X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Ftarget%2Ftgt_handler.c;h=4558c37405192f8b061a1c9709ec7fd881958466;hb=f44413717eaf5cb938d9c9b2b62d312f064d282a;hp=745f92dcf4e2061af70cfbb41a64ba86f499e417;hpb=28b8f3d9296d632a37e36c68c6000dc0c5e96e5a;p=fs%2Flustre-release.git diff --git a/lustre/target/tgt_handler.c b/lustre/target/tgt_handler.c index 745f92d..4558c37 100644 --- a/lustre/target/tgt_handler.c +++ b/lustre/target/tgt_handler.c @@ -35,10 +35,10 @@ #define DEBUG_SUBSYSTEM S_CLASS #include -#ifdef HAVE_UIDGID_HEADER -# include -#endif +#include +#include +#include #include #include #include @@ -63,7 +63,7 @@ EXPORT_SYMBOL(tgt_name); * - create lu_object, corresponding to the fid in mdt_body, and save it in * @tsi; * - * - if HABEO_CORPUS flag is set for this request type check whether object + * - if HAS_BODY flag is set for this request type check whether object * actually exists on storage (lu_object_exists()). * */ @@ -100,7 +100,7 @@ static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags) &tsi->tsi_tgt->lut_bottom->dd_lu_dev, &body->mbo_fid1, NULL); if (!IS_ERR(obj)) { - if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) { + if ((flags & HAS_BODY) && !lu_object_exists(obj)) { lu_object_put(tsi->tsi_env, obj); rc = -ENOENT; } else { @@ -277,9 +277,8 @@ static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags) } if (!(body->oa.o_valid & OBD_MD_FLID)) { - if (flags & HABEO_CORPUS) { - CERROR("%s: OBD_MD_FLID flag is not set in ost_body " - "but OID/FID is mandatory with HABEO_CORPUS\n", + if (flags & HAS_BODY) { + CERROR("%s: OBD_MD_FLID flag is not set in ost_body but OID/FID is mandatory with HAS_BODY\n", tgt_name(tsi->tsi_tgt)); RETURN(-EPROTO); } else { @@ -317,15 +316,19 @@ static int tgt_request_preprocess(struct tgt_session_info *tsi, LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg)); LASSERT(current->journal_info == NULL); - LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO), + LASSERT(ergo(flags & (HAS_BODY | HAS_REPLY), h->th_fmt != NULL)); if (h->th_fmt != NULL) { req_capsule_set(pill, h->th_fmt); - if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) { + if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT) && + req_capsule_field_present(pill, &RMF_MDT_BODY, + RCL_CLIENT)) { rc = tgt_mdt_body_unpack(tsi, flags); if (rc < 0) RETURN(rc); } else if (req_capsule_has_field(pill, &RMF_OST_BODY, + RCL_CLIENT) && + req_capsule_field_present(pill, &RMF_OST_BODY, RCL_CLIENT)) { rc = tgt_ost_body_unpack(tsi, flags); if (rc < 0) @@ -333,10 +336,10 @@ static int tgt_request_preprocess(struct tgt_session_info *tsi, } } - if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY) + if (flags & IS_MUTABLE && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY) RETURN(-EROFS); - if (flags & HABEO_CLAVIS) { + if (flags & HAS_KEY) { struct ldlm_request *dlm_req; LASSERT(h->th_fmt != NULL); @@ -416,9 +419,14 @@ static int tgt_handle_request0(struct tgt_session_info *tsi, OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET))) RETURN(0); + /* drop OUT_UPDATE rpc */ + if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == OUT_UPDATE && + OBD_FAIL_CHECK(OBD_FAIL_OUT_UPDATE_DROP))) + RETURN(0); + rc = tgt_request_preprocess(tsi, h, req); /* pack reply if reply format is fixed */ - if (rc == 0 && h->th_flags & HABEO_REFERO) { + if (rc == 0 && h->th_flags & HAS_REPLY) { /* Pack reply */ if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD, RCL_SERVER)) @@ -443,7 +451,8 @@ static int tgt_handle_request0(struct tgt_session_info *tsi, req_capsule_set_size(tsi->tsi_pill, &RMF_SHORT_IO, RCL_SERVER, - (body->oa.o_flags & OBD_FL_SHORT_IO) ? + (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_SHORT_IO) ? remote_nb[0].rnb_len : 0); } @@ -461,8 +470,8 @@ static int tgt_handle_request0(struct tgt_session_info *tsi, rc = h->th_act(tsi); if (!is_serious(rc) && !req->rq_no_reply && req->rq_reply_state == NULL) { - DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not " - "pack reply and returned 0 error\n", + DEBUG_REQ(D_ERROR, req, + "%s: %s handler did not pack reply but returned no error", tgt_name(tsi->tsi_tgt), h->th_name); LBUG(); } @@ -518,6 +527,7 @@ static int tgt_filter_recovery_request(struct ptlrpc_request *req, case MDS_HSM_PROGRESS: case MDS_HSM_STATE_SET: case MDS_HSM_REQUEST: + case OST_FALLOCATE: *process = target_queue_recovery_request(req, obd); RETURN(0); @@ -553,12 +563,12 @@ static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id) /* sanity check: if the xid matches, the request must be marked as a * resent or replayed */ - if (req_can_reconstruct(req, NULL)) { + if (req_can_reconstruct(req, NULL) == 1) { if (!(lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY))) { - DEBUG_REQ(D_WARNING, req, "rq_xid %llu matches " - "saved xid, expected REPLAY or RESENT flag " - "(%x)", req->rq_xid, + DEBUG_REQ(D_WARNING, req, + "rq_xid=%llu matches saved XID, expected REPLAY or RESENT flag (%x)", + req->rq_xid, lustre_msg_get_flags(req->rq_reqmsg)); req->rq_status = -ENOTCONN; RETURN(-ENOTCONN); @@ -601,7 +611,7 @@ static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req) tgt = class_exp2tgt(req->rq_export); if (unlikely(tgt == NULL)) { - DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n", + DEBUG_REQ(D_ERROR, req, "%s: no target for connected export", class_exp2obd(req->rq_export)->obd_name); RETURN(ERR_PTR(-EINVAL)); } @@ -630,18 +640,20 @@ static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req) static int process_req_last_xid(struct ptlrpc_request *req) { __u64 last_xid; + int rc = 0; + struct obd_export *exp = req->rq_export; + struct tg_export_data *ted = &exp->exp_target_data; + bool need_lock = tgt_is_multimodrpcs_client(exp); ENTRY; + if (need_lock) + mutex_lock(&ted->ted_lcd_lock); /* check request's xid is consistent with export's last_xid */ last_xid = lustre_msg_get_last_xid(req->rq_reqmsg); - if (last_xid > req->rq_export->exp_last_xid) - req->rq_export->exp_last_xid = last_xid; - - if (req->rq_xid == 0 || - (req->rq_xid <= req->rq_export->exp_last_xid)) { - DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. " - "last_xid %llx\n", req->rq_xid, - req->rq_export->exp_last_xid); + if (last_xid > exp->exp_last_xid) + exp->exp_last_xid = last_xid; + + if (req->rq_xid == 0 || req->rq_xid <= exp->exp_last_xid) { /* Some request is allowed to be sent during replay, * such as OUT update requests, FLD requests, so it * is possible that replay requests has smaller XID @@ -658,19 +670,46 @@ static int process_req_last_xid(struct ptlrpc_request *req) * - The former RPC got chance to be processed; */ if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)) - RETURN(-EPROTO); + rc = -EPROTO; + + DEBUG_REQ(D_WARNING, req, + "unexpected xid=%llx != exp_last_xid=%llx, rc = %d", + req->rq_xid, exp->exp_last_xid, rc); + if (rc) + GOTO(out, rc); + } + + /* The "last_xid" is the minimum xid among unreplied requests, + * if the request is from the previous connection, its xid can + * still be larger than "exp_last_xid", then the above check of + * xid is not enough to determine whether the request is delayed. + * + * For example, if some replay request was delayed and caused + * timeout at client and the replay is restarted, the delayed + * replay request will have the larger xid than "exp_last_xid" + */ + if (req->rq_export->exp_conn_cnt > + lustre_msg_get_conn_cnt(req->rq_reqmsg)) { + CDEBUG(D_RPCTRACE, + "Dropping request %llu from an old epoch %u/%u\n", + req->rq_xid, + lustre_msg_get_conn_cnt(req->rq_reqmsg), + req->rq_export->exp_conn_cnt); + req->rq_no_reply = 1; + GOTO(out, rc = -ESTALE); } /* try to release in-memory reply data */ - if (tgt_is_multimodrpcs_client(req->rq_export)) { - tgt_handle_received_xid(req->rq_export, - lustre_msg_get_last_xid(req->rq_reqmsg)); - if (!(lustre_msg_get_flags(req->rq_reqmsg) & - (MSG_RESENT | MSG_REPLAY))) - tgt_handle_tag(req->rq_export, - lustre_msg_get_tag(req->rq_reqmsg)); + if (tgt_is_multimodrpcs_client(exp)) { + tgt_handle_received_xid(exp, last_xid); + rc = tgt_handle_tag(req); } - RETURN(0); + +out: + if (need_lock) + mutex_unlock(&ted->ted_lcd_lock); + + RETURN(rc); } int tgt_request_handle(struct ptlrpc_request *req) @@ -687,8 +726,15 @@ int tgt_request_handle(struct ptlrpc_request *req) bool is_connect = false; ENTRY; - /* Refill the context, to make sure all thread keys are allocated */ - lu_env_refill(req->rq_svc_thread->t_env); + if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_TGT_RECOVERY_REQ_RACE))) { + if (cfs_fail_val == 0 && + lustre_msg_get_opc(msg) != OBD_PING && + lustre_msg_get_flags(msg) & MSG_REQ_REPLAY_DONE) { + cfs_fail_val = 1; + cfs_race_state = 0; + wait_event_idle(cfs_race_waitq, (cfs_race_state == 1)); + } + } req_capsule_init(&req->rq_pill, req, RCL_SERVER); tsi->tsi_pill = &req->rq_pill; @@ -735,7 +781,7 @@ int tgt_request_handle(struct ptlrpc_request *req) tsi->tsi_jobid = NULL; if (tgt == NULL) { - DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n", + DEBUG_REQ(D_ERROR, req, "%s: No target for connected export", class_exp2obd(req->rq_export)->obd_name); req->rq_status = -EINVAL; rc = ptlrpc_error(req); @@ -751,7 +797,7 @@ int tgt_request_handle(struct ptlrpc_request *req) /* reset the exp_last_xid on each connection. */ req->rq_export->exp_last_xid = 0; } else if (obd->obd_recovery_data.trd_processing_task != - current_pid()) { + current->pid) { rc = process_req_last_xid(req); if (rc) { req->rq_status = rc; @@ -773,14 +819,16 @@ int tgt_request_handle(struct ptlrpc_request *req) LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n", h->th_opc, opc); - if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE)) + if ((cfs_fail_val == 0 || cfs_fail_val == opc) && + CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE)) GOTO(out, rc = 0); rc = lustre_msg_check_version(msg, h->th_version); if (unlikely(rc)) { - DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version" - " %08x, expecting %08x\n", tgt_name(tgt), - lustre_msg_get_version(msg), h->th_version); + DEBUG_REQ(D_ERROR, req, + "%s: drop malformed request version=%08x expect=%08x", + tgt_name(tgt), lustre_msg_get_version(msg), + h->th_version); req->rq_status = -EINVAL; rc = ptlrpc_error(req); GOTO(out, rc); @@ -852,9 +900,9 @@ EXPORT_SYMBOL(tgt_counter_incr); int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp) { - struct lu_target *tgt = class_exp2tgt(exp); - struct sptlrpc_flavor flvr; - int rc = 0; + struct lu_target *tgt = class_exp2tgt(exp); + struct sptlrpc_flavor flvr; + int rc = 0; LASSERT(tgt); LASSERT(tgt->lut_obd); @@ -879,13 +927,13 @@ int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp exp->exp_sp_peer = req->rq_sp_from; exp->exp_flvr = flvr; - /* when on mgs, if no restriction is set, or if client - * is loopback, allow any flavor */ + /* when on mgs, if no restriction is set, or if the client + * NID is on the local node, allow any flavor + */ if ((strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_MGS_NAME) == 0) && (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL || - LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid)) - == LOLND)) + LNetIsPeerLocal(exp->exp_connection->c_peer.nid))) exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY; if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY && @@ -965,9 +1013,40 @@ int tgt_connect(struct tgt_session_info *tsi) reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA); spin_lock(&tsi->tsi_exp->exp_lock); *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags; + if (reply->ocd_connect_flags & OBD_CONNECT_FLAGS2) + *exp_connect_flags2_ptr(tsi->tsi_exp) = + reply->ocd_connect_flags2; tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size; spin_unlock(&tsi->tsi_exp->exp_lock); + if (strcmp(tsi->tsi_exp->exp_obd->obd_type->typ_name, + LUSTRE_MDT_NAME) == 0) { + rc = req_check_sepol(tsi->tsi_pill); + if (rc) + GOTO(out, rc); + + if (reply->ocd_connect_flags & OBD_CONNECT_FLAGS2 && + reply->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT && + tsi->tsi_pill->rc_req->rq_export) { + bool forbid_encrypt = true; + struct lu_nodemap *nm = + nodemap_get_from_exp(tsi->tsi_pill->rc_req->rq_export); + + if (!nm) { + /* nodemap_get_from_exp returns NULL in case + * nodemap is not active, so we do not forbid + */ + forbid_encrypt = false; + } else if (!IS_ERR(nm)) { + forbid_encrypt = nm->nmf_forbid_encryption; + nodemap_putref(nm); + } + + if (forbid_encrypt) + GOTO(out, rc = -EACCES); + } + } + RETURN(0); out: obd_disconnect(class_export_get(tsi->tsi_exp)); @@ -981,6 +1060,8 @@ int tgt_disconnect(struct tgt_session_info *tsi) ENTRY; + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DISCONNECT_DELAY, cfs_fail_val); + rc = target_handle_disconnect(tgt_ses_req(tsi)); if (rc) RETURN(err_serious(rc)); @@ -998,7 +1079,16 @@ int tgt_obd_ping(struct tgt_session_info *tsi) ENTRY; - rc = target_handle_ping(tgt_ses_req(tsi)); + /* The target-specific part of OBD_PING request handling. + * It controls Filter Modification Data (FMD) expiration each time + * PING is received. + * + * Valid only for replayable targets, e.g. MDT and OFD + */ + if (tsi->tsi_exp->exp_obd->obd_replayable) + tgt_fmd_expire(tsi->tsi_exp); + + rc = req_capsule_server_pack(tsi->tsi_pill); if (rc) RETURN(err_serious(rc)); @@ -1013,19 +1103,27 @@ int tgt_obd_log_cancel(struct tgt_session_info *tsi) int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf) { - struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); struct ptlrpc_request *req = tgt_ses_req(tsi); struct obd_export *exp = req->rq_export; struct ptlrpc_bulk_desc *desc; - struct l_wait_info *lwi = &tti->tti_u.update.tti_wait_info; int i; int rc; + int pages = 0; ENTRY; - desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1, - PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC, - MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops); + for (i = 0; i < rdbuf->rb_nbufs; i++) { + unsigned int offset; + + offset = (unsigned long)rdbuf->rb_bufs[i].lb_buf & ~PAGE_MASK; + pages += DIV_ROUND_UP(rdbuf->rb_bufs[i].lb_len + offset, + PAGE_SIZE); + } + + desc = ptlrpc_prep_bulk_exp(req, pages, 1, + PTLRPC_BULK_PUT_SOURCE, + MDS_BULK_PORTAL, + &ptlrpc_bulk_kiov_nopin_ops); if (desc == NULL) RETURN(-ENOMEM); @@ -1034,7 +1132,7 @@ int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf) rdbuf->rb_bufs[i].lb_buf, rdbuf->rb_bufs[i].lb_len); - rc = target_bulk_io(exp, desc, lwi); + rc = target_bulk_io(exp, desc); ptlrpc_free_bulk(desc); RETURN(rc); } @@ -1042,11 +1140,9 @@ EXPORT_SYMBOL(tgt_send_buffer); int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob) { - struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env); struct ptlrpc_request *req = tgt_ses_req(tsi); struct obd_export *exp = req->rq_export; struct ptlrpc_bulk_desc *desc; - struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info; int tmpcount; int tmpsize; int i; @@ -1055,8 +1151,7 @@ int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob) ENTRY; desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, - PTLRPC_BULK_PUT_SOURCE | - PTLRPC_BULK_BUF_KIOV, + PTLRPC_BULK_PUT_SOURCE, MDS_BULK_PORTAL, &ptlrpc_bulk_kiov_pin_ops); if (desc == NULL) @@ -1075,7 +1170,7 @@ int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob) } LASSERT(desc->bd_nob == nob); - rc = target_bulk_io(exp, desc, lwi); + rc = target_bulk_io(exp, desc); ptlrpc_free_bulk(desc); RETURN(rc); } @@ -1133,7 +1228,7 @@ static int tgt_obd_idx_read(struct tgt_session_info *tsi) rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> PAGE_SHIFT; /* allocate pages to store the containers */ - OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0])); + OBD_ALLOC_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages); if (rdpg->rp_pages == NULL) GOTO(out, rc = -ENOMEM); for (i = 0; i < rdpg->rp_npages; i++) { @@ -1160,15 +1255,13 @@ out: for (i = 0; i < rdpg->rp_npages; i++) if (rdpg->rp_pages[i]) __free_page(rdpg->rp_pages[i]); - OBD_FREE(rdpg->rp_pages, - rdpg->rp_npages * sizeof(rdpg->rp_pages[0])); + OBD_FREE_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages); } return rc; } struct tgt_handler tgt_obd_handlers[] = { TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping), -TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel), TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read) }; EXPORT_SYMBOL(tgt_obd_handlers); @@ -1232,8 +1325,8 @@ static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, if (flag == LDLM_CB_CANCELING && (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) && - (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL || - (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL && + (tgt->lut_sync_lock_cancel == SYNC_LOCK_CANCEL_ALWAYS || + (tgt->lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING && ldlm_is_cbpending(lock))) && ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) || lock->l_resource->lr_type == LDLM_EXTENT)) { @@ -1242,7 +1335,7 @@ static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, rc = lu_env_init(&env, LCT_DT_THREAD); if (unlikely(rc != 0)) - RETURN(rc); + GOTO(err, rc); ost_fid_from_resid(&fid, &lock->l_resource->lr_name, tgt->lut_lsd.lsd_osd_index); @@ -1273,7 +1366,7 @@ err_put: err_env: lu_env_fini(&env); } - +err: rc = ldlm_server_blocking_ast(lock, desc, data, flag); RETURN(rc); } @@ -1344,10 +1437,10 @@ int tgt_cp_callback(struct tgt_session_info *tsi) /* generic LDLM target handler */ struct tgt_handler tgt_dlm_handlers[] = { -TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue), -TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert), -TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback), -TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback) +TGT_DLM_HDL(HAS_KEY, LDLM_ENQUEUE, tgt_enqueue), +TGT_DLM_HDL(HAS_KEY, LDLM_CONVERT, tgt_convert), +TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback), +TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback) }; EXPORT_SYMBOL(tgt_dlm_handlers); @@ -1366,30 +1459,6 @@ int tgt_llog_open(struct tgt_session_info *tsi) } EXPORT_SYMBOL(tgt_llog_open); -int tgt_llog_close(struct tgt_session_info *tsi) -{ - int rc; - - ENTRY; - - rc = llog_origin_handle_close(tgt_ses_req(tsi)); - - RETURN(rc); -} -EXPORT_SYMBOL(tgt_llog_close); - - -int tgt_llog_destroy(struct tgt_session_info *tsi) -{ - int rc; - - ENTRY; - - rc = llog_origin_handle_destroy(tgt_ses_req(tsi)); - - RETURN(rc); -} - int tgt_llog_read_header(struct tgt_session_info *tsi) { int rc; @@ -1432,8 +1501,6 @@ TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open), TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block), TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header), TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block), -TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy), -TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close), }; EXPORT_SYMBOL(tgt_llog_handlers); @@ -1535,8 +1602,8 @@ static int tgt_handle_lfsck_query(struct tgt_session_info *tsi) } struct tgt_handler tgt_lfsck_handlers[] = { -TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify), -TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query), +TGT_LFSCK_HDL(HAS_REPLY, LFSCK_NOTIFY, tgt_handle_lfsck_notify), +TGT_LFSCK_HDL(HAS_REPLY, LFSCK_QUERY, tgt_handle_lfsck_query), }; EXPORT_SYMBOL(tgt_lfsck_handlers); @@ -1602,7 +1669,7 @@ int tgt_mdt_data_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, LASSERT(ns != NULL); LASSERT(!lustre_handle_is_used(lh)); - rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, &policy, mode, + rc = ldlm_cli_enqueue_local(NULL, ns, res_id, LDLM_IBITS, &policy, mode, flags, ldlm_blocking_ast, ldlm_completion_ast, ldlm_glimpse_ast, NULL, 0, LVB_T_NONE, NULL, lh); @@ -1611,20 +1678,13 @@ int tgt_mdt_data_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, } EXPORT_SYMBOL(tgt_mdt_data_lock); -void tgt_mdt_data_unlock(struct lustre_handle *lh, enum ldlm_mode mode) -{ - LASSERT(lustre_handle_is_used(lh)); - ldlm_lock_decref(lh, mode); -} -EXPORT_SYMBOL(tgt_mdt_data_unlock); - /** * Helper function for getting server side [start, start+count] DLM lock * if asked by client. */ -int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, - __u64 start, __u64 end, struct lustre_handle *lh, - int mode, __u64 *flags) +int tgt_extent_lock(const struct lu_env *env, struct ldlm_namespace *ns, + struct ldlm_res_id *res_id, __u64 start, __u64 end, + struct lustre_handle *lh, int mode, __u64 *flags) { union ldlm_policy_data policy; int rc; @@ -1647,37 +1707,49 @@ int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id, else policy.l_extent.end = end | ~PAGE_MASK; - rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode, - flags, ldlm_blocking_ast, + rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_EXTENT, &policy, + mode, flags, ldlm_blocking_ast, ldlm_completion_ast, ldlm_glimpse_ast, NULL, 0, LVB_T_NONE, NULL, lh); RETURN(rc == ELDLM_OK ? 0 : -EIO); } EXPORT_SYMBOL(tgt_extent_lock); -void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode) +static int tgt_data_lock(const struct lu_env *env, struct obd_export *exp, + struct ldlm_res_id *res_id, __u64 start, __u64 end, + struct lustre_handle *lh, enum ldlm_mode mode) +{ + struct ldlm_namespace *ns = exp->exp_obd->obd_namespace; + __u64 flags = 0; + + /* MDT IO for data-on-mdt */ + if (exp->exp_connect_data.ocd_connect_flags & OBD_CONNECT_IBITS) + return tgt_mdt_data_lock(ns, res_id, lh, mode, &flags); + + return tgt_extent_lock(env, ns, res_id, start, end, lh, mode, &flags); +} + +void tgt_data_unlock(struct lustre_handle *lh, enum ldlm_mode mode) { LASSERT(lustre_handle_is_used(lh)); ldlm_lock_decref(lh, mode); } -EXPORT_SYMBOL(tgt_extent_unlock); +EXPORT_SYMBOL(tgt_data_unlock); -static int tgt_brw_lock(struct obd_export *exp, struct ldlm_res_id *res_id, - struct obd_ioobj *obj, struct niobuf_remote *nb, - struct lustre_handle *lh, enum ldlm_mode mode) +static int tgt_brw_lock(const struct lu_env *env, struct obd_export *exp, + struct ldlm_res_id *res_id, struct obd_ioobj *obj, + struct niobuf_remote *nb, struct lustre_handle *lh, + enum ldlm_mode mode) { - struct ldlm_namespace *ns = exp->exp_obd->obd_namespace; - __u64 flags = 0; - int nrbufs = obj->ioo_bufcnt; - int i; - int rc; + int nrbufs = obj->ioo_bufcnt; + int i; ENTRY; LASSERT(mode == LCK_PR || mode == LCK_PW); LASSERT(!lustre_handle_is_used(lh)); - if (ns->ns_obd->obd_recovering) + if (exp->exp_obd->obd_recovering) RETURN(0); if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK)) @@ -1687,46 +1759,42 @@ static int tgt_brw_lock(struct obd_export *exp, struct ldlm_res_id *res_id, if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK)) RETURN(-EFAULT); - /* MDT IO for data-on-mdt */ - if (exp->exp_connect_data.ocd_connect_flags & OBD_CONNECT_IBITS) - rc = tgt_mdt_data_lock(ns, res_id, lh, mode, &flags); - else - rc = tgt_extent_lock(ns, res_id, nb[0].rnb_offset, - nb[nrbufs - 1].rnb_offset + - nb[nrbufs - 1].rnb_len - 1, - lh, mode, &flags); - RETURN(rc); + return tgt_data_lock(env, exp, res_id, nb[0].rnb_offset, + nb[nrbufs - 1].rnb_offset + + nb[nrbufs - 1].rnb_len - 1, lh, mode); } -static void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob, +static void tgt_brw_unlock(struct obd_export *exp, struct obd_ioobj *obj, + struct niobuf_remote *niob, struct lustre_handle *lh, enum ldlm_mode mode) { ENTRY; LASSERT(mode == LCK_PR || mode == LCK_PW); - LASSERT((obj->ioo_bufcnt > 0 && - (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) == + LASSERT((!exp->exp_obd->obd_recovering && obj->ioo_bufcnt && + niob[0].rnb_flags & OBD_BRW_SRVLOCK) == lustre_handle_is_used(lh)); if (lustre_handle_is_used(lh)) - tgt_extent_unlock(lh, mode); + tgt_data_unlock(lh, mode); EXIT; } + static int tgt_checksum_niobuf(struct lu_target *tgt, struct niobuf_local *local_nb, int npages, int opc, enum cksum_types cksum_type, __u32 *cksum) { - struct cfs_crypto_hash_desc *hdesc; + struct ahash_request *req; unsigned int bufsize; int i, err; unsigned char cfs_alg = cksum_obd2cfs(cksum_type); - 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("%s: unable to initialize checksum hash %s\n", tgt_name(tgt), cfs_crypto_hash_name(cfs_alg)); - return PTR_ERR(hdesc); + return PTR_ERR(req); } CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg)); @@ -1740,19 +1808,18 @@ static int tgt_checksum_niobuf(struct lu_target *tgt, struct page *np = tgt_page_to_corrupt; if (np) { - char *ptr = ll_kmap_atomic(local_nb[i].lnb_page, - KM_USER0); + char *ptr = kmap_atomic(local_nb[i].lnb_page); char *ptr2 = page_address(np); memcpy(ptr2 + off, ptr + off, len); memcpy(ptr2 + off, "bad3", min(4, len)); - ll_kunmap_atomic(ptr, KM_USER0); + kunmap_atomic(ptr); /* LU-8376 to preserve original index for * display in dump_all_bulk_pages() */ np->index = i; - cfs_crypto_hash_update_page(hdesc, np, off, + cfs_crypto_hash_update_page(req, np, off, len); continue; } else { @@ -1760,7 +1827,7 @@ static int tgt_checksum_niobuf(struct lu_target *tgt, tgt_name(tgt)); } } - cfs_crypto_hash_update_page(hdesc, local_nb[i].lnb_page, + cfs_crypto_hash_update_page(req, local_nb[i].lnb_page, local_nb[i].lnb_page_offset & ~PAGE_MASK, local_nb[i].lnb_len); @@ -1773,19 +1840,18 @@ static int tgt_checksum_niobuf(struct lu_target *tgt, struct page *np = tgt_page_to_corrupt; if (np) { - char *ptr = ll_kmap_atomic(local_nb[i].lnb_page, - KM_USER0); + char *ptr = kmap_atomic(local_nb[i].lnb_page); char *ptr2 = page_address(np); memcpy(ptr2 + off, ptr + off, len); memcpy(ptr2 + off, "bad4", min(4, len)); - ll_kunmap_atomic(ptr, KM_USER0); + kunmap_atomic(ptr); /* LU-8376 to preserve original index for * display in dump_all_bulk_pages() */ np->index = i; - cfs_crypto_hash_update_page(hdesc, np, off, + cfs_crypto_hash_update_page(req, np, off, len); continue; } else { @@ -1796,7 +1862,7 @@ static int tgt_checksum_niobuf(struct lu_target *tgt, } bufsize = sizeof(*cksum); - err = cfs_crypto_hash_final(hdesc, (unsigned char *)cksum, &bufsize); + err = cfs_crypto_hash_final(req, (unsigned char *)cksum, &bufsize); return 0; } @@ -1811,7 +1877,6 @@ static void dump_all_bulk_pages(struct obdo *oa, int count, int rc, i; unsigned int len; char *buf; - mm_segment_t oldfs; /* will only keep dump of pages on first error for the same range in * file/fid, not during the resends/retries. */ @@ -1840,14 +1905,11 @@ static void dump_all_bulk_pages(struct obdo *oa, int count, return; } - oldfs = get_fs(); - set_fs(KERNEL_DS); for (i = 0; i < count; i++) { len = local_nb[i].lnb_len; buf = kmap(local_nb[i].lnb_page); while (len != 0) { - rc = vfs_write(filp, (__force const char __user *)buf, - len, &filp->f_pos); + rc = cfs_kernel_write(filp, buf, len, &filp->f_pos); if (rc < 0) { CERROR("%s: wanted to write %u but got %d " "error\n", dbgcksum_file_name, len, rc); @@ -1860,13 +1922,11 @@ static void dump_all_bulk_pages(struct obdo *oa, int count, } kunmap(local_nb[i].lnb_page); } - set_fs(oldfs); - 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 check_read_checksum(struct niobuf_local *local_nb, int npages, @@ -1891,8 +1951,8 @@ static int check_read_checksum(struct niobuf_local *local_nb, int npages, dump_all_bulk_pages(oa, npages, local_nb, server_cksum, client_cksum); - cksum_type = cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ? - oa->o_flags : 0); + cksum_type = obd_cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ? + oa->o_flags : 0); if (cksum_type != server_cksum_type) msg = "the server may have not used the checksum type specified" @@ -1934,15 +1994,202 @@ static int tgt_pages2shortio(struct niobuf_local *local, int npages, if (len > size) return -EINVAL; - ptr = ll_kmap_atomic(local[i].lnb_page, KM_USER0); - memcpy(buf + off, ptr, len); - ll_kunmap_atomic(ptr, KM_USER0); + ptr = kmap_atomic(local[i].lnb_page); + memcpy(buf, ptr + off, len); + kunmap_atomic(ptr); buf += len; size -= len; } return copied - size; } +static int tgt_checksum_niobuf_t10pi(struct lu_target *tgt, + struct niobuf_local *local_nb, + int npages, int opc, + obd_dif_csum_fn *fn, + int sector_size, + u32 *check_sum) +{ + enum cksum_types t10_cksum_type = tgt->lut_dt_conf.ddp_t10_cksum_type; + unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP); + const char *obd_name = tgt->lut_obd->obd_name; + struct ahash_request *req; + unsigned int bufsize; + unsigned char *buffer; + struct page *__page; + __u16 *guard_start; + int guard_number; + int used_number = 0; + __u32 cksum; + int rc = 0; + int used; + int i; + + __page = alloc_page(GFP_KERNEL); + if (__page == NULL) + return -ENOMEM; + + req = cfs_crypto_hash_init(cfs_alg, NULL, 0); + if (IS_ERR(req)) { + CERROR("%s: unable to initialize checksum hash %s\n", + tgt_name(tgt), cfs_crypto_hash_name(cfs_alg)); + return PTR_ERR(req); + } + + buffer = kmap(__page); + guard_start = (__u16 *)buffer; + guard_number = PAGE_SIZE / sizeof(*guard_start); + for (i = 0; i < npages; i++) { + /* corrupt the data before we compute the checksum, to + * simulate a client->OST data error */ + if (i == 0 && opc == OST_WRITE && + OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) { + int off = local_nb[i].lnb_page_offset & ~PAGE_MASK; + int len = local_nb[i].lnb_len; + struct page *np = tgt_page_to_corrupt; + + if (np) { + char *ptr = kmap_atomic(local_nb[i].lnb_page); + char *ptr2 = page_address(np); + + memcpy(ptr2 + off, ptr + off, len); + memcpy(ptr2 + off, "bad3", min(4, len)); + kunmap_atomic(ptr); + + /* LU-8376 to preserve original index for + * display in dump_all_bulk_pages() */ + np->index = i; + + cfs_crypto_hash_update_page(req, np, off, + len); + continue; + } else { + CERROR("%s: can't alloc page for corruption\n", + tgt_name(tgt)); + } + } + + /* + * The left guard number should be able to hold checksums of a + * whole page + */ + if (t10_cksum_type && opc == OST_READ && + local_nb[i].lnb_len == PAGE_SIZE && + local_nb[i].lnb_guard_disk) { + used = DIV_ROUND_UP(local_nb[i].lnb_len, sector_size); + if (used > (guard_number - used_number)) { + rc = -E2BIG; + break; + } + memcpy(guard_start + used_number, + local_nb[i].lnb_guards, + used * sizeof(*local_nb[i].lnb_guards)); + } else { + rc = obd_page_dif_generate_buffer(obd_name, + local_nb[i].lnb_page, + local_nb[i].lnb_page_offset & ~PAGE_MASK, + local_nb[i].lnb_len, guard_start + used_number, + guard_number - used_number, &used, sector_size, + fn); + if (rc) + break; + } + + LASSERT(used <= MAX_GUARD_NUMBER); + /* + * If disk support T10PI checksum, copy guards to local_nb. + * If the write is partial page, do not use the guards for bio + * submission since the data might not be full-sector. The bio + * guards will be generated later based on the full sectors. If + * the sector size is 512B rather than 4 KB, or the page size + * is larger than 4KB, this might drop some useful guards for + * partial page write, but it will only add minimal extra time + * of checksum calculation. + */ + if (t10_cksum_type && opc == OST_WRITE && + local_nb[i].lnb_len == PAGE_SIZE) { + local_nb[i].lnb_guard_rpc = 1; + memcpy(local_nb[i].lnb_guards, + guard_start + used_number, + used * sizeof(*local_nb[i].lnb_guards)); + } + + used_number += used; + if (used_number == guard_number) { + cfs_crypto_hash_update_page(req, __page, 0, + used_number * sizeof(*guard_start)); + used_number = 0; + } + + /* corrupt the data after we compute the checksum, to + * simulate an OST->client data error */ + if (unlikely(i == 0 && opc == OST_READ && + OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND))) { + int off = local_nb[i].lnb_page_offset & ~PAGE_MASK; + int len = local_nb[i].lnb_len; + struct page *np = tgt_page_to_corrupt; + + if (np) { + char *ptr = kmap_atomic(local_nb[i].lnb_page); + char *ptr2 = page_address(np); + + memcpy(ptr2 + off, ptr + off, len); + memcpy(ptr2 + off, "bad4", min(4, len)); + kunmap_atomic(ptr); + + /* LU-8376 to preserve original index for + * display in dump_all_bulk_pages() */ + np->index = i; + + cfs_crypto_hash_update_page(req, np, off, + len); + continue; + } else { + CERROR("%s: can't alloc page for corruption\n", + tgt_name(tgt)); + } + } + } + kunmap(__page); + if (rc) + GOTO(out, rc); + + if (used_number != 0) + cfs_crypto_hash_update_page(req, __page, 0, + used_number * sizeof(*guard_start)); + + bufsize = sizeof(cksum); + rc = cfs_crypto_hash_final(req, (unsigned char *)&cksum, &bufsize); + + if (rc == 0) + *check_sum = cksum; +out: + __free_page(__page); + return rc; +} + +static int tgt_checksum_niobuf_rw(struct lu_target *tgt, + enum cksum_types cksum_type, + struct niobuf_local *local_nb, + int npages, int opc, u32 *check_sum) +{ + obd_dif_csum_fn *fn = NULL; + int sector_size = 0; + int rc; + + ENTRY; + obd_t10_cksum2dif(cksum_type, &fn, §or_size); + + if (fn) + rc = tgt_checksum_niobuf_t10pi(tgt, local_nb, npages, + opc, fn, sector_size, + check_sum); + else + rc = tgt_checksum_niobuf(tgt, local_nb, npages, opc, + cksum_type, check_sum); + RETURN(rc); +} + int tgt_brw_read(struct tgt_session_info *tsi) { struct ptlrpc_request *req = tgt_ses_req(tsi); @@ -1952,11 +2199,11 @@ int tgt_brw_read(struct tgt_session_info *tsi) struct niobuf_local *local_nb; struct obd_ioobj *ioo; struct ost_body *body, *repbody; - struct l_wait_info lwi; struct lustre_handle lockh = { 0 }; int npages, nob = 0, rc, i, no_reply = 0, npages_read; struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data; + const char *obd_name = exp->exp_obd->obd_name; ENTRY; @@ -1981,10 +2228,9 @@ int tgt_brw_read(struct tgt_session_info *tsi) * finish */ if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) { /* We do not care how long it takes */ - lwi = LWI_INTR(NULL, NULL); - rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq, - !atomic_read(&exp->exp_obd->obd_evict_inprogress), - &lwi); + wait_event_idle( + exp->exp_obd->obd_evict_inprogress_waitq, + !atomic_read(&exp->exp_obd->obd_evict_inprogress)); } /* There must be big cache in current thread to process this request @@ -2004,8 +2250,8 @@ int tgt_brw_read(struct tgt_session_info *tsi) local_nb = tbc->local; - rc = tgt_brw_lock(exp, &tsi->tsi_resid, ioo, remote_nb, &lockh, - LCK_PR); + rc = tgt_brw_lock(tsi->tsi_env, exp, &tsi->tsi_resid, ioo, remote_nb, + &lockh, LCK_PR); if (rc != 0) RETURN(rc); @@ -2032,12 +2278,12 @@ int tgt_brw_read(struct tgt_session_info *tsi) if (rc != 0) GOTO(out_lock, rc); - if (body->oa.o_flags & OBD_FL_SHORT_IO) { + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_SHORT_IO) { desc = NULL; } else { desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo), - PTLRPC_BULK_PUT_SOURCE | - PTLRPC_BULK_BUF_KIOV, + PTLRPC_BULK_PUT_SOURCE, OST_BULK_PORTAL, &ptlrpc_bulk_kiov_nopin_ops); if (desc == NULL) @@ -2073,23 +2319,21 @@ int tgt_brw_read(struct tgt_session_info *tsi) break; } } - if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) && - nob != cfs_fail_val) - rc = -E2BIG; if (body->oa.o_valid & OBD_MD_FLCKSUM) { - enum cksum_types cksum_type = - cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ? - body->oa.o_flags : 0); + u32 flag = body->oa.o_valid & OBD_MD_FLFLAGS ? + body->oa.o_flags : 0; + enum cksum_types cksum_type = obd_cksum_type_unpack(flag); - repbody->oa.o_flags = cksum_type_pack(cksum_type); + repbody->oa.o_flags = obd_cksum_type_pack(obd_name, + cksum_type); repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS; - rc = tgt_checksum_niobuf(tsi->tsi_tgt, local_nb, - npages_read, OST_READ, cksum_type, - &repbody->oa.o_cksum); + + rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type, + local_nb, npages_read, OST_READ, + &repbody->oa.o_cksum); if (rc < 0) GOTO(out_commitrw, rc); - CDEBUG(D_PAGE, "checksum at read origin: %x\n", repbody->oa.o_cksum); @@ -2110,7 +2354,8 @@ int tgt_brw_read(struct tgt_session_info *tsi) /* Check if client was evicted while we were doing i/o before touching * network */ if (rc == 0) { - if (body->oa.o_flags & OBD_FL_SHORT_IO) { + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_SHORT_IO) { unsigned char *short_io_buf; int short_io_size; @@ -2127,11 +2372,12 @@ int tgt_brw_read(struct tgt_session_info *tsi) RCL_SERVER); rc = rc > 0 ? 0 : rc; } else if (!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) { - rc = target_bulk_io(exp, desc, &lwi); + rc = target_bulk_io(exp, desc); } no_reply = rc != 0; } else { - if (body->oa.o_flags & OBD_FL_SHORT_IO) + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_SHORT_IO) req_capsule_shrink(&req->rq_pill, &RMF_SHORT_IO, 0, RCL_SERVER); } @@ -2141,7 +2387,7 @@ out_commitrw: rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo, remote_nb, npages, local_nb, rc); out_lock: - tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR); + tgt_brw_unlock(exp, ioo, remote_nb, &lockh, LCK_PR); if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) ptlrpc_free_bulk(desc); @@ -2156,7 +2402,7 @@ out_lock: ptlrpc_req_drop_rs(req); LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), " "client will retry: rc %d\n", - exp->exp_obd->obd_name, + obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp), rc); } @@ -2165,15 +2411,10 @@ out_lock: * to reorder. */ if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) && desc) { - wait_queue_head_t waitq; - struct l_wait_info lwi1; - CDEBUG(D_INFO, "reorder BULK\n"); - init_waitqueue_head(&waitq); - lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL); - l_wait_event(waitq, 0, &lwi1); - target_bulk_io(exp, desc, &lwi); + ssleep(3); + target_bulk_io(exp, desc); ptlrpc_free_bulk(desc); } @@ -2182,7 +2423,7 @@ out_lock: EXPORT_SYMBOL(tgt_brw_read); static int tgt_shortio2pages(struct niobuf_local *local, int npages, - unsigned char *buf, int size) + unsigned char *buf, unsigned int size) { int i, off, len; char *ptr; @@ -2196,11 +2437,11 @@ static int tgt_shortio2pages(struct niobuf_local *local, int npages, CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n", i, off, len, size); - ptr = ll_kmap_atomic(local[i].lnb_page, KM_USER0); + ptr = kmap_atomic(local[i].lnb_page); if (ptr == NULL) return -EINVAL; memcpy(ptr + off, buf, len < size ? len : size); - ll_kunmap_atomic(ptr, KM_USER0); + kunmap_atomic(ptr); buf += len; size -= len; } @@ -2263,7 +2504,6 @@ int tgt_brw_write(struct tgt_session_info *tsi) struct niobuf_local *local_nb; struct obd_ioobj *ioo; struct ost_body *body, *repbody; - struct l_wait_info lwi; struct lustre_handle lockh = {0}; __u32 *rcs; int objcount, niocount, npages; @@ -2272,6 +2512,9 @@ int tgt_brw_write(struct tgt_session_info *tsi) bool no_reply = false, mmap; struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data; bool wait_sync = false; + const char *obd_name = exp->exp_obd->obd_name; + /* '1' for consistency with code that checks !mpflag to restore */ + unsigned int mpflags = 1; ENTRY; @@ -2300,6 +2543,9 @@ int tgt_brw_write(struct tgt_session_info *tsi) CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ? cfs_fail_val : (obd_timeout + 1) / 4); + /* Delay write commit to show stale size information */ + CFS_FAIL_TIMEOUT(OBD_FAIL_OSC_NO_SIZE_DATA, cfs_fail_val); + /* There must be big cache in current thread to process this request * if it is NULL then something went wrong and it wasn't allocated, * report -ENOMEM in that case */ @@ -2327,7 +2573,7 @@ int tgt_brw_write(struct tgt_session_info *tsi) if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) && ptlrpc_connection_is_local(exp->exp_connection)) - memory_pressure_set(); + mpflags = memalloc_noreclaim_save(); req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER, niocount * sizeof(*rcs)); @@ -2340,8 +2586,8 @@ int tgt_brw_write(struct tgt_session_info *tsi) local_nb = tbc->local; - rc = tgt_brw_lock(exp, &tsi->tsi_resid, ioo, remote_nb, &lockh, - LCK_PW); + rc = tgt_brw_lock(tsi->tsi_env, exp, &tsi->tsi_resid, ioo, remote_nb, + &lockh, LCK_PW); if (rc != 0) GOTO(out, rc); @@ -2378,8 +2624,9 @@ int tgt_brw_write(struct tgt_session_info *tsi) objcount, ioo, remote_nb, &npages, local_nb); if (rc < 0) GOTO(out_lock, rc); - if (body->oa.o_flags & OBD_FL_SHORT_IO) { - int short_io_size; + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_SHORT_IO) { + unsigned int short_io_size; unsigned char *short_io_buf; short_io_size = req_capsule_get_size(&req->rq_pill, @@ -2396,8 +2643,7 @@ int tgt_brw_write(struct tgt_session_info *tsi) desc = NULL; } else { desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo), - PTLRPC_BULK_GET_SINK | - PTLRPC_BULK_BUF_KIOV, + PTLRPC_BULK_GET_SINK, OST_BULK_PORTAL, &ptlrpc_bulk_kiov_nopin_ops); if (desc == NULL) @@ -2414,7 +2660,7 @@ int tgt_brw_write(struct tgt_session_info *tsi) if (rc != 0) GOTO(skip_transfer, rc); - rc = target_bulk_io(exp, desc, &lwi); + rc = target_bulk_io(exp, desc); } no_reply = rc != 0; @@ -2424,14 +2670,16 @@ skip_transfer: static int cksum_counter; if (body->oa.o_valid & OBD_MD_FLFLAGS) - cksum_type = cksum_type_unpack(body->oa.o_flags); + cksum_type = obd_cksum_type_unpack(body->oa.o_flags); repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS; repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL; - repbody->oa.o_flags |= cksum_type_pack(cksum_type); - rc = tgt_checksum_niobuf(tsi->tsi_tgt, local_nb, - npages, OST_WRITE, cksum_type, - &repbody->oa.o_cksum); + repbody->oa.o_flags |= obd_cksum_type_pack(obd_name, + cksum_type); + + rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type, + local_nb, npages, OST_WRITE, + &repbody->oa.o_cksum); if (rc < 0) GOTO(out_commitrw, rc); @@ -2453,6 +2701,8 @@ skip_transfer: } } + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK2, cfs_fail_val); + out_commitrw: /* Must commit after prep above in all cases */ rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa, @@ -2498,7 +2748,7 @@ out_commitrw: ptlrpc_lprocfs_brw(req, nob); } out_lock: - tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW); + tgt_brw_unlock(exp, ioo, remote_nb, &lockh, LCK_PW); if (desc) ptlrpc_free_bulk(desc); out: @@ -2509,24 +2759,93 @@ out: if (!exp->exp_obd->obd_no_transno) LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s)," " client will retry: rc = %d\n", - exp->exp_obd->obd_name, + obd_name, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp), rc); } - memory_pressure_clr(); + + if (mpflags) + memalloc_noreclaim_restore(mpflags); + RETURN(rc); } EXPORT_SYMBOL(tgt_brw_write); +/** + * Common request handler for OST_SEEK RPC. + * + * Unified request handling for OST_SEEK RPC. + * It takes object by its FID, does needed lseek and packs result + * into reply. Only SEEK_HOLE and SEEK_DATA are supported. + * + * \param[in] tsi target session environment for this request + * + * \retval 0 if successful + * \retval negative value on error + */ +int tgt_lseek(struct tgt_session_info *tsi) +{ + struct lustre_handle lh = { 0 }; + struct dt_object *dob; + struct ost_body *repbody; + loff_t offset = tsi->tsi_ost_body->oa.o_size; + int whence = tsi->tsi_ost_body->oa.o_mode; + bool srvlock; + int rc = 0; + + ENTRY; + + if (whence != SEEK_HOLE && whence != SEEK_DATA) + RETURN(-EPROTO); + + /* Negative offset is prohibited on wire and must be handled on client + * prior sending RPC. + */ + if (offset < 0) + RETURN(-EPROTO); + + repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY); + if (repbody == NULL) + RETURN(-ENOMEM); + repbody->oa = tsi->tsi_ost_body->oa; + + srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS && + tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK; + if (srvlock) { + rc = tgt_data_lock(tsi->tsi_env, tsi->tsi_exp, &tsi->tsi_resid, + offset, OBD_OBJECT_EOF, &lh, LCK_PR); + if (rc) + RETURN(rc); + } + + dob = dt_locate(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, &tsi->tsi_fid); + if (IS_ERR(dob)) + GOTO(out, rc = PTR_ERR(dob)); + + if (!dt_object_exists(dob)) + GOTO(obj_put, rc = -ENOENT); + + repbody->oa.o_size = dt_lseek(tsi->tsi_env, dob, offset, whence); + rc = 0; +obj_put: + dt_object_put(tsi->tsi_env, dob); +out: + if (srvlock) + tgt_data_unlock(&lh, LCK_PR); + + RETURN(rc); +} +EXPORT_SYMBOL(tgt_lseek); + /* Check if request can be reconstructed from saved reply data * A copy of the reply data is returned in @trd if the pointer is not NULL */ -bool req_can_reconstruct(struct ptlrpc_request *req, +int req_can_reconstruct(struct ptlrpc_request *req, struct tg_reply_data *trd) { struct tg_export_data *ted = &req->rq_export->exp_target_data; struct lsd_client_data *lcd = ted->ted_lcd; - bool found; + int found; if (tgt_is_multimodrpcs_client(req->rq_export)) return tgt_lookup_reply(req, trd);