X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fost%2Fost_handler.c;h=cb520bb40ccf479dc586f3432c24e159a90d1fd2;hp=b3eec4582665e6897b6eb045259e05a425651cbc;hb=e6ba48a08c9fe831b3f60985a72bb31b9806bb34;hpb=113303973ec9f8484eb2355a1a6ef3c4c7fd6a56 diff --git a/lustre/ost/ost_handler.c b/lustre/ost/ost_handler.c index b3eec45..cb520bb 100644 --- a/lustre/ost/ost_handler.c +++ b/lustre/ost/ost_handler.c @@ -1,36 +1,42 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Copyright (C) 2001-2003 Cluster File Systems, Inc. - * Author: Peter J. Braam - * Author: Phil Schwan + * GPL HEADER START * - * This file is part of the Lustre file system, http://www.lustre.org - * Lustre is a trademark of Cluster File Systems, Inc. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * You may have signed or agreed to another license before downloading - * this software. If so, you are bound by the terms and conditions - * of that agreement, and the following does not apply to you. See the - * LICENSE file included with this distribution for more information. + * This program is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License version 2 only, + * as published by the Free Software Foundation. * - * If you did not agree to a different license, then this copy of Lustre - * is open source software; you can redistribute it and/or modify it - * under the terms of version 2 of the GNU General Public License as - * published by the Free Software Foundation. + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License version 2 for more details (a copy is included + * in the LICENSE file that accompanied this code). * - * In either case, Lustre is distributed in the hope that it will be - * useful, but WITHOUT ANY WARRANTY; without even the implied warranty - * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * license text for more details. + * You should have received a copy of the GNU General Public License + * version 2 along with this program; If not, see + * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * - * Storage Target Handling functions - * Lustre Object Server Module (OST) + * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, + * CA 95054 USA or visit www.sun.com if you need additional information or + * have any questions. * - * This server is single threaded at present (but can easily be multi - * threaded). For testing and management it is treated as an - * obd_device, although it does not export a full OBD method table - * (the requests are coming in over the wire, so object target - * modules do not have a full method table.) + * GPL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Use is subject to license terms. + */ +/* + * This file is part of Lustre, http://www.lustre.org/ + * Lustre is a trademark of Sun Microsystems, Inc. + * + * lustre/ost/ost_handler.c + * + * Author: Peter J. Braam + * Author: Phil Schwan */ #ifndef EXPORT_SYMTAB @@ -39,6 +45,7 @@ #define DEBUG_SUBSYSTEM S_OST #include +#include #include #include #include @@ -46,7 +53,6 @@ #include #include #include -#include #include #include #include "ost_internal.h" @@ -55,6 +61,26 @@ static int oss_num_threads; CFS_MODULE_PARM(oss_num_threads, "i", int, 0444, "number of OSS service threads to start"); +static int ost_num_threads; +CFS_MODULE_PARM(ost_num_threads, "i", int, 0444, + "number of OST service threads to start (deprecated)"); + +static int oss_num_create_threads; +CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444, + "number of OSS create threads to start"); + +/** + * Do not return server-side uid/gid to remote client + */ +static void ost_drop_id(struct obd_export *exp, struct obdo *oa) +{ + if (exp_connect_rmtclient(exp)) { + oa->o_uid = -1; + oa->o_gid = -1; + oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID); + } +} + void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req) { struct oti_req_ack_lock *ack_lock; @@ -72,7 +98,7 @@ void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req) if (!ack_lock->mode) break; /* XXX not even calling target_send_reply in some cases... */ - ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode); + ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0); } } @@ -80,7 +106,9 @@ static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req, struct obd_trans_info *oti) { struct ost_body *body, *repbody; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + struct lustre_capa *capa = NULL; + int rc; ENTRY; body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), @@ -88,16 +116,31 @@ static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req, if (body == NULL) RETURN(-EFAULT); + if (body->oa.o_id == 0) + RETURN(-EPROTO); + + if (lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1)) { + struct ldlm_request *dlm; + dlm = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*dlm), + lustre_swab_ldlm_request); + if (dlm == NULL) + RETURN (-EFAULT); + ldlm_request_cancel(req, dlm, 0); + } + + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) + capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 2); + rc = lustre_pack_reply(req, 2, size, NULL); if (rc) RETURN(rc); if (body->oa.o_valid & OBD_MD_FLCOOKIE) - oti->oti_logcookies = obdo_logcookie(&body->oa); + oti->oti_logcookies = &body->oa.o_lcookie; repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repbody)); memcpy(&repbody->oa, &body->oa, sizeof(body->oa)); - req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL); + req->rq_status = obd_destroy(exp, &body->oa, NULL, oti, NULL, capa); RETURN(0); } @@ -105,7 +148,8 @@ static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req) { struct ost_body *body, *repbody; struct obd_info oinfo = { { { 0 } } }; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + int rc; ENTRY; body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), @@ -119,17 +163,22 @@ static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req) repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repbody)); - memcpy(&repbody->oa, &body->oa, sizeof(body->oa)); + repbody->oa = body->oa; oinfo.oi_oa = &repbody->oa; + if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) + oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg, + REQ_REC_OFF + 1); req->rq_status = obd_getattr(exp, &oinfo); + ost_drop_id(exp, &repbody->oa); RETURN(0); } static int ost_statfs(struct ptlrpc_request *req) { struct obd_statfs *osfs; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*osfs) }; + int rc; ENTRY; rc = lustre_pack_reply(req, 2, size, NULL); @@ -138,9 +187,9 @@ static int ost_statfs(struct ptlrpc_request *req) osfs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*osfs)); - req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs, - cfs_time_current_64() - HZ); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_ENOSPC)) + req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs, + cfs_time_current_64() - HZ, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC)) osfs->os_bfree = osfs->os_bavail = 64; if (req->rq_status != 0) CERROR("ost: statfs failed: rc %d\n", req->rq_status); @@ -152,7 +201,8 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req, struct obd_trans_info *oti) { struct ost_body *body, *repbody; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + int rc; ENTRY; body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), @@ -167,7 +217,7 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req, repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repbody)); memcpy(&repbody->oa, &body->oa, sizeof(body->oa)); - oti->oti_logcookies = obdo_logcookie(&repbody->oa); + oti->oti_logcookies = &repbody->oa.o_lcookie; req->rq_status = obd_create(exp, &repbody->oa, NULL, oti); //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0); RETURN(0); @@ -181,13 +231,14 @@ static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa, struct lustre_handle *lh) { int flags; - struct ldlm_res_id res_id = { .name = { oa->o_id } }; + struct ldlm_res_id res_id; ldlm_policy_data_t policy; __u64 start; __u64 finis; ENTRY; + osc_build_res_name(oa->o_id, oa->o_gr, &res_id); LASSERT(!lustre_handle_is_used(lh)); if (!(oa->o_valid & OBD_MD_FLFLAGS) || @@ -216,10 +267,11 @@ static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa, else policy.l_extent.end = finis | ~CFS_PAGE_MASK; - RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, res_id, + RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id, LDLM_EXTENT, &policy, LCK_PW, &flags, ldlm_blocking_ast, ldlm_completion_ast, - ldlm_glimpse_ast, NULL, 0, NULL, lh)); + ldlm_glimpse_ast, NULL, 0, NULL, + NULL, lh)); } /* @@ -240,17 +292,17 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, { struct obd_info oinfo = { { { 0 } } }; struct ost_body *body, *repbody; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + int rc; struct lustre_handle lh = {0,}; ENTRY; /* check that we do support OBD_CONNECT_TRUNCLOCK. */ CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK); - body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), - lustre_swab_ost_body); - if (body == NULL) - RETURN(-EFAULT); + /* ost_body is varified and swabbed in ost_hpreq_handler() */ + body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body)); + LASSERT(body != NULL); oinfo.oi_oa = &body->oa; oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size; @@ -258,7 +310,7 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) != (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) - RETURN(-EINVAL); + RETURN(-EPROTO); rc = lustre_pack_reply(req, 2, size, NULL); if (rc) @@ -277,17 +329,23 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, */ oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS; + if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) + oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg, + REQ_REC_OFF + 1); req->rq_status = obd_punch(exp, &oinfo, oti, NULL); ost_punch_lock_put(exp, oinfo.oi_oa, &lh); } repbody->oa = *oinfo.oi_oa; + ost_drop_id(exp, &repbody->oa); RETURN(rc); } static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) { struct ost_body *body, *repbody; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + struct lustre_capa *capa = NULL; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + int rc; ENTRY; body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), @@ -295,6 +353,9 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) if (body == NULL) RETURN(-EFAULT); + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) + capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 1); + rc = lustre_pack_reply(req, 2, size, NULL); if (rc) RETURN(rc); @@ -303,7 +364,8 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) sizeof(*repbody)); 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); + repbody->oa.o_blocks, capa); + ost_drop_id(exp, &repbody->oa); RETURN(0); } @@ -311,7 +373,8 @@ static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req, struct obd_trans_info *oti) { struct ost_body *body, *repbody; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repbody) }; + int rc; struct obd_info oinfo = { { { 0 } } }; ENTRY; @@ -326,10 +389,14 @@ static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req, repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repbody)); - memcpy(&repbody->oa, &body->oa, sizeof(body->oa)); + repbody->oa = body->oa; oinfo.oi_oa = &repbody->oa; + if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) + oinfo.oi_capa = lustre_unpack_capa(req->rq_reqmsg, + REQ_REC_OFF + 1); req->rq_status = obd_setattr(exp, &oinfo, oti); + ost_drop_id(exp, &repbody->oa); RETURN(0); } @@ -342,105 +409,13 @@ static int ost_bulk_timeout(void *data) RETURN(1); } -static int get_per_page_niobufs(struct obd_ioobj *ioo, int nioo, - struct niobuf_remote *rnb, int nrnb, - struct niobuf_remote **pp_rnbp) +static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc, + cksum_type_t cksum_type) { - /* Copy a remote niobuf, splitting it into page-sized chunks - * and setting ioo[i].ioo_bufcnt accordingly */ - struct niobuf_remote *pp_rnb; - int i; - int j; - int page; - int rnbidx = 0; - int npages = 0; - - /* - * array of sufficient size already preallocated by caller - */ - LASSERT(pp_rnbp != NULL); - LASSERT(*pp_rnbp != NULL); - - /* first count and check the number of pages required */ - for (i = 0; i < nioo; i++) - for (j = 0; j < ioo->ioo_bufcnt; j++, rnbidx++) { - obd_off offset = rnb[rnbidx].offset; - obd_off p0 = offset >> CFS_PAGE_SHIFT; - obd_off pn = (offset + rnb[rnbidx].len - 1)>>CFS_PAGE_SHIFT; - - LASSERT(rnbidx < nrnb); - - npages += (pn + 1 - p0); - - if (rnb[rnbidx].len == 0) { - CERROR("zero len BRW: obj %d objid "LPX64 - " buf %u\n", i, ioo[i].ioo_id, j); - return -EINVAL; - } - if (j > 0 && - rnb[rnbidx].offset <= rnb[rnbidx-1].offset) { - CERROR("unordered BRW: obj %d objid "LPX64 - " buf %u offset "LPX64" <= "LPX64"\n", - i, ioo[i].ioo_id, j, rnb[rnbidx].offset, - rnb[rnbidx].offset); - return -EINVAL; - } - } - - LASSERT(rnbidx == nrnb); - - if (npages == nrnb) { /* all niobufs are for single pages */ - *pp_rnbp = rnb; - return npages; - } - - pp_rnb = *pp_rnbp; - - /* now do the actual split */ - page = rnbidx = 0; - for (i = 0; i < nioo; i++) { - int obj_pages = 0; - - for (j = 0; j < ioo[i].ioo_bufcnt; j++, rnbidx++) { - obd_off off = rnb[rnbidx].offset; - int nob = rnb[rnbidx].len; - - LASSERT(rnbidx < nrnb); - do { - obd_off poff = off & ~CFS_PAGE_MASK; - int pnob = (poff + nob > CFS_PAGE_SIZE) ? - CFS_PAGE_SIZE - poff : nob; - - LASSERT(page < npages); - pp_rnb[page].len = pnob; - pp_rnb[page].offset = off; - pp_rnb[page].flags = rnb[rnbidx].flags; - - CDEBUG(0, " obj %d id "LPX64 - "page %d(%d) "LPX64" for %d, flg %x\n", - i, ioo[i].ioo_id, obj_pages, page, - pp_rnb[page].offset, pp_rnb[page].len, - pp_rnb[page].flags); - page++; - obj_pages++; - - off += pnob; - nob -= pnob; - } while (nob > 0); - LASSERT(nob == 0); - } - ioo[i].ioo_bufcnt = obj_pages; - } - LASSERT(page == npages); - - return npages; -} - -static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc) -{ - __u32 cksum = ~0; + __u32 cksum; int i; + cksum = init_checksum(cksum_type); for (i = 0; i < desc->bd_iov_count; i++) { struct page *page = desc->bd_iov[i].kiov_page; int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK; @@ -449,79 +424,36 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc) /* corrupt the data before we compute the checksum, to * simulate a client->OST data error */ - if (i == 0 &&OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_CHECKSUM_RECEIVE)) + if (i == 0 && opc == OST_WRITE && + OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) memcpy(ptr, "bad3", min(4, len)); - cksum = crc32_le(cksum, ptr, len); + cksum = compute_checksum(cksum, ptr, len, cksum_type); /* corrupt the data after we compute the checksum, to * simulate an OST->client data error */ - if (i == 0 && OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_CHECKSUM_SEND)) + if (i == 0 && opc == OST_READ && + OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) { memcpy(ptr, "bad4", min(4, len)); + /* nobody should use corrupted page again */ + ClearPageUptodate(page); + } kunmap(page); } return cksum; } -/* - * populate @nio by @nrpages pages from per-thread page pool - */ -static void ost_nio_pages_get(struct ptlrpc_request *req, - struct niobuf_local *nio, int nrpages) -{ - int i; - struct ost_thread_local_cache *tls; - - ENTRY; - - LASSERT(nrpages <= OST_THREAD_POOL_SIZE); - LASSERT(req != NULL); - LASSERT(req->rq_svc_thread != NULL); - - tls = ost_tls(req); - LASSERT(tls != NULL); - - memset(nio, 0, nrpages * sizeof *nio); - for (i = 0; i < nrpages; ++ i) { - struct page *page; - - page = tls->page[i]; - LASSERT(page != NULL); - POISON_PAGE(page, 0xf1); - nio[i].page = page; - LL_CDEBUG_PAGE(D_INFO, page, "%d\n", i); - } - EXIT; -} - -/* - * Dual for ost_nio_pages_get(). Poison pages in pool for debugging - */ -static void ost_nio_pages_put(struct ptlrpc_request *req, - struct niobuf_local *nio, int nrpages) -{ - int i; - - ENTRY; - - LASSERT(nrpages <= OST_THREAD_POOL_SIZE); - - for (i = 0; i < nrpages; ++ i) - POISON_PAGE(nio[i].page, 0xf2); - EXIT; -} - static int ost_brw_lock_get(int mode, struct obd_export *exp, struct obd_ioobj *obj, struct niobuf_remote *nb, struct lustre_handle *lh) { int flags = 0; int nrbufs = obj->ioo_bufcnt; - struct ldlm_res_id res_id = { .name = { obj->ioo_id } }; + struct ldlm_res_id res_id; ldlm_policy_data_t policy; int i; - ENTRY; + osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id); LASSERT(mode == LCK_PR || mode == LCK_PW); LASSERT(!lustre_handle_is_used(lh)); @@ -537,10 +469,11 @@ static int ost_brw_lock_get(int mode, struct obd_export *exp, policy.l_extent.end = (nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK; - RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, res_id, + RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id, LDLM_EXTENT, &policy, mode, &flags, ldlm_blocking_ast, ldlm_completion_ast, - ldlm_glimpse_ast, NULL, 0, NULL, lh)); + ldlm_glimpse_ast, NULL, 0, NULL, + NULL, lh)); } static void ost_brw_lock_put(int mode, @@ -559,7 +492,10 @@ static void ost_brw_lock_put(int mode, struct ost_prolong_data { struct obd_export *opd_exp; ldlm_policy_data_t opd_policy; + struct obdo *opd_oa; ldlm_mode_t opd_mode; + int opd_lock_match; + int opd_timeout; }; static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data) @@ -589,6 +525,14 @@ static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data) return LDLM_ITER_CONTINUE; } + /* Fill the obdo with the matched lock handle. + * XXX: it is possible in some cases the IO RPC is covered by several + * locks, even for the write case, so it may need to be a lock list. */ + if (opd->opd_oa && !(opd->opd_oa->o_valid & OBD_MD_FLHANDLE)) { + opd->opd_oa->o_handle.cookie = lock->l_handle.h_cookie; + opd->opd_oa->o_valid |= OBD_MD_FLHANDLE; + } + if (!(lock->l_flags & LDLM_FL_AST_SENT)) { /* ignore locks not being cancelled */ return LDLM_ITER_CONTINUE; @@ -596,86 +540,118 @@ static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data) /* OK. this is a possible lock the user holds doing I/O * let's refresh eviction timer for it */ - ldlm_refresh_waiting_lock(lock); + ldlm_refresh_waiting_lock(lock, opd->opd_timeout); + opd->opd_lock_match = 1; return LDLM_ITER_CONTINUE; } -static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj, - struct niobuf_remote *nb, ldlm_mode_t mode) +static int ost_rw_prolong_locks(struct ptlrpc_request *req, struct obd_ioobj *obj, + struct niobuf_remote *nb, struct obdo *oa, + ldlm_mode_t mode) { - struct ldlm_res_id res_id = { .name = { obj->ioo_id } }; + struct ldlm_res_id res_id; int nrbufs = obj->ioo_bufcnt; - struct ost_prolong_data opd; - + struct ost_prolong_data opd = { 0 }; ENTRY; + osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id); + opd.opd_mode = mode; - opd.opd_exp = exp; + opd.opd_exp = req->rq_export; opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK; opd.opd_policy.l_extent.end = (nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK; + /* prolong locks for the current service time of the corresponding + * portal (= OST_IO_PORTAL) */ + opd.opd_timeout = AT_OFF ? obd_timeout / 2: + max(at_est2timeout(at_get(&req->rq_rqbd-> + rqbd_service->srv_at_estimate)), ldlm_timeout); + CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n", res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start, opd.opd_policy.l_extent.end); - ldlm_resource_iterate(exp->exp_obd->obd_namespace, &res_id, + + if (oa->o_valid & OBD_MD_FLHANDLE) { + struct ldlm_lock *lock; + + lock = ldlm_handle2lock(&oa->o_handle); + if (lock != NULL) { + ost_prolong_locks_iter(lock, &opd); + if (opd.opd_lock_match) { + LDLM_LOCK_PUT(lock); + RETURN(1); + } + + /* Check if the lock covers the whole IO region, + * otherwise iterate through the resource. */ + if (lock->l_policy_data.l_extent.end >= + opd.opd_policy.l_extent.end && + lock->l_policy_data.l_extent.start <= + opd.opd_policy.l_extent.start) { + LDLM_LOCK_PUT(lock); + RETURN(0); + } + LDLM_LOCK_PUT(lock); + } + } + + opd.opd_oa = oa; + ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id, ost_prolong_locks_iter, &opd); + RETURN(opd.opd_lock_match); } static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) { - struct ptlrpc_bulk_desc *desc; + struct ptlrpc_bulk_desc *desc = NULL; + struct obd_export *exp = req->rq_export; struct niobuf_remote *remote_nb; - struct niobuf_remote *pp_rnb = NULL; struct niobuf_local *local_nb; struct obd_ioobj *ioo; struct ost_body *body, *repbody; + struct lustre_capa *capa = NULL; struct l_wait_info lwi; struct lustre_handle lockh = { 0 }; - int size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; - int comms_error = 0, niocount, npages, nob = 0, rc, i; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + int niocount, npages, nob = 0, rc, i; int no_reply = 0; ENTRY; + req->rq_bulk_read = 1; + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK)) GOTO(out, rc = -EIO); - OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE, - (obd_timeout + 1) / 4); + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4); - body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), - lustre_swab_ost_body); - if (body == NULL) { - CERROR("Missing/short ost_body\n"); - GOTO(out, rc = -EFAULT); + /* Check if there is eviction in progress, and if so, wait for it to + * finish */ + if (unlikely(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, + !atomic_read(&exp->exp_obd->obd_evict_inprogress), + &lwi); } + if (exp->exp_failed) + GOTO(out, rc = -ENOTCONN); - ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*ioo), - lustre_swab_obd_ioobj); - if (ioo == NULL) { - CERROR("Missing/short ioobj\n"); - GOTO(out, rc = -EFAULT); - } + /* ost_body, ioobj & noibuf_remote are verified and swabbed in + * ost_rw_hpreq_check(). */ + body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body)); + LASSERT(body != NULL); + + ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, sizeof(*ioo)); + LASSERT(ioo != NULL); niocount = ioo->ioo_bufcnt; - if (niocount > PTLRPC_MAX_BRW_PAGES) { - DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)\n", - niocount); - GOTO(out, rc = -EFAULT); - } + remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, + niocount * sizeof(*remote_nb)); + LASSERT(remote_nb != NULL); - remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2, - niocount * sizeof(*remote_nb), - lustre_swab_niobuf_remote); - if (remote_nb == NULL) { - CERROR("Missing/short niobuf\n"); - GOTO(out, rc = -EFAULT); - } - if (lustre_msg_swabbed(req->rq_reqmsg)) { /* swab remaining niobufs */ - for (i = 1; i < niocount; i++) - lustre_swab_niobuf_remote (&remote_nb[i]); - } + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) + capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3); rc = lustre_pack_reply(req, 2, size, NULL); if (rc) @@ -686,47 +662,38 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) * ost_thread_init(). */ local_nb = ost_tls(req)->local; - pp_rnb = ost_tls(req)->remote; - - /* FIXME all niobuf splitting should be done in obdfilter if needed */ - /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */ - npages = get_per_page_niobufs(ioo, 1, remote_nb, niocount, &pp_rnb); - if (npages < 0) - GOTO(out, rc = npages); - - LASSERT(npages <= OST_THREAD_POOL_SIZE); - - ost_nio_pages_get(req, local_nb, npages); - - desc = ptlrpc_prep_bulk_exp(req, npages, - BULK_PUT_SOURCE, OST_BULK_PORTAL); - if (desc == NULL) - GOTO(out, rc = -ENOMEM); - rc = ost_brw_lock_get(LCK_PR, req->rq_export, ioo, pp_rnb, &lockh); + rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh); if (rc != 0) GOTO(out_bulk, rc); - /* + /* * If getting the lock took more time than * client was willing to wait, drop it. b=11330 */ - if (cfs_time_current_sec() > req->rq_arrival_time.tv_sec + obd_timeout || + if (cfs_time_current_sec() > req->rq_deadline || OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) { no_reply = 1; CERROR("Dropping timed-out read from %s because locking" - "object "LPX64" took %ld seconds.\n", + "object "LPX64" took %ld seconds (limit was %ld).\n", libcfs_id2str(req->rq_peer), ioo->ioo_id, - cfs_time_current_sec() - req->rq_arrival_time.tv_sec); - goto out_lock; + cfs_time_current_sec() - req->rq_arrival_time.tv_sec, + req->rq_deadline - req->rq_arrival_time.tv_sec); + GOTO(out_lock, rc = -ETIMEDOUT); } - rc = obd_preprw(OBD_BRW_READ, req->rq_export, &body->oa, 1, - ioo, npages, pp_rnb, local_nb, oti); + npages = OST_THREAD_POOL_SIZE; + rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, ioo, + remote_nb, &npages, local_nb, oti, capa); if (rc != 0) GOTO(out_lock, rc); - ost_prolong_locks(req->rq_export, ioo, pp_rnb, LCK_PW | LCK_PR); + desc = ptlrpc_prep_bulk_exp(req, npages, + BULK_PUT_SOURCE, OST_BULK_PORTAL); + if (desc == NULL) /* XXX: check all cleanup stuff */ + GOTO(out, rc = -ENOMEM); + + ost_rw_prolong_locks(req, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR); nob = 0; for (i = 0; i < npages; i++) { @@ -737,17 +704,15 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) break; } - LASSERTF(page_rc <= pp_rnb[i].len, "page_rc (%d) > " - "pp_rnb[%d].len (%d)\n", page_rc, i, pp_rnb[i].len); nob += page_rc; if (page_rc != 0) { /* some data! */ LASSERT (local_nb[i].page != NULL); ptlrpc_prep_bulk_page(desc, local_nb[i].page, - pp_rnb[i].offset & ~CFS_PAGE_MASK, + local_nb[i].offset & ~CFS_PAGE_MASK, page_rc); } - if (page_rc != pp_rnb[i].len) { /* short read */ + if (page_rc != local_nb[i].len) { /* short read */ /* All subsequent pages should be 0 */ while(++i < npages) LASSERT(local_nb[i].rc == 0); @@ -755,9 +720,14 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) } } - if (unlikely(body->oa.o_valid & OBD_MD_FLCKSUM)) { - body->oa.o_cksum = ost_checksum_bulk(desc); - body->oa.o_valid = OBD_MD_FLCKSUM; + 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); + 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); CDEBUG(D_PAGE,"checksum at read origin: %x\n",body->oa.o_cksum); } else { body->oa.o_valid = 0; @@ -767,21 +737,52 @@ 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) { - if (desc->bd_export->exp_failed) + /* Check if there is eviction in progress, and if so, wait for + * it to finish */ + if (unlikely(atomic_read(&exp->exp_obd-> + obd_evict_inprogress))) { + 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); + } + if (exp->exp_failed) rc = -ENOTCONN; - else - rc = ptlrpc_start_bulk_transfer(desc); + else { + rc = sptlrpc_svc_wrap_bulk(req, desc); + if (rc == 0) + rc = ptlrpc_start_bulk_transfer(desc); + } + if (rc == 0) { - lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ, - ost_bulk_timeout, desc); - rc = l_wait_event(desc->bd_waitq, - !ptlrpc_bulk_active(desc) || - desc->bd_export->exp_failed, &lwi); - LASSERT(rc == 0 || rc == -ETIMEDOUT); + time_t start = cfs_time_current_sec(); + do { + long timeoutl = req->rq_deadline - + cfs_time_current_sec(); + cfs_duration_t timeout = (timeoutl <= 0 || rc) ? + 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, &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"); + 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 (desc->bd_export->exp_failed) { + } else if (exp->exp_failed) { DEBUG_REQ(D_ERROR, req, "Eviction on bulk PUT"); rc = -ENOTCONN; ptlrpc_abort_bulk(desc); @@ -796,51 +797,47 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) rc = -ETIMEDOUT; } } else { - DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d\n", rc); + DEBUG_REQ(D_ERROR, req, "bulk PUT failed: rc %d", rc); } - comms_error = rc != 0; + no_reply = rc != 0; } /* Must commit after prep above in all cases */ - rc = obd_commitrw(OBD_BRW_READ, req->rq_export, &body->oa, 1, - ioo, npages, local_nb, oti, rc); - - ost_nio_pages_put(req, local_nb, npages); + rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo, + remote_nb, npages, local_nb, oti, rc); if (rc == 0) { repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repbody)); memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa)); + ost_drop_id(exp, &repbody->oa); } - out_lock: - ost_brw_lock_put(LCK_PR, ioo, pp_rnb, &lockh); - out_bulk: - ptlrpc_free_bulk(desc); - if (no_reply) - RETURN(rc); - out: +out_lock: + ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh); +out_bulk: + if (desc) + ptlrpc_free_bulk(desc); +out: LASSERT(rc <= 0); if (rc == 0) { req->rq_status = nob; + ptlrpc_lprocfs_brw(req, nob); target_committed_to_req(req); ptlrpc_reply(req); - } else if (!comms_error) { + } else if (!no_reply) { /* Only reply if there was no comms problem with bulk */ target_committed_to_req(req); req->rq_status = rc; ptlrpc_error(req); } else { - if (req->rq_reply_state != NULL) { - /* reply out callback would free */ - ptlrpc_rs_decref(req->rq_reply_state); - req->rq_reply_state = NULL; - } + /* reply out callback would free */ + ptlrpc_req_drop_rs(req); CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - " "client will retry\n", - req->rq_export->exp_obd->obd_name, - req->rq_export->exp_client_uuid.uuid, - req->rq_export->exp_connection->c_remote_uuid.uuid, + exp->exp_obd->obd_name, + exp->exp_client_uuid.uuid, + exp->exp_connection->c_remote_uuid.uuid, libcfs_id2str(req->rq_peer)); } @@ -849,84 +846,71 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) { - struct ptlrpc_bulk_desc *desc; + struct ptlrpc_bulk_desc *desc = NULL; + struct obd_export *exp = req->rq_export; struct niobuf_remote *remote_nb; - struct niobuf_remote *pp_rnb; struct niobuf_local *local_nb; struct obd_ioobj *ioo; struct ost_body *body, *repbody; struct l_wait_info lwi; struct lustre_handle lockh = {0}; + struct lustre_capa *capa = NULL; __u32 *rcs; - int size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) }; - int objcount, niocount, npages, comms_error = 0; - int rc, swab, i, j; - obd_count client_cksum, server_cksum = 0; - int no_reply = 0; + __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + int objcount, niocount, npages; + int rc, i, j; + obd_count client_cksum = 0, server_cksum = 0; + cksum_type_t cksum_type = OBD_CKSUM_CRC32; + int no_reply = 0; + __u32 o_uid = 0, o_gid = 0; ENTRY; + req->rq_bulk_write = 1; + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK)) GOTO(out, rc = -EIO); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2)) + GOTO(out, rc = -EFAULT); /* pause before transaction has been started */ - OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK | OBD_FAIL_ONCE, - (obd_timeout + 1) / 4); - - swab = lustre_msg_swabbed(req->rq_reqmsg); - body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), - lustre_swab_ost_body); - if (body == NULL) { - CERROR("Missing/short ost_body\n"); - GOTO(out, rc = -EFAULT); + 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(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, + !atomic_read(&exp->exp_obd->obd_evict_inprogress), + &lwi); } + if (exp->exp_failed) + GOTO(out, rc = -ENOTCONN); - LASSERT_REQSWAB(req, REQ_REC_OFF + 1); - objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) / - sizeof(*ioo); - if (objcount == 0) { - CERROR("Missing/short ioobj\n"); - GOTO(out, rc = -EFAULT); - } - if (objcount > 1) { - CERROR("too many ioobjs (%d)\n", objcount); - GOTO(out, rc = -EFAULT); - } + /* ost_body, ioobj & noibuf_remote are verified and swabbed in + * ost_rw_hpreq_check(). */ + body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body)); + LASSERT(body != NULL); + objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) / + sizeof(*ioo); ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, objcount * sizeof(*ioo)); - LASSERT (ioo != NULL); - for (niocount = i = 0; i < objcount; i++) { - if (swab) - lustre_swab_obd_ioobj(&ioo[i]); - if (ioo[i].ioo_bufcnt == 0) { - CERROR("ioo[%d] has zero bufcnt\n", i); - GOTO(out, rc = -EFAULT); - } + LASSERT(ioo != NULL); + for (niocount = i = 0; i < objcount; i++) niocount += ioo[i].ioo_bufcnt; - } - if (niocount > PTLRPC_MAX_BRW_PAGES) { - DEBUG_REQ(D_ERROR, req, "bulk has too many pages (%d)\n", - niocount); - GOTO(out, rc = -EFAULT); - } + remote_nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, + niocount * sizeof(*remote_nb)); + LASSERT(remote_nb != NULL); - remote_nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2, - niocount * sizeof(*remote_nb), - lustre_swab_niobuf_remote); - if (remote_nb == NULL) { - CERROR("Missing/short niobuf\n"); - GOTO(out, rc = -EFAULT); - } - if (swab) { /* swab the remaining niobufs */ - for (i = 1; i < niocount; i++) - lustre_swab_niobuf_remote (&remote_nb[i]); - } + if (body->oa.o_valid & OBD_MD_FLOSSCAPA) + capa = lustre_unpack_capa(req->rq_reqmsg, REQ_REC_OFF + 3); size[REPLY_REC_OFF + 1] = niocount * sizeof(*rcs); rc = lustre_pack_reply(req, 3, size, NULL); if (rc != 0) GOTO(out, rc); + OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, obd_fail_val); rcs = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF + 1, niocount * sizeof(*rcs)); @@ -935,90 +919,114 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) * ost_thread_init(). */ local_nb = ost_tls(req)->local; - pp_rnb = ost_tls(req)->remote; - - /* FIXME all niobuf splitting should be done in obdfilter if needed */ - /* CAVEAT EMPTOR this sets ioo->ioo_bufcnt to # pages */ - npages = get_per_page_niobufs(ioo, objcount,remote_nb,niocount,&pp_rnb); - if (npages < 0) - GOTO(out, rc = npages); - LASSERT(npages <= OST_THREAD_POOL_SIZE); - - ost_nio_pages_get(req, local_nb, npages); - - desc = ptlrpc_prep_bulk_exp(req, npages, - BULK_GET_SINK, OST_BULK_PORTAL); - if (desc == NULL) - GOTO(out, rc = -ENOMEM); - - rc = ost_brw_lock_get(LCK_PW, req->rq_export, ioo, pp_rnb, &lockh); + rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh); if (rc != 0) GOTO(out_bulk, rc); - /* + /* * If getting the lock took more time than * client was willing to wait, drop it. b=11330 */ - if (cfs_time_current_sec() > req->rq_arrival_time.tv_sec + obd_timeout || + if (cfs_time_current_sec() > req->rq_deadline || OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) { no_reply = 1; - CERROR("Dropping timed-out write from %s because locking" - "object "LPX64" took %ld seconds.\n", + CERROR("Dropping timed-out write from %s because locking " + "object "LPX64" took %ld seconds (limit was %ld).\n", libcfs_id2str(req->rq_peer), ioo->ioo_id, - cfs_time_current_sec() - req->rq_arrival_time.tv_sec); - goto out_lock; + cfs_time_current_sec() - req->rq_arrival_time.tv_sec, + req->rq_deadline - req->rq_arrival_time.tv_sec); + GOTO(out_lock, rc = -ETIMEDOUT); } - ost_prolong_locks(req->rq_export, ioo, pp_rnb, LCK_PW); + ost_rw_prolong_locks(req, ioo, remote_nb,&body->oa, LCK_PW); /* obd_preprw clobbers oa->valid, so save what we need */ - client_cksum = body->oa.o_valid & OBD_MD_FLCKSUM ? body->oa.o_cksum : 0; + if (body->oa.o_valid & OBD_MD_FLCKSUM) { + client_cksum = body->oa.o_cksum; + if (body->oa.o_valid & OBD_MD_FLFLAGS) + cksum_type = cksum_type_unpack(body->oa.o_flags); + } - rc = obd_preprw(OBD_BRW_WRITE, req->rq_export, &body->oa, objcount, - ioo, npages, pp_rnb, local_nb, oti); + /* Because we already sync grant info with client when reconnect, + * grant info will be cleared for resent req, then fed_grant and + * total_grant will not be modified in following preprw_write */ + if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) { + DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info"); + body->oa.o_valid &= ~OBD_MD_FLGRANT; + } + + if (exp_connect_rmtclient(exp)) { + o_uid = body->oa.o_uid; + o_gid = body->oa.o_gid; + } + npages = OST_THREAD_POOL_SIZE; + rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount, + ioo, remote_nb, &npages, local_nb, oti, capa); if (rc != 0) GOTO(out_lock, rc); + desc = ptlrpc_prep_bulk_exp(req, npages, + BULK_GET_SINK, OST_BULK_PORTAL); + if (desc == NULL) + GOTO(out, rc = -ENOMEM); + /* NB Having prepped, we must commit... */ for (i = 0; i < npages; i++) ptlrpc_prep_bulk_page(desc, local_nb[i].page, - pp_rnb[i].offset & ~CFS_PAGE_MASK, - pp_rnb[i].len); + local_nb[i].offset & ~CFS_PAGE_MASK, + local_nb[i].len); + + rc = sptlrpc_svc_prep_bulk(req, desc); + if (rc != 0) + GOTO(out_lock, rc); /* Check if client was evicted while we were doing i/o before touching network */ if (desc->bd_export->exp_failed) rc = -ENOTCONN; else - rc = ptlrpc_start_bulk_transfer (desc); + rc = ptlrpc_start_bulk_transfer(desc); if (rc == 0) { - lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 2, HZ, - ost_bulk_timeout, desc); - rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) || - desc->bd_export->exp_failed, &lwi); - LASSERT(rc == 0 || rc == -ETIMEDOUT); + time_t start = cfs_time_current_sec(); + do { + long timeoutl = req->rq_deadline - + cfs_time_current_sec(); + cfs_duration_t timeout = (timeoutl <= 0 || rc) ? + 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, &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"); + 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_success || - desc->bd_nob_transferred != desc->bd_nob) { - DEBUG_REQ(D_ERROR, req, "%s bulk GET %d(%d)", - desc->bd_success ? - "truncated" : "network error on", - desc->bd_nob_transferred, desc->bd_nob); + } 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\n", rc); + DEBUG_REQ(D_ERROR, req, "ptlrpc_bulk_get failed: rc %d", rc); } - comms_error = rc != 0; + no_reply = rc != 0; repbody = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repbody)); @@ -1026,9 +1034,10 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) if (unlikely(client_cksum != 0 && rc == 0)) { static int cksum_counter; - - server_cksum = ost_checksum_bulk(desc); - repbody->oa.o_valid |= OBD_MD_FLCKSUM; + 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); + server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type); repbody->oa.o_cksum = server_cksum; cksum_counter++; if (unlikely(client_cksum != server_cksum)) { @@ -1043,12 +1052,18 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) } /* Must commit after prep above in all cases */ - rc = obd_commitrw(OBD_BRW_WRITE, req->rq_export, &repbody->oa, - objcount, ioo, npages, local_nb, oti, rc); + rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo, + remote_nb, npages, local_nb, oti, rc); + if (exp_connect_rmtclient(exp)) { + repbody->oa.o_uid = o_uid; + repbody->oa.o_gid = o_gid; + } if (unlikely(client_cksum != server_cksum && rc == 0)) { - int new_cksum = ost_checksum_bulk(desc); + int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type); char *msg; + char *via; + char *router; if (new_cksum == server_cksum) msg = "changed in transit before arrival at OST"; @@ -1057,72 +1072,79 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) else msg = "changed in transit AND after initial checksum"; - LCONSOLE_ERROR("%s: BAD WRITE CHECKSUM: %s from %s inum " - LPU64"/"LPU64" object "LPU64"/"LPU64 - " extent ["LPU64"-"LPU64"]\n", - req->rq_export->exp_obd->obd_name, msg, - libcfs_id2str(req->rq_peer), - body->oa.o_valid & OBD_MD_FLFID ? + if (req->rq_peer.nid == desc->bd_sender) { + via = router = ""; + } else { + via = " via "; + router = libcfs_nid2str(desc->bd_sender); + } + + LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from " + "%s%s%s inum "LPU64"/"LPU64" 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_valid & OBD_MD_FLFID ? + body->oa.o_valid & OBD_MD_FLFID ? body->oa.o_generation :(__u64)0, - body->oa.o_id, - body->oa.o_valid & OBD_MD_FLGROUP ? + body->oa.o_id, + body->oa.o_valid & OBD_MD_FLGROUP ? body->oa.o_gr : (__u64)0, - pp_rnb[0].offset, - pp_rnb[npages-1].offset+pp_rnb[npages-1].len-1); + local_nb[0].offset, + local_nb[npages-1].offset + + local_nb[npages-1].len - 1 ); CERROR("client csum %x, original server csum %x, " "server csum now %x\n", client_cksum, server_cksum, new_cksum); } - ost_nio_pages_put(req, local_nb, npages); - if (rc == 0) { + int nob = 0; + /* set per-requested niobuf return codes */ for (i = j = 0; i < niocount; i++) { - int nob = remote_nb[i].len; + int len = remote_nb[i].len; + nob += len; rcs[i] = 0; do { LASSERT(j < npages); if (local_nb[j].rc < 0) rcs[i] = local_nb[j].rc; - nob -= pp_rnb[j].len; + len -= local_nb[j].len; j++; - } while (nob > 0); - LASSERT(nob == 0); + } while (len > 0); + LASSERT(len == 0); } LASSERT(j == npages); + ptlrpc_lprocfs_brw(req, nob); } - out_lock: - ost_brw_lock_put(LCK_PW, ioo, pp_rnb, &lockh); - out_bulk: - ptlrpc_free_bulk(desc); - if (no_reply) - RETURN(rc); - out: +out_lock: + ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh); +out_bulk: + if (desc) + ptlrpc_free_bulk(desc); +out: if (rc == 0) { oti_to_request(oti, req); target_committed_to_req(req); rc = ptlrpc_reply(req); - } else if (!comms_error) { + } else if (!no_reply) { /* Only reply if there was no comms problem with bulk */ target_committed_to_req(req); req->rq_status = rc; ptlrpc_error(req); } else { - if (req->rq_reply_state != NULL) { - /* reply out callback would free */ - ptlrpc_rs_decref(req->rq_reply_state); - req->rq_reply_state = NULL; - } + /* reply out callback would free */ + ptlrpc_req_drop_rs(req); CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - " "client will retry\n", - req->rq_export->exp_obd->obd_name, - req->rq_export->exp_client_uuid.uuid, - req->rq_export->exp_connection->c_remote_uuid.uuid, + exp->exp_obd->obd_name, + exp->exp_client_uuid.uuid, + exp->exp_connection->c_remote_uuid.uuid, libcfs_id2str(req->rq_peer)); } RETURN(rc); @@ -1149,7 +1171,7 @@ static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req) if (vallen) val = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, 0); - if (KEY_IS("evict_by_nid")) { + if (KEY_IS(KEY_EVICT_BY_NID)) { if (val && vallen) obd_export_evict_by_nid(exp->exp_obd, val); @@ -1164,51 +1186,62 @@ out: static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req) { - char *key; - int keylen, rc = 0; - int size[2] = { sizeof(struct ptlrpc_body), sizeof(obd_id) }; - obd_id *reply; + void *key, *reply; + int keylen, replylen, rc = 0; + struct req_capsule *pill = &req->rq_pill; ENTRY; - key = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, 1); + req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC); + + /* this common part for get_info rpc */ + key = req_capsule_client_get(pill, &RMF_SETINFO_KEY); if (key == NULL) { DEBUG_REQ(D_HA, req, "no get_info key"); RETURN(-EFAULT); } - keylen = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF); + keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT); - if (keylen < strlen("last_id") || memcmp(key, "last_id", 7) != 0) - RETURN(-EPROTO); + rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL); + if (rc) + RETURN(rc); - rc = lustre_pack_reply(req, 2, size, NULL); + req_capsule_set_size(pill, &RMF_GENERIC_DATA, + RCL_SERVER, replylen); + + rc = req_capsule_server_pack(pill); if (rc) RETURN(rc); - reply = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*reply)); - rc = obd_get_info(exp, keylen, key, size, reply); + reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA); + if (reply == NULL) + RETURN(-ENOMEM); + + /* call again to fill in the reply buffer */ + rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL); + lustre_msg_set_status(req->rq_repmsg, 0); RETURN(rc); } +#ifdef HAVE_QUOTA_SUPPORT static int ost_handle_quotactl(struct ptlrpc_request *req) { struct obd_quotactl *oqctl, *repoqc; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) }; + int rc; ENTRY; - oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl), - lustre_swab_obd_quotactl); + oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL); if (oqctl == NULL) GOTO(out, rc = -EPROTO); - rc = lustre_pack_reply(req, 2, size, NULL); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) GOTO(out, rc); - repoqc = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*repoqc)); - + repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL); req->rq_status = obd_quotactl(req->rq_export, oqctl); *repoqc = *oqctl; + out: RETURN(rc); } @@ -1219,21 +1252,224 @@ static int ost_handle_quotacheck(struct ptlrpc_request *req) int rc; ENTRY; - oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl), - lustre_swab_obd_quotactl); - if (oqctl == NULL) + oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL); + if (oqctl == NULL) RETURN(-EPROTO); - rc = lustre_pack_reply(req, 1, NULL, NULL); - if (rc) { - CERROR("ost: out of memory while packing quotacheck reply\n"); + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) RETURN(-ENOMEM); - } req->rq_status = obd_quotacheck(req->rq_export, oqctl); RETURN(0); } +static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req) +{ + struct quota_adjust_qunit *oqaq, *repoqa; + struct lustre_quota_ctxt *qctxt; + int rc; + ENTRY; + + qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt; + oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT); + if (oqaq == NULL) + GOTO(out, rc = -EPROTO); + + rc = req_capsule_server_pack(&req->rq_pill); + if (rc) + 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); + *repoqa = *oqaq; + + out: + RETURN(rc); +} +#endif + +static int ost_llog_handle_connect(struct obd_export *exp, + struct ptlrpc_request *req) +{ + struct llogd_conn_body *body; + int rc; + ENTRY; + + body = lustre_msg_buf(req->rq_reqmsg, 1, sizeof(*body)); + rc = obd_llog_connect(exp, body); + RETURN(rc); +} + +#define ost_init_sec_none(reply, exp) \ +do { \ + reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \ + OBD_CONNECT_RMT_CLIENT_FORCE | \ + OBD_CONNECT_OSS_CAPA); \ + spin_lock(&exp->exp_lock); \ + exp->exp_connect_flags = reply->ocd_connect_flags; \ + spin_unlock(&exp->exp_lock); \ +} while (0) + +static int ost_init_sec_level(struct ptlrpc_request *req) +{ + struct obd_export *exp = req->rq_export; + struct req_capsule *pill = &req->rq_pill; + struct obd_device *obd = exp->exp_obd; + struct filter_obd *filter = &obd->u.filter; + char *client = libcfs_nid2str(req->rq_peer.nid); + struct obd_connect_data *data, *reply; + int rc = 0, remote; + ENTRY; + + data = req_capsule_client_get(pill, &RMF_CONNECT_DATA); + reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA); + if (data == NULL || reply == NULL) + RETURN(-EFAULT); + + /* connection from MDT is always trusted */ + if (req->rq_auth_usr_mdt) { + ost_init_sec_none(reply, exp); + RETURN(0); + } + + /* no GSS support case */ + if (!req->rq_auth_gss) { + if (filter->fo_sec_level > LUSTRE_SEC_NONE) { + CWARN("client %s -> target %s does not user GSS, " + "can not run under security level %d.\n", + client, obd->obd_name, filter->fo_sec_level); + RETURN(-EACCES); + } else { + ost_init_sec_none(reply, exp); + RETURN(0); + } + } + + /* old version case */ + if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) || + !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) { + if (filter->fo_sec_level > LUSTRE_SEC_NONE) { + CWARN("client %s -> target %s uses old version, " + "can not run under security level %d.\n", + client, obd->obd_name, filter->fo_sec_level); + RETURN(-EACCES); + } else { + CWARN("client %s -> target %s uses old version, " + "run under security level %d.\n", + client, obd->obd_name, filter->fo_sec_level); + ost_init_sec_none(reply, exp); + RETURN(0); + } + } + + remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE; + if (remote) { + if (!req->rq_auth_remote) + CDEBUG(D_SEC, "client (local realm) %s -> target %s " + "asked to be remote.\n", client, obd->obd_name); + } else if (req->rq_auth_remote) { + remote = 1; + CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set " + "as remote by default.\n", client, obd->obd_name); + } + + if (remote) { + if (!filter->fo_fl_oss_capa) { + CDEBUG(D_SEC, "client %s -> target %s is set as remote," + " but OSS capabilities are not enabled: %d.\n", + client, obd->obd_name, filter->fo_fl_oss_capa); + RETURN(-EACCES); + } + } + + switch (filter->fo_sec_level) { + case LUSTRE_SEC_NONE: + if (!remote) { + ost_init_sec_none(reply, exp); + break; + } else { + CDEBUG(D_SEC, "client %s -> target %s is set as remote, " + "can not run under security level %d.\n", + client, obd->obd_name, filter->fo_sec_level); + RETURN(-EACCES); + } + case LUSTRE_SEC_REMOTE: + if (!remote) + ost_init_sec_none(reply, exp); + break; + case LUSTRE_SEC_ALL: + if (!remote) { + reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | + OBD_CONNECT_RMT_CLIENT_FORCE); + if (!filter->fo_fl_oss_capa) + reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA; + + spin_lock(&exp->exp_lock); + exp->exp_connect_flags = reply->ocd_connect_flags; + spin_unlock(&exp->exp_lock); + } + break; + default: + RETURN(-EINVAL); + } + + RETURN(rc); +} + +/* + * FIXME + * this should be done in filter_connect()/filter_reconnect(), but + * we can't obtain information like NID, which stored in incoming + * request, thus can't decide what flavor to use. so we do it here. + * + * This hack should be removed after the OST stack be rewritten, just + * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect(). + */ +static int ost_connect_check_sptlrpc(struct ptlrpc_request *req) +{ + struct obd_export *exp = req->rq_export; + struct filter_obd *filter = &exp->exp_obd->u.filter; + struct sptlrpc_flavor flvr; + int rc = 0; + + if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { + read_lock(&filter->fo_sptlrpc_lock); + sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset, + req->rq_sp_from, + req->rq_peer.nid, + &flvr); + read_unlock(&filter->fo_sptlrpc_lock); + + spin_lock(&exp->exp_lock); + + exp->exp_sp_peer = req->rq_sp_from; + exp->exp_flvr = flvr; + + if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY && + exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) { + CERROR("unauthorized rpc flavor %x from %s, " + "expect %x\n", req->rq_flvr.sf_rpc, + libcfs_nid2str(req->rq_peer.nid), + exp->exp_flvr.sf_rpc); + rc = -EACCES; + } + + spin_unlock(&exp->exp_lock); + } else { + if (exp->exp_sp_peer != req->rq_sp_from) { + CERROR("RPC source %s doesn't match %s\n", + sptlrpc_part2name(req->rq_sp_from), + sptlrpc_part2name(exp->exp_sp_peer)); + rc = -EACCES; + } else { + rc = sptlrpc_target_export_check(exp, req); + } + } + + return rc; +} + static int ost_filter_recovery_request(struct ptlrpc_request *req, struct obd_device *obd, int *process) { @@ -1257,10 +1493,8 @@ static int ost_filter_recovery_request(struct ptlrpc_request *req, default: DEBUG_REQ(D_ERROR, req, "not permitted during recovery"); - *process = 0; - /* XXX what should we set rq_status to here? */ - req->rq_status = -EAGAIN; - RETURN(ptlrpc_error(req)); + *process = -EAGAIN; + RETURN(0); } } @@ -1272,6 +1506,9 @@ int ost_msg_check_version(struct lustre_msg *msg) case OST_CONNECT: case OST_DISCONNECT: case OBD_PING: + case SEC_CTX_INIT: + case SEC_CTX_INIT_CONT: + case SEC_CTX_FINI: rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION); if (rc) CERROR("bad opc %u version %08x, expecting %08x\n", @@ -1290,8 +1527,11 @@ int ost_msg_check_version(struct lustre_msg *msg) case OST_SYNC: case OST_SET_INFO: case OST_GET_INFO: +#ifdef HAVE_QUOTA_SUPPORT case OST_QUOTACHECK: case OST_QUOTACTL: + case OST_QUOTA_ADJUST_QUNIT: +#endif rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION); if (rc) CERROR("bad opc %u version %08x, expecting %08x\n", @@ -1327,7 +1567,263 @@ int ost_msg_check_version(struct lustre_msg *msg) return rc; } -static int ost_handle(struct ptlrpc_request *req) +static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req, + struct ldlm_lock *lock) +{ + struct niobuf_remote *nb; + struct obd_ioobj *ioo; + struct ost_body *body; + int objcount, niocount; + int mode, opc, i; + __u64 start, end; + ENTRY; + + opc = lustre_msg_get_opc(req->rq_reqmsg); + LASSERT(opc == OST_READ || opc == OST_WRITE); + + /* As the request may be covered by several locks, do not look at + * o_handle, look at the RPC IO region. */ + body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), + lustre_swab_obdo); + objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) / + sizeof(*ioo); + ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, + objcount * sizeof(*ioo)); + LASSERT(ioo != NULL); + for (niocount = i = 0; i < objcount; i++) + niocount += ioo[i].ioo_bufcnt; + + nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, + niocount * sizeof(*nb)); + LASSERT(nb != NULL); + + mode = LCK_PW; + if (opc == OST_READ) + mode |= LCK_PR; + + start = nb[0].offset & CFS_PAGE_MASK; + end = (nb[ioo->ioo_bufcnt - 1].offset + + 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, + &lock->l_resource->lr_name)) + RETURN(0); + + if (!(lock->l_granted_mode & mode)) + RETURN(0); + + if (lock->l_policy_data.l_extent.end < start || + lock->l_policy_data.l_extent.start > end) + RETURN(0); + + RETURN(1); +} + +/** + * Swab buffers needed to call ost_rw_prolong_locks() and call it. + * Return the value from ost_rw_prolong_locks() which is non-zero if + * there is a cancelled lock which is waiting for this IO request. + */ +static int ost_rw_hpreq_check(struct ptlrpc_request *req) +{ + struct niobuf_remote *nb; + struct obd_ioobj *ioo; + struct ost_body *body; + int objcount, niocount; + int mode, opc, i; + ENTRY; + + opc = lustre_msg_get_opc(req->rq_reqmsg); + LASSERT(opc == OST_READ || opc == OST_WRITE); + + body = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF, sizeof(*body)); + LASSERT(body != NULL); + + objcount = lustre_msg_buflen(req->rq_reqmsg, REQ_REC_OFF + 1) / + sizeof(*ioo); + ioo = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 1, + objcount * sizeof(*ioo)); + LASSERT(ioo != NULL); + + for (niocount = i = 0; i < objcount; i++) + niocount += ioo[i].ioo_bufcnt; + nb = lustre_msg_buf(req->rq_reqmsg, REQ_REC_OFF + 2, + niocount * sizeof(*nb)); + LASSERT(nb != NULL); + LASSERT(niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK)); + + mode = LCK_PW; + if (opc == OST_READ) + mode |= LCK_PR; + RETURN(ost_rw_prolong_locks(req, ioo, nb, &body->oa, mode)); +} + +static int ost_punch_prolong_locks(struct ptlrpc_request *req, struct obdo *oa) +{ + struct ldlm_res_id res_id = { .name = { oa->o_id } }; + struct ost_prolong_data opd = { 0 }; + __u64 start, end; + ENTRY; + + start = oa->o_size; + end = start + oa->o_blocks; + + opd.opd_mode = LCK_PW; + opd.opd_exp = req->rq_export; + opd.opd_policy.l_extent.start = start & CFS_PAGE_MASK; + if (oa->o_blocks == OBD_OBJECT_EOF || end < start) + opd.opd_policy.l_extent.end = OBD_OBJECT_EOF; + else + opd.opd_policy.l_extent.end = end | ~CFS_PAGE_MASK; + + /* prolong locks for the current service time of the corresponding + * portal (= OST_IO_PORTAL) */ + opd.opd_timeout = AT_OFF ? obd_timeout / 2: + max(at_est2timeout(at_get(&req->rq_rqbd-> + rqbd_service->srv_at_estimate)), ldlm_timeout); + + CDEBUG(D_DLMTRACE,"refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n", + res_id.name[0], res_id.name[1], opd.opd_policy.l_extent.start, + opd.opd_policy.l_extent.end); + + opd.opd_oa = oa; + ldlm_resource_iterate(req->rq_export->exp_obd->obd_namespace, &res_id, + ost_prolong_locks_iter, &opd); + RETURN(opd.opd_lock_match); +} + +static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req, + struct ldlm_lock *lock) +{ + struct ost_body *body; + ENTRY; + + body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), + lustre_swab_obdo); + LASSERT(body != NULL); + + if (body->oa.o_valid & OBD_MD_FLHANDLE && + body->oa.o_handle.cookie == lock->l_handle.h_cookie) + RETURN(1); + RETURN(0); +} + +static int ost_punch_hpreq_check(struct ptlrpc_request *req) +{ + struct ost_body *body = lustre_msg_buf(req->rq_reqmsg, + REQ_REC_OFF, sizeof(*body)); + LASSERT(body != NULL); + LASSERT(!(body->oa.o_valid & OBD_MD_FLFLAGS) || + !(body->oa.o_flags & OBD_FL_TRUNCLOCK)); + + RETURN(ost_punch_prolong_locks(req, &body->oa)); +} + +struct ptlrpc_hpreq_ops ost_hpreq_rw = { + .hpreq_lock_match = ost_rw_hpreq_lock_match, + .hpreq_check = ost_rw_hpreq_check, +}; + +struct ptlrpc_hpreq_ops ost_hpreq_punch = { + .hpreq_lock_match = ost_punch_hpreq_lock_match, + .hpreq_check = ost_punch_hpreq_check, +}; + +/** Assign high priority operations to the request if needed. */ +static int ost_hpreq_handler(struct ptlrpc_request *req) +{ + ENTRY; + if (req->rq_export) { + int opc = lustre_msg_get_opc(req->rq_reqmsg); + struct ost_body *body; + + if (opc == OST_READ || opc == OST_WRITE) { + struct niobuf_remote *nb; + struct obd_ioobj *ioo; + int objcount, niocount; + int swab, i; + + body = lustre_swab_reqbuf(req, REQ_REC_OFF, + sizeof(*body), + lustre_swab_obdo); + if (!body) { + CERROR("Missing/short ost_body\n"); + RETURN(-EFAULT); + } + objcount = lustre_msg_buflen(req->rq_reqmsg, + REQ_REC_OFF + 1) / + sizeof(*ioo); + if (objcount == 0) { + CERROR("Missing/short ioobj\n"); + RETURN(-EFAULT); + } + if (objcount > 1) { + CERROR("too many ioobjs (%d)\n", objcount); + RETURN(-EFAULT); + } + + swab = !lustre_req_swabbed(req, REQ_REC_OFF + 1) && + lustre_msg_swabbed(req->rq_reqmsg); + ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, + objcount * sizeof(*ioo), + lustre_swab_obd_ioobj); + if (!ioo) { + CERROR("Missing/short ioobj\n"); + RETURN(-EFAULT); + } + for (niocount = i = 0; i < objcount; i++) { + if (i > 0 && swab) + lustre_swab_obd_ioobj(&ioo[i]); + if (ioo[i].ioo_bufcnt == 0) { + CERROR("ioo[%d] has zero bufcnt\n", i); + RETURN(-EFAULT); + } + niocount += ioo[i].ioo_bufcnt; + } + if (niocount > PTLRPC_MAX_BRW_PAGES) { + DEBUG_REQ(D_ERROR, req, "bulk has too many " + "pages (%d)", niocount); + RETURN(-EFAULT); + } + + swab = !lustre_req_swabbed(req, REQ_REC_OFF + 2) && + lustre_msg_swabbed(req->rq_reqmsg); + nb = lustre_swab_reqbuf(req, REQ_REC_OFF + 2, + niocount * sizeof(*nb), + lustre_swab_niobuf_remote); + if (!nb) { + CERROR("Missing/short niobuf\n"); + RETURN(-EFAULT); + } + + if (swab) { + /* swab remaining niobufs */ + for (i = 1; i < niocount; i++) + lustre_swab_niobuf_remote(&nb[i]); + } + + if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK)) + req->rq_ops = &ost_hpreq_rw; + } else if (opc == OST_PUNCH) { + body = lustre_swab_reqbuf(req, REQ_REC_OFF, + sizeof(*body), + lustre_swab_obdo); + if (!body) { + CERROR("Missing/short ost_body\n"); + RETURN(-EFAULT); + } + + if (!(body->oa.o_valid & OBD_MD_FLFLAGS) || + !(body->oa.o_flags & OBD_FL_TRUNCLOCK)) + req->rq_ops = &ost_hpreq_punch; + } + } + RETURN(0); +} + +/* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */ +int ost_handle(struct ptlrpc_request *req) { struct obd_trans_info trans_info = { 0, }; struct obd_trans_info *oti = &trans_info; @@ -1336,9 +1832,20 @@ static int ost_handle(struct ptlrpc_request *req) ENTRY; LASSERT(current->journal_info == NULL); + + /* primordial rpcs don't affect server recovery */ + switch (lustre_msg_get_opc(req->rq_reqmsg)) { + case SEC_CTX_INIT: + case SEC_CTX_INIT_CONT: + case SEC_CTX_FINI: + GOTO(out, rc = 0); + } + + req_capsule_init(&req->rq_pill, req, RCL_SERVER); + /* XXX identical to MDS */ if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) { - int abort_recovery, recovering; + int recovering; if (req->rq_export == NULL) { CDEBUG(D_HA,"operation %d on unconnected OST from %s\n", @@ -1352,23 +1859,22 @@ static int ost_handle(struct ptlrpc_request *req) /* Check for aborted recovery. */ spin_lock_bh(&obd->obd_processing_task_lock); - abort_recovery = obd->obd_abort_recovery; recovering = obd->obd_recovering; spin_unlock_bh(&obd->obd_processing_task_lock); - if (abort_recovery) { - target_abort_recovery(obd); - } else if (recovering) { + if (recovering) { rc = ost_filter_recovery_request(req, obd, &should_process); if (rc || !should_process) RETURN(rc); + else if (should_process < 0) { + req->rq_status = should_process; + rc = ptlrpc_error(req); + RETURN(rc); + } } } oti_init(oti, req); - rc = ost_msg_check_version(req->rq_reqmsg); - if (rc) - RETURN(rc); rc = ost_msg_check_version(req->rq_reqmsg); if (rc) @@ -1377,49 +1883,71 @@ static int ost_handle(struct ptlrpc_request *req) switch (lustre_msg_get_opc(req->rq_reqmsg)) { case OST_CONNECT: { CDEBUG(D_INODE, "connect\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0); - rc = target_handle_connect(req, ost_handle); - if (!rc) - obd = req->rq_export->exp_obd; + req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET)) + RETURN(0); + rc = target_handle_connect(req); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)) + RETURN(0); + if (!rc) { + rc = ost_init_sec_level(req); + if (!rc) + rc = ost_connect_check_sptlrpc(req); + } break; } case OST_DISCONNECT: CDEBUG(D_INODE, "disconnect\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0); + req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET)) + RETURN(0); rc = target_handle_disconnect(req); break; case OST_CREATE: CDEBUG(D_INODE, "create\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_ENOSPC)) + if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET)) + RETURN(0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC)) GOTO(out, rc = -ENOSPC); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS)) + if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) GOTO(out, rc = -EROFS); rc = ost_create(req->rq_export, req, oti); break; case OST_DESTROY: CDEBUG(D_INODE, "destroy\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS)) + if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET)) + RETURN(0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) GOTO(out, rc = -EROFS); rc = ost_destroy(req->rq_export, req, oti); break; case OST_GETATTR: CDEBUG(D_INODE, "getattr\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET)) + RETURN(0); rc = ost_getattr(req->rq_export, req); break; case OST_SETATTR: CDEBUG(D_INODE, "setattr\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET)) + RETURN(0); rc = ost_setattr(req->rq_export, req, oti); break; case OST_WRITE: CDEBUG(D_INODE, "write\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_ENOSPC)) + /* req->rq_request_portal would be nice, if it was set */ + if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){ + CERROR("%s: deny write request from %s to portal %u\n", + req->rq_export->exp_obd->obd_name, + obd_export_nid2str(req->rq_export), + req->rq_rqbd->rqbd_service->srv_req_portal); + GOTO(out, rc = -EPROTO); + } + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET)) + RETURN(0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC)) GOTO(out, rc = -ENOSPC); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS)) + if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) GOTO(out, rc = -EROFS); rc = ost_brw_write(req, oti); LASSERT(current->journal_info == NULL); @@ -1427,71 +1955,103 @@ static int ost_handle(struct ptlrpc_request *req) RETURN(rc); case OST_READ: CDEBUG(D_INODE, "read\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0); + /* req->rq_request_portal would be nice, if it was set */ + if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){ + CERROR("%s: deny read request from %s to portal %u\n", + req->rq_export->exp_obd->obd_name, + obd_export_nid2str(req->rq_export), + req->rq_rqbd->rqbd_service->srv_req_portal); + GOTO(out, rc = -EPROTO); + } + if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET)) + RETURN(0); rc = ost_brw_read(req, oti); LASSERT(current->journal_info == NULL); /* ost_brw_read sends its own replies */ RETURN(rc); case OST_PUNCH: CDEBUG(D_INODE, "punch\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0); - if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_OST_EROFS)) + if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET)) + RETURN(0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) GOTO(out, rc = -EROFS); rc = ost_punch(req->rq_export, req, oti); break; case OST_STATFS: CDEBUG(D_INODE, "statfs\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET)) + RETURN(0); rc = ost_statfs(req); break; case OST_SYNC: CDEBUG(D_INODE, "sync\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_SYNC_NET, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET)) + RETURN(0); rc = ost_sync(req->rq_export, req); break; case OST_SET_INFO: DEBUG_REQ(D_INODE, req, "set_info"); + req_capsule_set(&req->rq_pill, &RQF_OST_SET_INFO); rc = ost_set_info(req->rq_export, req); break; case OST_GET_INFO: DEBUG_REQ(D_INODE, req, "get_info"); rc = ost_get_info(req->rq_export, req); break; +#ifdef HAVE_QUOTA_SUPPORT case OST_QUOTACHECK: CDEBUG(D_INODE, "quotacheck\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_QUOTACHECK_NET, 0); + req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET)) + RETURN(0); rc = ost_handle_quotacheck(req); break; case OST_QUOTACTL: CDEBUG(D_INODE, "quotactl\n"); - OBD_FAIL_RETURN(OBD_FAIL_OST_QUOTACTL_NET, 0); + req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET)) + RETURN(0); rc = ost_handle_quotactl(req); break; + case OST_QUOTA_ADJUST_QUNIT: + CDEBUG(D_INODE, "quota_adjust_qunit\n"); + req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT); + rc = ost_handle_quota_adjust_qunit(req); + break; +#endif case OBD_PING: DEBUG_REQ(D_INODE, req, "ping"); + req_capsule_set(&req->rq_pill, &RQF_OBD_PING); rc = target_handle_ping(req); break; /* FIXME - just reply status */ case LLOG_ORIGIN_CONNECT: - DEBUG_REQ(D_INODE, req, "log connect\n"); - rc = llog_handle_connect(req); + DEBUG_REQ(D_INODE, req, "log connect"); + req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT); + rc = ost_llog_handle_connect(req->rq_export, req); req->rq_status = rc; - rc = lustre_pack_reply(req, 1, NULL, NULL); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) RETURN(rc); RETURN(ptlrpc_reply(req)); case OBD_LOG_CANCEL: CDEBUG(D_INODE, "log cancel\n"); - OBD_FAIL_RETURN(OBD_FAIL_OBD_LOG_CANCEL_NET, 0); + req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET)) + RETURN(0); rc = llog_origin_handle_cancel(req); + if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP)) + RETURN(0); req->rq_status = rc; - rc = lustre_pack_reply(req, 1, NULL, NULL); + rc = req_capsule_server_pack(&req->rq_pill); if (rc) RETURN(rc); RETURN(ptlrpc_reply(req)); case LDLM_ENQUEUE: CDEBUG(D_INODE, "enqueue\n"); - OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0); + req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE)) + RETURN(0); rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast, ldlm_server_blocking_ast, ldlm_server_glimpse_ast); @@ -1499,12 +2059,16 @@ static int ost_handle(struct ptlrpc_request *req) break; case LDLM_CONVERT: CDEBUG(D_INODE, "convert\n"); - OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0); + req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT)) + RETURN(0); rc = ldlm_handle_convert(req); break; case LDLM_CANCEL: CDEBUG(D_INODE, "cancel\n"); - OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0); + req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL)) + RETURN(0); rc = ldlm_handle_cancel(req); break; case LDLM_BL_CALLBACK: @@ -1528,28 +2092,18 @@ static int ost_handle(struct ptlrpc_request *req) target_committed_to_req(req); out: - if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_LAST_REPLAY) { - if (obd && obd->obd_recovering) { - DEBUG_REQ(D_HA, req, "LAST_REPLAY, queuing reply"); - return target_queue_final_reply(req, rc); - } - /* Lost a race with recovery; let the error path DTRT. */ - rc = req->rq_status = -ENOTCONN; - } - if (!rc) oti_to_request(oti, req); target_send_reply(req, rc, fail); return 0; } - +EXPORT_SYMBOL(ost_handle); /* * free per-thread pool created by ost_thread_init(). */ static void ost_thread_done(struct ptlrpc_thread *thread) { - int i; struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local * Storage */ @@ -1563,10 +2117,6 @@ static void ost_thread_done(struct ptlrpc_thread *thread) */ tls = thread->t_data; if (tls != NULL) { - for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) { - if (tls->page[i] != NULL) - __free_page(tls->page[i]); - } OBD_FREE_PTR(tls); thread->t_data = NULL; } @@ -1578,43 +2128,32 @@ static void ost_thread_done(struct ptlrpc_thread *thread) */ static int ost_thread_init(struct ptlrpc_thread *thread) { - int result; - int i; struct ost_thread_local_cache *tls; ENTRY; LASSERT(thread != NULL); LASSERT(thread->t_data == NULL); - LASSERT(thread->t_id < OSS_THREADS_MAX); + LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id); OBD_ALLOC_PTR(tls); - if (tls != NULL) { - result = 0; - thread->t_data = tls; - /* - * populate pool - */ - for (i = 0; i < OST_THREAD_POOL_SIZE; ++ i) { - tls->page[i] = alloc_page(OST_THREAD_POOL_GFP); - if (tls->page[i] == NULL) { - ost_thread_done(thread); - result = -ENOMEM; - break; - } - } - } else - result = -ENOMEM; - RETURN(result); + if (tls == NULL) + RETURN(-ENOMEM); + thread->t_data = tls; + RETURN(0); } +#define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000) + /* Sigh - really, this is an OSS, the _server_, not the _target_ */ -static int ost_setup(struct obd_device *obd, obd_count len, void *buf) +static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) { struct ost_obd *ost = &obd->u.ost; struct lprocfs_static_vars lvars; - int oss_min_threads = OSS_THREADS_AUTO_MIN; - int oss_max_threads = OSS_THREADS_AUTO_MAX; + int oss_min_threads; + int oss_max_threads; + int oss_min_create_threads; + int oss_max_create_threads; int rc; ENTRY; @@ -1622,33 +2161,40 @@ static int ost_setup(struct obd_device *obd, obd_count len, void *buf) if (rc) RETURN(rc); - rc = llog_start_commit_thread(); - if (rc < 0) - RETURN(rc); - - lprocfs_init_vars(ost, &lvars); + lprocfs_ost_init_vars(&lvars); lprocfs_obd_setup(obd, lvars.obd_vars); sema_init(&ost->ost_health_sem, 1); if (oss_num_threads) { /* If oss_num_threads is set, it is the min and the max. */ - if (oss_num_threads > OSS_THREADS_MAX) + if (oss_num_threads > OSS_THREADS_MAX) oss_num_threads = OSS_THREADS_MAX; if (oss_num_threads < OSS_THREADS_MIN) oss_num_threads = OSS_THREADS_MIN; oss_max_threads = oss_min_threads = oss_num_threads; + } else { + /* Base min threads on memory and cpus */ + oss_min_threads = num_possible_cpus() * num_physpages >> + (27 - CFS_PAGE_SHIFT); + if (oss_min_threads < OSS_THREADS_MIN) + oss_min_threads = OSS_THREADS_MIN; + /* Insure a 4x range for dynamic threads */ + if (oss_min_threads > OSS_THREADS_MAX / 4) + oss_min_threads = OSS_THREADS_MAX / 4; + oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1); } ost->ost_service = ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE, OST_MAXREPSIZE, OST_REQUEST_PORTAL, - OSC_REPLY_PORTAL, - obd_timeout * 1000, ost_handle, LUSTRE_OSS_NAME, - obd->obd_proc_entry, ost_print_req, - oss_min_threads, oss_max_threads, "ll_ost"); + OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR, + ost_handle, LUSTRE_OSS_NAME, + obd->obd_proc_entry, target_print_req, + oss_min_threads, oss_max_threads, + "ll_ost", LCT_DT_THREAD, NULL); if (ost->ost_service == NULL) { - CERROR("failed to start OST service\n"); + CERROR("failed to start service\n"); GOTO(out_lprocfs, rc = -ENOMEM); } @@ -1656,13 +2202,26 @@ static int ost_setup(struct obd_device *obd, obd_count len, void *buf) if (rc) GOTO(out_service, rc = -EINVAL); + if (oss_num_create_threads) { + if (oss_num_create_threads > OSS_MAX_CREATE_THREADS) + oss_num_create_threads = OSS_MAX_CREATE_THREADS; + if (oss_num_create_threads < OSS_MIN_CREATE_THREADS) + oss_num_create_threads = OSS_MIN_CREATE_THREADS; + oss_min_create_threads = oss_max_create_threads = + oss_num_create_threads; + } else { + oss_min_create_threads = OSS_MIN_CREATE_THREADS; + oss_max_create_threads = OSS_MAX_CREATE_THREADS; + } + ost->ost_create_service = ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE, OST_MAXREPSIZE, OST_CREATE_PORTAL, - OSC_REPLY_PORTAL, - obd_timeout * 1000, ost_handle, "ost_create", - obd->obd_proc_entry, ost_print_req, - 1, 1, "ll_ost_creat"); + OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR, + ost_handle, "ost_create", + obd->obd_proc_entry, target_print_req, + oss_min_create_threads, oss_max_create_threads, + "ll_ost_creat", LCT_DT_THREAD, NULL); if (ost->ost_create_service == NULL) { CERROR("failed to start OST create service\n"); GOTO(out_service, rc = -ENOMEM); @@ -1675,10 +2234,11 @@ static int ost_setup(struct obd_device *obd, obd_count len, void *buf) ost->ost_io_service = ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE, OST_MAXREPSIZE, OST_IO_PORTAL, - OSC_REPLY_PORTAL, - obd_timeout * 1000, ost_handle, "ost_io", - obd->obd_proc_entry, ost_print_req, - oss_min_threads, oss_max_threads, "ll_ost_io"); + OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR, + ost_handle, "ost_io", + obd->obd_proc_entry, target_print_req, + oss_min_threads, oss_max_threads, + "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler); if (ost->ost_io_service == NULL) { CERROR("failed to start OST I/O service\n"); GOTO(out_create, rc = -ENOMEM); @@ -1778,9 +2338,17 @@ static int __init ost_init(void) int rc; ENTRY; - lprocfs_init_vars(ost, &lvars); - rc = class_register_type(&ost_obd_ops, lvars.module_vars, - LUSTRE_OSS_NAME); + lprocfs_ost_init_vars(&lvars); + rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars, + LUSTRE_OSS_NAME, NULL); + + if (ost_num_threads != 0 && oss_num_threads == 0) { + LCONSOLE_INFO("ost_num_threads module parameter is deprecated, " + "use oss_num_threads instead or unset both for " + "dynamic thread startup\n"); + oss_num_threads = ost_num_threads; + } + RETURN(rc); } @@ -1789,7 +2357,7 @@ static void /*__exit*/ ost_exit(void) class_unregister_type(LUSTRE_OSS_NAME); } -MODULE_AUTHOR("Cluster File Systems, Inc. "); +MODULE_AUTHOR("Sun Microsystems, Inc. "); MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01"); MODULE_LICENSE("GPL");