]>
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; | |
776a75fa | 184 | mark_inode_dirty_sync(vn_to_inode(ioend->io_vnode)); |
ba87ea69 LM |
185 | } |
186 | ||
187 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
188 | } | |
189 | ||
0829c360 | 190 | /* |
f6d6d4fc | 191 | * Buffered IO write completion for delayed allocate extents. |
f6d6d4fc CH |
192 | */ |
193 | STATIC void | |
194 | xfs_end_bio_delalloc( | |
c4028958 | 195 | struct work_struct *work) |
f6d6d4fc | 196 | { |
c4028958 DH |
197 | xfs_ioend_t *ioend = |
198 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 199 | |
ba87ea69 | 200 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
201 | xfs_destroy_ioend(ioend); |
202 | } | |
203 | ||
204 | /* | |
205 | * Buffered IO write completion for regular, written extents. | |
206 | */ | |
207 | STATIC void | |
208 | xfs_end_bio_written( | |
c4028958 | 209 | struct work_struct *work) |
f6d6d4fc | 210 | { |
c4028958 DH |
211 | xfs_ioend_t *ioend = |
212 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 213 | |
ba87ea69 | 214 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
215 | xfs_destroy_ioend(ioend); |
216 | } | |
217 | ||
218 | /* | |
219 | * IO write completion for unwritten extents. | |
220 | * | |
0829c360 | 221 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 222 | * to written extents. |
0829c360 CH |
223 | */ |
224 | STATIC void | |
225 | xfs_end_bio_unwritten( | |
c4028958 | 226 | struct work_struct *work) |
0829c360 | 227 | { |
c4028958 DH |
228 | xfs_ioend_t *ioend = |
229 | container_of(work, xfs_ioend_t, io_work); | |
67fcaa73 | 230 | bhv_vnode_t *vp = ioend->io_vnode; |
0829c360 CH |
231 | xfs_off_t offset = ioend->io_offset; |
232 | size_t size = ioend->io_size; | |
0829c360 | 233 | |
ba87ea69 | 234 | if (likely(!ioend->io_error)) { |
67fcaa73 | 235 | bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL); |
ba87ea69 LM |
236 | xfs_setfilesize(ioend); |
237 | } | |
238 | xfs_destroy_ioend(ioend); | |
239 | } | |
240 | ||
241 | /* | |
242 | * IO read completion for regular, written extents. | |
243 | */ | |
244 | STATIC void | |
245 | xfs_end_bio_read( | |
246 | struct work_struct *work) | |
247 | { | |
248 | xfs_ioend_t *ioend = | |
249 | container_of(work, xfs_ioend_t, io_work); | |
250 | ||
0829c360 CH |
251 | xfs_destroy_ioend(ioend); |
252 | } | |
253 | ||
254 | /* | |
255 | * Allocate and initialise an IO completion structure. | |
256 | * We need to track unwritten extent write completion here initially. | |
257 | * We'll need to extend this for updating the ondisk inode size later | |
258 | * (vs. incore size). | |
259 | */ | |
260 | STATIC xfs_ioend_t * | |
261 | xfs_alloc_ioend( | |
f6d6d4fc CH |
262 | struct inode *inode, |
263 | unsigned int type) | |
0829c360 CH |
264 | { |
265 | xfs_ioend_t *ioend; | |
266 | ||
267 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
268 | ||
269 | /* | |
270 | * Set the count to 1 initially, which will prevent an I/O | |
271 | * completion callback from happening before we have started | |
272 | * all the I/O from calling the completion routine too early. | |
273 | */ | |
274 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 275 | ioend->io_error = 0; |
f6d6d4fc CH |
276 | ioend->io_list = NULL; |
277 | ioend->io_type = type; | |
ec86dc02 | 278 | ioend->io_vnode = vn_from_inode(inode); |
c1a073bd | 279 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 280 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
281 | atomic_inc(&ioend->io_vnode->v_iocount); |
282 | ioend->io_offset = 0; | |
283 | ioend->io_size = 0; | |
284 | ||
f6d6d4fc | 285 | if (type == IOMAP_UNWRITTEN) |
c4028958 | 286 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten); |
f6d6d4fc | 287 | else if (type == IOMAP_DELAY) |
c4028958 | 288 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc); |
ba87ea69 LM |
289 | else if (type == IOMAP_READ) |
290 | INIT_WORK(&ioend->io_work, xfs_end_bio_read); | |
f6d6d4fc | 291 | else |
c4028958 | 292 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); |
0829c360 CH |
293 | |
294 | return ioend; | |
295 | } | |
296 | ||
1da177e4 LT |
297 | STATIC int |
298 | xfs_map_blocks( | |
299 | struct inode *inode, | |
300 | loff_t offset, | |
301 | ssize_t count, | |
302 | xfs_iomap_t *mapp, | |
303 | int flags) | |
304 | { | |
67fcaa73 | 305 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
306 | int error, nmaps = 1; |
307 | ||
67fcaa73 | 308 | error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps); |
1da177e4 LT |
309 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) |
310 | VMODIFY(vp); | |
311 | return -error; | |
312 | } | |
313 | ||
7989cb8e | 314 | STATIC_INLINE int |
1defeac9 | 315 | xfs_iomap_valid( |
1da177e4 | 316 | xfs_iomap_t *iomapp, |
1defeac9 | 317 | loff_t offset) |
1da177e4 | 318 | { |
1defeac9 CH |
319 | return offset >= iomapp->iomap_offset && |
320 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
321 | } |
322 | ||
f6d6d4fc CH |
323 | /* |
324 | * BIO completion handler for buffered IO. | |
325 | */ | |
782e3b3b | 326 | STATIC void |
f6d6d4fc CH |
327 | xfs_end_bio( |
328 | struct bio *bio, | |
f6d6d4fc CH |
329 | int error) |
330 | { | |
331 | xfs_ioend_t *ioend = bio->bi_private; | |
332 | ||
f6d6d4fc | 333 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 334 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
335 | |
336 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
337 | bio->bi_private = NULL; |
338 | bio->bi_end_io = NULL; | |
f6d6d4fc | 339 | bio_put(bio); |
7d04a335 | 340 | |
e927af90 | 341 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
342 | } |
343 | ||
344 | STATIC void | |
345 | xfs_submit_ioend_bio( | |
346 | xfs_ioend_t *ioend, | |
347 | struct bio *bio) | |
348 | { | |
349 | atomic_inc(&ioend->io_remaining); | |
350 | ||
351 | bio->bi_private = ioend; | |
352 | bio->bi_end_io = xfs_end_bio; | |
353 | ||
354 | submit_bio(WRITE, bio); | |
355 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
356 | bio_put(bio); | |
357 | } | |
358 | ||
359 | STATIC struct bio * | |
360 | xfs_alloc_ioend_bio( | |
361 | struct buffer_head *bh) | |
362 | { | |
363 | struct bio *bio; | |
364 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
365 | ||
366 | do { | |
367 | bio = bio_alloc(GFP_NOIO, nvecs); | |
368 | nvecs >>= 1; | |
369 | } while (!bio); | |
370 | ||
371 | ASSERT(bio->bi_private == NULL); | |
372 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
373 | bio->bi_bdev = bh->b_bdev; | |
374 | bio_get(bio); | |
375 | return bio; | |
376 | } | |
377 | ||
378 | STATIC void | |
379 | xfs_start_buffer_writeback( | |
380 | struct buffer_head *bh) | |
381 | { | |
382 | ASSERT(buffer_mapped(bh)); | |
383 | ASSERT(buffer_locked(bh)); | |
384 | ASSERT(!buffer_delay(bh)); | |
385 | ASSERT(!buffer_unwritten(bh)); | |
386 | ||
387 | mark_buffer_async_write(bh); | |
388 | set_buffer_uptodate(bh); | |
389 | clear_buffer_dirty(bh); | |
390 | } | |
391 | ||
392 | STATIC void | |
393 | xfs_start_page_writeback( | |
394 | struct page *page, | |
395 | struct writeback_control *wbc, | |
396 | int clear_dirty, | |
397 | int buffers) | |
398 | { | |
399 | ASSERT(PageLocked(page)); | |
400 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 401 | if (clear_dirty) |
92132021 DC |
402 | clear_page_dirty_for_io(page); |
403 | set_page_writeback(page); | |
f6d6d4fc CH |
404 | unlock_page(page); |
405 | if (!buffers) { | |
406 | end_page_writeback(page); | |
407 | wbc->pages_skipped++; /* We didn't write this page */ | |
408 | } | |
409 | } | |
410 | ||
411 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
412 | { | |
413 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
414 | } | |
415 | ||
416 | /* | |
d88992f6 DC |
417 | * Submit all of the bios for all of the ioends we have saved up, covering the |
418 | * initial writepage page and also any probed pages. | |
419 | * | |
420 | * Because we may have multiple ioends spanning a page, we need to start | |
421 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
422 | * buffers as we got, then we can end up with a page that only has buffers | |
423 | * marked async write and I/O complete on can occur before we mark the other | |
424 | * buffers async write. | |
425 | * | |
426 | * The end result of this is that we trip a bug in end_page_writeback() because | |
427 | * we call it twice for the one page as the code in end_buffer_async_write() | |
428 | * assumes that all buffers on the page are started at the same time. | |
429 | * | |
430 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 431 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
432 | */ |
433 | STATIC void | |
434 | xfs_submit_ioend( | |
435 | xfs_ioend_t *ioend) | |
436 | { | |
d88992f6 | 437 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
438 | xfs_ioend_t *next; |
439 | struct buffer_head *bh; | |
440 | struct bio *bio; | |
441 | sector_t lastblock = 0; | |
442 | ||
d88992f6 DC |
443 | /* Pass 1 - start writeback */ |
444 | do { | |
445 | next = ioend->io_list; | |
446 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
447 | xfs_start_buffer_writeback(bh); | |
448 | } | |
449 | } while ((ioend = next) != NULL); | |
450 | ||
451 | /* Pass 2 - submit I/O */ | |
452 | ioend = head; | |
f6d6d4fc CH |
453 | do { |
454 | next = ioend->io_list; | |
455 | bio = NULL; | |
456 | ||
457 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
458 | |
459 | if (!bio) { | |
460 | retry: | |
461 | bio = xfs_alloc_ioend_bio(bh); | |
462 | } else if (bh->b_blocknr != lastblock + 1) { | |
463 | xfs_submit_ioend_bio(ioend, bio); | |
464 | goto retry; | |
465 | } | |
466 | ||
467 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
468 | xfs_submit_ioend_bio(ioend, bio); | |
469 | goto retry; | |
470 | } | |
471 | ||
472 | lastblock = bh->b_blocknr; | |
473 | } | |
474 | if (bio) | |
475 | xfs_submit_ioend_bio(ioend, bio); | |
e927af90 | 476 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
477 | } while ((ioend = next) != NULL); |
478 | } | |
479 | ||
480 | /* | |
481 | * Cancel submission of all buffer_heads so far in this endio. | |
482 | * Toss the endio too. Only ever called for the initial page | |
483 | * in a writepage request, so only ever one page. | |
484 | */ | |
485 | STATIC void | |
486 | xfs_cancel_ioend( | |
487 | xfs_ioend_t *ioend) | |
488 | { | |
489 | xfs_ioend_t *next; | |
490 | struct buffer_head *bh, *next_bh; | |
491 | ||
492 | do { | |
493 | next = ioend->io_list; | |
494 | bh = ioend->io_buffer_head; | |
495 | do { | |
496 | next_bh = bh->b_private; | |
497 | clear_buffer_async_write(bh); | |
498 | unlock_buffer(bh); | |
499 | } while ((bh = next_bh) != NULL); | |
500 | ||
501 | vn_iowake(ioend->io_vnode); | |
502 | mempool_free(ioend, xfs_ioend_pool); | |
503 | } while ((ioend = next) != NULL); | |
504 | } | |
505 | ||
506 | /* | |
507 | * Test to see if we've been building up a completion structure for | |
508 | * earlier buffers -- if so, we try to append to this ioend if we | |
509 | * can, otherwise we finish off any current ioend and start another. | |
510 | * Return true if we've finished the given ioend. | |
511 | */ | |
512 | STATIC void | |
513 | xfs_add_to_ioend( | |
514 | struct inode *inode, | |
515 | struct buffer_head *bh, | |
7336cea8 | 516 | xfs_off_t offset, |
f6d6d4fc CH |
517 | unsigned int type, |
518 | xfs_ioend_t **result, | |
519 | int need_ioend) | |
520 | { | |
521 | xfs_ioend_t *ioend = *result; | |
522 | ||
523 | if (!ioend || need_ioend || type != ioend->io_type) { | |
524 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 525 | |
f6d6d4fc CH |
526 | ioend = xfs_alloc_ioend(inode, type); |
527 | ioend->io_offset = offset; | |
528 | ioend->io_buffer_head = bh; | |
529 | ioend->io_buffer_tail = bh; | |
530 | if (previous) | |
531 | previous->io_list = ioend; | |
532 | *result = ioend; | |
533 | } else { | |
534 | ioend->io_buffer_tail->b_private = bh; | |
535 | ioend->io_buffer_tail = bh; | |
536 | } | |
537 | ||
538 | bh->b_private = NULL; | |
539 | ioend->io_size += bh->b_size; | |
540 | } | |
541 | ||
87cbc49c NS |
542 | STATIC void |
543 | xfs_map_buffer( | |
544 | struct buffer_head *bh, | |
545 | xfs_iomap_t *mp, | |
546 | xfs_off_t offset, | |
547 | uint block_bits) | |
548 | { | |
549 | sector_t bn; | |
550 | ||
551 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
552 | ||
553 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
554 | ((offset - mp->iomap_offset) >> block_bits); | |
555 | ||
556 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
557 | ||
558 | bh->b_blocknr = bn; | |
559 | set_buffer_mapped(bh); | |
560 | } | |
561 | ||
1da177e4 LT |
562 | STATIC void |
563 | xfs_map_at_offset( | |
1da177e4 | 564 | struct buffer_head *bh, |
1defeac9 | 565 | loff_t offset, |
1da177e4 | 566 | int block_bits, |
1defeac9 | 567 | xfs_iomap_t *iomapp) |
1da177e4 | 568 | { |
1da177e4 LT |
569 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
570 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
571 | |
572 | lock_buffer(bh); | |
87cbc49c | 573 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 574 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
575 | set_buffer_mapped(bh); |
576 | clear_buffer_delay(bh); | |
f6d6d4fc | 577 | clear_buffer_unwritten(bh); |
1da177e4 LT |
578 | } |
579 | ||
580 | /* | |
6c4fe19f | 581 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
582 | */ |
583 | STATIC unsigned int | |
6c4fe19f | 584 | xfs_probe_page( |
10ce4444 | 585 | struct page *page, |
6c4fe19f CH |
586 | unsigned int pg_offset, |
587 | int mapped) | |
1da177e4 | 588 | { |
1da177e4 LT |
589 | int ret = 0; |
590 | ||
1da177e4 | 591 | if (PageWriteback(page)) |
10ce4444 | 592 | return 0; |
1da177e4 LT |
593 | |
594 | if (page->mapping && PageDirty(page)) { | |
595 | if (page_has_buffers(page)) { | |
596 | struct buffer_head *bh, *head; | |
597 | ||
598 | bh = head = page_buffers(page); | |
599 | do { | |
6c4fe19f CH |
600 | if (!buffer_uptodate(bh)) |
601 | break; | |
602 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
603 | break; |
604 | ret += bh->b_size; | |
605 | if (ret >= pg_offset) | |
606 | break; | |
607 | } while ((bh = bh->b_this_page) != head); | |
608 | } else | |
6c4fe19f | 609 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
610 | } |
611 | ||
1da177e4 LT |
612 | return ret; |
613 | } | |
614 | ||
f6d6d4fc | 615 | STATIC size_t |
6c4fe19f | 616 | xfs_probe_cluster( |
1da177e4 LT |
617 | struct inode *inode, |
618 | struct page *startpage, | |
619 | struct buffer_head *bh, | |
6c4fe19f CH |
620 | struct buffer_head *head, |
621 | int mapped) | |
1da177e4 | 622 | { |
10ce4444 | 623 | struct pagevec pvec; |
1da177e4 | 624 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
625 | size_t total = 0; |
626 | int done = 0, i; | |
1da177e4 LT |
627 | |
628 | /* First sum forwards in this page */ | |
629 | do { | |
2353e8e9 | 630 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 631 | return total; |
1da177e4 LT |
632 | total += bh->b_size; |
633 | } while ((bh = bh->b_this_page) != head); | |
634 | ||
10ce4444 CH |
635 | /* if we reached the end of the page, sum forwards in following pages */ |
636 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
637 | tindex = startpage->index + 1; | |
638 | ||
639 | /* Prune this back to avoid pathological behavior */ | |
640 | tloff = min(tlast, startpage->index + 64); | |
641 | ||
642 | pagevec_init(&pvec, 0); | |
643 | while (!done && tindex <= tloff) { | |
644 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
645 | ||
646 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
647 | break; | |
648 | ||
649 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
650 | struct page *page = pvec.pages[i]; | |
265c1fac | 651 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
652 | |
653 | if (tindex == tlast) { | |
654 | pg_offset = | |
655 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
656 | if (!pg_offset) { |
657 | done = 1; | |
10ce4444 | 658 | break; |
1defeac9 | 659 | } |
10ce4444 CH |
660 | } else |
661 | pg_offset = PAGE_CACHE_SIZE; | |
662 | ||
663 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
265c1fac | 664 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
665 | unlock_page(page); |
666 | } | |
667 | ||
265c1fac | 668 | if (!pg_len) { |
10ce4444 CH |
669 | done = 1; |
670 | break; | |
671 | } | |
672 | ||
265c1fac | 673 | total += pg_len; |
1defeac9 | 674 | tindex++; |
1da177e4 | 675 | } |
10ce4444 CH |
676 | |
677 | pagevec_release(&pvec); | |
678 | cond_resched(); | |
1da177e4 | 679 | } |
10ce4444 | 680 | |
1da177e4 LT |
681 | return total; |
682 | } | |
683 | ||
684 | /* | |
10ce4444 CH |
685 | * Test if a given page is suitable for writing as part of an unwritten |
686 | * or delayed allocate extent. | |
1da177e4 | 687 | */ |
10ce4444 CH |
688 | STATIC int |
689 | xfs_is_delayed_page( | |
690 | struct page *page, | |
f6d6d4fc | 691 | unsigned int type) |
1da177e4 | 692 | { |
1da177e4 | 693 | if (PageWriteback(page)) |
10ce4444 | 694 | return 0; |
1da177e4 LT |
695 | |
696 | if (page->mapping && page_has_buffers(page)) { | |
697 | struct buffer_head *bh, *head; | |
698 | int acceptable = 0; | |
699 | ||
700 | bh = head = page_buffers(page); | |
701 | do { | |
f6d6d4fc CH |
702 | if (buffer_unwritten(bh)) |
703 | acceptable = (type == IOMAP_UNWRITTEN); | |
704 | else if (buffer_delay(bh)) | |
705 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 706 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 707 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 708 | else |
1da177e4 | 709 | break; |
1da177e4 LT |
710 | } while ((bh = bh->b_this_page) != head); |
711 | ||
712 | if (acceptable) | |
10ce4444 | 713 | return 1; |
1da177e4 LT |
714 | } |
715 | ||
10ce4444 | 716 | return 0; |
1da177e4 LT |
717 | } |
718 | ||
1da177e4 LT |
719 | /* |
720 | * Allocate & map buffers for page given the extent map. Write it out. | |
721 | * except for the original page of a writepage, this is called on | |
722 | * delalloc/unwritten pages only, for the original page it is possible | |
723 | * that the page has no mapping at all. | |
724 | */ | |
f6d6d4fc | 725 | STATIC int |
1da177e4 LT |
726 | xfs_convert_page( |
727 | struct inode *inode, | |
728 | struct page *page, | |
10ce4444 | 729 | loff_t tindex, |
1defeac9 | 730 | xfs_iomap_t *mp, |
f6d6d4fc | 731 | xfs_ioend_t **ioendp, |
1da177e4 | 732 | struct writeback_control *wbc, |
1da177e4 LT |
733 | int startio, |
734 | int all_bh) | |
735 | { | |
f6d6d4fc | 736 | struct buffer_head *bh, *head; |
9260dc6b CH |
737 | xfs_off_t end_offset; |
738 | unsigned long p_offset; | |
f6d6d4fc | 739 | unsigned int type; |
1da177e4 | 740 | int bbits = inode->i_blkbits; |
24e17b5f | 741 | int len, page_dirty; |
f6d6d4fc | 742 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 743 | xfs_off_t offset = page_offset(page); |
1da177e4 | 744 | |
10ce4444 CH |
745 | if (page->index != tindex) |
746 | goto fail; | |
747 | if (TestSetPageLocked(page)) | |
748 | goto fail; | |
749 | if (PageWriteback(page)) | |
750 | goto fail_unlock_page; | |
751 | if (page->mapping != inode->i_mapping) | |
752 | goto fail_unlock_page; | |
753 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
754 | goto fail_unlock_page; | |
755 | ||
24e17b5f NS |
756 | /* |
757 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 758 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
759 | * |
760 | * Derivation: | |
761 | * | |
762 | * End offset is the highest offset that this page should represent. | |
763 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
764 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
765 | * hence give us the correct page_dirty count. On any other page, | |
766 | * it will be zero and in that case we need page_dirty to be the | |
767 | * count of buffers on the page. | |
24e17b5f | 768 | */ |
9260dc6b CH |
769 | end_offset = min_t(unsigned long long, |
770 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
771 | i_size_read(inode)); | |
772 | ||
24e17b5f | 773 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
774 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
775 | PAGE_CACHE_SIZE); | |
776 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
777 | page_dirty = p_offset / len; | |
24e17b5f | 778 | |
1da177e4 LT |
779 | bh = head = page_buffers(page); |
780 | do { | |
9260dc6b | 781 | if (offset >= end_offset) |
1da177e4 | 782 | break; |
f6d6d4fc CH |
783 | if (!buffer_uptodate(bh)) |
784 | uptodate = 0; | |
785 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
786 | done = 1; | |
1da177e4 | 787 | continue; |
f6d6d4fc CH |
788 | } |
789 | ||
9260dc6b CH |
790 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
791 | if (buffer_unwritten(bh)) | |
792 | type = IOMAP_UNWRITTEN; | |
793 | else | |
794 | type = IOMAP_DELAY; | |
795 | ||
796 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 797 | done = 1; |
9260dc6b CH |
798 | continue; |
799 | } | |
800 | ||
801 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
802 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
803 | ||
804 | xfs_map_at_offset(bh, offset, bbits, mp); | |
805 | if (startio) { | |
7336cea8 | 806 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
807 | type, ioendp, done); |
808 | } else { | |
809 | set_buffer_dirty(bh); | |
810 | unlock_buffer(bh); | |
811 | mark_buffer_dirty(bh); | |
812 | } | |
813 | page_dirty--; | |
814 | count++; | |
815 | } else { | |
df3c7244 | 816 | type = IOMAP_NEW; |
9260dc6b | 817 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 818 | lock_buffer(bh); |
7336cea8 | 819 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
820 | type, ioendp, done); |
821 | count++; | |
24e17b5f | 822 | page_dirty--; |
9260dc6b CH |
823 | } else { |
824 | done = 1; | |
1da177e4 | 825 | } |
1da177e4 | 826 | } |
7336cea8 | 827 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 828 | |
f6d6d4fc CH |
829 | if (uptodate && bh == head) |
830 | SetPageUptodate(page); | |
831 | ||
832 | if (startio) { | |
f5e596bb CH |
833 | if (count) { |
834 | struct backing_dev_info *bdi; | |
835 | ||
836 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 837 | wbc->nr_to_write--; |
f5e596bb CH |
838 | if (bdi_write_congested(bdi)) { |
839 | wbc->encountered_congestion = 1; | |
840 | done = 1; | |
9fddaca2 | 841 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
842 | done = 1; |
843 | } | |
844 | } | |
f6d6d4fc | 845 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 846 | } |
f6d6d4fc CH |
847 | |
848 | return done; | |
10ce4444 CH |
849 | fail_unlock_page: |
850 | unlock_page(page); | |
851 | fail: | |
852 | return 1; | |
1da177e4 LT |
853 | } |
854 | ||
855 | /* | |
856 | * Convert & write out a cluster of pages in the same extent as defined | |
857 | * by mp and following the start page. | |
858 | */ | |
859 | STATIC void | |
860 | xfs_cluster_write( | |
861 | struct inode *inode, | |
862 | pgoff_t tindex, | |
863 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 864 | xfs_ioend_t **ioendp, |
1da177e4 LT |
865 | struct writeback_control *wbc, |
866 | int startio, | |
867 | int all_bh, | |
868 | pgoff_t tlast) | |
869 | { | |
10ce4444 CH |
870 | struct pagevec pvec; |
871 | int done = 0, i; | |
1da177e4 | 872 | |
10ce4444 CH |
873 | pagevec_init(&pvec, 0); |
874 | while (!done && tindex <= tlast) { | |
875 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
876 | ||
877 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 878 | break; |
10ce4444 CH |
879 | |
880 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
881 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
882 | iomapp, ioendp, wbc, startio, all_bh); | |
883 | if (done) | |
884 | break; | |
885 | } | |
886 | ||
887 | pagevec_release(&pvec); | |
888 | cond_resched(); | |
1da177e4 LT |
889 | } |
890 | } | |
891 | ||
892 | /* | |
893 | * Calling this without startio set means we are being asked to make a dirty | |
894 | * page ready for freeing it's buffers. When called with startio set then | |
895 | * we are coming from writepage. | |
896 | * | |
897 | * When called with startio set it is important that we write the WHOLE | |
898 | * page if possible. | |
899 | * The bh->b_state's cannot know if any of the blocks or which block for | |
900 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 901 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
902 | * may clear the page dirty flag prior to calling write page, under the |
903 | * assumption the entire page will be written out; by not writing out the | |
904 | * whole page the page can be reused before all valid dirty data is | |
905 | * written out. Note: in the case of a page that has been dirty'd by | |
906 | * mapwrite and but partially setup by block_prepare_write the | |
907 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
908 | * valid state, thus the whole page must be written out thing. | |
909 | */ | |
910 | ||
911 | STATIC int | |
912 | xfs_page_state_convert( | |
913 | struct inode *inode, | |
914 | struct page *page, | |
915 | struct writeback_control *wbc, | |
916 | int startio, | |
917 | int unmapped) /* also implies page uptodate */ | |
918 | { | |
f6d6d4fc | 919 | struct buffer_head *bh, *head; |
1defeac9 | 920 | xfs_iomap_t iomap; |
f6d6d4fc | 921 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
922 | loff_t offset; |
923 | unsigned long p_offset = 0; | |
f6d6d4fc | 924 | unsigned int type; |
1da177e4 LT |
925 | __uint64_t end_offset; |
926 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
927 | ssize_t size, len; |
928 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
929 | int page_dirty, count = 0; |
930 | int trylock = 0; | |
6c4fe19f | 931 | int all_bh = unmapped; |
1da177e4 | 932 | |
8272145c NS |
933 | if (startio) { |
934 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
935 | trylock |= BMAPI_TRYLOCK; | |
936 | } | |
3ba0815a | 937 | |
1da177e4 LT |
938 | /* Is this page beyond the end of the file? */ |
939 | offset = i_size_read(inode); | |
940 | end_index = offset >> PAGE_CACHE_SHIFT; | |
941 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
942 | if (page->index >= end_index) { | |
943 | if ((page->index >= end_index + 1) || | |
944 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
945 | if (startio) |
946 | unlock_page(page); | |
947 | return 0; | |
1da177e4 LT |
948 | } |
949 | } | |
950 | ||
1da177e4 | 951 | /* |
24e17b5f | 952 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 953 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
954 | * |
955 | * Derivation: | |
956 | * | |
957 | * End offset is the highest offset that this page should represent. | |
958 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
959 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
960 | * hence give us the correct page_dirty count. On any other page, | |
961 | * it will be zero and in that case we need page_dirty to be the | |
962 | * count of buffers on the page. | |
963 | */ | |
964 | end_offset = min_t(unsigned long long, | |
965 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 966 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
967 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
968 | PAGE_CACHE_SIZE); | |
969 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
970 | page_dirty = p_offset / len; |
971 | ||
24e17b5f | 972 | bh = head = page_buffers(page); |
f6d6d4fc | 973 | offset = page_offset(page); |
df3c7244 DC |
974 | flags = BMAPI_READ; |
975 | type = IOMAP_NEW; | |
f6d6d4fc | 976 | |
f6d6d4fc | 977 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
978 | |
979 | do { | |
980 | if (offset >= end_offset) | |
981 | break; | |
982 | if (!buffer_uptodate(bh)) | |
983 | uptodate = 0; | |
f6d6d4fc | 984 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
985 | /* |
986 | * the iomap is actually still valid, but the ioend | |
987 | * isn't. shouldn't happen too often. | |
988 | */ | |
989 | iomap_valid = 0; | |
1da177e4 | 990 | continue; |
f6d6d4fc | 991 | } |
1da177e4 | 992 | |
1defeac9 CH |
993 | if (iomap_valid) |
994 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
995 | |
996 | /* | |
997 | * First case, map an unwritten extent and prepare for | |
998 | * extent state conversion transaction on completion. | |
f6d6d4fc | 999 | * |
1da177e4 LT |
1000 | * Second case, allocate space for a delalloc buffer. |
1001 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1002 | * |
1003 | * Third case, an unmapped buffer was found, and we are | |
1004 | * in a path where we need to write the whole page out. | |
df3c7244 | 1005 | */ |
d5cb48aa CH |
1006 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1007 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1008 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1009 | int new_ioend = 0; |
1010 | ||
df3c7244 | 1011 | /* |
6c4fe19f CH |
1012 | * Make sure we don't use a read-only iomap |
1013 | */ | |
df3c7244 | 1014 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1015 | iomap_valid = 0; |
1016 | ||
f6d6d4fc CH |
1017 | if (buffer_unwritten(bh)) { |
1018 | type = IOMAP_UNWRITTEN; | |
8272145c | 1019 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1020 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1021 | type = IOMAP_DELAY; |
8272145c | 1022 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1023 | } else { |
6c4fe19f | 1024 | type = IOMAP_NEW; |
8272145c | 1025 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1026 | } |
1027 | ||
1defeac9 | 1028 | if (!iomap_valid) { |
effd120e DC |
1029 | /* |
1030 | * if we didn't have a valid mapping then we | |
1031 | * need to ensure that we put the new mapping | |
1032 | * in a new ioend structure. This needs to be | |
1033 | * done to ensure that the ioends correctly | |
1034 | * reflect the block mappings at io completion | |
1035 | * for unwritten extent conversion. | |
1036 | */ | |
1037 | new_ioend = 1; | |
6c4fe19f CH |
1038 | if (type == IOMAP_NEW) { |
1039 | size = xfs_probe_cluster(inode, | |
1040 | page, bh, head, 0); | |
d5cb48aa CH |
1041 | } else { |
1042 | size = len; | |
1043 | } | |
1044 | ||
1045 | err = xfs_map_blocks(inode, offset, size, | |
1046 | &iomap, flags); | |
f6d6d4fc | 1047 | if (err) |
1da177e4 | 1048 | goto error; |
1defeac9 | 1049 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1050 | } |
1defeac9 CH |
1051 | if (iomap_valid) { |
1052 | xfs_map_at_offset(bh, offset, | |
1053 | inode->i_blkbits, &iomap); | |
1da177e4 | 1054 | if (startio) { |
7336cea8 | 1055 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1056 | type, &ioend, |
effd120e | 1057 | new_ioend); |
1da177e4 LT |
1058 | } else { |
1059 | set_buffer_dirty(bh); | |
1060 | unlock_buffer(bh); | |
1061 | mark_buffer_dirty(bh); | |
1062 | } | |
1063 | page_dirty--; | |
f6d6d4fc | 1064 | count++; |
1da177e4 | 1065 | } |
d5cb48aa | 1066 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1067 | /* |
1068 | * we got here because the buffer is already mapped. | |
1069 | * That means it must already have extents allocated | |
1070 | * underneath it. Map the extent by reading it. | |
1071 | */ | |
df3c7244 | 1072 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1073 | flags = BMAPI_READ; |
1074 | size = xfs_probe_cluster(inode, page, bh, | |
1075 | head, 1); | |
1076 | err = xfs_map_blocks(inode, offset, size, | |
1077 | &iomap, flags); | |
1078 | if (err) | |
1079 | goto error; | |
1080 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1081 | } | |
d5cb48aa | 1082 | |
df3c7244 DC |
1083 | /* |
1084 | * We set the type to IOMAP_NEW in case we are doing a | |
1085 | * small write at EOF that is extending the file but | |
1086 | * without needing an allocation. We need to update the | |
1087 | * file size on I/O completion in this case so it is | |
1088 | * the same case as having just allocated a new extent | |
1089 | * that we are writing into for the first time. | |
1090 | */ | |
1091 | type = IOMAP_NEW; | |
d5cb48aa CH |
1092 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
1093 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
1094 | if (iomap_valid) |
1095 | all_bh = 1; | |
7336cea8 | 1096 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1097 | &ioend, !iomap_valid); |
1098 | page_dirty--; | |
1099 | count++; | |
f6d6d4fc | 1100 | } else { |
1defeac9 | 1101 | iomap_valid = 0; |
1da177e4 | 1102 | } |
d5cb48aa CH |
1103 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1104 | (unmapped || startio)) { | |
1105 | iomap_valid = 0; | |
1da177e4 | 1106 | } |
f6d6d4fc CH |
1107 | |
1108 | if (!iohead) | |
1109 | iohead = ioend; | |
1110 | ||
1111 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1112 | |
1113 | if (uptodate && bh == head) | |
1114 | SetPageUptodate(page); | |
1115 | ||
f6d6d4fc CH |
1116 | if (startio) |
1117 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1118 | |
1defeac9 CH |
1119 | if (ioend && iomap_valid) { |
1120 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1121 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1122 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1123 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1124 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1125 | } |
1126 | ||
f6d6d4fc CH |
1127 | if (iohead) |
1128 | xfs_submit_ioend(iohead); | |
1129 | ||
1da177e4 LT |
1130 | return page_dirty; |
1131 | ||
1132 | error: | |
f6d6d4fc CH |
1133 | if (iohead) |
1134 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1135 | |
1136 | /* | |
1137 | * If it's delalloc and we have nowhere to put it, | |
1138 | * throw it away, unless the lower layers told | |
1139 | * us to try again. | |
1140 | */ | |
1141 | if (err != -EAGAIN) { | |
f6d6d4fc | 1142 | if (!unmapped) |
1da177e4 | 1143 | block_invalidatepage(page, 0); |
1da177e4 LT |
1144 | ClearPageUptodate(page); |
1145 | } | |
1146 | return err; | |
1147 | } | |
1148 | ||
f51623b2 NS |
1149 | /* |
1150 | * writepage: Called from one of two places: | |
1151 | * | |
1152 | * 1. we are flushing a delalloc buffer head. | |
1153 | * | |
1154 | * 2. we are writing out a dirty page. Typically the page dirty | |
1155 | * state is cleared before we get here. In this case is it | |
1156 | * conceivable we have no buffer heads. | |
1157 | * | |
1158 | * For delalloc space on the page we need to allocate space and | |
1159 | * flush it. For unmapped buffer heads on the page we should | |
1160 | * allocate space if the page is uptodate. For any other dirty | |
1161 | * buffer heads on the page we should flush them. | |
1162 | * | |
1163 | * If we detect that a transaction would be required to flush | |
1164 | * the page, we have to check the process flags first, if we | |
1165 | * are already in a transaction or disk I/O during allocations | |
1166 | * is off, we need to fail the writepage and redirty the page. | |
1167 | */ | |
1168 | ||
1169 | STATIC int | |
e4c573bb | 1170 | xfs_vm_writepage( |
f51623b2 NS |
1171 | struct page *page, |
1172 | struct writeback_control *wbc) | |
1173 | { | |
1174 | int error; | |
1175 | int need_trans; | |
1176 | int delalloc, unmapped, unwritten; | |
1177 | struct inode *inode = page->mapping->host; | |
1178 | ||
1179 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1180 | ||
1181 | /* | |
1182 | * We need a transaction if: | |
1183 | * 1. There are delalloc buffers on the page | |
1184 | * 2. The page is uptodate and we have unmapped buffers | |
1185 | * 3. The page is uptodate and we have no buffers | |
1186 | * 4. There are unwritten buffers on the page | |
1187 | */ | |
1188 | ||
1189 | if (!page_has_buffers(page)) { | |
1190 | unmapped = 1; | |
1191 | need_trans = 1; | |
1192 | } else { | |
1193 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1194 | if (!PageUptodate(page)) | |
1195 | unmapped = 0; | |
1196 | need_trans = delalloc + unmapped + unwritten; | |
1197 | } | |
1198 | ||
1199 | /* | |
1200 | * If we need a transaction and the process flags say | |
1201 | * we are already in a transaction, or no IO is allowed | |
1202 | * then mark the page dirty again and leave the page | |
1203 | * as is. | |
1204 | */ | |
59c1b082 | 1205 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1206 | goto out_fail; |
1207 | ||
1208 | /* | |
1209 | * Delay hooking up buffer heads until we have | |
1210 | * made our go/no-go decision. | |
1211 | */ | |
1212 | if (!page_has_buffers(page)) | |
1213 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1214 | ||
1215 | /* | |
1216 | * Convert delayed allocate, unwritten or unmapped space | |
1217 | * to real space and flush out to disk. | |
1218 | */ | |
1219 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1220 | if (error == -EAGAIN) | |
1221 | goto out_fail; | |
1222 | if (unlikely(error < 0)) | |
1223 | goto out_unlock; | |
1224 | ||
1225 | return 0; | |
1226 | ||
1227 | out_fail: | |
1228 | redirty_page_for_writepage(wbc, page); | |
1229 | unlock_page(page); | |
1230 | return 0; | |
1231 | out_unlock: | |
1232 | unlock_page(page); | |
1233 | return error; | |
1234 | } | |
1235 | ||
7d4fb40a NS |
1236 | STATIC int |
1237 | xfs_vm_writepages( | |
1238 | struct address_space *mapping, | |
1239 | struct writeback_control *wbc) | |
1240 | { | |
67fcaa73 | 1241 | struct bhv_vnode *vp = vn_from_inode(mapping->host); |
7d4fb40a NS |
1242 | |
1243 | if (VN_TRUNC(vp)) | |
1244 | VUNTRUNCATE(vp); | |
1245 | return generic_writepages(mapping, wbc); | |
1246 | } | |
1247 | ||
f51623b2 NS |
1248 | /* |
1249 | * Called to move a page into cleanable state - and from there | |
1250 | * to be released. Possibly the page is already clean. We always | |
1251 | * have buffer heads in this call. | |
1252 | * | |
1253 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1254 | * | |
1255 | * Possible scenarios are: | |
1256 | * | |
1257 | * 1. We are being called to release a page which has been written | |
1258 | * to via regular I/O. buffer heads will be dirty and possibly | |
1259 | * delalloc. If no delalloc buffer heads in this case then we | |
1260 | * can just return zero. | |
1261 | * | |
1262 | * 2. We are called to release a page which has been written via | |
1263 | * mmap, all we need to do is ensure there is no delalloc | |
1264 | * state in the buffer heads, if not we can let the caller | |
1265 | * free them and we should come back later via writepage. | |
1266 | */ | |
1267 | STATIC int | |
238f4c54 | 1268 | xfs_vm_releasepage( |
f51623b2 NS |
1269 | struct page *page, |
1270 | gfp_t gfp_mask) | |
1271 | { | |
1272 | struct inode *inode = page->mapping->host; | |
1273 | int dirty, delalloc, unmapped, unwritten; | |
1274 | struct writeback_control wbc = { | |
1275 | .sync_mode = WB_SYNC_ALL, | |
1276 | .nr_to_write = 1, | |
1277 | }; | |
1278 | ||
ed9d88f7 | 1279 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0); |
f51623b2 | 1280 | |
238f4c54 NS |
1281 | if (!page_has_buffers(page)) |
1282 | return 0; | |
1283 | ||
f51623b2 NS |
1284 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1285 | if (!delalloc && !unwritten) | |
1286 | goto free_buffers; | |
1287 | ||
1288 | if (!(gfp_mask & __GFP_FS)) | |
1289 | return 0; | |
1290 | ||
1291 | /* If we are already inside a transaction or the thread cannot | |
1292 | * do I/O, we cannot release this page. | |
1293 | */ | |
59c1b082 | 1294 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1295 | return 0; |
1296 | ||
1297 | /* | |
1298 | * Convert delalloc space to real space, do not flush the | |
1299 | * data out to disk, that will be done by the caller. | |
1300 | * Never need to allocate space here - we will always | |
1301 | * come back to writepage in that case. | |
1302 | */ | |
1303 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1304 | if (dirty == 0 && !unwritten) | |
1305 | goto free_buffers; | |
1306 | return 0; | |
1307 | ||
1308 | free_buffers: | |
1309 | return try_to_free_buffers(page); | |
1310 | } | |
1311 | ||
1da177e4 | 1312 | STATIC int |
c2536668 | 1313 | __xfs_get_blocks( |
1da177e4 LT |
1314 | struct inode *inode, |
1315 | sector_t iblock, | |
1da177e4 LT |
1316 | struct buffer_head *bh_result, |
1317 | int create, | |
1318 | int direct, | |
1319 | bmapi_flags_t flags) | |
1320 | { | |
67fcaa73 | 1321 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 1322 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1323 | xfs_off_t offset; |
1324 | ssize_t size; | |
c2536668 | 1325 | int niomap = 1; |
1da177e4 | 1326 | int error; |
1da177e4 | 1327 | |
fdc7ed75 | 1328 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1329 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1330 | size = bh_result->b_size; | |
67fcaa73 NS |
1331 | error = bhv_vop_bmap(vp, offset, size, |
1332 | create ? flags : BMAPI_READ, &iomap, &niomap); | |
1da177e4 LT |
1333 | if (error) |
1334 | return -error; | |
c2536668 | 1335 | if (niomap == 0) |
1da177e4 LT |
1336 | return 0; |
1337 | ||
1338 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1339 | /* |
1340 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1341 | * the read case (treat as if we're reading into a hole). |
1342 | */ | |
1343 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1344 | xfs_map_buffer(bh_result, &iomap, offset, |
1345 | inode->i_blkbits); | |
1da177e4 LT |
1346 | } |
1347 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1348 | if (direct) | |
1349 | bh_result->b_private = inode; | |
1350 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1351 | } |
1352 | } | |
1353 | ||
c2536668 NS |
1354 | /* |
1355 | * If this is a realtime file, data may be on a different device. | |
1356 | * to that pointed to from the buffer_head b_bdev currently. | |
1357 | */ | |
ce8e922c | 1358 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1359 | |
c2536668 | 1360 | /* |
549054af DC |
1361 | * If we previously allocated a block out beyond eof and we are now |
1362 | * coming back to use it then we will need to flag it as new even if it | |
1363 | * has a disk address. | |
1364 | * | |
1365 | * With sub-block writes into unwritten extents we also need to mark | |
1366 | * the buffer as new so that the unwritten parts of the buffer gets | |
1367 | * correctly zeroed. | |
1da177e4 LT |
1368 | */ |
1369 | if (create && | |
1370 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1371 | (offset >= i_size_read(inode)) || |
1372 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1373 | set_buffer_new(bh_result); |
1da177e4 LT |
1374 | |
1375 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1376 | BUG_ON(direct); | |
1377 | if (create) { | |
1378 | set_buffer_uptodate(bh_result); | |
1379 | set_buffer_mapped(bh_result); | |
1380 | set_buffer_delay(bh_result); | |
1381 | } | |
1382 | } | |
1383 | ||
c2536668 | 1384 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1385 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1386 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1387 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1388 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1389 | } |
1390 | ||
1391 | return 0; | |
1392 | } | |
1393 | ||
1394 | int | |
c2536668 | 1395 | xfs_get_blocks( |
1da177e4 LT |
1396 | struct inode *inode, |
1397 | sector_t iblock, | |
1398 | struct buffer_head *bh_result, | |
1399 | int create) | |
1400 | { | |
c2536668 | 1401 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1402 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1403 | } |
1404 | ||
1405 | STATIC int | |
e4c573bb | 1406 | xfs_get_blocks_direct( |
1da177e4 LT |
1407 | struct inode *inode, |
1408 | sector_t iblock, | |
1da177e4 LT |
1409 | struct buffer_head *bh_result, |
1410 | int create) | |
1411 | { | |
c2536668 | 1412 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1413 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1414 | } |
1415 | ||
f0973863 | 1416 | STATIC void |
e4c573bb | 1417 | xfs_end_io_direct( |
f0973863 CH |
1418 | struct kiocb *iocb, |
1419 | loff_t offset, | |
1420 | ssize_t size, | |
1421 | void *private) | |
1422 | { | |
1423 | xfs_ioend_t *ioend = iocb->private; | |
1424 | ||
1425 | /* | |
1426 | * Non-NULL private data means we need to issue a transaction to | |
1427 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1428 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1429 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1430 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1431 | * it anyway to keep the code uniform and simpler. |
1432 | * | |
e927af90 DC |
1433 | * Well, if only it were that simple. Because synchronous direct I/O |
1434 | * requires extent conversion to occur *before* we return to userspace, | |
1435 | * we have to wait for extent conversion to complete. Look at the | |
1436 | * iocb that has been passed to us to determine if this is AIO or | |
1437 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1438 | * workqueue and wait for it to complete. | |
1439 | * | |
f0973863 CH |
1440 | * The core direct I/O code might be changed to always call the |
1441 | * completion handler in the future, in which case all this can | |
1442 | * go away. | |
1443 | */ | |
ba87ea69 LM |
1444 | ioend->io_offset = offset; |
1445 | ioend->io_size = size; | |
1446 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1447 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1448 | } else if (private && size > 0) { |
e927af90 | 1449 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1450 | } else { |
ba87ea69 LM |
1451 | /* |
1452 | * A direct I/O write ioend starts it's life in unwritten | |
1453 | * state in case they map an unwritten extent. This write | |
1454 | * didn't map an unwritten extent so switch it's completion | |
1455 | * handler. | |
1456 | */ | |
1457 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); | |
e927af90 | 1458 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1459 | } |
1460 | ||
1461 | /* | |
c41564b5 | 1462 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1463 | * completion handler was called. Thus we need to protect |
1464 | * against double-freeing. | |
1465 | */ | |
1466 | iocb->private = NULL; | |
1467 | } | |
1468 | ||
1da177e4 | 1469 | STATIC ssize_t |
e4c573bb | 1470 | xfs_vm_direct_IO( |
1da177e4 LT |
1471 | int rw, |
1472 | struct kiocb *iocb, | |
1473 | const struct iovec *iov, | |
1474 | loff_t offset, | |
1475 | unsigned long nr_segs) | |
1476 | { | |
1477 | struct file *file = iocb->ki_filp; | |
1478 | struct inode *inode = file->f_mapping->host; | |
67fcaa73 | 1479 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 LT |
1480 | xfs_iomap_t iomap; |
1481 | int maps = 1; | |
1482 | int error; | |
f0973863 | 1483 | ssize_t ret; |
1da177e4 | 1484 | |
67fcaa73 | 1485 | error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps); |
1da177e4 LT |
1486 | if (error) |
1487 | return -error; | |
1488 | ||
721259bc | 1489 | if (rw == WRITE) { |
ba87ea69 | 1490 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
721259bc LM |
1491 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, |
1492 | iomap.iomap_target->bt_bdev, | |
1493 | iov, offset, nr_segs, | |
1494 | xfs_get_blocks_direct, | |
1495 | xfs_end_io_direct); | |
1496 | } else { | |
ba87ea69 | 1497 | iocb->private = xfs_alloc_ioend(inode, IOMAP_READ); |
721259bc LM |
1498 | ret = blockdev_direct_IO_no_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 | } | |
f0973863 | 1504 | |
8459d86a | 1505 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1506 | xfs_destroy_ioend(iocb->private); |
1507 | return ret; | |
1da177e4 LT |
1508 | } |
1509 | ||
f51623b2 | 1510 | STATIC int |
e4c573bb | 1511 | xfs_vm_prepare_write( |
f51623b2 NS |
1512 | struct file *file, |
1513 | struct page *page, | |
1514 | unsigned int from, | |
1515 | unsigned int to) | |
1516 | { | |
c2536668 | 1517 | return block_prepare_write(page, from, to, xfs_get_blocks); |
f51623b2 | 1518 | } |
1da177e4 LT |
1519 | |
1520 | STATIC sector_t | |
e4c573bb | 1521 | xfs_vm_bmap( |
1da177e4 LT |
1522 | struct address_space *mapping, |
1523 | sector_t block) | |
1524 | { | |
1525 | struct inode *inode = (struct inode *)mapping->host; | |
67fcaa73 | 1526 | bhv_vnode_t *vp = vn_from_inode(inode); |
1da177e4 | 1527 | |
e4c573bb | 1528 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
67fcaa73 NS |
1529 | bhv_vop_rwlock(vp, VRWLOCK_READ); |
1530 | bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF); | |
1531 | bhv_vop_rwunlock(vp, VRWLOCK_READ); | |
c2536668 | 1532 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1533 | } |
1534 | ||
1535 | STATIC int | |
e4c573bb | 1536 | xfs_vm_readpage( |
1da177e4 LT |
1537 | struct file *unused, |
1538 | struct page *page) | |
1539 | { | |
c2536668 | 1540 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1541 | } |
1542 | ||
1543 | STATIC int | |
e4c573bb | 1544 | xfs_vm_readpages( |
1da177e4 LT |
1545 | struct file *unused, |
1546 | struct address_space *mapping, | |
1547 | struct list_head *pages, | |
1548 | unsigned nr_pages) | |
1549 | { | |
c2536668 | 1550 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1551 | } |
1552 | ||
2ff28e22 | 1553 | STATIC void |
238f4c54 | 1554 | xfs_vm_invalidatepage( |
bcec2b7f NS |
1555 | struct page *page, |
1556 | unsigned long offset) | |
1557 | { | |
1558 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1559 | page->mapping->host, page, offset); | |
2ff28e22 | 1560 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1561 | } |
1562 | ||
f5e54d6e | 1563 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1564 | .readpage = xfs_vm_readpage, |
1565 | .readpages = xfs_vm_readpages, | |
1566 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1567 | .writepages = xfs_vm_writepages, |
1da177e4 | 1568 | .sync_page = block_sync_page, |
238f4c54 NS |
1569 | .releasepage = xfs_vm_releasepage, |
1570 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb | 1571 | .prepare_write = xfs_vm_prepare_write, |
1da177e4 | 1572 | .commit_write = generic_commit_write, |
e4c573bb NS |
1573 | .bmap = xfs_vm_bmap, |
1574 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1575 | .migratepage = buffer_migrate_page, |
1da177e4 | 1576 | }; |