]>
Commit | Line | Data |
---|---|---|
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_shared.h" | |
20 | #include "xfs_format.h" | |
21 | #include "xfs_log_format.h" | |
22 | #include "xfs_trans_resv.h" | |
23 | #include "xfs_mount.h" | |
24 | #include "xfs_inode.h" | |
25 | #include "xfs_trans.h" | |
26 | #include "xfs_inode_item.h" | |
27 | #include "xfs_alloc.h" | |
28 | #include "xfs_error.h" | |
29 | #include "xfs_iomap.h" | |
30 | #include "xfs_trace.h" | |
31 | #include "xfs_bmap.h" | |
32 | #include "xfs_bmap_util.h" | |
33 | #include "xfs_bmap_btree.h" | |
34 | #include <linux/aio.h> | |
35 | #include <linux/gfp.h> | |
36 | #include <linux/mpage.h> | |
37 | #include <linux/pagevec.h> | |
38 | #include <linux/writeback.h> | |
39 | ||
40 | void | |
41 | xfs_count_page_state( | |
42 | struct page *page, | |
43 | int *delalloc, | |
44 | int *unwritten) | |
45 | { | |
46 | struct buffer_head *bh, *head; | |
47 | ||
48 | *delalloc = *unwritten = 0; | |
49 | ||
50 | bh = head = page_buffers(page); | |
51 | do { | |
52 | if (buffer_unwritten(bh)) | |
53 | (*unwritten) = 1; | |
54 | else if (buffer_delay(bh)) | |
55 | (*delalloc) = 1; | |
56 | } while ((bh = bh->b_this_page) != head); | |
57 | } | |
58 | ||
59 | STATIC struct block_device * | |
60 | xfs_find_bdev_for_inode( | |
61 | struct inode *inode) | |
62 | { | |
63 | struct xfs_inode *ip = XFS_I(inode); | |
64 | struct xfs_mount *mp = ip->i_mount; | |
65 | ||
66 | if (XFS_IS_REALTIME_INODE(ip)) | |
67 | return mp->m_rtdev_targp->bt_bdev; | |
68 | else | |
69 | return mp->m_ddev_targp->bt_bdev; | |
70 | } | |
71 | ||
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 | */ | |
78 | STATIC void | |
79 | xfs_destroy_ioend( | |
80 | xfs_ioend_t *ioend) | |
81 | { | |
82 | struct buffer_head *bh, *next; | |
83 | ||
84 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
85 | next = bh->b_private; | |
86 | bh->b_end_io(bh, !ioend->io_error); | |
87 | } | |
88 | ||
89 | mempool_free(ioend, xfs_ioend_pool); | |
90 | } | |
91 | ||
92 | /* | |
93 | * Fast and loose check if this write could update the on-disk inode size. | |
94 | */ | |
95 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
96 | { | |
97 | return ioend->io_offset + ioend->io_size > | |
98 | XFS_I(ioend->io_inode)->i_d.di_size; | |
99 | } | |
100 | ||
101 | STATIC int | |
102 | xfs_setfilesize_trans_alloc( | |
103 | struct xfs_ioend *ioend) | |
104 | { | |
105 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
106 | struct xfs_trans *tp; | |
107 | int error; | |
108 | ||
109 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
110 | ||
111 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); | |
112 | if (error) { | |
113 | xfs_trans_cancel(tp, 0); | |
114 | return error; | |
115 | } | |
116 | ||
117 | ioend->io_append_trans = tp; | |
118 | ||
119 | /* | |
120 | * We may pass freeze protection with a transaction. So tell lockdep | |
121 | * we released it. | |
122 | */ | |
123 | rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
124 | 1, _THIS_IP_); | |
125 | /* | |
126 | * We hand off the transaction to the completion thread now, so | |
127 | * clear the flag here. | |
128 | */ | |
129 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
130 | return 0; | |
131 | } | |
132 | ||
133 | /* | |
134 | * Update on-disk file size now that data has been written to disk. | |
135 | */ | |
136 | STATIC int | |
137 | xfs_setfilesize( | |
138 | struct xfs_inode *ip, | |
139 | struct xfs_trans *tp, | |
140 | xfs_off_t offset, | |
141 | size_t size) | |
142 | { | |
143 | xfs_fsize_t isize; | |
144 | ||
145 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
146 | isize = xfs_new_eof(ip, offset + size); | |
147 | if (!isize) { | |
148 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
149 | xfs_trans_cancel(tp, 0); | |
150 | return 0; | |
151 | } | |
152 | ||
153 | trace_xfs_setfilesize(ip, offset, size); | |
154 | ||
155 | ip->i_d.di_size = isize; | |
156 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
157 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
158 | ||
159 | return xfs_trans_commit(tp, 0); | |
160 | } | |
161 | ||
162 | STATIC int | |
163 | xfs_setfilesize_ioend( | |
164 | struct xfs_ioend *ioend) | |
165 | { | |
166 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
167 | struct xfs_trans *tp = ioend->io_append_trans; | |
168 | ||
169 | /* | |
170 | * The transaction may have been allocated in the I/O submission thread, | |
171 | * thus we need to mark ourselves as being in a transaction manually. | |
172 | * Similarly for freeze protection. | |
173 | */ | |
174 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
175 | rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
176 | 0, 1, _THIS_IP_); | |
177 | ||
178 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); | |
179 | } | |
180 | ||
181 | /* | |
182 | * Schedule IO completion handling on the final put of an ioend. | |
183 | * | |
184 | * If there is no work to do we might as well call it a day and free the | |
185 | * ioend right now. | |
186 | */ | |
187 | STATIC void | |
188 | xfs_finish_ioend( | |
189 | struct xfs_ioend *ioend) | |
190 | { | |
191 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
192 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
193 | ||
194 | if (ioend->io_type == XFS_IO_UNWRITTEN) | |
195 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); | |
196 | else if (ioend->io_append_trans) | |
197 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
198 | else | |
199 | xfs_destroy_ioend(ioend); | |
200 | } | |
201 | } | |
202 | ||
203 | /* | |
204 | * IO write completion. | |
205 | */ | |
206 | STATIC void | |
207 | xfs_end_io( | |
208 | struct work_struct *work) | |
209 | { | |
210 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); | |
211 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
212 | int error = 0; | |
213 | ||
214 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
215 | ioend->io_error = -EIO; | |
216 | goto done; | |
217 | } | |
218 | if (ioend->io_error) | |
219 | goto done; | |
220 | ||
221 | /* | |
222 | * For unwritten extents we need to issue transactions to convert a | |
223 | * range to normal written extens after the data I/O has finished. | |
224 | */ | |
225 | if (ioend->io_type == XFS_IO_UNWRITTEN) { | |
226 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
227 | ioend->io_size); | |
228 | } else if (ioend->io_append_trans) { | |
229 | error = xfs_setfilesize_ioend(ioend); | |
230 | } else { | |
231 | ASSERT(!xfs_ioend_is_append(ioend)); | |
232 | } | |
233 | ||
234 | done: | |
235 | if (error) | |
236 | ioend->io_error = error; | |
237 | xfs_destroy_ioend(ioend); | |
238 | } | |
239 | ||
240 | /* | |
241 | * Allocate and initialise an IO completion structure. | |
242 | * We need to track unwritten extent write completion here initially. | |
243 | * We'll need to extend this for updating the ondisk inode size later | |
244 | * (vs. incore size). | |
245 | */ | |
246 | STATIC xfs_ioend_t * | |
247 | xfs_alloc_ioend( | |
248 | struct inode *inode, | |
249 | unsigned int type) | |
250 | { | |
251 | xfs_ioend_t *ioend; | |
252 | ||
253 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
254 | ||
255 | /* | |
256 | * Set the count to 1 initially, which will prevent an I/O | |
257 | * completion callback from happening before we have started | |
258 | * all the I/O from calling the completion routine too early. | |
259 | */ | |
260 | atomic_set(&ioend->io_remaining, 1); | |
261 | ioend->io_error = 0; | |
262 | ioend->io_list = NULL; | |
263 | ioend->io_type = type; | |
264 | ioend->io_inode = inode; | |
265 | ioend->io_buffer_head = NULL; | |
266 | ioend->io_buffer_tail = NULL; | |
267 | ioend->io_offset = 0; | |
268 | ioend->io_size = 0; | |
269 | ioend->io_append_trans = NULL; | |
270 | ||
271 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
272 | return ioend; | |
273 | } | |
274 | ||
275 | STATIC int | |
276 | xfs_map_blocks( | |
277 | struct inode *inode, | |
278 | loff_t offset, | |
279 | struct xfs_bmbt_irec *imap, | |
280 | int type, | |
281 | int nonblocking) | |
282 | { | |
283 | struct xfs_inode *ip = XFS_I(inode); | |
284 | struct xfs_mount *mp = ip->i_mount; | |
285 | ssize_t count = 1 << inode->i_blkbits; | |
286 | xfs_fileoff_t offset_fsb, end_fsb; | |
287 | int error = 0; | |
288 | int bmapi_flags = XFS_BMAPI_ENTIRE; | |
289 | int nimaps = 1; | |
290 | ||
291 | if (XFS_FORCED_SHUTDOWN(mp)) | |
292 | return -EIO; | |
293 | ||
294 | if (type == XFS_IO_UNWRITTEN) | |
295 | bmapi_flags |= XFS_BMAPI_IGSTATE; | |
296 | ||
297 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
298 | if (nonblocking) | |
299 | return -EAGAIN; | |
300 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
301 | } | |
302 | ||
303 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || | |
304 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
305 | ASSERT(offset <= mp->m_super->s_maxbytes); | |
306 | ||
307 | if (offset + count > mp->m_super->s_maxbytes) | |
308 | count = mp->m_super->s_maxbytes - offset; | |
309 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
310 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
311 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, | |
312 | imap, &nimaps, bmapi_flags); | |
313 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
314 | ||
315 | if (error) | |
316 | return error; | |
317 | ||
318 | if (type == XFS_IO_DELALLOC && | |
319 | (!nimaps || isnullstartblock(imap->br_startblock))) { | |
320 | error = xfs_iomap_write_allocate(ip, offset, imap); | |
321 | if (!error) | |
322 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
323 | return error; | |
324 | } | |
325 | ||
326 | #ifdef DEBUG | |
327 | if (type == XFS_IO_UNWRITTEN) { | |
328 | ASSERT(nimaps); | |
329 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
330 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
331 | } | |
332 | #endif | |
333 | if (nimaps) | |
334 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
335 | return 0; | |
336 | } | |
337 | ||
338 | STATIC int | |
339 | xfs_imap_valid( | |
340 | struct inode *inode, | |
341 | struct xfs_bmbt_irec *imap, | |
342 | xfs_off_t offset) | |
343 | { | |
344 | offset >>= inode->i_blkbits; | |
345 | ||
346 | return offset >= imap->br_startoff && | |
347 | offset < imap->br_startoff + imap->br_blockcount; | |
348 | } | |
349 | ||
350 | /* | |
351 | * BIO completion handler for buffered IO. | |
352 | */ | |
353 | STATIC void | |
354 | xfs_end_bio( | |
355 | struct bio *bio, | |
356 | int error) | |
357 | { | |
358 | xfs_ioend_t *ioend = bio->bi_private; | |
359 | ||
360 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); | |
361 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; | |
362 | ||
363 | /* Toss bio and pass work off to an xfsdatad thread */ | |
364 | bio->bi_private = NULL; | |
365 | bio->bi_end_io = NULL; | |
366 | bio_put(bio); | |
367 | ||
368 | xfs_finish_ioend(ioend); | |
369 | } | |
370 | ||
371 | STATIC void | |
372 | xfs_submit_ioend_bio( | |
373 | struct writeback_control *wbc, | |
374 | xfs_ioend_t *ioend, | |
375 | struct bio *bio) | |
376 | { | |
377 | atomic_inc(&ioend->io_remaining); | |
378 | bio->bi_private = ioend; | |
379 | bio->bi_end_io = xfs_end_bio; | |
380 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); | |
381 | } | |
382 | ||
383 | STATIC struct bio * | |
384 | xfs_alloc_ioend_bio( | |
385 | struct buffer_head *bh) | |
386 | { | |
387 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
388 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); | |
389 | ||
390 | ASSERT(bio->bi_private == NULL); | |
391 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
392 | bio->bi_bdev = bh->b_bdev; | |
393 | return bio; | |
394 | } | |
395 | ||
396 | STATIC void | |
397 | xfs_start_buffer_writeback( | |
398 | struct buffer_head *bh) | |
399 | { | |
400 | ASSERT(buffer_mapped(bh)); | |
401 | ASSERT(buffer_locked(bh)); | |
402 | ASSERT(!buffer_delay(bh)); | |
403 | ASSERT(!buffer_unwritten(bh)); | |
404 | ||
405 | mark_buffer_async_write(bh); | |
406 | set_buffer_uptodate(bh); | |
407 | clear_buffer_dirty(bh); | |
408 | } | |
409 | ||
410 | STATIC void | |
411 | xfs_start_page_writeback( | |
412 | struct page *page, | |
413 | int clear_dirty, | |
414 | int buffers) | |
415 | { | |
416 | ASSERT(PageLocked(page)); | |
417 | ASSERT(!PageWriteback(page)); | |
418 | ||
419 | /* | |
420 | * if the page was not fully cleaned, we need to ensure that the higher | |
421 | * layers come back to it correctly. That means we need to keep the page | |
422 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
423 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
424 | * write this page in this writeback sweep will be made. | |
425 | */ | |
426 | if (clear_dirty) { | |
427 | clear_page_dirty_for_io(page); | |
428 | set_page_writeback(page); | |
429 | } else | |
430 | set_page_writeback_keepwrite(page); | |
431 | ||
432 | unlock_page(page); | |
433 | ||
434 | /* If no buffers on the page are to be written, finish it here */ | |
435 | if (!buffers) | |
436 | end_page_writeback(page); | |
437 | } | |
438 | ||
439 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
440 | { | |
441 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
442 | } | |
443 | ||
444 | /* | |
445 | * Submit all of the bios for all of the ioends we have saved up, covering the | |
446 | * initial writepage page and also any probed pages. | |
447 | * | |
448 | * Because we may have multiple ioends spanning a page, we need to start | |
449 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
450 | * buffers as we got, then we can end up with a page that only has buffers | |
451 | * marked async write and I/O complete on can occur before we mark the other | |
452 | * buffers async write. | |
453 | * | |
454 | * The end result of this is that we trip a bug in end_page_writeback() because | |
455 | * we call it twice for the one page as the code in end_buffer_async_write() | |
456 | * assumes that all buffers on the page are started at the same time. | |
457 | * | |
458 | * The fix is two passes across the ioend list - one to start writeback on the | |
459 | * buffer_heads, and then submit them for I/O on the second pass. | |
460 | * | |
461 | * If @fail is non-zero, it means that we have a situation where some part of | |
462 | * the submission process has failed after we have marked paged for writeback | |
463 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
464 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
465 | */ | |
466 | STATIC void | |
467 | xfs_submit_ioend( | |
468 | struct writeback_control *wbc, | |
469 | xfs_ioend_t *ioend, | |
470 | int fail) | |
471 | { | |
472 | xfs_ioend_t *head = ioend; | |
473 | xfs_ioend_t *next; | |
474 | struct buffer_head *bh; | |
475 | struct bio *bio; | |
476 | sector_t lastblock = 0; | |
477 | ||
478 | /* Pass 1 - start writeback */ | |
479 | do { | |
480 | next = ioend->io_list; | |
481 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) | |
482 | xfs_start_buffer_writeback(bh); | |
483 | } while ((ioend = next) != NULL); | |
484 | ||
485 | /* Pass 2 - submit I/O */ | |
486 | ioend = head; | |
487 | do { | |
488 | next = ioend->io_list; | |
489 | bio = NULL; | |
490 | ||
491 | /* | |
492 | * If we are failing the IO now, just mark the ioend with an | |
493 | * error and finish it. This will run IO completion immediately | |
494 | * as there is only one reference to the ioend at this point in | |
495 | * time. | |
496 | */ | |
497 | if (fail) { | |
498 | ioend->io_error = fail; | |
499 | xfs_finish_ioend(ioend); | |
500 | continue; | |
501 | } | |
502 | ||
503 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
504 | ||
505 | if (!bio) { | |
506 | retry: | |
507 | bio = xfs_alloc_ioend_bio(bh); | |
508 | } else if (bh->b_blocknr != lastblock + 1) { | |
509 | xfs_submit_ioend_bio(wbc, ioend, bio); | |
510 | goto retry; | |
511 | } | |
512 | ||
513 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { | |
514 | xfs_submit_ioend_bio(wbc, ioend, bio); | |
515 | goto retry; | |
516 | } | |
517 | ||
518 | lastblock = bh->b_blocknr; | |
519 | } | |
520 | if (bio) | |
521 | xfs_submit_ioend_bio(wbc, ioend, bio); | |
522 | xfs_finish_ioend(ioend); | |
523 | } while ((ioend = next) != NULL); | |
524 | } | |
525 | ||
526 | /* | |
527 | * Cancel submission of all buffer_heads so far in this endio. | |
528 | * Toss the endio too. Only ever called for the initial page | |
529 | * in a writepage request, so only ever one page. | |
530 | */ | |
531 | STATIC void | |
532 | xfs_cancel_ioend( | |
533 | xfs_ioend_t *ioend) | |
534 | { | |
535 | xfs_ioend_t *next; | |
536 | struct buffer_head *bh, *next_bh; | |
537 | ||
538 | do { | |
539 | next = ioend->io_list; | |
540 | bh = ioend->io_buffer_head; | |
541 | do { | |
542 | next_bh = bh->b_private; | |
543 | clear_buffer_async_write(bh); | |
544 | /* | |
545 | * The unwritten flag is cleared when added to the | |
546 | * ioend. We're not submitting for I/O so mark the | |
547 | * buffer unwritten again for next time around. | |
548 | */ | |
549 | if (ioend->io_type == XFS_IO_UNWRITTEN) | |
550 | set_buffer_unwritten(bh); | |
551 | unlock_buffer(bh); | |
552 | } while ((bh = next_bh) != NULL); | |
553 | ||
554 | mempool_free(ioend, xfs_ioend_pool); | |
555 | } while ((ioend = next) != NULL); | |
556 | } | |
557 | ||
558 | /* | |
559 | * Test to see if we've been building up a completion structure for | |
560 | * earlier buffers -- if so, we try to append to this ioend if we | |
561 | * can, otherwise we finish off any current ioend and start another. | |
562 | * Return true if we've finished the given ioend. | |
563 | */ | |
564 | STATIC void | |
565 | xfs_add_to_ioend( | |
566 | struct inode *inode, | |
567 | struct buffer_head *bh, | |
568 | xfs_off_t offset, | |
569 | unsigned int type, | |
570 | xfs_ioend_t **result, | |
571 | int need_ioend) | |
572 | { | |
573 | xfs_ioend_t *ioend = *result; | |
574 | ||
575 | if (!ioend || need_ioend || type != ioend->io_type) { | |
576 | xfs_ioend_t *previous = *result; | |
577 | ||
578 | ioend = xfs_alloc_ioend(inode, type); | |
579 | ioend->io_offset = offset; | |
580 | ioend->io_buffer_head = bh; | |
581 | ioend->io_buffer_tail = bh; | |
582 | if (previous) | |
583 | previous->io_list = ioend; | |
584 | *result = ioend; | |
585 | } else { | |
586 | ioend->io_buffer_tail->b_private = bh; | |
587 | ioend->io_buffer_tail = bh; | |
588 | } | |
589 | ||
590 | bh->b_private = NULL; | |
591 | ioend->io_size += bh->b_size; | |
592 | } | |
593 | ||
594 | STATIC void | |
595 | xfs_map_buffer( | |
596 | struct inode *inode, | |
597 | struct buffer_head *bh, | |
598 | struct xfs_bmbt_irec *imap, | |
599 | xfs_off_t offset) | |
600 | { | |
601 | sector_t bn; | |
602 | struct xfs_mount *m = XFS_I(inode)->i_mount; | |
603 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); | |
604 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
605 | ||
606 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
607 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
608 | ||
609 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + | |
610 | ((offset - iomap_offset) >> inode->i_blkbits); | |
611 | ||
612 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); | |
613 | ||
614 | bh->b_blocknr = bn; | |
615 | set_buffer_mapped(bh); | |
616 | } | |
617 | ||
618 | STATIC void | |
619 | xfs_map_at_offset( | |
620 | struct inode *inode, | |
621 | struct buffer_head *bh, | |
622 | struct xfs_bmbt_irec *imap, | |
623 | xfs_off_t offset) | |
624 | { | |
625 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
626 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
627 | ||
628 | xfs_map_buffer(inode, bh, imap, offset); | |
629 | set_buffer_mapped(bh); | |
630 | clear_buffer_delay(bh); | |
631 | clear_buffer_unwritten(bh); | |
632 | } | |
633 | ||
634 | /* | |
635 | * Test if a given page contains at least one buffer of a given @type. | |
636 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
637 | * try to find one of the type passed in. If it is not set, then the caller only | |
638 | * needs to check the first buffer on the page for a match. | |
639 | */ | |
640 | STATIC bool | |
641 | xfs_check_page_type( | |
642 | struct page *page, | |
643 | unsigned int type, | |
644 | bool check_all_buffers) | |
645 | { | |
646 | struct buffer_head *bh; | |
647 | struct buffer_head *head; | |
648 | ||
649 | if (PageWriteback(page)) | |
650 | return false; | |
651 | if (!page->mapping) | |
652 | return false; | |
653 | if (!page_has_buffers(page)) | |
654 | return false; | |
655 | ||
656 | bh = head = page_buffers(page); | |
657 | do { | |
658 | if (buffer_unwritten(bh)) { | |
659 | if (type == XFS_IO_UNWRITTEN) | |
660 | return true; | |
661 | } else if (buffer_delay(bh)) { | |
662 | if (type == XFS_IO_DELALLOC) | |
663 | return true; | |
664 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
665 | if (type == XFS_IO_OVERWRITE) | |
666 | return true; | |
667 | } | |
668 | ||
669 | /* If we are only checking the first buffer, we are done now. */ | |
670 | if (!check_all_buffers) | |
671 | break; | |
672 | } while ((bh = bh->b_this_page) != head); | |
673 | ||
674 | return false; | |
675 | } | |
676 | ||
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 | */ | |
683 | STATIC int | |
684 | xfs_convert_page( | |
685 | struct inode *inode, | |
686 | struct page *page, | |
687 | loff_t tindex, | |
688 | struct xfs_bmbt_irec *imap, | |
689 | xfs_ioend_t **ioendp, | |
690 | struct writeback_control *wbc) | |
691 | { | |
692 | struct buffer_head *bh, *head; | |
693 | xfs_off_t end_offset; | |
694 | unsigned long p_offset; | |
695 | unsigned int type; | |
696 | int len, page_dirty; | |
697 | int count = 0, done = 0, uptodate = 1; | |
698 | xfs_off_t offset = page_offset(page); | |
699 | ||
700 | if (page->index != tindex) | |
701 | goto fail; | |
702 | if (!trylock_page(page)) | |
703 | goto fail; | |
704 | if (PageWriteback(page)) | |
705 | goto fail_unlock_page; | |
706 | if (page->mapping != inode->i_mapping) | |
707 | goto fail_unlock_page; | |
708 | if (!xfs_check_page_type(page, (*ioendp)->io_type, false)) | |
709 | goto fail_unlock_page; | |
710 | ||
711 | /* | |
712 | * page_dirty is initially a count of buffers on the page before | |
713 | * EOF and is decremented as we move each into a cleanable state. | |
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. | |
723 | */ | |
724 | end_offset = min_t(unsigned long long, | |
725 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
726 | i_size_read(inode)); | |
727 | ||
728 | /* | |
729 | * If the current map does not span the entire page we are about to try | |
730 | * to write, then give up. The only way we can write a page that spans | |
731 | * multiple mappings in a single writeback iteration is via the | |
732 | * xfs_vm_writepage() function. Data integrity writeback requires the | |
733 | * entire page to be written in a single attempt, otherwise the part of | |
734 | * the page we don't write here doesn't get written as part of the data | |
735 | * integrity sync. | |
736 | * | |
737 | * For normal writeback, we also don't attempt to write partial pages | |
738 | * here as it simply means that write_cache_pages() will see it under | |
739 | * writeback and ignore the page until some point in the future, at | |
740 | * which time this will be the only page in the file that needs | |
741 | * writeback. Hence for more optimal IO patterns, we should always | |
742 | * avoid partial page writeback due to multiple mappings on a page here. | |
743 | */ | |
744 | if (!xfs_imap_valid(inode, imap, end_offset)) | |
745 | goto fail_unlock_page; | |
746 | ||
747 | len = 1 << inode->i_blkbits; | |
748 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), | |
749 | PAGE_CACHE_SIZE); | |
750 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
751 | page_dirty = p_offset / len; | |
752 | ||
753 | /* | |
754 | * The moment we find a buffer that doesn't match our current type | |
755 | * specification or can't be written, abort the loop and start | |
756 | * writeback. As per the above xfs_imap_valid() check, only | |
757 | * xfs_vm_writepage() can handle partial page writeback fully - we are | |
758 | * limited here to the buffers that are contiguous with the current | |
759 | * ioend, and hence a buffer we can't write breaks that contiguity and | |
760 | * we have to defer the rest of the IO to xfs_vm_writepage(). | |
761 | */ | |
762 | bh = head = page_buffers(page); | |
763 | do { | |
764 | if (offset >= end_offset) | |
765 | break; | |
766 | if (!buffer_uptodate(bh)) | |
767 | uptodate = 0; | |
768 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
769 | done = 1; | |
770 | break; | |
771 | } | |
772 | ||
773 | if (buffer_unwritten(bh) || buffer_delay(bh) || | |
774 | buffer_mapped(bh)) { | |
775 | if (buffer_unwritten(bh)) | |
776 | type = XFS_IO_UNWRITTEN; | |
777 | else if (buffer_delay(bh)) | |
778 | type = XFS_IO_DELALLOC; | |
779 | else | |
780 | type = XFS_IO_OVERWRITE; | |
781 | ||
782 | /* | |
783 | * imap should always be valid because of the above | |
784 | * partial page end_offset check on the imap. | |
785 | */ | |
786 | ASSERT(xfs_imap_valid(inode, imap, offset)); | |
787 | ||
788 | lock_buffer(bh); | |
789 | if (type != XFS_IO_OVERWRITE) | |
790 | xfs_map_at_offset(inode, bh, imap, offset); | |
791 | xfs_add_to_ioend(inode, bh, offset, type, | |
792 | ioendp, done); | |
793 | ||
794 | page_dirty--; | |
795 | count++; | |
796 | } else { | |
797 | done = 1; | |
798 | break; | |
799 | } | |
800 | } while (offset += len, (bh = bh->b_this_page) != head); | |
801 | ||
802 | if (uptodate && bh == head) | |
803 | SetPageUptodate(page); | |
804 | ||
805 | if (count) { | |
806 | if (--wbc->nr_to_write <= 0 && | |
807 | wbc->sync_mode == WB_SYNC_NONE) | |
808 | done = 1; | |
809 | } | |
810 | xfs_start_page_writeback(page, !page_dirty, count); | |
811 | ||
812 | return done; | |
813 | fail_unlock_page: | |
814 | unlock_page(page); | |
815 | fail: | |
816 | return 1; | |
817 | } | |
818 | ||
819 | /* | |
820 | * Convert & write out a cluster of pages in the same extent as defined | |
821 | * by mp and following the start page. | |
822 | */ | |
823 | STATIC void | |
824 | xfs_cluster_write( | |
825 | struct inode *inode, | |
826 | pgoff_t tindex, | |
827 | struct xfs_bmbt_irec *imap, | |
828 | xfs_ioend_t **ioendp, | |
829 | struct writeback_control *wbc, | |
830 | pgoff_t tlast) | |
831 | { | |
832 | struct pagevec pvec; | |
833 | int done = 0, i; | |
834 | ||
835 | pagevec_init(&pvec, 0); | |
836 | while (!done && tindex <= tlast) { | |
837 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
838 | ||
839 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
840 | break; | |
841 | ||
842 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
843 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
844 | imap, ioendp, wbc); | |
845 | if (done) | |
846 | break; | |
847 | } | |
848 | ||
849 | pagevec_release(&pvec); | |
850 | cond_resched(); | |
851 | } | |
852 | } | |
853 | ||
854 | STATIC void | |
855 | xfs_vm_invalidatepage( | |
856 | struct page *page, | |
857 | unsigned int offset, | |
858 | unsigned int length) | |
859 | { | |
860 | trace_xfs_invalidatepage(page->mapping->host, page, offset, | |
861 | length); | |
862 | block_invalidatepage(page, offset, length); | |
863 | } | |
864 | ||
865 | /* | |
866 | * If the page has delalloc buffers on it, we need to punch them out before we | |
867 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
868 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
869 | * is done on that same region - the delalloc extent is returned when none is | |
870 | * supposed to be there. | |
871 | * | |
872 | * We prevent this by truncating away the delalloc regions on the page before | |
873 | * invalidating it. Because they are delalloc, we can do this without needing a | |
874 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
875 | * truncation without a transaction as there is no space left for block | |
876 | * reservation (typically why we see a ENOSPC in writeback). | |
877 | * | |
878 | * This is not a performance critical path, so for now just do the punching a | |
879 | * buffer head at a time. | |
880 | */ | |
881 | STATIC void | |
882 | xfs_aops_discard_page( | |
883 | struct page *page) | |
884 | { | |
885 | struct inode *inode = page->mapping->host; | |
886 | struct xfs_inode *ip = XFS_I(inode); | |
887 | struct buffer_head *bh, *head; | |
888 | loff_t offset = page_offset(page); | |
889 | ||
890 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) | |
891 | goto out_invalidate; | |
892 | ||
893 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
894 | goto out_invalidate; | |
895 | ||
896 | xfs_alert(ip->i_mount, | |
897 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
898 | page, ip->i_ino, offset); | |
899 | ||
900 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
901 | bh = head = page_buffers(page); | |
902 | do { | |
903 | int error; | |
904 | xfs_fileoff_t start_fsb; | |
905 | ||
906 | if (!buffer_delay(bh)) | |
907 | goto next_buffer; | |
908 | ||
909 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
910 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
911 | if (error) { | |
912 | /* something screwed, just bail */ | |
913 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
914 | xfs_alert(ip->i_mount, | |
915 | "page discard unable to remove delalloc mapping."); | |
916 | } | |
917 | break; | |
918 | } | |
919 | next_buffer: | |
920 | offset += 1 << inode->i_blkbits; | |
921 | ||
922 | } while ((bh = bh->b_this_page) != head); | |
923 | ||
924 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
925 | out_invalidate: | |
926 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
927 | return; | |
928 | } | |
929 | ||
930 | /* | |
931 | * Write out a dirty page. | |
932 | * | |
933 | * For delalloc space on the page we need to allocate space and flush it. | |
934 | * For unwritten space on the page we need to start the conversion to | |
935 | * regular allocated space. | |
936 | * For any other dirty buffer heads on the page we should flush them. | |
937 | */ | |
938 | STATIC int | |
939 | xfs_vm_writepage( | |
940 | struct page *page, | |
941 | struct writeback_control *wbc) | |
942 | { | |
943 | struct inode *inode = page->mapping->host; | |
944 | struct buffer_head *bh, *head; | |
945 | struct xfs_bmbt_irec imap; | |
946 | xfs_ioend_t *ioend = NULL, *iohead = NULL; | |
947 | loff_t offset; | |
948 | unsigned int type; | |
949 | __uint64_t end_offset; | |
950 | pgoff_t end_index, last_index; | |
951 | ssize_t len; | |
952 | int err, imap_valid = 0, uptodate = 1; | |
953 | int count = 0; | |
954 | int nonblocking = 0; | |
955 | ||
956 | trace_xfs_writepage(inode, page, 0, 0); | |
957 | ||
958 | ASSERT(page_has_buffers(page)); | |
959 | ||
960 | /* | |
961 | * Refuse to write the page out if we are called from reclaim context. | |
962 | * | |
963 | * This avoids stack overflows when called from deeply used stacks in | |
964 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
965 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
966 | * | |
967 | * This should never happen except in the case of a VM regression so | |
968 | * warn about it. | |
969 | */ | |
970 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == | |
971 | PF_MEMALLOC)) | |
972 | goto redirty; | |
973 | ||
974 | /* | |
975 | * Given that we do not allow direct reclaim to call us, we should | |
976 | * never be called while in a filesystem transaction. | |
977 | */ | |
978 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) | |
979 | goto redirty; | |
980 | ||
981 | /* Is this page beyond the end of the file? */ | |
982 | offset = i_size_read(inode); | |
983 | end_index = offset >> PAGE_CACHE_SHIFT; | |
984 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
985 | ||
986 | /* | |
987 | * The page index is less than the end_index, adjust the end_offset | |
988 | * to the highest offset that this page should represent. | |
989 | * ----------------------------------------------------- | |
990 | * | file mapping | <EOF> | | |
991 | * ----------------------------------------------------- | |
992 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
993 | * ^--------------------------------^----------|-------- | |
994 | * | desired writeback range | see else | | |
995 | * ---------------------------------^------------------| | |
996 | */ | |
997 | if (page->index < end_index) | |
998 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; | |
999 | else { | |
1000 | /* | |
1001 | * Check whether the page to write out is beyond or straddles | |
1002 | * i_size or not. | |
1003 | * ------------------------------------------------------- | |
1004 | * | file mapping | <EOF> | | |
1005 | * ------------------------------------------------------- | |
1006 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1007 | * ^--------------------------------^-----------|--------- | |
1008 | * | | Straddles | | |
1009 | * ---------------------------------^-----------|--------| | |
1010 | */ | |
1011 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); | |
1012 | ||
1013 | /* | |
1014 | * Skip the page if it is fully outside i_size, e.g. due to a | |
1015 | * truncate operation that is in progress. We must redirty the | |
1016 | * page so that reclaim stops reclaiming it. Otherwise | |
1017 | * xfs_vm_releasepage() is called on it and gets confused. | |
1018 | * | |
1019 | * Note that the end_index is unsigned long, it would overflow | |
1020 | * if the given offset is greater than 16TB on 32-bit system | |
1021 | * and if we do check the page is fully outside i_size or not | |
1022 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1023 | * will be evaluated to 0. Hence this page will be redirtied | |
1024 | * and be written out repeatedly which would result in an | |
1025 | * infinite loop, the user program that perform this operation | |
1026 | * will hang. Instead, we can verify this situation by checking | |
1027 | * if the page to write is totally beyond the i_size or if it's | |
1028 | * offset is just equal to the EOF. | |
1029 | */ | |
1030 | if (page->index > end_index || | |
1031 | (page->index == end_index && offset_into_page == 0)) | |
1032 | goto redirty; | |
1033 | ||
1034 | /* | |
1035 | * The page straddles i_size. It must be zeroed out on each | |
1036 | * and every writepage invocation because it may be mmapped. | |
1037 | * "A file is mapped in multiples of the page size. For a file | |
1038 | * that is not a multiple of the page size, the remaining | |
1039 | * memory is zeroed when mapped, and writes to that region are | |
1040 | * not written out to the file." | |
1041 | */ | |
1042 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
1043 | ||
1044 | /* Adjust the end_offset to the end of file */ | |
1045 | end_offset = offset; | |
1046 | } | |
1047 | ||
1048 | len = 1 << inode->i_blkbits; | |
1049 | ||
1050 | bh = head = page_buffers(page); | |
1051 | offset = page_offset(page); | |
1052 | type = XFS_IO_OVERWRITE; | |
1053 | ||
1054 | if (wbc->sync_mode == WB_SYNC_NONE) | |
1055 | nonblocking = 1; | |
1056 | ||
1057 | do { | |
1058 | int new_ioend = 0; | |
1059 | ||
1060 | if (offset >= end_offset) | |
1061 | break; | |
1062 | if (!buffer_uptodate(bh)) | |
1063 | uptodate = 0; | |
1064 | ||
1065 | /* | |
1066 | * set_page_dirty dirties all buffers in a page, independent | |
1067 | * of their state. The dirty state however is entirely | |
1068 | * meaningless for holes (!mapped && uptodate), so skip | |
1069 | * buffers covering holes here. | |
1070 | */ | |
1071 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
1072 | imap_valid = 0; | |
1073 | continue; | |
1074 | } | |
1075 | ||
1076 | if (buffer_unwritten(bh)) { | |
1077 | if (type != XFS_IO_UNWRITTEN) { | |
1078 | type = XFS_IO_UNWRITTEN; | |
1079 | imap_valid = 0; | |
1080 | } | |
1081 | } else if (buffer_delay(bh)) { | |
1082 | if (type != XFS_IO_DELALLOC) { | |
1083 | type = XFS_IO_DELALLOC; | |
1084 | imap_valid = 0; | |
1085 | } | |
1086 | } else if (buffer_uptodate(bh)) { | |
1087 | if (type != XFS_IO_OVERWRITE) { | |
1088 | type = XFS_IO_OVERWRITE; | |
1089 | imap_valid = 0; | |
1090 | } | |
1091 | } else { | |
1092 | if (PageUptodate(page)) | |
1093 | ASSERT(buffer_mapped(bh)); | |
1094 | /* | |
1095 | * This buffer is not uptodate and will not be | |
1096 | * written to disk. Ensure that we will put any | |
1097 | * subsequent writeable buffers into a new | |
1098 | * ioend. | |
1099 | */ | |
1100 | imap_valid = 0; | |
1101 | continue; | |
1102 | } | |
1103 | ||
1104 | if (imap_valid) | |
1105 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1106 | if (!imap_valid) { | |
1107 | /* | |
1108 | * If we didn't have a valid mapping then we need to | |
1109 | * put the new mapping into a separate ioend structure. | |
1110 | * This ensures non-contiguous extents always have | |
1111 | * separate ioends, which is particularly important | |
1112 | * for unwritten extent conversion at I/O completion | |
1113 | * time. | |
1114 | */ | |
1115 | new_ioend = 1; | |
1116 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1117 | nonblocking); | |
1118 | if (err) | |
1119 | goto error; | |
1120 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1121 | } | |
1122 | if (imap_valid) { | |
1123 | lock_buffer(bh); | |
1124 | if (type != XFS_IO_OVERWRITE) | |
1125 | xfs_map_at_offset(inode, bh, &imap, offset); | |
1126 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1127 | new_ioend); | |
1128 | count++; | |
1129 | } | |
1130 | ||
1131 | if (!iohead) | |
1132 | iohead = ioend; | |
1133 | ||
1134 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1135 | ||
1136 | if (uptodate && bh == head) | |
1137 | SetPageUptodate(page); | |
1138 | ||
1139 | xfs_start_page_writeback(page, 1, count); | |
1140 | ||
1141 | /* if there is no IO to be submitted for this page, we are done */ | |
1142 | if (!ioend) | |
1143 | return 0; | |
1144 | ||
1145 | ASSERT(iohead); | |
1146 | ||
1147 | /* | |
1148 | * Any errors from this point onwards need tobe reported through the IO | |
1149 | * completion path as we have marked the initial page as under writeback | |
1150 | * and unlocked it. | |
1151 | */ | |
1152 | if (imap_valid) { | |
1153 | xfs_off_t end_index; | |
1154 | ||
1155 | end_index = imap.br_startoff + imap.br_blockcount; | |
1156 | ||
1157 | /* to bytes */ | |
1158 | end_index <<= inode->i_blkbits; | |
1159 | ||
1160 | /* to pages */ | |
1161 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1162 | ||
1163 | /* check against file size */ | |
1164 | if (end_index > last_index) | |
1165 | end_index = last_index; | |
1166 | ||
1167 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, | |
1168 | wbc, end_index); | |
1169 | } | |
1170 | ||
1171 | ||
1172 | /* | |
1173 | * Reserve log space if we might write beyond the on-disk inode size. | |
1174 | */ | |
1175 | err = 0; | |
1176 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) | |
1177 | err = xfs_setfilesize_trans_alloc(ioend); | |
1178 | ||
1179 | xfs_submit_ioend(wbc, iohead, err); | |
1180 | ||
1181 | return 0; | |
1182 | ||
1183 | error: | |
1184 | if (iohead) | |
1185 | xfs_cancel_ioend(iohead); | |
1186 | ||
1187 | if (err == -EAGAIN) | |
1188 | goto redirty; | |
1189 | ||
1190 | xfs_aops_discard_page(page); | |
1191 | ClearPageUptodate(page); | |
1192 | unlock_page(page); | |
1193 | return err; | |
1194 | ||
1195 | redirty: | |
1196 | redirty_page_for_writepage(wbc, page); | |
1197 | unlock_page(page); | |
1198 | return 0; | |
1199 | } | |
1200 | ||
1201 | STATIC int | |
1202 | xfs_vm_writepages( | |
1203 | struct address_space *mapping, | |
1204 | struct writeback_control *wbc) | |
1205 | { | |
1206 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); | |
1207 | return generic_writepages(mapping, wbc); | |
1208 | } | |
1209 | ||
1210 | /* | |
1211 | * Called to move a page into cleanable state - and from there | |
1212 | * to be released. The page should already be clean. We always | |
1213 | * have buffer heads in this call. | |
1214 | * | |
1215 | * Returns 1 if the page is ok to release, 0 otherwise. | |
1216 | */ | |
1217 | STATIC int | |
1218 | xfs_vm_releasepage( | |
1219 | struct page *page, | |
1220 | gfp_t gfp_mask) | |
1221 | { | |
1222 | int delalloc, unwritten; | |
1223 | ||
1224 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); | |
1225 | ||
1226 | xfs_count_page_state(page, &delalloc, &unwritten); | |
1227 | ||
1228 | if (WARN_ON_ONCE(delalloc)) | |
1229 | return 0; | |
1230 | if (WARN_ON_ONCE(unwritten)) | |
1231 | return 0; | |
1232 | ||
1233 | return try_to_free_buffers(page); | |
1234 | } | |
1235 | ||
1236 | /* | |
1237 | * When we map a DIO buffer, we need to attach an ioend that describes the type | |
1238 | * of write IO we are doing. This passes to the completion function the | |
1239 | * operations it needs to perform. | |
1240 | * | |
1241 | * If we get multiple mappings in a single IO, we might be mapping different | |
1242 | * types. But because the direct IO can only have a single private pointer, we | |
1243 | * need to ensure that: | |
1244 | * | |
1245 | * a) the ioend spans the entire region of the IO; and | |
1246 | * b) if it contains unwritten extents, it is *permanently* marked as such | |
1247 | * | |
1248 | * We could do this by chaining ioends like buffered IO does, but we only | |
1249 | * actually get one IO completion callback from the direct IO, and that spans | |
1250 | * the entire IO regardless of how many mappings and IOs are needed to complete | |
1251 | * the DIO. There is only going to be one reference to the ioend and its life | |
1252 | * cycle is constrained by the DIO completion code. hence we don't need | |
1253 | * reference counting here. | |
1254 | */ | |
1255 | static void | |
1256 | xfs_map_direct( | |
1257 | struct inode *inode, | |
1258 | struct buffer_head *bh_result, | |
1259 | struct xfs_bmbt_irec *imap, | |
1260 | xfs_off_t offset) | |
1261 | { | |
1262 | struct xfs_ioend *ioend; | |
1263 | xfs_off_t size = bh_result->b_size; | |
1264 | int type; | |
1265 | ||
1266 | if (ISUNWRITTEN(imap)) | |
1267 | type = XFS_IO_UNWRITTEN; | |
1268 | else | |
1269 | type = XFS_IO_OVERWRITE; | |
1270 | ||
1271 | trace_xfs_gbmap_direct(XFS_I(inode), offset, size, type, imap); | |
1272 | ||
1273 | if (bh_result->b_private) { | |
1274 | ioend = bh_result->b_private; | |
1275 | ASSERT(ioend->io_size > 0); | |
1276 | ASSERT(offset >= ioend->io_offset); | |
1277 | if (offset + size > ioend->io_offset + ioend->io_size) | |
1278 | ioend->io_size = offset - ioend->io_offset + size; | |
1279 | ||
1280 | if (type == XFS_IO_UNWRITTEN && type != ioend->io_type) | |
1281 | ioend->io_type = XFS_IO_UNWRITTEN; | |
1282 | ||
1283 | trace_xfs_gbmap_direct_update(XFS_I(inode), ioend->io_offset, | |
1284 | ioend->io_size, ioend->io_type, | |
1285 | imap); | |
1286 | } else { | |
1287 | ioend = xfs_alloc_ioend(inode, type); | |
1288 | ioend->io_offset = offset; | |
1289 | ioend->io_size = size; | |
1290 | bh_result->b_private = ioend; | |
1291 | ||
1292 | trace_xfs_gbmap_direct_new(XFS_I(inode), offset, size, type, | |
1293 | imap); | |
1294 | } | |
1295 | ||
1296 | if (ioend->io_type == XFS_IO_UNWRITTEN || xfs_ioend_is_append(ioend)) | |
1297 | set_buffer_defer_completion(bh_result); | |
1298 | } | |
1299 | ||
1300 | ||
1301 | /* | |
1302 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1303 | * is, so that we can avoid repeated get_blocks calls. | |
1304 | * | |
1305 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1306 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1307 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1308 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1309 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1310 | * to and including EOF, and a second mapping for beyond EOF. | |
1311 | */ | |
1312 | static void | |
1313 | xfs_map_trim_size( | |
1314 | struct inode *inode, | |
1315 | sector_t iblock, | |
1316 | struct buffer_head *bh_result, | |
1317 | struct xfs_bmbt_irec *imap, | |
1318 | xfs_off_t offset, | |
1319 | ssize_t size) | |
1320 | { | |
1321 | xfs_off_t mapping_size; | |
1322 | ||
1323 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1324 | mapping_size <<= inode->i_blkbits; | |
1325 | ||
1326 | ASSERT(mapping_size > 0); | |
1327 | if (mapping_size > size) | |
1328 | mapping_size = size; | |
1329 | if (offset < i_size_read(inode) && | |
1330 | offset + mapping_size >= i_size_read(inode)) { | |
1331 | /* limit mapping to block that spans EOF */ | |
1332 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1333 | 1 << inode->i_blkbits); | |
1334 | } | |
1335 | if (mapping_size > LONG_MAX) | |
1336 | mapping_size = LONG_MAX; | |
1337 | ||
1338 | bh_result->b_size = mapping_size; | |
1339 | } | |
1340 | ||
1341 | STATIC int | |
1342 | __xfs_get_blocks( | |
1343 | struct inode *inode, | |
1344 | sector_t iblock, | |
1345 | struct buffer_head *bh_result, | |
1346 | int create, | |
1347 | int direct) | |
1348 | { | |
1349 | struct xfs_inode *ip = XFS_I(inode); | |
1350 | struct xfs_mount *mp = ip->i_mount; | |
1351 | xfs_fileoff_t offset_fsb, end_fsb; | |
1352 | int error = 0; | |
1353 | int lockmode = 0; | |
1354 | struct xfs_bmbt_irec imap; | |
1355 | int nimaps = 1; | |
1356 | xfs_off_t offset; | |
1357 | ssize_t size; | |
1358 | int new = 0; | |
1359 | ||
1360 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1361 | return -EIO; | |
1362 | ||
1363 | offset = (xfs_off_t)iblock << inode->i_blkbits; | |
1364 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); | |
1365 | size = bh_result->b_size; | |
1366 | ||
1367 | if (!create && direct && offset >= i_size_read(inode)) | |
1368 | return 0; | |
1369 | ||
1370 | /* | |
1371 | * Direct I/O is usually done on preallocated files, so try getting | |
1372 | * a block mapping without an exclusive lock first. For buffered | |
1373 | * writes we already have the exclusive iolock anyway, so avoiding | |
1374 | * a lock roundtrip here by taking the ilock exclusive from the | |
1375 | * beginning is a useful micro optimization. | |
1376 | */ | |
1377 | if (create && !direct) { | |
1378 | lockmode = XFS_ILOCK_EXCL; | |
1379 | xfs_ilock(ip, lockmode); | |
1380 | } else { | |
1381 | lockmode = xfs_ilock_data_map_shared(ip); | |
1382 | } | |
1383 | ||
1384 | ASSERT(offset <= mp->m_super->s_maxbytes); | |
1385 | if (offset + size > mp->m_super->s_maxbytes) | |
1386 | size = mp->m_super->s_maxbytes - offset; | |
1387 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); | |
1388 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1389 | ||
1390 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, | |
1391 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1392 | if (error) | |
1393 | goto out_unlock; | |
1394 | ||
1395 | if (create && | |
1396 | (!nimaps || | |
1397 | (imap.br_startblock == HOLESTARTBLOCK || | |
1398 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
1399 | if (direct || xfs_get_extsz_hint(ip)) { | |
1400 | /* | |
1401 | * Drop the ilock in preparation for starting the block | |
1402 | * allocation transaction. It will be retaken | |
1403 | * exclusively inside xfs_iomap_write_direct for the | |
1404 | * actual allocation. | |
1405 | */ | |
1406 | xfs_iunlock(ip, lockmode); | |
1407 | error = xfs_iomap_write_direct(ip, offset, size, | |
1408 | &imap, nimaps); | |
1409 | if (error) | |
1410 | return error; | |
1411 | new = 1; | |
1412 | } else { | |
1413 | /* | |
1414 | * Delalloc reservations do not require a transaction, | |
1415 | * we can go on without dropping the lock here. If we | |
1416 | * are allocating a new delalloc block, make sure that | |
1417 | * we set the new flag so that we mark the buffer new so | |
1418 | * that we know that it is newly allocated if the write | |
1419 | * fails. | |
1420 | */ | |
1421 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) | |
1422 | new = 1; | |
1423 | error = xfs_iomap_write_delay(ip, offset, size, &imap); | |
1424 | if (error) | |
1425 | goto out_unlock; | |
1426 | ||
1427 | xfs_iunlock(ip, lockmode); | |
1428 | } | |
1429 | trace_xfs_get_blocks_alloc(ip, offset, size, | |
1430 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1431 | : XFS_IO_DELALLOC, &imap); | |
1432 | } else if (nimaps) { | |
1433 | trace_xfs_get_blocks_found(ip, offset, size, | |
1434 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1435 | : XFS_IO_OVERWRITE, &imap); | |
1436 | xfs_iunlock(ip, lockmode); | |
1437 | } else { | |
1438 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1439 | goto out_unlock; | |
1440 | } | |
1441 | ||
1442 | /* trim mapping down to size requested */ | |
1443 | if (direct || size > (1 << inode->i_blkbits)) | |
1444 | xfs_map_trim_size(inode, iblock, bh_result, | |
1445 | &imap, offset, size); | |
1446 | ||
1447 | /* | |
1448 | * For unwritten extents do not report a disk address in the buffered | |
1449 | * read case (treat as if we're reading into a hole). | |
1450 | */ | |
1451 | if (imap.br_startblock != HOLESTARTBLOCK && | |
1452 | imap.br_startblock != DELAYSTARTBLOCK && | |
1453 | (create || !ISUNWRITTEN(&imap))) { | |
1454 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1455 | if (ISUNWRITTEN(&imap)) | |
1456 | set_buffer_unwritten(bh_result); | |
1457 | /* direct IO needs special help */ | |
1458 | if (create && direct) | |
1459 | xfs_map_direct(inode, bh_result, &imap, offset); | |
1460 | } | |
1461 | ||
1462 | /* | |
1463 | * If this is a realtime file, data may be on a different device. | |
1464 | * to that pointed to from the buffer_head b_bdev currently. | |
1465 | */ | |
1466 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); | |
1467 | ||
1468 | /* | |
1469 | * If we previously allocated a block out beyond eof and we are now | |
1470 | * coming back to use it then we will need to flag it as new even if it | |
1471 | * has a disk address. | |
1472 | * | |
1473 | * With sub-block writes into unwritten extents we also need to mark | |
1474 | * the buffer as new so that the unwritten parts of the buffer gets | |
1475 | * correctly zeroed. | |
1476 | */ | |
1477 | if (create && | |
1478 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
1479 | (offset >= i_size_read(inode)) || | |
1480 | (new || ISUNWRITTEN(&imap)))) | |
1481 | set_buffer_new(bh_result); | |
1482 | ||
1483 | if (imap.br_startblock == DELAYSTARTBLOCK) { | |
1484 | BUG_ON(direct); | |
1485 | if (create) { | |
1486 | set_buffer_uptodate(bh_result); | |
1487 | set_buffer_mapped(bh_result); | |
1488 | set_buffer_delay(bh_result); | |
1489 | } | |
1490 | } | |
1491 | ||
1492 | return 0; | |
1493 | ||
1494 | out_unlock: | |
1495 | xfs_iunlock(ip, lockmode); | |
1496 | return error; | |
1497 | } | |
1498 | ||
1499 | int | |
1500 | xfs_get_blocks( | |
1501 | struct inode *inode, | |
1502 | sector_t iblock, | |
1503 | struct buffer_head *bh_result, | |
1504 | int create) | |
1505 | { | |
1506 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); | |
1507 | } | |
1508 | ||
1509 | STATIC int | |
1510 | xfs_get_blocks_direct( | |
1511 | struct inode *inode, | |
1512 | sector_t iblock, | |
1513 | struct buffer_head *bh_result, | |
1514 | int create) | |
1515 | { | |
1516 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); | |
1517 | } | |
1518 | ||
1519 | /* | |
1520 | * Complete a direct I/O write request. | |
1521 | * | |
1522 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1523 | * need to issue a transaction to convert the range from unwritten to written | |
1524 | * extents. | |
1525 | */ | |
1526 | STATIC void | |
1527 | xfs_end_io_direct_write( | |
1528 | struct kiocb *iocb, | |
1529 | loff_t offset, | |
1530 | ssize_t size, | |
1531 | void *private) | |
1532 | { | |
1533 | struct inode *inode = file_inode(iocb->ki_filp); | |
1534 | struct xfs_inode *ip = XFS_I(inode); | |
1535 | struct xfs_mount *mp = ip->i_mount; | |
1536 | struct xfs_ioend *ioend = private; | |
1537 | ||
1538 | trace_xfs_gbmap_direct_endio(ip, offset, size, ioend->io_type, NULL); | |
1539 | ||
1540 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1541 | goto out_end_io; | |
1542 | ||
1543 | /* | |
1544 | * dio completion end_io functions are only called on writes if more | |
1545 | * than 0 bytes was written. | |
1546 | */ | |
1547 | ASSERT(size > 0); | |
1548 | ||
1549 | /* | |
1550 | * The ioend only maps whole blocks, while the IO may be sector aligned. | |
1551 | * Hence the ioend offset/size may not match the IO offset/size exactly, | |
1552 | * but should span it completely. Write the IO sizes into the ioend so | |
1553 | * that completion processing does the right thing. | |
1554 | */ | |
1555 | ASSERT(size <= ioend->io_size); | |
1556 | ASSERT(offset >= ioend->io_offset); | |
1557 | ASSERT(offset + size <= ioend->io_offset + ioend->io_size); | |
1558 | ioend->io_size = size; | |
1559 | ioend->io_offset = offset; | |
1560 | ||
1561 | /* | |
1562 | * The ioend tells us whether we are doing unwritten extent conversion | |
1563 | * or an append transaction that updates the on-disk file size. These | |
1564 | * cases are the only cases where we should *potentially* be needing | |
1565 | * to update the VFS inode size. When the ioend indicates this, we | |
1566 | * are *guaranteed* to be running in non-interrupt context. | |
1567 | * | |
1568 | * We need to update the in-core inode size here so that we don't end up | |
1569 | * with the on-disk inode size being outside the in-core inode size. | |
1570 | * While we can do this in the process context after the IO has | |
1571 | * completed, this does not work for AIO and hence we always update | |
1572 | * the in-core inode size here if necessary. | |
1573 | */ | |
1574 | if (ioend->io_type == XFS_IO_UNWRITTEN || xfs_ioend_is_append(ioend)) { | |
1575 | if (offset + size > i_size_read(inode)) | |
1576 | i_size_write(inode, offset + size); | |
1577 | } else | |
1578 | ASSERT(offset + size <= i_size_read(inode)); | |
1579 | ||
1580 | /* | |
1581 | * If we are doing an append IO that needs to update the EOF on disk, | |
1582 | * do the transaction reserve now so we can use common end io | |
1583 | * processing. Stashing the error (if there is one) in the ioend will | |
1584 | * result in the ioend processing passing on the error if it is | |
1585 | * possible as we can't return it from here. | |
1586 | */ | |
1587 | if (ioend->io_type == XFS_IO_OVERWRITE && xfs_ioend_is_append(ioend)) | |
1588 | ioend->io_error = xfs_setfilesize_trans_alloc(ioend); | |
1589 | ||
1590 | out_end_io: | |
1591 | xfs_end_io(&ioend->io_work); | |
1592 | return; | |
1593 | } | |
1594 | ||
1595 | STATIC ssize_t | |
1596 | xfs_vm_direct_IO( | |
1597 | int rw, | |
1598 | struct kiocb *iocb, | |
1599 | struct iov_iter *iter, | |
1600 | loff_t offset) | |
1601 | { | |
1602 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
1603 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1604 | ||
1605 | if (rw & WRITE) { | |
1606 | return __blockdev_direct_IO(rw, iocb, inode, bdev, iter, | |
1607 | offset, xfs_get_blocks_direct, | |
1608 | xfs_end_io_direct_write, NULL, | |
1609 | DIO_ASYNC_EXTEND); | |
1610 | } | |
1611 | return __blockdev_direct_IO(rw, iocb, inode, bdev, iter, | |
1612 | offset, xfs_get_blocks_direct, | |
1613 | NULL, NULL, 0); | |
1614 | } | |
1615 | ||
1616 | /* | |
1617 | * Punch out the delalloc blocks we have already allocated. | |
1618 | * | |
1619 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1620 | * as the page is still locked at this point. | |
1621 | */ | |
1622 | STATIC void | |
1623 | xfs_vm_kill_delalloc_range( | |
1624 | struct inode *inode, | |
1625 | loff_t start, | |
1626 | loff_t end) | |
1627 | { | |
1628 | struct xfs_inode *ip = XFS_I(inode); | |
1629 | xfs_fileoff_t start_fsb; | |
1630 | xfs_fileoff_t end_fsb; | |
1631 | int error; | |
1632 | ||
1633 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1634 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1635 | if (end_fsb <= start_fsb) | |
1636 | return; | |
1637 | ||
1638 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1639 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1640 | end_fsb - start_fsb); | |
1641 | if (error) { | |
1642 | /* something screwed, just bail */ | |
1643 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1644 | xfs_alert(ip->i_mount, | |
1645 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1646 | ip->i_ino); | |
1647 | } | |
1648 | } | |
1649 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1650 | } | |
1651 | ||
1652 | STATIC void | |
1653 | xfs_vm_write_failed( | |
1654 | struct inode *inode, | |
1655 | struct page *page, | |
1656 | loff_t pos, | |
1657 | unsigned len) | |
1658 | { | |
1659 | loff_t block_offset; | |
1660 | loff_t block_start; | |
1661 | loff_t block_end; | |
1662 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1663 | loff_t to = from + len; | |
1664 | struct buffer_head *bh, *head; | |
1665 | ||
1666 | /* | |
1667 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1668 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1669 | * platform, the high 32-bit will be masked off if we evaluate the | |
1670 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1671 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1672 | * which could cause the following ASSERT failed in most cases. | |
1673 | * In order to avoid this, we can evaluate the block_offset of the | |
1674 | * start of the page by using shifts rather than masks the mismatch | |
1675 | * problem. | |
1676 | */ | |
1677 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1678 | ||
1679 | ASSERT(block_offset + from == pos); | |
1680 | ||
1681 | head = page_buffers(page); | |
1682 | block_start = 0; | |
1683 | for (bh = head; bh != head || !block_start; | |
1684 | bh = bh->b_this_page, block_start = block_end, | |
1685 | block_offset += bh->b_size) { | |
1686 | block_end = block_start + bh->b_size; | |
1687 | ||
1688 | /* skip buffers before the write */ | |
1689 | if (block_end <= from) | |
1690 | continue; | |
1691 | ||
1692 | /* if the buffer is after the write, we're done */ | |
1693 | if (block_start >= to) | |
1694 | break; | |
1695 | ||
1696 | if (!buffer_delay(bh)) | |
1697 | continue; | |
1698 | ||
1699 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1700 | continue; | |
1701 | ||
1702 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1703 | block_offset + bh->b_size); | |
1704 | ||
1705 | /* | |
1706 | * This buffer does not contain data anymore. make sure anyone | |
1707 | * who finds it knows that for certain. | |
1708 | */ | |
1709 | clear_buffer_delay(bh); | |
1710 | clear_buffer_uptodate(bh); | |
1711 | clear_buffer_mapped(bh); | |
1712 | clear_buffer_new(bh); | |
1713 | clear_buffer_dirty(bh); | |
1714 | } | |
1715 | ||
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1720 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1721 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1722 | * the appropriate point. | |
1723 | */ | |
1724 | STATIC int | |
1725 | xfs_vm_write_begin( | |
1726 | struct file *file, | |
1727 | struct address_space *mapping, | |
1728 | loff_t pos, | |
1729 | unsigned len, | |
1730 | unsigned flags, | |
1731 | struct page **pagep, | |
1732 | void **fsdata) | |
1733 | { | |
1734 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
1735 | struct page *page; | |
1736 | int status; | |
1737 | ||
1738 | ASSERT(len <= PAGE_CACHE_SIZE); | |
1739 | ||
1740 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1741 | if (!page) | |
1742 | return -ENOMEM; | |
1743 | ||
1744 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1745 | if (unlikely(status)) { | |
1746 | struct inode *inode = mapping->host; | |
1747 | size_t isize = i_size_read(inode); | |
1748 | ||
1749 | xfs_vm_write_failed(inode, page, pos, len); | |
1750 | unlock_page(page); | |
1751 | ||
1752 | /* | |
1753 | * If the write is beyond EOF, we only want to kill blocks | |
1754 | * allocated in this write, not blocks that were previously | |
1755 | * written successfully. | |
1756 | */ | |
1757 | if (pos + len > isize) { | |
1758 | ssize_t start = max_t(ssize_t, pos, isize); | |
1759 | ||
1760 | truncate_pagecache_range(inode, start, pos + len); | |
1761 | } | |
1762 | ||
1763 | page_cache_release(page); | |
1764 | page = NULL; | |
1765 | } | |
1766 | ||
1767 | *pagep = page; | |
1768 | return status; | |
1769 | } | |
1770 | ||
1771 | /* | |
1772 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of | |
1773 | * this specific write because they will never be written. Previous writes | |
1774 | * beyond EOF where block allocation succeeded do not need to be trashed, so | |
1775 | * only new blocks from this write should be trashed. For blocks within | |
1776 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be | |
1777 | * written with all the other valid data. | |
1778 | */ | |
1779 | STATIC int | |
1780 | xfs_vm_write_end( | |
1781 | struct file *file, | |
1782 | struct address_space *mapping, | |
1783 | loff_t pos, | |
1784 | unsigned len, | |
1785 | unsigned copied, | |
1786 | struct page *page, | |
1787 | void *fsdata) | |
1788 | { | |
1789 | int ret; | |
1790 | ||
1791 | ASSERT(len <= PAGE_CACHE_SIZE); | |
1792 | ||
1793 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1794 | if (unlikely(ret < len)) { | |
1795 | struct inode *inode = mapping->host; | |
1796 | size_t isize = i_size_read(inode); | |
1797 | loff_t to = pos + len; | |
1798 | ||
1799 | if (to > isize) { | |
1800 | /* only kill blocks in this write beyond EOF */ | |
1801 | if (pos > isize) | |
1802 | isize = pos; | |
1803 | xfs_vm_kill_delalloc_range(inode, isize, to); | |
1804 | truncate_pagecache_range(inode, isize, to); | |
1805 | } | |
1806 | } | |
1807 | return ret; | |
1808 | } | |
1809 | ||
1810 | STATIC sector_t | |
1811 | xfs_vm_bmap( | |
1812 | struct address_space *mapping, | |
1813 | sector_t block) | |
1814 | { | |
1815 | struct inode *inode = (struct inode *)mapping->host; | |
1816 | struct xfs_inode *ip = XFS_I(inode); | |
1817 | ||
1818 | trace_xfs_vm_bmap(XFS_I(inode)); | |
1819 | xfs_ilock(ip, XFS_IOLOCK_SHARED); | |
1820 | filemap_write_and_wait(mapping); | |
1821 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); | |
1822 | return generic_block_bmap(mapping, block, xfs_get_blocks); | |
1823 | } | |
1824 | ||
1825 | STATIC int | |
1826 | xfs_vm_readpage( | |
1827 | struct file *unused, | |
1828 | struct page *page) | |
1829 | { | |
1830 | return mpage_readpage(page, xfs_get_blocks); | |
1831 | } | |
1832 | ||
1833 | STATIC int | |
1834 | xfs_vm_readpages( | |
1835 | struct file *unused, | |
1836 | struct address_space *mapping, | |
1837 | struct list_head *pages, | |
1838 | unsigned nr_pages) | |
1839 | { | |
1840 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); | |
1841 | } | |
1842 | ||
1843 | /* | |
1844 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1845 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1846 | * dirty, we'll never be able to clean them because we don't write buffers | |
1847 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1848 | * that have been marked dirty. Further, the dirty state can leak into | |
1849 | * the file interior if the file is extended, resulting in all sorts of | |
1850 | * bad things happening as the state does not match the underlying data. | |
1851 | * | |
1852 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1853 | * this only exist because of bufferheads and how the generic code manages them. | |
1854 | */ | |
1855 | STATIC int | |
1856 | xfs_vm_set_page_dirty( | |
1857 | struct page *page) | |
1858 | { | |
1859 | struct address_space *mapping = page->mapping; | |
1860 | struct inode *inode = mapping->host; | |
1861 | loff_t end_offset; | |
1862 | loff_t offset; | |
1863 | int newly_dirty; | |
1864 | ||
1865 | if (unlikely(!mapping)) | |
1866 | return !TestSetPageDirty(page); | |
1867 | ||
1868 | end_offset = i_size_read(inode); | |
1869 | offset = page_offset(page); | |
1870 | ||
1871 | spin_lock(&mapping->private_lock); | |
1872 | if (page_has_buffers(page)) { | |
1873 | struct buffer_head *head = page_buffers(page); | |
1874 | struct buffer_head *bh = head; | |
1875 | ||
1876 | do { | |
1877 | if (offset < end_offset) | |
1878 | set_buffer_dirty(bh); | |
1879 | bh = bh->b_this_page; | |
1880 | offset += 1 << inode->i_blkbits; | |
1881 | } while (bh != head); | |
1882 | } | |
1883 | newly_dirty = !TestSetPageDirty(page); | |
1884 | spin_unlock(&mapping->private_lock); | |
1885 | ||
1886 | if (newly_dirty) { | |
1887 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1888 | unsigned long flags; | |
1889 | ||
1890 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1891 | if (page->mapping) { /* Race with truncate? */ | |
1892 | WARN_ON_ONCE(!PageUptodate(page)); | |
1893 | account_page_dirtied(page, mapping); | |
1894 | radix_tree_tag_set(&mapping->page_tree, | |
1895 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1896 | } | |
1897 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
1898 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1899 | } | |
1900 | return newly_dirty; | |
1901 | } | |
1902 | ||
1903 | const struct address_space_operations xfs_address_space_operations = { | |
1904 | .readpage = xfs_vm_readpage, | |
1905 | .readpages = xfs_vm_readpages, | |
1906 | .writepage = xfs_vm_writepage, | |
1907 | .writepages = xfs_vm_writepages, | |
1908 | .set_page_dirty = xfs_vm_set_page_dirty, | |
1909 | .releasepage = xfs_vm_releasepage, | |
1910 | .invalidatepage = xfs_vm_invalidatepage, | |
1911 | .write_begin = xfs_vm_write_begin, | |
1912 | .write_end = xfs_vm_write_end, | |
1913 | .bmap = xfs_vm_bmap, | |
1914 | .direct_IO = xfs_vm_direct_IO, | |
1915 | .migratepage = buffer_migrate_page, | |
1916 | .is_partially_uptodate = block_is_partially_uptodate, | |
1917 | .error_remove_page = generic_error_remove_page, | |
1918 | }; |