X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fosc%2Fosc_request.c;h=7cdf82f0510157a7b4fa5960e5939387b9470a80;hb=32ff1453bab3b1771d46b717df1ec479d156e5f8;hp=713b6a8711e4417142f6f883766aed2cb4f0b27a;hpb=0a859380c36ac24871f221b35042f76c56b04438;p=fs%2Flustre-release.git diff --git a/lustre/osc/osc_request.c b/lustre/osc/osc_request.c index 713b6a8..7cdf82f 100644 --- a/lustre/osc/osc_request.c +++ b/lustre/osc/osc_request.c @@ -26,7 +26,7 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* @@ -564,16 +564,40 @@ static int osc_punch(struct obd_export *exp, struct obd_info *oinfo, oinfo->oi_cb_up, oinfo, rqset); } -static int osc_sync(struct obd_export *exp, struct obdo *oa, - struct lov_stripe_md *md, obd_size start, obd_size end, - void *capa) +static int osc_sync_interpret(const struct lu_env *env, + struct ptlrpc_request *req, + void *arg, int rc) +{ + struct osc_async_args *aa = arg; + struct ost_body *body; + ENTRY; + + if (rc) + GOTO(out, rc); + + body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); + if (body == NULL) { + CERROR ("can't unpack ost_body\n"); + GOTO(out, rc = -EPROTO); + } + + *aa->aa_oi->oi_oa = body->oa; +out: + rc = aa->aa_oi->oi_cb_up(aa->aa_oi, rc); + RETURN(rc); +} + +static int osc_sync(struct obd_export *exp, struct obd_info *oinfo, + obd_size start, obd_size end, + struct ptlrpc_request_set *set) { struct ptlrpc_request *req; struct ost_body *body; + struct osc_async_args *aa; int rc; ENTRY; - if (!oa) { + if (!oinfo->oi_oa) { CDEBUG(D_INFO, "oa NULL\n"); RETURN(-EINVAL); } @@ -582,7 +606,7 @@ static int osc_sync(struct obd_export *exp, struct obdo *oa, if (req == NULL) RETURN(-ENOMEM); - osc_set_capa_size(req, &RMF_CAPA1, capa); + osc_set_capa_size(req, &RMF_CAPA1, oinfo->oi_capa); rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SYNC); if (rc) { ptlrpc_request_free(req); @@ -592,28 +616,21 @@ static int osc_sync(struct obd_export *exp, struct obdo *oa, /* overload the size and blocks fields in the oa with start/end */ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); LASSERT(body); - lustre_set_wire_obdo(&body->oa, oa); + lustre_set_wire_obdo(&body->oa, oinfo->oi_oa); body->oa.o_size = start; body->oa.o_blocks = end; body->oa.o_valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS); - osc_pack_capa(req, body, capa); + osc_pack_capa(req, body, oinfo->oi_capa); ptlrpc_request_set_replen(req); + req->rq_interpret_reply = osc_sync_interpret; - rc = ptlrpc_queue_wait(req); - if (rc) - 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(oa, &body->oa); + CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args)); + aa = ptlrpc_req_async_args(req); + aa->aa_oi = oinfo; - EXIT; - out: - ptlrpc_req_finished(req); - return rc; + ptlrpc_set_add_req(set, req); + RETURN (0); } /* Find and cancel locally locks matched by @mode in the resource found by @@ -943,7 +960,7 @@ static int osc_shrink_grant_interpret(const struct lu_env *env, LASSERT(body); osc_update_grant(cli, body); out: - OBD_FREE_PTR(oa); + OBDO_FREE(oa); return rc; } @@ -953,6 +970,10 @@ static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa) oa->o_grant = cli->cl_avail_grant / 4; cli->cl_avail_grant -= oa->o_grant; client_obd_list_unlock(&cli->cl_loi_list_lock); + if (!(oa->o_valid & OBD_MD_FLFLAGS)) { + oa->o_valid |= OBD_MD_FLFLAGS; + oa->o_flags = 0; + } oa->o_flags |= OBD_FL_SHRINK_GRANT; osc_update_next_shrink(cli); } @@ -1003,6 +1024,10 @@ int osc_shrink_grant_to_target(struct client_obd *cli, long target) body->oa.o_grant = cli->cl_avail_grant - target; cli->cl_avail_grant = target; client_obd_list_unlock(&cli->cl_loi_list_lock); + if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) { + body->oa.o_valid |= OBD_MD_FLFLAGS; + body->oa.o_flags = 0; + } body->oa.o_flags |= OBD_FL_SHRINK_GRANT; osc_update_next_shrink(cli); @@ -1020,6 +1045,11 @@ static int osc_should_shrink_grant(struct client_obd *client) { cfs_time_t time = cfs_time_current(); cfs_time_t next_shrink = client->cl_next_shrink_grant; + + if ((client->cl_import->imp_connect_data.ocd_connect_flags & + OBD_CONNECT_GRANT_SHRINK) == 0) + return 0; + if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) { if (client->cl_import->imp_state == LUSTRE_IMP_FULL && client->cl_avail_grant > GRANT_SHRINK_LIMIT) @@ -1082,11 +1112,21 @@ static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd) cli->cl_avail_grant = ocd->ocd_grant; else cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty; + + if (cli->cl_avail_grant < 0) { + CWARN("%s: available grant < 0, the OSS is probably not running" + " with patch from bug20278 (%ld) \n", + cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant); + /* workaround for 1.6 servers which do not have + * the patch from bug20278 */ + cli->cl_avail_grant = ocd->ocd_grant; + } + client_obd_list_unlock(&cli->cl_loi_list_lock); - CDEBUG(D_CACHE, "setting cl_avail_grant: %ld cl_lost_grant: %ld \n", + CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld \n", + cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant, cli->cl_lost_grant); - LASSERT(cli->cl_avail_grant >= 0); if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK && cfs_list_empty(&cli->cl_grant_shrink_list)) @@ -1172,7 +1212,7 @@ static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2) { if (p1->flag != p2->flag) { unsigned mask = ~(OBD_BRW_FROM_GRANT| - OBD_BRW_NOCACHE|OBD_BRW_SYNC); + OBD_BRW_NOCACHE|OBD_BRW_SYNC|OBD_BRW_ASYNC); /* warn if we try to combine flags that we don't know to be * safe to combine */ @@ -1225,7 +1265,8 @@ static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa, struct lov_stripe_md *lsm, obd_count page_count, struct brw_page **pga, struct ptlrpc_request **reqp, - struct obd_capa *ocapa, int reserve) + struct obd_capa *ocapa, int reserve, + int resend) { struct ptlrpc_request *req; struct ptlrpc_bulk_desc *desc; @@ -1341,6 +1382,14 @@ static int osc_brw_prep_request(int cmd, struct client_obd *cli,struct obdo *oa, &RMF_NIOBUF_REMOTE), (void *)(niobuf - niocount)); osc_announce_cached(cli, &body->oa, opc == OST_WRITE ? requested_nob:0); + if (resend) { + if ((body->oa.o_valid & OBD_MD_FLFLAGS) == 0) { + body->oa.o_valid |= OBD_MD_FLFLAGS; + body->oa.o_flags = 0; + } + body->oa.o_flags |= OBD_FL_RECOV_RESEND; + } + if (osc_should_shrink_grant(cli)) osc_shrink_grant_local(cli, &body->oa); @@ -1422,6 +1471,10 @@ static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer, return 0; } + /* If this is mmaped file - it can be changed at any time */ + if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & OBD_FL_MMAP) + return 1; + if (oa->o_valid & OBD_MD_FLFLAGS) cksum_type = cksum_type_unpack(oa->o_flags); else @@ -1495,14 +1548,14 @@ static int osc_brw_fini_request(struct ptlrpc_request *req, int rc) } #endif + osc_update_grant(cli, body); + if (rc < 0) RETURN(rc); if (aa->aa_oa->o_valid & OBD_MD_FLCKSUM) client_cksum = aa->aa_oa->o_cksum; /* save for later */ - osc_update_grant(cli, body); - if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) { if (rc > 0) { CERROR("Unexpected +ve rc %d\n", rc); @@ -1639,7 +1692,7 @@ static int osc_brw_internal(int cmd, struct obd_export *exp, struct obdo *oa, restart_bulk: rc = osc_brw_prep_request(cmd, &exp->exp_obd->u.cli, oa, lsm, - page_count, pga, &req, ocapa, 0); + page_count, pga, &req, ocapa, 0, resends); if (rc != 0) return (rc); @@ -1656,7 +1709,7 @@ restart_bulk: ptlrpc_req_finished(req); if (osc_recoverable_error(rc)) { resends++; - if (!osc_should_resend(resends, &exp->exp_obd->u.cli)) { + if (!client_should_resend(resends, &exp->exp_obd->u.cli)) { CERROR("too many resend retries, returning error\n"); RETURN(-EIO); } @@ -1680,8 +1733,8 @@ int osc_brw_redo_request(struct ptlrpc_request *request, int rc = 0; ENTRY; - if (!osc_should_resend(aa->aa_resends, aa->aa_cli)) { - CERROR("too many resend retries, returning error\n"); + if (!client_should_resend(aa->aa_resends, aa->aa_cli)) { + CERROR("too many resent retries, returning error\n"); RETURN(-EIO); } @@ -1692,7 +1745,7 @@ int osc_brw_redo_request(struct ptlrpc_request *request, aa->aa_cli, aa->aa_oa, NULL /* lsm unused by osc currently */, aa->aa_page_count, aa->aa_ppga, - &new_req, aa->aa_ocapa, 0); + &new_req, aa->aa_ocapa, 0, 1); if (rc) RETURN(rc); @@ -1938,7 +1991,7 @@ static int lop_makes_rpc(struct client_obd *cli, struct loi_oap_pages *lop, if (cmd & OBD_BRW_WRITE) { /* trigger a write rpc stream as long as there are dirtiers * waiting for space. as they're waiting, they're not going to - * create more pages to coallesce with what's waiting.. */ + * create more pages to coalesce with what's waiting.. */ if (!cfs_list_empty(&cli->cl_cache_waiters)) { CDEBUG(D_CACHE, "cache waiters forcing RPC\n"); RETURN(1); @@ -2162,9 +2215,20 @@ static int brw_interpret(const struct lu_env *env, rc = osc_brw_fini_request(req, rc); CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc); if (osc_recoverable_error(rc)) { - rc = osc_brw_redo_request(req, aa); - if (rc == 0) - RETURN(0); + /* Only retry once for mmaped files since the mmaped page + * might be modified at anytime. We have to retry at least + * once in case there WAS really a corruption of the page + * on the network, that was not caused by mmap() modifying + * the page. Bug11742 */ + if ((rc == -EAGAIN) && (aa->aa_resends > 0) && + aa->aa_oa->o_valid & OBD_MD_FLFLAGS && + aa->aa_oa->o_flags & OBD_FL_MMAP) { + rc = 0; + } else { + rc = osc_brw_redo_request(req, aa); + if (rc == 0) + RETURN(0); + } } if (aa->aa_ocapa) { @@ -2196,12 +2260,9 @@ static int brw_interpret(const struct lu_env *env, } OBDO_FREE(aa->aa_oa); } else { /* from async_internal() */ - int i; + obd_count i; for (i = 0; i < aa->aa_page_count; i++) osc_release_write_grant(aa->aa_cli, aa->aa_ppga[i], 1); - - if (aa->aa_oa->o_flags & OBD_FL_TEMPORARY) - OBDO_FREE(aa->aa_oa); } osc_wake_cache_waiters(cli); osc_check_rpcs(env, cli); @@ -2209,6 +2270,7 @@ static int brw_interpret(const struct lu_env *env, if (!async) cl_req_completion(env, aa->aa_clerq, rc); osc_release_ppga(aa->aa_ppga, aa->aa_page_count); + RETURN(rc); } @@ -2230,11 +2292,14 @@ static struct ptlrpc_request *osc_build_req(const struct lu_env *env, enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ; struct ldlm_lock *lock = NULL; struct cl_req_attr crattr; - int i, rc; + int i, rc, mpflag = 0; ENTRY; LASSERT(!cfs_list_empty(rpc_list)); + if (cmd & OBD_BRW_MEMALLOC) + mpflag = cfs_memory_pressure_get_and_set(); + memset(&crattr, 0, sizeof crattr); OBD_ALLOC(pga, sizeof(*pga) * page_count); if (pga == NULL) @@ -2284,12 +2349,15 @@ static struct ptlrpc_request *osc_build_req(const struct lu_env *env, sort_brw_pages(pga, page_count); rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count, - pga, &req, crattr.cra_capa, 1); + pga, &req, crattr.cra_capa, 1, 0); if (rc != 0) { CERROR("prep_req failed: %d\n", rc); GOTO(out, req = ERR_PTR(rc)); } + if (cmd & OBD_BRW_MEMALLOC) + req->rq_memalloc = 1; + /* Need to update the timestamps after the request is built in case * we race with setattr (locally or in queue at OST). If OST gets * later setattr before earlier BRW (as determined by the request xid), @@ -2306,6 +2374,9 @@ static struct ptlrpc_request *osc_build_req(const struct lu_env *env, CFS_INIT_LIST_HEAD(rpc_list); aa->aa_clerq = clerq; out: + if (cmd & OBD_BRW_MEMALLOC) + cfs_memory_pressure_restore(mpflag); + capa_put(crattr.cra_capa); if (IS_ERR(req)) { if (oa) @@ -2340,8 +2411,9 @@ out: * \param cmd OBD_BRW_* macroses * \param lop pending pages * - * \return zero if pages successfully add to send queue. - * \return not zere if error occurring. + * \return zero if no page added to send queue. + * \return 1 if pages successfully added to send queue. + * \return negative on errors. */ static int osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli, @@ -2357,7 +2429,7 @@ osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli, CFS_LIST_HEAD(tmp_list); unsigned int ending_offset; unsigned starting_offset = 0; - int srvlock = 0; + int srvlock = 0, mem_tight = 0; struct cl_object *clob = NULL; ENTRY; @@ -2409,7 +2481,7 @@ osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli, * until completion unlocks it. commit_write submits a page * as not ready because its unlock will happen unconditionally * as the call returns. if we race with commit_write giving - * us that page we dont' want to create a hole in the page + * us that page we don't want to create a hole in the page * stream, so we stop and leave the rpc to be fired by * another dirtier or kupdated interval (the not ready page * will still be on the dirty list). we could call in @@ -2501,6 +2573,8 @@ osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli, /* now put the page back in our accounting */ cfs_list_add_tail(&oap->oap_rpc_item, &rpc_list); + if (oap->oap_brw_flags & OBD_BRW_MEMALLOC) + mem_tight = 1; if (page_count == 0) srvlock = !!(oap->oap_brw_flags & OBD_BRW_SRVLOCK); if (++page_count >= cli->cl_max_pages_per_rpc) @@ -2536,7 +2610,8 @@ osc_send_oap_rpc(const struct lu_env *env, struct client_obd *cli, RETURN(0); } - req = osc_build_req(env, cli, &rpc_list, page_count, cmd); + req = osc_build_req(env, cli, &rpc_list, page_count, + mem_tight ? (cmd | OBD_BRW_MEMALLOC) : cmd); if (IS_ERR(req)) { LASSERT(cfs_list_empty(&rpc_list)); loi_list_maint(cli, loi); @@ -2720,7 +2795,7 @@ void osc_check_rpcs(const struct lu_env *env, struct client_obd *cli) race_counter++; } - /* attempt some inter-object balancing by issueing rpcs + /* attempt some inter-object balancing by issuing rpcs * for each object in turn */ if (!cfs_list_empty(&loi->loi_hp_ready_item)) cfs_list_del_init(&loi->loi_hp_ready_item); @@ -2791,7 +2866,7 @@ static int osc_enter_cache(const struct lu_env *env, struct osc_async_page *oap) { struct osc_cache_waiter ocw; - struct l_wait_info lwi = { 0 }; + struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL); ENTRY; @@ -2883,28 +2958,18 @@ int osc_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm, RETURN(0); } -struct osc_async_page *oap_from_cookie(void *cookie) -{ - struct osc_async_page *oap = cookie; - if (oap->oap_magic != OAP_MAGIC) - return ERR_PTR(-EINVAL); - return oap; -}; - -int osc_queue_async_io(const struct lu_env *env, - struct obd_export *exp, struct lov_stripe_md *lsm, - struct lov_oinfo *loi, void *cookie, - int cmd, obd_off off, int count, - obd_flag brw_flags, enum async_flags async_flags) +int osc_queue_async_io(const struct lu_env *env, struct obd_export *exp, + struct lov_stripe_md *lsm, struct lov_oinfo *loi, + struct osc_async_page *oap, int cmd, obd_off off, + int count, obd_flag brw_flags, + enum async_flags async_flags) { struct client_obd *cli = &exp->exp_obd->u.cli; - struct osc_async_page *oap; int rc = 0; ENTRY; - oap = oap_from_cookie(cookie); - if (IS_ERR(oap)) - RETURN(PTR_ERR(oap)); + if (oap->oap_magic != OAP_MAGIC) + RETURN(-EINVAL); if (cli->cl_import == NULL || cli->cl_import->imp_invalid) RETURN(-EIO); @@ -2946,7 +3011,7 @@ int osc_queue_async_io(const struct lu_env *env, oap->oap_count = count; oap->oap_brw_flags = brw_flags; /* Give a hint to OST that requests are coming from kswapd - bug19529 */ - if (libcfs_memory_pressure_get()) + if (cfs_memory_pressure_get()) oap->oap_brw_flags |= OBD_BRW_MEMALLOC; cfs_spin_lock(&oap->oap_lock); oap->oap_async_flags = async_flags; @@ -3016,19 +3081,16 @@ int osc_set_async_flags_base(struct client_obd *cli, RETURN(0); } -int osc_teardown_async_page(struct obd_export *exp, - struct lov_stripe_md *lsm, - struct lov_oinfo *loi, void *cookie) +int osc_teardown_async_page(struct obd_export *exp, struct lov_stripe_md *lsm, + struct lov_oinfo *loi, struct osc_async_page *oap) { struct client_obd *cli = &exp->exp_obd->u.cli; struct loi_oap_pages *lop; - struct osc_async_page *oap; int rc = 0; ENTRY; - oap = oap_from_cookie(cookie); - if (IS_ERR(oap)) - RETURN(PTR_ERR(oap)); + if (oap->oap_magic != OAP_MAGIC) + RETURN(-EINVAL); if (loi == NULL) loi = lsm->lsm_oinfo[0]; @@ -3064,11 +3126,11 @@ out: RETURN(rc); } -static void osc_set_lock_data_with_check(struct ldlm_lock *lock, - struct ldlm_enqueue_info *einfo, - int flags) +static int osc_set_lock_data_with_check(struct ldlm_lock *lock, + struct ldlm_enqueue_info *einfo) { void *data = einfo->ei_cbdata; + int set = 0; LASSERT(lock != NULL); LASSERT(lock->l_blocking_ast == einfo->ei_cb_bl); @@ -3078,24 +3140,31 @@ static void osc_set_lock_data_with_check(struct ldlm_lock *lock, lock_res_and_lock(lock); cfs_spin_lock(&osc_ast_guard); - LASSERT(lock->l_ast_data == NULL || lock->l_ast_data == data); - lock->l_ast_data = data; + + if (lock->l_ast_data == NULL) + lock->l_ast_data = data; + if (lock->l_ast_data == data) + set = 1; + cfs_spin_unlock(&osc_ast_guard); unlock_res_and_lock(lock); + + return set; } -static void osc_set_data_with_check(struct lustre_handle *lockh, - struct ldlm_enqueue_info *einfo, - int flags) +static int osc_set_data_with_check(struct lustre_handle *lockh, + struct ldlm_enqueue_info *einfo) { struct ldlm_lock *lock = ldlm_handle2lock(lockh); + int set = 0; if (lock != NULL) { - osc_set_lock_data_with_check(lock, einfo, flags); + set = osc_set_lock_data_with_check(lock, einfo); LDLM_LOCK_PUT(lock); } else CERROR("lockh %p, data %p - client evicted?\n", lockh, einfo->ei_cbdata); + return set; } static int osc_change_cbdata(struct obd_export *exp, struct lov_stripe_md *lsm, @@ -3183,6 +3252,9 @@ static int osc_enqueue_interpret(const struct lu_env *env, * osc_enqueue_fini(). */ ldlm_lock_addref(&handle, mode); + /* Let CP AST to grant the lock first. */ + OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1); + /* Complete obtaining the lock procedure. */ rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1, mode, aa->oa_flags, aa->oa_lvb, @@ -3304,13 +3376,11 @@ int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id, if (mode) { struct ldlm_lock *matched = ldlm_handle2lock(lockh); - if (matched->l_ast_data == NULL || - matched->l_ast_data == einfo->ei_cbdata) { + if (osc_set_lock_data_with_check(matched, einfo)) { /* addref the lock only if not async requests and PW * lock is matched whereas we asked for PR. */ if (!rqset && einfo->ei_mode != mode) ldlm_lock_addref(lockh, LCK_PR); - osc_set_lock_data_with_check(matched, einfo, *flags); if (intent) { /* I would like to be able to ASSERT here that * rss <= kms, but I can't, for reasons which @@ -3341,8 +3411,10 @@ int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id, RETURN(-ENOMEM); rc = ldlm_prep_enqueue_req(exp, req, &cancels, 0); - if (rc) + if (rc) { + ptlrpc_request_free(req); RETURN(rc); + } req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, sizeof *lvb); @@ -3433,8 +3505,13 @@ int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id, rc = ldlm_lock_match(obd->obd_namespace, lflags, res_id, type, policy, rc, lockh, unref); if (rc) { - if (data != NULL) - osc_set_data_with_check(lockh, data, lflags); + if (data != NULL) { + if (!osc_set_data_with_check(lockh, data)) { + if (!(lflags & LDLM_FL_TEST_LOCK)) + ldlm_lock_decref(lockh, rc); + RETURN(0); + } + } if (!(lflags & LDLM_FL_TEST_LOCK) && mode != rc) { ldlm_lock_addref(lockh, LCK_PR); ldlm_lock_decref(lockh, LCK_PW); @@ -4042,7 +4119,7 @@ static int osc_set_info_async(struct obd_export *exp, obd_count keylen, CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args)); aa = ptlrpc_req_async_args(req); - OBD_ALLOC_PTR(oa); + OBDO_ALLOC(oa); if (!oa) { ptlrpc_req_finished(req); RETURN(-ENOMEM); @@ -4309,6 +4386,14 @@ static int osc_import_event(struct obd_device *obd, rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL); break; } + case IMP_EVENT_DEACTIVATE: { + rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DEACTIVATE, NULL); + break; + } + case IMP_EVENT_ACTIVATE: { + rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVATE, NULL); + break; + } default: CERROR("Unknown import event %d\n", event); LBUG(); @@ -4316,6 +4401,32 @@ static int osc_import_event(struct obd_device *obd, RETURN(rc); } +/** + * Determine whether the lock can be canceled before replaying the lock + * during recovery, see bug16774 for detailed information. + * + * \retval zero the lock can't be canceled + * \retval other ok to cancel + */ +static int osc_cancel_for_recovery(struct ldlm_lock *lock) +{ + check_res_locked(lock->l_resource); + + /* + * Cancel all unused extent lock in granted mode LCK_PR or LCK_CR. + * + * XXX as a future improvement, we can also cancel unused write lock + * if it doesn't have dirty data and active mmaps. + */ + if (lock->l_resource->lr_type == LDLM_EXTENT && + (lock->l_granted_mode == LCK_PR || + lock->l_granted_mode == LCK_CR) && + (osc_dlm_lock_pageref(lock) == 0)) + RETURN(1); + + RETURN(0); +} + int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { int rc; @@ -4354,6 +4465,8 @@ int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg) CFS_INIT_LIST_HEAD(&cli->cl_grant_shrink_list); cfs_sema_init(&cli->cl_grant_sem, 1); + + ns_register_cancel(obd->obd_namespace, osc_cancel_for_recovery); } RETURN(rc); @@ -4377,6 +4490,16 @@ static int osc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) break; } case OBD_CLEANUP_EXPORTS: { + /* LU-464 + * for echo client, export may be on zombie list, wait for + * zombie thread to cull it, because cli.cl_import will be + * cleared in client_disconnect_export(): + * class_export_destroy() -> obd_cleanup() -> + * echo_device_free() -> echo_client_cleanup() -> + * obd_disconnect() -> osc_disconnect() -> + * client_disconnect_export() + */ + obd_zombie_barrier(); /* If we set up but never connected, the client import will not have been cleaned. */ if (obd->u.cli.cl_import) {