blob: 4f8cdc59bc38154b45f1adfd69c0a371df4394e3 [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_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110025#include "xfs_trans.h"
Christoph Hellwig281627d2012-03-13 08:41:05 +000026#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_iomap.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000030#include "xfs_trace.h"
Dave Chinner3ed3a432010-03-05 02:00:42 +000031#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100032#include "xfs_bmap_util.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110033#include "xfs_bmap_btree.h"
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110036#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/writeback.h>
38
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000039void
Nathan Scottf51623b2006-03-14 13:26:27 +110040xfs_count_page_state(
41 struct page *page,
42 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110043 int *unwritten)
44{
45 struct buffer_head *bh, *head;
46
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100047 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110048
49 bh = head = page_buffers(page);
50 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100051 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110052 (*unwritten) = 1;
53 else if (buffer_delay(bh))
54 (*delalloc) = 1;
55 } while ((bh = bh->b_this_page) != head);
56}
57
Christoph Hellwig6214ed42007-09-14 15:23:17 +100058STATIC struct block_device *
59xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000060 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100061{
Christoph Hellwig046f1682010-04-28 12:28:52 +000062 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100063 struct xfs_mount *mp = ip->i_mount;
64
Eric Sandeen71ddabb2007-11-23 16:29:42 +110065 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100066 return mp->m_rtdev_targp->bt_bdev;
67 else
68 return mp->m_ddev_targp->bt_bdev;
69}
70
Christoph Hellwig0829c362005-09-02 16:58:49 +100071/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110072 * We're now finished for good with this ioend structure.
73 * Update the page state via the associated buffer_heads,
74 * release holds on the inode and bio, and finally free
75 * up memory. Do not use the ioend after this.
76 */
Christoph Hellwig0829c362005-09-02 16:58:49 +100077STATIC void
78xfs_destroy_ioend(
79 xfs_ioend_t *ioend)
80{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110081 struct buffer_head *bh, *next;
82
83 for (bh = ioend->io_buffer_head; bh; bh = next) {
84 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +100085 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110086 }
Christoph Hellwig583fa582008-12-03 12:20:38 +010087
Christoph Hellwig0829c362005-09-02 16:58:49 +100088 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089}
90
91/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +000092 * Fast and loose check if this write could update the on-disk inode size.
93 */
94static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
95{
96 return ioend->io_offset + ioend->io_size >
97 XFS_I(ioend->io_inode)->i_d.di_size;
98}
99
Christoph Hellwig281627d2012-03-13 08:41:05 +0000100STATIC int
101xfs_setfilesize_trans_alloc(
102 struct xfs_ioend *ioend)
103{
104 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
105 struct xfs_trans *tp;
106 int error;
107
108 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
109
Jie Liu3d3c8b52013-08-12 20:49:59 +1000110 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000111 if (error) {
112 xfs_trans_cancel(tp, 0);
113 return error;
114 }
115
116 ioend->io_append_trans = tp;
117
118 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100119 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200120 * we released it.
121 */
122 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
123 1, _THIS_IP_);
124 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000125 * We hand off the transaction to the completion thread now, so
126 * clear the flag here.
127 */
128 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
129 return 0;
130}
131
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000132/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000133 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000134 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000135STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000136xfs_setfilesize(
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100137 struct xfs_inode *ip,
138 struct xfs_trans *tp,
139 xfs_off_t offset,
140 size_t size)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000141{
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000142 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000143
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000144 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100145 isize = xfs_new_eof(ip, offset + size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000146 if (!isize) {
147 xfs_iunlock(ip, XFS_ILOCK_EXCL);
148 xfs_trans_cancel(tp, 0);
149 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000150 }
151
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100152 trace_xfs_setfilesize(ip, offset, size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000153
154 ip->i_d.di_size = isize;
155 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
156 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
157
158 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000159}
160
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100161STATIC int
162xfs_setfilesize_ioend(
163 struct xfs_ioend *ioend)
164{
165 struct xfs_inode *ip = XFS_I(ioend->io_inode);
166 struct xfs_trans *tp = ioend->io_append_trans;
167
168 /*
169 * The transaction may have been allocated in the I/O submission thread,
170 * thus we need to mark ourselves as being in a transaction manually.
171 * Similarly for freeze protection.
172 */
173 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
174 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
175 0, 1, _THIS_IP_);
176
177 return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
178}
179
Christoph Hellwig0829c362005-09-02 16:58:49 +1000180/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000181 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000182 *
183 * If there is no work to do we might as well call it a day and free the
184 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200185 */
186STATIC void
187xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000188 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200189{
190 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000191 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
192
Alain Renaud0d882a32012-05-22 15:56:21 -0500193 if (ioend->io_type == XFS_IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000194 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100195 else if (ioend->io_append_trans)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000196 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000197 else
198 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200199 }
200}
201
202/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000203 * IO write completion.
204 */
205STATIC void
206xfs_end_io(
207 struct work_struct *work)
208{
209 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
210 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000211 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000212
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000213 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000214 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000215 goto done;
216 }
217 if (ioend->io_error)
218 goto done;
219
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000220 /*
221 * For unwritten extents we need to issue transactions to convert a
222 * range to normal written extens after the data I/O has finished.
223 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500224 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000225 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100226 ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000227 } else if (ioend->io_append_trans) {
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100228 error = xfs_setfilesize_ioend(ioend);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000229 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000230 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000231 }
232
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000233done:
Dave Chinner437a2552012-11-28 13:01:00 +1100234 if (error)
Dave Chinner24513372014-06-25 14:58:08 +1000235 ioend->io_error = error;
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000236 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000237}
238
239/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000240 * Allocate and initialise an IO completion structure.
241 * We need to track unwritten extent write completion here initially.
242 * We'll need to extend this for updating the ondisk inode size later
243 * (vs. incore size).
244 */
245STATIC xfs_ioend_t *
246xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100247 struct inode *inode,
248 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000249{
250 xfs_ioend_t *ioend;
251
252 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
253
254 /*
255 * Set the count to 1 initially, which will prevent an I/O
256 * completion callback from happening before we have started
257 * all the I/O from calling the completion routine too early.
258 */
259 atomic_set(&ioend->io_remaining, 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000260 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100261 ioend->io_list = NULL;
262 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000263 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000264 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100265 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000266 ioend->io_offset = 0;
267 ioend->io_size = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000268 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000269
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000270 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000271 return ioend;
272}
273
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274STATIC int
275xfs_map_blocks(
276 struct inode *inode,
277 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000278 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000279 int type,
280 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000282 struct xfs_inode *ip = XFS_I(inode);
283 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000284 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000285 xfs_fileoff_t offset_fsb, end_fsb;
286 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000287 int bmapi_flags = XFS_BMAPI_ENTIRE;
288 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289
Christoph Hellwiga206c812010-12-10 08:42:20 +0000290 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000291 return -EIO;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000292
Alain Renaud0d882a32012-05-22 15:56:21 -0500293 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000294 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000295
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000296 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
297 if (nonblocking)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000298 return -EAGAIN;
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000299 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000300 }
301
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000302 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
303 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000304 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000305
Dave Chinnerd2c28192012-06-08 15:44:53 +1000306 if (offset + count > mp->m_super->s_maxbytes)
307 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000308 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
309 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000310 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
311 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000312 xfs_iunlock(ip, XFS_ILOCK_SHARED);
313
Christoph Hellwiga206c812010-12-10 08:42:20 +0000314 if (error)
Dave Chinner24513372014-06-25 14:58:08 +1000315 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000316
Alain Renaud0d882a32012-05-22 15:56:21 -0500317 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000318 (!nimaps || isnullstartblock(imap->br_startblock))) {
Jie Liu0799a3e2013-09-29 18:56:04 +0800319 error = xfs_iomap_write_allocate(ip, offset, imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000320 if (!error)
321 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Dave Chinner24513372014-06-25 14:58:08 +1000322 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000323 }
324
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000325#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500326 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000327 ASSERT(nimaps);
328 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
329 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
330 }
331#endif
332 if (nimaps)
333 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
334 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335}
336
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000337STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000338xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000339 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000340 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000341 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000343 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000344
Christoph Hellwig558e6892010-04-28 12:28:58 +0000345 return offset >= imap->br_startoff &&
346 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100349/*
350 * BIO completion handler for buffered IO.
351 */
Al Viro782e3b32007-10-12 07:17:47 +0100352STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100353xfs_end_bio(
354 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100355 int error)
356{
357 xfs_ioend_t *ioend = bio->bi_private;
358
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100359 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000360 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100361
362 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100363 bio->bi_private = NULL;
364 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100365 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000366
Christoph Hellwig209fb872010-07-18 21:17:11 +0000367 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100368}
369
370STATIC void
371xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000372 struct writeback_control *wbc,
373 xfs_ioend_t *ioend,
374 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100375{
376 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100377 bio->bi_private = ioend;
378 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100379 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100380}
381
382STATIC struct bio *
383xfs_alloc_ioend_bio(
384 struct buffer_head *bh)
385{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100386 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000387 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100388
389 ASSERT(bio->bi_private == NULL);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700390 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100391 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100392 return bio;
393}
394
395STATIC void
396xfs_start_buffer_writeback(
397 struct buffer_head *bh)
398{
399 ASSERT(buffer_mapped(bh));
400 ASSERT(buffer_locked(bh));
401 ASSERT(!buffer_delay(bh));
402 ASSERT(!buffer_unwritten(bh));
403
404 mark_buffer_async_write(bh);
405 set_buffer_uptodate(bh);
406 clear_buffer_dirty(bh);
407}
408
409STATIC void
410xfs_start_page_writeback(
411 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100412 int clear_dirty,
413 int buffers)
414{
415 ASSERT(PageLocked(page));
416 ASSERT(!PageWriteback(page));
Dave Chinner0d085a52014-09-23 15:36:27 +1000417
418 /*
419 * if the page was not fully cleaned, we need to ensure that the higher
420 * layers come back to it correctly. That means we need to keep the page
421 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
422 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
423 * write this page in this writeback sweep will be made.
424 */
425 if (clear_dirty) {
David Chinner92132022006-12-21 10:24:01 +1100426 clear_page_dirty_for_io(page);
Dave Chinner0d085a52014-09-23 15:36:27 +1000427 set_page_writeback(page);
428 } else
429 set_page_writeback_keepwrite(page);
430
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100431 unlock_page(page);
Dave Chinner0d085a52014-09-23 15:36:27 +1000432
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700433 /* If no buffers on the page are to be written, finish it here */
434 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100435 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100436}
437
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000438static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100439{
440 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
441}
442
443/*
David Chinnerd88992f2006-01-18 13:38:12 +1100444 * Submit all of the bios for all of the ioends we have saved up, covering the
445 * initial writepage page and also any probed pages.
446 *
447 * Because we may have multiple ioends spanning a page, we need to start
448 * writeback on all the buffers before we submit them for I/O. If we mark the
449 * buffers as we got, then we can end up with a page that only has buffers
450 * marked async write and I/O complete on can occur before we mark the other
451 * buffers async write.
452 *
453 * The end result of this is that we trip a bug in end_page_writeback() because
454 * we call it twice for the one page as the code in end_buffer_async_write()
455 * assumes that all buffers on the page are started at the same time.
456 *
457 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000458 * buffer_heads, and then submit them for I/O on the second pass.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100459 *
460 * If @fail is non-zero, it means that we have a situation where some part of
461 * the submission process has failed after we have marked paged for writeback
462 * and unlocked them. In this situation, we need to fail the ioend chain rather
463 * than submit it to IO. This typically only happens on a filesystem shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100464 */
465STATIC void
466xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000467 struct writeback_control *wbc,
Dave Chinner7bf7f352012-11-12 22:09:45 +1100468 xfs_ioend_t *ioend,
469 int fail)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100470{
David Chinnerd88992f2006-01-18 13:38:12 +1100471 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100472 xfs_ioend_t *next;
473 struct buffer_head *bh;
474 struct bio *bio;
475 sector_t lastblock = 0;
476
David Chinnerd88992f2006-01-18 13:38:12 +1100477 /* Pass 1 - start writeback */
478 do {
479 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000480 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100481 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100482 } while ((ioend = next) != NULL);
483
484 /* Pass 2 - submit I/O */
485 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100486 do {
487 next = ioend->io_list;
488 bio = NULL;
489
Dave Chinner7bf7f352012-11-12 22:09:45 +1100490 /*
491 * If we are failing the IO now, just mark the ioend with an
492 * error and finish it. This will run IO completion immediately
493 * as there is only one reference to the ioend at this point in
494 * time.
495 */
496 if (fail) {
Dave Chinner24513372014-06-25 14:58:08 +1000497 ioend->io_error = fail;
Dave Chinner7bf7f352012-11-12 22:09:45 +1100498 xfs_finish_ioend(ioend);
499 continue;
500 }
501
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100502 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100503
504 if (!bio) {
505 retry:
506 bio = xfs_alloc_ioend_bio(bh);
507 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000508 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100509 goto retry;
510 }
511
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000512 if (xfs_bio_add_buffer(bio, bh) != bh->b_size) {
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
517 lastblock = bh->b_blocknr;
518 }
519 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000520 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000521 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100522 } while ((ioend = next) != NULL);
523}
524
525/*
526 * Cancel submission of all buffer_heads so far in this endio.
527 * Toss the endio too. Only ever called for the initial page
528 * in a writepage request, so only ever one page.
529 */
530STATIC void
531xfs_cancel_ioend(
532 xfs_ioend_t *ioend)
533{
534 xfs_ioend_t *next;
535 struct buffer_head *bh, *next_bh;
536
537 do {
538 next = ioend->io_list;
539 bh = ioend->io_buffer_head;
540 do {
541 next_bh = bh->b_private;
542 clear_buffer_async_write(bh);
Brian Foster07d08682014-10-02 09:42:06 +1000543 /*
544 * The unwritten flag is cleared when added to the
545 * ioend. We're not submitting for I/O so mark the
546 * buffer unwritten again for next time around.
547 */
548 if (ioend->io_type == XFS_IO_UNWRITTEN)
549 set_buffer_unwritten(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100550 unlock_buffer(bh);
551 } while ((bh = next_bh) != NULL);
552
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100553 mempool_free(ioend, xfs_ioend_pool);
554 } while ((ioend = next) != NULL);
555}
556
557/*
558 * Test to see if we've been building up a completion structure for
559 * earlier buffers -- if so, we try to append to this ioend if we
560 * can, otherwise we finish off any current ioend and start another.
561 * Return true if we've finished the given ioend.
562 */
563STATIC void
564xfs_add_to_ioend(
565 struct inode *inode,
566 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100567 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100568 unsigned int type,
569 xfs_ioend_t **result,
570 int need_ioend)
571{
572 xfs_ioend_t *ioend = *result;
573
574 if (!ioend || need_ioend || type != ioend->io_type) {
575 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100576
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100577 ioend = xfs_alloc_ioend(inode, type);
578 ioend->io_offset = offset;
579 ioend->io_buffer_head = bh;
580 ioend->io_buffer_tail = bh;
581 if (previous)
582 previous->io_list = ioend;
583 *result = ioend;
584 } else {
585 ioend->io_buffer_tail->b_private = bh;
586 ioend->io_buffer_tail = bh;
587 }
588
589 bh->b_private = NULL;
590 ioend->io_size += bh->b_size;
591}
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100594xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000595 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100596 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000597 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000598 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100599{
600 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000601 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000602 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
603 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100604
Christoph Hellwig207d0412010-04-28 12:28:56 +0000605 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
606 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100607
Christoph Hellwige5131822010-04-28 12:28:55 +0000608 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000609 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100610
Christoph Hellwig046f1682010-04-28 12:28:52 +0000611 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100612
613 bh->b_blocknr = bn;
614 set_buffer_mapped(bh);
615}
616
617STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000619 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000621 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000622 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000624 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
625 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Christoph Hellwig207d0412010-04-28 12:28:56 +0000627 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 set_buffer_mapped(bh);
629 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100630 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631}
632
633/*
Dave Chinnera49935f2014-03-07 16:19:14 +1100634 * Test if a given page contains at least one buffer of a given @type.
635 * If @check_all_buffers is true, then we walk all the buffers in the page to
636 * try to find one of the type passed in. If it is not set, then the caller only
637 * needs to check the first buffer on the page for a match.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 */
Dave Chinnera49935f2014-03-07 16:19:14 +1100639STATIC bool
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000640xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100641 struct page *page,
Dave Chinnera49935f2014-03-07 16:19:14 +1100642 unsigned int type,
643 bool check_all_buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644{
Dave Chinnera49935f2014-03-07 16:19:14 +1100645 struct buffer_head *bh;
646 struct buffer_head *head;
647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 if (PageWriteback(page))
Dave Chinnera49935f2014-03-07 16:19:14 +1100649 return false;
650 if (!page->mapping)
651 return false;
652 if (!page_has_buffers(page))
653 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Dave Chinnera49935f2014-03-07 16:19:14 +1100655 bh = head = page_buffers(page);
656 do {
657 if (buffer_unwritten(bh)) {
658 if (type == XFS_IO_UNWRITTEN)
659 return true;
660 } else if (buffer_delay(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100661 if (type == XFS_IO_DELALLOC)
Dave Chinnera49935f2014-03-07 16:19:14 +1100662 return true;
663 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100664 if (type == XFS_IO_OVERWRITE)
Dave Chinnera49935f2014-03-07 16:19:14 +1100665 return true;
666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Dave Chinnera49935f2014-03-07 16:19:14 +1100668 /* If we are only checking the first buffer, we are done now. */
669 if (!check_all_buffers)
670 break;
671 } while ((bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672
Dave Chinnera49935f2014-03-07 16:19:14 +1100673 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674}
675
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676/*
677 * Allocate & map buffers for page given the extent map. Write it out.
678 * except for the original page of a writepage, this is called on
679 * delalloc/unwritten pages only, for the original page it is possible
680 * that the page has no mapping at all.
681 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100682STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683xfs_convert_page(
684 struct inode *inode,
685 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100686 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000687 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100688 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000689 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100691 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100692 xfs_off_t end_offset;
693 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100694 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700695 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100696 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100697 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100699 if (page->index != tindex)
700 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200701 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100702 goto fail;
703 if (PageWriteback(page))
704 goto fail_unlock_page;
705 if (page->mapping != inode->i_mapping)
706 goto fail_unlock_page;
Dave Chinnera49935f2014-03-07 16:19:14 +1100707 if (!xfs_check_page_type(page, (*ioendp)->io_type, false))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100708 goto fail_unlock_page;
709
Nathan Scott24e17b52005-05-05 13:33:20 -0700710 /*
711 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000712 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100713 *
714 * Derivation:
715 *
716 * End offset is the highest offset that this page should represent.
717 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
718 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
719 * hence give us the correct page_dirty count. On any other page,
720 * it will be zero and in that case we need page_dirty to be the
721 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700722 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100723 end_offset = min_t(unsigned long long,
724 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
725 i_size_read(inode));
726
Dave Chinner480d7462013-05-20 09:51:08 +1000727 /*
728 * If the current map does not span the entire page we are about to try
729 * to write, then give up. The only way we can write a page that spans
730 * multiple mappings in a single writeback iteration is via the
731 * xfs_vm_writepage() function. Data integrity writeback requires the
732 * entire page to be written in a single attempt, otherwise the part of
733 * the page we don't write here doesn't get written as part of the data
734 * integrity sync.
735 *
736 * For normal writeback, we also don't attempt to write partial pages
737 * here as it simply means that write_cache_pages() will see it under
738 * writeback and ignore the page until some point in the future, at
739 * which time this will be the only page in the file that needs
740 * writeback. Hence for more optimal IO patterns, we should always
741 * avoid partial page writeback due to multiple mappings on a page here.
742 */
743 if (!xfs_imap_valid(inode, imap, end_offset))
744 goto fail_unlock_page;
745
Nathan Scott24e17b52005-05-05 13:33:20 -0700746 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100747 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
748 PAGE_CACHE_SIZE);
749 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
750 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700751
Dave Chinnera49935f2014-03-07 16:19:14 +1100752 /*
753 * The moment we find a buffer that doesn't match our current type
754 * specification or can't be written, abort the loop and start
755 * writeback. As per the above xfs_imap_valid() check, only
756 * xfs_vm_writepage() can handle partial page writeback fully - we are
757 * limited here to the buffers that are contiguous with the current
758 * ioend, and hence a buffer we can't write breaks that contiguity and
759 * we have to defer the rest of the IO to xfs_vm_writepage().
760 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 bh = head = page_buffers(page);
762 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100763 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100765 if (!buffer_uptodate(bh))
766 uptodate = 0;
767 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
768 done = 1;
Dave Chinnera49935f2014-03-07 16:19:14 +1100769 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100770 }
771
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000772 if (buffer_unwritten(bh) || buffer_delay(bh) ||
773 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100774 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500775 type = XFS_IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000776 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500777 type = XFS_IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000778 else
Alain Renaud0d882a32012-05-22 15:56:21 -0500779 type = XFS_IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100780
Dave Chinnera49935f2014-03-07 16:19:14 +1100781 /*
782 * imap should always be valid because of the above
783 * partial page end_offset check on the imap.
784 */
785 ASSERT(xfs_imap_valid(inode, imap, offset));
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100786
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000787 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -0500788 if (type != XFS_IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000789 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000790 xfs_add_to_ioend(inode, bh, offset, type,
791 ioendp, done);
792
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100793 page_dirty--;
794 count++;
795 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000796 done = 1;
Dave Chinnera49935f2014-03-07 16:19:14 +1100797 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100799 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100801 if (uptodate && bh == head)
802 SetPageUptodate(page);
803
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000804 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000805 if (--wbc->nr_to_write <= 0 &&
806 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000807 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000809 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100810
811 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100812 fail_unlock_page:
813 unlock_page(page);
814 fail:
815 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816}
817
818/*
819 * Convert & write out a cluster of pages in the same extent as defined
820 * by mp and following the start page.
821 */
822STATIC void
823xfs_cluster_write(
824 struct inode *inode,
825 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000826 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100827 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 pgoff_t tlast)
830{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100831 struct pagevec pvec;
832 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100834 pagevec_init(&pvec, 0);
835 while (!done && tindex <= tlast) {
836 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
837
838 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100840
841 for (i = 0; i < pagevec_count(&pvec); i++) {
842 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000843 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100844 if (done)
845 break;
846 }
847
848 pagevec_release(&pvec);
849 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 }
851}
852
Dave Chinner3ed3a432010-03-05 02:00:42 +0000853STATIC void
854xfs_vm_invalidatepage(
855 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400856 unsigned int offset,
857 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000858{
Lukas Czerner34097df2013-05-21 23:58:01 -0400859 trace_xfs_invalidatepage(page->mapping->host, page, offset,
860 length);
861 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000862}
863
864/*
865 * If the page has delalloc buffers on it, we need to punch them out before we
866 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
867 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
868 * is done on that same region - the delalloc extent is returned when none is
869 * supposed to be there.
870 *
871 * We prevent this by truncating away the delalloc regions on the page before
872 * invalidating it. Because they are delalloc, we can do this without needing a
873 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
874 * truncation without a transaction as there is no space left for block
875 * reservation (typically why we see a ENOSPC in writeback).
876 *
877 * This is not a performance critical path, so for now just do the punching a
878 * buffer head at a time.
879 */
880STATIC void
881xfs_aops_discard_page(
882 struct page *page)
883{
884 struct inode *inode = page->mapping->host;
885 struct xfs_inode *ip = XFS_I(inode);
886 struct buffer_head *bh, *head;
887 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000888
Dave Chinnera49935f2014-03-07 16:19:14 +1100889 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000890 goto out_invalidate;
891
Dave Chinnere8c37532010-03-15 02:36:35 +0000892 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
893 goto out_invalidate;
894
Dave Chinner4f107002011-03-07 10:00:35 +1100895 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000896 "page discard on page %p, inode 0x%llx, offset %llu.",
897 page, ip->i_ino, offset);
898
899 xfs_ilock(ip, XFS_ILOCK_EXCL);
900 bh = head = page_buffers(page);
901 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000902 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100903 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000904
905 if (!buffer_delay(bh))
906 goto next_buffer;
907
Dave Chinnerc726de42010-11-30 15:14:39 +1100908 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
909 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000910 if (error) {
911 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000912 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100913 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000914 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000915 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000916 break;
917 }
918next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100919 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000920
921 } while ((bh = bh->b_this_page) != head);
922
923 xfs_iunlock(ip, XFS_ILOCK_EXCL);
924out_invalidate:
Lukas Czernerd47992f2013-05-21 23:17:23 -0400925 xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000926 return;
927}
928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000930 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000932 * For delalloc space on the page we need to allocate space and flush it.
933 * For unwritten space on the page we need to start the conversion to
934 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000935 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937STATIC int
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000938xfs_vm_writepage(
939 struct page *page,
940 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941{
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000942 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100943 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000944 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100945 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100947 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000949 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000950 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000951 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000952 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000953 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
Lukas Czerner34097df2013-05-21 23:58:01 -0400955 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000956
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000957 ASSERT(page_has_buffers(page));
958
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000959 /*
960 * Refuse to write the page out if we are called from reclaim context.
961 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400962 * This avoids stack overflows when called from deeply used stacks in
963 * random callers for direct reclaim or memcg reclaim. We explicitly
964 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000965 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700966 * This should never happen except in the case of a VM regression so
967 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000968 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700969 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
970 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000971 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000972
973 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200974 * Given that we do not allow direct reclaim to call us, we should
975 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000976 */
Christoph Hellwig448011e2014-06-06 16:05:15 +1000977 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000978 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 /* Is this page beyond the end of the file? */
981 offset = i_size_read(inode);
982 end_index = offset >> PAGE_CACHE_SHIFT;
983 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +1000984
985 /*
986 * The page index is less than the end_index, adjust the end_offset
987 * to the highest offset that this page should represent.
988 * -----------------------------------------------------
989 * | file mapping | <EOF> |
990 * -----------------------------------------------------
991 * | Page ... | Page N-2 | Page N-1 | Page N | |
992 * ^--------------------------------^----------|--------
993 * | desired writeback range | see else |
994 * ---------------------------------^------------------|
995 */
996 if (page->index < end_index)
997 end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT;
998 else {
999 /*
1000 * Check whether the page to write out is beyond or straddles
1001 * i_size or not.
1002 * -------------------------------------------------------
1003 * | file mapping | <EOF> |
1004 * -------------------------------------------------------
1005 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1006 * ^--------------------------------^-----------|---------
1007 * | | Straddles |
1008 * ---------------------------------^-----------|--------|
1009 */
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001010 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
1011
1012 /*
Jan Karaff9a28f2013-03-14 14:30:54 +01001013 * Skip the page if it is fully outside i_size, e.g. due to a
1014 * truncate operation that is in progress. We must redirty the
1015 * page so that reclaim stops reclaiming it. Otherwise
1016 * xfs_vm_releasepage() is called on it and gets confused.
Jie Liu8695d272014-05-20 08:24:26 +10001017 *
1018 * Note that the end_index is unsigned long, it would overflow
1019 * if the given offset is greater than 16TB on 32-bit system
1020 * and if we do check the page is fully outside i_size or not
1021 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1022 * will be evaluated to 0. Hence this page will be redirtied
1023 * and be written out repeatedly which would result in an
1024 * infinite loop, the user program that perform this operation
1025 * will hang. Instead, we can verify this situation by checking
1026 * if the page to write is totally beyond the i_size or if it's
1027 * offset is just equal to the EOF.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001028 */
Jie Liu8695d272014-05-20 08:24:26 +10001029 if (page->index > end_index ||
1030 (page->index == end_index && offset_into_page == 0))
Jan Karaff9a28f2013-03-14 14:30:54 +01001031 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001032
1033 /*
1034 * The page straddles i_size. It must be zeroed out on each
1035 * and every writepage invocation because it may be mmapped.
1036 * "A file is mapped in multiples of the page size. For a file
Jie Liu8695d272014-05-20 08:24:26 +10001037 * that is not a multiple of the page size, the remaining
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001038 * memory is zeroed when mapped, and writes to that region are
1039 * not written out to the file."
1040 */
1041 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
Jie Liu8695d272014-05-20 08:24:26 +10001042
1043 /* Adjust the end_offset to the end of file */
1044 end_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 }
1046
Nathan Scott24e17b52005-05-05 13:33:20 -07001047 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -07001048
Nathan Scott24e17b52005-05-05 13:33:20 -07001049 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001050 offset = page_offset(page);
Alain Renaud0d882a32012-05-22 15:56:21 -05001051 type = XFS_IO_OVERWRITE;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001052
Christoph Hellwigdbcdde32011-07-08 14:34:14 +02001053 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001054 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001055
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +00001057 int new_ioend = 0;
1058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (offset >= end_offset)
1060 break;
1061 if (!buffer_uptodate(bh))
1062 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001064 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +00001065 * set_page_dirty dirties all buffers in a page, independent
1066 * of their state. The dirty state however is entirely
1067 * meaningless for holes (!mapped && uptodate), so skip
1068 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001069 */
1070 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001071 imap_valid = 0;
1072 continue;
1073 }
1074
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001075 if (buffer_unwritten(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001076 if (type != XFS_IO_UNWRITTEN) {
1077 type = XFS_IO_UNWRITTEN;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001078 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001079 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001080 } else if (buffer_delay(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001081 if (type != XFS_IO_DELALLOC) {
1082 type = XFS_IO_DELALLOC;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001083 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001085 } else if (buffer_uptodate(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001086 if (type != XFS_IO_OVERWRITE) {
1087 type = XFS_IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +00001088 imap_valid = 0;
1089 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001090 } else {
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001091 if (PageUptodate(page))
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001092 ASSERT(buffer_mapped(bh));
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001093 /*
1094 * This buffer is not uptodate and will not be
1095 * written to disk. Ensure that we will put any
1096 * subsequent writeable buffers into a new
1097 * ioend.
1098 */
1099 imap_valid = 0;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001100 continue;
1101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001103 if (imap_valid)
1104 imap_valid = xfs_imap_valid(inode, &imap, offset);
1105 if (!imap_valid) {
1106 /*
1107 * If we didn't have a valid mapping then we need to
1108 * put the new mapping into a separate ioend structure.
1109 * This ensures non-contiguous extents always have
1110 * separate ioends, which is particularly important
1111 * for unwritten extent conversion at I/O completion
1112 * time.
1113 */
1114 new_ioend = 1;
1115 err = xfs_map_blocks(inode, offset, &imap, type,
1116 nonblocking);
1117 if (err)
1118 goto error;
1119 imap_valid = xfs_imap_valid(inode, &imap, offset);
1120 }
1121 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001122 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -05001123 if (type != XFS_IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001124 xfs_map_at_offset(inode, bh, &imap, offset);
1125 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1126 new_ioend);
1127 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001129
1130 if (!iohead)
1131 iohead = ioend;
1132
1133 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135 if (uptodate && bh == head)
1136 SetPageUptodate(page);
1137
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001138 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139
Dave Chinner7bf7f352012-11-12 22:09:45 +11001140 /* if there is no IO to be submitted for this page, we are done */
1141 if (!ioend)
1142 return 0;
1143
1144 ASSERT(iohead);
1145
1146 /*
1147 * Any errors from this point onwards need tobe reported through the IO
1148 * completion path as we have marked the initial page as under writeback
1149 * and unlocked it.
1150 */
1151 if (imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001152 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001153
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001154 end_index = imap.br_startoff + imap.br_blockcount;
1155
1156 /* to bytes */
1157 end_index <<= inode->i_blkbits;
1158
1159 /* to pages */
1160 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1161
1162 /* check against file size */
1163 if (end_index > last_index)
1164 end_index = last_index;
1165
Christoph Hellwig207d0412010-04-28 12:28:56 +00001166 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001167 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 }
1169
Christoph Hellwig281627d2012-03-13 08:41:05 +00001170
Dave Chinner7bf7f352012-11-12 22:09:45 +11001171 /*
1172 * Reserve log space if we might write beyond the on-disk inode size.
1173 */
1174 err = 0;
1175 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1176 err = xfs_setfilesize_trans_alloc(ioend);
1177
1178 xfs_submit_ioend(wbc, iohead, err);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001179
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001180 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181
1182error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001183 if (iohead)
1184 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001186 if (err == -EAGAIN)
1187 goto redirty;
1188
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001189 xfs_aops_discard_page(page);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001190 ClearPageUptodate(page);
1191 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001193
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001194redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001195 redirty_page_for_writepage(wbc, page);
1196 unlock_page(page);
1197 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001198}
1199
Nathan Scott7d4fb402006-06-09 15:27:16 +10001200STATIC int
1201xfs_vm_writepages(
1202 struct address_space *mapping,
1203 struct writeback_control *wbc)
1204{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001205 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001206 return generic_writepages(mapping, wbc);
1207}
1208
Nathan Scottf51623b2006-03-14 13:26:27 +11001209/*
1210 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001211 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001212 * have buffer heads in this call.
1213 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001214 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001215 */
1216STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001217xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001218 struct page *page,
1219 gfp_t gfp_mask)
1220{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001221 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001222
Lukas Czerner34097df2013-05-21 23:58:01 -04001223 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001224
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001225 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001226
Christoph Hellwig448011e2014-06-06 16:05:15 +10001227 if (WARN_ON_ONCE(delalloc))
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001228 return 0;
Christoph Hellwig448011e2014-06-06 16:05:15 +10001229 if (WARN_ON_ONCE(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001230 return 0;
1231
Nathan Scottf51623b2006-03-14 13:26:27 +11001232 return try_to_free_buffers(page);
1233}
1234
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001236__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 struct inode *inode,
1238 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 struct buffer_head *bh_result,
1240 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001241 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001243 struct xfs_inode *ip = XFS_I(inode);
1244 struct xfs_mount *mp = ip->i_mount;
1245 xfs_fileoff_t offset_fsb, end_fsb;
1246 int error = 0;
1247 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001248 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001249 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001250 xfs_off_t offset;
1251 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001252 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001253
1254 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001255 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001257 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001258 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1259 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001260
1261 if (!create && direct && offset >= i_size_read(inode))
1262 return 0;
1263
Dave Chinner507630b2012-03-27 10:34:50 -04001264 /*
1265 * Direct I/O is usually done on preallocated files, so try getting
1266 * a block mapping without an exclusive lock first. For buffered
1267 * writes we already have the exclusive iolock anyway, so avoiding
1268 * a lock roundtrip here by taking the ilock exclusive from the
1269 * beginning is a useful micro optimization.
1270 */
1271 if (create && !direct) {
Christoph Hellwiga206c812010-12-10 08:42:20 +00001272 lockmode = XFS_ILOCK_EXCL;
1273 xfs_ilock(ip, lockmode);
1274 } else {
Christoph Hellwig309ecac82013-12-06 12:30:09 -08001275 lockmode = xfs_ilock_data_map_shared(ip);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001276 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001277
Dave Chinnerd2c28192012-06-08 15:44:53 +10001278 ASSERT(offset <= mp->m_super->s_maxbytes);
1279 if (offset + size > mp->m_super->s_maxbytes)
1280 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001281 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1282 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1283
Dave Chinner5c8ed202011-09-18 20:40:45 +00001284 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1285 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001287 goto out_unlock;
1288
1289 if (create &&
1290 (!nimaps ||
1291 (imap.br_startblock == HOLESTARTBLOCK ||
1292 imap.br_startblock == DELAYSTARTBLOCK))) {
Dave Chinneraff3a9e2012-04-23 15:58:44 +10001293 if (direct || xfs_get_extsz_hint(ip)) {
Dave Chinner507630b2012-03-27 10:34:50 -04001294 /*
1295 * Drop the ilock in preparation for starting the block
1296 * allocation transaction. It will be retaken
1297 * exclusively inside xfs_iomap_write_direct for the
1298 * actual allocation.
1299 */
1300 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001301 error = xfs_iomap_write_direct(ip, offset, size,
1302 &imap, nimaps);
Dave Chinner507630b2012-03-27 10:34:50 -04001303 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10001304 return error;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001305 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001306 } else {
Dave Chinner507630b2012-03-27 10:34:50 -04001307 /*
1308 * Delalloc reservations do not require a transaction,
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001309 * we can go on without dropping the lock here. If we
1310 * are allocating a new delalloc block, make sure that
1311 * we set the new flag so that we mark the buffer new so
1312 * that we know that it is newly allocated if the write
1313 * fails.
Dave Chinner507630b2012-03-27 10:34:50 -04001314 */
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001315 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1316 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001317 error = xfs_iomap_write_delay(ip, offset, size, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001318 if (error)
1319 goto out_unlock;
1320
1321 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001322 }
Christoph Hellwiga206c812010-12-10 08:42:20 +00001323
1324 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1325 } else if (nimaps) {
1326 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001327 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001328 } else {
1329 trace_xfs_get_blocks_notfound(ip, offset, size);
1330 goto out_unlock;
1331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
Christoph Hellwig207d0412010-04-28 12:28:56 +00001333 if (imap.br_startblock != HOLESTARTBLOCK &&
1334 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001335 /*
1336 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 * the read case (treat as if we're reading into a hole).
1338 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001339 if (create || !ISUNWRITTEN(&imap))
1340 xfs_map_buffer(inode, bh_result, &imap, offset);
1341 if (create && ISUNWRITTEN(&imap)) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001342 if (direct) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 bh_result->b_private = inode;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001344 set_buffer_defer_completion(bh_result);
1345 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 }
1348 }
1349
Nathan Scottc2536662006-03-29 10:44:40 +10001350 /*
1351 * If this is a realtime file, data may be on a different device.
1352 * to that pointed to from the buffer_head b_bdev currently.
1353 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001354 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
Nathan Scottc2536662006-03-29 10:44:40 +10001356 /*
David Chinner549054a2007-02-10 18:36:35 +11001357 * If we previously allocated a block out beyond eof and we are now
1358 * coming back to use it then we will need to flag it as new even if it
1359 * has a disk address.
1360 *
1361 * With sub-block writes into unwritten extents we also need to mark
1362 * the buffer as new so that the unwritten parts of the buffer gets
1363 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 */
1365 if (create &&
1366 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001367 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001368 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370
Christoph Hellwig207d0412010-04-28 12:28:56 +00001371 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 BUG_ON(direct);
1373 if (create) {
1374 set_buffer_uptodate(bh_result);
1375 set_buffer_mapped(bh_result);
1376 set_buffer_delay(bh_result);
1377 }
1378 }
1379
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001380 /*
1381 * If this is O_DIRECT or the mpage code calling tell them how large
1382 * the mapping is, so that we can avoid repeated get_blocks calls.
Dave Chinner0e1f7892014-04-17 08:15:19 +10001383 *
1384 * If the mapping spans EOF, then we have to break the mapping up as the
1385 * mapping for blocks beyond EOF must be marked new so that sub block
1386 * regions can be correctly zeroed. We can't do this for mappings within
1387 * EOF unless the mapping was just allocated or is unwritten, otherwise
1388 * the callers would overwrite existing data with zeros. Hence we have
1389 * to split the mapping into a range up to and including EOF, and a
1390 * second mapping for beyond EOF.
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001391 */
Nathan Scottc2536662006-03-29 10:44:40 +10001392 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001393 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001394
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001395 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1396 mapping_size <<= inode->i_blkbits;
1397
1398 ASSERT(mapping_size > 0);
1399 if (mapping_size > size)
1400 mapping_size = size;
Dave Chinner0e1f7892014-04-17 08:15:19 +10001401 if (offset < i_size_read(inode) &&
1402 offset + mapping_size >= i_size_read(inode)) {
1403 /* limit mapping to block that spans EOF */
1404 mapping_size = roundup_64(i_size_read(inode) - offset,
1405 1 << inode->i_blkbits);
1406 }
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001407 if (mapping_size > LONG_MAX)
1408 mapping_size = LONG_MAX;
1409
1410 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 }
1412
1413 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001414
1415out_unlock:
1416 xfs_iunlock(ip, lockmode);
Dave Chinner24513372014-06-25 14:58:08 +10001417 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418}
1419
1420int
Nathan Scottc2536662006-03-29 10:44:40 +10001421xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 struct inode *inode,
1423 sector_t iblock,
1424 struct buffer_head *bh_result,
1425 int create)
1426{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001427 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428}
1429
1430STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001431xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 struct inode *inode,
1433 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 struct buffer_head *bh_result,
1435 int create)
1436{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001437 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438}
1439
Christoph Hellwig209fb872010-07-18 21:17:11 +00001440/*
1441 * Complete a direct I/O write request.
1442 *
1443 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1444 * need to issue a transaction to convert the range from unwritten to written
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001445 * extents.
Christoph Hellwig209fb872010-07-18 21:17:11 +00001446 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001447STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001448xfs_end_io_direct_write(
1449 struct kiocb *iocb,
1450 loff_t offset,
1451 ssize_t size,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001452 void *private)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001453{
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001454 struct inode *inode = file_inode(iocb->ki_filp);
1455 struct xfs_inode *ip = XFS_I(inode);
1456 struct xfs_mount *mp = ip->i_mount;
1457
1458 if (XFS_FORCED_SHUTDOWN(mp))
1459 return;
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 */
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001467 if (offset + size > i_size_read(inode))
1468 i_size_write(inode, offset + size);
Christoph Hellwig2813d682011-12-18 20:00:12 +00001469
1470 /*
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001471 * For direct I/O we do not know if we need to allocate blocks or not,
1472 * so we can't preallocate an append transaction, as that results in
1473 * nested reservations and log space deadlocks. Hence allocate the
1474 * transaction here. While this is sub-optimal and can block IO
1475 * completion for some time, we're stuck with doing it this way until
1476 * we can pass the ioend to the direct IO allocation callbacks and
1477 * avoid nesting that way.
Christoph Hellwigf0973862005-09-05 08:22:52 +10001478 */
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001479 if (private && size > 0) {
1480 xfs_iomap_write_unwritten(ip, offset, size);
1481 } else if (offset + size > ip->i_d.di_size) {
1482 struct xfs_trans *tp;
1483 int error;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001484
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001485 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
1486 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1487 if (error) {
1488 xfs_trans_cancel(tp, 0);
1489 return;
1490 }
Christoph Hellwig209fb872010-07-18 21:17:11 +00001491
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001492 xfs_setfilesize(ip, tp, offset, size);
1493 }
Christoph Hellwigf0973862005-09-05 08:22:52 +10001494}
1495
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001497xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 int rw,
1499 struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05001500 struct iov_iter *iter,
1501 loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001503 struct inode *inode = iocb->ki_filp->f_mapping->host;
1504 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
Christoph Hellwig209fb872010-07-18 21:17:11 +00001506 if (rw & WRITE) {
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001507 return __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
Al Viro31b14032014-03-05 01:33:16 -05001508 offset, xfs_get_blocks_direct,
Christoph Hellwig9862f622014-02-10 10:28:04 +11001509 xfs_end_io_direct_write, NULL,
1510 DIO_ASYNC_EXTEND);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001511 }
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001512 return __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
1513 offset, xfs_get_blocks_direct,
1514 NULL, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515}
1516
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001517/*
1518 * Punch out the delalloc blocks we have already allocated.
1519 *
1520 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1521 * as the page is still locked at this point.
1522 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001523STATIC void
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001524xfs_vm_kill_delalloc_range(
1525 struct inode *inode,
1526 loff_t start,
1527 loff_t end)
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001528{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001529 struct xfs_inode *ip = XFS_I(inode);
1530 xfs_fileoff_t start_fsb;
1531 xfs_fileoff_t end_fsb;
1532 int error;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001533
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001534 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1535 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1536 if (end_fsb <= start_fsb)
1537 return;
Dave Chinnerc726de42010-11-30 15:14:39 +11001538
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001539 xfs_ilock(ip, XFS_ILOCK_EXCL);
1540 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1541 end_fsb - start_fsb);
1542 if (error) {
1543 /* something screwed, just bail */
1544 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1545 xfs_alert(ip->i_mount,
1546 "xfs_vm_write_failed: unable to clean up ino %lld",
1547 ip->i_ino);
Dave Chinnerc726de42010-11-30 15:14:39 +11001548 }
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001549 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001550 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001551}
1552
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001553STATIC void
1554xfs_vm_write_failed(
1555 struct inode *inode,
1556 struct page *page,
1557 loff_t pos,
1558 unsigned len)
1559{
Jie Liu58e59852013-07-16 13:11:16 +08001560 loff_t block_offset;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001561 loff_t block_start;
1562 loff_t block_end;
1563 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1564 loff_t to = from + len;
1565 struct buffer_head *bh, *head;
1566
Jie Liu58e59852013-07-16 13:11:16 +08001567 /*
1568 * The request pos offset might be 32 or 64 bit, this is all fine
1569 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1570 * platform, the high 32-bit will be masked off if we evaluate the
1571 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1572 * 0xfffff000 as an unsigned long, hence the result is incorrect
1573 * which could cause the following ASSERT failed in most cases.
1574 * In order to avoid this, we can evaluate the block_offset of the
1575 * start of the page by using shifts rather than masks the mismatch
1576 * problem.
1577 */
1578 block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT;
1579
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001580 ASSERT(block_offset + from == pos);
1581
1582 head = page_buffers(page);
1583 block_start = 0;
1584 for (bh = head; bh != head || !block_start;
1585 bh = bh->b_this_page, block_start = block_end,
1586 block_offset += bh->b_size) {
1587 block_end = block_start + bh->b_size;
1588
1589 /* skip buffers before the write */
1590 if (block_end <= from)
1591 continue;
1592
1593 /* if the buffer is after the write, we're done */
1594 if (block_start >= to)
1595 break;
1596
1597 if (!buffer_delay(bh))
1598 continue;
1599
1600 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1601 continue;
1602
1603 xfs_vm_kill_delalloc_range(inode, block_offset,
1604 block_offset + bh->b_size);
Dave Chinner4ab9ed52014-04-14 18:11:58 +10001605
1606 /*
1607 * This buffer does not contain data anymore. make sure anyone
1608 * who finds it knows that for certain.
1609 */
1610 clear_buffer_delay(bh);
1611 clear_buffer_uptodate(bh);
1612 clear_buffer_mapped(bh);
1613 clear_buffer_new(bh);
1614 clear_buffer_dirty(bh);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001615 }
1616
1617}
1618
1619/*
1620 * This used to call block_write_begin(), but it unlocks and releases the page
1621 * on error, and we need that page to be able to punch stale delalloc blocks out
1622 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1623 * the appropriate point.
1624 */
Nathan Scottf51623b2006-03-14 13:26:27 +11001625STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001626xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001627 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001628 struct address_space *mapping,
1629 loff_t pos,
1630 unsigned len,
1631 unsigned flags,
1632 struct page **pagep,
1633 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001634{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001635 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1636 struct page *page;
1637 int status;
Christoph Hellwig155130a2010-06-04 11:29:58 +02001638
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001639 ASSERT(len <= PAGE_CACHE_SIZE);
1640
Dave Chinnerad22c7a2013-10-29 22:11:57 +11001641 page = grab_cache_page_write_begin(mapping, index, flags);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001642 if (!page)
1643 return -ENOMEM;
1644
1645 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1646 if (unlikely(status)) {
1647 struct inode *inode = mapping->host;
Dave Chinner72ab70a2014-04-14 18:13:29 +10001648 size_t isize = i_size_read(inode);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001649
1650 xfs_vm_write_failed(inode, page, pos, len);
1651 unlock_page(page);
1652
Dave Chinner72ab70a2014-04-14 18:13:29 +10001653 /*
1654 * If the write is beyond EOF, we only want to kill blocks
1655 * allocated in this write, not blocks that were previously
1656 * written successfully.
1657 */
1658 if (pos + len > isize) {
1659 ssize_t start = max_t(ssize_t, pos, isize);
1660
1661 truncate_pagecache_range(inode, start, pos + len);
1662 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001663
1664 page_cache_release(page);
1665 page = NULL;
1666 }
1667
1668 *pagep = page;
1669 return status;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001670}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001671
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001672/*
Dave Chinneraad3f372014-04-14 18:14:11 +10001673 * On failure, we only need to kill delalloc blocks beyond EOF in the range of
1674 * this specific write because they will never be written. Previous writes
1675 * beyond EOF where block allocation succeeded do not need to be trashed, so
1676 * only new blocks from this write should be trashed. For blocks within
1677 * EOF, generic_write_end() zeros them so they are safe to leave alone and be
1678 * written with all the other valid data.
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001679 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001680STATIC int
1681xfs_vm_write_end(
1682 struct file *file,
1683 struct address_space *mapping,
1684 loff_t pos,
1685 unsigned len,
1686 unsigned copied,
1687 struct page *page,
1688 void *fsdata)
1689{
1690 int ret;
1691
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001692 ASSERT(len <= PAGE_CACHE_SIZE);
1693
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001694 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001695 if (unlikely(ret < len)) {
1696 struct inode *inode = mapping->host;
1697 size_t isize = i_size_read(inode);
1698 loff_t to = pos + len;
1699
1700 if (to > isize) {
Dave Chinneraad3f372014-04-14 18:14:11 +10001701 /* only kill blocks in this write beyond EOF */
1702 if (pos > isize)
1703 isize = pos;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001704 xfs_vm_kill_delalloc_range(inode, isize, to);
Dave Chinneraad3f372014-04-14 18:14:11 +10001705 truncate_pagecache_range(inode, isize, to);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001706 }
1707 }
Christoph Hellwig155130a2010-06-04 11:29:58 +02001708 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001709}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001712xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 struct address_space *mapping,
1714 sector_t block)
1715{
1716 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001717 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001719 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001720 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001721 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001722 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001723 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724}
1725
1726STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001727xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 struct file *unused,
1729 struct page *page)
1730{
Nathan Scottc2536662006-03-29 10:44:40 +10001731 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732}
1733
1734STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001735xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 struct file *unused,
1737 struct address_space *mapping,
1738 struct list_head *pages,
1739 unsigned nr_pages)
1740{
Nathan Scottc2536662006-03-29 10:44:40 +10001741 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742}
1743
Dave Chinner22e757a2014-09-02 12:12:51 +10001744/*
1745 * This is basically a copy of __set_page_dirty_buffers() with one
1746 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1747 * dirty, we'll never be able to clean them because we don't write buffers
1748 * beyond EOF, and that means we can't invalidate pages that span EOF
1749 * that have been marked dirty. Further, the dirty state can leak into
1750 * the file interior if the file is extended, resulting in all sorts of
1751 * bad things happening as the state does not match the underlying data.
1752 *
1753 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1754 * this only exist because of bufferheads and how the generic code manages them.
1755 */
1756STATIC int
1757xfs_vm_set_page_dirty(
1758 struct page *page)
1759{
1760 struct address_space *mapping = page->mapping;
1761 struct inode *inode = mapping->host;
1762 loff_t end_offset;
1763 loff_t offset;
1764 int newly_dirty;
1765
1766 if (unlikely(!mapping))
1767 return !TestSetPageDirty(page);
1768
1769 end_offset = i_size_read(inode);
1770 offset = page_offset(page);
1771
1772 spin_lock(&mapping->private_lock);
1773 if (page_has_buffers(page)) {
1774 struct buffer_head *head = page_buffers(page);
1775 struct buffer_head *bh = head;
1776
1777 do {
1778 if (offset < end_offset)
1779 set_buffer_dirty(bh);
1780 bh = bh->b_this_page;
1781 offset += 1 << inode->i_blkbits;
1782 } while (bh != head);
1783 }
1784 newly_dirty = !TestSetPageDirty(page);
1785 spin_unlock(&mapping->private_lock);
1786
1787 if (newly_dirty) {
1788 /* sigh - __set_page_dirty() is static, so copy it here, too */
1789 unsigned long flags;
1790
1791 spin_lock_irqsave(&mapping->tree_lock, flags);
1792 if (page->mapping) { /* Race with truncate? */
1793 WARN_ON_ONCE(!PageUptodate(page));
1794 account_page_dirtied(page, mapping);
1795 radix_tree_tag_set(&mapping->page_tree,
1796 page_index(page), PAGECACHE_TAG_DIRTY);
1797 }
1798 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1799 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1800 }
1801 return newly_dirty;
1802}
1803
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001804const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001805 .readpage = xfs_vm_readpage,
1806 .readpages = xfs_vm_readpages,
1807 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001808 .writepages = xfs_vm_writepages,
Dave Chinner22e757a2014-09-02 12:12:51 +10001809 .set_page_dirty = xfs_vm_set_page_dirty,
Nathan Scott238f4c52006-03-17 17:26:25 +11001810 .releasepage = xfs_vm_releasepage,
1811 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001812 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001813 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001814 .bmap = xfs_vm_bmap,
1815 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001816 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001817 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001818 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819};