xen/tmem: Fix compile warning.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / xfs / xfs_aops.c
CommitLineData
1da177e4 1/*
7b718769
NS
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
7b718769
NS
9 * 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.
1da177e4 13 *
7b718769
NS
14 * 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
1da177e4 17 */
1da177e4 18#include "xfs.h"
1da177e4
LT
19#include "xfs_log.h"
20#include "xfs_sb.h"
a844f451 21#include "xfs_ag.h"
1da177e4 22#include "xfs_trans.h"
1da177e4
LT
23#include "xfs_mount.h"
24#include "xfs_bmap_btree.h"
1da177e4
LT
25#include "xfs_dinode.h"
26#include "xfs_inode.h"
281627df 27#include "xfs_inode_item.h"
a844f451 28#include "xfs_alloc.h"
1da177e4 29#include "xfs_error.h"
1da177e4 30#include "xfs_iomap.h"
739bfb2a 31#include "xfs_vnodeops.h"
0b1b213f 32#include "xfs_trace.h"
3ed3a434 33#include "xfs_bmap.h"
a27bb332 34#include <linux/aio.h>
5a0e3ad6 35#include <linux/gfp.h>
1da177e4 36#include <linux/mpage.h>
10ce4444 37#include <linux/pagevec.h>
1da177e4
LT
38#include <linux/writeback.h>
39
0b1b213f 40void
f51623b2
NS
41xfs_count_page_state(
42 struct page *page,
43 int *delalloc,
f51623b2
NS
44 int *unwritten)
45{
46 struct buffer_head *bh, *head;
47
20cb52eb 48 *delalloc = *unwritten = 0;
f51623b2
NS
49
50 bh = head = page_buffers(page);
51 do {
20cb52eb 52 if (buffer_unwritten(bh))
f51623b2
NS
53 (*unwritten) = 1;
54 else if (buffer_delay(bh))
55 (*delalloc) = 1;
56 } while ((bh = bh->b_this_page) != head);
57}
58
6214ed44
CH
59STATIC struct block_device *
60xfs_find_bdev_for_inode(
046f1685 61 struct inode *inode)
6214ed44 62{
046f1685 63 struct xfs_inode *ip = XFS_I(inode);
6214ed44
CH
64 struct xfs_mount *mp = ip->i_mount;
65
71ddabb9 66 if (XFS_IS_REALTIME_INODE(ip))
6214ed44
CH
67 return mp->m_rtdev_targp->bt_bdev;
68 else
69 return mp->m_ddev_targp->bt_bdev;
70}
71
f6d6d4fc
CH
72/*
73 * We're now finished for good with this ioend structure.
74 * Update the page state via the associated buffer_heads,
75 * release holds on the inode and bio, and finally free
76 * up memory. Do not use the ioend after this.
77 */
0829c360
CH
78STATIC void
79xfs_destroy_ioend(
80 xfs_ioend_t *ioend)
81{
f6d6d4fc
CH
82 struct buffer_head *bh, *next;
83
84 for (bh = ioend->io_buffer_head; bh; bh = next) {
85 next = bh->b_private;
7d04a335 86 bh->b_end_io(bh, !ioend->io_error);
f6d6d4fc 87 }
583fa586 88
c859cdd1 89 if (ioend->io_iocb) {
4b05d09c 90 inode_dio_done(ioend->io_inode);
04f658ee
CH
91 if (ioend->io_isasync) {
92 aio_complete(ioend->io_iocb, ioend->io_error ?
93 ioend->io_error : ioend->io_result, 0);
94 }
c859cdd1 95 }
4a06fd26 96
0829c360
CH
97 mempool_free(ioend, xfs_ioend_pool);
98}
99
fc0063c4
CH
100/*
101 * Fast and loose check if this write could update the on-disk inode size.
102 */
103static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
104{
105 return ioend->io_offset + ioend->io_size >
106 XFS_I(ioend->io_inode)->i_d.di_size;
107}
108
281627df
CH
109STATIC int
110xfs_setfilesize_trans_alloc(
111 struct xfs_ioend *ioend)
112{
113 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
114 struct xfs_trans *tp;
115 int error;
116
117 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
118
119 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
120 if (error) {
121 xfs_trans_cancel(tp, 0);
122 return error;
123 }
124
125 ioend->io_append_trans = tp;
126
d9457dc0 127 /*
437a255a 128 * We may pass freeze protection with a transaction. So tell lockdep
d9457dc0
JK
129 * we released it.
130 */
131 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
132 1, _THIS_IP_);
281627df
CH
133 /*
134 * We hand off the transaction to the completion thread now, so
135 * clear the flag here.
136 */
137 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
138 return 0;
139}
140
ba87ea69 141/*
2813d682 142 * Update on-disk file size now that data has been written to disk.
ba87ea69 143 */
281627df 144STATIC int
ba87ea69 145xfs_setfilesize(
aa6bf01d 146 struct xfs_ioend *ioend)
ba87ea69 147{
aa6bf01d 148 struct xfs_inode *ip = XFS_I(ioend->io_inode);
281627df 149 struct xfs_trans *tp = ioend->io_append_trans;
ba87ea69 150 xfs_fsize_t isize;
ba87ea69 151
281627df 152 /*
437a255a
DC
153 * The transaction may have been allocated in the I/O submission thread,
154 * thus we need to mark ourselves as beeing in a transaction manually.
155 * Similarly for freeze protection.
281627df
CH
156 */
157 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
437a255a
DC
158 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
159 0, 1, _THIS_IP_);
281627df 160
aa6bf01d 161 xfs_ilock(ip, XFS_ILOCK_EXCL);
6923e686 162 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
281627df
CH
163 if (!isize) {
164 xfs_iunlock(ip, XFS_ILOCK_EXCL);
165 xfs_trans_cancel(tp, 0);
166 return 0;
ba87ea69
LM
167 }
168
281627df
CH
169 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
170
171 ip->i_d.di_size = isize;
172 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
173 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
174
175 return xfs_trans_commit(tp, 0);
77d7a0c2
DC
176}
177
178/*
209fb87a 179 * Schedule IO completion handling on the final put of an ioend.
fc0063c4
CH
180 *
181 * If there is no work to do we might as well call it a day and free the
182 * ioend right now.
77d7a0c2
DC
183 */
184STATIC void
185xfs_finish_ioend(
209fb87a 186 struct xfs_ioend *ioend)
77d7a0c2
DC
187{
188 if (atomic_dec_and_test(&ioend->io_remaining)) {
aa6bf01d
CH
189 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
190
0d882a36 191 if (ioend->io_type == XFS_IO_UNWRITTEN)
aa6bf01d 192 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
437a255a
DC
193 else if (ioend->io_append_trans ||
194 (ioend->io_isdirect && xfs_ioend_is_append(ioend)))
aa6bf01d 195 queue_work(mp->m_data_workqueue, &ioend->io_work);
fc0063c4
CH
196 else
197 xfs_destroy_ioend(ioend);
77d7a0c2 198 }
ba87ea69
LM
199}
200
0829c360 201/*
5ec4fabb 202 * IO write completion.
f6d6d4fc
CH
203 */
204STATIC void
5ec4fabb 205xfs_end_io(
77d7a0c2 206 struct work_struct *work)
0829c360 207{
77d7a0c2
DC
208 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
209 struct xfs_inode *ip = XFS_I(ioend->io_inode);
69418932 210 int error = 0;
ba87ea69 211
04f658ee 212 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
810627d9 213 ioend->io_error = -EIO;
04f658ee
CH
214 goto done;
215 }
216 if (ioend->io_error)
217 goto done;
218
5ec4fabb
CH
219 /*
220 * For unwritten extents we need to issue transactions to convert a
221 * range to normal written extens after the data I/O has finished.
222 */
0d882a36 223 if (ioend->io_type == XFS_IO_UNWRITTEN) {
437a255a
DC
224 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
225 ioend->io_size);
226 } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) {
281627df 227 /*
437a255a
DC
228 * For direct I/O we do not know if we need to allocate blocks
229 * or not so we can't preallocate an append transaction as that
230 * results in nested reservations and log space deadlocks. Hence
231 * allocate the transaction here. While this is sub-optimal and
232 * can block IO completion for some time, we're stuck with doing
233 * it this way until we can pass the ioend to the direct IO
234 * allocation callbacks and avoid nesting that way.
281627df 235 */
437a255a
DC
236 error = xfs_setfilesize_trans_alloc(ioend);
237 if (error)
04f658ee 238 goto done;
437a255a 239 error = xfs_setfilesize(ioend);
281627df
CH
240 } else if (ioend->io_append_trans) {
241 error = xfs_setfilesize(ioend);
84803fb7 242 } else {
281627df 243 ASSERT(!xfs_ioend_is_append(ioend));
5ec4fabb 244 }
ba87ea69 245
04f658ee 246done:
437a255a
DC
247 if (error)
248 ioend->io_error = -error;
aa6bf01d 249 xfs_destroy_ioend(ioend);
c626d174
DC
250}
251
209fb87a
CH
252/*
253 * Call IO completion handling in caller context on the final put of an ioend.
254 */
255STATIC void
256xfs_finish_ioend_sync(
257 struct xfs_ioend *ioend)
258{
259 if (atomic_dec_and_test(&ioend->io_remaining))
260 xfs_end_io(&ioend->io_work);
261}
262
0829c360
CH
263/*
264 * Allocate and initialise an IO completion structure.
265 * We need to track unwritten extent write completion here initially.
266 * We'll need to extend this for updating the ondisk inode size later
267 * (vs. incore size).
268 */
269STATIC xfs_ioend_t *
270xfs_alloc_ioend(
f6d6d4fc
CH
271 struct inode *inode,
272 unsigned int type)
0829c360
CH
273{
274 xfs_ioend_t *ioend;
275
276 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
277
278 /*
279 * Set the count to 1 initially, which will prevent an I/O
280 * completion callback from happening before we have started
281 * all the I/O from calling the completion routine too early.
282 */
283 atomic_set(&ioend->io_remaining, 1);
c859cdd1 284 ioend->io_isasync = 0;
281627df 285 ioend->io_isdirect = 0;
7d04a335 286 ioend->io_error = 0;
f6d6d4fc
CH
287 ioend->io_list = NULL;
288 ioend->io_type = type;
b677c210 289 ioend->io_inode = inode;
c1a073bd 290 ioend->io_buffer_head = NULL;
f6d6d4fc 291 ioend->io_buffer_tail = NULL;
0829c360
CH
292 ioend->io_offset = 0;
293 ioend->io_size = 0;
fb511f21
CH
294 ioend->io_iocb = NULL;
295 ioend->io_result = 0;
281627df 296 ioend->io_append_trans = NULL;
0829c360 297
5ec4fabb 298 INIT_WORK(&ioend->io_work, xfs_end_io);
0829c360
CH
299 return ioend;
300}
301
1da177e4
LT
302STATIC int
303xfs_map_blocks(
304 struct inode *inode,
305 loff_t offset,
207d0416 306 struct xfs_bmbt_irec *imap,
a206c817
CH
307 int type,
308 int nonblocking)
1da177e4 309{
a206c817
CH
310 struct xfs_inode *ip = XFS_I(inode);
311 struct xfs_mount *mp = ip->i_mount;
ed1e7b7e 312 ssize_t count = 1 << inode->i_blkbits;
a206c817
CH
313 xfs_fileoff_t offset_fsb, end_fsb;
314 int error = 0;
a206c817
CH
315 int bmapi_flags = XFS_BMAPI_ENTIRE;
316 int nimaps = 1;
317
318 if (XFS_FORCED_SHUTDOWN(mp))
319 return -XFS_ERROR(EIO);
320
0d882a36 321 if (type == XFS_IO_UNWRITTEN)
a206c817 322 bmapi_flags |= XFS_BMAPI_IGSTATE;
8ff2957d
CH
323
324 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
325 if (nonblocking)
326 return -XFS_ERROR(EAGAIN);
327 xfs_ilock(ip, XFS_ILOCK_SHARED);
a206c817
CH
328 }
329
8ff2957d
CH
330 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
331 (ip->i_df.if_flags & XFS_IFEXTENTS));
d2c28191 332 ASSERT(offset <= mp->m_super->s_maxbytes);
8ff2957d 333
d2c28191
DC
334 if (offset + count > mp->m_super->s_maxbytes)
335 count = mp->m_super->s_maxbytes - offset;
a206c817
CH
336 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
337 offset_fsb = XFS_B_TO_FSBT(mp, offset);
5c8ed202
DC
338 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
339 imap, &nimaps, bmapi_flags);
8ff2957d 340 xfs_iunlock(ip, XFS_ILOCK_SHARED);
a206c817 341
8ff2957d
CH
342 if (error)
343 return -XFS_ERROR(error);
a206c817 344
0d882a36 345 if (type == XFS_IO_DELALLOC &&
8ff2957d 346 (!nimaps || isnullstartblock(imap->br_startblock))) {
a206c817
CH
347 error = xfs_iomap_write_allocate(ip, offset, count, imap);
348 if (!error)
349 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
8ff2957d 350 return -XFS_ERROR(error);
a206c817
CH
351 }
352
8ff2957d 353#ifdef DEBUG
0d882a36 354 if (type == XFS_IO_UNWRITTEN) {
8ff2957d
CH
355 ASSERT(nimaps);
356 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
357 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
358 }
359#endif
360 if (nimaps)
361 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
362 return 0;
1da177e4
LT
363}
364
b8f82a4a 365STATIC int
558e6891 366xfs_imap_valid(
8699bb0a 367 struct inode *inode,
207d0416 368 struct xfs_bmbt_irec *imap,
558e6891 369 xfs_off_t offset)
1da177e4 370{
558e6891 371 offset >>= inode->i_blkbits;
8699bb0a 372
558e6891
CH
373 return offset >= imap->br_startoff &&
374 offset < imap->br_startoff + imap->br_blockcount;
1da177e4
LT
375}
376
f6d6d4fc
CH
377/*
378 * BIO completion handler for buffered IO.
379 */
782e3b3b 380STATIC void
f6d6d4fc
CH
381xfs_end_bio(
382 struct bio *bio,
f6d6d4fc
CH
383 int error)
384{
385 xfs_ioend_t *ioend = bio->bi_private;
386
f6d6d4fc 387 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
7d04a335 388 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
f6d6d4fc
CH
389
390 /* Toss bio and pass work off to an xfsdatad thread */
f6d6d4fc
CH
391 bio->bi_private = NULL;
392 bio->bi_end_io = NULL;
f6d6d4fc 393 bio_put(bio);
7d04a335 394
209fb87a 395 xfs_finish_ioend(ioend);
f6d6d4fc
CH
396}
397
398STATIC void
399xfs_submit_ioend_bio(
06342cf8
CH
400 struct writeback_control *wbc,
401 xfs_ioend_t *ioend,
402 struct bio *bio)
f6d6d4fc
CH
403{
404 atomic_inc(&ioend->io_remaining);
f6d6d4fc
CH
405 bio->bi_private = ioend;
406 bio->bi_end_io = xfs_end_bio;
721a9602 407 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
f6d6d4fc
CH
408}
409
410STATIC struct bio *
411xfs_alloc_ioend_bio(
412 struct buffer_head *bh)
413{
f6d6d4fc 414 int nvecs = bio_get_nr_vecs(bh->b_bdev);
221cb251 415 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
f6d6d4fc
CH
416
417 ASSERT(bio->bi_private == NULL);
418 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
419 bio->bi_bdev = bh->b_bdev;
f6d6d4fc
CH
420 return bio;
421}
422
423STATIC void
424xfs_start_buffer_writeback(
425 struct buffer_head *bh)
426{
427 ASSERT(buffer_mapped(bh));
428 ASSERT(buffer_locked(bh));
429 ASSERT(!buffer_delay(bh));
430 ASSERT(!buffer_unwritten(bh));
431
432 mark_buffer_async_write(bh);
433 set_buffer_uptodate(bh);
434 clear_buffer_dirty(bh);
435}
436
437STATIC void
438xfs_start_page_writeback(
439 struct page *page,
f6d6d4fc
CH
440 int clear_dirty,
441 int buffers)
442{
443 ASSERT(PageLocked(page));
444 ASSERT(!PageWriteback(page));
f6d6d4fc 445 if (clear_dirty)
92132021
DC
446 clear_page_dirty_for_io(page);
447 set_page_writeback(page);
f6d6d4fc 448 unlock_page(page);
1f7decf6
FW
449 /* If no buffers on the page are to be written, finish it here */
450 if (!buffers)
f6d6d4fc 451 end_page_writeback(page);
f6d6d4fc
CH
452}
453
454static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
455{
456 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
457}
458
459/*
d88992f6
DC
460 * Submit all of the bios for all of the ioends we have saved up, covering the
461 * initial writepage page and also any probed pages.
462 *
463 * Because we may have multiple ioends spanning a page, we need to start
464 * writeback on all the buffers before we submit them for I/O. If we mark the
465 * buffers as we got, then we can end up with a page that only has buffers
466 * marked async write and I/O complete on can occur before we mark the other
467 * buffers async write.
468 *
469 * The end result of this is that we trip a bug in end_page_writeback() because
470 * we call it twice for the one page as the code in end_buffer_async_write()
471 * assumes that all buffers on the page are started at the same time.
472 *
473 * The fix is two passes across the ioend list - one to start writeback on the
c41564b5 474 * buffer_heads, and then submit them for I/O on the second pass.
7bf7f352
DC
475 *
476 * If @fail is non-zero, it means that we have a situation where some part of
477 * the submission process has failed after we have marked paged for writeback
478 * and unlocked them. In this situation, we need to fail the ioend chain rather
479 * than submit it to IO. This typically only happens on a filesystem shutdown.
f6d6d4fc
CH
480 */
481STATIC void
482xfs_submit_ioend(
06342cf8 483 struct writeback_control *wbc,
7bf7f352
DC
484 xfs_ioend_t *ioend,
485 int fail)
f6d6d4fc 486{
d88992f6 487 xfs_ioend_t *head = ioend;
f6d6d4fc
CH
488 xfs_ioend_t *next;
489 struct buffer_head *bh;
490 struct bio *bio;
491 sector_t lastblock = 0;
492
d88992f6
DC
493 /* Pass 1 - start writeback */
494 do {
495 next = ioend->io_list;
221cb251 496 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
d88992f6 497 xfs_start_buffer_writeback(bh);
d88992f6
DC
498 } while ((ioend = next) != NULL);
499
500 /* Pass 2 - submit I/O */
501 ioend = head;
f6d6d4fc
CH
502 do {
503 next = ioend->io_list;
504 bio = NULL;
505
7bf7f352
DC
506 /*
507 * If we are failing the IO now, just mark the ioend with an
508 * error and finish it. This will run IO completion immediately
509 * as there is only one reference to the ioend at this point in
510 * time.
511 */
512 if (fail) {
513 ioend->io_error = -fail;
514 xfs_finish_ioend(ioend);
515 continue;
516 }
517
f6d6d4fc 518 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
f6d6d4fc
CH
519
520 if (!bio) {
521 retry:
522 bio = xfs_alloc_ioend_bio(bh);
523 } else if (bh->b_blocknr != lastblock + 1) {
06342cf8 524 xfs_submit_ioend_bio(wbc, ioend, bio);
f6d6d4fc
CH
525 goto retry;
526 }
527
528 if (bio_add_buffer(bio, bh) != bh->b_size) {
06342cf8 529 xfs_submit_ioend_bio(wbc, ioend, bio);
f6d6d4fc
CH
530 goto retry;
531 }
532
533 lastblock = bh->b_blocknr;
534 }
535 if (bio)
06342cf8 536 xfs_submit_ioend_bio(wbc, ioend, bio);
209fb87a 537 xfs_finish_ioend(ioend);
f6d6d4fc
CH
538 } while ((ioend = next) != NULL);
539}
540
541/*
542 * Cancel submission of all buffer_heads so far in this endio.
543 * Toss the endio too. Only ever called for the initial page
544 * in a writepage request, so only ever one page.
545 */
546STATIC void
547xfs_cancel_ioend(
548 xfs_ioend_t *ioend)
549{
550 xfs_ioend_t *next;
551 struct buffer_head *bh, *next_bh;
552
553 do {
554 next = ioend->io_list;
555 bh = ioend->io_buffer_head;
556 do {
557 next_bh = bh->b_private;
558 clear_buffer_async_write(bh);
559 unlock_buffer(bh);
560 } while ((bh = next_bh) != NULL);
561
f6d6d4fc
CH
562 mempool_free(ioend, xfs_ioend_pool);
563 } while ((ioend = next) != NULL);
564}
565
566/*
567 * Test to see if we've been building up a completion structure for
568 * earlier buffers -- if so, we try to append to this ioend if we
569 * can, otherwise we finish off any current ioend and start another.
570 * Return true if we've finished the given ioend.
571 */
572STATIC void
573xfs_add_to_ioend(
574 struct inode *inode,
575 struct buffer_head *bh,
7336cea8 576 xfs_off_t offset,
f6d6d4fc
CH
577 unsigned int type,
578 xfs_ioend_t **result,
579 int need_ioend)
580{
581 xfs_ioend_t *ioend = *result;
582
583 if (!ioend || need_ioend || type != ioend->io_type) {
584 xfs_ioend_t *previous = *result;
f6d6d4fc 585
f6d6d4fc
CH
586 ioend = xfs_alloc_ioend(inode, type);
587 ioend->io_offset = offset;
588 ioend->io_buffer_head = bh;
589 ioend->io_buffer_tail = bh;
590 if (previous)
591 previous->io_list = ioend;
592 *result = ioend;
593 } else {
594 ioend->io_buffer_tail->b_private = bh;
595 ioend->io_buffer_tail = bh;
596 }
597
598 bh->b_private = NULL;
599 ioend->io_size += bh->b_size;
600}
601
87cbc49c
NS
602STATIC void
603xfs_map_buffer(
046f1685 604 struct inode *inode,
87cbc49c 605 struct buffer_head *bh,
207d0416 606 struct xfs_bmbt_irec *imap,
046f1685 607 xfs_off_t offset)
87cbc49c
NS
608{
609 sector_t bn;
8699bb0a 610 struct xfs_mount *m = XFS_I(inode)->i_mount;
207d0416
CH
611 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
612 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
87cbc49c 613
207d0416
CH
614 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
615 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
87cbc49c 616
e513182d 617 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
8699bb0a 618 ((offset - iomap_offset) >> inode->i_blkbits);
87cbc49c 619
046f1685 620 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
87cbc49c
NS
621
622 bh->b_blocknr = bn;
623 set_buffer_mapped(bh);
624}
625
1da177e4
LT
626STATIC void
627xfs_map_at_offset(
046f1685 628 struct inode *inode,
1da177e4 629 struct buffer_head *bh,
207d0416 630 struct xfs_bmbt_irec *imap,
046f1685 631 xfs_off_t offset)
1da177e4 632{
207d0416
CH
633 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
634 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
1da177e4 635
207d0416 636 xfs_map_buffer(inode, bh, imap, offset);
1da177e4
LT
637 set_buffer_mapped(bh);
638 clear_buffer_delay(bh);
f6d6d4fc 639 clear_buffer_unwritten(bh);
1da177e4
LT
640}
641
1da177e4 642/*
10ce4444
CH
643 * Test if a given page is suitable for writing as part of an unwritten
644 * or delayed allocate extent.
1da177e4 645 */
10ce4444 646STATIC int
6ffc4db5 647xfs_check_page_type(
10ce4444 648 struct page *page,
f6d6d4fc 649 unsigned int type)
1da177e4 650{
1da177e4 651 if (PageWriteback(page))
10ce4444 652 return 0;
1da177e4
LT
653
654 if (page->mapping && page_has_buffers(page)) {
655 struct buffer_head *bh, *head;
656 int acceptable = 0;
657
658 bh = head = page_buffers(page);
659 do {
f6d6d4fc 660 if (buffer_unwritten(bh))
0d882a36 661 acceptable += (type == XFS_IO_UNWRITTEN);
f6d6d4fc 662 else if (buffer_delay(bh))
0d882a36 663 acceptable += (type == XFS_IO_DELALLOC);
2ddee844 664 else if (buffer_dirty(bh) && buffer_mapped(bh))
0d882a36 665 acceptable += (type == XFS_IO_OVERWRITE);
f6d6d4fc 666 else
1da177e4 667 break;
1da177e4
LT
668 } while ((bh = bh->b_this_page) != head);
669
670 if (acceptable)
10ce4444 671 return 1;
1da177e4
LT
672 }
673
10ce4444 674 return 0;
1da177e4
LT
675}
676
1da177e4
LT
677/*
678 * Allocate & map buffers for page given the extent map. Write it out.
679 * except for the original page of a writepage, this is called on
680 * delalloc/unwritten pages only, for the original page it is possible
681 * that the page has no mapping at all.
682 */
f6d6d4fc 683STATIC int
1da177e4
LT
684xfs_convert_page(
685 struct inode *inode,
686 struct page *page,
10ce4444 687 loff_t tindex,
207d0416 688 struct xfs_bmbt_irec *imap,
f6d6d4fc 689 xfs_ioend_t **ioendp,
2fa24f92 690 struct writeback_control *wbc)
1da177e4 691{
f6d6d4fc 692 struct buffer_head *bh, *head;
9260dc6b
CH
693 xfs_off_t end_offset;
694 unsigned long p_offset;
f6d6d4fc 695 unsigned int type;
24e17b5f 696 int len, page_dirty;
f6d6d4fc 697 int count = 0, done = 0, uptodate = 1;
9260dc6b 698 xfs_off_t offset = page_offset(page);
1da177e4 699
10ce4444
CH
700 if (page->index != tindex)
701 goto fail;
529ae9aa 702 if (!trylock_page(page))
10ce4444
CH
703 goto fail;
704 if (PageWriteback(page))
705 goto fail_unlock_page;
706 if (page->mapping != inode->i_mapping)
707 goto fail_unlock_page;
6ffc4db5 708 if (!xfs_check_page_type(page, (*ioendp)->io_type))
10ce4444
CH
709 goto fail_unlock_page;
710
24e17b5f
NS
711 /*
712 * page_dirty is initially a count of buffers on the page before
c41564b5 713 * EOF and is decremented as we move each into a cleanable state.
9260dc6b
CH
714 *
715 * Derivation:
716 *
717 * End offset is the highest offset that this page should represent.
718 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
719 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
720 * hence give us the correct page_dirty count. On any other page,
721 * it will be zero and in that case we need page_dirty to be the
722 * count of buffers on the page.
24e17b5f 723 */
9260dc6b
CH
724 end_offset = min_t(unsigned long long,
725 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
726 i_size_read(inode));
727
24e17b5f 728 len = 1 << inode->i_blkbits;
9260dc6b
CH
729 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
730 PAGE_CACHE_SIZE);
731 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
732 page_dirty = p_offset / len;
24e17b5f 733
1da177e4
LT
734 bh = head = page_buffers(page);
735 do {
9260dc6b 736 if (offset >= end_offset)
1da177e4 737 break;
f6d6d4fc
CH
738 if (!buffer_uptodate(bh))
739 uptodate = 0;
740 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
741 done = 1;
1da177e4 742 continue;
f6d6d4fc
CH
743 }
744
2fa24f92
CH
745 if (buffer_unwritten(bh) || buffer_delay(bh) ||
746 buffer_mapped(bh)) {
9260dc6b 747 if (buffer_unwritten(bh))
0d882a36 748 type = XFS_IO_UNWRITTEN;
2fa24f92 749 else if (buffer_delay(bh))
0d882a36 750 type = XFS_IO_DELALLOC;
2fa24f92 751 else
0d882a36 752 type = XFS_IO_OVERWRITE;
9260dc6b 753
558e6891 754 if (!xfs_imap_valid(inode, imap, offset)) {
f6d6d4fc 755 done = 1;
9260dc6b
CH
756 continue;
757 }
758
ecff71e6 759 lock_buffer(bh);
0d882a36 760 if (type != XFS_IO_OVERWRITE)
2fa24f92 761 xfs_map_at_offset(inode, bh, imap, offset);
89f3b363
CH
762 xfs_add_to_ioend(inode, bh, offset, type,
763 ioendp, done);
764
9260dc6b
CH
765 page_dirty--;
766 count++;
767 } else {
2fa24f92 768 done = 1;
1da177e4 769 }
7336cea8 770 } while (offset += len, (bh = bh->b_this_page) != head);
1da177e4 771
f6d6d4fc
CH
772 if (uptodate && bh == head)
773 SetPageUptodate(page);
774
89f3b363 775 if (count) {
efceab1d
DC
776 if (--wbc->nr_to_write <= 0 &&
777 wbc->sync_mode == WB_SYNC_NONE)
89f3b363 778 done = 1;
1da177e4 779 }
89f3b363 780 xfs_start_page_writeback(page, !page_dirty, count);
f6d6d4fc
CH
781
782 return done;
10ce4444
CH
783 fail_unlock_page:
784 unlock_page(page);
785 fail:
786 return 1;
1da177e4
LT
787}
788
789/*
790 * Convert & write out a cluster of pages in the same extent as defined
791 * by mp and following the start page.
792 */
793STATIC void
794xfs_cluster_write(
795 struct inode *inode,
796 pgoff_t tindex,
207d0416 797 struct xfs_bmbt_irec *imap,
f6d6d4fc 798 xfs_ioend_t **ioendp,
1da177e4 799 struct writeback_control *wbc,
1da177e4
LT
800 pgoff_t tlast)
801{
10ce4444
CH
802 struct pagevec pvec;
803 int done = 0, i;
1da177e4 804
10ce4444
CH
805 pagevec_init(&pvec, 0);
806 while (!done && tindex <= tlast) {
807 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
808
809 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
1da177e4 810 break;
10ce4444
CH
811
812 for (i = 0; i < pagevec_count(&pvec); i++) {
813 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
2fa24f92 814 imap, ioendp, wbc);
10ce4444
CH
815 if (done)
816 break;
817 }
818
819 pagevec_release(&pvec);
820 cond_resched();
1da177e4
LT
821 }
822}
823
3ed3a434
DC
824STATIC void
825xfs_vm_invalidatepage(
826 struct page *page,
827 unsigned long offset)
828{
829 trace_xfs_invalidatepage(page->mapping->host, page, offset);
830 block_invalidatepage(page, offset);
831}
832
833/*
834 * If the page has delalloc buffers on it, we need to punch them out before we
835 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
836 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
837 * is done on that same region - the delalloc extent is returned when none is
838 * supposed to be there.
839 *
840 * We prevent this by truncating away the delalloc regions on the page before
841 * invalidating it. Because they are delalloc, we can do this without needing a
842 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
843 * truncation without a transaction as there is no space left for block
844 * reservation (typically why we see a ENOSPC in writeback).
845 *
846 * This is not a performance critical path, so for now just do the punching a
847 * buffer head at a time.
848 */
849STATIC void
850xfs_aops_discard_page(
851 struct page *page)
852{
853 struct inode *inode = page->mapping->host;
854 struct xfs_inode *ip = XFS_I(inode);
855 struct buffer_head *bh, *head;
856 loff_t offset = page_offset(page);
3ed3a434 857
0d882a36 858 if (!xfs_check_page_type(page, XFS_IO_DELALLOC))
3ed3a434
DC
859 goto out_invalidate;
860
e8c3753c
DC
861 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
862 goto out_invalidate;
863
4f10700a 864 xfs_alert(ip->i_mount,
3ed3a434
DC
865 "page discard on page %p, inode 0x%llx, offset %llu.",
866 page, ip->i_ino, offset);
867
868 xfs_ilock(ip, XFS_ILOCK_EXCL);
869 bh = head = page_buffers(page);
870 do {
3ed3a434 871 int error;
c726de44 872 xfs_fileoff_t start_fsb;
3ed3a434
DC
873
874 if (!buffer_delay(bh))
875 goto next_buffer;
876
c726de44
DC
877 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
878 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
3ed3a434
DC
879 if (error) {
880 /* something screwed, just bail */
e8c3753c 881 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
4f10700a 882 xfs_alert(ip->i_mount,
3ed3a434 883 "page discard unable to remove delalloc mapping.");
e8c3753c 884 }
3ed3a434
DC
885 break;
886 }
887next_buffer:
c726de44 888 offset += 1 << inode->i_blkbits;
3ed3a434
DC
889
890 } while ((bh = bh->b_this_page) != head);
891
892 xfs_iunlock(ip, XFS_ILOCK_EXCL);
893out_invalidate:
894 xfs_vm_invalidatepage(page, 0);
895 return;
896}
897
1da177e4 898/*
89f3b363
CH
899 * Write out a dirty page.
900 *
901 * For delalloc space on the page we need to allocate space and flush it.
902 * For unwritten space on the page we need to start the conversion to
903 * regular allocated space.
89f3b363 904 * For any other dirty buffer heads on the page we should flush them.
1da177e4 905 */
1da177e4 906STATIC int
89f3b363
CH
907xfs_vm_writepage(
908 struct page *page,
909 struct writeback_control *wbc)
1da177e4 910{
89f3b363 911 struct inode *inode = page->mapping->host;
f6d6d4fc 912 struct buffer_head *bh, *head;
207d0416 913 struct xfs_bmbt_irec imap;
f6d6d4fc 914 xfs_ioend_t *ioend = NULL, *iohead = NULL;
1da177e4 915 loff_t offset;
f6d6d4fc 916 unsigned int type;
1da177e4 917 __uint64_t end_offset;
bd1556a1 918 pgoff_t end_index, last_index;
ed1e7b7e 919 ssize_t len;
a206c817 920 int err, imap_valid = 0, uptodate = 1;
89f3b363 921 int count = 0;
a206c817 922 int nonblocking = 0;
89f3b363
CH
923
924 trace_xfs_writepage(inode, page, 0);
925
20cb52eb
CH
926 ASSERT(page_has_buffers(page));
927
89f3b363
CH
928 /*
929 * Refuse to write the page out if we are called from reclaim context.
930 *
d4f7a5cb
CH
931 * This avoids stack overflows when called from deeply used stacks in
932 * random callers for direct reclaim or memcg reclaim. We explicitly
933 * allow reclaim from kswapd as the stack usage there is relatively low.
89f3b363 934 *
94054fa3
MG
935 * This should never happen except in the case of a VM regression so
936 * warn about it.
89f3b363 937 */
94054fa3
MG
938 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
939 PF_MEMALLOC))
b5420f23 940 goto redirty;
1da177e4 941
89f3b363 942 /*
680a647b
CH
943 * Given that we do not allow direct reclaim to call us, we should
944 * never be called while in a filesystem transaction.
89f3b363 945 */
680a647b 946 if (WARN_ON(current->flags & PF_FSTRANS))
b5420f23 947 goto redirty;
89f3b363 948
1da177e4
LT
949 /* Is this page beyond the end of the file? */
950 offset = i_size_read(inode);
951 end_index = offset >> PAGE_CACHE_SHIFT;
952 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
953 if (page->index >= end_index) {
6b7a03f0
CH
954 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
955
956 /*
ff9a28f6
JK
957 * Skip the page if it is fully outside i_size, e.g. due to a
958 * truncate operation that is in progress. We must redirty the
959 * page so that reclaim stops reclaiming it. Otherwise
960 * xfs_vm_releasepage() is called on it and gets confused.
6b7a03f0 961 */
ff9a28f6
JK
962 if (page->index >= end_index + 1 || offset_into_page == 0)
963 goto redirty;
6b7a03f0
CH
964
965 /*
966 * The page straddles i_size. It must be zeroed out on each
967 * and every writepage invocation because it may be mmapped.
968 * "A file is mapped in multiples of the page size. For a file
969 * that is not a multiple of the page size, the remaining
970 * memory is zeroed when mapped, and writes to that region are
971 * not written out to the file."
972 */
973 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
1da177e4
LT
974 }
975
f6d6d4fc 976 end_offset = min_t(unsigned long long,
20cb52eb
CH
977 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
978 offset);
24e17b5f 979 len = 1 << inode->i_blkbits;
24e17b5f 980
24e17b5f 981 bh = head = page_buffers(page);
f6d6d4fc 982 offset = page_offset(page);
0d882a36 983 type = XFS_IO_OVERWRITE;
a206c817 984
dbcdde3e 985 if (wbc->sync_mode == WB_SYNC_NONE)
a206c817 986 nonblocking = 1;
f6d6d4fc 987
1da177e4 988 do {
6ac7248e
CH
989 int new_ioend = 0;
990
1da177e4
LT
991 if (offset >= end_offset)
992 break;
993 if (!buffer_uptodate(bh))
994 uptodate = 0;
1da177e4 995
3d9b02e3 996 /*
ece413f5
CH
997 * set_page_dirty dirties all buffers in a page, independent
998 * of their state. The dirty state however is entirely
999 * meaningless for holes (!mapped && uptodate), so skip
1000 * buffers covering holes here.
3d9b02e3
ES
1001 */
1002 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
3d9b02e3
ES
1003 imap_valid = 0;
1004 continue;
1005 }
1006
aeea1b1f 1007 if (buffer_unwritten(bh)) {
0d882a36
AR
1008 if (type != XFS_IO_UNWRITTEN) {
1009 type = XFS_IO_UNWRITTEN;
aeea1b1f 1010 imap_valid = 0;
1da177e4 1011 }
aeea1b1f 1012 } else if (buffer_delay(bh)) {
0d882a36
AR
1013 if (type != XFS_IO_DELALLOC) {
1014 type = XFS_IO_DELALLOC;
aeea1b1f 1015 imap_valid = 0;
1da177e4 1016 }
89f3b363 1017 } else if (buffer_uptodate(bh)) {
0d882a36
AR
1018 if (type != XFS_IO_OVERWRITE) {
1019 type = XFS_IO_OVERWRITE;
85da94c6
CH
1020 imap_valid = 0;
1021 }
aeea1b1f 1022 } else {
7d0fa3ec 1023 if (PageUptodate(page))
aeea1b1f 1024 ASSERT(buffer_mapped(bh));
7d0fa3ec
AR
1025 /*
1026 * This buffer is not uptodate and will not be
1027 * written to disk. Ensure that we will put any
1028 * subsequent writeable buffers into a new
1029 * ioend.
1030 */
1031 imap_valid = 0;
aeea1b1f
CH
1032 continue;
1033 }
d5cb48aa 1034
aeea1b1f
CH
1035 if (imap_valid)
1036 imap_valid = xfs_imap_valid(inode, &imap, offset);
1037 if (!imap_valid) {
1038 /*
1039 * If we didn't have a valid mapping then we need to
1040 * put the new mapping into a separate ioend structure.
1041 * This ensures non-contiguous extents always have
1042 * separate ioends, which is particularly important
1043 * for unwritten extent conversion at I/O completion
1044 * time.
1045 */
1046 new_ioend = 1;
1047 err = xfs_map_blocks(inode, offset, &imap, type,
1048 nonblocking);
1049 if (err)
1050 goto error;
1051 imap_valid = xfs_imap_valid(inode, &imap, offset);
1052 }
1053 if (imap_valid) {
ecff71e6 1054 lock_buffer(bh);
0d882a36 1055 if (type != XFS_IO_OVERWRITE)
aeea1b1f
CH
1056 xfs_map_at_offset(inode, bh, &imap, offset);
1057 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1058 new_ioend);
1059 count++;
1da177e4 1060 }
f6d6d4fc
CH
1061
1062 if (!iohead)
1063 iohead = ioend;
1064
1065 } while (offset += len, ((bh = bh->b_this_page) != head));
1da177e4
LT
1066
1067 if (uptodate && bh == head)
1068 SetPageUptodate(page);
1069
89f3b363 1070 xfs_start_page_writeback(page, 1, count);
1da177e4 1071
7bf7f352
DC
1072 /* if there is no IO to be submitted for this page, we are done */
1073 if (!ioend)
1074 return 0;
1075
1076 ASSERT(iohead);
1077
1078 /*
1079 * Any errors from this point onwards need tobe reported through the IO
1080 * completion path as we have marked the initial page as under writeback
1081 * and unlocked it.
1082 */
1083 if (imap_valid) {
bd1556a1
CH
1084 xfs_off_t end_index;
1085
1086 end_index = imap.br_startoff + imap.br_blockcount;
1087
1088 /* to bytes */
1089 end_index <<= inode->i_blkbits;
1090
1091 /* to pages */
1092 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1093
1094 /* check against file size */
1095 if (end_index > last_index)
1096 end_index = last_index;
8699bb0a 1097
207d0416 1098 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
2fa24f92 1099 wbc, end_index);
1da177e4
LT
1100 }
1101
281627df 1102
7bf7f352
DC
1103 /*
1104 * Reserve log space if we might write beyond the on-disk inode size.
1105 */
1106 err = 0;
1107 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1108 err = xfs_setfilesize_trans_alloc(ioend);
1109
1110 xfs_submit_ioend(wbc, iohead, err);
f6d6d4fc 1111
89f3b363 1112 return 0;
1da177e4
LT
1113
1114error:
f6d6d4fc
CH
1115 if (iohead)
1116 xfs_cancel_ioend(iohead);
1da177e4 1117
b5420f23
CH
1118 if (err == -EAGAIN)
1119 goto redirty;
1120
20cb52eb 1121 xfs_aops_discard_page(page);
89f3b363
CH
1122 ClearPageUptodate(page);
1123 unlock_page(page);
1da177e4 1124 return err;
f51623b2 1125
b5420f23 1126redirty:
f51623b2
NS
1127 redirty_page_for_writepage(wbc, page);
1128 unlock_page(page);
1129 return 0;
f51623b2
NS
1130}
1131
7d4fb40a
NS
1132STATIC int
1133xfs_vm_writepages(
1134 struct address_space *mapping,
1135 struct writeback_control *wbc)
1136{
b3aea4ed 1137 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
7d4fb40a
NS
1138 return generic_writepages(mapping, wbc);
1139}
1140
f51623b2
NS
1141/*
1142 * Called to move a page into cleanable state - and from there
89f3b363 1143 * to be released. The page should already be clean. We always
f51623b2
NS
1144 * have buffer heads in this call.
1145 *
89f3b363 1146 * Returns 1 if the page is ok to release, 0 otherwise.
f51623b2
NS
1147 */
1148STATIC int
238f4c54 1149xfs_vm_releasepage(
f51623b2
NS
1150 struct page *page,
1151 gfp_t gfp_mask)
1152{
20cb52eb 1153 int delalloc, unwritten;
f51623b2 1154
89f3b363 1155 trace_xfs_releasepage(page->mapping->host, page, 0);
238f4c54 1156
20cb52eb 1157 xfs_count_page_state(page, &delalloc, &unwritten);
f51623b2 1158
89f3b363 1159 if (WARN_ON(delalloc))
f51623b2 1160 return 0;
89f3b363 1161 if (WARN_ON(unwritten))
f51623b2
NS
1162 return 0;
1163
f51623b2
NS
1164 return try_to_free_buffers(page);
1165}
1166
1da177e4 1167STATIC int
c2536668 1168__xfs_get_blocks(
1da177e4
LT
1169 struct inode *inode,
1170 sector_t iblock,
1da177e4
LT
1171 struct buffer_head *bh_result,
1172 int create,
f2bde9b8 1173 int direct)
1da177e4 1174{
a206c817
CH
1175 struct xfs_inode *ip = XFS_I(inode);
1176 struct xfs_mount *mp = ip->i_mount;
1177 xfs_fileoff_t offset_fsb, end_fsb;
1178 int error = 0;
1179 int lockmode = 0;
207d0416 1180 struct xfs_bmbt_irec imap;
a206c817 1181 int nimaps = 1;
fdc7ed75
NS
1182 xfs_off_t offset;
1183 ssize_t size;
207d0416 1184 int new = 0;
a206c817
CH
1185
1186 if (XFS_FORCED_SHUTDOWN(mp))
1187 return -XFS_ERROR(EIO);
1da177e4 1188
fdc7ed75 1189 offset = (xfs_off_t)iblock << inode->i_blkbits;
c2536668
NS
1190 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1191 size = bh_result->b_size;
364f358a
LM
1192
1193 if (!create && direct && offset >= i_size_read(inode))
1194 return 0;
1195
507630b2
DC
1196 /*
1197 * Direct I/O is usually done on preallocated files, so try getting
1198 * a block mapping without an exclusive lock first. For buffered
1199 * writes we already have the exclusive iolock anyway, so avoiding
1200 * a lock roundtrip here by taking the ilock exclusive from the
1201 * beginning is a useful micro optimization.
1202 */
1203 if (create && !direct) {
a206c817
CH
1204 lockmode = XFS_ILOCK_EXCL;
1205 xfs_ilock(ip, lockmode);
1206 } else {
1207 lockmode = xfs_ilock_map_shared(ip);
1208 }
f2bde9b8 1209
d2c28191
DC
1210 ASSERT(offset <= mp->m_super->s_maxbytes);
1211 if (offset + size > mp->m_super->s_maxbytes)
1212 size = mp->m_super->s_maxbytes - offset;
a206c817
CH
1213 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1214 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1215
5c8ed202
DC
1216 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1217 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1da177e4 1218 if (error)
a206c817
CH
1219 goto out_unlock;
1220
1221 if (create &&
1222 (!nimaps ||
1223 (imap.br_startblock == HOLESTARTBLOCK ||
1224 imap.br_startblock == DELAYSTARTBLOCK))) {
aff3a9ed 1225 if (direct || xfs_get_extsz_hint(ip)) {
507630b2
DC
1226 /*
1227 * Drop the ilock in preparation for starting the block
1228 * allocation transaction. It will be retaken
1229 * exclusively inside xfs_iomap_write_direct for the
1230 * actual allocation.
1231 */
1232 xfs_iunlock(ip, lockmode);
a206c817
CH
1233 error = xfs_iomap_write_direct(ip, offset, size,
1234 &imap, nimaps);
507630b2
DC
1235 if (error)
1236 return -error;
d3bc815a 1237 new = 1;
a206c817 1238 } else {
507630b2
DC
1239 /*
1240 * Delalloc reservations do not require a transaction,
d3bc815a
DC
1241 * we can go on without dropping the lock here. If we
1242 * are allocating a new delalloc block, make sure that
1243 * we set the new flag so that we mark the buffer new so
1244 * that we know that it is newly allocated if the write
1245 * fails.
507630b2 1246 */
d3bc815a
DC
1247 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1248 new = 1;
a206c817 1249 error = xfs_iomap_write_delay(ip, offset, size, &imap);
507630b2
DC
1250 if (error)
1251 goto out_unlock;
1252
1253 xfs_iunlock(ip, lockmode);
a206c817 1254 }
a206c817
CH
1255
1256 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1257 } else if (nimaps) {
1258 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
507630b2 1259 xfs_iunlock(ip, lockmode);
a206c817
CH
1260 } else {
1261 trace_xfs_get_blocks_notfound(ip, offset, size);
1262 goto out_unlock;
1263 }
1da177e4 1264
207d0416
CH
1265 if (imap.br_startblock != HOLESTARTBLOCK &&
1266 imap.br_startblock != DELAYSTARTBLOCK) {
87cbc49c
NS
1267 /*
1268 * For unwritten extents do not report a disk address on
1da177e4
LT
1269 * the read case (treat as if we're reading into a hole).
1270 */
207d0416
CH
1271 if (create || !ISUNWRITTEN(&imap))
1272 xfs_map_buffer(inode, bh_result, &imap, offset);
1273 if (create && ISUNWRITTEN(&imap)) {
1da177e4
LT
1274 if (direct)
1275 bh_result->b_private = inode;
1276 set_buffer_unwritten(bh_result);
1da177e4
LT
1277 }
1278 }
1279
c2536668
NS
1280 /*
1281 * If this is a realtime file, data may be on a different device.
1282 * to that pointed to from the buffer_head b_bdev currently.
1283 */
046f1685 1284 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1da177e4 1285
c2536668 1286 /*
549054af
DC
1287 * If we previously allocated a block out beyond eof and we are now
1288 * coming back to use it then we will need to flag it as new even if it
1289 * has a disk address.
1290 *
1291 * With sub-block writes into unwritten extents we also need to mark
1292 * the buffer as new so that the unwritten parts of the buffer gets
1293 * correctly zeroed.
1da177e4
LT
1294 */
1295 if (create &&
1296 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
549054af 1297 (offset >= i_size_read(inode)) ||
207d0416 1298 (new || ISUNWRITTEN(&imap))))
1da177e4 1299 set_buffer_new(bh_result);
1da177e4 1300
207d0416 1301 if (imap.br_startblock == DELAYSTARTBLOCK) {
1da177e4
LT
1302 BUG_ON(direct);
1303 if (create) {
1304 set_buffer_uptodate(bh_result);
1305 set_buffer_mapped(bh_result);
1306 set_buffer_delay(bh_result);
1307 }
1308 }
1309
2b8f12b7
CH
1310 /*
1311 * If this is O_DIRECT or the mpage code calling tell them how large
1312 * the mapping is, so that we can avoid repeated get_blocks calls.
1313 */
c2536668 1314 if (direct || size > (1 << inode->i_blkbits)) {
2b8f12b7
CH
1315 xfs_off_t mapping_size;
1316
1317 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1318 mapping_size <<= inode->i_blkbits;
1319
1320 ASSERT(mapping_size > 0);
1321 if (mapping_size > size)
1322 mapping_size = size;
1323 if (mapping_size > LONG_MAX)
1324 mapping_size = LONG_MAX;
1325
1326 bh_result->b_size = mapping_size;
1da177e4
LT
1327 }
1328
1329 return 0;
a206c817
CH
1330
1331out_unlock:
1332 xfs_iunlock(ip, lockmode);
1333 return -error;
1da177e4
LT
1334}
1335
1336int
c2536668 1337xfs_get_blocks(
1da177e4
LT
1338 struct inode *inode,
1339 sector_t iblock,
1340 struct buffer_head *bh_result,
1341 int create)
1342{
f2bde9b8 1343 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1da177e4
LT
1344}
1345
1346STATIC int
e4c573bb 1347xfs_get_blocks_direct(
1da177e4
LT
1348 struct inode *inode,
1349 sector_t iblock,
1da177e4
LT
1350 struct buffer_head *bh_result,
1351 int create)
1352{
f2bde9b8 1353 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1da177e4
LT
1354}
1355
209fb87a
CH
1356/*
1357 * Complete a direct I/O write request.
1358 *
1359 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1360 * need to issue a transaction to convert the range from unwritten to written
1361 * extents. In case this is regular synchronous I/O we just call xfs_end_io
25985edc 1362 * to do this and we are done. But in case this was a successful AIO
209fb87a
CH
1363 * request this handler is called from interrupt context, from which we
1364 * can't start transactions. In that case offload the I/O completion to
1365 * the workqueues we also use for buffered I/O completion.
1366 */
f0973863 1367STATIC void
209fb87a
CH
1368xfs_end_io_direct_write(
1369 struct kiocb *iocb,
1370 loff_t offset,
1371 ssize_t size,
1372 void *private,
1373 int ret,
1374 bool is_async)
f0973863 1375{
209fb87a 1376 struct xfs_ioend *ioend = iocb->private;
f0973863 1377
2813d682
CH
1378 /*
1379 * While the generic direct I/O code updates the inode size, it does
1380 * so only after the end_io handler is called, which means our
1381 * end_io handler thinks the on-disk size is outside the in-core
1382 * size. To prevent this just update it a little bit earlier here.
1383 */
1384 if (offset + size > i_size_read(ioend->io_inode))
1385 i_size_write(ioend->io_inode, offset + size);
1386
f0973863 1387 /*
209fb87a
CH
1388 * blockdev_direct_IO can return an error even after the I/O
1389 * completion handler was called. Thus we need to protect
1390 * against double-freeing.
f0973863 1391 */
209fb87a
CH
1392 iocb->private = NULL;
1393
ba87ea69
LM
1394 ioend->io_offset = offset;
1395 ioend->io_size = size;
c859cdd1
CH
1396 ioend->io_iocb = iocb;
1397 ioend->io_result = ret;
209fb87a 1398 if (private && size > 0)
0d882a36 1399 ioend->io_type = XFS_IO_UNWRITTEN;
209fb87a
CH
1400
1401 if (is_async) {
c859cdd1 1402 ioend->io_isasync = 1;
209fb87a 1403 xfs_finish_ioend(ioend);
f0973863 1404 } else {
209fb87a 1405 xfs_finish_ioend_sync(ioend);
f0973863 1406 }
f0973863
CH
1407}
1408
1da177e4 1409STATIC ssize_t
e4c573bb 1410xfs_vm_direct_IO(
1da177e4
LT
1411 int rw,
1412 struct kiocb *iocb,
1413 const struct iovec *iov,
1414 loff_t offset,
1415 unsigned long nr_segs)
1416{
209fb87a
CH
1417 struct inode *inode = iocb->ki_filp->f_mapping->host;
1418 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
281627df 1419 struct xfs_ioend *ioend = NULL;
209fb87a
CH
1420 ssize_t ret;
1421
1422 if (rw & WRITE) {
281627df
CH
1423 size_t size = iov_length(iov, nr_segs);
1424
1425 /*
437a255a
DC
1426 * We cannot preallocate a size update transaction here as we
1427 * don't know whether allocation is necessary or not. Hence we
1428 * can only tell IO completion that one is necessary if we are
1429 * not doing unwritten extent conversion.
281627df 1430 */
0d882a36 1431 iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT);
437a255a 1432 if (offset + size > XFS_I(inode)->i_d.di_size)
281627df 1433 ioend->io_isdirect = 1;
209fb87a 1434
eafdc7d1
CH
1435 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1436 offset, nr_segs,
1437 xfs_get_blocks_direct,
1438 xfs_end_io_direct_write, NULL, 0);
209fb87a 1439 if (ret != -EIOCBQUEUED && iocb->private)
437a255a 1440 goto out_destroy_ioend;
209fb87a 1441 } else {
eafdc7d1
CH
1442 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1443 offset, nr_segs,
1444 xfs_get_blocks_direct,
1445 NULL, NULL, 0);
209fb87a 1446 }
f0973863 1447
f0973863 1448 return ret;
281627df 1449
281627df
CH
1450out_destroy_ioend:
1451 xfs_destroy_ioend(ioend);
1452 return ret;
1da177e4
LT
1453}
1454
d3bc815a
DC
1455/*
1456 * Punch out the delalloc blocks we have already allocated.
1457 *
1458 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1459 * as the page is still locked at this point.
1460 */
1461STATIC void
1462xfs_vm_kill_delalloc_range(
1463 struct inode *inode,
1464 loff_t start,
1465 loff_t end)
1466{
1467 struct xfs_inode *ip = XFS_I(inode);
1468 xfs_fileoff_t start_fsb;
1469 xfs_fileoff_t end_fsb;
1470 int error;
1471
1472 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1473 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1474 if (end_fsb <= start_fsb)
1475 return;
1476
1477 xfs_ilock(ip, XFS_ILOCK_EXCL);
1478 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1479 end_fsb - start_fsb);
1480 if (error) {
1481 /* something screwed, just bail */
1482 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1483 xfs_alert(ip->i_mount,
1484 "xfs_vm_write_failed: unable to clean up ino %lld",
1485 ip->i_ino);
1486 }
1487 }
1488 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1489}
1490
fa9b227e
CH
1491STATIC void
1492xfs_vm_write_failed(
d3bc815a
DC
1493 struct inode *inode,
1494 struct page *page,
1495 loff_t pos,
1496 unsigned len)
fa9b227e 1497{
d3bc815a
DC
1498 loff_t block_offset = pos & PAGE_MASK;
1499 loff_t block_start;
1500 loff_t block_end;
1501 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1502 loff_t to = from + len;
1503 struct buffer_head *bh, *head;
fa9b227e 1504
d3bc815a 1505 ASSERT(block_offset + from == pos);
c726de44 1506
d3bc815a
DC
1507 head = page_buffers(page);
1508 block_start = 0;
1509 for (bh = head; bh != head || !block_start;
1510 bh = bh->b_this_page, block_start = block_end,
1511 block_offset += bh->b_size) {
1512 block_end = block_start + bh->b_size;
c726de44 1513
d3bc815a
DC
1514 /* skip buffers before the write */
1515 if (block_end <= from)
1516 continue;
1517
1518 /* if the buffer is after the write, we're done */
1519 if (block_start >= to)
1520 break;
1521
1522 if (!buffer_delay(bh))
1523 continue;
1524
1525 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1526 continue;
1527
1528 xfs_vm_kill_delalloc_range(inode, block_offset,
1529 block_offset + bh->b_size);
fa9b227e 1530 }
d3bc815a 1531
fa9b227e
CH
1532}
1533
d3bc815a
DC
1534/*
1535 * This used to call block_write_begin(), but it unlocks and releases the page
1536 * on error, and we need that page to be able to punch stale delalloc blocks out
1537 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1538 * the appropriate point.
1539 */
f51623b2 1540STATIC int
d79689c7 1541xfs_vm_write_begin(
f51623b2 1542 struct file *file,
d79689c7
NP
1543 struct address_space *mapping,
1544 loff_t pos,
1545 unsigned len,
1546 unsigned flags,
1547 struct page **pagep,
1548 void **fsdata)
f51623b2 1549{
d3bc815a
DC
1550 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1551 struct page *page;
1552 int status;
155130a4 1553
d3bc815a
DC
1554 ASSERT(len <= PAGE_CACHE_SIZE);
1555
1556 page = grab_cache_page_write_begin(mapping, index,
1557 flags | AOP_FLAG_NOFS);
1558 if (!page)
1559 return -ENOMEM;
1560
1561 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1562 if (unlikely(status)) {
1563 struct inode *inode = mapping->host;
1564
1565 xfs_vm_write_failed(inode, page, pos, len);
1566 unlock_page(page);
1567
1568 if (pos + len > i_size_read(inode))
1569 truncate_pagecache(inode, pos + len, i_size_read(inode));
1570
1571 page_cache_release(page);
1572 page = NULL;
1573 }
1574
1575 *pagep = page;
1576 return status;
fa9b227e
CH
1577}
1578
d3bc815a
DC
1579/*
1580 * On failure, we only need to kill delalloc blocks beyond EOF because they
1581 * will never be written. For blocks within EOF, generic_write_end() zeros them
1582 * so they are safe to leave alone and be written with all the other valid data.
1583 */
fa9b227e
CH
1584STATIC int
1585xfs_vm_write_end(
1586 struct file *file,
1587 struct address_space *mapping,
1588 loff_t pos,
1589 unsigned len,
1590 unsigned copied,
1591 struct page *page,
1592 void *fsdata)
1593{
1594 int ret;
155130a4 1595
d3bc815a
DC
1596 ASSERT(len <= PAGE_CACHE_SIZE);
1597
fa9b227e 1598 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
d3bc815a
DC
1599 if (unlikely(ret < len)) {
1600 struct inode *inode = mapping->host;
1601 size_t isize = i_size_read(inode);
1602 loff_t to = pos + len;
1603
1604 if (to > isize) {
1605 truncate_pagecache(inode, to, isize);
1606 xfs_vm_kill_delalloc_range(inode, isize, to);
1607 }
1608 }
155130a4 1609 return ret;
f51623b2 1610}
1da177e4
LT
1611
1612STATIC sector_t
e4c573bb 1613xfs_vm_bmap(
1da177e4
LT
1614 struct address_space *mapping,
1615 sector_t block)
1616{
1617 struct inode *inode = (struct inode *)mapping->host;
739bfb2a 1618 struct xfs_inode *ip = XFS_I(inode);
1da177e4 1619
cca28fb8 1620 trace_xfs_vm_bmap(XFS_I(inode));
126468b1 1621 xfs_ilock(ip, XFS_IOLOCK_SHARED);
4bc1ea6b 1622 filemap_write_and_wait(mapping);
126468b1 1623 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
c2536668 1624 return generic_block_bmap(mapping, block, xfs_get_blocks);
1da177e4
LT
1625}
1626
1627STATIC int
e4c573bb 1628xfs_vm_readpage(
1da177e4
LT
1629 struct file *unused,
1630 struct page *page)
1631{
c2536668 1632 return mpage_readpage(page, xfs_get_blocks);
1da177e4
LT
1633}
1634
1635STATIC int
e4c573bb 1636xfs_vm_readpages(
1da177e4
LT
1637 struct file *unused,
1638 struct address_space *mapping,
1639 struct list_head *pages,
1640 unsigned nr_pages)
1641{
c2536668 1642 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1da177e4
LT
1643}
1644
f5e54d6e 1645const struct address_space_operations xfs_address_space_operations = {
e4c573bb
NS
1646 .readpage = xfs_vm_readpage,
1647 .readpages = xfs_vm_readpages,
1648 .writepage = xfs_vm_writepage,
7d4fb40a 1649 .writepages = xfs_vm_writepages,
238f4c54
NS
1650 .releasepage = xfs_vm_releasepage,
1651 .invalidatepage = xfs_vm_invalidatepage,
d79689c7 1652 .write_begin = xfs_vm_write_begin,
fa9b227e 1653 .write_end = xfs_vm_write_end,
e4c573bb
NS
1654 .bmap = xfs_vm_bmap,
1655 .direct_IO = xfs_vm_direct_IO,
e965f963 1656 .migratepage = buffer_migrate_page,
bddaafa1 1657 .is_partially_uptodate = block_is_partially_uptodate,
aa261f54 1658 .error_remove_page = generic_error_remove_page,
1da177e4 1659};