blob: 007a520b5ddc74324efcc7673882ced7356cdd47 [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 Chinner70a9883c2013-10-23 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 Hellwig273dda72016-02-08 14:40:51 +110039/* flags for direct write completions */
40#define XFS_DIO_FLAG_UNWRITTEN (1 << 0)
41#define XFS_DIO_FLAG_APPEND (1 << 1)
42
Dave Chinnerfbcc0252016-02-15 17:21:19 +110043/*
44 * structure owned by writepages passed to individual writepage calls
45 */
46struct xfs_writepage_ctx {
47 struct xfs_bmbt_irec imap;
48 bool imap_valid;
49 unsigned int io_type;
Dave Chinnerfbcc0252016-02-15 17:21:19 +110050 struct xfs_ioend *ioend;
51 sector_t last_block;
52};
53
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000054void
Nathan Scottf51623b2006-03-14 13:26:27 +110055xfs_count_page_state(
56 struct page *page,
57 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110058 int *unwritten)
59{
60 struct buffer_head *bh, *head;
61
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100062 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110063
64 bh = head = page_buffers(page);
65 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100066 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110067 (*unwritten) = 1;
68 else if (buffer_delay(bh))
69 (*delalloc) = 1;
70 } while ((bh = bh->b_this_page) != head);
71}
72
Ross Zwisler20a90f52016-02-26 15:19:52 -080073struct block_device *
Christoph Hellwig6214ed42007-09-14 15:23:17 +100074xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000075 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100076{
Christoph Hellwig046f1682010-04-28 12:28:52 +000077 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100078 struct xfs_mount *mp = ip->i_mount;
79
Eric Sandeen71ddabb2007-11-23 16:29:42 +110080 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100081 return mp->m_rtdev_targp->bt_bdev;
82 else
83 return mp->m_ddev_targp->bt_bdev;
84}
85
Christoph Hellwig0829c362005-09-02 16:58:49 +100086/*
Dave Chinner37992c12016-04-06 08:12:28 +100087 * We're now finished for good with this page. Update the page state via the
88 * associated buffer_heads, paying attention to the start and end offsets that
89 * we need to process on the page.
Dave Chinner28b783e2016-07-22 09:56:38 +100090 *
91 * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last
92 * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or
93 * the page at all, as we may be racing with memory reclaim and it can free both
94 * the bufferhead chain and the page as it will see the page as clean and
95 * unused.
Dave Chinner37992c12016-04-06 08:12:28 +100096 */
97static void
98xfs_finish_page_writeback(
99 struct inode *inode,
100 struct bio_vec *bvec,
101 int error)
102{
Dave Chinner37992c12016-04-06 08:12:28 +1000103 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
Dave Chinner28b783e2016-07-22 09:56:38 +1000104 struct buffer_head *head, *bh, *next;
Dave Chinner37992c12016-04-06 08:12:28 +1000105 unsigned int off = 0;
Dave Chinner28b783e2016-07-22 09:56:38 +1000106 unsigned int bsize;
Dave Chinner37992c12016-04-06 08:12:28 +1000107
108 ASSERT(bvec->bv_offset < PAGE_SIZE);
Christoph Hellwig690a7872016-05-20 10:29:15 +1000109 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
Dave Chinner37992c12016-04-06 08:12:28 +1000110 ASSERT(end < PAGE_SIZE);
Christoph Hellwig690a7872016-05-20 10:29:15 +1000111 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
Dave Chinner37992c12016-04-06 08:12:28 +1000112
113 bh = head = page_buffers(bvec->bv_page);
114
Dave Chinner28b783e2016-07-22 09:56:38 +1000115 bsize = bh->b_size;
Dave Chinner37992c12016-04-06 08:12:28 +1000116 do {
Dave Chinner28b783e2016-07-22 09:56:38 +1000117 next = bh->b_this_page;
Dave Chinner37992c12016-04-06 08:12:28 +1000118 if (off < bvec->bv_offset)
119 goto next_bh;
120 if (off > end)
121 break;
122 bh->b_end_io(bh, !error);
123next_bh:
Dave Chinner28b783e2016-07-22 09:56:38 +1000124 off += bsize;
125 } while ((bh = next) != head);
Dave Chinner37992c12016-04-06 08:12:28 +1000126}
127
128/*
129 * We're now finished for good with this ioend structure. Update the page
130 * state, release holds on bios, and finally free up memory. Do not use the
131 * ioend after this.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100132 */
Christoph Hellwig0829c362005-09-02 16:58:49 +1000133STATIC void
134xfs_destroy_ioend(
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000135 struct xfs_ioend *ioend,
136 int error)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000137{
Dave Chinner37992c12016-04-06 08:12:28 +1000138 struct inode *inode = ioend->io_inode;
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000139 struct bio *last = ioend->io_bio;
Dave Chinner37992c12016-04-06 08:12:28 +1000140 struct bio *bio, *next;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100141
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000142 for (bio = &ioend->io_inline_bio; bio; bio = next) {
Dave Chinner37992c12016-04-06 08:12:28 +1000143 struct bio_vec *bvec;
144 int i;
145
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000146 /*
147 * For the last bio, bi_private points to the ioend, so we
148 * need to explicitly end the iteration here.
149 */
150 if (bio == last)
151 next = NULL;
152 else
153 next = bio->bi_private;
Dave Chinner37992c12016-04-06 08:12:28 +1000154
155 /* walk each page on bio, ending page IO on them */
156 bio_for_each_segment_all(bvec, bio, i)
157 xfs_finish_page_writeback(inode, bvec, error);
158
159 bio_put(bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
162
163/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000164 * Fast and loose check if this write could update the on-disk inode size.
165 */
166static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
167{
168 return ioend->io_offset + ioend->io_size >
169 XFS_I(ioend->io_inode)->i_d.di_size;
170}
171
Christoph Hellwig281627d2012-03-13 08:41:05 +0000172STATIC int
173xfs_setfilesize_trans_alloc(
174 struct xfs_ioend *ioend)
175{
176 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
177 struct xfs_trans *tp;
178 int error;
179
Christoph Hellwig253f4912016-04-06 09:19:55 +1000180 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
181 if (error)
Christoph Hellwig281627d2012-03-13 08:41:05 +0000182 return error;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000183
184 ioend->io_append_trans = tp;
185
186 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100187 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200188 * we released it.
189 */
Oleg Nesterovbee91822015-07-19 23:48:20 +0200190 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
Jan Karad9457dc2012-06-12 16:20:39 +0200191 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000192 * We hand off the transaction to the completion thread now, so
193 * clear the flag here.
194 */
195 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
196 return 0;
197}
198
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000199/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000200 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000201 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000202STATIC int
Christoph Hellwige3728432016-09-19 11:26:41 +1000203__xfs_setfilesize(
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100204 struct xfs_inode *ip,
205 struct xfs_trans *tp,
206 xfs_off_t offset,
207 size_t size)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000208{
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000209 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000210
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000211 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100212 isize = xfs_new_eof(ip, offset + size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000213 if (!isize) {
214 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig4906e212015-06-04 13:47:56 +1000215 xfs_trans_cancel(tp);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000216 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000217 }
218
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100219 trace_xfs_setfilesize(ip, offset, size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000220
221 ip->i_d.di_size = isize;
222 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
223 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
224
Christoph Hellwig70393312015-06-04 13:48:08 +1000225 return xfs_trans_commit(tp);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000226}
227
Christoph Hellwige3728432016-09-19 11:26:41 +1000228int
229xfs_setfilesize(
230 struct xfs_inode *ip,
231 xfs_off_t offset,
232 size_t size)
233{
234 struct xfs_mount *mp = ip->i_mount;
235 struct xfs_trans *tp;
236 int error;
237
238 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
239 if (error)
240 return error;
241
242 return __xfs_setfilesize(ip, tp, offset, size);
243}
244
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100245STATIC int
246xfs_setfilesize_ioend(
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000247 struct xfs_ioend *ioend,
248 int error)
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100249{
250 struct xfs_inode *ip = XFS_I(ioend->io_inode);
251 struct xfs_trans *tp = ioend->io_append_trans;
252
253 /*
254 * The transaction may have been allocated in the I/O submission thread,
255 * thus we need to mark ourselves as being in a transaction manually.
256 * Similarly for freeze protection.
257 */
258 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
Oleg Nesterovbee91822015-07-19 23:48:20 +0200259 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100260
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100261 /* we abort the update if there was an IO error */
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000262 if (error) {
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100263 xfs_trans_cancel(tp);
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000264 return error;
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100265 }
266
Christoph Hellwige3728432016-09-19 11:26:41 +1000267 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100268}
269
Christoph Hellwig0829c362005-09-02 16:58:49 +1000270/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000271 * IO write completion.
272 */
273STATIC void
274xfs_end_io(
275 struct work_struct *work)
276{
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000277 struct xfs_ioend *ioend =
278 container_of(work, struct xfs_ioend, io_work);
279 struct xfs_inode *ip = XFS_I(ioend->io_inode);
280 int error = ioend->io_bio->bi_error;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000281
Brian Fosteraf055e32016-02-08 15:00:02 +1100282 /*
283 * Set an error if the mount has shut down and proceed with end I/O
284 * processing so it can perform whatever cleanups are necessary.
285 */
286 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000287 error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000288
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000289 /*
290 * For unwritten extents we need to issue transactions to convert a
291 * range to normal written extens after the data I/O has finished.
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100292 * Detecting and handling completion IO errors is done individually
293 * for each case as different cleanup operations need to be performed
294 * on error.
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000295 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500296 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000297 if (error)
Zhaohongjiang5cb13dc2015-10-12 15:28:39 +1100298 goto done;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000299 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100300 ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000301 } else if (ioend->io_append_trans) {
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000302 error = xfs_setfilesize_ioend(ioend, error);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000303 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000304 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000305 }
306
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000307done:
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000308 xfs_destroy_ioend(ioend, error);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000309}
310
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000311STATIC void
312xfs_end_bio(
313 struct bio *bio)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000314{
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000315 struct xfs_ioend *ioend = bio->bi_private;
316 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000317
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000318 if (ioend->io_type == XFS_IO_UNWRITTEN)
319 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
320 else if (ioend->io_append_trans)
321 queue_work(mp->m_data_workqueue, &ioend->io_work);
322 else
323 xfs_destroy_ioend(ioend, bio->bi_error);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000324}
325
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326STATIC int
327xfs_map_blocks(
328 struct inode *inode,
329 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000330 struct xfs_bmbt_irec *imap,
Dave Chinner988ef922016-02-15 17:20:50 +1100331 int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000333 struct xfs_inode *ip = XFS_I(inode);
334 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000335 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000336 xfs_fileoff_t offset_fsb, end_fsb;
337 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000338 int bmapi_flags = XFS_BMAPI_ENTIRE;
339 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Christoph Hellwiga206c812010-12-10 08:42:20 +0000341 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000342 return -EIO;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000343
Alain Renaud0d882a32012-05-22 15:56:21 -0500344 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000345 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000346
Dave Chinner988ef922016-02-15 17:20:50 +1100347 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000348 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
349 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000350 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000351
Dave Chinnerd2c28192012-06-08 15:44:53 +1000352 if (offset + count > mp->m_super->s_maxbytes)
353 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000354 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
355 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000356 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
357 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000358 xfs_iunlock(ip, XFS_ILOCK_SHARED);
359
Christoph Hellwiga206c812010-12-10 08:42:20 +0000360 if (error)
Dave Chinner24513372014-06-25 14:58:08 +1000361 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000362
Alain Renaud0d882a32012-05-22 15:56:21 -0500363 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000364 (!nimaps || isnullstartblock(imap->br_startblock))) {
Darrick J. Wong60b49842016-10-03 09:11:34 -0700365 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
366 imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000367 if (!error)
Darrick J. Wong60b49842016-10-03 09:11:34 -0700368 trace_xfs_map_blocks_alloc(ip, offset, count, type,
369 imap);
Dave Chinner24513372014-06-25 14:58:08 +1000370 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000371 }
372
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000373#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500374 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000375 ASSERT(nimaps);
376 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
377 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
378 }
379#endif
380 if (nimaps)
381 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
382 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100385STATIC bool
Christoph Hellwig558e6892010-04-28 12:28:58 +0000386xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000387 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000388 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000389 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000391 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000392
Christoph Hellwig558e6892010-04-28 12:28:58 +0000393 return offset >= imap->br_startoff &&
394 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100397STATIC void
398xfs_start_buffer_writeback(
399 struct buffer_head *bh)
400{
401 ASSERT(buffer_mapped(bh));
402 ASSERT(buffer_locked(bh));
403 ASSERT(!buffer_delay(bh));
404 ASSERT(!buffer_unwritten(bh));
405
406 mark_buffer_async_write(bh);
407 set_buffer_uptodate(bh);
408 clear_buffer_dirty(bh);
409}
410
411STATIC void
412xfs_start_page_writeback(
413 struct page *page,
Dave Chinnere10de372016-02-15 17:23:12 +1100414 int clear_dirty)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100415{
416 ASSERT(PageLocked(page));
417 ASSERT(!PageWriteback(page));
Dave Chinner0d085a52014-09-23 15:36:27 +1000418
419 /*
420 * if the page was not fully cleaned, we need to ensure that the higher
421 * layers come back to it correctly. That means we need to keep the page
422 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
423 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
424 * write this page in this writeback sweep will be made.
425 */
426 if (clear_dirty) {
David Chinner92132022006-12-21 10:24:01 +1100427 clear_page_dirty_for_io(page);
Dave Chinner0d085a52014-09-23 15:36:27 +1000428 set_page_writeback(page);
429 } else
430 set_page_writeback_keepwrite(page);
431
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100432 unlock_page(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100433}
434
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000435static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100436{
437 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
438}
439
440/*
Dave Chinnerbb187822016-04-06 08:11:25 +1000441 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
442 * it, and we submit that bio. The ioend may be used for multiple bio
443 * submissions, so we only want to allocate an append transaction for the ioend
444 * once. In the case of multiple bio submission, each bio will take an IO
445 * reference to the ioend to ensure that the ioend completion is only done once
446 * all bios have been submitted and the ioend is really done.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100447 *
448 * If @fail is non-zero, it means that we have a situation where some part of
449 * the submission process has failed after we have marked paged for writeback
Dave Chinnerbb187822016-04-06 08:11:25 +1000450 * and unlocked them. In this situation, we need to fail the bio and ioend
451 * rather than submit it to IO. This typically only happens on a filesystem
452 * shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100453 */
Dave Chinnere10de372016-02-15 17:23:12 +1100454STATIC int
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100455xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000456 struct writeback_control *wbc,
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000457 struct xfs_ioend *ioend,
Dave Chinnere10de372016-02-15 17:23:12 +1100458 int status)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100459{
Dave Chinnere10de372016-02-15 17:23:12 +1100460 /* Reserve log space if we might write beyond the on-disk inode size. */
461 if (!status &&
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000462 ioend->io_type != XFS_IO_UNWRITTEN &&
Dave Chinnerbb187822016-04-06 08:11:25 +1000463 xfs_ioend_is_append(ioend) &&
464 !ioend->io_append_trans)
Dave Chinnere10de372016-02-15 17:23:12 +1100465 status = xfs_setfilesize_trans_alloc(ioend);
Dave Chinnerbb187822016-04-06 08:11:25 +1000466
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000467 ioend->io_bio->bi_private = ioend;
468 ioend->io_bio->bi_end_io = xfs_end_bio;
Mike Christie50bfcd02016-06-05 14:31:57 -0500469 bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE,
470 (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0);
Dave Chinnere10de372016-02-15 17:23:12 +1100471 /*
472 * If we are failing the IO now, just mark the ioend with an
473 * error and finish it. This will run IO completion immediately
474 * as there is only one reference to the ioend at this point in
475 * time.
476 */
477 if (status) {
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000478 ioend->io_bio->bi_error = status;
479 bio_endio(ioend->io_bio);
Dave Chinnere10de372016-02-15 17:23:12 +1100480 return status;
481 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100482
Mike Christie4e49ea42016-06-05 14:31:41 -0500483 submit_bio(ioend->io_bio);
Dave Chinnere10de372016-02-15 17:23:12 +1100484 return 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100485}
486
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000487static void
488xfs_init_bio_from_bh(
489 struct bio *bio,
490 struct buffer_head *bh)
491{
492 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
493 bio->bi_bdev = bh->b_bdev;
494}
495
496static struct xfs_ioend *
497xfs_alloc_ioend(
498 struct inode *inode,
499 unsigned int type,
500 xfs_off_t offset,
501 struct buffer_head *bh)
502{
503 struct xfs_ioend *ioend;
504 struct bio *bio;
505
506 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
507 xfs_init_bio_from_bh(bio, bh);
508
509 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
510 INIT_LIST_HEAD(&ioend->io_list);
511 ioend->io_type = type;
512 ioend->io_inode = inode;
513 ioend->io_size = 0;
514 ioend->io_offset = offset;
515 INIT_WORK(&ioend->io_work, xfs_end_io);
516 ioend->io_append_trans = NULL;
517 ioend->io_bio = bio;
518 return ioend;
519}
520
521/*
522 * Allocate a new bio, and chain the old bio to the new one.
523 *
524 * Note that we have to do perform the chaining in this unintuitive order
525 * so that the bi_private linkage is set up in the right direction for the
526 * traversal in xfs_destroy_ioend().
527 */
528static void
529xfs_chain_bio(
530 struct xfs_ioend *ioend,
531 struct writeback_control *wbc,
532 struct buffer_head *bh)
533{
534 struct bio *new;
535
536 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
537 xfs_init_bio_from_bh(new, bh);
538
539 bio_chain(ioend->io_bio, new);
540 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
Mike Christie50bfcd02016-06-05 14:31:57 -0500541 bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE,
542 (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0);
Mike Christie4e49ea42016-06-05 14:31:41 -0500543 submit_bio(ioend->io_bio);
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000544 ioend->io_bio = new;
545}
546
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100547/*
548 * Test to see if we've been building up a completion structure for
549 * earlier buffers -- if so, we try to append to this ioend if we
550 * can, otherwise we finish off any current ioend and start another.
Dave Chinnere10de372016-02-15 17:23:12 +1100551 * Return the ioend we finished off so that the caller can submit it
552 * once it has finished processing the dirty page.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100553 */
554STATIC void
555xfs_add_to_ioend(
556 struct inode *inode,
557 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100558 xfs_off_t offset,
Dave Chinnere10de372016-02-15 17:23:12 +1100559 struct xfs_writepage_ctx *wpc,
Dave Chinnerbb187822016-04-06 08:11:25 +1000560 struct writeback_control *wbc,
Dave Chinnere10de372016-02-15 17:23:12 +1100561 struct list_head *iolist)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100562{
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100563 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
Darrick J. Wong0df61da2016-03-07 09:32:14 +1100564 bh->b_blocknr != wpc->last_block + 1 ||
565 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
Dave Chinnere10de372016-02-15 17:23:12 +1100566 if (wpc->ioend)
567 list_add(&wpc->ioend->io_list, iolist);
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000568 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100569 }
570
Christoph Hellwig0e51a8e2016-04-06 08:34:30 +1000571 /*
572 * If the buffer doesn't fit into the bio we need to allocate a new
573 * one. This shouldn't happen more than once for a given buffer.
574 */
575 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
576 xfs_chain_bio(wpc->ioend, wbc, bh);
Dave Chinnerbb187822016-04-06 08:11:25 +1000577
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100578 wpc->ioend->io_size += bh->b_size;
579 wpc->last_block = bh->b_blocknr;
Dave Chinnere10de372016-02-15 17:23:12 +1100580 xfs_start_buffer_writeback(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100581}
582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100584xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000585 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100586 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000587 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000588 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100589{
590 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000591 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000592 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
593 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100594
Christoph Hellwig207d0412010-04-28 12:28:56 +0000595 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
596 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100597
Christoph Hellwige5131822010-04-28 12:28:55 +0000598 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000599 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100600
Christoph Hellwig046f1682010-04-28 12:28:52 +0000601 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100602
603 bh->b_blocknr = bn;
604 set_buffer_mapped(bh);
605}
606
607STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000609 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000611 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000612 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000614 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
615 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
Christoph Hellwig207d0412010-04-28 12:28:56 +0000617 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 set_buffer_mapped(bh);
619 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100620 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621}
622
623/*
Dave Chinnera49935f2014-03-07 16:19:14 +1100624 * Test if a given page contains at least one buffer of a given @type.
625 * If @check_all_buffers is true, then we walk all the buffers in the page to
626 * try to find one of the type passed in. If it is not set, then the caller only
627 * needs to check the first buffer on the page for a match.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
Dave Chinnera49935f2014-03-07 16:19:14 +1100629STATIC bool
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000630xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100631 struct page *page,
Dave Chinnera49935f2014-03-07 16:19:14 +1100632 unsigned int type,
633 bool check_all_buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Dave Chinnera49935f2014-03-07 16:19:14 +1100635 struct buffer_head *bh;
636 struct buffer_head *head;
637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (PageWriteback(page))
Dave Chinnera49935f2014-03-07 16:19:14 +1100639 return false;
640 if (!page->mapping)
641 return false;
642 if (!page_has_buffers(page))
643 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Dave Chinnera49935f2014-03-07 16:19:14 +1100645 bh = head = page_buffers(page);
646 do {
647 if (buffer_unwritten(bh)) {
648 if (type == XFS_IO_UNWRITTEN)
649 return true;
650 } else if (buffer_delay(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100651 if (type == XFS_IO_DELALLOC)
Dave Chinnera49935f2014-03-07 16:19:14 +1100652 return true;
653 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100654 if (type == XFS_IO_OVERWRITE)
Dave Chinnera49935f2014-03-07 16:19:14 +1100655 return true;
656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Dave Chinnera49935f2014-03-07 16:19:14 +1100658 /* If we are only checking the first buffer, we are done now. */
659 if (!check_all_buffers)
660 break;
661 } while ((bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Dave Chinnera49935f2014-03-07 16:19:14 +1100663 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
Dave Chinner3ed3a432010-03-05 02:00:42 +0000666STATIC void
667xfs_vm_invalidatepage(
668 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400669 unsigned int offset,
670 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000671{
Lukas Czerner34097df2013-05-21 23:58:01 -0400672 trace_xfs_invalidatepage(page->mapping->host, page, offset,
673 length);
674 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000675}
676
677/*
678 * If the page has delalloc buffers on it, we need to punch them out before we
679 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
680 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
681 * is done on that same region - the delalloc extent is returned when none is
682 * supposed to be there.
683 *
684 * We prevent this by truncating away the delalloc regions on the page before
685 * invalidating it. Because they are delalloc, we can do this without needing a
686 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
687 * truncation without a transaction as there is no space left for block
688 * reservation (typically why we see a ENOSPC in writeback).
689 *
690 * This is not a performance critical path, so for now just do the punching a
691 * buffer head at a time.
692 */
693STATIC void
694xfs_aops_discard_page(
695 struct page *page)
696{
697 struct inode *inode = page->mapping->host;
698 struct xfs_inode *ip = XFS_I(inode);
699 struct buffer_head *bh, *head;
700 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000701
Dave Chinnera49935f2014-03-07 16:19:14 +1100702 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000703 goto out_invalidate;
704
Dave Chinnere8c37532010-03-15 02:36:35 +0000705 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
706 goto out_invalidate;
707
Dave Chinner4f107002011-03-07 10:00:35 +1100708 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000709 "page discard on page %p, inode 0x%llx, offset %llu.",
710 page, ip->i_ino, offset);
711
712 xfs_ilock(ip, XFS_ILOCK_EXCL);
713 bh = head = page_buffers(page);
714 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000715 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100716 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000717
718 if (!buffer_delay(bh))
719 goto next_buffer;
720
Dave Chinnerc726de42010-11-30 15:14:39 +1100721 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
722 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000723 if (error) {
724 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000725 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100726 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000727 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000728 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000729 break;
730 }
731next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100732 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000733
734 } while ((bh = bh->b_this_page) != head);
735
736 xfs_iunlock(ip, XFS_ILOCK_EXCL);
737out_invalidate:
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300738 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000739 return;
740}
741
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742/*
Dave Chinnere10de372016-02-15 17:23:12 +1100743 * We implement an immediate ioend submission policy here to avoid needing to
744 * chain multiple ioends and hence nest mempool allocations which can violate
745 * forward progress guarantees we need to provide. The current ioend we are
746 * adding buffers to is cached on the writepage context, and if the new buffer
747 * does not append to the cached ioend it will create a new ioend and cache that
748 * instead.
749 *
750 * If a new ioend is created and cached, the old ioend is returned and queued
751 * locally for submission once the entire page is processed or an error has been
752 * detected. While ioends are submitted immediately after they are completed,
753 * batching optimisations are provided by higher level block plugging.
754 *
755 * At the end of a writeback pass, there will be a cached ioend remaining on the
756 * writepage context that the caller will need to submit.
757 */
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100758static int
759xfs_writepage_map(
760 struct xfs_writepage_ctx *wpc,
Dave Chinnere10de372016-02-15 17:23:12 +1100761 struct writeback_control *wbc,
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100762 struct inode *inode,
763 struct page *page,
764 loff_t offset,
765 __uint64_t end_offset)
766{
Dave Chinnere10de372016-02-15 17:23:12 +1100767 LIST_HEAD(submit_list);
768 struct xfs_ioend *ioend, *next;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100769 struct buffer_head *bh, *head;
770 ssize_t len = 1 << inode->i_blkbits;
771 int error = 0;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100772 int count = 0;
Dave Chinnere10de372016-02-15 17:23:12 +1100773 int uptodate = 1;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100774
775 bh = head = page_buffers(page);
776 offset = page_offset(page);
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100777 do {
778 if (offset >= end_offset)
779 break;
780 if (!buffer_uptodate(bh))
781 uptodate = 0;
782
783 /*
784 * set_page_dirty dirties all buffers in a page, independent
785 * of their state. The dirty state however is entirely
786 * meaningless for holes (!mapped && uptodate), so skip
787 * buffers covering holes here.
788 */
789 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
790 wpc->imap_valid = false;
791 continue;
792 }
793
794 if (buffer_unwritten(bh)) {
795 if (wpc->io_type != XFS_IO_UNWRITTEN) {
796 wpc->io_type = XFS_IO_UNWRITTEN;
797 wpc->imap_valid = false;
798 }
799 } else if (buffer_delay(bh)) {
800 if (wpc->io_type != XFS_IO_DELALLOC) {
801 wpc->io_type = XFS_IO_DELALLOC;
802 wpc->imap_valid = false;
803 }
804 } else if (buffer_uptodate(bh)) {
805 if (wpc->io_type != XFS_IO_OVERWRITE) {
806 wpc->io_type = XFS_IO_OVERWRITE;
807 wpc->imap_valid = false;
808 }
809 } else {
810 if (PageUptodate(page))
811 ASSERT(buffer_mapped(bh));
812 /*
813 * This buffer is not uptodate and will not be
814 * written to disk. Ensure that we will put any
815 * subsequent writeable buffers into a new
816 * ioend.
817 */
818 wpc->imap_valid = false;
819 continue;
820 }
821
822 if (wpc->imap_valid)
823 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
824 offset);
825 if (!wpc->imap_valid) {
826 error = xfs_map_blocks(inode, offset, &wpc->imap,
827 wpc->io_type);
828 if (error)
Dave Chinnere10de372016-02-15 17:23:12 +1100829 goto out;
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100830 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
831 offset);
832 }
833 if (wpc->imap_valid) {
834 lock_buffer(bh);
835 if (wpc->io_type != XFS_IO_OVERWRITE)
836 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
Dave Chinnerbb187822016-04-06 08:11:25 +1000837 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100838 count++;
839 }
840
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100841 } while (offset += len, ((bh = bh->b_this_page) != head));
842
843 if (uptodate && bh == head)
844 SetPageUptodate(page);
845
Dave Chinnere10de372016-02-15 17:23:12 +1100846 ASSERT(wpc->ioend || list_empty(&submit_list));
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100847
Dave Chinnere10de372016-02-15 17:23:12 +1100848out:
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100849 /*
Dave Chinnere10de372016-02-15 17:23:12 +1100850 * On error, we have to fail the ioend here because we have locked
851 * buffers in the ioend. If we don't do this, we'll deadlock
852 * invalidating the page as that tries to lock the buffers on the page.
853 * Also, because we may have set pages under writeback, we have to make
854 * sure we run IO completion to mark the error state of the IO
855 * appropriately, so we can't cancel the ioend directly here. That means
856 * we have to mark this page as under writeback if we included any
857 * buffers from it in the ioend chain so that completion treats it
858 * correctly.
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100859 *
Dave Chinnere10de372016-02-15 17:23:12 +1100860 * If we didn't include the page in the ioend, the on error we can
861 * simply discard and unlock it as there are no other users of the page
862 * or it's buffers right now. The caller will still need to trigger
863 * submission of outstanding ioends on the writepage context so they are
864 * treated correctly on error.
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100865 */
Dave Chinnere10de372016-02-15 17:23:12 +1100866 if (count) {
867 xfs_start_page_writeback(page, !error);
868
869 /*
870 * Preserve the original error if there was one, otherwise catch
871 * submission errors here and propagate into subsequent ioend
872 * submissions.
873 */
874 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
875 int error2;
876
877 list_del_init(&ioend->io_list);
878 error2 = xfs_submit_ioend(wbc, ioend, error);
879 if (error2 && !error)
880 error = error2;
881 }
882 } else if (error) {
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100883 xfs_aops_discard_page(page);
884 ClearPageUptodate(page);
885 unlock_page(page);
Dave Chinnere10de372016-02-15 17:23:12 +1100886 } else {
887 /*
888 * We can end up here with no error and nothing to write if we
889 * race with a partial page truncate on a sub-page block sized
890 * filesystem. In that case we need to mark the page clean.
891 */
892 xfs_start_page_writeback(page, 1);
893 end_page_writeback(page);
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100894 }
Dave Chinnere10de372016-02-15 17:23:12 +1100895
Dave Chinnerbfce7d22016-02-15 17:21:37 +1100896 mapping_set_error(page->mapping, error);
897 return error;
898}
899
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000901 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000903 * For delalloc space on the page we need to allocate space and flush it.
904 * For unwritten space on the page we need to start the conversion to
905 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000906 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908STATIC int
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100909xfs_do_writepage(
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000910 struct page *page,
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100911 struct writeback_control *wbc,
912 void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
Dave Chinnerfbcc0252016-02-15 17:21:19 +1100914 struct xfs_writepage_ctx *wpc = data;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000915 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 loff_t offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 __uint64_t end_offset;
Dave Chinnerad689722016-02-15 17:21:31 +1100918 pgoff_t end_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Lukas Czerner34097df2013-05-21 23:58:01 -0400920 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000921
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000922 ASSERT(page_has_buffers(page));
923
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000924 /*
925 * Refuse to write the page out if we are called from reclaim context.
926 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400927 * This avoids stack overflows when called from deeply used stacks in
928 * random callers for direct reclaim or memcg reclaim. We explicitly
929 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000930 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700931 * This should never happen except in the case of a VM regression so
932 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000933 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700934 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
935 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000936 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000937
938 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200939 * Given that we do not allow direct reclaim to call us, we should
940 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000941 */
Christoph Hellwig448011e2014-06-06 16:05:15 +1000942 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000943 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000944
Jie Liu8695d272014-05-20 08:24:26 +1000945 /*
Dave Chinnerad689722016-02-15 17:21:31 +1100946 * Is this page beyond the end of the file?
947 *
Jie Liu8695d272014-05-20 08:24:26 +1000948 * The page index is less than the end_index, adjust the end_offset
949 * to the highest offset that this page should represent.
950 * -----------------------------------------------------
951 * | file mapping | <EOF> |
952 * -----------------------------------------------------
953 * | Page ... | Page N-2 | Page N-1 | Page N | |
954 * ^--------------------------------^----------|--------
955 * | desired writeback range | see else |
956 * ---------------------------------^------------------|
957 */
Dave Chinnerad689722016-02-15 17:21:31 +1100958 offset = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300959 end_index = offset >> PAGE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +1000960 if (page->index < end_index)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300961 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +1000962 else {
963 /*
964 * Check whether the page to write out is beyond or straddles
965 * i_size or not.
966 * -------------------------------------------------------
967 * | file mapping | <EOF> |
968 * -------------------------------------------------------
969 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
970 * ^--------------------------------^-----------|---------
971 * | | Straddles |
972 * ---------------------------------^-----------|--------|
973 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300974 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400975
976 /*
Jan Karaff9a28f2013-03-14 14:30:54 +0100977 * Skip the page if it is fully outside i_size, e.g. due to a
978 * truncate operation that is in progress. We must redirty the
979 * page so that reclaim stops reclaiming it. Otherwise
980 * xfs_vm_releasepage() is called on it and gets confused.
Jie Liu8695d272014-05-20 08:24:26 +1000981 *
982 * Note that the end_index is unsigned long, it would overflow
983 * if the given offset is greater than 16TB on 32-bit system
984 * and if we do check the page is fully outside i_size or not
985 * via "if (page->index >= end_index + 1)" as "end_index + 1"
986 * will be evaluated to 0. Hence this page will be redirtied
987 * and be written out repeatedly which would result in an
988 * infinite loop, the user program that perform this operation
989 * will hang. Instead, we can verify this situation by checking
990 * if the page to write is totally beyond the i_size or if it's
991 * offset is just equal to the EOF.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400992 */
Jie Liu8695d272014-05-20 08:24:26 +1000993 if (page->index > end_index ||
994 (page->index == end_index && offset_into_page == 0))
Jan Karaff9a28f2013-03-14 14:30:54 +0100995 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -0400996
997 /*
998 * The page straddles i_size. It must be zeroed out on each
999 * and every writepage invocation because it may be mmapped.
1000 * "A file is mapped in multiples of the page size. For a file
Jie Liu8695d272014-05-20 08:24:26 +10001001 * that is not a multiple of the page size, the remaining
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001002 * memory is zeroed when mapped, and writes to that region are
1003 * not written out to the file."
1004 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001005 zero_user_segment(page, offset_into_page, PAGE_SIZE);
Jie Liu8695d272014-05-20 08:24:26 +10001006
1007 /* Adjust the end_offset to the end of file */
1008 end_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010
Dave Chinnere10de372016-02-15 17:23:12 +11001011 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
Nathan Scottf51623b2006-03-14 13:26:27 +11001012
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001013redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001014 redirty_page_for_writepage(wbc, page);
1015 unlock_page(page);
1016 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001017}
1018
Nathan Scott7d4fb402006-06-09 15:27:16 +10001019STATIC int
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001020xfs_vm_writepage(
1021 struct page *page,
1022 struct writeback_control *wbc)
1023{
1024 struct xfs_writepage_ctx wpc = {
1025 .io_type = XFS_IO_INVALID,
1026 };
1027 int ret;
1028
1029 ret = xfs_do_writepage(page, wbc, &wpc);
Dave Chinnere10de372016-02-15 17:23:12 +11001030 if (wpc.ioend)
1031 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1032 return ret;
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001033}
1034
1035STATIC int
Nathan Scott7d4fb402006-06-09 15:27:16 +10001036xfs_vm_writepages(
1037 struct address_space *mapping,
1038 struct writeback_control *wbc)
1039{
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001040 struct xfs_writepage_ctx wpc = {
1041 .io_type = XFS_IO_INVALID,
1042 };
1043 int ret;
1044
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001045 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Ross Zwisler7f6d5b52016-02-26 15:19:55 -08001046 if (dax_mapping(mapping))
1047 return dax_writeback_mapping_range(mapping,
1048 xfs_find_bdev_for_inode(mapping->host), wbc);
1049
Dave Chinnerfbcc0252016-02-15 17:21:19 +11001050 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
Dave Chinnere10de372016-02-15 17:23:12 +11001051 if (wpc.ioend)
1052 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1053 return ret;
Nathan Scott7d4fb402006-06-09 15:27:16 +10001054}
1055
Nathan Scottf51623b2006-03-14 13:26:27 +11001056/*
1057 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001058 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001059 * have buffer heads in this call.
1060 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001061 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001062 */
1063STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001064xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001065 struct page *page,
1066 gfp_t gfp_mask)
1067{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001068 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001069
Lukas Czerner34097df2013-05-21 23:58:01 -04001070 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001071
Brian Foster99579cc2016-07-22 09:50:38 +10001072 /*
1073 * mm accommodates an old ext3 case where clean pages might not have had
1074 * the dirty bit cleared. Thus, it can send actual dirty pages to
1075 * ->releasepage() via shrink_active_list(). Conversely,
1076 * block_invalidatepage() can send pages that are still marked dirty
1077 * but otherwise have invalidated buffers.
1078 *
1079 * We've historically freed buffers on the latter. Instead, quietly
1080 * filter out all dirty pages to avoid spurious buffer state warnings.
1081 * This can likely be removed once shrink_active_list() is fixed.
1082 */
1083 if (PageDirty(page))
1084 return 0;
1085
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001086 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001087
Christoph Hellwig448011e2014-06-06 16:05:15 +10001088 if (WARN_ON_ONCE(delalloc))
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001089 return 0;
Christoph Hellwig448011e2014-06-06 16:05:15 +10001090 if (WARN_ON_ONCE(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001091 return 0;
1092
Nathan Scottf51623b2006-03-14 13:26:27 +11001093 return try_to_free_buffers(page);
1094}
1095
Dave Chinnera7193702015-04-16 21:57:48 +10001096/*
Christoph Hellwig273dda72016-02-08 14:40:51 +11001097 * When we map a DIO buffer, we may need to pass flags to
1098 * xfs_end_io_direct_write to tell it what kind of write IO we are doing.
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001099 *
1100 * Note that for DIO, an IO to the highest supported file block offset (i.e.
1101 * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64
1102 * bit variable. Hence if we see this overflow, we have to assume that the IO is
1103 * extending the file size. We won't know for sure until IO completion is run
1104 * and the actual max write offset is communicated to the IO completion
1105 * routine.
Dave Chinnera7193702015-04-16 21:57:48 +10001106 */
1107static void
1108xfs_map_direct(
1109 struct inode *inode,
1110 struct buffer_head *bh_result,
1111 struct xfs_bmbt_irec *imap,
Christoph Hellwig273dda72016-02-08 14:40:51 +11001112 xfs_off_t offset)
Dave Chinnera7193702015-04-16 21:57:48 +10001113{
Christoph Hellwig273dda72016-02-08 14:40:51 +11001114 uintptr_t *flags = (uintptr_t *)&bh_result->b_private;
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001115 xfs_off_t size = bh_result->b_size;
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001116
Christoph Hellwig273dda72016-02-08 14:40:51 +11001117 trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size,
1118 ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap);
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001119
Christoph Hellwig273dda72016-02-08 14:40:51 +11001120 if (ISUNWRITTEN(imap)) {
1121 *flags |= XFS_DIO_FLAG_UNWRITTEN;
Dave Chinnera06c2772015-04-16 22:00:00 +10001122 set_buffer_defer_completion(bh_result);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001123 } else if (offset + size > i_size_read(inode) || offset + size < 0) {
1124 *flags |= XFS_DIO_FLAG_APPEND;
1125 set_buffer_defer_completion(bh_result);
Dave Chinnera7193702015-04-16 21:57:48 +10001126 }
1127}
1128
Dave Chinner1fdca9c2015-04-16 21:58:21 +10001129/*
1130 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1131 * is, so that we can avoid repeated get_blocks calls.
1132 *
1133 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1134 * for blocks beyond EOF must be marked new so that sub block regions can be
1135 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1136 * was just allocated or is unwritten, otherwise the callers would overwrite
1137 * existing data with zeros. Hence we have to split the mapping into a range up
1138 * to and including EOF, and a second mapping for beyond EOF.
1139 */
1140static void
1141xfs_map_trim_size(
1142 struct inode *inode,
1143 sector_t iblock,
1144 struct buffer_head *bh_result,
1145 struct xfs_bmbt_irec *imap,
1146 xfs_off_t offset,
1147 ssize_t size)
1148{
1149 xfs_off_t mapping_size;
1150
1151 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1152 mapping_size <<= inode->i_blkbits;
1153
1154 ASSERT(mapping_size > 0);
1155 if (mapping_size > size)
1156 mapping_size = size;
1157 if (offset < i_size_read(inode) &&
1158 offset + mapping_size >= i_size_read(inode)) {
1159 /* limit mapping to block that spans EOF */
1160 mapping_size = roundup_64(i_size_read(inode) - offset,
1161 1 << inode->i_blkbits);
1162 }
1163 if (mapping_size > LONG_MAX)
1164 mapping_size = LONG_MAX;
1165
1166 bh_result->b_size = mapping_size;
1167}
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001170__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 struct inode *inode,
1172 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 struct buffer_head *bh_result,
1174 int create,
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001175 bool direct,
1176 bool dax_fault)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001178 struct xfs_inode *ip = XFS_I(inode);
1179 struct xfs_mount *mp = ip->i_mount;
1180 xfs_fileoff_t offset_fsb, end_fsb;
1181 int error = 0;
1182 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001183 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001184 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001185 xfs_off_t offset;
1186 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001187 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001188
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001189 BUG_ON(create && !direct);
1190
Christoph Hellwiga206c812010-12-10 08:42:20 +00001191 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001192 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001194 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001195 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1196 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001197
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001198 if (!create && offset >= i_size_read(inode))
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001199 return 0;
1200
Dave Chinner507630b2012-03-27 10:34:50 -04001201 /*
1202 * Direct I/O is usually done on preallocated files, so try getting
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001203 * a block mapping without an exclusive lock first.
Dave Chinner507630b2012-03-27 10:34:50 -04001204 */
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001205 lockmode = xfs_ilock_data_map_shared(ip);
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001206
Dave Chinnerd2c28192012-06-08 15:44:53 +10001207 ASSERT(offset <= mp->m_super->s_maxbytes);
1208 if (offset + size > mp->m_super->s_maxbytes)
1209 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001210 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1211 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1212
Dave Chinner5c8ed202011-09-18 20:40:45 +00001213 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1214 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001216 goto out_unlock;
1217
Dave Chinner1ca19152015-11-03 12:37:00 +11001218 /* for DAX, we convert unwritten extents directly */
Christoph Hellwiga206c812010-12-10 08:42:20 +00001219 if (create &&
1220 (!nimaps ||
1221 (imap.br_startblock == HOLESTARTBLOCK ||
Dave Chinner1ca19152015-11-03 12:37:00 +11001222 imap.br_startblock == DELAYSTARTBLOCK) ||
1223 (IS_DAX(inode) && ISUNWRITTEN(&imap)))) {
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001224 /*
1225 * xfs_iomap_write_direct() expects the shared lock. It
1226 * is unlocked on return.
1227 */
1228 if (lockmode == XFS_ILOCK_EXCL)
1229 xfs_ilock_demote(ip, lockmode);
Brian Foster009c6e82015-10-12 15:34:20 +11001230
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001231 error = xfs_iomap_write_direct(ip, offset, size,
1232 &imap, nimaps);
1233 if (error)
1234 return error;
1235 new = 1;
Dave Chinner6b698ed2015-06-04 09:18:53 +10001236
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001237 trace_xfs_get_blocks_alloc(ip, offset, size,
1238 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1239 : XFS_IO_DELALLOC, &imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001240 } else if (nimaps) {
Dave Chinnerd5cc2e32015-04-16 21:59:07 +10001241 trace_xfs_get_blocks_found(ip, offset, size,
1242 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1243 : XFS_IO_OVERWRITE, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001244 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001245 } else {
1246 trace_xfs_get_blocks_notfound(ip, offset, size);
1247 goto out_unlock;
1248 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
Dave Chinner1ca19152015-11-03 12:37:00 +11001250 if (IS_DAX(inode) && create) {
1251 ASSERT(!ISUNWRITTEN(&imap));
1252 /* zeroing is not needed at a higher layer */
1253 new = 0;
1254 }
1255
Dave Chinner1fdca9c2015-04-16 21:58:21 +10001256 /* trim mapping down to size requested */
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001257 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
Dave Chinner1fdca9c2015-04-16 21:58:21 +10001258
Dave Chinnera7193702015-04-16 21:57:48 +10001259 /*
1260 * For unwritten extents do not report a disk address in the buffered
1261 * read case (treat as if we're reading into a hole).
1262 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001263 if (imap.br_startblock != HOLESTARTBLOCK &&
Dave Chinnera7193702015-04-16 21:57:48 +10001264 imap.br_startblock != DELAYSTARTBLOCK &&
1265 (create || !ISUNWRITTEN(&imap))) {
1266 xfs_map_buffer(inode, bh_result, &imap, offset);
1267 if (ISUNWRITTEN(&imap))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 set_buffer_unwritten(bh_result);
Dave Chinnera7193702015-04-16 21:57:48 +10001269 /* direct IO needs special help */
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001270 if (create) {
Christoph Hellwig273dda72016-02-08 14:40:51 +11001271 if (dax_fault)
1272 ASSERT(!ISUNWRITTEN(&imap));
1273 else
1274 xfs_map_direct(inode, bh_result, &imap, offset);
1275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 }
1277
Nathan Scottc2536662006-03-29 10:44:40 +10001278 /*
1279 * If this is a realtime file, data may be on a different device.
1280 * to that pointed to from the buffer_head b_bdev currently.
1281 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001282 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Nathan Scottc2536662006-03-29 10:44:40 +10001284 /*
David Chinner549054a2007-02-10 18:36:35 +11001285 * If we previously allocated a block out beyond eof and we are now
1286 * coming back to use it then we will need to flag it as new even if it
1287 * has a disk address.
1288 *
1289 * With sub-block writes into unwritten extents we also need to mark
1290 * the buffer as new so that the unwritten parts of the buffer gets
1291 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 */
1293 if (create &&
1294 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001295 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001296 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
Christoph Hellwig6e8a27a2016-06-21 09:53:45 +10001299 BUG_ON(direct && imap.br_startblock == DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001302
1303out_unlock:
1304 xfs_iunlock(ip, lockmode);
Dave Chinner24513372014-06-25 14:58:08 +10001305 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306}
1307
1308int
Nathan Scottc2536662006-03-29 10:44:40 +10001309xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 struct inode *inode,
1311 sector_t iblock,
1312 struct buffer_head *bh_result,
1313 int create)
1314{
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001315 return __xfs_get_blocks(inode, iblock, bh_result, create, false, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316}
1317
Dave Chinner6b698ed2015-06-04 09:18:53 +10001318int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001319xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 struct inode *inode,
1321 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 struct buffer_head *bh_result,
1323 int create)
1324{
Dave Chinner3e12dbb2015-11-03 12:27:22 +11001325 return __xfs_get_blocks(inode, iblock, bh_result, create, true, false);
1326}
1327
1328int
1329xfs_get_blocks_dax_fault(
1330 struct inode *inode,
1331 sector_t iblock,
1332 struct buffer_head *bh_result,
1333 int create)
1334{
1335 return __xfs_get_blocks(inode, iblock, bh_result, create, true, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337
Christoph Hellwig273dda72016-02-08 14:40:51 +11001338/*
1339 * Complete a direct I/O write request.
1340 *
1341 * xfs_map_direct passes us some flags in the private data to tell us what to
1342 * do. If no flags are set, then the write IO is an overwrite wholly within
1343 * the existing allocated file size and so there is nothing for us to do.
1344 *
1345 * Note that in this case the completion can be called in interrupt context,
1346 * whereas if we have flags set we will always be called in task context
1347 * (i.e. from a workqueue).
1348 */
Christoph Hellwigfa8d9722016-07-20 11:38:01 +10001349int
Christoph Hellwig273dda72016-02-08 14:40:51 +11001350xfs_end_io_direct_write(
1351 struct kiocb *iocb,
Christoph Hellwig209fb872010-07-18 21:17:11 +00001352 loff_t offset,
Christoph Hellwig273dda72016-02-08 14:40:51 +11001353 ssize_t size,
1354 void *private)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001355{
Christoph Hellwig273dda72016-02-08 14:40:51 +11001356 struct inode *inode = file_inode(iocb->ki_filp);
1357 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001358 uintptr_t flags = (uintptr_t)private;
1359 int error = 0;
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001360
Christoph Hellwig273dda72016-02-08 14:40:51 +11001361 trace_xfs_end_io_direct_write(ip, offset, size);
1362
Christoph Hellwige3728432016-09-19 11:26:41 +10001363 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
Christoph Hellwig273dda72016-02-08 14:40:51 +11001364 return -EIO;
1365
1366 if (size <= 0)
1367 return size;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001368
1369 /*
Christoph Hellwig273dda72016-02-08 14:40:51 +11001370 * The flags tell us whether we are doing unwritten extent conversions
Dave Chinner6dfa1b62015-04-16 21:59:34 +10001371 * or an append transaction that updates the on-disk file size. These
1372 * cases are the only cases where we should *potentially* be needing
Dave Chinnera06c2772015-04-16 22:00:00 +10001373 * to update the VFS inode size.
Christoph Hellwig273dda72016-02-08 14:40:51 +11001374 */
1375 if (flags == 0) {
1376 ASSERT(offset + size <= i_size_read(inode));
1377 return 0;
1378 }
1379
1380 /*
Dave Chinner6dfa1b62015-04-16 21:59:34 +10001381 * We need to update the in-core inode size here so that we don't end up
Dave Chinnera06c2772015-04-16 22:00:00 +10001382 * with the on-disk inode size being outside the in-core inode size. We
1383 * have no other method of updating EOF for AIO, so always do it here
1384 * if necessary.
Dave Chinnerb9d59842015-04-16 22:03:07 +10001385 *
1386 * We need to lock the test/set EOF update as we can be racing with
1387 * other IO completions here to update the EOF. Failing to serialise
1388 * here can result in EOF moving backwards and Bad Things Happen when
1389 * that occurs.
Christoph Hellwig2813d682011-12-18 20:00:12 +00001390 */
Christoph Hellwig273dda72016-02-08 14:40:51 +11001391 spin_lock(&ip->i_flags_lock);
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001392 if (offset + size > i_size_read(inode))
1393 i_size_write(inode, offset + size);
Christoph Hellwig273dda72016-02-08 14:40:51 +11001394 spin_unlock(&ip->i_flags_lock);
Christoph Hellwig2813d682011-12-18 20:00:12 +00001395
Christoph Hellwig273dda72016-02-08 14:40:51 +11001396 if (flags & XFS_DIO_FLAG_UNWRITTEN) {
1397 trace_xfs_end_io_direct_write_unwritten(ip, offset, size);
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001398
Christoph Hellwig273dda72016-02-08 14:40:51 +11001399 error = xfs_iomap_write_unwritten(ip, offset, size);
1400 } else if (flags & XFS_DIO_FLAG_APPEND) {
Christoph Hellwig273dda72016-02-08 14:40:51 +11001401 trace_xfs_end_io_direct_write_append(ip, offset, size);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001402
Christoph Hellwige3728432016-09-19 11:26:41 +10001403 error = xfs_setfilesize(ip, offset, size);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001404 }
1405
Christoph Hellwig273dda72016-02-08 14:40:51 +11001406 return error;
Dave Chinner6e1ba0b2015-06-04 09:19:15 +10001407}
1408
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001410xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 struct kiocb *iocb,
Christoph Hellwigc8b8e322016-04-07 08:51:58 -07001412 struct iov_iter *iter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
Jie Liu58e59852013-07-16 13:11:16 +08001414 /*
Christoph Hellwigfa8d9722016-07-20 11:38:01 +10001415 * We just need the method present so that open/fcntl allow direct I/O.
Jie Liu58e59852013-07-16 13:11:16 +08001416 */
Christoph Hellwigfa8d9722016-07-20 11:38:01 +10001417 return -EINVAL;
Nathan Scottf51623b2006-03-14 13:26:27 +11001418}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
1420STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001421xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 struct address_space *mapping,
1423 sector_t block)
1424{
1425 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001426 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001428 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001429 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001430 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001431 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001432 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433}
1434
1435STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001436xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 struct file *unused,
1438 struct page *page)
1439{
Dave Chinner121e2132016-01-08 11:28:35 +11001440 trace_xfs_vm_readpage(page->mapping->host, 1);
Nathan Scottc2536662006-03-29 10:44:40 +10001441 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442}
1443
1444STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001445xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 struct file *unused,
1447 struct address_space *mapping,
1448 struct list_head *pages,
1449 unsigned nr_pages)
1450{
Dave Chinner121e2132016-01-08 11:28:35 +11001451 trace_xfs_vm_readpages(mapping->host, nr_pages);
Nathan Scottc2536662006-03-29 10:44:40 +10001452 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453}
1454
Dave Chinner22e757a2014-09-02 12:12:51 +10001455/*
1456 * This is basically a copy of __set_page_dirty_buffers() with one
1457 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1458 * dirty, we'll never be able to clean them because we don't write buffers
1459 * beyond EOF, and that means we can't invalidate pages that span EOF
1460 * that have been marked dirty. Further, the dirty state can leak into
1461 * the file interior if the file is extended, resulting in all sorts of
1462 * bad things happening as the state does not match the underlying data.
1463 *
1464 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1465 * this only exist because of bufferheads and how the generic code manages them.
1466 */
1467STATIC int
1468xfs_vm_set_page_dirty(
1469 struct page *page)
1470{
1471 struct address_space *mapping = page->mapping;
1472 struct inode *inode = mapping->host;
1473 loff_t end_offset;
1474 loff_t offset;
1475 int newly_dirty;
1476
1477 if (unlikely(!mapping))
1478 return !TestSetPageDirty(page);
1479
1480 end_offset = i_size_read(inode);
1481 offset = page_offset(page);
1482
1483 spin_lock(&mapping->private_lock);
1484 if (page_has_buffers(page)) {
1485 struct buffer_head *head = page_buffers(page);
1486 struct buffer_head *bh = head;
1487
1488 do {
1489 if (offset < end_offset)
1490 set_buffer_dirty(bh);
1491 bh = bh->b_this_page;
1492 offset += 1 << inode->i_blkbits;
1493 } while (bh != head);
1494 }
Greg Thelenc4843a72015-05-22 17:13:16 -04001495 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -07001496 * Lock out page->mem_cgroup migration to keep PageDirty
1497 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -04001498 */
Johannes Weiner62cccb82016-03-15 14:57:22 -07001499 lock_page_memcg(page);
Dave Chinner22e757a2014-09-02 12:12:51 +10001500 newly_dirty = !TestSetPageDirty(page);
1501 spin_unlock(&mapping->private_lock);
1502
1503 if (newly_dirty) {
1504 /* sigh - __set_page_dirty() is static, so copy it here, too */
1505 unsigned long flags;
1506
1507 spin_lock_irqsave(&mapping->tree_lock, flags);
1508 if (page->mapping) { /* Race with truncate? */
1509 WARN_ON_ONCE(!PageUptodate(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -07001510 account_page_dirtied(page, mapping);
Dave Chinner22e757a2014-09-02 12:12:51 +10001511 radix_tree_tag_set(&mapping->page_tree,
1512 page_index(page), PAGECACHE_TAG_DIRTY);
1513 }
1514 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Dave Chinner22e757a2014-09-02 12:12:51 +10001515 }
Johannes Weiner62cccb82016-03-15 14:57:22 -07001516 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04001517 if (newly_dirty)
1518 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Dave Chinner22e757a2014-09-02 12:12:51 +10001519 return newly_dirty;
1520}
1521
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001522const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001523 .readpage = xfs_vm_readpage,
1524 .readpages = xfs_vm_readpages,
1525 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001526 .writepages = xfs_vm_writepages,
Dave Chinner22e757a2014-09-02 12:12:51 +10001527 .set_page_dirty = xfs_vm_set_page_dirty,
Nathan Scott238f4c52006-03-17 17:26:25 +11001528 .releasepage = xfs_vm_releasepage,
1529 .invalidatepage = xfs_vm_invalidatepage,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001530 .bmap = xfs_vm_bmap,
1531 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001532 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001533 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001534 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535};