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 | /* | |
205 | * Schedule IO completion handling on a xfsdatad if this was | |
206 | * the final hold on this ioend. If we are asked to wait, | |
207 | * flush the workqueue. | |
208 | */ | |
209 | STATIC void | |
210 | xfs_finish_ioend( | |
211 | xfs_ioend_t *ioend, | |
212 | int wait) | |
213 | { | |
214 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
215 | struct workqueue_struct *wq; | |
216 | ||
34a52c6c | 217 | wq = (ioend->io_type == IO_UNWRITTEN) ? |
77d7a0c2 DC |
218 | xfsconvertd_workqueue : xfsdatad_workqueue; |
219 | queue_work(wq, &ioend->io_work); | |
220 | if (wait) | |
221 | flush_workqueue(wq); | |
222 | } | |
ba87ea69 LM |
223 | } |
224 | ||
0829c360 | 225 | /* |
5ec4fabb | 226 | * IO write completion. |
f6d6d4fc CH |
227 | */ |
228 | STATIC void | |
5ec4fabb | 229 | xfs_end_io( |
77d7a0c2 | 230 | struct work_struct *work) |
0829c360 | 231 | { |
77d7a0c2 DC |
232 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
233 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 234 | int error = 0; |
ba87ea69 | 235 | |
5ec4fabb CH |
236 | /* |
237 | * For unwritten extents we need to issue transactions to convert a | |
238 | * range to normal written extens after the data I/O has finished. | |
239 | */ | |
34a52c6c | 240 | if (ioend->io_type == IO_UNWRITTEN && |
5ec4fabb | 241 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { |
5ec4fabb CH |
242 | |
243 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
244 | ioend->io_size); | |
245 | if (error) | |
246 | ioend->io_error = error; | |
247 | } | |
ba87ea69 | 248 | |
5ec4fabb CH |
249 | /* |
250 | * We might have to update the on-disk file size after extending | |
251 | * writes. | |
252 | */ | |
34a52c6c | 253 | if (ioend->io_type != IO_READ) { |
77d7a0c2 DC |
254 | error = xfs_setfilesize(ioend); |
255 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 256 | } |
77d7a0c2 DC |
257 | |
258 | /* | |
259 | * If we didn't complete processing of the ioend, requeue it to the | |
260 | * tail of the workqueue for another attempt later. Otherwise destroy | |
261 | * it. | |
262 | */ | |
263 | if (error == EAGAIN) { | |
264 | atomic_inc(&ioend->io_remaining); | |
265 | xfs_finish_ioend(ioend, 0); | |
266 | /* ensure we don't spin on blocked ioends */ | |
267 | delay(1); | |
fb511f21 CH |
268 | } else { |
269 | if (ioend->io_iocb) | |
270 | aio_complete(ioend->io_iocb, ioend->io_result, 0); | |
77d7a0c2 | 271 | xfs_destroy_ioend(ioend); |
fb511f21 | 272 | } |
c626d174 DC |
273 | } |
274 | ||
0829c360 CH |
275 | /* |
276 | * Allocate and initialise an IO completion structure. | |
277 | * We need to track unwritten extent write completion here initially. | |
278 | * We'll need to extend this for updating the ondisk inode size later | |
279 | * (vs. incore size). | |
280 | */ | |
281 | STATIC xfs_ioend_t * | |
282 | xfs_alloc_ioend( | |
f6d6d4fc CH |
283 | struct inode *inode, |
284 | unsigned int type) | |
0829c360 CH |
285 | { |
286 | xfs_ioend_t *ioend; | |
287 | ||
288 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
289 | ||
290 | /* | |
291 | * Set the count to 1 initially, which will prevent an I/O | |
292 | * completion callback from happening before we have started | |
293 | * all the I/O from calling the completion routine too early. | |
294 | */ | |
295 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 296 | ioend->io_error = 0; |
f6d6d4fc CH |
297 | ioend->io_list = NULL; |
298 | ioend->io_type = type; | |
b677c210 | 299 | ioend->io_inode = inode; |
c1a073bd | 300 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 301 | ioend->io_buffer_tail = NULL; |
b677c210 | 302 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
303 | ioend->io_offset = 0; |
304 | ioend->io_size = 0; | |
fb511f21 CH |
305 | ioend->io_iocb = NULL; |
306 | ioend->io_result = 0; | |
0829c360 | 307 | |
5ec4fabb | 308 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
309 | return ioend; |
310 | } | |
311 | ||
1da177e4 LT |
312 | STATIC int |
313 | xfs_map_blocks( | |
314 | struct inode *inode, | |
315 | loff_t offset, | |
316 | ssize_t count, | |
207d0416 | 317 | struct xfs_bmbt_irec *imap, |
1da177e4 LT |
318 | int flags) |
319 | { | |
6bd16ff2 | 320 | int nmaps = 1; |
207d0416 | 321 | int new = 0; |
6bd16ff2 | 322 | |
207d0416 | 323 | return -xfs_iomap(XFS_I(inode), offset, count, flags, imap, &nmaps, &new); |
1da177e4 LT |
324 | } |
325 | ||
b8f82a4a | 326 | STATIC int |
558e6891 | 327 | xfs_imap_valid( |
8699bb0a | 328 | struct inode *inode, |
207d0416 | 329 | struct xfs_bmbt_irec *imap, |
558e6891 | 330 | xfs_off_t offset) |
1da177e4 | 331 | { |
558e6891 | 332 | offset >>= inode->i_blkbits; |
8699bb0a | 333 | |
558e6891 CH |
334 | return offset >= imap->br_startoff && |
335 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
336 | } |
337 | ||
f6d6d4fc CH |
338 | /* |
339 | * BIO completion handler for buffered IO. | |
340 | */ | |
782e3b3b | 341 | STATIC void |
f6d6d4fc CH |
342 | xfs_end_bio( |
343 | struct bio *bio, | |
f6d6d4fc CH |
344 | int error) |
345 | { | |
346 | xfs_ioend_t *ioend = bio->bi_private; | |
347 | ||
f6d6d4fc | 348 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 349 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
350 | |
351 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
352 | bio->bi_private = NULL; |
353 | bio->bi_end_io = NULL; | |
f6d6d4fc | 354 | bio_put(bio); |
7d04a335 | 355 | |
e927af90 | 356 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
357 | } |
358 | ||
359 | STATIC void | |
360 | xfs_submit_ioend_bio( | |
06342cf8 CH |
361 | struct writeback_control *wbc, |
362 | xfs_ioend_t *ioend, | |
363 | struct bio *bio) | |
f6d6d4fc CH |
364 | { |
365 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
366 | bio->bi_private = ioend; |
367 | bio->bi_end_io = xfs_end_bio; | |
368 | ||
932640e8 DC |
369 | /* |
370 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
371 | * but don't update the inode size until I/O completion. | |
372 | */ | |
373 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 374 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 375 | |
06342cf8 CH |
376 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
377 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
378 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
379 | bio_put(bio); | |
380 | } | |
381 | ||
382 | STATIC struct bio * | |
383 | xfs_alloc_ioend_bio( | |
384 | struct buffer_head *bh) | |
385 | { | |
386 | struct bio *bio; | |
387 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
388 | ||
389 | do { | |
390 | bio = bio_alloc(GFP_NOIO, nvecs); | |
391 | nvecs >>= 1; | |
392 | } while (!bio); | |
393 | ||
394 | ASSERT(bio->bi_private == NULL); | |
395 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
396 | bio->bi_bdev = bh->b_bdev; | |
397 | bio_get(bio); | |
398 | return bio; | |
399 | } | |
400 | ||
401 | STATIC void | |
402 | xfs_start_buffer_writeback( | |
403 | struct buffer_head *bh) | |
404 | { | |
405 | ASSERT(buffer_mapped(bh)); | |
406 | ASSERT(buffer_locked(bh)); | |
407 | ASSERT(!buffer_delay(bh)); | |
408 | ASSERT(!buffer_unwritten(bh)); | |
409 | ||
410 | mark_buffer_async_write(bh); | |
411 | set_buffer_uptodate(bh); | |
412 | clear_buffer_dirty(bh); | |
413 | } | |
414 | ||
415 | STATIC void | |
416 | xfs_start_page_writeback( | |
417 | struct page *page, | |
f6d6d4fc CH |
418 | int clear_dirty, |
419 | int buffers) | |
420 | { | |
421 | ASSERT(PageLocked(page)); | |
422 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 423 | if (clear_dirty) |
92132021 DC |
424 | clear_page_dirty_for_io(page); |
425 | set_page_writeback(page); | |
f6d6d4fc | 426 | unlock_page(page); |
1f7decf6 FW |
427 | /* If no buffers on the page are to be written, finish it here */ |
428 | if (!buffers) | |
f6d6d4fc | 429 | end_page_writeback(page); |
f6d6d4fc CH |
430 | } |
431 | ||
432 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
433 | { | |
434 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
435 | } | |
436 | ||
437 | /* | |
d88992f6 DC |
438 | * Submit all of the bios for all of the ioends we have saved up, covering the |
439 | * initial writepage page and also any probed pages. | |
440 | * | |
441 | * Because we may have multiple ioends spanning a page, we need to start | |
442 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
443 | * buffers as we got, then we can end up with a page that only has buffers | |
444 | * marked async write and I/O complete on can occur before we mark the other | |
445 | * buffers async write. | |
446 | * | |
447 | * The end result of this is that we trip a bug in end_page_writeback() because | |
448 | * we call it twice for the one page as the code in end_buffer_async_write() | |
449 | * assumes that all buffers on the page are started at the same time. | |
450 | * | |
451 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 452 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
453 | */ |
454 | STATIC void | |
455 | xfs_submit_ioend( | |
06342cf8 | 456 | struct writeback_control *wbc, |
f6d6d4fc CH |
457 | xfs_ioend_t *ioend) |
458 | { | |
d88992f6 | 459 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
460 | xfs_ioend_t *next; |
461 | struct buffer_head *bh; | |
462 | struct bio *bio; | |
463 | sector_t lastblock = 0; | |
464 | ||
d88992f6 DC |
465 | /* Pass 1 - start writeback */ |
466 | do { | |
467 | next = ioend->io_list; | |
468 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
469 | xfs_start_buffer_writeback(bh); | |
470 | } | |
471 | } while ((ioend = next) != NULL); | |
472 | ||
473 | /* Pass 2 - submit I/O */ | |
474 | ioend = head; | |
f6d6d4fc CH |
475 | do { |
476 | next = ioend->io_list; | |
477 | bio = NULL; | |
478 | ||
479 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
480 | |
481 | if (!bio) { | |
482 | retry: | |
483 | bio = xfs_alloc_ioend_bio(bh); | |
484 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 485 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
486 | goto retry; |
487 | } | |
488 | ||
489 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 490 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
491 | goto retry; |
492 | } | |
493 | ||
494 | lastblock = bh->b_blocknr; | |
495 | } | |
496 | if (bio) | |
06342cf8 | 497 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 498 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
499 | } while ((ioend = next) != NULL); |
500 | } | |
501 | ||
502 | /* | |
503 | * Cancel submission of all buffer_heads so far in this endio. | |
504 | * Toss the endio too. Only ever called for the initial page | |
505 | * in a writepage request, so only ever one page. | |
506 | */ | |
507 | STATIC void | |
508 | xfs_cancel_ioend( | |
509 | xfs_ioend_t *ioend) | |
510 | { | |
511 | xfs_ioend_t *next; | |
512 | struct buffer_head *bh, *next_bh; | |
513 | ||
514 | do { | |
515 | next = ioend->io_list; | |
516 | bh = ioend->io_buffer_head; | |
517 | do { | |
518 | next_bh = bh->b_private; | |
519 | clear_buffer_async_write(bh); | |
520 | unlock_buffer(bh); | |
521 | } while ((bh = next_bh) != NULL); | |
522 | ||
25e41b3d | 523 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
524 | mempool_free(ioend, xfs_ioend_pool); |
525 | } while ((ioend = next) != NULL); | |
526 | } | |
527 | ||
528 | /* | |
529 | * Test to see if we've been building up a completion structure for | |
530 | * earlier buffers -- if so, we try to append to this ioend if we | |
531 | * can, otherwise we finish off any current ioend and start another. | |
532 | * Return true if we've finished the given ioend. | |
533 | */ | |
534 | STATIC void | |
535 | xfs_add_to_ioend( | |
536 | struct inode *inode, | |
537 | struct buffer_head *bh, | |
7336cea8 | 538 | xfs_off_t offset, |
f6d6d4fc CH |
539 | unsigned int type, |
540 | xfs_ioend_t **result, | |
541 | int need_ioend) | |
542 | { | |
543 | xfs_ioend_t *ioend = *result; | |
544 | ||
545 | if (!ioend || need_ioend || type != ioend->io_type) { | |
546 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 547 | |
f6d6d4fc CH |
548 | ioend = xfs_alloc_ioend(inode, type); |
549 | ioend->io_offset = offset; | |
550 | ioend->io_buffer_head = bh; | |
551 | ioend->io_buffer_tail = bh; | |
552 | if (previous) | |
553 | previous->io_list = ioend; | |
554 | *result = ioend; | |
555 | } else { | |
556 | ioend->io_buffer_tail->b_private = bh; | |
557 | ioend->io_buffer_tail = bh; | |
558 | } | |
559 | ||
560 | bh->b_private = NULL; | |
561 | ioend->io_size += bh->b_size; | |
562 | } | |
563 | ||
87cbc49c NS |
564 | STATIC void |
565 | xfs_map_buffer( | |
046f1685 | 566 | struct inode *inode, |
87cbc49c | 567 | struct buffer_head *bh, |
207d0416 | 568 | struct xfs_bmbt_irec *imap, |
046f1685 | 569 | xfs_off_t offset) |
87cbc49c NS |
570 | { |
571 | sector_t bn; | |
8699bb0a | 572 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
573 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
574 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 575 | |
207d0416 CH |
576 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
577 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 578 | |
e513182d | 579 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 580 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 581 | |
046f1685 | 582 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
583 | |
584 | bh->b_blocknr = bn; | |
585 | set_buffer_mapped(bh); | |
586 | } | |
587 | ||
1da177e4 LT |
588 | STATIC void |
589 | xfs_map_at_offset( | |
046f1685 | 590 | struct inode *inode, |
1da177e4 | 591 | struct buffer_head *bh, |
207d0416 | 592 | struct xfs_bmbt_irec *imap, |
046f1685 | 593 | xfs_off_t offset) |
1da177e4 | 594 | { |
207d0416 CH |
595 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
596 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 LT |
597 | |
598 | lock_buffer(bh); | |
207d0416 | 599 | xfs_map_buffer(inode, bh, imap, offset); |
046f1685 | 600 | bh->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 LT |
601 | set_buffer_mapped(bh); |
602 | clear_buffer_delay(bh); | |
f6d6d4fc | 603 | clear_buffer_unwritten(bh); |
1da177e4 LT |
604 | } |
605 | ||
606 | /* | |
6c4fe19f | 607 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
608 | */ |
609 | STATIC unsigned int | |
6c4fe19f | 610 | xfs_probe_page( |
10ce4444 | 611 | struct page *page, |
20cb52eb | 612 | unsigned int pg_offset) |
1da177e4 | 613 | { |
20cb52eb | 614 | struct buffer_head *bh, *head; |
1da177e4 LT |
615 | int ret = 0; |
616 | ||
1da177e4 | 617 | if (PageWriteback(page)) |
10ce4444 | 618 | return 0; |
20cb52eb CH |
619 | if (!PageDirty(page)) |
620 | return 0; | |
621 | if (!page->mapping) | |
622 | return 0; | |
623 | if (!page_has_buffers(page)) | |
624 | return 0; | |
1da177e4 | 625 | |
20cb52eb CH |
626 | bh = head = page_buffers(page); |
627 | do { | |
628 | if (!buffer_uptodate(bh)) | |
629 | break; | |
630 | if (!buffer_mapped(bh)) | |
631 | break; | |
632 | ret += bh->b_size; | |
633 | if (ret >= pg_offset) | |
634 | break; | |
635 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 636 | |
1da177e4 LT |
637 | return ret; |
638 | } | |
639 | ||
f6d6d4fc | 640 | STATIC size_t |
6c4fe19f | 641 | xfs_probe_cluster( |
1da177e4 LT |
642 | struct inode *inode, |
643 | struct page *startpage, | |
644 | struct buffer_head *bh, | |
20cb52eb | 645 | struct buffer_head *head) |
1da177e4 | 646 | { |
10ce4444 | 647 | struct pagevec pvec; |
1da177e4 | 648 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
649 | size_t total = 0; |
650 | int done = 0, i; | |
1da177e4 LT |
651 | |
652 | /* First sum forwards in this page */ | |
653 | do { | |
20cb52eb | 654 | if (!buffer_uptodate(bh) || !buffer_mapped(bh)) |
10ce4444 | 655 | return total; |
1da177e4 LT |
656 | total += bh->b_size; |
657 | } while ((bh = bh->b_this_page) != head); | |
658 | ||
10ce4444 CH |
659 | /* if we reached the end of the page, sum forwards in following pages */ |
660 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
661 | tindex = startpage->index + 1; | |
662 | ||
663 | /* Prune this back to avoid pathological behavior */ | |
664 | tloff = min(tlast, startpage->index + 64); | |
665 | ||
666 | pagevec_init(&pvec, 0); | |
667 | while (!done && tindex <= tloff) { | |
668 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
669 | ||
670 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
671 | break; | |
672 | ||
673 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
674 | struct page *page = pvec.pages[i]; | |
265c1fac | 675 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
676 | |
677 | if (tindex == tlast) { | |
678 | pg_offset = | |
679 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
680 | if (!pg_offset) { |
681 | done = 1; | |
10ce4444 | 682 | break; |
1defeac9 | 683 | } |
10ce4444 CH |
684 | } else |
685 | pg_offset = PAGE_CACHE_SIZE; | |
686 | ||
529ae9aa | 687 | if (page->index == tindex && trylock_page(page)) { |
20cb52eb | 688 | pg_len = xfs_probe_page(page, pg_offset); |
10ce4444 CH |
689 | unlock_page(page); |
690 | } | |
691 | ||
265c1fac | 692 | if (!pg_len) { |
10ce4444 CH |
693 | done = 1; |
694 | break; | |
695 | } | |
696 | ||
265c1fac | 697 | total += pg_len; |
1defeac9 | 698 | tindex++; |
1da177e4 | 699 | } |
10ce4444 CH |
700 | |
701 | pagevec_release(&pvec); | |
702 | cond_resched(); | |
1da177e4 | 703 | } |
10ce4444 | 704 | |
1da177e4 LT |
705 | return total; |
706 | } | |
707 | ||
708 | /* | |
10ce4444 CH |
709 | * Test if a given page is suitable for writing as part of an unwritten |
710 | * or delayed allocate extent. | |
1da177e4 | 711 | */ |
10ce4444 CH |
712 | STATIC int |
713 | xfs_is_delayed_page( | |
714 | struct page *page, | |
f6d6d4fc | 715 | unsigned int type) |
1da177e4 | 716 | { |
1da177e4 | 717 | if (PageWriteback(page)) |
10ce4444 | 718 | return 0; |
1da177e4 LT |
719 | |
720 | if (page->mapping && page_has_buffers(page)) { | |
721 | struct buffer_head *bh, *head; | |
722 | int acceptable = 0; | |
723 | ||
724 | bh = head = page_buffers(page); | |
725 | do { | |
f6d6d4fc | 726 | if (buffer_unwritten(bh)) |
34a52c6c | 727 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 728 | else if (buffer_delay(bh)) |
34a52c6c | 729 | acceptable = (type == IO_DELAY); |
2ddee844 | 730 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
34a52c6c | 731 | acceptable = (type == IO_NEW); |
f6d6d4fc | 732 | else |
1da177e4 | 733 | break; |
1da177e4 LT |
734 | } while ((bh = bh->b_this_page) != head); |
735 | ||
736 | if (acceptable) | |
10ce4444 | 737 | return 1; |
1da177e4 LT |
738 | } |
739 | ||
10ce4444 | 740 | return 0; |
1da177e4 LT |
741 | } |
742 | ||
1da177e4 LT |
743 | /* |
744 | * Allocate & map buffers for page given the extent map. Write it out. | |
745 | * except for the original page of a writepage, this is called on | |
746 | * delalloc/unwritten pages only, for the original page it is possible | |
747 | * that the page has no mapping at all. | |
748 | */ | |
f6d6d4fc | 749 | STATIC int |
1da177e4 LT |
750 | xfs_convert_page( |
751 | struct inode *inode, | |
752 | struct page *page, | |
10ce4444 | 753 | loff_t tindex, |
207d0416 | 754 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 755 | xfs_ioend_t **ioendp, |
1da177e4 | 756 | struct writeback_control *wbc, |
1da177e4 LT |
757 | int all_bh) |
758 | { | |
f6d6d4fc | 759 | struct buffer_head *bh, *head; |
9260dc6b CH |
760 | xfs_off_t end_offset; |
761 | unsigned long p_offset; | |
f6d6d4fc | 762 | unsigned int type; |
24e17b5f | 763 | int len, page_dirty; |
f6d6d4fc | 764 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 765 | xfs_off_t offset = page_offset(page); |
1da177e4 | 766 | |
10ce4444 CH |
767 | if (page->index != tindex) |
768 | goto fail; | |
529ae9aa | 769 | if (!trylock_page(page)) |
10ce4444 CH |
770 | goto fail; |
771 | if (PageWriteback(page)) | |
772 | goto fail_unlock_page; | |
773 | if (page->mapping != inode->i_mapping) | |
774 | goto fail_unlock_page; | |
775 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
776 | goto fail_unlock_page; | |
777 | ||
24e17b5f NS |
778 | /* |
779 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 780 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
781 | * |
782 | * Derivation: | |
783 | * | |
784 | * End offset is the highest offset that this page should represent. | |
785 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
786 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
787 | * hence give us the correct page_dirty count. On any other page, | |
788 | * it will be zero and in that case we need page_dirty to be the | |
789 | * count of buffers on the page. | |
24e17b5f | 790 | */ |
9260dc6b CH |
791 | end_offset = min_t(unsigned long long, |
792 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
793 | i_size_read(inode)); | |
794 | ||
24e17b5f | 795 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
796 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
797 | PAGE_CACHE_SIZE); | |
798 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
799 | page_dirty = p_offset / len; | |
24e17b5f | 800 | |
1da177e4 LT |
801 | bh = head = page_buffers(page); |
802 | do { | |
9260dc6b | 803 | if (offset >= end_offset) |
1da177e4 | 804 | break; |
f6d6d4fc CH |
805 | if (!buffer_uptodate(bh)) |
806 | uptodate = 0; | |
807 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
808 | done = 1; | |
1da177e4 | 809 | continue; |
f6d6d4fc CH |
810 | } |
811 | ||
9260dc6b CH |
812 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
813 | if (buffer_unwritten(bh)) | |
34a52c6c | 814 | type = IO_UNWRITTEN; |
9260dc6b | 815 | else |
34a52c6c | 816 | type = IO_DELAY; |
9260dc6b | 817 | |
558e6891 | 818 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 819 | done = 1; |
9260dc6b CH |
820 | continue; |
821 | } | |
822 | ||
207d0416 CH |
823 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
824 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
9260dc6b | 825 | |
207d0416 | 826 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
827 | xfs_add_to_ioend(inode, bh, offset, type, |
828 | ioendp, done); | |
829 | ||
9260dc6b CH |
830 | page_dirty--; |
831 | count++; | |
832 | } else { | |
34a52c6c | 833 | type = IO_NEW; |
89f3b363 | 834 | if (buffer_mapped(bh) && all_bh) { |
1da177e4 | 835 | lock_buffer(bh); |
7336cea8 | 836 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
837 | type, ioendp, done); |
838 | count++; | |
24e17b5f | 839 | page_dirty--; |
9260dc6b CH |
840 | } else { |
841 | done = 1; | |
1da177e4 | 842 | } |
1da177e4 | 843 | } |
7336cea8 | 844 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 845 | |
f6d6d4fc CH |
846 | if (uptodate && bh == head) |
847 | SetPageUptodate(page); | |
848 | ||
89f3b363 CH |
849 | if (count) { |
850 | wbc->nr_to_write--; | |
851 | if (wbc->nr_to_write <= 0) | |
852 | done = 1; | |
1da177e4 | 853 | } |
89f3b363 | 854 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
855 | |
856 | return done; | |
10ce4444 CH |
857 | fail_unlock_page: |
858 | unlock_page(page); | |
859 | fail: | |
860 | return 1; | |
1da177e4 LT |
861 | } |
862 | ||
863 | /* | |
864 | * Convert & write out a cluster of pages in the same extent as defined | |
865 | * by mp and following the start page. | |
866 | */ | |
867 | STATIC void | |
868 | xfs_cluster_write( | |
869 | struct inode *inode, | |
870 | pgoff_t tindex, | |
207d0416 | 871 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 872 | xfs_ioend_t **ioendp, |
1da177e4 | 873 | struct writeback_control *wbc, |
1da177e4 LT |
874 | int all_bh, |
875 | pgoff_t tlast) | |
876 | { | |
10ce4444 CH |
877 | struct pagevec pvec; |
878 | int done = 0, i; | |
1da177e4 | 879 | |
10ce4444 CH |
880 | pagevec_init(&pvec, 0); |
881 | while (!done && tindex <= tlast) { | |
882 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
883 | ||
884 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 885 | break; |
10ce4444 CH |
886 | |
887 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
888 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
89f3b363 | 889 | imap, ioendp, wbc, all_bh); |
10ce4444 CH |
890 | if (done) |
891 | break; | |
892 | } | |
893 | ||
894 | pagevec_release(&pvec); | |
895 | cond_resched(); | |
1da177e4 LT |
896 | } |
897 | } | |
898 | ||
3ed3a434 DC |
899 | STATIC void |
900 | xfs_vm_invalidatepage( | |
901 | struct page *page, | |
902 | unsigned long offset) | |
903 | { | |
904 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
905 | block_invalidatepage(page, offset); | |
906 | } | |
907 | ||
908 | /* | |
909 | * If the page has delalloc buffers on it, we need to punch them out before we | |
910 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
911 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
912 | * is done on that same region - the delalloc extent is returned when none is | |
913 | * supposed to be there. | |
914 | * | |
915 | * We prevent this by truncating away the delalloc regions on the page before | |
916 | * invalidating it. Because they are delalloc, we can do this without needing a | |
917 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
918 | * truncation without a transaction as there is no space left for block | |
919 | * reservation (typically why we see a ENOSPC in writeback). | |
920 | * | |
921 | * This is not a performance critical path, so for now just do the punching a | |
922 | * buffer head at a time. | |
923 | */ | |
924 | STATIC void | |
925 | xfs_aops_discard_page( | |
926 | struct page *page) | |
927 | { | |
928 | struct inode *inode = page->mapping->host; | |
929 | struct xfs_inode *ip = XFS_I(inode); | |
930 | struct buffer_head *bh, *head; | |
931 | loff_t offset = page_offset(page); | |
932 | ssize_t len = 1 << inode->i_blkbits; | |
933 | ||
34a52c6c | 934 | if (!xfs_is_delayed_page(page, IO_DELAY)) |
3ed3a434 DC |
935 | goto out_invalidate; |
936 | ||
e8c3753c DC |
937 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
938 | goto out_invalidate; | |
939 | ||
3ed3a434 DC |
940 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
941 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
942 | page, ip->i_ino, offset); | |
943 | ||
944 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
945 | bh = head = page_buffers(page); | |
946 | do { | |
947 | int done; | |
948 | xfs_fileoff_t offset_fsb; | |
949 | xfs_bmbt_irec_t imap; | |
950 | int nimaps = 1; | |
951 | int error; | |
952 | xfs_fsblock_t firstblock; | |
953 | xfs_bmap_free_t flist; | |
954 | ||
955 | if (!buffer_delay(bh)) | |
956 | goto next_buffer; | |
957 | ||
958 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
959 | ||
960 | /* | |
961 | * Map the range first and check that it is a delalloc extent | |
962 | * before trying to unmap the range. Otherwise we will be | |
963 | * trying to remove a real extent (which requires a | |
964 | * transaction) or a hole, which is probably a bad idea... | |
965 | */ | |
966 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
967 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
b4e9181e | 968 | &nimaps, NULL); |
3ed3a434 DC |
969 | |
970 | if (error) { | |
971 | /* something screwed, just bail */ | |
e8c3753c DC |
972 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
973 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
974 | "page discard failed delalloc mapping lookup."); | |
975 | } | |
3ed3a434 DC |
976 | break; |
977 | } | |
978 | if (!nimaps) { | |
979 | /* nothing there */ | |
980 | goto next_buffer; | |
981 | } | |
982 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
983 | /* been converted, ignore */ | |
984 | goto next_buffer; | |
985 | } | |
986 | WARN_ON(imap.br_blockcount == 0); | |
987 | ||
988 | /* | |
989 | * Note: while we initialise the firstblock/flist pair, they | |
990 | * should never be used because blocks should never be | |
991 | * allocated or freed for a delalloc extent and hence we need | |
992 | * don't cancel or finish them after the xfs_bunmapi() call. | |
993 | */ | |
994 | xfs_bmap_init(&flist, &firstblock); | |
995 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
b4e9181e | 996 | &flist, &done); |
3ed3a434 DC |
997 | |
998 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
999 | if (error) { | |
1000 | /* something screwed, just bail */ | |
e8c3753c DC |
1001 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1002 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 1003 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 1004 | } |
3ed3a434 DC |
1005 | break; |
1006 | } | |
1007 | next_buffer: | |
1008 | offset += len; | |
1009 | ||
1010 | } while ((bh = bh->b_this_page) != head); | |
1011 | ||
1012 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1013 | out_invalidate: | |
1014 | xfs_vm_invalidatepage(page, 0); | |
1015 | return; | |
1016 | } | |
1017 | ||
1da177e4 | 1018 | /* |
89f3b363 CH |
1019 | * Write out a dirty page. |
1020 | * | |
1021 | * For delalloc space on the page we need to allocate space and flush it. | |
1022 | * For unwritten space on the page we need to start the conversion to | |
1023 | * regular allocated space. | |
89f3b363 | 1024 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1025 | * |
89f3b363 CH |
1026 | * If we detect that a transaction would be required to flush the page, we |
1027 | * have to check the process flags first, if we are already in a transaction | |
1028 | * or disk I/O during allocations is off, we need to fail the writepage and | |
1029 | * redirty the page. | |
1da177e4 | 1030 | */ |
1da177e4 | 1031 | STATIC int |
89f3b363 CH |
1032 | xfs_vm_writepage( |
1033 | struct page *page, | |
1034 | struct writeback_control *wbc) | |
1da177e4 | 1035 | { |
89f3b363 | 1036 | struct inode *inode = page->mapping->host; |
20cb52eb | 1037 | int delalloc, unwritten; |
f6d6d4fc | 1038 | struct buffer_head *bh, *head; |
207d0416 | 1039 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 1040 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 1041 | loff_t offset; |
f6d6d4fc | 1042 | unsigned int type; |
1da177e4 | 1043 | __uint64_t end_offset; |
bd1556a1 | 1044 | pgoff_t end_index, last_index; |
d5cb48aa | 1045 | ssize_t size, len; |
558e6891 | 1046 | int flags, err, imap_valid = 0, uptodate = 1; |
89f3b363 | 1047 | int count = 0; |
20cb52eb | 1048 | int all_bh = 0; |
89f3b363 CH |
1049 | |
1050 | trace_xfs_writepage(inode, page, 0); | |
1051 | ||
20cb52eb CH |
1052 | ASSERT(page_has_buffers(page)); |
1053 | ||
89f3b363 CH |
1054 | /* |
1055 | * Refuse to write the page out if we are called from reclaim context. | |
1056 | * | |
d4f7a5cb CH |
1057 | * This avoids stack overflows when called from deeply used stacks in |
1058 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1059 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 CH |
1060 | * |
1061 | * This should really be done by the core VM, but until that happens | |
1062 | * filesystems like XFS, btrfs and ext4 have to take care of this | |
1063 | * by themselves. | |
1064 | */ | |
d4f7a5cb | 1065 | if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC) |
89f3b363 | 1066 | goto out_fail; |
1da177e4 | 1067 | |
89f3b363 | 1068 | /* |
20cb52eb CH |
1069 | * We need a transaction if there are delalloc or unwritten buffers |
1070 | * on the page. | |
1071 | * | |
1072 | * If we need a transaction and the process flags say we are already | |
1073 | * in a transaction, or no IO is allowed then mark the page dirty | |
1074 | * again and leave the page as is. | |
89f3b363 | 1075 | */ |
20cb52eb CH |
1076 | xfs_count_page_state(page, &delalloc, &unwritten); |
1077 | if ((current->flags & PF_FSTRANS) && (delalloc || unwritten)) | |
89f3b363 CH |
1078 | goto out_fail; |
1079 | ||
1da177e4 LT |
1080 | /* Is this page beyond the end of the file? */ |
1081 | offset = i_size_read(inode); | |
1082 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1083 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1084 | if (page->index >= end_index) { | |
1085 | if ((page->index >= end_index + 1) || | |
1086 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 1087 | unlock_page(page); |
19d5bcf3 | 1088 | return 0; |
1da177e4 LT |
1089 | } |
1090 | } | |
1091 | ||
f6d6d4fc | 1092 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
1093 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
1094 | offset); | |
24e17b5f | 1095 | len = 1 << inode->i_blkbits; |
24e17b5f | 1096 | |
24e17b5f | 1097 | bh = head = page_buffers(page); |
f6d6d4fc | 1098 | offset = page_offset(page); |
df3c7244 | 1099 | flags = BMAPI_READ; |
34a52c6c | 1100 | type = IO_NEW; |
f6d6d4fc | 1101 | |
1da177e4 LT |
1102 | do { |
1103 | if (offset >= end_offset) | |
1104 | break; | |
1105 | if (!buffer_uptodate(bh)) | |
1106 | uptodate = 0; | |
1da177e4 | 1107 | |
3d9b02e3 ES |
1108 | /* |
1109 | * A hole may still be marked uptodate because discard_buffer | |
1110 | * leaves the flag set. | |
1111 | */ | |
1112 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
1113 | ASSERT(!buffer_dirty(bh)); | |
1114 | imap_valid = 0; | |
1115 | continue; | |
1116 | } | |
1117 | ||
558e6891 CH |
1118 | if (imap_valid) |
1119 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1da177e4 | 1120 | |
20cb52eb | 1121 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
effd120e DC |
1122 | int new_ioend = 0; |
1123 | ||
df3c7244 | 1124 | /* |
6c4fe19f CH |
1125 | * Make sure we don't use a read-only iomap |
1126 | */ | |
df3c7244 | 1127 | if (flags == BMAPI_READ) |
558e6891 | 1128 | imap_valid = 0; |
6c4fe19f | 1129 | |
f6d6d4fc | 1130 | if (buffer_unwritten(bh)) { |
34a52c6c | 1131 | type = IO_UNWRITTEN; |
8272145c | 1132 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1133 | } else if (buffer_delay(bh)) { |
34a52c6c | 1134 | type = IO_DELAY; |
89f3b363 CH |
1135 | flags = BMAPI_ALLOCATE; |
1136 | ||
1137 | if (wbc->sync_mode == WB_SYNC_NONE && | |
1138 | wbc->nonblocking) | |
1139 | flags |= BMAPI_TRYLOCK; | |
f6d6d4fc CH |
1140 | } |
1141 | ||
558e6891 | 1142 | if (!imap_valid) { |
effd120e | 1143 | /* |
20cb52eb | 1144 | * If we didn't have a valid mapping then we |
effd120e DC |
1145 | * need to ensure that we put the new mapping |
1146 | * in a new ioend structure. This needs to be | |
1147 | * done to ensure that the ioends correctly | |
1148 | * reflect the block mappings at io completion | |
1149 | * for unwritten extent conversion. | |
1150 | */ | |
1151 | new_ioend = 1; | |
20cb52eb | 1152 | err = xfs_map_blocks(inode, offset, len, |
207d0416 | 1153 | &imap, flags); |
f6d6d4fc | 1154 | if (err) |
1da177e4 | 1155 | goto error; |
558e6891 CH |
1156 | imap_valid = xfs_imap_valid(inode, &imap, |
1157 | offset); | |
1da177e4 | 1158 | } |
558e6891 | 1159 | if (imap_valid) { |
207d0416 | 1160 | xfs_map_at_offset(inode, bh, &imap, offset); |
89f3b363 CH |
1161 | xfs_add_to_ioend(inode, bh, offset, type, |
1162 | &ioend, new_ioend); | |
f6d6d4fc | 1163 | count++; |
1da177e4 | 1164 | } |
89f3b363 | 1165 | } else if (buffer_uptodate(bh)) { |
6c4fe19f CH |
1166 | /* |
1167 | * we got here because the buffer is already mapped. | |
1168 | * That means it must already have extents allocated | |
1169 | * underneath it. Map the extent by reading it. | |
1170 | */ | |
558e6891 | 1171 | if (!imap_valid || flags != BMAPI_READ) { |
6c4fe19f | 1172 | flags = BMAPI_READ; |
20cb52eb | 1173 | size = xfs_probe_cluster(inode, page, bh, head); |
6c4fe19f | 1174 | err = xfs_map_blocks(inode, offset, size, |
207d0416 | 1175 | &imap, flags); |
6c4fe19f CH |
1176 | if (err) |
1177 | goto error; | |
558e6891 CH |
1178 | imap_valid = xfs_imap_valid(inode, &imap, |
1179 | offset); | |
6c4fe19f | 1180 | } |
d5cb48aa | 1181 | |
df3c7244 | 1182 | /* |
34a52c6c | 1183 | * We set the type to IO_NEW in case we are doing a |
df3c7244 DC |
1184 | * small write at EOF that is extending the file but |
1185 | * without needing an allocation. We need to update the | |
1186 | * file size on I/O completion in this case so it is | |
1187 | * the same case as having just allocated a new extent | |
1188 | * that we are writing into for the first time. | |
1189 | */ | |
34a52c6c | 1190 | type = IO_NEW; |
ca5de404 | 1191 | if (trylock_buffer(bh)) { |
558e6891 | 1192 | if (imap_valid) |
6c4fe19f | 1193 | all_bh = 1; |
7336cea8 | 1194 | xfs_add_to_ioend(inode, bh, offset, type, |
558e6891 | 1195 | &ioend, !imap_valid); |
d5cb48aa | 1196 | count++; |
f6d6d4fc | 1197 | } else { |
558e6891 | 1198 | imap_valid = 0; |
1da177e4 | 1199 | } |
89f3b363 | 1200 | } else if (PageUptodate(page)) { |
20cb52eb | 1201 | ASSERT(buffer_mapped(bh)); |
558e6891 | 1202 | imap_valid = 0; |
1da177e4 | 1203 | } |
f6d6d4fc CH |
1204 | |
1205 | if (!iohead) | |
1206 | iohead = ioend; | |
1207 | ||
1208 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1209 | |
1210 | if (uptodate && bh == head) | |
1211 | SetPageUptodate(page); | |
1212 | ||
89f3b363 | 1213 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1214 | |
558e6891 | 1215 | if (ioend && imap_valid) { |
bd1556a1 CH |
1216 | xfs_off_t end_index; |
1217 | ||
1218 | end_index = imap.br_startoff + imap.br_blockcount; | |
1219 | ||
1220 | /* to bytes */ | |
1221 | end_index <<= inode->i_blkbits; | |
1222 | ||
1223 | /* to pages */ | |
1224 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1225 | ||
1226 | /* check against file size */ | |
1227 | if (end_index > last_index) | |
1228 | end_index = last_index; | |
8699bb0a | 1229 | |
207d0416 | 1230 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
89f3b363 | 1231 | wbc, all_bh, end_index); |
1da177e4 LT |
1232 | } |
1233 | ||
f6d6d4fc | 1234 | if (iohead) |
06342cf8 | 1235 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1236 | |
89f3b363 | 1237 | return 0; |
1da177e4 LT |
1238 | |
1239 | error: | |
f6d6d4fc CH |
1240 | if (iohead) |
1241 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1242 | |
20cb52eb | 1243 | xfs_aops_discard_page(page); |
89f3b363 CH |
1244 | ClearPageUptodate(page); |
1245 | unlock_page(page); | |
1da177e4 | 1246 | return err; |
f51623b2 NS |
1247 | |
1248 | out_fail: | |
1249 | redirty_page_for_writepage(wbc, page); | |
1250 | unlock_page(page); | |
1251 | return 0; | |
f51623b2 NS |
1252 | } |
1253 | ||
7d4fb40a NS |
1254 | STATIC int |
1255 | xfs_vm_writepages( | |
1256 | struct address_space *mapping, | |
1257 | struct writeback_control *wbc) | |
1258 | { | |
b3aea4ed | 1259 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1260 | return generic_writepages(mapping, wbc); |
1261 | } | |
1262 | ||
f51623b2 NS |
1263 | /* |
1264 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1265 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1266 | * have buffer heads in this call. |
1267 | * | |
89f3b363 | 1268 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1269 | */ |
1270 | STATIC int | |
238f4c54 | 1271 | xfs_vm_releasepage( |
f51623b2 NS |
1272 | struct page *page, |
1273 | gfp_t gfp_mask) | |
1274 | { | |
20cb52eb | 1275 | int delalloc, unwritten; |
f51623b2 | 1276 | |
89f3b363 | 1277 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1278 | |
20cb52eb | 1279 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1280 | |
89f3b363 | 1281 | if (WARN_ON(delalloc)) |
f51623b2 | 1282 | return 0; |
89f3b363 | 1283 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1284 | return 0; |
1285 | ||
f51623b2 NS |
1286 | return try_to_free_buffers(page); |
1287 | } | |
1288 | ||
1da177e4 | 1289 | STATIC int |
c2536668 | 1290 | __xfs_get_blocks( |
1da177e4 LT |
1291 | struct inode *inode, |
1292 | sector_t iblock, | |
1da177e4 LT |
1293 | struct buffer_head *bh_result, |
1294 | int create, | |
f2bde9b8 | 1295 | int direct) |
1da177e4 | 1296 | { |
f2bde9b8 | 1297 | int flags = create ? BMAPI_WRITE : BMAPI_READ; |
207d0416 | 1298 | struct xfs_bmbt_irec imap; |
fdc7ed75 NS |
1299 | xfs_off_t offset; |
1300 | ssize_t size; | |
207d0416 CH |
1301 | int nimap = 1; |
1302 | int new = 0; | |
1da177e4 | 1303 | int error; |
1da177e4 | 1304 | |
fdc7ed75 | 1305 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1306 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1307 | size = bh_result->b_size; | |
364f358a LM |
1308 | |
1309 | if (!create && direct && offset >= i_size_read(inode)) | |
1310 | return 0; | |
1311 | ||
f2bde9b8 CH |
1312 | if (direct && create) |
1313 | flags |= BMAPI_DIRECT; | |
1314 | ||
1315 | error = xfs_iomap(XFS_I(inode), offset, size, flags, &imap, &nimap, | |
1316 | &new); | |
1da177e4 LT |
1317 | if (error) |
1318 | return -error; | |
207d0416 | 1319 | if (nimap == 0) |
1da177e4 LT |
1320 | return 0; |
1321 | ||
207d0416 CH |
1322 | if (imap.br_startblock != HOLESTARTBLOCK && |
1323 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1324 | /* |
1325 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1326 | * the read case (treat as if we're reading into a hole). |
1327 | */ | |
207d0416 CH |
1328 | if (create || !ISUNWRITTEN(&imap)) |
1329 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1330 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1331 | if (direct) |
1332 | bh_result->b_private = inode; | |
1333 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1334 | } |
1335 | } | |
1336 | ||
c2536668 NS |
1337 | /* |
1338 | * If this is a realtime file, data may be on a different device. | |
1339 | * to that pointed to from the buffer_head b_bdev currently. | |
1340 | */ | |
046f1685 | 1341 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1342 | |
c2536668 | 1343 | /* |
549054af DC |
1344 | * If we previously allocated a block out beyond eof and we are now |
1345 | * coming back to use it then we will need to flag it as new even if it | |
1346 | * has a disk address. | |
1347 | * | |
1348 | * With sub-block writes into unwritten extents we also need to mark | |
1349 | * the buffer as new so that the unwritten parts of the buffer gets | |
1350 | * correctly zeroed. | |
1da177e4 LT |
1351 | */ |
1352 | if (create && | |
1353 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1354 | (offset >= i_size_read(inode)) || |
207d0416 | 1355 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1356 | set_buffer_new(bh_result); |
1da177e4 | 1357 | |
207d0416 | 1358 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1359 | BUG_ON(direct); |
1360 | if (create) { | |
1361 | set_buffer_uptodate(bh_result); | |
1362 | set_buffer_mapped(bh_result); | |
1363 | set_buffer_delay(bh_result); | |
1364 | } | |
1365 | } | |
1366 | ||
2b8f12b7 CH |
1367 | /* |
1368 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1369 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1370 | */ | |
c2536668 | 1371 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1372 | xfs_off_t mapping_size; |
1373 | ||
1374 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1375 | mapping_size <<= inode->i_blkbits; | |
1376 | ||
1377 | ASSERT(mapping_size > 0); | |
1378 | if (mapping_size > size) | |
1379 | mapping_size = size; | |
1380 | if (mapping_size > LONG_MAX) | |
1381 | mapping_size = LONG_MAX; | |
1382 | ||
1383 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1384 | } |
1385 | ||
1386 | return 0; | |
1387 | } | |
1388 | ||
1389 | int | |
c2536668 | 1390 | xfs_get_blocks( |
1da177e4 LT |
1391 | struct inode *inode, |
1392 | sector_t iblock, | |
1393 | struct buffer_head *bh_result, | |
1394 | int create) | |
1395 | { | |
f2bde9b8 | 1396 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1397 | } |
1398 | ||
1399 | STATIC int | |
e4c573bb | 1400 | xfs_get_blocks_direct( |
1da177e4 LT |
1401 | struct inode *inode, |
1402 | sector_t iblock, | |
1da177e4 LT |
1403 | struct buffer_head *bh_result, |
1404 | int create) | |
1405 | { | |
f2bde9b8 | 1406 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1407 | } |
1408 | ||
f0973863 | 1409 | STATIC void |
e4c573bb | 1410 | xfs_end_io_direct( |
f0973863 CH |
1411 | struct kiocb *iocb, |
1412 | loff_t offset, | |
1413 | ssize_t size, | |
40e2e973 CH |
1414 | void *private, |
1415 | int ret, | |
1416 | bool is_async) | |
f0973863 CH |
1417 | { |
1418 | xfs_ioend_t *ioend = iocb->private; | |
fb511f21 | 1419 | bool complete_aio = is_async; |
f0973863 CH |
1420 | |
1421 | /* | |
1422 | * Non-NULL private data means we need to issue a transaction to | |
1423 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1424 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1425 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1426 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1427 | * it anyway to keep the code uniform and simpler. |
1428 | * | |
e927af90 DC |
1429 | * Well, if only it were that simple. Because synchronous direct I/O |
1430 | * requires extent conversion to occur *before* we return to userspace, | |
1431 | * we have to wait for extent conversion to complete. Look at the | |
1432 | * iocb that has been passed to us to determine if this is AIO or | |
1433 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1434 | * workqueue and wait for it to complete. | |
1435 | * | |
f0973863 CH |
1436 | * The core direct I/O code might be changed to always call the |
1437 | * completion handler in the future, in which case all this can | |
1438 | * go away. | |
1439 | */ | |
ba87ea69 LM |
1440 | ioend->io_offset = offset; |
1441 | ioend->io_size = size; | |
34a52c6c | 1442 | if (ioend->io_type == IO_READ) { |
e927af90 | 1443 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1444 | } else if (private && size > 0) { |
fb511f21 CH |
1445 | if (is_async) { |
1446 | ioend->io_iocb = iocb; | |
1447 | ioend->io_result = ret; | |
1448 | complete_aio = false; | |
1449 | xfs_finish_ioend(ioend, 0); | |
1450 | } else { | |
1451 | xfs_finish_ioend(ioend, 1); | |
1452 | } | |
f0973863 | 1453 | } else { |
ba87ea69 LM |
1454 | /* |
1455 | * A direct I/O write ioend starts it's life in unwritten | |
1456 | * state in case they map an unwritten extent. This write | |
1457 | * didn't map an unwritten extent so switch it's completion | |
1458 | * handler. | |
1459 | */ | |
34a52c6c | 1460 | ioend->io_type = IO_NEW; |
e927af90 | 1461 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1462 | } |
1463 | ||
1464 | /* | |
c41564b5 | 1465 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1466 | * completion handler was called. Thus we need to protect |
1467 | * against double-freeing. | |
1468 | */ | |
1469 | iocb->private = NULL; | |
40e2e973 | 1470 | |
fb511f21 | 1471 | if (complete_aio) |
40e2e973 | 1472 | aio_complete(iocb, ret, 0); |
f0973863 CH |
1473 | } |
1474 | ||
1da177e4 | 1475 | STATIC ssize_t |
e4c573bb | 1476 | xfs_vm_direct_IO( |
1da177e4 LT |
1477 | int rw, |
1478 | struct kiocb *iocb, | |
1479 | const struct iovec *iov, | |
1480 | loff_t offset, | |
1481 | unsigned long nr_segs) | |
1482 | { | |
1483 | struct file *file = iocb->ki_filp; | |
1484 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1485 | struct block_device *bdev; |
f0973863 | 1486 | ssize_t ret; |
1da177e4 | 1487 | |
046f1685 | 1488 | bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1489 | |
5fe878ae | 1490 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
34a52c6c | 1491 | IO_UNWRITTEN : IO_READ); |
5fe878ae CH |
1492 | |
1493 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1494 | offset, nr_segs, | |
1495 | xfs_get_blocks_direct, | |
1496 | xfs_end_io_direct); | |
f0973863 | 1497 | |
8459d86a | 1498 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1499 | xfs_destroy_ioend(iocb->private); |
1500 | return ret; | |
1da177e4 LT |
1501 | } |
1502 | ||
f51623b2 | 1503 | STATIC int |
d79689c7 | 1504 | xfs_vm_write_begin( |
f51623b2 | 1505 | struct file *file, |
d79689c7 NP |
1506 | struct address_space *mapping, |
1507 | loff_t pos, | |
1508 | unsigned len, | |
1509 | unsigned flags, | |
1510 | struct page **pagep, | |
1511 | void **fsdata) | |
f51623b2 | 1512 | { |
d79689c7 | 1513 | *pagep = NULL; |
aea1b953 DC |
1514 | return block_write_begin(file, mapping, pos, len, flags | AOP_FLAG_NOFS, |
1515 | pagep, fsdata, xfs_get_blocks); | |
f51623b2 | 1516 | } |
1da177e4 LT |
1517 | |
1518 | STATIC sector_t | |
e4c573bb | 1519 | xfs_vm_bmap( |
1da177e4 LT |
1520 | struct address_space *mapping, |
1521 | sector_t block) | |
1522 | { | |
1523 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1524 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1525 | |
cca28fb8 | 1526 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1527 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1528 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1529 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1530 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1531 | } |
1532 | ||
1533 | STATIC int | |
e4c573bb | 1534 | xfs_vm_readpage( |
1da177e4 LT |
1535 | struct file *unused, |
1536 | struct page *page) | |
1537 | { | |
c2536668 | 1538 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1539 | } |
1540 | ||
1541 | STATIC int | |
e4c573bb | 1542 | xfs_vm_readpages( |
1da177e4 LT |
1543 | struct file *unused, |
1544 | struct address_space *mapping, | |
1545 | struct list_head *pages, | |
1546 | unsigned nr_pages) | |
1547 | { | |
c2536668 | 1548 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1549 | } |
1550 | ||
f5e54d6e | 1551 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1552 | .readpage = xfs_vm_readpage, |
1553 | .readpages = xfs_vm_readpages, | |
1554 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1555 | .writepages = xfs_vm_writepages, |
1da177e4 | 1556 | .sync_page = block_sync_page, |
238f4c54 NS |
1557 | .releasepage = xfs_vm_releasepage, |
1558 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1559 | .write_begin = xfs_vm_write_begin, |
1560 | .write_end = generic_write_end, | |
e4c573bb NS |
1561 | .bmap = xfs_vm_bmap, |
1562 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1563 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1564 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1565 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1566 | }; |