blob: 5f707e5371717a331131f72f5babff24670d9bef [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"
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110036#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/writeback.h>
38
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000039void
Nathan Scottf51623b2006-03-14 13:26:27 +110040xfs_count_page_state(
41 struct page *page,
42 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110043 int *unwritten)
44{
45 struct buffer_head *bh, *head;
46
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100047 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110048
49 bh = head = page_buffers(page);
50 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100051 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110052 (*unwritten) = 1;
53 else if (buffer_delay(bh))
54 (*delalloc) = 1;
55 } while ((bh = bh->b_this_page) != head);
56}
57
Christoph Hellwig6214ed42007-09-14 15:23:17 +100058STATIC struct block_device *
59xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000060 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100061{
Christoph Hellwig046f1682010-04-28 12:28:52 +000062 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100063 struct xfs_mount *mp = ip->i_mount;
64
Eric Sandeen71ddabb2007-11-23 16:29:42 +110065 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100066 return mp->m_rtdev_targp->bt_bdev;
67 else
68 return mp->m_ddev_targp->bt_bdev;
69}
70
Christoph Hellwig0829c362005-09-02 16:58:49 +100071/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110072 * We're now finished for good with this ioend structure.
73 * Update the page state via the associated buffer_heads,
74 * release holds on the inode and bio, and finally free
75 * up memory. Do not use the ioend after this.
76 */
Christoph Hellwig0829c362005-09-02 16:58:49 +100077STATIC void
78xfs_destroy_ioend(
79 xfs_ioend_t *ioend)
80{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110081 struct buffer_head *bh, *next;
82
83 for (bh = ioend->io_buffer_head; bh; bh = next) {
84 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +100085 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110086 }
Christoph Hellwig583fa582008-12-03 12:20:38 +010087
Christoph Hellwigc859cdd2011-08-23 08:28:10 +000088 if (ioend->io_iocb) {
Jan Kara4b05d092013-01-23 13:56:18 +010089 inode_dio_done(ioend->io_inode);
Christoph Hellwig04f658e2011-08-24 05:59:25 +000090 if (ioend->io_isasync) {
91 aio_complete(ioend->io_iocb, ioend->io_error ?
92 ioend->io_error : ioend->io_result, 0);
93 }
Christoph Hellwigc859cdd2011-08-23 08:28:10 +000094 }
Christoph Hellwig4a06fd22011-08-23 08:28:13 +000095
Christoph Hellwig0829c362005-09-02 16:58:49 +100096 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097}
98
99/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000100 * Fast and loose check if this write could update the on-disk inode size.
101 */
102static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
103{
104 return ioend->io_offset + ioend->io_size >
105 XFS_I(ioend->io_inode)->i_d.di_size;
106}
107
Christoph Hellwig281627d2012-03-13 08:41:05 +0000108STATIC int
109xfs_setfilesize_trans_alloc(
110 struct xfs_ioend *ioend)
111{
112 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
113 struct xfs_trans *tp;
114 int error;
115
116 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
117
118 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
119 if (error) {
120 xfs_trans_cancel(tp, 0);
121 return error;
122 }
123
124 ioend->io_append_trans = tp;
125
126 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100127 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200128 * we released it.
129 */
130 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
131 1, _THIS_IP_);
132 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000133 * We hand off the transaction to the completion thread now, so
134 * clear the flag here.
135 */
136 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
137 return 0;
138}
139
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000140/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000141 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000142 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000143STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000144xfs_setfilesize(
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000145 struct xfs_ioend *ioend)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000146{
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000147 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000148 struct xfs_trans *tp = ioend->io_append_trans;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000149 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000150
Christoph Hellwig281627d2012-03-13 08:41:05 +0000151 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100152 * The transaction may have been allocated in the I/O submission thread,
153 * thus we need to mark ourselves as beeing in a transaction manually.
154 * Similarly for freeze protection.
Christoph Hellwig281627d2012-03-13 08:41:05 +0000155 */
156 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner437a2552012-11-28 13:01:00 +1100157 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
158 0, 1, _THIS_IP_);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000159
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000160 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig6923e682012-02-29 09:53:49 +0000161 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000162 if (!isize) {
163 xfs_iunlock(ip, XFS_ILOCK_EXCL);
164 xfs_trans_cancel(tp, 0);
165 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000166 }
167
Christoph Hellwig281627d2012-03-13 08:41:05 +0000168 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
169
170 ip->i_d.di_size = isize;
171 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
172 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
173
174 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000175}
176
177/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000178 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000179 *
180 * If there is no work to do we might as well call it a day and free the
181 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200182 */
183STATIC void
184xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000185 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200186{
187 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000188 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
189
Alain Renaud0d882a32012-05-22 15:56:21 -0500190 if (ioend->io_type == XFS_IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000191 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Dave Chinner437a2552012-11-28 13:01:00 +1100192 else if (ioend->io_append_trans ||
193 (ioend->io_isdirect && xfs_ioend_is_append(ioend)))
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000194 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000195 else
196 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200197 }
198}
199
200/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000201 * IO write completion.
202 */
203STATIC void
204xfs_end_io(
205 struct work_struct *work)
206{
207 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
208 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000209 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000210
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000211 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000212 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000213 goto done;
214 }
215 if (ioend->io_error)
216 goto done;
217
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000218 /*
219 * For unwritten extents we need to issue transactions to convert a
220 * range to normal written extens after the data I/O has finished.
221 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500222 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000223 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100224 ioend->io_size);
225 } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
226 /*
227 * For direct I/O we do not know if we need to allocate blocks
228 * or not so we can't preallocate an append transaction as that
229 * results in nested reservations and log space deadlocks. Hence
230 * allocate the transaction here. While this is sub-optimal and
231 * can block IO completion for some time, we're stuck with doing
232 * it this way until we can pass the ioend to the direct IO
233 * allocation callbacks and avoid nesting that way.
234 */
235 error = xfs_setfilesize_trans_alloc(ioend);
236 if (error)
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000237 goto done;
Dave Chinner437a2552012-11-28 13:01:00 +1100238 error = xfs_setfilesize(ioend);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000239 } else if (ioend->io_append_trans) {
240 error = xfs_setfilesize(ioend);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000241 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000242 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000243 }
244
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000245done:
Dave Chinner437a2552012-11-28 13:01:00 +1100246 if (error)
247 ioend->io_error = -error;
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000248 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000249}
250
251/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000252 * Call IO completion handling in caller context on the final put of an ioend.
253 */
254STATIC void
255xfs_finish_ioend_sync(
256 struct xfs_ioend *ioend)
257{
258 if (atomic_dec_and_test(&ioend->io_remaining))
259 xfs_end_io(&ioend->io_work);
260}
261
262/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000263 * Allocate and initialise an IO completion structure.
264 * We need to track unwritten extent write completion here initially.
265 * We'll need to extend this for updating the ondisk inode size later
266 * (vs. incore size).
267 */
268STATIC xfs_ioend_t *
269xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100270 struct inode *inode,
271 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000272{
273 xfs_ioend_t *ioend;
274
275 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
276
277 /*
278 * Set the count to 1 initially, which will prevent an I/O
279 * completion callback from happening before we have started
280 * all the I/O from calling the completion routine too early.
281 */
282 atomic_set(&ioend->io_remaining, 1);
Christoph Hellwigc859cdd2011-08-23 08:28:10 +0000283 ioend->io_isasync = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000284 ioend->io_isdirect = 0;
Nathan Scott7d04a332006-06-09 14:58:38 +1000285 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100286 ioend->io_list = NULL;
287 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000288 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000289 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100290 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000291 ioend->io_offset = 0;
292 ioend->io_size = 0;
Christoph Hellwigfb511f22010-07-18 21:17:10 +0000293 ioend->io_iocb = NULL;
294 ioend->io_result = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000295 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000296
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000297 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000298 return ioend;
299}
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301STATIC int
302xfs_map_blocks(
303 struct inode *inode,
304 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000305 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000306 int type,
307 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000309 struct xfs_inode *ip = XFS_I(inode);
310 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000311 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000312 xfs_fileoff_t offset_fsb, end_fsb;
313 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000314 int bmapi_flags = XFS_BMAPI_ENTIRE;
315 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Christoph Hellwiga206c812010-12-10 08:42:20 +0000317 if (XFS_FORCED_SHUTDOWN(mp))
318 return -XFS_ERROR(EIO);
319
Alain Renaud0d882a32012-05-22 15:56:21 -0500320 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000321 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000322
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000323 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
324 if (nonblocking)
325 return -XFS_ERROR(EAGAIN);
326 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000327 }
328
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000329 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
330 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000331 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000332
Dave Chinnerd2c28192012-06-08 15:44:53 +1000333 if (offset + count > mp->m_super->s_maxbytes)
334 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000335 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
336 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000337 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
338 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000339 xfs_iunlock(ip, XFS_ILOCK_SHARED);
340
Christoph Hellwiga206c812010-12-10 08:42:20 +0000341 if (error)
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000342 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000343
Alain Renaud0d882a32012-05-22 15:56:21 -0500344 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000345 (!nimaps || isnullstartblock(imap->br_startblock))) {
Christoph Hellwiga206c812010-12-10 08:42:20 +0000346 error = xfs_iomap_write_allocate(ip, offset, count, imap);
347 if (!error)
348 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000349 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000350 }
351
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000352#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500353 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000354 ASSERT(nimaps);
355 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
356 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
357 }
358#endif
359 if (nimaps)
360 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
361 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000364STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000365xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000366 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000367 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000368 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000370 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000371
Christoph Hellwig558e6892010-04-28 12:28:58 +0000372 return offset >= imap->br_startoff &&
373 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374}
375
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100376/*
377 * BIO completion handler for buffered IO.
378 */
Al Viro782e3b32007-10-12 07:17:47 +0100379STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100380xfs_end_bio(
381 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100382 int error)
383{
384 xfs_ioend_t *ioend = bio->bi_private;
385
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100386 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000387 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100388
389 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100390 bio->bi_private = NULL;
391 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100392 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000393
Christoph Hellwig209fb872010-07-18 21:17:11 +0000394 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100395}
396
397STATIC void
398xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000399 struct writeback_control *wbc,
400 xfs_ioend_t *ioend,
401 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100402{
403 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100404 bio->bi_private = ioend;
405 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100406 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100407}
408
409STATIC struct bio *
410xfs_alloc_ioend_bio(
411 struct buffer_head *bh)
412{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100413 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000414 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100415
416 ASSERT(bio->bi_private == NULL);
417 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
418 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100419 return bio;
420}
421
422STATIC void
423xfs_start_buffer_writeback(
424 struct buffer_head *bh)
425{
426 ASSERT(buffer_mapped(bh));
427 ASSERT(buffer_locked(bh));
428 ASSERT(!buffer_delay(bh));
429 ASSERT(!buffer_unwritten(bh));
430
431 mark_buffer_async_write(bh);
432 set_buffer_uptodate(bh);
433 clear_buffer_dirty(bh);
434}
435
436STATIC void
437xfs_start_page_writeback(
438 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100439 int clear_dirty,
440 int buffers)
441{
442 ASSERT(PageLocked(page));
443 ASSERT(!PageWriteback(page));
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100444 if (clear_dirty)
David Chinner92132022006-12-21 10:24:01 +1100445 clear_page_dirty_for_io(page);
446 set_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100447 unlock_page(page);
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700448 /* If no buffers on the page are to be written, finish it here */
449 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100450 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100451}
452
453static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
454{
455 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
456}
457
458/*
David Chinnerd88992f2006-01-18 13:38:12 +1100459 * Submit all of the bios for all of the ioends we have saved up, covering the
460 * initial writepage page and also any probed pages.
461 *
462 * Because we may have multiple ioends spanning a page, we need to start
463 * writeback on all the buffers before we submit them for I/O. If we mark the
464 * buffers as we got, then we can end up with a page that only has buffers
465 * marked async write and I/O complete on can occur before we mark the other
466 * buffers async write.
467 *
468 * The end result of this is that we trip a bug in end_page_writeback() because
469 * we call it twice for the one page as the code in end_buffer_async_write()
470 * assumes that all buffers on the page are started at the same time.
471 *
472 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000473 * buffer_heads, and then submit them for I/O on the second pass.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100474 *
475 * If @fail is non-zero, it means that we have a situation where some part of
476 * the submission process has failed after we have marked paged for writeback
477 * and unlocked them. In this situation, we need to fail the ioend chain rather
478 * than submit it to IO. This typically only happens on a filesystem shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100479 */
480STATIC void
481xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000482 struct writeback_control *wbc,
Dave Chinner7bf7f352012-11-12 22:09:45 +1100483 xfs_ioend_t *ioend,
484 int fail)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100485{
David Chinnerd88992f2006-01-18 13:38:12 +1100486 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100487 xfs_ioend_t *next;
488 struct buffer_head *bh;
489 struct bio *bio;
490 sector_t lastblock = 0;
491
David Chinnerd88992f2006-01-18 13:38:12 +1100492 /* Pass 1 - start writeback */
493 do {
494 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000495 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100496 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100497 } while ((ioend = next) != NULL);
498
499 /* Pass 2 - submit I/O */
500 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100501 do {
502 next = ioend->io_list;
503 bio = NULL;
504
Dave Chinner7bf7f352012-11-12 22:09:45 +1100505 /*
506 * If we are failing the IO now, just mark the ioend with an
507 * error and finish it. This will run IO completion immediately
508 * as there is only one reference to the ioend at this point in
509 * time.
510 */
511 if (fail) {
512 ioend->io_error = -fail;
513 xfs_finish_ioend(ioend);
514 continue;
515 }
516
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100517 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100518
519 if (!bio) {
520 retry:
521 bio = xfs_alloc_ioend_bio(bh);
522 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000523 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100524 goto retry;
525 }
526
527 if (bio_add_buffer(bio, bh) != bh->b_size) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000528 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100529 goto retry;
530 }
531
532 lastblock = bh->b_blocknr;
533 }
534 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000535 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000536 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100537 } while ((ioend = next) != NULL);
538}
539
540/*
541 * Cancel submission of all buffer_heads so far in this endio.
542 * Toss the endio too. Only ever called for the initial page
543 * in a writepage request, so only ever one page.
544 */
545STATIC void
546xfs_cancel_ioend(
547 xfs_ioend_t *ioend)
548{
549 xfs_ioend_t *next;
550 struct buffer_head *bh, *next_bh;
551
552 do {
553 next = ioend->io_list;
554 bh = ioend->io_buffer_head;
555 do {
556 next_bh = bh->b_private;
557 clear_buffer_async_write(bh);
558 unlock_buffer(bh);
559 } while ((bh = next_bh) != NULL);
560
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100561 mempool_free(ioend, xfs_ioend_pool);
562 } while ((ioend = next) != NULL);
563}
564
565/*
566 * Test to see if we've been building up a completion structure for
567 * earlier buffers -- if so, we try to append to this ioend if we
568 * can, otherwise we finish off any current ioend and start another.
569 * Return true if we've finished the given ioend.
570 */
571STATIC void
572xfs_add_to_ioend(
573 struct inode *inode,
574 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100575 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100576 unsigned int type,
577 xfs_ioend_t **result,
578 int need_ioend)
579{
580 xfs_ioend_t *ioend = *result;
581
582 if (!ioend || need_ioend || type != ioend->io_type) {
583 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100584
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100585 ioend = xfs_alloc_ioend(inode, type);
586 ioend->io_offset = offset;
587 ioend->io_buffer_head = bh;
588 ioend->io_buffer_tail = bh;
589 if (previous)
590 previous->io_list = ioend;
591 *result = ioend;
592 } else {
593 ioend->io_buffer_tail->b_private = bh;
594 ioend->io_buffer_tail = bh;
595 }
596
597 bh->b_private = NULL;
598 ioend->io_size += bh->b_size;
599}
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100602xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000603 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100604 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000605 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000606 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100607{
608 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000609 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000610 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
611 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100612
Christoph Hellwig207d0412010-04-28 12:28:56 +0000613 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
614 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100615
Christoph Hellwige5131822010-04-28 12:28:55 +0000616 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000617 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100618
Christoph Hellwig046f1682010-04-28 12:28:52 +0000619 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100620
621 bh->b_blocknr = bn;
622 set_buffer_mapped(bh);
623}
624
625STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000627 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000629 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000630 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000632 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
633 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
Christoph Hellwig207d0412010-04-28 12:28:56 +0000635 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 set_buffer_mapped(bh);
637 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100638 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639}
640
641/*
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100642 * Test if a given page is suitable for writing as part of an unwritten
643 * or delayed allocate extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 */
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100645STATIC int
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000646xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100647 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100648 unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 if (PageWriteback(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100651 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
653 if (page->mapping && page_has_buffers(page)) {
654 struct buffer_head *bh, *head;
655 int acceptable = 0;
656
657 bh = head = page_buffers(page);
658 do {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100659 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500660 acceptable += (type == XFS_IO_UNWRITTEN);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100661 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500662 acceptable += (type == XFS_IO_DELALLOC);
David Chinner2ddee842006-03-22 12:47:40 +1100663 else if (buffer_dirty(bh) && buffer_mapped(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500664 acceptable += (type == XFS_IO_OVERWRITE);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100665 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 } while ((bh = bh->b_this_page) != head);
668
669 if (acceptable)
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100670 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 }
672
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100673 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676/*
677 * Allocate & map buffers for page given the extent map. Write it out.
678 * except for the original page of a writepage, this is called on
679 * delalloc/unwritten pages only, for the original page it is possible
680 * that the page has no mapping at all.
681 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100682STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683xfs_convert_page(
684 struct inode *inode,
685 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100686 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000687 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100688 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000689 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100691 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100692 xfs_off_t end_offset;
693 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100694 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700695 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100696 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100697 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100699 if (page->index != tindex)
700 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200701 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100702 goto fail;
703 if (PageWriteback(page))
704 goto fail_unlock_page;
705 if (page->mapping != inode->i_mapping)
706 goto fail_unlock_page;
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000707 if (!xfs_check_page_type(page, (*ioendp)->io_type))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100708 goto fail_unlock_page;
709
Nathan Scott24e17b52005-05-05 13:33:20 -0700710 /*
711 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000712 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100713 *
714 * Derivation:
715 *
716 * End offset is the highest offset that this page should represent.
717 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
718 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
719 * hence give us the correct page_dirty count. On any other page,
720 * it will be zero and in that case we need page_dirty to be the
721 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700722 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100723 end_offset = min_t(unsigned long long,
724 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
725 i_size_read(inode));
726
Nathan Scott24e17b52005-05-05 13:33:20 -0700727 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100728 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
729 PAGE_CACHE_SIZE);
730 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
731 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 bh = head = page_buffers(page);
734 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100735 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100737 if (!buffer_uptodate(bh))
738 uptodate = 0;
739 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
740 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 continue;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100742 }
743
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000744 if (buffer_unwritten(bh) || buffer_delay(bh) ||
745 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100746 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500747 type = XFS_IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000748 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500749 type = XFS_IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000750 else
Alain Renaud0d882a32012-05-22 15:56:21 -0500751 type = XFS_IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100752
Christoph Hellwig558e6892010-04-28 12:28:58 +0000753 if (!xfs_imap_valid(inode, imap, offset)) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100754 done = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100755 continue;
756 }
757
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000758 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -0500759 if (type != XFS_IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000760 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000761 xfs_add_to_ioend(inode, bh, offset, type,
762 ioendp, done);
763
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100764 page_dirty--;
765 count++;
766 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000767 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100769 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100771 if (uptodate && bh == head)
772 SetPageUptodate(page);
773
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000774 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000775 if (--wbc->nr_to_write <= 0 &&
776 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000777 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000779 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100780
781 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100782 fail_unlock_page:
783 unlock_page(page);
784 fail:
785 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786}
787
788/*
789 * Convert & write out a cluster of pages in the same extent as defined
790 * by mp and following the start page.
791 */
792STATIC void
793xfs_cluster_write(
794 struct inode *inode,
795 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000796 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100797 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 pgoff_t tlast)
800{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100801 struct pagevec pvec;
802 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100804 pagevec_init(&pvec, 0);
805 while (!done && tindex <= tlast) {
806 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
807
808 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100810
811 for (i = 0; i < pagevec_count(&pvec); i++) {
812 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000813 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100814 if (done)
815 break;
816 }
817
818 pagevec_release(&pvec);
819 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 }
821}
822
Dave Chinner3ed3a432010-03-05 02:00:42 +0000823STATIC void
824xfs_vm_invalidatepage(
825 struct page *page,
826 unsigned long offset)
827{
828 trace_xfs_invalidatepage(page->mapping->host, page, offset);
829 block_invalidatepage(page, offset);
830}
831
832/*
833 * If the page has delalloc buffers on it, we need to punch them out before we
834 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
835 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
836 * is done on that same region - the delalloc extent is returned when none is
837 * supposed to be there.
838 *
839 * We prevent this by truncating away the delalloc regions on the page before
840 * invalidating it. Because they are delalloc, we can do this without needing a
841 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
842 * truncation without a transaction as there is no space left for block
843 * reservation (typically why we see a ENOSPC in writeback).
844 *
845 * This is not a performance critical path, so for now just do the punching a
846 * buffer head at a time.
847 */
848STATIC void
849xfs_aops_discard_page(
850 struct page *page)
851{
852 struct inode *inode = page->mapping->host;
853 struct xfs_inode *ip = XFS_I(inode);
854 struct buffer_head *bh, *head;
855 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000856
Alain Renaud0d882a32012-05-22 15:56:21 -0500857 if (!xfs_check_page_type(page, XFS_IO_DELALLOC))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000858 goto out_invalidate;
859
Dave Chinnere8c37532010-03-15 02:36:35 +0000860 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
861 goto out_invalidate;
862
Dave Chinner4f107002011-03-07 10:00:35 +1100863 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000864 "page discard on page %p, inode 0x%llx, offset %llu.",
865 page, ip->i_ino, offset);
866
867 xfs_ilock(ip, XFS_ILOCK_EXCL);
868 bh = head = page_buffers(page);
869 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000870 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100871 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000872
873 if (!buffer_delay(bh))
874 goto next_buffer;
875
Dave Chinnerc726de42010-11-30 15:14:39 +1100876 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
877 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000878 if (error) {
879 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000880 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100881 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000882 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000883 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000884 break;
885 }
886next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100887 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000888
889 } while ((bh = bh->b_this_page) != head);
890
891 xfs_iunlock(ip, XFS_ILOCK_EXCL);
892out_invalidate:
893 xfs_vm_invalidatepage(page, 0);
894 return;
895}
896
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000898 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000900 * For delalloc space on the page we need to allocate space and flush it.
901 * For unwritten space on the page we need to start the conversion to
902 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000903 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905STATIC int
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000906xfs_vm_writepage(
907 struct page *page,
908 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909{
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000910 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100911 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000912 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100913 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100915 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000917 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000918 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000919 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000920 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000921 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000923 trace_xfs_writepage(inode, page, 0);
924
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000925 ASSERT(page_has_buffers(page));
926
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000927 /*
928 * Refuse to write the page out if we are called from reclaim context.
929 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400930 * This avoids stack overflows when called from deeply used stacks in
931 * random callers for direct reclaim or memcg reclaim. We explicitly
932 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000933 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700934 * This should never happen except in the case of a VM regression so
935 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000936 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700937 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
938 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000939 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000940
941 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200942 * Given that we do not allow direct reclaim to call us, we should
943 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000944 */
Christoph Hellwig680a6472011-07-08 14:34:05 +0200945 if (WARN_ON(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000946 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 /* Is this page beyond the end of the file? */
949 offset = i_size_read(inode);
950 end_index = offset >> PAGE_CACHE_SHIFT;
951 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
952 if (page->index >= end_index) {
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400953 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
954
955 /*
956 * Just skip the page if it is fully outside i_size, e.g. due
957 * to a truncate operation that is in progress.
958 */
959 if (page->index >= end_index + 1 || offset_into_page == 0) {
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000960 unlock_page(page);
Nathan Scott19d5bcf2005-11-02 15:14:09 +1100961 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 }
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400963
964 /*
965 * The page straddles i_size. It must be zeroed out on each
966 * and every writepage invocation because it may be mmapped.
967 * "A file is mapped in multiples of the page size. For a file
968 * that is not a multiple of the page size, the remaining
969 * memory is zeroed when mapped, and writes to that region are
970 * not written out to the file."
971 */
972 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 }
974
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100975 end_offset = min_t(unsigned long long,
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000976 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
977 offset);
Nathan Scott24e17b52005-05-05 13:33:20 -0700978 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -0700979
Nathan Scott24e17b52005-05-05 13:33:20 -0700980 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100981 offset = page_offset(page);
Alain Renaud0d882a32012-05-22 15:56:21 -0500982 type = XFS_IO_OVERWRITE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000983
Christoph Hellwigdbcdde32011-07-08 14:34:14 +0200984 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000985 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100986
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +0000988 int new_ioend = 0;
989
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 if (offset >= end_offset)
991 break;
992 if (!buffer_uptodate(bh))
993 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Eric Sandeen3d9b02e2010-06-24 09:45:30 +1000995 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +0000996 * set_page_dirty dirties all buffers in a page, independent
997 * of their state. The dirty state however is entirely
998 * meaningless for holes (!mapped && uptodate), so skip
999 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001000 */
1001 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001002 imap_valid = 0;
1003 continue;
1004 }
1005
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001006 if (buffer_unwritten(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001007 if (type != XFS_IO_UNWRITTEN) {
1008 type = XFS_IO_UNWRITTEN;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001009 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001010 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001011 } else if (buffer_delay(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001012 if (type != XFS_IO_DELALLOC) {
1013 type = XFS_IO_DELALLOC;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001014 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001016 } else if (buffer_uptodate(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001017 if (type != XFS_IO_OVERWRITE) {
1018 type = XFS_IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +00001019 imap_valid = 0;
1020 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001021 } else {
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001022 if (PageUptodate(page))
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001023 ASSERT(buffer_mapped(bh));
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001024 /*
1025 * This buffer is not uptodate and will not be
1026 * written to disk. Ensure that we will put any
1027 * subsequent writeable buffers into a new
1028 * ioend.
1029 */
1030 imap_valid = 0;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001031 continue;
1032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001034 if (imap_valid)
1035 imap_valid = xfs_imap_valid(inode, &imap, offset);
1036 if (!imap_valid) {
1037 /*
1038 * If we didn't have a valid mapping then we need to
1039 * put the new mapping into a separate ioend structure.
1040 * This ensures non-contiguous extents always have
1041 * separate ioends, which is particularly important
1042 * for unwritten extent conversion at I/O completion
1043 * time.
1044 */
1045 new_ioend = 1;
1046 err = xfs_map_blocks(inode, offset, &imap, type,
1047 nonblocking);
1048 if (err)
1049 goto error;
1050 imap_valid = xfs_imap_valid(inode, &imap, offset);
1051 }
1052 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001053 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -05001054 if (type != XFS_IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001055 xfs_map_at_offset(inode, bh, &imap, offset);
1056 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1057 new_ioend);
1058 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001060
1061 if (!iohead)
1062 iohead = ioend;
1063
1064 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
1066 if (uptodate && bh == head)
1067 SetPageUptodate(page);
1068
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001069 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Dave Chinner7bf7f352012-11-12 22:09:45 +11001071 /* if there is no IO to be submitted for this page, we are done */
1072 if (!ioend)
1073 return 0;
1074
1075 ASSERT(iohead);
1076
1077 /*
1078 * Any errors from this point onwards need tobe reported through the IO
1079 * completion path as we have marked the initial page as under writeback
1080 * and unlocked it.
1081 */
1082 if (imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001083 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001084
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001085 end_index = imap.br_startoff + imap.br_blockcount;
1086
1087 /* to bytes */
1088 end_index <<= inode->i_blkbits;
1089
1090 /* to pages */
1091 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1092
1093 /* check against file size */
1094 if (end_index > last_index)
1095 end_index = last_index;
1096
Christoph Hellwig207d0412010-04-28 12:28:56 +00001097 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001098 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 }
1100
Christoph Hellwig281627d2012-03-13 08:41:05 +00001101
Dave Chinner7bf7f352012-11-12 22:09:45 +11001102 /*
1103 * Reserve log space if we might write beyond the on-disk inode size.
1104 */
1105 err = 0;
1106 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1107 err = xfs_setfilesize_trans_alloc(ioend);
1108
1109 xfs_submit_ioend(wbc, iohead, err);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001110
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001111 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001114 if (iohead)
1115 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001117 if (err == -EAGAIN)
1118 goto redirty;
1119
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001120 xfs_aops_discard_page(page);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001121 ClearPageUptodate(page);
1122 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001124
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001125redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001126 redirty_page_for_writepage(wbc, page);
1127 unlock_page(page);
1128 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001129}
1130
Nathan Scott7d4fb402006-06-09 15:27:16 +10001131STATIC int
1132xfs_vm_writepages(
1133 struct address_space *mapping,
1134 struct writeback_control *wbc)
1135{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001136 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001137 return generic_writepages(mapping, wbc);
1138}
1139
Nathan Scottf51623b2006-03-14 13:26:27 +11001140/*
1141 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001142 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001143 * have buffer heads in this call.
1144 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001145 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001146 */
1147STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001148xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001149 struct page *page,
1150 gfp_t gfp_mask)
1151{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001152 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001153
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001154 trace_xfs_releasepage(page->mapping->host, page, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001155
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001156 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001157
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001158 if (WARN_ON(delalloc))
1159 return 0;
1160 if (WARN_ON(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001161 return 0;
1162
Nathan Scottf51623b2006-03-14 13:26:27 +11001163 return try_to_free_buffers(page);
1164}
1165
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001167__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 struct inode *inode,
1169 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 struct buffer_head *bh_result,
1171 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001172 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001174 struct xfs_inode *ip = XFS_I(inode);
1175 struct xfs_mount *mp = ip->i_mount;
1176 xfs_fileoff_t offset_fsb, end_fsb;
1177 int error = 0;
1178 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001179 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001180 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001181 xfs_off_t offset;
1182 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001183 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001184
1185 if (XFS_FORCED_SHUTDOWN(mp))
1186 return -XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001188 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001189 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1190 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001191
1192 if (!create && direct && offset >= i_size_read(inode))
1193 return 0;
1194
Dave Chinner507630b2012-03-27 10:34:50 -04001195 /*
1196 * Direct I/O is usually done on preallocated files, so try getting
1197 * a block mapping without an exclusive lock first. For buffered
1198 * writes we already have the exclusive iolock anyway, so avoiding
1199 * a lock roundtrip here by taking the ilock exclusive from the
1200 * beginning is a useful micro optimization.
1201 */
1202 if (create && !direct) {
Christoph Hellwiga206c812010-12-10 08:42:20 +00001203 lockmode = XFS_ILOCK_EXCL;
1204 xfs_ilock(ip, lockmode);
1205 } else {
1206 lockmode = xfs_ilock_map_shared(ip);
1207 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001208
Dave Chinnerd2c28192012-06-08 15:44:53 +10001209 ASSERT(offset <= mp->m_super->s_maxbytes);
1210 if (offset + size > mp->m_super->s_maxbytes)
1211 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001212 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1213 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1214
Dave Chinner5c8ed202011-09-18 20:40:45 +00001215 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1216 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001218 goto out_unlock;
1219
1220 if (create &&
1221 (!nimaps ||
1222 (imap.br_startblock == HOLESTARTBLOCK ||
1223 imap.br_startblock == DELAYSTARTBLOCK))) {
Dave Chinneraff3a9e2012-04-23 15:58:44 +10001224 if (direct || xfs_get_extsz_hint(ip)) {
Dave Chinner507630b2012-03-27 10:34:50 -04001225 /*
1226 * Drop the ilock in preparation for starting the block
1227 * allocation transaction. It will be retaken
1228 * exclusively inside xfs_iomap_write_direct for the
1229 * actual allocation.
1230 */
1231 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001232 error = xfs_iomap_write_direct(ip, offset, size,
1233 &imap, nimaps);
Dave Chinner507630b2012-03-27 10:34:50 -04001234 if (error)
1235 return -error;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001236 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001237 } else {
Dave Chinner507630b2012-03-27 10:34:50 -04001238 /*
1239 * Delalloc reservations do not require a transaction,
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001240 * we can go on without dropping the lock here. If we
1241 * are allocating a new delalloc block, make sure that
1242 * we set the new flag so that we mark the buffer new so
1243 * that we know that it is newly allocated if the write
1244 * fails.
Dave Chinner507630b2012-03-27 10:34:50 -04001245 */
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001246 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1247 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001248 error = xfs_iomap_write_delay(ip, offset, size, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001249 if (error)
1250 goto out_unlock;
1251
1252 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001253 }
Christoph Hellwiga206c812010-12-10 08:42:20 +00001254
1255 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1256 } else if (nimaps) {
1257 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001258 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001259 } else {
1260 trace_xfs_get_blocks_notfound(ip, offset, size);
1261 goto out_unlock;
1262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
Christoph Hellwig207d0412010-04-28 12:28:56 +00001264 if (imap.br_startblock != HOLESTARTBLOCK &&
1265 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001266 /*
1267 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 * the read case (treat as if we're reading into a hole).
1269 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001270 if (create || !ISUNWRITTEN(&imap))
1271 xfs_map_buffer(inode, bh_result, &imap, offset);
1272 if (create && ISUNWRITTEN(&imap)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 if (direct)
1274 bh_result->b_private = inode;
1275 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
1277 }
1278
Nathan Scottc2536662006-03-29 10:44:40 +10001279 /*
1280 * If this is a realtime file, data may be on a different device.
1281 * to that pointed to from the buffer_head b_bdev currently.
1282 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001283 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Nathan Scottc2536662006-03-29 10:44:40 +10001285 /*
David Chinner549054a2007-02-10 18:36:35 +11001286 * If we previously allocated a block out beyond eof and we are now
1287 * coming back to use it then we will need to flag it as new even if it
1288 * has a disk address.
1289 *
1290 * With sub-block writes into unwritten extents we also need to mark
1291 * the buffer as new so that the unwritten parts of the buffer gets
1292 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 */
1294 if (create &&
1295 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001296 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001297 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Christoph Hellwig207d0412010-04-28 12:28:56 +00001300 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 BUG_ON(direct);
1302 if (create) {
1303 set_buffer_uptodate(bh_result);
1304 set_buffer_mapped(bh_result);
1305 set_buffer_delay(bh_result);
1306 }
1307 }
1308
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001309 /*
1310 * If this is O_DIRECT or the mpage code calling tell them how large
1311 * the mapping is, so that we can avoid repeated get_blocks calls.
1312 */
Nathan Scottc2536662006-03-29 10:44:40 +10001313 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001314 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001315
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001316 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1317 mapping_size <<= inode->i_blkbits;
1318
1319 ASSERT(mapping_size > 0);
1320 if (mapping_size > size)
1321 mapping_size = size;
1322 if (mapping_size > LONG_MAX)
1323 mapping_size = LONG_MAX;
1324
1325 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 }
1327
1328 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001329
1330out_unlock:
1331 xfs_iunlock(ip, lockmode);
1332 return -error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333}
1334
1335int
Nathan Scottc2536662006-03-29 10:44:40 +10001336xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 struct inode *inode,
1338 sector_t iblock,
1339 struct buffer_head *bh_result,
1340 int create)
1341{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001342 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343}
1344
1345STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001346xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 struct inode *inode,
1348 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 struct buffer_head *bh_result,
1350 int create)
1351{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001352 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353}
1354
Christoph Hellwig209fb872010-07-18 21:17:11 +00001355/*
1356 * Complete a direct I/O write request.
1357 *
1358 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1359 * need to issue a transaction to convert the range from unwritten to written
1360 * extents. In case this is regular synchronous I/O we just call xfs_end_io
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001361 * to do this and we are done. But in case this was a successful AIO
Christoph Hellwig209fb872010-07-18 21:17:11 +00001362 * request this handler is called from interrupt context, from which we
1363 * can't start transactions. In that case offload the I/O completion to
1364 * the workqueues we also use for buffered I/O completion.
1365 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001366STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001367xfs_end_io_direct_write(
1368 struct kiocb *iocb,
1369 loff_t offset,
1370 ssize_t size,
1371 void *private,
1372 int ret,
1373 bool is_async)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001374{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001375 struct xfs_ioend *ioend = iocb->private;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001376
1377 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001378 * While the generic direct I/O code updates the inode size, it does
1379 * so only after the end_io handler is called, which means our
1380 * end_io handler thinks the on-disk size is outside the in-core
1381 * size. To prevent this just update it a little bit earlier here.
1382 */
1383 if (offset + size > i_size_read(ioend->io_inode))
1384 i_size_write(ioend->io_inode, offset + size);
1385
1386 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001387 * blockdev_direct_IO can return an error even after the I/O
Christoph Hellwigf0973862005-09-05 08:22:52 +10001388 * completion handler was called. Thus we need to protect
1389 * against double-freeing.
1390 */
1391 iocb->private = NULL;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001392
Christoph Hellwig209fb872010-07-18 21:17:11 +00001393 ioend->io_offset = offset;
1394 ioend->io_size = size;
Christoph Hellwigc859cdd2011-08-23 08:28:10 +00001395 ioend->io_iocb = iocb;
1396 ioend->io_result = ret;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001397 if (private && size > 0)
Alain Renaud0d882a32012-05-22 15:56:21 -05001398 ioend->io_type = XFS_IO_UNWRITTEN;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001399
1400 if (is_async) {
Christoph Hellwigc859cdd2011-08-23 08:28:10 +00001401 ioend->io_isasync = 1;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001402 xfs_finish_ioend(ioend);
1403 } else {
1404 xfs_finish_ioend_sync(ioend);
1405 }
Christoph Hellwigf0973862005-09-05 08:22:52 +10001406}
1407
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001409xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 int rw,
1411 struct kiocb *iocb,
1412 const struct iovec *iov,
1413 loff_t offset,
1414 unsigned long nr_segs)
1415{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001416 struct inode *inode = iocb->ki_filp->f_mapping->host;
1417 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001418 struct xfs_ioend *ioend = NULL;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001419 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420
Christoph Hellwig209fb872010-07-18 21:17:11 +00001421 if (rw & WRITE) {
Christoph Hellwig281627d2012-03-13 08:41:05 +00001422 size_t size = iov_length(iov, nr_segs);
1423
1424 /*
Dave Chinner437a2552012-11-28 13:01:00 +11001425 * We cannot preallocate a size update transaction here as we
1426 * don't know whether allocation is necessary or not. Hence we
1427 * can only tell IO completion that one is necessary if we are
1428 * not doing unwritten extent conversion.
Christoph Hellwig281627d2012-03-13 08:41:05 +00001429 */
Alain Renaud0d882a32012-05-22 15:56:21 -05001430 iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
Dave Chinner437a2552012-11-28 13:01:00 +11001431 if (offset + size > XFS_I(inode)->i_d.di_size)
Christoph Hellwig281627d2012-03-13 08:41:05 +00001432 ioend->io_isdirect = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001434 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1435 offset, nr_segs,
1436 xfs_get_blocks_direct,
1437 xfs_end_io_direct_write, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001438 if (ret != -EIOCBQUEUED && iocb->private)
Dave Chinner437a2552012-11-28 13:01:00 +11001439 goto out_destroy_ioend;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001440 } else {
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001441 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1442 offset, nr_segs,
1443 xfs_get_blocks_direct,
1444 NULL, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001445 }
Christoph Hellwig5fe878a2009-12-15 16:47:50 -08001446
Christoph Hellwigf0973862005-09-05 08:22:52 +10001447 return ret;
Christoph Hellwig281627d2012-03-13 08:41:05 +00001448
Christoph Hellwig281627d2012-03-13 08:41:05 +00001449out_destroy_ioend:
1450 xfs_destroy_ioend(ioend);
1451 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001454/*
1455 * Punch out the delalloc blocks we have already allocated.
1456 *
1457 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1458 * as the page is still locked at this point.
1459 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001460STATIC void
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001461xfs_vm_kill_delalloc_range(
1462 struct inode *inode,
1463 loff_t start,
1464 loff_t end)
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001465{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001466 struct xfs_inode *ip = XFS_I(inode);
1467 xfs_fileoff_t start_fsb;
1468 xfs_fileoff_t end_fsb;
1469 int error;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001470
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001471 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1472 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1473 if (end_fsb <= start_fsb)
1474 return;
Dave Chinnerc726de42010-11-30 15:14:39 +11001475
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001476 xfs_ilock(ip, XFS_ILOCK_EXCL);
1477 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1478 end_fsb - start_fsb);
1479 if (error) {
1480 /* something screwed, just bail */
1481 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1482 xfs_alert(ip->i_mount,
1483 "xfs_vm_write_failed: unable to clean up ino %lld",
1484 ip->i_ino);
Dave Chinnerc726de42010-11-30 15:14:39 +11001485 }
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001486 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001487 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001488}
1489
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001490STATIC void
1491xfs_vm_write_failed(
1492 struct inode *inode,
1493 struct page *page,
1494 loff_t pos,
1495 unsigned len)
1496{
1497 loff_t block_offset = pos & PAGE_MASK;
1498 loff_t block_start;
1499 loff_t block_end;
1500 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1501 loff_t to = from + len;
1502 struct buffer_head *bh, *head;
1503
1504 ASSERT(block_offset + from == pos);
1505
1506 head = page_buffers(page);
1507 block_start = 0;
1508 for (bh = head; bh != head || !block_start;
1509 bh = bh->b_this_page, block_start = block_end,
1510 block_offset += bh->b_size) {
1511 block_end = block_start + bh->b_size;
1512
1513 /* skip buffers before the write */
1514 if (block_end <= from)
1515 continue;
1516
1517 /* if the buffer is after the write, we're done */
1518 if (block_start >= to)
1519 break;
1520
1521 if (!buffer_delay(bh))
1522 continue;
1523
1524 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1525 continue;
1526
1527 xfs_vm_kill_delalloc_range(inode, block_offset,
1528 block_offset + bh->b_size);
1529 }
1530
1531}
1532
1533/*
1534 * This used to call block_write_begin(), but it unlocks and releases the page
1535 * on error, and we need that page to be able to punch stale delalloc blocks out
1536 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1537 * the appropriate point.
1538 */
Nathan Scottf51623b2006-03-14 13:26:27 +11001539STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001540xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001541 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001542 struct address_space *mapping,
1543 loff_t pos,
1544 unsigned len,
1545 unsigned flags,
1546 struct page **pagep,
1547 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001548{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001549 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1550 struct page *page;
1551 int status;
Christoph Hellwig155130a2010-06-04 11:29:58 +02001552
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001553 ASSERT(len <= PAGE_CACHE_SIZE);
1554
1555 page = grab_cache_page_write_begin(mapping, index,
1556 flags | AOP_FLAG_NOFS);
1557 if (!page)
1558 return -ENOMEM;
1559
1560 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1561 if (unlikely(status)) {
1562 struct inode *inode = mapping->host;
1563
1564 xfs_vm_write_failed(inode, page, pos, len);
1565 unlock_page(page);
1566
1567 if (pos + len > i_size_read(inode))
1568 truncate_pagecache(inode, pos + len, i_size_read(inode));
1569
1570 page_cache_release(page);
1571 page = NULL;
1572 }
1573
1574 *pagep = page;
1575 return status;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001576}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001577
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001578/*
1579 * On failure, we only need to kill delalloc blocks beyond EOF because they
1580 * will never be written. For blocks within EOF, generic_write_end() zeros them
1581 * so they are safe to leave alone and be written with all the other valid data.
1582 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001583STATIC int
1584xfs_vm_write_end(
1585 struct file *file,
1586 struct address_space *mapping,
1587 loff_t pos,
1588 unsigned len,
1589 unsigned copied,
1590 struct page *page,
1591 void *fsdata)
1592{
1593 int ret;
1594
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001595 ASSERT(len <= PAGE_CACHE_SIZE);
1596
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001597 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001598 if (unlikely(ret < len)) {
1599 struct inode *inode = mapping->host;
1600 size_t isize = i_size_read(inode);
1601 loff_t to = pos + len;
1602
1603 if (to > isize) {
1604 truncate_pagecache(inode, to, isize);
1605 xfs_vm_kill_delalloc_range(inode, isize, to);
1606 }
1607 }
Christoph Hellwig155130a2010-06-04 11:29:58 +02001608 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001609}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001612xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 struct address_space *mapping,
1614 sector_t block)
1615{
1616 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001617 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001619 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001620 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001621 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001622 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001623 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624}
1625
1626STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001627xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 struct file *unused,
1629 struct page *page)
1630{
Nathan Scottc2536662006-03-29 10:44:40 +10001631 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632}
1633
1634STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001635xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 struct file *unused,
1637 struct address_space *mapping,
1638 struct list_head *pages,
1639 unsigned nr_pages)
1640{
Nathan Scottc2536662006-03-29 10:44:40 +10001641 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001644const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001645 .readpage = xfs_vm_readpage,
1646 .readpages = xfs_vm_readpages,
1647 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001648 .writepages = xfs_vm_writepages,
Nathan Scott238f4c52006-03-17 17:26:25 +11001649 .releasepage = xfs_vm_releasepage,
1650 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001651 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001652 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001653 .bmap = xfs_vm_bmap,
1654 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001655 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001656 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001657 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658};