projects / deliverable / linux.git / blob
? search:


fa47e43b8b41a525843c57cb4591ed9ffd334141
[deliverable/linux.git] / fs / xfs / linux-2.6 / xfs_aops.c
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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
7 * published by the Free Software Foundation.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_bit.h"
20 #include "xfs_log.h"
21 #include "xfs_inum.h"
22 #include "xfs_sb.h"
23 #include "xfs_ag.h"
24 #include "xfs_dir2.h"
25 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dir2_sf.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_alloc.h"
36 #include "xfs_btree.h"
37 #include "xfs_error.h"
38 #include "xfs_rw.h"
39 #include "xfs_iomap.h"
40 #include "xfs_vnodeops.h"
41 #include <linux/mpage.h>
42 #include <linux/pagevec.h>
43 #include <linux/writeback.h>
44
45 STATIC void
46 xfs_count_page_state(
47 struct page *page,
48 int *delalloc,
49 int *unmapped,
50 int *unwritten)
51 {
52 struct buffer_head *bh, *head;
53
54 *delalloc = *unmapped = *unwritten = 0;
55
56 bh = head = page_buffers(page);
57 do {
58 if (buffer_uptodate(bh) && !buffer_mapped(bh))
59 (*unmapped) = 1;
60 else if (buffer_unwritten(bh))
61 (*unwritten) = 1;
62 else if (buffer_delay(bh))
63 (*delalloc) = 1;
64 } while ((bh = bh->b_this_page) != head);
65 }
66
67 #if defined(XFS_RW_TRACE)
68 void
69 xfs_page_trace(
70 int tag,
71 struct inode *inode,
72 struct page *page,
73 unsigned long pgoff)
74 {
75 xfs_inode_t *ip;
76 bhv_vnode_t *vp = vn_from_inode(inode);
77 loff_t isize = i_size_read(inode);
78 loff_t offset = page_offset(page);
79 int delalloc = -1, unmapped = -1, unwritten = -1;
80
81 if (page_has_buffers(page))
82 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
83
84 ip = xfs_vtoi(vp);
85 if (!ip->i_rwtrace)
86 return;
87
88 ktrace_enter(ip->i_rwtrace,
89 (void *)((unsigned long)tag),
90 (void *)ip,
91 (void *)inode,
92 (void *)page,
93 (void *)pgoff,
94 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
95 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
96 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
97 (void *)((unsigned long)(isize & 0xffffffff)),
98 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
99 (void *)((unsigned long)(offset & 0xffffffff)),
100 (void *)((unsigned long)delalloc),
101 (void *)((unsigned long)unmapped),
102 (void *)((unsigned long)unwritten),
103 (void *)((unsigned long)current_pid()),
104 (void *)NULL);
105 }
106 #else
107 #define xfs_page_trace(tag, inode, page, pgoff)
108 #endif
109
110 STATIC struct block_device *
111 xfs_find_bdev_for_inode(
112 struct xfs_inode *ip)
113 {
114 struct xfs_mount *mp = ip->i_mount;
115
116 if (XFS_IS_REALTIME_INODE(ip))
117 return mp->m_rtdev_targp->bt_bdev;
118 else
119 return mp->m_ddev_targp->bt_bdev;
120 }
121
122 /*
123 * Schedule IO completion handling on a xfsdatad if this was
124 * the final hold on this ioend. If we are asked to wait,
125 * flush the workqueue.
126 */
127 STATIC void
128 xfs_finish_ioend(
129 xfs_ioend_t *ioend,
130 int wait)
131 {
132 if (atomic_dec_and_test(&ioend->io_remaining)) {
133 queue_work(xfsdatad_workqueue, &ioend->io_work);
134 if (wait)
135 flush_workqueue(xfsdatad_workqueue);
136 }
137 }
138
139 /*
140 * We're now finished for good with this ioend structure.
141 * Update the page state via the associated buffer_heads,
142 * release holds on the inode and bio, and finally free
143 * up memory. Do not use the ioend after this.
144 */
145 STATIC void
146 xfs_destroy_ioend(
147 xfs_ioend_t *ioend)
148 {
149 struct buffer_head *bh, *next;
150
151 for (bh = ioend->io_buffer_head; bh; bh = next) {
152 next = bh->b_private;
153 bh->b_end_io(bh, !ioend->io_error);
154 }
155 if (unlikely(ioend->io_error)) {
156 vn_ioerror(XFS_I(ioend->io_inode), ioend->io_error,
157 __FILE__,__LINE__);
158 }
159 vn_iowake(XFS_I(ioend->io_inode));
160 mempool_free(ioend, xfs_ioend_pool);
161 }
162
163 /*
164 * Update on-disk file size now that data has been written to disk.
165 * The current in-memory file size is i_size. If a write is beyond
166 * eof i_new_size will be the intended file size until i_size is
167 * updated. If this write does not extend all the way to the valid
168 * file size then restrict this update to the end of the write.
169 */
170 STATIC void
171 xfs_setfilesize(
172 xfs_ioend_t *ioend)
173 {
174 xfs_inode_t *ip = XFS_I(ioend->io_inode);
175 xfs_fsize_t isize;
176 xfs_fsize_t bsize;
177
178 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
179 ASSERT(ioend->io_type != IOMAP_READ);
180
181 if (unlikely(ioend->io_error))
182 return;
183
184 bsize = ioend->io_offset + ioend->io_size;
185
186 xfs_ilock(ip, XFS_ILOCK_EXCL);
187
188 isize = MAX(ip->i_size, ip->i_new_size);
189 isize = MIN(isize, bsize);
190
191 if (ip->i_d.di_size < isize) {
192 ip->i_d.di_size = isize;
193 ip->i_update_core = 1;
194 ip->i_update_size = 1;
195 mark_inode_dirty_sync(ioend->io_inode);
196 }
197
198 xfs_iunlock(ip, XFS_ILOCK_EXCL);
199 }
200
201 /*
202 * Buffered IO write completion for delayed allocate extents.
203 */
204 STATIC void
205 xfs_end_bio_delalloc(
206 struct work_struct *work)
207 {
208 xfs_ioend_t *ioend =
209 container_of(work, xfs_ioend_t, io_work);
210
211 xfs_setfilesize(ioend);
212 xfs_destroy_ioend(ioend);
213 }
214
215 /*
216 * Buffered IO write completion for regular, written extents.
217 */
218 STATIC void
219 xfs_end_bio_written(
220 struct work_struct *work)
221 {
222 xfs_ioend_t *ioend =
223 container_of(work, xfs_ioend_t, io_work);
224
225 xfs_setfilesize(ioend);
226 xfs_destroy_ioend(ioend);
227 }
228
229 /*
230 * IO write completion for unwritten extents.
231 *
232 * Issue transactions to convert a buffer range from unwritten
233 * to written extents.
234 */
235 STATIC void
236 xfs_end_bio_unwritten(
237 struct work_struct *work)
238 {
239 xfs_ioend_t *ioend =
240 container_of(work, xfs_ioend_t, io_work);
241 struct xfs_inode *ip = XFS_I(ioend->io_inode);
242 xfs_off_t offset = ioend->io_offset;
243 size_t size = ioend->io_size;
244
245 if (likely(!ioend->io_error)) {
246 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
247 int error;
248 error = xfs_iomap_write_unwritten(ip, offset, size);
249 if (error)
250 ioend->io_error = error;
251 }
252 xfs_setfilesize(ioend);
253 }
254 xfs_destroy_ioend(ioend);
255 }
256
257 /*
258 * IO read completion for regular, written extents.
259 */
260 STATIC void
261 xfs_end_bio_read(
262 struct work_struct *work)
263 {
264 xfs_ioend_t *ioend =
265 container_of(work, xfs_ioend_t, io_work);
266
267 xfs_destroy_ioend(ioend);
268 }
269
270 /*
271 * Allocate and initialise an IO completion structure.
272 * We need to track unwritten extent write completion here initially.
273 * We'll need to extend this for updating the ondisk inode size later
274 * (vs. incore size).
275 */
276 STATIC xfs_ioend_t *
277 xfs_alloc_ioend(
278 struct inode *inode,
279 unsigned int type)
280 {
281 xfs_ioend_t *ioend;
282
283 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
284
285 /*
286 * Set the count to 1 initially, which will prevent an I/O
287 * completion callback from happening before we have started
288 * all the I/O from calling the completion routine too early.
289 */
290 atomic_set(&ioend->io_remaining, 1);
291 ioend->io_error = 0;
292 ioend->io_list = NULL;
293 ioend->io_type = type;
294 ioend->io_inode = inode;
295 ioend->io_buffer_head = NULL;
296 ioend->io_buffer_tail = NULL;
297 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
298 ioend->io_offset = 0;
299 ioend->io_size = 0;
300
301 if (type == IOMAP_UNWRITTEN)
302 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
303 else if (type == IOMAP_DELAY)
304 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
305 else if (type == IOMAP_READ)
306 INIT_WORK(&ioend->io_work, xfs_end_bio_read);
307 else
308 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
309
310 return ioend;
311 }
312
313 STATIC int
314 xfs_map_blocks(
315 struct inode *inode,
316 loff_t offset,
317 ssize_t count,
318 xfs_iomap_t *mapp,
319 int flags)
320 {
321 xfs_inode_t *ip = XFS_I(inode);
322 int error, nmaps = 1;
323
324 error = xfs_iomap(ip, offset, count,
325 flags, mapp, &nmaps);
326 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
327 xfs_iflags_set(ip, XFS_IMODIFIED);
328 return -error;
329 }
330
331 STATIC_INLINE int
332 xfs_iomap_valid(
333 xfs_iomap_t *iomapp,
334 loff_t offset)
335 {
336 return offset >= iomapp->iomap_offset &&
337 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
338 }
339
340 /*
341 * BIO completion handler for buffered IO.
342 */
343 STATIC void
344 xfs_end_bio(
345 struct bio *bio,
346 int error)
347 {
348 xfs_ioend_t *ioend = bio->bi_private;
349
350 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
351 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
352
353 /* Toss bio and pass work off to an xfsdatad thread */
354 bio->bi_private = NULL;
355 bio->bi_end_io = NULL;
356 bio_put(bio);
357
358 xfs_finish_ioend(ioend, 0);
359 }
360
361 STATIC void
362 xfs_submit_ioend_bio(
363 xfs_ioend_t *ioend,
364 struct bio *bio)
365 {
366 atomic_inc(&ioend->io_remaining);
367
368 bio->bi_private = ioend;
369 bio->bi_end_io = xfs_end_bio;
370
371 submit_bio(WRITE, bio);
372 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
373 bio_put(bio);
374 }
375
376 STATIC struct bio *
377 xfs_alloc_ioend_bio(
378 struct buffer_head *bh)
379 {
380 struct bio *bio;
381 int nvecs = bio_get_nr_vecs(bh->b_bdev);
382
383 do {
384 bio = bio_alloc(GFP_NOIO, nvecs);
385 nvecs >>= 1;
386 } while (!bio);
387
388 ASSERT(bio->bi_private == NULL);
389 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
390 bio->bi_bdev = bh->b_bdev;
391 bio_get(bio);
392 return bio;
393 }
394
395 STATIC void
396 xfs_start_buffer_writeback(
397 struct buffer_head *bh)
398 {
399 ASSERT(buffer_mapped(bh));
400 ASSERT(buffer_locked(bh));
401 ASSERT(!buffer_delay(bh));
402 ASSERT(!buffer_unwritten(bh));
403
404 mark_buffer_async_write(bh);
405 set_buffer_uptodate(bh);
406 clear_buffer_dirty(bh);
407 }
408
409 STATIC void
410 xfs_start_page_writeback(
411 struct page *page,
412 int clear_dirty,
413 int buffers)
414 {
415 ASSERT(PageLocked(page));
416 ASSERT(!PageWriteback(page));
417 if (clear_dirty)
418 clear_page_dirty_for_io(page);
419 set_page_writeback(page);
420 unlock_page(page);
421 /* If no buffers on the page are to be written, finish it here */
422 if (!buffers)
423 end_page_writeback(page);
424 }
425
426 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
427 {
428 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
429 }
430
431 /*
432 * Submit all of the bios for all of the ioends we have saved up, covering the
433 * initial writepage page and also any probed pages.
434 *
435 * Because we may have multiple ioends spanning a page, we need to start
436 * writeback on all the buffers before we submit them for I/O. If we mark the
437 * buffers as we got, then we can end up with a page that only has buffers
438 * marked async write and I/O complete on can occur before we mark the other
439 * buffers async write.
440 *
441 * The end result of this is that we trip a bug in end_page_writeback() because
442 * we call it twice for the one page as the code in end_buffer_async_write()
443 * assumes that all buffers on the page are started at the same time.
444 *
445 * The fix is two passes across the ioend list - one to start writeback on the
446 * buffer_heads, and then submit them for I/O on the second pass.
447 */
448 STATIC void
449 xfs_submit_ioend(
450 xfs_ioend_t *ioend)
451 {
452 xfs_ioend_t *head = ioend;
453 xfs_ioend_t *next;
454 struct buffer_head *bh;
455 struct bio *bio;
456 sector_t lastblock = 0;
457
458 /* Pass 1 - start writeback */
459 do {
460 next = ioend->io_list;
461 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
462 xfs_start_buffer_writeback(bh);
463 }
464 } while ((ioend = next) != NULL);
465
466 /* Pass 2 - submit I/O */
467 ioend = head;
468 do {
469 next = ioend->io_list;
470 bio = NULL;
471
472 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
473
474 if (!bio) {
475 retry:
476 bio = xfs_alloc_ioend_bio(bh);
477 } else if (bh->b_blocknr != lastblock + 1) {
478 xfs_submit_ioend_bio(ioend, bio);
479 goto retry;
480 }
481
482 if (bio_add_buffer(bio, bh) != bh->b_size) {
483 xfs_submit_ioend_bio(ioend, bio);
484 goto retry;
485 }
486
487 lastblock = bh->b_blocknr;
488 }
489 if (bio)
490 xfs_submit_ioend_bio(ioend, bio);
491 xfs_finish_ioend(ioend, 0);
492 } while ((ioend = next) != NULL);
493 }
494
495 /*
496 * Cancel submission of all buffer_heads so far in this endio.
497 * Toss the endio too. Only ever called for the initial page
498 * in a writepage request, so only ever one page.
499 */
500 STATIC void
501 xfs_cancel_ioend(
502 xfs_ioend_t *ioend)
503 {
504 xfs_ioend_t *next;
505 struct buffer_head *bh, *next_bh;
506
507 do {
508 next = ioend->io_list;
509 bh = ioend->io_buffer_head;
510 do {
511 next_bh = bh->b_private;
512 clear_buffer_async_write(bh);
513 unlock_buffer(bh);
514 } while ((bh = next_bh) != NULL);
515
516 vn_iowake(XFS_I(ioend->io_inode));
517 mempool_free(ioend, xfs_ioend_pool);
518 } while ((ioend = next) != NULL);
519 }
520
521 /*
522 * Test to see if we've been building up a completion structure for
523 * earlier buffers -- if so, we try to append to this ioend if we
524 * can, otherwise we finish off any current ioend and start another.
525 * Return true if we've finished the given ioend.
526 */
527 STATIC void
528 xfs_add_to_ioend(
529 struct inode *inode,
530 struct buffer_head *bh,
531 xfs_off_t offset,
532 unsigned int type,
533 xfs_ioend_t **result,
534 int need_ioend)
535 {
536 xfs_ioend_t *ioend = *result;
537
538 if (!ioend || need_ioend || type != ioend->io_type) {
539 xfs_ioend_t *previous = *result;
540
541 ioend = xfs_alloc_ioend(inode, type);
542 ioend->io_offset = offset;
543 ioend->io_buffer_head = bh;
544 ioend->io_buffer_tail = bh;
545 if (previous)
546 previous->io_list = ioend;
547 *result = ioend;
548 } else {
549 ioend->io_buffer_tail->b_private = bh;
550 ioend->io_buffer_tail = bh;
551 }
552
553 bh->b_private = NULL;
554 ioend->io_size += bh->b_size;
555 }
556
557 STATIC void
558 xfs_map_buffer(
559 struct buffer_head *bh,
560 xfs_iomap_t *mp,
561 xfs_off_t offset,
562 uint block_bits)
563 {
564 sector_t bn;
565
566 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
567
568 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
569 ((offset - mp->iomap_offset) >> block_bits);
570
571 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
572
573 bh->b_blocknr = bn;
574 set_buffer_mapped(bh);
575 }
576
577 STATIC void
578 xfs_map_at_offset(
579 struct buffer_head *bh,
580 loff_t offset,
581 int block_bits,
582 xfs_iomap_t *iomapp)
583 {
584 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
585 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
586
587 lock_buffer(bh);
588 xfs_map_buffer(bh, iomapp, offset, block_bits);
589 bh->b_bdev = iomapp->iomap_target->bt_bdev;
590 set_buffer_mapped(bh);
591 clear_buffer_delay(bh);
592 clear_buffer_unwritten(bh);
593 }
594
595 /*
596 * Look for a page at index that is suitable for clustering.
597 */
598 STATIC unsigned int
599 xfs_probe_page(
600 struct page *page,
601 unsigned int pg_offset,
602 int mapped)
603 {
604 int ret = 0;
605
606 if (PageWriteback(page))
607 return 0;
608
609 if (page->mapping && PageDirty(page)) {
610 if (page_has_buffers(page)) {
611 struct buffer_head *bh, *head;
612
613 bh = head = page_buffers(page);
614 do {
615 if (!buffer_uptodate(bh))
616 break;
617 if (mapped != buffer_mapped(bh))
618 break;
619 ret += bh->b_size;
620 if (ret >= pg_offset)
621 break;
622 } while ((bh = bh->b_this_page) != head);
623 } else
624 ret = mapped ? 0 : PAGE_CACHE_SIZE;
625 }
626
627 return ret;
628 }
629
630 STATIC size_t
631 xfs_probe_cluster(
632 struct inode *inode,
633 struct page *startpage,
634 struct buffer_head *bh,
635 struct buffer_head *head,
636 int mapped)
637 {
638 struct pagevec pvec;
639 pgoff_t tindex, tlast, tloff;
640 size_t total = 0;
641 int done = 0, i;
642
643 /* First sum forwards in this page */
644 do {
645 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
646 return total;
647 total += bh->b_size;
648 } while ((bh = bh->b_this_page) != head);
649
650 /* if we reached the end of the page, sum forwards in following pages */
651 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
652 tindex = startpage->index + 1;
653
654 /* Prune this back to avoid pathological behavior */
655 tloff = min(tlast, startpage->index + 64);
656
657 pagevec_init(&pvec, 0);
658 while (!done && tindex <= tloff) {
659 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
660
661 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
662 break;
663
664 for (i = 0; i < pagevec_count(&pvec); i++) {
665 struct page *page = pvec.pages[i];
666 size_t pg_offset, pg_len = 0;
667
668 if (tindex == tlast) {
669 pg_offset =
670 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
671 if (!pg_offset) {
672 done = 1;
673 break;
674 }
675 } else
676 pg_offset = PAGE_CACHE_SIZE;
677
678 if (page->index == tindex && trylock_page(page)) {
679 pg_len = xfs_probe_page(page, pg_offset, mapped);
680 unlock_page(page);
681 }
682
683 if (!pg_len) {
684 done = 1;
685 break;
686 }
687
688 total += pg_len;
689 tindex++;
690 }
691
692 pagevec_release(&pvec);
693 cond_resched();
694 }
695
696 return total;
697 }
698
699 /*
700 * Test if a given page is suitable for writing as part of an unwritten
701 * or delayed allocate extent.
702 */
703 STATIC int
704 xfs_is_delayed_page(
705 struct page *page,
706 unsigned int type)
707 {
708 if (PageWriteback(page))
709 return 0;
710
711 if (page->mapping && page_has_buffers(page)) {
712 struct buffer_head *bh, *head;
713 int acceptable = 0;
714
715 bh = head = page_buffers(page);
716 do {
717 if (buffer_unwritten(bh))
718 acceptable = (type == IOMAP_UNWRITTEN);
719 else if (buffer_delay(bh))
720 acceptable = (type == IOMAP_DELAY);
721 else if (buffer_dirty(bh) && buffer_mapped(bh))
722 acceptable = (type == IOMAP_NEW);
723 else
724 break;
725 } while ((bh = bh->b_this_page) != head);
726
727 if (acceptable)
728 return 1;
729 }
730
731 return 0;
732 }
733
734 /*
735 * Allocate & map buffers for page given the extent map. Write it out.
736 * except for the original page of a writepage, this is called on
737 * delalloc/unwritten pages only, for the original page it is possible
738 * that the page has no mapping at all.
739 */
740 STATIC int
741 xfs_convert_page(
742 struct inode *inode,
743 struct page *page,
744 loff_t tindex,
745 xfs_iomap_t *mp,
746 xfs_ioend_t **ioendp,
747 struct writeback_control *wbc,
748 int startio,
749 int all_bh)
750 {
751 struct buffer_head *bh, *head;
752 xfs_off_t end_offset;
753 unsigned long p_offset;
754 unsigned int type;
755 int bbits = inode->i_blkbits;
756 int len, page_dirty;
757 int count = 0, done = 0, uptodate = 1;
758 xfs_off_t offset = page_offset(page);
759
760 if (page->index != tindex)
761 goto fail;
762 if (!trylock_page(page))
763 goto fail;
764 if (PageWriteback(page))
765 goto fail_unlock_page;
766 if (page->mapping != inode->i_mapping)
767 goto fail_unlock_page;
768 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
769 goto fail_unlock_page;
770
771 /*
772 * page_dirty is initially a count of buffers on the page before
773 * EOF and is decremented as we move each into a cleanable state.
774 *
775 * Derivation:
776 *
777 * End offset is the highest offset that this page should represent.
778 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
779 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
780 * hence give us the correct page_dirty count. On any other page,
781 * it will be zero and in that case we need page_dirty to be the
782 * count of buffers on the page.
783 */
784 end_offset = min_t(unsigned long long,
785 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
786 i_size_read(inode));
787
788 len = 1 << inode->i_blkbits;
789 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
790 PAGE_CACHE_SIZE);
791 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
792 page_dirty = p_offset / len;
793
794 bh = head = page_buffers(page);
795 do {
796 if (offset >= end_offset)
797 break;
798 if (!buffer_uptodate(bh))
799 uptodate = 0;
800 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
801 done = 1;
802 continue;
803 }
804
805 if (buffer_unwritten(bh) || buffer_delay(bh)) {
806 if (buffer_unwritten(bh))
807 type = IOMAP_UNWRITTEN;
808 else
809 type = IOMAP_DELAY;
810
811 if (!xfs_iomap_valid(mp, offset)) {
812 done = 1;
813 continue;
814 }
815
816 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
817 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
818
819 xfs_map_at_offset(bh, offset, bbits, mp);
820 if (startio) {
821 xfs_add_to_ioend(inode, bh, offset,
822 type, ioendp, done);
823 } else {
824 set_buffer_dirty(bh);
825 unlock_buffer(bh);
826 mark_buffer_dirty(bh);
827 }
828 page_dirty--;
829 count++;
830 } else {
831 type = IOMAP_NEW;
832 if (buffer_mapped(bh) && all_bh && startio) {
833 lock_buffer(bh);
834 xfs_add_to_ioend(inode, bh, offset,
835 type, ioendp, done);
836 count++;
837 page_dirty--;
838 } else {
839 done = 1;
840 }
841 }
842 } while (offset += len, (bh = bh->b_this_page) != head);
843
844 if (uptodate && bh == head)
845 SetPageUptodate(page);
846
847 if (startio) {
848 if (count) {
849 struct backing_dev_info *bdi;
850
851 bdi = inode->i_mapping->backing_dev_info;
852 wbc->nr_to_write--;
853 if (bdi_write_congested(bdi)) {
854 wbc->encountered_congestion = 1;
855 done = 1;
856 } else if (wbc->nr_to_write <= 0) {
857 done = 1;
858 }
859 }
860 xfs_start_page_writeback(page, !page_dirty, count);
861 }
862
863 return done;
864 fail_unlock_page:
865 unlock_page(page);
866 fail:
867 return 1;
868 }
869
870 /*
871 * Convert & write out a cluster of pages in the same extent as defined
872 * by mp and following the start page.
873 */
874 STATIC void
875 xfs_cluster_write(
876 struct inode *inode,
877 pgoff_t tindex,
878 xfs_iomap_t *iomapp,
879 xfs_ioend_t **ioendp,
880 struct writeback_control *wbc,
881 int startio,
882 int all_bh,
883 pgoff_t tlast)
884 {
885 struct pagevec pvec;
886 int done = 0, i;
887
888 pagevec_init(&pvec, 0);
889 while (!done && tindex <= tlast) {
890 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
891
892 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
893 break;
894
895 for (i = 0; i < pagevec_count(&pvec); i++) {
896 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
897 iomapp, ioendp, wbc, startio, all_bh);
898 if (done)
899 break;
900 }
901
902 pagevec_release(&pvec);
903 cond_resched();
904 }
905 }
906
907 /*
908 * Calling this without startio set means we are being asked to make a dirty
909 * page ready for freeing it's buffers. When called with startio set then
910 * we are coming from writepage.
911 *
912 * When called with startio set it is important that we write the WHOLE
913 * page if possible.
914 * The bh->b_state's cannot know if any of the blocks or which block for
915 * that matter are dirty due to mmap writes, and therefore bh uptodate is
916 * only valid if the page itself isn't completely uptodate. Some layers
917 * may clear the page dirty flag prior to calling write page, under the
918 * assumption the entire page will be written out; by not writing out the
919 * whole page the page can be reused before all valid dirty data is
920 * written out. Note: in the case of a page that has been dirty'd by
921 * mapwrite and but partially setup by block_prepare_write the
922 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
923 * valid state, thus the whole page must be written out thing.
924 */
925
926 STATIC int
927 xfs_page_state_convert(
928 struct inode *inode,
929 struct page *page,
930 struct writeback_control *wbc,
931 int startio,
932 int unmapped) /* also implies page uptodate */
933 {
934 struct buffer_head *bh, *head;
935 xfs_iomap_t iomap;
936 xfs_ioend_t *ioend = NULL, *iohead = NULL;
937 loff_t offset;
938 unsigned long p_offset = 0;
939 unsigned int type;
940 __uint64_t end_offset;
941 pgoff_t end_index, last_index, tlast;
942 ssize_t size, len;
943 int flags, err, iomap_valid = 0, uptodate = 1;
944 int page_dirty, count = 0;
945 int trylock = 0;
946 int all_bh = unmapped;
947
948 if (startio) {
949 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
950 trylock |= BMAPI_TRYLOCK;
951 }
952
953 /* Is this page beyond the end of the file? */
954 offset = i_size_read(inode);
955 end_index = offset >> PAGE_CACHE_SHIFT;
956 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
957 if (page->index >= end_index) {
958 if ((page->index >= end_index + 1) ||
959 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
960 if (startio)
961 unlock_page(page);
962 return 0;
963 }
964 }
965
966 /*
967 * page_dirty is initially a count of buffers on the page before
968 * EOF and is decremented as we move each into a cleanable state.
969 *
970 * Derivation:
971 *
972 * End offset is the highest offset that this page should represent.
973 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
974 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
975 * hence give us the correct page_dirty count. On any other page,
976 * it will be zero and in that case we need page_dirty to be the
977 * count of buffers on the page.
978 */
979 end_offset = min_t(unsigned long long,
980 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
981 len = 1 << inode->i_blkbits;
982 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
983 PAGE_CACHE_SIZE);
984 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
985 page_dirty = p_offset / len;
986
987 bh = head = page_buffers(page);
988 offset = page_offset(page);
989 flags = BMAPI_READ;
990 type = IOMAP_NEW;
991
992 /* TODO: cleanup count and page_dirty */
993
994 do {
995 if (offset >= end_offset)
996 break;
997 if (!buffer_uptodate(bh))
998 uptodate = 0;
999 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
1000 /*
1001 * the iomap is actually still valid, but the ioend
1002 * isn't. shouldn't happen too often.
1003 */
1004 iomap_valid = 0;
1005 continue;
1006 }
1007
1008 if (iomap_valid)
1009 iomap_valid = xfs_iomap_valid(&iomap, offset);
1010
1011 /*
1012 * First case, map an unwritten extent and prepare for
1013 * extent state conversion transaction on completion.
1014 *
1015 * Second case, allocate space for a delalloc buffer.
1016 * We can return EAGAIN here in the release page case.
1017 *
1018 * Third case, an unmapped buffer was found, and we are
1019 * in a path where we need to write the whole page out.
1020 */
1021 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1022 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1023 !buffer_mapped(bh) && (unmapped || startio))) {
1024 int new_ioend = 0;
1025
1026 /*
1027 * Make sure we don't use a read-only iomap
1028 */
1029 if (flags == BMAPI_READ)
1030 iomap_valid = 0;
1031
1032 if (buffer_unwritten(bh)) {
1033 type = IOMAP_UNWRITTEN;
1034 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
1035 } else if (buffer_delay(bh)) {
1036 type = IOMAP_DELAY;
1037 flags = BMAPI_ALLOCATE | trylock;
1038 } else {
1039 type = IOMAP_NEW;
1040 flags = BMAPI_WRITE | BMAPI_MMAP;
1041 }
1042
1043 if (!iomap_valid) {
1044 /*
1045 * if we didn't have a valid mapping then we
1046 * need to ensure that we put the new mapping
1047 * in a new ioend structure. This needs to be
1048 * done to ensure that the ioends correctly
1049 * reflect the block mappings at io completion
1050 * for unwritten extent conversion.
1051 */
1052 new_ioend = 1;
1053 if (type == IOMAP_NEW) {
1054 size = xfs_probe_cluster(inode,
1055 page, bh, head, 0);
1056 } else {
1057 size = len;
1058 }
1059
1060 err = xfs_map_blocks(inode, offset, size,
1061 &iomap, flags);
1062 if (err)
1063 goto error;
1064 iomap_valid = xfs_iomap_valid(&iomap, offset);
1065 }
1066 if (iomap_valid) {
1067 xfs_map_at_offset(bh, offset,
1068 inode->i_blkbits, &iomap);
1069 if (startio) {
1070 xfs_add_to_ioend(inode, bh, offset,
1071 type, &ioend,
1072 new_ioend);
1073 } else {
1074 set_buffer_dirty(bh);
1075 unlock_buffer(bh);
1076 mark_buffer_dirty(bh);
1077 }
1078 page_dirty--;
1079 count++;
1080 }
1081 } else if (buffer_uptodate(bh) && startio) {
1082 /*
1083 * we got here because the buffer is already mapped.
1084 * That means it must already have extents allocated
1085 * underneath it. Map the extent by reading it.
1086 */
1087 if (!iomap_valid || flags != BMAPI_READ) {
1088 flags = BMAPI_READ;
1089 size = xfs_probe_cluster(inode, page, bh,
1090 head, 1);
1091 err = xfs_map_blocks(inode, offset, size,
1092 &iomap, flags);
1093 if (err)
1094 goto error;
1095 iomap_valid = xfs_iomap_valid(&iomap, offset);
1096 }
1097
1098 /*
1099 * We set the type to IOMAP_NEW in case we are doing a
1100 * small write at EOF that is extending the file but
1101 * without needing an allocation. We need to update the
1102 * file size on I/O completion in this case so it is
1103 * the same case as having just allocated a new extent
1104 * that we are writing into for the first time.
1105 */
1106 type = IOMAP_NEW;
1107 if (trylock_buffer(bh)) {
1108 ASSERT(buffer_mapped(bh));
1109 if (iomap_valid)
1110 all_bh = 1;
1111 xfs_add_to_ioend(inode, bh, offset, type,
1112 &ioend, !iomap_valid);
1113 page_dirty--;
1114 count++;
1115 } else {
1116 iomap_valid = 0;
1117 }
1118 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1119 (unmapped || startio)) {
1120 iomap_valid = 0;
1121 }
1122
1123 if (!iohead)
1124 iohead = ioend;
1125
1126 } while (offset += len, ((bh = bh->b_this_page) != head));
1127
1128 if (uptodate && bh == head)
1129 SetPageUptodate(page);
1130
1131 if (startio)
1132 xfs_start_page_writeback(page, 1, count);
1133
1134 if (ioend && iomap_valid) {
1135 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
1136 PAGE_CACHE_SHIFT;
1137 tlast = min_t(pgoff_t, offset, last_index);
1138 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
1139 wbc, startio, all_bh, tlast);
1140 }
1141
1142 if (iohead)
1143 xfs_submit_ioend(iohead);
1144
1145 return page_dirty;
1146
1147 error:
1148 if (iohead)
1149 xfs_cancel_ioend(iohead);
1150
1151 /*
1152 * If it's delalloc and we have nowhere to put it,
1153 * throw it away, unless the lower layers told
1154 * us to try again.
1155 */
1156 if (err != -EAGAIN) {
1157 if (!unmapped)
1158 block_invalidatepage(page, 0);
1159 ClearPageUptodate(page);
1160 }
1161 return err;
1162 }
1163
1164 /*
1165 * writepage: Called from one of two places:
1166 *
1167 * 1. we are flushing a delalloc buffer head.
1168 *
1169 * 2. we are writing out a dirty page. Typically the page dirty
1170 * state is cleared before we get here. In this case is it
1171 * conceivable we have no buffer heads.
1172 *
1173 * For delalloc space on the page we need to allocate space and
1174 * flush it. For unmapped buffer heads on the page we should
1175 * allocate space if the page is uptodate. For any other dirty
1176 * buffer heads on the page we should flush them.
1177 *
1178 * If we detect that a transaction would be required to flush
1179 * the page, we have to check the process flags first, if we
1180 * are already in a transaction or disk I/O during allocations
1181 * is off, we need to fail the writepage and redirty the page.
1182 */
1183
1184 STATIC int
1185 xfs_vm_writepage(
1186 struct page *page,
1187 struct writeback_control *wbc)
1188 {
1189 int error;
1190 int need_trans;
1191 int delalloc, unmapped, unwritten;
1192 struct inode *inode = page->mapping->host;
1193
1194 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1195
1196 /*
1197 * We need a transaction if:
1198 * 1. There are delalloc buffers on the page
1199 * 2. The page is uptodate and we have unmapped buffers
1200 * 3. The page is uptodate and we have no buffers
1201 * 4. There are unwritten buffers on the page
1202 */
1203
1204 if (!page_has_buffers(page)) {
1205 unmapped = 1;
1206 need_trans = 1;
1207 } else {
1208 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1209 if (!PageUptodate(page))
1210 unmapped = 0;
1211 need_trans = delalloc + unmapped + unwritten;
1212 }
1213
1214 /*
1215 * If we need a transaction and the process flags say
1216 * we are already in a transaction, or no IO is allowed
1217 * then mark the page dirty again and leave the page
1218 * as is.
1219 */
1220 if (current_test_flags(PF_FSTRANS) && need_trans)
1221 goto out_fail;
1222
1223 /*
1224 * Delay hooking up buffer heads until we have
1225 * made our go/no-go decision.
1226 */
1227 if (!page_has_buffers(page))
1228 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1229
1230 /*
1231 * Convert delayed allocate, unwritten or unmapped space
1232 * to real space and flush out to disk.
1233 */
1234 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1235 if (error == -EAGAIN)
1236 goto out_fail;
1237 if (unlikely(error < 0))
1238 goto out_unlock;
1239
1240 return 0;
1241
1242 out_fail:
1243 redirty_page_for_writepage(wbc, page);
1244 unlock_page(page);
1245 return 0;
1246 out_unlock:
1247 unlock_page(page);
1248 return error;
1249 }
1250
1251 STATIC int
1252 xfs_vm_writepages(
1253 struct address_space *mapping,
1254 struct writeback_control *wbc)
1255 {
1256 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1257 return generic_writepages(mapping, wbc);
1258 }
1259
1260 /*
1261 * Called to move a page into cleanable state - and from there
1262 * to be released. Possibly the page is already clean. We always
1263 * have buffer heads in this call.
1264 *
1265 * Returns 0 if the page is ok to release, 1 otherwise.
1266 *
1267 * Possible scenarios are:
1268 *
1269 * 1. We are being called to release a page which has been written
1270 * to via regular I/O. buffer heads will be dirty and possibly
1271 * delalloc. If no delalloc buffer heads in this case then we
1272 * can just return zero.
1273 *
1274 * 2. We are called to release a page which has been written via
1275 * mmap, all we need to do is ensure there is no delalloc
1276 * state in the buffer heads, if not we can let the caller
1277 * free them and we should come back later via writepage.
1278 */
1279 STATIC int
1280 xfs_vm_releasepage(
1281 struct page *page,
1282 gfp_t gfp_mask)
1283 {
1284 struct inode *inode = page->mapping->host;
1285 int dirty, delalloc, unmapped, unwritten;
1286 struct writeback_control wbc = {
1287 .sync_mode = WB_SYNC_ALL,
1288 .nr_to_write = 1,
1289 };
1290
1291 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
1292
1293 if (!page_has_buffers(page))
1294 return 0;
1295
1296 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1297 if (!delalloc && !unwritten)
1298 goto free_buffers;
1299
1300 if (!(gfp_mask & __GFP_FS))
1301 return 0;
1302
1303 /* If we are already inside a transaction or the thread cannot
1304 * do I/O, we cannot release this page.
1305 */
1306 if (current_test_flags(PF_FSTRANS))
1307 return 0;
1308
1309 /*
1310 * Convert delalloc space to real space, do not flush the
1311 * data out to disk, that will be done by the caller.
1312 * Never need to allocate space here - we will always
1313 * come back to writepage in that case.
1314 */
1315 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1316 if (dirty == 0 && !unwritten)
1317 goto free_buffers;
1318 return 0;
1319
1320 free_buffers:
1321 return try_to_free_buffers(page);
1322 }
1323
1324 STATIC int
1325 __xfs_get_blocks(
1326 struct inode *inode,
1327 sector_t iblock,
1328 struct buffer_head *bh_result,
1329 int create,
1330 int direct,
1331 bmapi_flags_t flags)
1332 {
1333 xfs_iomap_t iomap;
1334 xfs_off_t offset;
1335 ssize_t size;
1336 int niomap = 1;
1337 int error;
1338
1339 offset = (xfs_off_t)iblock << inode->i_blkbits;
1340 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1341 size = bh_result->b_size;
1342 error = xfs_iomap(XFS_I(inode), offset, size,
1343 create ? flags : BMAPI_READ, &iomap, &niomap);
1344 if (error)
1345 return -error;
1346 if (niomap == 0)
1347 return 0;
1348
1349 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
1350 /*
1351 * For unwritten extents do not report a disk address on
1352 * the read case (treat as if we're reading into a hole).
1353 */
1354 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1355 xfs_map_buffer(bh_result, &iomap, offset,
1356 inode->i_blkbits);
1357 }
1358 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1359 if (direct)
1360 bh_result->b_private = inode;
1361 set_buffer_unwritten(bh_result);
1362 }
1363 }
1364
1365 /*
1366 * If this is a realtime file, data may be on a different device.
1367 * to that pointed to from the buffer_head b_bdev currently.
1368 */
1369 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
1370
1371 /*
1372 * If we previously allocated a block out beyond eof and we are now
1373 * coming back to use it then we will need to flag it as new even if it
1374 * has a disk address.
1375 *
1376 * With sub-block writes into unwritten extents we also need to mark
1377 * the buffer as new so that the unwritten parts of the buffer gets
1378 * correctly zeroed.
1379 */
1380 if (create &&
1381 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1382 (offset >= i_size_read(inode)) ||
1383 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
1384 set_buffer_new(bh_result);
1385
1386 if (iomap.iomap_flags & IOMAP_DELAY) {
1387 BUG_ON(direct);
1388 if (create) {
1389 set_buffer_uptodate(bh_result);
1390 set_buffer_mapped(bh_result);
1391 set_buffer_delay(bh_result);
1392 }
1393 }
1394
1395 if (direct || size > (1 << inode->i_blkbits)) {
1396 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1397 offset = min_t(xfs_off_t,
1398 iomap.iomap_bsize - iomap.iomap_delta, size);
1399 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
1400 }
1401
1402 return 0;
1403 }
1404
1405 int
1406 xfs_get_blocks(
1407 struct inode *inode,
1408 sector_t iblock,
1409 struct buffer_head *bh_result,
1410 int create)
1411 {
1412 return __xfs_get_blocks(inode, iblock,
1413 bh_result, create, 0, BMAPI_WRITE);
1414 }
1415
1416 STATIC int
1417 xfs_get_blocks_direct(
1418 struct inode *inode,
1419 sector_t iblock,
1420 struct buffer_head *bh_result,
1421 int create)
1422 {
1423 return __xfs_get_blocks(inode, iblock,
1424 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
1425 }
1426
1427 STATIC void
1428 xfs_end_io_direct(
1429 struct kiocb *iocb,
1430 loff_t offset,
1431 ssize_t size,
1432 void *private)
1433 {
1434 xfs_ioend_t *ioend = iocb->private;
1435
1436 /*
1437 * Non-NULL private data means we need to issue a transaction to
1438 * convert a range from unwritten to written extents. This needs
1439 * to happen from process context but aio+dio I/O completion
1440 * happens from irq context so we need to defer it to a workqueue.
1441 * This is not necessary for synchronous direct I/O, but we do
1442 * it anyway to keep the code uniform and simpler.
1443 *
1444 * Well, if only it were that simple. Because synchronous direct I/O
1445 * requires extent conversion to occur *before* we return to userspace,
1446 * we have to wait for extent conversion to complete. Look at the
1447 * iocb that has been passed to us to determine if this is AIO or
1448 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1449 * workqueue and wait for it to complete.
1450 *
1451 * The core direct I/O code might be changed to always call the
1452 * completion handler in the future, in which case all this can
1453 * go away.
1454 */
1455 ioend->io_offset = offset;
1456 ioend->io_size = size;
1457 if (ioend->io_type == IOMAP_READ) {
1458 xfs_finish_ioend(ioend, 0);
1459 } else if (private && size > 0) {
1460 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
1461 } else {
1462 /*
1463 * A direct I/O write ioend starts it's life in unwritten
1464 * state in case they map an unwritten extent. This write
1465 * didn't map an unwritten extent so switch it's completion
1466 * handler.
1467 */
1468 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
1469 xfs_finish_ioend(ioend, 0);
1470 }
1471
1472 /*
1473 * blockdev_direct_IO can return an error even after the I/O
1474 * completion handler was called. Thus we need to protect
1475 * against double-freeing.
1476 */
1477 iocb->private = NULL;
1478 }
1479
1480 STATIC ssize_t
1481 xfs_vm_direct_IO(
1482 int rw,
1483 struct kiocb *iocb,
1484 const struct iovec *iov,
1485 loff_t offset,
1486 unsigned long nr_segs)
1487 {
1488 struct file *file = iocb->ki_filp;
1489 struct inode *inode = file->f_mapping->host;
1490 struct block_device *bdev;
1491 ssize_t ret;
1492
1493 bdev = xfs_find_bdev_for_inode(XFS_I(inode));
1494
1495 if (rw == WRITE) {
1496 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
1497 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1498 bdev, iov, offset, nr_segs,
1499 xfs_get_blocks_direct,
1500 xfs_end_io_direct);
1501 } else {
1502 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
1503 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1504 bdev, iov, offset, nr_segs,
1505 xfs_get_blocks_direct,
1506 xfs_end_io_direct);
1507 }
1508
1509 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
1510 xfs_destroy_ioend(iocb->private);
1511 return ret;
1512 }
1513
1514 STATIC int
1515 xfs_vm_write_begin(
1516 struct file *file,
1517 struct address_space *mapping,
1518 loff_t pos,
1519 unsigned len,
1520 unsigned flags,
1521 struct page **pagep,
1522 void **fsdata)
1523 {
1524 *pagep = NULL;
1525 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1526 xfs_get_blocks);
1527 }
1528
1529 STATIC sector_t
1530 xfs_vm_bmap(
1531 struct address_space *mapping,
1532 sector_t block)
1533 {
1534 struct inode *inode = (struct inode *)mapping->host;
1535 struct xfs_inode *ip = XFS_I(inode);
1536
1537 xfs_itrace_entry(XFS_I(inode));
1538 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1539 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1540 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1541 return generic_block_bmap(mapping, block, xfs_get_blocks);
1542 }
1543
1544 STATIC int
1545 xfs_vm_readpage(
1546 struct file *unused,
1547 struct page *page)
1548 {
1549 return mpage_readpage(page, xfs_get_blocks);
1550 }
1551
1552 STATIC int
1553 xfs_vm_readpages(
1554 struct file *unused,
1555 struct address_space *mapping,
1556 struct list_head *pages,
1557 unsigned nr_pages)
1558 {
1559 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1560 }
1561
1562 STATIC void
1563 xfs_vm_invalidatepage(
1564 struct page *page,
1565 unsigned long offset)
1566 {
1567 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1568 page->mapping->host, page, offset);
1569 block_invalidatepage(page, offset);
1570 }
1571
1572 const struct address_space_operations xfs_address_space_operations = {
1573 .readpage = xfs_vm_readpage,
1574 .readpages = xfs_vm_readpages,
1575 .writepage = xfs_vm_writepage,
1576 .writepages = xfs_vm_writepages,
1577 .sync_page = block_sync_page,
1578 .releasepage = xfs_vm_releasepage,
1579 .invalidatepage = xfs_vm_invalidatepage,
1580 .write_begin = xfs_vm_write_begin,
1581 .write_end = generic_write_end,
1582 .bmap = xfs_vm_bmap,
1583 .direct_IO = xfs_vm_direct_IO,
1584 .migratepage = buffer_migrate_page,
1585 };
Linux kernel - Deliverables
This page took 0.064106 seconds and 4 git commands to generate.