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" |
1da177e4 | 41 | #include <linux/mpage.h> |
10ce4444 | 42 | #include <linux/pagevec.h> |
1da177e4 LT |
43 | #include <linux/writeback.h> |
44 | ||
25e41b3d CH |
45 | |
46 | /* | |
47 | * Prime number of hash buckets since address is used as the key. | |
48 | */ | |
49 | #define NVSYNC 37 | |
50 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
51 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
52 | ||
53 | void __init | |
54 | xfs_ioend_init(void) | |
55 | { | |
56 | int i; | |
57 | ||
58 | for (i = 0; i < NVSYNC; i++) | |
59 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
60 | } | |
61 | ||
62 | void | |
63 | xfs_ioend_wait( | |
64 | xfs_inode_t *ip) | |
65 | { | |
66 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
67 | ||
68 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
69 | } | |
70 | ||
71 | STATIC void | |
72 | xfs_ioend_wake( | |
73 | xfs_inode_t *ip) | |
74 | { | |
75 | if (atomic_dec_and_test(&ip->i_iocount)) | |
76 | wake_up(to_ioend_wq(ip)); | |
77 | } | |
78 | ||
f51623b2 NS |
79 | STATIC void |
80 | xfs_count_page_state( | |
81 | struct page *page, | |
82 | int *delalloc, | |
83 | int *unmapped, | |
84 | int *unwritten) | |
85 | { | |
86 | struct buffer_head *bh, *head; | |
87 | ||
88 | *delalloc = *unmapped = *unwritten = 0; | |
89 | ||
90 | bh = head = page_buffers(page); | |
91 | do { | |
92 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
93 | (*unmapped) = 1; | |
f51623b2 NS |
94 | else if (buffer_unwritten(bh)) |
95 | (*unwritten) = 1; | |
96 | else if (buffer_delay(bh)) | |
97 | (*delalloc) = 1; | |
98 | } while ((bh = bh->b_this_page) != head); | |
99 | } | |
100 | ||
1da177e4 LT |
101 | #if defined(XFS_RW_TRACE) |
102 | void | |
103 | xfs_page_trace( | |
104 | int tag, | |
105 | struct inode *inode, | |
106 | struct page *page, | |
ed9d88f7 | 107 | unsigned long pgoff) |
1da177e4 LT |
108 | { |
109 | xfs_inode_t *ip; | |
1da177e4 | 110 | loff_t isize = i_size_read(inode); |
f6d6d4fc | 111 | loff_t offset = page_offset(page); |
1da177e4 LT |
112 | int delalloc = -1, unmapped = -1, unwritten = -1; |
113 | ||
114 | if (page_has_buffers(page)) | |
115 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
116 | ||
e6064d30 | 117 | ip = XFS_I(inode); |
1da177e4 LT |
118 | if (!ip->i_rwtrace) |
119 | return; | |
120 | ||
121 | ktrace_enter(ip->i_rwtrace, | |
122 | (void *)((unsigned long)tag), | |
123 | (void *)ip, | |
124 | (void *)inode, | |
125 | (void *)page, | |
ed9d88f7 | 126 | (void *)pgoff, |
1da177e4 LT |
127 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), |
128 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
129 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
130 | (void *)((unsigned long)(isize & 0xffffffff)), | |
131 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
132 | (void *)((unsigned long)(offset & 0xffffffff)), | |
133 | (void *)((unsigned long)delalloc), | |
134 | (void *)((unsigned long)unmapped), | |
135 | (void *)((unsigned long)unwritten), | |
f1fdc848 | 136 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
137 | (void *)NULL); |
138 | } | |
139 | #else | |
ed9d88f7 | 140 | #define xfs_page_trace(tag, inode, page, pgoff) |
1da177e4 LT |
141 | #endif |
142 | ||
6214ed44 CH |
143 | STATIC struct block_device * |
144 | xfs_find_bdev_for_inode( | |
145 | struct xfs_inode *ip) | |
146 | { | |
147 | struct xfs_mount *mp = ip->i_mount; | |
148 | ||
71ddabb9 | 149 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
150 | return mp->m_rtdev_targp->bt_bdev; |
151 | else | |
152 | return mp->m_ddev_targp->bt_bdev; | |
153 | } | |
154 | ||
0829c360 CH |
155 | /* |
156 | * Schedule IO completion handling on a xfsdatad if this was | |
e927af90 DC |
157 | * the final hold on this ioend. If we are asked to wait, |
158 | * flush the workqueue. | |
0829c360 CH |
159 | */ |
160 | STATIC void | |
161 | xfs_finish_ioend( | |
e927af90 DC |
162 | xfs_ioend_t *ioend, |
163 | int wait) | |
0829c360 | 164 | { |
e927af90 | 165 | if (atomic_dec_and_test(&ioend->io_remaining)) { |
0829c360 | 166 | queue_work(xfsdatad_workqueue, &ioend->io_work); |
e927af90 DC |
167 | if (wait) |
168 | flush_workqueue(xfsdatad_workqueue); | |
169 | } | |
0829c360 CH |
170 | } |
171 | ||
f6d6d4fc CH |
172 | /* |
173 | * We're now finished for good with this ioend structure. | |
174 | * Update the page state via the associated buffer_heads, | |
175 | * release holds on the inode and bio, and finally free | |
176 | * up memory. Do not use the ioend after this. | |
177 | */ | |
0829c360 CH |
178 | STATIC void |
179 | xfs_destroy_ioend( | |
180 | xfs_ioend_t *ioend) | |
181 | { | |
f6d6d4fc | 182 | struct buffer_head *bh, *next; |
583fa586 | 183 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
184 | |
185 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
186 | next = bh->b_private; | |
7d04a335 | 187 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 188 | } |
583fa586 CH |
189 | |
190 | /* | |
191 | * Volume managers supporting multiple paths can send back ENODEV | |
192 | * when the final path disappears. In this case continuing to fill | |
193 | * the page cache with dirty data which cannot be written out is | |
194 | * evil, so prevent that. | |
195 | */ | |
196 | if (unlikely(ioend->io_error == -ENODEV)) { | |
197 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
198 | __FILE__, __LINE__); | |
b677c210 | 199 | } |
583fa586 | 200 | |
25e41b3d | 201 | xfs_ioend_wake(ip); |
0829c360 CH |
202 | mempool_free(ioend, xfs_ioend_pool); |
203 | } | |
204 | ||
ba87ea69 LM |
205 | /* |
206 | * Update on-disk file size now that data has been written to disk. | |
207 | * The current in-memory file size is i_size. If a write is beyond | |
613d7043 | 208 | * eof i_new_size will be the intended file size until i_size is |
ba87ea69 LM |
209 | * updated. If this write does not extend all the way to the valid |
210 | * file size then restrict this update to the end of the write. | |
211 | */ | |
212 | STATIC void | |
213 | xfs_setfilesize( | |
214 | xfs_ioend_t *ioend) | |
215 | { | |
b677c210 | 216 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 LM |
217 | xfs_fsize_t isize; |
218 | xfs_fsize_t bsize; | |
219 | ||
ba87ea69 LM |
220 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
221 | ASSERT(ioend->io_type != IOMAP_READ); | |
222 | ||
223 | if (unlikely(ioend->io_error)) | |
224 | return; | |
225 | ||
226 | bsize = ioend->io_offset + ioend->io_size; | |
227 | ||
228 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
229 | ||
613d7043 | 230 | isize = MAX(ip->i_size, ip->i_new_size); |
ba87ea69 LM |
231 | isize = MIN(isize, bsize); |
232 | ||
233 | if (ip->i_d.di_size < isize) { | |
234 | ip->i_d.di_size = isize; | |
235 | ip->i_update_core = 1; | |
236 | ip->i_update_size = 1; | |
94b97e39 | 237 | xfs_mark_inode_dirty_sync(ip); |
ba87ea69 LM |
238 | } |
239 | ||
240 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
241 | } | |
242 | ||
0829c360 | 243 | /* |
f6d6d4fc | 244 | * Buffered IO write completion for delayed allocate extents. |
f6d6d4fc CH |
245 | */ |
246 | STATIC void | |
247 | xfs_end_bio_delalloc( | |
c4028958 | 248 | struct work_struct *work) |
f6d6d4fc | 249 | { |
c4028958 DH |
250 | xfs_ioend_t *ioend = |
251 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 252 | |
ba87ea69 | 253 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
254 | xfs_destroy_ioend(ioend); |
255 | } | |
256 | ||
257 | /* | |
258 | * Buffered IO write completion for regular, written extents. | |
259 | */ | |
260 | STATIC void | |
261 | xfs_end_bio_written( | |
c4028958 | 262 | struct work_struct *work) |
f6d6d4fc | 263 | { |
c4028958 DH |
264 | xfs_ioend_t *ioend = |
265 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 266 | |
ba87ea69 | 267 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
268 | xfs_destroy_ioend(ioend); |
269 | } | |
270 | ||
271 | /* | |
272 | * IO write completion for unwritten extents. | |
273 | * | |
0829c360 | 274 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 275 | * to written extents. |
0829c360 CH |
276 | */ |
277 | STATIC void | |
278 | xfs_end_bio_unwritten( | |
c4028958 | 279 | struct work_struct *work) |
0829c360 | 280 | { |
c4028958 DH |
281 | xfs_ioend_t *ioend = |
282 | container_of(work, xfs_ioend_t, io_work); | |
7642861b | 283 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
0829c360 CH |
284 | xfs_off_t offset = ioend->io_offset; |
285 | size_t size = ioend->io_size; | |
0829c360 | 286 | |
ba87ea69 | 287 | if (likely(!ioend->io_error)) { |
cc88466f DC |
288 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
289 | int error; | |
290 | error = xfs_iomap_write_unwritten(ip, offset, size); | |
291 | if (error) | |
292 | ioend->io_error = error; | |
293 | } | |
ba87ea69 LM |
294 | xfs_setfilesize(ioend); |
295 | } | |
296 | xfs_destroy_ioend(ioend); | |
297 | } | |
298 | ||
299 | /* | |
300 | * IO read completion for regular, written extents. | |
301 | */ | |
302 | STATIC void | |
303 | xfs_end_bio_read( | |
304 | struct work_struct *work) | |
305 | { | |
306 | xfs_ioend_t *ioend = | |
307 | container_of(work, xfs_ioend_t, io_work); | |
308 | ||
0829c360 CH |
309 | xfs_destroy_ioend(ioend); |
310 | } | |
311 | ||
312 | /* | |
313 | * Allocate and initialise an IO completion structure. | |
314 | * We need to track unwritten extent write completion here initially. | |
315 | * We'll need to extend this for updating the ondisk inode size later | |
316 | * (vs. incore size). | |
317 | */ | |
318 | STATIC xfs_ioend_t * | |
319 | xfs_alloc_ioend( | |
f6d6d4fc CH |
320 | struct inode *inode, |
321 | unsigned int type) | |
0829c360 CH |
322 | { |
323 | xfs_ioend_t *ioend; | |
324 | ||
325 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
326 | ||
327 | /* | |
328 | * Set the count to 1 initially, which will prevent an I/O | |
329 | * completion callback from happening before we have started | |
330 | * all the I/O from calling the completion routine too early. | |
331 | */ | |
332 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 333 | ioend->io_error = 0; |
f6d6d4fc CH |
334 | ioend->io_list = NULL; |
335 | ioend->io_type = type; | |
b677c210 | 336 | ioend->io_inode = inode; |
c1a073bd | 337 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 338 | ioend->io_buffer_tail = NULL; |
b677c210 | 339 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
340 | ioend->io_offset = 0; |
341 | ioend->io_size = 0; | |
342 | ||
f6d6d4fc | 343 | if (type == IOMAP_UNWRITTEN) |
c4028958 | 344 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten); |
f6d6d4fc | 345 | else if (type == IOMAP_DELAY) |
c4028958 | 346 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc); |
ba87ea69 LM |
347 | else if (type == IOMAP_READ) |
348 | INIT_WORK(&ioend->io_work, xfs_end_bio_read); | |
f6d6d4fc | 349 | else |
c4028958 | 350 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); |
0829c360 CH |
351 | |
352 | return ioend; | |
353 | } | |
354 | ||
1da177e4 LT |
355 | STATIC int |
356 | xfs_map_blocks( | |
357 | struct inode *inode, | |
358 | loff_t offset, | |
359 | ssize_t count, | |
360 | xfs_iomap_t *mapp, | |
361 | int flags) | |
362 | { | |
6bd16ff2 CH |
363 | int nmaps = 1; |
364 | ||
365 | return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps); | |
1da177e4 LT |
366 | } |
367 | ||
7989cb8e | 368 | STATIC_INLINE int |
1defeac9 | 369 | xfs_iomap_valid( |
1da177e4 | 370 | xfs_iomap_t *iomapp, |
1defeac9 | 371 | loff_t offset) |
1da177e4 | 372 | { |
1defeac9 CH |
373 | return offset >= iomapp->iomap_offset && |
374 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
375 | } |
376 | ||
f6d6d4fc CH |
377 | /* |
378 | * BIO completion handler for buffered IO. | |
379 | */ | |
782e3b3b | 380 | STATIC void |
f6d6d4fc CH |
381 | xfs_end_bio( |
382 | struct bio *bio, | |
f6d6d4fc CH |
383 | int error) |
384 | { | |
385 | xfs_ioend_t *ioend = bio->bi_private; | |
386 | ||
f6d6d4fc | 387 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 388 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
389 | |
390 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
391 | bio->bi_private = NULL; |
392 | bio->bi_end_io = NULL; | |
f6d6d4fc | 393 | bio_put(bio); |
7d04a335 | 394 | |
e927af90 | 395 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
396 | } |
397 | ||
398 | STATIC void | |
399 | xfs_submit_ioend_bio( | |
400 | xfs_ioend_t *ioend, | |
401 | struct bio *bio) | |
402 | { | |
403 | atomic_inc(&ioend->io_remaining); | |
404 | ||
405 | bio->bi_private = ioend; | |
406 | bio->bi_end_io = xfs_end_bio; | |
407 | ||
408 | submit_bio(WRITE, bio); | |
409 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
410 | bio_put(bio); | |
411 | } | |
412 | ||
413 | STATIC struct bio * | |
414 | xfs_alloc_ioend_bio( | |
415 | struct buffer_head *bh) | |
416 | { | |
417 | struct bio *bio; | |
418 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
419 | ||
420 | do { | |
421 | bio = bio_alloc(GFP_NOIO, nvecs); | |
422 | nvecs >>= 1; | |
423 | } while (!bio); | |
424 | ||
425 | ASSERT(bio->bi_private == NULL); | |
426 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
427 | bio->bi_bdev = bh->b_bdev; | |
428 | bio_get(bio); | |
429 | return bio; | |
430 | } | |
431 | ||
432 | STATIC void | |
433 | xfs_start_buffer_writeback( | |
434 | struct buffer_head *bh) | |
435 | { | |
436 | ASSERT(buffer_mapped(bh)); | |
437 | ASSERT(buffer_locked(bh)); | |
438 | ASSERT(!buffer_delay(bh)); | |
439 | ASSERT(!buffer_unwritten(bh)); | |
440 | ||
441 | mark_buffer_async_write(bh); | |
442 | set_buffer_uptodate(bh); | |
443 | clear_buffer_dirty(bh); | |
444 | } | |
445 | ||
446 | STATIC void | |
447 | xfs_start_page_writeback( | |
448 | struct page *page, | |
f6d6d4fc CH |
449 | int clear_dirty, |
450 | int buffers) | |
451 | { | |
452 | ASSERT(PageLocked(page)); | |
453 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 454 | if (clear_dirty) |
92132021 DC |
455 | clear_page_dirty_for_io(page); |
456 | set_page_writeback(page); | |
f6d6d4fc | 457 | unlock_page(page); |
1f7decf6 FW |
458 | /* If no buffers on the page are to be written, finish it here */ |
459 | if (!buffers) | |
f6d6d4fc | 460 | end_page_writeback(page); |
f6d6d4fc CH |
461 | } |
462 | ||
463 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
464 | { | |
465 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
466 | } | |
467 | ||
468 | /* | |
d88992f6 DC |
469 | * Submit all of the bios for all of the ioends we have saved up, covering the |
470 | * initial writepage page and also any probed pages. | |
471 | * | |
472 | * Because we may have multiple ioends spanning a page, we need to start | |
473 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
474 | * buffers as we got, then we can end up with a page that only has buffers | |
475 | * marked async write and I/O complete on can occur before we mark the other | |
476 | * buffers async write. | |
477 | * | |
478 | * The end result of this is that we trip a bug in end_page_writeback() because | |
479 | * we call it twice for the one page as the code in end_buffer_async_write() | |
480 | * assumes that all buffers on the page are started at the same time. | |
481 | * | |
482 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 483 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
484 | */ |
485 | STATIC void | |
486 | xfs_submit_ioend( | |
487 | xfs_ioend_t *ioend) | |
488 | { | |
d88992f6 | 489 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
490 | xfs_ioend_t *next; |
491 | struct buffer_head *bh; | |
492 | struct bio *bio; | |
493 | sector_t lastblock = 0; | |
494 | ||
d88992f6 DC |
495 | /* Pass 1 - start writeback */ |
496 | do { | |
497 | next = ioend->io_list; | |
498 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
499 | xfs_start_buffer_writeback(bh); | |
500 | } | |
501 | } while ((ioend = next) != NULL); | |
502 | ||
503 | /* Pass 2 - submit I/O */ | |
504 | ioend = head; | |
f6d6d4fc CH |
505 | do { |
506 | next = ioend->io_list; | |
507 | bio = NULL; | |
508 | ||
509 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
510 | |
511 | if (!bio) { | |
512 | retry: | |
513 | bio = xfs_alloc_ioend_bio(bh); | |
514 | } else if (bh->b_blocknr != lastblock + 1) { | |
515 | xfs_submit_ioend_bio(ioend, bio); | |
516 | goto retry; | |
517 | } | |
518 | ||
519 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
520 | xfs_submit_ioend_bio(ioend, bio); | |
521 | goto retry; | |
522 | } | |
523 | ||
524 | lastblock = bh->b_blocknr; | |
525 | } | |
526 | if (bio) | |
527 | xfs_submit_ioend_bio(ioend, bio); | |
e927af90 | 528 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
529 | } while ((ioend = next) != NULL); |
530 | } | |
531 | ||
532 | /* | |
533 | * Cancel submission of all buffer_heads so far in this endio. | |
534 | * Toss the endio too. Only ever called for the initial page | |
535 | * in a writepage request, so only ever one page. | |
536 | */ | |
537 | STATIC void | |
538 | xfs_cancel_ioend( | |
539 | xfs_ioend_t *ioend) | |
540 | { | |
541 | xfs_ioend_t *next; | |
542 | struct buffer_head *bh, *next_bh; | |
543 | ||
544 | do { | |
545 | next = ioend->io_list; | |
546 | bh = ioend->io_buffer_head; | |
547 | do { | |
548 | next_bh = bh->b_private; | |
549 | clear_buffer_async_write(bh); | |
550 | unlock_buffer(bh); | |
551 | } while ((bh = next_bh) != NULL); | |
552 | ||
25e41b3d | 553 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
554 | mempool_free(ioend, xfs_ioend_pool); |
555 | } while ((ioend = next) != NULL); | |
556 | } | |
557 | ||
558 | /* | |
559 | * Test to see if we've been building up a completion structure for | |
560 | * earlier buffers -- if so, we try to append to this ioend if we | |
561 | * can, otherwise we finish off any current ioend and start another. | |
562 | * Return true if we've finished the given ioend. | |
563 | */ | |
564 | STATIC void | |
565 | xfs_add_to_ioend( | |
566 | struct inode *inode, | |
567 | struct buffer_head *bh, | |
7336cea8 | 568 | xfs_off_t offset, |
f6d6d4fc CH |
569 | unsigned int type, |
570 | xfs_ioend_t **result, | |
571 | int need_ioend) | |
572 | { | |
573 | xfs_ioend_t *ioend = *result; | |
574 | ||
575 | if (!ioend || need_ioend || type != ioend->io_type) { | |
576 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 577 | |
f6d6d4fc CH |
578 | ioend = xfs_alloc_ioend(inode, type); |
579 | ioend->io_offset = offset; | |
580 | ioend->io_buffer_head = bh; | |
581 | ioend->io_buffer_tail = bh; | |
582 | if (previous) | |
583 | previous->io_list = ioend; | |
584 | *result = ioend; | |
585 | } else { | |
586 | ioend->io_buffer_tail->b_private = bh; | |
587 | ioend->io_buffer_tail = bh; | |
588 | } | |
589 | ||
590 | bh->b_private = NULL; | |
591 | ioend->io_size += bh->b_size; | |
592 | } | |
593 | ||
87cbc49c NS |
594 | STATIC void |
595 | xfs_map_buffer( | |
596 | struct buffer_head *bh, | |
597 | xfs_iomap_t *mp, | |
598 | xfs_off_t offset, | |
599 | uint block_bits) | |
600 | { | |
601 | sector_t bn; | |
602 | ||
603 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
604 | ||
605 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
606 | ((offset - mp->iomap_offset) >> block_bits); | |
607 | ||
608 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
609 | ||
610 | bh->b_blocknr = bn; | |
611 | set_buffer_mapped(bh); | |
612 | } | |
613 | ||
1da177e4 LT |
614 | STATIC void |
615 | xfs_map_at_offset( | |
1da177e4 | 616 | struct buffer_head *bh, |
1defeac9 | 617 | loff_t offset, |
1da177e4 | 618 | int block_bits, |
1defeac9 | 619 | xfs_iomap_t *iomapp) |
1da177e4 | 620 | { |
1da177e4 LT |
621 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
622 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
623 | |
624 | lock_buffer(bh); | |
87cbc49c | 625 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 626 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
627 | set_buffer_mapped(bh); |
628 | clear_buffer_delay(bh); | |
f6d6d4fc | 629 | clear_buffer_unwritten(bh); |
1da177e4 LT |
630 | } |
631 | ||
632 | /* | |
6c4fe19f | 633 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
634 | */ |
635 | STATIC unsigned int | |
6c4fe19f | 636 | xfs_probe_page( |
10ce4444 | 637 | struct page *page, |
6c4fe19f CH |
638 | unsigned int pg_offset, |
639 | int mapped) | |
1da177e4 | 640 | { |
1da177e4 LT |
641 | int ret = 0; |
642 | ||
1da177e4 | 643 | if (PageWriteback(page)) |
10ce4444 | 644 | return 0; |
1da177e4 LT |
645 | |
646 | if (page->mapping && PageDirty(page)) { | |
647 | if (page_has_buffers(page)) { | |
648 | struct buffer_head *bh, *head; | |
649 | ||
650 | bh = head = page_buffers(page); | |
651 | do { | |
6c4fe19f CH |
652 | if (!buffer_uptodate(bh)) |
653 | break; | |
654 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
655 | break; |
656 | ret += bh->b_size; | |
657 | if (ret >= pg_offset) | |
658 | break; | |
659 | } while ((bh = bh->b_this_page) != head); | |
660 | } else | |
6c4fe19f | 661 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
662 | } |
663 | ||
1da177e4 LT |
664 | return ret; |
665 | } | |
666 | ||
f6d6d4fc | 667 | STATIC size_t |
6c4fe19f | 668 | xfs_probe_cluster( |
1da177e4 LT |
669 | struct inode *inode, |
670 | struct page *startpage, | |
671 | struct buffer_head *bh, | |
6c4fe19f CH |
672 | struct buffer_head *head, |
673 | int mapped) | |
1da177e4 | 674 | { |
10ce4444 | 675 | struct pagevec pvec; |
1da177e4 | 676 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
677 | size_t total = 0; |
678 | int done = 0, i; | |
1da177e4 LT |
679 | |
680 | /* First sum forwards in this page */ | |
681 | do { | |
2353e8e9 | 682 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 683 | return total; |
1da177e4 LT |
684 | total += bh->b_size; |
685 | } while ((bh = bh->b_this_page) != head); | |
686 | ||
10ce4444 CH |
687 | /* if we reached the end of the page, sum forwards in following pages */ |
688 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
689 | tindex = startpage->index + 1; | |
690 | ||
691 | /* Prune this back to avoid pathological behavior */ | |
692 | tloff = min(tlast, startpage->index + 64); | |
693 | ||
694 | pagevec_init(&pvec, 0); | |
695 | while (!done && tindex <= tloff) { | |
696 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
697 | ||
698 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
699 | break; | |
700 | ||
701 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
702 | struct page *page = pvec.pages[i]; | |
265c1fac | 703 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
704 | |
705 | if (tindex == tlast) { | |
706 | pg_offset = | |
707 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
708 | if (!pg_offset) { |
709 | done = 1; | |
10ce4444 | 710 | break; |
1defeac9 | 711 | } |
10ce4444 CH |
712 | } else |
713 | pg_offset = PAGE_CACHE_SIZE; | |
714 | ||
529ae9aa | 715 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 716 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
717 | unlock_page(page); |
718 | } | |
719 | ||
265c1fac | 720 | if (!pg_len) { |
10ce4444 CH |
721 | done = 1; |
722 | break; | |
723 | } | |
724 | ||
265c1fac | 725 | total += pg_len; |
1defeac9 | 726 | tindex++; |
1da177e4 | 727 | } |
10ce4444 CH |
728 | |
729 | pagevec_release(&pvec); | |
730 | cond_resched(); | |
1da177e4 | 731 | } |
10ce4444 | 732 | |
1da177e4 LT |
733 | return total; |
734 | } | |
735 | ||
736 | /* | |
10ce4444 CH |
737 | * Test if a given page is suitable for writing as part of an unwritten |
738 | * or delayed allocate extent. | |
1da177e4 | 739 | */ |
10ce4444 CH |
740 | STATIC int |
741 | xfs_is_delayed_page( | |
742 | struct page *page, | |
f6d6d4fc | 743 | unsigned int type) |
1da177e4 | 744 | { |
1da177e4 | 745 | if (PageWriteback(page)) |
10ce4444 | 746 | return 0; |
1da177e4 LT |
747 | |
748 | if (page->mapping && page_has_buffers(page)) { | |
749 | struct buffer_head *bh, *head; | |
750 | int acceptable = 0; | |
751 | ||
752 | bh = head = page_buffers(page); | |
753 | do { | |
f6d6d4fc CH |
754 | if (buffer_unwritten(bh)) |
755 | acceptable = (type == IOMAP_UNWRITTEN); | |
756 | else if (buffer_delay(bh)) | |
757 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 758 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 759 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 760 | else |
1da177e4 | 761 | break; |
1da177e4 LT |
762 | } while ((bh = bh->b_this_page) != head); |
763 | ||
764 | if (acceptable) | |
10ce4444 | 765 | return 1; |
1da177e4 LT |
766 | } |
767 | ||
10ce4444 | 768 | return 0; |
1da177e4 LT |
769 | } |
770 | ||
1da177e4 LT |
771 | /* |
772 | * Allocate & map buffers for page given the extent map. Write it out. | |
773 | * except for the original page of a writepage, this is called on | |
774 | * delalloc/unwritten pages only, for the original page it is possible | |
775 | * that the page has no mapping at all. | |
776 | */ | |
f6d6d4fc | 777 | STATIC int |
1da177e4 LT |
778 | xfs_convert_page( |
779 | struct inode *inode, | |
780 | struct page *page, | |
10ce4444 | 781 | loff_t tindex, |
1defeac9 | 782 | xfs_iomap_t *mp, |
f6d6d4fc | 783 | xfs_ioend_t **ioendp, |
1da177e4 | 784 | struct writeback_control *wbc, |
1da177e4 LT |
785 | int startio, |
786 | int all_bh) | |
787 | { | |
f6d6d4fc | 788 | struct buffer_head *bh, *head; |
9260dc6b CH |
789 | xfs_off_t end_offset; |
790 | unsigned long p_offset; | |
f6d6d4fc | 791 | unsigned int type; |
1da177e4 | 792 | int bbits = inode->i_blkbits; |
24e17b5f | 793 | int len, page_dirty; |
f6d6d4fc | 794 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 795 | xfs_off_t offset = page_offset(page); |
1da177e4 | 796 | |
10ce4444 CH |
797 | if (page->index != tindex) |
798 | goto fail; | |
529ae9aa | 799 | if (!trylock_page(page)) |
10ce4444 CH |
800 | goto fail; |
801 | if (PageWriteback(page)) | |
802 | goto fail_unlock_page; | |
803 | if (page->mapping != inode->i_mapping) | |
804 | goto fail_unlock_page; | |
805 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
806 | goto fail_unlock_page; | |
807 | ||
24e17b5f NS |
808 | /* |
809 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 810 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
811 | * |
812 | * Derivation: | |
813 | * | |
814 | * End offset is the highest offset that this page should represent. | |
815 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
816 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
817 | * hence give us the correct page_dirty count. On any other page, | |
818 | * it will be zero and in that case we need page_dirty to be the | |
819 | * count of buffers on the page. | |
24e17b5f | 820 | */ |
9260dc6b CH |
821 | end_offset = min_t(unsigned long long, |
822 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
823 | i_size_read(inode)); | |
824 | ||
24e17b5f | 825 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
826 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
827 | PAGE_CACHE_SIZE); | |
828 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
829 | page_dirty = p_offset / len; | |
24e17b5f | 830 | |
1da177e4 LT |
831 | bh = head = page_buffers(page); |
832 | do { | |
9260dc6b | 833 | if (offset >= end_offset) |
1da177e4 | 834 | break; |
f6d6d4fc CH |
835 | if (!buffer_uptodate(bh)) |
836 | uptodate = 0; | |
837 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
838 | done = 1; | |
1da177e4 | 839 | continue; |
f6d6d4fc CH |
840 | } |
841 | ||
9260dc6b CH |
842 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
843 | if (buffer_unwritten(bh)) | |
844 | type = IOMAP_UNWRITTEN; | |
845 | else | |
846 | type = IOMAP_DELAY; | |
847 | ||
848 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 849 | done = 1; |
9260dc6b CH |
850 | continue; |
851 | } | |
852 | ||
853 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
854 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
855 | ||
856 | xfs_map_at_offset(bh, offset, bbits, mp); | |
857 | if (startio) { | |
7336cea8 | 858 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
859 | type, ioendp, done); |
860 | } else { | |
861 | set_buffer_dirty(bh); | |
862 | unlock_buffer(bh); | |
863 | mark_buffer_dirty(bh); | |
864 | } | |
865 | page_dirty--; | |
866 | count++; | |
867 | } else { | |
df3c7244 | 868 | type = IOMAP_NEW; |
9260dc6b | 869 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 870 | lock_buffer(bh); |
7336cea8 | 871 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
872 | type, ioendp, done); |
873 | count++; | |
24e17b5f | 874 | page_dirty--; |
9260dc6b CH |
875 | } else { |
876 | done = 1; | |
1da177e4 | 877 | } |
1da177e4 | 878 | } |
7336cea8 | 879 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 880 | |
f6d6d4fc CH |
881 | if (uptodate && bh == head) |
882 | SetPageUptodate(page); | |
883 | ||
884 | if (startio) { | |
f5e596bb CH |
885 | if (count) { |
886 | struct backing_dev_info *bdi; | |
887 | ||
888 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 889 | wbc->nr_to_write--; |
f5e596bb CH |
890 | if (bdi_write_congested(bdi)) { |
891 | wbc->encountered_congestion = 1; | |
892 | done = 1; | |
9fddaca2 | 893 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
894 | done = 1; |
895 | } | |
896 | } | |
b41759cf | 897 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 898 | } |
f6d6d4fc CH |
899 | |
900 | return done; | |
10ce4444 CH |
901 | fail_unlock_page: |
902 | unlock_page(page); | |
903 | fail: | |
904 | return 1; | |
1da177e4 LT |
905 | } |
906 | ||
907 | /* | |
908 | * Convert & write out a cluster of pages in the same extent as defined | |
909 | * by mp and following the start page. | |
910 | */ | |
911 | STATIC void | |
912 | xfs_cluster_write( | |
913 | struct inode *inode, | |
914 | pgoff_t tindex, | |
915 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 916 | xfs_ioend_t **ioendp, |
1da177e4 LT |
917 | struct writeback_control *wbc, |
918 | int startio, | |
919 | int all_bh, | |
920 | pgoff_t tlast) | |
921 | { | |
10ce4444 CH |
922 | struct pagevec pvec; |
923 | int done = 0, i; | |
1da177e4 | 924 | |
10ce4444 CH |
925 | pagevec_init(&pvec, 0); |
926 | while (!done && tindex <= tlast) { | |
927 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
928 | ||
929 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 930 | break; |
10ce4444 CH |
931 | |
932 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
933 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
934 | iomapp, ioendp, wbc, startio, all_bh); | |
935 | if (done) | |
936 | break; | |
937 | } | |
938 | ||
939 | pagevec_release(&pvec); | |
940 | cond_resched(); | |
1da177e4 LT |
941 | } |
942 | } | |
943 | ||
944 | /* | |
945 | * Calling this without startio set means we are being asked to make a dirty | |
946 | * page ready for freeing it's buffers. When called with startio set then | |
947 | * we are coming from writepage. | |
948 | * | |
949 | * When called with startio set it is important that we write the WHOLE | |
950 | * page if possible. | |
951 | * The bh->b_state's cannot know if any of the blocks or which block for | |
952 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 953 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
954 | * may clear the page dirty flag prior to calling write page, under the |
955 | * assumption the entire page will be written out; by not writing out the | |
956 | * whole page the page can be reused before all valid dirty data is | |
957 | * written out. Note: in the case of a page that has been dirty'd by | |
958 | * mapwrite and but partially setup by block_prepare_write the | |
959 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
960 | * valid state, thus the whole page must be written out thing. | |
961 | */ | |
962 | ||
963 | STATIC int | |
964 | xfs_page_state_convert( | |
965 | struct inode *inode, | |
966 | struct page *page, | |
967 | struct writeback_control *wbc, | |
968 | int startio, | |
969 | int unmapped) /* also implies page uptodate */ | |
970 | { | |
f6d6d4fc | 971 | struct buffer_head *bh, *head; |
1defeac9 | 972 | xfs_iomap_t iomap; |
f6d6d4fc | 973 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
974 | loff_t offset; |
975 | unsigned long p_offset = 0; | |
f6d6d4fc | 976 | unsigned int type; |
1da177e4 LT |
977 | __uint64_t end_offset; |
978 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
979 | ssize_t size, len; |
980 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
981 | int page_dirty, count = 0; |
982 | int trylock = 0; | |
6c4fe19f | 983 | int all_bh = unmapped; |
1da177e4 | 984 | |
8272145c NS |
985 | if (startio) { |
986 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
987 | trylock |= BMAPI_TRYLOCK; | |
988 | } | |
3ba0815a | 989 | |
1da177e4 LT |
990 | /* Is this page beyond the end of the file? */ |
991 | offset = i_size_read(inode); | |
992 | end_index = offset >> PAGE_CACHE_SHIFT; | |
993 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
994 | if (page->index >= end_index) { | |
995 | if ((page->index >= end_index + 1) || | |
996 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
997 | if (startio) |
998 | unlock_page(page); | |
999 | return 0; | |
1da177e4 LT |
1000 | } |
1001 | } | |
1002 | ||
1da177e4 | 1003 | /* |
24e17b5f | 1004 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 1005 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
1006 | * |
1007 | * Derivation: | |
1008 | * | |
1009 | * End offset is the highest offset that this page should represent. | |
1010 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
1011 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
1012 | * hence give us the correct page_dirty count. On any other page, | |
1013 | * it will be zero and in that case we need page_dirty to be the | |
1014 | * count of buffers on the page. | |
1015 | */ | |
1016 | end_offset = min_t(unsigned long long, | |
1017 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 1018 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
1019 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
1020 | PAGE_CACHE_SIZE); | |
1021 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
1022 | page_dirty = p_offset / len; |
1023 | ||
24e17b5f | 1024 | bh = head = page_buffers(page); |
f6d6d4fc | 1025 | offset = page_offset(page); |
df3c7244 DC |
1026 | flags = BMAPI_READ; |
1027 | type = IOMAP_NEW; | |
f6d6d4fc | 1028 | |
f6d6d4fc | 1029 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
1030 | |
1031 | do { | |
1032 | if (offset >= end_offset) | |
1033 | break; | |
1034 | if (!buffer_uptodate(bh)) | |
1035 | uptodate = 0; | |
f6d6d4fc | 1036 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
1037 | /* |
1038 | * the iomap is actually still valid, but the ioend | |
1039 | * isn't. shouldn't happen too often. | |
1040 | */ | |
1041 | iomap_valid = 0; | |
1da177e4 | 1042 | continue; |
f6d6d4fc | 1043 | } |
1da177e4 | 1044 | |
1defeac9 CH |
1045 | if (iomap_valid) |
1046 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
1047 | |
1048 | /* | |
1049 | * First case, map an unwritten extent and prepare for | |
1050 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1051 | * |
1da177e4 LT |
1052 | * Second case, allocate space for a delalloc buffer. |
1053 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1054 | * |
1055 | * Third case, an unmapped buffer was found, and we are | |
1056 | * in a path where we need to write the whole page out. | |
df3c7244 | 1057 | */ |
d5cb48aa CH |
1058 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1059 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1060 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1061 | int new_ioend = 0; |
1062 | ||
df3c7244 | 1063 | /* |
6c4fe19f CH |
1064 | * Make sure we don't use a read-only iomap |
1065 | */ | |
df3c7244 | 1066 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1067 | iomap_valid = 0; |
1068 | ||
f6d6d4fc CH |
1069 | if (buffer_unwritten(bh)) { |
1070 | type = IOMAP_UNWRITTEN; | |
8272145c | 1071 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1072 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1073 | type = IOMAP_DELAY; |
8272145c | 1074 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1075 | } else { |
6c4fe19f | 1076 | type = IOMAP_NEW; |
8272145c | 1077 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1078 | } |
1079 | ||
1defeac9 | 1080 | if (!iomap_valid) { |
effd120e DC |
1081 | /* |
1082 | * if we didn't have a valid mapping then we | |
1083 | * need to ensure that we put the new mapping | |
1084 | * in a new ioend structure. This needs to be | |
1085 | * done to ensure that the ioends correctly | |
1086 | * reflect the block mappings at io completion | |
1087 | * for unwritten extent conversion. | |
1088 | */ | |
1089 | new_ioend = 1; | |
6c4fe19f CH |
1090 | if (type == IOMAP_NEW) { |
1091 | size = xfs_probe_cluster(inode, | |
1092 | page, bh, head, 0); | |
d5cb48aa CH |
1093 | } else { |
1094 | size = len; | |
1095 | } | |
1096 | ||
1097 | err = xfs_map_blocks(inode, offset, size, | |
1098 | &iomap, flags); | |
f6d6d4fc | 1099 | if (err) |
1da177e4 | 1100 | goto error; |
1defeac9 | 1101 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1102 | } |
1defeac9 CH |
1103 | if (iomap_valid) { |
1104 | xfs_map_at_offset(bh, offset, | |
1105 | inode->i_blkbits, &iomap); | |
1da177e4 | 1106 | if (startio) { |
7336cea8 | 1107 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1108 | type, &ioend, |
effd120e | 1109 | new_ioend); |
1da177e4 LT |
1110 | } else { |
1111 | set_buffer_dirty(bh); | |
1112 | unlock_buffer(bh); | |
1113 | mark_buffer_dirty(bh); | |
1114 | } | |
1115 | page_dirty--; | |
f6d6d4fc | 1116 | count++; |
1da177e4 | 1117 | } |
d5cb48aa | 1118 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1119 | /* |
1120 | * we got here because the buffer is already mapped. | |
1121 | * That means it must already have extents allocated | |
1122 | * underneath it. Map the extent by reading it. | |
1123 | */ | |
df3c7244 | 1124 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1125 | flags = BMAPI_READ; |
1126 | size = xfs_probe_cluster(inode, page, bh, | |
1127 | head, 1); | |
1128 | err = xfs_map_blocks(inode, offset, size, | |
1129 | &iomap, flags); | |
1130 | if (err) | |
1131 | goto error; | |
1132 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1133 | } | |
d5cb48aa | 1134 | |
df3c7244 DC |
1135 | /* |
1136 | * We set the type to IOMAP_NEW in case we are doing a | |
1137 | * small write at EOF that is extending the file but | |
1138 | * without needing an allocation. We need to update the | |
1139 | * file size on I/O completion in this case so it is | |
1140 | * the same case as having just allocated a new extent | |
1141 | * that we are writing into for the first time. | |
1142 | */ | |
1143 | type = IOMAP_NEW; | |
ca5de404 | 1144 | if (trylock_buffer(bh)) { |
d5cb48aa | 1145 | ASSERT(buffer_mapped(bh)); |
6c4fe19f CH |
1146 | if (iomap_valid) |
1147 | all_bh = 1; | |
7336cea8 | 1148 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1149 | &ioend, !iomap_valid); |
1150 | page_dirty--; | |
1151 | count++; | |
f6d6d4fc | 1152 | } else { |
1defeac9 | 1153 | iomap_valid = 0; |
1da177e4 | 1154 | } |
d5cb48aa CH |
1155 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1156 | (unmapped || startio)) { | |
1157 | iomap_valid = 0; | |
1da177e4 | 1158 | } |
f6d6d4fc CH |
1159 | |
1160 | if (!iohead) | |
1161 | iohead = ioend; | |
1162 | ||
1163 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1164 | |
1165 | if (uptodate && bh == head) | |
1166 | SetPageUptodate(page); | |
1167 | ||
f6d6d4fc | 1168 | if (startio) |
b41759cf | 1169 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1170 | |
1defeac9 CH |
1171 | if (ioend && iomap_valid) { |
1172 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1173 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1174 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1175 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1176 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1177 | } |
1178 | ||
f6d6d4fc CH |
1179 | if (iohead) |
1180 | xfs_submit_ioend(iohead); | |
1181 | ||
1da177e4 LT |
1182 | return page_dirty; |
1183 | ||
1184 | error: | |
f6d6d4fc CH |
1185 | if (iohead) |
1186 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1187 | |
1188 | /* | |
1189 | * If it's delalloc and we have nowhere to put it, | |
1190 | * throw it away, unless the lower layers told | |
1191 | * us to try again. | |
1192 | */ | |
1193 | if (err != -EAGAIN) { | |
f6d6d4fc | 1194 | if (!unmapped) |
1da177e4 | 1195 | block_invalidatepage(page, 0); |
1da177e4 LT |
1196 | ClearPageUptodate(page); |
1197 | } | |
1198 | return err; | |
1199 | } | |
1200 | ||
f51623b2 NS |
1201 | /* |
1202 | * writepage: Called from one of two places: | |
1203 | * | |
1204 | * 1. we are flushing a delalloc buffer head. | |
1205 | * | |
1206 | * 2. we are writing out a dirty page. Typically the page dirty | |
1207 | * state is cleared before we get here. In this case is it | |
1208 | * conceivable we have no buffer heads. | |
1209 | * | |
1210 | * For delalloc space on the page we need to allocate space and | |
1211 | * flush it. For unmapped buffer heads on the page we should | |
1212 | * allocate space if the page is uptodate. For any other dirty | |
1213 | * buffer heads on the page we should flush them. | |
1214 | * | |
1215 | * If we detect that a transaction would be required to flush | |
1216 | * the page, we have to check the process flags first, if we | |
1217 | * are already in a transaction or disk I/O during allocations | |
1218 | * is off, we need to fail the writepage and redirty the page. | |
1219 | */ | |
1220 | ||
1221 | STATIC int | |
e4c573bb | 1222 | xfs_vm_writepage( |
f51623b2 NS |
1223 | struct page *page, |
1224 | struct writeback_control *wbc) | |
1225 | { | |
1226 | int error; | |
1227 | int need_trans; | |
1228 | int delalloc, unmapped, unwritten; | |
1229 | struct inode *inode = page->mapping->host; | |
1230 | ||
1231 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1232 | ||
1233 | /* | |
1234 | * We need a transaction if: | |
1235 | * 1. There are delalloc buffers on the page | |
1236 | * 2. The page is uptodate and we have unmapped buffers | |
1237 | * 3. The page is uptodate and we have no buffers | |
1238 | * 4. There are unwritten buffers on the page | |
1239 | */ | |
1240 | ||
1241 | if (!page_has_buffers(page)) { | |
1242 | unmapped = 1; | |
1243 | need_trans = 1; | |
1244 | } else { | |
1245 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1246 | if (!PageUptodate(page)) | |
1247 | unmapped = 0; | |
1248 | need_trans = delalloc + unmapped + unwritten; | |
1249 | } | |
1250 | ||
1251 | /* | |
1252 | * If we need a transaction and the process flags say | |
1253 | * we are already in a transaction, or no IO is allowed | |
1254 | * then mark the page dirty again and leave the page | |
1255 | * as is. | |
1256 | */ | |
59c1b082 | 1257 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1258 | goto out_fail; |
1259 | ||
1260 | /* | |
1261 | * Delay hooking up buffer heads until we have | |
1262 | * made our go/no-go decision. | |
1263 | */ | |
1264 | if (!page_has_buffers(page)) | |
1265 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1266 | ||
1267 | /* | |
1268 | * Convert delayed allocate, unwritten or unmapped space | |
1269 | * to real space and flush out to disk. | |
1270 | */ | |
1271 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1272 | if (error == -EAGAIN) | |
1273 | goto out_fail; | |
1274 | if (unlikely(error < 0)) | |
1275 | goto out_unlock; | |
1276 | ||
1277 | return 0; | |
1278 | ||
1279 | out_fail: | |
1280 | redirty_page_for_writepage(wbc, page); | |
1281 | unlock_page(page); | |
1282 | return 0; | |
1283 | out_unlock: | |
1284 | unlock_page(page); | |
1285 | return error; | |
1286 | } | |
1287 | ||
7d4fb40a NS |
1288 | STATIC int |
1289 | xfs_vm_writepages( | |
1290 | struct address_space *mapping, | |
1291 | struct writeback_control *wbc) | |
1292 | { | |
b3aea4ed | 1293 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1294 | return generic_writepages(mapping, wbc); |
1295 | } | |
1296 | ||
f51623b2 NS |
1297 | /* |
1298 | * Called to move a page into cleanable state - and from there | |
1299 | * to be released. Possibly the page is already clean. We always | |
1300 | * have buffer heads in this call. | |
1301 | * | |
1302 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1303 | * | |
1304 | * Possible scenarios are: | |
1305 | * | |
1306 | * 1. We are being called to release a page which has been written | |
1307 | * to via regular I/O. buffer heads will be dirty and possibly | |
1308 | * delalloc. If no delalloc buffer heads in this case then we | |
1309 | * can just return zero. | |
1310 | * | |
1311 | * 2. We are called to release a page which has been written via | |
1312 | * mmap, all we need to do is ensure there is no delalloc | |
1313 | * state in the buffer heads, if not we can let the caller | |
1314 | * free them and we should come back later via writepage. | |
1315 | */ | |
1316 | STATIC int | |
238f4c54 | 1317 | xfs_vm_releasepage( |
f51623b2 NS |
1318 | struct page *page, |
1319 | gfp_t gfp_mask) | |
1320 | { | |
1321 | struct inode *inode = page->mapping->host; | |
1322 | int dirty, delalloc, unmapped, unwritten; | |
1323 | struct writeback_control wbc = { | |
1324 | .sync_mode = WB_SYNC_ALL, | |
1325 | .nr_to_write = 1, | |
1326 | }; | |
1327 | ||
ed9d88f7 | 1328 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0); |
f51623b2 | 1329 | |
238f4c54 NS |
1330 | if (!page_has_buffers(page)) |
1331 | return 0; | |
1332 | ||
f51623b2 NS |
1333 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1334 | if (!delalloc && !unwritten) | |
1335 | goto free_buffers; | |
1336 | ||
1337 | if (!(gfp_mask & __GFP_FS)) | |
1338 | return 0; | |
1339 | ||
1340 | /* If we are already inside a transaction or the thread cannot | |
1341 | * do I/O, we cannot release this page. | |
1342 | */ | |
59c1b082 | 1343 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1344 | return 0; |
1345 | ||
1346 | /* | |
1347 | * Convert delalloc space to real space, do not flush the | |
1348 | * data out to disk, that will be done by the caller. | |
1349 | * Never need to allocate space here - we will always | |
1350 | * come back to writepage in that case. | |
1351 | */ | |
1352 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1353 | if (dirty == 0 && !unwritten) | |
1354 | goto free_buffers; | |
1355 | return 0; | |
1356 | ||
1357 | free_buffers: | |
1358 | return try_to_free_buffers(page); | |
1359 | } | |
1360 | ||
1da177e4 | 1361 | STATIC int |
c2536668 | 1362 | __xfs_get_blocks( |
1da177e4 LT |
1363 | struct inode *inode, |
1364 | sector_t iblock, | |
1da177e4 LT |
1365 | struct buffer_head *bh_result, |
1366 | int create, | |
1367 | int direct, | |
1368 | bmapi_flags_t flags) | |
1369 | { | |
1da177e4 | 1370 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1371 | xfs_off_t offset; |
1372 | ssize_t size; | |
c2536668 | 1373 | int niomap = 1; |
1da177e4 | 1374 | int error; |
1da177e4 | 1375 | |
fdc7ed75 | 1376 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1377 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1378 | size = bh_result->b_size; | |
364f358a LM |
1379 | |
1380 | if (!create && direct && offset >= i_size_read(inode)) | |
1381 | return 0; | |
1382 | ||
541d7d3c | 1383 | error = xfs_iomap(XFS_I(inode), offset, size, |
67fcaa73 | 1384 | create ? flags : BMAPI_READ, &iomap, &niomap); |
1da177e4 LT |
1385 | if (error) |
1386 | return -error; | |
c2536668 | 1387 | if (niomap == 0) |
1da177e4 LT |
1388 | return 0; |
1389 | ||
1390 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1391 | /* |
1392 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1393 | * the read case (treat as if we're reading into a hole). |
1394 | */ | |
1395 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1396 | xfs_map_buffer(bh_result, &iomap, offset, |
1397 | inode->i_blkbits); | |
1da177e4 LT |
1398 | } |
1399 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1400 | if (direct) | |
1401 | bh_result->b_private = inode; | |
1402 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1403 | } |
1404 | } | |
1405 | ||
c2536668 NS |
1406 | /* |
1407 | * If this is a realtime file, data may be on a different device. | |
1408 | * to that pointed to from the buffer_head b_bdev currently. | |
1409 | */ | |
ce8e922c | 1410 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1411 | |
c2536668 | 1412 | /* |
549054af DC |
1413 | * If we previously allocated a block out beyond eof and we are now |
1414 | * coming back to use it then we will need to flag it as new even if it | |
1415 | * has a disk address. | |
1416 | * | |
1417 | * With sub-block writes into unwritten extents we also need to mark | |
1418 | * the buffer as new so that the unwritten parts of the buffer gets | |
1419 | * correctly zeroed. | |
1da177e4 LT |
1420 | */ |
1421 | if (create && | |
1422 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1423 | (offset >= i_size_read(inode)) || |
1424 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1425 | set_buffer_new(bh_result); |
1da177e4 LT |
1426 | |
1427 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1428 | BUG_ON(direct); | |
1429 | if (create) { | |
1430 | set_buffer_uptodate(bh_result); | |
1431 | set_buffer_mapped(bh_result); | |
1432 | set_buffer_delay(bh_result); | |
1433 | } | |
1434 | } | |
1435 | ||
c2536668 | 1436 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1437 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1438 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1439 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1440 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1441 | } |
1442 | ||
1443 | return 0; | |
1444 | } | |
1445 | ||
1446 | int | |
c2536668 | 1447 | xfs_get_blocks( |
1da177e4 LT |
1448 | struct inode *inode, |
1449 | sector_t iblock, | |
1450 | struct buffer_head *bh_result, | |
1451 | int create) | |
1452 | { | |
c2536668 | 1453 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1454 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1455 | } |
1456 | ||
1457 | STATIC int | |
e4c573bb | 1458 | xfs_get_blocks_direct( |
1da177e4 LT |
1459 | struct inode *inode, |
1460 | sector_t iblock, | |
1da177e4 LT |
1461 | struct buffer_head *bh_result, |
1462 | int create) | |
1463 | { | |
c2536668 | 1464 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1465 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1466 | } |
1467 | ||
f0973863 | 1468 | STATIC void |
e4c573bb | 1469 | xfs_end_io_direct( |
f0973863 CH |
1470 | struct kiocb *iocb, |
1471 | loff_t offset, | |
1472 | ssize_t size, | |
1473 | void *private) | |
1474 | { | |
1475 | xfs_ioend_t *ioend = iocb->private; | |
1476 | ||
1477 | /* | |
1478 | * Non-NULL private data means we need to issue a transaction to | |
1479 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1480 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1481 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1482 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1483 | * it anyway to keep the code uniform and simpler. |
1484 | * | |
e927af90 DC |
1485 | * Well, if only it were that simple. Because synchronous direct I/O |
1486 | * requires extent conversion to occur *before* we return to userspace, | |
1487 | * we have to wait for extent conversion to complete. Look at the | |
1488 | * iocb that has been passed to us to determine if this is AIO or | |
1489 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1490 | * workqueue and wait for it to complete. | |
1491 | * | |
f0973863 CH |
1492 | * The core direct I/O code might be changed to always call the |
1493 | * completion handler in the future, in which case all this can | |
1494 | * go away. | |
1495 | */ | |
ba87ea69 LM |
1496 | ioend->io_offset = offset; |
1497 | ioend->io_size = size; | |
1498 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1499 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1500 | } else if (private && size > 0) { |
e927af90 | 1501 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1502 | } else { |
ba87ea69 LM |
1503 | /* |
1504 | * A direct I/O write ioend starts it's life in unwritten | |
1505 | * state in case they map an unwritten extent. This write | |
1506 | * didn't map an unwritten extent so switch it's completion | |
1507 | * handler. | |
1508 | */ | |
1509 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); | |
e927af90 | 1510 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1511 | } |
1512 | ||
1513 | /* | |
c41564b5 | 1514 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1515 | * completion handler was called. Thus we need to protect |
1516 | * against double-freeing. | |
1517 | */ | |
1518 | iocb->private = NULL; | |
1519 | } | |
1520 | ||
1da177e4 | 1521 | STATIC ssize_t |
e4c573bb | 1522 | xfs_vm_direct_IO( |
1da177e4 LT |
1523 | int rw, |
1524 | struct kiocb *iocb, | |
1525 | const struct iovec *iov, | |
1526 | loff_t offset, | |
1527 | unsigned long nr_segs) | |
1528 | { | |
1529 | struct file *file = iocb->ki_filp; | |
1530 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1531 | struct block_device *bdev; |
f0973863 | 1532 | ssize_t ret; |
1da177e4 | 1533 | |
6214ed44 | 1534 | bdev = xfs_find_bdev_for_inode(XFS_I(inode)); |
1da177e4 | 1535 | |
721259bc | 1536 | if (rw == WRITE) { |
ba87ea69 | 1537 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
721259bc | 1538 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, |
6214ed44 | 1539 | bdev, iov, offset, nr_segs, |
721259bc LM |
1540 | xfs_get_blocks_direct, |
1541 | xfs_end_io_direct); | |
1542 | } else { | |
ba87ea69 | 1543 | iocb->private = xfs_alloc_ioend(inode, IOMAP_READ); |
721259bc | 1544 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
6214ed44 | 1545 | bdev, iov, offset, nr_segs, |
721259bc LM |
1546 | xfs_get_blocks_direct, |
1547 | xfs_end_io_direct); | |
1548 | } | |
f0973863 | 1549 | |
8459d86a | 1550 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1551 | xfs_destroy_ioend(iocb->private); |
1552 | return ret; | |
1da177e4 LT |
1553 | } |
1554 | ||
f51623b2 | 1555 | STATIC int |
d79689c7 | 1556 | xfs_vm_write_begin( |
f51623b2 | 1557 | struct file *file, |
d79689c7 NP |
1558 | struct address_space *mapping, |
1559 | loff_t pos, | |
1560 | unsigned len, | |
1561 | unsigned flags, | |
1562 | struct page **pagep, | |
1563 | void **fsdata) | |
f51623b2 | 1564 | { |
d79689c7 NP |
1565 | *pagep = NULL; |
1566 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1567 | xfs_get_blocks); | |
f51623b2 | 1568 | } |
1da177e4 LT |
1569 | |
1570 | STATIC sector_t | |
e4c573bb | 1571 | xfs_vm_bmap( |
1da177e4 LT |
1572 | struct address_space *mapping, |
1573 | sector_t block) | |
1574 | { | |
1575 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1576 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1577 | |
cf441eeb | 1578 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1579 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1580 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1581 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1582 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1583 | } |
1584 | ||
1585 | STATIC int | |
e4c573bb | 1586 | xfs_vm_readpage( |
1da177e4 LT |
1587 | struct file *unused, |
1588 | struct page *page) | |
1589 | { | |
c2536668 | 1590 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1591 | } |
1592 | ||
1593 | STATIC int | |
e4c573bb | 1594 | xfs_vm_readpages( |
1da177e4 LT |
1595 | struct file *unused, |
1596 | struct address_space *mapping, | |
1597 | struct list_head *pages, | |
1598 | unsigned nr_pages) | |
1599 | { | |
c2536668 | 1600 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1601 | } |
1602 | ||
2ff28e22 | 1603 | STATIC void |
238f4c54 | 1604 | xfs_vm_invalidatepage( |
bcec2b7f NS |
1605 | struct page *page, |
1606 | unsigned long offset) | |
1607 | { | |
1608 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1609 | page->mapping->host, page, offset); | |
2ff28e22 | 1610 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1611 | } |
1612 | ||
f5e54d6e | 1613 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1614 | .readpage = xfs_vm_readpage, |
1615 | .readpages = xfs_vm_readpages, | |
1616 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1617 | .writepages = xfs_vm_writepages, |
1da177e4 | 1618 | .sync_page = block_sync_page, |
238f4c54 NS |
1619 | .releasepage = xfs_vm_releasepage, |
1620 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1621 | .write_begin = xfs_vm_write_begin, |
1622 | .write_end = generic_write_end, | |
e4c573bb NS |
1623 | .bmap = xfs_vm_bmap, |
1624 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1625 | .migratepage = buffer_migrate_page, |
1da177e4 | 1626 | }; |