X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Ftarget%2Ftgt_handler.c;h=1a903dbca59191fbf9eb461aa26a3acbebc0999e;hp=e41c59c7f084f2f66dd9ee6d0ad81286eab0bef5;hb=4894683342d77964daeded9fbc608fc46aa479ee;hpb=23773b32bfe15fccc9c029d722dc1bd5b0144779 diff --git a/lustre/target/tgt_handler.c b/lustre/target/tgt_handler.c index e41c59c..1a903db 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 @@ -265,6 +265,9 @@ static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags) body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID, NODEMAP_CLIENT_TO_FS, body->oa.o_gid); + body->oa.o_projid = nodemap_map_id(nodemap, NODEMAP_PROJID, + NODEMAP_CLIENT_TO_FS, + body->oa.o_projid); nodemap_putref(nodemap); tsi->tsi_ost_body = body; @@ -320,11 +323,15 @@ static int tgt_request_preprocess(struct tgt_session_info *tsi, 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) @@ -415,6 +422,11 @@ 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 & HAS_REPLY) { @@ -442,7 +454,8 @@ 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); } @@ -517,6 +530,7 @@ 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); @@ -611,8 +625,14 @@ static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req) /* opcode was not found in slice */ if (unlikely(s->tos_hs == NULL)) { - CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt), - opc); + static bool printed; + + /* don't spew error messages for unhandled RPCs */ + if (!printed) { + CERROR("%s: no handler for opcode 0x%x from %s\n", + tgt_name(tgt), opc, libcfs_id2str(req->rq_peer)); + printed = true; + } RETURN(ERR_PTR(-ENOTSUPP)); } @@ -678,8 +698,15 @@ static int process_req_last_xid(struct ptlrpc_request *req) * 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)) + 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(exp)) { @@ -712,18 +739,12 @@ int tgt_request_handle(struct ptlrpc_request *req) if (cfs_fail_val == 0 && lustre_msg_get_opc(msg) != OBD_PING && lustre_msg_get_flags(msg) & MSG_REQ_REPLAY_DONE) { - struct l_wait_info lwi = { 0 }; - cfs_fail_val = 1; cfs_race_state = 0; - l_wait_event(cfs_race_waitq, (cfs_race_state == 1), - &lwi); + wait_event_idle(cfs_race_waitq, (cfs_race_state == 1)); } } - /* Refill the context, to make sure all thread keys are allocated */ - lu_env_refill(req->rq_svc_thread->t_env); - req_capsule_init(&req->rq_pill, req, RCL_SERVER); tsi->tsi_pill = &req->rq_pill; tsi->tsi_env = req->rq_svc_thread->t_env; @@ -785,7 +806,7 @@ 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; @@ -888,9 +909,9 @@ 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); @@ -915,13 +936,13 @@ 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 && @@ -955,7 +976,7 @@ int tgt_adapt_sptlrpc_conf(struct lu_target *tgt) int rc; if (unlikely(tgt == NULL)) { - CERROR("No target passed"); + CERROR("No target passed\n"); return -EINVAL; } @@ -1012,6 +1033,27 @@ int tgt_connect(struct tgt_session_info *tsi) 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); @@ -1075,12 +1117,22 @@ int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf) struct ptlrpc_bulk_desc *desc; 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); @@ -1108,8 +1160,7 @@ 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) @@ -1186,7 +1237,7 @@ 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++) { @@ -1213,8 +1264,7 @@ 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; } @@ -1284,8 +1334,8 @@ 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)) { @@ -1637,13 +1687,6 @@ 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. @@ -1681,30 +1724,41 @@ int tgt_extent_lock(const struct lu_env *env, struct ldlm_namespace *ns, } 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(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)) @@ -1714,31 +1768,27 @@ static int tgt_brw_lock(const struct lu_env *env, struct obd_export *exp, 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(env, 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, @@ -1767,13 +1817,12 @@ 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() */ @@ -1800,13 +1849,12 @@ 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() */ @@ -1843,15 +1891,15 @@ static void dump_all_bulk_pages(struct obdo *oa, int count, * file/fid, not during the resends/retries. */ snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name), "%s-checksum_dump-ost-"DFID":[%llu-%llu]-%x-%x", - (strncmp(libcfs_debug_file_path_arr, "NONE", 4) != 0 ? - libcfs_debug_file_path_arr : - LIBCFS_DEBUG_FILE_PATH_DEFAULT), + (strncmp(libcfs_debug_file_path, "NONE", 4) != 0 ? + libcfs_debug_file_path : LIBCFS_DEBUG_FILE_PATH_DEFAULT), oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0, oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0, oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0, local_nb[0].lnb_file_offset, local_nb[count-1].lnb_file_offset + local_nb[count-1].lnb_len - 1, client_cksum, server_cksum); + CWARN("dumping checksum data to %s\n", dbgcksum_file_name); filp = filp_open(dbgcksum_file_name, O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600); if (IS_ERR(filp)) { @@ -1878,8 +1926,6 @@ static void dump_all_bulk_pages(struct obdo *oa, int count, } len -= rc; buf += rc; - CDEBUG(D_INFO, "%s: wrote %d bytes\n", - dbgcksum_file_name, rc); } kunmap(local_nb[i].lnb_page); } @@ -1888,7 +1934,8 @@ static void dump_all_bulk_pages(struct obdo *oa, int count, if (rc) CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc); filp_close(filp, NULL); - return; + + libcfs_debug_dumplog(); } static int check_read_checksum(struct niobuf_local *local_nb, int npages, @@ -1956,9 +2003,9 @@ 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; } @@ -1966,11 +2013,10 @@ static int tgt_pages2shortio(struct niobuf_local *local, int npages, } 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) + struct niobuf_local *local_nb, int npages, + int opc, obd_dif_csum_fn *fn, + int sector_size, u32 *check_sum, + bool resend) { enum cksum_types t10_cksum_type = tgt->lut_dt_conf.ddp_t10_cksum_type; unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP); @@ -2001,7 +2047,11 @@ static int tgt_checksum_niobuf_t10pi(struct lu_target *tgt, buffer = kmap(__page); guard_start = (__u16 *)buffer; guard_number = PAGE_SIZE / sizeof(*guard_start); + if (unlikely(resend)) + CDEBUG(D_PAGE | D_HA, "GRD tags per page = %u\n", guard_number); for (i = 0; i < npages; i++) { + bool use_t10_grd; + /* corrupt the data before we compute the checksum, to * simulate a client->OST data error */ if (i == 0 && opc == OST_WRITE && @@ -2011,13 +2061,12 @@ static int tgt_checksum_niobuf_t10pi(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() */ @@ -2036,8 +2085,10 @@ static int tgt_checksum_niobuf_t10pi(struct lu_target *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_guard_disk) { + use_t10_grd = t10_cksum_type && opc == OST_READ && + local_nb[i].lnb_len == PAGE_SIZE && + local_nb[i].lnb_guard_disk; + if (use_t10_grd) { used = DIV_ROUND_UP(local_nb[i].lnb_len, sector_size); if (used > (guard_number - used_number)) { rc = -E2BIG; @@ -2045,14 +2096,52 @@ static int tgt_checksum_niobuf_t10pi(struct lu_target *tgt, } memcpy(guard_start + used_number, local_nb[i].lnb_guards, - used * sizeof(*local_nb[i].lnb_guards)); - } else { + used * sizeof(*guard_start)); + if (unlikely(resend)) + CDEBUG(D_PAGE | D_HA, + "lnb[%u]: used %u off %u+%u lnb checksum: %*phN\n", + i, used, + local_nb[i].lnb_page_offset, + local_nb[i].lnb_len, + (int)(used * sizeof(*guard_start)), + guard_start + used_number); + } + if (!use_t10_grd || unlikely(resend)) { + __u16 guard_tmp[MAX_GUARD_NUMBER]; + __u16 *guards = guard_start + used_number; + int used_tmp = -1, *usedp = &used; + + if (unlikely(use_t10_grd)) { + guards = guard_tmp; + usedp = &used_tmp; + } 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, + local_nb[i].lnb_len, guards, + guard_number - used_number, usedp, sector_size, fn); + if (unlikely(resend)) { + bool bad = use_t10_grd && + memcmp(guard_tmp, + local_nb[i].lnb_guards, + used_tmp * sizeof(*guard_tmp)); + + if (bad) + CERROR("lnb[%u]: used %u/%u off %u+%u tmp checksum: %*phN\n", + i, used, used_tmp, + local_nb[i].lnb_page_offset, + local_nb[i].lnb_len, + (int)(used_tmp * sizeof(*guard_start)), + guard_tmp); + CDEBUG_LIMIT(D_PAGE | D_HA | (bad ? D_ERROR : 0), + "lnb[%u]: used %u/%u off %u+%u gen checksum: %*phN\n", + i, used, used_tmp, + local_nb[i].lnb_page_offset, + local_nb[i].lnb_len, + (int)(used * sizeof(*guard_start)), + guard_start + used_number); + } if (rc) break; } @@ -2092,13 +2181,12 @@ static int tgt_checksum_niobuf_t10pi(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() */ @@ -2134,7 +2222,8 @@ out: 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) + int npages, int opc, u32 *check_sum, + bool resend) { obd_dif_csum_fn *fn = NULL; int sector_size = 0; @@ -2146,10 +2235,11 @@ static int tgt_checksum_niobuf_rw(struct lu_target *tgt, if (fn) rc = tgt_checksum_niobuf_t10pi(tgt, local_nb, npages, opc, fn, sector_size, - check_sum); + check_sum, resend); else rc = tgt_checksum_niobuf(tgt, local_nb, npages, opc, cksum_type, check_sum); + RETURN(rc); } @@ -2162,7 +2252,6 @@ 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; @@ -2182,8 +2271,22 @@ int tgt_brw_read(struct tgt_session_info *tsi) req->rq_bulk_read = 1; - if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK)) - RETURN(-EIO); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK)) { + /* optionally use cfs_fail_val - 1 to select a specific OST on + * this server to fail requests. + */ + char fail_ost_name[MAX_OBD_NAME]; + + if (cfs_fail_val > 0) { + snprintf(fail_ost_name, MAX_OBD_NAME, "OST%04X", + cfs_fail_val - 1); + + if (strstr(obd_name, fail_ost_name)) + RETURN(err_serious(-EIO)); + } else { + RETURN(err_serious(-EIO)); + } + } OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ? cfs_fail_val : (obd_timeout + 1) / 4); @@ -2192,10 +2295,9 @@ 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 @@ -2207,6 +2309,10 @@ int tgt_brw_read(struct tgt_session_info *tsi) body = tsi->tsi_ost_body; LASSERT(body != NULL); + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_NORPC) + RETURN(0); + ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ); LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */ @@ -2243,12 +2349,12 @@ 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) @@ -2284,14 +2390,13 @@ 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) { u32 flag = body->oa.o_valid & OBD_MD_FLFLAGS ? body->oa.o_flags : 0; enum cksum_types cksum_type = obd_cksum_type_unpack(flag); + bool resend = (body->oa.o_valid & OBD_MD_FLFLAGS) && + (body->oa.o_flags & OBD_FL_RECOV_RESEND); repbody->oa.o_flags = obd_cksum_type_pack(obd_name, cksum_type); @@ -2299,17 +2404,17 @@ int tgt_brw_read(struct tgt_session_info *tsi) rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type, local_nb, npages_read, OST_READ, - &repbody->oa.o_cksum); + &repbody->oa.o_cksum, resend); if (rc < 0) GOTO(out_commitrw, rc); - CDEBUG(D_PAGE, "checksum at read origin: %x\n", - repbody->oa.o_cksum); + CDEBUG(D_PAGE | (resend ? D_HA : 0), + "checksum at read origin: %x (%x)\n", + repbody->oa.o_cksum, cksum_type); /* if a resend it could be for a cksum error, so check Server * cksum with returned Client cksum (this should even cover * zero-cksum case) */ - if ((body->oa.o_valid & OBD_MD_FLFLAGS) && - (body->oa.o_flags & OBD_FL_RECOV_RESEND)) + if (resend) check_read_checksum(local_nb, npages_read, exp, &body->oa, &req->rq_peer, body->oa.o_cksum, @@ -2317,12 +2422,15 @@ int tgt_brw_read(struct tgt_session_info *tsi) } else { repbody->oa.o_valid = 0; } + if (body->oa.o_valid & OBD_MD_FLGRANT) + repbody->oa.o_valid |= OBD_MD_FLGRANT; /* We're finishing using body->oa as an input variable */ /* 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; @@ -2343,7 +2451,8 @@ int tgt_brw_read(struct tgt_session_info *tsi) } 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); } @@ -2353,7 +2462,7 @@ 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); @@ -2377,14 +2486,22 @@ 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; + /* Calculate checksum before request transfer, original + * it is done by target_bulk_io() */ + rc = sptlrpc_svc_wrap_bulk(req, desc); + if (OCD_HAS_FLAG(&exp->exp_connect_data, BULK_MBITS)) + req->rq_mbits = lustre_msg_get_mbits(req->rq_reqmsg); + else /* old version, bulk matchbits is rq_xid */ + req->rq_mbits = req->rq_xid; + + req->rq_status = rc; + target_committed_to_req(req); + target_send_reply(req, 0, 0); CDEBUG(D_INFO, "reorder BULK\n"); - init_waitqueue_head(&waitq); + OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2, + cfs_fail_val ? : 3); - lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL); - l_wait_event(waitq, 0, &lwi1); target_bulk_io(exp, desc); ptlrpc_free_bulk(desc); } @@ -2408,11 +2525,11 @@ 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; } @@ -2478,12 +2595,15 @@ int tgt_brw_write(struct tgt_session_info *tsi) struct lustre_handle lockh = {0}; __u32 *rcs; int objcount, niocount, npages; - int rc, i, j; + int rc = 0; + int i, j; enum cksum_types cksum_type = OBD_CKSUM_CRC32; 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; @@ -2504,9 +2624,25 @@ int tgt_brw_write(struct tgt_session_info *tsi) req->rq_bulk_write = 1; if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK)) - RETURN(err_serious(-EIO)); + rc = -EIO; if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2)) - RETURN(err_serious(-EFAULT)); + rc = -EFAULT; + if (rc < 0) { + /* optionally use cfs_fail_val - 1 to select a specific OST on + * this server to fail requests. + */ + char fail_ost_name[MAX_OBD_NAME]; + + if (cfs_fail_val > 0) { + snprintf(fail_ost_name, MAX_OBD_NAME, "OST%04X", + cfs_fail_val - 1); + + if (strstr(obd_name, fail_ost_name)) + RETURN(err_serious(rc)); + } else { + RETURN(err_serious(rc)); + } + } /* pause before transaction has been started */ CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ? @@ -2524,6 +2660,11 @@ int tgt_brw_write(struct tgt_session_info *tsi) body = tsi->tsi_ost_body; LASSERT(body != NULL); + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags & OBD_FL_NORPC) + RETURN(0); + + ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ); LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */ @@ -2542,7 +2683,7 @@ 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)); @@ -2593,7 +2734,8 @@ 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) { + 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; @@ -2611,8 +2753,7 @@ 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) @@ -2648,7 +2789,7 @@ skip_transfer: rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type, local_nb, npages, OST_WRITE, - &repbody->oa.o_cksum); + &repbody->oa.o_cksum, false); if (rc < 0) GOTO(out_commitrw, rc); @@ -2670,6 +2811,8 @@ 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, @@ -2715,7 +2858,7 @@ 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: @@ -2730,11 +2873,80 @@ out: 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 */