blob: 596ec71da00e8d4f005ab264142e927d03ac667d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include "xfs_log.h"
20#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_mount.h"
24#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_dinode.h"
26#include "xfs_inode.h"
Christoph Hellwig281627d2012-03-13 08:41:05 +000027#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_iomap.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100031#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000032#include "xfs_trace.h"
Dave Chinner3ed3a432010-03-05 02:00:42 +000033#include "xfs_bmap.h"
Kent Overstreeta27bb332013-05-07 16:19:08 -070034#include <linux/aio.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090035#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110037#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/writeback.h>
39
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000040void
Nathan Scottf51623b2006-03-14 13:26:27 +110041xfs_count_page_state(
42 struct page *page,
43 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110044 int *unwritten)
45{
46 struct buffer_head *bh, *head;
47
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100048 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110049
50 bh = head = page_buffers(page);
51 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100052 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110053 (*unwritten) = 1;
54 else if (buffer_delay(bh))
55 (*delalloc) = 1;
56 } while ((bh = bh->b_this_page) != head);
57}
58
Christoph Hellwig6214ed42007-09-14 15:23:17 +100059STATIC struct block_device *
60xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000061 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100062{
Christoph Hellwig046f1682010-04-28 12:28:52 +000063 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100064 struct xfs_mount *mp = ip->i_mount;
65
Eric Sandeen71ddabb2007-11-23 16:29:42 +110066 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100067 return mp->m_rtdev_targp->bt_bdev;
68 else
69 return mp->m_ddev_targp->bt_bdev;
70}
71
Christoph Hellwig0829c362005-09-02 16:58:49 +100072/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110073 * We're now finished for good with this ioend structure.
74 * Update the page state via the associated buffer_heads,
75 * release holds on the inode and bio, and finally free
76 * up memory. Do not use the ioend after this.
77 */
Christoph Hellwig0829c362005-09-02 16:58:49 +100078STATIC void
79xfs_destroy_ioend(
80 xfs_ioend_t *ioend)
81{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110082 struct buffer_head *bh, *next;
83
84 for (bh = ioend->io_buffer_head; bh; bh = next) {
85 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +100086 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110087 }
Christoph Hellwig583fa582008-12-03 12:20:38 +010088
Christoph Hellwigc859cdd2011-08-23 08:28:10 +000089 if (ioend->io_iocb) {
Jan Kara4b05d092013-01-23 13:56:18 +010090 inode_dio_done(ioend->io_inode);
Christoph Hellwig04f658e2011-08-24 05:59:25 +000091 if (ioend->io_isasync) {
92 aio_complete(ioend->io_iocb, ioend->io_error ?
93 ioend->io_error : ioend->io_result, 0);
94 }
Christoph Hellwigc859cdd2011-08-23 08:28:10 +000095 }
Christoph Hellwig4a06fd22011-08-23 08:28:13 +000096
Christoph Hellwig0829c362005-09-02 16:58:49 +100097 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
100/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000101 * Fast and loose check if this write could update the on-disk inode size.
102 */
103static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
104{
105 return ioend->io_offset + ioend->io_size >
106 XFS_I(ioend->io_inode)->i_d.di_size;
107}
108
Christoph Hellwig281627d2012-03-13 08:41:05 +0000109STATIC int
110xfs_setfilesize_trans_alloc(
111 struct xfs_ioend *ioend)
112{
113 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
114 struct xfs_trans *tp;
115 int error;
116
117 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
118
119 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
120 if (error) {
121 xfs_trans_cancel(tp, 0);
122 return error;
123 }
124
125 ioend->io_append_trans = tp;
126
127 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100128 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200129 * we released it.
130 */
131 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
132 1, _THIS_IP_);
133 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000134 * We hand off the transaction to the completion thread now, so
135 * clear the flag here.
136 */
137 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
138 return 0;
139}
140
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000141/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000142 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000143 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000144STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000145xfs_setfilesize(
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000146 struct xfs_ioend *ioend)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000147{
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000148 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000149 struct xfs_trans *tp = ioend->io_append_trans;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000150 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000151
Christoph Hellwig281627d2012-03-13 08:41:05 +0000152 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100153 * The transaction may have been allocated in the I/O submission thread,
154 * thus we need to mark ourselves as beeing in a transaction manually.
155 * Similarly for freeze protection.
Christoph Hellwig281627d2012-03-13 08:41:05 +0000156 */
157 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner437a2552012-11-28 13:01:00 +1100158 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
159 0, 1, _THIS_IP_);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000160
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000161 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig6923e682012-02-29 09:53:49 +0000162 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000163 if (!isize) {
164 xfs_iunlock(ip, XFS_ILOCK_EXCL);
165 xfs_trans_cancel(tp, 0);
166 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000167 }
168
Christoph Hellwig281627d2012-03-13 08:41:05 +0000169 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
170
171 ip->i_d.di_size = isize;
172 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
173 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
174
175 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000176}
177
178/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000179 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000180 *
181 * If there is no work to do we might as well call it a day and free the
182 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200183 */
184STATIC void
185xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000186 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200187{
188 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000189 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
190
Alain Renaud0d882a32012-05-22 15:56:21 -0500191 if (ioend->io_type == XFS_IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000192 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Dave Chinner437a2552012-11-28 13:01:00 +1100193 else if (ioend->io_append_trans ||
194 (ioend->io_isdirect && xfs_ioend_is_append(ioend)))
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000195 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000196 else
197 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200198 }
199}
200
201/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000202 * IO write completion.
203 */
204STATIC void
205xfs_end_io(
206 struct work_struct *work)
207{
208 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
209 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000210 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000211
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000212 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000213 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000214 goto done;
215 }
216 if (ioend->io_error)
217 goto done;
218
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000219 /*
220 * For unwritten extents we need to issue transactions to convert a
221 * range to normal written extens after the data I/O has finished.
222 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500223 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000224 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100225 ioend->io_size);
226 } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
227 /*
228 * For direct I/O we do not know if we need to allocate blocks
229 * or not so we can't preallocate an append transaction as that
230 * results in nested reservations and log space deadlocks. Hence
231 * allocate the transaction here. While this is sub-optimal and
232 * can block IO completion for some time, we're stuck with doing
233 * it this way until we can pass the ioend to the direct IO
234 * allocation callbacks and avoid nesting that way.
235 */
236 error = xfs_setfilesize_trans_alloc(ioend);
237 if (error)
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000238 goto done;
Dave Chinner437a2552012-11-28 13:01:00 +1100239 error = xfs_setfilesize(ioend);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000240 } else if (ioend->io_append_trans) {
241 error = xfs_setfilesize(ioend);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000242 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000243 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000244 }
245
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000246done:
Dave Chinner437a2552012-11-28 13:01:00 +1100247 if (error)
248 ioend->io_error = -error;
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000249 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000250}
251
252/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000253 * Call IO completion handling in caller context on the final put of an ioend.
254 */
255STATIC void
256xfs_finish_ioend_sync(
257 struct xfs_ioend *ioend)
258{
259 if (atomic_dec_and_test(&ioend->io_remaining))
260 xfs_end_io(&ioend->io_work);
261}
262
263/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000264 * Allocate and initialise an IO completion structure.
265 * We need to track unwritten extent write completion here initially.
266 * We'll need to extend this for updating the ondisk inode size later
267 * (vs. incore size).
268 */
269STATIC xfs_ioend_t *
270xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100271 struct inode *inode,
272 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000273{
274 xfs_ioend_t *ioend;
275
276 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
277
278 /*
279 * Set the count to 1 initially, which will prevent an I/O
280 * completion callback from happening before we have started
281 * all the I/O from calling the completion routine too early.
282 */
283 atomic_set(&ioend->io_remaining, 1);
Christoph Hellwigc859cdd2011-08-23 08:28:10 +0000284 ioend->io_isasync = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000285 ioend->io_isdirect = 0;
Nathan Scott7d04a332006-06-09 14:58:38 +1000286 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100287 ioend->io_list = NULL;
288 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000289 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000290 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100291 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000292 ioend->io_offset = 0;
293 ioend->io_size = 0;
Christoph Hellwigfb511f22010-07-18 21:17:10 +0000294 ioend->io_iocb = NULL;
295 ioend->io_result = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000296 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000297
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000298 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000299 return ioend;
300}
301
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302STATIC int
303xfs_map_blocks(
304 struct inode *inode,
305 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000306 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000307 int type,
308 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000310 struct xfs_inode *ip = XFS_I(inode);
311 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000312 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000313 xfs_fileoff_t offset_fsb, end_fsb;
314 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000315 int bmapi_flags = XFS_BMAPI_ENTIRE;
316 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
Christoph Hellwiga206c812010-12-10 08:42:20 +0000318 if (XFS_FORCED_SHUTDOWN(mp))
319 return -XFS_ERROR(EIO);
320
Alain Renaud0d882a32012-05-22 15:56:21 -0500321 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000322 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000323
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000324 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
325 if (nonblocking)
326 return -XFS_ERROR(EAGAIN);
327 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000328 }
329
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000330 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
331 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000332 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000333
Dave Chinnerd2c28192012-06-08 15:44:53 +1000334 if (offset + count > mp->m_super->s_maxbytes)
335 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000336 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
337 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000338 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
339 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000340 xfs_iunlock(ip, XFS_ILOCK_SHARED);
341
Christoph Hellwiga206c812010-12-10 08:42:20 +0000342 if (error)
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000343 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000344
Alain Renaud0d882a32012-05-22 15:56:21 -0500345 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000346 (!nimaps || isnullstartblock(imap->br_startblock))) {
Christoph Hellwiga206c812010-12-10 08:42:20 +0000347 error = xfs_iomap_write_allocate(ip, offset, count, imap);
348 if (!error)
349 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000350 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000351 }
352
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000353#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500354 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000355 ASSERT(nimaps);
356 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
357 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
358 }
359#endif
360 if (nimaps)
361 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
362 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000365STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000366xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000367 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000368 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000369 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000371 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000372
Christoph Hellwig558e6892010-04-28 12:28:58 +0000373 return offset >= imap->br_startoff &&
374 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375}
376
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100377/*
378 * BIO completion handler for buffered IO.
379 */
Al Viro782e3b32007-10-12 07:17:47 +0100380STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100381xfs_end_bio(
382 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100383 int error)
384{
385 xfs_ioend_t *ioend = bio->bi_private;
386
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100387 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000388 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100389
390 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100391 bio->bi_private = NULL;
392 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100393 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000394
Christoph Hellwig209fb872010-07-18 21:17:11 +0000395 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100396}
397
398STATIC void
399xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000400 struct writeback_control *wbc,
401 xfs_ioend_t *ioend,
402 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100403{
404 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100405 bio->bi_private = ioend;
406 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100407 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100408}
409
410STATIC struct bio *
411xfs_alloc_ioend_bio(
412 struct buffer_head *bh)
413{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100414 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000415 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100416
417 ASSERT(bio->bi_private == NULL);
418 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
419 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100420 return bio;
421}
422
423STATIC void
424xfs_start_buffer_writeback(
425 struct buffer_head *bh)
426{
427 ASSERT(buffer_mapped(bh));
428 ASSERT(buffer_locked(bh));
429 ASSERT(!buffer_delay(bh));
430 ASSERT(!buffer_unwritten(bh));
431
432 mark_buffer_async_write(bh);
433 set_buffer_uptodate(bh);
434 clear_buffer_dirty(bh);
435}
436
437STATIC void
438xfs_start_page_writeback(
439 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100440 int clear_dirty,
441 int buffers)
442{
443 ASSERT(PageLocked(page));
444 ASSERT(!PageWriteback(page));
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100445 if (clear_dirty)
David Chinner92132022006-12-21 10:24:01 +1100446 clear_page_dirty_for_io(page);
447 set_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100448 unlock_page(page);
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700449 /* If no buffers on the page are to be written, finish it here */
450 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100451 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100452}
453
454static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
455{
456 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
457}
458
459/*
David Chinnerd88992f2006-01-18 13:38:12 +1100460 * Submit all of the bios for all of the ioends we have saved up, covering the
461 * initial writepage page and also any probed pages.
462 *
463 * Because we may have multiple ioends spanning a page, we need to start
464 * writeback on all the buffers before we submit them for I/O. If we mark the
465 * buffers as we got, then we can end up with a page that only has buffers
466 * marked async write and I/O complete on can occur before we mark the other
467 * buffers async write.
468 *
469 * The end result of this is that we trip a bug in end_page_writeback() because
470 * we call it twice for the one page as the code in end_buffer_async_write()
471 * assumes that all buffers on the page are started at the same time.
472 *
473 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000474 * buffer_heads, and then submit them for I/O on the second pass.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100475 *
476 * If @fail is non-zero, it means that we have a situation where some part of
477 * the submission process has failed after we have marked paged for writeback
478 * and unlocked them. In this situation, we need to fail the ioend chain rather
479 * than submit it to IO. This typically only happens on a filesystem shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100480 */
481STATIC void
482xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000483 struct writeback_control *wbc,
Dave Chinner7bf7f352012-11-12 22:09:45 +1100484 xfs_ioend_t *ioend,
485 int fail)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100486{
David Chinnerd88992f2006-01-18 13:38:12 +1100487 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100488 xfs_ioend_t *next;
489 struct buffer_head *bh;
490 struct bio *bio;
491 sector_t lastblock = 0;
492
David Chinnerd88992f2006-01-18 13:38:12 +1100493 /* Pass 1 - start writeback */
494 do {
495 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000496 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100497 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100498 } while ((ioend = next) != NULL);
499
500 /* Pass 2 - submit I/O */
501 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100502 do {
503 next = ioend->io_list;
504 bio = NULL;
505
Dave Chinner7bf7f352012-11-12 22:09:45 +1100506 /*
507 * If we are failing the IO now, just mark the ioend with an
508 * error and finish it. This will run IO completion immediately
509 * as there is only one reference to the ioend at this point in
510 * time.
511 */
512 if (fail) {
513 ioend->io_error = -fail;
514 xfs_finish_ioend(ioend);
515 continue;
516 }
517
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100518 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100519
520 if (!bio) {
521 retry:
522 bio = xfs_alloc_ioend_bio(bh);
523 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000524 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100525 goto retry;
526 }
527
528 if (bio_add_buffer(bio, bh) != bh->b_size) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000529 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100530 goto retry;
531 }
532
533 lastblock = bh->b_blocknr;
534 }
535 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000536 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000537 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100538 } while ((ioend = next) != NULL);
539}
540
541/*
542 * Cancel submission of all buffer_heads so far in this endio.
543 * Toss the endio too. Only ever called for the initial page
544 * in a writepage request, so only ever one page.
545 */
546STATIC void
547xfs_cancel_ioend(
548 xfs_ioend_t *ioend)
549{
550 xfs_ioend_t *next;
551 struct buffer_head *bh, *next_bh;
552
553 do {
554 next = ioend->io_list;
555 bh = ioend->io_buffer_head;
556 do {
557 next_bh = bh->b_private;
558 clear_buffer_async_write(bh);
559 unlock_buffer(bh);
560 } while ((bh = next_bh) != NULL);
561
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100562 mempool_free(ioend, xfs_ioend_pool);
563 } while ((ioend = next) != NULL);
564}
565
566/*
567 * Test to see if we've been building up a completion structure for
568 * earlier buffers -- if so, we try to append to this ioend if we
569 * can, otherwise we finish off any current ioend and start another.
570 * Return true if we've finished the given ioend.
571 */
572STATIC void
573xfs_add_to_ioend(
574 struct inode *inode,
575 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100576 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100577 unsigned int type,
578 xfs_ioend_t **result,
579 int need_ioend)
580{
581 xfs_ioend_t *ioend = *result;
582
583 if (!ioend || need_ioend || type != ioend->io_type) {
584 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100585
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100586 ioend = xfs_alloc_ioend(inode, type);
587 ioend->io_offset = offset;
588 ioend->io_buffer_head = bh;
589 ioend->io_buffer_tail = bh;
590 if (previous)
591 previous->io_list = ioend;
592 *result = ioend;
593 } else {
594 ioend->io_buffer_tail->b_private = bh;
595 ioend->io_buffer_tail = bh;
596 }
597
598 bh->b_private = NULL;
599 ioend->io_size += bh->b_size;
600}
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100603xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000604 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100605 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000606 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000607 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100608{
609 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000610 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000611 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
612 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100613
Christoph Hellwig207d0412010-04-28 12:28:56 +0000614 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
615 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100616
Christoph Hellwige5131822010-04-28 12:28:55 +0000617 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000618 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100619
Christoph Hellwig046f1682010-04-28 12:28:52 +0000620 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100621
622 bh->b_blocknr = bn;
623 set_buffer_mapped(bh);
624}
625
626STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000628 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000630 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000631 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000633 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
634 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635
Christoph Hellwig207d0412010-04-28 12:28:56 +0000636 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 set_buffer_mapped(bh);
638 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100639 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640}
641
642/*
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100643 * Test if a given page is suitable for writing as part of an unwritten
644 * or delayed allocate extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 */
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100646STATIC int
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000647xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100648 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100649 unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (PageWriteback(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100652 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
654 if (page->mapping && page_has_buffers(page)) {
655 struct buffer_head *bh, *head;
656 int acceptable = 0;
657
658 bh = head = page_buffers(page);
659 do {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100660 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500661 acceptable += (type == XFS_IO_UNWRITTEN);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100662 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500663 acceptable += (type == XFS_IO_DELALLOC);
David Chinner2ddee842006-03-22 12:47:40 +1100664 else if (buffer_dirty(bh) && buffer_mapped(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500665 acceptable += (type == XFS_IO_OVERWRITE);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100666 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 } while ((bh = bh->b_this_page) != head);
669
670 if (acceptable)
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100671 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 }
673
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100674 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677/*
678 * Allocate & map buffers for page given the extent map. Write it out.
679 * except for the original page of a writepage, this is called on
680 * delalloc/unwritten pages only, for the original page it is possible
681 * that the page has no mapping at all.
682 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100683STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684xfs_convert_page(
685 struct inode *inode,
686 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100687 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000688 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100689 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000690 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100692 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100693 xfs_off_t end_offset;
694 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100695 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700696 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100697 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100698 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100700 if (page->index != tindex)
701 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200702 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100703 goto fail;
704 if (PageWriteback(page))
705 goto fail_unlock_page;
706 if (page->mapping != inode->i_mapping)
707 goto fail_unlock_page;
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000708 if (!xfs_check_page_type(page, (*ioendp)->io_type))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100709 goto fail_unlock_page;
710
Nathan Scott24e17b52005-05-05 13:33:20 -0700711 /*
712 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000713 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100714 *
715 * Derivation:
716 *
717 * End offset is the highest offset that this page should represent.
718 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
719 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
720 * hence give us the correct page_dirty count. On any other page,
721 * it will be zero and in that case we need page_dirty to be the
722 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700723 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100724 end_offset = min_t(unsigned long long,
725 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
726 i_size_read(inode));
727
Dave Chinner480d7462013-05-20 09:51:08 +1000728 /*
729 * If the current map does not span the entire page we are about to try
730 * to write, then give up. The only way we can write a page that spans
731 * multiple mappings in a single writeback iteration is via the
732 * xfs_vm_writepage() function. Data integrity writeback requires the
733 * entire page to be written in a single attempt, otherwise the part of
734 * the page we don't write here doesn't get written as part of the data
735 * integrity sync.
736 *
737 * For normal writeback, we also don't attempt to write partial pages
738 * here as it simply means that write_cache_pages() will see it under
739 * writeback and ignore the page until some point in the future, at
740 * which time this will be the only page in the file that needs
741 * writeback. Hence for more optimal IO patterns, we should always
742 * avoid partial page writeback due to multiple mappings on a page here.
743 */
744 if (!xfs_imap_valid(inode, imap, end_offset))
745 goto fail_unlock_page;
746
Nathan Scott24e17b52005-05-05 13:33:20 -0700747 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100748 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
749 PAGE_CACHE_SIZE);
750 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
751 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700752
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 bh = head = page_buffers(page);
754 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100755 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100757 if (!buffer_uptodate(bh))
758 uptodate = 0;
759 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
760 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 continue;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100762 }
763
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000764 if (buffer_unwritten(bh) || buffer_delay(bh) ||
765 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100766 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500767 type = XFS_IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000768 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500769 type = XFS_IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000770 else
Alain Renaud0d882a32012-05-22 15:56:21 -0500771 type = XFS_IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100772
Christoph Hellwig558e6892010-04-28 12:28:58 +0000773 if (!xfs_imap_valid(inode, imap, offset)) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100774 done = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100775 continue;
776 }
777
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000778 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -0500779 if (type != XFS_IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000780 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000781 xfs_add_to_ioend(inode, bh, offset, type,
782 ioendp, done);
783
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100784 page_dirty--;
785 count++;
786 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000787 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100789 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100791 if (uptodate && bh == head)
792 SetPageUptodate(page);
793
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000794 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000795 if (--wbc->nr_to_write <= 0 &&
796 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000797 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000799 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100800
801 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100802 fail_unlock_page:
803 unlock_page(page);
804 fail:
805 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
808/*
809 * Convert & write out a cluster of pages in the same extent as defined
810 * by mp and following the start page.
811 */
812STATIC void
813xfs_cluster_write(
814 struct inode *inode,
815 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000816 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100817 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 pgoff_t tlast)
820{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100821 struct pagevec pvec;
822 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100824 pagevec_init(&pvec, 0);
825 while (!done && tindex <= tlast) {
826 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
827
828 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100830
831 for (i = 0; i < pagevec_count(&pvec); i++) {
832 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000833 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100834 if (done)
835 break;
836 }
837
838 pagevec_release(&pvec);
839 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 }
841}
842
Dave Chinner3ed3a432010-03-05 02:00:42 +0000843STATIC void
844xfs_vm_invalidatepage(
845 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400846 unsigned int offset,
847 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000848{
Lukas Czerner34097df2013-05-21 23:58:01 -0400849 trace_xfs_invalidatepage(page->mapping->host, page, offset,
850 length);
851 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000852}
853
854/*
855 * If the page has delalloc buffers on it, we need to punch them out before we
856 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
857 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
858 * is done on that same region - the delalloc extent is returned when none is
859 * supposed to be there.
860 *
861 * We prevent this by truncating away the delalloc regions on the page before
862 * invalidating it. Because they are delalloc, we can do this without needing a
863 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
864 * truncation without a transaction as there is no space left for block
865 * reservation (typically why we see a ENOSPC in writeback).
866 *
867 * This is not a performance critical path, so for now just do the punching a
868 * buffer head at a time.
869 */
870STATIC void
871xfs_aops_discard_page(
872 struct page *page)
873{
874 struct inode *inode = page->mapping->host;
875 struct xfs_inode *ip = XFS_I(inode);
876 struct buffer_head *bh, *head;
877 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000878
Alain Renaud0d882a32012-05-22 15:56:21 -0500879 if (!xfs_check_page_type(page, XFS_IO_DELALLOC))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000880 goto out_invalidate;
881
Dave Chinnere8c37532010-03-15 02:36:35 +0000882 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
883 goto out_invalidate;
884
Dave Chinner4f107002011-03-07 10:00:35 +1100885 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000886 "page discard on page %p, inode 0x%llx, offset %llu.",
887 page, ip->i_ino, offset);
888
889 xfs_ilock(ip, XFS_ILOCK_EXCL);
890 bh = head = page_buffers(page);
891 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000892 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100893 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000894
895 if (!buffer_delay(bh))
896 goto next_buffer;
897
Dave Chinnerc726de42010-11-30 15:14:39 +1100898 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
899 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000900 if (error) {
901 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000902 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100903 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000904 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000905 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000906 break;
907 }
908next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100909 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000910
911 } while ((bh = bh->b_this_page) != head);
912
913 xfs_iunlock(ip, XFS_ILOCK_EXCL);
914out_invalidate:
Lukas Czernerd47992f2013-05-21 23:17:23 -0400915 xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000916 return;
917}
918
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000920 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000922 * For delalloc space on the page we need to allocate space and flush it.
923 * For unwritten space on the page we need to start the conversion to
924 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000925 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927STATIC int
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000928xfs_vm_writepage(
929 struct page *page,
930 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000932 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100933 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000934 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100935 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100937 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000939 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000940 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000941 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000942 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000943 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944
Lukas Czerner34097df2013-05-21 23:58:01 -0400945 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000946
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000947 ASSERT(page_has_buffers(page));
948
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000949 /*
950 * Refuse to write the page out if we are called from reclaim context.
951 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400952 * This avoids stack overflows when called from deeply used stacks in
953 * random callers for direct reclaim or memcg reclaim. We explicitly
954 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000955 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700956 * This should never happen except in the case of a VM regression so
957 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000958 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700959 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
960 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000961 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000962
963 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200964 * Given that we do not allow direct reclaim to call us, we should
965 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000966 */
Christoph Hellwig680a6472011-07-08 14:34:05 +0200967 if (WARN_ON(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000968 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 /* Is this page beyond the end of the file? */
971 offset = i_size_read(inode);
972 end_index = offset >> PAGE_CACHE_SHIFT;
973 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
974 if (page->index >= end_index) {
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400975 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
976
977 /*
Jan Karaff9a28f2013-03-14 14:30:54 +0100978 * Skip the page if it is fully outside i_size, e.g. due to a
979 * truncate operation that is in progress. We must redirty the
980 * page so that reclaim stops reclaiming it. Otherwise
981 * xfs_vm_releasepage() is called on it and gets confused.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400982 */
Jan Karaff9a28f2013-03-14 14:30:54 +0100983 if (page->index >= end_index + 1 || offset_into_page == 0)
984 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400985
986 /*
987 * The page straddles i_size. It must be zeroed out on each
988 * and every writepage invocation because it may be mmapped.
989 * "A file is mapped in multiples of the page size. For a file
990 * that is not a multiple of the page size, the remaining
991 * memory is zeroed when mapped, and writes to that region are
992 * not written out to the file."
993 */
994 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 }
996
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100997 end_offset = min_t(unsigned long long,
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000998 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
999 offset);
Nathan Scott24e17b52005-05-05 13:33:20 -07001000 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -07001001
Nathan Scott24e17b52005-05-05 13:33:20 -07001002 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001003 offset = page_offset(page);
Alain Renaud0d882a32012-05-22 15:56:21 -05001004 type = XFS_IO_OVERWRITE;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001005
Christoph Hellwigdbcdde32011-07-08 14:34:14 +02001006 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001007 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001008
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +00001010 int new_ioend = 0;
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 if (offset >= end_offset)
1013 break;
1014 if (!buffer_uptodate(bh))
1015 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001017 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +00001018 * set_page_dirty dirties all buffers in a page, independent
1019 * of their state. The dirty state however is entirely
1020 * meaningless for holes (!mapped && uptodate), so skip
1021 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001022 */
1023 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001024 imap_valid = 0;
1025 continue;
1026 }
1027
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001028 if (buffer_unwritten(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001029 if (type != XFS_IO_UNWRITTEN) {
1030 type = XFS_IO_UNWRITTEN;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001031 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001032 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001033 } else if (buffer_delay(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001034 if (type != XFS_IO_DELALLOC) {
1035 type = XFS_IO_DELALLOC;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001036 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001038 } else if (buffer_uptodate(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001039 if (type != XFS_IO_OVERWRITE) {
1040 type = XFS_IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +00001041 imap_valid = 0;
1042 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001043 } else {
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001044 if (PageUptodate(page))
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001045 ASSERT(buffer_mapped(bh));
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001046 /*
1047 * This buffer is not uptodate and will not be
1048 * written to disk. Ensure that we will put any
1049 * subsequent writeable buffers into a new
1050 * ioend.
1051 */
1052 imap_valid = 0;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001053 continue;
1054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001056 if (imap_valid)
1057 imap_valid = xfs_imap_valid(inode, &imap, offset);
1058 if (!imap_valid) {
1059 /*
1060 * If we didn't have a valid mapping then we need to
1061 * put the new mapping into a separate ioend structure.
1062 * This ensures non-contiguous extents always have
1063 * separate ioends, which is particularly important
1064 * for unwritten extent conversion at I/O completion
1065 * time.
1066 */
1067 new_ioend = 1;
1068 err = xfs_map_blocks(inode, offset, &imap, type,
1069 nonblocking);
1070 if (err)
1071 goto error;
1072 imap_valid = xfs_imap_valid(inode, &imap, offset);
1073 }
1074 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001075 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -05001076 if (type != XFS_IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001077 xfs_map_at_offset(inode, bh, &imap, offset);
1078 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1079 new_ioend);
1080 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001082
1083 if (!iohead)
1084 iohead = ioend;
1085
1086 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 if (uptodate && bh == head)
1089 SetPageUptodate(page);
1090
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001091 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Dave Chinner7bf7f352012-11-12 22:09:45 +11001093 /* if there is no IO to be submitted for this page, we are done */
1094 if (!ioend)
1095 return 0;
1096
1097 ASSERT(iohead);
1098
1099 /*
1100 * Any errors from this point onwards need tobe reported through the IO
1101 * completion path as we have marked the initial page as under writeback
1102 * and unlocked it.
1103 */
1104 if (imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001105 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001106
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001107 end_index = imap.br_startoff + imap.br_blockcount;
1108
1109 /* to bytes */
1110 end_index <<= inode->i_blkbits;
1111
1112 /* to pages */
1113 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1114
1115 /* check against file size */
1116 if (end_index > last_index)
1117 end_index = last_index;
1118
Christoph Hellwig207d0412010-04-28 12:28:56 +00001119 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001120 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 }
1122
Christoph Hellwig281627d2012-03-13 08:41:05 +00001123
Dave Chinner7bf7f352012-11-12 22:09:45 +11001124 /*
1125 * Reserve log space if we might write beyond the on-disk inode size.
1126 */
1127 err = 0;
1128 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1129 err = xfs_setfilesize_trans_alloc(ioend);
1130
1131 xfs_submit_ioend(wbc, iohead, err);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001132
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001133 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001136 if (iohead)
1137 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001139 if (err == -EAGAIN)
1140 goto redirty;
1141
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001142 xfs_aops_discard_page(page);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001143 ClearPageUptodate(page);
1144 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001146
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001147redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001148 redirty_page_for_writepage(wbc, page);
1149 unlock_page(page);
1150 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001151}
1152
Nathan Scott7d4fb402006-06-09 15:27:16 +10001153STATIC int
1154xfs_vm_writepages(
1155 struct address_space *mapping,
1156 struct writeback_control *wbc)
1157{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001158 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001159 return generic_writepages(mapping, wbc);
1160}
1161
Nathan Scottf51623b2006-03-14 13:26:27 +11001162/*
1163 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001164 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001165 * have buffer heads in this call.
1166 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001167 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001168 */
1169STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001170xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001171 struct page *page,
1172 gfp_t gfp_mask)
1173{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001174 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001175
Lukas Czerner34097df2013-05-21 23:58:01 -04001176 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001177
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001178 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001179
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001180 if (WARN_ON(delalloc))
1181 return 0;
1182 if (WARN_ON(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001183 return 0;
1184
Nathan Scottf51623b2006-03-14 13:26:27 +11001185 return try_to_free_buffers(page);
1186}
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001189__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 struct inode *inode,
1191 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 struct buffer_head *bh_result,
1193 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001194 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001196 struct xfs_inode *ip = XFS_I(inode);
1197 struct xfs_mount *mp = ip->i_mount;
1198 xfs_fileoff_t offset_fsb, end_fsb;
1199 int error = 0;
1200 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001201 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001202 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001203 xfs_off_t offset;
1204 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001205 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001206
1207 if (XFS_FORCED_SHUTDOWN(mp))
1208 return -XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001210 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001211 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1212 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001213
1214 if (!create && direct && offset >= i_size_read(inode))
1215 return 0;
1216
Dave Chinner507630b2012-03-27 10:34:50 -04001217 /*
1218 * Direct I/O is usually done on preallocated files, so try getting
1219 * a block mapping without an exclusive lock first. For buffered
1220 * writes we already have the exclusive iolock anyway, so avoiding
1221 * a lock roundtrip here by taking the ilock exclusive from the
1222 * beginning is a useful micro optimization.
1223 */
1224 if (create && !direct) {
Christoph Hellwiga206c812010-12-10 08:42:20 +00001225 lockmode = XFS_ILOCK_EXCL;
1226 xfs_ilock(ip, lockmode);
1227 } else {
1228 lockmode = xfs_ilock_map_shared(ip);
1229 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001230
Dave Chinnerd2c28192012-06-08 15:44:53 +10001231 ASSERT(offset <= mp->m_super->s_maxbytes);
1232 if (offset + size > mp->m_super->s_maxbytes)
1233 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001234 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1235 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1236
Dave Chinner5c8ed202011-09-18 20:40:45 +00001237 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1238 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001240 goto out_unlock;
1241
1242 if (create &&
1243 (!nimaps ||
1244 (imap.br_startblock == HOLESTARTBLOCK ||
1245 imap.br_startblock == DELAYSTARTBLOCK))) {
Dave Chinneraff3a9e2012-04-23 15:58:44 +10001246 if (direct || xfs_get_extsz_hint(ip)) {
Dave Chinner507630b2012-03-27 10:34:50 -04001247 /*
1248 * Drop the ilock in preparation for starting the block
1249 * allocation transaction. It will be retaken
1250 * exclusively inside xfs_iomap_write_direct for the
1251 * actual allocation.
1252 */
1253 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001254 error = xfs_iomap_write_direct(ip, offset, size,
1255 &imap, nimaps);
Dave Chinner507630b2012-03-27 10:34:50 -04001256 if (error)
1257 return -error;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001258 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001259 } else {
Dave Chinner507630b2012-03-27 10:34:50 -04001260 /*
1261 * Delalloc reservations do not require a transaction,
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001262 * we can go on without dropping the lock here. If we
1263 * are allocating a new delalloc block, make sure that
1264 * we set the new flag so that we mark the buffer new so
1265 * that we know that it is newly allocated if the write
1266 * fails.
Dave Chinner507630b2012-03-27 10:34:50 -04001267 */
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001268 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1269 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001270 error = xfs_iomap_write_delay(ip, offset, size, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001271 if (error)
1272 goto out_unlock;
1273
1274 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001275 }
Christoph Hellwiga206c812010-12-10 08:42:20 +00001276
1277 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1278 } else if (nimaps) {
1279 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001280 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001281 } else {
1282 trace_xfs_get_blocks_notfound(ip, offset, size);
1283 goto out_unlock;
1284 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
Christoph Hellwig207d0412010-04-28 12:28:56 +00001286 if (imap.br_startblock != HOLESTARTBLOCK &&
1287 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001288 /*
1289 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 * the read case (treat as if we're reading into a hole).
1291 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001292 if (create || !ISUNWRITTEN(&imap))
1293 xfs_map_buffer(inode, bh_result, &imap, offset);
1294 if (create && ISUNWRITTEN(&imap)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 if (direct)
1296 bh_result->b_private = inode;
1297 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 }
1299 }
1300
Nathan Scottc2536662006-03-29 10:44:40 +10001301 /*
1302 * If this is a realtime file, data may be on a different device.
1303 * to that pointed to from the buffer_head b_bdev currently.
1304 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001305 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
Nathan Scottc2536662006-03-29 10:44:40 +10001307 /*
David Chinner549054a2007-02-10 18:36:35 +11001308 * If we previously allocated a block out beyond eof and we are now
1309 * coming back to use it then we will need to flag it as new even if it
1310 * has a disk address.
1311 *
1312 * With sub-block writes into unwritten extents we also need to mark
1313 * the buffer as new so that the unwritten parts of the buffer gets
1314 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 */
1316 if (create &&
1317 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001318 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001319 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321
Christoph Hellwig207d0412010-04-28 12:28:56 +00001322 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 BUG_ON(direct);
1324 if (create) {
1325 set_buffer_uptodate(bh_result);
1326 set_buffer_mapped(bh_result);
1327 set_buffer_delay(bh_result);
1328 }
1329 }
1330
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001331 /*
1332 * If this is O_DIRECT or the mpage code calling tell them how large
1333 * the mapping is, so that we can avoid repeated get_blocks calls.
1334 */
Nathan Scottc2536662006-03-29 10:44:40 +10001335 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001336 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001337
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001338 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1339 mapping_size <<= inode->i_blkbits;
1340
1341 ASSERT(mapping_size > 0);
1342 if (mapping_size > size)
1343 mapping_size = size;
1344 if (mapping_size > LONG_MAX)
1345 mapping_size = LONG_MAX;
1346
1347 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
1349
1350 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001351
1352out_unlock:
1353 xfs_iunlock(ip, lockmode);
1354 return -error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355}
1356
1357int
Nathan Scottc2536662006-03-29 10:44:40 +10001358xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359 struct inode *inode,
1360 sector_t iblock,
1361 struct buffer_head *bh_result,
1362 int create)
1363{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001364 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365}
1366
1367STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001368xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 struct inode *inode,
1370 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 struct buffer_head *bh_result,
1372 int create)
1373{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001374 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
Christoph Hellwig209fb872010-07-18 21:17:11 +00001377/*
1378 * Complete a direct I/O write request.
1379 *
1380 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1381 * need to issue a transaction to convert the range from unwritten to written
1382 * extents. In case this is regular synchronous I/O we just call xfs_end_io
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001383 * to do this and we are done. But in case this was a successful AIO
Christoph Hellwig209fb872010-07-18 21:17:11 +00001384 * request this handler is called from interrupt context, from which we
1385 * can't start transactions. In that case offload the I/O completion to
1386 * the workqueues we also use for buffered I/O completion.
1387 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001388STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001389xfs_end_io_direct_write(
1390 struct kiocb *iocb,
1391 loff_t offset,
1392 ssize_t size,
1393 void *private,
1394 int ret,
1395 bool is_async)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001396{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001397 struct xfs_ioend *ioend = iocb->private;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001398
1399 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001400 * While the generic direct I/O code updates the inode size, it does
1401 * so only after the end_io handler is called, which means our
1402 * end_io handler thinks the on-disk size is outside the in-core
1403 * size. To prevent this just update it a little bit earlier here.
1404 */
1405 if (offset + size > i_size_read(ioend->io_inode))
1406 i_size_write(ioend->io_inode, offset + size);
1407
1408 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001409 * blockdev_direct_IO can return an error even after the I/O
Christoph Hellwigf0973862005-09-05 08:22:52 +10001410 * completion handler was called. Thus we need to protect
1411 * against double-freeing.
1412 */
1413 iocb->private = NULL;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001414
Christoph Hellwig209fb872010-07-18 21:17:11 +00001415 ioend->io_offset = offset;
1416 ioend->io_size = size;
Christoph Hellwigc859cdd2011-08-23 08:28:10 +00001417 ioend->io_iocb = iocb;
1418 ioend->io_result = ret;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001419 if (private && size > 0)
Alain Renaud0d882a32012-05-22 15:56:21 -05001420 ioend->io_type = XFS_IO_UNWRITTEN;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001421
1422 if (is_async) {
Christoph Hellwigc859cdd2011-08-23 08:28:10 +00001423 ioend->io_isasync = 1;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001424 xfs_finish_ioend(ioend);
1425 } else {
1426 xfs_finish_ioend_sync(ioend);
1427 }
Christoph Hellwigf0973862005-09-05 08:22:52 +10001428}
1429
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001431xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 int rw,
1433 struct kiocb *iocb,
1434 const struct iovec *iov,
1435 loff_t offset,
1436 unsigned long nr_segs)
1437{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001438 struct inode *inode = iocb->ki_filp->f_mapping->host;
1439 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001440 struct xfs_ioend *ioend = NULL;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001441 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Christoph Hellwig209fb872010-07-18 21:17:11 +00001443 if (rw & WRITE) {
Christoph Hellwig281627d2012-03-13 08:41:05 +00001444 size_t size = iov_length(iov, nr_segs);
1445
1446 /*
Dave Chinner437a2552012-11-28 13:01:00 +11001447 * We cannot preallocate a size update transaction here as we
1448 * don't know whether allocation is necessary or not. Hence we
1449 * can only tell IO completion that one is necessary if we are
1450 * not doing unwritten extent conversion.
Christoph Hellwig281627d2012-03-13 08:41:05 +00001451 */
Alain Renaud0d882a32012-05-22 15:56:21 -05001452 iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
Dave Chinner437a2552012-11-28 13:01:00 +11001453 if (offset + size > XFS_I(inode)->i_d.di_size)
Christoph Hellwig281627d2012-03-13 08:41:05 +00001454 ioend->io_isdirect = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001456 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1457 offset, nr_segs,
1458 xfs_get_blocks_direct,
1459 xfs_end_io_direct_write, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001460 if (ret != -EIOCBQUEUED && iocb->private)
Dave Chinner437a2552012-11-28 13:01:00 +11001461 goto out_destroy_ioend;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001462 } else {
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001463 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1464 offset, nr_segs,
1465 xfs_get_blocks_direct,
1466 NULL, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001467 }
Christoph Hellwig5fe878a2009-12-15 16:47:50 -08001468
Christoph Hellwigf0973862005-09-05 08:22:52 +10001469 return ret;
Christoph Hellwig281627d2012-03-13 08:41:05 +00001470
Christoph Hellwig281627d2012-03-13 08:41:05 +00001471out_destroy_ioend:
1472 xfs_destroy_ioend(ioend);
1473 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474}
1475
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001476/*
1477 * Punch out the delalloc blocks we have already allocated.
1478 *
1479 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1480 * as the page is still locked at this point.
1481 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001482STATIC void
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001483xfs_vm_kill_delalloc_range(
1484 struct inode *inode,
1485 loff_t start,
1486 loff_t end)
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001487{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001488 struct xfs_inode *ip = XFS_I(inode);
1489 xfs_fileoff_t start_fsb;
1490 xfs_fileoff_t end_fsb;
1491 int error;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001492
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001493 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1494 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1495 if (end_fsb <= start_fsb)
1496 return;
Dave Chinnerc726de42010-11-30 15:14:39 +11001497
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001498 xfs_ilock(ip, XFS_ILOCK_EXCL);
1499 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1500 end_fsb - start_fsb);
1501 if (error) {
1502 /* something screwed, just bail */
1503 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1504 xfs_alert(ip->i_mount,
1505 "xfs_vm_write_failed: unable to clean up ino %lld",
1506 ip->i_ino);
Dave Chinnerc726de42010-11-30 15:14:39 +11001507 }
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001508 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001509 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001510}
1511
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001512STATIC void
1513xfs_vm_write_failed(
1514 struct inode *inode,
1515 struct page *page,
1516 loff_t pos,
1517 unsigned len)
1518{
1519 loff_t block_offset = pos & PAGE_MASK;
1520 loff_t block_start;
1521 loff_t block_end;
1522 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1523 loff_t to = from + len;
1524 struct buffer_head *bh, *head;
1525
1526 ASSERT(block_offset + from == pos);
1527
1528 head = page_buffers(page);
1529 block_start = 0;
1530 for (bh = head; bh != head || !block_start;
1531 bh = bh->b_this_page, block_start = block_end,
1532 block_offset += bh->b_size) {
1533 block_end = block_start + bh->b_size;
1534
1535 /* skip buffers before the write */
1536 if (block_end <= from)
1537 continue;
1538
1539 /* if the buffer is after the write, we're done */
1540 if (block_start >= to)
1541 break;
1542
1543 if (!buffer_delay(bh))
1544 continue;
1545
1546 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1547 continue;
1548
1549 xfs_vm_kill_delalloc_range(inode, block_offset,
1550 block_offset + bh->b_size);
1551 }
1552
1553}
1554
1555/*
1556 * This used to call block_write_begin(), but it unlocks and releases the page
1557 * on error, and we need that page to be able to punch stale delalloc blocks out
1558 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1559 * the appropriate point.
1560 */
Nathan Scottf51623b2006-03-14 13:26:27 +11001561STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001562xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001563 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001564 struct address_space *mapping,
1565 loff_t pos,
1566 unsigned len,
1567 unsigned flags,
1568 struct page **pagep,
1569 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001570{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001571 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1572 struct page *page;
1573 int status;
Christoph Hellwig155130a2010-06-04 11:29:58 +02001574
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001575 ASSERT(len <= PAGE_CACHE_SIZE);
1576
1577 page = grab_cache_page_write_begin(mapping, index,
1578 flags | AOP_FLAG_NOFS);
1579 if (!page)
1580 return -ENOMEM;
1581
1582 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1583 if (unlikely(status)) {
1584 struct inode *inode = mapping->host;
1585
1586 xfs_vm_write_failed(inode, page, pos, len);
1587 unlock_page(page);
1588
1589 if (pos + len > i_size_read(inode))
1590 truncate_pagecache(inode, pos + len, i_size_read(inode));
1591
1592 page_cache_release(page);
1593 page = NULL;
1594 }
1595
1596 *pagep = page;
1597 return status;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001598}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001599
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001600/*
1601 * On failure, we only need to kill delalloc blocks beyond EOF because they
1602 * will never be written. For blocks within EOF, generic_write_end() zeros them
1603 * so they are safe to leave alone and be written with all the other valid data.
1604 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001605STATIC int
1606xfs_vm_write_end(
1607 struct file *file,
1608 struct address_space *mapping,
1609 loff_t pos,
1610 unsigned len,
1611 unsigned copied,
1612 struct page *page,
1613 void *fsdata)
1614{
1615 int ret;
1616
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001617 ASSERT(len <= PAGE_CACHE_SIZE);
1618
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001619 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001620 if (unlikely(ret < len)) {
1621 struct inode *inode = mapping->host;
1622 size_t isize = i_size_read(inode);
1623 loff_t to = pos + len;
1624
1625 if (to > isize) {
1626 truncate_pagecache(inode, to, isize);
1627 xfs_vm_kill_delalloc_range(inode, isize, to);
1628 }
1629 }
Christoph Hellwig155130a2010-06-04 11:29:58 +02001630 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001631}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
1633STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001634xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 struct address_space *mapping,
1636 sector_t block)
1637{
1638 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001639 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001641 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001642 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001643 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001644 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001645 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646}
1647
1648STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001649xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 struct file *unused,
1651 struct page *page)
1652{
Nathan Scottc2536662006-03-29 10:44:40 +10001653 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654}
1655
1656STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001657xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 struct file *unused,
1659 struct address_space *mapping,
1660 struct list_head *pages,
1661 unsigned nr_pages)
1662{
Nathan Scottc2536662006-03-29 10:44:40 +10001663 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664}
1665
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001666const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001667 .readpage = xfs_vm_readpage,
1668 .readpages = xfs_vm_readpages,
1669 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001670 .writepages = xfs_vm_writepages,
Nathan Scott238f4c52006-03-17 17:26:25 +11001671 .releasepage = xfs_vm_releasepage,
1672 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001673 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001674 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001675 .bmap = xfs_vm_bmap,
1676 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001677 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001678 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001679 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680};