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 LT |
24 | #include "xfs_dir2.h" |
25 | #include "xfs_trans.h" | |
26 | #include "xfs_dmapi.h" | |
27 | #include "xfs_mount.h" | |
28 | #include "xfs_bmap_btree.h" | |
29 | #include "xfs_alloc_btree.h" | |
30 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 31 | #include "xfs_dir2_sf.h" |
a844f451 | 32 | #include "xfs_attr_sf.h" |
1da177e4 LT |
33 | #include "xfs_dinode.h" |
34 | #include "xfs_inode.h" | |
a844f451 NS |
35 | #include "xfs_alloc.h" |
36 | #include "xfs_btree.h" | |
1da177e4 LT |
37 | #include "xfs_error.h" |
38 | #include "xfs_rw.h" | |
39 | #include "xfs_iomap.h" | |
739bfb2a | 40 | #include "xfs_vnodeops.h" |
0b1b213f | 41 | #include "xfs_trace.h" |
1da177e4 | 42 | #include <linux/mpage.h> |
10ce4444 | 43 | #include <linux/pagevec.h> |
1da177e4 LT |
44 | #include <linux/writeback.h> |
45 | ||
25e41b3d CH |
46 | |
47 | /* | |
48 | * Prime number of hash buckets since address is used as the key. | |
49 | */ | |
50 | #define NVSYNC 37 | |
51 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
52 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
53 | ||
54 | void __init | |
55 | xfs_ioend_init(void) | |
56 | { | |
57 | int i; | |
58 | ||
59 | for (i = 0; i < NVSYNC; i++) | |
60 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
61 | } | |
62 | ||
63 | void | |
64 | xfs_ioend_wait( | |
65 | xfs_inode_t *ip) | |
66 | { | |
67 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
68 | ||
69 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
70 | } | |
71 | ||
72 | STATIC void | |
73 | xfs_ioend_wake( | |
74 | xfs_inode_t *ip) | |
75 | { | |
76 | if (atomic_dec_and_test(&ip->i_iocount)) | |
77 | wake_up(to_ioend_wq(ip)); | |
78 | } | |
79 | ||
0b1b213f | 80 | void |
f51623b2 NS |
81 | xfs_count_page_state( |
82 | struct page *page, | |
83 | int *delalloc, | |
84 | int *unmapped, | |
85 | int *unwritten) | |
86 | { | |
87 | struct buffer_head *bh, *head; | |
88 | ||
89 | *delalloc = *unmapped = *unwritten = 0; | |
90 | ||
91 | bh = head = page_buffers(page); | |
92 | do { | |
93 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
94 | (*unmapped) = 1; | |
f51623b2 NS |
95 | else if (buffer_unwritten(bh)) |
96 | (*unwritten) = 1; | |
97 | else if (buffer_delay(bh)) | |
98 | (*delalloc) = 1; | |
99 | } while ((bh = bh->b_this_page) != head); | |
100 | } | |
101 | ||
6214ed44 CH |
102 | STATIC struct block_device * |
103 | xfs_find_bdev_for_inode( | |
104 | struct xfs_inode *ip) | |
105 | { | |
106 | struct xfs_mount *mp = ip->i_mount; | |
107 | ||
71ddabb9 | 108 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
109 | return mp->m_rtdev_targp->bt_bdev; |
110 | else | |
111 | return mp->m_ddev_targp->bt_bdev; | |
112 | } | |
113 | ||
f6d6d4fc CH |
114 | /* |
115 | * We're now finished for good with this ioend structure. | |
116 | * Update the page state via the associated buffer_heads, | |
117 | * release holds on the inode and bio, and finally free | |
118 | * up memory. Do not use the ioend after this. | |
119 | */ | |
0829c360 CH |
120 | STATIC void |
121 | xfs_destroy_ioend( | |
122 | xfs_ioend_t *ioend) | |
123 | { | |
f6d6d4fc | 124 | struct buffer_head *bh, *next; |
583fa586 | 125 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
126 | |
127 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
128 | next = bh->b_private; | |
7d04a335 | 129 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 130 | } |
583fa586 CH |
131 | |
132 | /* | |
133 | * Volume managers supporting multiple paths can send back ENODEV | |
134 | * when the final path disappears. In this case continuing to fill | |
135 | * the page cache with dirty data which cannot be written out is | |
136 | * evil, so prevent that. | |
137 | */ | |
138 | if (unlikely(ioend->io_error == -ENODEV)) { | |
139 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
140 | __FILE__, __LINE__); | |
b677c210 | 141 | } |
583fa586 | 142 | |
25e41b3d | 143 | xfs_ioend_wake(ip); |
0829c360 CH |
144 | mempool_free(ioend, xfs_ioend_pool); |
145 | } | |
146 | ||
932640e8 DC |
147 | /* |
148 | * If the end of the current ioend is beyond the current EOF, | |
149 | * return the new EOF value, otherwise zero. | |
150 | */ | |
151 | STATIC xfs_fsize_t | |
152 | xfs_ioend_new_eof( | |
153 | xfs_ioend_t *ioend) | |
154 | { | |
155 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
156 | xfs_fsize_t isize; | |
157 | xfs_fsize_t bsize; | |
158 | ||
159 | bsize = ioend->io_offset + ioend->io_size; | |
160 | isize = MAX(ip->i_size, ip->i_new_size); | |
161 | isize = MIN(isize, bsize); | |
162 | return isize > ip->i_d.di_size ? isize : 0; | |
163 | } | |
164 | ||
ba87ea69 | 165 | /* |
77d7a0c2 DC |
166 | * Update on-disk file size now that data has been written to disk. The |
167 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
168 | * will be the intended file size until i_size is updated. If this write does | |
169 | * not extend all the way to the valid file size then restrict this update to | |
170 | * the end of the write. | |
171 | * | |
172 | * This function does not block as blocking on the inode lock in IO completion | |
173 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
174 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 175 | */ |
77d7a0c2 | 176 | STATIC int |
ba87ea69 LM |
177 | xfs_setfilesize( |
178 | xfs_ioend_t *ioend) | |
179 | { | |
b677c210 | 180 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 181 | xfs_fsize_t isize; |
ba87ea69 | 182 | |
ba87ea69 LM |
183 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
184 | ASSERT(ioend->io_type != IOMAP_READ); | |
185 | ||
186 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
187 | return 0; |
188 | ||
189 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
190 | return EAGAIN; | |
ba87ea69 | 191 | |
932640e8 DC |
192 | isize = xfs_ioend_new_eof(ioend); |
193 | if (isize) { | |
ba87ea69 | 194 | ip->i_d.di_size = isize; |
66d834ea | 195 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
196 | } |
197 | ||
198 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
199 | return 0; |
200 | } | |
201 | ||
202 | /* | |
203 | * Schedule IO completion handling on a xfsdatad if this was | |
204 | * the final hold on this ioend. If we are asked to wait, | |
205 | * flush the workqueue. | |
206 | */ | |
207 | STATIC void | |
208 | xfs_finish_ioend( | |
209 | xfs_ioend_t *ioend, | |
210 | int wait) | |
211 | { | |
212 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
213 | struct workqueue_struct *wq; | |
214 | ||
215 | wq = (ioend->io_type == IOMAP_UNWRITTEN) ? | |
216 | xfsconvertd_workqueue : xfsdatad_workqueue; | |
217 | queue_work(wq, &ioend->io_work); | |
218 | if (wait) | |
219 | flush_workqueue(wq); | |
220 | } | |
ba87ea69 LM |
221 | } |
222 | ||
0829c360 | 223 | /* |
5ec4fabb | 224 | * IO write completion. |
f6d6d4fc CH |
225 | */ |
226 | STATIC void | |
5ec4fabb | 227 | xfs_end_io( |
77d7a0c2 | 228 | struct work_struct *work) |
0829c360 | 229 | { |
77d7a0c2 DC |
230 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
231 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 232 | int error = 0; |
ba87ea69 | 233 | |
5ec4fabb CH |
234 | /* |
235 | * For unwritten extents we need to issue transactions to convert a | |
236 | * range to normal written extens after the data I/O has finished. | |
237 | */ | |
238 | if (ioend->io_type == IOMAP_UNWRITTEN && | |
239 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { | |
5ec4fabb CH |
240 | |
241 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
242 | ioend->io_size); | |
243 | if (error) | |
244 | ioend->io_error = error; | |
245 | } | |
ba87ea69 | 246 | |
5ec4fabb CH |
247 | /* |
248 | * We might have to update the on-disk file size after extending | |
249 | * writes. | |
250 | */ | |
77d7a0c2 DC |
251 | if (ioend->io_type != IOMAP_READ) { |
252 | error = xfs_setfilesize(ioend); | |
253 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 254 | } |
77d7a0c2 DC |
255 | |
256 | /* | |
257 | * If we didn't complete processing of the ioend, requeue it to the | |
258 | * tail of the workqueue for another attempt later. Otherwise destroy | |
259 | * it. | |
260 | */ | |
261 | if (error == EAGAIN) { | |
262 | atomic_inc(&ioend->io_remaining); | |
263 | xfs_finish_ioend(ioend, 0); | |
264 | /* ensure we don't spin on blocked ioends */ | |
265 | delay(1); | |
266 | } else | |
267 | xfs_destroy_ioend(ioend); | |
c626d174 DC |
268 | } |
269 | ||
0829c360 CH |
270 | /* |
271 | * Allocate and initialise an IO completion structure. | |
272 | * We need to track unwritten extent write completion here initially. | |
273 | * We'll need to extend this for updating the ondisk inode size later | |
274 | * (vs. incore size). | |
275 | */ | |
276 | STATIC xfs_ioend_t * | |
277 | xfs_alloc_ioend( | |
f6d6d4fc CH |
278 | struct inode *inode, |
279 | unsigned int type) | |
0829c360 CH |
280 | { |
281 | xfs_ioend_t *ioend; | |
282 | ||
283 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
284 | ||
285 | /* | |
286 | * Set the count to 1 initially, which will prevent an I/O | |
287 | * completion callback from happening before we have started | |
288 | * all the I/O from calling the completion routine too early. | |
289 | */ | |
290 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 291 | ioend->io_error = 0; |
f6d6d4fc CH |
292 | ioend->io_list = NULL; |
293 | ioend->io_type = type; | |
b677c210 | 294 | ioend->io_inode = inode; |
c1a073bd | 295 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 296 | ioend->io_buffer_tail = NULL; |
b677c210 | 297 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
298 | ioend->io_offset = 0; |
299 | ioend->io_size = 0; | |
300 | ||
5ec4fabb | 301 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
302 | return ioend; |
303 | } | |
304 | ||
1da177e4 LT |
305 | STATIC int |
306 | xfs_map_blocks( | |
307 | struct inode *inode, | |
308 | loff_t offset, | |
309 | ssize_t count, | |
310 | xfs_iomap_t *mapp, | |
311 | int flags) | |
312 | { | |
6bd16ff2 CH |
313 | int nmaps = 1; |
314 | ||
315 | return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps); | |
1da177e4 LT |
316 | } |
317 | ||
b8f82a4a | 318 | STATIC int |
1defeac9 | 319 | xfs_iomap_valid( |
1da177e4 | 320 | xfs_iomap_t *iomapp, |
1defeac9 | 321 | loff_t offset) |
1da177e4 | 322 | { |
1defeac9 CH |
323 | return offset >= iomapp->iomap_offset && |
324 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
325 | } |
326 | ||
f6d6d4fc CH |
327 | /* |
328 | * BIO completion handler for buffered IO. | |
329 | */ | |
782e3b3b | 330 | STATIC void |
f6d6d4fc CH |
331 | xfs_end_bio( |
332 | struct bio *bio, | |
f6d6d4fc CH |
333 | int error) |
334 | { | |
335 | xfs_ioend_t *ioend = bio->bi_private; | |
336 | ||
f6d6d4fc | 337 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 338 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
339 | |
340 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
341 | bio->bi_private = NULL; |
342 | bio->bi_end_io = NULL; | |
f6d6d4fc | 343 | bio_put(bio); |
7d04a335 | 344 | |
e927af90 | 345 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
346 | } |
347 | ||
348 | STATIC void | |
349 | xfs_submit_ioend_bio( | |
06342cf8 CH |
350 | struct writeback_control *wbc, |
351 | xfs_ioend_t *ioend, | |
352 | struct bio *bio) | |
f6d6d4fc CH |
353 | { |
354 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
355 | bio->bi_private = ioend; |
356 | bio->bi_end_io = xfs_end_bio; | |
357 | ||
932640e8 DC |
358 | /* |
359 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
360 | * but don't update the inode size until I/O completion. | |
361 | */ | |
362 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 363 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 364 | |
06342cf8 CH |
365 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
366 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
367 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
368 | bio_put(bio); | |
369 | } | |
370 | ||
371 | STATIC struct bio * | |
372 | xfs_alloc_ioend_bio( | |
373 | struct buffer_head *bh) | |
374 | { | |
375 | struct bio *bio; | |
376 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
377 | ||
378 | do { | |
379 | bio = bio_alloc(GFP_NOIO, nvecs); | |
380 | nvecs >>= 1; | |
381 | } while (!bio); | |
382 | ||
383 | ASSERT(bio->bi_private == NULL); | |
384 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
385 | bio->bi_bdev = bh->b_bdev; | |
386 | bio_get(bio); | |
387 | return bio; | |
388 | } | |
389 | ||
390 | STATIC void | |
391 | xfs_start_buffer_writeback( | |
392 | struct buffer_head *bh) | |
393 | { | |
394 | ASSERT(buffer_mapped(bh)); | |
395 | ASSERT(buffer_locked(bh)); | |
396 | ASSERT(!buffer_delay(bh)); | |
397 | ASSERT(!buffer_unwritten(bh)); | |
398 | ||
399 | mark_buffer_async_write(bh); | |
400 | set_buffer_uptodate(bh); | |
401 | clear_buffer_dirty(bh); | |
402 | } | |
403 | ||
404 | STATIC void | |
405 | xfs_start_page_writeback( | |
406 | struct page *page, | |
f6d6d4fc CH |
407 | int clear_dirty, |
408 | int buffers) | |
409 | { | |
410 | ASSERT(PageLocked(page)); | |
411 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 412 | if (clear_dirty) |
92132021 DC |
413 | clear_page_dirty_for_io(page); |
414 | set_page_writeback(page); | |
f6d6d4fc | 415 | unlock_page(page); |
1f7decf6 FW |
416 | /* If no buffers on the page are to be written, finish it here */ |
417 | if (!buffers) | |
f6d6d4fc | 418 | end_page_writeback(page); |
f6d6d4fc CH |
419 | } |
420 | ||
421 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
422 | { | |
423 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
424 | } | |
425 | ||
426 | /* | |
d88992f6 DC |
427 | * Submit all of the bios for all of the ioends we have saved up, covering the |
428 | * initial writepage page and also any probed pages. | |
429 | * | |
430 | * Because we may have multiple ioends spanning a page, we need to start | |
431 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
432 | * buffers as we got, then we can end up with a page that only has buffers | |
433 | * marked async write and I/O complete on can occur before we mark the other | |
434 | * buffers async write. | |
435 | * | |
436 | * The end result of this is that we trip a bug in end_page_writeback() because | |
437 | * we call it twice for the one page as the code in end_buffer_async_write() | |
438 | * assumes that all buffers on the page are started at the same time. | |
439 | * | |
440 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 441 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
442 | */ |
443 | STATIC void | |
444 | xfs_submit_ioend( | |
06342cf8 | 445 | struct writeback_control *wbc, |
f6d6d4fc CH |
446 | xfs_ioend_t *ioend) |
447 | { | |
d88992f6 | 448 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
449 | xfs_ioend_t *next; |
450 | struct buffer_head *bh; | |
451 | struct bio *bio; | |
452 | sector_t lastblock = 0; | |
453 | ||
d88992f6 DC |
454 | /* Pass 1 - start writeback */ |
455 | do { | |
456 | next = ioend->io_list; | |
457 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
458 | xfs_start_buffer_writeback(bh); | |
459 | } | |
460 | } while ((ioend = next) != NULL); | |
461 | ||
462 | /* Pass 2 - submit I/O */ | |
463 | ioend = head; | |
f6d6d4fc CH |
464 | do { |
465 | next = ioend->io_list; | |
466 | bio = NULL; | |
467 | ||
468 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
469 | |
470 | if (!bio) { | |
471 | retry: | |
472 | bio = xfs_alloc_ioend_bio(bh); | |
473 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 474 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
475 | goto retry; |
476 | } | |
477 | ||
478 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 479 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
480 | goto retry; |
481 | } | |
482 | ||
483 | lastblock = bh->b_blocknr; | |
484 | } | |
485 | if (bio) | |
06342cf8 | 486 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 487 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
488 | } while ((ioend = next) != NULL); |
489 | } | |
490 | ||
491 | /* | |
492 | * Cancel submission of all buffer_heads so far in this endio. | |
493 | * Toss the endio too. Only ever called for the initial page | |
494 | * in a writepage request, so only ever one page. | |
495 | */ | |
496 | STATIC void | |
497 | xfs_cancel_ioend( | |
498 | xfs_ioend_t *ioend) | |
499 | { | |
500 | xfs_ioend_t *next; | |
501 | struct buffer_head *bh, *next_bh; | |
502 | ||
503 | do { | |
504 | next = ioend->io_list; | |
505 | bh = ioend->io_buffer_head; | |
506 | do { | |
507 | next_bh = bh->b_private; | |
508 | clear_buffer_async_write(bh); | |
509 | unlock_buffer(bh); | |
510 | } while ((bh = next_bh) != NULL); | |
511 | ||
25e41b3d | 512 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
513 | mempool_free(ioend, xfs_ioend_pool); |
514 | } while ((ioend = next) != NULL); | |
515 | } | |
516 | ||
517 | /* | |
518 | * Test to see if we've been building up a completion structure for | |
519 | * earlier buffers -- if so, we try to append to this ioend if we | |
520 | * can, otherwise we finish off any current ioend and start another. | |
521 | * Return true if we've finished the given ioend. | |
522 | */ | |
523 | STATIC void | |
524 | xfs_add_to_ioend( | |
525 | struct inode *inode, | |
526 | struct buffer_head *bh, | |
7336cea8 | 527 | xfs_off_t offset, |
f6d6d4fc CH |
528 | unsigned int type, |
529 | xfs_ioend_t **result, | |
530 | int need_ioend) | |
531 | { | |
532 | xfs_ioend_t *ioend = *result; | |
533 | ||
534 | if (!ioend || need_ioend || type != ioend->io_type) { | |
535 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 536 | |
f6d6d4fc CH |
537 | ioend = xfs_alloc_ioend(inode, type); |
538 | ioend->io_offset = offset; | |
539 | ioend->io_buffer_head = bh; | |
540 | ioend->io_buffer_tail = bh; | |
541 | if (previous) | |
542 | previous->io_list = ioend; | |
543 | *result = ioend; | |
544 | } else { | |
545 | ioend->io_buffer_tail->b_private = bh; | |
546 | ioend->io_buffer_tail = bh; | |
547 | } | |
548 | ||
549 | bh->b_private = NULL; | |
550 | ioend->io_size += bh->b_size; | |
551 | } | |
552 | ||
87cbc49c NS |
553 | STATIC void |
554 | xfs_map_buffer( | |
555 | struct buffer_head *bh, | |
556 | xfs_iomap_t *mp, | |
557 | xfs_off_t offset, | |
558 | uint block_bits) | |
559 | { | |
560 | sector_t bn; | |
561 | ||
562 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
563 | ||
564 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
565 | ((offset - mp->iomap_offset) >> block_bits); | |
566 | ||
567 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
568 | ||
569 | bh->b_blocknr = bn; | |
570 | set_buffer_mapped(bh); | |
571 | } | |
572 | ||
1da177e4 LT |
573 | STATIC void |
574 | xfs_map_at_offset( | |
1da177e4 | 575 | struct buffer_head *bh, |
1defeac9 | 576 | loff_t offset, |
1da177e4 | 577 | int block_bits, |
1defeac9 | 578 | xfs_iomap_t *iomapp) |
1da177e4 | 579 | { |
1da177e4 LT |
580 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
581 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
582 | |
583 | lock_buffer(bh); | |
87cbc49c | 584 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 585 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
586 | set_buffer_mapped(bh); |
587 | clear_buffer_delay(bh); | |
f6d6d4fc | 588 | clear_buffer_unwritten(bh); |
1da177e4 LT |
589 | } |
590 | ||
591 | /* | |
6c4fe19f | 592 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
593 | */ |
594 | STATIC unsigned int | |
6c4fe19f | 595 | xfs_probe_page( |
10ce4444 | 596 | struct page *page, |
6c4fe19f CH |
597 | unsigned int pg_offset, |
598 | int mapped) | |
1da177e4 | 599 | { |
1da177e4 LT |
600 | int ret = 0; |
601 | ||
1da177e4 | 602 | if (PageWriteback(page)) |
10ce4444 | 603 | return 0; |
1da177e4 LT |
604 | |
605 | if (page->mapping && PageDirty(page)) { | |
606 | if (page_has_buffers(page)) { | |
607 | struct buffer_head *bh, *head; | |
608 | ||
609 | bh = head = page_buffers(page); | |
610 | do { | |
6c4fe19f CH |
611 | if (!buffer_uptodate(bh)) |
612 | break; | |
613 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
614 | break; |
615 | ret += bh->b_size; | |
616 | if (ret >= pg_offset) | |
617 | break; | |
618 | } while ((bh = bh->b_this_page) != head); | |
619 | } else | |
6c4fe19f | 620 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
621 | } |
622 | ||
1da177e4 LT |
623 | return ret; |
624 | } | |
625 | ||
f6d6d4fc | 626 | STATIC size_t |
6c4fe19f | 627 | xfs_probe_cluster( |
1da177e4 LT |
628 | struct inode *inode, |
629 | struct page *startpage, | |
630 | struct buffer_head *bh, | |
6c4fe19f CH |
631 | struct buffer_head *head, |
632 | int mapped) | |
1da177e4 | 633 | { |
10ce4444 | 634 | struct pagevec pvec; |
1da177e4 | 635 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
636 | size_t total = 0; |
637 | int done = 0, i; | |
1da177e4 LT |
638 | |
639 | /* First sum forwards in this page */ | |
640 | do { | |
2353e8e9 | 641 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 642 | return total; |
1da177e4 LT |
643 | total += bh->b_size; |
644 | } while ((bh = bh->b_this_page) != head); | |
645 | ||
10ce4444 CH |
646 | /* if we reached the end of the page, sum forwards in following pages */ |
647 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
648 | tindex = startpage->index + 1; | |
649 | ||
650 | /* Prune this back to avoid pathological behavior */ | |
651 | tloff = min(tlast, startpage->index + 64); | |
652 | ||
653 | pagevec_init(&pvec, 0); | |
654 | while (!done && tindex <= tloff) { | |
655 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
656 | ||
657 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
658 | break; | |
659 | ||
660 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
661 | struct page *page = pvec.pages[i]; | |
265c1fac | 662 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
663 | |
664 | if (tindex == tlast) { | |
665 | pg_offset = | |
666 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
667 | if (!pg_offset) { |
668 | done = 1; | |
10ce4444 | 669 | break; |
1defeac9 | 670 | } |
10ce4444 CH |
671 | } else |
672 | pg_offset = PAGE_CACHE_SIZE; | |
673 | ||
529ae9aa | 674 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 675 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
676 | unlock_page(page); |
677 | } | |
678 | ||
265c1fac | 679 | if (!pg_len) { |
10ce4444 CH |
680 | done = 1; |
681 | break; | |
682 | } | |
683 | ||
265c1fac | 684 | total += pg_len; |
1defeac9 | 685 | tindex++; |
1da177e4 | 686 | } |
10ce4444 CH |
687 | |
688 | pagevec_release(&pvec); | |
689 | cond_resched(); | |
1da177e4 | 690 | } |
10ce4444 | 691 | |
1da177e4 LT |
692 | return total; |
693 | } | |
694 | ||
695 | /* | |
10ce4444 CH |
696 | * Test if a given page is suitable for writing as part of an unwritten |
697 | * or delayed allocate extent. | |
1da177e4 | 698 | */ |
10ce4444 CH |
699 | STATIC int |
700 | xfs_is_delayed_page( | |
701 | struct page *page, | |
f6d6d4fc | 702 | unsigned int type) |
1da177e4 | 703 | { |
1da177e4 | 704 | if (PageWriteback(page)) |
10ce4444 | 705 | return 0; |
1da177e4 LT |
706 | |
707 | if (page->mapping && page_has_buffers(page)) { | |
708 | struct buffer_head *bh, *head; | |
709 | int acceptable = 0; | |
710 | ||
711 | bh = head = page_buffers(page); | |
712 | do { | |
f6d6d4fc CH |
713 | if (buffer_unwritten(bh)) |
714 | acceptable = (type == IOMAP_UNWRITTEN); | |
715 | else if (buffer_delay(bh)) | |
716 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 717 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 718 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 719 | else |
1da177e4 | 720 | break; |
1da177e4 LT |
721 | } while ((bh = bh->b_this_page) != head); |
722 | ||
723 | if (acceptable) | |
10ce4444 | 724 | return 1; |
1da177e4 LT |
725 | } |
726 | ||
10ce4444 | 727 | return 0; |
1da177e4 LT |
728 | } |
729 | ||
1da177e4 LT |
730 | /* |
731 | * Allocate & map buffers for page given the extent map. Write it out. | |
732 | * except for the original page of a writepage, this is called on | |
733 | * delalloc/unwritten pages only, for the original page it is possible | |
734 | * that the page has no mapping at all. | |
735 | */ | |
f6d6d4fc | 736 | STATIC int |
1da177e4 LT |
737 | xfs_convert_page( |
738 | struct inode *inode, | |
739 | struct page *page, | |
10ce4444 | 740 | loff_t tindex, |
1defeac9 | 741 | xfs_iomap_t *mp, |
f6d6d4fc | 742 | xfs_ioend_t **ioendp, |
1da177e4 | 743 | struct writeback_control *wbc, |
1da177e4 LT |
744 | int startio, |
745 | int all_bh) | |
746 | { | |
f6d6d4fc | 747 | struct buffer_head *bh, *head; |
9260dc6b CH |
748 | xfs_off_t end_offset; |
749 | unsigned long p_offset; | |
f6d6d4fc | 750 | unsigned int type; |
1da177e4 | 751 | int bbits = inode->i_blkbits; |
24e17b5f | 752 | int len, page_dirty; |
f6d6d4fc | 753 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 754 | xfs_off_t offset = page_offset(page); |
1da177e4 | 755 | |
10ce4444 CH |
756 | if (page->index != tindex) |
757 | goto fail; | |
529ae9aa | 758 | if (!trylock_page(page)) |
10ce4444 CH |
759 | goto fail; |
760 | if (PageWriteback(page)) | |
761 | goto fail_unlock_page; | |
762 | if (page->mapping != inode->i_mapping) | |
763 | goto fail_unlock_page; | |
764 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
765 | goto fail_unlock_page; | |
766 | ||
24e17b5f NS |
767 | /* |
768 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 769 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
770 | * |
771 | * Derivation: | |
772 | * | |
773 | * End offset is the highest offset that this page should represent. | |
774 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
775 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
776 | * hence give us the correct page_dirty count. On any other page, | |
777 | * it will be zero and in that case we need page_dirty to be the | |
778 | * count of buffers on the page. | |
24e17b5f | 779 | */ |
9260dc6b CH |
780 | end_offset = min_t(unsigned long long, |
781 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
782 | i_size_read(inode)); | |
783 | ||
24e17b5f | 784 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
785 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
786 | PAGE_CACHE_SIZE); | |
787 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
788 | page_dirty = p_offset / len; | |
24e17b5f | 789 | |
1da177e4 LT |
790 | bh = head = page_buffers(page); |
791 | do { | |
9260dc6b | 792 | if (offset >= end_offset) |
1da177e4 | 793 | break; |
f6d6d4fc CH |
794 | if (!buffer_uptodate(bh)) |
795 | uptodate = 0; | |
796 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
797 | done = 1; | |
1da177e4 | 798 | continue; |
f6d6d4fc CH |
799 | } |
800 | ||
9260dc6b CH |
801 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
802 | if (buffer_unwritten(bh)) | |
803 | type = IOMAP_UNWRITTEN; | |
804 | else | |
805 | type = IOMAP_DELAY; | |
806 | ||
807 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 808 | done = 1; |
9260dc6b CH |
809 | continue; |
810 | } | |
811 | ||
812 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
813 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
814 | ||
815 | xfs_map_at_offset(bh, offset, bbits, mp); | |
816 | if (startio) { | |
7336cea8 | 817 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
818 | type, ioendp, done); |
819 | } else { | |
820 | set_buffer_dirty(bh); | |
821 | unlock_buffer(bh); | |
822 | mark_buffer_dirty(bh); | |
823 | } | |
824 | page_dirty--; | |
825 | count++; | |
826 | } else { | |
df3c7244 | 827 | type = IOMAP_NEW; |
9260dc6b | 828 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 829 | lock_buffer(bh); |
7336cea8 | 830 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
831 | type, ioendp, done); |
832 | count++; | |
24e17b5f | 833 | page_dirty--; |
9260dc6b CH |
834 | } else { |
835 | done = 1; | |
1da177e4 | 836 | } |
1da177e4 | 837 | } |
7336cea8 | 838 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 839 | |
f6d6d4fc CH |
840 | if (uptodate && bh == head) |
841 | SetPageUptodate(page); | |
842 | ||
843 | if (startio) { | |
f5e596bb | 844 | if (count) { |
9fddaca2 | 845 | wbc->nr_to_write--; |
0d99519e | 846 | if (wbc->nr_to_write <= 0) |
f5e596bb | 847 | done = 1; |
f5e596bb | 848 | } |
b41759cf | 849 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 850 | } |
f6d6d4fc CH |
851 | |
852 | return done; | |
10ce4444 CH |
853 | fail_unlock_page: |
854 | unlock_page(page); | |
855 | fail: | |
856 | return 1; | |
1da177e4 LT |
857 | } |
858 | ||
859 | /* | |
860 | * Convert & write out a cluster of pages in the same extent as defined | |
861 | * by mp and following the start page. | |
862 | */ | |
863 | STATIC void | |
864 | xfs_cluster_write( | |
865 | struct inode *inode, | |
866 | pgoff_t tindex, | |
867 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 868 | xfs_ioend_t **ioendp, |
1da177e4 LT |
869 | struct writeback_control *wbc, |
870 | int startio, | |
871 | int all_bh, | |
872 | pgoff_t tlast) | |
873 | { | |
10ce4444 CH |
874 | struct pagevec pvec; |
875 | int done = 0, i; | |
1da177e4 | 876 | |
10ce4444 CH |
877 | pagevec_init(&pvec, 0); |
878 | while (!done && tindex <= tlast) { | |
879 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
880 | ||
881 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 882 | break; |
10ce4444 CH |
883 | |
884 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
885 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
886 | iomapp, ioendp, wbc, startio, all_bh); | |
887 | if (done) | |
888 | break; | |
889 | } | |
890 | ||
891 | pagevec_release(&pvec); | |
892 | cond_resched(); | |
1da177e4 LT |
893 | } |
894 | } | |
895 | ||
896 | /* | |
897 | * Calling this without startio set means we are being asked to make a dirty | |
898 | * page ready for freeing it's buffers. When called with startio set then | |
899 | * we are coming from writepage. | |
900 | * | |
901 | * When called with startio set it is important that we write the WHOLE | |
902 | * page if possible. | |
903 | * The bh->b_state's cannot know if any of the blocks or which block for | |
904 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 905 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
906 | * may clear the page dirty flag prior to calling write page, under the |
907 | * assumption the entire page will be written out; by not writing out the | |
908 | * whole page the page can be reused before all valid dirty data is | |
909 | * written out. Note: in the case of a page that has been dirty'd by | |
910 | * mapwrite and but partially setup by block_prepare_write the | |
911 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
912 | * valid state, thus the whole page must be written out thing. | |
913 | */ | |
914 | ||
915 | STATIC int | |
916 | xfs_page_state_convert( | |
917 | struct inode *inode, | |
918 | struct page *page, | |
919 | struct writeback_control *wbc, | |
920 | int startio, | |
921 | int unmapped) /* also implies page uptodate */ | |
922 | { | |
f6d6d4fc | 923 | struct buffer_head *bh, *head; |
1defeac9 | 924 | xfs_iomap_t iomap; |
f6d6d4fc | 925 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
926 | loff_t offset; |
927 | unsigned long p_offset = 0; | |
f6d6d4fc | 928 | unsigned int type; |
1da177e4 LT |
929 | __uint64_t end_offset; |
930 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
931 | ssize_t size, len; |
932 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
933 | int page_dirty, count = 0; |
934 | int trylock = 0; | |
6c4fe19f | 935 | int all_bh = unmapped; |
1da177e4 | 936 | |
8272145c NS |
937 | if (startio) { |
938 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
939 | trylock |= BMAPI_TRYLOCK; | |
940 | } | |
3ba0815a | 941 | |
1da177e4 LT |
942 | /* Is this page beyond the end of the file? */ |
943 | offset = i_size_read(inode); | |
944 | end_index = offset >> PAGE_CACHE_SHIFT; | |
945 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
946 | if (page->index >= end_index) { | |
947 | if ((page->index >= end_index + 1) || | |
948 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
949 | if (startio) |
950 | unlock_page(page); | |
951 | return 0; | |
1da177e4 LT |
952 | } |
953 | } | |
954 | ||
1da177e4 | 955 | /* |
24e17b5f | 956 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 957 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
958 | * |
959 | * Derivation: | |
960 | * | |
961 | * End offset is the highest offset that this page should represent. | |
962 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
963 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
964 | * hence give us the correct page_dirty count. On any other page, | |
965 | * it will be zero and in that case we need page_dirty to be the | |
966 | * count of buffers on the page. | |
967 | */ | |
968 | end_offset = min_t(unsigned long long, | |
969 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 970 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
971 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
972 | PAGE_CACHE_SIZE); | |
973 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
974 | page_dirty = p_offset / len; |
975 | ||
24e17b5f | 976 | bh = head = page_buffers(page); |
f6d6d4fc | 977 | offset = page_offset(page); |
df3c7244 DC |
978 | flags = BMAPI_READ; |
979 | type = IOMAP_NEW; | |
f6d6d4fc | 980 | |
f6d6d4fc | 981 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
982 | |
983 | do { | |
984 | if (offset >= end_offset) | |
985 | break; | |
986 | if (!buffer_uptodate(bh)) | |
987 | uptodate = 0; | |
f6d6d4fc | 988 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
989 | /* |
990 | * the iomap is actually still valid, but the ioend | |
991 | * isn't. shouldn't happen too often. | |
992 | */ | |
993 | iomap_valid = 0; | |
1da177e4 | 994 | continue; |
f6d6d4fc | 995 | } |
1da177e4 | 996 | |
1defeac9 CH |
997 | if (iomap_valid) |
998 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
999 | |
1000 | /* | |
1001 | * First case, map an unwritten extent and prepare for | |
1002 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1003 | * |
1da177e4 LT |
1004 | * Second case, allocate space for a delalloc buffer. |
1005 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1006 | * |
1007 | * Third case, an unmapped buffer was found, and we are | |
1008 | * in a path where we need to write the whole page out. | |
df3c7244 | 1009 | */ |
d5cb48aa CH |
1010 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1011 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1012 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1013 | int new_ioend = 0; |
1014 | ||
df3c7244 | 1015 | /* |
6c4fe19f CH |
1016 | * Make sure we don't use a read-only iomap |
1017 | */ | |
df3c7244 | 1018 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1019 | iomap_valid = 0; |
1020 | ||
f6d6d4fc CH |
1021 | if (buffer_unwritten(bh)) { |
1022 | type = IOMAP_UNWRITTEN; | |
8272145c | 1023 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1024 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1025 | type = IOMAP_DELAY; |
8272145c | 1026 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1027 | } else { |
6c4fe19f | 1028 | type = IOMAP_NEW; |
8272145c | 1029 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1030 | } |
1031 | ||
1defeac9 | 1032 | if (!iomap_valid) { |
effd120e DC |
1033 | /* |
1034 | * if we didn't have a valid mapping then we | |
1035 | * need to ensure that we put the new mapping | |
1036 | * in a new ioend structure. This needs to be | |
1037 | * done to ensure that the ioends correctly | |
1038 | * reflect the block mappings at io completion | |
1039 | * for unwritten extent conversion. | |
1040 | */ | |
1041 | new_ioend = 1; | |
6c4fe19f CH |
1042 | if (type == IOMAP_NEW) { |
1043 | size = xfs_probe_cluster(inode, | |
1044 | page, bh, head, 0); | |
d5cb48aa CH |
1045 | } else { |
1046 | size = len; | |
1047 | } | |
1048 | ||
1049 | err = xfs_map_blocks(inode, offset, size, | |
1050 | &iomap, flags); | |
f6d6d4fc | 1051 | if (err) |
1da177e4 | 1052 | goto error; |
1defeac9 | 1053 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1054 | } |
1defeac9 CH |
1055 | if (iomap_valid) { |
1056 | xfs_map_at_offset(bh, offset, | |
1057 | inode->i_blkbits, &iomap); | |
1da177e4 | 1058 | if (startio) { |
7336cea8 | 1059 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1060 | type, &ioend, |
effd120e | 1061 | new_ioend); |
1da177e4 LT |
1062 | } else { |
1063 | set_buffer_dirty(bh); | |
1064 | unlock_buffer(bh); | |
1065 | mark_buffer_dirty(bh); | |
1066 | } | |
1067 | page_dirty--; | |
f6d6d4fc | 1068 | count++; |
1da177e4 | 1069 | } |
d5cb48aa | 1070 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1071 | /* |
1072 | * we got here because the buffer is already mapped. | |
1073 | * That means it must already have extents allocated | |
1074 | * underneath it. Map the extent by reading it. | |
1075 | */ | |
df3c7244 | 1076 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1077 | flags = BMAPI_READ; |
1078 | size = xfs_probe_cluster(inode, page, bh, | |
1079 | head, 1); | |
1080 | err = xfs_map_blocks(inode, offset, size, | |
1081 | &iomap, flags); | |
1082 | if (err) | |
1083 | goto error; | |
1084 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1085 | } | |
d5cb48aa | 1086 | |
df3c7244 DC |
1087 | /* |
1088 | * We set the type to IOMAP_NEW in case we are doing a | |
1089 | * small write at EOF that is extending the file but | |
1090 | * without needing an allocation. We need to update the | |
1091 | * file size on I/O completion in this case so it is | |
1092 | * the same case as having just allocated a new extent | |
1093 | * that we are writing into for the first time. | |
1094 | */ | |
1095 | type = IOMAP_NEW; | |
ca5de404 | 1096 | if (trylock_buffer(bh)) { |
d5cb48aa | 1097 | ASSERT(buffer_mapped(bh)); |
6c4fe19f CH |
1098 | if (iomap_valid) |
1099 | all_bh = 1; | |
7336cea8 | 1100 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1101 | &ioend, !iomap_valid); |
1102 | page_dirty--; | |
1103 | count++; | |
f6d6d4fc | 1104 | } else { |
1defeac9 | 1105 | iomap_valid = 0; |
1da177e4 | 1106 | } |
d5cb48aa CH |
1107 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1108 | (unmapped || startio)) { | |
1109 | iomap_valid = 0; | |
1da177e4 | 1110 | } |
f6d6d4fc CH |
1111 | |
1112 | if (!iohead) | |
1113 | iohead = ioend; | |
1114 | ||
1115 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1116 | |
1117 | if (uptodate && bh == head) | |
1118 | SetPageUptodate(page); | |
1119 | ||
f6d6d4fc | 1120 | if (startio) |
b41759cf | 1121 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1122 | |
1defeac9 CH |
1123 | if (ioend && iomap_valid) { |
1124 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1125 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1126 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1127 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1128 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1129 | } |
1130 | ||
f6d6d4fc | 1131 | if (iohead) |
06342cf8 | 1132 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1133 | |
1da177e4 LT |
1134 | return page_dirty; |
1135 | ||
1136 | error: | |
f6d6d4fc CH |
1137 | if (iohead) |
1138 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1139 | |
1140 | /* | |
1141 | * If it's delalloc and we have nowhere to put it, | |
1142 | * throw it away, unless the lower layers told | |
1143 | * us to try again. | |
1144 | */ | |
1145 | if (err != -EAGAIN) { | |
f6d6d4fc | 1146 | if (!unmapped) |
1da177e4 | 1147 | block_invalidatepage(page, 0); |
1da177e4 LT |
1148 | ClearPageUptodate(page); |
1149 | } | |
1150 | return err; | |
1151 | } | |
1152 | ||
f51623b2 NS |
1153 | /* |
1154 | * writepage: Called from one of two places: | |
1155 | * | |
1156 | * 1. we are flushing a delalloc buffer head. | |
1157 | * | |
1158 | * 2. we are writing out a dirty page. Typically the page dirty | |
1159 | * state is cleared before we get here. In this case is it | |
1160 | * conceivable we have no buffer heads. | |
1161 | * | |
1162 | * For delalloc space on the page we need to allocate space and | |
1163 | * flush it. For unmapped buffer heads on the page we should | |
1164 | * allocate space if the page is uptodate. For any other dirty | |
1165 | * buffer heads on the page we should flush them. | |
1166 | * | |
1167 | * If we detect that a transaction would be required to flush | |
1168 | * the page, we have to check the process flags first, if we | |
1169 | * are already in a transaction or disk I/O during allocations | |
1170 | * is off, we need to fail the writepage and redirty the page. | |
1171 | */ | |
1172 | ||
1173 | STATIC int | |
e4c573bb | 1174 | xfs_vm_writepage( |
f51623b2 NS |
1175 | struct page *page, |
1176 | struct writeback_control *wbc) | |
1177 | { | |
1178 | int error; | |
1179 | int need_trans; | |
1180 | int delalloc, unmapped, unwritten; | |
1181 | struct inode *inode = page->mapping->host; | |
1182 | ||
0b1b213f | 1183 | trace_xfs_writepage(inode, page, 0); |
f51623b2 NS |
1184 | |
1185 | /* | |
1186 | * We need a transaction if: | |
1187 | * 1. There are delalloc buffers on the page | |
1188 | * 2. The page is uptodate and we have unmapped buffers | |
1189 | * 3. The page is uptodate and we have no buffers | |
1190 | * 4. There are unwritten buffers on the page | |
1191 | */ | |
1192 | ||
1193 | if (!page_has_buffers(page)) { | |
1194 | unmapped = 1; | |
1195 | need_trans = 1; | |
1196 | } else { | |
1197 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1198 | if (!PageUptodate(page)) | |
1199 | unmapped = 0; | |
1200 | need_trans = delalloc + unmapped + unwritten; | |
1201 | } | |
1202 | ||
1203 | /* | |
1204 | * If we need a transaction and the process flags say | |
1205 | * we are already in a transaction, or no IO is allowed | |
1206 | * then mark the page dirty again and leave the page | |
1207 | * as is. | |
1208 | */ | |
59c1b082 | 1209 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1210 | goto out_fail; |
1211 | ||
1212 | /* | |
1213 | * Delay hooking up buffer heads until we have | |
1214 | * made our go/no-go decision. | |
1215 | */ | |
1216 | if (!page_has_buffers(page)) | |
1217 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1218 | ||
c8a4051c ES |
1219 | |
1220 | /* | |
1221 | * VM calculation for nr_to_write seems off. Bump it way | |
1222 | * up, this gets simple streaming writes zippy again. | |
1223 | * To be reviewed again after Jens' writeback changes. | |
1224 | */ | |
1225 | wbc->nr_to_write *= 4; | |
1226 | ||
f51623b2 NS |
1227 | /* |
1228 | * Convert delayed allocate, unwritten or unmapped space | |
1229 | * to real space and flush out to disk. | |
1230 | */ | |
1231 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1232 | if (error == -EAGAIN) | |
1233 | goto out_fail; | |
1234 | if (unlikely(error < 0)) | |
1235 | goto out_unlock; | |
1236 | ||
1237 | return 0; | |
1238 | ||
1239 | out_fail: | |
1240 | redirty_page_for_writepage(wbc, page); | |
1241 | unlock_page(page); | |
1242 | return 0; | |
1243 | out_unlock: | |
1244 | unlock_page(page); | |
1245 | return error; | |
1246 | } | |
1247 | ||
7d4fb40a NS |
1248 | STATIC int |
1249 | xfs_vm_writepages( | |
1250 | struct address_space *mapping, | |
1251 | struct writeback_control *wbc) | |
1252 | { | |
b3aea4ed | 1253 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1254 | return generic_writepages(mapping, wbc); |
1255 | } | |
1256 | ||
f51623b2 NS |
1257 | /* |
1258 | * Called to move a page into cleanable state - and from there | |
1259 | * to be released. Possibly the page is already clean. We always | |
1260 | * have buffer heads in this call. | |
1261 | * | |
1262 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1263 | * | |
1264 | * Possible scenarios are: | |
1265 | * | |
1266 | * 1. We are being called to release a page which has been written | |
1267 | * to via regular I/O. buffer heads will be dirty and possibly | |
1268 | * delalloc. If no delalloc buffer heads in this case then we | |
1269 | * can just return zero. | |
1270 | * | |
1271 | * 2. We are called to release a page which has been written via | |
1272 | * mmap, all we need to do is ensure there is no delalloc | |
1273 | * state in the buffer heads, if not we can let the caller | |
1274 | * free them and we should come back later via writepage. | |
1275 | */ | |
1276 | STATIC int | |
238f4c54 | 1277 | xfs_vm_releasepage( |
f51623b2 NS |
1278 | struct page *page, |
1279 | gfp_t gfp_mask) | |
1280 | { | |
1281 | struct inode *inode = page->mapping->host; | |
1282 | int dirty, delalloc, unmapped, unwritten; | |
1283 | struct writeback_control wbc = { | |
1284 | .sync_mode = WB_SYNC_ALL, | |
1285 | .nr_to_write = 1, | |
1286 | }; | |
1287 | ||
0b1b213f | 1288 | trace_xfs_releasepage(inode, page, 0); |
f51623b2 | 1289 | |
238f4c54 NS |
1290 | if (!page_has_buffers(page)) |
1291 | return 0; | |
1292 | ||
f51623b2 NS |
1293 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1294 | if (!delalloc && !unwritten) | |
1295 | goto free_buffers; | |
1296 | ||
1297 | if (!(gfp_mask & __GFP_FS)) | |
1298 | return 0; | |
1299 | ||
1300 | /* If we are already inside a transaction or the thread cannot | |
1301 | * do I/O, we cannot release this page. | |
1302 | */ | |
59c1b082 | 1303 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1304 | return 0; |
1305 | ||
1306 | /* | |
1307 | * Convert delalloc space to real space, do not flush the | |
1308 | * data out to disk, that will be done by the caller. | |
1309 | * Never need to allocate space here - we will always | |
1310 | * come back to writepage in that case. | |
1311 | */ | |
1312 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1313 | if (dirty == 0 && !unwritten) | |
1314 | goto free_buffers; | |
1315 | return 0; | |
1316 | ||
1317 | free_buffers: | |
1318 | return try_to_free_buffers(page); | |
1319 | } | |
1320 | ||
1da177e4 | 1321 | STATIC int |
c2536668 | 1322 | __xfs_get_blocks( |
1da177e4 LT |
1323 | struct inode *inode, |
1324 | sector_t iblock, | |
1da177e4 LT |
1325 | struct buffer_head *bh_result, |
1326 | int create, | |
1327 | int direct, | |
1328 | bmapi_flags_t flags) | |
1329 | { | |
1da177e4 | 1330 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1331 | xfs_off_t offset; |
1332 | ssize_t size; | |
c2536668 | 1333 | int niomap = 1; |
1da177e4 | 1334 | int error; |
1da177e4 | 1335 | |
fdc7ed75 | 1336 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1337 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1338 | size = bh_result->b_size; | |
364f358a LM |
1339 | |
1340 | if (!create && direct && offset >= i_size_read(inode)) | |
1341 | return 0; | |
1342 | ||
541d7d3c | 1343 | error = xfs_iomap(XFS_I(inode), offset, size, |
67fcaa73 | 1344 | create ? flags : BMAPI_READ, &iomap, &niomap); |
1da177e4 LT |
1345 | if (error) |
1346 | return -error; | |
c2536668 | 1347 | if (niomap == 0) |
1da177e4 LT |
1348 | return 0; |
1349 | ||
1350 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1351 | /* |
1352 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1353 | * the read case (treat as if we're reading into a hole). |
1354 | */ | |
1355 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1356 | xfs_map_buffer(bh_result, &iomap, offset, |
1357 | inode->i_blkbits); | |
1da177e4 LT |
1358 | } |
1359 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1360 | if (direct) | |
1361 | bh_result->b_private = inode; | |
1362 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1363 | } |
1364 | } | |
1365 | ||
c2536668 NS |
1366 | /* |
1367 | * If this is a realtime file, data may be on a different device. | |
1368 | * to that pointed to from the buffer_head b_bdev currently. | |
1369 | */ | |
ce8e922c | 1370 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1371 | |
c2536668 | 1372 | /* |
549054af DC |
1373 | * If we previously allocated a block out beyond eof and we are now |
1374 | * coming back to use it then we will need to flag it as new even if it | |
1375 | * has a disk address. | |
1376 | * | |
1377 | * With sub-block writes into unwritten extents we also need to mark | |
1378 | * the buffer as new so that the unwritten parts of the buffer gets | |
1379 | * correctly zeroed. | |
1da177e4 LT |
1380 | */ |
1381 | if (create && | |
1382 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1383 | (offset >= i_size_read(inode)) || |
1384 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1385 | set_buffer_new(bh_result); |
1da177e4 LT |
1386 | |
1387 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1388 | BUG_ON(direct); | |
1389 | if (create) { | |
1390 | set_buffer_uptodate(bh_result); | |
1391 | set_buffer_mapped(bh_result); | |
1392 | set_buffer_delay(bh_result); | |
1393 | } | |
1394 | } | |
1395 | ||
c2536668 | 1396 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1397 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1398 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1399 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1400 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1401 | } |
1402 | ||
1403 | return 0; | |
1404 | } | |
1405 | ||
1406 | int | |
c2536668 | 1407 | xfs_get_blocks( |
1da177e4 LT |
1408 | struct inode *inode, |
1409 | sector_t iblock, | |
1410 | struct buffer_head *bh_result, | |
1411 | int create) | |
1412 | { | |
c2536668 | 1413 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1414 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1415 | } |
1416 | ||
1417 | STATIC int | |
e4c573bb | 1418 | xfs_get_blocks_direct( |
1da177e4 LT |
1419 | struct inode *inode, |
1420 | sector_t iblock, | |
1da177e4 LT |
1421 | struct buffer_head *bh_result, |
1422 | int create) | |
1423 | { | |
c2536668 | 1424 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1425 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1426 | } |
1427 | ||
f0973863 | 1428 | STATIC void |
e4c573bb | 1429 | xfs_end_io_direct( |
f0973863 CH |
1430 | struct kiocb *iocb, |
1431 | loff_t offset, | |
1432 | ssize_t size, | |
1433 | void *private) | |
1434 | { | |
1435 | xfs_ioend_t *ioend = iocb->private; | |
1436 | ||
1437 | /* | |
1438 | * Non-NULL private data means we need to issue a transaction to | |
1439 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1440 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1441 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1442 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1443 | * it anyway to keep the code uniform and simpler. |
1444 | * | |
e927af90 DC |
1445 | * Well, if only it were that simple. Because synchronous direct I/O |
1446 | * requires extent conversion to occur *before* we return to userspace, | |
1447 | * we have to wait for extent conversion to complete. Look at the | |
1448 | * iocb that has been passed to us to determine if this is AIO or | |
1449 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1450 | * workqueue and wait for it to complete. | |
1451 | * | |
f0973863 CH |
1452 | * The core direct I/O code might be changed to always call the |
1453 | * completion handler in the future, in which case all this can | |
1454 | * go away. | |
1455 | */ | |
ba87ea69 LM |
1456 | ioend->io_offset = offset; |
1457 | ioend->io_size = size; | |
1458 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1459 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1460 | } else if (private && size > 0) { |
e927af90 | 1461 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1462 | } else { |
ba87ea69 LM |
1463 | /* |
1464 | * A direct I/O write ioend starts it's life in unwritten | |
1465 | * state in case they map an unwritten extent. This write | |
1466 | * didn't map an unwritten extent so switch it's completion | |
1467 | * handler. | |
1468 | */ | |
5ec4fabb | 1469 | ioend->io_type = IOMAP_NEW; |
e927af90 | 1470 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1471 | } |
1472 | ||
1473 | /* | |
c41564b5 | 1474 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1475 | * completion handler was called. Thus we need to protect |
1476 | * against double-freeing. | |
1477 | */ | |
1478 | iocb->private = NULL; | |
1479 | } | |
1480 | ||
1da177e4 | 1481 | STATIC ssize_t |
e4c573bb | 1482 | xfs_vm_direct_IO( |
1da177e4 LT |
1483 | int rw, |
1484 | struct kiocb *iocb, | |
1485 | const struct iovec *iov, | |
1486 | loff_t offset, | |
1487 | unsigned long nr_segs) | |
1488 | { | |
1489 | struct file *file = iocb->ki_filp; | |
1490 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1491 | struct block_device *bdev; |
f0973863 | 1492 | ssize_t ret; |
1da177e4 | 1493 | |
6214ed44 | 1494 | bdev = xfs_find_bdev_for_inode(XFS_I(inode)); |
1da177e4 | 1495 | |
5fe878ae CH |
1496 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
1497 | IOMAP_UNWRITTEN : IOMAP_READ); | |
1498 | ||
1499 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1500 | offset, nr_segs, | |
1501 | xfs_get_blocks_direct, | |
1502 | xfs_end_io_direct); | |
f0973863 | 1503 | |
8459d86a | 1504 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1505 | xfs_destroy_ioend(iocb->private); |
1506 | return ret; | |
1da177e4 LT |
1507 | } |
1508 | ||
f51623b2 | 1509 | STATIC int |
d79689c7 | 1510 | xfs_vm_write_begin( |
f51623b2 | 1511 | struct file *file, |
d79689c7 NP |
1512 | struct address_space *mapping, |
1513 | loff_t pos, | |
1514 | unsigned len, | |
1515 | unsigned flags, | |
1516 | struct page **pagep, | |
1517 | void **fsdata) | |
f51623b2 | 1518 | { |
d79689c7 NP |
1519 | *pagep = NULL; |
1520 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1521 | xfs_get_blocks); | |
f51623b2 | 1522 | } |
1da177e4 LT |
1523 | |
1524 | STATIC sector_t | |
e4c573bb | 1525 | xfs_vm_bmap( |
1da177e4 LT |
1526 | struct address_space *mapping, |
1527 | sector_t block) | |
1528 | { | |
1529 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1530 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1531 | |
cf441eeb | 1532 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1533 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1534 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1535 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1536 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1537 | } |
1538 | ||
1539 | STATIC int | |
e4c573bb | 1540 | xfs_vm_readpage( |
1da177e4 LT |
1541 | struct file *unused, |
1542 | struct page *page) | |
1543 | { | |
c2536668 | 1544 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | STATIC int | |
e4c573bb | 1548 | xfs_vm_readpages( |
1da177e4 LT |
1549 | struct file *unused, |
1550 | struct address_space *mapping, | |
1551 | struct list_head *pages, | |
1552 | unsigned nr_pages) | |
1553 | { | |
c2536668 | 1554 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1555 | } |
1556 | ||
2ff28e22 | 1557 | STATIC void |
238f4c54 | 1558 | xfs_vm_invalidatepage( |
bcec2b7f NS |
1559 | struct page *page, |
1560 | unsigned long offset) | |
1561 | { | |
0b1b213f | 1562 | trace_xfs_invalidatepage(page->mapping->host, page, offset); |
2ff28e22 | 1563 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1564 | } |
1565 | ||
f5e54d6e | 1566 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1567 | .readpage = xfs_vm_readpage, |
1568 | .readpages = xfs_vm_readpages, | |
1569 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1570 | .writepages = xfs_vm_writepages, |
1da177e4 | 1571 | .sync_page = block_sync_page, |
238f4c54 NS |
1572 | .releasepage = xfs_vm_releasepage, |
1573 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1574 | .write_begin = xfs_vm_write_begin, |
1575 | .write_end = generic_write_end, | |
e4c573bb NS |
1576 | .bmap = xfs_vm_bmap, |
1577 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1578 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1579 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1580 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1581 | }; |