X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fost%2Fost_handler.c;h=d8fcccb88772b069419e8abf32b1e4b871ff0225;hp=ec5675952b2ac1e73918f6d140ff921c85d18ec6;hb=a6db2409b0282b0a0e44d5d4e6eeb154b5c62dff;hpb=d800fc41a1abdaf7aaf6c0e3e7ddcdec489985a8 diff --git a/lustre/ost/ost_handler.c b/lustre/ost/ost_handler.c index ec56759..d8fcccb 100644 --- a/lustre/ost/ost_handler.c +++ b/lustre/ost/ost_handler.c @@ -27,7 +27,7 @@ * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Whamcloud, Inc. + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -51,8 +51,8 @@ #include #include #include -#include #include "ost_internal.h" +#include static int oss_num_threads; CFS_MODULE_PARM(oss_num_threads, "i", int, 0444, @@ -74,6 +74,15 @@ static char *oss_io_cpts; CFS_MODULE_PARM(oss_io_cpts, "s", charp, 0444, "CPU partitions OSS IO threads should run on"); +/* + * this page is allocated statically when module is initializing + * it is used to simulate data corruptions, see ost_checksum_bulk() + * for details. as the original pages provided by the layers below + * can be remain in the internal cache, we do not want to modify + * them. + */ +static struct page *ost_page_to_corrupt = NULL; + /** * Do not return server-side uid/gid to remote client */ @@ -203,7 +212,7 @@ static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req, */ static int ost_lock_get(struct obd_export *exp, struct obdo *oa, __u64 start, __u64 count, struct lustre_handle *lh, - int mode, int flags) + int mode, __u64 flags) { struct ldlm_res_id res_id; ldlm_policy_data_t policy; @@ -237,7 +246,8 @@ static int ost_lock_get(struct obd_export *exp, struct obdo *oa, 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, LVB_T_NONE, + NULL, lh)); } /* Helper function: release lock, if any. */ @@ -322,6 +332,9 @@ static int ost_statfs(struct ptlrpc_request *req) if (req->rq_status != 0) CERROR("ost: statfs failed: rc %d\n", req->rq_status); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_EINPROGRESS)) + req->rq_status = -EINPROGRESS; + RETURN(0); } @@ -358,8 +371,9 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req, struct obd_trans_info *oti) { struct ost_body *body, *repbody; - int rc, flags = 0; + __u64 flags = 0; struct lustre_handle lh = {0,}; + int rc; ENTRY; /* check that we do support OBD_CONNECT_TRUNCLOCK. */ @@ -552,9 +566,19 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc, OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) { int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK; int len = desc->bd_iov[i].kiov_len; + struct page *np = ost_page_to_corrupt; char *ptr = kmap(desc->bd_iov[i].kiov_page) + off; - memcpy(ptr, "bad3", min(4, len)); - kunmap(desc->bd_iov[i].kiov_page); + + if (np) { + char *ptr2 = kmap(np) + off; + + memcpy(ptr2, ptr, len); + memcpy(ptr2, "bad3", min(4, len)); + kunmap(np); + desc->bd_iov[i].kiov_page = np; + } else { + CERROR("can't alloc page for corruption\n"); + } } cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page, desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK, @@ -566,11 +590,19 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc, OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) { int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK; int len = desc->bd_iov[i].kiov_len; + struct page *np = ost_page_to_corrupt; char *ptr = kmap(desc->bd_iov[i].kiov_page) + off; - memcpy(ptr, "bad4", min(4, len)); - kunmap(desc->bd_iov[i].kiov_page); - /* nobody should use corrupted page again */ - ClearPageUptodate(desc->bd_iov[i].kiov_page); + + if (np) { + char *ptr2 = kmap(np) + off; + + memcpy(ptr2, ptr, len); + memcpy(ptr2, "bad4", min(4, len)); + kunmap(np); + desc->bd_iov[i].kiov_page = np; + } else { + CERROR("can't alloc page for corruption\n"); + } } } @@ -586,7 +618,7 @@ 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; + __u64 flags = 0; int nrbufs = obj->ioo_bufcnt; struct ldlm_res_id res_id; ldlm_policy_data_t policy; @@ -612,7 +644,8 @@ 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, LVB_T_NONE, + NULL, lh)); } static void ost_brw_lock_put(int mode, @@ -636,7 +669,7 @@ static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r) /* In normal mode of operation an I/O request is serviced only * by ll_ost_io threads each of them has own tls buffers allocated by - * ost_thread_init(). + * ost_io_thread_init(). * During recovery, an I/O request may be queued until any of the ost * service threads process it. Not necessary it should be one of * ll_ost_io threads. In that case we dynamically allocating tls @@ -786,9 +819,9 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti) nob += page_rc; if (page_rc != 0) { /* some data! */ LASSERT (local_nb[i].page != NULL); - ptlrpc_prep_bulk_page(desc, local_nb[i].page, - local_nb[i].offset & ~CFS_PAGE_MASK, - page_rc); + ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page, + local_nb[i].lnb_page_offset, + page_rc); } if (page_rc != local_nb[i].len) { /* short read */ @@ -836,7 +869,7 @@ out_tls: ost_tls_put(req); out_bulk: if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) - ptlrpc_free_bulk(desc); + ptlrpc_free_bulk_nopin(desc); out: LASSERT(rc <= 0); if (rc == 0) { @@ -870,12 +903,59 @@ out: lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL); l_wait_event(waitq, 0, &lwi1); rc = target_bulk_io(exp, desc, &lwi); - ptlrpc_free_bulk(desc); + ptlrpc_free_bulk_nopin(desc); } RETURN(rc); } +static void ost_warn_on_cksum(struct ptlrpc_request *req, + struct ptlrpc_bulk_desc *desc, + struct niobuf_local *local_nb, int npages, + obd_count client_cksum, obd_count server_cksum, + int mmap) +{ + struct obd_export *exp = req->rq_export; + struct ost_body *body; + char *router; + char *via; + + body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY); + LASSERT (body != NULL); + + if (req->rq_peer.nid == desc->bd_sender) { + via = router = ""; + } else { + via = " via "; + router = libcfs_nid2str(desc->bd_sender); + } + + if (mmap) { + CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n", + client_cksum, server_cksum); + return; + } + + LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode " + DFID" object "LPU64"/"LPU64" extent ["LPU64"-"LPU64 + "]: client csum %x, server csum %x\n", + exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer), + via, router, + body->oa.o_valid & OBD_MD_FLFID ? + body->oa.o_parent_seq : (__u64)0, + body->oa.o_valid & OBD_MD_FLFID ? + body->oa.o_parent_oid : 0, + body->oa.o_valid & OBD_MD_FLFID ? + body->oa.o_parent_ver : 0, + body->oa.o_id, + body->oa.o_valid & OBD_MD_FLGROUP ? + body->oa.o_seq : (__u64)0, + local_nb[0].lnb_file_offset, + local_nb[npages-1].lnb_file_offset + + local_nb[npages-1].len - 1, + client_cksum, server_cksum); +} + static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) { struct ptlrpc_bulk_desc *desc = NULL; @@ -1020,9 +1100,9 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti) /* NB Having prepped, we must commit... */ for (i = 0; i < npages; i++) - ptlrpc_prep_bulk_page(desc, local_nb[i].page, - local_nb[i].offset & ~CFS_PAGE_MASK, - local_nb[i].len); + ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page, + local_nb[i].lnb_page_offset, + local_nb[i].len); rc = sptlrpc_svc_prep_bulk(req, desc); if (rc != 0) @@ -1041,10 +1121,10 @@ skip_transfer: repbody->oa.o_cksum = server_cksum; cksum_counter++; if (unlikely(client_cksum != server_cksum)) { - CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR, - "client csum %x, server csum %x\n", - client_cksum, server_cksum); + ost_warn_on_cksum(req, desc, local_nb, npages, + client_cksum, server_cksum, mmap); cksum_counter = 0; + } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){ CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n", cksum_counter, libcfs_id2str(req->rq_peer), @@ -1074,49 +1154,6 @@ skip_transfer: */ repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME); - if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) { - 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"; - else if (new_cksum == client_cksum) - msg = "initial checksum before message complete"; - else - msg = "changed in transit AND after initial checksum"; - - 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 inode "DFID" object " - LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n", - exp->exp_obd->obd_name, msg, - libcfs_id2str(req->rq_peer), - via, router, - body->oa.o_valid & OBD_MD_FLFID ? - body->oa.o_parent_seq : (__u64)0, - body->oa.o_valid & OBD_MD_FLFID ? - body->oa.o_parent_oid : 0, - body->oa.o_valid & OBD_MD_FLFID ? - body->oa.o_parent_ver : 0, - body->oa.o_id, - body->oa.o_valid & OBD_MD_FLGROUP ? - body->oa.o_seq : (__u64)0, - 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); - } - if (rc == 0) { int nob = 0; @@ -1145,7 +1182,7 @@ out_tls: ost_tls_put(req); out_bulk: if (desc) - ptlrpc_free_bulk(desc); + ptlrpc_free_bulk_nopin(desc); out: if (rc == 0) { oti_to_request(oti, req); @@ -1296,7 +1333,6 @@ static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req) RETURN(rc); } -#ifdef HAVE_QUOTA_SUPPORT static int ost_handle_quotactl(struct ptlrpc_request *req) { struct obd_quotactl *oqctl, *repoqc; @@ -1333,34 +1369,10 @@ static int ost_handle_quotacheck(struct ptlrpc_request *req) 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, NULL); - *repoqa = *oqaq; - - out: - RETURN(rc); + /* deprecated, not used any more */ + req->rq_status = -EOPNOTSUPP; + RETURN(-EOPNOTSUPP); } -#endif static int ost_llog_handle_connect(struct obd_export *exp, struct ptlrpc_request *req) @@ -1374,14 +1386,14 @@ static int ost_llog_handle_connect(struct obd_export *exp, 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); \ - cfs_spin_lock(&exp->exp_lock); \ - exp->exp_connect_flags = reply->ocd_connect_flags; \ - cfs_spin_unlock(&exp->exp_lock); \ +#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) @@ -1478,9 +1490,9 @@ static int ost_init_sec_level(struct ptlrpc_request *req) if (!filter->fo_fl_oss_capa) reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA; - cfs_spin_lock(&exp->exp_lock); - exp->exp_connect_flags = reply->ocd_connect_flags; - cfs_spin_unlock(&exp->exp_lock); + spin_lock(&exp->exp_lock); + exp->exp_connect_flags = reply->ocd_connect_flags; + spin_unlock(&exp->exp_lock); } break; default: @@ -1513,14 +1525,14 @@ static int ost_connect_check_sptlrpc(struct ptlrpc_request *req) } if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { - cfs_read_lock(&filter->fo_sptlrpc_lock); - sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset, - req->rq_sp_from, - req->rq_peer.nid, - &flvr); - cfs_read_unlock(&filter->fo_sptlrpc_lock); + 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); - cfs_spin_lock(&exp->exp_lock); + spin_lock(&exp->exp_lock); exp->exp_sp_peer = req->rq_sp_from; exp->exp_flvr = flvr; @@ -1534,7 +1546,7 @@ static int ost_connect_check_sptlrpc(struct ptlrpc_request *req) rc = -EACCES; } - cfs_spin_unlock(&exp->exp_lock); + spin_unlock(&exp->exp_lock); } else { if (exp->exp_sp_peer != req->rq_sp_from) { CERROR("RPC source %s doesn't match %s\n", @@ -1588,6 +1600,7 @@ int ost_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, oa->o_seq = lock->l_resource->lr_name.name[1]; oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP; oinfo->oi_oa = oa; + oinfo->oi_capa = BYPASS_CAPA; rc = obd_sync(&env, lock->l_export, oinfo, lock->l_policy_data.l_extent.start, @@ -1651,6 +1664,15 @@ int ost_msg_check_version(struct lustre_msg *msg) lustre_msg_get_version(msg), LUSTRE_OBD_VERSION); break; + case SEQ_QUERY: + /* Note: client always use MDS_VERSION for FID request */ + rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION); + if (rc) + CERROR("bad opc %u version %08x, expecting %08x\n", + lustre_msg_get_opc(msg), + lustre_msg_get_version(msg), + LUSTRE_MDS_VERSION); + break; case OST_CREATE: case OST_DESTROY: case OST_GETATTR: @@ -1662,11 +1684,8 @@ 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", @@ -1695,6 +1714,10 @@ int ost_msg_check_version(struct lustre_msg *msg) lustre_msg_get_version(msg), LUSTRE_LOG_VERSION); break; + case OST_QUOTA_ADJUST_QUNIT: + rc = -ENOTSUPP; + CERROR("Quota adjust is deprecated as of 2.4.0\n"); + break; default: CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg)); rc = -ENOTSUPP; @@ -1781,7 +1804,7 @@ static void ost_prolong_locks(struct ost_prolong_data *data) } - cfs_spin_lock_bh(&exp->exp_bl_list_lock); + spin_lock_bh(&exp->exp_bl_list_lock); cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) { LASSERT(lock->l_flags & LDLM_FL_AST_SENT); LASSERT(lock->l_resource->lr_type == LDLM_EXTENT); @@ -1795,9 +1818,9 @@ static void ost_prolong_locks(struct ost_prolong_data *data) ost_prolong_lock_one(data, lock); } - cfs_spin_unlock_bh(&exp->exp_bl_list_lock); + spin_unlock_bh(&exp->exp_bl_list_lock); - EXIT; + EXIT; } /** @@ -1995,7 +2018,7 @@ struct ptlrpc_hpreq_ops ost_hpreq_punch = { }; /** Assign high priority operations to the request if needed. */ -static int ost_hpreq_handler(struct ptlrpc_request *req) +static int ost_io_hpreq_handler(struct ptlrpc_request *req) { ENTRY; if (req->rq_export) { @@ -2230,7 +2253,7 @@ int ost_handle(struct ptlrpc_request *req) RETURN(0); if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC)) GOTO(out, rc = -ENOSPC); - if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS)) + if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1)) GOTO(out, rc = -EROFS); rc = ost_brw_write(req, oti); LASSERT(current->journal_info == NULL); @@ -2286,7 +2309,10 @@ int ost_handle(struct ptlrpc_request *req) req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC); rc = ost_get_info(req->rq_export, req); break; -#ifdef HAVE_QUOTA_SUPPORT + case SEQ_QUERY: + CDEBUG(D_INODE, "seq\n"); + rc = seq_handle(req); + break; case OST_QUOTACHECK: CDEBUG(D_INODE, "quotacheck\n"); req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK); @@ -2301,12 +2327,6 @@ int ost_handle(struct ptlrpc_request *req) 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); @@ -2335,30 +2355,30 @@ int ost_handle(struct ptlrpc_request *req) if (rc) RETURN(rc); RETURN(ptlrpc_reply(req)); - case LDLM_ENQUEUE: - CDEBUG(D_INODE, "enqueue\n"); - 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, - ost_blocking_ast, - ldlm_server_glimpse_ast); - fail = OBD_FAIL_OST_LDLM_REPLY_NET; - break; - case LDLM_CONVERT: - CDEBUG(D_INODE, "convert\n"); - 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"); - 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_ENQUEUE: + CDEBUG(D_INODE, "enqueue\n"); + req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE_NET)) + RETURN(0); + rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast, + ost_blocking_ast, + ldlm_server_glimpse_ast); + fail = OBD_FAIL_OST_LDLM_REPLY_NET; + break; + case LDLM_CONVERT: + CDEBUG(D_INODE, "convert\n"); + req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT_NET)) + RETURN(0); + rc = ldlm_handle_convert(req); + break; + case LDLM_CANCEL: + CDEBUG(D_INODE, "cancel\n"); + req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL); + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL_NET)) + RETURN(0); + rc = ldlm_handle_cancel(req); + break; case LDLM_BL_CALLBACK: case LDLM_CP_CALLBACK: CDEBUG(D_INODE, "callback\n"); @@ -2387,10 +2407,11 @@ out: return 0; } EXPORT_SYMBOL(ost_handle); + /* - * free per-thread pool created by ost_thread_init(). + * free per-thread pool created by ost_io_thread_init(). */ -static void ost_thread_done(struct ptlrpc_thread *thread) +static void ost_io_thread_done(struct ptlrpc_thread *thread) { struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local * Storage */ @@ -2401,7 +2422,7 @@ static void ost_thread_done(struct ptlrpc_thread *thread) /* * be prepared to handle partially-initialized pools (because this is - * called from ost_thread_init() for cleanup. + * called from ost_io_thread_init() for cleanup. */ tls = thread->t_data; if (tls != NULL) { @@ -2414,7 +2435,7 @@ static void ost_thread_done(struct ptlrpc_thread *thread) /* * initialize per-thread page pool (bug 5137). */ -static int ost_thread_init(struct ptlrpc_thread *thread) +static int ost_io_thread_init(struct ptlrpc_thread *thread) { struct ost_thread_local_cache *tls; @@ -2451,7 +2472,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) lprocfs_ost_init_vars(&lvars); lprocfs_obd_setup(obd, lvars.obd_vars); - cfs_mutex_init(&ost->ost_health_mutex); + mutex_init(&ost->ost_health_mutex); svc_conf = (typeof(svc_conf)) { .psc_name = LUSTRE_OSS_NAME, @@ -2480,6 +2501,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) .psc_ops = { .so_req_handler = ost_handle, .so_req_printer = target_print_req, + .so_hpreq_handler = ptlrpc_hpreq_handler, }, }; ost->ost_service = ptlrpc_register_service(&svc_conf, @@ -2581,10 +2603,10 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) oss_io_cpts : NULL, }, .psc_ops = { - .so_thr_init = ost_thread_init, - .so_thr_done = ost_thread_done, + .so_thr_init = ost_io_thread_init, + .so_thr_done = ost_io_thread_done, .so_req_handler = ost_handle, - .so_hpreq_handler = ost_hpreq_handler, + .so_hpreq_handler = ost_io_hpreq_handler, .so_req_printer = target_print_req, }, }; @@ -2597,10 +2619,53 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg) GOTO(out_create, rc); } + memset(&svc_conf, 0, sizeof(svc_conf)); + svc_conf = (typeof(svc_conf)) { + .psc_name = "ost_seq", + .psc_watchdog_factor = OSS_SERVICE_WATCHDOG_FACTOR, + .psc_buf = { + .bc_nbufs = OST_NBUFS, + .bc_buf_size = OST_BUFSIZE, + .bc_req_max_size = OST_MAXREQSIZE, + .bc_rep_max_size = OST_MAXREPSIZE, + .bc_req_portal = SEQ_DATA_PORTAL, + .bc_rep_portal = OSC_REPLY_PORTAL, + }, + .psc_thr = { + .tc_thr_name = "ll_ost_seq", + .tc_thr_factor = OSS_CR_THR_FACTOR, + .tc_nthrs_init = OSS_CR_NTHRS_INIT, + .tc_nthrs_base = OSS_CR_NTHRS_BASE, + .tc_nthrs_max = OSS_CR_NTHRS_MAX, + .tc_nthrs_user = oss_num_create_threads, + .tc_cpu_affinity = 1, + .tc_ctx_tags = LCT_DT_THREAD, + }, + + .psc_cpt = { + .cc_pattern = oss_cpts, + }, + .psc_ops = { + .so_req_handler = ost_handle, + .so_req_printer = target_print_req, + .so_hpreq_handler = NULL, + }, + }; + ost->ost_seq_service = ptlrpc_register_service(&svc_conf, + obd->obd_proc_entry); + if (IS_ERR(ost->ost_seq_service)) { + rc = PTR_ERR(ost->ost_seq_service); + CERROR("failed to start OST seq service: %d\n", rc); + ost->ost_seq_service = NULL; + GOTO(out_io, rc); + } + ping_evictor_start(); RETURN(0); - +out_io: + ptlrpc_unregister_service(ost->ost_io_service); + ost->ost_io_service = NULL; out_create: ptlrpc_unregister_service(ost->ost_create_service); ost->ost_create_service = NULL; @@ -2614,24 +2679,26 @@ out_lprocfs: static int ost_cleanup(struct obd_device *obd) { - struct ost_obd *ost = &obd->u.ost; - int err = 0; - ENTRY; - - ping_evictor_stop(); + struct ost_obd *ost = &obd->u.ost; + int err = 0; + ENTRY; - /* there is no recovery for OST OBD, all recovery is controlled by - * obdfilter OBD */ - LASSERT(obd->obd_recovering == 0); - cfs_mutex_lock(&ost->ost_health_mutex); - ptlrpc_unregister_service(ost->ost_service); - ptlrpc_unregister_service(ost->ost_create_service); - ptlrpc_unregister_service(ost->ost_io_service); - ost->ost_service = NULL; - ost->ost_create_service = NULL; + ping_evictor_stop(); + + /* there is no recovery for OST OBD, all recovery is controlled by + * obdfilter OBD */ + LASSERT(obd->obd_recovering == 0); + mutex_lock(&ost->ost_health_mutex); + ptlrpc_unregister_service(ost->ost_service); + ptlrpc_unregister_service(ost->ost_create_service); + ptlrpc_unregister_service(ost->ost_io_service); + ptlrpc_unregister_service(ost->ost_seq_service); + ost->ost_service = NULL; + ost->ost_create_service = NULL; ost->ost_io_service = NULL; + ost->ost_seq_service = NULL; - cfs_mutex_unlock(&ost->ost_health_mutex); + mutex_unlock(&ost->ost_health_mutex); lprocfs_obd_cleanup(obd); @@ -2648,11 +2715,11 @@ static int ost_health_check(const struct lu_env *env, struct obd_device *obd) struct ost_obd *ost = &obd->u.ost; int rc = 0; - cfs_mutex_lock(&ost->ost_health_mutex); + mutex_lock(&ost->ost_health_mutex); rc |= ptlrpc_service_health_check(ost->ost_service); rc |= ptlrpc_service_health_check(ost->ost_create_service); rc |= ptlrpc_service_health_check(ost->ost_io_service); - cfs_mutex_unlock(&ost->ost_health_mutex); + mutex_unlock(&ost->ost_health_mutex); /* * health_check to return 0 on healthy @@ -2684,6 +2751,8 @@ static int __init ost_init(void) int rc; ENTRY; + ost_page_to_corrupt = cfs_alloc_page(CFS_ALLOC_STD); + lprocfs_ost_init_vars(&lvars); rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars, LUSTRE_OSS_NAME, NULL); @@ -2700,6 +2769,9 @@ static int __init ost_init(void) static void /*__exit*/ ost_exit(void) { + if (ost_page_to_corrupt) + page_cache_release(ost_page_to_corrupt); + class_unregister_type(LUSTRE_OSS_NAME); }