]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_bit.h" |
1da177e4 | 20 | #include "xfs_log.h" |
a844f451 | 21 | #include "xfs_inum.h" |
1da177e4 | 22 | #include "xfs_sb.h" |
a844f451 | 23 | #include "xfs_ag.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
1da177e4 LT |
25 | #include "xfs_mount.h" |
26 | #include "xfs_bmap_btree.h" | |
1da177e4 LT |
27 | #include "xfs_dinode.h" |
28 | #include "xfs_inode.h" | |
a844f451 | 29 | #include "xfs_alloc.h" |
1da177e4 LT |
30 | #include "xfs_error.h" |
31 | #include "xfs_rw.h" | |
32 | #include "xfs_iomap.h" | |
739bfb2a | 33 | #include "xfs_vnodeops.h" |
0b1b213f | 34 | #include "xfs_trace.h" |
3ed3a434 | 35 | #include "xfs_bmap.h" |
5a0e3ad6 | 36 | #include <linux/gfp.h> |
1da177e4 | 37 | #include <linux/mpage.h> |
10ce4444 | 38 | #include <linux/pagevec.h> |
1da177e4 LT |
39 | #include <linux/writeback.h> |
40 | ||
34a52c6c CH |
41 | /* |
42 | * Types of I/O for bmap clustering and I/O completion tracking. | |
43 | */ | |
44 | enum { | |
45 | IO_READ, /* mapping for a read */ | |
46 | IO_DELAY, /* mapping covers delalloc region */ | |
47 | IO_UNWRITTEN, /* mapping covers allocated but uninitialized data */ | |
48 | IO_NEW /* just allocated */ | |
49 | }; | |
25e41b3d CH |
50 | |
51 | /* | |
52 | * Prime number of hash buckets since address is used as the key. | |
53 | */ | |
54 | #define NVSYNC 37 | |
55 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
56 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
57 | ||
58 | void __init | |
59 | xfs_ioend_init(void) | |
60 | { | |
61 | int i; | |
62 | ||
63 | for (i = 0; i < NVSYNC; i++) | |
64 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
65 | } | |
66 | ||
67 | void | |
68 | xfs_ioend_wait( | |
69 | xfs_inode_t *ip) | |
70 | { | |
71 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
72 | ||
73 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
74 | } | |
75 | ||
76 | STATIC void | |
77 | xfs_ioend_wake( | |
78 | xfs_inode_t *ip) | |
79 | { | |
80 | if (atomic_dec_and_test(&ip->i_iocount)) | |
81 | wake_up(to_ioend_wq(ip)); | |
82 | } | |
83 | ||
0b1b213f | 84 | void |
f51623b2 NS |
85 | xfs_count_page_state( |
86 | struct page *page, | |
87 | int *delalloc, | |
f51623b2 NS |
88 | int *unwritten) |
89 | { | |
90 | struct buffer_head *bh, *head; | |
91 | ||
20cb52eb | 92 | *delalloc = *unwritten = 0; |
f51623b2 NS |
93 | |
94 | bh = head = page_buffers(page); | |
95 | do { | |
20cb52eb | 96 | if (buffer_unwritten(bh)) |
f51623b2 NS |
97 | (*unwritten) = 1; |
98 | else if (buffer_delay(bh)) | |
99 | (*delalloc) = 1; | |
100 | } while ((bh = bh->b_this_page) != head); | |
101 | } | |
102 | ||
6214ed44 CH |
103 | STATIC struct block_device * |
104 | xfs_find_bdev_for_inode( | |
046f1685 | 105 | struct inode *inode) |
6214ed44 | 106 | { |
046f1685 | 107 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
108 | struct xfs_mount *mp = ip->i_mount; |
109 | ||
71ddabb9 | 110 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
111 | return mp->m_rtdev_targp->bt_bdev; |
112 | else | |
113 | return mp->m_ddev_targp->bt_bdev; | |
114 | } | |
115 | ||
f6d6d4fc CH |
116 | /* |
117 | * We're now finished for good with this ioend structure. | |
118 | * Update the page state via the associated buffer_heads, | |
119 | * release holds on the inode and bio, and finally free | |
120 | * up memory. Do not use the ioend after this. | |
121 | */ | |
0829c360 CH |
122 | STATIC void |
123 | xfs_destroy_ioend( | |
124 | xfs_ioend_t *ioend) | |
125 | { | |
f6d6d4fc | 126 | struct buffer_head *bh, *next; |
583fa586 | 127 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
128 | |
129 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
130 | next = bh->b_private; | |
7d04a335 | 131 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 132 | } |
583fa586 CH |
133 | |
134 | /* | |
135 | * Volume managers supporting multiple paths can send back ENODEV | |
136 | * when the final path disappears. In this case continuing to fill | |
137 | * the page cache with dirty data which cannot be written out is | |
138 | * evil, so prevent that. | |
139 | */ | |
140 | if (unlikely(ioend->io_error == -ENODEV)) { | |
141 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
142 | __FILE__, __LINE__); | |
b677c210 | 143 | } |
583fa586 | 144 | |
25e41b3d | 145 | xfs_ioend_wake(ip); |
0829c360 CH |
146 | mempool_free(ioend, xfs_ioend_pool); |
147 | } | |
148 | ||
932640e8 DC |
149 | /* |
150 | * If the end of the current ioend is beyond the current EOF, | |
151 | * return the new EOF value, otherwise zero. | |
152 | */ | |
153 | STATIC xfs_fsize_t | |
154 | xfs_ioend_new_eof( | |
155 | xfs_ioend_t *ioend) | |
156 | { | |
157 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
158 | xfs_fsize_t isize; | |
159 | xfs_fsize_t bsize; | |
160 | ||
161 | bsize = ioend->io_offset + ioend->io_size; | |
162 | isize = MAX(ip->i_size, ip->i_new_size); | |
163 | isize = MIN(isize, bsize); | |
164 | return isize > ip->i_d.di_size ? isize : 0; | |
165 | } | |
166 | ||
ba87ea69 | 167 | /* |
77d7a0c2 DC |
168 | * Update on-disk file size now that data has been written to disk. The |
169 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
170 | * will be the intended file size until i_size is updated. If this write does | |
171 | * not extend all the way to the valid file size then restrict this update to | |
172 | * the end of the write. | |
173 | * | |
174 | * This function does not block as blocking on the inode lock in IO completion | |
175 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
176 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 177 | */ |
77d7a0c2 | 178 | STATIC int |
ba87ea69 LM |
179 | xfs_setfilesize( |
180 | xfs_ioend_t *ioend) | |
181 | { | |
b677c210 | 182 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 183 | xfs_fsize_t isize; |
ba87ea69 | 184 | |
ba87ea69 | 185 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
34a52c6c | 186 | ASSERT(ioend->io_type != IO_READ); |
ba87ea69 LM |
187 | |
188 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
189 | return 0; |
190 | ||
191 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
192 | return EAGAIN; | |
ba87ea69 | 193 | |
932640e8 DC |
194 | isize = xfs_ioend_new_eof(ioend); |
195 | if (isize) { | |
ba87ea69 | 196 | ip->i_d.di_size = isize; |
66d834ea | 197 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
198 | } |
199 | ||
200 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
201 | return 0; |
202 | } | |
203 | ||
204 | /* | |
205 | * Schedule IO completion handling on a xfsdatad if this was | |
206 | * the final hold on this ioend. If we are asked to wait, | |
207 | * flush the workqueue. | |
208 | */ | |
209 | STATIC void | |
210 | xfs_finish_ioend( | |
211 | xfs_ioend_t *ioend, | |
212 | int wait) | |
213 | { | |
214 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
215 | struct workqueue_struct *wq; | |
216 | ||
34a52c6c | 217 | wq = (ioend->io_type == IO_UNWRITTEN) ? |
77d7a0c2 DC |
218 | xfsconvertd_workqueue : xfsdatad_workqueue; |
219 | queue_work(wq, &ioend->io_work); | |
220 | if (wait) | |
221 | flush_workqueue(wq); | |
222 | } | |
ba87ea69 LM |
223 | } |
224 | ||
0829c360 | 225 | /* |
5ec4fabb | 226 | * IO write completion. |
f6d6d4fc CH |
227 | */ |
228 | STATIC void | |
5ec4fabb | 229 | xfs_end_io( |
77d7a0c2 | 230 | struct work_struct *work) |
0829c360 | 231 | { |
77d7a0c2 DC |
232 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
233 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 234 | int error = 0; |
ba87ea69 | 235 | |
5ec4fabb CH |
236 | /* |
237 | * For unwritten extents we need to issue transactions to convert a | |
238 | * range to normal written extens after the data I/O has finished. | |
239 | */ | |
34a52c6c | 240 | if (ioend->io_type == IO_UNWRITTEN && |
5ec4fabb | 241 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { |
5ec4fabb CH |
242 | |
243 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
244 | ioend->io_size); | |
245 | if (error) | |
246 | ioend->io_error = error; | |
247 | } | |
ba87ea69 | 248 | |
5ec4fabb CH |
249 | /* |
250 | * We might have to update the on-disk file size after extending | |
251 | * writes. | |
252 | */ | |
34a52c6c | 253 | if (ioend->io_type != IO_READ) { |
77d7a0c2 DC |
254 | error = xfs_setfilesize(ioend); |
255 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 256 | } |
77d7a0c2 DC |
257 | |
258 | /* | |
259 | * If we didn't complete processing of the ioend, requeue it to the | |
260 | * tail of the workqueue for another attempt later. Otherwise destroy | |
261 | * it. | |
262 | */ | |
263 | if (error == EAGAIN) { | |
264 | atomic_inc(&ioend->io_remaining); | |
265 | xfs_finish_ioend(ioend, 0); | |
266 | /* ensure we don't spin on blocked ioends */ | |
267 | delay(1); | |
268 | } else | |
269 | xfs_destroy_ioend(ioend); | |
c626d174 DC |
270 | } |
271 | ||
0829c360 CH |
272 | /* |
273 | * Allocate and initialise an IO completion structure. | |
274 | * We need to track unwritten extent write completion here initially. | |
275 | * We'll need to extend this for updating the ondisk inode size later | |
276 | * (vs. incore size). | |
277 | */ | |
278 | STATIC xfs_ioend_t * | |
279 | xfs_alloc_ioend( | |
f6d6d4fc CH |
280 | struct inode *inode, |
281 | unsigned int type) | |
0829c360 CH |
282 | { |
283 | xfs_ioend_t *ioend; | |
284 | ||
285 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
286 | ||
287 | /* | |
288 | * Set the count to 1 initially, which will prevent an I/O | |
289 | * completion callback from happening before we have started | |
290 | * all the I/O from calling the completion routine too early. | |
291 | */ | |
292 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 293 | ioend->io_error = 0; |
f6d6d4fc CH |
294 | ioend->io_list = NULL; |
295 | ioend->io_type = type; | |
b677c210 | 296 | ioend->io_inode = inode; |
c1a073bd | 297 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 298 | ioend->io_buffer_tail = NULL; |
b677c210 | 299 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
300 | ioend->io_offset = 0; |
301 | ioend->io_size = 0; | |
302 | ||
5ec4fabb | 303 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
304 | return ioend; |
305 | } | |
306 | ||
1da177e4 LT |
307 | STATIC int |
308 | xfs_map_blocks( | |
309 | struct inode *inode, | |
310 | loff_t offset, | |
311 | ssize_t count, | |
207d0416 | 312 | struct xfs_bmbt_irec *imap, |
1da177e4 LT |
313 | int flags) |
314 | { | |
6bd16ff2 | 315 | int nmaps = 1; |
207d0416 | 316 | int new = 0; |
6bd16ff2 | 317 | |
207d0416 | 318 | return -xfs_iomap(XFS_I(inode), offset, count, flags, imap, &nmaps, &new); |
1da177e4 LT |
319 | } |
320 | ||
b8f82a4a | 321 | STATIC int |
558e6891 | 322 | xfs_imap_valid( |
8699bb0a | 323 | struct inode *inode, |
207d0416 | 324 | struct xfs_bmbt_irec *imap, |
558e6891 | 325 | xfs_off_t offset) |
1da177e4 | 326 | { |
558e6891 | 327 | offset >>= inode->i_blkbits; |
8699bb0a | 328 | |
558e6891 CH |
329 | return offset >= imap->br_startoff && |
330 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
331 | } |
332 | ||
f6d6d4fc CH |
333 | /* |
334 | * BIO completion handler for buffered IO. | |
335 | */ | |
782e3b3b | 336 | STATIC void |
f6d6d4fc CH |
337 | xfs_end_bio( |
338 | struct bio *bio, | |
f6d6d4fc CH |
339 | int error) |
340 | { | |
341 | xfs_ioend_t *ioend = bio->bi_private; | |
342 | ||
f6d6d4fc | 343 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 344 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
345 | |
346 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
347 | bio->bi_private = NULL; |
348 | bio->bi_end_io = NULL; | |
f6d6d4fc | 349 | bio_put(bio); |
7d04a335 | 350 | |
e927af90 | 351 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
352 | } |
353 | ||
354 | STATIC void | |
355 | xfs_submit_ioend_bio( | |
06342cf8 CH |
356 | struct writeback_control *wbc, |
357 | xfs_ioend_t *ioend, | |
358 | struct bio *bio) | |
f6d6d4fc CH |
359 | { |
360 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
361 | bio->bi_private = ioend; |
362 | bio->bi_end_io = xfs_end_bio; | |
363 | ||
932640e8 DC |
364 | /* |
365 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
366 | * but don't update the inode size until I/O completion. | |
367 | */ | |
368 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 369 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 370 | |
06342cf8 CH |
371 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
372 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
373 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
374 | bio_put(bio); | |
375 | } | |
376 | ||
377 | STATIC struct bio * | |
378 | xfs_alloc_ioend_bio( | |
379 | struct buffer_head *bh) | |
380 | { | |
381 | struct bio *bio; | |
382 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
383 | ||
384 | do { | |
385 | bio = bio_alloc(GFP_NOIO, nvecs); | |
386 | nvecs >>= 1; | |
387 | } while (!bio); | |
388 | ||
389 | ASSERT(bio->bi_private == NULL); | |
390 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
391 | bio->bi_bdev = bh->b_bdev; | |
392 | bio_get(bio); | |
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, | |
f6d6d4fc CH |
413 | int clear_dirty, |
414 | int buffers) | |
415 | { | |
416 | ASSERT(PageLocked(page)); | |
417 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 418 | if (clear_dirty) |
92132021 DC |
419 | clear_page_dirty_for_io(page); |
420 | set_page_writeback(page); | |
f6d6d4fc | 421 | unlock_page(page); |
1f7decf6 FW |
422 | /* If no buffers on the page are to be written, finish it here */ |
423 | if (!buffers) | |
f6d6d4fc | 424 | end_page_writeback(page); |
f6d6d4fc CH |
425 | } |
426 | ||
427 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
428 | { | |
429 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
430 | } | |
431 | ||
432 | /* | |
d88992f6 DC |
433 | * Submit all of the bios for all of the ioends we have saved up, covering the |
434 | * initial writepage page and also any probed pages. | |
435 | * | |
436 | * Because we may have multiple ioends spanning a page, we need to start | |
437 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
438 | * buffers as we got, then we can end up with a page that only has buffers | |
439 | * marked async write and I/O complete on can occur before we mark the other | |
440 | * buffers async write. | |
441 | * | |
442 | * The end result of this is that we trip a bug in end_page_writeback() because | |
443 | * we call it twice for the one page as the code in end_buffer_async_write() | |
444 | * assumes that all buffers on the page are started at the same time. | |
445 | * | |
446 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 447 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
448 | */ |
449 | STATIC void | |
450 | xfs_submit_ioend( | |
06342cf8 | 451 | struct writeback_control *wbc, |
f6d6d4fc CH |
452 | xfs_ioend_t *ioend) |
453 | { | |
d88992f6 | 454 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
455 | xfs_ioend_t *next; |
456 | struct buffer_head *bh; | |
457 | struct bio *bio; | |
458 | sector_t lastblock = 0; | |
459 | ||
d88992f6 DC |
460 | /* Pass 1 - start writeback */ |
461 | do { | |
462 | next = ioend->io_list; | |
463 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
464 | xfs_start_buffer_writeback(bh); | |
465 | } | |
466 | } while ((ioend = next) != NULL); | |
467 | ||
468 | /* Pass 2 - submit I/O */ | |
469 | ioend = head; | |
f6d6d4fc CH |
470 | do { |
471 | next = ioend->io_list; | |
472 | bio = NULL; | |
473 | ||
474 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
475 | |
476 | if (!bio) { | |
477 | retry: | |
478 | bio = xfs_alloc_ioend_bio(bh); | |
479 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 480 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
481 | goto retry; |
482 | } | |
483 | ||
484 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 485 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
486 | goto retry; |
487 | } | |
488 | ||
489 | lastblock = bh->b_blocknr; | |
490 | } | |
491 | if (bio) | |
06342cf8 | 492 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 493 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
494 | } while ((ioend = next) != NULL); |
495 | } | |
496 | ||
497 | /* | |
498 | * Cancel submission of all buffer_heads so far in this endio. | |
499 | * Toss the endio too. Only ever called for the initial page | |
500 | * in a writepage request, so only ever one page. | |
501 | */ | |
502 | STATIC void | |
503 | xfs_cancel_ioend( | |
504 | xfs_ioend_t *ioend) | |
505 | { | |
506 | xfs_ioend_t *next; | |
507 | struct buffer_head *bh, *next_bh; | |
508 | ||
509 | do { | |
510 | next = ioend->io_list; | |
511 | bh = ioend->io_buffer_head; | |
512 | do { | |
513 | next_bh = bh->b_private; | |
514 | clear_buffer_async_write(bh); | |
515 | unlock_buffer(bh); | |
516 | } while ((bh = next_bh) != NULL); | |
517 | ||
25e41b3d | 518 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
519 | mempool_free(ioend, xfs_ioend_pool); |
520 | } while ((ioend = next) != NULL); | |
521 | } | |
522 | ||
523 | /* | |
524 | * Test to see if we've been building up a completion structure for | |
525 | * earlier buffers -- if so, we try to append to this ioend if we | |
526 | * can, otherwise we finish off any current ioend and start another. | |
527 | * Return true if we've finished the given ioend. | |
528 | */ | |
529 | STATIC void | |
530 | xfs_add_to_ioend( | |
531 | struct inode *inode, | |
532 | struct buffer_head *bh, | |
7336cea8 | 533 | xfs_off_t offset, |
f6d6d4fc CH |
534 | unsigned int type, |
535 | xfs_ioend_t **result, | |
536 | int need_ioend) | |
537 | { | |
538 | xfs_ioend_t *ioend = *result; | |
539 | ||
540 | if (!ioend || need_ioend || type != ioend->io_type) { | |
541 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 542 | |
f6d6d4fc CH |
543 | ioend = xfs_alloc_ioend(inode, type); |
544 | ioend->io_offset = offset; | |
545 | ioend->io_buffer_head = bh; | |
546 | ioend->io_buffer_tail = bh; | |
547 | if (previous) | |
548 | previous->io_list = ioend; | |
549 | *result = ioend; | |
550 | } else { | |
551 | ioend->io_buffer_tail->b_private = bh; | |
552 | ioend->io_buffer_tail = bh; | |
553 | } | |
554 | ||
555 | bh->b_private = NULL; | |
556 | ioend->io_size += bh->b_size; | |
557 | } | |
558 | ||
87cbc49c NS |
559 | STATIC void |
560 | xfs_map_buffer( | |
046f1685 | 561 | struct inode *inode, |
87cbc49c | 562 | struct buffer_head *bh, |
207d0416 | 563 | struct xfs_bmbt_irec *imap, |
046f1685 | 564 | xfs_off_t offset) |
87cbc49c NS |
565 | { |
566 | sector_t bn; | |
8699bb0a | 567 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
568 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
569 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 570 | |
207d0416 CH |
571 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
572 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 573 | |
e513182d | 574 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 575 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 576 | |
046f1685 | 577 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
578 | |
579 | bh->b_blocknr = bn; | |
580 | set_buffer_mapped(bh); | |
581 | } | |
582 | ||
1da177e4 LT |
583 | STATIC void |
584 | xfs_map_at_offset( | |
046f1685 | 585 | struct inode *inode, |
1da177e4 | 586 | struct buffer_head *bh, |
207d0416 | 587 | struct xfs_bmbt_irec *imap, |
046f1685 | 588 | xfs_off_t offset) |
1da177e4 | 589 | { |
207d0416 CH |
590 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
591 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 LT |
592 | |
593 | lock_buffer(bh); | |
207d0416 | 594 | xfs_map_buffer(inode, bh, imap, offset); |
046f1685 | 595 | bh->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 LT |
596 | set_buffer_mapped(bh); |
597 | clear_buffer_delay(bh); | |
f6d6d4fc | 598 | clear_buffer_unwritten(bh); |
1da177e4 LT |
599 | } |
600 | ||
601 | /* | |
6c4fe19f | 602 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
603 | */ |
604 | STATIC unsigned int | |
6c4fe19f | 605 | xfs_probe_page( |
10ce4444 | 606 | struct page *page, |
20cb52eb | 607 | unsigned int pg_offset) |
1da177e4 | 608 | { |
20cb52eb | 609 | struct buffer_head *bh, *head; |
1da177e4 LT |
610 | int ret = 0; |
611 | ||
1da177e4 | 612 | if (PageWriteback(page)) |
10ce4444 | 613 | return 0; |
20cb52eb CH |
614 | if (!PageDirty(page)) |
615 | return 0; | |
616 | if (!page->mapping) | |
617 | return 0; | |
618 | if (!page_has_buffers(page)) | |
619 | return 0; | |
1da177e4 | 620 | |
20cb52eb CH |
621 | bh = head = page_buffers(page); |
622 | do { | |
623 | if (!buffer_uptodate(bh)) | |
624 | break; | |
625 | if (!buffer_mapped(bh)) | |
626 | break; | |
627 | ret += bh->b_size; | |
628 | if (ret >= pg_offset) | |
629 | break; | |
630 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 631 | |
1da177e4 LT |
632 | return ret; |
633 | } | |
634 | ||
f6d6d4fc | 635 | STATIC size_t |
6c4fe19f | 636 | xfs_probe_cluster( |
1da177e4 LT |
637 | struct inode *inode, |
638 | struct page *startpage, | |
639 | struct buffer_head *bh, | |
20cb52eb | 640 | struct buffer_head *head) |
1da177e4 | 641 | { |
10ce4444 | 642 | struct pagevec pvec; |
1da177e4 | 643 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
644 | size_t total = 0; |
645 | int done = 0, i; | |
1da177e4 LT |
646 | |
647 | /* First sum forwards in this page */ | |
648 | do { | |
20cb52eb | 649 | if (!buffer_uptodate(bh) || !buffer_mapped(bh)) |
10ce4444 | 650 | return total; |
1da177e4 LT |
651 | total += bh->b_size; |
652 | } while ((bh = bh->b_this_page) != head); | |
653 | ||
10ce4444 CH |
654 | /* if we reached the end of the page, sum forwards in following pages */ |
655 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
656 | tindex = startpage->index + 1; | |
657 | ||
658 | /* Prune this back to avoid pathological behavior */ | |
659 | tloff = min(tlast, startpage->index + 64); | |
660 | ||
661 | pagevec_init(&pvec, 0); | |
662 | while (!done && tindex <= tloff) { | |
663 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
664 | ||
665 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
666 | break; | |
667 | ||
668 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
669 | struct page *page = pvec.pages[i]; | |
265c1fac | 670 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
671 | |
672 | if (tindex == tlast) { | |
673 | pg_offset = | |
674 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
675 | if (!pg_offset) { |
676 | done = 1; | |
10ce4444 | 677 | break; |
1defeac9 | 678 | } |
10ce4444 CH |
679 | } else |
680 | pg_offset = PAGE_CACHE_SIZE; | |
681 | ||
529ae9aa | 682 | if (page->index == tindex && trylock_page(page)) { |
20cb52eb | 683 | pg_len = xfs_probe_page(page, pg_offset); |
10ce4444 CH |
684 | unlock_page(page); |
685 | } | |
686 | ||
265c1fac | 687 | if (!pg_len) { |
10ce4444 CH |
688 | done = 1; |
689 | break; | |
690 | } | |
691 | ||
265c1fac | 692 | total += pg_len; |
1defeac9 | 693 | tindex++; |
1da177e4 | 694 | } |
10ce4444 CH |
695 | |
696 | pagevec_release(&pvec); | |
697 | cond_resched(); | |
1da177e4 | 698 | } |
10ce4444 | 699 | |
1da177e4 LT |
700 | return total; |
701 | } | |
702 | ||
703 | /* | |
10ce4444 CH |
704 | * Test if a given page is suitable for writing as part of an unwritten |
705 | * or delayed allocate extent. | |
1da177e4 | 706 | */ |
10ce4444 CH |
707 | STATIC int |
708 | xfs_is_delayed_page( | |
709 | struct page *page, | |
f6d6d4fc | 710 | unsigned int type) |
1da177e4 | 711 | { |
1da177e4 | 712 | if (PageWriteback(page)) |
10ce4444 | 713 | return 0; |
1da177e4 LT |
714 | |
715 | if (page->mapping && page_has_buffers(page)) { | |
716 | struct buffer_head *bh, *head; | |
717 | int acceptable = 0; | |
718 | ||
719 | bh = head = page_buffers(page); | |
720 | do { | |
f6d6d4fc | 721 | if (buffer_unwritten(bh)) |
34a52c6c | 722 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 723 | else if (buffer_delay(bh)) |
34a52c6c | 724 | acceptable = (type == IO_DELAY); |
2ddee844 | 725 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
34a52c6c | 726 | acceptable = (type == IO_NEW); |
f6d6d4fc | 727 | else |
1da177e4 | 728 | break; |
1da177e4 LT |
729 | } while ((bh = bh->b_this_page) != head); |
730 | ||
731 | if (acceptable) | |
10ce4444 | 732 | return 1; |
1da177e4 LT |
733 | } |
734 | ||
10ce4444 | 735 | return 0; |
1da177e4 LT |
736 | } |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Allocate & map buffers for page given the extent map. Write it out. | |
740 | * except for the original page of a writepage, this is called on | |
741 | * delalloc/unwritten pages only, for the original page it is possible | |
742 | * that the page has no mapping at all. | |
743 | */ | |
f6d6d4fc | 744 | STATIC int |
1da177e4 LT |
745 | xfs_convert_page( |
746 | struct inode *inode, | |
747 | struct page *page, | |
10ce4444 | 748 | loff_t tindex, |
207d0416 | 749 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 750 | xfs_ioend_t **ioendp, |
1da177e4 | 751 | struct writeback_control *wbc, |
1da177e4 LT |
752 | int all_bh) |
753 | { | |
f6d6d4fc | 754 | struct buffer_head *bh, *head; |
9260dc6b CH |
755 | xfs_off_t end_offset; |
756 | unsigned long p_offset; | |
f6d6d4fc | 757 | unsigned int type; |
24e17b5f | 758 | int len, page_dirty; |
f6d6d4fc | 759 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 760 | xfs_off_t offset = page_offset(page); |
1da177e4 | 761 | |
10ce4444 CH |
762 | if (page->index != tindex) |
763 | goto fail; | |
529ae9aa | 764 | if (!trylock_page(page)) |
10ce4444 CH |
765 | goto fail; |
766 | if (PageWriteback(page)) | |
767 | goto fail_unlock_page; | |
768 | if (page->mapping != inode->i_mapping) | |
769 | goto fail_unlock_page; | |
770 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
771 | goto fail_unlock_page; | |
772 | ||
24e17b5f NS |
773 | /* |
774 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 775 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
776 | * |
777 | * Derivation: | |
778 | * | |
779 | * End offset is the highest offset that this page should represent. | |
780 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
781 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
782 | * hence give us the correct page_dirty count. On any other page, | |
783 | * it will be zero and in that case we need page_dirty to be the | |
784 | * count of buffers on the page. | |
24e17b5f | 785 | */ |
9260dc6b CH |
786 | end_offset = min_t(unsigned long long, |
787 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
788 | i_size_read(inode)); | |
789 | ||
24e17b5f | 790 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
791 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
792 | PAGE_CACHE_SIZE); | |
793 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
794 | page_dirty = p_offset / len; | |
24e17b5f | 795 | |
1da177e4 LT |
796 | bh = head = page_buffers(page); |
797 | do { | |
9260dc6b | 798 | if (offset >= end_offset) |
1da177e4 | 799 | break; |
f6d6d4fc CH |
800 | if (!buffer_uptodate(bh)) |
801 | uptodate = 0; | |
802 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
803 | done = 1; | |
1da177e4 | 804 | continue; |
f6d6d4fc CH |
805 | } |
806 | ||
9260dc6b CH |
807 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
808 | if (buffer_unwritten(bh)) | |
34a52c6c | 809 | type = IO_UNWRITTEN; |
9260dc6b | 810 | else |
34a52c6c | 811 | type = IO_DELAY; |
9260dc6b | 812 | |
558e6891 | 813 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 814 | done = 1; |
9260dc6b CH |
815 | continue; |
816 | } | |
817 | ||
207d0416 CH |
818 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
819 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
9260dc6b | 820 | |
207d0416 | 821 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
822 | xfs_add_to_ioend(inode, bh, offset, type, |
823 | ioendp, done); | |
824 | ||
9260dc6b CH |
825 | page_dirty--; |
826 | count++; | |
827 | } else { | |
34a52c6c | 828 | type = IO_NEW; |
89f3b363 | 829 | if (buffer_mapped(bh) && all_bh) { |
1da177e4 | 830 | lock_buffer(bh); |
7336cea8 | 831 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
832 | type, ioendp, done); |
833 | count++; | |
24e17b5f | 834 | page_dirty--; |
9260dc6b CH |
835 | } else { |
836 | done = 1; | |
1da177e4 | 837 | } |
1da177e4 | 838 | } |
7336cea8 | 839 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 840 | |
f6d6d4fc CH |
841 | if (uptodate && bh == head) |
842 | SetPageUptodate(page); | |
843 | ||
89f3b363 CH |
844 | if (count) { |
845 | wbc->nr_to_write--; | |
846 | if (wbc->nr_to_write <= 0) | |
847 | done = 1; | |
1da177e4 | 848 | } |
89f3b363 | 849 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
850 | |
851 | return done; | |
10ce4444 CH |
852 | fail_unlock_page: |
853 | unlock_page(page); | |
854 | fail: | |
855 | return 1; | |
1da177e4 LT |
856 | } |
857 | ||
858 | /* | |
859 | * Convert & write out a cluster of pages in the same extent as defined | |
860 | * by mp and following the start page. | |
861 | */ | |
862 | STATIC void | |
863 | xfs_cluster_write( | |
864 | struct inode *inode, | |
865 | pgoff_t tindex, | |
207d0416 | 866 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 867 | xfs_ioend_t **ioendp, |
1da177e4 | 868 | struct writeback_control *wbc, |
1da177e4 LT |
869 | int all_bh, |
870 | pgoff_t tlast) | |
871 | { | |
10ce4444 CH |
872 | struct pagevec pvec; |
873 | int done = 0, i; | |
1da177e4 | 874 | |
10ce4444 CH |
875 | pagevec_init(&pvec, 0); |
876 | while (!done && tindex <= tlast) { | |
877 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
878 | ||
879 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 880 | break; |
10ce4444 CH |
881 | |
882 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
883 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
89f3b363 | 884 | imap, ioendp, wbc, all_bh); |
10ce4444 CH |
885 | if (done) |
886 | break; | |
887 | } | |
888 | ||
889 | pagevec_release(&pvec); | |
890 | cond_resched(); | |
1da177e4 LT |
891 | } |
892 | } | |
893 | ||
3ed3a434 DC |
894 | STATIC void |
895 | xfs_vm_invalidatepage( | |
896 | struct page *page, | |
897 | unsigned long offset) | |
898 | { | |
899 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
900 | block_invalidatepage(page, offset); | |
901 | } | |
902 | ||
903 | /* | |
904 | * If the page has delalloc buffers on it, we need to punch them out before we | |
905 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
906 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
907 | * is done on that same region - the delalloc extent is returned when none is | |
908 | * supposed to be there. | |
909 | * | |
910 | * We prevent this by truncating away the delalloc regions on the page before | |
911 | * invalidating it. Because they are delalloc, we can do this without needing a | |
912 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
913 | * truncation without a transaction as there is no space left for block | |
914 | * reservation (typically why we see a ENOSPC in writeback). | |
915 | * | |
916 | * This is not a performance critical path, so for now just do the punching a | |
917 | * buffer head at a time. | |
918 | */ | |
919 | STATIC void | |
920 | xfs_aops_discard_page( | |
921 | struct page *page) | |
922 | { | |
923 | struct inode *inode = page->mapping->host; | |
924 | struct xfs_inode *ip = XFS_I(inode); | |
925 | struct buffer_head *bh, *head; | |
926 | loff_t offset = page_offset(page); | |
927 | ssize_t len = 1 << inode->i_blkbits; | |
928 | ||
34a52c6c | 929 | if (!xfs_is_delayed_page(page, IO_DELAY)) |
3ed3a434 DC |
930 | goto out_invalidate; |
931 | ||
e8c3753c DC |
932 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
933 | goto out_invalidate; | |
934 | ||
3ed3a434 DC |
935 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
936 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
937 | page, ip->i_ino, offset); | |
938 | ||
939 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
940 | bh = head = page_buffers(page); | |
941 | do { | |
942 | int done; | |
943 | xfs_fileoff_t offset_fsb; | |
944 | xfs_bmbt_irec_t imap; | |
945 | int nimaps = 1; | |
946 | int error; | |
947 | xfs_fsblock_t firstblock; | |
948 | xfs_bmap_free_t flist; | |
949 | ||
950 | if (!buffer_delay(bh)) | |
951 | goto next_buffer; | |
952 | ||
953 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
954 | ||
955 | /* | |
956 | * Map the range first and check that it is a delalloc extent | |
957 | * before trying to unmap the range. Otherwise we will be | |
958 | * trying to remove a real extent (which requires a | |
959 | * transaction) or a hole, which is probably a bad idea... | |
960 | */ | |
961 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
962 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
b4e9181e | 963 | &nimaps, NULL); |
3ed3a434 DC |
964 | |
965 | if (error) { | |
966 | /* something screwed, just bail */ | |
e8c3753c DC |
967 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
968 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
969 | "page discard failed delalloc mapping lookup."); | |
970 | } | |
3ed3a434 DC |
971 | break; |
972 | } | |
973 | if (!nimaps) { | |
974 | /* nothing there */ | |
975 | goto next_buffer; | |
976 | } | |
977 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
978 | /* been converted, ignore */ | |
979 | goto next_buffer; | |
980 | } | |
981 | WARN_ON(imap.br_blockcount == 0); | |
982 | ||
983 | /* | |
984 | * Note: while we initialise the firstblock/flist pair, they | |
985 | * should never be used because blocks should never be | |
986 | * allocated or freed for a delalloc extent and hence we need | |
987 | * don't cancel or finish them after the xfs_bunmapi() call. | |
988 | */ | |
989 | xfs_bmap_init(&flist, &firstblock); | |
990 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
b4e9181e | 991 | &flist, &done); |
3ed3a434 DC |
992 | |
993 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
994 | if (error) { | |
995 | /* something screwed, just bail */ | |
e8c3753c DC |
996 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
997 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 998 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 999 | } |
3ed3a434 DC |
1000 | break; |
1001 | } | |
1002 | next_buffer: | |
1003 | offset += len; | |
1004 | ||
1005 | } while ((bh = bh->b_this_page) != head); | |
1006 | ||
1007 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1008 | out_invalidate: | |
1009 | xfs_vm_invalidatepage(page, 0); | |
1010 | return; | |
1011 | } | |
1012 | ||
1da177e4 | 1013 | /* |
89f3b363 CH |
1014 | * Write out a dirty page. |
1015 | * | |
1016 | * For delalloc space on the page we need to allocate space and flush it. | |
1017 | * For unwritten space on the page we need to start the conversion to | |
1018 | * regular allocated space. | |
89f3b363 | 1019 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1020 | * |
89f3b363 CH |
1021 | * If we detect that a transaction would be required to flush the page, we |
1022 | * have to check the process flags first, if we are already in a transaction | |
1023 | * or disk I/O during allocations is off, we need to fail the writepage and | |
1024 | * redirty the page. | |
1da177e4 | 1025 | */ |
1da177e4 | 1026 | STATIC int |
89f3b363 CH |
1027 | xfs_vm_writepage( |
1028 | struct page *page, | |
1029 | struct writeback_control *wbc) | |
1da177e4 | 1030 | { |
89f3b363 | 1031 | struct inode *inode = page->mapping->host; |
20cb52eb | 1032 | int delalloc, unwritten; |
f6d6d4fc | 1033 | struct buffer_head *bh, *head; |
207d0416 | 1034 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 1035 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 1036 | loff_t offset; |
f6d6d4fc | 1037 | unsigned int type; |
1da177e4 | 1038 | __uint64_t end_offset; |
bd1556a1 | 1039 | pgoff_t end_index, last_index; |
d5cb48aa | 1040 | ssize_t size, len; |
558e6891 | 1041 | int flags, err, imap_valid = 0, uptodate = 1; |
89f3b363 | 1042 | int count = 0; |
20cb52eb | 1043 | int all_bh = 0; |
89f3b363 CH |
1044 | |
1045 | trace_xfs_writepage(inode, page, 0); | |
1046 | ||
20cb52eb CH |
1047 | ASSERT(page_has_buffers(page)); |
1048 | ||
89f3b363 CH |
1049 | /* |
1050 | * Refuse to write the page out if we are called from reclaim context. | |
1051 | * | |
d4f7a5cb CH |
1052 | * This avoids stack overflows when called from deeply used stacks in |
1053 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1054 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 CH |
1055 | * |
1056 | * This should really be done by the core VM, but until that happens | |
1057 | * filesystems like XFS, btrfs and ext4 have to take care of this | |
1058 | * by themselves. | |
1059 | */ | |
d4f7a5cb | 1060 | if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC) |
89f3b363 | 1061 | goto out_fail; |
1da177e4 | 1062 | |
89f3b363 | 1063 | /* |
20cb52eb CH |
1064 | * We need a transaction if there are delalloc or unwritten buffers |
1065 | * on the page. | |
1066 | * | |
1067 | * If we need a transaction and the process flags say we are already | |
1068 | * in a transaction, or no IO is allowed then mark the page dirty | |
1069 | * again and leave the page as is. | |
89f3b363 | 1070 | */ |
20cb52eb CH |
1071 | xfs_count_page_state(page, &delalloc, &unwritten); |
1072 | if ((current->flags & PF_FSTRANS) && (delalloc || unwritten)) | |
89f3b363 CH |
1073 | goto out_fail; |
1074 | ||
1da177e4 LT |
1075 | /* Is this page beyond the end of the file? */ |
1076 | offset = i_size_read(inode); | |
1077 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1078 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1079 | if (page->index >= end_index) { | |
1080 | if ((page->index >= end_index + 1) || | |
1081 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 1082 | unlock_page(page); |
19d5bcf3 | 1083 | return 0; |
1da177e4 LT |
1084 | } |
1085 | } | |
1086 | ||
f6d6d4fc | 1087 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
1088 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
1089 | offset); | |
24e17b5f | 1090 | len = 1 << inode->i_blkbits; |
24e17b5f | 1091 | |
24e17b5f | 1092 | bh = head = page_buffers(page); |
f6d6d4fc | 1093 | offset = page_offset(page); |
df3c7244 | 1094 | flags = BMAPI_READ; |
34a52c6c | 1095 | type = IO_NEW; |
f6d6d4fc | 1096 | |
1da177e4 LT |
1097 | do { |
1098 | if (offset >= end_offset) | |
1099 | break; | |
1100 | if (!buffer_uptodate(bh)) | |
1101 | uptodate = 0; | |
1da177e4 | 1102 | |
3d9b02e3 ES |
1103 | /* |
1104 | * A hole may still be marked uptodate because discard_buffer | |
1105 | * leaves the flag set. | |
1106 | */ | |
1107 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
1108 | ASSERT(!buffer_dirty(bh)); | |
1109 | imap_valid = 0; | |
1110 | continue; | |
1111 | } | |
1112 | ||
558e6891 CH |
1113 | if (imap_valid) |
1114 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1da177e4 | 1115 | |
20cb52eb | 1116 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
effd120e DC |
1117 | int new_ioend = 0; |
1118 | ||
df3c7244 | 1119 | /* |
6c4fe19f CH |
1120 | * Make sure we don't use a read-only iomap |
1121 | */ | |
df3c7244 | 1122 | if (flags == BMAPI_READ) |
558e6891 | 1123 | imap_valid = 0; |
6c4fe19f | 1124 | |
f6d6d4fc | 1125 | if (buffer_unwritten(bh)) { |
34a52c6c | 1126 | type = IO_UNWRITTEN; |
8272145c | 1127 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1128 | } else if (buffer_delay(bh)) { |
34a52c6c | 1129 | type = IO_DELAY; |
89f3b363 CH |
1130 | flags = BMAPI_ALLOCATE; |
1131 | ||
1132 | if (wbc->sync_mode == WB_SYNC_NONE && | |
1133 | wbc->nonblocking) | |
1134 | flags |= BMAPI_TRYLOCK; | |
f6d6d4fc CH |
1135 | } |
1136 | ||
558e6891 | 1137 | if (!imap_valid) { |
effd120e | 1138 | /* |
20cb52eb | 1139 | * If we didn't have a valid mapping then we |
effd120e DC |
1140 | * need to ensure that we put the new mapping |
1141 | * in a new ioend structure. This needs to be | |
1142 | * done to ensure that the ioends correctly | |
1143 | * reflect the block mappings at io completion | |
1144 | * for unwritten extent conversion. | |
1145 | */ | |
1146 | new_ioend = 1; | |
20cb52eb | 1147 | err = xfs_map_blocks(inode, offset, len, |
207d0416 | 1148 | &imap, flags); |
f6d6d4fc | 1149 | if (err) |
1da177e4 | 1150 | goto error; |
558e6891 CH |
1151 | imap_valid = xfs_imap_valid(inode, &imap, |
1152 | offset); | |
1da177e4 | 1153 | } |
558e6891 | 1154 | if (imap_valid) { |
207d0416 | 1155 | xfs_map_at_offset(inode, bh, &imap, offset); |
89f3b363 CH |
1156 | xfs_add_to_ioend(inode, bh, offset, type, |
1157 | &ioend, new_ioend); | |
f6d6d4fc | 1158 | count++; |
1da177e4 | 1159 | } |
89f3b363 | 1160 | } else if (buffer_uptodate(bh)) { |
6c4fe19f CH |
1161 | /* |
1162 | * we got here because the buffer is already mapped. | |
1163 | * That means it must already have extents allocated | |
1164 | * underneath it. Map the extent by reading it. | |
1165 | */ | |
558e6891 | 1166 | if (!imap_valid || flags != BMAPI_READ) { |
6c4fe19f | 1167 | flags = BMAPI_READ; |
20cb52eb | 1168 | size = xfs_probe_cluster(inode, page, bh, head); |
6c4fe19f | 1169 | err = xfs_map_blocks(inode, offset, size, |
207d0416 | 1170 | &imap, flags); |
6c4fe19f CH |
1171 | if (err) |
1172 | goto error; | |
558e6891 CH |
1173 | imap_valid = xfs_imap_valid(inode, &imap, |
1174 | offset); | |
6c4fe19f | 1175 | } |
d5cb48aa | 1176 | |
df3c7244 | 1177 | /* |
34a52c6c | 1178 | * We set the type to IO_NEW in case we are doing a |
df3c7244 DC |
1179 | * small write at EOF that is extending the file but |
1180 | * without needing an allocation. We need to update the | |
1181 | * file size on I/O completion in this case so it is | |
1182 | * the same case as having just allocated a new extent | |
1183 | * that we are writing into for the first time. | |
1184 | */ | |
34a52c6c | 1185 | type = IO_NEW; |
ca5de404 | 1186 | if (trylock_buffer(bh)) { |
558e6891 | 1187 | if (imap_valid) |
6c4fe19f | 1188 | all_bh = 1; |
7336cea8 | 1189 | xfs_add_to_ioend(inode, bh, offset, type, |
558e6891 | 1190 | &ioend, !imap_valid); |
d5cb48aa | 1191 | count++; |
f6d6d4fc | 1192 | } else { |
558e6891 | 1193 | imap_valid = 0; |
1da177e4 | 1194 | } |
89f3b363 | 1195 | } else if (PageUptodate(page)) { |
20cb52eb | 1196 | ASSERT(buffer_mapped(bh)); |
558e6891 | 1197 | imap_valid = 0; |
1da177e4 | 1198 | } |
f6d6d4fc CH |
1199 | |
1200 | if (!iohead) | |
1201 | iohead = ioend; | |
1202 | ||
1203 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1204 | |
1205 | if (uptodate && bh == head) | |
1206 | SetPageUptodate(page); | |
1207 | ||
89f3b363 | 1208 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1209 | |
558e6891 | 1210 | if (ioend && imap_valid) { |
bd1556a1 CH |
1211 | xfs_off_t end_index; |
1212 | ||
1213 | end_index = imap.br_startoff + imap.br_blockcount; | |
1214 | ||
1215 | /* to bytes */ | |
1216 | end_index <<= inode->i_blkbits; | |
1217 | ||
1218 | /* to pages */ | |
1219 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1220 | ||
1221 | /* check against file size */ | |
1222 | if (end_index > last_index) | |
1223 | end_index = last_index; | |
8699bb0a | 1224 | |
207d0416 | 1225 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
89f3b363 | 1226 | wbc, all_bh, end_index); |
1da177e4 LT |
1227 | } |
1228 | ||
f6d6d4fc | 1229 | if (iohead) |
06342cf8 | 1230 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1231 | |
89f3b363 | 1232 | return 0; |
1da177e4 LT |
1233 | |
1234 | error: | |
f6d6d4fc CH |
1235 | if (iohead) |
1236 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1237 | |
20cb52eb | 1238 | xfs_aops_discard_page(page); |
89f3b363 CH |
1239 | ClearPageUptodate(page); |
1240 | unlock_page(page); | |
1da177e4 | 1241 | return err; |
f51623b2 NS |
1242 | |
1243 | out_fail: | |
1244 | redirty_page_for_writepage(wbc, page); | |
1245 | unlock_page(page); | |
1246 | return 0; | |
f51623b2 NS |
1247 | } |
1248 | ||
7d4fb40a NS |
1249 | STATIC int |
1250 | xfs_vm_writepages( | |
1251 | struct address_space *mapping, | |
1252 | struct writeback_control *wbc) | |
1253 | { | |
b3aea4ed | 1254 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1255 | return generic_writepages(mapping, wbc); |
1256 | } | |
1257 | ||
f51623b2 NS |
1258 | /* |
1259 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1260 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1261 | * have buffer heads in this call. |
1262 | * | |
89f3b363 | 1263 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1264 | */ |
1265 | STATIC int | |
238f4c54 | 1266 | xfs_vm_releasepage( |
f51623b2 NS |
1267 | struct page *page, |
1268 | gfp_t gfp_mask) | |
1269 | { | |
20cb52eb | 1270 | int delalloc, unwritten; |
f51623b2 | 1271 | |
89f3b363 | 1272 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1273 | |
20cb52eb | 1274 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1275 | |
89f3b363 | 1276 | if (WARN_ON(delalloc)) |
f51623b2 | 1277 | return 0; |
89f3b363 | 1278 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1279 | return 0; |
1280 | ||
f51623b2 NS |
1281 | return try_to_free_buffers(page); |
1282 | } | |
1283 | ||
1da177e4 | 1284 | STATIC int |
c2536668 | 1285 | __xfs_get_blocks( |
1da177e4 LT |
1286 | struct inode *inode, |
1287 | sector_t iblock, | |
1da177e4 LT |
1288 | struct buffer_head *bh_result, |
1289 | int create, | |
f2bde9b8 | 1290 | int direct) |
1da177e4 | 1291 | { |
f2bde9b8 | 1292 | int flags = create ? BMAPI_WRITE : BMAPI_READ; |
207d0416 | 1293 | struct xfs_bmbt_irec imap; |
fdc7ed75 NS |
1294 | xfs_off_t offset; |
1295 | ssize_t size; | |
207d0416 CH |
1296 | int nimap = 1; |
1297 | int new = 0; | |
1da177e4 | 1298 | int error; |
1da177e4 | 1299 | |
fdc7ed75 | 1300 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1301 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1302 | size = bh_result->b_size; | |
364f358a LM |
1303 | |
1304 | if (!create && direct && offset >= i_size_read(inode)) | |
1305 | return 0; | |
1306 | ||
f2bde9b8 CH |
1307 | if (direct && create) |
1308 | flags |= BMAPI_DIRECT; | |
1309 | ||
1310 | error = xfs_iomap(XFS_I(inode), offset, size, flags, &imap, &nimap, | |
1311 | &new); | |
1da177e4 LT |
1312 | if (error) |
1313 | return -error; | |
207d0416 | 1314 | if (nimap == 0) |
1da177e4 LT |
1315 | return 0; |
1316 | ||
207d0416 CH |
1317 | if (imap.br_startblock != HOLESTARTBLOCK && |
1318 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1319 | /* |
1320 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1321 | * the read case (treat as if we're reading into a hole). |
1322 | */ | |
207d0416 CH |
1323 | if (create || !ISUNWRITTEN(&imap)) |
1324 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1325 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1326 | if (direct) |
1327 | bh_result->b_private = inode; | |
1328 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1329 | } |
1330 | } | |
1331 | ||
c2536668 NS |
1332 | /* |
1333 | * If this is a realtime file, data may be on a different device. | |
1334 | * to that pointed to from the buffer_head b_bdev currently. | |
1335 | */ | |
046f1685 | 1336 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1337 | |
c2536668 | 1338 | /* |
549054af DC |
1339 | * If we previously allocated a block out beyond eof and we are now |
1340 | * coming back to use it then we will need to flag it as new even if it | |
1341 | * has a disk address. | |
1342 | * | |
1343 | * With sub-block writes into unwritten extents we also need to mark | |
1344 | * the buffer as new so that the unwritten parts of the buffer gets | |
1345 | * correctly zeroed. | |
1da177e4 LT |
1346 | */ |
1347 | if (create && | |
1348 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1349 | (offset >= i_size_read(inode)) || |
207d0416 | 1350 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1351 | set_buffer_new(bh_result); |
1da177e4 | 1352 | |
207d0416 | 1353 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1354 | BUG_ON(direct); |
1355 | if (create) { | |
1356 | set_buffer_uptodate(bh_result); | |
1357 | set_buffer_mapped(bh_result); | |
1358 | set_buffer_delay(bh_result); | |
1359 | } | |
1360 | } | |
1361 | ||
2b8f12b7 CH |
1362 | /* |
1363 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1364 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1365 | */ | |
c2536668 | 1366 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1367 | xfs_off_t mapping_size; |
1368 | ||
1369 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1370 | mapping_size <<= inode->i_blkbits; | |
1371 | ||
1372 | ASSERT(mapping_size > 0); | |
1373 | if (mapping_size > size) | |
1374 | mapping_size = size; | |
1375 | if (mapping_size > LONG_MAX) | |
1376 | mapping_size = LONG_MAX; | |
1377 | ||
1378 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1379 | } |
1380 | ||
1381 | return 0; | |
1382 | } | |
1383 | ||
1384 | int | |
c2536668 | 1385 | xfs_get_blocks( |
1da177e4 LT |
1386 | struct inode *inode, |
1387 | sector_t iblock, | |
1388 | struct buffer_head *bh_result, | |
1389 | int create) | |
1390 | { | |
f2bde9b8 | 1391 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1392 | } |
1393 | ||
1394 | STATIC int | |
e4c573bb | 1395 | xfs_get_blocks_direct( |
1da177e4 LT |
1396 | struct inode *inode, |
1397 | sector_t iblock, | |
1da177e4 LT |
1398 | struct buffer_head *bh_result, |
1399 | int create) | |
1400 | { | |
f2bde9b8 | 1401 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1402 | } |
1403 | ||
f0973863 | 1404 | STATIC void |
e4c573bb | 1405 | xfs_end_io_direct( |
f0973863 CH |
1406 | struct kiocb *iocb, |
1407 | loff_t offset, | |
1408 | ssize_t size, | |
40e2e973 CH |
1409 | void *private, |
1410 | int ret, | |
1411 | bool is_async) | |
f0973863 CH |
1412 | { |
1413 | xfs_ioend_t *ioend = iocb->private; | |
1414 | ||
1415 | /* | |
1416 | * Non-NULL private data means we need to issue a transaction to | |
1417 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1418 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1419 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1420 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1421 | * it anyway to keep the code uniform and simpler. |
1422 | * | |
e927af90 DC |
1423 | * Well, if only it were that simple. Because synchronous direct I/O |
1424 | * requires extent conversion to occur *before* we return to userspace, | |
1425 | * we have to wait for extent conversion to complete. Look at the | |
1426 | * iocb that has been passed to us to determine if this is AIO or | |
1427 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1428 | * workqueue and wait for it to complete. | |
1429 | * | |
f0973863 CH |
1430 | * The core direct I/O code might be changed to always call the |
1431 | * completion handler in the future, in which case all this can | |
1432 | * go away. | |
1433 | */ | |
ba87ea69 LM |
1434 | ioend->io_offset = offset; |
1435 | ioend->io_size = size; | |
34a52c6c | 1436 | if (ioend->io_type == IO_READ) { |
e927af90 | 1437 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1438 | } else if (private && size > 0) { |
e927af90 | 1439 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1440 | } else { |
ba87ea69 LM |
1441 | /* |
1442 | * A direct I/O write ioend starts it's life in unwritten | |
1443 | * state in case they map an unwritten extent. This write | |
1444 | * didn't map an unwritten extent so switch it's completion | |
1445 | * handler. | |
1446 | */ | |
34a52c6c | 1447 | ioend->io_type = IO_NEW; |
e927af90 | 1448 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1449 | } |
1450 | ||
1451 | /* | |
c41564b5 | 1452 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1453 | * completion handler was called. Thus we need to protect |
1454 | * against double-freeing. | |
1455 | */ | |
1456 | iocb->private = NULL; | |
40e2e973 CH |
1457 | |
1458 | if (is_async) | |
1459 | aio_complete(iocb, ret, 0); | |
f0973863 CH |
1460 | } |
1461 | ||
1da177e4 | 1462 | STATIC ssize_t |
e4c573bb | 1463 | xfs_vm_direct_IO( |
1da177e4 LT |
1464 | int rw, |
1465 | struct kiocb *iocb, | |
1466 | const struct iovec *iov, | |
1467 | loff_t offset, | |
1468 | unsigned long nr_segs) | |
1469 | { | |
1470 | struct file *file = iocb->ki_filp; | |
1471 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1472 | struct block_device *bdev; |
f0973863 | 1473 | ssize_t ret; |
1da177e4 | 1474 | |
046f1685 | 1475 | bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1476 | |
5fe878ae | 1477 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
34a52c6c | 1478 | IO_UNWRITTEN : IO_READ); |
5fe878ae CH |
1479 | |
1480 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1481 | offset, nr_segs, | |
1482 | xfs_get_blocks_direct, | |
1483 | xfs_end_io_direct); | |
f0973863 | 1484 | |
8459d86a | 1485 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1486 | xfs_destroy_ioend(iocb->private); |
1487 | return ret; | |
1da177e4 LT |
1488 | } |
1489 | ||
f51623b2 | 1490 | STATIC int |
d79689c7 | 1491 | xfs_vm_write_begin( |
f51623b2 | 1492 | struct file *file, |
d79689c7 NP |
1493 | struct address_space *mapping, |
1494 | loff_t pos, | |
1495 | unsigned len, | |
1496 | unsigned flags, | |
1497 | struct page **pagep, | |
1498 | void **fsdata) | |
f51623b2 | 1499 | { |
d79689c7 | 1500 | *pagep = NULL; |
aea1b953 DC |
1501 | return block_write_begin(file, mapping, pos, len, flags | AOP_FLAG_NOFS, |
1502 | pagep, fsdata, xfs_get_blocks); | |
f51623b2 | 1503 | } |
1da177e4 LT |
1504 | |
1505 | STATIC sector_t | |
e4c573bb | 1506 | xfs_vm_bmap( |
1da177e4 LT |
1507 | struct address_space *mapping, |
1508 | sector_t block) | |
1509 | { | |
1510 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1511 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1512 | |
cca28fb8 | 1513 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1514 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1515 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1516 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1517 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1518 | } |
1519 | ||
1520 | STATIC int | |
e4c573bb | 1521 | xfs_vm_readpage( |
1da177e4 LT |
1522 | struct file *unused, |
1523 | struct page *page) | |
1524 | { | |
c2536668 | 1525 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1526 | } |
1527 | ||
1528 | STATIC int | |
e4c573bb | 1529 | xfs_vm_readpages( |
1da177e4 LT |
1530 | struct file *unused, |
1531 | struct address_space *mapping, | |
1532 | struct list_head *pages, | |
1533 | unsigned nr_pages) | |
1534 | { | |
c2536668 | 1535 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1536 | } |
1537 | ||
f5e54d6e | 1538 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1539 | .readpage = xfs_vm_readpage, |
1540 | .readpages = xfs_vm_readpages, | |
1541 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1542 | .writepages = xfs_vm_writepages, |
1da177e4 | 1543 | .sync_page = block_sync_page, |
238f4c54 NS |
1544 | .releasepage = xfs_vm_releasepage, |
1545 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1546 | .write_begin = xfs_vm_write_begin, |
1547 | .write_end = generic_write_end, | |
e4c573bb NS |
1548 | .bmap = xfs_vm_bmap, |
1549 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1550 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1551 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1552 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1553 | }; |