]>
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 | } |
384 | ||
385 | STATIC struct bio * | |
386 | xfs_alloc_ioend_bio( | |
387 | struct buffer_head *bh) | |
388 | { | |
f6d6d4fc | 389 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 390 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
391 | |
392 | ASSERT(bio->bi_private == NULL); | |
393 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
394 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
395 | return bio; |
396 | } | |
397 | ||
398 | STATIC void | |
399 | xfs_start_buffer_writeback( | |
400 | struct buffer_head *bh) | |
401 | { | |
402 | ASSERT(buffer_mapped(bh)); | |
403 | ASSERT(buffer_locked(bh)); | |
404 | ASSERT(!buffer_delay(bh)); | |
405 | ASSERT(!buffer_unwritten(bh)); | |
406 | ||
407 | mark_buffer_async_write(bh); | |
408 | set_buffer_uptodate(bh); | |
409 | clear_buffer_dirty(bh); | |
410 | } | |
411 | ||
412 | STATIC void | |
413 | xfs_start_page_writeback( | |
414 | struct page *page, | |
f6d6d4fc CH |
415 | int clear_dirty, |
416 | int buffers) | |
417 | { | |
418 | ASSERT(PageLocked(page)); | |
419 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 420 | if (clear_dirty) |
92132021 DC |
421 | clear_page_dirty_for_io(page); |
422 | set_page_writeback(page); | |
f6d6d4fc | 423 | unlock_page(page); |
1f7decf6 FW |
424 | /* If no buffers on the page are to be written, finish it here */ |
425 | if (!buffers) | |
f6d6d4fc | 426 | end_page_writeback(page); |
f6d6d4fc CH |
427 | } |
428 | ||
429 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
430 | { | |
431 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
432 | } | |
433 | ||
434 | /* | |
d88992f6 DC |
435 | * Submit all of the bios for all of the ioends we have saved up, covering the |
436 | * initial writepage page and also any probed pages. | |
437 | * | |
438 | * Because we may have multiple ioends spanning a page, we need to start | |
439 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
440 | * buffers as we got, then we can end up with a page that only has buffers | |
441 | * marked async write and I/O complete on can occur before we mark the other | |
442 | * buffers async write. | |
443 | * | |
444 | * The end result of this is that we trip a bug in end_page_writeback() because | |
445 | * we call it twice for the one page as the code in end_buffer_async_write() | |
446 | * assumes that all buffers on the page are started at the same time. | |
447 | * | |
448 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 449 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
450 | */ |
451 | STATIC void | |
452 | xfs_submit_ioend( | |
06342cf8 | 453 | struct writeback_control *wbc, |
f6d6d4fc CH |
454 | xfs_ioend_t *ioend) |
455 | { | |
d88992f6 | 456 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
457 | xfs_ioend_t *next; |
458 | struct buffer_head *bh; | |
459 | struct bio *bio; | |
460 | sector_t lastblock = 0; | |
461 | ||
d88992f6 DC |
462 | /* Pass 1 - start writeback */ |
463 | do { | |
464 | next = ioend->io_list; | |
221cb251 | 465 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 466 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
467 | } while ((ioend = next) != NULL); |
468 | ||
469 | /* Pass 2 - submit I/O */ | |
470 | ioend = head; | |
f6d6d4fc CH |
471 | do { |
472 | next = ioend->io_list; | |
473 | bio = NULL; | |
474 | ||
475 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
476 | |
477 | if (!bio) { | |
478 | retry: | |
479 | bio = xfs_alloc_ioend_bio(bh); | |
480 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 481 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
482 | goto retry; |
483 | } | |
484 | ||
485 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 486 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
487 | goto retry; |
488 | } | |
489 | ||
490 | lastblock = bh->b_blocknr; | |
491 | } | |
492 | if (bio) | |
06342cf8 | 493 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 494 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
495 | } while ((ioend = next) != NULL); |
496 | } | |
497 | ||
498 | /* | |
499 | * Cancel submission of all buffer_heads so far in this endio. | |
500 | * Toss the endio too. Only ever called for the initial page | |
501 | * in a writepage request, so only ever one page. | |
502 | */ | |
503 | STATIC void | |
504 | xfs_cancel_ioend( | |
505 | xfs_ioend_t *ioend) | |
506 | { | |
507 | xfs_ioend_t *next; | |
508 | struct buffer_head *bh, *next_bh; | |
509 | ||
510 | do { | |
511 | next = ioend->io_list; | |
512 | bh = ioend->io_buffer_head; | |
513 | do { | |
514 | next_bh = bh->b_private; | |
515 | clear_buffer_async_write(bh); | |
516 | unlock_buffer(bh); | |
517 | } while ((bh = next_bh) != NULL); | |
518 | ||
25e41b3d | 519 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
520 | mempool_free(ioend, xfs_ioend_pool); |
521 | } while ((ioend = next) != NULL); | |
522 | } | |
523 | ||
524 | /* | |
525 | * Test to see if we've been building up a completion structure for | |
526 | * earlier buffers -- if so, we try to append to this ioend if we | |
527 | * can, otherwise we finish off any current ioend and start another. | |
528 | * Return true if we've finished the given ioend. | |
529 | */ | |
530 | STATIC void | |
531 | xfs_add_to_ioend( | |
532 | struct inode *inode, | |
533 | struct buffer_head *bh, | |
7336cea8 | 534 | xfs_off_t offset, |
f6d6d4fc CH |
535 | unsigned int type, |
536 | xfs_ioend_t **result, | |
537 | int need_ioend) | |
538 | { | |
539 | xfs_ioend_t *ioend = *result; | |
540 | ||
541 | if (!ioend || need_ioend || type != ioend->io_type) { | |
542 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 543 | |
f6d6d4fc CH |
544 | ioend = xfs_alloc_ioend(inode, type); |
545 | ioend->io_offset = offset; | |
546 | ioend->io_buffer_head = bh; | |
547 | ioend->io_buffer_tail = bh; | |
548 | if (previous) | |
549 | previous->io_list = ioend; | |
550 | *result = ioend; | |
551 | } else { | |
552 | ioend->io_buffer_tail->b_private = bh; | |
553 | ioend->io_buffer_tail = bh; | |
554 | } | |
555 | ||
556 | bh->b_private = NULL; | |
557 | ioend->io_size += bh->b_size; | |
558 | } | |
559 | ||
87cbc49c NS |
560 | STATIC void |
561 | xfs_map_buffer( | |
046f1685 | 562 | struct inode *inode, |
87cbc49c | 563 | struct buffer_head *bh, |
207d0416 | 564 | struct xfs_bmbt_irec *imap, |
046f1685 | 565 | xfs_off_t offset) |
87cbc49c NS |
566 | { |
567 | sector_t bn; | |
8699bb0a | 568 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
569 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
570 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 571 | |
207d0416 CH |
572 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
573 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 574 | |
e513182d | 575 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 576 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 577 | |
046f1685 | 578 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
579 | |
580 | bh->b_blocknr = bn; | |
581 | set_buffer_mapped(bh); | |
582 | } | |
583 | ||
1da177e4 LT |
584 | STATIC void |
585 | xfs_map_at_offset( | |
046f1685 | 586 | struct inode *inode, |
1da177e4 | 587 | struct buffer_head *bh, |
207d0416 | 588 | struct xfs_bmbt_irec *imap, |
046f1685 | 589 | xfs_off_t offset) |
1da177e4 | 590 | { |
207d0416 CH |
591 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
592 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 LT |
593 | |
594 | lock_buffer(bh); | |
207d0416 | 595 | xfs_map_buffer(inode, bh, imap, offset); |
046f1685 | 596 | bh->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 LT |
597 | set_buffer_mapped(bh); |
598 | clear_buffer_delay(bh); | |
f6d6d4fc | 599 | clear_buffer_unwritten(bh); |
1da177e4 LT |
600 | } |
601 | ||
602 | /* | |
6c4fe19f | 603 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
604 | */ |
605 | STATIC unsigned int | |
6c4fe19f | 606 | xfs_probe_page( |
10ce4444 | 607 | struct page *page, |
20cb52eb | 608 | unsigned int pg_offset) |
1da177e4 | 609 | { |
20cb52eb | 610 | struct buffer_head *bh, *head; |
1da177e4 LT |
611 | int ret = 0; |
612 | ||
1da177e4 | 613 | if (PageWriteback(page)) |
10ce4444 | 614 | return 0; |
20cb52eb CH |
615 | if (!PageDirty(page)) |
616 | return 0; | |
617 | if (!page->mapping) | |
618 | return 0; | |
619 | if (!page_has_buffers(page)) | |
620 | return 0; | |
1da177e4 | 621 | |
20cb52eb CH |
622 | bh = head = page_buffers(page); |
623 | do { | |
624 | if (!buffer_uptodate(bh)) | |
625 | break; | |
626 | if (!buffer_mapped(bh)) | |
627 | break; | |
628 | ret += bh->b_size; | |
629 | if (ret >= pg_offset) | |
630 | break; | |
631 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 632 | |
1da177e4 LT |
633 | return ret; |
634 | } | |
635 | ||
f6d6d4fc | 636 | STATIC size_t |
6c4fe19f | 637 | xfs_probe_cluster( |
1da177e4 LT |
638 | struct inode *inode, |
639 | struct page *startpage, | |
640 | struct buffer_head *bh, | |
20cb52eb | 641 | struct buffer_head *head) |
1da177e4 | 642 | { |
10ce4444 | 643 | struct pagevec pvec; |
1da177e4 | 644 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
645 | size_t total = 0; |
646 | int done = 0, i; | |
1da177e4 LT |
647 | |
648 | /* First sum forwards in this page */ | |
649 | do { | |
20cb52eb | 650 | if (!buffer_uptodate(bh) || !buffer_mapped(bh)) |
10ce4444 | 651 | return total; |
1da177e4 LT |
652 | total += bh->b_size; |
653 | } while ((bh = bh->b_this_page) != head); | |
654 | ||
10ce4444 CH |
655 | /* if we reached the end of the page, sum forwards in following pages */ |
656 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
657 | tindex = startpage->index + 1; | |
658 | ||
659 | /* Prune this back to avoid pathological behavior */ | |
660 | tloff = min(tlast, startpage->index + 64); | |
661 | ||
662 | pagevec_init(&pvec, 0); | |
663 | while (!done && tindex <= tloff) { | |
664 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
665 | ||
666 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
667 | break; | |
668 | ||
669 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
670 | struct page *page = pvec.pages[i]; | |
265c1fac | 671 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
672 | |
673 | if (tindex == tlast) { | |
674 | pg_offset = | |
675 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
676 | if (!pg_offset) { |
677 | done = 1; | |
10ce4444 | 678 | break; |
1defeac9 | 679 | } |
10ce4444 CH |
680 | } else |
681 | pg_offset = PAGE_CACHE_SIZE; | |
682 | ||
529ae9aa | 683 | if (page->index == tindex && trylock_page(page)) { |
20cb52eb | 684 | pg_len = xfs_probe_page(page, pg_offset); |
10ce4444 CH |
685 | unlock_page(page); |
686 | } | |
687 | ||
265c1fac | 688 | if (!pg_len) { |
10ce4444 CH |
689 | done = 1; |
690 | break; | |
691 | } | |
692 | ||
265c1fac | 693 | total += pg_len; |
1defeac9 | 694 | tindex++; |
1da177e4 | 695 | } |
10ce4444 CH |
696 | |
697 | pagevec_release(&pvec); | |
698 | cond_resched(); | |
1da177e4 | 699 | } |
10ce4444 | 700 | |
1da177e4 LT |
701 | return total; |
702 | } | |
703 | ||
704 | /* | |
10ce4444 CH |
705 | * Test if a given page is suitable for writing as part of an unwritten |
706 | * or delayed allocate extent. | |
1da177e4 | 707 | */ |
10ce4444 CH |
708 | STATIC int |
709 | xfs_is_delayed_page( | |
710 | struct page *page, | |
f6d6d4fc | 711 | unsigned int type) |
1da177e4 | 712 | { |
1da177e4 | 713 | if (PageWriteback(page)) |
10ce4444 | 714 | return 0; |
1da177e4 LT |
715 | |
716 | if (page->mapping && page_has_buffers(page)) { | |
717 | struct buffer_head *bh, *head; | |
718 | int acceptable = 0; | |
719 | ||
720 | bh = head = page_buffers(page); | |
721 | do { | |
f6d6d4fc | 722 | if (buffer_unwritten(bh)) |
34a52c6c | 723 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 724 | else if (buffer_delay(bh)) |
34a52c6c | 725 | acceptable = (type == IO_DELAY); |
2ddee844 | 726 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
34a52c6c | 727 | acceptable = (type == IO_NEW); |
f6d6d4fc | 728 | else |
1da177e4 | 729 | break; |
1da177e4 LT |
730 | } while ((bh = bh->b_this_page) != head); |
731 | ||
732 | if (acceptable) | |
10ce4444 | 733 | return 1; |
1da177e4 LT |
734 | } |
735 | ||
10ce4444 | 736 | return 0; |
1da177e4 LT |
737 | } |
738 | ||
1da177e4 LT |
739 | /* |
740 | * Allocate & map buffers for page given the extent map. Write it out. | |
741 | * except for the original page of a writepage, this is called on | |
742 | * delalloc/unwritten pages only, for the original page it is possible | |
743 | * that the page has no mapping at all. | |
744 | */ | |
f6d6d4fc | 745 | STATIC int |
1da177e4 LT |
746 | xfs_convert_page( |
747 | struct inode *inode, | |
748 | struct page *page, | |
10ce4444 | 749 | loff_t tindex, |
207d0416 | 750 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 751 | xfs_ioend_t **ioendp, |
1da177e4 | 752 | struct writeback_control *wbc, |
1da177e4 LT |
753 | int all_bh) |
754 | { | |
f6d6d4fc | 755 | struct buffer_head *bh, *head; |
9260dc6b CH |
756 | xfs_off_t end_offset; |
757 | unsigned long p_offset; | |
f6d6d4fc | 758 | unsigned int type; |
24e17b5f | 759 | int len, page_dirty; |
f6d6d4fc | 760 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 761 | xfs_off_t offset = page_offset(page); |
1da177e4 | 762 | |
10ce4444 CH |
763 | if (page->index != tindex) |
764 | goto fail; | |
529ae9aa | 765 | if (!trylock_page(page)) |
10ce4444 CH |
766 | goto fail; |
767 | if (PageWriteback(page)) | |
768 | goto fail_unlock_page; | |
769 | if (page->mapping != inode->i_mapping) | |
770 | goto fail_unlock_page; | |
771 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
772 | goto fail_unlock_page; | |
773 | ||
24e17b5f NS |
774 | /* |
775 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 776 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
777 | * |
778 | * Derivation: | |
779 | * | |
780 | * End offset is the highest offset that this page should represent. | |
781 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
782 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
783 | * hence give us the correct page_dirty count. On any other page, | |
784 | * it will be zero and in that case we need page_dirty to be the | |
785 | * count of buffers on the page. | |
24e17b5f | 786 | */ |
9260dc6b CH |
787 | end_offset = min_t(unsigned long long, |
788 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
789 | i_size_read(inode)); | |
790 | ||
24e17b5f | 791 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
792 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
793 | PAGE_CACHE_SIZE); | |
794 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
795 | page_dirty = p_offset / len; | |
24e17b5f | 796 | |
1da177e4 LT |
797 | bh = head = page_buffers(page); |
798 | do { | |
9260dc6b | 799 | if (offset >= end_offset) |
1da177e4 | 800 | break; |
f6d6d4fc CH |
801 | if (!buffer_uptodate(bh)) |
802 | uptodate = 0; | |
803 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
804 | done = 1; | |
1da177e4 | 805 | continue; |
f6d6d4fc CH |
806 | } |
807 | ||
9260dc6b CH |
808 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
809 | if (buffer_unwritten(bh)) | |
34a52c6c | 810 | type = IO_UNWRITTEN; |
9260dc6b | 811 | else |
34a52c6c | 812 | type = IO_DELAY; |
9260dc6b | 813 | |
558e6891 | 814 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 815 | done = 1; |
9260dc6b CH |
816 | continue; |
817 | } | |
818 | ||
207d0416 CH |
819 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
820 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
9260dc6b | 821 | |
207d0416 | 822 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
823 | xfs_add_to_ioend(inode, bh, offset, type, |
824 | ioendp, done); | |
825 | ||
9260dc6b CH |
826 | page_dirty--; |
827 | count++; | |
828 | } else { | |
34a52c6c | 829 | type = IO_NEW; |
89f3b363 | 830 | if (buffer_mapped(bh) && all_bh) { |
1da177e4 | 831 | lock_buffer(bh); |
7336cea8 | 832 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
833 | type, ioendp, done); |
834 | count++; | |
24e17b5f | 835 | page_dirty--; |
9260dc6b CH |
836 | } else { |
837 | done = 1; | |
1da177e4 | 838 | } |
1da177e4 | 839 | } |
7336cea8 | 840 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 841 | |
f6d6d4fc CH |
842 | if (uptodate && bh == head) |
843 | SetPageUptodate(page); | |
844 | ||
89f3b363 | 845 | if (count) { |
efceab1d DC |
846 | if (--wbc->nr_to_write <= 0 && |
847 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 848 | done = 1; |
1da177e4 | 849 | } |
89f3b363 | 850 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
851 | |
852 | return done; | |
10ce4444 CH |
853 | fail_unlock_page: |
854 | unlock_page(page); | |
855 | fail: | |
856 | return 1; | |
1da177e4 LT |
857 | } |
858 | ||
859 | /* | |
860 | * Convert & write out a cluster of pages in the same extent as defined | |
861 | * by mp and following the start page. | |
862 | */ | |
863 | STATIC void | |
864 | xfs_cluster_write( | |
865 | struct inode *inode, | |
866 | pgoff_t tindex, | |
207d0416 | 867 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 868 | xfs_ioend_t **ioendp, |
1da177e4 | 869 | struct writeback_control *wbc, |
1da177e4 LT |
870 | int all_bh, |
871 | pgoff_t tlast) | |
872 | { | |
10ce4444 CH |
873 | struct pagevec pvec; |
874 | int done = 0, i; | |
1da177e4 | 875 | |
10ce4444 CH |
876 | pagevec_init(&pvec, 0); |
877 | while (!done && tindex <= tlast) { | |
878 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
879 | ||
880 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 881 | break; |
10ce4444 CH |
882 | |
883 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
884 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
89f3b363 | 885 | imap, ioendp, wbc, all_bh); |
10ce4444 CH |
886 | if (done) |
887 | break; | |
888 | } | |
889 | ||
890 | pagevec_release(&pvec); | |
891 | cond_resched(); | |
1da177e4 LT |
892 | } |
893 | } | |
894 | ||
3ed3a434 DC |
895 | STATIC void |
896 | xfs_vm_invalidatepage( | |
897 | struct page *page, | |
898 | unsigned long offset) | |
899 | { | |
900 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
901 | block_invalidatepage(page, offset); | |
902 | } | |
903 | ||
904 | /* | |
905 | * If the page has delalloc buffers on it, we need to punch them out before we | |
906 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
907 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
908 | * is done on that same region - the delalloc extent is returned when none is | |
909 | * supposed to be there. | |
910 | * | |
911 | * We prevent this by truncating away the delalloc regions on the page before | |
912 | * invalidating it. Because they are delalloc, we can do this without needing a | |
913 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
914 | * truncation without a transaction as there is no space left for block | |
915 | * reservation (typically why we see a ENOSPC in writeback). | |
916 | * | |
917 | * This is not a performance critical path, so for now just do the punching a | |
918 | * buffer head at a time. | |
919 | */ | |
920 | STATIC void | |
921 | xfs_aops_discard_page( | |
922 | struct page *page) | |
923 | { | |
924 | struct inode *inode = page->mapping->host; | |
925 | struct xfs_inode *ip = XFS_I(inode); | |
926 | struct buffer_head *bh, *head; | |
927 | loff_t offset = page_offset(page); | |
3ed3a434 | 928 | |
34a52c6c | 929 | if (!xfs_is_delayed_page(page, IO_DELAY)) |
3ed3a434 DC |
930 | goto out_invalidate; |
931 | ||
e8c3753c DC |
932 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
933 | goto out_invalidate; | |
934 | ||
3ed3a434 DC |
935 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
936 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
937 | page, ip->i_ino, offset); | |
938 | ||
939 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
940 | bh = head = page_buffers(page); | |
941 | do { | |
3ed3a434 | 942 | int error; |
c726de44 | 943 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
944 | |
945 | if (!buffer_delay(bh)) | |
946 | goto next_buffer; | |
947 | ||
c726de44 DC |
948 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
949 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
950 | if (error) { |
951 | /* something screwed, just bail */ | |
e8c3753c DC |
952 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
953 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 954 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 955 | } |
3ed3a434 DC |
956 | break; |
957 | } | |
958 | next_buffer: | |
c726de44 | 959 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
960 | |
961 | } while ((bh = bh->b_this_page) != head); | |
962 | ||
963 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
964 | out_invalidate: | |
965 | xfs_vm_invalidatepage(page, 0); | |
966 | return; | |
967 | } | |
968 | ||
1da177e4 | 969 | /* |
89f3b363 CH |
970 | * Write out a dirty page. |
971 | * | |
972 | * For delalloc space on the page we need to allocate space and flush it. | |
973 | * For unwritten space on the page we need to start the conversion to | |
974 | * regular allocated space. | |
89f3b363 | 975 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 976 | * |
89f3b363 CH |
977 | * If we detect that a transaction would be required to flush the page, we |
978 | * have to check the process flags first, if we are already in a transaction | |
979 | * or disk I/O during allocations is off, we need to fail the writepage and | |
980 | * redirty the page. | |
1da177e4 | 981 | */ |
1da177e4 | 982 | STATIC int |
89f3b363 CH |
983 | xfs_vm_writepage( |
984 | struct page *page, | |
985 | struct writeback_control *wbc) | |
1da177e4 | 986 | { |
89f3b363 | 987 | struct inode *inode = page->mapping->host; |
20cb52eb | 988 | int delalloc, unwritten; |
f6d6d4fc | 989 | struct buffer_head *bh, *head; |
207d0416 | 990 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 991 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 992 | loff_t offset; |
f6d6d4fc | 993 | unsigned int type; |
1da177e4 | 994 | __uint64_t end_offset; |
bd1556a1 | 995 | pgoff_t end_index, last_index; |
d5cb48aa | 996 | ssize_t size, len; |
558e6891 | 997 | int flags, err, imap_valid = 0, uptodate = 1; |
89f3b363 | 998 | int count = 0; |
20cb52eb | 999 | int all_bh = 0; |
89f3b363 CH |
1000 | |
1001 | trace_xfs_writepage(inode, page, 0); | |
1002 | ||
20cb52eb CH |
1003 | ASSERT(page_has_buffers(page)); |
1004 | ||
89f3b363 CH |
1005 | /* |
1006 | * Refuse to write the page out if we are called from reclaim context. | |
1007 | * | |
d4f7a5cb CH |
1008 | * This avoids stack overflows when called from deeply used stacks in |
1009 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1010 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 CH |
1011 | * |
1012 | * This should really be done by the core VM, but until that happens | |
1013 | * filesystems like XFS, btrfs and ext4 have to take care of this | |
1014 | * by themselves. | |
1015 | */ | |
d4f7a5cb | 1016 | if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC) |
b5420f23 | 1017 | goto redirty; |
1da177e4 | 1018 | |
89f3b363 | 1019 | /* |
20cb52eb CH |
1020 | * We need a transaction if there are delalloc or unwritten buffers |
1021 | * on the page. | |
1022 | * | |
1023 | * If we need a transaction and the process flags say we are already | |
1024 | * in a transaction, or no IO is allowed then mark the page dirty | |
1025 | * again and leave the page as is. | |
89f3b363 | 1026 | */ |
20cb52eb CH |
1027 | xfs_count_page_state(page, &delalloc, &unwritten); |
1028 | if ((current->flags & PF_FSTRANS) && (delalloc || unwritten)) | |
b5420f23 | 1029 | goto redirty; |
89f3b363 | 1030 | |
1da177e4 LT |
1031 | /* Is this page beyond the end of the file? */ |
1032 | offset = i_size_read(inode); | |
1033 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1034 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1035 | if (page->index >= end_index) { | |
1036 | if ((page->index >= end_index + 1) || | |
1037 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 1038 | unlock_page(page); |
19d5bcf3 | 1039 | return 0; |
1da177e4 LT |
1040 | } |
1041 | } | |
1042 | ||
f6d6d4fc | 1043 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
1044 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
1045 | offset); | |
24e17b5f | 1046 | len = 1 << inode->i_blkbits; |
24e17b5f | 1047 | |
24e17b5f | 1048 | bh = head = page_buffers(page); |
f6d6d4fc | 1049 | offset = page_offset(page); |
df3c7244 | 1050 | flags = BMAPI_READ; |
34a52c6c | 1051 | type = IO_NEW; |
f6d6d4fc | 1052 | |
1da177e4 | 1053 | do { |
6ac7248e CH |
1054 | int new_ioend = 0; |
1055 | ||
1da177e4 LT |
1056 | if (offset >= end_offset) |
1057 | break; | |
1058 | if (!buffer_uptodate(bh)) | |
1059 | uptodate = 0; | |
1da177e4 | 1060 | |
3d9b02e3 | 1061 | /* |
ece413f5 CH |
1062 | * set_page_dirty dirties all buffers in a page, independent |
1063 | * of their state. The dirty state however is entirely | |
1064 | * meaningless for holes (!mapped && uptodate), so skip | |
1065 | * buffers covering holes here. | |
3d9b02e3 ES |
1066 | */ |
1067 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
1068 | imap_valid = 0; |
1069 | continue; | |
1070 | } | |
1071 | ||
558e6891 CH |
1072 | if (imap_valid) |
1073 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1da177e4 | 1074 | |
20cb52eb | 1075 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
f6d6d4fc | 1076 | if (buffer_unwritten(bh)) { |
85da94c6 CH |
1077 | if (type != IO_UNWRITTEN) { |
1078 | type = IO_UNWRITTEN; | |
1079 | imap_valid = 0; | |
1080 | } | |
8272145c | 1081 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1082 | } else if (buffer_delay(bh)) { |
85da94c6 CH |
1083 | if (type != IO_DELAY) { |
1084 | type = IO_DELAY; | |
1085 | imap_valid = 0; | |
1086 | } | |
89f3b363 CH |
1087 | flags = BMAPI_ALLOCATE; |
1088 | ||
1b430bee | 1089 | if (wbc->sync_mode == WB_SYNC_NONE) |
89f3b363 | 1090 | flags |= BMAPI_TRYLOCK; |
f6d6d4fc CH |
1091 | } |
1092 | ||
558e6891 | 1093 | if (!imap_valid) { |
effd120e | 1094 | /* |
20cb52eb | 1095 | * If we didn't have a valid mapping then we |
effd120e DC |
1096 | * need to ensure that we put the new mapping |
1097 | * in a new ioend structure. This needs to be | |
1098 | * done to ensure that the ioends correctly | |
1099 | * reflect the block mappings at io completion | |
1100 | * for unwritten extent conversion. | |
1101 | */ | |
1102 | new_ioend = 1; | |
20cb52eb | 1103 | err = xfs_map_blocks(inode, offset, len, |
207d0416 | 1104 | &imap, flags); |
f6d6d4fc | 1105 | if (err) |
1da177e4 | 1106 | goto error; |
558e6891 CH |
1107 | imap_valid = xfs_imap_valid(inode, &imap, |
1108 | offset); | |
1da177e4 | 1109 | } |
558e6891 | 1110 | if (imap_valid) { |
207d0416 | 1111 | xfs_map_at_offset(inode, bh, &imap, offset); |
89f3b363 CH |
1112 | xfs_add_to_ioend(inode, bh, offset, type, |
1113 | &ioend, new_ioend); | |
f6d6d4fc | 1114 | count++; |
1da177e4 | 1115 | } |
89f3b363 | 1116 | } else if (buffer_uptodate(bh)) { |
6c4fe19f CH |
1117 | /* |
1118 | * we got here because the buffer is already mapped. | |
1119 | * That means it must already have extents allocated | |
1120 | * underneath it. Map the extent by reading it. | |
1121 | */ | |
85da94c6 | 1122 | if (flags != BMAPI_READ) { |
6c4fe19f | 1123 | flags = BMAPI_READ; |
85da94c6 CH |
1124 | imap_valid = 0; |
1125 | } | |
1126 | if (!imap_valid) { | |
6ac7248e | 1127 | new_ioend = 1; |
20cb52eb | 1128 | size = xfs_probe_cluster(inode, page, bh, head); |
6c4fe19f | 1129 | err = xfs_map_blocks(inode, offset, size, |
207d0416 | 1130 | &imap, flags); |
6c4fe19f CH |
1131 | if (err) |
1132 | goto error; | |
558e6891 CH |
1133 | imap_valid = xfs_imap_valid(inode, &imap, |
1134 | offset); | |
6c4fe19f | 1135 | } |
d5cb48aa | 1136 | |
df3c7244 | 1137 | /* |
34a52c6c | 1138 | * We set the type to IO_NEW in case we are doing a |
df3c7244 DC |
1139 | * small write at EOF that is extending the file but |
1140 | * without needing an allocation. We need to update the | |
1141 | * file size on I/O completion in this case so it is | |
1142 | * the same case as having just allocated a new extent | |
1143 | * that we are writing into for the first time. | |
1144 | */ | |
34a52c6c | 1145 | type = IO_NEW; |
6ac7248e CH |
1146 | if (imap_valid) { |
1147 | all_bh = 1; | |
1148 | lock_buffer(bh); | |
7336cea8 | 1149 | xfs_add_to_ioend(inode, bh, offset, type, |
6ac7248e | 1150 | &ioend, new_ioend); |
d5cb48aa | 1151 | count++; |
1da177e4 | 1152 | } |
89f3b363 | 1153 | } else if (PageUptodate(page)) { |
20cb52eb | 1154 | ASSERT(buffer_mapped(bh)); |
558e6891 | 1155 | imap_valid = 0; |
1da177e4 | 1156 | } |
f6d6d4fc CH |
1157 | |
1158 | if (!iohead) | |
1159 | iohead = ioend; | |
1160 | ||
1161 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1162 | |
1163 | if (uptodate && bh == head) | |
1164 | SetPageUptodate(page); | |
1165 | ||
89f3b363 | 1166 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1167 | |
558e6891 | 1168 | if (ioend && imap_valid) { |
bd1556a1 CH |
1169 | xfs_off_t end_index; |
1170 | ||
1171 | end_index = imap.br_startoff + imap.br_blockcount; | |
1172 | ||
1173 | /* to bytes */ | |
1174 | end_index <<= inode->i_blkbits; | |
1175 | ||
1176 | /* to pages */ | |
1177 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1178 | ||
1179 | /* check against file size */ | |
1180 | if (end_index > last_index) | |
1181 | end_index = last_index; | |
8699bb0a | 1182 | |
207d0416 | 1183 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
89f3b363 | 1184 | wbc, all_bh, end_index); |
1da177e4 LT |
1185 | } |
1186 | ||
f6d6d4fc | 1187 | if (iohead) |
06342cf8 | 1188 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1189 | |
89f3b363 | 1190 | return 0; |
1da177e4 LT |
1191 | |
1192 | error: | |
f6d6d4fc CH |
1193 | if (iohead) |
1194 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1195 | |
b5420f23 CH |
1196 | if (err == -EAGAIN) |
1197 | goto redirty; | |
1198 | ||
20cb52eb | 1199 | xfs_aops_discard_page(page); |
89f3b363 CH |
1200 | ClearPageUptodate(page); |
1201 | unlock_page(page); | |
1da177e4 | 1202 | return err; |
f51623b2 | 1203 | |
b5420f23 | 1204 | redirty: |
f51623b2 NS |
1205 | redirty_page_for_writepage(wbc, page); |
1206 | unlock_page(page); | |
1207 | return 0; | |
f51623b2 NS |
1208 | } |
1209 | ||
7d4fb40a NS |
1210 | STATIC int |
1211 | xfs_vm_writepages( | |
1212 | struct address_space *mapping, | |
1213 | struct writeback_control *wbc) | |
1214 | { | |
b3aea4ed | 1215 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1216 | return generic_writepages(mapping, wbc); |
1217 | } | |
1218 | ||
f51623b2 NS |
1219 | /* |
1220 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1221 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1222 | * have buffer heads in this call. |
1223 | * | |
89f3b363 | 1224 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1225 | */ |
1226 | STATIC int | |
238f4c54 | 1227 | xfs_vm_releasepage( |
f51623b2 NS |
1228 | struct page *page, |
1229 | gfp_t gfp_mask) | |
1230 | { | |
20cb52eb | 1231 | int delalloc, unwritten; |
f51623b2 | 1232 | |
89f3b363 | 1233 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1234 | |
20cb52eb | 1235 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1236 | |
89f3b363 | 1237 | if (WARN_ON(delalloc)) |
f51623b2 | 1238 | return 0; |
89f3b363 | 1239 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1240 | return 0; |
1241 | ||
f51623b2 NS |
1242 | return try_to_free_buffers(page); |
1243 | } | |
1244 | ||
1da177e4 | 1245 | STATIC int |
c2536668 | 1246 | __xfs_get_blocks( |
1da177e4 LT |
1247 | struct inode *inode, |
1248 | sector_t iblock, | |
1da177e4 LT |
1249 | struct buffer_head *bh_result, |
1250 | int create, | |
f2bde9b8 | 1251 | int direct) |
1da177e4 | 1252 | { |
f2bde9b8 | 1253 | int flags = create ? BMAPI_WRITE : BMAPI_READ; |
207d0416 | 1254 | struct xfs_bmbt_irec imap; |
fdc7ed75 NS |
1255 | xfs_off_t offset; |
1256 | ssize_t size; | |
207d0416 CH |
1257 | int nimap = 1; |
1258 | int new = 0; | |
1da177e4 | 1259 | int error; |
1da177e4 | 1260 | |
fdc7ed75 | 1261 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1262 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1263 | size = bh_result->b_size; | |
364f358a LM |
1264 | |
1265 | if (!create && direct && offset >= i_size_read(inode)) | |
1266 | return 0; | |
1267 | ||
f2bde9b8 CH |
1268 | if (direct && create) |
1269 | flags |= BMAPI_DIRECT; | |
1270 | ||
1271 | error = xfs_iomap(XFS_I(inode), offset, size, flags, &imap, &nimap, | |
1272 | &new); | |
1da177e4 LT |
1273 | if (error) |
1274 | return -error; | |
207d0416 | 1275 | if (nimap == 0) |
1da177e4 LT |
1276 | return 0; |
1277 | ||
207d0416 CH |
1278 | if (imap.br_startblock != HOLESTARTBLOCK && |
1279 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1280 | /* |
1281 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1282 | * the read case (treat as if we're reading into a hole). |
1283 | */ | |
207d0416 CH |
1284 | if (create || !ISUNWRITTEN(&imap)) |
1285 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1286 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1287 | if (direct) |
1288 | bh_result->b_private = inode; | |
1289 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1290 | } |
1291 | } | |
1292 | ||
c2536668 NS |
1293 | /* |
1294 | * If this is a realtime file, data may be on a different device. | |
1295 | * to that pointed to from the buffer_head b_bdev currently. | |
1296 | */ | |
046f1685 | 1297 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1298 | |
c2536668 | 1299 | /* |
549054af DC |
1300 | * If we previously allocated a block out beyond eof and we are now |
1301 | * coming back to use it then we will need to flag it as new even if it | |
1302 | * has a disk address. | |
1303 | * | |
1304 | * With sub-block writes into unwritten extents we also need to mark | |
1305 | * the buffer as new so that the unwritten parts of the buffer gets | |
1306 | * correctly zeroed. | |
1da177e4 LT |
1307 | */ |
1308 | if (create && | |
1309 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1310 | (offset >= i_size_read(inode)) || |
207d0416 | 1311 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1312 | set_buffer_new(bh_result); |
1da177e4 | 1313 | |
207d0416 | 1314 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1315 | BUG_ON(direct); |
1316 | if (create) { | |
1317 | set_buffer_uptodate(bh_result); | |
1318 | set_buffer_mapped(bh_result); | |
1319 | set_buffer_delay(bh_result); | |
1320 | } | |
1321 | } | |
1322 | ||
2b8f12b7 CH |
1323 | /* |
1324 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1325 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1326 | */ | |
c2536668 | 1327 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1328 | xfs_off_t mapping_size; |
1329 | ||
1330 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1331 | mapping_size <<= inode->i_blkbits; | |
1332 | ||
1333 | ASSERT(mapping_size > 0); | |
1334 | if (mapping_size > size) | |
1335 | mapping_size = size; | |
1336 | if (mapping_size > LONG_MAX) | |
1337 | mapping_size = LONG_MAX; | |
1338 | ||
1339 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1340 | } |
1341 | ||
1342 | return 0; | |
1343 | } | |
1344 | ||
1345 | int | |
c2536668 | 1346 | xfs_get_blocks( |
1da177e4 LT |
1347 | struct inode *inode, |
1348 | sector_t iblock, | |
1349 | struct buffer_head *bh_result, | |
1350 | int create) | |
1351 | { | |
f2bde9b8 | 1352 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1353 | } |
1354 | ||
1355 | STATIC int | |
e4c573bb | 1356 | xfs_get_blocks_direct( |
1da177e4 LT |
1357 | struct inode *inode, |
1358 | sector_t iblock, | |
1da177e4 LT |
1359 | struct buffer_head *bh_result, |
1360 | int create) | |
1361 | { | |
f2bde9b8 | 1362 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1363 | } |
1364 | ||
209fb87a CH |
1365 | /* |
1366 | * Complete a direct I/O write request. | |
1367 | * | |
1368 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1369 | * need to issue a transaction to convert the range from unwritten to written | |
1370 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
1371 | * to do this and we are done. But in case this was a successfull AIO | |
1372 | * request this handler is called from interrupt context, from which we | |
1373 | * can't start transactions. In that case offload the I/O completion to | |
1374 | * the workqueues we also use for buffered I/O completion. | |
1375 | */ | |
f0973863 | 1376 | STATIC void |
209fb87a CH |
1377 | xfs_end_io_direct_write( |
1378 | struct kiocb *iocb, | |
1379 | loff_t offset, | |
1380 | ssize_t size, | |
1381 | void *private, | |
1382 | int ret, | |
1383 | bool is_async) | |
f0973863 | 1384 | { |
209fb87a | 1385 | struct xfs_ioend *ioend = iocb->private; |
f0973863 CH |
1386 | |
1387 | /* | |
209fb87a CH |
1388 | * blockdev_direct_IO can return an error even after the I/O |
1389 | * completion handler was called. Thus we need to protect | |
1390 | * against double-freeing. | |
f0973863 | 1391 | */ |
209fb87a CH |
1392 | iocb->private = NULL; |
1393 | ||
ba87ea69 LM |
1394 | ioend->io_offset = offset; |
1395 | ioend->io_size = size; | |
209fb87a CH |
1396 | if (private && size > 0) |
1397 | ioend->io_type = IO_UNWRITTEN; | |
1398 | ||
1399 | if (is_async) { | |
1400 | /* | |
1401 | * If we are converting an unwritten extent we need to delay | |
1402 | * the AIO completion until after the unwrittent extent | |
1403 | * conversion has completed, otherwise do it ASAP. | |
1404 | */ | |
1405 | if (ioend->io_type == IO_UNWRITTEN) { | |
fb511f21 CH |
1406 | ioend->io_iocb = iocb; |
1407 | ioend->io_result = ret; | |
fb511f21 | 1408 | } else { |
209fb87a | 1409 | aio_complete(iocb, ret, 0); |
fb511f21 | 1410 | } |
209fb87a | 1411 | xfs_finish_ioend(ioend); |
f0973863 | 1412 | } else { |
209fb87a | 1413 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1414 | } |
f0973863 CH |
1415 | } |
1416 | ||
1da177e4 | 1417 | STATIC ssize_t |
e4c573bb | 1418 | xfs_vm_direct_IO( |
1da177e4 LT |
1419 | int rw, |
1420 | struct kiocb *iocb, | |
1421 | const struct iovec *iov, | |
1422 | loff_t offset, | |
1423 | unsigned long nr_segs) | |
1424 | { | |
209fb87a CH |
1425 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1426 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1427 | ssize_t ret; | |
1428 | ||
1429 | if (rw & WRITE) { | |
1430 | iocb->private = xfs_alloc_ioend(inode, IO_NEW); | |
1431 | ||
eafdc7d1 CH |
1432 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1433 | offset, nr_segs, | |
1434 | xfs_get_blocks_direct, | |
1435 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a CH |
1436 | if (ret != -EIOCBQUEUED && iocb->private) |
1437 | xfs_destroy_ioend(iocb->private); | |
1438 | } else { | |
eafdc7d1 CH |
1439 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1440 | offset, nr_segs, | |
1441 | xfs_get_blocks_direct, | |
1442 | NULL, NULL, 0); | |
209fb87a | 1443 | } |
f0973863 | 1444 | |
f0973863 | 1445 | return ret; |
1da177e4 LT |
1446 | } |
1447 | ||
fa9b227e CH |
1448 | STATIC void |
1449 | xfs_vm_write_failed( | |
1450 | struct address_space *mapping, | |
1451 | loff_t to) | |
1452 | { | |
1453 | struct inode *inode = mapping->host; | |
1454 | ||
1455 | if (to > inode->i_size) { | |
c726de44 DC |
1456 | /* |
1457 | * punch out the delalloc blocks we have already allocated. We | |
1458 | * don't call xfs_setattr() to do this as we may be in the | |
1459 | * middle of a multi-iovec write and so the vfs inode->i_size | |
1460 | * will not match the xfs ip->i_size and so it will zero too | |
1461 | * much. Hence we jus truncate the page cache to zero what is | |
1462 | * necessary and punch the delalloc blocks directly. | |
1463 | */ | |
1464 | struct xfs_inode *ip = XFS_I(inode); | |
1465 | xfs_fileoff_t start_fsb; | |
1466 | xfs_fileoff_t end_fsb; | |
1467 | int error; | |
1468 | ||
1469 | truncate_pagecache(inode, to, inode->i_size); | |
1470 | ||
1471 | /* | |
1472 | * Check if there are any blocks that are outside of i_size | |
1473 | * that need to be trimmed back. | |
1474 | */ | |
1475 | start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1; | |
1476 | end_fsb = XFS_B_TO_FSB(ip->i_mount, to); | |
1477 | if (end_fsb <= start_fsb) | |
1478 | return; | |
1479 | ||
1480 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1481 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1482 | end_fsb - start_fsb); | |
1483 | if (error) { | |
1484 | /* something screwed, just bail */ | |
1485 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1486 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
1487 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1488 | ip->i_ino); | |
1489 | } | |
1490 | } | |
1491 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
fa9b227e CH |
1492 | } |
1493 | } | |
1494 | ||
f51623b2 | 1495 | STATIC int |
d79689c7 | 1496 | xfs_vm_write_begin( |
f51623b2 | 1497 | struct file *file, |
d79689c7 NP |
1498 | struct address_space *mapping, |
1499 | loff_t pos, | |
1500 | unsigned len, | |
1501 | unsigned flags, | |
1502 | struct page **pagep, | |
1503 | void **fsdata) | |
f51623b2 | 1504 | { |
155130a4 CH |
1505 | int ret; |
1506 | ||
1507 | ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS, | |
1508 | pagep, xfs_get_blocks); | |
fa9b227e CH |
1509 | if (unlikely(ret)) |
1510 | xfs_vm_write_failed(mapping, pos + len); | |
1511 | return ret; | |
1512 | } | |
1513 | ||
1514 | STATIC int | |
1515 | xfs_vm_write_end( | |
1516 | struct file *file, | |
1517 | struct address_space *mapping, | |
1518 | loff_t pos, | |
1519 | unsigned len, | |
1520 | unsigned copied, | |
1521 | struct page *page, | |
1522 | void *fsdata) | |
1523 | { | |
1524 | int ret; | |
155130a4 | 1525 | |
fa9b227e CH |
1526 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
1527 | if (unlikely(ret < len)) | |
1528 | xfs_vm_write_failed(mapping, pos + len); | |
155130a4 | 1529 | return ret; |
f51623b2 | 1530 | } |
1da177e4 LT |
1531 | |
1532 | STATIC sector_t | |
e4c573bb | 1533 | xfs_vm_bmap( |
1da177e4 LT |
1534 | struct address_space *mapping, |
1535 | sector_t block) | |
1536 | { | |
1537 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1538 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1539 | |
cca28fb8 | 1540 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1541 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1542 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1543 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1544 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | STATIC int | |
e4c573bb | 1548 | xfs_vm_readpage( |
1da177e4 LT |
1549 | struct file *unused, |
1550 | struct page *page) | |
1551 | { | |
c2536668 | 1552 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1553 | } |
1554 | ||
1555 | STATIC int | |
e4c573bb | 1556 | xfs_vm_readpages( |
1da177e4 LT |
1557 | struct file *unused, |
1558 | struct address_space *mapping, | |
1559 | struct list_head *pages, | |
1560 | unsigned nr_pages) | |
1561 | { | |
c2536668 | 1562 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1563 | } |
1564 | ||
f5e54d6e | 1565 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1566 | .readpage = xfs_vm_readpage, |
1567 | .readpages = xfs_vm_readpages, | |
1568 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1569 | .writepages = xfs_vm_writepages, |
1da177e4 | 1570 | .sync_page = block_sync_page, |
238f4c54 NS |
1571 | .releasepage = xfs_vm_releasepage, |
1572 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1573 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1574 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1575 | .bmap = xfs_vm_bmap, |
1576 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1577 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1578 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1579 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1580 | }; |