blob: e51e581454e93113c7ead8d81136ab82bd29da94 [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 Hellwig0b1b2132009-12-14 23:14:59 +000031#include "xfs_trace.h"
Dave Chinner3ed3a432010-03-05 02:00:42 +000032#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100033#include "xfs_bmap_util.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 Hellwig0829c362005-09-02 16:58:49 +100089 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090}
91
92/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +000093 * Fast and loose check if this write could update the on-disk inode size.
94 */
95static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
96{
97 return ioend->io_offset + ioend->io_size >
98 XFS_I(ioend->io_inode)->i_d.di_size;
99}
100
Christoph Hellwig281627d2012-03-13 08:41:05 +0000101STATIC int
102xfs_setfilesize_trans_alloc(
103 struct xfs_ioend *ioend)
104{
105 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
106 struct xfs_trans *tp;
107 int error;
108
109 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
110
Jie Liu3d3c8b52013-08-12 20:49:59 +1000111 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000112 if (error) {
113 xfs_trans_cancel(tp, 0);
114 return error;
115 }
116
117 ioend->io_append_trans = tp;
118
119 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100120 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200121 * we released it.
122 */
123 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
124 1, _THIS_IP_);
125 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000126 * We hand off the transaction to the completion thread now, so
127 * clear the flag here.
128 */
129 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
130 return 0;
131}
132
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000133/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000134 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000135 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000136STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000137xfs_setfilesize(
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000138 struct xfs_ioend *ioend)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000139{
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000140 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000141 struct xfs_trans *tp = ioend->io_append_trans;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000142 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000143
Christoph Hellwig281627d2012-03-13 08:41:05 +0000144 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100145 * The transaction may have been allocated in the I/O submission thread,
146 * thus we need to mark ourselves as beeing in a transaction manually.
147 * Similarly for freeze protection.
Christoph Hellwig281627d2012-03-13 08:41:05 +0000148 */
149 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner437a2552012-11-28 13:01:00 +1100150 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
151 0, 1, _THIS_IP_);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000152
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000153 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig6923e682012-02-29 09:53:49 +0000154 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000155 if (!isize) {
156 xfs_iunlock(ip, XFS_ILOCK_EXCL);
157 xfs_trans_cancel(tp, 0);
158 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000159 }
160
Christoph Hellwig281627d2012-03-13 08:41:05 +0000161 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
162
163 ip->i_d.di_size = isize;
164 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
165 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
166
167 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000168}
169
170/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000171 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000172 *
173 * If there is no work to do we might as well call it a day and free the
174 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200175 */
176STATIC void
177xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000178 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200179{
180 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000181 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
182
Alain Renaud0d882a32012-05-22 15:56:21 -0500183 if (ioend->io_type == XFS_IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000184 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Dave Chinner437a2552012-11-28 13:01:00 +1100185 else if (ioend->io_append_trans ||
186 (ioend->io_isdirect && xfs_ioend_is_append(ioend)))
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000187 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000188 else
189 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200190 }
191}
192
193/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000194 * IO write completion.
195 */
196STATIC void
197xfs_end_io(
198 struct work_struct *work)
199{
200 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
201 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000202 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000203
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000204 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000205 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000206 goto done;
207 }
208 if (ioend->io_error)
209 goto done;
210
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000211 /*
212 * For unwritten extents we need to issue transactions to convert a
213 * range to normal written extens after the data I/O has finished.
214 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500215 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000216 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100217 ioend->io_size);
218 } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
219 /*
220 * For direct I/O we do not know if we need to allocate blocks
221 * or not so we can't preallocate an append transaction as that
222 * results in nested reservations and log space deadlocks. Hence
223 * allocate the transaction here. While this is sub-optimal and
224 * can block IO completion for some time, we're stuck with doing
225 * it this way until we can pass the ioend to the direct IO
226 * allocation callbacks and avoid nesting that way.
227 */
228 error = xfs_setfilesize_trans_alloc(ioend);
229 if (error)
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000230 goto done;
Dave Chinner437a2552012-11-28 13:01:00 +1100231 error = xfs_setfilesize(ioend);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000232 } else if (ioend->io_append_trans) {
233 error = xfs_setfilesize(ioend);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000234 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000235 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000236 }
237
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000238done:
Dave Chinner437a2552012-11-28 13:01:00 +1100239 if (error)
240 ioend->io_error = -error;
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000241 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000242}
243
244/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000245 * Call IO completion handling in caller context on the final put of an ioend.
246 */
247STATIC void
248xfs_finish_ioend_sync(
249 struct xfs_ioend *ioend)
250{
251 if (atomic_dec_and_test(&ioend->io_remaining))
252 xfs_end_io(&ioend->io_work);
253}
254
255/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000256 * Allocate and initialise an IO completion structure.
257 * We need to track unwritten extent write completion here initially.
258 * We'll need to extend this for updating the ondisk inode size later
259 * (vs. incore size).
260 */
261STATIC xfs_ioend_t *
262xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100263 struct inode *inode,
264 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000265{
266 xfs_ioend_t *ioend;
267
268 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
269
270 /*
271 * Set the count to 1 initially, which will prevent an I/O
272 * completion callback from happening before we have started
273 * all the I/O from calling the completion routine too early.
274 */
275 atomic_set(&ioend->io_remaining, 1);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000276 ioend->io_isdirect = 0;
Nathan Scott7d04a332006-06-09 14:58:38 +1000277 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100278 ioend->io_list = NULL;
279 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000280 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000281 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100282 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000283 ioend->io_offset = 0;
284 ioend->io_size = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000285 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000286
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000287 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000288 return ioend;
289}
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291STATIC int
292xfs_map_blocks(
293 struct inode *inode,
294 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000295 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000296 int type,
297 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000299 struct xfs_inode *ip = XFS_I(inode);
300 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000301 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000302 xfs_fileoff_t offset_fsb, end_fsb;
303 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000304 int bmapi_flags = XFS_BMAPI_ENTIRE;
305 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Christoph Hellwiga206c812010-12-10 08:42:20 +0000307 if (XFS_FORCED_SHUTDOWN(mp))
308 return -XFS_ERROR(EIO);
309
Alain Renaud0d882a32012-05-22 15:56:21 -0500310 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000311 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000312
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000313 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
314 if (nonblocking)
315 return -XFS_ERROR(EAGAIN);
316 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000317 }
318
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000319 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
320 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000321 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000322
Dave Chinnerd2c28192012-06-08 15:44:53 +1000323 if (offset + count > mp->m_super->s_maxbytes)
324 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000325 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
326 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000327 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
328 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000329 xfs_iunlock(ip, XFS_ILOCK_SHARED);
330
Christoph Hellwiga206c812010-12-10 08:42:20 +0000331 if (error)
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000332 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000333
Alain Renaud0d882a32012-05-22 15:56:21 -0500334 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000335 (!nimaps || isnullstartblock(imap->br_startblock))) {
Christoph Hellwiga206c812010-12-10 08:42:20 +0000336 error = xfs_iomap_write_allocate(ip, offset, count, imap);
337 if (!error)
338 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000339 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000340 }
341
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000342#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500343 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000344 ASSERT(nimaps);
345 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
346 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
347 }
348#endif
349 if (nimaps)
350 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
351 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352}
353
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000354STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000355xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000356 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000357 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000358 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000360 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000361
Christoph Hellwig558e6892010-04-28 12:28:58 +0000362 return offset >= imap->br_startoff &&
363 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
365
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100366/*
367 * BIO completion handler for buffered IO.
368 */
Al Viro782e3b32007-10-12 07:17:47 +0100369STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100370xfs_end_bio(
371 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100372 int error)
373{
374 xfs_ioend_t *ioend = bio->bi_private;
375
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100376 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000377 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100378
379 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100380 bio->bi_private = NULL;
381 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100382 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000383
Christoph Hellwig209fb872010-07-18 21:17:11 +0000384 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100385}
386
387STATIC void
388xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000389 struct writeback_control *wbc,
390 xfs_ioend_t *ioend,
391 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100392{
393 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100394 bio->bi_private = ioend;
395 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100396 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100397}
398
399STATIC struct bio *
400xfs_alloc_ioend_bio(
401 struct buffer_head *bh)
402{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100403 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000404 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100405
406 ASSERT(bio->bi_private == NULL);
407 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
408 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100409 return bio;
410}
411
412STATIC void
413xfs_start_buffer_writeback(
414 struct buffer_head *bh)
415{
416 ASSERT(buffer_mapped(bh));
417 ASSERT(buffer_locked(bh));
418 ASSERT(!buffer_delay(bh));
419 ASSERT(!buffer_unwritten(bh));
420
421 mark_buffer_async_write(bh);
422 set_buffer_uptodate(bh);
423 clear_buffer_dirty(bh);
424}
425
426STATIC void
427xfs_start_page_writeback(
428 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100429 int clear_dirty,
430 int buffers)
431{
432 ASSERT(PageLocked(page));
433 ASSERT(!PageWriteback(page));
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100434 if (clear_dirty)
David Chinner92132022006-12-21 10:24:01 +1100435 clear_page_dirty_for_io(page);
436 set_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100437 unlock_page(page);
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700438 /* If no buffers on the page are to be written, finish it here */
439 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100440 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100441}
442
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000443static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100444{
445 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
446}
447
448/*
David Chinnerd88992f2006-01-18 13:38:12 +1100449 * Submit all of the bios for all of the ioends we have saved up, covering the
450 * initial writepage page and also any probed pages.
451 *
452 * Because we may have multiple ioends spanning a page, we need to start
453 * writeback on all the buffers before we submit them for I/O. If we mark the
454 * buffers as we got, then we can end up with a page that only has buffers
455 * marked async write and I/O complete on can occur before we mark the other
456 * buffers async write.
457 *
458 * The end result of this is that we trip a bug in end_page_writeback() because
459 * we call it twice for the one page as the code in end_buffer_async_write()
460 * assumes that all buffers on the page are started at the same time.
461 *
462 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000463 * buffer_heads, and then submit them for I/O on the second pass.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100464 *
465 * If @fail is non-zero, it means that we have a situation where some part of
466 * the submission process has failed after we have marked paged for writeback
467 * and unlocked them. In this situation, we need to fail the ioend chain rather
468 * than submit it to IO. This typically only happens on a filesystem shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100469 */
470STATIC void
471xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000472 struct writeback_control *wbc,
Dave Chinner7bf7f352012-11-12 22:09:45 +1100473 xfs_ioend_t *ioend,
474 int fail)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100475{
David Chinnerd88992f2006-01-18 13:38:12 +1100476 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100477 xfs_ioend_t *next;
478 struct buffer_head *bh;
479 struct bio *bio;
480 sector_t lastblock = 0;
481
David Chinnerd88992f2006-01-18 13:38:12 +1100482 /* Pass 1 - start writeback */
483 do {
484 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000485 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100486 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100487 } while ((ioend = next) != NULL);
488
489 /* Pass 2 - submit I/O */
490 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100491 do {
492 next = ioend->io_list;
493 bio = NULL;
494
Dave Chinner7bf7f352012-11-12 22:09:45 +1100495 /*
496 * If we are failing the IO now, just mark the ioend with an
497 * error and finish it. This will run IO completion immediately
498 * as there is only one reference to the ioend at this point in
499 * time.
500 */
501 if (fail) {
502 ioend->io_error = -fail;
503 xfs_finish_ioend(ioend);
504 continue;
505 }
506
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100507 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100508
509 if (!bio) {
510 retry:
511 bio = xfs_alloc_ioend_bio(bh);
512 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000513 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100514 goto retry;
515 }
516
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000517 if (xfs_bio_add_buffer(bio, bh) != bh->b_size) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000518 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100519 goto retry;
520 }
521
522 lastblock = bh->b_blocknr;
523 }
524 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000525 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000526 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100527 } while ((ioend = next) != NULL);
528}
529
530/*
531 * Cancel submission of all buffer_heads so far in this endio.
532 * Toss the endio too. Only ever called for the initial page
533 * in a writepage request, so only ever one page.
534 */
535STATIC void
536xfs_cancel_ioend(
537 xfs_ioend_t *ioend)
538{
539 xfs_ioend_t *next;
540 struct buffer_head *bh, *next_bh;
541
542 do {
543 next = ioend->io_list;
544 bh = ioend->io_buffer_head;
545 do {
546 next_bh = bh->b_private;
547 clear_buffer_async_write(bh);
548 unlock_buffer(bh);
549 } while ((bh = next_bh) != NULL);
550
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100551 mempool_free(ioend, xfs_ioend_pool);
552 } while ((ioend = next) != NULL);
553}
554
555/*
556 * Test to see if we've been building up a completion structure for
557 * earlier buffers -- if so, we try to append to this ioend if we
558 * can, otherwise we finish off any current ioend and start another.
559 * Return true if we've finished the given ioend.
560 */
561STATIC void
562xfs_add_to_ioend(
563 struct inode *inode,
564 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100565 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100566 unsigned int type,
567 xfs_ioend_t **result,
568 int need_ioend)
569{
570 xfs_ioend_t *ioend = *result;
571
572 if (!ioend || need_ioend || type != ioend->io_type) {
573 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100574
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100575 ioend = xfs_alloc_ioend(inode, type);
576 ioend->io_offset = offset;
577 ioend->io_buffer_head = bh;
578 ioend->io_buffer_tail = bh;
579 if (previous)
580 previous->io_list = ioend;
581 *result = ioend;
582 } else {
583 ioend->io_buffer_tail->b_private = bh;
584 ioend->io_buffer_tail = bh;
585 }
586
587 bh->b_private = NULL;
588 ioend->io_size += bh->b_size;
589}
590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100592xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000593 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100594 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000595 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000596 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100597{
598 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000599 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000600 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
601 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100602
Christoph Hellwig207d0412010-04-28 12:28:56 +0000603 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
604 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100605
Christoph Hellwige5131822010-04-28 12:28:55 +0000606 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000607 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100608
Christoph Hellwig046f1682010-04-28 12:28:52 +0000609 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100610
611 bh->b_blocknr = bn;
612 set_buffer_mapped(bh);
613}
614
615STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000617 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000619 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000620 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000622 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
623 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
Christoph Hellwig207d0412010-04-28 12:28:56 +0000625 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 set_buffer_mapped(bh);
627 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100628 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
631/*
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100632 * Test if a given page is suitable for writing as part of an unwritten
633 * or delayed allocate extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 */
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100635STATIC int
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000636xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100637 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100638 unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 if (PageWriteback(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100641 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
643 if (page->mapping && page_has_buffers(page)) {
644 struct buffer_head *bh, *head;
645 int acceptable = 0;
646
647 bh = head = page_buffers(page);
648 do {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100649 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500650 acceptable += (type == XFS_IO_UNWRITTEN);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100651 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500652 acceptable += (type == XFS_IO_DELALLOC);
David Chinner2ddee842006-03-22 12:47:40 +1100653 else if (buffer_dirty(bh) && buffer_mapped(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500654 acceptable += (type == XFS_IO_OVERWRITE);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100655 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 } while ((bh = bh->b_this_page) != head);
658
659 if (acceptable)
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100660 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 }
662
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100663 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666/*
667 * Allocate & map buffers for page given the extent map. Write it out.
668 * except for the original page of a writepage, this is called on
669 * delalloc/unwritten pages only, for the original page it is possible
670 * that the page has no mapping at all.
671 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100672STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673xfs_convert_page(
674 struct inode *inode,
675 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100676 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000677 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100678 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000679 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100681 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100682 xfs_off_t end_offset;
683 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100684 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700685 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100686 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100687 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100689 if (page->index != tindex)
690 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200691 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100692 goto fail;
693 if (PageWriteback(page))
694 goto fail_unlock_page;
695 if (page->mapping != inode->i_mapping)
696 goto fail_unlock_page;
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000697 if (!xfs_check_page_type(page, (*ioendp)->io_type))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100698 goto fail_unlock_page;
699
Nathan Scott24e17b52005-05-05 13:33:20 -0700700 /*
701 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000702 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100703 *
704 * Derivation:
705 *
706 * End offset is the highest offset that this page should represent.
707 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
708 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
709 * hence give us the correct page_dirty count. On any other page,
710 * it will be zero and in that case we need page_dirty to be the
711 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700712 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100713 end_offset = min_t(unsigned long long,
714 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
715 i_size_read(inode));
716
Dave Chinner480d7462013-05-20 09:51:08 +1000717 /*
718 * If the current map does not span the entire page we are about to try
719 * to write, then give up. The only way we can write a page that spans
720 * multiple mappings in a single writeback iteration is via the
721 * xfs_vm_writepage() function. Data integrity writeback requires the
722 * entire page to be written in a single attempt, otherwise the part of
723 * the page we don't write here doesn't get written as part of the data
724 * integrity sync.
725 *
726 * For normal writeback, we also don't attempt to write partial pages
727 * here as it simply means that write_cache_pages() will see it under
728 * writeback and ignore the page until some point in the future, at
729 * which time this will be the only page in the file that needs
730 * writeback. Hence for more optimal IO patterns, we should always
731 * avoid partial page writeback due to multiple mappings on a page here.
732 */
733 if (!xfs_imap_valid(inode, imap, end_offset))
734 goto fail_unlock_page;
735
Nathan Scott24e17b52005-05-05 13:33:20 -0700736 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100737 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
738 PAGE_CACHE_SIZE);
739 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
740 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 bh = head = page_buffers(page);
743 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100744 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100746 if (!buffer_uptodate(bh))
747 uptodate = 0;
748 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
749 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 continue;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100751 }
752
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000753 if (buffer_unwritten(bh) || buffer_delay(bh) ||
754 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100755 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500756 type = XFS_IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000757 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500758 type = XFS_IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000759 else
Alain Renaud0d882a32012-05-22 15:56:21 -0500760 type = XFS_IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100761
Christoph Hellwig558e6892010-04-28 12:28:58 +0000762 if (!xfs_imap_valid(inode, imap, offset)) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100763 done = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100764 continue;
765 }
766
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000767 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -0500768 if (type != XFS_IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000769 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000770 xfs_add_to_ioend(inode, bh, offset, type,
771 ioendp, done);
772
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100773 page_dirty--;
774 count++;
775 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000776 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100778 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100780 if (uptodate && bh == head)
781 SetPageUptodate(page);
782
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000783 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000784 if (--wbc->nr_to_write <= 0 &&
785 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000786 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000788 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100789
790 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100791 fail_unlock_page:
792 unlock_page(page);
793 fail:
794 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
796
797/*
798 * Convert & write out a cluster of pages in the same extent as defined
799 * by mp and following the start page.
800 */
801STATIC void
802xfs_cluster_write(
803 struct inode *inode,
804 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000805 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100806 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 pgoff_t tlast)
809{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100810 struct pagevec pvec;
811 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100813 pagevec_init(&pvec, 0);
814 while (!done && tindex <= tlast) {
815 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
816
817 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100819
820 for (i = 0; i < pagevec_count(&pvec); i++) {
821 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000822 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100823 if (done)
824 break;
825 }
826
827 pagevec_release(&pvec);
828 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 }
830}
831
Dave Chinner3ed3a432010-03-05 02:00:42 +0000832STATIC void
833xfs_vm_invalidatepage(
834 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400835 unsigned int offset,
836 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000837{
Lukas Czerner34097df2013-05-21 23:58:01 -0400838 trace_xfs_invalidatepage(page->mapping->host, page, offset,
839 length);
840 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000841}
842
843/*
844 * If the page has delalloc buffers on it, we need to punch them out before we
845 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
846 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
847 * is done on that same region - the delalloc extent is returned when none is
848 * supposed to be there.
849 *
850 * We prevent this by truncating away the delalloc regions on the page before
851 * invalidating it. Because they are delalloc, we can do this without needing a
852 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
853 * truncation without a transaction as there is no space left for block
854 * reservation (typically why we see a ENOSPC in writeback).
855 *
856 * This is not a performance critical path, so for now just do the punching a
857 * buffer head at a time.
858 */
859STATIC void
860xfs_aops_discard_page(
861 struct page *page)
862{
863 struct inode *inode = page->mapping->host;
864 struct xfs_inode *ip = XFS_I(inode);
865 struct buffer_head *bh, *head;
866 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000867
Alain Renaud0d882a32012-05-22 15:56:21 -0500868 if (!xfs_check_page_type(page, XFS_IO_DELALLOC))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000869 goto out_invalidate;
870
Dave Chinnere8c37532010-03-15 02:36:35 +0000871 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
872 goto out_invalidate;
873
Dave Chinner4f107002011-03-07 10:00:35 +1100874 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000875 "page discard on page %p, inode 0x%llx, offset %llu.",
876 page, ip->i_ino, offset);
877
878 xfs_ilock(ip, XFS_ILOCK_EXCL);
879 bh = head = page_buffers(page);
880 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000881 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100882 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000883
884 if (!buffer_delay(bh))
885 goto next_buffer;
886
Dave Chinnerc726de42010-11-30 15:14:39 +1100887 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
888 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000889 if (error) {
890 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000891 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100892 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000893 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000894 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000895 break;
896 }
897next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100898 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000899
900 } while ((bh = bh->b_this_page) != head);
901
902 xfs_iunlock(ip, XFS_ILOCK_EXCL);
903out_invalidate:
Lukas Czernerd47992f2013-05-21 23:17:23 -0400904 xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000905 return;
906}
907
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000909 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000911 * For delalloc space on the page we need to allocate space and flush it.
912 * For unwritten space on the page we need to start the conversion to
913 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000914 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916STATIC int
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000917xfs_vm_writepage(
918 struct page *page,
919 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920{
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000921 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100922 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000923 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100924 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100926 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000928 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000929 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000930 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000931 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000932 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Lukas Czerner34097df2013-05-21 23:58:01 -0400934 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000935
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000936 ASSERT(page_has_buffers(page));
937
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000938 /*
939 * Refuse to write the page out if we are called from reclaim context.
940 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400941 * This avoids stack overflows when called from deeply used stacks in
942 * random callers for direct reclaim or memcg reclaim. We explicitly
943 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000944 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700945 * This should never happen except in the case of a VM regression so
946 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000947 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700948 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
949 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000950 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000951
952 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200953 * Given that we do not allow direct reclaim to call us, we should
954 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000955 */
Christoph Hellwig680a6472011-07-08 14:34:05 +0200956 if (WARN_ON(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000957 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 /* Is this page beyond the end of the file? */
960 offset = i_size_read(inode);
961 end_index = offset >> PAGE_CACHE_SHIFT;
962 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
963 if (page->index >= end_index) {
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400964 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
965
966 /*
Jan Karaff9a28f2013-03-14 14:30:54 +0100967 * Skip the page if it is fully outside i_size, e.g. due to a
968 * truncate operation that is in progress. We must redirty the
969 * page so that reclaim stops reclaiming it. Otherwise
970 * xfs_vm_releasepage() is called on it and gets confused.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400971 */
Jan Karaff9a28f2013-03-14 14:30:54 +0100972 if (page->index >= end_index + 1 || offset_into_page == 0)
973 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400974
975 /*
976 * The page straddles i_size. It must be zeroed out on each
977 * and every writepage invocation because it may be mmapped.
978 * "A file is mapped in multiples of the page size. For a file
979 * that is not a multiple of the page size, the remaining
980 * memory is zeroed when mapped, and writes to that region are
981 * not written out to the file."
982 */
983 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 }
985
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100986 end_offset = min_t(unsigned long long,
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000987 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
988 offset);
Nathan Scott24e17b52005-05-05 13:33:20 -0700989 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -0700990
Nathan Scott24e17b52005-05-05 13:33:20 -0700991 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100992 offset = page_offset(page);
Alain Renaud0d882a32012-05-22 15:56:21 -0500993 type = XFS_IO_OVERWRITE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000994
Christoph Hellwigdbcdde32011-07-08 14:34:14 +0200995 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000996 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100997
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +0000999 int new_ioend = 0;
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 if (offset >= end_offset)
1002 break;
1003 if (!buffer_uptodate(bh))
1004 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001006 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +00001007 * set_page_dirty dirties all buffers in a page, independent
1008 * of their state. The dirty state however is entirely
1009 * meaningless for holes (!mapped && uptodate), so skip
1010 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001011 */
1012 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001013 imap_valid = 0;
1014 continue;
1015 }
1016
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001017 if (buffer_unwritten(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001018 if (type != XFS_IO_UNWRITTEN) {
1019 type = XFS_IO_UNWRITTEN;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001020 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001021 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001022 } else if (buffer_delay(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001023 if (type != XFS_IO_DELALLOC) {
1024 type = XFS_IO_DELALLOC;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001025 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001027 } else if (buffer_uptodate(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001028 if (type != XFS_IO_OVERWRITE) {
1029 type = XFS_IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +00001030 imap_valid = 0;
1031 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001032 } else {
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001033 if (PageUptodate(page))
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001034 ASSERT(buffer_mapped(bh));
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001035 /*
1036 * This buffer is not uptodate and will not be
1037 * written to disk. Ensure that we will put any
1038 * subsequent writeable buffers into a new
1039 * ioend.
1040 */
1041 imap_valid = 0;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001042 continue;
1043 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001045 if (imap_valid)
1046 imap_valid = xfs_imap_valid(inode, &imap, offset);
1047 if (!imap_valid) {
1048 /*
1049 * If we didn't have a valid mapping then we need to
1050 * put the new mapping into a separate ioend structure.
1051 * This ensures non-contiguous extents always have
1052 * separate ioends, which is particularly important
1053 * for unwritten extent conversion at I/O completion
1054 * time.
1055 */
1056 new_ioend = 1;
1057 err = xfs_map_blocks(inode, offset, &imap, type,
1058 nonblocking);
1059 if (err)
1060 goto error;
1061 imap_valid = xfs_imap_valid(inode, &imap, offset);
1062 }
1063 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001064 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -05001065 if (type != XFS_IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001066 xfs_map_at_offset(inode, bh, &imap, offset);
1067 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1068 new_ioend);
1069 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001071
1072 if (!iohead)
1073 iohead = ioend;
1074
1075 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 if (uptodate && bh == head)
1078 SetPageUptodate(page);
1079
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001080 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Dave Chinner7bf7f352012-11-12 22:09:45 +11001082 /* if there is no IO to be submitted for this page, we are done */
1083 if (!ioend)
1084 return 0;
1085
1086 ASSERT(iohead);
1087
1088 /*
1089 * Any errors from this point onwards need tobe reported through the IO
1090 * completion path as we have marked the initial page as under writeback
1091 * and unlocked it.
1092 */
1093 if (imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001094 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001095
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001096 end_index = imap.br_startoff + imap.br_blockcount;
1097
1098 /* to bytes */
1099 end_index <<= inode->i_blkbits;
1100
1101 /* to pages */
1102 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1103
1104 /* check against file size */
1105 if (end_index > last_index)
1106 end_index = last_index;
1107
Christoph Hellwig207d0412010-04-28 12:28:56 +00001108 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001109 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
1111
Christoph Hellwig281627d2012-03-13 08:41:05 +00001112
Dave Chinner7bf7f352012-11-12 22:09:45 +11001113 /*
1114 * Reserve log space if we might write beyond the on-disk inode size.
1115 */
1116 err = 0;
1117 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1118 err = xfs_setfilesize_trans_alloc(ioend);
1119
1120 xfs_submit_ioend(wbc, iohead, err);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001121
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001122 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001125 if (iohead)
1126 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001128 if (err == -EAGAIN)
1129 goto redirty;
1130
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001131 xfs_aops_discard_page(page);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001132 ClearPageUptodate(page);
1133 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001135
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001136redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001137 redirty_page_for_writepage(wbc, page);
1138 unlock_page(page);
1139 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001140}
1141
Nathan Scott7d4fb402006-06-09 15:27:16 +10001142STATIC int
1143xfs_vm_writepages(
1144 struct address_space *mapping,
1145 struct writeback_control *wbc)
1146{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001147 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001148 return generic_writepages(mapping, wbc);
1149}
1150
Nathan Scottf51623b2006-03-14 13:26:27 +11001151/*
1152 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001153 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001154 * have buffer heads in this call.
1155 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001156 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001157 */
1158STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001159xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001160 struct page *page,
1161 gfp_t gfp_mask)
1162{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001163 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001164
Lukas Czerner34097df2013-05-21 23:58:01 -04001165 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001166
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001167 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001168
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001169 if (WARN_ON(delalloc))
1170 return 0;
1171 if (WARN_ON(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001172 return 0;
1173
Nathan Scottf51623b2006-03-14 13:26:27 +11001174 return try_to_free_buffers(page);
1175}
1176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001178__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 struct inode *inode,
1180 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 struct buffer_head *bh_result,
1182 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001183 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001185 struct xfs_inode *ip = XFS_I(inode);
1186 struct xfs_mount *mp = ip->i_mount;
1187 xfs_fileoff_t offset_fsb, end_fsb;
1188 int error = 0;
1189 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001190 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001191 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001192 xfs_off_t offset;
1193 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001194 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001195
1196 if (XFS_FORCED_SHUTDOWN(mp))
1197 return -XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001199 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001200 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1201 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001202
1203 if (!create && direct && offset >= i_size_read(inode))
1204 return 0;
1205
Dave Chinner507630b2012-03-27 10:34:50 -04001206 /*
1207 * Direct I/O is usually done on preallocated files, so try getting
1208 * a block mapping without an exclusive lock first. For buffered
1209 * writes we already have the exclusive iolock anyway, so avoiding
1210 * a lock roundtrip here by taking the ilock exclusive from the
1211 * beginning is a useful micro optimization.
1212 */
1213 if (create && !direct) {
Christoph Hellwiga206c812010-12-10 08:42:20 +00001214 lockmode = XFS_ILOCK_EXCL;
1215 xfs_ilock(ip, lockmode);
1216 } else {
1217 lockmode = xfs_ilock_map_shared(ip);
1218 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001219
Dave Chinnerd2c28192012-06-08 15:44:53 +10001220 ASSERT(offset <= mp->m_super->s_maxbytes);
1221 if (offset + size > mp->m_super->s_maxbytes)
1222 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001223 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1224 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1225
Dave Chinner5c8ed202011-09-18 20:40:45 +00001226 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1227 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001229 goto out_unlock;
1230
1231 if (create &&
1232 (!nimaps ||
1233 (imap.br_startblock == HOLESTARTBLOCK ||
1234 imap.br_startblock == DELAYSTARTBLOCK))) {
Dave Chinneraff3a9e2012-04-23 15:58:44 +10001235 if (direct || xfs_get_extsz_hint(ip)) {
Dave Chinner507630b2012-03-27 10:34:50 -04001236 /*
1237 * Drop the ilock in preparation for starting the block
1238 * allocation transaction. It will be retaken
1239 * exclusively inside xfs_iomap_write_direct for the
1240 * actual allocation.
1241 */
1242 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001243 error = xfs_iomap_write_direct(ip, offset, size,
1244 &imap, nimaps);
Dave Chinner507630b2012-03-27 10:34:50 -04001245 if (error)
1246 return -error;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001247 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001248 } else {
Dave Chinner507630b2012-03-27 10:34:50 -04001249 /*
1250 * Delalloc reservations do not require a transaction,
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001251 * we can go on without dropping the lock here. If we
1252 * are allocating a new delalloc block, make sure that
1253 * we set the new flag so that we mark the buffer new so
1254 * that we know that it is newly allocated if the write
1255 * fails.
Dave Chinner507630b2012-03-27 10:34:50 -04001256 */
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001257 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1258 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001259 error = xfs_iomap_write_delay(ip, offset, size, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001260 if (error)
1261 goto out_unlock;
1262
1263 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001264 }
Christoph Hellwiga206c812010-12-10 08:42:20 +00001265
1266 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1267 } else if (nimaps) {
1268 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001269 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001270 } else {
1271 trace_xfs_get_blocks_notfound(ip, offset, size);
1272 goto out_unlock;
1273 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Christoph Hellwig207d0412010-04-28 12:28:56 +00001275 if (imap.br_startblock != HOLESTARTBLOCK &&
1276 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001277 /*
1278 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 * the read case (treat as if we're reading into a hole).
1280 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001281 if (create || !ISUNWRITTEN(&imap))
1282 xfs_map_buffer(inode, bh_result, &imap, offset);
1283 if (create && ISUNWRITTEN(&imap)) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001284 if (direct) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 bh_result->b_private = inode;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001286 set_buffer_defer_completion(bh_result);
1287 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 }
1290 }
1291
Nathan Scottc2536662006-03-29 10:44:40 +10001292 /*
1293 * If this is a realtime file, data may be on a different device.
1294 * to that pointed to from the buffer_head b_bdev currently.
1295 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001296 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297
Nathan Scottc2536662006-03-29 10:44:40 +10001298 /*
David Chinner549054a2007-02-10 18:36:35 +11001299 * If we previously allocated a block out beyond eof and we are now
1300 * coming back to use it then we will need to flag it as new even if it
1301 * has a disk address.
1302 *
1303 * With sub-block writes into unwritten extents we also need to mark
1304 * the buffer as new so that the unwritten parts of the buffer gets
1305 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 */
1307 if (create &&
1308 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001309 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001310 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Christoph Hellwig207d0412010-04-28 12:28:56 +00001313 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 BUG_ON(direct);
1315 if (create) {
1316 set_buffer_uptodate(bh_result);
1317 set_buffer_mapped(bh_result);
1318 set_buffer_delay(bh_result);
1319 }
1320 }
1321
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001322 /*
1323 * If this is O_DIRECT or the mpage code calling tell them how large
1324 * the mapping is, so that we can avoid repeated get_blocks calls.
1325 */
Nathan Scottc2536662006-03-29 10:44:40 +10001326 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001327 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001328
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001329 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1330 mapping_size <<= inode->i_blkbits;
1331
1332 ASSERT(mapping_size > 0);
1333 if (mapping_size > size)
1334 mapping_size = size;
1335 if (mapping_size > LONG_MAX)
1336 mapping_size = LONG_MAX;
1337
1338 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
1340
1341 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001342
1343out_unlock:
1344 xfs_iunlock(ip, lockmode);
1345 return -error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346}
1347
1348int
Nathan Scottc2536662006-03-29 10:44:40 +10001349xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 struct inode *inode,
1351 sector_t iblock,
1352 struct buffer_head *bh_result,
1353 int create)
1354{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001355 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356}
1357
1358STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001359xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 struct inode *inode,
1361 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 struct buffer_head *bh_result,
1363 int create)
1364{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001365 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366}
1367
Christoph Hellwig209fb872010-07-18 21:17:11 +00001368/*
1369 * Complete a direct I/O write request.
1370 *
1371 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1372 * need to issue a transaction to convert the range from unwritten to written
1373 * extents. In case this is regular synchronous I/O we just call xfs_end_io
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001374 * to do this and we are done. But in case this was a successful AIO
Christoph Hellwig209fb872010-07-18 21:17:11 +00001375 * request this handler is called from interrupt context, from which we
1376 * can't start transactions. In that case offload the I/O completion to
1377 * the workqueues we also use for buffered I/O completion.
1378 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001379STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001380xfs_end_io_direct_write(
1381 struct kiocb *iocb,
1382 loff_t offset,
1383 ssize_t size,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001384 void *private)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001385{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001386 struct xfs_ioend *ioend = iocb->private;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001387
1388 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001389 * While the generic direct I/O code updates the inode size, it does
1390 * so only after the end_io handler is called, which means our
1391 * end_io handler thinks the on-disk size is outside the in-core
1392 * size. To prevent this just update it a little bit earlier here.
1393 */
1394 if (offset + size > i_size_read(ioend->io_inode))
1395 i_size_write(ioend->io_inode, offset + size);
1396
1397 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001398 * blockdev_direct_IO can return an error even after the I/O
Christoph Hellwigf0973862005-09-05 08:22:52 +10001399 * completion handler was called. Thus we need to protect
1400 * against double-freeing.
1401 */
1402 iocb->private = NULL;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001403
Christoph Hellwig209fb872010-07-18 21:17:11 +00001404 ioend->io_offset = offset;
1405 ioend->io_size = size;
1406 if (private && size > 0)
Alain Renaud0d882a32012-05-22 15:56:21 -05001407 ioend->io_type = XFS_IO_UNWRITTEN;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001408
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001409 xfs_finish_ioend_sync(ioend);
Christoph Hellwigf0973862005-09-05 08:22:52 +10001410}
1411
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001413xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 int rw,
1415 struct kiocb *iocb,
1416 const struct iovec *iov,
1417 loff_t offset,
1418 unsigned long nr_segs)
1419{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001420 struct inode *inode = iocb->ki_filp->f_mapping->host;
1421 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001422 struct xfs_ioend *ioend = NULL;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001423 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Christoph Hellwig209fb872010-07-18 21:17:11 +00001425 if (rw & WRITE) {
Christoph Hellwig281627d2012-03-13 08:41:05 +00001426 size_t size = iov_length(iov, nr_segs);
1427
1428 /*
Dave Chinner437a2552012-11-28 13:01:00 +11001429 * We cannot preallocate a size update transaction here as we
1430 * don't know whether allocation is necessary or not. Hence we
1431 * can only tell IO completion that one is necessary if we are
1432 * not doing unwritten extent conversion.
Christoph Hellwig281627d2012-03-13 08:41:05 +00001433 */
Alain Renaud0d882a32012-05-22 15:56:21 -05001434 iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
Dave Chinner437a2552012-11-28 13:01:00 +11001435 if (offset + size > XFS_I(inode)->i_d.di_size)
Christoph Hellwig281627d2012-03-13 08:41:05 +00001436 ioend->io_isdirect = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001438 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1439 offset, nr_segs,
1440 xfs_get_blocks_direct,
1441 xfs_end_io_direct_write, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001442 if (ret != -EIOCBQUEUED && iocb->private)
Dave Chinner437a2552012-11-28 13:01:00 +11001443 goto out_destroy_ioend;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001444 } else {
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001445 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1446 offset, nr_segs,
1447 xfs_get_blocks_direct,
1448 NULL, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001449 }
Christoph Hellwig5fe878a2009-12-15 16:47:50 -08001450
Christoph Hellwigf0973862005-09-05 08:22:52 +10001451 return ret;
Christoph Hellwig281627d2012-03-13 08:41:05 +00001452
Christoph Hellwig281627d2012-03-13 08:41:05 +00001453out_destroy_ioend:
1454 xfs_destroy_ioend(ioend);
1455 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001458/*
1459 * Punch out the delalloc blocks we have already allocated.
1460 *
1461 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1462 * as the page is still locked at this point.
1463 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001464STATIC void
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001465xfs_vm_kill_delalloc_range(
1466 struct inode *inode,
1467 loff_t start,
1468 loff_t end)
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001469{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001470 struct xfs_inode *ip = XFS_I(inode);
1471 xfs_fileoff_t start_fsb;
1472 xfs_fileoff_t end_fsb;
1473 int error;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001474
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001475 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1476 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1477 if (end_fsb <= start_fsb)
1478 return;
Dave Chinnerc726de42010-11-30 15:14:39 +11001479
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001480 xfs_ilock(ip, XFS_ILOCK_EXCL);
1481 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1482 end_fsb - start_fsb);
1483 if (error) {
1484 /* something screwed, just bail */
1485 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1486 xfs_alert(ip->i_mount,
1487 "xfs_vm_write_failed: unable to clean up ino %lld",
1488 ip->i_ino);
Dave Chinnerc726de42010-11-30 15:14:39 +11001489 }
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001490 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001491 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001492}
1493
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001494STATIC void
1495xfs_vm_write_failed(
1496 struct inode *inode,
1497 struct page *page,
1498 loff_t pos,
1499 unsigned len)
1500{
Jie Liu58e59852013-07-16 13:11:16 +08001501 loff_t block_offset;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001502 loff_t block_start;
1503 loff_t block_end;
1504 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1505 loff_t to = from + len;
1506 struct buffer_head *bh, *head;
1507
Jie Liu58e59852013-07-16 13:11:16 +08001508 /*
1509 * The request pos offset might be 32 or 64 bit, this is all fine
1510 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1511 * platform, the high 32-bit will be masked off if we evaluate the
1512 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1513 * 0xfffff000 as an unsigned long, hence the result is incorrect
1514 * which could cause the following ASSERT failed in most cases.
1515 * In order to avoid this, we can evaluate the block_offset of the
1516 * start of the page by using shifts rather than masks the mismatch
1517 * problem.
1518 */
1519 block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT;
1520
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001521 ASSERT(block_offset + from == pos);
1522
1523 head = page_buffers(page);
1524 block_start = 0;
1525 for (bh = head; bh != head || !block_start;
1526 bh = bh->b_this_page, block_start = block_end,
1527 block_offset += bh->b_size) {
1528 block_end = block_start + bh->b_size;
1529
1530 /* skip buffers before the write */
1531 if (block_end <= from)
1532 continue;
1533
1534 /* if the buffer is after the write, we're done */
1535 if (block_start >= to)
1536 break;
1537
1538 if (!buffer_delay(bh))
1539 continue;
1540
1541 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1542 continue;
1543
1544 xfs_vm_kill_delalloc_range(inode, block_offset,
1545 block_offset + bh->b_size);
1546 }
1547
1548}
1549
1550/*
1551 * This used to call block_write_begin(), but it unlocks and releases the page
1552 * on error, and we need that page to be able to punch stale delalloc blocks out
1553 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1554 * the appropriate point.
1555 */
Nathan Scottf51623b2006-03-14 13:26:27 +11001556STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001557xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001558 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001559 struct address_space *mapping,
1560 loff_t pos,
1561 unsigned len,
1562 unsigned flags,
1563 struct page **pagep,
1564 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001565{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001566 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1567 struct page *page;
1568 int status;
Christoph Hellwig155130a2010-06-04 11:29:58 +02001569
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001570 ASSERT(len <= PAGE_CACHE_SIZE);
1571
1572 page = grab_cache_page_write_begin(mapping, index,
1573 flags | AOP_FLAG_NOFS);
1574 if (!page)
1575 return -ENOMEM;
1576
1577 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1578 if (unlikely(status)) {
1579 struct inode *inode = mapping->host;
1580
1581 xfs_vm_write_failed(inode, page, pos, len);
1582 unlock_page(page);
1583
1584 if (pos + len > i_size_read(inode))
Kirill A. Shutemov7caef262013-09-12 15:13:56 -07001585 truncate_pagecache(inode, i_size_read(inode));
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001586
1587 page_cache_release(page);
1588 page = NULL;
1589 }
1590
1591 *pagep = page;
1592 return status;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001593}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001594
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001595/*
1596 * On failure, we only need to kill delalloc blocks beyond EOF because they
1597 * will never be written. For blocks within EOF, generic_write_end() zeros them
1598 * so they are safe to leave alone and be written with all the other valid data.
1599 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001600STATIC int
1601xfs_vm_write_end(
1602 struct file *file,
1603 struct address_space *mapping,
1604 loff_t pos,
1605 unsigned len,
1606 unsigned copied,
1607 struct page *page,
1608 void *fsdata)
1609{
1610 int ret;
1611
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001612 ASSERT(len <= PAGE_CACHE_SIZE);
1613
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001614 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001615 if (unlikely(ret < len)) {
1616 struct inode *inode = mapping->host;
1617 size_t isize = i_size_read(inode);
1618 loff_t to = pos + len;
1619
1620 if (to > isize) {
Kirill A. Shutemov7caef262013-09-12 15:13:56 -07001621 truncate_pagecache(inode, isize);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001622 xfs_vm_kill_delalloc_range(inode, isize, to);
1623 }
1624 }
Christoph Hellwig155130a2010-06-04 11:29:58 +02001625 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001626}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001629xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 struct address_space *mapping,
1631 sector_t block)
1632{
1633 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001634 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001636 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001637 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001638 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001639 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001640 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641}
1642
1643STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001644xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 struct file *unused,
1646 struct page *page)
1647{
Nathan Scottc2536662006-03-29 10:44:40 +10001648 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649}
1650
1651STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001652xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 struct file *unused,
1654 struct address_space *mapping,
1655 struct list_head *pages,
1656 unsigned nr_pages)
1657{
Nathan Scottc2536662006-03-29 10:44:40 +10001658 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001661const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001662 .readpage = xfs_vm_readpage,
1663 .readpages = xfs_vm_readpages,
1664 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001665 .writepages = xfs_vm_writepages,
Nathan Scott238f4c52006-03-17 17:26:25 +11001666 .releasepage = xfs_vm_releasepage,
1667 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001668 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001669 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001670 .bmap = xfs_vm_bmap,
1671 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001672 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001673 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001674 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675};