]>
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" | |
40 | #include <linux/mpage.h> | |
10ce4444 | 41 | #include <linux/pagevec.h> |
1da177e4 LT |
42 | #include <linux/writeback.h> |
43 | ||
f51623b2 NS |
44 | STATIC void |
45 | xfs_count_page_state( | |
46 | struct page *page, | |
47 | int *delalloc, | |
48 | int *unmapped, | |
49 | int *unwritten) | |
50 | { | |
51 | struct buffer_head *bh, *head; | |
52 | ||
53 | *delalloc = *unmapped = *unwritten = 0; | |
54 | ||
55 | bh = head = page_buffers(page); | |
56 | do { | |
57 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
58 | (*unmapped) = 1; | |
f51623b2 NS |
59 | else if (buffer_unwritten(bh)) |
60 | (*unwritten) = 1; | |
61 | else if (buffer_delay(bh)) | |
62 | (*delalloc) = 1; | |
63 | } while ((bh = bh->b_this_page) != head); | |
64 | } | |
65 | ||
1da177e4 LT |
66 | #if defined(XFS_RW_TRACE) |
67 | void | |
68 | xfs_page_trace( | |
69 | int tag, | |
70 | struct inode *inode, | |
71 | struct page *page, | |
ed9d88f7 | 72 | unsigned long pgoff) |
1da177e4 LT |
73 | { |
74 | xfs_inode_t *ip; | |
67fcaa73 | 75 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 76 | loff_t isize = i_size_read(inode); |
f6d6d4fc | 77 | loff_t offset = page_offset(page); |
1da177e4 LT |
78 | int delalloc = -1, unmapped = -1, unwritten = -1; |
79 | ||
80 | if (page_has_buffers(page)) | |
81 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
82 | ||
75e17b3c | 83 | ip = xfs_vtoi(vp); |
1da177e4 LT |
84 | if (!ip->i_rwtrace) |
85 | return; | |
86 | ||
87 | ktrace_enter(ip->i_rwtrace, | |
88 | (void *)((unsigned long)tag), | |
89 | (void *)ip, | |
90 | (void *)inode, | |
91 | (void *)page, | |
ed9d88f7 | 92 | (void *)pgoff, |
1da177e4 LT |
93 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), |
94 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
95 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
96 | (void *)((unsigned long)(isize & 0xffffffff)), | |
97 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
98 | (void *)((unsigned long)(offset & 0xffffffff)), | |
99 | (void *)((unsigned long)delalloc), | |
100 | (void *)((unsigned long)unmapped), | |
101 | (void *)((unsigned long)unwritten), | |
f1fdc848 | 102 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
103 | (void *)NULL); |
104 | } | |
105 | #else | |
ed9d88f7 | 106 | #define xfs_page_trace(tag, inode, page, pgoff) |
1da177e4 LT |
107 | #endif |
108 | ||
0829c360 CH |
109 | /* |
110 | * Schedule IO completion handling on a xfsdatad if this was | |
e927af90 DC |
111 | * the final hold on this ioend. If we are asked to wait, |
112 | * flush the workqueue. | |
0829c360 CH |
113 | */ |
114 | STATIC void | |
115 | xfs_finish_ioend( | |
e927af90 DC |
116 | xfs_ioend_t *ioend, |
117 | int wait) | |
0829c360 | 118 | { |
e927af90 | 119 | if (atomic_dec_and_test(&ioend->io_remaining)) { |
0829c360 | 120 | queue_work(xfsdatad_workqueue, &ioend->io_work); |
e927af90 DC |
121 | if (wait) |
122 | flush_workqueue(xfsdatad_workqueue); | |
123 | } | |
0829c360 CH |
124 | } |
125 | ||
f6d6d4fc CH |
126 | /* |
127 | * We're now finished for good with this ioend structure. | |
128 | * Update the page state via the associated buffer_heads, | |
129 | * release holds on the inode and bio, and finally free | |
130 | * up memory. Do not use the ioend after this. | |
131 | */ | |
0829c360 CH |
132 | STATIC void |
133 | xfs_destroy_ioend( | |
134 | xfs_ioend_t *ioend) | |
135 | { | |
f6d6d4fc CH |
136 | struct buffer_head *bh, *next; |
137 | ||
138 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
139 | next = bh->b_private; | |
7d04a335 | 140 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 141 | } |
7d04a335 NS |
142 | if (unlikely(ioend->io_error)) |
143 | vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__); | |
0829c360 CH |
144 | vn_iowake(ioend->io_vnode); |
145 | mempool_free(ioend, xfs_ioend_pool); | |
146 | } | |
147 | ||
ba87ea69 LM |
148 | /* |
149 | * Update on-disk file size now that data has been written to disk. | |
150 | * The current in-memory file size is i_size. If a write is beyond | |
151 | * eof io_new_size will be the intended file size until i_size is | |
152 | * updated. If this write does not extend all the way to the valid | |
153 | * file size then restrict this update to the end of the write. | |
154 | */ | |
155 | STATIC void | |
156 | xfs_setfilesize( | |
157 | xfs_ioend_t *ioend) | |
158 | { | |
159 | xfs_inode_t *ip; | |
160 | xfs_fsize_t isize; | |
161 | xfs_fsize_t bsize; | |
162 | ||
163 | ip = xfs_vtoi(ioend->io_vnode); | |
b2826136 DC |
164 | if (!ip) |
165 | return; | |
ba87ea69 LM |
166 | |
167 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); | |
168 | ASSERT(ioend->io_type != IOMAP_READ); | |
169 | ||
170 | if (unlikely(ioend->io_error)) | |
171 | return; | |
172 | ||
173 | bsize = ioend->io_offset + ioend->io_size; | |
174 | ||
175 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
176 | ||
177 | isize = MAX(ip->i_size, ip->i_iocore.io_new_size); | |
178 | isize = MIN(isize, bsize); | |
179 | ||
180 | if (ip->i_d.di_size < isize) { | |
181 | ip->i_d.di_size = isize; | |
182 | ip->i_update_core = 1; | |
183 | ip->i_update_size = 1; | |
184 | } | |
185 | ||
186 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
187 | } | |
188 | ||
0829c360 | 189 | /* |
f6d6d4fc | 190 | * Buffered IO write completion for delayed allocate extents. |
f6d6d4fc CH |
191 | */ |
192 | STATIC void | |
193 | xfs_end_bio_delalloc( | |
c4028958 | 194 | struct work_struct *work) |
f6d6d4fc | 195 | { |
c4028958 DH |
196 | xfs_ioend_t *ioend = |
197 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 198 | |
ba87ea69 | 199 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
200 | xfs_destroy_ioend(ioend); |
201 | } | |
202 | ||
203 | /* | |
204 | * Buffered IO write completion for regular, written extents. | |
205 | */ | |
206 | STATIC void | |
207 | xfs_end_bio_written( | |
c4028958 | 208 | struct work_struct *work) |
f6d6d4fc | 209 | { |
c4028958 DH |
210 | xfs_ioend_t *ioend = |
211 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 212 | |
ba87ea69 | 213 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
214 | xfs_destroy_ioend(ioend); |
215 | } | |
216 | ||
217 | /* | |
218 | * IO write completion for unwritten extents. | |
219 | * | |
0829c360 | 220 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 221 | * to written extents. |
0829c360 CH |
222 | */ |
223 | STATIC void | |
224 | xfs_end_bio_unwritten( | |
c4028958 | 225 | struct work_struct *work) |
0829c360 | 226 | { |
c4028958 DH |
227 | xfs_ioend_t *ioend = |
228 | container_of(work, xfs_ioend_t, io_work); | |
67fcaa73 | 229 | bhv_vnode_t *vp = ioend->io_vnode; |
0829c360 CH |
230 | xfs_off_t offset = ioend->io_offset; |
231 | size_t size = ioend->io_size; | |
0829c360 | 232 | |
ba87ea69 | 233 | if (likely(!ioend->io_error)) { |
67fcaa73 | 234 | bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL); |
ba87ea69 LM |
235 | xfs_setfilesize(ioend); |
236 | } | |
237 | xfs_destroy_ioend(ioend); | |
238 | } | |
239 | ||
240 | /* | |
241 | * IO read completion for regular, written extents. | |
242 | */ | |
243 | STATIC void | |
244 | xfs_end_bio_read( | |
245 | struct work_struct *work) | |
246 | { | |
247 | xfs_ioend_t *ioend = | |
248 | container_of(work, xfs_ioend_t, io_work); | |
249 | ||
0829c360 CH |
250 | xfs_destroy_ioend(ioend); |
251 | } | |
252 | ||
253 | /* | |
254 | * Allocate and initialise an IO completion structure. | |
255 | * We need to track unwritten extent write completion here initially. | |
256 | * We'll need to extend this for updating the ondisk inode size later | |
257 | * (vs. incore size). | |
258 | */ | |
259 | STATIC xfs_ioend_t * | |
260 | xfs_alloc_ioend( | |
f6d6d4fc CH |
261 | struct inode *inode, |
262 | unsigned int type) | |
0829c360 CH |
263 | { |
264 | xfs_ioend_t *ioend; | |
265 | ||
266 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
267 | ||
268 | /* | |
269 | * Set the count to 1 initially, which will prevent an I/O | |
270 | * completion callback from happening before we have started | |
271 | * all the I/O from calling the completion routine too early. | |
272 | */ | |
273 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 274 | ioend->io_error = 0; |
f6d6d4fc CH |
275 | ioend->io_list = NULL; |
276 | ioend->io_type = type; | |
ec86dc02 | 277 | ioend->io_vnode = vn_from_inode(inode); |
c1a073bd | 278 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 279 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
280 | atomic_inc(&ioend->io_vnode->v_iocount); |
281 | ioend->io_offset = 0; | |
282 | ioend->io_size = 0; | |
283 | ||
f6d6d4fc | 284 | if (type == IOMAP_UNWRITTEN) |
c4028958 | 285 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten); |
f6d6d4fc | 286 | else if (type == IOMAP_DELAY) |
c4028958 | 287 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc); |
ba87ea69 LM |
288 | else if (type == IOMAP_READ) |
289 | INIT_WORK(&ioend->io_work, xfs_end_bio_read); | |
f6d6d4fc | 290 | else |
c4028958 | 291 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); |
0829c360 CH |
292 | |
293 | return ioend; | |
294 | } | |
295 | ||
1da177e4 LT |
296 | STATIC int |
297 | xfs_map_blocks( | |
298 | struct inode *inode, | |
299 | loff_t offset, | |
300 | ssize_t count, | |
301 | xfs_iomap_t *mapp, | |
302 | int flags) | |
303 | { | |
67fcaa73 | 304 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
305 | int error, nmaps = 1; |
306 | ||
67fcaa73 | 307 | error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps); |
1da177e4 LT |
308 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) |
309 | VMODIFY(vp); | |
310 | return -error; | |
311 | } | |
312 | ||
7989cb8e | 313 | STATIC_INLINE int |
1defeac9 | 314 | xfs_iomap_valid( |
1da177e4 | 315 | xfs_iomap_t *iomapp, |
1defeac9 | 316 | loff_t offset) |
1da177e4 | 317 | { |
1defeac9 CH |
318 | return offset >= iomapp->iomap_offset && |
319 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
320 | } |
321 | ||
f6d6d4fc CH |
322 | /* |
323 | * BIO completion handler for buffered IO. | |
324 | */ | |
325 | STATIC int | |
326 | xfs_end_bio( | |
327 | struct bio *bio, | |
328 | unsigned int bytes_done, | |
329 | int error) | |
330 | { | |
331 | xfs_ioend_t *ioend = bio->bi_private; | |
332 | ||
333 | if (bio->bi_size) | |
334 | return 1; | |
335 | ||
f6d6d4fc | 336 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 337 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
338 | |
339 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
340 | bio->bi_private = NULL; |
341 | bio->bi_end_io = NULL; | |
f6d6d4fc | 342 | bio_put(bio); |
7d04a335 | 343 | |
e927af90 | 344 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
345 | return 0; |
346 | } | |
347 | ||
348 | STATIC void | |
349 | xfs_submit_ioend_bio( | |
350 | xfs_ioend_t *ioend, | |
351 | struct bio *bio) | |
352 | { | |
353 | atomic_inc(&ioend->io_remaining); | |
354 | ||
355 | bio->bi_private = ioend; | |
356 | bio->bi_end_io = xfs_end_bio; | |
357 | ||
358 | submit_bio(WRITE, bio); | |
359 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
360 | bio_put(bio); | |
361 | } | |
362 | ||
363 | STATIC struct bio * | |
364 | xfs_alloc_ioend_bio( | |
365 | struct buffer_head *bh) | |
366 | { | |
367 | struct bio *bio; | |
368 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
369 | ||
370 | do { | |
371 | bio = bio_alloc(GFP_NOIO, nvecs); | |
372 | nvecs >>= 1; | |
373 | } while (!bio); | |
374 | ||
375 | ASSERT(bio->bi_private == NULL); | |
376 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
377 | bio->bi_bdev = bh->b_bdev; | |
378 | bio_get(bio); | |
379 | return bio; | |
380 | } | |
381 | ||
382 | STATIC void | |
383 | xfs_start_buffer_writeback( | |
384 | struct buffer_head *bh) | |
385 | { | |
386 | ASSERT(buffer_mapped(bh)); | |
387 | ASSERT(buffer_locked(bh)); | |
388 | ASSERT(!buffer_delay(bh)); | |
389 | ASSERT(!buffer_unwritten(bh)); | |
390 | ||
391 | mark_buffer_async_write(bh); | |
392 | set_buffer_uptodate(bh); | |
393 | clear_buffer_dirty(bh); | |
394 | } | |
395 | ||
396 | STATIC void | |
397 | xfs_start_page_writeback( | |
398 | struct page *page, | |
399 | struct writeback_control *wbc, | |
400 | int clear_dirty, | |
401 | int buffers) | |
402 | { | |
403 | ASSERT(PageLocked(page)); | |
404 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 405 | if (clear_dirty) |
92132021 DC |
406 | clear_page_dirty_for_io(page); |
407 | set_page_writeback(page); | |
f6d6d4fc CH |
408 | unlock_page(page); |
409 | if (!buffers) { | |
410 | end_page_writeback(page); | |
411 | wbc->pages_skipped++; /* We didn't write this page */ | |
412 | } | |
413 | } | |
414 | ||
415 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
416 | { | |
417 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
418 | } | |
419 | ||
420 | /* | |
d88992f6 DC |
421 | * Submit all of the bios for all of the ioends we have saved up, covering the |
422 | * initial writepage page and also any probed pages. | |
423 | * | |
424 | * Because we may have multiple ioends spanning a page, we need to start | |
425 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
426 | * buffers as we got, then we can end up with a page that only has buffers | |
427 | * marked async write and I/O complete on can occur before we mark the other | |
428 | * buffers async write. | |
429 | * | |
430 | * The end result of this is that we trip a bug in end_page_writeback() because | |
431 | * we call it twice for the one page as the code in end_buffer_async_write() | |
432 | * assumes that all buffers on the page are started at the same time. | |
433 | * | |
434 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 435 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
436 | */ |
437 | STATIC void | |
438 | xfs_submit_ioend( | |
439 | xfs_ioend_t *ioend) | |
440 | { | |
d88992f6 | 441 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
442 | xfs_ioend_t *next; |
443 | struct buffer_head *bh; | |
444 | struct bio *bio; | |
445 | sector_t lastblock = 0; | |
446 | ||
d88992f6 DC |
447 | /* Pass 1 - start writeback */ |
448 | do { | |
449 | next = ioend->io_list; | |
450 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
451 | xfs_start_buffer_writeback(bh); | |
452 | } | |
453 | } while ((ioend = next) != NULL); | |
454 | ||
455 | /* Pass 2 - submit I/O */ | |
456 | ioend = head; | |
f6d6d4fc CH |
457 | do { |
458 | next = ioend->io_list; | |
459 | bio = NULL; | |
460 | ||
461 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
462 | |
463 | if (!bio) { | |
464 | retry: | |
465 | bio = xfs_alloc_ioend_bio(bh); | |
466 | } else if (bh->b_blocknr != lastblock + 1) { | |
467 | xfs_submit_ioend_bio(ioend, bio); | |
468 | goto retry; | |
469 | } | |
470 | ||
471 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
472 | xfs_submit_ioend_bio(ioend, bio); | |
473 | goto retry; | |
474 | } | |
475 | ||
476 | lastblock = bh->b_blocknr; | |
477 | } | |
478 | if (bio) | |
479 | xfs_submit_ioend_bio(ioend, bio); | |
e927af90 | 480 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
481 | } while ((ioend = next) != NULL); |
482 | } | |
483 | ||
484 | /* | |
485 | * Cancel submission of all buffer_heads so far in this endio. | |
486 | * Toss the endio too. Only ever called for the initial page | |
487 | * in a writepage request, so only ever one page. | |
488 | */ | |
489 | STATIC void | |
490 | xfs_cancel_ioend( | |
491 | xfs_ioend_t *ioend) | |
492 | { | |
493 | xfs_ioend_t *next; | |
494 | struct buffer_head *bh, *next_bh; | |
495 | ||
496 | do { | |
497 | next = ioend->io_list; | |
498 | bh = ioend->io_buffer_head; | |
499 | do { | |
500 | next_bh = bh->b_private; | |
501 | clear_buffer_async_write(bh); | |
502 | unlock_buffer(bh); | |
503 | } while ((bh = next_bh) != NULL); | |
504 | ||
505 | vn_iowake(ioend->io_vnode); | |
506 | mempool_free(ioend, xfs_ioend_pool); | |
507 | } while ((ioend = next) != NULL); | |
508 | } | |
509 | ||
510 | /* | |
511 | * Test to see if we've been building up a completion structure for | |
512 | * earlier buffers -- if so, we try to append to this ioend if we | |
513 | * can, otherwise we finish off any current ioend and start another. | |
514 | * Return true if we've finished the given ioend. | |
515 | */ | |
516 | STATIC void | |
517 | xfs_add_to_ioend( | |
518 | struct inode *inode, | |
519 | struct buffer_head *bh, | |
7336cea8 | 520 | xfs_off_t offset, |
f6d6d4fc CH |
521 | unsigned int type, |
522 | xfs_ioend_t **result, | |
523 | int need_ioend) | |
524 | { | |
525 | xfs_ioend_t *ioend = *result; | |
526 | ||
527 | if (!ioend || need_ioend || type != ioend->io_type) { | |
528 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 529 | |
f6d6d4fc CH |
530 | ioend = xfs_alloc_ioend(inode, type); |
531 | ioend->io_offset = offset; | |
532 | ioend->io_buffer_head = bh; | |
533 | ioend->io_buffer_tail = bh; | |
534 | if (previous) | |
535 | previous->io_list = ioend; | |
536 | *result = ioend; | |
537 | } else { | |
538 | ioend->io_buffer_tail->b_private = bh; | |
539 | ioend->io_buffer_tail = bh; | |
540 | } | |
541 | ||
542 | bh->b_private = NULL; | |
543 | ioend->io_size += bh->b_size; | |
544 | } | |
545 | ||
87cbc49c NS |
546 | STATIC void |
547 | xfs_map_buffer( | |
548 | struct buffer_head *bh, | |
549 | xfs_iomap_t *mp, | |
550 | xfs_off_t offset, | |
551 | uint block_bits) | |
552 | { | |
553 | sector_t bn; | |
554 | ||
555 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
556 | ||
557 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
558 | ((offset - mp->iomap_offset) >> block_bits); | |
559 | ||
560 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
561 | ||
562 | bh->b_blocknr = bn; | |
563 | set_buffer_mapped(bh); | |
564 | } | |
565 | ||
1da177e4 LT |
566 | STATIC void |
567 | xfs_map_at_offset( | |
1da177e4 | 568 | struct buffer_head *bh, |
1defeac9 | 569 | loff_t offset, |
1da177e4 | 570 | int block_bits, |
1defeac9 | 571 | xfs_iomap_t *iomapp) |
1da177e4 | 572 | { |
1da177e4 LT |
573 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
574 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
575 | |
576 | lock_buffer(bh); | |
87cbc49c | 577 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 578 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
579 | set_buffer_mapped(bh); |
580 | clear_buffer_delay(bh); | |
f6d6d4fc | 581 | clear_buffer_unwritten(bh); |
1da177e4 LT |
582 | } |
583 | ||
584 | /* | |
6c4fe19f | 585 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
586 | */ |
587 | STATIC unsigned int | |
6c4fe19f | 588 | xfs_probe_page( |
10ce4444 | 589 | struct page *page, |
6c4fe19f CH |
590 | unsigned int pg_offset, |
591 | int mapped) | |
1da177e4 | 592 | { |
1da177e4 LT |
593 | int ret = 0; |
594 | ||
1da177e4 | 595 | if (PageWriteback(page)) |
10ce4444 | 596 | return 0; |
1da177e4 LT |
597 | |
598 | if (page->mapping && PageDirty(page)) { | |
599 | if (page_has_buffers(page)) { | |
600 | struct buffer_head *bh, *head; | |
601 | ||
602 | bh = head = page_buffers(page); | |
603 | do { | |
6c4fe19f CH |
604 | if (!buffer_uptodate(bh)) |
605 | break; | |
606 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
607 | break; |
608 | ret += bh->b_size; | |
609 | if (ret >= pg_offset) | |
610 | break; | |
611 | } while ((bh = bh->b_this_page) != head); | |
612 | } else | |
6c4fe19f | 613 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
614 | } |
615 | ||
1da177e4 LT |
616 | return ret; |
617 | } | |
618 | ||
f6d6d4fc | 619 | STATIC size_t |
6c4fe19f | 620 | xfs_probe_cluster( |
1da177e4 LT |
621 | struct inode *inode, |
622 | struct page *startpage, | |
623 | struct buffer_head *bh, | |
6c4fe19f CH |
624 | struct buffer_head *head, |
625 | int mapped) | |
1da177e4 | 626 | { |
10ce4444 | 627 | struct pagevec pvec; |
1da177e4 | 628 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
629 | size_t total = 0; |
630 | int done = 0, i; | |
1da177e4 LT |
631 | |
632 | /* First sum forwards in this page */ | |
633 | do { | |
2353e8e9 | 634 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 635 | return total; |
1da177e4 LT |
636 | total += bh->b_size; |
637 | } while ((bh = bh->b_this_page) != head); | |
638 | ||
10ce4444 CH |
639 | /* if we reached the end of the page, sum forwards in following pages */ |
640 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
641 | tindex = startpage->index + 1; | |
642 | ||
643 | /* Prune this back to avoid pathological behavior */ | |
644 | tloff = min(tlast, startpage->index + 64); | |
645 | ||
646 | pagevec_init(&pvec, 0); | |
647 | while (!done && tindex <= tloff) { | |
648 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
649 | ||
650 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
651 | break; | |
652 | ||
653 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
654 | struct page *page = pvec.pages[i]; | |
265c1fac | 655 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
656 | |
657 | if (tindex == tlast) { | |
658 | pg_offset = | |
659 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
660 | if (!pg_offset) { |
661 | done = 1; | |
10ce4444 | 662 | break; |
1defeac9 | 663 | } |
10ce4444 CH |
664 | } else |
665 | pg_offset = PAGE_CACHE_SIZE; | |
666 | ||
667 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
265c1fac | 668 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
669 | unlock_page(page); |
670 | } | |
671 | ||
265c1fac | 672 | if (!pg_len) { |
10ce4444 CH |
673 | done = 1; |
674 | break; | |
675 | } | |
676 | ||
265c1fac | 677 | total += pg_len; |
1defeac9 | 678 | tindex++; |
1da177e4 | 679 | } |
10ce4444 CH |
680 | |
681 | pagevec_release(&pvec); | |
682 | cond_resched(); | |
1da177e4 | 683 | } |
10ce4444 | 684 | |
1da177e4 LT |
685 | return total; |
686 | } | |
687 | ||
688 | /* | |
10ce4444 CH |
689 | * Test if a given page is suitable for writing as part of an unwritten |
690 | * or delayed allocate extent. | |
1da177e4 | 691 | */ |
10ce4444 CH |
692 | STATIC int |
693 | xfs_is_delayed_page( | |
694 | struct page *page, | |
f6d6d4fc | 695 | unsigned int type) |
1da177e4 | 696 | { |
1da177e4 | 697 | if (PageWriteback(page)) |
10ce4444 | 698 | return 0; |
1da177e4 LT |
699 | |
700 | if (page->mapping && page_has_buffers(page)) { | |
701 | struct buffer_head *bh, *head; | |
702 | int acceptable = 0; | |
703 | ||
704 | bh = head = page_buffers(page); | |
705 | do { | |
f6d6d4fc CH |
706 | if (buffer_unwritten(bh)) |
707 | acceptable = (type == IOMAP_UNWRITTEN); | |
708 | else if (buffer_delay(bh)) | |
709 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 710 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 711 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 712 | else |
1da177e4 | 713 | break; |
1da177e4 LT |
714 | } while ((bh = bh->b_this_page) != head); |
715 | ||
716 | if (acceptable) | |
10ce4444 | 717 | return 1; |
1da177e4 LT |
718 | } |
719 | ||
10ce4444 | 720 | return 0; |
1da177e4 LT |
721 | } |
722 | ||
1da177e4 LT |
723 | /* |
724 | * Allocate & map buffers for page given the extent map. Write it out. | |
725 | * except for the original page of a writepage, this is called on | |
726 | * delalloc/unwritten pages only, for the original page it is possible | |
727 | * that the page has no mapping at all. | |
728 | */ | |
f6d6d4fc | 729 | STATIC int |
1da177e4 LT |
730 | xfs_convert_page( |
731 | struct inode *inode, | |
732 | struct page *page, | |
10ce4444 | 733 | loff_t tindex, |
1defeac9 | 734 | xfs_iomap_t *mp, |
f6d6d4fc | 735 | xfs_ioend_t **ioendp, |
1da177e4 | 736 | struct writeback_control *wbc, |
1da177e4 LT |
737 | int startio, |
738 | int all_bh) | |
739 | { | |
f6d6d4fc | 740 | struct buffer_head *bh, *head; |
9260dc6b CH |
741 | xfs_off_t end_offset; |
742 | unsigned long p_offset; | |
f6d6d4fc | 743 | unsigned int type; |
1da177e4 | 744 | int bbits = inode->i_blkbits; |
24e17b5f | 745 | int len, page_dirty; |
f6d6d4fc | 746 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 747 | xfs_off_t offset = page_offset(page); |
1da177e4 | 748 | |
10ce4444 CH |
749 | if (page->index != tindex) |
750 | goto fail; | |
751 | if (TestSetPageLocked(page)) | |
752 | goto fail; | |
753 | if (PageWriteback(page)) | |
754 | goto fail_unlock_page; | |
755 | if (page->mapping != inode->i_mapping) | |
756 | goto fail_unlock_page; | |
757 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
758 | goto fail_unlock_page; | |
759 | ||
24e17b5f NS |
760 | /* |
761 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 762 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
763 | * |
764 | * Derivation: | |
765 | * | |
766 | * End offset is the highest offset that this page should represent. | |
767 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
768 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
769 | * hence give us the correct page_dirty count. On any other page, | |
770 | * it will be zero and in that case we need page_dirty to be the | |
771 | * count of buffers on the page. | |
24e17b5f | 772 | */ |
9260dc6b CH |
773 | end_offset = min_t(unsigned long long, |
774 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
775 | i_size_read(inode)); | |
776 | ||
24e17b5f | 777 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
778 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
779 | PAGE_CACHE_SIZE); | |
780 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
781 | page_dirty = p_offset / len; | |
24e17b5f | 782 | |
1da177e4 LT |
783 | bh = head = page_buffers(page); |
784 | do { | |
9260dc6b | 785 | if (offset >= end_offset) |
1da177e4 | 786 | break; |
f6d6d4fc CH |
787 | if (!buffer_uptodate(bh)) |
788 | uptodate = 0; | |
789 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
790 | done = 1; | |
1da177e4 | 791 | continue; |
f6d6d4fc CH |
792 | } |
793 | ||
9260dc6b CH |
794 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
795 | if (buffer_unwritten(bh)) | |
796 | type = IOMAP_UNWRITTEN; | |
797 | else | |
798 | type = IOMAP_DELAY; | |
799 | ||
800 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 801 | done = 1; |
9260dc6b CH |
802 | continue; |
803 | } | |
804 | ||
805 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
806 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
807 | ||
808 | xfs_map_at_offset(bh, offset, bbits, mp); | |
809 | if (startio) { | |
7336cea8 | 810 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
811 | type, ioendp, done); |
812 | } else { | |
813 | set_buffer_dirty(bh); | |
814 | unlock_buffer(bh); | |
815 | mark_buffer_dirty(bh); | |
816 | } | |
817 | page_dirty--; | |
818 | count++; | |
819 | } else { | |
df3c7244 | 820 | type = IOMAP_NEW; |
9260dc6b | 821 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 822 | lock_buffer(bh); |
7336cea8 | 823 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
824 | type, ioendp, done); |
825 | count++; | |
24e17b5f | 826 | page_dirty--; |
9260dc6b CH |
827 | } else { |
828 | done = 1; | |
1da177e4 | 829 | } |
1da177e4 | 830 | } |
7336cea8 | 831 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 832 | |
f6d6d4fc CH |
833 | if (uptodate && bh == head) |
834 | SetPageUptodate(page); | |
835 | ||
836 | if (startio) { | |
f5e596bb CH |
837 | if (count) { |
838 | struct backing_dev_info *bdi; | |
839 | ||
840 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 841 | wbc->nr_to_write--; |
f5e596bb CH |
842 | if (bdi_write_congested(bdi)) { |
843 | wbc->encountered_congestion = 1; | |
844 | done = 1; | |
9fddaca2 | 845 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
846 | done = 1; |
847 | } | |
848 | } | |
f6d6d4fc | 849 | xfs_start_page_writeback(page, wbc, !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; | |
d5cb48aa CH |
1096 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
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 CH |
1120 | if (startio) |
1121 | xfs_start_page_writeback(page, wbc, 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 CH |
1131 | if (iohead) |
1132 | xfs_submit_ioend(iohead); | |
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 | ||
1183 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
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 | ||
1219 | /* | |
1220 | * Convert delayed allocate, unwritten or unmapped space | |
1221 | * to real space and flush out to disk. | |
1222 | */ | |
1223 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1224 | if (error == -EAGAIN) | |
1225 | goto out_fail; | |
1226 | if (unlikely(error < 0)) | |
1227 | goto out_unlock; | |
1228 | ||
1229 | return 0; | |
1230 | ||
1231 | out_fail: | |
1232 | redirty_page_for_writepage(wbc, page); | |
1233 | unlock_page(page); | |
1234 | return 0; | |
1235 | out_unlock: | |
1236 | unlock_page(page); | |
1237 | return error; | |
1238 | } | |
1239 | ||
7d4fb40a NS |
1240 | STATIC int |
1241 | xfs_vm_writepages( | |
1242 | struct address_space *mapping, | |
1243 | struct writeback_control *wbc) | |
1244 | { | |
67fcaa73 | 1245 | struct bhv_vnode *vp = vn_from_inode(mapping->host); |
7d4fb40a NS |
1246 | |
1247 | if (VN_TRUNC(vp)) | |
1248 | VUNTRUNCATE(vp); | |
1249 | return generic_writepages(mapping, wbc); | |
1250 | } | |
1251 | ||
f51623b2 NS |
1252 | /* |
1253 | * Called to move a page into cleanable state - and from there | |
1254 | * to be released. Possibly the page is already clean. We always | |
1255 | * have buffer heads in this call. | |
1256 | * | |
1257 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1258 | * | |
1259 | * Possible scenarios are: | |
1260 | * | |
1261 | * 1. We are being called to release a page which has been written | |
1262 | * to via regular I/O. buffer heads will be dirty and possibly | |
1263 | * delalloc. If no delalloc buffer heads in this case then we | |
1264 | * can just return zero. | |
1265 | * | |
1266 | * 2. We are called to release a page which has been written via | |
1267 | * mmap, all we need to do is ensure there is no delalloc | |
1268 | * state in the buffer heads, if not we can let the caller | |
1269 | * free them and we should come back later via writepage. | |
1270 | */ | |
1271 | STATIC int | |
238f4c54 | 1272 | xfs_vm_releasepage( |
f51623b2 NS |
1273 | struct page *page, |
1274 | gfp_t gfp_mask) | |
1275 | { | |
1276 | struct inode *inode = page->mapping->host; | |
1277 | int dirty, delalloc, unmapped, unwritten; | |
1278 | struct writeback_control wbc = { | |
1279 | .sync_mode = WB_SYNC_ALL, | |
1280 | .nr_to_write = 1, | |
1281 | }; | |
1282 | ||
ed9d88f7 | 1283 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0); |
f51623b2 | 1284 | |
238f4c54 NS |
1285 | if (!page_has_buffers(page)) |
1286 | return 0; | |
1287 | ||
f51623b2 NS |
1288 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1289 | if (!delalloc && !unwritten) | |
1290 | goto free_buffers; | |
1291 | ||
1292 | if (!(gfp_mask & __GFP_FS)) | |
1293 | return 0; | |
1294 | ||
1295 | /* If we are already inside a transaction or the thread cannot | |
1296 | * do I/O, we cannot release this page. | |
1297 | */ | |
59c1b082 | 1298 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1299 | return 0; |
1300 | ||
1301 | /* | |
1302 | * Convert delalloc space to real space, do not flush the | |
1303 | * data out to disk, that will be done by the caller. | |
1304 | * Never need to allocate space here - we will always | |
1305 | * come back to writepage in that case. | |
1306 | */ | |
1307 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1308 | if (dirty == 0 && !unwritten) | |
1309 | goto free_buffers; | |
1310 | return 0; | |
1311 | ||
1312 | free_buffers: | |
1313 | return try_to_free_buffers(page); | |
1314 | } | |
1315 | ||
1da177e4 | 1316 | STATIC int |
c2536668 | 1317 | __xfs_get_blocks( |
1da177e4 LT |
1318 | struct inode *inode, |
1319 | sector_t iblock, | |
1da177e4 LT |
1320 | struct buffer_head *bh_result, |
1321 | int create, | |
1322 | int direct, | |
1323 | bmapi_flags_t flags) | |
1324 | { | |
67fcaa73 | 1325 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 1326 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1327 | xfs_off_t offset; |
1328 | ssize_t size; | |
c2536668 | 1329 | int niomap = 1; |
1da177e4 | 1330 | int error; |
1da177e4 | 1331 | |
fdc7ed75 | 1332 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1333 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1334 | size = bh_result->b_size; | |
67fcaa73 NS |
1335 | error = bhv_vop_bmap(vp, offset, size, |
1336 | create ? flags : BMAPI_READ, &iomap, &niomap); | |
1da177e4 LT |
1337 | if (error) |
1338 | return -error; | |
c2536668 | 1339 | if (niomap == 0) |
1da177e4 LT |
1340 | return 0; |
1341 | ||
1342 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1343 | /* |
1344 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1345 | * the read case (treat as if we're reading into a hole). |
1346 | */ | |
1347 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1348 | xfs_map_buffer(bh_result, &iomap, offset, |
1349 | inode->i_blkbits); | |
1da177e4 LT |
1350 | } |
1351 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1352 | if (direct) | |
1353 | bh_result->b_private = inode; | |
1354 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1355 | } |
1356 | } | |
1357 | ||
c2536668 NS |
1358 | /* |
1359 | * If this is a realtime file, data may be on a different device. | |
1360 | * to that pointed to from the buffer_head b_bdev currently. | |
1361 | */ | |
ce8e922c | 1362 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1363 | |
c2536668 | 1364 | /* |
549054af DC |
1365 | * If we previously allocated a block out beyond eof and we are now |
1366 | * coming back to use it then we will need to flag it as new even if it | |
1367 | * has a disk address. | |
1368 | * | |
1369 | * With sub-block writes into unwritten extents we also need to mark | |
1370 | * the buffer as new so that the unwritten parts of the buffer gets | |
1371 | * correctly zeroed. | |
1da177e4 LT |
1372 | */ |
1373 | if (create && | |
1374 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1375 | (offset >= i_size_read(inode)) || |
1376 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1377 | set_buffer_new(bh_result); |
1da177e4 LT |
1378 | |
1379 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1380 | BUG_ON(direct); | |
1381 | if (create) { | |
1382 | set_buffer_uptodate(bh_result); | |
1383 | set_buffer_mapped(bh_result); | |
1384 | set_buffer_delay(bh_result); | |
1385 | } | |
1386 | } | |
1387 | ||
c2536668 | 1388 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1389 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1390 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1391 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1392 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1393 | } |
1394 | ||
1395 | return 0; | |
1396 | } | |
1397 | ||
1398 | int | |
c2536668 | 1399 | xfs_get_blocks( |
1da177e4 LT |
1400 | struct inode *inode, |
1401 | sector_t iblock, | |
1402 | struct buffer_head *bh_result, | |
1403 | int create) | |
1404 | { | |
c2536668 | 1405 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1406 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | STATIC int | |
e4c573bb | 1410 | xfs_get_blocks_direct( |
1da177e4 LT |
1411 | struct inode *inode, |
1412 | sector_t iblock, | |
1da177e4 LT |
1413 | struct buffer_head *bh_result, |
1414 | int create) | |
1415 | { | |
c2536668 | 1416 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1417 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1418 | } |
1419 | ||
f0973863 | 1420 | STATIC void |
e4c573bb | 1421 | xfs_end_io_direct( |
f0973863 CH |
1422 | struct kiocb *iocb, |
1423 | loff_t offset, | |
1424 | ssize_t size, | |
1425 | void *private) | |
1426 | { | |
1427 | xfs_ioend_t *ioend = iocb->private; | |
1428 | ||
1429 | /* | |
1430 | * Non-NULL private data means we need to issue a transaction to | |
1431 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1432 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1433 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1434 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1435 | * it anyway to keep the code uniform and simpler. |
1436 | * | |
e927af90 DC |
1437 | * Well, if only it were that simple. Because synchronous direct I/O |
1438 | * requires extent conversion to occur *before* we return to userspace, | |
1439 | * we have to wait for extent conversion to complete. Look at the | |
1440 | * iocb that has been passed to us to determine if this is AIO or | |
1441 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1442 | * workqueue and wait for it to complete. | |
1443 | * | |
f0973863 CH |
1444 | * The core direct I/O code might be changed to always call the |
1445 | * completion handler in the future, in which case all this can | |
1446 | * go away. | |
1447 | */ | |
ba87ea69 LM |
1448 | ioend->io_offset = offset; |
1449 | ioend->io_size = size; | |
1450 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1451 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1452 | } else if (private && size > 0) { |
e927af90 | 1453 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1454 | } else { |
ba87ea69 LM |
1455 | /* |
1456 | * A direct I/O write ioend starts it's life in unwritten | |
1457 | * state in case they map an unwritten extent. This write | |
1458 | * didn't map an unwritten extent so switch it's completion | |
1459 | * handler. | |
1460 | */ | |
1461 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); | |
e927af90 | 1462 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1463 | } |
1464 | ||
1465 | /* | |
c41564b5 | 1466 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1467 | * completion handler was called. Thus we need to protect |
1468 | * against double-freeing. | |
1469 | */ | |
1470 | iocb->private = NULL; | |
1471 | } | |
1472 | ||
1da177e4 | 1473 | STATIC ssize_t |
e4c573bb | 1474 | xfs_vm_direct_IO( |
1da177e4 LT |
1475 | int rw, |
1476 | struct kiocb *iocb, | |
1477 | const struct iovec *iov, | |
1478 | loff_t offset, | |
1479 | unsigned long nr_segs) | |
1480 | { | |
1481 | struct file *file = iocb->ki_filp; | |
1482 | struct inode *inode = file->f_mapping->host; | |
67fcaa73 | 1483 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
1484 | xfs_iomap_t iomap; |
1485 | int maps = 1; | |
1486 | int error; | |
f0973863 | 1487 | ssize_t ret; |
1da177e4 | 1488 | |
67fcaa73 | 1489 | error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps); |
1da177e4 LT |
1490 | if (error) |
1491 | return -error; | |
1492 | ||
721259bc | 1493 | if (rw == WRITE) { |
ba87ea69 | 1494 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
721259bc LM |
1495 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, |
1496 | iomap.iomap_target->bt_bdev, | |
1497 | iov, offset, nr_segs, | |
1498 | xfs_get_blocks_direct, | |
1499 | xfs_end_io_direct); | |
1500 | } else { | |
ba87ea69 | 1501 | iocb->private = xfs_alloc_ioend(inode, IOMAP_READ); |
721259bc LM |
1502 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
1503 | iomap.iomap_target->bt_bdev, | |
1504 | iov, offset, nr_segs, | |
1505 | xfs_get_blocks_direct, | |
1506 | xfs_end_io_direct); | |
1507 | } | |
f0973863 | 1508 | |
8459d86a | 1509 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1510 | xfs_destroy_ioend(iocb->private); |
1511 | return ret; | |
1da177e4 LT |
1512 | } |
1513 | ||
f51623b2 | 1514 | STATIC int |
e4c573bb | 1515 | xfs_vm_prepare_write( |
f51623b2 NS |
1516 | struct file *file, |
1517 | struct page *page, | |
1518 | unsigned int from, | |
1519 | unsigned int to) | |
1520 | { | |
c2536668 | 1521 | return block_prepare_write(page, from, to, 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; | |
67fcaa73 | 1530 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 1531 | |
e4c573bb | 1532 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
67fcaa73 NS |
1533 | bhv_vop_rwlock(vp, VRWLOCK_READ); |
1534 | bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF); | |
1535 | bhv_vop_rwunlock(vp, VRWLOCK_READ); | |
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 | { | |
1562 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1563 | page->mapping->host, page, offset); | |
2ff28e22 | 1564 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1565 | } |
1566 | ||
f5e54d6e | 1567 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1568 | .readpage = xfs_vm_readpage, |
1569 | .readpages = xfs_vm_readpages, | |
1570 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1571 | .writepages = xfs_vm_writepages, |
1da177e4 | 1572 | .sync_page = block_sync_page, |
238f4c54 NS |
1573 | .releasepage = xfs_vm_releasepage, |
1574 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb | 1575 | .prepare_write = xfs_vm_prepare_write, |
1da177e4 | 1576 | .commit_write = generic_commit_write, |
e4c573bb NS |
1577 | .bmap = xfs_vm_bmap, |
1578 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1579 | .migratepage = buffer_migrate_page, |
1da177e4 | 1580 | }; |