X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fost%2Fost_handler.c;h=ee41fe46a3115d8fb6781c6f8e9d1122d5c79c05;hp=6712d4c1c56dc49e628fbfb4d73a516a2cfb8804;hb=864516e29129cb209435b4cc3ba513a9f453d383;hpb=88cbf2754651d051288a2b68ff0cf27aea6b502a diff --git a/lustre/ost/ost_handler.c b/lustre/ost/ost_handler.c index 6712d4c..ee41fe4 100644 --- a/lustre/ost/ost_handler.c +++ b/lustre/ost/ost_handler.c @@ -26,8 +26,11 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011 Whamcloud, Inc. + * */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -72,7 +75,7 @@ CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444, /** * Do not return server-side uid/gid to remote client */ -static void ost_drop_id(struct obd_export *exp, struct obdo *oa) +static void ost_drop_id(struct obd_export *exp, struct obdo *oa) { if (exp_connect_rmtclient(exp)) { oa->o_uid = -1; @@ -81,6 +84,34 @@ static void ost_drop_id(struct obd_export *exp, struct obdo *oa) } } +/** + * Validate oa from client. + * If the request comes from 2.0 clients, currently only RSVD seq and IDIF + * req are valid. + * a. for single MDS seq = FID_SEQ_OST_MDT0, + * b. for CMD, seq = FID_SEQ_OST_MDT0, FID_SEQ_OST_MDT1 - FID_SEQ_OST_MAX + */ +static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa, + struct obd_ioobj *ioobj) +{ + if (oa != NULL && !(oa->o_valid & OBD_MD_FLGROUP)) { + oa->o_seq = FID_SEQ_OST_MDT0; + if (ioobj) + ioobj->ioo_seq = FID_SEQ_OST_MDT0; + /* remove fid_seq_is_rsvd() after FID-on-OST allows SEQ > 9 */ + } else if (oa == NULL || + !(fid_seq_is_rsvd(oa->o_seq) || fid_seq_is_idif(oa->o_seq))) { + CERROR("%s: client %s sent invalid object "POSTID"\n", + exp->exp_obd->obd_name, obd_export_nid2str(exp), + oa ? oa->o_id : -1, oa ? oa->o_seq : -1); + return -EPROTO; + } + obdo_from_ostid(oa, &oa->o_oi); + if (ioobj) + ioobj_from_obdo(ioobj, oa); + return 0; +} + void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req) { struct oti_req_ack_lock *ack_lock; @@ -122,6 +153,10 @@ static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req, if (body->oa.o_id == 0) RETURN(-EPROTO); + rc = ost_validate_obdo(exp, &body->oa, NULL); + if (rc) + RETURN(rc); + /* If there's a DLM request, cancel the locks mentioned in it*/ if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) { struct ldlm_request *dlm; @@ -167,20 +202,24 @@ static int ost_lock_get(struct obd_export *exp, struct obdo *oa, __u64 start, __u64 count, struct lustre_handle *lh, int mode, int flags) { - struct ldlm_res_id res_id = { .name = { oa->o_id, 0, oa->o_gr, 0} }; + struct ldlm_res_id res_id; ldlm_policy_data_t policy; __u64 end = start + count; ENTRY; LASSERT(!lustre_handle_is_used(lh)); - LASSERT((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) == - (OBD_MD_FLID | OBD_MD_FLGROUP)); + /* o_id and o_gr are used for localizing resource, if client miss to set + * them, do not trigger ASSERTION. */ + if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) != + (OBD_MD_FLID | OBD_MD_FLGROUP))) + RETURN(-EPROTO); if (!(oa->o_valid & OBD_MD_FLFLAGS) || !(oa->o_flags & OBD_FL_SRVLOCK)) RETURN(0); + osc_build_res_name(oa->o_id, oa->o_seq, &res_id); CDEBUG(D_INODE, "OST-side extent lock.\n"); policy.l_extent.start = start & CFS_PAGE_MASK; @@ -211,8 +250,9 @@ static void ost_lock_put(struct obd_export *exp, static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req) { struct ost_body *body, *repbody; - struct obd_info oinfo = { { { 0 } } }; + struct obd_info *oinfo; struct lustre_handle lh = { 0 }; + struct lustre_capa *capa = NULL; int rc; ENTRY; @@ -220,31 +260,43 @@ static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req) if (body == NULL) RETURN(-EFAULT); - rc = req_capsule_server_pack(&req->rq_pill); + rc = ost_validate_obdo(exp, &body->oa, NULL); if (rc) RETURN(rc); - repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); - repbody->oa = body->oa; + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) + RETURN(rc); rc = ost_lock_get(exp, &body->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0); if (rc) RETURN(rc); - oinfo.oi_oa = &repbody->oa; - if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) { - oinfo.oi_capa = req_capsule_client_get(&req->rq_pill, - &RMF_CAPA1); - if (oinfo.oi_capa == NULL) { + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) { + capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1); + if (capa == NULL) { CERROR("Missing capability for OST GETATTR"); - RETURN (-EFAULT); + GOTO(unlock, rc = -EFAULT); } } - req->rq_status = obd_getattr(exp, &oinfo); - ost_lock_put(exp, &lh, LCK_PR); + OBD_ALLOC_PTR(oinfo); + if (!oinfo) + GOTO(unlock, rc = -ENOMEM); + oinfo->oi_oa = &body->oa; + oinfo->oi_capa = capa; + + req->rq_status = obd_getattr(exp, oinfo); + + OBD_FREE_PTR(oinfo); + repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); + repbody->oa = body->oa; ost_drop_id(exp, &repbody->oa); + +unlock: + ost_lock_put(exp, &lh, LCK_PR); + RETURN(0); } @@ -261,7 +313,8 @@ static int ost_statfs(struct ptlrpc_request *req) osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS); req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs, - cfs_time_current_64() - CFS_HZ, 0); + cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), + 0); if (req->rq_status != 0) CERROR("ost: statfs failed: rc %d\n", req->rq_status); @@ -279,13 +332,17 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req, if (body == NULL) RETURN(-EFAULT); + rc = ost_validate_obdo(req->rq_export, &body->oa, NULL); + if (rc) + RETURN(rc); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) RETURN(rc); repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); - memcpy(&repbody->oa, &body->oa, sizeof(body->oa)); - oti->oti_logcookies = &repbody->oa.o_lcookie; + repbody->oa = body->oa; + oti->oti_logcookies = &body->oa.o_lcookie; req->rq_status = obd_create(exp, &repbody->oa, NULL, oti); //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0); @@ -295,7 +352,6 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req, static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, struct obd_trans_info *oti) { - struct obd_info oinfo = { { { 0 } } }; struct ost_body *body, *repbody; int rc, flags = 0; struct lustre_handle lh = {0,}; @@ -308,11 +364,11 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, if (body == NULL) RETURN(-EFAULT); - oinfo.oi_oa = &body->oa; - oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size; - oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks; + rc = ost_validate_obdo(exp, &body->oa, NULL); + if (rc) + RETURN(rc); - if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != + if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) RETURN(-EPROTO); @@ -322,34 +378,50 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, /* standard truncate optimization: if file body is completely * destroyed, don't send data back to the server. */ - if (oinfo.oi_oa->o_size == 0) + if (body->oa.o_size == 0) flags |= LDLM_AST_DISCARD_DATA; - repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); - rc = ost_lock_get(exp, oinfo.oi_oa, oinfo.oi_oa->o_size, - oinfo.oi_oa->o_blocks, &lh, LCK_PW, flags); + rc = ost_lock_get(exp, &body->oa, body->oa.o_size, body->oa.o_blocks, + &lh, LCK_PW, flags); if (rc == 0) { - if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS && - oinfo.oi_oa->o_flags == OBD_FL_SRVLOCK) + struct obd_info *oinfo; + struct lustre_capa *capa = NULL; + + if (body->oa.o_valid & OBD_MD_FLFLAGS && + body->oa.o_flags == OBD_FL_SRVLOCK) /* * If OBD_FL_SRVLOCK is the only bit set in * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling * through filter_setattr() to filter_iocontrol(). */ - oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS; + body->oa.o_valid &= ~OBD_MD_FLFLAGS; - if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) { - oinfo.oi_capa = req_capsule_client_get(&req->rq_pill, - &RMF_CAPA1); - if (oinfo.oi_capa == NULL) { + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) { + capa = req_capsule_client_get(&req->rq_pill, + &RMF_CAPA1); + if (capa == NULL) { CERROR("Missing capability for OST PUNCH"); - RETURN (-EFAULT); + GOTO(unlock, rc = -EFAULT); } } - req->rq_status = obd_punch(exp, &oinfo, oti, NULL); + + OBD_ALLOC_PTR(oinfo); + if (!oinfo) + GOTO(unlock, rc = -ENOMEM); + oinfo->oi_oa = &body->oa; + oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size; + oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks; + oinfo->oi_capa = capa; + oinfo->oi_flags = OBD_FL_PUNCH; + + req->rq_status = obd_punch(exp, oinfo, oti, NULL); + OBD_FREE_PTR(oinfo); +unlock: ost_lock_put(exp, &lh, LCK_PW); } - repbody->oa = *oinfo.oi_oa; + + repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); + repbody->oa = body->oa; ost_drop_id(exp, &repbody->oa); RETURN(rc); } @@ -357,6 +429,7 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) { struct ost_body *body, *repbody; + struct obd_info *oinfo; struct lustre_capa *capa = NULL; int rc; ENTRY; @@ -365,6 +438,10 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) if (body == NULL) RETURN(-EFAULT); + rc = ost_validate_obdo(exp, &body->oa, NULL); + if (rc) + RETURN(rc); + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) { capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1); if (capa == NULL) { @@ -377,10 +454,18 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) if (rc) RETURN(rc); + OBD_ALLOC_PTR(oinfo); + if (!oinfo) + RETURN(-ENOMEM); + + oinfo->oi_oa = &body->oa; + oinfo->oi_capa = capa; + req->rq_status = obd_sync(exp, oinfo, body->oa.o_size, + body->oa.o_blocks, NULL); + OBD_FREE_PTR(oinfo); + repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); - memcpy(&repbody->oa, &body->oa, sizeof(body->oa)); - req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size, - repbody->oa.o_blocks, capa); + repbody->oa = body->oa; ost_drop_id(exp, &repbody->oa); RETURN(0); } @@ -389,44 +474,47 @@ static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req, struct obd_trans_info *oti) { struct ost_body *body, *repbody; + struct obd_info *oinfo; + struct lustre_capa *capa = NULL; int rc; - struct obd_info oinfo = { { { 0 } } }; ENTRY; body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); if (body == NULL) RETURN(-EFAULT); - rc = req_capsule_server_pack(&req->rq_pill); + rc = ost_validate_obdo(req->rq_export, &body->oa, NULL); if (rc) RETURN(rc); - repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); - repbody->oa = body->oa; + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) + RETURN(rc); - oinfo.oi_oa = &repbody->oa; - if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) { - oinfo.oi_capa = req_capsule_client_get(&req->rq_pill, - &RMF_CAPA1); - if (oinfo.oi_capa == NULL) { + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) { + capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1); + if (capa == NULL) { CERROR("Missing capability for OST SETATTR"); RETURN (-EFAULT); } } - req->rq_status = obd_setattr(exp, &oinfo, oti); + + OBD_ALLOC_PTR(oinfo); + if (!oinfo) + RETURN(-ENOMEM); + oinfo->oi_oa = &body->oa; + oinfo->oi_capa = capa; + + req->rq_status = obd_setattr(exp, oinfo, oti); + + OBD_FREE_PTR(oinfo); + + repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); + repbody->oa = body->oa; ost_drop_id(exp, &repbody->oa); RETURN(0); } -static int ost_bulk_timeout(void *data) -{ - ENTRY; - /* We don't fail the connection here, because having the export - * killed makes the (vital) call to commitrw very sad. - */ - RETURN(1); -} - static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc, cksum_type_t cksum_type) { @@ -457,7 +545,7 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc, kunmap(page); } - return cksum; + return fini_checksum(cksum, cksum_type); } static int ost_brw_lock_get(int mode, struct obd_export *exp, @@ -471,7 +559,7 @@ static int ost_brw_lock_get(int mode, struct obd_export *exp, int i; ENTRY; - osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id); + osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id); LASSERT(mode == LCK_PR || mode == LCK_PW); LASSERT(!lustre_handle_is_used(lh)); @@ -576,7 +664,7 @@ static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *ob struct ost_prolong_data opd = { 0 }; ENTRY; - osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id); + osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id); opd.opd_mode = mode; opd.opd_exp = req->rq_export; @@ -710,6 +798,10 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) if (ioo == NULL) GOTO(out, rc = -EFAULT); + rc = ost_validate_obdo(exp, &body->oa, ioo); + if (rc) + RETURN(rc); + niocount = ioo->ioo_bufcnt; remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE); if (remote_nb == NULL) @@ -760,7 +852,7 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) desc = ptlrpc_prep_bulk_exp(req, npages, BULK_PUT_SOURCE, OST_BULK_PORTAL); if (desc == NULL) - GOTO(out_lock, rc = -ENOMEM); + GOTO(out_commitrw, rc = -ENOMEM); if (!lustre_handle_is_used(&lockh)) /* no needs to try to prolong lock if server is asked @@ -794,13 +886,12 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) } if (body->oa.o_valid & OBD_MD_FLCKSUM) { - cksum_type_t cksum_type = OBD_CKSUM_CRC32; - - if (body->oa.o_valid & OBD_MD_FLFLAGS) - cksum_type = cksum_type_unpack(body->oa.o_flags); + cksum_type_t cksum_type = + cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ? + body->oa.o_flags : 0); body->oa.o_flags = cksum_type_pack(cksum_type); body->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS; - body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ, cksum_type); + body->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type); CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum); } else { body->oa.o_valid = 0; @@ -810,79 +901,11 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) /* Check if client was evicted while we were doing i/o before touching network */ if (rc == 0) { - /* Check if there is eviction in progress, and if so, wait for - * it to finish */ - if (unlikely(cfs_atomic_read(&exp->exp_obd-> - obd_evict_inprogress))) { - lwi = LWI_INTR(NULL, NULL); - rc = l_wait_event(exp->exp_obd-> - obd_evict_inprogress_waitq, - !cfs_atomic_read(&exp->exp_obd-> - obd_evict_inprogress), - &lwi); - } - /* Check if client was evicted or tried to reconnect already */ - if (exp->exp_failed || exp->exp_abort_active_req) - rc = -ENOTCONN; - else { - rc = sptlrpc_svc_wrap_bulk(req, desc); - if (rc == 0) - rc = ptlrpc_start_bulk_transfer(desc); - } - - if (rc == 0) { - time_t start = cfs_time_current_sec(); - do { - long timeoutl = req->rq_deadline - - cfs_time_current_sec(); - cfs_duration_t timeout = timeoutl <= 0 ? - CFS_TICK : cfs_time_seconds(timeoutl); - lwi = LWI_TIMEOUT_INTERVAL(timeout, - cfs_time_seconds(1), - ost_bulk_timeout, - desc); - rc = l_wait_event(desc->bd_waitq, - !ptlrpc_server_bulk_active(desc) || - exp->exp_failed || - exp->exp_abort_active_req, - &lwi); - LASSERT(rc == 0 || rc == -ETIMEDOUT); - /* Wait again if we changed deadline */ - } while ((rc == -ETIMEDOUT) && - (req->rq_deadline > cfs_time_current_sec())); - - if (rc == -ETIMEDOUT) { - DEBUG_REQ(D_ERROR, req, - "timeout on bulk PUT after %ld%+lds", - req->rq_deadline - start, - cfs_time_current_sec() - - req->rq_deadline); - ptlrpc_abort_bulk(desc); - } else if (exp->exp_failed) { - DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT"); - rc = -ENOTCONN; - ptlrpc_abort_bulk(desc); - } else if (exp->exp_abort_active_req) { - DEBUG_REQ(D_ERROR, req, "Reconnect on bulk PUT"); - /* we don't reply anyway */ - rc = -ETIMEDOUT; - ptlrpc_abort_bulk(desc); - } else if (!desc->bd_success || - desc->bd_nob_transferred != desc->bd_nob) { - DEBUG_REQ(D_ERROR, req, "%s bulk PUT %d(%d)", - desc->bd_success ? - "truncated" : "network error on", - desc->bd_nob_transferred, - desc->bd_nob); - /* XXX should this be a different errno? */ - rc = -ETIMEDOUT; - } - } else { - DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc); - } + rc = target_bulk_io(exp, desc, &lwi); no_reply = rc != 0; } +out_commitrw: /* Must commit after prep above in all cases */ rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo, remote_nb, npages, local_nb, oti, rc); @@ -942,7 +965,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) int rc, i, j; obd_count client_cksum = 0, server_cksum = 0; cksum_type_t cksum_type = OBD_CKSUM_CRC32; - int no_reply = 0; + int no_reply = 0, mmap = 0; __u32 o_uid = 0, o_gid = 0; struct ost_thread_local_cache *tls; ENTRY; @@ -957,32 +980,22 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) /* pause before transaction has been started */ OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4); - /* Check if there is eviction in progress, and if so, wait for it to - * finish */ - if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) { - lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes - rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq, - !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress), - &lwi); - } - if (exp->exp_failed) - GOTO(out, rc = -ENOTCONN); - /* ost_body, ioobj & noibuf_remote are verified and swabbed in * ost_rw_hpreq_check(). */ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); if (body == NULL) GOTO(out, rc = -EFAULT); - if ((body->oa.o_flags & OBD_BRW_MEMALLOC) && - (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self)) - libcfs_memory_pressure_set(); - objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ, RCL_CLIENT) / sizeof(*ioo); ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ); if (ioo == NULL) GOTO(out, rc = -EFAULT); + + rc = ost_validate_obdo(exp, &body->oa, ioo); + if (rc) + RETURN(rc); + for (niocount = i = 0; i < objcount; i++) niocount += ioo[i].ioo_bufcnt; @@ -996,6 +1009,10 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb))) GOTO(out, rc = -EFAULT); + if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) && + (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self)) + cfs_memory_pressure_set(); + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) { capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1); if (capa == NULL) { @@ -1009,7 +1026,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) rc = req_capsule_server_pack(&req->rq_pill); if (rc != 0) GOTO(out, rc); - OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val); + CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val); rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS); tls = ost_tls_get(req); @@ -1047,6 +1064,8 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) if (body->oa.o_valid & OBD_MD_FLFLAGS) cksum_type = cksum_type_unpack(body->oa.o_flags); } + if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP) + mmap = 1; /* Because we already sync grant info with client when reconnect, * grant info will be cleared for resent req, then fed_grant and @@ -1069,7 +1088,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) desc = ptlrpc_prep_bulk_exp(req, npages, BULK_GET_SINK, OST_BULK_PORTAL); if (desc == NULL) - GOTO(out_lock, rc = -ENOMEM); + GOTO(skip_transfer, rc = -ENOMEM); /* NB Having prepped, we must commit... */ @@ -1082,60 +1101,10 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) if (rc != 0) GOTO(out_lock, rc); - /* Check if client was evicted or tried to reconnect while we - * were doing i/o before touching network */ - if (desc->bd_export->exp_failed || - desc->bd_export->exp_abort_active_req) - rc = -ENOTCONN; - else - rc = ptlrpc_start_bulk_transfer(desc); - if (rc == 0) { - time_t start = cfs_time_current_sec(); - do { - long timeoutl = req->rq_deadline - - cfs_time_current_sec(); - cfs_duration_t timeout = timeoutl <= 0 ? - CFS_TICK : cfs_time_seconds(timeoutl); - lwi = LWI_TIMEOUT_INTERVAL(timeout, cfs_time_seconds(1), - ost_bulk_timeout, desc); - rc = l_wait_event(desc->bd_waitq, - !ptlrpc_server_bulk_active(desc) || - desc->bd_export->exp_failed || - desc->bd_export->exp_abort_active_req, - &lwi); - LASSERT(rc == 0 || rc == -ETIMEDOUT); - /* Wait again if we changed deadline */ - } while ((rc == -ETIMEDOUT) && - (req->rq_deadline > cfs_time_current_sec())); - - if (rc == -ETIMEDOUT) { - DEBUG_REQ(D_ERROR, req, - "timeout on bulk GET after %ld%+lds", - req->rq_deadline - start, - cfs_time_current_sec() - - req->rq_deadline); - ptlrpc_abort_bulk(desc); - } else if (desc->bd_export->exp_failed) { - DEBUG_REQ(D_ERROR, req, "Eviction on bulk GET"); - rc = -ENOTCONN; - ptlrpc_abort_bulk(desc); - } else if (desc->bd_export->exp_abort_active_req) { - DEBUG_REQ(D_ERROR, req, "Reconnect on bulk GET"); - /* we don't reply anyway */ - rc = -ETIMEDOUT; - ptlrpc_abort_bulk(desc); - } else if (!desc->bd_success) { - DEBUG_REQ(D_ERROR, req, "network error on bulk GET"); - /* XXX should this be a different errno? */ - rc = -ETIMEDOUT; - } else { - rc = sptlrpc_svc_unwrap_bulk(req, desc); - } - } else { - DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc); - } + rc = target_bulk_io(exp, desc, &lwi); no_reply = rc != 0; +skip_transfer: repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY); memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa)); @@ -1148,8 +1117,9 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) repbody->oa.o_cksum = server_cksum; cksum_counter++; if (unlikely(client_cksum != server_cksum)) { - CERROR("client csum %x, server csum %x\n", - client_cksum, server_cksum); + CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR, + "client csum %x, server csum %x\n", + client_cksum, server_cksum); cksum_counter = 0; } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){ CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n", @@ -1179,7 +1149,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) */ repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME); - if (unlikely(client_cksum != server_cksum && rc == 0)) { + if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) { int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type); char *msg; char *via; @@ -1200,18 +1170,20 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) } LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from " - "%s%s%s inum "LPU64"/"LPU64" object " + "%s%s%s inode "DFID" object " LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n", exp->exp_obd->obd_name, msg, libcfs_id2str(req->rq_peer), via, router, body->oa.o_valid & OBD_MD_FLFID ? - body->oa.o_fid : (__u64)0, + body->oa.o_parent_seq : (__u64)0, body->oa.o_valid & OBD_MD_FLFID ? - body->oa.o_generation :(__u64)0, + body->oa.o_parent_oid : 0, + body->oa.o_valid & OBD_MD_FLFID ? + body->oa.o_parent_ver : 0, body->oa.o_id, body->oa.o_valid & OBD_MD_FLGROUP ? - body->oa.o_gr : (__u64)0, + body->oa.o_seq : (__u64)0, local_nb[0].offset, local_nb[npages-1].offset + local_nb[npages-1].len - 1 ); @@ -1250,14 +1222,6 @@ out_bulk: if (desc) ptlrpc_free_bulk(desc); out: - /* XXX: don't send reply if obd rdonly mode, this can cause data loss - * on client, see bug 22190. Remove this when async bulk will be done. - * Meanwhile, if this is umount then don't reply anything. */ - if (req->rq_export->exp_obd->obd_no_transno) { - no_reply = req->rq_export->exp_obd->obd_stopping; - rc = -EIO; - } - if (rc == 0) { oti_to_request(oti, req); target_committed_to_req(req); @@ -1277,7 +1241,7 @@ out: exp->exp_connection->c_remote_uuid.uuid, libcfs_id2str(req->rq_peer)); } - libcfs_memory_pressure_clr(); + cfs_memory_pressure_clr(); RETURN(rc); } @@ -1374,6 +1338,15 @@ static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req) } keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT); + if (KEY_IS(KEY_FIEMAP)) { + struct ll_fiemap_info_key *fm_key = key; + int rc; + + rc = ost_validate_obdo(exp, &fm_key->oa, NULL); + if (rc) + RETURN(rc); + } + rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL); if (rc) RETURN(rc); @@ -1454,7 +1427,7 @@ static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req) GOTO(out, rc); repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT); - req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt); + req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt, NULL); *repoqa = *oqaq; out: @@ -1649,6 +1622,56 @@ static int ost_connect_check_sptlrpc(struct ptlrpc_request *req) return rc; } +/* Ensure that data and metadata are synced to the disk when lock is cancelled + * (if requested) */ +int ost_blocking_ast(struct ldlm_lock *lock, + struct ldlm_lock_desc *desc, + void *data, int flag) +{ + __u32 sync_lock_cancel = 0; + __u32 len = sizeof(sync_lock_cancel); + int rc = 0; + ENTRY; + + rc = obd_get_info(lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL), + KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL); + + if (!rc && flag == LDLM_CB_CANCELING && + (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) && + (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL || + (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL && + lock->l_flags & LDLM_FL_CBPENDING))) { + struct obd_info *oinfo; + struct obdo *oa; + int rc; + + OBD_ALLOC_PTR(oinfo); + if (!oinfo) + RETURN(-ENOMEM); + OBDO_ALLOC(oa); + if (!oa) { + OBD_FREE_PTR(oinfo); + RETURN(-ENOMEM); + } + oa->o_id = lock->l_resource->lr_name.name[0]; + oa->o_seq = lock->l_resource->lr_name.name[1]; + oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP; + oinfo->oi_oa = oa; + + rc = obd_sync(lock->l_export, oinfo, + lock->l_policy_data.l_extent.start, + lock->l_policy_data.l_extent.end, NULL); + if (rc) + CERROR("Error %d syncing data on lock cancel\n", rc); + + OBDO_FREE(oa); + OBD_FREE_PTR(oinfo); + } + + rc = ldlm_server_blocking_ast(lock, desc, data, flag); + RETURN(rc); +} + static int ost_filter_recovery_request(struct ptlrpc_request *req, struct obd_device *obd, int *process) { @@ -1757,7 +1780,7 @@ static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req, struct obd_ioobj *ioo; struct ost_body *body; int objcount, niocount; - int mode, opc, i; + int mode, opc, i, rc; __u64 start, end; ENTRY; @@ -1776,6 +1799,10 @@ static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req, if (ioo == NULL) RETURN(0); + rc = ost_validate_obdo(req->rq_export, &body->oa, ioo); + if (rc) + RETURN(rc); + for (niocount = i = 0; i < objcount; i++) niocount += ioo[i].ioo_bufcnt; @@ -1794,7 +1821,7 @@ static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req, nb[ioo->ioo_bufcnt - 1].len - 1) | ~CFS_PAGE_MASK; LASSERT(lock->l_resource != NULL); - if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_gr, + if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq, &lock->l_resource->lr_name)) RETURN(0); @@ -1824,7 +1851,7 @@ static int ost_rw_hpreq_check(struct ptlrpc_request *req) struct obd_ioobj *ioo; struct ost_body *body; int objcount, niocount; - int mode, opc, i; + int mode, opc, i, rc; ENTRY; opc = lustre_msg_get_opc(req->rq_reqmsg); @@ -1840,6 +1867,10 @@ static int ost_rw_hpreq_check(struct ptlrpc_request *req) if (ioo == NULL) RETURN(-EFAULT); + rc = ost_validate_obdo(req->rq_export, &body->oa, ioo); + if (rc) + RETURN(rc); + for (niocount = i = 0; i < objcount; i++) niocount += ioo[i].ioo_bufcnt; nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE); @@ -1897,12 +1928,17 @@ static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req, struct ldlm_lock *lock) { struct ost_body *body; + int rc; ENTRY; body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); if (body == NULL) RETURN(0); /* can't return -EFAULT here */ + rc = ost_validate_obdo(req->rq_export, &body->oa, NULL); + if (rc) + RETURN(rc); + if (body->oa.o_valid & OBD_MD_FLHANDLE && body->oa.o_handle.cookie == lock->l_handle.h_cookie) RETURN(1); @@ -1915,11 +1951,16 @@ static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req, static int ost_punch_hpreq_check(struct ptlrpc_request *req) { struct ost_body *body; + int rc; body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); if (body == NULL) RETURN(-EFAULT); + rc = ost_validate_obdo(req->rq_export, &body->oa, NULL); + if (rc) + RETURN(rc); + LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) || !(body->oa.o_flags & OBD_FL_SRVLOCK)); @@ -1970,6 +2011,7 @@ static int ost_hpreq_handler(struct ptlrpc_request *req) CERROR("Missing/short ost_body\n"); RETURN(-EFAULT); } + objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ, RCL_CLIENT) / @@ -2265,7 +2307,7 @@ int ost_handle(struct ptlrpc_request *req) if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE)) RETURN(0); rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast, - ldlm_server_blocking_ast, + ost_blocking_ast, ldlm_server_glimpse_ast); fail = OBD_FAIL_OST_LDLM_REPLY_NET; break; @@ -2411,7 +2453,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) GOTO(out_lprocfs, rc = -ENOMEM); } - rc = ptlrpc_start_threads(obd, ost->ost_service); + rc = ptlrpc_start_threads(ost->ost_service); if (rc) GOTO(out_service, rc = -EINVAL); @@ -2440,7 +2482,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) GOTO(out_service, rc = -ENOMEM); } - rc = ptlrpc_start_threads(obd, ost->ost_create_service); + rc = ptlrpc_start_threads(ost->ost_create_service); if (rc) GOTO(out_create, rc = -EINVAL); @@ -2460,7 +2502,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) ost->ost_io_service->srv_init = ost_thread_init; ost->ost_io_service->srv_done = ost_thread_done; ost->ost_io_service->srv_cpu_affinity = 1; - rc = ptlrpc_start_threads(obd, ost->ost_io_service); + rc = ptlrpc_start_threads(ost->ost_io_service); if (rc) GOTO(out_io, rc = -EINVAL); @@ -2490,13 +2532,9 @@ static int ost_cleanup(struct obd_device *obd) ping_evictor_stop(); - cfs_spin_lock_bh(&obd->obd_processing_task_lock); - if (obd->obd_recovering) { - target_cancel_recovery_timer(obd); - obd->obd_recovering = 0; - } - cfs_spin_unlock_bh(&obd->obd_processing_task_lock); - + /* there is no recovery for OST OBD, all recovery is controlled by + * obdfilter OBD */ + LASSERT(obd->obd_recovering == 0); cfs_down(&ost->ost_health_sem); ptlrpc_unregister_service(ost->ost_service); ptlrpc_unregister_service(ost->ost_create_service);