]>
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 | 24 | #include "xfs_trans.h" |
1da177e4 LT |
25 | #include "xfs_mount.h" |
26 | #include "xfs_bmap_btree.h" | |
1da177e4 LT |
27 | #include "xfs_dinode.h" |
28 | #include "xfs_inode.h" | |
a844f451 | 29 | #include "xfs_alloc.h" |
1da177e4 LT |
30 | #include "xfs_error.h" |
31 | #include "xfs_rw.h" | |
32 | #include "xfs_iomap.h" | |
739bfb2a | 33 | #include "xfs_vnodeops.h" |
0b1b213f | 34 | #include "xfs_trace.h" |
3ed3a434 | 35 | #include "xfs_bmap.h" |
5a0e3ad6 | 36 | #include <linux/gfp.h> |
1da177e4 | 37 | #include <linux/mpage.h> |
10ce4444 | 38 | #include <linux/pagevec.h> |
1da177e4 LT |
39 | #include <linux/writeback.h> |
40 | ||
34a52c6c CH |
41 | /* |
42 | * Types of I/O for bmap clustering and I/O completion tracking. | |
43 | */ | |
44 | enum { | |
45 | IO_READ, /* mapping for a read */ | |
46 | IO_DELAY, /* mapping covers delalloc region */ | |
47 | IO_UNWRITTEN, /* mapping covers allocated but uninitialized data */ | |
48 | IO_NEW /* just allocated */ | |
49 | }; | |
25e41b3d CH |
50 | |
51 | /* | |
52 | * Prime number of hash buckets since address is used as the key. | |
53 | */ | |
54 | #define NVSYNC 37 | |
55 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
56 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
57 | ||
58 | void __init | |
59 | xfs_ioend_init(void) | |
60 | { | |
61 | int i; | |
62 | ||
63 | for (i = 0; i < NVSYNC; i++) | |
64 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
65 | } | |
66 | ||
67 | void | |
68 | xfs_ioend_wait( | |
69 | xfs_inode_t *ip) | |
70 | { | |
71 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
72 | ||
73 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
74 | } | |
75 | ||
76 | STATIC void | |
77 | xfs_ioend_wake( | |
78 | xfs_inode_t *ip) | |
79 | { | |
80 | if (atomic_dec_and_test(&ip->i_iocount)) | |
81 | wake_up(to_ioend_wq(ip)); | |
82 | } | |
83 | ||
0b1b213f | 84 | void |
f51623b2 NS |
85 | xfs_count_page_state( |
86 | struct page *page, | |
87 | int *delalloc, | |
f51623b2 NS |
88 | int *unwritten) |
89 | { | |
90 | struct buffer_head *bh, *head; | |
91 | ||
20cb52eb | 92 | *delalloc = *unwritten = 0; |
f51623b2 NS |
93 | |
94 | bh = head = page_buffers(page); | |
95 | do { | |
20cb52eb | 96 | if (buffer_unwritten(bh)) |
f51623b2 NS |
97 | (*unwritten) = 1; |
98 | else if (buffer_delay(bh)) | |
99 | (*delalloc) = 1; | |
100 | } while ((bh = bh->b_this_page) != head); | |
101 | } | |
102 | ||
6214ed44 CH |
103 | STATIC struct block_device * |
104 | xfs_find_bdev_for_inode( | |
046f1685 | 105 | struct inode *inode) |
6214ed44 | 106 | { |
046f1685 | 107 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
108 | struct xfs_mount *mp = ip->i_mount; |
109 | ||
71ddabb9 | 110 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
111 | return mp->m_rtdev_targp->bt_bdev; |
112 | else | |
113 | return mp->m_ddev_targp->bt_bdev; | |
114 | } | |
115 | ||
f6d6d4fc CH |
116 | /* |
117 | * We're now finished for good with this ioend structure. | |
118 | * Update the page state via the associated buffer_heads, | |
119 | * release holds on the inode and bio, and finally free | |
120 | * up memory. Do not use the ioend after this. | |
121 | */ | |
0829c360 CH |
122 | STATIC void |
123 | xfs_destroy_ioend( | |
124 | xfs_ioend_t *ioend) | |
125 | { | |
f6d6d4fc | 126 | struct buffer_head *bh, *next; |
583fa586 | 127 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
128 | |
129 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
130 | next = bh->b_private; | |
7d04a335 | 131 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 132 | } |
583fa586 CH |
133 | |
134 | /* | |
135 | * Volume managers supporting multiple paths can send back ENODEV | |
136 | * when the final path disappears. In this case continuing to fill | |
137 | * the page cache with dirty data which cannot be written out is | |
138 | * evil, so prevent that. | |
139 | */ | |
140 | if (unlikely(ioend->io_error == -ENODEV)) { | |
141 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
142 | __FILE__, __LINE__); | |
b677c210 | 143 | } |
583fa586 | 144 | |
25e41b3d | 145 | xfs_ioend_wake(ip); |
0829c360 CH |
146 | mempool_free(ioend, xfs_ioend_pool); |
147 | } | |
148 | ||
932640e8 DC |
149 | /* |
150 | * If the end of the current ioend is beyond the current EOF, | |
151 | * return the new EOF value, otherwise zero. | |
152 | */ | |
153 | STATIC xfs_fsize_t | |
154 | xfs_ioend_new_eof( | |
155 | xfs_ioend_t *ioend) | |
156 | { | |
157 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
158 | xfs_fsize_t isize; | |
159 | xfs_fsize_t bsize; | |
160 | ||
161 | bsize = ioend->io_offset + ioend->io_size; | |
162 | isize = MAX(ip->i_size, ip->i_new_size); | |
163 | isize = MIN(isize, bsize); | |
164 | return isize > ip->i_d.di_size ? isize : 0; | |
165 | } | |
166 | ||
ba87ea69 | 167 | /* |
77d7a0c2 DC |
168 | * Update on-disk file size now that data has been written to disk. The |
169 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
170 | * will be the intended file size until i_size is updated. If this write does | |
171 | * not extend all the way to the valid file size then restrict this update to | |
172 | * the end of the write. | |
173 | * | |
174 | * This function does not block as blocking on the inode lock in IO completion | |
175 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
176 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 177 | */ |
77d7a0c2 | 178 | STATIC int |
ba87ea69 LM |
179 | xfs_setfilesize( |
180 | xfs_ioend_t *ioend) | |
181 | { | |
b677c210 | 182 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 183 | xfs_fsize_t isize; |
ba87ea69 | 184 | |
ba87ea69 | 185 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
34a52c6c | 186 | ASSERT(ioend->io_type != IO_READ); |
ba87ea69 LM |
187 | |
188 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
189 | return 0; |
190 | ||
191 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
192 | return EAGAIN; | |
ba87ea69 | 193 | |
932640e8 DC |
194 | isize = xfs_ioend_new_eof(ioend); |
195 | if (isize) { | |
ba87ea69 | 196 | ip->i_d.di_size = isize; |
66d834ea | 197 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
198 | } |
199 | ||
200 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
201 | return 0; |
202 | } | |
203 | ||
204 | /* | |
209fb87a | 205 | * Schedule IO completion handling on the final put of an ioend. |
77d7a0c2 DC |
206 | */ |
207 | STATIC void | |
208 | xfs_finish_ioend( | |
209fb87a | 209 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
210 | { |
211 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
209fb87a CH |
212 | if (ioend->io_type == IO_UNWRITTEN) |
213 | queue_work(xfsconvertd_workqueue, &ioend->io_work); | |
214 | else | |
215 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
77d7a0c2 | 216 | } |
ba87ea69 LM |
217 | } |
218 | ||
0829c360 | 219 | /* |
5ec4fabb | 220 | * IO write completion. |
f6d6d4fc CH |
221 | */ |
222 | STATIC void | |
5ec4fabb | 223 | xfs_end_io( |
77d7a0c2 | 224 | struct work_struct *work) |
0829c360 | 225 | { |
77d7a0c2 DC |
226 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
227 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 228 | int error = 0; |
ba87ea69 | 229 | |
5ec4fabb CH |
230 | /* |
231 | * For unwritten extents we need to issue transactions to convert a | |
232 | * range to normal written extens after the data I/O has finished. | |
233 | */ | |
34a52c6c | 234 | if (ioend->io_type == IO_UNWRITTEN && |
5ec4fabb | 235 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { |
5ec4fabb CH |
236 | |
237 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
238 | ioend->io_size); | |
239 | if (error) | |
240 | ioend->io_error = error; | |
241 | } | |
ba87ea69 | 242 | |
5ec4fabb CH |
243 | /* |
244 | * We might have to update the on-disk file size after extending | |
245 | * writes. | |
246 | */ | |
34a52c6c | 247 | if (ioend->io_type != IO_READ) { |
77d7a0c2 DC |
248 | error = xfs_setfilesize(ioend); |
249 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 250 | } |
77d7a0c2 DC |
251 | |
252 | /* | |
253 | * If we didn't complete processing of the ioend, requeue it to the | |
254 | * tail of the workqueue for another attempt later. Otherwise destroy | |
255 | * it. | |
256 | */ | |
257 | if (error == EAGAIN) { | |
258 | atomic_inc(&ioend->io_remaining); | |
209fb87a | 259 | xfs_finish_ioend(ioend); |
77d7a0c2 DC |
260 | /* ensure we don't spin on blocked ioends */ |
261 | delay(1); | |
fb511f21 CH |
262 | } else { |
263 | if (ioend->io_iocb) | |
264 | aio_complete(ioend->io_iocb, ioend->io_result, 0); | |
77d7a0c2 | 265 | xfs_destroy_ioend(ioend); |
fb511f21 | 266 | } |
c626d174 DC |
267 | } |
268 | ||
209fb87a CH |
269 | /* |
270 | * Call IO completion handling in caller context on the final put of an ioend. | |
271 | */ | |
272 | STATIC void | |
273 | xfs_finish_ioend_sync( | |
274 | struct xfs_ioend *ioend) | |
275 | { | |
276 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
277 | xfs_end_io(&ioend->io_work); | |
278 | } | |
279 | ||
0829c360 CH |
280 | /* |
281 | * Allocate and initialise an IO completion structure. | |
282 | * We need to track unwritten extent write completion here initially. | |
283 | * We'll need to extend this for updating the ondisk inode size later | |
284 | * (vs. incore size). | |
285 | */ | |
286 | STATIC xfs_ioend_t * | |
287 | xfs_alloc_ioend( | |
f6d6d4fc CH |
288 | struct inode *inode, |
289 | unsigned int type) | |
0829c360 CH |
290 | { |
291 | xfs_ioend_t *ioend; | |
292 | ||
293 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
294 | ||
295 | /* | |
296 | * Set the count to 1 initially, which will prevent an I/O | |
297 | * completion callback from happening before we have started | |
298 | * all the I/O from calling the completion routine too early. | |
299 | */ | |
300 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 301 | ioend->io_error = 0; |
f6d6d4fc CH |
302 | ioend->io_list = NULL; |
303 | ioend->io_type = type; | |
b677c210 | 304 | ioend->io_inode = inode; |
c1a073bd | 305 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 306 | ioend->io_buffer_tail = NULL; |
b677c210 | 307 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
308 | ioend->io_offset = 0; |
309 | ioend->io_size = 0; | |
fb511f21 CH |
310 | ioend->io_iocb = NULL; |
311 | ioend->io_result = 0; | |
0829c360 | 312 | |
5ec4fabb | 313 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
314 | return ioend; |
315 | } | |
316 | ||
1da177e4 LT |
317 | STATIC int |
318 | xfs_map_blocks( | |
319 | struct inode *inode, | |
320 | loff_t offset, | |
321 | ssize_t count, | |
207d0416 | 322 | struct xfs_bmbt_irec *imap, |
1da177e4 LT |
323 | int flags) |
324 | { | |
6bd16ff2 | 325 | int nmaps = 1; |
207d0416 | 326 | int new = 0; |
6bd16ff2 | 327 | |
207d0416 | 328 | return -xfs_iomap(XFS_I(inode), offset, count, flags, imap, &nmaps, &new); |
1da177e4 LT |
329 | } |
330 | ||
b8f82a4a | 331 | STATIC int |
558e6891 | 332 | xfs_imap_valid( |
8699bb0a | 333 | struct inode *inode, |
207d0416 | 334 | struct xfs_bmbt_irec *imap, |
558e6891 | 335 | xfs_off_t offset) |
1da177e4 | 336 | { |
558e6891 | 337 | offset >>= inode->i_blkbits; |
8699bb0a | 338 | |
558e6891 CH |
339 | return offset >= imap->br_startoff && |
340 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
341 | } |
342 | ||
f6d6d4fc CH |
343 | /* |
344 | * BIO completion handler for buffered IO. | |
345 | */ | |
782e3b3b | 346 | STATIC void |
f6d6d4fc CH |
347 | xfs_end_bio( |
348 | struct bio *bio, | |
f6d6d4fc CH |
349 | int error) |
350 | { | |
351 | xfs_ioend_t *ioend = bio->bi_private; | |
352 | ||
f6d6d4fc | 353 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 354 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
355 | |
356 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
357 | bio->bi_private = NULL; |
358 | bio->bi_end_io = NULL; | |
f6d6d4fc | 359 | bio_put(bio); |
7d04a335 | 360 | |
209fb87a | 361 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
362 | } |
363 | ||
364 | STATIC void | |
365 | xfs_submit_ioend_bio( | |
06342cf8 CH |
366 | struct writeback_control *wbc, |
367 | xfs_ioend_t *ioend, | |
368 | struct bio *bio) | |
f6d6d4fc CH |
369 | { |
370 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
371 | bio->bi_private = ioend; |
372 | bio->bi_end_io = xfs_end_bio; | |
373 | ||
932640e8 DC |
374 | /* |
375 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
376 | * but don't update the inode size until I/O completion. | |
377 | */ | |
378 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 379 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 380 | |
06342cf8 CH |
381 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
382 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
383 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
384 | bio_put(bio); | |
385 | } | |
386 | ||
387 | STATIC struct bio * | |
388 | xfs_alloc_ioend_bio( | |
389 | struct buffer_head *bh) | |
390 | { | |
391 | struct bio *bio; | |
392 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
393 | ||
394 | do { | |
395 | bio = bio_alloc(GFP_NOIO, nvecs); | |
396 | nvecs >>= 1; | |
397 | } while (!bio); | |
398 | ||
399 | ASSERT(bio->bi_private == NULL); | |
400 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
401 | bio->bi_bdev = bh->b_bdev; | |
402 | bio_get(bio); | |
403 | return bio; | |
404 | } | |
405 | ||
406 | STATIC void | |
407 | xfs_start_buffer_writeback( | |
408 | struct buffer_head *bh) | |
409 | { | |
410 | ASSERT(buffer_mapped(bh)); | |
411 | ASSERT(buffer_locked(bh)); | |
412 | ASSERT(!buffer_delay(bh)); | |
413 | ASSERT(!buffer_unwritten(bh)); | |
414 | ||
415 | mark_buffer_async_write(bh); | |
416 | set_buffer_uptodate(bh); | |
417 | clear_buffer_dirty(bh); | |
418 | } | |
419 | ||
420 | STATIC void | |
421 | xfs_start_page_writeback( | |
422 | struct page *page, | |
f6d6d4fc CH |
423 | int clear_dirty, |
424 | int buffers) | |
425 | { | |
426 | ASSERT(PageLocked(page)); | |
427 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 428 | if (clear_dirty) |
92132021 DC |
429 | clear_page_dirty_for_io(page); |
430 | set_page_writeback(page); | |
f6d6d4fc | 431 | unlock_page(page); |
1f7decf6 FW |
432 | /* If no buffers on the page are to be written, finish it here */ |
433 | if (!buffers) | |
f6d6d4fc | 434 | end_page_writeback(page); |
f6d6d4fc CH |
435 | } |
436 | ||
437 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
438 | { | |
439 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
440 | } | |
441 | ||
442 | /* | |
d88992f6 DC |
443 | * Submit all of the bios for all of the ioends we have saved up, covering the |
444 | * initial writepage page and also any probed pages. | |
445 | * | |
446 | * Because we may have multiple ioends spanning a page, we need to start | |
447 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
448 | * buffers as we got, then we can end up with a page that only has buffers | |
449 | * marked async write and I/O complete on can occur before we mark the other | |
450 | * buffers async write. | |
451 | * | |
452 | * The end result of this is that we trip a bug in end_page_writeback() because | |
453 | * we call it twice for the one page as the code in end_buffer_async_write() | |
454 | * assumes that all buffers on the page are started at the same time. | |
455 | * | |
456 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 457 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
458 | */ |
459 | STATIC void | |
460 | xfs_submit_ioend( | |
06342cf8 | 461 | struct writeback_control *wbc, |
f6d6d4fc CH |
462 | xfs_ioend_t *ioend) |
463 | { | |
d88992f6 | 464 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
465 | xfs_ioend_t *next; |
466 | struct buffer_head *bh; | |
467 | struct bio *bio; | |
468 | sector_t lastblock = 0; | |
469 | ||
d88992f6 DC |
470 | /* Pass 1 - start writeback */ |
471 | do { | |
472 | next = ioend->io_list; | |
473 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
474 | xfs_start_buffer_writeback(bh); | |
475 | } | |
476 | } while ((ioend = next) != NULL); | |
477 | ||
478 | /* Pass 2 - submit I/O */ | |
479 | ioend = head; | |
f6d6d4fc CH |
480 | do { |
481 | next = ioend->io_list; | |
482 | bio = NULL; | |
483 | ||
484 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
485 | |
486 | if (!bio) { | |
487 | retry: | |
488 | bio = xfs_alloc_ioend_bio(bh); | |
489 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 490 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
491 | goto retry; |
492 | } | |
493 | ||
494 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 495 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
496 | goto retry; |
497 | } | |
498 | ||
499 | lastblock = bh->b_blocknr; | |
500 | } | |
501 | if (bio) | |
06342cf8 | 502 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 503 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
504 | } while ((ioend = next) != NULL); |
505 | } | |
506 | ||
507 | /* | |
508 | * Cancel submission of all buffer_heads so far in this endio. | |
509 | * Toss the endio too. Only ever called for the initial page | |
510 | * in a writepage request, so only ever one page. | |
511 | */ | |
512 | STATIC void | |
513 | xfs_cancel_ioend( | |
514 | xfs_ioend_t *ioend) | |
515 | { | |
516 | xfs_ioend_t *next; | |
517 | struct buffer_head *bh, *next_bh; | |
518 | ||
519 | do { | |
520 | next = ioend->io_list; | |
521 | bh = ioend->io_buffer_head; | |
522 | do { | |
523 | next_bh = bh->b_private; | |
524 | clear_buffer_async_write(bh); | |
525 | unlock_buffer(bh); | |
526 | } while ((bh = next_bh) != NULL); | |
527 | ||
25e41b3d | 528 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
529 | mempool_free(ioend, xfs_ioend_pool); |
530 | } while ((ioend = next) != NULL); | |
531 | } | |
532 | ||
533 | /* | |
534 | * Test to see if we've been building up a completion structure for | |
535 | * earlier buffers -- if so, we try to append to this ioend if we | |
536 | * can, otherwise we finish off any current ioend and start another. | |
537 | * Return true if we've finished the given ioend. | |
538 | */ | |
539 | STATIC void | |
540 | xfs_add_to_ioend( | |
541 | struct inode *inode, | |
542 | struct buffer_head *bh, | |
7336cea8 | 543 | xfs_off_t offset, |
f6d6d4fc CH |
544 | unsigned int type, |
545 | xfs_ioend_t **result, | |
546 | int need_ioend) | |
547 | { | |
548 | xfs_ioend_t *ioend = *result; | |
549 | ||
550 | if (!ioend || need_ioend || type != ioend->io_type) { | |
551 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 552 | |
f6d6d4fc CH |
553 | ioend = xfs_alloc_ioend(inode, type); |
554 | ioend->io_offset = offset; | |
555 | ioend->io_buffer_head = bh; | |
556 | ioend->io_buffer_tail = bh; | |
557 | if (previous) | |
558 | previous->io_list = ioend; | |
559 | *result = ioend; | |
560 | } else { | |
561 | ioend->io_buffer_tail->b_private = bh; | |
562 | ioend->io_buffer_tail = bh; | |
563 | } | |
564 | ||
565 | bh->b_private = NULL; | |
566 | ioend->io_size += bh->b_size; | |
567 | } | |
568 | ||
87cbc49c NS |
569 | STATIC void |
570 | xfs_map_buffer( | |
046f1685 | 571 | struct inode *inode, |
87cbc49c | 572 | struct buffer_head *bh, |
207d0416 | 573 | struct xfs_bmbt_irec *imap, |
046f1685 | 574 | xfs_off_t offset) |
87cbc49c NS |
575 | { |
576 | sector_t bn; | |
8699bb0a | 577 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
578 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
579 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 580 | |
207d0416 CH |
581 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
582 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 583 | |
e513182d | 584 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 585 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 586 | |
046f1685 | 587 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
588 | |
589 | bh->b_blocknr = bn; | |
590 | set_buffer_mapped(bh); | |
591 | } | |
592 | ||
1da177e4 LT |
593 | STATIC void |
594 | xfs_map_at_offset( | |
046f1685 | 595 | struct inode *inode, |
1da177e4 | 596 | struct buffer_head *bh, |
207d0416 | 597 | struct xfs_bmbt_irec *imap, |
046f1685 | 598 | xfs_off_t offset) |
1da177e4 | 599 | { |
207d0416 CH |
600 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
601 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 LT |
602 | |
603 | lock_buffer(bh); | |
207d0416 | 604 | xfs_map_buffer(inode, bh, imap, offset); |
046f1685 | 605 | bh->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 LT |
606 | set_buffer_mapped(bh); |
607 | clear_buffer_delay(bh); | |
f6d6d4fc | 608 | clear_buffer_unwritten(bh); |
1da177e4 LT |
609 | } |
610 | ||
611 | /* | |
6c4fe19f | 612 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
613 | */ |
614 | STATIC unsigned int | |
6c4fe19f | 615 | xfs_probe_page( |
10ce4444 | 616 | struct page *page, |
20cb52eb | 617 | unsigned int pg_offset) |
1da177e4 | 618 | { |
20cb52eb | 619 | struct buffer_head *bh, *head; |
1da177e4 LT |
620 | int ret = 0; |
621 | ||
1da177e4 | 622 | if (PageWriteback(page)) |
10ce4444 | 623 | return 0; |
20cb52eb CH |
624 | if (!PageDirty(page)) |
625 | return 0; | |
626 | if (!page->mapping) | |
627 | return 0; | |
628 | if (!page_has_buffers(page)) | |
629 | return 0; | |
1da177e4 | 630 | |
20cb52eb CH |
631 | bh = head = page_buffers(page); |
632 | do { | |
633 | if (!buffer_uptodate(bh)) | |
634 | break; | |
635 | if (!buffer_mapped(bh)) | |
636 | break; | |
637 | ret += bh->b_size; | |
638 | if (ret >= pg_offset) | |
639 | break; | |
640 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 641 | |
1da177e4 LT |
642 | return ret; |
643 | } | |
644 | ||
f6d6d4fc | 645 | STATIC size_t |
6c4fe19f | 646 | xfs_probe_cluster( |
1da177e4 LT |
647 | struct inode *inode, |
648 | struct page *startpage, | |
649 | struct buffer_head *bh, | |
20cb52eb | 650 | struct buffer_head *head) |
1da177e4 | 651 | { |
10ce4444 | 652 | struct pagevec pvec; |
1da177e4 | 653 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
654 | size_t total = 0; |
655 | int done = 0, i; | |
1da177e4 LT |
656 | |
657 | /* First sum forwards in this page */ | |
658 | do { | |
20cb52eb | 659 | if (!buffer_uptodate(bh) || !buffer_mapped(bh)) |
10ce4444 | 660 | return total; |
1da177e4 LT |
661 | total += bh->b_size; |
662 | } while ((bh = bh->b_this_page) != head); | |
663 | ||
10ce4444 CH |
664 | /* if we reached the end of the page, sum forwards in following pages */ |
665 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
666 | tindex = startpage->index + 1; | |
667 | ||
668 | /* Prune this back to avoid pathological behavior */ | |
669 | tloff = min(tlast, startpage->index + 64); | |
670 | ||
671 | pagevec_init(&pvec, 0); | |
672 | while (!done && tindex <= tloff) { | |
673 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
674 | ||
675 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
676 | break; | |
677 | ||
678 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
679 | struct page *page = pvec.pages[i]; | |
265c1fac | 680 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
681 | |
682 | if (tindex == tlast) { | |
683 | pg_offset = | |
684 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
685 | if (!pg_offset) { |
686 | done = 1; | |
10ce4444 | 687 | break; |
1defeac9 | 688 | } |
10ce4444 CH |
689 | } else |
690 | pg_offset = PAGE_CACHE_SIZE; | |
691 | ||
529ae9aa | 692 | if (page->index == tindex && trylock_page(page)) { |
20cb52eb | 693 | pg_len = xfs_probe_page(page, pg_offset); |
10ce4444 CH |
694 | unlock_page(page); |
695 | } | |
696 | ||
265c1fac | 697 | if (!pg_len) { |
10ce4444 CH |
698 | done = 1; |
699 | break; | |
700 | } | |
701 | ||
265c1fac | 702 | total += pg_len; |
1defeac9 | 703 | tindex++; |
1da177e4 | 704 | } |
10ce4444 CH |
705 | |
706 | pagevec_release(&pvec); | |
707 | cond_resched(); | |
1da177e4 | 708 | } |
10ce4444 | 709 | |
1da177e4 LT |
710 | return total; |
711 | } | |
712 | ||
713 | /* | |
10ce4444 CH |
714 | * Test if a given page is suitable for writing as part of an unwritten |
715 | * or delayed allocate extent. | |
1da177e4 | 716 | */ |
10ce4444 CH |
717 | STATIC int |
718 | xfs_is_delayed_page( | |
719 | struct page *page, | |
f6d6d4fc | 720 | unsigned int type) |
1da177e4 | 721 | { |
1da177e4 | 722 | if (PageWriteback(page)) |
10ce4444 | 723 | return 0; |
1da177e4 LT |
724 | |
725 | if (page->mapping && page_has_buffers(page)) { | |
726 | struct buffer_head *bh, *head; | |
727 | int acceptable = 0; | |
728 | ||
729 | bh = head = page_buffers(page); | |
730 | do { | |
f6d6d4fc | 731 | if (buffer_unwritten(bh)) |
34a52c6c | 732 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 733 | else if (buffer_delay(bh)) |
34a52c6c | 734 | acceptable = (type == IO_DELAY); |
2ddee844 | 735 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
34a52c6c | 736 | acceptable = (type == IO_NEW); |
f6d6d4fc | 737 | else |
1da177e4 | 738 | break; |
1da177e4 LT |
739 | } while ((bh = bh->b_this_page) != head); |
740 | ||
741 | if (acceptable) | |
10ce4444 | 742 | return 1; |
1da177e4 LT |
743 | } |
744 | ||
10ce4444 | 745 | return 0; |
1da177e4 LT |
746 | } |
747 | ||
1da177e4 LT |
748 | /* |
749 | * Allocate & map buffers for page given the extent map. Write it out. | |
750 | * except for the original page of a writepage, this is called on | |
751 | * delalloc/unwritten pages only, for the original page it is possible | |
752 | * that the page has no mapping at all. | |
753 | */ | |
f6d6d4fc | 754 | STATIC int |
1da177e4 LT |
755 | xfs_convert_page( |
756 | struct inode *inode, | |
757 | struct page *page, | |
10ce4444 | 758 | loff_t tindex, |
207d0416 | 759 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 760 | xfs_ioend_t **ioendp, |
1da177e4 | 761 | struct writeback_control *wbc, |
1da177e4 LT |
762 | int all_bh) |
763 | { | |
f6d6d4fc | 764 | struct buffer_head *bh, *head; |
9260dc6b CH |
765 | xfs_off_t end_offset; |
766 | unsigned long p_offset; | |
f6d6d4fc | 767 | unsigned int type; |
24e17b5f | 768 | int len, page_dirty; |
f6d6d4fc | 769 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 770 | xfs_off_t offset = page_offset(page); |
1da177e4 | 771 | |
10ce4444 CH |
772 | if (page->index != tindex) |
773 | goto fail; | |
529ae9aa | 774 | if (!trylock_page(page)) |
10ce4444 CH |
775 | goto fail; |
776 | if (PageWriteback(page)) | |
777 | goto fail_unlock_page; | |
778 | if (page->mapping != inode->i_mapping) | |
779 | goto fail_unlock_page; | |
780 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
781 | goto fail_unlock_page; | |
782 | ||
24e17b5f NS |
783 | /* |
784 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 785 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
786 | * |
787 | * Derivation: | |
788 | * | |
789 | * End offset is the highest offset that this page should represent. | |
790 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
791 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
792 | * hence give us the correct page_dirty count. On any other page, | |
793 | * it will be zero and in that case we need page_dirty to be the | |
794 | * count of buffers on the page. | |
24e17b5f | 795 | */ |
9260dc6b CH |
796 | end_offset = min_t(unsigned long long, |
797 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
798 | i_size_read(inode)); | |
799 | ||
24e17b5f | 800 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
801 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
802 | PAGE_CACHE_SIZE); | |
803 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
804 | page_dirty = p_offset / len; | |
24e17b5f | 805 | |
1da177e4 LT |
806 | bh = head = page_buffers(page); |
807 | do { | |
9260dc6b | 808 | if (offset >= end_offset) |
1da177e4 | 809 | break; |
f6d6d4fc CH |
810 | if (!buffer_uptodate(bh)) |
811 | uptodate = 0; | |
812 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
813 | done = 1; | |
1da177e4 | 814 | continue; |
f6d6d4fc CH |
815 | } |
816 | ||
9260dc6b CH |
817 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
818 | if (buffer_unwritten(bh)) | |
34a52c6c | 819 | type = IO_UNWRITTEN; |
9260dc6b | 820 | else |
34a52c6c | 821 | type = IO_DELAY; |
9260dc6b | 822 | |
558e6891 | 823 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 824 | done = 1; |
9260dc6b CH |
825 | continue; |
826 | } | |
827 | ||
207d0416 CH |
828 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
829 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
9260dc6b | 830 | |
207d0416 | 831 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
832 | xfs_add_to_ioend(inode, bh, offset, type, |
833 | ioendp, done); | |
834 | ||
9260dc6b CH |
835 | page_dirty--; |
836 | count++; | |
837 | } else { | |
34a52c6c | 838 | type = IO_NEW; |
89f3b363 | 839 | if (buffer_mapped(bh) && all_bh) { |
1da177e4 | 840 | lock_buffer(bh); |
7336cea8 | 841 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
842 | type, ioendp, done); |
843 | count++; | |
24e17b5f | 844 | page_dirty--; |
9260dc6b CH |
845 | } else { |
846 | done = 1; | |
1da177e4 | 847 | } |
1da177e4 | 848 | } |
7336cea8 | 849 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 850 | |
f6d6d4fc CH |
851 | if (uptodate && bh == head) |
852 | SetPageUptodate(page); | |
853 | ||
89f3b363 CH |
854 | if (count) { |
855 | wbc->nr_to_write--; | |
856 | if (wbc->nr_to_write <= 0) | |
857 | done = 1; | |
1da177e4 | 858 | } |
89f3b363 | 859 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
860 | |
861 | return done; | |
10ce4444 CH |
862 | fail_unlock_page: |
863 | unlock_page(page); | |
864 | fail: | |
865 | return 1; | |
1da177e4 LT |
866 | } |
867 | ||
868 | /* | |
869 | * Convert & write out a cluster of pages in the same extent as defined | |
870 | * by mp and following the start page. | |
871 | */ | |
872 | STATIC void | |
873 | xfs_cluster_write( | |
874 | struct inode *inode, | |
875 | pgoff_t tindex, | |
207d0416 | 876 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 877 | xfs_ioend_t **ioendp, |
1da177e4 | 878 | struct writeback_control *wbc, |
1da177e4 LT |
879 | int all_bh, |
880 | pgoff_t tlast) | |
881 | { | |
10ce4444 CH |
882 | struct pagevec pvec; |
883 | int done = 0, i; | |
1da177e4 | 884 | |
10ce4444 CH |
885 | pagevec_init(&pvec, 0); |
886 | while (!done && tindex <= tlast) { | |
887 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
888 | ||
889 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 890 | break; |
10ce4444 CH |
891 | |
892 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
893 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
89f3b363 | 894 | imap, ioendp, wbc, all_bh); |
10ce4444 CH |
895 | if (done) |
896 | break; | |
897 | } | |
898 | ||
899 | pagevec_release(&pvec); | |
900 | cond_resched(); | |
1da177e4 LT |
901 | } |
902 | } | |
903 | ||
3ed3a434 DC |
904 | STATIC void |
905 | xfs_vm_invalidatepage( | |
906 | struct page *page, | |
907 | unsigned long offset) | |
908 | { | |
909 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
910 | block_invalidatepage(page, offset); | |
911 | } | |
912 | ||
913 | /* | |
914 | * If the page has delalloc buffers on it, we need to punch them out before we | |
915 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
916 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
917 | * is done on that same region - the delalloc extent is returned when none is | |
918 | * supposed to be there. | |
919 | * | |
920 | * We prevent this by truncating away the delalloc regions on the page before | |
921 | * invalidating it. Because they are delalloc, we can do this without needing a | |
922 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
923 | * truncation without a transaction as there is no space left for block | |
924 | * reservation (typically why we see a ENOSPC in writeback). | |
925 | * | |
926 | * This is not a performance critical path, so for now just do the punching a | |
927 | * buffer head at a time. | |
928 | */ | |
929 | STATIC void | |
930 | xfs_aops_discard_page( | |
931 | struct page *page) | |
932 | { | |
933 | struct inode *inode = page->mapping->host; | |
934 | struct xfs_inode *ip = XFS_I(inode); | |
935 | struct buffer_head *bh, *head; | |
936 | loff_t offset = page_offset(page); | |
937 | ssize_t len = 1 << inode->i_blkbits; | |
938 | ||
34a52c6c | 939 | if (!xfs_is_delayed_page(page, IO_DELAY)) |
3ed3a434 DC |
940 | goto out_invalidate; |
941 | ||
e8c3753c DC |
942 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
943 | goto out_invalidate; | |
944 | ||
3ed3a434 DC |
945 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
946 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
947 | page, ip->i_ino, offset); | |
948 | ||
949 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
950 | bh = head = page_buffers(page); | |
951 | do { | |
952 | int done; | |
953 | xfs_fileoff_t offset_fsb; | |
954 | xfs_bmbt_irec_t imap; | |
955 | int nimaps = 1; | |
956 | int error; | |
957 | xfs_fsblock_t firstblock; | |
958 | xfs_bmap_free_t flist; | |
959 | ||
960 | if (!buffer_delay(bh)) | |
961 | goto next_buffer; | |
962 | ||
963 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
964 | ||
965 | /* | |
966 | * Map the range first and check that it is a delalloc extent | |
967 | * before trying to unmap the range. Otherwise we will be | |
968 | * trying to remove a real extent (which requires a | |
969 | * transaction) or a hole, which is probably a bad idea... | |
970 | */ | |
971 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
972 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
b4e9181e | 973 | &nimaps, NULL); |
3ed3a434 DC |
974 | |
975 | if (error) { | |
976 | /* something screwed, just bail */ | |
e8c3753c DC |
977 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
978 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
979 | "page discard failed delalloc mapping lookup."); | |
980 | } | |
3ed3a434 DC |
981 | break; |
982 | } | |
983 | if (!nimaps) { | |
984 | /* nothing there */ | |
985 | goto next_buffer; | |
986 | } | |
987 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
988 | /* been converted, ignore */ | |
989 | goto next_buffer; | |
990 | } | |
991 | WARN_ON(imap.br_blockcount == 0); | |
992 | ||
993 | /* | |
994 | * Note: while we initialise the firstblock/flist pair, they | |
995 | * should never be used because blocks should never be | |
996 | * allocated or freed for a delalloc extent and hence we need | |
997 | * don't cancel or finish them after the xfs_bunmapi() call. | |
998 | */ | |
999 | xfs_bmap_init(&flist, &firstblock); | |
1000 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
b4e9181e | 1001 | &flist, &done); |
3ed3a434 DC |
1002 | |
1003 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
1004 | if (error) { | |
1005 | /* something screwed, just bail */ | |
e8c3753c DC |
1006 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1007 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 1008 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 1009 | } |
3ed3a434 DC |
1010 | break; |
1011 | } | |
1012 | next_buffer: | |
1013 | offset += len; | |
1014 | ||
1015 | } while ((bh = bh->b_this_page) != head); | |
1016 | ||
1017 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1018 | out_invalidate: | |
1019 | xfs_vm_invalidatepage(page, 0); | |
1020 | return; | |
1021 | } | |
1022 | ||
1da177e4 | 1023 | /* |
89f3b363 CH |
1024 | * Write out a dirty page. |
1025 | * | |
1026 | * For delalloc space on the page we need to allocate space and flush it. | |
1027 | * For unwritten space on the page we need to start the conversion to | |
1028 | * regular allocated space. | |
89f3b363 | 1029 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1030 | * |
89f3b363 CH |
1031 | * If we detect that a transaction would be required to flush the page, we |
1032 | * have to check the process flags first, if we are already in a transaction | |
1033 | * or disk I/O during allocations is off, we need to fail the writepage and | |
1034 | * redirty the page. | |
1da177e4 | 1035 | */ |
1da177e4 | 1036 | STATIC int |
89f3b363 CH |
1037 | xfs_vm_writepage( |
1038 | struct page *page, | |
1039 | struct writeback_control *wbc) | |
1da177e4 | 1040 | { |
89f3b363 | 1041 | struct inode *inode = page->mapping->host; |
20cb52eb | 1042 | int delalloc, unwritten; |
f6d6d4fc | 1043 | struct buffer_head *bh, *head; |
207d0416 | 1044 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 1045 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 1046 | loff_t offset; |
f6d6d4fc | 1047 | unsigned int type; |
1da177e4 | 1048 | __uint64_t end_offset; |
bd1556a1 | 1049 | pgoff_t end_index, last_index; |
d5cb48aa | 1050 | ssize_t size, len; |
558e6891 | 1051 | int flags, err, imap_valid = 0, uptodate = 1; |
89f3b363 | 1052 | int count = 0; |
20cb52eb | 1053 | int all_bh = 0; |
89f3b363 CH |
1054 | |
1055 | trace_xfs_writepage(inode, page, 0); | |
1056 | ||
20cb52eb CH |
1057 | ASSERT(page_has_buffers(page)); |
1058 | ||
89f3b363 CH |
1059 | /* |
1060 | * Refuse to write the page out if we are called from reclaim context. | |
1061 | * | |
d4f7a5cb CH |
1062 | * This avoids stack overflows when called from deeply used stacks in |
1063 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1064 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 CH |
1065 | * |
1066 | * This should really be done by the core VM, but until that happens | |
1067 | * filesystems like XFS, btrfs and ext4 have to take care of this | |
1068 | * by themselves. | |
1069 | */ | |
d4f7a5cb | 1070 | if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC) |
89f3b363 | 1071 | goto out_fail; |
1da177e4 | 1072 | |
89f3b363 | 1073 | /* |
20cb52eb CH |
1074 | * We need a transaction if there are delalloc or unwritten buffers |
1075 | * on the page. | |
1076 | * | |
1077 | * If we need a transaction and the process flags say we are already | |
1078 | * in a transaction, or no IO is allowed then mark the page dirty | |
1079 | * again and leave the page as is. | |
89f3b363 | 1080 | */ |
20cb52eb CH |
1081 | xfs_count_page_state(page, &delalloc, &unwritten); |
1082 | if ((current->flags & PF_FSTRANS) && (delalloc || unwritten)) | |
89f3b363 CH |
1083 | goto out_fail; |
1084 | ||
1da177e4 LT |
1085 | /* Is this page beyond the end of the file? */ |
1086 | offset = i_size_read(inode); | |
1087 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1088 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1089 | if (page->index >= end_index) { | |
1090 | if ((page->index >= end_index + 1) || | |
1091 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 1092 | unlock_page(page); |
19d5bcf3 | 1093 | return 0; |
1da177e4 LT |
1094 | } |
1095 | } | |
1096 | ||
f6d6d4fc | 1097 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
1098 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
1099 | offset); | |
24e17b5f | 1100 | len = 1 << inode->i_blkbits; |
24e17b5f | 1101 | |
24e17b5f | 1102 | bh = head = page_buffers(page); |
f6d6d4fc | 1103 | offset = page_offset(page); |
df3c7244 | 1104 | flags = BMAPI_READ; |
34a52c6c | 1105 | type = IO_NEW; |
f6d6d4fc | 1106 | |
1da177e4 LT |
1107 | do { |
1108 | if (offset >= end_offset) | |
1109 | break; | |
1110 | if (!buffer_uptodate(bh)) | |
1111 | uptodate = 0; | |
1da177e4 | 1112 | |
3d9b02e3 ES |
1113 | /* |
1114 | * A hole may still be marked uptodate because discard_buffer | |
1115 | * leaves the flag set. | |
1116 | */ | |
1117 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
1118 | ASSERT(!buffer_dirty(bh)); | |
1119 | imap_valid = 0; | |
1120 | continue; | |
1121 | } | |
1122 | ||
558e6891 CH |
1123 | if (imap_valid) |
1124 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1da177e4 | 1125 | |
20cb52eb | 1126 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
effd120e DC |
1127 | int new_ioend = 0; |
1128 | ||
df3c7244 | 1129 | /* |
6c4fe19f CH |
1130 | * Make sure we don't use a read-only iomap |
1131 | */ | |
df3c7244 | 1132 | if (flags == BMAPI_READ) |
558e6891 | 1133 | imap_valid = 0; |
6c4fe19f | 1134 | |
f6d6d4fc | 1135 | if (buffer_unwritten(bh)) { |
34a52c6c | 1136 | type = IO_UNWRITTEN; |
8272145c | 1137 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1138 | } else if (buffer_delay(bh)) { |
34a52c6c | 1139 | type = IO_DELAY; |
89f3b363 CH |
1140 | flags = BMAPI_ALLOCATE; |
1141 | ||
1142 | if (wbc->sync_mode == WB_SYNC_NONE && | |
1143 | wbc->nonblocking) | |
1144 | flags |= BMAPI_TRYLOCK; | |
f6d6d4fc CH |
1145 | } |
1146 | ||
558e6891 | 1147 | if (!imap_valid) { |
effd120e | 1148 | /* |
20cb52eb | 1149 | * If we didn't have a valid mapping then we |
effd120e DC |
1150 | * need to ensure that we put the new mapping |
1151 | * in a new ioend structure. This needs to be | |
1152 | * done to ensure that the ioends correctly | |
1153 | * reflect the block mappings at io completion | |
1154 | * for unwritten extent conversion. | |
1155 | */ | |
1156 | new_ioend = 1; | |
20cb52eb | 1157 | err = xfs_map_blocks(inode, offset, len, |
207d0416 | 1158 | &imap, flags); |
f6d6d4fc | 1159 | if (err) |
1da177e4 | 1160 | goto error; |
558e6891 CH |
1161 | imap_valid = xfs_imap_valid(inode, &imap, |
1162 | offset); | |
1da177e4 | 1163 | } |
558e6891 | 1164 | if (imap_valid) { |
207d0416 | 1165 | xfs_map_at_offset(inode, bh, &imap, offset); |
89f3b363 CH |
1166 | xfs_add_to_ioend(inode, bh, offset, type, |
1167 | &ioend, new_ioend); | |
f6d6d4fc | 1168 | count++; |
1da177e4 | 1169 | } |
89f3b363 | 1170 | } else if (buffer_uptodate(bh)) { |
6c4fe19f CH |
1171 | /* |
1172 | * we got here because the buffer is already mapped. | |
1173 | * That means it must already have extents allocated | |
1174 | * underneath it. Map the extent by reading it. | |
1175 | */ | |
558e6891 | 1176 | if (!imap_valid || flags != BMAPI_READ) { |
6c4fe19f | 1177 | flags = BMAPI_READ; |
20cb52eb | 1178 | size = xfs_probe_cluster(inode, page, bh, head); |
6c4fe19f | 1179 | err = xfs_map_blocks(inode, offset, size, |
207d0416 | 1180 | &imap, flags); |
6c4fe19f CH |
1181 | if (err) |
1182 | goto error; | |
558e6891 CH |
1183 | imap_valid = xfs_imap_valid(inode, &imap, |
1184 | offset); | |
6c4fe19f | 1185 | } |
d5cb48aa | 1186 | |
df3c7244 | 1187 | /* |
34a52c6c | 1188 | * We set the type to IO_NEW in case we are doing a |
df3c7244 DC |
1189 | * small write at EOF that is extending the file but |
1190 | * without needing an allocation. We need to update the | |
1191 | * file size on I/O completion in this case so it is | |
1192 | * the same case as having just allocated a new extent | |
1193 | * that we are writing into for the first time. | |
1194 | */ | |
34a52c6c | 1195 | type = IO_NEW; |
ca5de404 | 1196 | if (trylock_buffer(bh)) { |
558e6891 | 1197 | if (imap_valid) |
6c4fe19f | 1198 | all_bh = 1; |
7336cea8 | 1199 | xfs_add_to_ioend(inode, bh, offset, type, |
558e6891 | 1200 | &ioend, !imap_valid); |
d5cb48aa | 1201 | count++; |
f6d6d4fc | 1202 | } else { |
558e6891 | 1203 | imap_valid = 0; |
1da177e4 | 1204 | } |
89f3b363 | 1205 | } else if (PageUptodate(page)) { |
20cb52eb | 1206 | ASSERT(buffer_mapped(bh)); |
558e6891 | 1207 | imap_valid = 0; |
1da177e4 | 1208 | } |
f6d6d4fc CH |
1209 | |
1210 | if (!iohead) | |
1211 | iohead = ioend; | |
1212 | ||
1213 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1214 | |
1215 | if (uptodate && bh == head) | |
1216 | SetPageUptodate(page); | |
1217 | ||
89f3b363 | 1218 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1219 | |
558e6891 | 1220 | if (ioend && imap_valid) { |
bd1556a1 CH |
1221 | xfs_off_t end_index; |
1222 | ||
1223 | end_index = imap.br_startoff + imap.br_blockcount; | |
1224 | ||
1225 | /* to bytes */ | |
1226 | end_index <<= inode->i_blkbits; | |
1227 | ||
1228 | /* to pages */ | |
1229 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1230 | ||
1231 | /* check against file size */ | |
1232 | if (end_index > last_index) | |
1233 | end_index = last_index; | |
8699bb0a | 1234 | |
207d0416 | 1235 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
89f3b363 | 1236 | wbc, all_bh, end_index); |
1da177e4 LT |
1237 | } |
1238 | ||
f6d6d4fc | 1239 | if (iohead) |
06342cf8 | 1240 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1241 | |
89f3b363 | 1242 | return 0; |
1da177e4 LT |
1243 | |
1244 | error: | |
f6d6d4fc CH |
1245 | if (iohead) |
1246 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1247 | |
20cb52eb | 1248 | xfs_aops_discard_page(page); |
89f3b363 CH |
1249 | ClearPageUptodate(page); |
1250 | unlock_page(page); | |
1da177e4 | 1251 | return err; |
f51623b2 NS |
1252 | |
1253 | out_fail: | |
1254 | redirty_page_for_writepage(wbc, page); | |
1255 | unlock_page(page); | |
1256 | return 0; | |
f51623b2 NS |
1257 | } |
1258 | ||
7d4fb40a NS |
1259 | STATIC int |
1260 | xfs_vm_writepages( | |
1261 | struct address_space *mapping, | |
1262 | struct writeback_control *wbc) | |
1263 | { | |
b3aea4ed | 1264 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1265 | return generic_writepages(mapping, wbc); |
1266 | } | |
1267 | ||
f51623b2 NS |
1268 | /* |
1269 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1270 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1271 | * have buffer heads in this call. |
1272 | * | |
89f3b363 | 1273 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1274 | */ |
1275 | STATIC int | |
238f4c54 | 1276 | xfs_vm_releasepage( |
f51623b2 NS |
1277 | struct page *page, |
1278 | gfp_t gfp_mask) | |
1279 | { | |
20cb52eb | 1280 | int delalloc, unwritten; |
f51623b2 | 1281 | |
89f3b363 | 1282 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1283 | |
20cb52eb | 1284 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1285 | |
89f3b363 | 1286 | if (WARN_ON(delalloc)) |
f51623b2 | 1287 | return 0; |
89f3b363 | 1288 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1289 | return 0; |
1290 | ||
f51623b2 NS |
1291 | return try_to_free_buffers(page); |
1292 | } | |
1293 | ||
1da177e4 | 1294 | STATIC int |
c2536668 | 1295 | __xfs_get_blocks( |
1da177e4 LT |
1296 | struct inode *inode, |
1297 | sector_t iblock, | |
1da177e4 LT |
1298 | struct buffer_head *bh_result, |
1299 | int create, | |
f2bde9b8 | 1300 | int direct) |
1da177e4 | 1301 | { |
f2bde9b8 | 1302 | int flags = create ? BMAPI_WRITE : BMAPI_READ; |
207d0416 | 1303 | struct xfs_bmbt_irec imap; |
fdc7ed75 NS |
1304 | xfs_off_t offset; |
1305 | ssize_t size; | |
207d0416 CH |
1306 | int nimap = 1; |
1307 | int new = 0; | |
1da177e4 | 1308 | int error; |
1da177e4 | 1309 | |
fdc7ed75 | 1310 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1311 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1312 | size = bh_result->b_size; | |
364f358a LM |
1313 | |
1314 | if (!create && direct && offset >= i_size_read(inode)) | |
1315 | return 0; | |
1316 | ||
f2bde9b8 CH |
1317 | if (direct && create) |
1318 | flags |= BMAPI_DIRECT; | |
1319 | ||
1320 | error = xfs_iomap(XFS_I(inode), offset, size, flags, &imap, &nimap, | |
1321 | &new); | |
1da177e4 LT |
1322 | if (error) |
1323 | return -error; | |
207d0416 | 1324 | if (nimap == 0) |
1da177e4 LT |
1325 | return 0; |
1326 | ||
207d0416 CH |
1327 | if (imap.br_startblock != HOLESTARTBLOCK && |
1328 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1329 | /* |
1330 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1331 | * the read case (treat as if we're reading into a hole). |
1332 | */ | |
207d0416 CH |
1333 | if (create || !ISUNWRITTEN(&imap)) |
1334 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1335 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1336 | if (direct) |
1337 | bh_result->b_private = inode; | |
1338 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1339 | } |
1340 | } | |
1341 | ||
c2536668 NS |
1342 | /* |
1343 | * If this is a realtime file, data may be on a different device. | |
1344 | * to that pointed to from the buffer_head b_bdev currently. | |
1345 | */ | |
046f1685 | 1346 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1347 | |
c2536668 | 1348 | /* |
549054af DC |
1349 | * If we previously allocated a block out beyond eof and we are now |
1350 | * coming back to use it then we will need to flag it as new even if it | |
1351 | * has a disk address. | |
1352 | * | |
1353 | * With sub-block writes into unwritten extents we also need to mark | |
1354 | * the buffer as new so that the unwritten parts of the buffer gets | |
1355 | * correctly zeroed. | |
1da177e4 LT |
1356 | */ |
1357 | if (create && | |
1358 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1359 | (offset >= i_size_read(inode)) || |
207d0416 | 1360 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1361 | set_buffer_new(bh_result); |
1da177e4 | 1362 | |
207d0416 | 1363 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1364 | BUG_ON(direct); |
1365 | if (create) { | |
1366 | set_buffer_uptodate(bh_result); | |
1367 | set_buffer_mapped(bh_result); | |
1368 | set_buffer_delay(bh_result); | |
1369 | } | |
1370 | } | |
1371 | ||
2b8f12b7 CH |
1372 | /* |
1373 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1374 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1375 | */ | |
c2536668 | 1376 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1377 | xfs_off_t mapping_size; |
1378 | ||
1379 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1380 | mapping_size <<= inode->i_blkbits; | |
1381 | ||
1382 | ASSERT(mapping_size > 0); | |
1383 | if (mapping_size > size) | |
1384 | mapping_size = size; | |
1385 | if (mapping_size > LONG_MAX) | |
1386 | mapping_size = LONG_MAX; | |
1387 | ||
1388 | bh_result->b_size = mapping_size; | |
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 | { | |
f2bde9b8 | 1401 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1402 | } |
1403 | ||
1404 | STATIC int | |
e4c573bb | 1405 | xfs_get_blocks_direct( |
1da177e4 LT |
1406 | struct inode *inode, |
1407 | sector_t iblock, | |
1da177e4 LT |
1408 | struct buffer_head *bh_result, |
1409 | int create) | |
1410 | { | |
f2bde9b8 | 1411 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1412 | } |
1413 | ||
209fb87a CH |
1414 | /* |
1415 | * Complete a direct I/O write request. | |
1416 | * | |
1417 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1418 | * need to issue a transaction to convert the range from unwritten to written | |
1419 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
1420 | * to do this and we are done. But in case this was a successfull AIO | |
1421 | * request this handler is called from interrupt context, from which we | |
1422 | * can't start transactions. In that case offload the I/O completion to | |
1423 | * the workqueues we also use for buffered I/O completion. | |
1424 | */ | |
f0973863 | 1425 | STATIC void |
209fb87a CH |
1426 | xfs_end_io_direct_write( |
1427 | struct kiocb *iocb, | |
1428 | loff_t offset, | |
1429 | ssize_t size, | |
1430 | void *private, | |
1431 | int ret, | |
1432 | bool is_async) | |
f0973863 | 1433 | { |
209fb87a | 1434 | struct xfs_ioend *ioend = iocb->private; |
f0973863 CH |
1435 | |
1436 | /* | |
209fb87a CH |
1437 | * blockdev_direct_IO can return an error even after the I/O |
1438 | * completion handler was called. Thus we need to protect | |
1439 | * against double-freeing. | |
f0973863 | 1440 | */ |
209fb87a CH |
1441 | iocb->private = NULL; |
1442 | ||
ba87ea69 LM |
1443 | ioend->io_offset = offset; |
1444 | ioend->io_size = size; | |
209fb87a CH |
1445 | if (private && size > 0) |
1446 | ioend->io_type = IO_UNWRITTEN; | |
1447 | ||
1448 | if (is_async) { | |
1449 | /* | |
1450 | * If we are converting an unwritten extent we need to delay | |
1451 | * the AIO completion until after the unwrittent extent | |
1452 | * conversion has completed, otherwise do it ASAP. | |
1453 | */ | |
1454 | if (ioend->io_type == IO_UNWRITTEN) { | |
fb511f21 CH |
1455 | ioend->io_iocb = iocb; |
1456 | ioend->io_result = ret; | |
fb511f21 | 1457 | } else { |
209fb87a | 1458 | aio_complete(iocb, ret, 0); |
fb511f21 | 1459 | } |
209fb87a | 1460 | xfs_finish_ioend(ioend); |
f0973863 | 1461 | } else { |
209fb87a | 1462 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1463 | } |
f0973863 CH |
1464 | } |
1465 | ||
1da177e4 | 1466 | STATIC ssize_t |
e4c573bb | 1467 | xfs_vm_direct_IO( |
1da177e4 LT |
1468 | int rw, |
1469 | struct kiocb *iocb, | |
1470 | const struct iovec *iov, | |
1471 | loff_t offset, | |
1472 | unsigned long nr_segs) | |
1473 | { | |
209fb87a CH |
1474 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1475 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1476 | ssize_t ret; | |
1477 | ||
1478 | if (rw & WRITE) { | |
1479 | iocb->private = xfs_alloc_ioend(inode, IO_NEW); | |
1480 | ||
eafdc7d1 CH |
1481 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1482 | offset, nr_segs, | |
1483 | xfs_get_blocks_direct, | |
1484 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a CH |
1485 | if (ret != -EIOCBQUEUED && iocb->private) |
1486 | xfs_destroy_ioend(iocb->private); | |
1487 | } else { | |
eafdc7d1 CH |
1488 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1489 | offset, nr_segs, | |
1490 | xfs_get_blocks_direct, | |
1491 | NULL, NULL, 0); | |
209fb87a | 1492 | } |
f0973863 | 1493 | |
f0973863 | 1494 | return ret; |
1da177e4 LT |
1495 | } |
1496 | ||
fa9b227e CH |
1497 | STATIC void |
1498 | xfs_vm_write_failed( | |
1499 | struct address_space *mapping, | |
1500 | loff_t to) | |
1501 | { | |
1502 | struct inode *inode = mapping->host; | |
1503 | ||
1504 | if (to > inode->i_size) { | |
1505 | struct iattr ia = { | |
1506 | .ia_valid = ATTR_SIZE | ATTR_FORCE, | |
1507 | .ia_size = inode->i_size, | |
1508 | }; | |
1509 | xfs_setattr(XFS_I(inode), &ia, XFS_ATTR_NOLOCK); | |
1510 | } | |
1511 | } | |
1512 | ||
f51623b2 | 1513 | STATIC int |
d79689c7 | 1514 | xfs_vm_write_begin( |
f51623b2 | 1515 | struct file *file, |
d79689c7 NP |
1516 | struct address_space *mapping, |
1517 | loff_t pos, | |
1518 | unsigned len, | |
1519 | unsigned flags, | |
1520 | struct page **pagep, | |
1521 | void **fsdata) | |
f51623b2 | 1522 | { |
155130a4 CH |
1523 | int ret; |
1524 | ||
1525 | ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS, | |
1526 | pagep, xfs_get_blocks); | |
fa9b227e CH |
1527 | if (unlikely(ret)) |
1528 | xfs_vm_write_failed(mapping, pos + len); | |
1529 | return ret; | |
1530 | } | |
1531 | ||
1532 | STATIC int | |
1533 | xfs_vm_write_end( | |
1534 | struct file *file, | |
1535 | struct address_space *mapping, | |
1536 | loff_t pos, | |
1537 | unsigned len, | |
1538 | unsigned copied, | |
1539 | struct page *page, | |
1540 | void *fsdata) | |
1541 | { | |
1542 | int ret; | |
155130a4 | 1543 | |
fa9b227e CH |
1544 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
1545 | if (unlikely(ret < len)) | |
1546 | xfs_vm_write_failed(mapping, pos + len); | |
155130a4 | 1547 | return ret; |
f51623b2 | 1548 | } |
1da177e4 LT |
1549 | |
1550 | STATIC sector_t | |
e4c573bb | 1551 | xfs_vm_bmap( |
1da177e4 LT |
1552 | struct address_space *mapping, |
1553 | sector_t block) | |
1554 | { | |
1555 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1556 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1557 | |
cca28fb8 | 1558 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1559 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1560 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1561 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1562 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1563 | } |
1564 | ||
1565 | STATIC int | |
e4c573bb | 1566 | xfs_vm_readpage( |
1da177e4 LT |
1567 | struct file *unused, |
1568 | struct page *page) | |
1569 | { | |
c2536668 | 1570 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1571 | } |
1572 | ||
1573 | STATIC int | |
e4c573bb | 1574 | xfs_vm_readpages( |
1da177e4 LT |
1575 | struct file *unused, |
1576 | struct address_space *mapping, | |
1577 | struct list_head *pages, | |
1578 | unsigned nr_pages) | |
1579 | { | |
c2536668 | 1580 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1581 | } |
1582 | ||
f5e54d6e | 1583 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1584 | .readpage = xfs_vm_readpage, |
1585 | .readpages = xfs_vm_readpages, | |
1586 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1587 | .writepages = xfs_vm_writepages, |
1da177e4 | 1588 | .sync_page = block_sync_page, |
238f4c54 NS |
1589 | .releasepage = xfs_vm_releasepage, |
1590 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1591 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1592 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1593 | .bmap = xfs_vm_bmap, |
1594 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1595 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1596 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1597 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1598 | }; |