X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fost%2Fost_handler.c;h=abb4f66b89c6f9fe0ebdb6e13e3df164d134a43f;hp=45c510273802d4ef9fd794580d57a616b1ea2fe3;hb=62499626b81f83f9e2ceceaa11d9b33861581cb6;hpb=de3c3fdf2597ba110a5247e271470b5dd626df75 diff --git a/lustre/ost/ost_handler.c b/lustre/ost/ost_handler.c index 45c5102..abb4f66 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 @@ -47,7 +53,6 @@ #include #include #include -#include #include #include #include "ost_internal.h" @@ -81,7 +86,7 @@ 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); } } @@ -89,7 +94,8 @@ 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) }; + int rc; ENTRY; body = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*body), @@ -108,13 +114,13 @@ static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req, RETURN (-EFAULT); ldlm_request_cancel(req, dlm, 0); } - + 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)); @@ -126,7 +132,8 @@ 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), @@ -153,7 +160,8 @@ static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req) 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); @@ -176,7 +184,8 @@ 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), @@ -191,7 +200,7 @@ 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); @@ -205,13 +214,14 @@ 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, 0, oa->o_gr, 0} }; + 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) || @@ -243,7 +253,8 @@ static int ost_punch_lock_get(struct obd_export *exp, struct obdo *oa, 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)); } /* @@ -264,7 +275,8 @@ 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; @@ -282,7 +294,7 @@ 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) @@ -315,7 +327,8 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req) { struct ost_body *body, *repbody; struct lustre_capa *capa = NULL; - 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), @@ -342,7 +355,8 @@ 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; @@ -376,101 +390,6 @@ 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) -{ - /* 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, int opc, cksum_type_t cksum_type) { @@ -493,75 +412,29 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc, /* corrupt the data after we compute the checksum, to * simulate an OST->client data error */ if (i == 0 && opc == OST_READ && - OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) + 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, 0, - obj->ioo_gr, 0} }; + 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)); @@ -580,7 +453,8 @@ static int ost_brw_lock_get(int mode, struct obd_export *exp, 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, @@ -642,15 +516,16 @@ static int ost_prolong_locks_iter(struct ldlm_lock *lock, void *data) } static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj, - struct niobuf_remote *nb, ldlm_mode_t mode) + struct niobuf_remote *nb, struct obdo *oa, + ldlm_mode_t mode) { - struct ldlm_res_id res_id = { .name = { obj->ioo_id, 0, - obj->ioo_gr, 0} }; + struct ldlm_res_id res_id; int nrbufs = obj->ioo_bufcnt; struct ost_prolong_data opd; - ENTRY; + osc_build_res_name(obj->ioo_id, obj->ioo_gr, &res_id); + opd.opd_mode = mode; opd.opd_exp = exp; opd.opd_policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK; @@ -660,24 +535,38 @@ static void ost_prolong_locks(struct obd_export *exp, struct obd_ioobj *obj, 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); + + 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); + LDLM_LOCK_PUT(lock); + EXIT; + return; + } + } + ldlm_resource_iterate(exp->exp_obd->obd_namespace, &res_id, ost_prolong_locks_iter, &opd); + + EXIT; } 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 niocount, npages, nob = 0, rc, i; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + int objcount, niocount, npages, nob = 0, rc, i; int no_reply = 0; ENTRY; @@ -706,6 +595,17 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) GOTO(out, rc = -EFAULT); } + 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); + } + ioo = lustre_swab_reqbuf(req, REQ_REC_OFF + 1, sizeof(*ioo), lustre_swab_obd_ioobj); if (ioo == NULL) { @@ -744,28 +644,12 @@ 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, exp, 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 */ @@ -780,12 +664,18 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) GOTO(out_lock, rc = -ETIMEDOUT); } - rc = obd_preprw(OBD_BRW_READ, exp, &body->oa, 1, - ioo, npages, pp_rnb, local_nb, oti, capa); + 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(exp, 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_prolong_locks(exp, ioo, remote_nb, &body->oa, LCK_PW | LCK_PR); nob = 0; for (i = 0; i < npages; i++) { @@ -796,17 +686,15 @@ 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); @@ -897,10 +785,8 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) } /* Must commit after prep above in all cases */ - rc = obd_commitrw(OBD_BRW_READ, exp, &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, @@ -909,13 +795,15 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) } out_lock: - ost_brw_lock_put(LCK_PR, ioo, pp_rnb, &lockh); + ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh); out_bulk: - ptlrpc_free_bulk(desc); + 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 (!no_reply) { @@ -939,10 +827,9 @@ out: 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; @@ -950,12 +837,12 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) struct lustre_handle lockh = {0}; struct lustre_capa *capa = NULL; __u32 *rcs; - int size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) }; + __u32 size[3] = { sizeof(struct ptlrpc_body), sizeof(*body) }; int objcount, niocount, npages; int rc, swab, 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; ENTRY; req->rq_bulk_write = 1; @@ -1046,28 +933,12 @@ 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, exp, 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 */ @@ -1082,7 +953,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) GOTO(out_lock, rc = -ETIMEDOUT); } - ost_prolong_locks(exp, ioo, pp_rnb, LCK_PW); + ost_prolong_locks(exp, ioo, remote_nb,&body->oa, LCK_PW); /* obd_preprw clobbers oa->valid, so save what we need */ if (body->oa.o_valid & OBD_MD_FLCKSUM) { @@ -1090,26 +961,32 @@ 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); } - + /* 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 */ + * 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; } + npages = OST_THREAD_POOL_SIZE; rc = obd_preprw(OBD_BRW_WRITE, exp, &body->oa, objcount, - ioo, npages, pp_rnb, local_nb, oti, capa); + 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); /* Check if client was evicted while we were doing i/o before touching network */ @@ -1186,8 +1063,8 @@ 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, exp, &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 (unlikely(client_cksum != server_cksum && rc == 0)) { int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type); @@ -1208,7 +1085,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) 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", @@ -1222,38 +1099,41 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) 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); + ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh); out_bulk: - ptlrpc_free_bulk(desc); + if (desc) + ptlrpc_free_bulk(desc); out: if (rc == 0) { oti_to_request(oti, req); @@ -1328,7 +1208,7 @@ static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req) } keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT); - rc = obd_get_info(exp, keylen, key, &replylen, NULL); + rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL); if (rc) RETURN(rc); @@ -1344,7 +1224,7 @@ static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req) RETURN(-ENOMEM); /* call again to fill in the reply buffer */ - rc = obd_get_info(exp, keylen, key, &replylen, reply); + rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL); lustre_msg_set_status(req->rq_repmsg, 0); RETURN(rc); @@ -1353,7 +1233,8 @@ static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req) static int ost_handle_quotactl(struct ptlrpc_request *req) { struct obd_quotactl *oqctl, *repoqc; - int rc, size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) }; + __u32 size[2] = { sizeof(struct ptlrpc_body), sizeof(*repoqc) }; + int rc; ENTRY; oqctl = lustre_swab_reqbuf(req, REQ_REC_OFF, sizeof(*oqctl), @@ -1591,7 +1472,7 @@ int ost_handle(struct ptlrpc_request *req) } oti_init(oti, req); - + rc = ost_msg_check_version(req->rq_reqmsg); if (rc) RETURN(rc); @@ -1752,6 +1633,8 @@ int ost_handle(struct ptlrpc_request *req) 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 = req_capsule_server_pack(&req->rq_pill); if (rc) @@ -1814,7 +1697,6 @@ EXPORT_SYMBOL(ost_handle); */ static void ost_thread_done(struct ptlrpc_thread *thread) { - int i; struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local * Storage */ @@ -1828,10 +1710,6 @@ 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) - OBD_PAGE_FREE(tls->page[i]); - } OBD_FREE_PTR(tls); thread->t_data = NULL; } @@ -1843,8 +1721,6 @@ 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; @@ -1854,23 +1730,10 @@ static int ost_thread_init(struct ptlrpc_thread *thread) 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) { - OBD_PAGE_ALLOC(tls->page[i], 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) @@ -1898,19 +1761,19 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) 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 >> + 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) + 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); } @@ -2087,7 +1950,7 @@ 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");