1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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 <lustre_mdc.h>
46 #include <lustre_lite.h>
47 #include "llite_internal.h"
49 /** records that a write is in flight */
50 void vvp_write_pending(struct ccc_object *club, struct ccc_page *page)
52 struct ll_inode_info *lli = ll_i2info(club->cob_inode);
55 spin_lock(&lli->lli_lock);
56 lli->lli_flags |= LLIF_SOM_DIRTY;
57 if (page != NULL && list_empty(&page->cpg_pending_linkage))
58 list_add(&page->cpg_pending_linkage, &club->cob_pending_list);
59 spin_unlock(&lli->lli_lock);
63 /** records that a write has completed */
64 void vvp_write_complete(struct ccc_object *club, struct ccc_page *page)
66 struct ll_inode_info *lli = ll_i2info(club->cob_inode);
70 spin_lock(&lli->lli_lock);
71 if (page != NULL && !list_empty(&page->cpg_pending_linkage)) {
72 list_del_init(&page->cpg_pending_linkage);
75 spin_unlock(&lli->lli_lock);
77 ll_queue_done_writing(club->cob_inode, 0);
81 /** Queues DONE_WRITING if
82 * - done writing is allowed;
83 * - inode has no no dirty pages; */
84 void ll_queue_done_writing(struct inode *inode, unsigned long flags)
86 struct ll_inode_info *lli = ll_i2info(inode);
87 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
90 spin_lock(&lli->lli_lock);
91 lli->lli_flags |= flags;
93 if ((lli->lli_flags & LLIF_DONE_WRITING) &&
94 list_empty(&club->cob_pending_list)) {
95 struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
97 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
98 CWARN("ino %lu/%u(flags %lu) som valid it just after "
100 inode->i_ino, inode->i_generation,
102 /* DONE_WRITING is allowed and inode has no dirty page. */
103 spin_lock(&lcq->lcq_lock);
105 LASSERT(list_empty(&lli->lli_close_list));
106 CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
107 inode->i_ino, inode->i_generation);
108 list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
110 /* Avoid a concurrent insertion into the close thread queue:
111 * an inode is already in the close thread, open(), write(),
112 * close() happen, epoch is closed as the inode is marked as
113 * LLIF_EPOCH_PENDING. When pages are written inode should not
114 * be inserted into the queue again, clear this flag to avoid
116 lli->lli_flags &= ~LLIF_DONE_WRITING;
118 wake_up(&lcq->lcq_waitq);
119 spin_unlock(&lcq->lcq_lock);
121 spin_unlock(&lli->lli_lock);
125 void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
127 struct ll_inode_info *lli = ll_i2info(inode);
130 op_data->op_flags |= MF_SOM_CHANGE;
131 /* Check if Size-on-MDS attributes are valid. */
132 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
133 CERROR("ino %lu/%u(flags %lu) som valid it just after "
134 "recovery\n", inode->i_ino, inode->i_generation,
137 if (!cl_local_size(inode)) {
138 /* Send Size-on-MDS Attributes if valid. */
139 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
140 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
145 /** Closes epoch and sends Size-on-MDS attribute update if possible. Call
146 * this under ll_inode_info::lli_lock spinlock. */
147 void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
148 struct obd_client_handle **och, unsigned long flags)
150 struct ll_inode_info *lli = ll_i2info(inode);
151 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
154 spin_lock(&lli->lli_lock);
155 if (!(list_empty(&club->cob_pending_list))) {
156 if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
157 LASSERT(*och != NULL);
158 LASSERT(lli->lli_pending_och == NULL);
159 /* Inode is dirty and there is no pending write done
160 * request yet, DONE_WRITE is to be sent later. */
161 lli->lli_flags |= LLIF_EPOCH_PENDING;
162 lli->lli_pending_och = *och;
163 spin_unlock(&lli->lli_lock);
165 inode = igrab(inode);
169 if (flags & LLIF_DONE_WRITING) {
170 /* Some pages are still dirty, it is early to send
171 * DONE_WRITE. Wait untill all pages will be flushed
172 * and try DONE_WRITE again later. */
173 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
174 lli->lli_flags |= LLIF_DONE_WRITING;
175 spin_unlock(&lli->lli_lock);
177 inode = igrab(inode);
182 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n",
183 ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
184 op_data->op_flags |= MF_EPOCH_CLOSE;
186 if (flags & LLIF_DONE_WRITING) {
187 LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
188 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
189 *och = lli->lli_pending_och;
190 lli->lli_pending_och = NULL;
191 lli->lli_flags &= ~LLIF_EPOCH_PENDING;
193 /* Pack Size-on-MDS inode attributes only if they has changed */
194 if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
195 spin_unlock(&lli->lli_lock);
199 /* There is a pending DONE_WRITE -- close epoch with no
200 * attribute change. */
201 if (lli->lli_flags & LLIF_EPOCH_PENDING) {
202 spin_unlock(&lli->lli_lock);
207 LASSERT(list_empty(&club->cob_pending_list));
208 lli->lli_flags &= ~LLIF_SOM_DIRTY;
209 spin_unlock(&lli->lli_lock);
210 ll_done_writing_attr(inode, op_data);
217 int ll_sizeonmds_update(struct inode *inode, struct lustre_handle *fh,
220 struct ll_inode_info *lli = ll_i2info(inode);
221 struct md_op_data *op_data;
226 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
227 CERROR("ino %lu/%u(flags %lu) som valid it just after "
228 "recovery\n", inode->i_ino, inode->i_generation,
232 OBD_ALLOC_PTR(op_data);
233 if (!oa || !op_data) {
234 CERROR("can't allocate memory for Size-on-MDS update.\n");
237 /* If inode is already in another epoch, skip getattr from OSTs. */
238 if (lli->lli_ioepoch == ioepoch) {
239 rc = ll_inode_getattr(inode, oa);
242 CDEBUG(D_INODE, "objid "LPX64" is already destroyed\n",
243 lli->lli_smd->lsm_object_id);
245 CERROR("inode_getattr failed (%d): unable to send a "
246 "Size-on-MDS attribute update for inode "
247 "%lu/%u\n", rc, inode->i_ino,
248 inode->i_generation);
251 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
252 PFID(&lli->lli_fid));
253 md_from_obdo(op_data, oa, oa->o_valid);
255 memcpy(&op_data->op_handle, fh, sizeof(*fh));
257 op_data->op_ioepoch = ioepoch;
258 op_data->op_flags |= MF_SOM_CHANGE;
260 rc = ll_md_setattr(inode, op_data, NULL);
266 ll_finish_md_op_data(op_data);
270 /** Sends a DONE_WRITING rpc, packs Size-on-MDS attributes into it, if
272 static void ll_done_writing(struct inode *inode)
274 struct obd_client_handle *och = NULL;
275 struct md_op_data *op_data;
279 LASSERT(exp_connect_som(ll_i2mdexp(inode)));
281 OBD_ALLOC_PTR(op_data);
282 if (op_data == NULL) {
283 CERROR("can't allocate op_data\n");
288 ll_ioepoch_close(inode, op_data, &och, LLIF_DONE_WRITING);
289 /* If there is no @och, we do not do D_W yet. */
293 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
295 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
297 /* MDS has instructed us to obtain Size-on-MDS attribute from
298 * OSTs and send setattr to back to MDS. */
299 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
300 op_data->op_ioepoch);
302 CERROR("inode %lu mdc done_writing failed: rc = %d\n",
306 ll_finish_md_op_data(op_data);
308 md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
314 static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
316 struct ll_inode_info *lli = NULL;
318 spin_lock(&lcq->lcq_lock);
320 if (!list_empty(&lcq->lcq_head)) {
321 lli = list_entry(lcq->lcq_head.next, struct ll_inode_info,
323 list_del_init(&lli->lli_close_list);
324 } else if (atomic_read(&lcq->lcq_stop))
325 lli = ERR_PTR(-EALREADY);
327 spin_unlock(&lcq->lcq_lock);
331 static int ll_close_thread(void *arg)
333 struct ll_close_queue *lcq = arg;
337 char name[CFS_CURPROC_COMM_MAX];
338 snprintf(name, sizeof(name) - 1, "ll_close");
342 complete(&lcq->lcq_comp);
345 struct l_wait_info lwi = { 0 };
346 struct ll_inode_info *lli;
349 l_wait_event_exclusive(lcq->lcq_waitq,
350 (lli = ll_close_next_lli(lcq)) != NULL,
355 inode = ll_info2i(lli);
356 CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
357 inode->i_ino, inode->i_generation);
358 ll_done_writing(inode);
362 CDEBUG(D_INFO, "ll_close exiting\n");
363 complete(&lcq->lcq_comp);
367 int ll_close_thread_start(struct ll_close_queue **lcq_ret)
369 struct ll_close_queue *lcq;
372 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
375 OBD_ALLOC(lcq, sizeof(*lcq));
379 spin_lock_init(&lcq->lcq_lock);
380 INIT_LIST_HEAD(&lcq->lcq_head);
381 init_waitqueue_head(&lcq->lcq_waitq);
382 init_completion(&lcq->lcq_comp);
384 pid = kernel_thread(ll_close_thread, lcq, 0);
386 OBD_FREE(lcq, sizeof(*lcq));
390 wait_for_completion(&lcq->lcq_comp);
395 void ll_close_thread_shutdown(struct ll_close_queue *lcq)
397 init_completion(&lcq->lcq_comp);
398 atomic_inc(&lcq->lcq_stop);
399 wake_up(&lcq->lcq_waitq);
400 wait_for_completion(&lcq->lcq_comp);
401 OBD_FREE(lcq, sizeof(*lcq));