blob: faaf716e2080ad5d41cd86dd05c1ac8f4e3e2fad [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"
Dave Chinner70a98832013-10-23 10:36:05 +110019#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110020#include "xfs_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110024#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110027#include "xfs_trans.h"
Christoph Hellwig281627d2012-03-13 08:41:05 +000028#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_iomap.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"
Dave Chinner68988112013-08-12 20:49:42 +100034#include "xfs_bmap_util.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110035#include "xfs_bmap_btree.h"
36#include "xfs_dinode.h"
Kent Overstreeta27bb332013-05-07 16:19:08 -070037#include <linux/aio.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090038#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110040#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/writeback.h>
42
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000043void
Nathan Scottf51623b2006-03-14 13:26:27 +110044xfs_count_page_state(
45 struct page *page,
46 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110047 int *unwritten)
48{
49 struct buffer_head *bh, *head;
50
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100051 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110052
53 bh = head = page_buffers(page);
54 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100055 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110056 (*unwritten) = 1;
57 else if (buffer_delay(bh))
58 (*delalloc) = 1;
59 } while ((bh = bh->b_this_page) != head);
60}
61
Christoph Hellwig6214ed42007-09-14 15:23:17 +100062STATIC struct block_device *
63xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000064 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100065{
Christoph Hellwig046f1682010-04-28 12:28:52 +000066 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100067 struct xfs_mount *mp = ip->i_mount;
68
Eric Sandeen71ddabb2007-11-23 16:29:42 +110069 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100070 return mp->m_rtdev_targp->bt_bdev;
71 else
72 return mp->m_ddev_targp->bt_bdev;
73}
74
Christoph Hellwig0829c362005-09-02 16:58:49 +100075/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110076 * We're now finished for good with this ioend structure.
77 * Update the page state via the associated buffer_heads,
78 * release holds on the inode and bio, and finally free
79 * up memory. Do not use the ioend after this.
80 */
Christoph Hellwig0829c362005-09-02 16:58:49 +100081STATIC void
82xfs_destroy_ioend(
83 xfs_ioend_t *ioend)
84{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110085 struct buffer_head *bh, *next;
86
87 for (bh = ioend->io_buffer_head; bh; bh = next) {
88 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +100089 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110090 }
Christoph Hellwig583fa582008-12-03 12:20:38 +010091
Christoph Hellwig0829c362005-09-02 16:58:49 +100092 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093}
94
95/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +000096 * Fast and loose check if this write could update the on-disk inode size.
97 */
98static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
99{
100 return ioend->io_offset + ioend->io_size >
101 XFS_I(ioend->io_inode)->i_d.di_size;
102}
103
Christoph Hellwig281627d2012-03-13 08:41:05 +0000104STATIC int
105xfs_setfilesize_trans_alloc(
106 struct xfs_ioend *ioend)
107{
108 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
109 struct xfs_trans *tp;
110 int error;
111
112 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
113
Jie Liu3d3c8b52013-08-12 20:49:59 +1000114 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000115 if (error) {
116 xfs_trans_cancel(tp, 0);
117 return error;
118 }
119
120 ioend->io_append_trans = tp;
121
122 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100123 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200124 * we released it.
125 */
126 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
127 1, _THIS_IP_);
128 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000129 * We hand off the transaction to the completion thread now, so
130 * clear the flag here.
131 */
132 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
133 return 0;
134}
135
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000136/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000137 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000138 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000139STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000140xfs_setfilesize(
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000141 struct xfs_ioend *ioend)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000142{
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000143 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000144 struct xfs_trans *tp = ioend->io_append_trans;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000145 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000146
Christoph Hellwig281627d2012-03-13 08:41:05 +0000147 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100148 * The transaction may have been allocated in the I/O submission thread,
149 * thus we need to mark ourselves as beeing in a transaction manually.
150 * Similarly for freeze protection.
Christoph Hellwig281627d2012-03-13 08:41:05 +0000151 */
152 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Dave Chinner437a2552012-11-28 13:01:00 +1100153 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
154 0, 1, _THIS_IP_);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000155
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000156 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig6923e682012-02-29 09:53:49 +0000157 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000158 if (!isize) {
159 xfs_iunlock(ip, XFS_ILOCK_EXCL);
160 xfs_trans_cancel(tp, 0);
161 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000162 }
163
Christoph Hellwig281627d2012-03-13 08:41:05 +0000164 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
165
166 ip->i_d.di_size = isize;
167 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
168 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
169
170 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000171}
172
173/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000174 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000175 *
176 * If there is no work to do we might as well call it a day and free the
177 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200178 */
179STATIC void
180xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000181 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200182{
183 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000184 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
185
Alain Renaud0d882a32012-05-22 15:56:21 -0500186 if (ioend->io_type == XFS_IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000187 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Dave Chinner437a2552012-11-28 13:01:00 +1100188 else if (ioend->io_append_trans ||
189 (ioend->io_isdirect && xfs_ioend_is_append(ioend)))
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000190 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000191 else
192 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200193 }
194}
195
196/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000197 * IO write completion.
198 */
199STATIC void
200xfs_end_io(
201 struct work_struct *work)
202{
203 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
204 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000205 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000206
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000207 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000208 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000209 goto done;
210 }
211 if (ioend->io_error)
212 goto done;
213
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000214 /*
215 * For unwritten extents we need to issue transactions to convert a
216 * range to normal written extens after the data I/O has finished.
217 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500218 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000219 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100220 ioend->io_size);
221 } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
222 /*
223 * For direct I/O we do not know if we need to allocate blocks
224 * or not so we can't preallocate an append transaction as that
225 * results in nested reservations and log space deadlocks. Hence
226 * allocate the transaction here. While this is sub-optimal and
227 * can block IO completion for some time, we're stuck with doing
228 * it this way until we can pass the ioend to the direct IO
229 * allocation callbacks and avoid nesting that way.
230 */
231 error = xfs_setfilesize_trans_alloc(ioend);
232 if (error)
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000233 goto done;
Dave Chinner437a2552012-11-28 13:01:00 +1100234 error = xfs_setfilesize(ioend);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000235 } else if (ioend->io_append_trans) {
236 error = xfs_setfilesize(ioend);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000237 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000238 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000239 }
240
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000241done:
Dave Chinner437a2552012-11-28 13:01:00 +1100242 if (error)
243 ioend->io_error = -error;
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000244 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000245}
246
247/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000248 * Call IO completion handling in caller context on the final put of an ioend.
249 */
250STATIC void
251xfs_finish_ioend_sync(
252 struct xfs_ioend *ioend)
253{
254 if (atomic_dec_and_test(&ioend->io_remaining))
255 xfs_end_io(&ioend->io_work);
256}
257
258/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000259 * Allocate and initialise an IO completion structure.
260 * We need to track unwritten extent write completion here initially.
261 * We'll need to extend this for updating the ondisk inode size later
262 * (vs. incore size).
263 */
264STATIC xfs_ioend_t *
265xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100266 struct inode *inode,
267 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000268{
269 xfs_ioend_t *ioend;
270
271 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
272
273 /*
274 * Set the count to 1 initially, which will prevent an I/O
275 * completion callback from happening before we have started
276 * all the I/O from calling the completion routine too early.
277 */
278 atomic_set(&ioend->io_remaining, 1);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000279 ioend->io_isdirect = 0;
Nathan Scott7d04a332006-06-09 14:58:38 +1000280 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100281 ioend->io_list = NULL;
282 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000283 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000284 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100285 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000286 ioend->io_offset = 0;
287 ioend->io_size = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000288 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000289
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000290 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000291 return ioend;
292}
293
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294STATIC int
295xfs_map_blocks(
296 struct inode *inode,
297 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000298 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000299 int type,
300 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000302 struct xfs_inode *ip = XFS_I(inode);
303 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000304 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000305 xfs_fileoff_t offset_fsb, end_fsb;
306 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000307 int bmapi_flags = XFS_BMAPI_ENTIRE;
308 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
Christoph Hellwiga206c812010-12-10 08:42:20 +0000310 if (XFS_FORCED_SHUTDOWN(mp))
311 return -XFS_ERROR(EIO);
312
Alain Renaud0d882a32012-05-22 15:56:21 -0500313 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000314 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000315
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000316 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
317 if (nonblocking)
318 return -XFS_ERROR(EAGAIN);
319 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000320 }
321
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000322 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
323 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000324 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000325
Dave Chinnerd2c28192012-06-08 15:44:53 +1000326 if (offset + count > mp->m_super->s_maxbytes)
327 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000328 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
329 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000330 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
331 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000332 xfs_iunlock(ip, XFS_ILOCK_SHARED);
333
Christoph Hellwiga206c812010-12-10 08:42:20 +0000334 if (error)
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000335 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000336
Alain Renaud0d882a32012-05-22 15:56:21 -0500337 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000338 (!nimaps || isnullstartblock(imap->br_startblock))) {
Jie Liu0799a3e2013-09-29 18:56:04 +0800339 error = xfs_iomap_write_allocate(ip, offset, imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000340 if (!error)
341 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000342 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000343 }
344
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000345#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500346 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000347 ASSERT(nimaps);
348 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
349 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
350 }
351#endif
352 if (nimaps)
353 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
354 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000357STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000358xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000359 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000360 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000361 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000363 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000364
Christoph Hellwig558e6892010-04-28 12:28:58 +0000365 return offset >= imap->br_startoff &&
366 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100369/*
370 * BIO completion handler for buffered IO.
371 */
Al Viro782e3b32007-10-12 07:17:47 +0100372STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100373xfs_end_bio(
374 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100375 int error)
376{
377 xfs_ioend_t *ioend = bio->bi_private;
378
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100379 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000380 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100381
382 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100383 bio->bi_private = NULL;
384 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100385 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000386
Christoph Hellwig209fb872010-07-18 21:17:11 +0000387 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100388}
389
390STATIC void
391xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000392 struct writeback_control *wbc,
393 xfs_ioend_t *ioend,
394 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100395{
396 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100397 bio->bi_private = ioend;
398 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100399 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100400}
401
402STATIC struct bio *
403xfs_alloc_ioend_bio(
404 struct buffer_head *bh)
405{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100406 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000407 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100408
409 ASSERT(bio->bi_private == NULL);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700410 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100411 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100412 return bio;
413}
414
415STATIC void
416xfs_start_buffer_writeback(
417 struct buffer_head *bh)
418{
419 ASSERT(buffer_mapped(bh));
420 ASSERT(buffer_locked(bh));
421 ASSERT(!buffer_delay(bh));
422 ASSERT(!buffer_unwritten(bh));
423
424 mark_buffer_async_write(bh);
425 set_buffer_uptodate(bh);
426 clear_buffer_dirty(bh);
427}
428
429STATIC void
430xfs_start_page_writeback(
431 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100432 int clear_dirty,
433 int buffers)
434{
435 ASSERT(PageLocked(page));
436 ASSERT(!PageWriteback(page));
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100437 if (clear_dirty)
David Chinner92132022006-12-21 10:24:01 +1100438 clear_page_dirty_for_io(page);
439 set_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100440 unlock_page(page);
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700441 /* If no buffers on the page are to be written, finish it here */
442 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100443 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100444}
445
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000446static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100447{
448 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
449}
450
451/*
David Chinnerd88992f2006-01-18 13:38:12 +1100452 * Submit all of the bios for all of the ioends we have saved up, covering the
453 * initial writepage page and also any probed pages.
454 *
455 * Because we may have multiple ioends spanning a page, we need to start
456 * writeback on all the buffers before we submit them for I/O. If we mark the
457 * buffers as we got, then we can end up with a page that only has buffers
458 * marked async write and I/O complete on can occur before we mark the other
459 * buffers async write.
460 *
461 * The end result of this is that we trip a bug in end_page_writeback() because
462 * we call it twice for the one page as the code in end_buffer_async_write()
463 * assumes that all buffers on the page are started at the same time.
464 *
465 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000466 * buffer_heads, and then submit them for I/O on the second pass.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100467 *
468 * If @fail is non-zero, it means that we have a situation where some part of
469 * the submission process has failed after we have marked paged for writeback
470 * and unlocked them. In this situation, we need to fail the ioend chain rather
471 * than submit it to IO. This typically only happens on a filesystem shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100472 */
473STATIC void
474xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000475 struct writeback_control *wbc,
Dave Chinner7bf7f352012-11-12 22:09:45 +1100476 xfs_ioend_t *ioend,
477 int fail)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100478{
David Chinnerd88992f2006-01-18 13:38:12 +1100479 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100480 xfs_ioend_t *next;
481 struct buffer_head *bh;
482 struct bio *bio;
483 sector_t lastblock = 0;
484
David Chinnerd88992f2006-01-18 13:38:12 +1100485 /* Pass 1 - start writeback */
486 do {
487 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000488 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100489 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100490 } while ((ioend = next) != NULL);
491
492 /* Pass 2 - submit I/O */
493 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100494 do {
495 next = ioend->io_list;
496 bio = NULL;
497
Dave Chinner7bf7f352012-11-12 22:09:45 +1100498 /*
499 * If we are failing the IO now, just mark the ioend with an
500 * error and finish it. This will run IO completion immediately
501 * as there is only one reference to the ioend at this point in
502 * time.
503 */
504 if (fail) {
505 ioend->io_error = -fail;
506 xfs_finish_ioend(ioend);
507 continue;
508 }
509
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100510 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100511
512 if (!bio) {
513 retry:
514 bio = xfs_alloc_ioend_bio(bh);
515 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000516 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100517 goto retry;
518 }
519
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000520 if (xfs_bio_add_buffer(bio, bh) != bh->b_size) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000521 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100522 goto retry;
523 }
524
525 lastblock = bh->b_blocknr;
526 }
527 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000528 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000529 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100530 } while ((ioend = next) != NULL);
531}
532
533/*
534 * Cancel submission of all buffer_heads so far in this endio.
535 * Toss the endio too. Only ever called for the initial page
536 * in a writepage request, so only ever one page.
537 */
538STATIC void
539xfs_cancel_ioend(
540 xfs_ioend_t *ioend)
541{
542 xfs_ioend_t *next;
543 struct buffer_head *bh, *next_bh;
544
545 do {
546 next = ioend->io_list;
547 bh = ioend->io_buffer_head;
548 do {
549 next_bh = bh->b_private;
550 clear_buffer_async_write(bh);
551 unlock_buffer(bh);
552 } while ((bh = next_bh) != NULL);
553
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100554 mempool_free(ioend, xfs_ioend_pool);
555 } while ((ioend = next) != NULL);
556}
557
558/*
559 * Test to see if we've been building up a completion structure for
560 * earlier buffers -- if so, we try to append to this ioend if we
561 * can, otherwise we finish off any current ioend and start another.
562 * Return true if we've finished the given ioend.
563 */
564STATIC void
565xfs_add_to_ioend(
566 struct inode *inode,
567 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100568 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100569 unsigned int type,
570 xfs_ioend_t **result,
571 int need_ioend)
572{
573 xfs_ioend_t *ioend = *result;
574
575 if (!ioend || need_ioend || type != ioend->io_type) {
576 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100577
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100578 ioend = xfs_alloc_ioend(inode, type);
579 ioend->io_offset = offset;
580 ioend->io_buffer_head = bh;
581 ioend->io_buffer_tail = bh;
582 if (previous)
583 previous->io_list = ioend;
584 *result = ioend;
585 } else {
586 ioend->io_buffer_tail->b_private = bh;
587 ioend->io_buffer_tail = bh;
588 }
589
590 bh->b_private = NULL;
591 ioend->io_size += bh->b_size;
592}
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100595xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000596 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100597 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000598 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000599 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100600{
601 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000602 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000603 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
604 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100605
Christoph Hellwig207d0412010-04-28 12:28:56 +0000606 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
607 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100608
Christoph Hellwige5131822010-04-28 12:28:55 +0000609 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000610 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100611
Christoph Hellwig046f1682010-04-28 12:28:52 +0000612 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100613
614 bh->b_blocknr = bn;
615 set_buffer_mapped(bh);
616}
617
618STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000620 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000622 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000623 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000625 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
626 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Christoph Hellwig207d0412010-04-28 12:28:56 +0000628 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 set_buffer_mapped(bh);
630 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100631 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
633
634/*
Dave Chinnera49935f2014-03-07 16:19:14 +1100635 * Test if a given page contains at least one buffer of a given @type.
636 * If @check_all_buffers is true, then we walk all the buffers in the page to
637 * try to find one of the type passed in. If it is not set, then the caller only
638 * needs to check the first buffer on the page for a match.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 */
Dave Chinnera49935f2014-03-07 16:19:14 +1100640STATIC bool
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000641xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100642 struct page *page,
Dave Chinnera49935f2014-03-07 16:19:14 +1100643 unsigned int type,
644 bool check_all_buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Dave Chinnera49935f2014-03-07 16:19:14 +1100646 struct buffer_head *bh;
647 struct buffer_head *head;
648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (PageWriteback(page))
Dave Chinnera49935f2014-03-07 16:19:14 +1100650 return false;
651 if (!page->mapping)
652 return false;
653 if (!page_has_buffers(page))
654 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Dave Chinnera49935f2014-03-07 16:19:14 +1100656 bh = head = page_buffers(page);
657 do {
658 if (buffer_unwritten(bh)) {
659 if (type == XFS_IO_UNWRITTEN)
660 return true;
661 } else if (buffer_delay(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100662 if (type == XFS_IO_DELALLOC)
Dave Chinnera49935f2014-03-07 16:19:14 +1100663 return true;
664 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100665 if (type == XFS_IO_OVERWRITE)
Dave Chinnera49935f2014-03-07 16:19:14 +1100666 return true;
667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Dave Chinnera49935f2014-03-07 16:19:14 +1100669 /* If we are only checking the first buffer, we are done now. */
670 if (!check_all_buffers)
671 break;
672 } while ((bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Dave Chinnera49935f2014-03-07 16:19:14 +1100674 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677/*
678 * Allocate & map buffers for page given the extent map. Write it out.
679 * except for the original page of a writepage, this is called on
680 * delalloc/unwritten pages only, for the original page it is possible
681 * that the page has no mapping at all.
682 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100683STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684xfs_convert_page(
685 struct inode *inode,
686 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100687 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000688 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100689 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000690 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100692 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100693 xfs_off_t end_offset;
694 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100695 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700696 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100697 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100698 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100700 if (page->index != tindex)
701 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200702 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100703 goto fail;
704 if (PageWriteback(page))
705 goto fail_unlock_page;
706 if (page->mapping != inode->i_mapping)
707 goto fail_unlock_page;
Dave Chinnera49935f2014-03-07 16:19:14 +1100708 if (!xfs_check_page_type(page, (*ioendp)->io_type, false))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100709 goto fail_unlock_page;
710
Nathan Scott24e17b52005-05-05 13:33:20 -0700711 /*
712 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000713 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100714 *
715 * Derivation:
716 *
717 * End offset is the highest offset that this page should represent.
718 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
719 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
720 * hence give us the correct page_dirty count. On any other page,
721 * it will be zero and in that case we need page_dirty to be the
722 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700723 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100724 end_offset = min_t(unsigned long long,
725 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
726 i_size_read(inode));
727
Dave Chinner480d7462013-05-20 09:51:08 +1000728 /*
729 * If the current map does not span the entire page we are about to try
730 * to write, then give up. The only way we can write a page that spans
731 * multiple mappings in a single writeback iteration is via the
732 * xfs_vm_writepage() function. Data integrity writeback requires the
733 * entire page to be written in a single attempt, otherwise the part of
734 * the page we don't write here doesn't get written as part of the data
735 * integrity sync.
736 *
737 * For normal writeback, we also don't attempt to write partial pages
738 * here as it simply means that write_cache_pages() will see it under
739 * writeback and ignore the page until some point in the future, at
740 * which time this will be the only page in the file that needs
741 * writeback. Hence for more optimal IO patterns, we should always
742 * avoid partial page writeback due to multiple mappings on a page here.
743 */
744 if (!xfs_imap_valid(inode, imap, end_offset))
745 goto fail_unlock_page;
746
Nathan Scott24e17b52005-05-05 13:33:20 -0700747 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100748 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
749 PAGE_CACHE_SIZE);
750 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
751 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700752
Dave Chinnera49935f2014-03-07 16:19:14 +1100753 /*
754 * The moment we find a buffer that doesn't match our current type
755 * specification or can't be written, abort the loop and start
756 * writeback. As per the above xfs_imap_valid() check, only
757 * xfs_vm_writepage() can handle partial page writeback fully - we are
758 * limited here to the buffers that are contiguous with the current
759 * ioend, and hence a buffer we can't write breaks that contiguity and
760 * we have to defer the rest of the IO to xfs_vm_writepage().
761 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 bh = head = page_buffers(page);
763 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100764 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100766 if (!buffer_uptodate(bh))
767 uptodate = 0;
768 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
769 done = 1;
Dave Chinnera49935f2014-03-07 16:19:14 +1100770 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100771 }
772
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000773 if (buffer_unwritten(bh) || buffer_delay(bh) ||
774 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100775 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500776 type = XFS_IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000777 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500778 type = XFS_IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000779 else
Alain Renaud0d882a32012-05-22 15:56:21 -0500780 type = XFS_IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100781
Dave Chinnera49935f2014-03-07 16:19:14 +1100782 /*
783 * imap should always be valid because of the above
784 * partial page end_offset check on the imap.
785 */
786 ASSERT(xfs_imap_valid(inode, imap, offset));
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100787
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000788 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -0500789 if (type != XFS_IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000790 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000791 xfs_add_to_ioend(inode, bh, offset, type,
792 ioendp, done);
793
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100794 page_dirty--;
795 count++;
796 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000797 done = 1;
Dave Chinnera49935f2014-03-07 16:19:14 +1100798 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100800 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100802 if (uptodate && bh == head)
803 SetPageUptodate(page);
804
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000805 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000806 if (--wbc->nr_to_write <= 0 &&
807 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000808 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000810 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100811
812 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100813 fail_unlock_page:
814 unlock_page(page);
815 fail:
816 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
819/*
820 * Convert & write out a cluster of pages in the same extent as defined
821 * by mp and following the start page.
822 */
823STATIC void
824xfs_cluster_write(
825 struct inode *inode,
826 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000827 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100828 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 pgoff_t tlast)
831{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100832 struct pagevec pvec;
833 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100835 pagevec_init(&pvec, 0);
836 while (!done && tindex <= tlast) {
837 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
838
839 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100841
842 for (i = 0; i < pagevec_count(&pvec); i++) {
843 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000844 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100845 if (done)
846 break;
847 }
848
849 pagevec_release(&pvec);
850 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 }
852}
853
Dave Chinner3ed3a432010-03-05 02:00:42 +0000854STATIC void
855xfs_vm_invalidatepage(
856 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400857 unsigned int offset,
858 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000859{
Lukas Czerner34097df2013-05-21 23:58:01 -0400860 trace_xfs_invalidatepage(page->mapping->host, page, offset,
861 length);
862 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000863}
864
865/*
866 * If the page has delalloc buffers on it, we need to punch them out before we
867 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
868 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
869 * is done on that same region - the delalloc extent is returned when none is
870 * supposed to be there.
871 *
872 * We prevent this by truncating away the delalloc regions on the page before
873 * invalidating it. Because they are delalloc, we can do this without needing a
874 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
875 * truncation without a transaction as there is no space left for block
876 * reservation (typically why we see a ENOSPC in writeback).
877 *
878 * This is not a performance critical path, so for now just do the punching a
879 * buffer head at a time.
880 */
881STATIC void
882xfs_aops_discard_page(
883 struct page *page)
884{
885 struct inode *inode = page->mapping->host;
886 struct xfs_inode *ip = XFS_I(inode);
887 struct buffer_head *bh, *head;
888 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000889
Dave Chinnera49935f2014-03-07 16:19:14 +1100890 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000891 goto out_invalidate;
892
Dave Chinnere8c37532010-03-15 02:36:35 +0000893 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
894 goto out_invalidate;
895
Dave Chinner4f107002011-03-07 10:00:35 +1100896 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000897 "page discard on page %p, inode 0x%llx, offset %llu.",
898 page, ip->i_ino, offset);
899
900 xfs_ilock(ip, XFS_ILOCK_EXCL);
901 bh = head = page_buffers(page);
902 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000903 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100904 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000905
906 if (!buffer_delay(bh))
907 goto next_buffer;
908
Dave Chinnerc726de42010-11-30 15:14:39 +1100909 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
910 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000911 if (error) {
912 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000913 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100914 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000915 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000916 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000917 break;
918 }
919next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100920 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000921
922 } while ((bh = bh->b_this_page) != head);
923
924 xfs_iunlock(ip, XFS_ILOCK_EXCL);
925out_invalidate:
Lukas Czernerd47992f2013-05-21 23:17:23 -0400926 xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000927 return;
928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000931 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000933 * For delalloc space on the page we need to allocate space and flush it.
934 * For unwritten space on the page we need to start the conversion to
935 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000936 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938STATIC int
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000939xfs_vm_writepage(
940 struct page *page,
941 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000943 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100944 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000945 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100946 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100948 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000950 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000951 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000952 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000953 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000954 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Lukas Czerner34097df2013-05-21 23:58:01 -0400956 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000957
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000958 ASSERT(page_has_buffers(page));
959
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000960 /*
961 * Refuse to write the page out if we are called from reclaim context.
962 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400963 * This avoids stack overflows when called from deeply used stacks in
964 * random callers for direct reclaim or memcg reclaim. We explicitly
965 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000966 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700967 * This should never happen except in the case of a VM regression so
968 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000969 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700970 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
971 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000972 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000973
974 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200975 * Given that we do not allow direct reclaim to call us, we should
976 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000977 */
Christoph Hellwig448011e2014-06-06 16:05:15 +1000978 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000979 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 /* Is this page beyond the end of the file? */
982 offset = i_size_read(inode);
983 end_index = offset >> PAGE_CACHE_SHIFT;
984 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +1000985
986 /*
987 * The page index is less than the end_index, adjust the end_offset
988 * to the highest offset that this page should represent.
989 * -----------------------------------------------------
990 * | file mapping | <EOF> |
991 * -----------------------------------------------------
992 * | Page ... | Page N-2 | Page N-1 | Page N | |
993 * ^--------------------------------^----------|--------
994 * | desired writeback range | see else |
995 * ---------------------------------^------------------|
996 */
997 if (page->index < end_index)
998 end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT;
999 else {
1000 /*
1001 * Check whether the page to write out is beyond or straddles
1002 * i_size or not.
1003 * -------------------------------------------------------
1004 * | file mapping | <EOF> |
1005 * -------------------------------------------------------
1006 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1007 * ^--------------------------------^-----------|---------
1008 * | | Straddles |
1009 * ---------------------------------^-----------|--------|
1010 */
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001011 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
1012
1013 /*
Jan Karaff9a28f2013-03-14 14:30:54 +01001014 * Skip the page if it is fully outside i_size, e.g. due to a
1015 * truncate operation that is in progress. We must redirty the
1016 * page so that reclaim stops reclaiming it. Otherwise
1017 * xfs_vm_releasepage() is called on it and gets confused.
Jie Liu8695d272014-05-20 08:24:26 +10001018 *
1019 * Note that the end_index is unsigned long, it would overflow
1020 * if the given offset is greater than 16TB on 32-bit system
1021 * and if we do check the page is fully outside i_size or not
1022 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1023 * will be evaluated to 0. Hence this page will be redirtied
1024 * and be written out repeatedly which would result in an
1025 * infinite loop, the user program that perform this operation
1026 * will hang. Instead, we can verify this situation by checking
1027 * if the page to write is totally beyond the i_size or if it's
1028 * offset is just equal to the EOF.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001029 */
Jie Liu8695d272014-05-20 08:24:26 +10001030 if (page->index > end_index ||
1031 (page->index == end_index && offset_into_page == 0))
Jan Karaff9a28f2013-03-14 14:30:54 +01001032 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001033
1034 /*
1035 * The page straddles i_size. It must be zeroed out on each
1036 * and every writepage invocation because it may be mmapped.
1037 * "A file is mapped in multiples of the page size. For a file
Jie Liu8695d272014-05-20 08:24:26 +10001038 * that is not a multiple of the page size, the remaining
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001039 * memory is zeroed when mapped, and writes to that region are
1040 * not written out to the file."
1041 */
1042 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
Jie Liu8695d272014-05-20 08:24:26 +10001043
1044 /* Adjust the end_offset to the end of file */
1045 end_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
1047
Nathan Scott24e17b52005-05-05 13:33:20 -07001048 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -07001049
Nathan Scott24e17b52005-05-05 13:33:20 -07001050 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001051 offset = page_offset(page);
Alain Renaud0d882a32012-05-22 15:56:21 -05001052 type = XFS_IO_OVERWRITE;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001053
Christoph Hellwigdbcdde32011-07-08 14:34:14 +02001054 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001055 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +00001058 int new_ioend = 0;
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (offset >= end_offset)
1061 break;
1062 if (!buffer_uptodate(bh))
1063 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001065 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +00001066 * set_page_dirty dirties all buffers in a page, independent
1067 * of their state. The dirty state however is entirely
1068 * meaningless for holes (!mapped && uptodate), so skip
1069 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001070 */
1071 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001072 imap_valid = 0;
1073 continue;
1074 }
1075
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001076 if (buffer_unwritten(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001077 if (type != XFS_IO_UNWRITTEN) {
1078 type = XFS_IO_UNWRITTEN;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001079 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001080 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001081 } else if (buffer_delay(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001082 if (type != XFS_IO_DELALLOC) {
1083 type = XFS_IO_DELALLOC;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001084 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001086 } else if (buffer_uptodate(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001087 if (type != XFS_IO_OVERWRITE) {
1088 type = XFS_IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +00001089 imap_valid = 0;
1090 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001091 } else {
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001092 if (PageUptodate(page))
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001093 ASSERT(buffer_mapped(bh));
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001094 /*
1095 * This buffer is not uptodate and will not be
1096 * written to disk. Ensure that we will put any
1097 * subsequent writeable buffers into a new
1098 * ioend.
1099 */
1100 imap_valid = 0;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001101 continue;
1102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001104 if (imap_valid)
1105 imap_valid = xfs_imap_valid(inode, &imap, offset);
1106 if (!imap_valid) {
1107 /*
1108 * If we didn't have a valid mapping then we need to
1109 * put the new mapping into a separate ioend structure.
1110 * This ensures non-contiguous extents always have
1111 * separate ioends, which is particularly important
1112 * for unwritten extent conversion at I/O completion
1113 * time.
1114 */
1115 new_ioend = 1;
1116 err = xfs_map_blocks(inode, offset, &imap, type,
1117 nonblocking);
1118 if (err)
1119 goto error;
1120 imap_valid = xfs_imap_valid(inode, &imap, offset);
1121 }
1122 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001123 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -05001124 if (type != XFS_IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001125 xfs_map_at_offset(inode, bh, &imap, offset);
1126 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1127 new_ioend);
1128 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001130
1131 if (!iohead)
1132 iohead = ioend;
1133
1134 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 if (uptodate && bh == head)
1137 SetPageUptodate(page);
1138
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001139 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Dave Chinner7bf7f352012-11-12 22:09:45 +11001141 /* if there is no IO to be submitted for this page, we are done */
1142 if (!ioend)
1143 return 0;
1144
1145 ASSERT(iohead);
1146
1147 /*
1148 * Any errors from this point onwards need tobe reported through the IO
1149 * completion path as we have marked the initial page as under writeback
1150 * and unlocked it.
1151 */
1152 if (imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001153 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001154
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001155 end_index = imap.br_startoff + imap.br_blockcount;
1156
1157 /* to bytes */
1158 end_index <<= inode->i_blkbits;
1159
1160 /* to pages */
1161 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1162
1163 /* check against file size */
1164 if (end_index > last_index)
1165 end_index = last_index;
1166
Christoph Hellwig207d0412010-04-28 12:28:56 +00001167 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001168 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
1170
Christoph Hellwig281627d2012-03-13 08:41:05 +00001171
Dave Chinner7bf7f352012-11-12 22:09:45 +11001172 /*
1173 * Reserve log space if we might write beyond the on-disk inode size.
1174 */
1175 err = 0;
1176 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1177 err = xfs_setfilesize_trans_alloc(ioend);
1178
1179 xfs_submit_ioend(wbc, iohead, err);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001180
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001181 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001184 if (iohead)
1185 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001187 if (err == -EAGAIN)
1188 goto redirty;
1189
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001190 xfs_aops_discard_page(page);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001191 ClearPageUptodate(page);
1192 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001194
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001195redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001196 redirty_page_for_writepage(wbc, page);
1197 unlock_page(page);
1198 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001199}
1200
Nathan Scott7d4fb402006-06-09 15:27:16 +10001201STATIC int
1202xfs_vm_writepages(
1203 struct address_space *mapping,
1204 struct writeback_control *wbc)
1205{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001206 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001207 return generic_writepages(mapping, wbc);
1208}
1209
Nathan Scottf51623b2006-03-14 13:26:27 +11001210/*
1211 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001212 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001213 * have buffer heads in this call.
1214 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001215 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001216 */
1217STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001218xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001219 struct page *page,
1220 gfp_t gfp_mask)
1221{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001222 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001223
Lukas Czerner34097df2013-05-21 23:58:01 -04001224 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001225
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001226 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001227
Christoph Hellwig448011e2014-06-06 16:05:15 +10001228 if (WARN_ON_ONCE(delalloc))
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001229 return 0;
Christoph Hellwig448011e2014-06-06 16:05:15 +10001230 if (WARN_ON_ONCE(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001231 return 0;
1232
Nathan Scottf51623b2006-03-14 13:26:27 +11001233 return try_to_free_buffers(page);
1234}
1235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001237__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 struct inode *inode,
1239 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 struct buffer_head *bh_result,
1241 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001242 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001244 struct xfs_inode *ip = XFS_I(inode);
1245 struct xfs_mount *mp = ip->i_mount;
1246 xfs_fileoff_t offset_fsb, end_fsb;
1247 int error = 0;
1248 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001249 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001250 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001251 xfs_off_t offset;
1252 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001253 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001254
1255 if (XFS_FORCED_SHUTDOWN(mp))
1256 return -XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001258 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001259 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1260 size = bh_result->b_size;
Lachlan McIlroy364f358a2008-09-17 16:50:14 +10001261
1262 if (!create && direct && offset >= i_size_read(inode))
1263 return 0;
1264
Dave Chinner507630b2012-03-27 10:34:50 -04001265 /*
1266 * Direct I/O is usually done on preallocated files, so try getting
1267 * a block mapping without an exclusive lock first. For buffered
1268 * writes we already have the exclusive iolock anyway, so avoiding
1269 * a lock roundtrip here by taking the ilock exclusive from the
1270 * beginning is a useful micro optimization.
1271 */
1272 if (create && !direct) {
Christoph Hellwiga206c812010-12-10 08:42:20 +00001273 lockmode = XFS_ILOCK_EXCL;
1274 xfs_ilock(ip, lockmode);
1275 } else {
Christoph Hellwig309ecac82013-12-06 12:30:09 -08001276 lockmode = xfs_ilock_data_map_shared(ip);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001277 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001278
Dave Chinnerd2c28192012-06-08 15:44:53 +10001279 ASSERT(offset <= mp->m_super->s_maxbytes);
1280 if (offset + size > mp->m_super->s_maxbytes)
1281 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001282 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1283 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1284
Dave Chinner5c8ed202011-09-18 20:40:45 +00001285 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1286 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001288 goto out_unlock;
1289
1290 if (create &&
1291 (!nimaps ||
1292 (imap.br_startblock == HOLESTARTBLOCK ||
1293 imap.br_startblock == DELAYSTARTBLOCK))) {
Dave Chinneraff3a9e2012-04-23 15:58:44 +10001294 if (direct || xfs_get_extsz_hint(ip)) {
Dave Chinner507630b2012-03-27 10:34:50 -04001295 /*
1296 * Drop the ilock in preparation for starting the block
1297 * allocation transaction. It will be retaken
1298 * exclusively inside xfs_iomap_write_direct for the
1299 * actual allocation.
1300 */
1301 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001302 error = xfs_iomap_write_direct(ip, offset, size,
1303 &imap, nimaps);
Dave Chinner507630b2012-03-27 10:34:50 -04001304 if (error)
1305 return -error;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001306 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001307 } else {
Dave Chinner507630b2012-03-27 10:34:50 -04001308 /*
1309 * Delalloc reservations do not require a transaction,
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001310 * we can go on without dropping the lock here. If we
1311 * are allocating a new delalloc block, make sure that
1312 * we set the new flag so that we mark the buffer new so
1313 * that we know that it is newly allocated if the write
1314 * fails.
Dave Chinner507630b2012-03-27 10:34:50 -04001315 */
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001316 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1317 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001318 error = xfs_iomap_write_delay(ip, offset, size, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001319 if (error)
1320 goto out_unlock;
1321
1322 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001323 }
Christoph Hellwiga206c812010-12-10 08:42:20 +00001324
1325 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1326 } else if (nimaps) {
1327 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001328 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001329 } else {
1330 trace_xfs_get_blocks_notfound(ip, offset, size);
1331 goto out_unlock;
1332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Christoph Hellwig207d0412010-04-28 12:28:56 +00001334 if (imap.br_startblock != HOLESTARTBLOCK &&
1335 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001336 /*
1337 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 * the read case (treat as if we're reading into a hole).
1339 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001340 if (create || !ISUNWRITTEN(&imap))
1341 xfs_map_buffer(inode, bh_result, &imap, offset);
1342 if (create && ISUNWRITTEN(&imap)) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001343 if (direct) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 bh_result->b_private = inode;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001345 set_buffer_defer_completion(bh_result);
1346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
1349 }
1350
Nathan Scottc2536662006-03-29 10:44:40 +10001351 /*
1352 * If this is a realtime file, data may be on a different device.
1353 * to that pointed to from the buffer_head b_bdev currently.
1354 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001355 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Nathan Scottc2536662006-03-29 10:44:40 +10001357 /*
David Chinner549054a2007-02-10 18:36:35 +11001358 * If we previously allocated a block out beyond eof and we are now
1359 * coming back to use it then we will need to flag it as new even if it
1360 * has a disk address.
1361 *
1362 * With sub-block writes into unwritten extents we also need to mark
1363 * the buffer as new so that the unwritten parts of the buffer gets
1364 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 */
1366 if (create &&
1367 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001368 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001369 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Christoph Hellwig207d0412010-04-28 12:28:56 +00001372 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 BUG_ON(direct);
1374 if (create) {
1375 set_buffer_uptodate(bh_result);
1376 set_buffer_mapped(bh_result);
1377 set_buffer_delay(bh_result);
1378 }
1379 }
1380
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001381 /*
1382 * If this is O_DIRECT or the mpage code calling tell them how large
1383 * the mapping is, so that we can avoid repeated get_blocks calls.
Dave Chinner0e1f7892014-04-17 08:15:19 +10001384 *
1385 * If the mapping spans EOF, then we have to break the mapping up as the
1386 * mapping for blocks beyond EOF must be marked new so that sub block
1387 * regions can be correctly zeroed. We can't do this for mappings within
1388 * EOF unless the mapping was just allocated or is unwritten, otherwise
1389 * the callers would overwrite existing data with zeros. Hence we have
1390 * to split the mapping into a range up to and including EOF, and a
1391 * second mapping for beyond EOF.
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001392 */
Nathan Scottc2536662006-03-29 10:44:40 +10001393 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001394 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001395
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001396 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1397 mapping_size <<= inode->i_blkbits;
1398
1399 ASSERT(mapping_size > 0);
1400 if (mapping_size > size)
1401 mapping_size = size;
Dave Chinner0e1f7892014-04-17 08:15:19 +10001402 if (offset < i_size_read(inode) &&
1403 offset + mapping_size >= i_size_read(inode)) {
1404 /* limit mapping to block that spans EOF */
1405 mapping_size = roundup_64(i_size_read(inode) - offset,
1406 1 << inode->i_blkbits);
1407 }
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001408 if (mapping_size > LONG_MAX)
1409 mapping_size = LONG_MAX;
1410
1411 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 }
1413
1414 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001415
1416out_unlock:
1417 xfs_iunlock(ip, lockmode);
1418 return -error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419}
1420
1421int
Nathan Scottc2536662006-03-29 10:44:40 +10001422xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 struct inode *inode,
1424 sector_t iblock,
1425 struct buffer_head *bh_result,
1426 int create)
1427{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001428 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429}
1430
1431STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001432xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 struct inode *inode,
1434 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 struct buffer_head *bh_result,
1436 int create)
1437{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001438 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439}
1440
Christoph Hellwig209fb872010-07-18 21:17:11 +00001441/*
1442 * Complete a direct I/O write request.
1443 *
1444 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1445 * need to issue a transaction to convert the range from unwritten to written
1446 * extents. In case this is regular synchronous I/O we just call xfs_end_io
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001447 * to do this and we are done. But in case this was a successful AIO
Christoph Hellwig209fb872010-07-18 21:17:11 +00001448 * request this handler is called from interrupt context, from which we
1449 * can't start transactions. In that case offload the I/O completion to
1450 * the workqueues we also use for buffered I/O completion.
1451 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001452STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001453xfs_end_io_direct_write(
1454 struct kiocb *iocb,
1455 loff_t offset,
1456 ssize_t size,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001457 void *private)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001458{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001459 struct xfs_ioend *ioend = iocb->private;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001460
1461 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001462 * While the generic direct I/O code updates the inode size, it does
1463 * so only after the end_io handler is called, which means our
1464 * end_io handler thinks the on-disk size is outside the in-core
1465 * size. To prevent this just update it a little bit earlier here.
1466 */
1467 if (offset + size > i_size_read(ioend->io_inode))
1468 i_size_write(ioend->io_inode, offset + size);
1469
1470 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001471 * blockdev_direct_IO can return an error even after the I/O
Christoph Hellwigf0973862005-09-05 08:22:52 +10001472 * completion handler was called. Thus we need to protect
1473 * against double-freeing.
1474 */
1475 iocb->private = NULL;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001476
Christoph Hellwig209fb872010-07-18 21:17:11 +00001477 ioend->io_offset = offset;
1478 ioend->io_size = size;
1479 if (private && size > 0)
Alain Renaud0d882a32012-05-22 15:56:21 -05001480 ioend->io_type = XFS_IO_UNWRITTEN;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001481
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001482 xfs_finish_ioend_sync(ioend);
Christoph Hellwigf0973862005-09-05 08:22:52 +10001483}
1484
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001486xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 int rw,
1488 struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05001489 struct iov_iter *iter,
1490 loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001492 struct inode *inode = iocb->ki_filp->f_mapping->host;
1493 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001494 struct xfs_ioend *ioend = NULL;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001495 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496
Christoph Hellwig209fb872010-07-18 21:17:11 +00001497 if (rw & WRITE) {
Al Viroa6cbcd42014-03-04 22:38:00 -05001498 size_t size = iov_iter_count(iter);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001499
1500 /*
Dave Chinner437a2552012-11-28 13:01:00 +11001501 * We cannot preallocate a size update transaction here as we
1502 * don't know whether allocation is necessary or not. Hence we
1503 * can only tell IO completion that one is necessary if we are
1504 * not doing unwritten extent conversion.
Christoph Hellwig281627d2012-03-13 08:41:05 +00001505 */
Alain Renaud0d882a32012-05-22 15:56:21 -05001506 iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
Dave Chinner437a2552012-11-28 13:01:00 +11001507 if (offset + size > XFS_I(inode)->i_d.di_size)
Christoph Hellwig281627d2012-03-13 08:41:05 +00001508 ioend->io_isdirect = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
Al Viro31b14032014-03-05 01:33:16 -05001510 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
1511 offset, xfs_get_blocks_direct,
Christoph Hellwig9862f622014-02-10 10:28:04 +11001512 xfs_end_io_direct_write, NULL,
1513 DIO_ASYNC_EXTEND);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001514 if (ret != -EIOCBQUEUED && iocb->private)
Dave Chinner437a2552012-11-28 13:01:00 +11001515 goto out_destroy_ioend;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001516 } else {
Al Viro31b14032014-03-05 01:33:16 -05001517 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
1518 offset, xfs_get_blocks_direct,
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001519 NULL, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001520 }
Christoph Hellwig5fe878ae2009-12-15 16:47:50 -08001521
Christoph Hellwigf0973862005-09-05 08:22:52 +10001522 return ret;
Christoph Hellwig281627d2012-03-13 08:41:05 +00001523
Christoph Hellwig281627d2012-03-13 08:41:05 +00001524out_destroy_ioend:
1525 xfs_destroy_ioend(ioend);
1526 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
1528
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001529/*
1530 * Punch out the delalloc blocks we have already allocated.
1531 *
1532 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1533 * as the page is still locked at this point.
1534 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001535STATIC void
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001536xfs_vm_kill_delalloc_range(
1537 struct inode *inode,
1538 loff_t start,
1539 loff_t end)
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001540{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001541 struct xfs_inode *ip = XFS_I(inode);
1542 xfs_fileoff_t start_fsb;
1543 xfs_fileoff_t end_fsb;
1544 int error;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001545
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001546 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1547 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1548 if (end_fsb <= start_fsb)
1549 return;
Dave Chinnerc726de42010-11-30 15:14:39 +11001550
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001551 xfs_ilock(ip, XFS_ILOCK_EXCL);
1552 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1553 end_fsb - start_fsb);
1554 if (error) {
1555 /* something screwed, just bail */
1556 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1557 xfs_alert(ip->i_mount,
1558 "xfs_vm_write_failed: unable to clean up ino %lld",
1559 ip->i_ino);
Dave Chinnerc726de42010-11-30 15:14:39 +11001560 }
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001561 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001562 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001563}
1564
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001565STATIC void
1566xfs_vm_write_failed(
1567 struct inode *inode,
1568 struct page *page,
1569 loff_t pos,
1570 unsigned len)
1571{
Jie Liu58e59852013-07-16 13:11:16 +08001572 loff_t block_offset;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001573 loff_t block_start;
1574 loff_t block_end;
1575 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1576 loff_t to = from + len;
1577 struct buffer_head *bh, *head;
1578
Jie Liu58e59852013-07-16 13:11:16 +08001579 /*
1580 * The request pos offset might be 32 or 64 bit, this is all fine
1581 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1582 * platform, the high 32-bit will be masked off if we evaluate the
1583 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1584 * 0xfffff000 as an unsigned long, hence the result is incorrect
1585 * which could cause the following ASSERT failed in most cases.
1586 * In order to avoid this, we can evaluate the block_offset of the
1587 * start of the page by using shifts rather than masks the mismatch
1588 * problem.
1589 */
1590 block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT;
1591
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001592 ASSERT(block_offset + from == pos);
1593
1594 head = page_buffers(page);
1595 block_start = 0;
1596 for (bh = head; bh != head || !block_start;
1597 bh = bh->b_this_page, block_start = block_end,
1598 block_offset += bh->b_size) {
1599 block_end = block_start + bh->b_size;
1600
1601 /* skip buffers before the write */
1602 if (block_end <= from)
1603 continue;
1604
1605 /* if the buffer is after the write, we're done */
1606 if (block_start >= to)
1607 break;
1608
1609 if (!buffer_delay(bh))
1610 continue;
1611
1612 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1613 continue;
1614
1615 xfs_vm_kill_delalloc_range(inode, block_offset,
1616 block_offset + bh->b_size);
Dave Chinner4ab9ed52014-04-14 18:11:58 +10001617
1618 /*
1619 * This buffer does not contain data anymore. make sure anyone
1620 * who finds it knows that for certain.
1621 */
1622 clear_buffer_delay(bh);
1623 clear_buffer_uptodate(bh);
1624 clear_buffer_mapped(bh);
1625 clear_buffer_new(bh);
1626 clear_buffer_dirty(bh);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001627 }
1628
1629}
1630
1631/*
1632 * This used to call block_write_begin(), but it unlocks and releases the page
1633 * on error, and we need that page to be able to punch stale delalloc blocks out
1634 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1635 * the appropriate point.
1636 */
Nathan Scottf51623b2006-03-14 13:26:27 +11001637STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001638xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001639 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001640 struct address_space *mapping,
1641 loff_t pos,
1642 unsigned len,
1643 unsigned flags,
1644 struct page **pagep,
1645 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001646{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001647 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1648 struct page *page;
1649 int status;
Christoph Hellwig155130a2010-06-04 11:29:58 +02001650
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001651 ASSERT(len <= PAGE_CACHE_SIZE);
1652
Dave Chinnerad22c7a2013-10-29 22:11:57 +11001653 page = grab_cache_page_write_begin(mapping, index, flags);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001654 if (!page)
1655 return -ENOMEM;
1656
1657 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1658 if (unlikely(status)) {
1659 struct inode *inode = mapping->host;
Dave Chinner72ab70a2014-04-14 18:13:29 +10001660 size_t isize = i_size_read(inode);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001661
1662 xfs_vm_write_failed(inode, page, pos, len);
1663 unlock_page(page);
1664
Dave Chinner72ab70a2014-04-14 18:13:29 +10001665 /*
1666 * If the write is beyond EOF, we only want to kill blocks
1667 * allocated in this write, not blocks that were previously
1668 * written successfully.
1669 */
1670 if (pos + len > isize) {
1671 ssize_t start = max_t(ssize_t, pos, isize);
1672
1673 truncate_pagecache_range(inode, start, pos + len);
1674 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001675
1676 page_cache_release(page);
1677 page = NULL;
1678 }
1679
1680 *pagep = page;
1681 return status;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001682}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001683
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001684/*
Dave Chinneraad3f372014-04-14 18:14:11 +10001685 * On failure, we only need to kill delalloc blocks beyond EOF in the range of
1686 * this specific write because they will never be written. Previous writes
1687 * beyond EOF where block allocation succeeded do not need to be trashed, so
1688 * only new blocks from this write should be trashed. For blocks within
1689 * EOF, generic_write_end() zeros them so they are safe to leave alone and be
1690 * written with all the other valid data.
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001691 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001692STATIC int
1693xfs_vm_write_end(
1694 struct file *file,
1695 struct address_space *mapping,
1696 loff_t pos,
1697 unsigned len,
1698 unsigned copied,
1699 struct page *page,
1700 void *fsdata)
1701{
1702 int ret;
1703
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001704 ASSERT(len <= PAGE_CACHE_SIZE);
1705
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001706 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001707 if (unlikely(ret < len)) {
1708 struct inode *inode = mapping->host;
1709 size_t isize = i_size_read(inode);
1710 loff_t to = pos + len;
1711
1712 if (to > isize) {
Dave Chinneraad3f372014-04-14 18:14:11 +10001713 /* only kill blocks in this write beyond EOF */
1714 if (pos > isize)
1715 isize = pos;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001716 xfs_vm_kill_delalloc_range(inode, isize, to);
Dave Chinneraad3f372014-04-14 18:14:11 +10001717 truncate_pagecache_range(inode, isize, to);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001718 }
1719 }
Christoph Hellwig155130a2010-06-04 11:29:58 +02001720 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001721}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
1723STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001724xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 struct address_space *mapping,
1726 sector_t block)
1727{
1728 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001729 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001731 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001732 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001733 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001734 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001735 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736}
1737
1738STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001739xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 struct file *unused,
1741 struct page *page)
1742{
Nathan Scottc2536662006-03-29 10:44:40 +10001743 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744}
1745
1746STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001747xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 struct file *unused,
1749 struct address_space *mapping,
1750 struct list_head *pages,
1751 unsigned nr_pages)
1752{
Nathan Scottc2536662006-03-29 10:44:40 +10001753 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754}
1755
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001756const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001757 .readpage = xfs_vm_readpage,
1758 .readpages = xfs_vm_readpages,
1759 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001760 .writepages = xfs_vm_writepages,
Nathan Scott238f4c52006-03-17 17:26:25 +11001761 .releasepage = xfs_vm_releasepage,
1762 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001763 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001764 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001765 .bmap = xfs_vm_bmap,
1766 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001767 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001768 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001769 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770};