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);
88 spin_lock(&lli->lli_lock);
89 lli->lli_flags |= flags;
92 if ((lli->lli_flags & LLIF_DONE_WRITING) &&
93 list_empty(&club->cob_pending_list)) {
94 struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
96 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
97 CWARN("ino %lu/%u(flags %lu) som valid it just after "
99 inode->i_ino, inode->i_generation,
101 /* DONE_WRITING is allowed and inode has no dirty page. */
102 spin_lock(&lcq->lcq_lock);
104 LASSERT(list_empty(&lli->lli_close_list));
105 CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
106 inode->i_ino, inode->i_generation);
107 list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
109 /* Avoid a concurrent insertion into the close thread queue:
110 * an inode is already in the close thread, open(), write(),
111 * close() happen, epoch is closed as the inode is marked as
112 * LLIF_EPOCH_PENDING. When pages are written inode should not
113 * be inserted into the queue again, clear this flag to avoid
115 lli->lli_flags &= ~LLIF_DONE_WRITING;
117 wake_up(&lcq->lcq_waitq);
118 spin_unlock(&lcq->lcq_lock);
120 spin_unlock(&lli->lli_lock);
124 /** Closes epoch and sends Size-on-MDS attribute update if possible. Call
125 * this under ll_inode_info::lli_lock spinlock. */
126 void ll_epoch_close(struct inode *inode, struct md_op_data *op_data,
127 struct obd_client_handle **och, unsigned long flags)
129 struct ll_inode_info *lli = ll_i2info(inode);
130 struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
133 spin_lock(&lli->lli_lock);
134 if (!(list_empty(&club->cob_pending_list))) {
135 if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
136 LASSERT(*och != NULL);
137 LASSERT(lli->lli_pending_och == NULL);
138 /* Inode is dirty and there is no pending write done
139 * request yet, DONE_WRITE is to be sent later. */
140 lli->lli_flags |= LLIF_EPOCH_PENDING;
141 lli->lli_pending_och = *och;
142 spin_unlock(&lli->lli_lock);
144 inode = igrab(inode);
148 if (flags & LLIF_DONE_WRITING) {
149 /* Some pages are still dirty, it is early to send
150 * DONE_WRITE. Wait untill all pages will be flushed
151 * and try DONE_WRITE again later. */
152 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
153 lli->lli_flags |= LLIF_DONE_WRITING;
154 spin_unlock(&lli->lli_lock);
156 inode = igrab(inode);
161 CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n",
162 ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
163 op_data->op_flags |= MF_EPOCH_CLOSE;
165 if (flags & LLIF_DONE_WRITING) {
166 LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
167 LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
168 *och = lli->lli_pending_och;
169 lli->lli_pending_och = NULL;
170 lli->lli_flags &= ~LLIF_EPOCH_PENDING;
172 /* Pack Size-on-MDS inode attributes only if they has changed */
173 if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
174 spin_unlock(&lli->lli_lock);
178 /* There is a pending DONE_WRITE -- close epoch with no
179 * attribute change. */
180 if (lli->lli_flags & LLIF_EPOCH_PENDING) {
181 spin_unlock(&lli->lli_lock);
186 LASSERT(list_empty(&club->cob_pending_list));
187 lli->lli_flags &= ~LLIF_SOM_DIRTY;
188 spin_unlock(&lli->lli_lock);
189 op_data->op_flags |= MF_SOM_CHANGE;
191 /* Check if Size-on-MDS attributes are valid. */
192 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
193 CWARN("ino %lu/%u(flags %lu) som valid it just after "
195 inode->i_ino, inode->i_generation, lli->lli_flags);
197 if (!cl_local_size(inode)) {
198 /* Send Size-on-MDS Attributes if valid. Atime is sent along
199 * with all the attributes. */
200 op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
201 ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
208 int ll_sizeonmds_update(struct inode *inode, struct lustre_handle *fh,
211 struct ll_inode_info *lli = ll_i2info(inode);
212 struct md_op_data *op_data;
217 /* LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK)); */
218 /* After recovery that can be valid. */
219 if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
220 CWARN("ino %lu/%u(flags %lu) som valid it just after "
221 "recovery\n", inode->i_ino, inode->i_generation,
225 OBD_ALLOC_PTR(op_data);
226 if (!oa || !op_data) {
227 CERROR("can't allocate memory for Size-on-MDS update.\n");
230 rc = ll_inode_getattr(inode, oa);
233 CDEBUG(D_INODE, "objid "LPX64" is already destroyed\n",
234 lli->lli_smd->lsm_object_id);
236 CERROR("inode_getattr failed (%d): unable to send a "
237 "Size-on-MDS attribute update for inode %lu/%u\n",
238 rc, inode->i_ino, inode->i_generation);
241 CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n", PFID(&lli->lli_fid));
243 md_from_obdo(op_data, oa, oa->o_valid);
244 memcpy(&op_data->op_handle, fh, sizeof(*fh));
246 op_data->op_ioepoch = ioepoch;
247 op_data->op_flags |= MF_SOM_CHANGE;
249 rc = ll_md_setattr(inode, op_data, NULL);
255 ll_finish_md_op_data(op_data);
259 /** Sends a DONE_WRITING rpc, packs Size-on-MDS attributes into it, if
261 static void ll_done_writing(struct inode *inode)
263 struct obd_client_handle *och = NULL;
264 struct md_op_data *op_data;
268 LASSERT(ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM);
270 OBD_ALLOC_PTR(op_data);
271 if (op_data == NULL) {
272 CERROR("can't allocate op_data\n");
277 ll_epoch_close(inode, op_data, &och, LLIF_DONE_WRITING);
278 /* If there is no @och, we do not do D_W yet. */
282 ll_pack_inode2opdata(inode, op_data, &och->och_fh);
284 rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
286 /* MDS has instructed us to obtain Size-on-MDS attribute from
287 * OSTs and send setattr to back to MDS. */
288 rc = ll_sizeonmds_update(inode, &och->och_fh,
289 op_data->op_ioepoch);
291 CERROR("inode %lu mdc done_writing failed: rc = %d\n",
295 ll_finish_md_op_data(op_data);
297 md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
303 static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
305 struct ll_inode_info *lli = NULL;
307 spin_lock(&lcq->lcq_lock);
309 if (!list_empty(&lcq->lcq_head)) {
310 lli = list_entry(lcq->lcq_head.next, struct ll_inode_info,
312 list_del_init(&lli->lli_close_list);
313 } else if (atomic_read(&lcq->lcq_stop))
314 lli = ERR_PTR(-EALREADY);
316 spin_unlock(&lcq->lcq_lock);
320 static int ll_close_thread(void *arg)
322 struct ll_close_queue *lcq = arg;
326 char name[CFS_CURPROC_COMM_MAX];
327 snprintf(name, sizeof(name) - 1, "ll_close");
331 complete(&lcq->lcq_comp);
334 struct l_wait_info lwi = { 0 };
335 struct ll_inode_info *lli;
338 l_wait_event_exclusive(lcq->lcq_waitq,
339 (lli = ll_close_next_lli(lcq)) != NULL,
344 inode = ll_info2i(lli);
345 CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
346 inode->i_ino, inode->i_generation);
347 ll_done_writing(inode);
351 CDEBUG(D_INFO, "ll_close exiting\n");
352 complete(&lcq->lcq_comp);
356 int ll_close_thread_start(struct ll_close_queue **lcq_ret)
358 struct ll_close_queue *lcq;
361 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
364 OBD_ALLOC(lcq, sizeof(*lcq));
368 spin_lock_init(&lcq->lcq_lock);
369 INIT_LIST_HEAD(&lcq->lcq_head);
370 init_waitqueue_head(&lcq->lcq_waitq);
371 init_completion(&lcq->lcq_comp);
373 pid = kernel_thread(ll_close_thread, lcq, 0);
375 OBD_FREE(lcq, sizeof(*lcq));
379 wait_for_completion(&lcq->lcq_comp);
384 void ll_close_thread_shutdown(struct ll_close_queue *lcq)
386 init_completion(&lcq->lcq_comp);
387 atomic_inc(&lcq->lcq_stop);
388 wake_up(&lcq->lcq_waitq);
389 wait_for_completion(&lcq->lcq_comp);
390 OBD_FREE(lcq, sizeof(*lcq));