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 (c) 2003, 2010, Oracle and/or its affiliates. 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_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);
54 cfs_spin_lock(&lli->lli_lock);
55 lli->lli_flags |= LLIF_SOM_DIRTY;
56 if (page != NULL && cfs_list_empty(&page->cpg_pending_linkage))
57 cfs_list_add(&page->cpg_pending_linkage,
58 &club->cob_pending_list);
59 cfs_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 cfs_spin_lock(&lli->lli_lock);
71 if (page != NULL && !cfs_list_empty(&page->cpg_pending_linkage)) {
72 cfs_list_del_init(&page->cpg_pending_linkage);
75 cfs_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 cfs_spin_lock(&lli->lli_lock);
91 lli->lli_flags |= flags;
93 if ((lli->lli_flags & LLIF_DONE_WRITING) &&
94 cfs_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 cfs_spin_lock(&lcq->lcq_lock);
105 LASSERT(cfs_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 cfs_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 cfs_waitq_signal(&lcq->lcq_waitq);
119 cfs_spin_unlock(&lcq->lcq_lock);
121 cfs_spin_unlock(&lli->lli_lock);
125 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
126 void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
128 struct ll_inode_info *lli = ll_i2info(inode);
131 op_data->op_flags |= MF_SOM_CHANGE;
132 /* Check if Size-on-MDS attributes are valid. */
133 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
134 CERROR("ino %lu/%u(flags %lu) som valid it just after "
135 "recovery\n", inode->i_ino, inode->i_generation,
138 if (!cl_local_size(inode)) {
139 /* Send Size-on-MDS Attributes if valid. */
140 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
141 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
146 /** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
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 cfs_spin_lock(&lli->lli_lock);
155 if (!(cfs_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 cfs_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 cfs_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 cfs_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 cfs_spin_unlock(&lli->lli_lock);
207 LASSERT(cfs_list_empty(&club->cob_pending_list));
208 lli->lli_flags &= ~LLIF_SOM_DIRTY;
209 cfs_spin_unlock(&lli->lli_lock);
210 ll_done_writing_attr(inode, op_data);
218 * Cliens updates SOM attributes on MDS (including llog cookies):
219 * obd_getattr with no lock and md_setattr.
221 int ll_som_update(struct inode *inode, struct md_op_data *op_data)
223 struct ll_inode_info *lli = ll_i2info(inode);
224 struct ptlrpc_request *request = NULL;
230 LASSERT(op_data != NULL);
231 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
232 CERROR("ino %lu/%u(flags %lu) som valid it just after "
233 "recovery\n", inode->i_ino, inode->i_generation,
238 CERROR("can't allocate memory for Size-on-MDS update.\n");
242 old_flags = op_data->op_flags;
243 op_data->op_flags = MF_SOM_CHANGE;
245 /* If inode is already in another epoch, skip getattr from OSTs. */
246 if (lli->lli_ioepoch == op_data->op_ioepoch) {
247 rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
248 old_flags & MF_GETATTR_LOCK);
252 CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
253 lli->lli_smd->lsm_object_id);
255 CERROR("inode_getattr failed (%d): unable to "
256 "send a Size-on-MDS attribute update "
257 "for inode %lu/%u\n", rc, inode->i_ino,
258 inode->i_generation);
260 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
261 PFID(&lli->lli_fid));
263 /* Install attributes into op_data. */
264 md_from_obdo(op_data, oa, oa->o_valid);
267 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
268 NULL, 0, NULL, 0, &request, NULL);
269 ptlrpc_req_finished(request);
276 * Closes the ioepoch and packs all the attributes into @op_data for
279 static void ll_prepare_done_writing(struct inode *inode,
280 struct md_op_data *op_data,
281 struct obd_client_handle **och)
283 ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING);
284 /* If there is no @och, we do not do D_W yet. */
288 ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh);
289 ll_prep_md_op_data(op_data, inode, NULL, NULL,
290 0, 0, LUSTRE_OPC_ANY, NULL);
293 /** Send a DONE_WRITING rpc. */
294 static void ll_done_writing(struct inode *inode)
296 struct obd_client_handle *och = NULL;
297 struct md_op_data *op_data;
301 LASSERT(exp_connect_som(ll_i2mdexp(inode)));
303 OBD_ALLOC_PTR(op_data);
304 if (op_data == NULL) {
305 CERROR("can't allocate op_data\n");
310 ll_prepare_done_writing(inode, op_data, &och);
311 /* If there is no @och, we do not do D_W yet. */
315 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
317 /* MDS has instructed us to obtain Size-on-MDS attribute from
318 * OSTs and send setattr to back to MDS. */
319 rc = ll_som_update(inode, op_data);
321 CERROR("inode %lu mdc done_writing failed: rc = %d\n",
325 ll_finish_md_op_data(op_data);
327 md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
333 static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
335 struct ll_inode_info *lli = NULL;
337 cfs_spin_lock(&lcq->lcq_lock);
339 if (!cfs_list_empty(&lcq->lcq_head)) {
340 lli = cfs_list_entry(lcq->lcq_head.next, struct ll_inode_info,
342 cfs_list_del_init(&lli->lli_close_list);
343 } else if (cfs_atomic_read(&lcq->lcq_stop))
344 lli = ERR_PTR(-EALREADY);
346 cfs_spin_unlock(&lcq->lcq_lock);
350 static int ll_close_thread(void *arg)
352 struct ll_close_queue *lcq = arg;
356 char name[CFS_CURPROC_COMM_MAX];
357 snprintf(name, sizeof(name) - 1, "ll_close");
361 cfs_complete(&lcq->lcq_comp);
364 struct l_wait_info lwi = { 0 };
365 struct ll_inode_info *lli;
368 l_wait_event_exclusive(lcq->lcq_waitq,
369 (lli = ll_close_next_lli(lcq)) != NULL,
374 inode = ll_info2i(lli);
375 CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
376 inode->i_ino, inode->i_generation);
377 ll_done_writing(inode);
381 CDEBUG(D_INFO, "ll_close exiting\n");
382 cfs_complete(&lcq->lcq_comp);
386 int ll_close_thread_start(struct ll_close_queue **lcq_ret)
388 struct ll_close_queue *lcq;
391 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
394 OBD_ALLOC(lcq, sizeof(*lcq));
398 cfs_spin_lock_init(&lcq->lcq_lock);
399 CFS_INIT_LIST_HEAD(&lcq->lcq_head);
400 cfs_waitq_init(&lcq->lcq_waitq);
401 cfs_init_completion(&lcq->lcq_comp);
403 pid = cfs_create_thread(ll_close_thread, lcq, 0);
405 OBD_FREE(lcq, sizeof(*lcq));
409 cfs_wait_for_completion(&lcq->lcq_comp);
414 void ll_close_thread_shutdown(struct ll_close_queue *lcq)
416 cfs_init_completion(&lcq->lcq_comp);
417 cfs_atomic_inc(&lcq->lcq_stop);
418 cfs_waitq_signal(&lcq->lcq_waitq);
419 cfs_wait_for_completion(&lcq->lcq_comp);
420 OBD_FREE(lcq, sizeof(*lcq));