4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/llite/llite_close.c
38 * Lustre Lite routines to issue a secondary close after writeback
41 #include <linux/module.h>
43 #define DEBUG_SUBSYSTEM S_LLITE
45 #include "../include/lustre_lite.h"
46 #include "llite_internal.h"
48 /** records that a write is in flight */
49 void vvp_write_pending(struct ccc_object *club, struct ccc_page *page)
51 struct ll_inode_info *lli = ll_i2info(club->cob_inode);
53 spin_lock(&lli->lli_lock);
54 lli->lli_flags |= LLIF_SOM_DIRTY;
55 if (page && list_empty(&page->cpg_pending_linkage))
56 list_add(&page->cpg_pending_linkage,
57 &club->cob_pending_list);
58 spin_unlock(&lli->lli_lock);
61 /** records that a write has completed */
62 void vvp_write_complete(struct ccc_object *club, struct ccc_page *page)
64 struct ll_inode_info *lli = ll_i2info(club->cob_inode);
67 spin_lock(&lli->lli_lock);
68 if (page && !list_empty(&page->cpg_pending_linkage)) {
69 list_del_init(&page->cpg_pending_linkage);
72 spin_unlock(&lli->lli_lock);
74 ll_queue_done_writing(club->cob_inode, 0);
77 /** Queues DONE_WRITING if
78 * - done writing is allowed;
79 * - inode has no no dirty pages;
81 void ll_queue_done_writing(struct inode *inode, unsigned long flags)
83 struct ll_inode_info *lli = ll_i2info(inode);
84 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
86 spin_lock(&lli->lli_lock);
87 lli->lli_flags |= flags;
89 if ((lli->lli_flags & LLIF_DONE_WRITING) &&
90 list_empty(&club->cob_pending_list)) {
91 struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
93 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
94 CWARN("ino %lu/%u(flags %u) som valid it just after recovery\n",
95 inode->i_ino, inode->i_generation,
97 /* DONE_WRITING is allowed and inode has no dirty page. */
98 spin_lock(&lcq->lcq_lock);
100 LASSERT(list_empty(&lli->lli_close_list));
101 CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
102 inode->i_ino, inode->i_generation);
103 list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
105 /* Avoid a concurrent insertion into the close thread queue:
106 * an inode is already in the close thread, open(), write(),
107 * close() happen, epoch is closed as the inode is marked as
108 * LLIF_EPOCH_PENDING. When pages are written inode should not
109 * be inserted into the queue again, clear this flag to avoid
112 lli->lli_flags &= ~LLIF_DONE_WRITING;
114 wake_up(&lcq->lcq_waitq);
115 spin_unlock(&lcq->lcq_lock);
117 spin_unlock(&lli->lli_lock);
120 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
121 void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
123 struct ll_inode_info *lli = ll_i2info(inode);
125 op_data->op_flags |= MF_SOM_CHANGE;
126 /* Check if Size-on-MDS attributes are valid. */
127 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
128 CERROR("ino %lu/%u(flags %u) som valid it just after recovery\n",
129 inode->i_ino, inode->i_generation,
132 if (!cl_local_size(inode)) {
133 /* Send Size-on-MDS Attributes if valid. */
134 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
135 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
139 /** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
140 void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
141 struct obd_client_handle **och, unsigned long flags)
143 struct ll_inode_info *lli = ll_i2info(inode);
144 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
146 spin_lock(&lli->lli_lock);
147 if (!(list_empty(&club->cob_pending_list))) {
148 if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
150 LASSERT(!lli->lli_pending_och);
151 /* Inode is dirty and there is no pending write done
152 * request yet, DONE_WRITE is to be sent later.
154 lli->lli_flags |= LLIF_EPOCH_PENDING;
155 lli->lli_pending_och = *och;
156 spin_unlock(&lli->lli_lock);
158 inode = igrab(inode);
162 if (flags & LLIF_DONE_WRITING) {
163 /* Some pages are still dirty, it is early to send
164 * DONE_WRITE. Wait until all pages will be flushed
165 * and try DONE_WRITE again later.
167 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
168 lli->lli_flags |= LLIF_DONE_WRITING;
169 spin_unlock(&lli->lli_lock);
171 inode = igrab(inode);
176 CDEBUG(D_INODE, "Epoch %llu closed on "DFID"\n",
177 ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
178 op_data->op_flags |= MF_EPOCH_CLOSE;
180 if (flags & LLIF_DONE_WRITING) {
181 LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
182 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
183 *och = lli->lli_pending_och;
184 lli->lli_pending_och = NULL;
185 lli->lli_flags &= ~LLIF_EPOCH_PENDING;
187 /* Pack Size-on-MDS inode attributes only if they has changed */
188 if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
189 spin_unlock(&lli->lli_lock);
193 /* There is a pending DONE_WRITE -- close epoch with no
196 if (lli->lli_flags & LLIF_EPOCH_PENDING) {
197 spin_unlock(&lli->lli_lock);
202 LASSERT(list_empty(&club->cob_pending_list));
203 lli->lli_flags &= ~LLIF_SOM_DIRTY;
204 spin_unlock(&lli->lli_lock);
205 ll_done_writing_attr(inode, op_data);
212 * Cliens updates SOM attributes on MDS (including llog cookies):
213 * obd_getattr with no lock and md_setattr.
215 int ll_som_update(struct inode *inode, struct md_op_data *op_data)
217 struct ll_inode_info *lli = ll_i2info(inode);
218 struct ptlrpc_request *request = NULL;
224 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
225 CERROR("ino %lu/%u(flags %u) som valid it just after recovery\n",
226 inode->i_ino, inode->i_generation,
229 oa = kmem_cache_alloc(obdo_cachep, GFP_NOFS | __GFP_ZERO);
231 CERROR("can't allocate memory for Size-on-MDS update.\n");
235 old_flags = op_data->op_flags;
236 op_data->op_flags = MF_SOM_CHANGE;
238 /* If inode is already in another epoch, skip getattr from OSTs. */
239 if (lli->lli_ioepoch == op_data->op_ioepoch) {
240 rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
241 old_flags & MF_GETATTR_LOCK);
245 CERROR("inode_getattr failed (%d): unable to send a Size-on-MDS attribute update for inode %lu/%u\n",
247 inode->i_generation);
249 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
250 PFID(&lli->lli_fid));
252 /* Install attributes into op_data. */
253 md_from_obdo(op_data, oa, oa->o_valid);
256 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
257 NULL, 0, NULL, 0, &request, NULL);
258 ptlrpc_req_finished(request);
260 kmem_cache_free(obdo_cachep, oa);
265 * Closes the ioepoch and packs all the attributes into @op_data for
268 static void ll_prepare_done_writing(struct inode *inode,
269 struct md_op_data *op_data,
270 struct obd_client_handle **och)
272 ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING);
273 /* If there is no @och, we do not do D_W yet. */
277 ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh);
278 ll_prep_md_op_data(op_data, inode, NULL, NULL,
279 0, 0, LUSTRE_OPC_ANY, NULL);
282 /** Send a DONE_WRITING rpc. */
283 static void ll_done_writing(struct inode *inode)
285 struct obd_client_handle *och = NULL;
286 struct md_op_data *op_data;
289 LASSERT(exp_connect_som(ll_i2mdexp(inode)));
291 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
295 ll_prepare_done_writing(inode, op_data, &och);
296 /* If there is no @och, we do not do D_W yet. */
300 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
302 /* MDS has instructed us to obtain Size-on-MDS attribute from
303 * OSTs and send setattr to back to MDS.
305 rc = ll_som_update(inode, op_data);
307 CERROR("inode %lu mdc done_writing failed: rc = %d\n",
310 ll_finish_md_op_data(op_data);
312 md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
317 static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
319 struct ll_inode_info *lli = NULL;
321 spin_lock(&lcq->lcq_lock);
323 if (!list_empty(&lcq->lcq_head)) {
324 lli = list_entry(lcq->lcq_head.next, struct ll_inode_info,
326 list_del_init(&lli->lli_close_list);
327 } else if (atomic_read(&lcq->lcq_stop))
328 lli = ERR_PTR(-EALREADY);
330 spin_unlock(&lcq->lcq_lock);
334 static int ll_close_thread(void *arg)
336 struct ll_close_queue *lcq = arg;
338 complete(&lcq->lcq_comp);
341 struct l_wait_info lwi = { 0 };
342 struct ll_inode_info *lli;
345 l_wait_event_exclusive(lcq->lcq_waitq,
346 (lli = ll_close_next_lli(lcq)) != NULL,
351 inode = ll_info2i(lli);
352 CDEBUG(D_INFO, "done_writing for inode %lu/%u\n",
353 inode->i_ino, inode->i_generation);
354 ll_done_writing(inode);
358 CDEBUG(D_INFO, "ll_close exiting\n");
359 complete(&lcq->lcq_comp);
363 int ll_close_thread_start(struct ll_close_queue **lcq_ret)
365 struct ll_close_queue *lcq;
366 struct task_struct *task;
368 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
371 lcq = kzalloc(sizeof(*lcq), GFP_NOFS);
375 spin_lock_init(&lcq->lcq_lock);
376 INIT_LIST_HEAD(&lcq->lcq_head);
377 init_waitqueue_head(&lcq->lcq_waitq);
378 init_completion(&lcq->lcq_comp);
380 task = kthread_run(ll_close_thread, lcq, "ll_close");
383 return PTR_ERR(task);
386 wait_for_completion(&lcq->lcq_comp);
391 void ll_close_thread_shutdown(struct ll_close_queue *lcq)
393 init_completion(&lcq->lcq_comp);
394 atomic_inc(&lcq->lcq_stop);
395 wake_up(&lcq->lcq_waitq);
396 wait_for_completion(&lcq->lcq_comp);