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 cfs_spin_lock(&lli->lli_lock);
56 lli->lli_flags |= LLIF_SOM_DIRTY;
57 if (page != NULL && cfs_list_empty(&page->cpg_pending_linkage))
58 cfs_list_add(&page->cpg_pending_linkage,
59 &club->cob_pending_list);
60 cfs_spin_unlock(&lli->lli_lock);
64 /** records that a write has completed */
65 void vvp_write_complete(struct ccc_object *club, struct ccc_page *page)
67 struct ll_inode_info *lli = ll_i2info(club->cob_inode);
71 cfs_spin_lock(&lli->lli_lock);
72 if (page != NULL && !cfs_list_empty(&page->cpg_pending_linkage)) {
73 cfs_list_del_init(&page->cpg_pending_linkage);
76 cfs_spin_unlock(&lli->lli_lock);
78 ll_queue_done_writing(club->cob_inode, 0);
82 /** Queues DONE_WRITING if
83 * - done writing is allowed;
84 * - inode has no no dirty pages; */
85 void ll_queue_done_writing(struct inode *inode, unsigned long flags)
87 struct ll_inode_info *lli = ll_i2info(inode);
88 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
91 cfs_spin_lock(&lli->lli_lock);
92 lli->lli_flags |= flags;
94 if ((lli->lli_flags & LLIF_DONE_WRITING) &&
95 cfs_list_empty(&club->cob_pending_list)) {
96 struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
98 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
99 CWARN("ino %lu/%u(flags %lu) som valid it just after "
101 inode->i_ino, inode->i_generation,
103 /* DONE_WRITING is allowed and inode has no dirty page. */
104 cfs_spin_lock(&lcq->lcq_lock);
106 LASSERT(cfs_list_empty(&lli->lli_close_list));
107 CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
108 inode->i_ino, inode->i_generation);
109 cfs_list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
111 /* Avoid a concurrent insertion into the close thread queue:
112 * an inode is already in the close thread, open(), write(),
113 * close() happen, epoch is closed as the inode is marked as
114 * LLIF_EPOCH_PENDING. When pages are written inode should not
115 * be inserted into the queue again, clear this flag to avoid
117 lli->lli_flags &= ~LLIF_DONE_WRITING;
119 cfs_waitq_signal(&lcq->lcq_waitq);
120 cfs_spin_unlock(&lcq->lcq_lock);
122 cfs_spin_unlock(&lli->lli_lock);
126 /** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
127 void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
129 struct ll_inode_info *lli = ll_i2info(inode);
132 op_data->op_flags |= MF_SOM_CHANGE;
133 /* Check if Size-on-MDS attributes are valid. */
134 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
135 CERROR("ino %lu/%u(flags %lu) som valid it just after "
136 "recovery\n", inode->i_ino, inode->i_generation,
139 if (!cl_local_size(inode)) {
140 /* Send Size-on-MDS Attributes if valid. */
141 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
142 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
147 /** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
148 void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
149 struct obd_client_handle **och, unsigned long flags)
151 struct ll_inode_info *lli = ll_i2info(inode);
152 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
155 cfs_spin_lock(&lli->lli_lock);
156 if (!(cfs_list_empty(&club->cob_pending_list))) {
157 if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
158 LASSERT(*och != NULL);
159 LASSERT(lli->lli_pending_och == NULL);
160 /* Inode is dirty and there is no pending write done
161 * request yet, DONE_WRITE is to be sent later. */
162 lli->lli_flags |= LLIF_EPOCH_PENDING;
163 lli->lli_pending_och = *och;
164 cfs_spin_unlock(&lli->lli_lock);
166 inode = igrab(inode);
170 if (flags & LLIF_DONE_WRITING) {
171 /* Some pages are still dirty, it is early to send
172 * DONE_WRITE. Wait untill all pages will be flushed
173 * and try DONE_WRITE again later. */
174 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
175 lli->lli_flags |= LLIF_DONE_WRITING;
176 cfs_spin_unlock(&lli->lli_lock);
178 inode = igrab(inode);
183 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n",
184 ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
185 op_data->op_flags |= MF_EPOCH_CLOSE;
187 if (flags & LLIF_DONE_WRITING) {
188 LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
189 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
190 *och = lli->lli_pending_och;
191 lli->lli_pending_och = NULL;
192 lli->lli_flags &= ~LLIF_EPOCH_PENDING;
194 /* Pack Size-on-MDS inode attributes only if they has changed */
195 if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
196 cfs_spin_unlock(&lli->lli_lock);
200 /* There is a pending DONE_WRITE -- close epoch with no
201 * attribute change. */
202 if (lli->lli_flags & LLIF_EPOCH_PENDING) {
203 cfs_spin_unlock(&lli->lli_lock);
208 LASSERT(cfs_list_empty(&club->cob_pending_list));
209 lli->lli_flags &= ~LLIF_SOM_DIRTY;
210 cfs_spin_unlock(&lli->lli_lock);
211 ll_done_writing_attr(inode, op_data);
219 * Cliens updates SOM attributes on MDS (including llog cookies):
220 * obd_getattr with no lock and md_setattr.
222 int ll_som_update(struct inode *inode, struct md_op_data *op_data)
224 struct ll_inode_info *lli = ll_i2info(inode);
225 struct ptlrpc_request *request = NULL;
231 LASSERT(op_data != NULL);
232 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
233 CERROR("ino %lu/%u(flags %lu) som valid it just after "
234 "recovery\n", inode->i_ino, inode->i_generation,
239 CERROR("can't allocate memory for Size-on-MDS update.\n");
243 old_flags = op_data->op_flags;
244 op_data->op_flags = MF_SOM_CHANGE;
246 /* If inode is already in another epoch, skip getattr from OSTs. */
247 if (lli->lli_ioepoch == op_data->op_ioepoch) {
248 rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
249 old_flags & MF_GETATTR_LOCK);
253 CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
254 lli->lli_smd->lsm_object_id);
256 CERROR("inode_getattr failed (%d): unable to "
257 "send a Size-on-MDS attribute update "
258 "for inode %lu/%u\n", rc, inode->i_ino,
259 inode->i_generation);
261 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
262 PFID(&lli->lli_fid));
264 /* Install attributes into op_data. */
265 md_from_obdo(op_data, oa, oa->o_valid);
268 rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
269 NULL, 0, NULL, 0, &request, NULL);
270 ptlrpc_req_finished(request);
277 * Closes the ioepoch and packs all the attributes into @op_data for
280 static void ll_prepare_done_writing(struct inode *inode,
281 struct md_op_data *op_data,
282 struct obd_client_handle **och)
284 ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING);
285 /* If there is no @och, we do not do D_W yet. */
289 ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh);
290 ll_prep_md_op_data(op_data, inode, NULL, NULL,
291 0, 0, LUSTRE_OPC_ANY, NULL);
294 /** Send a DONE_WRITING rpc. */
295 static void ll_done_writing(struct inode *inode)
297 struct obd_client_handle *och = NULL;
298 struct md_op_data *op_data;
302 LASSERT(exp_connect_som(ll_i2mdexp(inode)));
304 OBD_ALLOC_PTR(op_data);
305 if (op_data == NULL) {
306 CERROR("can't allocate op_data\n");
311 ll_prepare_done_writing(inode, op_data, &och);
312 /* If there is no @och, we do not do D_W yet. */
316 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
318 /* MDS has instructed us to obtain Size-on-MDS attribute from
319 * OSTs and send setattr to back to MDS. */
320 rc = ll_som_update(inode, op_data);
322 CERROR("inode %lu mdc done_writing failed: rc = %d\n",
326 ll_finish_md_op_data(op_data);
328 md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
334 static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
336 struct ll_inode_info *lli = NULL;
338 cfs_spin_lock(&lcq->lcq_lock);
340 if (!cfs_list_empty(&lcq->lcq_head)) {
341 lli = cfs_list_entry(lcq->lcq_head.next, struct ll_inode_info,
343 cfs_list_del_init(&lli->lli_close_list);
344 } else if (cfs_atomic_read(&lcq->lcq_stop))
345 lli = ERR_PTR(-EALREADY);
347 cfs_spin_unlock(&lcq->lcq_lock);
351 static int ll_close_thread(void *arg)
353 struct ll_close_queue *lcq = arg;
357 char name[CFS_CURPROC_COMM_MAX];
358 snprintf(name, sizeof(name) - 1, "ll_close");
362 cfs_complete(&lcq->lcq_comp);
365 struct l_wait_info lwi = { 0 };
366 struct ll_inode_info *lli;
369 l_wait_event_exclusive(lcq->lcq_waitq,
370 (lli = ll_close_next_lli(lcq)) != NULL,
375 inode = ll_info2i(lli);
376 CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
377 inode->i_ino, inode->i_generation);
378 ll_done_writing(inode);
382 CDEBUG(D_INFO, "ll_close exiting\n");
383 cfs_complete(&lcq->lcq_comp);
387 int ll_close_thread_start(struct ll_close_queue **lcq_ret)
389 struct ll_close_queue *lcq;
392 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
395 OBD_ALLOC(lcq, sizeof(*lcq));
399 cfs_spin_lock_init(&lcq->lcq_lock);
400 CFS_INIT_LIST_HEAD(&lcq->lcq_head);
401 cfs_waitq_init(&lcq->lcq_waitq);
402 cfs_init_completion(&lcq->lcq_comp);
404 pid = cfs_kernel_thread(ll_close_thread, lcq, 0);
406 OBD_FREE(lcq, sizeof(*lcq));
410 cfs_wait_for_completion(&lcq->lcq_comp);
415 void ll_close_thread_shutdown(struct ll_close_queue *lcq)
417 cfs_init_completion(&lcq->lcq_comp);
418 cfs_atomic_inc(&lcq->lcq_stop);
419 cfs_waitq_signal(&lcq->lcq_waitq);
420 cfs_wait_for_completion(&lcq->lcq_comp);
421 OBD_FREE(lcq, sizeof(*lcq));