blob: 594e02c485b23d0359254ca31b03ffcebb4d90fb [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"
Darrick J. Wongef473662016-10-03 09:11:34 -070034#include "xfs_reflink.h"
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090035#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110037#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/writeback.h>
39
Christoph Hellwig273dda72016-02-08 14:40:51 +110040/* flags for direct write completions */
41#define XFS_DIO_FLAG_UNWRITTEN (1 << 0)
42#define XFS_DIO_FLAG_APPEND (1 << 1)
Darrick J. Wong0613f162016-10-03 09:11:37 -070043#define XFS_DIO_FLAG_COW (1 << 2)
Christoph Hellwig273dda72016-02-08 14:40:51 +110044
Dave Chinnerfbcc0252016-02-15 17:21:19 +110045/*
46 * structure owned by writepages passed to individual writepage calls
47 */
48struct xfs_writepage_ctx {
49 struct xfs_bmbt_irec imap;
50 bool imap_valid;
51 unsigned int io_type;
Dave Chinnerfbcc0252016-02-15 17:21:19 +110052 struct xfs_ioend *ioend;
53 sector_t last_block;
54};
55
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000056void
Nathan Scottf51623b2006-03-14 13:26:27 +110057xfs_count_page_state(
58 struct page *page,
59 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110060 int *unwritten)
61{
62 struct buffer_head *bh, *head;
63
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100064 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110065
66 bh = head = page_buffers(page);
67 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100068 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110069 (*unwritten) = 1;
70 else if (buffer_delay(bh))
71 (*delalloc) = 1;
72 } while ((bh = bh->b_this_page) != head);
73}
74
Ross Zwisler20a90f52016-02-26 15:19:52 -080075struct block_device *
Christoph Hellwig6214ed42007-09-14 15:23:17 +100076xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000077 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100078{
Christoph Hellwig046f1682010-04-28 12:28:52 +000079 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100080 struct xfs_mount *mp = ip->i_mount;
81
Eric Sandeen71ddabb2007-11-23 16:29:42 +110082 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100083 return mp->m_rtdev_targp->bt_bdev;
84 else
85 return mp->m_ddev_targp->bt_bdev;
86}
87
Christoph Hellwig0829c362005-09-02 16:58:49 +100088/*
Dave Chinner37992c12016-04-06 08:12:28 +100089 * We're now finished for good with this page. Update the page state via the
90 * associated buffer_heads, paying attention to the start and end offsets that
91 * we need to process on the page.
Dave Chinner28b783e2016-07-22 09:56:38 +100092 *
93 * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last
94 * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or
95 * the page at all, as we may be racing with memory reclaim and it can free both
96 * the bufferhead chain and the page as it will see the page as clean and
97 * unused.
Dave Chinner37992c12016-04-06 08:12:28 +100098 */
99static void
100xfs_finish_page_writeback(
101 struct inode *inode,
102 struct bio_vec *bvec,
103 int error)
104{
Dave Chinner37992c12016-04-06 08:12:28 +1000105 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
Dave Chinner28b783e2016-07-22 09:56:38 +1000106 struct buffer_head *head, *bh, *next;
Dave Chinner37992c12016-04-06 08:12:28 +1000107 unsigned int off = 0;
Dave Chinner28b783e2016-07-22 09:56:38 +1000108 unsigned int bsize;
Dave Chinner37992c12016-04-06 08:12:28 +1000109
110 ASSERT(bvec->bv_offset < PAGE_SIZE);
Christoph Hellwig690a7872016-05-20 10:29:15 +1000111 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
Dave Chinner37992c12016-04-06 08:12:28 +1000112 ASSERT(end < PAGE_SIZE);
Christoph Hellwig690a7872016-05-20 10:29:15 +1000113 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
Dave Chinner37992c12016-04-06 08:12:28 +1000114
115 bh = head = page_buffers(bvec->bv_page);
116
Dave Chinner28b783e2016-07-22 09:56:38 +1000117 bsize = bh->b_size;
Dave Chinner37992c12016-04-06 08:12:28 +1000118 do {
Dave Chinner28b783e2016-07-22 09:56:38 +1000119 next = bh->b_this_page;
Dave Chinner37992c12016-04-06 08:12:28 +1000120 if (off < bvec->bv_offset)
121 goto next_bh;
122 if (off > end)
123 break;
124 bh->b_end_io(bh, !error);
125next_bh:
Dave Chinner28b783e2016-07-22 09:56:38 +1000126 off += bsize;
127 } while ((bh = next) != head);
Dave Chinner37992c12016-04-06 08:12:28 +1000128}
129
130/*
131 * We're now finished for good with this ioend structure. Update the page
132 * state, release holds on bios, and finally free up memory. Do not use the
133 * ioend after this.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100134 */
Christoph Hellwig0829c362005-09-02 16:58:49 +1000135STATIC void
136xfs_destroy_ioend(
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000137 struct xfs_ioend *ioend,
138 int error)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000139{
Dave Chinner37992c12016-04-06 08:12:28 +1000140 struct inode *inode = ioend->io_inode;
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000141 struct bio *last = ioend->io_bio;
Dave Chinner37992c12016-04-06 08:12:28 +1000142 struct bio *bio, *next;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100143
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000144 for (bio = &ioend->io_inline_bio; bio; bio = next) {
Dave Chinner37992c12016-04-06 08:12:28 +1000145 struct bio_vec *bvec;
146 int i;
147
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000148 /*
149 * For the last bio, bi_private points to the ioend, so we
150 * need to explicitly end the iteration here.
151 */
152 if (bio == last)
153 next = NULL;
154 else
155 next = bio->bi_private;
Dave Chinner37992c12016-04-06 08:12:28 +1000156
157 /* walk each page on bio, ending page IO on them */
158 bio_for_each_segment_all(bvec, bio, i)
159 xfs_finish_page_writeback(inode, bvec, error);
160
161 bio_put(bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163}
164
165/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000166 * Fast and loose check if this write could update the on-disk inode size.
167 */
168static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
169{
170 return ioend->io_offset + ioend->io_size >
171 XFS_I(ioend->io_inode)->i_d.di_size;
172}
173
Christoph Hellwig281627d2012-03-13 08:41:05 +0000174STATIC int
175xfs_setfilesize_trans_alloc(
176 struct xfs_ioend *ioend)
177{
178 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
179 struct xfs_trans *tp;
180 int error;
181
Christoph Hellwig253f4912016-04-06 09:19:55 +1000182 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
183 if (error)
Christoph Hellwig281627d2012-03-13 08:41:05 +0000184 return error;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000185
186 ioend->io_append_trans = tp;
187
188 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100189 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200190 * we released it.
191 */
Oleg Nesterovbee91822015-07-19 23:48:20 +0200192 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
Jan Karad9457dc2012-06-12 16:20:39 +0200193 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000194 * We hand off the transaction to the completion thread now, so
195 * clear the flag here.
196 */
197 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
198 return 0;
199}
200
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000201/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000202 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000203 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000204STATIC int
Christoph Hellwige3728432016-09-19 11:26:41 +1000205__xfs_setfilesize(
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100206 struct xfs_inode *ip,
207 struct xfs_trans *tp,
208 xfs_off_t offset,
209 size_t size)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000210{
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000211 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000212
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000213 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100214 isize = xfs_new_eof(ip, offset + size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000215 if (!isize) {
216 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig4906e212015-06-04 13:47:56 +1000217 xfs_trans_cancel(tp);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000218 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000219 }
220
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100221 trace_xfs_setfilesize(ip, offset, size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000222
223 ip->i_d.di_size = isize;
224 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
225 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
226
Christoph Hellwig70393312015-06-04 13:48:08 +1000227 return xfs_trans_commit(tp);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000228}
229
Christoph Hellwige3728432016-09-19 11:26:41 +1000230int
231xfs_setfilesize(
232 struct xfs_inode *ip,
233 xfs_off_t offset,
234 size_t size)
235{
236 struct xfs_mount *mp = ip->i_mount;
237 struct xfs_trans *tp;
238 int error;
239
240 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
241 if (error)
242 return error;
243
244 return __xfs_setfilesize(ip, tp, offset, size);
245}
246
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100247STATIC int
248xfs_setfilesize_ioend(
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000249 struct xfs_ioend *ioend,
250 int error)
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100251{
252 struct xfs_inode *ip = XFS_I(ioend->io_inode);
253 struct xfs_trans *tp = ioend->io_append_trans;
254
255 /*
256 * The transaction may have been allocated in the I/O submission thread,
257 * thus we need to mark ourselves as being in a transaction manually.
258 * Similarly for freeze protection.
259 */
260 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Oleg Nesterovbee91822015-07-19 23:48:20 +0200261 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100262
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100263 /* we abort the update if there was an IO error */
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000264 if (error) {
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100265 xfs_trans_cancel(tp);
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000266 return error;
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100267 }
268
Christoph Hellwige3728432016-09-19 11:26:41 +1000269 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100270}
271
Christoph Hellwig0829c362005-09-02 16:58:49 +1000272/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000273 * IO write completion.
274 */
275STATIC void
276xfs_end_io(
277 struct work_struct *work)
278{
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000279 struct xfs_ioend *ioend =
280 container_of(work, struct xfs_ioend, io_work);
281 struct xfs_inode *ip = XFS_I(ioend->io_inode);
282 int error = ioend->io_bio->bi_error;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000283
Brian Fosteraf055e32016-02-08 15:00:02 +1100284 /*
285 * Set an error if the mount has shut down and proceed with end I/O
286 * processing so it can perform whatever cleanups are necessary.
287 */
288 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000289 error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000290
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000291 /*
Darrick J. Wong43caeb12016-10-03 09:11:35 -0700292 * For a CoW extent, we need to move the mapping from the CoW fork
293 * to the data fork. If instead an error happened, just dump the
294 * new blocks.
295 */
296 if (ioend->io_type == XFS_IO_COW) {
297 if (error)
298 goto done;
299 if (ioend->io_bio->bi_error) {
300 error = xfs_reflink_cancel_cow_range(ip,
301 ioend->io_offset, ioend->io_size);
302 goto done;
303 }
304 error = xfs_reflink_end_cow(ip, ioend->io_offset,
305 ioend->io_size);
306 if (error)
307 goto done;
308 }
309
310 /*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000311 * For unwritten extents we need to issue transactions to convert a
312 * range to normal written extens after the data I/O has finished.
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100313 * Detecting and handling completion IO errors is done individually
314 * for each case as different cleanup operations need to be performed
315 * on error.
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000316 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500317 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000318 if (error)
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100319 goto done;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000320 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100321 ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000322 } else if (ioend->io_append_trans) {
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000323 error = xfs_setfilesize_ioend(ioend, error);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000324 } else {
Darrick J. Wong43caeb12016-10-03 09:11:35 -0700325 ASSERT(!xfs_ioend_is_append(ioend) ||
326 ioend->io_type == XFS_IO_COW);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000327 }
328
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000329done:
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000330 xfs_destroy_ioend(ioend, error);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000331}
332
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000333STATIC void
334xfs_end_bio(
335 struct bio *bio)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000336{
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000337 struct xfs_ioend *ioend = bio->bi_private;
338 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000339
Darrick J. Wong43caeb12016-10-03 09:11:35 -0700340 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000341 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
342 else if (ioend->io_append_trans)
343 queue_work(mp->m_data_workqueue, &ioend->io_work);
344 else
345 xfs_destroy_ioend(ioend, bio->bi_error);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000346}
347
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348STATIC int
349xfs_map_blocks(
350 struct inode *inode,
351 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000352 struct xfs_bmbt_irec *imap,
Dave Chinner988ef922016-02-15 17:20:50 +1100353 int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000355 struct xfs_inode *ip = XFS_I(inode);
356 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000357 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000358 xfs_fileoff_t offset_fsb, end_fsb;
359 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000360 int bmapi_flags = XFS_BMAPI_ENTIRE;
361 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
Christoph Hellwiga206c812010-12-10 08:42:20 +0000363 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000364 return -EIO;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000365
Darrick J. Wongef473662016-10-03 09:11:34 -0700366 ASSERT(type != XFS_IO_COW);
Alain Renaud0d882a32012-05-22 15:56:21 -0500367 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000368 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000369
Dave Chinner988ef922016-02-15 17:20:50 +1100370 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000371 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
372 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000373 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000374
Dave Chinnerd2c28192012-06-08 15:44:53 +1000375 if (offset + count > mp->m_super->s_maxbytes)
376 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000377 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
378 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000379 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
380 imap, &nimaps, bmapi_flags);
Darrick J. Wongef473662016-10-03 09:11:34 -0700381 /*
382 * Truncate an overwrite extent if there's a pending CoW
383 * reservation before the end of this extent. This forces us
384 * to come back to writepage to take care of the CoW.
385 */
386 if (nimaps && type == XFS_IO_OVERWRITE)
387 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000388 xfs_iunlock(ip, XFS_ILOCK_SHARED);
389
Christoph Hellwiga206c812010-12-10 08:42:20 +0000390 if (error)
Dave Chinner24513372014-06-25 14:58:08 +1000391 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000392
Alain Renaud0d882a32012-05-22 15:56:21 -0500393 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000394 (!nimaps || isnullstartblock(imap->br_startblock))) {
Darrick J. Wong60b49842016-10-03 09:11:34 -0700395 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
396 imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000397 if (!error)
Darrick J. Wongef473662016-10-03 09:11:34 -0700398 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Dave Chinner24513372014-06-25 14:58:08 +1000399 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000400 }
401
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000402#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500403 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000404 ASSERT(nimaps);
405 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
406 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
407 }
408#endif
409 if (nimaps)
410 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
411 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412}
413
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100414STATIC bool
Christoph Hellwig558e6892010-04-28 12:28:58 +0000415xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000416 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000417 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000418 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000420 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000421
Christoph Hellwig558e6892010-04-28 12:28:58 +0000422 return offset >= imap->br_startoff &&
423 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100426STATIC void
427xfs_start_buffer_writeback(
428 struct buffer_head *bh)
429{
430 ASSERT(buffer_mapped(bh));
431 ASSERT(buffer_locked(bh));
432 ASSERT(!buffer_delay(bh));
433 ASSERT(!buffer_unwritten(bh));
434
435 mark_buffer_async_write(bh);
436 set_buffer_uptodate(bh);
437 clear_buffer_dirty(bh);
438}
439
440STATIC void
441xfs_start_page_writeback(
442 struct page *page,
Dave Chinnere10de372016-02-15 17:23:12 +1100443 int clear_dirty)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100444{
445 ASSERT(PageLocked(page));
446 ASSERT(!PageWriteback(page));
Dave Chinner0d085a52014-09-23 15:36:27 +1000447
448 /*
449 * if the page was not fully cleaned, we need to ensure that the higher
450 * layers come back to it correctly. That means we need to keep the page
451 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
452 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
453 * write this page in this writeback sweep will be made.
454 */
455 if (clear_dirty) {
David Chinner92132022006-12-21 10:24:01 +1100456 clear_page_dirty_for_io(page);
Dave Chinner0d085a52014-09-23 15:36:27 +1000457 set_page_writeback(page);
458 } else
459 set_page_writeback_keepwrite(page);
460
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100461 unlock_page(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100462}
463
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000464static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100465{
466 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
467}
468
469/*
Dave Chinnerbb187822016-04-06 08:11:25 +1000470 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
471 * it, and we submit that bio. The ioend may be used for multiple bio
472 * submissions, so we only want to allocate an append transaction for the ioend
473 * once. In the case of multiple bio submission, each bio will take an IO
474 * reference to the ioend to ensure that the ioend completion is only done once
475 * all bios have been submitted and the ioend is really done.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100476 *
477 * If @fail is non-zero, it means that we have a situation where some part of
478 * the submission process has failed after we have marked paged for writeback
Dave Chinnerbb187822016-04-06 08:11:25 +1000479 * and unlocked them. In this situation, we need to fail the bio and ioend
480 * rather than submit it to IO. This typically only happens on a filesystem
481 * shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100482 */
Dave Chinnere10de372016-02-15 17:23:12 +1100483STATIC int
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100484xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000485 struct writeback_control *wbc,
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000486 struct xfs_ioend *ioend,
Dave Chinnere10de372016-02-15 17:23:12 +1100487 int status)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100488{
Dave Chinnere10de372016-02-15 17:23:12 +1100489 /* Reserve log space if we might write beyond the on-disk inode size. */
490 if (!status &&
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000491 ioend->io_type != XFS_IO_UNWRITTEN &&
Dave Chinnerbb187822016-04-06 08:11:25 +1000492 xfs_ioend_is_append(ioend) &&
493 !ioend->io_append_trans)
Dave Chinnere10de372016-02-15 17:23:12 +1100494 status = xfs_setfilesize_trans_alloc(ioend);
Dave Chinnerbb187822016-04-06 08:11:25 +1000495
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000496 ioend->io_bio->bi_private = ioend;
497 ioend->io_bio->bi_end_io = xfs_end_bio;
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600498 ioend->io_bio->bi_opf = REQ_OP_WRITE;
499 if (wbc->sync_mode == WB_SYNC_ALL)
500 ioend->io_bio->bi_opf |= REQ_SYNC;
501
Dave Chinnere10de372016-02-15 17:23:12 +1100502 /*
503 * If we are failing the IO now, just mark the ioend with an
504 * error and finish it. This will run IO completion immediately
505 * as there is only one reference to the ioend at this point in
506 * time.
507 */
508 if (status) {
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000509 ioend->io_bio->bi_error = status;
510 bio_endio(ioend->io_bio);
Dave Chinnere10de372016-02-15 17:23:12 +1100511 return status;
512 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100513
Mike Christie4e49ea42016-06-05 14:31:41 -0500514 submit_bio(ioend->io_bio);
Dave Chinnere10de372016-02-15 17:23:12 +1100515 return 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100516}
517
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000518static void
519xfs_init_bio_from_bh(
520 struct bio *bio,
521 struct buffer_head *bh)
522{
523 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
524 bio->bi_bdev = bh->b_bdev;
525}
526
527static struct xfs_ioend *
528xfs_alloc_ioend(
529 struct inode *inode,
530 unsigned int type,
531 xfs_off_t offset,
532 struct buffer_head *bh)
533{
534 struct xfs_ioend *ioend;
535 struct bio *bio;
536
537 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
538 xfs_init_bio_from_bh(bio, bh);
539
540 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
541 INIT_LIST_HEAD(&ioend->io_list);
542 ioend->io_type = type;
543 ioend->io_inode = inode;
544 ioend->io_size = 0;
545 ioend->io_offset = offset;
546 INIT_WORK(&ioend->io_work, xfs_end_io);
547 ioend->io_append_trans = NULL;
548 ioend->io_bio = bio;
549 return ioend;
550}
551
552/*
553 * Allocate a new bio, and chain the old bio to the new one.
554 *
555 * Note that we have to do perform the chaining in this unintuitive order
556 * so that the bi_private linkage is set up in the right direction for the
557 * traversal in xfs_destroy_ioend().
558 */
559static void
560xfs_chain_bio(
561 struct xfs_ioend *ioend,
562 struct writeback_control *wbc,
563 struct buffer_head *bh)
564{
565 struct bio *new;
566
567 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
568 xfs_init_bio_from_bh(new, bh);
569
570 bio_chain(ioend->io_bio, new);
571 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600572 ioend->io_bio->bi_opf = REQ_OP_WRITE;
573 if (wbc->sync_mode == WB_SYNC_ALL)
574 ioend->io_bio->bi_opf |= REQ_SYNC;
Mike Christie4e49ea42016-06-05 14:31:41 -0500575 submit_bio(ioend->io_bio);
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000576 ioend->io_bio = new;
577}
578
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100579/*
580 * Test to see if we've been building up a completion structure for
581 * earlier buffers -- if so, we try to append to this ioend if we
582 * can, otherwise we finish off any current ioend and start another.
Dave Chinnere10de372016-02-15 17:23:12 +1100583 * Return the ioend we finished off so that the caller can submit it
584 * once it has finished processing the dirty page.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100585 */
586STATIC void
587xfs_add_to_ioend(
588 struct inode *inode,
589 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100590 xfs_off_t offset,
Dave Chinnere10de372016-02-15 17:23:12 +1100591 struct xfs_writepage_ctx *wpc,
Dave Chinnerbb187822016-04-06 08:11:25 +1000592 struct writeback_control *wbc,
Dave Chinnere10de372016-02-15 17:23:12 +1100593 struct list_head *iolist)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100594{
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100595 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
Darrick J. Wong0df61da2016-03-07 09:32:14 +1100596 bh->b_blocknr != wpc->last_block + 1 ||
597 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
Dave Chinnere10de372016-02-15 17:23:12 +1100598 if (wpc->ioend)
599 list_add(&wpc->ioend->io_list, iolist);
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000600 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100601 }
602
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000603 /*
604 * If the buffer doesn't fit into the bio we need to allocate a new
605 * one. This shouldn't happen more than once for a given buffer.
606 */
607 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
608 xfs_chain_bio(wpc->ioend, wbc, bh);
Dave Chinnerbb187822016-04-06 08:11:25 +1000609
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100610 wpc->ioend->io_size += bh->b_size;
611 wpc->last_block = bh->b_blocknr;
Dave Chinnere10de372016-02-15 17:23:12 +1100612 xfs_start_buffer_writeback(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100613}
614
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100616xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000617 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100618 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000619 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000620 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100621{
622 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000623 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000624 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
625 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100626
Christoph Hellwig207d0412010-04-28 12:28:56 +0000627 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
628 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100629
Christoph Hellwige5131822010-04-28 12:28:55 +0000630 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000631 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100632
Christoph Hellwig046f1682010-04-28 12:28:52 +0000633 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100634
635 bh->b_blocknr = bn;
636 set_buffer_mapped(bh);
637}
638
639STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000641 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000643 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000644 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000646 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
647 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
Christoph Hellwig207d0412010-04-28 12:28:56 +0000649 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 set_buffer_mapped(bh);
651 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100652 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653}
654
655/*
Dave Chinnera49935f2014-03-07 16:19:14 +1100656 * Test if a given page contains at least one buffer of a given @type.
657 * If @check_all_buffers is true, then we walk all the buffers in the page to
658 * try to find one of the type passed in. If it is not set, then the caller only
659 * needs to check the first buffer on the page for a match.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 */
Dave Chinnera49935f2014-03-07 16:19:14 +1100661STATIC bool
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000662xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100663 struct page *page,
Dave Chinnera49935f2014-03-07 16:19:14 +1100664 unsigned int type,
665 bool check_all_buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
Dave Chinnera49935f2014-03-07 16:19:14 +1100667 struct buffer_head *bh;
668 struct buffer_head *head;
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 if (PageWriteback(page))
Dave Chinnera49935f2014-03-07 16:19:14 +1100671 return false;
672 if (!page->mapping)
673 return false;
674 if (!page_has_buffers(page))
675 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Dave Chinnera49935f2014-03-07 16:19:14 +1100677 bh = head = page_buffers(page);
678 do {
679 if (buffer_unwritten(bh)) {
680 if (type == XFS_IO_UNWRITTEN)
681 return true;
682 } else if (buffer_delay(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100683 if (type == XFS_IO_DELALLOC)
Dave Chinnera49935f2014-03-07 16:19:14 +1100684 return true;
685 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100686 if (type == XFS_IO_OVERWRITE)
Dave Chinnera49935f2014-03-07 16:19:14 +1100687 return true;
688 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Dave Chinnera49935f2014-03-07 16:19:14 +1100690 /* If we are only checking the first buffer, we are done now. */
691 if (!check_all_buffers)
692 break;
693 } while ((bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
Dave Chinnera49935f2014-03-07 16:19:14 +1100695 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
Dave Chinner3ed3a432010-03-05 02:00:42 +0000698STATIC void
699xfs_vm_invalidatepage(
700 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400701 unsigned int offset,
702 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000703{
Lukas Czerner34097df2013-05-21 23:58:01 -0400704 trace_xfs_invalidatepage(page->mapping->host, page, offset,
705 length);
706 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000707}
708
709/*
710 * If the page has delalloc buffers on it, we need to punch them out before we
711 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
712 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
713 * is done on that same region - the delalloc extent is returned when none is
714 * supposed to be there.
715 *
716 * We prevent this by truncating away the delalloc regions on the page before
717 * invalidating it. Because they are delalloc, we can do this without needing a
718 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
719 * truncation without a transaction as there is no space left for block
720 * reservation (typically why we see a ENOSPC in writeback).
721 *
722 * This is not a performance critical path, so for now just do the punching a
723 * buffer head at a time.
724 */
725STATIC void
726xfs_aops_discard_page(
727 struct page *page)
728{
729 struct inode *inode = page->mapping->host;
730 struct xfs_inode *ip = XFS_I(inode);
731 struct buffer_head *bh, *head;
732 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000733
Dave Chinnera49935f2014-03-07 16:19:14 +1100734 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000735 goto out_invalidate;
736
Dave Chinnere8c37532010-03-15 02:36:35 +0000737 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
738 goto out_invalidate;
739
Dave Chinner4f107002011-03-07 10:00:35 +1100740 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000741 "page discard on page %p, inode 0x%llx, offset %llu.",
742 page, ip->i_ino, offset);
743
744 xfs_ilock(ip, XFS_ILOCK_EXCL);
745 bh = head = page_buffers(page);
746 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000747 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100748 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000749
750 if (!buffer_delay(bh))
751 goto next_buffer;
752
Dave Chinnerc726de42010-11-30 15:14:39 +1100753 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
754 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000755 if (error) {
756 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000757 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100758 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000759 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000760 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000761 break;
762 }
763next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100764 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000765
766 } while ((bh = bh->b_this_page) != head);
767
768 xfs_iunlock(ip, XFS_ILOCK_EXCL);
769out_invalidate:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300770 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000771 return;
772}
773
Darrick J. Wongef473662016-10-03 09:11:34 -0700774static int
775xfs_map_cow(
776 struct xfs_writepage_ctx *wpc,
777 struct inode *inode,
778 loff_t offset,
779 unsigned int *new_type)
780{
781 struct xfs_inode *ip = XFS_I(inode);
782 struct xfs_bmbt_irec imap;
783 bool is_cow = false, need_alloc = false;
784 int error;
785
786 /*
787 * If we already have a valid COW mapping keep using it.
788 */
789 if (wpc->io_type == XFS_IO_COW) {
790 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
791 if (wpc->imap_valid) {
792 *new_type = XFS_IO_COW;
793 return 0;
794 }
795 }
796
797 /*
798 * Else we need to check if there is a COW mapping at this offset.
799 */
800 xfs_ilock(ip, XFS_ILOCK_SHARED);
801 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap, &need_alloc);
802 xfs_iunlock(ip, XFS_ILOCK_SHARED);
803
804 if (!is_cow)
805 return 0;
806
807 /*
808 * And if the COW mapping has a delayed extent here we need to
809 * allocate real space for it now.
810 */
811 if (need_alloc) {
812 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
813 &imap);
814 if (error)
815 return error;
816 }
817
818 wpc->io_type = *new_type = XFS_IO_COW;
819 wpc->imap_valid = true;
820 wpc->imap = imap;
821 return 0;
822}
823
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824/*
Dave Chinnere10de372016-02-15 17:23:12 +1100825 * We implement an immediate ioend submission policy here to avoid needing to
826 * chain multiple ioends and hence nest mempool allocations which can violate
827 * forward progress guarantees we need to provide. The current ioend we are
828 * adding buffers to is cached on the writepage context, and if the new buffer
829 * does not append to the cached ioend it will create a new ioend and cache that
830 * instead.
831 *
832 * If a new ioend is created and cached, the old ioend is returned and queued
833 * locally for submission once the entire page is processed or an error has been
834 * detected. While ioends are submitted immediately after they are completed,
835 * batching optimisations are provided by higher level block plugging.
836 *
837 * At the end of a writeback pass, there will be a cached ioend remaining on the
838 * writepage context that the caller will need to submit.
839 */
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100840static int
841xfs_writepage_map(
842 struct xfs_writepage_ctx *wpc,
Dave Chinnere10de372016-02-15 17:23:12 +1100843 struct writeback_control *wbc,
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100844 struct inode *inode,
845 struct page *page,
846 loff_t offset,
847 __uint64_t end_offset)
848{
Dave Chinnere10de372016-02-15 17:23:12 +1100849 LIST_HEAD(submit_list);
850 struct xfs_ioend *ioend, *next;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100851 struct buffer_head *bh, *head;
852 ssize_t len = 1 << inode->i_blkbits;
853 int error = 0;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100854 int count = 0;
Dave Chinnere10de372016-02-15 17:23:12 +1100855 int uptodate = 1;
Darrick J. Wongef473662016-10-03 09:11:34 -0700856 unsigned int new_type;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100857
858 bh = head = page_buffers(page);
859 offset = page_offset(page);
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100860 do {
861 if (offset >= end_offset)
862 break;
863 if (!buffer_uptodate(bh))
864 uptodate = 0;
865
866 /*
867 * set_page_dirty dirties all buffers in a page, independent
868 * of their state. The dirty state however is entirely
869 * meaningless for holes (!mapped && uptodate), so skip
870 * buffers covering holes here.
871 */
872 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
873 wpc->imap_valid = false;
874 continue;
875 }
876
Darrick J. Wongef473662016-10-03 09:11:34 -0700877 if (buffer_unwritten(bh))
878 new_type = XFS_IO_UNWRITTEN;
879 else if (buffer_delay(bh))
880 new_type = XFS_IO_DELALLOC;
881 else if (buffer_uptodate(bh))
882 new_type = XFS_IO_OVERWRITE;
883 else {
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100884 if (PageUptodate(page))
885 ASSERT(buffer_mapped(bh));
886 /*
887 * This buffer is not uptodate and will not be
888 * written to disk. Ensure that we will put any
889 * subsequent writeable buffers into a new
890 * ioend.
891 */
892 wpc->imap_valid = false;
893 continue;
894 }
895
Darrick J. Wongef473662016-10-03 09:11:34 -0700896 if (xfs_is_reflink_inode(XFS_I(inode))) {
897 error = xfs_map_cow(wpc, inode, offset, &new_type);
898 if (error)
899 goto out;
900 }
901
902 if (wpc->io_type != new_type) {
903 wpc->io_type = new_type;
904 wpc->imap_valid = false;
905 }
906
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100907 if (wpc->imap_valid)
908 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
909 offset);
910 if (!wpc->imap_valid) {
911 error = xfs_map_blocks(inode, offset, &wpc->imap,
912 wpc->io_type);
913 if (error)
Dave Chinnere10de372016-02-15 17:23:12 +1100914 goto out;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100915 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
916 offset);
917 }
918 if (wpc->imap_valid) {
919 lock_buffer(bh);
920 if (wpc->io_type != XFS_IO_OVERWRITE)
921 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
Dave Chinnerbb187822016-04-06 08:11:25 +1000922 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100923 count++;
924 }
925
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100926 } while (offset += len, ((bh = bh->b_this_page) != head));
927
928 if (uptodate && bh == head)
929 SetPageUptodate(page);
930
Dave Chinnere10de372016-02-15 17:23:12 +1100931 ASSERT(wpc->ioend || list_empty(&submit_list));
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100932
Dave Chinnere10de372016-02-15 17:23:12 +1100933out:
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100934 /*
Dave Chinnere10de372016-02-15 17:23:12 +1100935 * On error, we have to fail the ioend here because we have locked
936 * buffers in the ioend. If we don't do this, we'll deadlock
937 * invalidating the page as that tries to lock the buffers on the page.
938 * Also, because we may have set pages under writeback, we have to make
939 * sure we run IO completion to mark the error state of the IO
940 * appropriately, so we can't cancel the ioend directly here. That means
941 * we have to mark this page as under writeback if we included any
942 * buffers from it in the ioend chain so that completion treats it
943 * correctly.
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100944 *
Dave Chinnere10de372016-02-15 17:23:12 +1100945 * If we didn't include the page in the ioend, the on error we can
946 * simply discard and unlock it as there are no other users of the page
947 * or it's buffers right now. The caller will still need to trigger
948 * submission of outstanding ioends on the writepage context so they are
949 * treated correctly on error.
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100950 */
Dave Chinnere10de372016-02-15 17:23:12 +1100951 if (count) {
952 xfs_start_page_writeback(page, !error);
953
954 /*
955 * Preserve the original error if there was one, otherwise catch
956 * submission errors here and propagate into subsequent ioend
957 * submissions.
958 */
959 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
960 int error2;
961
962 list_del_init(&ioend->io_list);
963 error2 = xfs_submit_ioend(wbc, ioend, error);
964 if (error2 && !error)
965 error = error2;
966 }
967 } else if (error) {
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100968 xfs_aops_discard_page(page);
969 ClearPageUptodate(page);
970 unlock_page(page);
Dave Chinnere10de372016-02-15 17:23:12 +1100971 } else {
972 /*
973 * We can end up here with no error and nothing to write if we
974 * race with a partial page truncate on a sub-page block sized
975 * filesystem. In that case we need to mark the page clean.
976 */
977 xfs_start_page_writeback(page, 1);
978 end_page_writeback(page);
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100979 }
Dave Chinnere10de372016-02-15 17:23:12 +1100980
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100981 mapping_set_error(page->mapping, error);
982 return error;
983}
984
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000986 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000988 * For delalloc space on the page we need to allocate space and flush it.
989 * For unwritten space on the page we need to start the conversion to
990 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000991 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993STATIC int
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100994xfs_do_writepage(
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000995 struct page *page,
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100996 struct writeback_control *wbc,
997 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100999 struct xfs_writepage_ctx *wpc = data;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001000 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 loff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 __uint64_t end_offset;
Dave Chinnerad689722016-02-15 17:21:31 +11001003 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
Lukas Czerner34097df2013-05-21 23:58:01 -04001005 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001006
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001007 ASSERT(page_has_buffers(page));
1008
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001009 /*
1010 * Refuse to write the page out if we are called from reclaim context.
1011 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -04001012 * This avoids stack overflows when called from deeply used stacks in
1013 * random callers for direct reclaim or memcg reclaim. We explicitly
1014 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001015 *
Mel Gorman94054fa2011-10-31 17:07:45 -07001016 * This should never happen except in the case of a VM regression so
1017 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001018 */
Mel Gorman94054fa2011-10-31 17:07:45 -07001019 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1020 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001021 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001022
1023 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +02001024 * Given that we do not allow direct reclaim to call us, we should
1025 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001026 */
Christoph Hellwig448011e2014-06-06 16:05:15 +10001027 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001028 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001029
Jie Liu8695d272014-05-20 08:24:26 +10001030 /*
Dave Chinnerad689722016-02-15 17:21:31 +11001031 * Is this page beyond the end of the file?
1032 *
Jie Liu8695d272014-05-20 08:24:26 +10001033 * The page index is less than the end_index, adjust the end_offset
1034 * to the highest offset that this page should represent.
1035 * -----------------------------------------------------
1036 * | file mapping | <EOF> |
1037 * -----------------------------------------------------
1038 * | Page ... | Page N-2 | Page N-1 | Page N | |
1039 * ^--------------------------------^----------|--------
1040 * | desired writeback range | see else |
1041 * ---------------------------------^------------------|
1042 */
Dave Chinnerad689722016-02-15 17:21:31 +11001043 offset = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001044 end_index = offset >> PAGE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +10001045 if (page->index < end_index)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001046 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +10001047 else {
1048 /*
1049 * Check whether the page to write out is beyond or straddles
1050 * i_size or not.
1051 * -------------------------------------------------------
1052 * | file mapping | <EOF> |
1053 * -------------------------------------------------------
1054 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1055 * ^--------------------------------^-----------|---------
1056 * | | Straddles |
1057 * ---------------------------------^-----------|--------|
1058 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001059 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001060
1061 /*
Jan Karaff9a28f2013-03-14 14:30:54 +01001062 * Skip the page if it is fully outside i_size, e.g. due to a
1063 * truncate operation that is in progress. We must redirty the
1064 * page so that reclaim stops reclaiming it. Otherwise
1065 * xfs_vm_releasepage() is called on it and gets confused.
Jie Liu8695d272014-05-20 08:24:26 +10001066 *
1067 * Note that the end_index is unsigned long, it would overflow
1068 * if the given offset is greater than 16TB on 32-bit system
1069 * and if we do check the page is fully outside i_size or not
1070 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1071 * will be evaluated to 0. Hence this page will be redirtied
1072 * and be written out repeatedly which would result in an
1073 * infinite loop, the user program that perform this operation
1074 * will hang. Instead, we can verify this situation by checking
1075 * if the page to write is totally beyond the i_size or if it's
1076 * offset is just equal to the EOF.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001077 */
Jie Liu8695d272014-05-20 08:24:26 +10001078 if (page->index > end_index ||
1079 (page->index == end_index && offset_into_page == 0))
Jan Karaff9a28f2013-03-14 14:30:54 +01001080 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001081
1082 /*
1083 * The page straddles i_size. It must be zeroed out on each
1084 * and every writepage invocation because it may be mmapped.
1085 * "A file is mapped in multiples of the page size. For a file
Jie Liu8695d272014-05-20 08:24:26 +10001086 * that is not a multiple of the page size, the remaining
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001087 * memory is zeroed when mapped, and writes to that region are
1088 * not written out to the file."
1089 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001090 zero_user_segment(page, offset_into_page, PAGE_SIZE);
Jie Liu8695d272014-05-20 08:24:26 +10001091
1092 /* Adjust the end_offset to the end of file */
1093 end_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095
Dave Chinnere10de372016-02-15 17:23:12 +11001096 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
Nathan Scottf51623b2006-03-14 13:26:27 +11001097
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001098redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001099 redirty_page_for_writepage(wbc, page);
1100 unlock_page(page);
1101 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001102}
1103
Nathan Scott7d4fb402006-06-09 15:27:16 +10001104STATIC int
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001105xfs_vm_writepage(
1106 struct page *page,
1107 struct writeback_control *wbc)
1108{
1109 struct xfs_writepage_ctx wpc = {
1110 .io_type = XFS_IO_INVALID,
1111 };
1112 int ret;
1113
1114 ret = xfs_do_writepage(page, wbc, &wpc);
Dave Chinnere10de372016-02-15 17:23:12 +11001115 if (wpc.ioend)
1116 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1117 return ret;
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001118}
1119
1120STATIC int
Nathan Scott7d4fb402006-06-09 15:27:16 +10001121xfs_vm_writepages(
1122 struct address_space *mapping,
1123 struct writeback_control *wbc)
1124{
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001125 struct xfs_writepage_ctx wpc = {
1126 .io_type = XFS_IO_INVALID,
1127 };
1128 int ret;
1129
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001130 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001131 if (dax_mapping(mapping))
1132 return dax_writeback_mapping_range(mapping,
1133 xfs_find_bdev_for_inode(mapping->host), wbc);
1134
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001135 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
Dave Chinnere10de372016-02-15 17:23:12 +11001136 if (wpc.ioend)
1137 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1138 return ret;
Nathan Scott7d4fb402006-06-09 15:27:16 +10001139}
1140
Nathan Scottf51623b2006-03-14 13:26:27 +11001141/*
1142 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001143 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001144 * have buffer heads in this call.
1145 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001146 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001147 */
1148STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001149xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001150 struct page *page,
1151 gfp_t gfp_mask)
1152{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001153 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001154
Lukas Czerner34097df2013-05-21 23:58:01 -04001155 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001156
Brian Foster99579cc2016-07-22 09:50:38 +10001157 /*
1158 * mm accommodates an old ext3 case where clean pages might not have had
1159 * the dirty bit cleared. Thus, it can send actual dirty pages to
1160 * ->releasepage() via shrink_active_list(). Conversely,
1161 * block_invalidatepage() can send pages that are still marked dirty
1162 * but otherwise have invalidated buffers.
1163 *
1164 * We've historically freed buffers on the latter. Instead, quietly
1165 * filter out all dirty pages to avoid spurious buffer state warnings.
1166 * This can likely be removed once shrink_active_list() is fixed.
1167 */
1168 if (PageDirty(page))
1169 return 0;
1170
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001171 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001172
Christoph Hellwig448011e2014-06-06 16:05:15 +10001173 if (WARN_ON_ONCE(delalloc))
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001174 return 0;
Christoph Hellwig448011e2014-06-06 16:05:15 +10001175 if (WARN_ON_ONCE(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001176 return 0;
1177
Nathan Scottf51623b2006-03-14 13:26:27 +11001178 return try_to_free_buffers(page);
1179}
1180
Dave Chinnera7193702015-04-16 21:57:48 +10001181/*
Christoph Hellwig273dda72016-02-08 14:40:51 +11001182 * When we map a DIO buffer, we may need to pass flags to
1183 * xfs_end_io_direct_write to tell it what kind of write IO we are doing.
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001184 *
1185 * Note that for DIO, an IO to the highest supported file block offset (i.e.
1186 * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64
1187 * bit variable. Hence if we see this overflow, we have to assume that the IO is
1188 * extending the file size. We won't know for sure until IO completion is run
1189 * and the actual max write offset is communicated to the IO completion
1190 * routine.
Dave Chinnera7193702015-04-16 21:57:48 +10001191 */
1192static void
1193xfs_map_direct(
1194 struct inode *inode,
1195 struct buffer_head *bh_result,
1196 struct xfs_bmbt_irec *imap,
Darrick J. Wong0613f162016-10-03 09:11:37 -07001197 xfs_off_t offset,
1198 bool is_cow)
Dave Chinnera7193702015-04-16 21:57:48 +10001199{
Christoph Hellwig273dda72016-02-08 14:40:51 +11001200 uintptr_t *flags = (uintptr_t *)&bh_result->b_private;
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001201 xfs_off_t size = bh_result->b_size;
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001202
Christoph Hellwig273dda72016-02-08 14:40:51 +11001203 trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size,
Darrick J. Wong0613f162016-10-03 09:11:37 -07001204 ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : is_cow ? XFS_IO_COW :
1205 XFS_IO_OVERWRITE, imap);
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001206
Christoph Hellwig273dda72016-02-08 14:40:51 +11001207 if (ISUNWRITTEN(imap)) {
1208 *flags |= XFS_DIO_FLAG_UNWRITTEN;
Dave Chinnera06c2772015-04-16 22:00:00 +10001209 set_buffer_defer_completion(bh_result);
Darrick J. Wong0613f162016-10-03 09:11:37 -07001210 } else if (is_cow) {
1211 *flags |= XFS_DIO_FLAG_COW;
1212 set_buffer_defer_completion(bh_result);
1213 }
1214 if (offset + size > i_size_read(inode) || offset + size < 0) {
Christoph Hellwig273dda72016-02-08 14:40:51 +11001215 *flags |= XFS_DIO_FLAG_APPEND;
1216 set_buffer_defer_completion(bh_result);
Dave Chinnera7193702015-04-16 21:57:48 +10001217 }
1218}
1219
Dave Chinner1fdca9c2015-04-16 21:58:21 +10001220/*
1221 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1222 * is, so that we can avoid repeated get_blocks calls.
1223 *
1224 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1225 * for blocks beyond EOF must be marked new so that sub block regions can be
1226 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1227 * was just allocated or is unwritten, otherwise the callers would overwrite
1228 * existing data with zeros. Hence we have to split the mapping into a range up
1229 * to and including EOF, and a second mapping for beyond EOF.
1230 */
1231static void
1232xfs_map_trim_size(
1233 struct inode *inode,
1234 sector_t iblock,
1235 struct buffer_head *bh_result,
1236 struct xfs_bmbt_irec *imap,
1237 xfs_off_t offset,
1238 ssize_t size)
1239{
1240 xfs_off_t mapping_size;
1241
1242 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1243 mapping_size <<= inode->i_blkbits;
1244
1245 ASSERT(mapping_size > 0);
1246 if (mapping_size > size)
1247 mapping_size = size;
1248 if (offset < i_size_read(inode) &&
1249 offset + mapping_size >= i_size_read(inode)) {
1250 /* limit mapping to block that spans EOF */
1251 mapping_size = roundup_64(i_size_read(inode) - offset,
1252 1 << inode->i_blkbits);
1253 }
1254 if (mapping_size > LONG_MAX)
1255 mapping_size = LONG_MAX;
1256
1257 bh_result->b_size = mapping_size;
1258}
1259
Darrick J. Wong0613f162016-10-03 09:11:37 -07001260/* Bounce unaligned directio writes to the page cache. */
1261static int
1262xfs_bounce_unaligned_dio_write(
1263 struct xfs_inode *ip,
1264 xfs_fileoff_t offset_fsb,
1265 struct xfs_bmbt_irec *imap)
1266{
1267 struct xfs_bmbt_irec irec;
1268 xfs_fileoff_t delta;
1269 bool shared;
1270 bool x;
1271 int error;
1272
1273 irec = *imap;
1274 if (offset_fsb > irec.br_startoff) {
1275 delta = offset_fsb - irec.br_startoff;
1276 irec.br_blockcount -= delta;
1277 irec.br_startblock += delta;
1278 irec.br_startoff = offset_fsb;
1279 }
1280 error = xfs_reflink_trim_around_shared(ip, &irec, &shared, &x);
1281 if (error)
1282 return error;
1283
1284 /*
1285 * We're here because we're trying to do a directio write to a
1286 * region that isn't aligned to a filesystem block. If any part
1287 * of the extent is shared, fall back to buffered mode to handle
1288 * the RMW. This is done by returning -EREMCHG ("remote addr
1289 * changed"), which is caught further up the call stack.
1290 */
1291 if (shared) {
1292 trace_xfs_reflink_bounce_dio_write(ip, imap);
1293 return -EREMCHG;
1294 }
1295 return 0;
1296}
1297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001299__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 struct inode *inode,
1301 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 struct buffer_head *bh_result,
1303 int create,
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001304 bool direct,
1305 bool dax_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001307 struct xfs_inode *ip = XFS_I(inode);
1308 struct xfs_mount *mp = ip->i_mount;
1309 xfs_fileoff_t offset_fsb, end_fsb;
1310 int error = 0;
1311 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001312 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001313 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001314 xfs_off_t offset;
1315 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001316 int new = 0;
Darrick J. Wong0613f162016-10-03 09:11:37 -07001317 bool is_cow = false;
1318 bool need_alloc = false;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001319
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001320 BUG_ON(create && !direct);
1321
Christoph Hellwiga206c812010-12-10 08:42:20 +00001322 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001323 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001325 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001326 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1327 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001328
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001329 if (!create && offset >= i_size_read(inode))
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001330 return 0;
1331
Dave Chinner507630b2012-03-27 10:34:50 -04001332 /*
1333 * Direct I/O is usually done on preallocated files, so try getting
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001334 * a block mapping without an exclusive lock first.
Dave Chinner507630b2012-03-27 10:34:50 -04001335 */
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001336 lockmode = xfs_ilock_data_map_shared(ip);
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001337
Dave Chinnerd2c28192012-06-08 15:44:53 +10001338 ASSERT(offset <= mp->m_super->s_maxbytes);
1339 if (offset + size > mp->m_super->s_maxbytes)
1340 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001341 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1342 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1343
Darrick J. Wong0613f162016-10-03 09:11:37 -07001344 if (create && direct && xfs_is_reflink_inode(ip))
1345 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap,
1346 &need_alloc);
1347 if (!is_cow) {
1348 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1349 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1350 /*
1351 * Truncate an overwrite extent if there's a pending CoW
1352 * reservation before the end of this extent. This
1353 * forces us to come back to get_blocks to take care of
1354 * the CoW.
1355 */
1356 if (create && direct && nimaps &&
1357 imap.br_startblock != HOLESTARTBLOCK &&
1358 imap.br_startblock != DELAYSTARTBLOCK &&
1359 !ISUNWRITTEN(&imap))
1360 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb,
1361 &imap);
1362 }
1363 ASSERT(!need_alloc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001365 goto out_unlock;
1366
Dave Chinner1ca19152015-11-03 12:37:00 +11001367 /* for DAX, we convert unwritten extents directly */
Christoph Hellwiga206c812010-12-10 08:42:20 +00001368 if (create &&
1369 (!nimaps ||
1370 (imap.br_startblock == HOLESTARTBLOCK ||
Dave Chinner1ca19152015-11-03 12:37:00 +11001371 imap.br_startblock == DELAYSTARTBLOCK) ||
1372 (IS_DAX(inode) && ISUNWRITTEN(&imap)))) {
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001373 /*
1374 * xfs_iomap_write_direct() expects the shared lock. It
1375 * is unlocked on return.
1376 */
1377 if (lockmode == XFS_ILOCK_EXCL)
1378 xfs_ilock_demote(ip, lockmode);
Brian Foster009c6e82015-10-12 15:34:20 +11001379
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001380 error = xfs_iomap_write_direct(ip, offset, size,
1381 &imap, nimaps);
1382 if (error)
1383 return error;
1384 new = 1;
Dave Chinner6b698ed2015-06-04 09:18:53 +10001385
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001386 trace_xfs_get_blocks_alloc(ip, offset, size,
1387 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1388 : XFS_IO_DELALLOC, &imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001389 } else if (nimaps) {
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001390 trace_xfs_get_blocks_found(ip, offset, size,
1391 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1392 : XFS_IO_OVERWRITE, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001393 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001394 } else {
1395 trace_xfs_get_blocks_notfound(ip, offset, size);
1396 goto out_unlock;
1397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Dave Chinner1ca19152015-11-03 12:37:00 +11001399 if (IS_DAX(inode) && create) {
1400 ASSERT(!ISUNWRITTEN(&imap));
1401 /* zeroing is not needed at a higher layer */
1402 new = 0;
1403 }
1404
Dave Chinner1fdca9c2015-04-16 21:58:21 +10001405 /* trim mapping down to size requested */
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001406 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
Dave Chinner1fdca9c2015-04-16 21:58:21 +10001407
Dave Chinnera7193702015-04-16 21:57:48 +10001408 /*
1409 * For unwritten extents do not report a disk address in the buffered
1410 * read case (treat as if we're reading into a hole).
1411 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001412 if (imap.br_startblock != HOLESTARTBLOCK &&
Dave Chinnera7193702015-04-16 21:57:48 +10001413 imap.br_startblock != DELAYSTARTBLOCK &&
1414 (create || !ISUNWRITTEN(&imap))) {
Darrick J. Wong0613f162016-10-03 09:11:37 -07001415 if (create && direct && !is_cow) {
1416 error = xfs_bounce_unaligned_dio_write(ip, offset_fsb,
1417 &imap);
1418 if (error)
1419 return error;
1420 }
1421
Dave Chinnera7193702015-04-16 21:57:48 +10001422 xfs_map_buffer(inode, bh_result, &imap, offset);
1423 if (ISUNWRITTEN(&imap))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 set_buffer_unwritten(bh_result);
Dave Chinnera7193702015-04-16 21:57:48 +10001425 /* direct IO needs special help */
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001426 if (create) {
Christoph Hellwig273dda72016-02-08 14:40:51 +11001427 if (dax_fault)
1428 ASSERT(!ISUNWRITTEN(&imap));
1429 else
Darrick J. Wong0613f162016-10-03 09:11:37 -07001430 xfs_map_direct(inode, bh_result, &imap, offset,
1431 is_cow);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001432 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 }
1434
Nathan Scottc2536662006-03-29 10:44:40 +10001435 /*
1436 * If this is a realtime file, data may be on a different device.
1437 * to that pointed to from the buffer_head b_bdev currently.
1438 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001439 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440
Nathan Scottc2536662006-03-29 10:44:40 +10001441 /*
David Chinner549054a2007-02-10 18:36:35 +11001442 * If we previously allocated a block out beyond eof and we are now
1443 * coming back to use it then we will need to flag it as new even if it
1444 * has a disk address.
1445 *
1446 * With sub-block writes into unwritten extents we also need to mark
1447 * the buffer as new so that the unwritten parts of the buffer gets
1448 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 */
1450 if (create &&
1451 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001452 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001453 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001456 BUG_ON(direct && imap.br_startblock == DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001459
1460out_unlock:
1461 xfs_iunlock(ip, lockmode);
Dave Chinner24513372014-06-25 14:58:08 +10001462 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463}
1464
1465int
Nathan Scottc2536662006-03-29 10:44:40 +10001466xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 struct inode *inode,
1468 sector_t iblock,
1469 struct buffer_head *bh_result,
1470 int create)
1471{
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001472 return __xfs_get_blocks(inode, iblock, bh_result, create, false, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473}
1474
Dave Chinner6b698ed2015-06-04 09:18:53 +10001475int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001476xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 struct inode *inode,
1478 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 struct buffer_head *bh_result,
1480 int create)
1481{
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001482 return __xfs_get_blocks(inode, iblock, bh_result, create, true, false);
1483}
1484
1485int
1486xfs_get_blocks_dax_fault(
1487 struct inode *inode,
1488 sector_t iblock,
1489 struct buffer_head *bh_result,
1490 int create)
1491{
1492 return __xfs_get_blocks(inode, iblock, bh_result, create, true, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494
Christoph Hellwig273dda72016-02-08 14:40:51 +11001495/*
1496 * Complete a direct I/O write request.
1497 *
1498 * xfs_map_direct passes us some flags in the private data to tell us what to
1499 * do. If no flags are set, then the write IO is an overwrite wholly within
1500 * the existing allocated file size and so there is nothing for us to do.
1501 *
1502 * Note that in this case the completion can be called in interrupt context,
1503 * whereas if we have flags set we will always be called in task context
1504 * (i.e. from a workqueue).
1505 */
Christoph Hellwigfa8d9722016-07-20 11:38:01 +10001506int
Christoph Hellwig273dda72016-02-08 14:40:51 +11001507xfs_end_io_direct_write(
1508 struct kiocb *iocb,
Christoph Hellwig209fb872010-07-18 21:17:11 +00001509 loff_t offset,
Christoph Hellwig273dda72016-02-08 14:40:51 +11001510 ssize_t size,
1511 void *private)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001512{
Christoph Hellwig273dda72016-02-08 14:40:51 +11001513 struct inode *inode = file_inode(iocb->ki_filp);
1514 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001515 uintptr_t flags = (uintptr_t)private;
1516 int error = 0;
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001517
Christoph Hellwig273dda72016-02-08 14:40:51 +11001518 trace_xfs_end_io_direct_write(ip, offset, size);
1519
Christoph Hellwige3728432016-09-19 11:26:41 +10001520 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Christoph Hellwig273dda72016-02-08 14:40:51 +11001521 return -EIO;
1522
1523 if (size <= 0)
1524 return size;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001525
1526 /*
Christoph Hellwig273dda72016-02-08 14:40:51 +11001527 * The flags tell us whether we are doing unwritten extent conversions
Dave Chinner6dfa1b62015-04-16 21:59:34 +10001528 * or an append transaction that updates the on-disk file size. These
1529 * cases are the only cases where we should *potentially* be needing
Dave Chinnera06c2772015-04-16 22:00:00 +10001530 * to update the VFS inode size.
Christoph Hellwig273dda72016-02-08 14:40:51 +11001531 */
1532 if (flags == 0) {
1533 ASSERT(offset + size <= i_size_read(inode));
1534 return 0;
1535 }
1536
1537 /*
Dave Chinner6dfa1b62015-04-16 21:59:34 +10001538 * We need to update the in-core inode size here so that we don't end up
Dave Chinnera06c2772015-04-16 22:00:00 +10001539 * with the on-disk inode size being outside the in-core inode size. We
1540 * have no other method of updating EOF for AIO, so always do it here
1541 * if necessary.
Dave Chinnerb9d59842015-04-16 22:03:07 +10001542 *
1543 * We need to lock the test/set EOF update as we can be racing with
1544 * other IO completions here to update the EOF. Failing to serialise
1545 * here can result in EOF moving backwards and Bad Things Happen when
1546 * that occurs.
Christoph Hellwig2813d682011-12-18 20:00:12 +00001547 */
Christoph Hellwig273dda72016-02-08 14:40:51 +11001548 spin_lock(&ip->i_flags_lock);
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001549 if (offset + size > i_size_read(inode))
1550 i_size_write(inode, offset + size);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001551 spin_unlock(&ip->i_flags_lock);
Christoph Hellwig2813d682011-12-18 20:00:12 +00001552
Christoph Hellwigfeac4702016-10-11 09:03:19 +11001553 if (flags & XFS_DIO_FLAG_COW)
1554 error = xfs_reflink_end_cow(ip, offset, size);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001555 if (flags & XFS_DIO_FLAG_UNWRITTEN) {
1556 trace_xfs_end_io_direct_write_unwritten(ip, offset, size);
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001557
Christoph Hellwig273dda72016-02-08 14:40:51 +11001558 error = xfs_iomap_write_unwritten(ip, offset, size);
Darrick J. Wong0613f162016-10-03 09:11:37 -07001559 }
Darrick J. Wong0613f162016-10-03 09:11:37 -07001560 if (flags & XFS_DIO_FLAG_APPEND) {
Christoph Hellwig273dda72016-02-08 14:40:51 +11001561 trace_xfs_end_io_direct_write_append(ip, offset, size);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001562
Christoph Hellwige3728432016-09-19 11:26:41 +10001563 error = xfs_setfilesize(ip, offset, size);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001564 }
1565
Christoph Hellwig273dda72016-02-08 14:40:51 +11001566 return error;
Dave Chinner6e1ba0b2015-06-04 09:19:15 +10001567}
1568
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001570xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 struct kiocb *iocb,
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07001572 struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Jie Liu58e59852013-07-16 13:11:16 +08001574 /*
Christoph Hellwigfa8d9722016-07-20 11:38:01 +10001575 * We just need the method present so that open/fcntl allow direct I/O.
Jie Liu58e59852013-07-16 13:11:16 +08001576 */
Christoph Hellwigfa8d9722016-07-20 11:38:01 +10001577 return -EINVAL;
Nathan Scottf51623b2006-03-14 13:26:27 +11001578}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001581xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 struct address_space *mapping,
1583 sector_t block)
1584{
1585 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001586 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001588 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001589 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Darrick J. Wongdb1327b2016-10-03 09:11:36 -07001590
1591 /*
1592 * The swap code (ab-)uses ->bmap to get a block mapping and then
1593 * bypasseѕ the file system for actual I/O. We really can't allow
1594 * that on reflinks inodes, so we have to skip out here. And yes,
1595 * 0 is the magic code for a bmap error..
1596 */
1597 if (xfs_is_reflink_inode(ip)) {
1598 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1599 return 0;
1600 }
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001601 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001602 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001603 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604}
1605
1606STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001607xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 struct file *unused,
1609 struct page *page)
1610{
Dave Chinner121e2132016-01-08 11:28:35 +11001611 trace_xfs_vm_readpage(page->mapping->host, 1);
Nathan Scottc2536662006-03-29 10:44:40 +10001612 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613}
1614
1615STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001616xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 struct file *unused,
1618 struct address_space *mapping,
1619 struct list_head *pages,
1620 unsigned nr_pages)
1621{
Dave Chinner121e2132016-01-08 11:28:35 +11001622 trace_xfs_vm_readpages(mapping->host, nr_pages);
Nathan Scottc2536662006-03-29 10:44:40 +10001623 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624}
1625
Dave Chinner22e757a2014-09-02 12:12:51 +10001626/*
1627 * This is basically a copy of __set_page_dirty_buffers() with one
1628 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1629 * dirty, we'll never be able to clean them because we don't write buffers
1630 * beyond EOF, and that means we can't invalidate pages that span EOF
1631 * that have been marked dirty. Further, the dirty state can leak into
1632 * the file interior if the file is extended, resulting in all sorts of
1633 * bad things happening as the state does not match the underlying data.
1634 *
1635 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1636 * this only exist because of bufferheads and how the generic code manages them.
1637 */
1638STATIC int
1639xfs_vm_set_page_dirty(
1640 struct page *page)
1641{
1642 struct address_space *mapping = page->mapping;
1643 struct inode *inode = mapping->host;
1644 loff_t end_offset;
1645 loff_t offset;
1646 int newly_dirty;
1647
1648 if (unlikely(!mapping))
1649 return !TestSetPageDirty(page);
1650
1651 end_offset = i_size_read(inode);
1652 offset = page_offset(page);
1653
1654 spin_lock(&mapping->private_lock);
1655 if (page_has_buffers(page)) {
1656 struct buffer_head *head = page_buffers(page);
1657 struct buffer_head *bh = head;
1658
1659 do {
1660 if (offset < end_offset)
1661 set_buffer_dirty(bh);
1662 bh = bh->b_this_page;
1663 offset += 1 << inode->i_blkbits;
1664 } while (bh != head);
1665 }
Greg Thelenc4843a72015-05-22 17:13:16 -04001666 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001667 * Lock out page->mem_cgroup migration to keep PageDirty
1668 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -04001669 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001670 lock_page_memcg(page);
Dave Chinner22e757a2014-09-02 12:12:51 +10001671 newly_dirty = !TestSetPageDirty(page);
1672 spin_unlock(&mapping->private_lock);
1673
1674 if (newly_dirty) {
1675 /* sigh - __set_page_dirty() is static, so copy it here, too */
1676 unsigned long flags;
1677
1678 spin_lock_irqsave(&mapping->tree_lock, flags);
1679 if (page->mapping) { /* Race with truncate? */
1680 WARN_ON_ONCE(!PageUptodate(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -07001681 account_page_dirtied(page, mapping);
Dave Chinner22e757a2014-09-02 12:12:51 +10001682 radix_tree_tag_set(&mapping->page_tree,
1683 page_index(page), PAGECACHE_TAG_DIRTY);
1684 }
1685 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Dave Chinner22e757a2014-09-02 12:12:51 +10001686 }
Johannes Weiner62cccb82016-03-15 14:57:22 -07001687 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04001688 if (newly_dirty)
1689 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Dave Chinner22e757a2014-09-02 12:12:51 +10001690 return newly_dirty;
1691}
1692
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001693const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001694 .readpage = xfs_vm_readpage,
1695 .readpages = xfs_vm_readpages,
1696 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001697 .writepages = xfs_vm_writepages,
Dave Chinner22e757a2014-09-02 12:12:51 +10001698 .set_page_dirty = xfs_vm_set_page_dirty,
Nathan Scott238f4c52006-03-17 17:26:25 +11001699 .releasepage = xfs_vm_releasepage,
1700 .invalidatepage = xfs_vm_invalidatepage,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001701 .bmap = xfs_vm_bmap,
1702 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001703 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001704 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001705 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706};