount; i++) {
-                if ((int)remote_rcs[i] < 0)
-                        return(remote_rcs[i]);
+               if ((int)remote_rcs[i] < 0) {
+                       CDEBUG(D_INFO, "rc[%d]: %d req %p\n",
+                              i, remote_rcs[i], req);
+                       return remote_rcs[i];
+               }
 
                 if (remote_rcs[i] != 0) {
                         CDEBUG(D_INFO, "rc[%d] invalid (%d) req %p\n",
@@ -1085,7 +1099,7 @@ static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
                  * safe to combine */
                 if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
                         CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
-                              "report this at https://jira.hpdd.intel.com/\n",
+                              "report this at https://jira.whamcloud.com/\n",
                               p1->flag, p2->flag);
                 }
                 return 0;
@@ -1094,13 +1108,14 @@ static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
         return (p1->off + p1->count == p2->off);
 }
 
+#if IS_ENABLED(CONFIG_CRC_T10DIF)
 static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
                                   size_t pg_count, struct brw_page **pga,
                                   int opc, obd_dif_csum_fn *fn,
                                   int sector_size,
                                   u32 *check_sum)
 {
-       struct cfs_crypto_hash_desc *hdesc;
+       struct ahash_request *req;
        /* Used Adler as the default checksum type on top of DIF tags */
        unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP);
        struct page *__page;
@@ -1120,9 +1135,9 @@ static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
        if (__page == NULL)
                return -ENOMEM;
 
-       hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
-       if (IS_ERR(hdesc)) {
-               rc = PTR_ERR(hdesc);
+       req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
+       if (IS_ERR(req)) {
+               rc = PTR_ERR(req);
                CERROR("%s: unable to initialize checksum hash %s: rc = %d\n",
                       obd_name, cfs_crypto_hash_name(cfs_alg), rc);
                GOTO(out, rc);
@@ -1149,7 +1164,8 @@ static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
                 * The left guard number should be able to hold checksums of a
                 * whole page
                 */
-               rc = obd_page_dif_generate_buffer(obd_name, pga[i]->pg, 0,
+               rc = obd_page_dif_generate_buffer(obd_name, pga[i]->pg,
+                                                 pga[i]->off & ~PAGE_MASK,
                                                  count,
                                                  guard_start + used_number,
                                                  guard_number - used_number,
@@ -1160,7 +1176,7 @@ static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
 
                used_number += used;
                if (used_number == guard_number) {
-                       cfs_crypto_hash_update_page(hdesc, __page, 0,
+                       cfs_crypto_hash_update_page(req, __page, 0,
                                used_number * sizeof(*guard_start));
                        used_number = 0;
                }
@@ -1174,11 +1190,11 @@ static int osc_checksum_bulk_t10pi(const char *obd_name, int nob,
                GOTO(out, rc);
 
        if (used_number != 0)
-               cfs_crypto_hash_update_page(hdesc, __page, 0,
+               cfs_crypto_hash_update_page(req, __page, 0,
                        used_number * sizeof(*guard_start));
 
        bufsize = sizeof(cksum);
-       cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
+       cfs_crypto_hash_final(req, (unsigned char *)&cksum, &bufsize);
 
        /* For sending we only compute the wrong checksum instead
         * of corrupting the data so it is still correct on a redo */
@@ -1190,6 +1206,12 @@ out:
        __free_page(__page);
        return rc;
 }
+#else /* !CONFIG_CRC_T10DIF */
+#define obd_dif_ip_fn NULL
+#define obd_dif_crc_fn NULL
+#define osc_checksum_bulk_t10pi(name, nob, pgc, pga, opc, fn, ssize, csum)  \
+       -EOPNOTSUPP
+#endif /* CONFIG_CRC_T10DIF */
 
 static int osc_checksum_bulk(int nob, size_t pg_count,
                             struct brw_page **pga, int opc,
@@ -1197,17 +1219,17 @@ static int osc_checksum_bulk(int nob, size_t pg_count,
                             u32 *cksum)
 {
        int                             i = 0;
-       struct cfs_crypto_hash_desc     *hdesc;
+       struct ahash_request           *req;
        unsigned int                    bufsize;
        unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
 
        LASSERT(pg_count > 0);
 
-       hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
-       if (IS_ERR(hdesc)) {
+       req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
+       if (IS_ERR(req)) {
                CERROR("Unable to initialize checksum hash %s\n",
                       cfs_crypto_hash_name(cfs_alg));
-               return PTR_ERR(hdesc);
+               return PTR_ERR(req);
        }
 
        while (nob > 0 && pg_count > 0) {
@@ -1223,7 +1245,7 @@ static int osc_checksum_bulk(int nob, size_t pg_count,
                        memcpy(ptr + off, "bad1", min_t(typeof(nob), 4, nob));
                        kunmap(pga[i]->pg);
                }
-               cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
+               cfs_crypto_hash_update_page(req, pga[i]->pg,
                                            pga[i]->off & ~PAGE_MASK,
                                            count);
                LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
@@ -1235,7 +1257,7 @@ static int osc_checksum_bulk(int nob, size_t pg_count,
        }
 
        bufsize = sizeof(*cksum);
-       cfs_crypto_hash_final(hdesc, (unsigned char *)cksum, &bufsize);
+       cfs_crypto_hash_final(req, (unsigned char *)cksum, &bufsize);
 
        /* For sending we only compute the wrong checksum instead
         * of corrupting the data so it is still correct on a redo */
@@ -1317,9 +1339,9 @@ osc_brw_prep_request(int cmd, struct client_obd *cli, struct obdo *oa,
        for (i = 0; i < page_count; i++)
                short_io_size += pga[i]->count;
 
-       /* Check if we can do a short io. */
-       if (!(short_io_size <= cli->cl_short_io_bytes && niocount == 1 &&
-           imp_connect_shortio(cli->cl_import)))
+       /* Check if read/write is small enough to be a short io. */
+       if (short_io_size > cli->cl_max_short_io_bytes || niocount > 1 ||
+           !imp_connect_shortio(cli->cl_import))
                short_io_size = 0;
 
        req_capsule_set_size(pill, &RMF_SHORT_IO, RCL_CLIENT,
@@ -1524,8 +1546,7 @@ 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;
@@ -1603,7 +1624,7 @@ 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);
@@ -1619,7 +1640,6 @@ check_write_checksum(struct obdo *oa, const struct lnet_process_id *peer,
        enum cksum_types cksum_type;
        obd_dif_csum_fn *fn = NULL;
        int sector_size = 0;
-       bool t10pi = false;
        __u32 new_cksum;
        char *msg;
        int rc;
@@ -1638,22 +1658,18 @@ check_write_checksum(struct obdo *oa, const struct lnet_process_id *peer,
 
        switch (cksum_type) {
        case OBD_CKSUM_T10IP512:
-               t10pi = true;
                fn = obd_dif_ip_fn;
                sector_size = 512;
                break;
        case OBD_CKSUM_T10IP4K:
-               t10pi = true;
                fn = obd_dif_ip_fn;
                sector_size = 4096;
                break;
        case OBD_CKSUM_T10CRC512:
-               t10pi = true;
                fn = obd_dif_crc_fn;
                sector_size = 512;
                break;
        case OBD_CKSUM_T10CRC4K:
-               t10pi = true;
                fn = obd_dif_crc_fn;
                sector_size = 4096;
                break;
@@ -1661,13 +1677,10 @@ check_write_checksum(struct obdo *oa, const struct lnet_process_id *peer,
                break;
        }
 
-       if (t10pi)
+       if (fn)
                rc = osc_checksum_bulk_t10pi(obd_name, aa->aa_requested_nob,
-                                            aa->aa_page_count,
-                                            aa->aa_ppga,
-                                            OST_WRITE,
-                                            fn,
-                                            sector_size,
+                                            aa->aa_page_count, aa->aa_ppga,
+                                            OST_WRITE, fn, sector_size,
                                             &new_cksum);
        else
                rc = osc_checksum_bulk(aa->aa_requested_nob, aa->aa_page_count,
@@ -1715,19 +1728,20 @@ 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 &&
@@ -1735,42 +1749,45 @@ 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, qid, body->oa.o_valid,
+                      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,
@@ -1780,20 +1797,20 @@ 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 */
@@ -1887,32 +1904,34 @@ 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);
 
@@ -1934,11 +1953,13 @@ static int osc_brw_redo_request(struct ptlrpc_request *request,
                         }
                 }
         }
-        /* New request takes over pga and oaps from old request.
-         * Note that copying a list_head doesn't work, need to move it... */
-        aa->aa_resends++;
-        new_req->rq_interpret_reply = request->rq_interpret_reply;
-        new_req->rq_async_args = request->rq_async_args;
+       /*
+        * New request takes over pga and oaps from old request.
+        * Note that copying a list_head doesn't work, need to move it...
+        */
+       aa->aa_resends++;
+       new_req->rq_interpret_reply = request->rq_interpret_reply;
+       new_req->rq_async_args = request->rq_async_args;
        new_req->rq_commit_cb = request->rq_commit_cb;
        /* cap resend delay to the current request timeout, this is similar to
         * what ptlrpc does (see after_reply()) */
@@ -1949,7 +1970,7 @@ 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);
@@ -2012,19 +2033,22 @@ static void osc_release_ppga(struct brw_page **ppga, size_t count)
 }
 
 static int brw_interpret(const struct lu_env *env,
-                         struct ptlrpc_request *req, void *data, int rc)
+                        struct ptlrpc_request *req, void *args, int rc)
 {
-       struct osc_brw_async_args *aa = data;
+       struct osc_brw_async_args *aa = args;
        struct osc_extent *ext;
        struct osc_extent *tmp;
        struct client_obd *cli = aa->aa_cli;
-       unsigned long           transferred = 0;
-        ENTRY;
+       unsigned long transferred = 0;
+
+       ENTRY;
 
-        rc = osc_brw_fini_request(req, rc);
-        CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
-        /* When server return -EINPROGRESS, client should always retry
-         * regardless of the number of times the bulk was resent already. */
+       rc = osc_brw_fini_request(req, rc);
+       CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
+       /*
+        * When server returns -EINPROGRESS, client should always retry
+        * regardless of the number of times the bulk was resent already.
+        */
        if (osc_recoverable_error(rc) && !req->rq_no_delay) {
                if (req->rq_import_generation !=
                    req->rq_import->imp_generation) {
@@ -2099,7 +2123,8 @@ static int brw_interpret(const struct lu_env *env,
                        cl_object_attr_update(env, obj, attr, valid);
                cl_object_attr_unlock(obj);
        }
-       OBDO_FREE(aa->aa_oa);
+       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);
@@ -2204,7 +2229,7 @@ int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
        if (pga == NULL)
                GOTO(out, rc = -ENOMEM);
 
-       OBDO_ALLOC(oa);
+       OBD_SLAB_ALLOC_PTR_GFP(oa, osc_obdo_kmem, GFP_NOFS);
        if (oa == NULL)
                GOTO(out, rc = -ENOMEM);
 
@@ -2291,8 +2316,7 @@ 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);
@@ -2315,7 +2339,7 @@ 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);
@@ -2332,7 +2356,7 @@ out:
                LASSERT(req == NULL);
 
                if (oa)
-                       OBDO_FREE(oa);
+                       OBD_SLAB_FREE_PTR(oa, osc_obdo_kmem);
                if (pga)
                        OBD_FREE(pga, sizeof(*pga) * page_count);
                /* this should happen rarely and is pretty bad, it makes the
@@ -2404,8 +2428,9 @@ int osc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
 }
 
 int osc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
-                         struct osc_enqueue_args *aa, int rc)
+                         void *args, int rc)
 {
+       struct osc_enqueue_args *aa = args;
        struct ldlm_lock *lock;
        struct lustre_handle *lockh = &aa->oa_lockh;
        enum ldlm_mode mode = aa->oa_mode;
@@ -2466,9 +2491,8 @@ 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)
 {
@@ -2486,15 +2510,6 @@ 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/
@@ -2557,7 +2572,6 @@ 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);
 
@@ -2586,8 +2600,7 @@ 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;
@@ -2606,8 +2619,7 @@ no_match:
                                aa->oa_flags  = NULL;
                        }
 
-                       req->rq_interpret_reply =
-                               (ptlrpc_interpterer_t)osc_enqueue_interpret;
+                       req->rq_interpret_reply = osc_enqueue_interpret;
                        if (rqset == PTLRPCD_SET)
                                ptlrpcd_add_req(req);
                        else
@@ -2670,36 +2682,36 @@ int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
 }
 
 static int osc_statfs_interpret(const struct lu_env *env,
-                                struct ptlrpc_request *req,
-                                struct osc_async_args *aa, int rc)
+                               struct ptlrpc_request *req, void *args, int rc)
 {
-        struct obd_statfs *msfs;
-        ENTRY;
+       struct osc_async_args *aa = args;
+       struct obd_statfs *msfs;
 
-        if (rc == -EBADR)
-                /* The request has in fact never been sent
-                 * due to issues at a higher level (LOV).
-                 * Exit immediately since the caller is
-                 * aware of the problem and takes care
-                 * of the clean up */
-                 RETURN(rc);
+       ENTRY;
+       if (rc == -EBADR)
+               /*
+                * The request has in fact never been sent due to issues at
+                * a higher level (LOV).  Exit immediately since the caller
+                * is aware of the problem and takes care of the clean up.
+                */
+               RETURN(rc);
 
-        if ((rc == -ENOTCONN || rc == -EAGAIN) &&
-            (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
-                GOTO(out, rc = 0);
+       if ((rc == -ENOTCONN || rc == -EAGAIN) &&
+           (aa->aa_oi->oi_flags & OBD_STATFS_NODELAY))
+               GOTO(out, rc = 0);
 
-        if (rc != 0)
-                GOTO(out, rc);
+       if (rc != 0)
+               GOTO(out, rc);
 
-        msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
-        if (msfs == NULL) {
+       msfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
+       if (msfs == NULL)
                 GOTO(out, rc = -EPROTO);
-        }
 
-        *aa->aa_oi->oi_osfs = *msfs;
+       *aa->aa_oi->oi_osfs = *msfs;
 out:
-        rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
-        RETURN(rc);
+       rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc);
+
+       RETURN(rc);
 }
 
 static int osc_statfs_async(struct obd_export *exp,
@@ -2712,6 +2724,22 @@ static int osc_statfs_async(struct obd_export *exp,
        int rc;
         ENTRY;
 
+       if (obd->obd_osfs_age >= max_age) {
+               CDEBUG(D_SUPER,
+                      "%s: use %p cache blocks %llu/%llu objects %llu/%llu\n",
+                      obd->obd_name, &obd->obd_osfs,
+                      obd->obd_osfs.os_bavail, obd->obd_osfs.os_blocks,
+                      obd->obd_osfs.os_ffree, obd->obd_osfs.os_files);
+               spin_lock(&obd->obd_osfs_lock);
+               memcpy(oinfo->oi_osfs, &obd->obd_osfs, sizeof(*oinfo->oi_osfs));
+               spin_unlock(&obd->obd_osfs_lock);
+               oinfo->oi_flags |= OBD_STATFS_FROM_CACHE;
+               if (oinfo->oi_cb_up)
+                       oinfo->oi_cb_up(oinfo, 0);
+
+               RETURN(0);
+       }
+
         /* We could possibly pass max_age in the request (as an absolute
          * timestamp or a "seconds.usec ago") so the target can avoid doing
          * extra calls into the filesystem if that isn't necessary (e.g.
@@ -2737,9 +2765,8 @@ static int osc_statfs_async(struct obd_export *exp,
                req->rq_no_delay = 1;
        }
 
-       req->rq_interpret_reply = (ptlrpc_interpterer_t)osc_statfs_interpret;
-       CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
-       aa = ptlrpc_req_async_args(req);
+       req->rq_interpret_reply = osc_statfs_interpret;
+       aa = ptlrpc_req_async_args(aa, req);
        aa->aa_oi = oinfo;
 
        ptlrpc_set_add_req(rqset, req);
@@ -2813,38 +2840,36 @@ out:
 static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                         void *karg, void __user *uarg)
 {
-        struct obd_device *obd = exp->exp_obd;
-        struct obd_ioctl_data *data = karg;
-        int err = 0;
-        ENTRY;
+       struct obd_device *obd = exp->exp_obd;
+       struct obd_ioctl_data *data = karg;
+       int rc = 0;
 
+       ENTRY;
        if (!try_module_get(THIS_MODULE)) {
                CERROR("%s: cannot get module '%s'\n", obd->obd_name,
                       module_name(THIS_MODULE));
                return -EINVAL;
        }
-        switch (cmd) {
-        case OBD_IOC_CLIENT_RECOVER:
-                err = ptlrpc_recover_import(obd->u.cli.cl_import,
-                                            data->ioc_inlbuf1, 0);
-                if (err > 0)
-                        err = 0;
-                GOTO(out, err);
-        case IOC_OSC_SET_ACTIVE:
-                err = ptlrpc_set_import_active(obd->u.cli.cl_import,
-                                               data->ioc_offset);
-                GOTO(out, err);
-        case OBD_IOC_PING_TARGET:
-                err = ptlrpc_obd_ping(obd);
-                GOTO(out, err);
+       switch (cmd) {
+       case OBD_IOC_CLIENT_RECOVER:
+               rc = ptlrpc_recover_import(obd->u.cli.cl_import,
+                                          data->ioc_inlbuf1, 0);
+               if (rc > 0)
+                       rc = 0;
+               break;
+       case IOC_OSC_SET_ACTIVE:
+               rc = ptlrpc_set_import_active(obd->u.cli.cl_import,
+                                             data->ioc_offset);
+               break;
        default:
-               CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
-                      cmd, current_comm());
-               GOTO(out, err = -ENOTTY);
+               rc = -ENOTTY;
+               CDEBUG(D_INODE, "%s: unrecognised ioctl %#x by %s: rc = %d\n",
+                      obd->obd_name, cmd, current_comm(), rc);
+               break;
        }
-out:
+
        module_put(THIS_MODULE);
-       return err;
+       return rc;
 }
 
 int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
@@ -2877,23 +2902,6 @@ int osc_set_info_async(const struct lu_env *env, struct obd_export *exp,
                 RETURN(0);
         }
 
-       if (KEY_IS(KEY_CACHE_SET)) {
-               struct client_obd *cli = &obd->u.cli;
-
-               LASSERT(cli->cl_cache == NULL); /* only once */
-               cli->cl_cache = (struct cl_client_cache *)val;
-               cl_cache_incref(cli->cl_cache);
-               cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
-
-               /* add this osc into entity list */
-               LASSERT(list_empty(&cli->cl_lru_osc));
-               spin_lock(&cli->cl_cache->ccc_lru_lock);
-               list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
-               spin_unlock(&cli->cl_cache->ccc_lru_lock);
-
-               RETURN(0);
-       }
-
        if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
                struct client_obd *cli = &obd->u.cli;
                long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
@@ -2942,9 +2950,8 @@ 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);
-               OBDO_ALLOC(oa);
+               aa = ptlrpc_req_async_args(aa, req);
+               OBD_SLAB_ALLOC_PTR_GFP(oa, osc_obdo_kmem, GFP_NOFS);
                if (!oa) {
                        ptlrpc_req_finished(req);
                        RETURN(-ENOMEM);
@@ -2979,10 +2986,13 @@ int osc_reconnect(const struct lu_env *env, struct obd_export *exp,
 
                spin_lock(&cli->cl_loi_list_lock);
                grant = cli->cl_avail_grant + cli->cl_reserved_grant;
-               if (data->ocd_connect_flags & OBD_CONNECT_GRANT_PARAM)
+               if (data->ocd_connect_flags & OBD_CONNECT_GRANT_PARAM) {
+                       /* restore ocd_grant_blkbits as client page bits */
+                       data->ocd_grant_blkbits = PAGE_SHIFT;
                        grant += cli->cl_dirty_grant;
-               else
+               } else {
                        grant += cli->cl_dirty_pages << PAGE_SHIFT;
+               }
                data->ocd_grant = grant ? : 2 * cli_brw_size(obd);
                lost_grant = cli->cl_lost_grant;
                cli->cl_lost_grant = 0;
@@ -3144,7 +3154,7 @@ static int osc_cancel_weight(struct ldlm_lock *lock)
         * Cancel all unused and granted extent lock.
         */
        if (lock->l_resource->lr_type == LDLM_EXTENT &&
-           lock->l_granted_mode == lock->l_req_mode &&
+           ldlm_is_granted(lock) &&
            osc_ldlm_weigh_ast(lock) == 0)
                RETURN(1);
 
@@ -3252,6 +3262,7 @@ int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
        list_add_tail(&cli->cl_shrink_list, &osc_shrink_list);
        spin_unlock(&osc_shrink_lock);
        cli->cl_import->imp_idle_timeout = osc_idle_timeout;
+       cli->cl_import->imp_idle_debug = D_HA;
 
        RETURN(0);
 }
@@ -3328,18 +3339,6 @@ int osc_cleanup_common(struct obd_device *obd)
 }
 EXPORT_SYMBOL(osc_cleanup_common);
 
-int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
-{
-       ssize_t count  = class_modify_config(lcfg, PARAM_OSC,
-                                            &obd->obd_kset.kobj);
-       return count > 0 ? 0 : count;
-}
-
-static int osc_process_config(struct obd_device *obd, size_t len, void *buf)
-{
-        return osc_process_config_base(obd, buf);
-}
-
 static struct obd_ops osc_obd_ops = {
         .o_owner                = THIS_MODULE,
         .o_setup                = osc_setup,
@@ -3347,7 +3346,7 @@ static struct obd_ops osc_obd_ops = {
        .o_cleanup              = osc_cleanup_common,
         .o_add_conn             = client_import_add_conn,
         .o_del_conn             = client_import_del_conn,
-        .o_connect              = client_connect_import,
+       .o_connect              = client_connect_import,
         .o_reconnect            = osc_reconnect,
         .o_disconnect           = osc_disconnect,
         .o_statfs               = osc_statfs,
@@ -3359,7 +3358,6 @@ static struct obd_ops osc_obd_ops = {
         .o_iocontrol            = osc_iocontrol,
         .o_set_info_async       = osc_set_info_async,
         .o_import_event         = osc_import_event,
-        .o_process_config       = osc_process_config,
         .o_quotactl             = osc_quotactl,
 };
 
@@ -3386,8 +3384,6 @@ static int osc_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
 
 static int __init osc_init(void)
 {
-       bool enable_proc = true;
-       struct obd_type *type;
        unsigned int reqpool_size;
        unsigned int reqsize;
        int rc;
@@ -3404,11 +3400,7 @@ static int __init osc_init(void)
        if (rc)
                RETURN(rc);
 
-       type = class_search_type(LUSTRE_OSP_NAME);
-       if (type != NULL && type->typ_procsym != NULL)
-               enable_proc = false;
-
-       rc = class_register_type(&osc_obd_ops, NULL, enable_proc, NULL,
+       rc = class_register_type(&osc_obd_ops, NULL, true, NULL,
                                 LUSTRE_OSC_NAME, &osc_device_type);
        if (rc)
                GOTO(out_kmem, rc);