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 | ||
f51623b2 NS |
45 | STATIC void |
46 | xfs_count_page_state( | |
47 | struct page *page, | |
48 | int *delalloc, | |
49 | int *unmapped, | |
50 | int *unwritten) | |
51 | { | |
52 | struct buffer_head *bh, *head; | |
53 | ||
54 | *delalloc = *unmapped = *unwritten = 0; | |
55 | ||
56 | bh = head = page_buffers(page); | |
57 | do { | |
58 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
59 | (*unmapped) = 1; | |
f51623b2 NS |
60 | else if (buffer_unwritten(bh)) |
61 | (*unwritten) = 1; | |
62 | else if (buffer_delay(bh)) | |
63 | (*delalloc) = 1; | |
64 | } while ((bh = bh->b_this_page) != head); | |
65 | } | |
66 | ||
1da177e4 LT |
67 | #if defined(XFS_RW_TRACE) |
68 | void | |
69 | xfs_page_trace( | |
70 | int tag, | |
71 | struct inode *inode, | |
72 | struct page *page, | |
ed9d88f7 | 73 | unsigned long pgoff) |
1da177e4 LT |
74 | { |
75 | xfs_inode_t *ip; | |
67fcaa73 | 76 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 77 | loff_t isize = i_size_read(inode); |
f6d6d4fc | 78 | loff_t offset = page_offset(page); |
1da177e4 LT |
79 | int delalloc = -1, unmapped = -1, unwritten = -1; |
80 | ||
81 | if (page_has_buffers(page)) | |
82 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
83 | ||
75e17b3c | 84 | ip = xfs_vtoi(vp); |
1da177e4 LT |
85 | if (!ip->i_rwtrace) |
86 | return; | |
87 | ||
88 | ktrace_enter(ip->i_rwtrace, | |
89 | (void *)((unsigned long)tag), | |
90 | (void *)ip, | |
91 | (void *)inode, | |
92 | (void *)page, | |
ed9d88f7 | 93 | (void *)pgoff, |
1da177e4 LT |
94 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), |
95 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
96 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
97 | (void *)((unsigned long)(isize & 0xffffffff)), | |
98 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
99 | (void *)((unsigned long)(offset & 0xffffffff)), | |
100 | (void *)((unsigned long)delalloc), | |
101 | (void *)((unsigned long)unmapped), | |
102 | (void *)((unsigned long)unwritten), | |
f1fdc848 | 103 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
104 | (void *)NULL); |
105 | } | |
106 | #else | |
ed9d88f7 | 107 | #define xfs_page_trace(tag, inode, page, pgoff) |
1da177e4 LT |
108 | #endif |
109 | ||
0829c360 CH |
110 | /* |
111 | * Schedule IO completion handling on a xfsdatad if this was | |
e927af90 DC |
112 | * the final hold on this ioend. If we are asked to wait, |
113 | * flush the workqueue. | |
0829c360 CH |
114 | */ |
115 | STATIC void | |
116 | xfs_finish_ioend( | |
e927af90 DC |
117 | xfs_ioend_t *ioend, |
118 | int wait) | |
0829c360 | 119 | { |
e927af90 | 120 | if (atomic_dec_and_test(&ioend->io_remaining)) { |
0829c360 | 121 | queue_work(xfsdatad_workqueue, &ioend->io_work); |
e927af90 DC |
122 | if (wait) |
123 | flush_workqueue(xfsdatad_workqueue); | |
124 | } | |
0829c360 CH |
125 | } |
126 | ||
f6d6d4fc CH |
127 | /* |
128 | * We're now finished for good with this ioend structure. | |
129 | * Update the page state via the associated buffer_heads, | |
130 | * release holds on the inode and bio, and finally free | |
131 | * up memory. Do not use the ioend after this. | |
132 | */ | |
0829c360 CH |
133 | STATIC void |
134 | xfs_destroy_ioend( | |
135 | xfs_ioend_t *ioend) | |
136 | { | |
f6d6d4fc CH |
137 | struct buffer_head *bh, *next; |
138 | ||
139 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
140 | next = bh->b_private; | |
7d04a335 | 141 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 142 | } |
7d04a335 NS |
143 | if (unlikely(ioend->io_error)) |
144 | vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__); | |
0829c360 CH |
145 | vn_iowake(ioend->io_vnode); |
146 | mempool_free(ioend, xfs_ioend_pool); | |
147 | } | |
148 | ||
ba87ea69 LM |
149 | /* |
150 | * Update on-disk file size now that data has been written to disk. | |
151 | * The current in-memory file size is i_size. If a write is beyond | |
152 | * eof io_new_size will be the intended file size until i_size is | |
153 | * updated. If this write does not extend all the way to the valid | |
154 | * file size then restrict this update to the end of the write. | |
155 | */ | |
156 | STATIC void | |
157 | xfs_setfilesize( | |
158 | xfs_ioend_t *ioend) | |
159 | { | |
160 | xfs_inode_t *ip; | |
161 | xfs_fsize_t isize; | |
162 | xfs_fsize_t bsize; | |
163 | ||
164 | ip = xfs_vtoi(ioend->io_vnode); | |
b2826136 DC |
165 | if (!ip) |
166 | return; | |
ba87ea69 LM |
167 | |
168 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); | |
169 | ASSERT(ioend->io_type != IOMAP_READ); | |
170 | ||
171 | if (unlikely(ioend->io_error)) | |
172 | return; | |
173 | ||
174 | bsize = ioend->io_offset + ioend->io_size; | |
175 | ||
176 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
177 | ||
178 | isize = MAX(ip->i_size, ip->i_iocore.io_new_size); | |
179 | isize = MIN(isize, bsize); | |
180 | ||
181 | if (ip->i_d.di_size < isize) { | |
182 | ip->i_d.di_size = isize; | |
183 | ip->i_update_core = 1; | |
184 | ip->i_update_size = 1; | |
776a75fa | 185 | mark_inode_dirty_sync(vn_to_inode(ioend->io_vnode)); |
ba87ea69 LM |
186 | } |
187 | ||
188 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
189 | } | |
190 | ||
0829c360 | 191 | /* |
f6d6d4fc | 192 | * Buffered IO write completion for delayed allocate extents. |
f6d6d4fc CH |
193 | */ |
194 | STATIC void | |
195 | xfs_end_bio_delalloc( | |
c4028958 | 196 | struct work_struct *work) |
f6d6d4fc | 197 | { |
c4028958 DH |
198 | xfs_ioend_t *ioend = |
199 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 200 | |
ba87ea69 | 201 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
202 | xfs_destroy_ioend(ioend); |
203 | } | |
204 | ||
205 | /* | |
206 | * Buffered IO write completion for regular, written extents. | |
207 | */ | |
208 | STATIC void | |
209 | xfs_end_bio_written( | |
c4028958 | 210 | struct work_struct *work) |
f6d6d4fc | 211 | { |
c4028958 DH |
212 | xfs_ioend_t *ioend = |
213 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 214 | |
ba87ea69 | 215 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
216 | xfs_destroy_ioend(ioend); |
217 | } | |
218 | ||
219 | /* | |
220 | * IO write completion for unwritten extents. | |
221 | * | |
0829c360 | 222 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 223 | * to written extents. |
0829c360 CH |
224 | */ |
225 | STATIC void | |
226 | xfs_end_bio_unwritten( | |
c4028958 | 227 | struct work_struct *work) |
0829c360 | 228 | { |
c4028958 DH |
229 | xfs_ioend_t *ioend = |
230 | container_of(work, xfs_ioend_t, io_work); | |
67fcaa73 | 231 | bhv_vnode_t *vp = ioend->io_vnode; |
0829c360 CH |
232 | xfs_off_t offset = ioend->io_offset; |
233 | size_t size = ioend->io_size; | |
0829c360 | 234 | |
ba87ea69 | 235 | if (likely(!ioend->io_error)) { |
739bfb2a CH |
236 | xfs_bmap(xfs_vtoi(vp), offset, size, |
237 | BMAPI_UNWRITTEN, NULL, NULL); | |
ba87ea69 LM |
238 | xfs_setfilesize(ioend); |
239 | } | |
240 | xfs_destroy_ioend(ioend); | |
241 | } | |
242 | ||
243 | /* | |
244 | * IO read completion for regular, written extents. | |
245 | */ | |
246 | STATIC void | |
247 | xfs_end_bio_read( | |
248 | struct work_struct *work) | |
249 | { | |
250 | xfs_ioend_t *ioend = | |
251 | container_of(work, xfs_ioend_t, io_work); | |
252 | ||
0829c360 CH |
253 | xfs_destroy_ioend(ioend); |
254 | } | |
255 | ||
256 | /* | |
257 | * Allocate and initialise an IO completion structure. | |
258 | * We need to track unwritten extent write completion here initially. | |
259 | * We'll need to extend this for updating the ondisk inode size later | |
260 | * (vs. incore size). | |
261 | */ | |
262 | STATIC xfs_ioend_t * | |
263 | xfs_alloc_ioend( | |
f6d6d4fc CH |
264 | struct inode *inode, |
265 | unsigned int type) | |
0829c360 CH |
266 | { |
267 | xfs_ioend_t *ioend; | |
268 | ||
269 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
270 | ||
271 | /* | |
272 | * Set the count to 1 initially, which will prevent an I/O | |
273 | * completion callback from happening before we have started | |
274 | * all the I/O from calling the completion routine too early. | |
275 | */ | |
276 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 277 | ioend->io_error = 0; |
f6d6d4fc CH |
278 | ioend->io_list = NULL; |
279 | ioend->io_type = type; | |
ec86dc02 | 280 | ioend->io_vnode = vn_from_inode(inode); |
c1a073bd | 281 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 282 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
283 | atomic_inc(&ioend->io_vnode->v_iocount); |
284 | ioend->io_offset = 0; | |
285 | ioend->io_size = 0; | |
286 | ||
f6d6d4fc | 287 | if (type == IOMAP_UNWRITTEN) |
c4028958 | 288 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten); |
f6d6d4fc | 289 | else if (type == IOMAP_DELAY) |
c4028958 | 290 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc); |
ba87ea69 LM |
291 | else if (type == IOMAP_READ) |
292 | INIT_WORK(&ioend->io_work, xfs_end_bio_read); | |
f6d6d4fc | 293 | else |
c4028958 | 294 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); |
0829c360 CH |
295 | |
296 | return ioend; | |
297 | } | |
298 | ||
1da177e4 LT |
299 | STATIC int |
300 | xfs_map_blocks( | |
301 | struct inode *inode, | |
302 | loff_t offset, | |
303 | ssize_t count, | |
304 | xfs_iomap_t *mapp, | |
305 | int flags) | |
306 | { | |
67fcaa73 | 307 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
308 | int error, nmaps = 1; |
309 | ||
739bfb2a CH |
310 | error = xfs_bmap(xfs_vtoi(vp), offset, count, |
311 | flags, mapp, &nmaps); | |
1da177e4 LT |
312 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) |
313 | VMODIFY(vp); | |
314 | return -error; | |
315 | } | |
316 | ||
7989cb8e | 317 | STATIC_INLINE int |
1defeac9 | 318 | xfs_iomap_valid( |
1da177e4 | 319 | xfs_iomap_t *iomapp, |
1defeac9 | 320 | loff_t offset) |
1da177e4 | 321 | { |
1defeac9 CH |
322 | return offset >= iomapp->iomap_offset && |
323 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
324 | } |
325 | ||
f6d6d4fc CH |
326 | /* |
327 | * BIO completion handler for buffered IO. | |
328 | */ | |
329 | STATIC int | |
330 | xfs_end_bio( | |
331 | struct bio *bio, | |
332 | unsigned int bytes_done, | |
333 | int error) | |
334 | { | |
335 | xfs_ioend_t *ioend = bio->bi_private; | |
336 | ||
337 | if (bio->bi_size) | |
338 | return 1; | |
339 | ||
f6d6d4fc | 340 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 341 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
342 | |
343 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
344 | bio->bi_private = NULL; |
345 | bio->bi_end_io = NULL; | |
f6d6d4fc | 346 | bio_put(bio); |
7d04a335 | 347 | |
e927af90 | 348 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
349 | return 0; |
350 | } | |
351 | ||
352 | STATIC void | |
353 | xfs_submit_ioend_bio( | |
354 | xfs_ioend_t *ioend, | |
355 | struct bio *bio) | |
356 | { | |
357 | atomic_inc(&ioend->io_remaining); | |
358 | ||
359 | bio->bi_private = ioend; | |
360 | bio->bi_end_io = xfs_end_bio; | |
361 | ||
362 | submit_bio(WRITE, bio); | |
363 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
364 | bio_put(bio); | |
365 | } | |
366 | ||
367 | STATIC struct bio * | |
368 | xfs_alloc_ioend_bio( | |
369 | struct buffer_head *bh) | |
370 | { | |
371 | struct bio *bio; | |
372 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
373 | ||
374 | do { | |
375 | bio = bio_alloc(GFP_NOIO, nvecs); | |
376 | nvecs >>= 1; | |
377 | } while (!bio); | |
378 | ||
379 | ASSERT(bio->bi_private == NULL); | |
380 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
381 | bio->bi_bdev = bh->b_bdev; | |
382 | bio_get(bio); | |
383 | return bio; | |
384 | } | |
385 | ||
386 | STATIC void | |
387 | xfs_start_buffer_writeback( | |
388 | struct buffer_head *bh) | |
389 | { | |
390 | ASSERT(buffer_mapped(bh)); | |
391 | ASSERT(buffer_locked(bh)); | |
392 | ASSERT(!buffer_delay(bh)); | |
393 | ASSERT(!buffer_unwritten(bh)); | |
394 | ||
395 | mark_buffer_async_write(bh); | |
396 | set_buffer_uptodate(bh); | |
397 | clear_buffer_dirty(bh); | |
398 | } | |
399 | ||
400 | STATIC void | |
401 | xfs_start_page_writeback( | |
402 | struct page *page, | |
403 | struct writeback_control *wbc, | |
404 | int clear_dirty, | |
405 | int buffers) | |
406 | { | |
407 | ASSERT(PageLocked(page)); | |
408 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 409 | if (clear_dirty) |
92132021 DC |
410 | clear_page_dirty_for_io(page); |
411 | set_page_writeback(page); | |
f6d6d4fc CH |
412 | unlock_page(page); |
413 | if (!buffers) { | |
414 | end_page_writeback(page); | |
415 | wbc->pages_skipped++; /* We didn't write this page */ | |
416 | } | |
417 | } | |
418 | ||
419 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
420 | { | |
421 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
422 | } | |
423 | ||
424 | /* | |
d88992f6 DC |
425 | * Submit all of the bios for all of the ioends we have saved up, covering the |
426 | * initial writepage page and also any probed pages. | |
427 | * | |
428 | * Because we may have multiple ioends spanning a page, we need to start | |
429 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
430 | * buffers as we got, then we can end up with a page that only has buffers | |
431 | * marked async write and I/O complete on can occur before we mark the other | |
432 | * buffers async write. | |
433 | * | |
434 | * The end result of this is that we trip a bug in end_page_writeback() because | |
435 | * we call it twice for the one page as the code in end_buffer_async_write() | |
436 | * assumes that all buffers on the page are started at the same time. | |
437 | * | |
438 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 439 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
440 | */ |
441 | STATIC void | |
442 | xfs_submit_ioend( | |
443 | xfs_ioend_t *ioend) | |
444 | { | |
d88992f6 | 445 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
446 | xfs_ioend_t *next; |
447 | struct buffer_head *bh; | |
448 | struct bio *bio; | |
449 | sector_t lastblock = 0; | |
450 | ||
d88992f6 DC |
451 | /* Pass 1 - start writeback */ |
452 | do { | |
453 | next = ioend->io_list; | |
454 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
455 | xfs_start_buffer_writeback(bh); | |
456 | } | |
457 | } while ((ioend = next) != NULL); | |
458 | ||
459 | /* Pass 2 - submit I/O */ | |
460 | ioend = head; | |
f6d6d4fc CH |
461 | do { |
462 | next = ioend->io_list; | |
463 | bio = NULL; | |
464 | ||
465 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
466 | |
467 | if (!bio) { | |
468 | retry: | |
469 | bio = xfs_alloc_ioend_bio(bh); | |
470 | } else if (bh->b_blocknr != lastblock + 1) { | |
471 | xfs_submit_ioend_bio(ioend, bio); | |
472 | goto retry; | |
473 | } | |
474 | ||
475 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
476 | xfs_submit_ioend_bio(ioend, bio); | |
477 | goto retry; | |
478 | } | |
479 | ||
480 | lastblock = bh->b_blocknr; | |
481 | } | |
482 | if (bio) | |
483 | xfs_submit_ioend_bio(ioend, bio); | |
e927af90 | 484 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
485 | } while ((ioend = next) != NULL); |
486 | } | |
487 | ||
488 | /* | |
489 | * Cancel submission of all buffer_heads so far in this endio. | |
490 | * Toss the endio too. Only ever called for the initial page | |
491 | * in a writepage request, so only ever one page. | |
492 | */ | |
493 | STATIC void | |
494 | xfs_cancel_ioend( | |
495 | xfs_ioend_t *ioend) | |
496 | { | |
497 | xfs_ioend_t *next; | |
498 | struct buffer_head *bh, *next_bh; | |
499 | ||
500 | do { | |
501 | next = ioend->io_list; | |
502 | bh = ioend->io_buffer_head; | |
503 | do { | |
504 | next_bh = bh->b_private; | |
505 | clear_buffer_async_write(bh); | |
506 | unlock_buffer(bh); | |
507 | } while ((bh = next_bh) != NULL); | |
508 | ||
509 | vn_iowake(ioend->io_vnode); | |
510 | mempool_free(ioend, xfs_ioend_pool); | |
511 | } while ((ioend = next) != NULL); | |
512 | } | |
513 | ||
514 | /* | |
515 | * Test to see if we've been building up a completion structure for | |
516 | * earlier buffers -- if so, we try to append to this ioend if we | |
517 | * can, otherwise we finish off any current ioend and start another. | |
518 | * Return true if we've finished the given ioend. | |
519 | */ | |
520 | STATIC void | |
521 | xfs_add_to_ioend( | |
522 | struct inode *inode, | |
523 | struct buffer_head *bh, | |
7336cea8 | 524 | xfs_off_t offset, |
f6d6d4fc CH |
525 | unsigned int type, |
526 | xfs_ioend_t **result, | |
527 | int need_ioend) | |
528 | { | |
529 | xfs_ioend_t *ioend = *result; | |
530 | ||
531 | if (!ioend || need_ioend || type != ioend->io_type) { | |
532 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 533 | |
f6d6d4fc CH |
534 | ioend = xfs_alloc_ioend(inode, type); |
535 | ioend->io_offset = offset; | |
536 | ioend->io_buffer_head = bh; | |
537 | ioend->io_buffer_tail = bh; | |
538 | if (previous) | |
539 | previous->io_list = ioend; | |
540 | *result = ioend; | |
541 | } else { | |
542 | ioend->io_buffer_tail->b_private = bh; | |
543 | ioend->io_buffer_tail = bh; | |
544 | } | |
545 | ||
546 | bh->b_private = NULL; | |
547 | ioend->io_size += bh->b_size; | |
548 | } | |
549 | ||
87cbc49c NS |
550 | STATIC void |
551 | xfs_map_buffer( | |
552 | struct buffer_head *bh, | |
553 | xfs_iomap_t *mp, | |
554 | xfs_off_t offset, | |
555 | uint block_bits) | |
556 | { | |
557 | sector_t bn; | |
558 | ||
559 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
560 | ||
561 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
562 | ((offset - mp->iomap_offset) >> block_bits); | |
563 | ||
564 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
565 | ||
566 | bh->b_blocknr = bn; | |
567 | set_buffer_mapped(bh); | |
568 | } | |
569 | ||
1da177e4 LT |
570 | STATIC void |
571 | xfs_map_at_offset( | |
1da177e4 | 572 | struct buffer_head *bh, |
1defeac9 | 573 | loff_t offset, |
1da177e4 | 574 | int block_bits, |
1defeac9 | 575 | xfs_iomap_t *iomapp) |
1da177e4 | 576 | { |
1da177e4 LT |
577 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
578 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
579 | |
580 | lock_buffer(bh); | |
87cbc49c | 581 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 582 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
583 | set_buffer_mapped(bh); |
584 | clear_buffer_delay(bh); | |
f6d6d4fc | 585 | clear_buffer_unwritten(bh); |
1da177e4 LT |
586 | } |
587 | ||
588 | /* | |
6c4fe19f | 589 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
590 | */ |
591 | STATIC unsigned int | |
6c4fe19f | 592 | xfs_probe_page( |
10ce4444 | 593 | struct page *page, |
6c4fe19f CH |
594 | unsigned int pg_offset, |
595 | int mapped) | |
1da177e4 | 596 | { |
1da177e4 LT |
597 | int ret = 0; |
598 | ||
1da177e4 | 599 | if (PageWriteback(page)) |
10ce4444 | 600 | return 0; |
1da177e4 LT |
601 | |
602 | if (page->mapping && PageDirty(page)) { | |
603 | if (page_has_buffers(page)) { | |
604 | struct buffer_head *bh, *head; | |
605 | ||
606 | bh = head = page_buffers(page); | |
607 | do { | |
6c4fe19f CH |
608 | if (!buffer_uptodate(bh)) |
609 | break; | |
610 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
611 | break; |
612 | ret += bh->b_size; | |
613 | if (ret >= pg_offset) | |
614 | break; | |
615 | } while ((bh = bh->b_this_page) != head); | |
616 | } else | |
6c4fe19f | 617 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
618 | } |
619 | ||
1da177e4 LT |
620 | return ret; |
621 | } | |
622 | ||
f6d6d4fc | 623 | STATIC size_t |
6c4fe19f | 624 | xfs_probe_cluster( |
1da177e4 LT |
625 | struct inode *inode, |
626 | struct page *startpage, | |
627 | struct buffer_head *bh, | |
6c4fe19f CH |
628 | struct buffer_head *head, |
629 | int mapped) | |
1da177e4 | 630 | { |
10ce4444 | 631 | struct pagevec pvec; |
1da177e4 | 632 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
633 | size_t total = 0; |
634 | int done = 0, i; | |
1da177e4 LT |
635 | |
636 | /* First sum forwards in this page */ | |
637 | do { | |
2353e8e9 | 638 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 639 | return total; |
1da177e4 LT |
640 | total += bh->b_size; |
641 | } while ((bh = bh->b_this_page) != head); | |
642 | ||
10ce4444 CH |
643 | /* if we reached the end of the page, sum forwards in following pages */ |
644 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
645 | tindex = startpage->index + 1; | |
646 | ||
647 | /* Prune this back to avoid pathological behavior */ | |
648 | tloff = min(tlast, startpage->index + 64); | |
649 | ||
650 | pagevec_init(&pvec, 0); | |
651 | while (!done && tindex <= tloff) { | |
652 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
653 | ||
654 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
655 | break; | |
656 | ||
657 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
658 | struct page *page = pvec.pages[i]; | |
265c1fac | 659 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
660 | |
661 | if (tindex == tlast) { | |
662 | pg_offset = | |
663 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
664 | if (!pg_offset) { |
665 | done = 1; | |
10ce4444 | 666 | break; |
1defeac9 | 667 | } |
10ce4444 CH |
668 | } else |
669 | pg_offset = PAGE_CACHE_SIZE; | |
670 | ||
671 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
265c1fac | 672 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
673 | unlock_page(page); |
674 | } | |
675 | ||
265c1fac | 676 | if (!pg_len) { |
10ce4444 CH |
677 | done = 1; |
678 | break; | |
679 | } | |
680 | ||
265c1fac | 681 | total += pg_len; |
1defeac9 | 682 | tindex++; |
1da177e4 | 683 | } |
10ce4444 CH |
684 | |
685 | pagevec_release(&pvec); | |
686 | cond_resched(); | |
1da177e4 | 687 | } |
10ce4444 | 688 | |
1da177e4 LT |
689 | return total; |
690 | } | |
691 | ||
692 | /* | |
10ce4444 CH |
693 | * Test if a given page is suitable for writing as part of an unwritten |
694 | * or delayed allocate extent. | |
1da177e4 | 695 | */ |
10ce4444 CH |
696 | STATIC int |
697 | xfs_is_delayed_page( | |
698 | struct page *page, | |
f6d6d4fc | 699 | unsigned int type) |
1da177e4 | 700 | { |
1da177e4 | 701 | if (PageWriteback(page)) |
10ce4444 | 702 | return 0; |
1da177e4 LT |
703 | |
704 | if (page->mapping && page_has_buffers(page)) { | |
705 | struct buffer_head *bh, *head; | |
706 | int acceptable = 0; | |
707 | ||
708 | bh = head = page_buffers(page); | |
709 | do { | |
f6d6d4fc CH |
710 | if (buffer_unwritten(bh)) |
711 | acceptable = (type == IOMAP_UNWRITTEN); | |
712 | else if (buffer_delay(bh)) | |
713 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 714 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 715 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 716 | else |
1da177e4 | 717 | break; |
1da177e4 LT |
718 | } while ((bh = bh->b_this_page) != head); |
719 | ||
720 | if (acceptable) | |
10ce4444 | 721 | return 1; |
1da177e4 LT |
722 | } |
723 | ||
10ce4444 | 724 | return 0; |
1da177e4 LT |
725 | } |
726 | ||
1da177e4 LT |
727 | /* |
728 | * Allocate & map buffers for page given the extent map. Write it out. | |
729 | * except for the original page of a writepage, this is called on | |
730 | * delalloc/unwritten pages only, for the original page it is possible | |
731 | * that the page has no mapping at all. | |
732 | */ | |
f6d6d4fc | 733 | STATIC int |
1da177e4 LT |
734 | xfs_convert_page( |
735 | struct inode *inode, | |
736 | struct page *page, | |
10ce4444 | 737 | loff_t tindex, |
1defeac9 | 738 | xfs_iomap_t *mp, |
f6d6d4fc | 739 | xfs_ioend_t **ioendp, |
1da177e4 | 740 | struct writeback_control *wbc, |
1da177e4 LT |
741 | int startio, |
742 | int all_bh) | |
743 | { | |
f6d6d4fc | 744 | struct buffer_head *bh, *head; |
9260dc6b CH |
745 | xfs_off_t end_offset; |
746 | unsigned long p_offset; | |
f6d6d4fc | 747 | unsigned int type; |
1da177e4 | 748 | int bbits = inode->i_blkbits; |
24e17b5f | 749 | int len, page_dirty; |
f6d6d4fc | 750 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 751 | xfs_off_t offset = page_offset(page); |
1da177e4 | 752 | |
10ce4444 CH |
753 | if (page->index != tindex) |
754 | goto fail; | |
755 | if (TestSetPageLocked(page)) | |
756 | goto fail; | |
757 | if (PageWriteback(page)) | |
758 | goto fail_unlock_page; | |
759 | if (page->mapping != inode->i_mapping) | |
760 | goto fail_unlock_page; | |
761 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
762 | goto fail_unlock_page; | |
763 | ||
24e17b5f NS |
764 | /* |
765 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 766 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
767 | * |
768 | * Derivation: | |
769 | * | |
770 | * End offset is the highest offset that this page should represent. | |
771 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
772 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
773 | * hence give us the correct page_dirty count. On any other page, | |
774 | * it will be zero and in that case we need page_dirty to be the | |
775 | * count of buffers on the page. | |
24e17b5f | 776 | */ |
9260dc6b CH |
777 | end_offset = min_t(unsigned long long, |
778 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
779 | i_size_read(inode)); | |
780 | ||
24e17b5f | 781 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
782 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
783 | PAGE_CACHE_SIZE); | |
784 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
785 | page_dirty = p_offset / len; | |
24e17b5f | 786 | |
1da177e4 LT |
787 | bh = head = page_buffers(page); |
788 | do { | |
9260dc6b | 789 | if (offset >= end_offset) |
1da177e4 | 790 | break; |
f6d6d4fc CH |
791 | if (!buffer_uptodate(bh)) |
792 | uptodate = 0; | |
793 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
794 | done = 1; | |
1da177e4 | 795 | continue; |
f6d6d4fc CH |
796 | } |
797 | ||
9260dc6b CH |
798 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
799 | if (buffer_unwritten(bh)) | |
800 | type = IOMAP_UNWRITTEN; | |
801 | else | |
802 | type = IOMAP_DELAY; | |
803 | ||
804 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 805 | done = 1; |
9260dc6b CH |
806 | continue; |
807 | } | |
808 | ||
809 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
810 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
811 | ||
812 | xfs_map_at_offset(bh, offset, bbits, mp); | |
813 | if (startio) { | |
7336cea8 | 814 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
815 | type, ioendp, done); |
816 | } else { | |
817 | set_buffer_dirty(bh); | |
818 | unlock_buffer(bh); | |
819 | mark_buffer_dirty(bh); | |
820 | } | |
821 | page_dirty--; | |
822 | count++; | |
823 | } else { | |
df3c7244 | 824 | type = IOMAP_NEW; |
9260dc6b | 825 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 826 | lock_buffer(bh); |
7336cea8 | 827 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
828 | type, ioendp, done); |
829 | count++; | |
24e17b5f | 830 | page_dirty--; |
9260dc6b CH |
831 | } else { |
832 | done = 1; | |
1da177e4 | 833 | } |
1da177e4 | 834 | } |
7336cea8 | 835 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 836 | |
f6d6d4fc CH |
837 | if (uptodate && bh == head) |
838 | SetPageUptodate(page); | |
839 | ||
840 | if (startio) { | |
f5e596bb CH |
841 | if (count) { |
842 | struct backing_dev_info *bdi; | |
843 | ||
844 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 845 | wbc->nr_to_write--; |
f5e596bb CH |
846 | if (bdi_write_congested(bdi)) { |
847 | wbc->encountered_congestion = 1; | |
848 | done = 1; | |
9fddaca2 | 849 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
850 | done = 1; |
851 | } | |
852 | } | |
f6d6d4fc | 853 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 854 | } |
f6d6d4fc CH |
855 | |
856 | return done; | |
10ce4444 CH |
857 | fail_unlock_page: |
858 | unlock_page(page); | |
859 | fail: | |
860 | return 1; | |
1da177e4 LT |
861 | } |
862 | ||
863 | /* | |
864 | * Convert & write out a cluster of pages in the same extent as defined | |
865 | * by mp and following the start page. | |
866 | */ | |
867 | STATIC void | |
868 | xfs_cluster_write( | |
869 | struct inode *inode, | |
870 | pgoff_t tindex, | |
871 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 872 | xfs_ioend_t **ioendp, |
1da177e4 LT |
873 | struct writeback_control *wbc, |
874 | int startio, | |
875 | int all_bh, | |
876 | pgoff_t tlast) | |
877 | { | |
10ce4444 CH |
878 | struct pagevec pvec; |
879 | int done = 0, i; | |
1da177e4 | 880 | |
10ce4444 CH |
881 | pagevec_init(&pvec, 0); |
882 | while (!done && tindex <= tlast) { | |
883 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
884 | ||
885 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 886 | break; |
10ce4444 CH |
887 | |
888 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
889 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
890 | iomapp, ioendp, wbc, startio, all_bh); | |
891 | if (done) | |
892 | break; | |
893 | } | |
894 | ||
895 | pagevec_release(&pvec); | |
896 | cond_resched(); | |
1da177e4 LT |
897 | } |
898 | } | |
899 | ||
900 | /* | |
901 | * Calling this without startio set means we are being asked to make a dirty | |
902 | * page ready for freeing it's buffers. When called with startio set then | |
903 | * we are coming from writepage. | |
904 | * | |
905 | * When called with startio set it is important that we write the WHOLE | |
906 | * page if possible. | |
907 | * The bh->b_state's cannot know if any of the blocks or which block for | |
908 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 909 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
910 | * may clear the page dirty flag prior to calling write page, under the |
911 | * assumption the entire page will be written out; by not writing out the | |
912 | * whole page the page can be reused before all valid dirty data is | |
913 | * written out. Note: in the case of a page that has been dirty'd by | |
914 | * mapwrite and but partially setup by block_prepare_write the | |
915 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
916 | * valid state, thus the whole page must be written out thing. | |
917 | */ | |
918 | ||
919 | STATIC int | |
920 | xfs_page_state_convert( | |
921 | struct inode *inode, | |
922 | struct page *page, | |
923 | struct writeback_control *wbc, | |
924 | int startio, | |
925 | int unmapped) /* also implies page uptodate */ | |
926 | { | |
f6d6d4fc | 927 | struct buffer_head *bh, *head; |
1defeac9 | 928 | xfs_iomap_t iomap; |
f6d6d4fc | 929 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
930 | loff_t offset; |
931 | unsigned long p_offset = 0; | |
f6d6d4fc | 932 | unsigned int type; |
1da177e4 LT |
933 | __uint64_t end_offset; |
934 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
935 | ssize_t size, len; |
936 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
937 | int page_dirty, count = 0; |
938 | int trylock = 0; | |
6c4fe19f | 939 | int all_bh = unmapped; |
1da177e4 | 940 | |
8272145c NS |
941 | if (startio) { |
942 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
943 | trylock |= BMAPI_TRYLOCK; | |
944 | } | |
3ba0815a | 945 | |
1da177e4 LT |
946 | /* Is this page beyond the end of the file? */ |
947 | offset = i_size_read(inode); | |
948 | end_index = offset >> PAGE_CACHE_SHIFT; | |
949 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
950 | if (page->index >= end_index) { | |
951 | if ((page->index >= end_index + 1) || | |
952 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
953 | if (startio) |
954 | unlock_page(page); | |
955 | return 0; | |
1da177e4 LT |
956 | } |
957 | } | |
958 | ||
1da177e4 | 959 | /* |
24e17b5f | 960 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 961 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
962 | * |
963 | * Derivation: | |
964 | * | |
965 | * End offset is the highest offset that this page should represent. | |
966 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
967 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
968 | * hence give us the correct page_dirty count. On any other page, | |
969 | * it will be zero and in that case we need page_dirty to be the | |
970 | * count of buffers on the page. | |
971 | */ | |
972 | end_offset = min_t(unsigned long long, | |
973 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 974 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
975 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
976 | PAGE_CACHE_SIZE); | |
977 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
978 | page_dirty = p_offset / len; |
979 | ||
24e17b5f | 980 | bh = head = page_buffers(page); |
f6d6d4fc | 981 | offset = page_offset(page); |
df3c7244 DC |
982 | flags = BMAPI_READ; |
983 | type = IOMAP_NEW; | |
f6d6d4fc | 984 | |
f6d6d4fc | 985 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
986 | |
987 | do { | |
988 | if (offset >= end_offset) | |
989 | break; | |
990 | if (!buffer_uptodate(bh)) | |
991 | uptodate = 0; | |
f6d6d4fc | 992 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
993 | /* |
994 | * the iomap is actually still valid, but the ioend | |
995 | * isn't. shouldn't happen too often. | |
996 | */ | |
997 | iomap_valid = 0; | |
1da177e4 | 998 | continue; |
f6d6d4fc | 999 | } |
1da177e4 | 1000 | |
1defeac9 CH |
1001 | if (iomap_valid) |
1002 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
1003 | |
1004 | /* | |
1005 | * First case, map an unwritten extent and prepare for | |
1006 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1007 | * |
1da177e4 LT |
1008 | * Second case, allocate space for a delalloc buffer. |
1009 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1010 | * |
1011 | * Third case, an unmapped buffer was found, and we are | |
1012 | * in a path where we need to write the whole page out. | |
df3c7244 | 1013 | */ |
d5cb48aa CH |
1014 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1015 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1016 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1017 | int new_ioend = 0; |
1018 | ||
df3c7244 | 1019 | /* |
6c4fe19f CH |
1020 | * Make sure we don't use a read-only iomap |
1021 | */ | |
df3c7244 | 1022 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1023 | iomap_valid = 0; |
1024 | ||
f6d6d4fc CH |
1025 | if (buffer_unwritten(bh)) { |
1026 | type = IOMAP_UNWRITTEN; | |
8272145c | 1027 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1028 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1029 | type = IOMAP_DELAY; |
8272145c | 1030 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1031 | } else { |
6c4fe19f | 1032 | type = IOMAP_NEW; |
8272145c | 1033 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1034 | } |
1035 | ||
1defeac9 | 1036 | if (!iomap_valid) { |
effd120e DC |
1037 | /* |
1038 | * if we didn't have a valid mapping then we | |
1039 | * need to ensure that we put the new mapping | |
1040 | * in a new ioend structure. This needs to be | |
1041 | * done to ensure that the ioends correctly | |
1042 | * reflect the block mappings at io completion | |
1043 | * for unwritten extent conversion. | |
1044 | */ | |
1045 | new_ioend = 1; | |
6c4fe19f CH |
1046 | if (type == IOMAP_NEW) { |
1047 | size = xfs_probe_cluster(inode, | |
1048 | page, bh, head, 0); | |
d5cb48aa CH |
1049 | } else { |
1050 | size = len; | |
1051 | } | |
1052 | ||
1053 | err = xfs_map_blocks(inode, offset, size, | |
1054 | &iomap, flags); | |
f6d6d4fc | 1055 | if (err) |
1da177e4 | 1056 | goto error; |
1defeac9 | 1057 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1058 | } |
1defeac9 CH |
1059 | if (iomap_valid) { |
1060 | xfs_map_at_offset(bh, offset, | |
1061 | inode->i_blkbits, &iomap); | |
1da177e4 | 1062 | if (startio) { |
7336cea8 | 1063 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1064 | type, &ioend, |
effd120e | 1065 | new_ioend); |
1da177e4 LT |
1066 | } else { |
1067 | set_buffer_dirty(bh); | |
1068 | unlock_buffer(bh); | |
1069 | mark_buffer_dirty(bh); | |
1070 | } | |
1071 | page_dirty--; | |
f6d6d4fc | 1072 | count++; |
1da177e4 | 1073 | } |
d5cb48aa | 1074 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1075 | /* |
1076 | * we got here because the buffer is already mapped. | |
1077 | * That means it must already have extents allocated | |
1078 | * underneath it. Map the extent by reading it. | |
1079 | */ | |
df3c7244 | 1080 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1081 | flags = BMAPI_READ; |
1082 | size = xfs_probe_cluster(inode, page, bh, | |
1083 | head, 1); | |
1084 | err = xfs_map_blocks(inode, offset, size, | |
1085 | &iomap, flags); | |
1086 | if (err) | |
1087 | goto error; | |
1088 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1089 | } | |
d5cb48aa | 1090 | |
df3c7244 DC |
1091 | /* |
1092 | * We set the type to IOMAP_NEW in case we are doing a | |
1093 | * small write at EOF that is extending the file but | |
1094 | * without needing an allocation. We need to update the | |
1095 | * file size on I/O completion in this case so it is | |
1096 | * the same case as having just allocated a new extent | |
1097 | * that we are writing into for the first time. | |
1098 | */ | |
1099 | type = IOMAP_NEW; | |
d5cb48aa CH |
1100 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
1101 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
1102 | if (iomap_valid) |
1103 | all_bh = 1; | |
7336cea8 | 1104 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1105 | &ioend, !iomap_valid); |
1106 | page_dirty--; | |
1107 | count++; | |
f6d6d4fc | 1108 | } else { |
1defeac9 | 1109 | iomap_valid = 0; |
1da177e4 | 1110 | } |
d5cb48aa CH |
1111 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1112 | (unmapped || startio)) { | |
1113 | iomap_valid = 0; | |
1da177e4 | 1114 | } |
f6d6d4fc CH |
1115 | |
1116 | if (!iohead) | |
1117 | iohead = ioend; | |
1118 | ||
1119 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1120 | |
1121 | if (uptodate && bh == head) | |
1122 | SetPageUptodate(page); | |
1123 | ||
f6d6d4fc CH |
1124 | if (startio) |
1125 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1126 | |
1defeac9 CH |
1127 | if (ioend && iomap_valid) { |
1128 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1129 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1130 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1131 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1132 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1133 | } |
1134 | ||
f6d6d4fc CH |
1135 | if (iohead) |
1136 | xfs_submit_ioend(iohead); | |
1137 | ||
1da177e4 LT |
1138 | return page_dirty; |
1139 | ||
1140 | error: | |
f6d6d4fc CH |
1141 | if (iohead) |
1142 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1143 | |
1144 | /* | |
1145 | * If it's delalloc and we have nowhere to put it, | |
1146 | * throw it away, unless the lower layers told | |
1147 | * us to try again. | |
1148 | */ | |
1149 | if (err != -EAGAIN) { | |
f6d6d4fc | 1150 | if (!unmapped) |
1da177e4 | 1151 | block_invalidatepage(page, 0); |
1da177e4 LT |
1152 | ClearPageUptodate(page); |
1153 | } | |
1154 | return err; | |
1155 | } | |
1156 | ||
f51623b2 NS |
1157 | /* |
1158 | * writepage: Called from one of two places: | |
1159 | * | |
1160 | * 1. we are flushing a delalloc buffer head. | |
1161 | * | |
1162 | * 2. we are writing out a dirty page. Typically the page dirty | |
1163 | * state is cleared before we get here. In this case is it | |
1164 | * conceivable we have no buffer heads. | |
1165 | * | |
1166 | * For delalloc space on the page we need to allocate space and | |
1167 | * flush it. For unmapped buffer heads on the page we should | |
1168 | * allocate space if the page is uptodate. For any other dirty | |
1169 | * buffer heads on the page we should flush them. | |
1170 | * | |
1171 | * If we detect that a transaction would be required to flush | |
1172 | * the page, we have to check the process flags first, if we | |
1173 | * are already in a transaction or disk I/O during allocations | |
1174 | * is off, we need to fail the writepage and redirty the page. | |
1175 | */ | |
1176 | ||
1177 | STATIC int | |
e4c573bb | 1178 | xfs_vm_writepage( |
f51623b2 NS |
1179 | struct page *page, |
1180 | struct writeback_control *wbc) | |
1181 | { | |
1182 | int error; | |
1183 | int need_trans; | |
1184 | int delalloc, unmapped, unwritten; | |
1185 | struct inode *inode = page->mapping->host; | |
1186 | ||
1187 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1188 | ||
1189 | /* | |
1190 | * We need a transaction if: | |
1191 | * 1. There are delalloc buffers on the page | |
1192 | * 2. The page is uptodate and we have unmapped buffers | |
1193 | * 3. The page is uptodate and we have no buffers | |
1194 | * 4. There are unwritten buffers on the page | |
1195 | */ | |
1196 | ||
1197 | if (!page_has_buffers(page)) { | |
1198 | unmapped = 1; | |
1199 | need_trans = 1; | |
1200 | } else { | |
1201 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1202 | if (!PageUptodate(page)) | |
1203 | unmapped = 0; | |
1204 | need_trans = delalloc + unmapped + unwritten; | |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * If we need a transaction and the process flags say | |
1209 | * we are already in a transaction, or no IO is allowed | |
1210 | * then mark the page dirty again and leave the page | |
1211 | * as is. | |
1212 | */ | |
59c1b082 | 1213 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1214 | goto out_fail; |
1215 | ||
1216 | /* | |
1217 | * Delay hooking up buffer heads until we have | |
1218 | * made our go/no-go decision. | |
1219 | */ | |
1220 | if (!page_has_buffers(page)) | |
1221 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1222 | ||
1223 | /* | |
1224 | * Convert delayed allocate, unwritten or unmapped space | |
1225 | * to real space and flush out to disk. | |
1226 | */ | |
1227 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1228 | if (error == -EAGAIN) | |
1229 | goto out_fail; | |
1230 | if (unlikely(error < 0)) | |
1231 | goto out_unlock; | |
1232 | ||
1233 | return 0; | |
1234 | ||
1235 | out_fail: | |
1236 | redirty_page_for_writepage(wbc, page); | |
1237 | unlock_page(page); | |
1238 | return 0; | |
1239 | out_unlock: | |
1240 | unlock_page(page); | |
1241 | return error; | |
1242 | } | |
1243 | ||
7d4fb40a NS |
1244 | STATIC int |
1245 | xfs_vm_writepages( | |
1246 | struct address_space *mapping, | |
1247 | struct writeback_control *wbc) | |
1248 | { | |
67fcaa73 | 1249 | struct bhv_vnode *vp = vn_from_inode(mapping->host); |
7d4fb40a NS |
1250 | |
1251 | if (VN_TRUNC(vp)) | |
1252 | VUNTRUNCATE(vp); | |
1253 | return generic_writepages(mapping, wbc); | |
1254 | } | |
1255 | ||
f51623b2 NS |
1256 | /* |
1257 | * Called to move a page into cleanable state - and from there | |
1258 | * to be released. Possibly the page is already clean. We always | |
1259 | * have buffer heads in this call. | |
1260 | * | |
1261 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1262 | * | |
1263 | * Possible scenarios are: | |
1264 | * | |
1265 | * 1. We are being called to release a page which has been written | |
1266 | * to via regular I/O. buffer heads will be dirty and possibly | |
1267 | * delalloc. If no delalloc buffer heads in this case then we | |
1268 | * can just return zero. | |
1269 | * | |
1270 | * 2. We are called to release a page which has been written via | |
1271 | * mmap, all we need to do is ensure there is no delalloc | |
1272 | * state in the buffer heads, if not we can let the caller | |
1273 | * free them and we should come back later via writepage. | |
1274 | */ | |
1275 | STATIC int | |
238f4c54 | 1276 | xfs_vm_releasepage( |
f51623b2 NS |
1277 | struct page *page, |
1278 | gfp_t gfp_mask) | |
1279 | { | |
1280 | struct inode *inode = page->mapping->host; | |
1281 | int dirty, delalloc, unmapped, unwritten; | |
1282 | struct writeback_control wbc = { | |
1283 | .sync_mode = WB_SYNC_ALL, | |
1284 | .nr_to_write = 1, | |
1285 | }; | |
1286 | ||
ed9d88f7 | 1287 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0); |
f51623b2 | 1288 | |
238f4c54 NS |
1289 | if (!page_has_buffers(page)) |
1290 | return 0; | |
1291 | ||
f51623b2 NS |
1292 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1293 | if (!delalloc && !unwritten) | |
1294 | goto free_buffers; | |
1295 | ||
1296 | if (!(gfp_mask & __GFP_FS)) | |
1297 | return 0; | |
1298 | ||
1299 | /* If we are already inside a transaction or the thread cannot | |
1300 | * do I/O, we cannot release this page. | |
1301 | */ | |
59c1b082 | 1302 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1303 | return 0; |
1304 | ||
1305 | /* | |
1306 | * Convert delalloc space to real space, do not flush the | |
1307 | * data out to disk, that will be done by the caller. | |
1308 | * Never need to allocate space here - we will always | |
1309 | * come back to writepage in that case. | |
1310 | */ | |
1311 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1312 | if (dirty == 0 && !unwritten) | |
1313 | goto free_buffers; | |
1314 | return 0; | |
1315 | ||
1316 | free_buffers: | |
1317 | return try_to_free_buffers(page); | |
1318 | } | |
1319 | ||
1da177e4 | 1320 | STATIC int |
c2536668 | 1321 | __xfs_get_blocks( |
1da177e4 LT |
1322 | struct inode *inode, |
1323 | sector_t iblock, | |
1da177e4 LT |
1324 | struct buffer_head *bh_result, |
1325 | int create, | |
1326 | int direct, | |
1327 | bmapi_flags_t flags) | |
1328 | { | |
1da177e4 | 1329 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1330 | xfs_off_t offset; |
1331 | ssize_t size; | |
c2536668 | 1332 | int niomap = 1; |
1da177e4 | 1333 | int error; |
1da177e4 | 1334 | |
fdc7ed75 | 1335 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1336 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1337 | size = bh_result->b_size; | |
739bfb2a | 1338 | error = xfs_bmap(XFS_I(inode), offset, size, |
67fcaa73 | 1339 | create ? flags : BMAPI_READ, &iomap, &niomap); |
1da177e4 LT |
1340 | if (error) |
1341 | return -error; | |
c2536668 | 1342 | if (niomap == 0) |
1da177e4 LT |
1343 | return 0; |
1344 | ||
1345 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1346 | /* |
1347 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1348 | * the read case (treat as if we're reading into a hole). |
1349 | */ | |
1350 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1351 | xfs_map_buffer(bh_result, &iomap, offset, |
1352 | inode->i_blkbits); | |
1da177e4 LT |
1353 | } |
1354 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1355 | if (direct) | |
1356 | bh_result->b_private = inode; | |
1357 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1358 | } |
1359 | } | |
1360 | ||
c2536668 NS |
1361 | /* |
1362 | * If this is a realtime file, data may be on a different device. | |
1363 | * to that pointed to from the buffer_head b_bdev currently. | |
1364 | */ | |
ce8e922c | 1365 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1366 | |
c2536668 | 1367 | /* |
549054af DC |
1368 | * If we previously allocated a block out beyond eof and we are now |
1369 | * coming back to use it then we will need to flag it as new even if it | |
1370 | * has a disk address. | |
1371 | * | |
1372 | * With sub-block writes into unwritten extents we also need to mark | |
1373 | * the buffer as new so that the unwritten parts of the buffer gets | |
1374 | * correctly zeroed. | |
1da177e4 LT |
1375 | */ |
1376 | if (create && | |
1377 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1378 | (offset >= i_size_read(inode)) || |
1379 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1380 | set_buffer_new(bh_result); |
1da177e4 LT |
1381 | |
1382 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1383 | BUG_ON(direct); | |
1384 | if (create) { | |
1385 | set_buffer_uptodate(bh_result); | |
1386 | set_buffer_mapped(bh_result); | |
1387 | set_buffer_delay(bh_result); | |
1388 | } | |
1389 | } | |
1390 | ||
c2536668 | 1391 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1392 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1393 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1394 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1395 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1396 | } |
1397 | ||
1398 | return 0; | |
1399 | } | |
1400 | ||
1401 | int | |
c2536668 | 1402 | xfs_get_blocks( |
1da177e4 LT |
1403 | struct inode *inode, |
1404 | sector_t iblock, | |
1405 | struct buffer_head *bh_result, | |
1406 | int create) | |
1407 | { | |
c2536668 | 1408 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1409 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1410 | } |
1411 | ||
1412 | STATIC int | |
e4c573bb | 1413 | xfs_get_blocks_direct( |
1da177e4 LT |
1414 | struct inode *inode, |
1415 | sector_t iblock, | |
1da177e4 LT |
1416 | struct buffer_head *bh_result, |
1417 | int create) | |
1418 | { | |
c2536668 | 1419 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1420 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1421 | } |
1422 | ||
f0973863 | 1423 | STATIC void |
e4c573bb | 1424 | xfs_end_io_direct( |
f0973863 CH |
1425 | struct kiocb *iocb, |
1426 | loff_t offset, | |
1427 | ssize_t size, | |
1428 | void *private) | |
1429 | { | |
1430 | xfs_ioend_t *ioend = iocb->private; | |
1431 | ||
1432 | /* | |
1433 | * Non-NULL private data means we need to issue a transaction to | |
1434 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1435 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1436 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1437 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1438 | * it anyway to keep the code uniform and simpler. |
1439 | * | |
e927af90 DC |
1440 | * Well, if only it were that simple. Because synchronous direct I/O |
1441 | * requires extent conversion to occur *before* we return to userspace, | |
1442 | * we have to wait for extent conversion to complete. Look at the | |
1443 | * iocb that has been passed to us to determine if this is AIO or | |
1444 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1445 | * workqueue and wait for it to complete. | |
1446 | * | |
f0973863 CH |
1447 | * The core direct I/O code might be changed to always call the |
1448 | * completion handler in the future, in which case all this can | |
1449 | * go away. | |
1450 | */ | |
ba87ea69 LM |
1451 | ioend->io_offset = offset; |
1452 | ioend->io_size = size; | |
1453 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1454 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1455 | } else if (private && size > 0) { |
e927af90 | 1456 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1457 | } else { |
ba87ea69 LM |
1458 | /* |
1459 | * A direct I/O write ioend starts it's life in unwritten | |
1460 | * state in case they map an unwritten extent. This write | |
1461 | * didn't map an unwritten extent so switch it's completion | |
1462 | * handler. | |
1463 | */ | |
1464 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); | |
e927af90 | 1465 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1466 | } |
1467 | ||
1468 | /* | |
c41564b5 | 1469 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1470 | * completion handler was called. Thus we need to protect |
1471 | * against double-freeing. | |
1472 | */ | |
1473 | iocb->private = NULL; | |
1474 | } | |
1475 | ||
1da177e4 | 1476 | STATIC ssize_t |
e4c573bb | 1477 | xfs_vm_direct_IO( |
1da177e4 LT |
1478 | int rw, |
1479 | struct kiocb *iocb, | |
1480 | const struct iovec *iov, | |
1481 | loff_t offset, | |
1482 | unsigned long nr_segs) | |
1483 | { | |
1484 | struct file *file = iocb->ki_filp; | |
1485 | struct inode *inode = file->f_mapping->host; | |
1da177e4 LT |
1486 | xfs_iomap_t iomap; |
1487 | int maps = 1; | |
1488 | int error; | |
f0973863 | 1489 | ssize_t ret; |
1da177e4 | 1490 | |
739bfb2a CH |
1491 | error = xfs_bmap(XFS_I(inode), offset, 0, |
1492 | BMAPI_DEVICE, &iomap, &maps); | |
1da177e4 LT |
1493 | if (error) |
1494 | return -error; | |
1495 | ||
721259bc | 1496 | if (rw == WRITE) { |
ba87ea69 | 1497 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
721259bc LM |
1498 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, |
1499 | iomap.iomap_target->bt_bdev, | |
1500 | iov, offset, nr_segs, | |
1501 | xfs_get_blocks_direct, | |
1502 | xfs_end_io_direct); | |
1503 | } else { | |
ba87ea69 | 1504 | iocb->private = xfs_alloc_ioend(inode, IOMAP_READ); |
721259bc LM |
1505 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
1506 | iomap.iomap_target->bt_bdev, | |
1507 | iov, offset, nr_segs, | |
1508 | xfs_get_blocks_direct, | |
1509 | xfs_end_io_direct); | |
1510 | } | |
f0973863 | 1511 | |
8459d86a | 1512 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1513 | xfs_destroy_ioend(iocb->private); |
1514 | return ret; | |
1da177e4 LT |
1515 | } |
1516 | ||
f51623b2 | 1517 | STATIC int |
e4c573bb | 1518 | xfs_vm_prepare_write( |
f51623b2 NS |
1519 | struct file *file, |
1520 | struct page *page, | |
1521 | unsigned int from, | |
1522 | unsigned int to) | |
1523 | { | |
c2536668 | 1524 | return block_prepare_write(page, from, to, xfs_get_blocks); |
f51623b2 | 1525 | } |
1da177e4 LT |
1526 | |
1527 | STATIC sector_t | |
e4c573bb | 1528 | xfs_vm_bmap( |
1da177e4 LT |
1529 | struct address_space *mapping, |
1530 | sector_t block) | |
1531 | { | |
1532 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1533 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1534 | |
739bfb2a CH |
1535 | vn_trace_entry(vn_from_inode(inode), __FUNCTION__, |
1536 | (inst_t *)__return_address); | |
1537 | xfs_rwlock(ip, VRWLOCK_READ); | |
1538 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); | |
1539 | xfs_rwunlock(ip, VRWLOCK_READ); | |
c2536668 | 1540 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1541 | } |
1542 | ||
1543 | STATIC int | |
e4c573bb | 1544 | xfs_vm_readpage( |
1da177e4 LT |
1545 | struct file *unused, |
1546 | struct page *page) | |
1547 | { | |
c2536668 | 1548 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1549 | } |
1550 | ||
1551 | STATIC int | |
e4c573bb | 1552 | xfs_vm_readpages( |
1da177e4 LT |
1553 | struct file *unused, |
1554 | struct address_space *mapping, | |
1555 | struct list_head *pages, | |
1556 | unsigned nr_pages) | |
1557 | { | |
c2536668 | 1558 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1559 | } |
1560 | ||
2ff28e22 | 1561 | STATIC void |
238f4c54 | 1562 | xfs_vm_invalidatepage( |
bcec2b7f NS |
1563 | struct page *page, |
1564 | unsigned long offset) | |
1565 | { | |
1566 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1567 | page->mapping->host, page, offset); | |
2ff28e22 | 1568 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1569 | } |
1570 | ||
f5e54d6e | 1571 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1572 | .readpage = xfs_vm_readpage, |
1573 | .readpages = xfs_vm_readpages, | |
1574 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1575 | .writepages = xfs_vm_writepages, |
1da177e4 | 1576 | .sync_page = block_sync_page, |
238f4c54 NS |
1577 | .releasepage = xfs_vm_releasepage, |
1578 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb | 1579 | .prepare_write = xfs_vm_prepare_write, |
1da177e4 | 1580 | .commit_write = generic_commit_write, |
e4c573bb NS |
1581 | .bmap = xfs_vm_bmap, |
1582 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1583 | .migratepage = buffer_migrate_page, |
1da177e4 | 1584 | }; |