X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fllite%2Fllite_close.c;h=a952fb11a7527c8d5ffc797e5f491fff994c8076;hp=615871375e8e817c70e2caa0656537b9fb4af21e;hb=7b778c9476a43a3e6c80414452e93bcb50c78013;hpb=2dc9c16e770415d56839e1996015fec5fab93f29 diff --git a/lustre/llite/llite_close.c b/lustre/llite/llite_close.c index 6158713..a952fb1 100644 --- a/lustre/llite/llite_close.c +++ b/lustre/llite/llite_close.c @@ -1,193 +1,350 @@ /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- * vim:expandtab:shiftwidth=8:tabstop=8: * - * Lustre Lite routines to issue a secondary close after writeback + * GPL HEADER START + * + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * - * Copyright (c) 2001-2003 Cluster File Systems, Inc. + * 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. * - * This file is part of Lustre, http://www.lustre.org. + * 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). * - * 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. + * 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 * - * 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. + * 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. * - * 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. + * 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/llite/llite_close.c + * + * Lustre Lite routines to issue a secondary close after writeback */ #include #define DEBUG_SUBSYSTEM S_LLITE -#include -#include +//#include +#include #include "llite_internal.h" -/* record that a write is in flight */ -void llap_write_pending(struct inode *inode, struct ll_async_page *llap) +/** records that a write is in flight */ +void vvp_write_pending(struct ccc_object *club, struct ccc_page *page) { - struct ll_inode_info *lli = ll_i2info(inode); - spin_lock(&lli->lli_lock); - list_add(&llap->llap_pending_write, &lli->lli_pending_write_llaps); - spin_unlock(&lli->lli_lock); -} + struct ll_inode_info *lli = ll_i2info(club->cob_inode); -/* record that a write has completed */ -void llap_write_complete(struct inode *inode, struct ll_async_page *llap) -{ - struct ll_inode_info *lli = ll_i2info(inode); - spin_lock(&lli->lli_lock); - if (!list_empty(&llap->llap_pending_write)) - list_del_init(&llap->llap_pending_write); - spin_unlock(&lli->lli_lock); + ENTRY; + cfs_spin_lock(&lli->lli_lock); + lli->lli_flags |= LLIF_SOM_DIRTY; + if (page != NULL && cfs_list_empty(&page->cpg_pending_linkage)) + cfs_list_add(&page->cpg_pending_linkage, + &club->cob_pending_list); + cfs_spin_unlock(&lli->lli_lock); + EXIT; } -void ll_open_complete(struct inode *inode) +/** records that a write has completed */ +void vvp_write_complete(struct ccc_object *club, struct ccc_page *page) { - struct ll_inode_info *lli = ll_i2info(inode); - spin_lock(&lli->lli_lock); - lli->lli_send_done_writing = 0; - spin_unlock(&lli->lli_lock); + struct ll_inode_info *lli = ll_i2info(club->cob_inode); + int rc = 0; + + ENTRY; + cfs_spin_lock(&lli->lli_lock); + if (page != NULL && !cfs_list_empty(&page->cpg_pending_linkage)) { + cfs_list_del_init(&page->cpg_pending_linkage); + rc = 1; + } + cfs_spin_unlock(&lli->lli_lock); + if (rc) + ll_queue_done_writing(club->cob_inode, 0); + EXIT; } -/* if we close with writes in flight then we want the completion or cancelation - * of those writes to send a DONE_WRITING rpc to the MDS */ -int ll_is_inode_dirty(struct inode *inode) +/** Queues DONE_WRITING if + * - done writing is allowed; + * - inode has no no dirty pages; */ +void ll_queue_done_writing(struct inode *inode, unsigned long flags) { struct ll_inode_info *lli = ll_i2info(inode); - int rc = 0; + struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob); ENTRY; - spin_lock(&lli->lli_lock); - if (!list_empty(&lli->lli_pending_write_llaps)) - rc = 1; - spin_unlock(&lli->lli_lock); - RETURN(rc); + cfs_spin_lock(&lli->lli_lock); + lli->lli_flags |= flags; + + if ((lli->lli_flags & LLIF_DONE_WRITING) && + cfs_list_empty(&club->cob_pending_list)) { + struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq; + + if (lli->lli_flags & LLIF_MDS_SIZE_LOCK) + CWARN("ino %lu/%u(flags %lu) som valid it just after " + "recovery\n", + inode->i_ino, inode->i_generation, + lli->lli_flags); + /* DONE_WRITING is allowed and inode has no dirty page. */ + cfs_spin_lock(&lcq->lcq_lock); + + LASSERT(cfs_list_empty(&lli->lli_close_list)); + CDEBUG(D_INODE, "adding inode %lu/%u to close list\n", + inode->i_ino, inode->i_generation); + cfs_list_add_tail(&lli->lli_close_list, &lcq->lcq_head); + + /* Avoid a concurrent insertion into the close thread queue: + * an inode is already in the close thread, open(), write(), + * close() happen, epoch is closed as the inode is marked as + * LLIF_EPOCH_PENDING. When pages are written inode should not + * be inserted into the queue again, clear this flag to avoid + * it. */ + lli->lli_flags &= ~LLIF_DONE_WRITING; + + cfs_waitq_signal(&lcq->lcq_waitq); + cfs_spin_unlock(&lcq->lcq_lock); + } + cfs_spin_unlock(&lli->lli_lock); + EXIT; } -void ll_try_done_writing(struct inode *inode) +/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */ +void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data) { struct ll_inode_info *lli = ll_i2info(inode); - struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq; - - spin_lock(&lli->lli_lock); - - if (lli->lli_send_done_writing && - list_empty(&lli->lli_pending_write_llaps)) { + ENTRY; - spin_lock(&lcq->lcq_lock); - if (list_empty(&lli->lli_close_item)) { - CDEBUG(D_INODE, "adding inode %lu/%u to close list\n", - inode->i_ino, inode->i_generation); - LASSERT(igrab(inode) == inode); - list_add_tail(&lli->lli_close_item, &lcq->lcq_list); - wake_up(&lcq->lcq_waitq); - } - spin_unlock(&lcq->lcq_lock); + op_data->op_flags |= MF_SOM_CHANGE; + /* Check if Size-on-MDS attributes are valid. */ + if (lli->lli_flags & LLIF_MDS_SIZE_LOCK) + CERROR("ino %lu/%u(flags %lu) som valid it just after " + "recovery\n", inode->i_ino, inode->i_generation, + lli->lli_flags); + + if (!cl_local_size(inode)) { + /* Send Size-on-MDS Attributes if valid. */ + op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET | + ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS; } - - spin_unlock(&lli->lli_lock); + EXIT; } -/* The MDS needs us to get the real file attributes, then send a DONE_WRITING */ -void ll_queue_done_writing(struct inode *inode) +/** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */ +void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data, + struct obd_client_handle **och, unsigned long flags) { struct ll_inode_info *lli = ll_i2info(inode); + struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob); ENTRY; - spin_lock(&lli->lli_lock); - lli->lli_send_done_writing = 1; - spin_unlock(&lli->lli_lock); + cfs_spin_lock(&lli->lli_lock); + if (!(cfs_list_empty(&club->cob_pending_list))) { + if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) { + LASSERT(*och != NULL); + LASSERT(lli->lli_pending_och == NULL); + /* Inode is dirty and there is no pending write done + * request yet, DONE_WRITE is to be sent later. */ + lli->lli_flags |= LLIF_EPOCH_PENDING; + lli->lli_pending_och = *och; + cfs_spin_unlock(&lli->lli_lock); + + inode = igrab(inode); + LASSERT(inode); + GOTO(out, 0); + } + if (flags & LLIF_DONE_WRITING) { + /* Some pages are still dirty, it is early to send + * DONE_WRITE. Wait untill all pages will be flushed + * and try DONE_WRITE again later. */ + LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING)); + lli->lli_flags |= LLIF_DONE_WRITING; + cfs_spin_unlock(&lli->lli_lock); + + inode = igrab(inode); + LASSERT(inode); + GOTO(out, 0); + } + } + CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n", + ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid)); + op_data->op_flags |= MF_EPOCH_CLOSE; + + if (flags & LLIF_DONE_WRITING) { + LASSERT(lli->lli_flags & LLIF_SOM_DIRTY); + LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING)); + *och = lli->lli_pending_och; + lli->lli_pending_och = NULL; + lli->lli_flags &= ~LLIF_EPOCH_PENDING; + } else { + /* Pack Size-on-MDS inode attributes only if they has changed */ + if (!(lli->lli_flags & LLIF_SOM_DIRTY)) { + cfs_spin_unlock(&lli->lli_lock); + GOTO(out, 0); + } + + /* There is a pending DONE_WRITE -- close epoch with no + * attribute change. */ + if (lli->lli_flags & LLIF_EPOCH_PENDING) { + cfs_spin_unlock(&lli->lli_lock); + GOTO(out, 0); + } + } + + LASSERT(cfs_list_empty(&club->cob_pending_list)); + lli->lli_flags &= ~LLIF_SOM_DIRTY; + cfs_spin_unlock(&lli->lli_lock); + ll_done_writing_attr(inode, op_data); - ll_try_done_writing(inode); EXIT; +out: + return; } -#if 0 -/* If we know the file size and have the cookies: - * - send a DONE_WRITING rpc - * - * Otherwise: - * - get a whole-file lock - * - get the authoritative size and all cookies with GETATTRs - * - send a DONE_WRITING rpc +/** + * Cliens updates SOM attributes on MDS (including llog cookies): + * obd_getattr with no lock and md_setattr. */ -static void ll_close_done_writing(struct inode *inode) +int ll_som_update(struct inode *inode, struct md_op_data *op_data) { struct ll_inode_info *lli = ll_i2info(inode); - ldlm_policy_data_t policy = { .l_extent = {0, OBD_OBJECT_EOF } }; - struct lustre_handle lockh = { 0 }; - struct obdo obdo; - obd_flag valid; - int rc, ast_flags = 0; + struct ptlrpc_request *request = NULL; + __u32 old_flags; + struct obdo *oa; + int rc; ENTRY; - memset(&obdo, 0, sizeof(obdo)); - if (test_bit(LLI_F_HAVE_OST_SIZE_LOCK, &lli->lli_flags)) - goto rpc; - - rc = ll_extent_lock(NULL, inode, lli->lli_smd, LCK_PW, &policy, &lockh, - ast_flags); - if (rc != 0) { - CERROR("lock acquisition failed (%d): unable to send " - "DONE_WRITING for inode %lu/%u\n", rc, inode->i_ino, - inode->i_generation); - GOTO(out, rc); + LASSERT(op_data != NULL); + if (lli->lli_flags & LLIF_MDS_SIZE_LOCK) + CERROR("ino %lu/%u(flags %lu) som valid it just after " + "recovery\n", inode->i_ino, inode->i_generation, + lli->lli_flags); + + OBDO_ALLOC(oa); + if (!oa) { + CERROR("can't allocate memory for Size-on-MDS update.\n"); + RETURN(-ENOMEM); } - rc = ll_lsm_getattr(ll_i2obdexp(inode), lli->lli_smd, &obdo); - if (rc) { - CERROR("inode_getattr failed (%d): unable to send DONE_WRITING " - "for inode %lu/%u\n", rc, inode->i_ino, - inode->i_generation); - ll_extent_unlock(NULL, inode, lli->lli_smd, LCK_PW, &lockh); - GOTO(out, rc); + old_flags = op_data->op_flags; + op_data->op_flags = MF_SOM_CHANGE; + + /* If inode is already in another epoch, skip getattr from OSTs. */ + if (lli->lli_ioepoch == op_data->op_ioepoch) { + rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch, + old_flags & MF_GETATTR_LOCK); + if (rc) { + oa->o_valid = 0; + if (rc == -ENOENT) + CDEBUG(D_INODE, "objid "LPX64" is destroyed\n", + lli->lli_smd->lsm_object_id); + else + CERROR("inode_getattr failed (%d): unable to " + "send a Size-on-MDS attribute update " + "for inode %lu/%u\n", rc, inode->i_ino, + inode->i_generation); + } else { + CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n", + PFID(&lli->lli_fid)); + } + /* Install attributes into op_data. */ + md_from_obdo(op_data, oa, oa->o_valid); } - obdo_refresh_inode(inode, &obdo, valid); + rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, + NULL, 0, NULL, 0, &request, NULL); + ptlrpc_req_finished(request); - CDEBUG(D_INODE, "objid "LPX64" size %Lu, blocks %lu, blksize %lu\n", - lli->lli_smd->lsm_object_id, inode->i_size, inode->i_blocks, - inode->i_blksize); + OBDO_FREE(oa); + RETURN(rc); +} - set_bit(LLI_F_HAVE_OST_SIZE_LOCK, &lli->lli_flags); +/** + * Closes the ioepoch and packs all the attributes into @op_data for + * DONE_WRITING rpc. + */ +static void ll_prepare_done_writing(struct inode *inode, + struct md_op_data *op_data, + struct obd_client_handle **och) +{ + ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING); + /* If there is no @och, we do not do D_W yet. */ + if (*och == NULL) + return; + + ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh); + ll_prep_md_op_data(op_data, inode, NULL, NULL, + 0, 0, LUSTRE_OPC_ANY, NULL); +} + +/** Send a DONE_WRITING rpc. */ +static void ll_done_writing(struct inode *inode) +{ + struct obd_client_handle *och = NULL; + struct md_op_data *op_data; + int rc; + ENTRY; - rc = ll_extent_unlock(NULL, inode, lli->lli_smd, LCK_PW, &lockh); - if (rc != ELDLM_OK) - CERROR("unlock failed (%d)? proceeding anyways...\n", rc); + LASSERT(exp_connect_som(ll_i2mdexp(inode))); - rpc: - obdo.o_id = inode->i_ino; - obdo.o_size = inode->i_size; - obdo.o_blocks = inode->i_blocks; - obdo.o_valid = OBD_MD_FLID | OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + OBD_ALLOC_PTR(op_data); + if (op_data == NULL) { + CERROR("can't allocate op_data\n"); + EXIT; + return; + } - rc = md_done_writing(ll_i2sbi(inode)->ll_mdc_exp, &obdo); - out: + ll_prepare_done_writing(inode, op_data, &och); + /* If there is no @och, we do not do D_W yet. */ + if (och == NULL) + GOTO(out, 0); + + rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL); + if (rc == -EAGAIN) { + /* MDS has instructed us to obtain Size-on-MDS attribute from + * OSTs and send setattr to back to MDS. */ + rc = ll_som_update(inode, op_data); + } else if (rc) { + CERROR("inode %lu mdc done_writing failed: rc = %d\n", + inode->i_ino, rc); + } +out: + ll_finish_md_op_data(op_data); + if (och) { + md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och); + OBD_FREE_PTR(och); + } + EXIT; } -#endif static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq) { struct ll_inode_info *lli = NULL; - spin_lock(&lcq->lcq_lock); + cfs_spin_lock(&lcq->lcq_lock); - if (lcq->lcq_list.next == NULL) - lli = ERR_PTR(-1); - else if (!list_empty(&lcq->lcq_list)) { - lli = list_entry(lcq->lcq_list.next, struct ll_inode_info, - lli_close_item); - list_del(&lli->lli_close_item); - } + if (!cfs_list_empty(&lcq->lcq_head)) { + lli = cfs_list_entry(lcq->lcq_head.next, struct ll_inode_info, + lli_close_list); + cfs_list_del_init(&lli->lli_close_list); + } else if (cfs_atomic_read(&lcq->lcq_stop)) + lli = ERR_PTR(-EALREADY); - spin_unlock(&lcq->lcq_lock); + cfs_spin_unlock(&lcq->lcq_lock); return lli; } @@ -196,24 +353,18 @@ static int ll_close_thread(void *arg) struct ll_close_queue *lcq = arg; ENTRY; - /* XXX boiler-plate */ { - char name[sizeof(current->comm)]; - unsigned long flags; + char name[CFS_CURPROC_COMM_MAX]; snprintf(name, sizeof(name) - 1, "ll_close"); - kportal_daemonize(name); - SIGNAL_MASK_LOCK(current, flags); - sigfillset(¤t->blocked); - RECALC_SIGPENDING; - SIGNAL_MASK_UNLOCK(current, flags); + cfs_daemonize(name); } - complete(&lcq->lcq_comp); + cfs_complete(&lcq->lcq_comp); while (1) { struct l_wait_info lwi = { 0 }; struct ll_inode_info *lli; - //struct inode *inode; + struct inode *inode; l_wait_event_exclusive(lcq->lcq_waitq, (lli = ll_close_next_lli(lcq)) != NULL, @@ -221,12 +372,15 @@ static int ll_close_thread(void *arg) if (IS_ERR(lli)) break; - //inode = ll_info2i(lli); - //ll_close_done_writing(inode); - //iput(inode); + inode = ll_info2i(lli); + CDEBUG(D_INFO, "done_writting for inode %lu/%u\n", + inode->i_ino, inode->i_generation); + ll_done_writing(inode); + iput(inode); } - complete(&lcq->lcq_comp); + CDEBUG(D_INFO, "ll_close exiting\n"); + cfs_complete(&lcq->lcq_comp); RETURN(0); } @@ -235,31 +389,34 @@ int ll_close_thread_start(struct ll_close_queue **lcq_ret) struct ll_close_queue *lcq; pid_t pid; + if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD)) + return -EINTR; + OBD_ALLOC(lcq, sizeof(*lcq)); if (lcq == NULL) return -ENOMEM; - spin_lock_init(&lcq->lcq_lock); - INIT_LIST_HEAD(&lcq->lcq_list); - init_waitqueue_head(&lcq->lcq_waitq); - init_completion(&lcq->lcq_comp); + cfs_spin_lock_init(&lcq->lcq_lock); + CFS_INIT_LIST_HEAD(&lcq->lcq_head); + cfs_waitq_init(&lcq->lcq_waitq); + cfs_init_completion(&lcq->lcq_comp); - pid = kernel_thread(ll_close_thread, lcq, 0); + pid = cfs_kernel_thread(ll_close_thread, lcq, 0); if (pid < 0) { OBD_FREE(lcq, sizeof(*lcq)); return pid; } - wait_for_completion(&lcq->lcq_comp); + cfs_wait_for_completion(&lcq->lcq_comp); *lcq_ret = lcq; return 0; } void ll_close_thread_shutdown(struct ll_close_queue *lcq) { - init_completion(&lcq->lcq_comp); - lcq->lcq_list.next = NULL; - wake_up(&lcq->lcq_waitq); - wait_for_completion(&lcq->lcq_comp); + cfs_init_completion(&lcq->lcq_comp); + cfs_atomic_inc(&lcq->lcq_stop); + cfs_waitq_signal(&lcq->lcq_waitq); + cfs_wait_for_completion(&lcq->lcq_comp); OBD_FREE(lcq, sizeof(*lcq)); }