]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_bit.h" |
1da177e4 | 20 | #include "xfs_log.h" |
a844f451 | 21 | #include "xfs_inum.h" |
1da177e4 | 22 | #include "xfs_sb.h" |
a844f451 | 23 | #include "xfs_ag.h" |
1da177e4 LT |
24 | #include "xfs_dir2.h" |
25 | #include "xfs_trans.h" | |
26 | #include "xfs_dmapi.h" | |
27 | #include "xfs_mount.h" | |
28 | #include "xfs_bmap_btree.h" | |
29 | #include "xfs_alloc_btree.h" | |
30 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 31 | #include "xfs_dir2_sf.h" |
a844f451 | 32 | #include "xfs_attr_sf.h" |
1da177e4 LT |
33 | #include "xfs_dinode.h" |
34 | #include "xfs_inode.h" | |
a844f451 NS |
35 | #include "xfs_alloc.h" |
36 | #include "xfs_btree.h" | |
1da177e4 LT |
37 | #include "xfs_error.h" |
38 | #include "xfs_rw.h" | |
39 | #include "xfs_iomap.h" | |
739bfb2a | 40 | #include "xfs_vnodeops.h" |
1da177e4 | 41 | #include <linux/mpage.h> |
10ce4444 | 42 | #include <linux/pagevec.h> |
1da177e4 LT |
43 | #include <linux/writeback.h> |
44 | ||
25e41b3d CH |
45 | |
46 | /* | |
47 | * Prime number of hash buckets since address is used as the key. | |
48 | */ | |
49 | #define NVSYNC 37 | |
50 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
51 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
52 | ||
53 | void __init | |
54 | xfs_ioend_init(void) | |
55 | { | |
56 | int i; | |
57 | ||
58 | for (i = 0; i < NVSYNC; i++) | |
59 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
60 | } | |
61 | ||
62 | void | |
63 | xfs_ioend_wait( | |
64 | xfs_inode_t *ip) | |
65 | { | |
66 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
67 | ||
68 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
69 | } | |
70 | ||
71 | STATIC void | |
72 | xfs_ioend_wake( | |
73 | xfs_inode_t *ip) | |
74 | { | |
75 | if (atomic_dec_and_test(&ip->i_iocount)) | |
76 | wake_up(to_ioend_wq(ip)); | |
77 | } | |
78 | ||
f51623b2 NS |
79 | STATIC void |
80 | xfs_count_page_state( | |
81 | struct page *page, | |
82 | int *delalloc, | |
83 | int *unmapped, | |
84 | int *unwritten) | |
85 | { | |
86 | struct buffer_head *bh, *head; | |
87 | ||
88 | *delalloc = *unmapped = *unwritten = 0; | |
89 | ||
90 | bh = head = page_buffers(page); | |
91 | do { | |
92 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
93 | (*unmapped) = 1; | |
f51623b2 NS |
94 | else if (buffer_unwritten(bh)) |
95 | (*unwritten) = 1; | |
96 | else if (buffer_delay(bh)) | |
97 | (*delalloc) = 1; | |
98 | } while ((bh = bh->b_this_page) != head); | |
99 | } | |
100 | ||
1da177e4 LT |
101 | #if defined(XFS_RW_TRACE) |
102 | void | |
103 | xfs_page_trace( | |
104 | int tag, | |
105 | struct inode *inode, | |
106 | struct page *page, | |
ed9d88f7 | 107 | unsigned long pgoff) |
1da177e4 LT |
108 | { |
109 | xfs_inode_t *ip; | |
1da177e4 | 110 | loff_t isize = i_size_read(inode); |
f6d6d4fc | 111 | loff_t offset = page_offset(page); |
1da177e4 LT |
112 | int delalloc = -1, unmapped = -1, unwritten = -1; |
113 | ||
114 | if (page_has_buffers(page)) | |
115 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
116 | ||
e6064d30 | 117 | ip = XFS_I(inode); |
1da177e4 LT |
118 | if (!ip->i_rwtrace) |
119 | return; | |
120 | ||
121 | ktrace_enter(ip->i_rwtrace, | |
122 | (void *)((unsigned long)tag), | |
123 | (void *)ip, | |
124 | (void *)inode, | |
125 | (void *)page, | |
ed9d88f7 | 126 | (void *)pgoff, |
1da177e4 LT |
127 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), |
128 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
129 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
130 | (void *)((unsigned long)(isize & 0xffffffff)), | |
131 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
132 | (void *)((unsigned long)(offset & 0xffffffff)), | |
133 | (void *)((unsigned long)delalloc), | |
134 | (void *)((unsigned long)unmapped), | |
135 | (void *)((unsigned long)unwritten), | |
f1fdc848 | 136 | (void *)((unsigned long)current_pid()), |
1da177e4 LT |
137 | (void *)NULL); |
138 | } | |
139 | #else | |
ed9d88f7 | 140 | #define xfs_page_trace(tag, inode, page, pgoff) |
1da177e4 LT |
141 | #endif |
142 | ||
6214ed44 CH |
143 | STATIC struct block_device * |
144 | xfs_find_bdev_for_inode( | |
145 | struct xfs_inode *ip) | |
146 | { | |
147 | struct xfs_mount *mp = ip->i_mount; | |
148 | ||
71ddabb9 | 149 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
150 | return mp->m_rtdev_targp->bt_bdev; |
151 | else | |
152 | return mp->m_ddev_targp->bt_bdev; | |
153 | } | |
154 | ||
f6d6d4fc CH |
155 | /* |
156 | * We're now finished for good with this ioend structure. | |
157 | * Update the page state via the associated buffer_heads, | |
158 | * release holds on the inode and bio, and finally free | |
159 | * up memory. Do not use the ioend after this. | |
160 | */ | |
0829c360 CH |
161 | STATIC void |
162 | xfs_destroy_ioend( | |
163 | xfs_ioend_t *ioend) | |
164 | { | |
f6d6d4fc | 165 | struct buffer_head *bh, *next; |
583fa586 | 166 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
167 | |
168 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
169 | next = bh->b_private; | |
7d04a335 | 170 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 171 | } |
583fa586 CH |
172 | |
173 | /* | |
174 | * Volume managers supporting multiple paths can send back ENODEV | |
175 | * when the final path disappears. In this case continuing to fill | |
176 | * the page cache with dirty data which cannot be written out is | |
177 | * evil, so prevent that. | |
178 | */ | |
179 | if (unlikely(ioend->io_error == -ENODEV)) { | |
180 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
181 | __FILE__, __LINE__); | |
b677c210 | 182 | } |
583fa586 | 183 | |
25e41b3d | 184 | xfs_ioend_wake(ip); |
0829c360 CH |
185 | mempool_free(ioend, xfs_ioend_pool); |
186 | } | |
187 | ||
932640e8 DC |
188 | /* |
189 | * If the end of the current ioend is beyond the current EOF, | |
190 | * return the new EOF value, otherwise zero. | |
191 | */ | |
192 | STATIC xfs_fsize_t | |
193 | xfs_ioend_new_eof( | |
194 | xfs_ioend_t *ioend) | |
195 | { | |
196 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
197 | xfs_fsize_t isize; | |
198 | xfs_fsize_t bsize; | |
199 | ||
200 | bsize = ioend->io_offset + ioend->io_size; | |
201 | isize = MAX(ip->i_size, ip->i_new_size); | |
202 | isize = MIN(isize, bsize); | |
203 | return isize > ip->i_d.di_size ? isize : 0; | |
204 | } | |
205 | ||
ba87ea69 LM |
206 | /* |
207 | * Update on-disk file size now that data has been written to disk. | |
208 | * The current in-memory file size is i_size. If a write is beyond | |
613d7043 | 209 | * eof i_new_size will be the intended file size until i_size is |
ba87ea69 LM |
210 | * updated. If this write does not extend all the way to the valid |
211 | * file size then restrict this update to the end of the write. | |
212 | */ | |
932640e8 | 213 | |
ba87ea69 LM |
214 | STATIC void |
215 | xfs_setfilesize( | |
216 | xfs_ioend_t *ioend) | |
217 | { | |
b677c210 | 218 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 219 | xfs_fsize_t isize; |
ba87ea69 | 220 | |
ba87ea69 LM |
221 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
222 | ASSERT(ioend->io_type != IOMAP_READ); | |
223 | ||
224 | if (unlikely(ioend->io_error)) | |
225 | return; | |
226 | ||
ba87ea69 | 227 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
932640e8 DC |
228 | isize = xfs_ioend_new_eof(ioend); |
229 | if (isize) { | |
ba87ea69 | 230 | ip->i_d.di_size = isize; |
94b97e39 | 231 | xfs_mark_inode_dirty_sync(ip); |
ba87ea69 LM |
232 | } |
233 | ||
234 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
235 | } | |
236 | ||
0829c360 | 237 | /* |
f6d6d4fc | 238 | * Buffered IO write completion for delayed allocate extents. |
f6d6d4fc CH |
239 | */ |
240 | STATIC void | |
241 | xfs_end_bio_delalloc( | |
c4028958 | 242 | struct work_struct *work) |
f6d6d4fc | 243 | { |
c4028958 DH |
244 | xfs_ioend_t *ioend = |
245 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 246 | |
ba87ea69 | 247 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
248 | xfs_destroy_ioend(ioend); |
249 | } | |
250 | ||
251 | /* | |
252 | * Buffered IO write completion for regular, written extents. | |
253 | */ | |
254 | STATIC void | |
255 | xfs_end_bio_written( | |
c4028958 | 256 | struct work_struct *work) |
f6d6d4fc | 257 | { |
c4028958 DH |
258 | xfs_ioend_t *ioend = |
259 | container_of(work, xfs_ioend_t, io_work); | |
f6d6d4fc | 260 | |
ba87ea69 | 261 | xfs_setfilesize(ioend); |
f6d6d4fc CH |
262 | xfs_destroy_ioend(ioend); |
263 | } | |
264 | ||
265 | /* | |
266 | * IO write completion for unwritten extents. | |
267 | * | |
0829c360 | 268 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 269 | * to written extents. |
0829c360 CH |
270 | */ |
271 | STATIC void | |
272 | xfs_end_bio_unwritten( | |
c4028958 | 273 | struct work_struct *work) |
0829c360 | 274 | { |
c4028958 DH |
275 | xfs_ioend_t *ioend = |
276 | container_of(work, xfs_ioend_t, io_work); | |
7642861b | 277 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
0829c360 CH |
278 | xfs_off_t offset = ioend->io_offset; |
279 | size_t size = ioend->io_size; | |
0829c360 | 280 | |
ba87ea69 | 281 | if (likely(!ioend->io_error)) { |
cc88466f DC |
282 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
283 | int error; | |
284 | error = xfs_iomap_write_unwritten(ip, offset, size); | |
285 | if (error) | |
286 | ioend->io_error = error; | |
287 | } | |
ba87ea69 LM |
288 | xfs_setfilesize(ioend); |
289 | } | |
290 | xfs_destroy_ioend(ioend); | |
291 | } | |
292 | ||
293 | /* | |
294 | * IO read completion for regular, written extents. | |
295 | */ | |
296 | STATIC void | |
297 | xfs_end_bio_read( | |
298 | struct work_struct *work) | |
299 | { | |
300 | xfs_ioend_t *ioend = | |
301 | container_of(work, xfs_ioend_t, io_work); | |
302 | ||
0829c360 CH |
303 | xfs_destroy_ioend(ioend); |
304 | } | |
305 | ||
c626d174 DC |
306 | /* |
307 | * Schedule IO completion handling on a xfsdatad if this was | |
308 | * the final hold on this ioend. If we are asked to wait, | |
309 | * flush the workqueue. | |
310 | */ | |
311 | STATIC void | |
312 | xfs_finish_ioend( | |
313 | xfs_ioend_t *ioend, | |
314 | int wait) | |
315 | { | |
316 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
317 | struct workqueue_struct *wq = xfsdatad_workqueue; | |
318 | if (ioend->io_work.func == xfs_end_bio_unwritten) | |
319 | wq = xfsconvertd_workqueue; | |
320 | ||
321 | queue_work(wq, &ioend->io_work); | |
322 | if (wait) | |
323 | flush_workqueue(wq); | |
324 | } | |
325 | } | |
326 | ||
0829c360 CH |
327 | /* |
328 | * Allocate and initialise an IO completion structure. | |
329 | * We need to track unwritten extent write completion here initially. | |
330 | * We'll need to extend this for updating the ondisk inode size later | |
331 | * (vs. incore size). | |
332 | */ | |
333 | STATIC xfs_ioend_t * | |
334 | xfs_alloc_ioend( | |
f6d6d4fc CH |
335 | struct inode *inode, |
336 | unsigned int type) | |
0829c360 CH |
337 | { |
338 | xfs_ioend_t *ioend; | |
339 | ||
340 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
341 | ||
342 | /* | |
343 | * Set the count to 1 initially, which will prevent an I/O | |
344 | * completion callback from happening before we have started | |
345 | * all the I/O from calling the completion routine too early. | |
346 | */ | |
347 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 348 | ioend->io_error = 0; |
f6d6d4fc CH |
349 | ioend->io_list = NULL; |
350 | ioend->io_type = type; | |
b677c210 | 351 | ioend->io_inode = inode; |
c1a073bd | 352 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 353 | ioend->io_buffer_tail = NULL; |
b677c210 | 354 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
355 | ioend->io_offset = 0; |
356 | ioend->io_size = 0; | |
357 | ||
f6d6d4fc | 358 | if (type == IOMAP_UNWRITTEN) |
c4028958 | 359 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten); |
f6d6d4fc | 360 | else if (type == IOMAP_DELAY) |
c4028958 | 361 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc); |
ba87ea69 LM |
362 | else if (type == IOMAP_READ) |
363 | INIT_WORK(&ioend->io_work, xfs_end_bio_read); | |
f6d6d4fc | 364 | else |
c4028958 | 365 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); |
0829c360 CH |
366 | |
367 | return ioend; | |
368 | } | |
369 | ||
1da177e4 LT |
370 | STATIC int |
371 | xfs_map_blocks( | |
372 | struct inode *inode, | |
373 | loff_t offset, | |
374 | ssize_t count, | |
375 | xfs_iomap_t *mapp, | |
376 | int flags) | |
377 | { | |
6bd16ff2 CH |
378 | int nmaps = 1; |
379 | ||
380 | return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps); | |
1da177e4 LT |
381 | } |
382 | ||
7989cb8e | 383 | STATIC_INLINE int |
1defeac9 | 384 | xfs_iomap_valid( |
1da177e4 | 385 | xfs_iomap_t *iomapp, |
1defeac9 | 386 | loff_t offset) |
1da177e4 | 387 | { |
1defeac9 CH |
388 | return offset >= iomapp->iomap_offset && |
389 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
390 | } |
391 | ||
f6d6d4fc CH |
392 | /* |
393 | * BIO completion handler for buffered IO. | |
394 | */ | |
782e3b3b | 395 | STATIC void |
f6d6d4fc CH |
396 | xfs_end_bio( |
397 | struct bio *bio, | |
f6d6d4fc CH |
398 | int error) |
399 | { | |
400 | xfs_ioend_t *ioend = bio->bi_private; | |
401 | ||
f6d6d4fc | 402 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 403 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
404 | |
405 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
406 | bio->bi_private = NULL; |
407 | bio->bi_end_io = NULL; | |
f6d6d4fc | 408 | bio_put(bio); |
7d04a335 | 409 | |
e927af90 | 410 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
411 | } |
412 | ||
413 | STATIC void | |
414 | xfs_submit_ioend_bio( | |
415 | xfs_ioend_t *ioend, | |
416 | struct bio *bio) | |
417 | { | |
418 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
419 | bio->bi_private = ioend; |
420 | bio->bi_end_io = xfs_end_bio; | |
421 | ||
932640e8 DC |
422 | /* |
423 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
424 | * but don't update the inode size until I/O completion. | |
425 | */ | |
426 | if (xfs_ioend_new_eof(ioend)) | |
427 | xfs_mark_inode_dirty_sync(XFS_I(ioend->io_inode)); | |
428 | ||
f6d6d4fc CH |
429 | submit_bio(WRITE, bio); |
430 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
431 | bio_put(bio); | |
432 | } | |
433 | ||
434 | STATIC struct bio * | |
435 | xfs_alloc_ioend_bio( | |
436 | struct buffer_head *bh) | |
437 | { | |
438 | struct bio *bio; | |
439 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
440 | ||
441 | do { | |
442 | bio = bio_alloc(GFP_NOIO, nvecs); | |
443 | nvecs >>= 1; | |
444 | } while (!bio); | |
445 | ||
446 | ASSERT(bio->bi_private == NULL); | |
447 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
448 | bio->bi_bdev = bh->b_bdev; | |
449 | bio_get(bio); | |
450 | return bio; | |
451 | } | |
452 | ||
453 | STATIC void | |
454 | xfs_start_buffer_writeback( | |
455 | struct buffer_head *bh) | |
456 | { | |
457 | ASSERT(buffer_mapped(bh)); | |
458 | ASSERT(buffer_locked(bh)); | |
459 | ASSERT(!buffer_delay(bh)); | |
460 | ASSERT(!buffer_unwritten(bh)); | |
461 | ||
462 | mark_buffer_async_write(bh); | |
463 | set_buffer_uptodate(bh); | |
464 | clear_buffer_dirty(bh); | |
465 | } | |
466 | ||
467 | STATIC void | |
468 | xfs_start_page_writeback( | |
469 | struct page *page, | |
f6d6d4fc CH |
470 | int clear_dirty, |
471 | int buffers) | |
472 | { | |
473 | ASSERT(PageLocked(page)); | |
474 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 475 | if (clear_dirty) |
92132021 DC |
476 | clear_page_dirty_for_io(page); |
477 | set_page_writeback(page); | |
f6d6d4fc | 478 | unlock_page(page); |
1f7decf6 FW |
479 | /* If no buffers on the page are to be written, finish it here */ |
480 | if (!buffers) | |
f6d6d4fc | 481 | end_page_writeback(page); |
f6d6d4fc CH |
482 | } |
483 | ||
484 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
485 | { | |
486 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
487 | } | |
488 | ||
489 | /* | |
d88992f6 DC |
490 | * Submit all of the bios for all of the ioends we have saved up, covering the |
491 | * initial writepage page and also any probed pages. | |
492 | * | |
493 | * Because we may have multiple ioends spanning a page, we need to start | |
494 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
495 | * buffers as we got, then we can end up with a page that only has buffers | |
496 | * marked async write and I/O complete on can occur before we mark the other | |
497 | * buffers async write. | |
498 | * | |
499 | * The end result of this is that we trip a bug in end_page_writeback() because | |
500 | * we call it twice for the one page as the code in end_buffer_async_write() | |
501 | * assumes that all buffers on the page are started at the same time. | |
502 | * | |
503 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 504 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
505 | */ |
506 | STATIC void | |
507 | xfs_submit_ioend( | |
508 | xfs_ioend_t *ioend) | |
509 | { | |
d88992f6 | 510 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
511 | xfs_ioend_t *next; |
512 | struct buffer_head *bh; | |
513 | struct bio *bio; | |
514 | sector_t lastblock = 0; | |
515 | ||
d88992f6 DC |
516 | /* Pass 1 - start writeback */ |
517 | do { | |
518 | next = ioend->io_list; | |
519 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
520 | xfs_start_buffer_writeback(bh); | |
521 | } | |
522 | } while ((ioend = next) != NULL); | |
523 | ||
524 | /* Pass 2 - submit I/O */ | |
525 | ioend = head; | |
f6d6d4fc CH |
526 | do { |
527 | next = ioend->io_list; | |
528 | bio = NULL; | |
529 | ||
530 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
531 | |
532 | if (!bio) { | |
533 | retry: | |
534 | bio = xfs_alloc_ioend_bio(bh); | |
535 | } else if (bh->b_blocknr != lastblock + 1) { | |
536 | xfs_submit_ioend_bio(ioend, bio); | |
537 | goto retry; | |
538 | } | |
539 | ||
540 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
541 | xfs_submit_ioend_bio(ioend, bio); | |
542 | goto retry; | |
543 | } | |
544 | ||
545 | lastblock = bh->b_blocknr; | |
546 | } | |
547 | if (bio) | |
548 | xfs_submit_ioend_bio(ioend, bio); | |
e927af90 | 549 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
550 | } while ((ioend = next) != NULL); |
551 | } | |
552 | ||
553 | /* | |
554 | * Cancel submission of all buffer_heads so far in this endio. | |
555 | * Toss the endio too. Only ever called for the initial page | |
556 | * in a writepage request, so only ever one page. | |
557 | */ | |
558 | STATIC void | |
559 | xfs_cancel_ioend( | |
560 | xfs_ioend_t *ioend) | |
561 | { | |
562 | xfs_ioend_t *next; | |
563 | struct buffer_head *bh, *next_bh; | |
564 | ||
565 | do { | |
566 | next = ioend->io_list; | |
567 | bh = ioend->io_buffer_head; | |
568 | do { | |
569 | next_bh = bh->b_private; | |
570 | clear_buffer_async_write(bh); | |
571 | unlock_buffer(bh); | |
572 | } while ((bh = next_bh) != NULL); | |
573 | ||
25e41b3d | 574 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
575 | mempool_free(ioend, xfs_ioend_pool); |
576 | } while ((ioend = next) != NULL); | |
577 | } | |
578 | ||
579 | /* | |
580 | * Test to see if we've been building up a completion structure for | |
581 | * earlier buffers -- if so, we try to append to this ioend if we | |
582 | * can, otherwise we finish off any current ioend and start another. | |
583 | * Return true if we've finished the given ioend. | |
584 | */ | |
585 | STATIC void | |
586 | xfs_add_to_ioend( | |
587 | struct inode *inode, | |
588 | struct buffer_head *bh, | |
7336cea8 | 589 | xfs_off_t offset, |
f6d6d4fc CH |
590 | unsigned int type, |
591 | xfs_ioend_t **result, | |
592 | int need_ioend) | |
593 | { | |
594 | xfs_ioend_t *ioend = *result; | |
595 | ||
596 | if (!ioend || need_ioend || type != ioend->io_type) { | |
597 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 598 | |
f6d6d4fc CH |
599 | ioend = xfs_alloc_ioend(inode, type); |
600 | ioend->io_offset = offset; | |
601 | ioend->io_buffer_head = bh; | |
602 | ioend->io_buffer_tail = bh; | |
603 | if (previous) | |
604 | previous->io_list = ioend; | |
605 | *result = ioend; | |
606 | } else { | |
607 | ioend->io_buffer_tail->b_private = bh; | |
608 | ioend->io_buffer_tail = bh; | |
609 | } | |
610 | ||
611 | bh->b_private = NULL; | |
612 | ioend->io_size += bh->b_size; | |
613 | } | |
614 | ||
87cbc49c NS |
615 | STATIC void |
616 | xfs_map_buffer( | |
617 | struct buffer_head *bh, | |
618 | xfs_iomap_t *mp, | |
619 | xfs_off_t offset, | |
620 | uint block_bits) | |
621 | { | |
622 | sector_t bn; | |
623 | ||
624 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
625 | ||
626 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
627 | ((offset - mp->iomap_offset) >> block_bits); | |
628 | ||
629 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
630 | ||
631 | bh->b_blocknr = bn; | |
632 | set_buffer_mapped(bh); | |
633 | } | |
634 | ||
1da177e4 LT |
635 | STATIC void |
636 | xfs_map_at_offset( | |
1da177e4 | 637 | struct buffer_head *bh, |
1defeac9 | 638 | loff_t offset, |
1da177e4 | 639 | int block_bits, |
1defeac9 | 640 | xfs_iomap_t *iomapp) |
1da177e4 | 641 | { |
1da177e4 LT |
642 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
643 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
644 | |
645 | lock_buffer(bh); | |
87cbc49c | 646 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 647 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
648 | set_buffer_mapped(bh); |
649 | clear_buffer_delay(bh); | |
f6d6d4fc | 650 | clear_buffer_unwritten(bh); |
1da177e4 LT |
651 | } |
652 | ||
653 | /* | |
6c4fe19f | 654 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
655 | */ |
656 | STATIC unsigned int | |
6c4fe19f | 657 | xfs_probe_page( |
10ce4444 | 658 | struct page *page, |
6c4fe19f CH |
659 | unsigned int pg_offset, |
660 | int mapped) | |
1da177e4 | 661 | { |
1da177e4 LT |
662 | int ret = 0; |
663 | ||
1da177e4 | 664 | if (PageWriteback(page)) |
10ce4444 | 665 | return 0; |
1da177e4 LT |
666 | |
667 | if (page->mapping && PageDirty(page)) { | |
668 | if (page_has_buffers(page)) { | |
669 | struct buffer_head *bh, *head; | |
670 | ||
671 | bh = head = page_buffers(page); | |
672 | do { | |
6c4fe19f CH |
673 | if (!buffer_uptodate(bh)) |
674 | break; | |
675 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
676 | break; |
677 | ret += bh->b_size; | |
678 | if (ret >= pg_offset) | |
679 | break; | |
680 | } while ((bh = bh->b_this_page) != head); | |
681 | } else | |
6c4fe19f | 682 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
683 | } |
684 | ||
1da177e4 LT |
685 | return ret; |
686 | } | |
687 | ||
f6d6d4fc | 688 | STATIC size_t |
6c4fe19f | 689 | xfs_probe_cluster( |
1da177e4 LT |
690 | struct inode *inode, |
691 | struct page *startpage, | |
692 | struct buffer_head *bh, | |
6c4fe19f CH |
693 | struct buffer_head *head, |
694 | int mapped) | |
1da177e4 | 695 | { |
10ce4444 | 696 | struct pagevec pvec; |
1da177e4 | 697 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
698 | size_t total = 0; |
699 | int done = 0, i; | |
1da177e4 LT |
700 | |
701 | /* First sum forwards in this page */ | |
702 | do { | |
2353e8e9 | 703 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 704 | return total; |
1da177e4 LT |
705 | total += bh->b_size; |
706 | } while ((bh = bh->b_this_page) != head); | |
707 | ||
10ce4444 CH |
708 | /* if we reached the end of the page, sum forwards in following pages */ |
709 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
710 | tindex = startpage->index + 1; | |
711 | ||
712 | /* Prune this back to avoid pathological behavior */ | |
713 | tloff = min(tlast, startpage->index + 64); | |
714 | ||
715 | pagevec_init(&pvec, 0); | |
716 | while (!done && tindex <= tloff) { | |
717 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
718 | ||
719 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
720 | break; | |
721 | ||
722 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
723 | struct page *page = pvec.pages[i]; | |
265c1fac | 724 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
725 | |
726 | if (tindex == tlast) { | |
727 | pg_offset = | |
728 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
729 | if (!pg_offset) { |
730 | done = 1; | |
10ce4444 | 731 | break; |
1defeac9 | 732 | } |
10ce4444 CH |
733 | } else |
734 | pg_offset = PAGE_CACHE_SIZE; | |
735 | ||
529ae9aa | 736 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 737 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
738 | unlock_page(page); |
739 | } | |
740 | ||
265c1fac | 741 | if (!pg_len) { |
10ce4444 CH |
742 | done = 1; |
743 | break; | |
744 | } | |
745 | ||
265c1fac | 746 | total += pg_len; |
1defeac9 | 747 | tindex++; |
1da177e4 | 748 | } |
10ce4444 CH |
749 | |
750 | pagevec_release(&pvec); | |
751 | cond_resched(); | |
1da177e4 | 752 | } |
10ce4444 | 753 | |
1da177e4 LT |
754 | return total; |
755 | } | |
756 | ||
757 | /* | |
10ce4444 CH |
758 | * Test if a given page is suitable for writing as part of an unwritten |
759 | * or delayed allocate extent. | |
1da177e4 | 760 | */ |
10ce4444 CH |
761 | STATIC int |
762 | xfs_is_delayed_page( | |
763 | struct page *page, | |
f6d6d4fc | 764 | unsigned int type) |
1da177e4 | 765 | { |
1da177e4 | 766 | if (PageWriteback(page)) |
10ce4444 | 767 | return 0; |
1da177e4 LT |
768 | |
769 | if (page->mapping && page_has_buffers(page)) { | |
770 | struct buffer_head *bh, *head; | |
771 | int acceptable = 0; | |
772 | ||
773 | bh = head = page_buffers(page); | |
774 | do { | |
f6d6d4fc CH |
775 | if (buffer_unwritten(bh)) |
776 | acceptable = (type == IOMAP_UNWRITTEN); | |
777 | else if (buffer_delay(bh)) | |
778 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 779 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 780 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 781 | else |
1da177e4 | 782 | break; |
1da177e4 LT |
783 | } while ((bh = bh->b_this_page) != head); |
784 | ||
785 | if (acceptable) | |
10ce4444 | 786 | return 1; |
1da177e4 LT |
787 | } |
788 | ||
10ce4444 | 789 | return 0; |
1da177e4 LT |
790 | } |
791 | ||
1da177e4 LT |
792 | /* |
793 | * Allocate & map buffers for page given the extent map. Write it out. | |
794 | * except for the original page of a writepage, this is called on | |
795 | * delalloc/unwritten pages only, for the original page it is possible | |
796 | * that the page has no mapping at all. | |
797 | */ | |
f6d6d4fc | 798 | STATIC int |
1da177e4 LT |
799 | xfs_convert_page( |
800 | struct inode *inode, | |
801 | struct page *page, | |
10ce4444 | 802 | loff_t tindex, |
1defeac9 | 803 | xfs_iomap_t *mp, |
f6d6d4fc | 804 | xfs_ioend_t **ioendp, |
1da177e4 | 805 | struct writeback_control *wbc, |
1da177e4 LT |
806 | int startio, |
807 | int all_bh) | |
808 | { | |
f6d6d4fc | 809 | struct buffer_head *bh, *head; |
9260dc6b CH |
810 | xfs_off_t end_offset; |
811 | unsigned long p_offset; | |
f6d6d4fc | 812 | unsigned int type; |
1da177e4 | 813 | int bbits = inode->i_blkbits; |
24e17b5f | 814 | int len, page_dirty; |
f6d6d4fc | 815 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 816 | xfs_off_t offset = page_offset(page); |
1da177e4 | 817 | |
10ce4444 CH |
818 | if (page->index != tindex) |
819 | goto fail; | |
529ae9aa | 820 | if (!trylock_page(page)) |
10ce4444 CH |
821 | goto fail; |
822 | if (PageWriteback(page)) | |
823 | goto fail_unlock_page; | |
824 | if (page->mapping != inode->i_mapping) | |
825 | goto fail_unlock_page; | |
826 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
827 | goto fail_unlock_page; | |
828 | ||
24e17b5f NS |
829 | /* |
830 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 831 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
832 | * |
833 | * Derivation: | |
834 | * | |
835 | * End offset is the highest offset that this page should represent. | |
836 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
837 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
838 | * hence give us the correct page_dirty count. On any other page, | |
839 | * it will be zero and in that case we need page_dirty to be the | |
840 | * count of buffers on the page. | |
24e17b5f | 841 | */ |
9260dc6b CH |
842 | end_offset = min_t(unsigned long long, |
843 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
844 | i_size_read(inode)); | |
845 | ||
24e17b5f | 846 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
847 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
848 | PAGE_CACHE_SIZE); | |
849 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
850 | page_dirty = p_offset / len; | |
24e17b5f | 851 | |
1da177e4 LT |
852 | bh = head = page_buffers(page); |
853 | do { | |
9260dc6b | 854 | if (offset >= end_offset) |
1da177e4 | 855 | break; |
f6d6d4fc CH |
856 | if (!buffer_uptodate(bh)) |
857 | uptodate = 0; | |
858 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
859 | done = 1; | |
1da177e4 | 860 | continue; |
f6d6d4fc CH |
861 | } |
862 | ||
9260dc6b CH |
863 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
864 | if (buffer_unwritten(bh)) | |
865 | type = IOMAP_UNWRITTEN; | |
866 | else | |
867 | type = IOMAP_DELAY; | |
868 | ||
869 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 870 | done = 1; |
9260dc6b CH |
871 | continue; |
872 | } | |
873 | ||
874 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
875 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
876 | ||
877 | xfs_map_at_offset(bh, offset, bbits, mp); | |
878 | if (startio) { | |
7336cea8 | 879 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
880 | type, ioendp, done); |
881 | } else { | |
882 | set_buffer_dirty(bh); | |
883 | unlock_buffer(bh); | |
884 | mark_buffer_dirty(bh); | |
885 | } | |
886 | page_dirty--; | |
887 | count++; | |
888 | } else { | |
df3c7244 | 889 | type = IOMAP_NEW; |
9260dc6b | 890 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 891 | lock_buffer(bh); |
7336cea8 | 892 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
893 | type, ioendp, done); |
894 | count++; | |
24e17b5f | 895 | page_dirty--; |
9260dc6b CH |
896 | } else { |
897 | done = 1; | |
1da177e4 | 898 | } |
1da177e4 | 899 | } |
7336cea8 | 900 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 901 | |
f6d6d4fc CH |
902 | if (uptodate && bh == head) |
903 | SetPageUptodate(page); | |
904 | ||
905 | if (startio) { | |
f5e596bb CH |
906 | if (count) { |
907 | struct backing_dev_info *bdi; | |
908 | ||
909 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 910 | wbc->nr_to_write--; |
f5e596bb CH |
911 | if (bdi_write_congested(bdi)) { |
912 | wbc->encountered_congestion = 1; | |
913 | done = 1; | |
9fddaca2 | 914 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
915 | done = 1; |
916 | } | |
917 | } | |
b41759cf | 918 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 919 | } |
f6d6d4fc CH |
920 | |
921 | return done; | |
10ce4444 CH |
922 | fail_unlock_page: |
923 | unlock_page(page); | |
924 | fail: | |
925 | return 1; | |
1da177e4 LT |
926 | } |
927 | ||
928 | /* | |
929 | * Convert & write out a cluster of pages in the same extent as defined | |
930 | * by mp and following the start page. | |
931 | */ | |
932 | STATIC void | |
933 | xfs_cluster_write( | |
934 | struct inode *inode, | |
935 | pgoff_t tindex, | |
936 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 937 | xfs_ioend_t **ioendp, |
1da177e4 LT |
938 | struct writeback_control *wbc, |
939 | int startio, | |
940 | int all_bh, | |
941 | pgoff_t tlast) | |
942 | { | |
10ce4444 CH |
943 | struct pagevec pvec; |
944 | int done = 0, i; | |
1da177e4 | 945 | |
10ce4444 CH |
946 | pagevec_init(&pvec, 0); |
947 | while (!done && tindex <= tlast) { | |
948 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
949 | ||
950 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 951 | break; |
10ce4444 CH |
952 | |
953 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
954 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
955 | iomapp, ioendp, wbc, startio, all_bh); | |
956 | if (done) | |
957 | break; | |
958 | } | |
959 | ||
960 | pagevec_release(&pvec); | |
961 | cond_resched(); | |
1da177e4 LT |
962 | } |
963 | } | |
964 | ||
965 | /* | |
966 | * Calling this without startio set means we are being asked to make a dirty | |
967 | * page ready for freeing it's buffers. When called with startio set then | |
968 | * we are coming from writepage. | |
969 | * | |
970 | * When called with startio set it is important that we write the WHOLE | |
971 | * page if possible. | |
972 | * The bh->b_state's cannot know if any of the blocks or which block for | |
973 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 974 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
975 | * may clear the page dirty flag prior to calling write page, under the |
976 | * assumption the entire page will be written out; by not writing out the | |
977 | * whole page the page can be reused before all valid dirty data is | |
978 | * written out. Note: in the case of a page that has been dirty'd by | |
979 | * mapwrite and but partially setup by block_prepare_write the | |
980 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
981 | * valid state, thus the whole page must be written out thing. | |
982 | */ | |
983 | ||
984 | STATIC int | |
985 | xfs_page_state_convert( | |
986 | struct inode *inode, | |
987 | struct page *page, | |
988 | struct writeback_control *wbc, | |
989 | int startio, | |
990 | int unmapped) /* also implies page uptodate */ | |
991 | { | |
f6d6d4fc | 992 | struct buffer_head *bh, *head; |
1defeac9 | 993 | xfs_iomap_t iomap; |
f6d6d4fc | 994 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
995 | loff_t offset; |
996 | unsigned long p_offset = 0; | |
f6d6d4fc | 997 | unsigned int type; |
1da177e4 LT |
998 | __uint64_t end_offset; |
999 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
1000 | ssize_t size, len; |
1001 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
1002 | int page_dirty, count = 0; |
1003 | int trylock = 0; | |
6c4fe19f | 1004 | int all_bh = unmapped; |
1da177e4 | 1005 | |
8272145c NS |
1006 | if (startio) { |
1007 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
1008 | trylock |= BMAPI_TRYLOCK; | |
1009 | } | |
3ba0815a | 1010 | |
1da177e4 LT |
1011 | /* Is this page beyond the end of the file? */ |
1012 | offset = i_size_read(inode); | |
1013 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1014 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1015 | if (page->index >= end_index) { | |
1016 | if ((page->index >= end_index + 1) || | |
1017 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
1018 | if (startio) |
1019 | unlock_page(page); | |
1020 | return 0; | |
1da177e4 LT |
1021 | } |
1022 | } | |
1023 | ||
1da177e4 | 1024 | /* |
24e17b5f | 1025 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 1026 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
1027 | * |
1028 | * Derivation: | |
1029 | * | |
1030 | * End offset is the highest offset that this page should represent. | |
1031 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
1032 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
1033 | * hence give us the correct page_dirty count. On any other page, | |
1034 | * it will be zero and in that case we need page_dirty to be the | |
1035 | * count of buffers on the page. | |
1036 | */ | |
1037 | end_offset = min_t(unsigned long long, | |
1038 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 1039 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
1040 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
1041 | PAGE_CACHE_SIZE); | |
1042 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
1043 | page_dirty = p_offset / len; |
1044 | ||
24e17b5f | 1045 | bh = head = page_buffers(page); |
f6d6d4fc | 1046 | offset = page_offset(page); |
df3c7244 DC |
1047 | flags = BMAPI_READ; |
1048 | type = IOMAP_NEW; | |
f6d6d4fc | 1049 | |
f6d6d4fc | 1050 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
1051 | |
1052 | do { | |
1053 | if (offset >= end_offset) | |
1054 | break; | |
1055 | if (!buffer_uptodate(bh)) | |
1056 | uptodate = 0; | |
f6d6d4fc | 1057 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
1058 | /* |
1059 | * the iomap is actually still valid, but the ioend | |
1060 | * isn't. shouldn't happen too often. | |
1061 | */ | |
1062 | iomap_valid = 0; | |
1da177e4 | 1063 | continue; |
f6d6d4fc | 1064 | } |
1da177e4 | 1065 | |
1defeac9 CH |
1066 | if (iomap_valid) |
1067 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
1068 | |
1069 | /* | |
1070 | * First case, map an unwritten extent and prepare for | |
1071 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1072 | * |
1da177e4 LT |
1073 | * Second case, allocate space for a delalloc buffer. |
1074 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1075 | * |
1076 | * Third case, an unmapped buffer was found, and we are | |
1077 | * in a path where we need to write the whole page out. | |
df3c7244 | 1078 | */ |
d5cb48aa CH |
1079 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1080 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1081 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1082 | int new_ioend = 0; |
1083 | ||
df3c7244 | 1084 | /* |
6c4fe19f CH |
1085 | * Make sure we don't use a read-only iomap |
1086 | */ | |
df3c7244 | 1087 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1088 | iomap_valid = 0; |
1089 | ||
f6d6d4fc CH |
1090 | if (buffer_unwritten(bh)) { |
1091 | type = IOMAP_UNWRITTEN; | |
8272145c | 1092 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1093 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1094 | type = IOMAP_DELAY; |
8272145c | 1095 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1096 | } else { |
6c4fe19f | 1097 | type = IOMAP_NEW; |
8272145c | 1098 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1099 | } |
1100 | ||
1defeac9 | 1101 | if (!iomap_valid) { |
effd120e DC |
1102 | /* |
1103 | * if we didn't have a valid mapping then we | |
1104 | * need to ensure that we put the new mapping | |
1105 | * in a new ioend structure. This needs to be | |
1106 | * done to ensure that the ioends correctly | |
1107 | * reflect the block mappings at io completion | |
1108 | * for unwritten extent conversion. | |
1109 | */ | |
1110 | new_ioend = 1; | |
6c4fe19f CH |
1111 | if (type == IOMAP_NEW) { |
1112 | size = xfs_probe_cluster(inode, | |
1113 | page, bh, head, 0); | |
d5cb48aa CH |
1114 | } else { |
1115 | size = len; | |
1116 | } | |
1117 | ||
1118 | err = xfs_map_blocks(inode, offset, size, | |
1119 | &iomap, flags); | |
f6d6d4fc | 1120 | if (err) |
1da177e4 | 1121 | goto error; |
1defeac9 | 1122 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1123 | } |
1defeac9 CH |
1124 | if (iomap_valid) { |
1125 | xfs_map_at_offset(bh, offset, | |
1126 | inode->i_blkbits, &iomap); | |
1da177e4 | 1127 | if (startio) { |
7336cea8 | 1128 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1129 | type, &ioend, |
effd120e | 1130 | new_ioend); |
1da177e4 LT |
1131 | } else { |
1132 | set_buffer_dirty(bh); | |
1133 | unlock_buffer(bh); | |
1134 | mark_buffer_dirty(bh); | |
1135 | } | |
1136 | page_dirty--; | |
f6d6d4fc | 1137 | count++; |
1da177e4 | 1138 | } |
d5cb48aa | 1139 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1140 | /* |
1141 | * we got here because the buffer is already mapped. | |
1142 | * That means it must already have extents allocated | |
1143 | * underneath it. Map the extent by reading it. | |
1144 | */ | |
df3c7244 | 1145 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1146 | flags = BMAPI_READ; |
1147 | size = xfs_probe_cluster(inode, page, bh, | |
1148 | head, 1); | |
1149 | err = xfs_map_blocks(inode, offset, size, | |
1150 | &iomap, flags); | |
1151 | if (err) | |
1152 | goto error; | |
1153 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1154 | } | |
d5cb48aa | 1155 | |
df3c7244 DC |
1156 | /* |
1157 | * We set the type to IOMAP_NEW in case we are doing a | |
1158 | * small write at EOF that is extending the file but | |
1159 | * without needing an allocation. We need to update the | |
1160 | * file size on I/O completion in this case so it is | |
1161 | * the same case as having just allocated a new extent | |
1162 | * that we are writing into for the first time. | |
1163 | */ | |
1164 | type = IOMAP_NEW; | |
ca5de404 | 1165 | if (trylock_buffer(bh)) { |
d5cb48aa | 1166 | ASSERT(buffer_mapped(bh)); |
6c4fe19f CH |
1167 | if (iomap_valid) |
1168 | all_bh = 1; | |
7336cea8 | 1169 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1170 | &ioend, !iomap_valid); |
1171 | page_dirty--; | |
1172 | count++; | |
f6d6d4fc | 1173 | } else { |
1defeac9 | 1174 | iomap_valid = 0; |
1da177e4 | 1175 | } |
d5cb48aa CH |
1176 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1177 | (unmapped || startio)) { | |
1178 | iomap_valid = 0; | |
1da177e4 | 1179 | } |
f6d6d4fc CH |
1180 | |
1181 | if (!iohead) | |
1182 | iohead = ioend; | |
1183 | ||
1184 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1185 | |
1186 | if (uptodate && bh == head) | |
1187 | SetPageUptodate(page); | |
1188 | ||
f6d6d4fc | 1189 | if (startio) |
b41759cf | 1190 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1191 | |
1defeac9 CH |
1192 | if (ioend && iomap_valid) { |
1193 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1194 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1195 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1196 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1197 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1198 | } |
1199 | ||
f6d6d4fc CH |
1200 | if (iohead) |
1201 | xfs_submit_ioend(iohead); | |
1202 | ||
1da177e4 LT |
1203 | return page_dirty; |
1204 | ||
1205 | error: | |
f6d6d4fc CH |
1206 | if (iohead) |
1207 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1208 | |
1209 | /* | |
1210 | * If it's delalloc and we have nowhere to put it, | |
1211 | * throw it away, unless the lower layers told | |
1212 | * us to try again. | |
1213 | */ | |
1214 | if (err != -EAGAIN) { | |
f6d6d4fc | 1215 | if (!unmapped) |
1da177e4 | 1216 | block_invalidatepage(page, 0); |
1da177e4 LT |
1217 | ClearPageUptodate(page); |
1218 | } | |
1219 | return err; | |
1220 | } | |
1221 | ||
f51623b2 NS |
1222 | /* |
1223 | * writepage: Called from one of two places: | |
1224 | * | |
1225 | * 1. we are flushing a delalloc buffer head. | |
1226 | * | |
1227 | * 2. we are writing out a dirty page. Typically the page dirty | |
1228 | * state is cleared before we get here. In this case is it | |
1229 | * conceivable we have no buffer heads. | |
1230 | * | |
1231 | * For delalloc space on the page we need to allocate space and | |
1232 | * flush it. For unmapped buffer heads on the page we should | |
1233 | * allocate space if the page is uptodate. For any other dirty | |
1234 | * buffer heads on the page we should flush them. | |
1235 | * | |
1236 | * If we detect that a transaction would be required to flush | |
1237 | * the page, we have to check the process flags first, if we | |
1238 | * are already in a transaction or disk I/O during allocations | |
1239 | * is off, we need to fail the writepage and redirty the page. | |
1240 | */ | |
1241 | ||
1242 | STATIC int | |
e4c573bb | 1243 | xfs_vm_writepage( |
f51623b2 NS |
1244 | struct page *page, |
1245 | struct writeback_control *wbc) | |
1246 | { | |
1247 | int error; | |
1248 | int need_trans; | |
1249 | int delalloc, unmapped, unwritten; | |
1250 | struct inode *inode = page->mapping->host; | |
1251 | ||
1252 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1253 | ||
1254 | /* | |
1255 | * We need a transaction if: | |
1256 | * 1. There are delalloc buffers on the page | |
1257 | * 2. The page is uptodate and we have unmapped buffers | |
1258 | * 3. The page is uptodate and we have no buffers | |
1259 | * 4. There are unwritten buffers on the page | |
1260 | */ | |
1261 | ||
1262 | if (!page_has_buffers(page)) { | |
1263 | unmapped = 1; | |
1264 | need_trans = 1; | |
1265 | } else { | |
1266 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1267 | if (!PageUptodate(page)) | |
1268 | unmapped = 0; | |
1269 | need_trans = delalloc + unmapped + unwritten; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * If we need a transaction and the process flags say | |
1274 | * we are already in a transaction, or no IO is allowed | |
1275 | * then mark the page dirty again and leave the page | |
1276 | * as is. | |
1277 | */ | |
59c1b082 | 1278 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1279 | goto out_fail; |
1280 | ||
1281 | /* | |
1282 | * Delay hooking up buffer heads until we have | |
1283 | * made our go/no-go decision. | |
1284 | */ | |
1285 | if (!page_has_buffers(page)) | |
1286 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1287 | ||
c8a4051c ES |
1288 | |
1289 | /* | |
1290 | * VM calculation for nr_to_write seems off. Bump it way | |
1291 | * up, this gets simple streaming writes zippy again. | |
1292 | * To be reviewed again after Jens' writeback changes. | |
1293 | */ | |
1294 | wbc->nr_to_write *= 4; | |
1295 | ||
f51623b2 NS |
1296 | /* |
1297 | * Convert delayed allocate, unwritten or unmapped space | |
1298 | * to real space and flush out to disk. | |
1299 | */ | |
1300 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1301 | if (error == -EAGAIN) | |
1302 | goto out_fail; | |
1303 | if (unlikely(error < 0)) | |
1304 | goto out_unlock; | |
1305 | ||
1306 | return 0; | |
1307 | ||
1308 | out_fail: | |
1309 | redirty_page_for_writepage(wbc, page); | |
1310 | unlock_page(page); | |
1311 | return 0; | |
1312 | out_unlock: | |
1313 | unlock_page(page); | |
1314 | return error; | |
1315 | } | |
1316 | ||
7d4fb40a NS |
1317 | STATIC int |
1318 | xfs_vm_writepages( | |
1319 | struct address_space *mapping, | |
1320 | struct writeback_control *wbc) | |
1321 | { | |
b3aea4ed | 1322 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1323 | return generic_writepages(mapping, wbc); |
1324 | } | |
1325 | ||
f51623b2 NS |
1326 | /* |
1327 | * Called to move a page into cleanable state - and from there | |
1328 | * to be released. Possibly the page is already clean. We always | |
1329 | * have buffer heads in this call. | |
1330 | * | |
1331 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1332 | * | |
1333 | * Possible scenarios are: | |
1334 | * | |
1335 | * 1. We are being called to release a page which has been written | |
1336 | * to via regular I/O. buffer heads will be dirty and possibly | |
1337 | * delalloc. If no delalloc buffer heads in this case then we | |
1338 | * can just return zero. | |
1339 | * | |
1340 | * 2. We are called to release a page which has been written via | |
1341 | * mmap, all we need to do is ensure there is no delalloc | |
1342 | * state in the buffer heads, if not we can let the caller | |
1343 | * free them and we should come back later via writepage. | |
1344 | */ | |
1345 | STATIC int | |
238f4c54 | 1346 | xfs_vm_releasepage( |
f51623b2 NS |
1347 | struct page *page, |
1348 | gfp_t gfp_mask) | |
1349 | { | |
1350 | struct inode *inode = page->mapping->host; | |
1351 | int dirty, delalloc, unmapped, unwritten; | |
1352 | struct writeback_control wbc = { | |
1353 | .sync_mode = WB_SYNC_ALL, | |
1354 | .nr_to_write = 1, | |
1355 | }; | |
1356 | ||
ed9d88f7 | 1357 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0); |
f51623b2 | 1358 | |
238f4c54 NS |
1359 | if (!page_has_buffers(page)) |
1360 | return 0; | |
1361 | ||
f51623b2 NS |
1362 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1363 | if (!delalloc && !unwritten) | |
1364 | goto free_buffers; | |
1365 | ||
1366 | if (!(gfp_mask & __GFP_FS)) | |
1367 | return 0; | |
1368 | ||
1369 | /* If we are already inside a transaction or the thread cannot | |
1370 | * do I/O, we cannot release this page. | |
1371 | */ | |
59c1b082 | 1372 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1373 | return 0; |
1374 | ||
1375 | /* | |
1376 | * Convert delalloc space to real space, do not flush the | |
1377 | * data out to disk, that will be done by the caller. | |
1378 | * Never need to allocate space here - we will always | |
1379 | * come back to writepage in that case. | |
1380 | */ | |
1381 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1382 | if (dirty == 0 && !unwritten) | |
1383 | goto free_buffers; | |
1384 | return 0; | |
1385 | ||
1386 | free_buffers: | |
1387 | return try_to_free_buffers(page); | |
1388 | } | |
1389 | ||
1da177e4 | 1390 | STATIC int |
c2536668 | 1391 | __xfs_get_blocks( |
1da177e4 LT |
1392 | struct inode *inode, |
1393 | sector_t iblock, | |
1da177e4 LT |
1394 | struct buffer_head *bh_result, |
1395 | int create, | |
1396 | int direct, | |
1397 | bmapi_flags_t flags) | |
1398 | { | |
1da177e4 | 1399 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1400 | xfs_off_t offset; |
1401 | ssize_t size; | |
c2536668 | 1402 | int niomap = 1; |
1da177e4 | 1403 | int error; |
1da177e4 | 1404 | |
fdc7ed75 | 1405 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1406 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1407 | size = bh_result->b_size; | |
364f358a LM |
1408 | |
1409 | if (!create && direct && offset >= i_size_read(inode)) | |
1410 | return 0; | |
1411 | ||
541d7d3c | 1412 | error = xfs_iomap(XFS_I(inode), offset, size, |
67fcaa73 | 1413 | create ? flags : BMAPI_READ, &iomap, &niomap); |
1da177e4 LT |
1414 | if (error) |
1415 | return -error; | |
c2536668 | 1416 | if (niomap == 0) |
1da177e4 LT |
1417 | return 0; |
1418 | ||
1419 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1420 | /* |
1421 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1422 | * the read case (treat as if we're reading into a hole). |
1423 | */ | |
1424 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1425 | xfs_map_buffer(bh_result, &iomap, offset, |
1426 | inode->i_blkbits); | |
1da177e4 LT |
1427 | } |
1428 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1429 | if (direct) | |
1430 | bh_result->b_private = inode; | |
1431 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1432 | } |
1433 | } | |
1434 | ||
c2536668 NS |
1435 | /* |
1436 | * If this is a realtime file, data may be on a different device. | |
1437 | * to that pointed to from the buffer_head b_bdev currently. | |
1438 | */ | |
ce8e922c | 1439 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1440 | |
c2536668 | 1441 | /* |
549054af DC |
1442 | * If we previously allocated a block out beyond eof and we are now |
1443 | * coming back to use it then we will need to flag it as new even if it | |
1444 | * has a disk address. | |
1445 | * | |
1446 | * With sub-block writes into unwritten extents we also need to mark | |
1447 | * the buffer as new so that the unwritten parts of the buffer gets | |
1448 | * correctly zeroed. | |
1da177e4 LT |
1449 | */ |
1450 | if (create && | |
1451 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1452 | (offset >= i_size_read(inode)) || |
1453 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1454 | set_buffer_new(bh_result); |
1da177e4 LT |
1455 | |
1456 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1457 | BUG_ON(direct); | |
1458 | if (create) { | |
1459 | set_buffer_uptodate(bh_result); | |
1460 | set_buffer_mapped(bh_result); | |
1461 | set_buffer_delay(bh_result); | |
1462 | } | |
1463 | } | |
1464 | ||
c2536668 | 1465 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1466 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1467 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1468 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1469 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1470 | } |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
1475 | int | |
c2536668 | 1476 | xfs_get_blocks( |
1da177e4 LT |
1477 | struct inode *inode, |
1478 | sector_t iblock, | |
1479 | struct buffer_head *bh_result, | |
1480 | int create) | |
1481 | { | |
c2536668 | 1482 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1483 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1484 | } |
1485 | ||
1486 | STATIC int | |
e4c573bb | 1487 | xfs_get_blocks_direct( |
1da177e4 LT |
1488 | struct inode *inode, |
1489 | sector_t iblock, | |
1da177e4 LT |
1490 | struct buffer_head *bh_result, |
1491 | int create) | |
1492 | { | |
c2536668 | 1493 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1494 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1495 | } |
1496 | ||
f0973863 | 1497 | STATIC void |
e4c573bb | 1498 | xfs_end_io_direct( |
f0973863 CH |
1499 | struct kiocb *iocb, |
1500 | loff_t offset, | |
1501 | ssize_t size, | |
1502 | void *private) | |
1503 | { | |
1504 | xfs_ioend_t *ioend = iocb->private; | |
1505 | ||
1506 | /* | |
1507 | * Non-NULL private data means we need to issue a transaction to | |
1508 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1509 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1510 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1511 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1512 | * it anyway to keep the code uniform and simpler. |
1513 | * | |
e927af90 DC |
1514 | * Well, if only it were that simple. Because synchronous direct I/O |
1515 | * requires extent conversion to occur *before* we return to userspace, | |
1516 | * we have to wait for extent conversion to complete. Look at the | |
1517 | * iocb that has been passed to us to determine if this is AIO or | |
1518 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1519 | * workqueue and wait for it to complete. | |
1520 | * | |
f0973863 CH |
1521 | * The core direct I/O code might be changed to always call the |
1522 | * completion handler in the future, in which case all this can | |
1523 | * go away. | |
1524 | */ | |
ba87ea69 LM |
1525 | ioend->io_offset = offset; |
1526 | ioend->io_size = size; | |
1527 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1528 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1529 | } else if (private && size > 0) { |
e927af90 | 1530 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1531 | } else { |
ba87ea69 LM |
1532 | /* |
1533 | * A direct I/O write ioend starts it's life in unwritten | |
1534 | * state in case they map an unwritten extent. This write | |
1535 | * didn't map an unwritten extent so switch it's completion | |
1536 | * handler. | |
1537 | */ | |
1538 | INIT_WORK(&ioend->io_work, xfs_end_bio_written); | |
e927af90 | 1539 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1540 | } |
1541 | ||
1542 | /* | |
c41564b5 | 1543 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1544 | * completion handler was called. Thus we need to protect |
1545 | * against double-freeing. | |
1546 | */ | |
1547 | iocb->private = NULL; | |
1548 | } | |
1549 | ||
1da177e4 | 1550 | STATIC ssize_t |
e4c573bb | 1551 | xfs_vm_direct_IO( |
1da177e4 LT |
1552 | int rw, |
1553 | struct kiocb *iocb, | |
1554 | const struct iovec *iov, | |
1555 | loff_t offset, | |
1556 | unsigned long nr_segs) | |
1557 | { | |
1558 | struct file *file = iocb->ki_filp; | |
1559 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1560 | struct block_device *bdev; |
f0973863 | 1561 | ssize_t ret; |
1da177e4 | 1562 | |
6214ed44 | 1563 | bdev = xfs_find_bdev_for_inode(XFS_I(inode)); |
1da177e4 | 1564 | |
721259bc | 1565 | if (rw == WRITE) { |
ba87ea69 | 1566 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
721259bc | 1567 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, |
6214ed44 | 1568 | bdev, iov, offset, nr_segs, |
721259bc LM |
1569 | xfs_get_blocks_direct, |
1570 | xfs_end_io_direct); | |
1571 | } else { | |
ba87ea69 | 1572 | iocb->private = xfs_alloc_ioend(inode, IOMAP_READ); |
721259bc | 1573 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, |
6214ed44 | 1574 | bdev, iov, offset, nr_segs, |
721259bc LM |
1575 | xfs_get_blocks_direct, |
1576 | xfs_end_io_direct); | |
1577 | } | |
f0973863 | 1578 | |
8459d86a | 1579 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1580 | xfs_destroy_ioend(iocb->private); |
1581 | return ret; | |
1da177e4 LT |
1582 | } |
1583 | ||
f51623b2 | 1584 | STATIC int |
d79689c7 | 1585 | xfs_vm_write_begin( |
f51623b2 | 1586 | struct file *file, |
d79689c7 NP |
1587 | struct address_space *mapping, |
1588 | loff_t pos, | |
1589 | unsigned len, | |
1590 | unsigned flags, | |
1591 | struct page **pagep, | |
1592 | void **fsdata) | |
f51623b2 | 1593 | { |
d79689c7 NP |
1594 | *pagep = NULL; |
1595 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1596 | xfs_get_blocks); | |
f51623b2 | 1597 | } |
1da177e4 LT |
1598 | |
1599 | STATIC sector_t | |
e4c573bb | 1600 | xfs_vm_bmap( |
1da177e4 LT |
1601 | struct address_space *mapping, |
1602 | sector_t block) | |
1603 | { | |
1604 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1605 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1606 | |
cf441eeb | 1607 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1608 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1609 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1610 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1611 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1612 | } |
1613 | ||
1614 | STATIC int | |
e4c573bb | 1615 | xfs_vm_readpage( |
1da177e4 LT |
1616 | struct file *unused, |
1617 | struct page *page) | |
1618 | { | |
c2536668 | 1619 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1620 | } |
1621 | ||
1622 | STATIC int | |
e4c573bb | 1623 | xfs_vm_readpages( |
1da177e4 LT |
1624 | struct file *unused, |
1625 | struct address_space *mapping, | |
1626 | struct list_head *pages, | |
1627 | unsigned nr_pages) | |
1628 | { | |
c2536668 | 1629 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1630 | } |
1631 | ||
2ff28e22 | 1632 | STATIC void |
238f4c54 | 1633 | xfs_vm_invalidatepage( |
bcec2b7f NS |
1634 | struct page *page, |
1635 | unsigned long offset) | |
1636 | { | |
1637 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1638 | page->mapping->host, page, offset); | |
2ff28e22 | 1639 | block_invalidatepage(page, offset); |
bcec2b7f NS |
1640 | } |
1641 | ||
f5e54d6e | 1642 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1643 | .readpage = xfs_vm_readpage, |
1644 | .readpages = xfs_vm_readpages, | |
1645 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1646 | .writepages = xfs_vm_writepages, |
1da177e4 | 1647 | .sync_page = block_sync_page, |
238f4c54 NS |
1648 | .releasepage = xfs_vm_releasepage, |
1649 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1650 | .write_begin = xfs_vm_write_begin, |
1651 | .write_end = generic_write_end, | |
e4c573bb NS |
1652 | .bmap = xfs_vm_bmap, |
1653 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1654 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1655 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1656 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1657 | }; |