X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosc%2Fosc_create.c;h=f3eb6bc1a19e4e70a3942969a6512300ee87a3bd;hp=184df58fa21994c244f6636f4d6738fdd1b3a16c;hb=6869932b552ac705f411de3362f01bd50c1f6f7d;hpb=fbb7ead129258897f5a5d5c9ce28d31fbbe5bca2 diff --git a/lustre/osc/osc_create.c b/lustre/osc/osc_create.c index 184df58..f3eb6bc 100644 --- a/lustre/osc/osc_create.c +++ b/lustre/osc/osc_create.c @@ -1,29 +1,45 @@ /* -*- 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 Braam + * GPL HEADER START * - * This file is part of Lustre, http://www.lustre.org. + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * Lustre is free 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 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. * - * 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 - * GNU General Public License for more details. + * 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). * - * You should have received a copy of the GNU General Public License - * along with Lustre; if not, write to the Free Software - * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + * 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 * - * 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.) + * 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. * + * 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/osc/osc_create.c + * 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.) + * + * Author: Peter Braam */ #ifndef EXPORT_SYMTAB @@ -32,18 +48,7 @@ #define DEBUG_SUBSYSTEM S_OSC #ifdef __KERNEL__ -# include -# include -# include -# include -# include -# include -# if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0)) -# include -# include -# else -# include -# endif +# include #else /* __KERNEL__ */ # include #endif @@ -52,8 +57,8 @@ # include #endif -# include -#include +# include +#include #include "osc_internal.h" static int osc_interpret_create(struct ptlrpc_request *req, void *data, int rc) @@ -63,16 +68,40 @@ static int osc_interpret_create(struct ptlrpc_request *req, void *data, int rc) ENTRY; if (req->rq_repmsg) { - body = lustre_swab_repbuf(req, 0, sizeof(*body), + body = lustre_swab_repbuf(req, REPLY_REC_OFF, sizeof(*body), lustre_swab_ost_body); if (body == NULL && rc == 0) rc = -EPROTO; } oscc = req->rq_async_args.pointer_arg[0]; + LASSERT(oscc && (oscc->oscc_obd != LP_POISON)); + spin_lock(&oscc->oscc_lock); oscc->oscc_flags &= ~OSCC_FLAG_CREATING; - if (rc == -ENOSPC || rc == -EROFS) { + switch (rc) { + case 0: { + if (body) { + int diff = body->oa.o_id - oscc->oscc_last_id; + + if (diff < oscc->oscc_grow_count) + oscc->oscc_grow_count = + max(diff/3, OST_MIN_PRECREATE); + else + oscc->oscc_flags &= ~OSCC_FLAG_LOW; + oscc->oscc_last_id = body->oa.o_id; + } + spin_unlock(&oscc->oscc_lock); + break; + } + case -EAGAIN: + /* valid race delorphan vs create, or somthing after resend */ + spin_unlock(&oscc->oscc_lock); + DEBUG_REQ(D_INODE, req, "Got EGAIN - resend \n"); + break; + case -ENOSPC: + case -EROFS: + case -EFBIG: { oscc->oscc_flags |= OSCC_FLAG_NOSPC; if (body && rc == -ENOSPC) { oscc->oscc_grow_count = OST_MIN_PRECREATE; @@ -80,31 +109,31 @@ static int osc_interpret_create(struct ptlrpc_request *req, void *data, int rc) } spin_unlock(&oscc->oscc_lock); DEBUG_REQ(D_INODE, req, "OST out of space, flagging"); - } else if (rc != 0 && rc != -EIO) { + break; + } + case -EIO: { + /* filter always set body->oa.o_id as the last_id + * of filter (see filter_handle_precreate for detail)*/ + if (body && body->oa.o_id > oscc->oscc_last_id) + oscc->oscc_last_id = body->oa.o_id; + spin_unlock(&oscc->oscc_lock); + break; + } + default: { oscc->oscc_flags |= OSCC_FLAG_RECOVERING; oscc->oscc_grow_count = OST_MIN_PRECREATE; spin_unlock(&oscc->oscc_lock); DEBUG_REQ(D_ERROR, req, - "unknown rc %d from async create: failing oscc", rc); - ptlrpc_fail_import(req->rq_import, req->rq_import_generation); - } else { - if (rc == 0) { - oscc->oscc_flags &= ~OSCC_FLAG_LOW; - if (body) { - int diff = body->oa.o_id - oscc->oscc_last_id; - if (diff != oscc->oscc_grow_count) - oscc->oscc_grow_count = - max(diff/3, OST_MIN_PRECREATE); - oscc->oscc_last_id = body->oa.o_id; - } - } - spin_unlock(&oscc->oscc_lock); + "Unknown rc %d from async create: failing oscc", rc); + ptlrpc_fail_import(req->rq_import, + lustre_msg_get_conn_cnt(req->rq_reqmsg)); + } } - CDEBUG(D_HA, "preallocated through id "LPU64" (last used "LPU64")\n", + CDEBUG(D_HA, "preallocated through id "LPU64" (next to use "LPU64")\n", oscc->oscc_last_id, oscc->oscc_next_id); - wake_up(&oscc->oscc_waitq); + cfs_waitq_signal(&oscc->oscc_waitq); RETURN(rc); } @@ -112,32 +141,34 @@ static int oscc_internal_create(struct osc_creator *oscc) { struct ptlrpc_request *request; struct ost_body *body; - int size = sizeof(*body); + __u32 size[] = { sizeof(struct ptlrpc_body), sizeof(*body) }; ENTRY; - spin_lock(&oscc->oscc_lock); - if (oscc->oscc_grow_count < OST_MAX_PRECREATE && - !(oscc->oscc_flags & (OSCC_FLAG_LOW | OSCC_FLAG_RECOVERING)) && + LASSERT_SPIN_LOCKED(&oscc->oscc_lock); + + if (oscc->oscc_flags & OSCC_FLAG_CREATING || + oscc->oscc_flags & OSCC_FLAG_RECOVERING) { + spin_unlock(&oscc->oscc_lock); + RETURN(0); + } + + if (oscc->oscc_grow_count < oscc->oscc_max_grow_count && + ((oscc->oscc_flags & OSCC_FLAG_LOW) == 0) && (__s64)(oscc->oscc_last_id - oscc->oscc_next_id) <= (oscc->oscc_grow_count / 4 + 1)) { oscc->oscc_flags |= OSCC_FLAG_LOW; oscc->oscc_grow_count *= 2; } - if (oscc->oscc_grow_count > OST_MAX_PRECREATE / 2) - oscc->oscc_grow_count = OST_MAX_PRECREATE / 2; + if (oscc->oscc_grow_count > oscc->oscc_max_grow_count / 2) + oscc->oscc_grow_count = oscc->oscc_max_grow_count / 2; - if (oscc->oscc_flags & OSCC_FLAG_CREATING || - oscc->oscc_flags & OSCC_FLAG_RECOVERING) { - spin_unlock(&oscc->oscc_lock); - RETURN(0); - } oscc->oscc_flags |= OSCC_FLAG_CREATING; spin_unlock(&oscc->oscc_lock); request = ptlrpc_prep_req(oscc->oscc_obd->u.cli.cl_import, - LUSTRE_OBD_VERSION, OST_CREATE, - 1, &size, NULL); + LUSTRE_OST_VERSION, OST_CREATE, 2, + size, NULL); if (request == NULL) { spin_lock(&oscc->oscc_lock); oscc->oscc_flags &= ~OSCC_FLAG_CREATING; @@ -145,25 +176,20 @@ static int oscc_internal_create(struct osc_creator *oscc) RETURN(-ENOMEM); } - request->rq_request_portal = OST_CREATE_PORTAL; //XXX FIXME bug 249 - body = lustre_msg_buf(request->rq_reqmsg, 0, sizeof(*body)); + request->rq_request_portal = OST_CREATE_PORTAL; + ptlrpc_at_set_req_timeout(request); + body = lustre_msg_buf(request->rq_reqmsg, REQ_REC_OFF, sizeof(*body)); spin_lock(&oscc->oscc_lock); body->oa.o_id = oscc->oscc_last_id + oscc->oscc_grow_count; - /* probably we should take frequence of request into account? -bzzz */ - if (oscc->oscc_grow_count < oscc->oscc_max_grow_count) { - oscc->oscc_grow_count *= 2; - if (oscc->oscc_grow_count > oscc->oscc_max_grow_count) - oscc->oscc_grow_count = oscc->oscc_max_grow_count; - } - body->oa.o_gr = oscc->oscc_gr; + body->oa.o_gr = oscc->oscc_oa.o_gr; LASSERT(body->oa.o_gr > 0); body->oa.o_valid |= OBD_MD_FLID | OBD_MD_FLGROUP; spin_unlock(&oscc->oscc_lock); - CDEBUG(D_INFO, "preallocating through id "LPU64" (last used "LPU64")\n", - body->oa.o_id, oscc->oscc_next_id); + CDEBUG(D_RPCTRACE, "prealloc through id "LPU64" (last seen "LPU64")\n", + body->oa.o_id, oscc->oscc_last_id); - request->rq_replen = lustre_msg_size(1, &size); + ptlrpc_req_set_repsize(request, 2, size); request->rq_async_args.pointer_arg[0] = oscc; request->rq_interpret_reply = osc_interpret_create; @@ -177,10 +203,12 @@ static int oscc_has_objects(struct osc_creator *oscc, int count) int have_objs; spin_lock(&oscc->oscc_lock); have_objs = ((__s64)(oscc->oscc_last_id - oscc->oscc_next_id) >= count); - spin_unlock(&oscc->oscc_lock); - if (!have_objs) + if (!have_objs) { oscc_internal_create(oscc); + } else { + spin_unlock(&oscc->oscc_lock); + } return have_objs; } @@ -237,30 +265,67 @@ int oscc_recovering(struct osc_creator *oscc) return recov; } +/* decide if the OST has remaining object, return value : + 0 : the OST has remaining object, and don't need to do precreate. + 1 : the OST has no remaining object, and will send a RPC for precreate. + 2 : the OST has no remaining object, and will not get any for + a potentially very long time + 1000 : unusable + */ +int osc_precreate(struct obd_export *exp) +{ + struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc; + struct obd_import *imp = exp->exp_imp_reverse; + ENTRY; + + LASSERT(oscc != NULL); + if (imp != NULL && imp->imp_deactive) + RETURN(1000); + + if (oscc->oscc_last_id < oscc->oscc_next_id) { + spin_lock(&oscc->oscc_lock); + if (oscc->oscc_flags & OSCC_FLAG_NOSPC) { + spin_unlock(&oscc->oscc_lock); + RETURN(1000); + } + if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) { + spin_unlock(&oscc->oscc_lock); + RETURN(1); + } + if (oscc->oscc_flags & OSCC_FLAG_RECOVERING) { + spin_unlock(&oscc->oscc_lock); + RETURN(2); + } + + if (oscc->oscc_flags & OSCC_FLAG_CREATING) { + spin_unlock(&oscc->oscc_lock); + RETURN(1); + } + + oscc_internal_create(oscc); + RETURN(1); + } + RETURN(0); +} + int osc_create(struct obd_export *exp, struct obdo *oa, struct lov_stripe_md **ea, struct obd_trans_info *oti) { - struct lov_stripe_md *lsm; struct osc_creator *oscc = &exp->exp_obd->u.cli.cl_oscc; + struct lov_stripe_md *lsm; int try_again = 1, rc = 0; ENTRY; + LASSERT(oa); LASSERT(ea); - LASSERT(oa->o_valid & OBD_MD_FLGROUP); LASSERT(oa->o_gr > 0); + LASSERT(oa->o_valid & OBD_MD_FLGROUP); if ((oa->o_valid & OBD_MD_FLFLAGS) && oa->o_flags == OBD_FL_RECREATE_OBJS) { - /* Exceptional case where we are trying to repair missing - * objects for various groups. We have already validated that - * this is a valid group for the file. Don't set oscc->oscc_gr. - */ RETURN(osc_real_create(exp, oa, ea, oti)); } - LASSERT(oscc->oscc_gr == 0 || oscc->oscc_gr == oa->o_gr); - oscc->oscc_gr = oa->o_gr; - if (oa->o_gr == FILTER_GROUP_LLOG || oa->o_gr == FILTER_GROUP_ECHO) RETURN(osc_real_create(exp, oa, ea, oti)); @@ -270,30 +335,25 @@ int osc_create(struct obd_export *exp, struct obdo *oa, spin_lock(&oscc->oscc_lock); if (oscc->oscc_flags & OSCC_FLAG_SYNC_IN_PROGRESS) { spin_unlock(&oscc->oscc_lock); - return -EBUSY; + RETURN(-EBUSY); } if (!(oscc->oscc_flags & OSCC_FLAG_RECOVERING)) { spin_unlock(&oscc->oscc_lock); - return 0; + RETURN(0); } oscc->oscc_flags |= OSCC_FLAG_SYNC_IN_PROGRESS; + /* seting flag LOW we prevent extra grow precreate size + * and enforce use last assigned size */ + oscc->oscc_flags |= OSCC_FLAG_LOW; spin_unlock(&oscc->oscc_lock); - CDEBUG(D_HA, "%s; oscc recovery started\n", - exp->exp_obd->obd_name); - LASSERT(oscc->oscc_flags & OSCC_FLAG_RECOVERING); + CDEBUG(D_HA, "%s: oscc recovery started - delete to "LPU64"\n", + oscc->oscc_obd->obd_name, oscc->oscc_next_id - 1); /* delete from next_id on up */ oa->o_valid |= OBD_MD_FLID; oa->o_id = oscc->oscc_next_id - 1; - CDEBUG(D_HA, "%s: deleting to next_id: "LPU64"\n", - oscc->oscc_obd->obd_name, oa->o_id); - rc = osc_real_create(exp, oa, ea, NULL); - if (oscc->oscc_obd == NULL) { - CWARN("the obd for oscc %p has been freed\n", oscc); - RETURN(rc); - } spin_lock(&oscc->oscc_lock); oscc->oscc_flags &= ~OSCC_FLAG_SYNC_IN_PROGRESS; @@ -302,17 +362,17 @@ int osc_create(struct obd_export *exp, struct obdo *oa, oscc->oscc_flags |= OSCC_FLAG_NOSPC; oscc->oscc_flags &= ~OSCC_FLAG_RECOVERING; oscc->oscc_last_id = oa->o_id; - - CDEBUG(D_HA, "%s: oscc recovery finished: %d\n", - oscc->oscc_obd->obd_name, rc); - wake_up(&oscc->oscc_waitq); - + CDEBUG(D_HA, "%s: oscc recovery finished, last_id: " + LPU64", rc: %d\n", oscc->oscc_obd->obd_name, + oscc->oscc_last_id, rc); + cfs_waitq_signal(&oscc->oscc_waitq); } else { CDEBUG(D_ERROR, "%s: oscc recovery failed: %d\n", oscc->oscc_obd->obd_name, rc); } spin_unlock(&oscc->oscc_lock); + RETURN(rc); } @@ -324,27 +384,28 @@ int osc_create(struct obd_export *exp, struct obdo *oa, } while (try_again) { - /* If orphans are being recovered, then we must wait until + /* If orphans are being recovered, then we must wait until it is finished before we can continue with create. */ if (oscc_recovering(oscc)) { struct l_wait_info lwi; - - CDEBUG(D_HA,"%p: oscc recovery in progress, waiting\n", - oscc); - lwi = LWI_TIMEOUT(MAX(obd_timeout*HZ/4, 1), NULL, NULL); + + CDEBUG(D_HA,"%s: oscc recovery in progress, waiting\n", + oscc->oscc_obd->obd_name); + + lwi = LWI_TIMEOUT(cfs_timeout_cap(cfs_time_seconds( + obd_timeout / 4)), NULL, NULL); rc = l_wait_event(oscc->oscc_waitq, !oscc_recovering(oscc), &lwi); - LASSERT(rc == 0 || rc == -ETIMEDOUT); if (rc == -ETIMEDOUT) { - CDEBUG(D_HA,"%p: timeout waiting on recovery\n", - oscc); + CDEBUG(D_HA,"%s: timeout waiting on recovery\n", + oscc->oscc_obd->obd_name); RETURN(rc); } CDEBUG(D_HA, "%s: oscc recovery over, waking up\n", - exp->exp_obd->obd_name); + oscc->oscc_obd->obd_name); } - + spin_lock(&oscc->oscc_lock); if (oscc->oscc_flags & OSCC_FLAG_EXITING) { spin_unlock(&oscc->oscc_lock); @@ -354,12 +415,13 @@ int osc_create(struct obd_export *exp, struct obdo *oa, if (oscc->oscc_last_id >= oscc->oscc_next_id) { memcpy(oa, &oscc->oscc_oa, sizeof(*oa)); oa->o_id = oscc->oscc_next_id; - oa->o_gr = oscc->oscc_gr; lsm->lsm_object_id = oscc->oscc_next_id; - lsm->lsm_object_gr = oscc->oscc_gr; *ea = lsm; oscc->oscc_next_id++; try_again = 0; + + CDEBUG(D_RPCTRACE, "%s: set oscc_next_id = "LPU64"\n", + exp->exp_obd->obd_name, oscc->oscc_next_id); } else if (oscc->oscc_flags & OSCC_FLAG_NOSPC) { rc = -ENOSPC; spin_unlock(&oscc->oscc_lock); @@ -372,9 +434,8 @@ int osc_create(struct obd_export *exp, struct obdo *oa, } if (rc == 0) - CDEBUG(D_HA, "%s: returning objid "LPU64"\n", - oscc->oscc_obd->u.cli.cl_import->imp_target_uuid.uuid, - lsm->lsm_object_id); + CDEBUG(D_INFO, "%s: returning objid "LPU64"\n", + obd2cli_tgt(oscc->oscc_obd), lsm->lsm_object_id); else if (*ea == NULL) obd_free_memmd(exp, &lsm); RETURN(rc); @@ -390,13 +451,12 @@ void oscc_init(struct obd_device *obd) oscc = &obd->u.cli.cl_oscc; memset(oscc, 0, sizeof(*oscc)); - INIT_LIST_HEAD(&oscc->oscc_list); - init_waitqueue_head(&oscc->oscc_waitq); + CFS_INIT_LIST_HEAD(&oscc->oscc_list); + cfs_waitq_init(&oscc->oscc_waitq); spin_lock_init(&oscc->oscc_lock); oscc->oscc_obd = obd; - oscc->oscc_kick_barrier = 100; - oscc->oscc_max_grow_count = 2000; oscc->oscc_grow_count = OST_MIN_PRECREATE; + oscc->oscc_max_grow_count = OST_MAX_PRECREATE; oscc->oscc_next_id = 2; oscc->oscc_last_id = 1;