]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_bit.h" |
1da177e4 | 20 | #include "xfs_log.h" |
a844f451 | 21 | #include "xfs_inum.h" |
1da177e4 | 22 | #include "xfs_sb.h" |
a844f451 | 23 | #include "xfs_ag.h" |
1da177e4 LT |
24 | #include "xfs_dir2.h" |
25 | #include "xfs_trans.h" | |
26 | #include "xfs_dmapi.h" | |
27 | #include "xfs_mount.h" | |
28 | #include "xfs_bmap_btree.h" | |
29 | #include "xfs_alloc_btree.h" | |
30 | #include "xfs_ialloc_btree.h" | |
1da177e4 | 31 | #include "xfs_dir2_sf.h" |
a844f451 | 32 | #include "xfs_attr_sf.h" |
1da177e4 LT |
33 | #include "xfs_dinode.h" |
34 | #include "xfs_inode.h" | |
a844f451 NS |
35 | #include "xfs_alloc.h" |
36 | #include "xfs_btree.h" | |
1da177e4 LT |
37 | #include "xfs_error.h" |
38 | #include "xfs_rw.h" | |
39 | #include "xfs_iomap.h" | |
739bfb2a | 40 | #include "xfs_vnodeops.h" |
0b1b213f | 41 | #include "xfs_trace.h" |
3ed3a434 | 42 | #include "xfs_bmap.h" |
1da177e4 | 43 | #include <linux/mpage.h> |
10ce4444 | 44 | #include <linux/pagevec.h> |
1da177e4 LT |
45 | #include <linux/writeback.h> |
46 | ||
25e41b3d CH |
47 | |
48 | /* | |
49 | * Prime number of hash buckets since address is used as the key. | |
50 | */ | |
51 | #define NVSYNC 37 | |
52 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
53 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
54 | ||
55 | void __init | |
56 | xfs_ioend_init(void) | |
57 | { | |
58 | int i; | |
59 | ||
60 | for (i = 0; i < NVSYNC; i++) | |
61 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
62 | } | |
63 | ||
64 | void | |
65 | xfs_ioend_wait( | |
66 | xfs_inode_t *ip) | |
67 | { | |
68 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
69 | ||
70 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
71 | } | |
72 | ||
73 | STATIC void | |
74 | xfs_ioend_wake( | |
75 | xfs_inode_t *ip) | |
76 | { | |
77 | if (atomic_dec_and_test(&ip->i_iocount)) | |
78 | wake_up(to_ioend_wq(ip)); | |
79 | } | |
80 | ||
0b1b213f | 81 | void |
f51623b2 NS |
82 | xfs_count_page_state( |
83 | struct page *page, | |
84 | int *delalloc, | |
85 | int *unmapped, | |
86 | int *unwritten) | |
87 | { | |
88 | struct buffer_head *bh, *head; | |
89 | ||
90 | *delalloc = *unmapped = *unwritten = 0; | |
91 | ||
92 | bh = head = page_buffers(page); | |
93 | do { | |
94 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
95 | (*unmapped) = 1; | |
f51623b2 NS |
96 | else if (buffer_unwritten(bh)) |
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( | |
105 | struct xfs_inode *ip) | |
106 | { | |
107 | struct xfs_mount *mp = ip->i_mount; | |
108 | ||
71ddabb9 | 109 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
110 | return mp->m_rtdev_targp->bt_bdev; |
111 | else | |
112 | return mp->m_ddev_targp->bt_bdev; | |
113 | } | |
114 | ||
f6d6d4fc CH |
115 | /* |
116 | * We're now finished for good with this ioend structure. | |
117 | * Update the page state via the associated buffer_heads, | |
118 | * release holds on the inode and bio, and finally free | |
119 | * up memory. Do not use the ioend after this. | |
120 | */ | |
0829c360 CH |
121 | STATIC void |
122 | xfs_destroy_ioend( | |
123 | xfs_ioend_t *ioend) | |
124 | { | |
f6d6d4fc | 125 | struct buffer_head *bh, *next; |
583fa586 | 126 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
127 | |
128 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
129 | next = bh->b_private; | |
7d04a335 | 130 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 131 | } |
583fa586 CH |
132 | |
133 | /* | |
134 | * Volume managers supporting multiple paths can send back ENODEV | |
135 | * when the final path disappears. In this case continuing to fill | |
136 | * the page cache with dirty data which cannot be written out is | |
137 | * evil, so prevent that. | |
138 | */ | |
139 | if (unlikely(ioend->io_error == -ENODEV)) { | |
140 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
141 | __FILE__, __LINE__); | |
b677c210 | 142 | } |
583fa586 | 143 | |
25e41b3d | 144 | xfs_ioend_wake(ip); |
0829c360 CH |
145 | mempool_free(ioend, xfs_ioend_pool); |
146 | } | |
147 | ||
932640e8 DC |
148 | /* |
149 | * If the end of the current ioend is beyond the current EOF, | |
150 | * return the new EOF value, otherwise zero. | |
151 | */ | |
152 | STATIC xfs_fsize_t | |
153 | xfs_ioend_new_eof( | |
154 | xfs_ioend_t *ioend) | |
155 | { | |
156 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
157 | xfs_fsize_t isize; | |
158 | xfs_fsize_t bsize; | |
159 | ||
160 | bsize = ioend->io_offset + ioend->io_size; | |
161 | isize = MAX(ip->i_size, ip->i_new_size); | |
162 | isize = MIN(isize, bsize); | |
163 | return isize > ip->i_d.di_size ? isize : 0; | |
164 | } | |
165 | ||
ba87ea69 | 166 | /* |
77d7a0c2 DC |
167 | * Update on-disk file size now that data has been written to disk. The |
168 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
169 | * will be the intended file size until i_size is updated. If this write does | |
170 | * not extend all the way to the valid file size then restrict this update to | |
171 | * the end of the write. | |
172 | * | |
173 | * This function does not block as blocking on the inode lock in IO completion | |
174 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
175 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 176 | */ |
77d7a0c2 | 177 | STATIC int |
ba87ea69 LM |
178 | xfs_setfilesize( |
179 | xfs_ioend_t *ioend) | |
180 | { | |
b677c210 | 181 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 182 | xfs_fsize_t isize; |
ba87ea69 | 183 | |
ba87ea69 LM |
184 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
185 | ASSERT(ioend->io_type != IOMAP_READ); | |
186 | ||
187 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
188 | return 0; |
189 | ||
190 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
191 | return EAGAIN; | |
ba87ea69 | 192 | |
932640e8 DC |
193 | isize = xfs_ioend_new_eof(ioend); |
194 | if (isize) { | |
ba87ea69 | 195 | ip->i_d.di_size = isize; |
66d834ea | 196 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
197 | } |
198 | ||
199 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
200 | return 0; |
201 | } | |
202 | ||
203 | /* | |
204 | * Schedule IO completion handling on a xfsdatad if this was | |
205 | * the final hold on this ioend. If we are asked to wait, | |
206 | * flush the workqueue. | |
207 | */ | |
208 | STATIC void | |
209 | xfs_finish_ioend( | |
210 | xfs_ioend_t *ioend, | |
211 | int wait) | |
212 | { | |
213 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
214 | struct workqueue_struct *wq; | |
215 | ||
216 | wq = (ioend->io_type == IOMAP_UNWRITTEN) ? | |
217 | xfsconvertd_workqueue : xfsdatad_workqueue; | |
218 | queue_work(wq, &ioend->io_work); | |
219 | if (wait) | |
220 | flush_workqueue(wq); | |
221 | } | |
ba87ea69 LM |
222 | } |
223 | ||
0829c360 | 224 | /* |
5ec4fabb | 225 | * IO write completion. |
f6d6d4fc CH |
226 | */ |
227 | STATIC void | |
5ec4fabb | 228 | xfs_end_io( |
77d7a0c2 | 229 | struct work_struct *work) |
0829c360 | 230 | { |
77d7a0c2 DC |
231 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
232 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 233 | int error = 0; |
ba87ea69 | 234 | |
5ec4fabb CH |
235 | /* |
236 | * For unwritten extents we need to issue transactions to convert a | |
237 | * range to normal written extens after the data I/O has finished. | |
238 | */ | |
239 | if (ioend->io_type == IOMAP_UNWRITTEN && | |
240 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { | |
5ec4fabb CH |
241 | |
242 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
243 | ioend->io_size); | |
244 | if (error) | |
245 | ioend->io_error = error; | |
246 | } | |
ba87ea69 | 247 | |
5ec4fabb CH |
248 | /* |
249 | * We might have to update the on-disk file size after extending | |
250 | * writes. | |
251 | */ | |
77d7a0c2 DC |
252 | if (ioend->io_type != IOMAP_READ) { |
253 | error = xfs_setfilesize(ioend); | |
254 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 255 | } |
77d7a0c2 DC |
256 | |
257 | /* | |
258 | * If we didn't complete processing of the ioend, requeue it to the | |
259 | * tail of the workqueue for another attempt later. Otherwise destroy | |
260 | * it. | |
261 | */ | |
262 | if (error == EAGAIN) { | |
263 | atomic_inc(&ioend->io_remaining); | |
264 | xfs_finish_ioend(ioend, 0); | |
265 | /* ensure we don't spin on blocked ioends */ | |
266 | delay(1); | |
267 | } else | |
268 | xfs_destroy_ioend(ioend); | |
c626d174 DC |
269 | } |
270 | ||
0829c360 CH |
271 | /* |
272 | * Allocate and initialise an IO completion structure. | |
273 | * We need to track unwritten extent write completion here initially. | |
274 | * We'll need to extend this for updating the ondisk inode size later | |
275 | * (vs. incore size). | |
276 | */ | |
277 | STATIC xfs_ioend_t * | |
278 | xfs_alloc_ioend( | |
f6d6d4fc CH |
279 | struct inode *inode, |
280 | unsigned int type) | |
0829c360 CH |
281 | { |
282 | xfs_ioend_t *ioend; | |
283 | ||
284 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
285 | ||
286 | /* | |
287 | * Set the count to 1 initially, which will prevent an I/O | |
288 | * completion callback from happening before we have started | |
289 | * all the I/O from calling the completion routine too early. | |
290 | */ | |
291 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 292 | ioend->io_error = 0; |
f6d6d4fc CH |
293 | ioend->io_list = NULL; |
294 | ioend->io_type = type; | |
b677c210 | 295 | ioend->io_inode = inode; |
c1a073bd | 296 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 297 | ioend->io_buffer_tail = NULL; |
b677c210 | 298 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
299 | ioend->io_offset = 0; |
300 | ioend->io_size = 0; | |
301 | ||
5ec4fabb | 302 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
303 | return ioend; |
304 | } | |
305 | ||
1da177e4 LT |
306 | STATIC int |
307 | xfs_map_blocks( | |
308 | struct inode *inode, | |
309 | loff_t offset, | |
310 | ssize_t count, | |
311 | xfs_iomap_t *mapp, | |
312 | int flags) | |
313 | { | |
6bd16ff2 CH |
314 | int nmaps = 1; |
315 | ||
316 | return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps); | |
1da177e4 LT |
317 | } |
318 | ||
b8f82a4a | 319 | STATIC int |
1defeac9 | 320 | xfs_iomap_valid( |
1da177e4 | 321 | xfs_iomap_t *iomapp, |
1defeac9 | 322 | loff_t offset) |
1da177e4 | 323 | { |
1defeac9 CH |
324 | return offset >= iomapp->iomap_offset && |
325 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
326 | } |
327 | ||
f6d6d4fc CH |
328 | /* |
329 | * BIO completion handler for buffered IO. | |
330 | */ | |
782e3b3b | 331 | STATIC void |
f6d6d4fc CH |
332 | xfs_end_bio( |
333 | struct bio *bio, | |
f6d6d4fc CH |
334 | int error) |
335 | { | |
336 | xfs_ioend_t *ioend = bio->bi_private; | |
337 | ||
f6d6d4fc | 338 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 339 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
340 | |
341 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
342 | bio->bi_private = NULL; |
343 | bio->bi_end_io = NULL; | |
f6d6d4fc | 344 | bio_put(bio); |
7d04a335 | 345 | |
e927af90 | 346 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
347 | } |
348 | ||
349 | STATIC void | |
350 | xfs_submit_ioend_bio( | |
06342cf8 CH |
351 | struct writeback_control *wbc, |
352 | xfs_ioend_t *ioend, | |
353 | struct bio *bio) | |
f6d6d4fc CH |
354 | { |
355 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
356 | bio->bi_private = ioend; |
357 | bio->bi_end_io = xfs_end_bio; | |
358 | ||
932640e8 DC |
359 | /* |
360 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
361 | * but don't update the inode size until I/O completion. | |
362 | */ | |
363 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 364 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 365 | |
06342cf8 CH |
366 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
367 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
368 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
369 | bio_put(bio); | |
370 | } | |
371 | ||
372 | STATIC struct bio * | |
373 | xfs_alloc_ioend_bio( | |
374 | struct buffer_head *bh) | |
375 | { | |
376 | struct bio *bio; | |
377 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
378 | ||
379 | do { | |
380 | bio = bio_alloc(GFP_NOIO, nvecs); | |
381 | nvecs >>= 1; | |
382 | } while (!bio); | |
383 | ||
384 | ASSERT(bio->bi_private == NULL); | |
385 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
386 | bio->bi_bdev = bh->b_bdev; | |
387 | bio_get(bio); | |
388 | return bio; | |
389 | } | |
390 | ||
391 | STATIC void | |
392 | xfs_start_buffer_writeback( | |
393 | struct buffer_head *bh) | |
394 | { | |
395 | ASSERT(buffer_mapped(bh)); | |
396 | ASSERT(buffer_locked(bh)); | |
397 | ASSERT(!buffer_delay(bh)); | |
398 | ASSERT(!buffer_unwritten(bh)); | |
399 | ||
400 | mark_buffer_async_write(bh); | |
401 | set_buffer_uptodate(bh); | |
402 | clear_buffer_dirty(bh); | |
403 | } | |
404 | ||
405 | STATIC void | |
406 | xfs_start_page_writeback( | |
407 | struct page *page, | |
f6d6d4fc CH |
408 | int clear_dirty, |
409 | int buffers) | |
410 | { | |
411 | ASSERT(PageLocked(page)); | |
412 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 413 | if (clear_dirty) |
92132021 DC |
414 | clear_page_dirty_for_io(page); |
415 | set_page_writeback(page); | |
f6d6d4fc | 416 | unlock_page(page); |
1f7decf6 FW |
417 | /* If no buffers on the page are to be written, finish it here */ |
418 | if (!buffers) | |
f6d6d4fc | 419 | end_page_writeback(page); |
f6d6d4fc CH |
420 | } |
421 | ||
422 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
423 | { | |
424 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
425 | } | |
426 | ||
427 | /* | |
d88992f6 DC |
428 | * Submit all of the bios for all of the ioends we have saved up, covering the |
429 | * initial writepage page and also any probed pages. | |
430 | * | |
431 | * Because we may have multiple ioends spanning a page, we need to start | |
432 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
433 | * buffers as we got, then we can end up with a page that only has buffers | |
434 | * marked async write and I/O complete on can occur before we mark the other | |
435 | * buffers async write. | |
436 | * | |
437 | * The end result of this is that we trip a bug in end_page_writeback() because | |
438 | * we call it twice for the one page as the code in end_buffer_async_write() | |
439 | * assumes that all buffers on the page are started at the same time. | |
440 | * | |
441 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 442 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
443 | */ |
444 | STATIC void | |
445 | xfs_submit_ioend( | |
06342cf8 | 446 | struct writeback_control *wbc, |
f6d6d4fc CH |
447 | xfs_ioend_t *ioend) |
448 | { | |
d88992f6 | 449 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
450 | xfs_ioend_t *next; |
451 | struct buffer_head *bh; | |
452 | struct bio *bio; | |
453 | sector_t lastblock = 0; | |
454 | ||
d88992f6 DC |
455 | /* Pass 1 - start writeback */ |
456 | do { | |
457 | next = ioend->io_list; | |
458 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
459 | xfs_start_buffer_writeback(bh); | |
460 | } | |
461 | } while ((ioend = next) != NULL); | |
462 | ||
463 | /* Pass 2 - submit I/O */ | |
464 | ioend = head; | |
f6d6d4fc CH |
465 | do { |
466 | next = ioend->io_list; | |
467 | bio = NULL; | |
468 | ||
469 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
470 | |
471 | if (!bio) { | |
472 | retry: | |
473 | bio = xfs_alloc_ioend_bio(bh); | |
474 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 475 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
476 | goto retry; |
477 | } | |
478 | ||
479 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 480 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
481 | goto retry; |
482 | } | |
483 | ||
484 | lastblock = bh->b_blocknr; | |
485 | } | |
486 | if (bio) | |
06342cf8 | 487 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 488 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
489 | } while ((ioend = next) != NULL); |
490 | } | |
491 | ||
492 | /* | |
493 | * Cancel submission of all buffer_heads so far in this endio. | |
494 | * Toss the endio too. Only ever called for the initial page | |
495 | * in a writepage request, so only ever one page. | |
496 | */ | |
497 | STATIC void | |
498 | xfs_cancel_ioend( | |
499 | xfs_ioend_t *ioend) | |
500 | { | |
501 | xfs_ioend_t *next; | |
502 | struct buffer_head *bh, *next_bh; | |
503 | ||
504 | do { | |
505 | next = ioend->io_list; | |
506 | bh = ioend->io_buffer_head; | |
507 | do { | |
508 | next_bh = bh->b_private; | |
509 | clear_buffer_async_write(bh); | |
510 | unlock_buffer(bh); | |
511 | } while ((bh = next_bh) != NULL); | |
512 | ||
25e41b3d | 513 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
514 | mempool_free(ioend, xfs_ioend_pool); |
515 | } while ((ioend = next) != NULL); | |
516 | } | |
517 | ||
518 | /* | |
519 | * Test to see if we've been building up a completion structure for | |
520 | * earlier buffers -- if so, we try to append to this ioend if we | |
521 | * can, otherwise we finish off any current ioend and start another. | |
522 | * Return true if we've finished the given ioend. | |
523 | */ | |
524 | STATIC void | |
525 | xfs_add_to_ioend( | |
526 | struct inode *inode, | |
527 | struct buffer_head *bh, | |
7336cea8 | 528 | xfs_off_t offset, |
f6d6d4fc CH |
529 | unsigned int type, |
530 | xfs_ioend_t **result, | |
531 | int need_ioend) | |
532 | { | |
533 | xfs_ioend_t *ioend = *result; | |
534 | ||
535 | if (!ioend || need_ioend || type != ioend->io_type) { | |
536 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 537 | |
f6d6d4fc CH |
538 | ioend = xfs_alloc_ioend(inode, type); |
539 | ioend->io_offset = offset; | |
540 | ioend->io_buffer_head = bh; | |
541 | ioend->io_buffer_tail = bh; | |
542 | if (previous) | |
543 | previous->io_list = ioend; | |
544 | *result = ioend; | |
545 | } else { | |
546 | ioend->io_buffer_tail->b_private = bh; | |
547 | ioend->io_buffer_tail = bh; | |
548 | } | |
549 | ||
550 | bh->b_private = NULL; | |
551 | ioend->io_size += bh->b_size; | |
552 | } | |
553 | ||
87cbc49c NS |
554 | STATIC void |
555 | xfs_map_buffer( | |
556 | struct buffer_head *bh, | |
557 | xfs_iomap_t *mp, | |
558 | xfs_off_t offset, | |
559 | uint block_bits) | |
560 | { | |
561 | sector_t bn; | |
562 | ||
563 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
564 | ||
565 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
566 | ((offset - mp->iomap_offset) >> block_bits); | |
567 | ||
568 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
569 | ||
570 | bh->b_blocknr = bn; | |
571 | set_buffer_mapped(bh); | |
572 | } | |
573 | ||
1da177e4 LT |
574 | STATIC void |
575 | xfs_map_at_offset( | |
1da177e4 | 576 | struct buffer_head *bh, |
1defeac9 | 577 | loff_t offset, |
1da177e4 | 578 | int block_bits, |
1defeac9 | 579 | xfs_iomap_t *iomapp) |
1da177e4 | 580 | { |
1da177e4 LT |
581 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
582 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
583 | |
584 | lock_buffer(bh); | |
87cbc49c | 585 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 586 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
587 | set_buffer_mapped(bh); |
588 | clear_buffer_delay(bh); | |
f6d6d4fc | 589 | clear_buffer_unwritten(bh); |
1da177e4 LT |
590 | } |
591 | ||
592 | /* | |
6c4fe19f | 593 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
594 | */ |
595 | STATIC unsigned int | |
6c4fe19f | 596 | xfs_probe_page( |
10ce4444 | 597 | struct page *page, |
6c4fe19f CH |
598 | unsigned int pg_offset, |
599 | int mapped) | |
1da177e4 | 600 | { |
1da177e4 LT |
601 | int ret = 0; |
602 | ||
1da177e4 | 603 | if (PageWriteback(page)) |
10ce4444 | 604 | return 0; |
1da177e4 LT |
605 | |
606 | if (page->mapping && PageDirty(page)) { | |
607 | if (page_has_buffers(page)) { | |
608 | struct buffer_head *bh, *head; | |
609 | ||
610 | bh = head = page_buffers(page); | |
611 | do { | |
6c4fe19f CH |
612 | if (!buffer_uptodate(bh)) |
613 | break; | |
614 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
615 | break; |
616 | ret += bh->b_size; | |
617 | if (ret >= pg_offset) | |
618 | break; | |
619 | } while ((bh = bh->b_this_page) != head); | |
620 | } else | |
6c4fe19f | 621 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
622 | } |
623 | ||
1da177e4 LT |
624 | return ret; |
625 | } | |
626 | ||
f6d6d4fc | 627 | STATIC size_t |
6c4fe19f | 628 | xfs_probe_cluster( |
1da177e4 LT |
629 | struct inode *inode, |
630 | struct page *startpage, | |
631 | struct buffer_head *bh, | |
6c4fe19f CH |
632 | struct buffer_head *head, |
633 | int mapped) | |
1da177e4 | 634 | { |
10ce4444 | 635 | struct pagevec pvec; |
1da177e4 | 636 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
637 | size_t total = 0; |
638 | int done = 0, i; | |
1da177e4 LT |
639 | |
640 | /* First sum forwards in this page */ | |
641 | do { | |
2353e8e9 | 642 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 643 | return total; |
1da177e4 LT |
644 | total += bh->b_size; |
645 | } while ((bh = bh->b_this_page) != head); | |
646 | ||
10ce4444 CH |
647 | /* if we reached the end of the page, sum forwards in following pages */ |
648 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
649 | tindex = startpage->index + 1; | |
650 | ||
651 | /* Prune this back to avoid pathological behavior */ | |
652 | tloff = min(tlast, startpage->index + 64); | |
653 | ||
654 | pagevec_init(&pvec, 0); | |
655 | while (!done && tindex <= tloff) { | |
656 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
657 | ||
658 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
659 | break; | |
660 | ||
661 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
662 | struct page *page = pvec.pages[i]; | |
265c1fac | 663 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
664 | |
665 | if (tindex == tlast) { | |
666 | pg_offset = | |
667 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
668 | if (!pg_offset) { |
669 | done = 1; | |
10ce4444 | 670 | break; |
1defeac9 | 671 | } |
10ce4444 CH |
672 | } else |
673 | pg_offset = PAGE_CACHE_SIZE; | |
674 | ||
529ae9aa | 675 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 676 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
677 | unlock_page(page); |
678 | } | |
679 | ||
265c1fac | 680 | if (!pg_len) { |
10ce4444 CH |
681 | done = 1; |
682 | break; | |
683 | } | |
684 | ||
265c1fac | 685 | total += pg_len; |
1defeac9 | 686 | tindex++; |
1da177e4 | 687 | } |
10ce4444 CH |
688 | |
689 | pagevec_release(&pvec); | |
690 | cond_resched(); | |
1da177e4 | 691 | } |
10ce4444 | 692 | |
1da177e4 LT |
693 | return total; |
694 | } | |
695 | ||
696 | /* | |
10ce4444 CH |
697 | * Test if a given page is suitable for writing as part of an unwritten |
698 | * or delayed allocate extent. | |
1da177e4 | 699 | */ |
10ce4444 CH |
700 | STATIC int |
701 | xfs_is_delayed_page( | |
702 | struct page *page, | |
f6d6d4fc | 703 | unsigned int type) |
1da177e4 | 704 | { |
1da177e4 | 705 | if (PageWriteback(page)) |
10ce4444 | 706 | return 0; |
1da177e4 LT |
707 | |
708 | if (page->mapping && page_has_buffers(page)) { | |
709 | struct buffer_head *bh, *head; | |
710 | int acceptable = 0; | |
711 | ||
712 | bh = head = page_buffers(page); | |
713 | do { | |
f6d6d4fc CH |
714 | if (buffer_unwritten(bh)) |
715 | acceptable = (type == IOMAP_UNWRITTEN); | |
716 | else if (buffer_delay(bh)) | |
717 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 718 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 719 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 720 | else |
1da177e4 | 721 | break; |
1da177e4 LT |
722 | } while ((bh = bh->b_this_page) != head); |
723 | ||
724 | if (acceptable) | |
10ce4444 | 725 | return 1; |
1da177e4 LT |
726 | } |
727 | ||
10ce4444 | 728 | return 0; |
1da177e4 LT |
729 | } |
730 | ||
1da177e4 LT |
731 | /* |
732 | * Allocate & map buffers for page given the extent map. Write it out. | |
733 | * except for the original page of a writepage, this is called on | |
734 | * delalloc/unwritten pages only, for the original page it is possible | |
735 | * that the page has no mapping at all. | |
736 | */ | |
f6d6d4fc | 737 | STATIC int |
1da177e4 LT |
738 | xfs_convert_page( |
739 | struct inode *inode, | |
740 | struct page *page, | |
10ce4444 | 741 | loff_t tindex, |
1defeac9 | 742 | xfs_iomap_t *mp, |
f6d6d4fc | 743 | xfs_ioend_t **ioendp, |
1da177e4 | 744 | struct writeback_control *wbc, |
1da177e4 LT |
745 | int startio, |
746 | int all_bh) | |
747 | { | |
f6d6d4fc | 748 | struct buffer_head *bh, *head; |
9260dc6b CH |
749 | xfs_off_t end_offset; |
750 | unsigned long p_offset; | |
f6d6d4fc | 751 | unsigned int type; |
1da177e4 | 752 | int bbits = inode->i_blkbits; |
24e17b5f | 753 | int len, page_dirty; |
f6d6d4fc | 754 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 755 | xfs_off_t offset = page_offset(page); |
1da177e4 | 756 | |
10ce4444 CH |
757 | if (page->index != tindex) |
758 | goto fail; | |
529ae9aa | 759 | if (!trylock_page(page)) |
10ce4444 CH |
760 | goto fail; |
761 | if (PageWriteback(page)) | |
762 | goto fail_unlock_page; | |
763 | if (page->mapping != inode->i_mapping) | |
764 | goto fail_unlock_page; | |
765 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
766 | goto fail_unlock_page; | |
767 | ||
24e17b5f NS |
768 | /* |
769 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 770 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
771 | * |
772 | * Derivation: | |
773 | * | |
774 | * End offset is the highest offset that this page should represent. | |
775 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
776 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
777 | * hence give us the correct page_dirty count. On any other page, | |
778 | * it will be zero and in that case we need page_dirty to be the | |
779 | * count of buffers on the page. | |
24e17b5f | 780 | */ |
9260dc6b CH |
781 | end_offset = min_t(unsigned long long, |
782 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
783 | i_size_read(inode)); | |
784 | ||
24e17b5f | 785 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
786 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
787 | PAGE_CACHE_SIZE); | |
788 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
789 | page_dirty = p_offset / len; | |
24e17b5f | 790 | |
1da177e4 LT |
791 | bh = head = page_buffers(page); |
792 | do { | |
9260dc6b | 793 | if (offset >= end_offset) |
1da177e4 | 794 | break; |
f6d6d4fc CH |
795 | if (!buffer_uptodate(bh)) |
796 | uptodate = 0; | |
797 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
798 | done = 1; | |
1da177e4 | 799 | continue; |
f6d6d4fc CH |
800 | } |
801 | ||
9260dc6b CH |
802 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
803 | if (buffer_unwritten(bh)) | |
804 | type = IOMAP_UNWRITTEN; | |
805 | else | |
806 | type = IOMAP_DELAY; | |
807 | ||
808 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 809 | done = 1; |
9260dc6b CH |
810 | continue; |
811 | } | |
812 | ||
813 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
814 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
815 | ||
816 | xfs_map_at_offset(bh, offset, bbits, mp); | |
817 | if (startio) { | |
7336cea8 | 818 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
819 | type, ioendp, done); |
820 | } else { | |
821 | set_buffer_dirty(bh); | |
822 | unlock_buffer(bh); | |
823 | mark_buffer_dirty(bh); | |
824 | } | |
825 | page_dirty--; | |
826 | count++; | |
827 | } else { | |
df3c7244 | 828 | type = IOMAP_NEW; |
9260dc6b | 829 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 830 | lock_buffer(bh); |
7336cea8 | 831 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
832 | type, ioendp, done); |
833 | count++; | |
24e17b5f | 834 | page_dirty--; |
9260dc6b CH |
835 | } else { |
836 | done = 1; | |
1da177e4 | 837 | } |
1da177e4 | 838 | } |
7336cea8 | 839 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 840 | |
f6d6d4fc CH |
841 | if (uptodate && bh == head) |
842 | SetPageUptodate(page); | |
843 | ||
844 | if (startio) { | |
f5e596bb | 845 | if (count) { |
9fddaca2 | 846 | wbc->nr_to_write--; |
0d99519e | 847 | if (wbc->nr_to_write <= 0) |
f5e596bb | 848 | done = 1; |
f5e596bb | 849 | } |
b41759cf | 850 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 851 | } |
f6d6d4fc CH |
852 | |
853 | return done; | |
10ce4444 CH |
854 | fail_unlock_page: |
855 | unlock_page(page); | |
856 | fail: | |
857 | return 1; | |
1da177e4 LT |
858 | } |
859 | ||
860 | /* | |
861 | * Convert & write out a cluster of pages in the same extent as defined | |
862 | * by mp and following the start page. | |
863 | */ | |
864 | STATIC void | |
865 | xfs_cluster_write( | |
866 | struct inode *inode, | |
867 | pgoff_t tindex, | |
868 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 869 | xfs_ioend_t **ioendp, |
1da177e4 LT |
870 | struct writeback_control *wbc, |
871 | int startio, | |
872 | int all_bh, | |
873 | pgoff_t tlast) | |
874 | { | |
10ce4444 CH |
875 | struct pagevec pvec; |
876 | int done = 0, i; | |
1da177e4 | 877 | |
10ce4444 CH |
878 | pagevec_init(&pvec, 0); |
879 | while (!done && tindex <= tlast) { | |
880 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
881 | ||
882 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 883 | break; |
10ce4444 CH |
884 | |
885 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
886 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
887 | iomapp, ioendp, wbc, startio, all_bh); | |
888 | if (done) | |
889 | break; | |
890 | } | |
891 | ||
892 | pagevec_release(&pvec); | |
893 | cond_resched(); | |
1da177e4 LT |
894 | } |
895 | } | |
896 | ||
3ed3a434 DC |
897 | STATIC void |
898 | xfs_vm_invalidatepage( | |
899 | struct page *page, | |
900 | unsigned long offset) | |
901 | { | |
902 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
903 | block_invalidatepage(page, offset); | |
904 | } | |
905 | ||
906 | /* | |
907 | * If the page has delalloc buffers on it, we need to punch them out before we | |
908 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
909 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
910 | * is done on that same region - the delalloc extent is returned when none is | |
911 | * supposed to be there. | |
912 | * | |
913 | * We prevent this by truncating away the delalloc regions on the page before | |
914 | * invalidating it. Because they are delalloc, we can do this without needing a | |
915 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
916 | * truncation without a transaction as there is no space left for block | |
917 | * reservation (typically why we see a ENOSPC in writeback). | |
918 | * | |
919 | * This is not a performance critical path, so for now just do the punching a | |
920 | * buffer head at a time. | |
921 | */ | |
922 | STATIC void | |
923 | xfs_aops_discard_page( | |
924 | struct page *page) | |
925 | { | |
926 | struct inode *inode = page->mapping->host; | |
927 | struct xfs_inode *ip = XFS_I(inode); | |
928 | struct buffer_head *bh, *head; | |
929 | loff_t offset = page_offset(page); | |
930 | ssize_t len = 1 << inode->i_blkbits; | |
931 | ||
932 | if (!xfs_is_delayed_page(page, IOMAP_DELAY)) | |
933 | goto out_invalidate; | |
934 | ||
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 { | |
942 | int done; | |
943 | xfs_fileoff_t offset_fsb; | |
944 | xfs_bmbt_irec_t imap; | |
945 | int nimaps = 1; | |
946 | int error; | |
947 | xfs_fsblock_t firstblock; | |
948 | xfs_bmap_free_t flist; | |
949 | ||
950 | if (!buffer_delay(bh)) | |
951 | goto next_buffer; | |
952 | ||
953 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
954 | ||
955 | /* | |
956 | * Map the range first and check that it is a delalloc extent | |
957 | * before trying to unmap the range. Otherwise we will be | |
958 | * trying to remove a real extent (which requires a | |
959 | * transaction) or a hole, which is probably a bad idea... | |
960 | */ | |
961 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
962 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
963 | &nimaps, NULL, NULL); | |
964 | ||
965 | if (error) { | |
966 | /* something screwed, just bail */ | |
967 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
968 | "page discard failed delalloc mapping lookup."); | |
969 | break; | |
970 | } | |
971 | if (!nimaps) { | |
972 | /* nothing there */ | |
973 | goto next_buffer; | |
974 | } | |
975 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
976 | /* been converted, ignore */ | |
977 | goto next_buffer; | |
978 | } | |
979 | WARN_ON(imap.br_blockcount == 0); | |
980 | ||
981 | /* | |
982 | * Note: while we initialise the firstblock/flist pair, they | |
983 | * should never be used because blocks should never be | |
984 | * allocated or freed for a delalloc extent and hence we need | |
985 | * don't cancel or finish them after the xfs_bunmapi() call. | |
986 | */ | |
987 | xfs_bmap_init(&flist, &firstblock); | |
988 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
989 | &flist, NULL, &done); | |
990 | ||
991 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
992 | if (error) { | |
993 | /* something screwed, just bail */ | |
994 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
995 | "page discard unable to remove delalloc mapping."); | |
996 | break; | |
997 | } | |
998 | next_buffer: | |
999 | offset += len; | |
1000 | ||
1001 | } while ((bh = bh->b_this_page) != head); | |
1002 | ||
1003 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1004 | out_invalidate: | |
1005 | xfs_vm_invalidatepage(page, 0); | |
1006 | return; | |
1007 | } | |
1008 | ||
1da177e4 LT |
1009 | /* |
1010 | * Calling this without startio set means we are being asked to make a dirty | |
1011 | * page ready for freeing it's buffers. When called with startio set then | |
1012 | * we are coming from writepage. | |
1013 | * | |
1014 | * When called with startio set it is important that we write the WHOLE | |
1015 | * page if possible. | |
1016 | * The bh->b_state's cannot know if any of the blocks or which block for | |
1017 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 1018 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
1019 | * may clear the page dirty flag prior to calling write page, under the |
1020 | * assumption the entire page will be written out; by not writing out the | |
1021 | * whole page the page can be reused before all valid dirty data is | |
1022 | * written out. Note: in the case of a page that has been dirty'd by | |
1023 | * mapwrite and but partially setup by block_prepare_write the | |
1024 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
1025 | * valid state, thus the whole page must be written out thing. | |
1026 | */ | |
1027 | ||
1028 | STATIC int | |
1029 | xfs_page_state_convert( | |
1030 | struct inode *inode, | |
1031 | struct page *page, | |
1032 | struct writeback_control *wbc, | |
1033 | int startio, | |
1034 | int unmapped) /* also implies page uptodate */ | |
1035 | { | |
f6d6d4fc | 1036 | struct buffer_head *bh, *head; |
1defeac9 | 1037 | xfs_iomap_t iomap; |
f6d6d4fc | 1038 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
1039 | loff_t offset; |
1040 | unsigned long p_offset = 0; | |
f6d6d4fc | 1041 | unsigned int type; |
1da177e4 LT |
1042 | __uint64_t end_offset; |
1043 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
1044 | ssize_t size, len; |
1045 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
1046 | int page_dirty, count = 0; |
1047 | int trylock = 0; | |
6c4fe19f | 1048 | int all_bh = unmapped; |
1da177e4 | 1049 | |
8272145c NS |
1050 | if (startio) { |
1051 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
1052 | trylock |= BMAPI_TRYLOCK; | |
1053 | } | |
3ba0815a | 1054 | |
1da177e4 LT |
1055 | /* Is this page beyond the end of the file? */ |
1056 | offset = i_size_read(inode); | |
1057 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1058 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1059 | if (page->index >= end_index) { | |
1060 | if ((page->index >= end_index + 1) || | |
1061 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
1062 | if (startio) |
1063 | unlock_page(page); | |
1064 | return 0; | |
1da177e4 LT |
1065 | } |
1066 | } | |
1067 | ||
1da177e4 | 1068 | /* |
24e17b5f | 1069 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 1070 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
1071 | * |
1072 | * Derivation: | |
1073 | * | |
1074 | * End offset is the highest offset that this page should represent. | |
1075 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
1076 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
1077 | * hence give us the correct page_dirty count. On any other page, | |
1078 | * it will be zero and in that case we need page_dirty to be the | |
1079 | * count of buffers on the page. | |
1080 | */ | |
1081 | end_offset = min_t(unsigned long long, | |
1082 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 1083 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
1084 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
1085 | PAGE_CACHE_SIZE); | |
1086 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
1087 | page_dirty = p_offset / len; |
1088 | ||
24e17b5f | 1089 | bh = head = page_buffers(page); |
f6d6d4fc | 1090 | offset = page_offset(page); |
df3c7244 DC |
1091 | flags = BMAPI_READ; |
1092 | type = IOMAP_NEW; | |
f6d6d4fc | 1093 | |
f6d6d4fc | 1094 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
1095 | |
1096 | do { | |
1097 | if (offset >= end_offset) | |
1098 | break; | |
1099 | if (!buffer_uptodate(bh)) | |
1100 | uptodate = 0; | |
f6d6d4fc | 1101 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
1102 | /* |
1103 | * the iomap is actually still valid, but the ioend | |
1104 | * isn't. shouldn't happen too often. | |
1105 | */ | |
1106 | iomap_valid = 0; | |
1da177e4 | 1107 | continue; |
f6d6d4fc | 1108 | } |
1da177e4 | 1109 | |
1defeac9 CH |
1110 | if (iomap_valid) |
1111 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
1112 | |
1113 | /* | |
1114 | * First case, map an unwritten extent and prepare for | |
1115 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1116 | * |
1da177e4 LT |
1117 | * Second case, allocate space for a delalloc buffer. |
1118 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1119 | * |
1120 | * Third case, an unmapped buffer was found, and we are | |
1121 | * in a path where we need to write the whole page out. | |
df3c7244 | 1122 | */ |
d5cb48aa CH |
1123 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1124 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1125 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1126 | int new_ioend = 0; |
1127 | ||
df3c7244 | 1128 | /* |
6c4fe19f CH |
1129 | * Make sure we don't use a read-only iomap |
1130 | */ | |
df3c7244 | 1131 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1132 | iomap_valid = 0; |
1133 | ||
f6d6d4fc CH |
1134 | if (buffer_unwritten(bh)) { |
1135 | type = IOMAP_UNWRITTEN; | |
8272145c | 1136 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1137 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1138 | type = IOMAP_DELAY; |
8272145c | 1139 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1140 | } else { |
6c4fe19f | 1141 | type = IOMAP_NEW; |
8272145c | 1142 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1143 | } |
1144 | ||
1defeac9 | 1145 | if (!iomap_valid) { |
effd120e DC |
1146 | /* |
1147 | * if we didn't have a valid mapping then we | |
1148 | * need to ensure that we put the new mapping | |
1149 | * in a new ioend structure. This needs to be | |
1150 | * done to ensure that the ioends correctly | |
1151 | * reflect the block mappings at io completion | |
1152 | * for unwritten extent conversion. | |
1153 | */ | |
1154 | new_ioend = 1; | |
6c4fe19f CH |
1155 | if (type == IOMAP_NEW) { |
1156 | size = xfs_probe_cluster(inode, | |
1157 | page, bh, head, 0); | |
d5cb48aa CH |
1158 | } else { |
1159 | size = len; | |
1160 | } | |
1161 | ||
1162 | err = xfs_map_blocks(inode, offset, size, | |
1163 | &iomap, flags); | |
f6d6d4fc | 1164 | if (err) |
1da177e4 | 1165 | goto error; |
1defeac9 | 1166 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1167 | } |
1defeac9 CH |
1168 | if (iomap_valid) { |
1169 | xfs_map_at_offset(bh, offset, | |
1170 | inode->i_blkbits, &iomap); | |
1da177e4 | 1171 | if (startio) { |
7336cea8 | 1172 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1173 | type, &ioend, |
effd120e | 1174 | new_ioend); |
1da177e4 LT |
1175 | } else { |
1176 | set_buffer_dirty(bh); | |
1177 | unlock_buffer(bh); | |
1178 | mark_buffer_dirty(bh); | |
1179 | } | |
1180 | page_dirty--; | |
f6d6d4fc | 1181 | count++; |
1da177e4 | 1182 | } |
d5cb48aa | 1183 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1184 | /* |
1185 | * we got here because the buffer is already mapped. | |
1186 | * That means it must already have extents allocated | |
1187 | * underneath it. Map the extent by reading it. | |
1188 | */ | |
df3c7244 | 1189 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1190 | flags = BMAPI_READ; |
1191 | size = xfs_probe_cluster(inode, page, bh, | |
1192 | head, 1); | |
1193 | err = xfs_map_blocks(inode, offset, size, | |
1194 | &iomap, flags); | |
1195 | if (err) | |
1196 | goto error; | |
1197 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1198 | } | |
d5cb48aa | 1199 | |
df3c7244 DC |
1200 | /* |
1201 | * We set the type to IOMAP_NEW in case we are doing a | |
1202 | * small write at EOF that is extending the file but | |
1203 | * without needing an allocation. We need to update the | |
1204 | * file size on I/O completion in this case so it is | |
1205 | * the same case as having just allocated a new extent | |
1206 | * that we are writing into for the first time. | |
1207 | */ | |
1208 | type = IOMAP_NEW; | |
ca5de404 | 1209 | if (trylock_buffer(bh)) { |
d5cb48aa | 1210 | ASSERT(buffer_mapped(bh)); |
6c4fe19f CH |
1211 | if (iomap_valid) |
1212 | all_bh = 1; | |
7336cea8 | 1213 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1214 | &ioend, !iomap_valid); |
1215 | page_dirty--; | |
1216 | count++; | |
f6d6d4fc | 1217 | } else { |
1defeac9 | 1218 | iomap_valid = 0; |
1da177e4 | 1219 | } |
d5cb48aa CH |
1220 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1221 | (unmapped || startio)) { | |
1222 | iomap_valid = 0; | |
1da177e4 | 1223 | } |
f6d6d4fc CH |
1224 | |
1225 | if (!iohead) | |
1226 | iohead = ioend; | |
1227 | ||
1228 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1229 | |
1230 | if (uptodate && bh == head) | |
1231 | SetPageUptodate(page); | |
1232 | ||
f6d6d4fc | 1233 | if (startio) |
b41759cf | 1234 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1235 | |
1defeac9 CH |
1236 | if (ioend && iomap_valid) { |
1237 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1238 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1239 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1240 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1241 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1242 | } |
1243 | ||
f6d6d4fc | 1244 | if (iohead) |
06342cf8 | 1245 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1246 | |
1da177e4 LT |
1247 | return page_dirty; |
1248 | ||
1249 | error: | |
f6d6d4fc CH |
1250 | if (iohead) |
1251 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1252 | |
1253 | /* | |
1254 | * If it's delalloc and we have nowhere to put it, | |
1255 | * throw it away, unless the lower layers told | |
1256 | * us to try again. | |
1257 | */ | |
1258 | if (err != -EAGAIN) { | |
f6d6d4fc | 1259 | if (!unmapped) |
3ed3a434 | 1260 | xfs_aops_discard_page(page); |
1da177e4 LT |
1261 | ClearPageUptodate(page); |
1262 | } | |
1263 | return err; | |
1264 | } | |
1265 | ||
f51623b2 NS |
1266 | /* |
1267 | * writepage: Called from one of two places: | |
1268 | * | |
1269 | * 1. we are flushing a delalloc buffer head. | |
1270 | * | |
1271 | * 2. we are writing out a dirty page. Typically the page dirty | |
1272 | * state is cleared before we get here. In this case is it | |
1273 | * conceivable we have no buffer heads. | |
1274 | * | |
1275 | * For delalloc space on the page we need to allocate space and | |
1276 | * flush it. For unmapped buffer heads on the page we should | |
1277 | * allocate space if the page is uptodate. For any other dirty | |
1278 | * buffer heads on the page we should flush them. | |
1279 | * | |
1280 | * If we detect that a transaction would be required to flush | |
1281 | * the page, we have to check the process flags first, if we | |
1282 | * are already in a transaction or disk I/O during allocations | |
1283 | * is off, we need to fail the writepage and redirty the page. | |
1284 | */ | |
1285 | ||
1286 | STATIC int | |
e4c573bb | 1287 | xfs_vm_writepage( |
f51623b2 NS |
1288 | struct page *page, |
1289 | struct writeback_control *wbc) | |
1290 | { | |
1291 | int error; | |
1292 | int need_trans; | |
1293 | int delalloc, unmapped, unwritten; | |
1294 | struct inode *inode = page->mapping->host; | |
1295 | ||
0b1b213f | 1296 | trace_xfs_writepage(inode, page, 0); |
f51623b2 NS |
1297 | |
1298 | /* | |
1299 | * We need a transaction if: | |
1300 | * 1. There are delalloc buffers on the page | |
1301 | * 2. The page is uptodate and we have unmapped buffers | |
1302 | * 3. The page is uptodate and we have no buffers | |
1303 | * 4. There are unwritten buffers on the page | |
1304 | */ | |
1305 | ||
1306 | if (!page_has_buffers(page)) { | |
1307 | unmapped = 1; | |
1308 | need_trans = 1; | |
1309 | } else { | |
1310 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1311 | if (!PageUptodate(page)) | |
1312 | unmapped = 0; | |
1313 | need_trans = delalloc + unmapped + unwritten; | |
1314 | } | |
1315 | ||
1316 | /* | |
1317 | * If we need a transaction and the process flags say | |
1318 | * we are already in a transaction, or no IO is allowed | |
1319 | * then mark the page dirty again and leave the page | |
1320 | * as is. | |
1321 | */ | |
59c1b082 | 1322 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1323 | goto out_fail; |
1324 | ||
1325 | /* | |
1326 | * Delay hooking up buffer heads until we have | |
1327 | * made our go/no-go decision. | |
1328 | */ | |
1329 | if (!page_has_buffers(page)) | |
1330 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1331 | ||
c8a4051c ES |
1332 | |
1333 | /* | |
1334 | * VM calculation for nr_to_write seems off. Bump it way | |
1335 | * up, this gets simple streaming writes zippy again. | |
1336 | * To be reviewed again after Jens' writeback changes. | |
1337 | */ | |
1338 | wbc->nr_to_write *= 4; | |
1339 | ||
f51623b2 NS |
1340 | /* |
1341 | * Convert delayed allocate, unwritten or unmapped space | |
1342 | * to real space and flush out to disk. | |
1343 | */ | |
1344 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1345 | if (error == -EAGAIN) | |
1346 | goto out_fail; | |
1347 | if (unlikely(error < 0)) | |
1348 | goto out_unlock; | |
1349 | ||
1350 | return 0; | |
1351 | ||
1352 | out_fail: | |
1353 | redirty_page_for_writepage(wbc, page); | |
1354 | unlock_page(page); | |
1355 | return 0; | |
1356 | out_unlock: | |
1357 | unlock_page(page); | |
1358 | return error; | |
1359 | } | |
1360 | ||
7d4fb40a NS |
1361 | STATIC int |
1362 | xfs_vm_writepages( | |
1363 | struct address_space *mapping, | |
1364 | struct writeback_control *wbc) | |
1365 | { | |
b3aea4ed | 1366 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1367 | return generic_writepages(mapping, wbc); |
1368 | } | |
1369 | ||
f51623b2 NS |
1370 | /* |
1371 | * Called to move a page into cleanable state - and from there | |
1372 | * to be released. Possibly the page is already clean. We always | |
1373 | * have buffer heads in this call. | |
1374 | * | |
1375 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1376 | * | |
1377 | * Possible scenarios are: | |
1378 | * | |
1379 | * 1. We are being called to release a page which has been written | |
1380 | * to via regular I/O. buffer heads will be dirty and possibly | |
1381 | * delalloc. If no delalloc buffer heads in this case then we | |
1382 | * can just return zero. | |
1383 | * | |
1384 | * 2. We are called to release a page which has been written via | |
1385 | * mmap, all we need to do is ensure there is no delalloc | |
1386 | * state in the buffer heads, if not we can let the caller | |
1387 | * free them and we should come back later via writepage. | |
1388 | */ | |
1389 | STATIC int | |
238f4c54 | 1390 | xfs_vm_releasepage( |
f51623b2 NS |
1391 | struct page *page, |
1392 | gfp_t gfp_mask) | |
1393 | { | |
1394 | struct inode *inode = page->mapping->host; | |
1395 | int dirty, delalloc, unmapped, unwritten; | |
1396 | struct writeback_control wbc = { | |
1397 | .sync_mode = WB_SYNC_ALL, | |
1398 | .nr_to_write = 1, | |
1399 | }; | |
1400 | ||
0b1b213f | 1401 | trace_xfs_releasepage(inode, page, 0); |
f51623b2 | 1402 | |
238f4c54 NS |
1403 | if (!page_has_buffers(page)) |
1404 | return 0; | |
1405 | ||
f51623b2 NS |
1406 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1407 | if (!delalloc && !unwritten) | |
1408 | goto free_buffers; | |
1409 | ||
1410 | if (!(gfp_mask & __GFP_FS)) | |
1411 | return 0; | |
1412 | ||
1413 | /* If we are already inside a transaction or the thread cannot | |
1414 | * do I/O, we cannot release this page. | |
1415 | */ | |
59c1b082 | 1416 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1417 | return 0; |
1418 | ||
1419 | /* | |
1420 | * Convert delalloc space to real space, do not flush the | |
1421 | * data out to disk, that will be done by the caller. | |
1422 | * Never need to allocate space here - we will always | |
1423 | * come back to writepage in that case. | |
1424 | */ | |
1425 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1426 | if (dirty == 0 && !unwritten) | |
1427 | goto free_buffers; | |
1428 | return 0; | |
1429 | ||
1430 | free_buffers: | |
1431 | return try_to_free_buffers(page); | |
1432 | } | |
1433 | ||
1da177e4 | 1434 | STATIC int |
c2536668 | 1435 | __xfs_get_blocks( |
1da177e4 LT |
1436 | struct inode *inode, |
1437 | sector_t iblock, | |
1da177e4 LT |
1438 | struct buffer_head *bh_result, |
1439 | int create, | |
1440 | int direct, | |
1441 | bmapi_flags_t flags) | |
1442 | { | |
1da177e4 | 1443 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1444 | xfs_off_t offset; |
1445 | ssize_t size; | |
c2536668 | 1446 | int niomap = 1; |
1da177e4 | 1447 | int error; |
1da177e4 | 1448 | |
fdc7ed75 | 1449 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1450 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1451 | size = bh_result->b_size; | |
364f358a LM |
1452 | |
1453 | if (!create && direct && offset >= i_size_read(inode)) | |
1454 | return 0; | |
1455 | ||
541d7d3c | 1456 | error = xfs_iomap(XFS_I(inode), offset, size, |
67fcaa73 | 1457 | create ? flags : BMAPI_READ, &iomap, &niomap); |
1da177e4 LT |
1458 | if (error) |
1459 | return -error; | |
c2536668 | 1460 | if (niomap == 0) |
1da177e4 LT |
1461 | return 0; |
1462 | ||
1463 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1464 | /* |
1465 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1466 | * the read case (treat as if we're reading into a hole). |
1467 | */ | |
1468 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1469 | xfs_map_buffer(bh_result, &iomap, offset, |
1470 | inode->i_blkbits); | |
1da177e4 LT |
1471 | } |
1472 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1473 | if (direct) | |
1474 | bh_result->b_private = inode; | |
1475 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1476 | } |
1477 | } | |
1478 | ||
c2536668 NS |
1479 | /* |
1480 | * If this is a realtime file, data may be on a different device. | |
1481 | * to that pointed to from the buffer_head b_bdev currently. | |
1482 | */ | |
ce8e922c | 1483 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1484 | |
c2536668 | 1485 | /* |
549054af DC |
1486 | * If we previously allocated a block out beyond eof and we are now |
1487 | * coming back to use it then we will need to flag it as new even if it | |
1488 | * has a disk address. | |
1489 | * | |
1490 | * With sub-block writes into unwritten extents we also need to mark | |
1491 | * the buffer as new so that the unwritten parts of the buffer gets | |
1492 | * correctly zeroed. | |
1da177e4 LT |
1493 | */ |
1494 | if (create && | |
1495 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1496 | (offset >= i_size_read(inode)) || |
1497 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1498 | set_buffer_new(bh_result); |
1da177e4 LT |
1499 | |
1500 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1501 | BUG_ON(direct); | |
1502 | if (create) { | |
1503 | set_buffer_uptodate(bh_result); | |
1504 | set_buffer_mapped(bh_result); | |
1505 | set_buffer_delay(bh_result); | |
1506 | } | |
1507 | } | |
1508 | ||
c2536668 | 1509 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1510 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1511 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1512 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1513 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1514 | } |
1515 | ||
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | int | |
c2536668 | 1520 | xfs_get_blocks( |
1da177e4 LT |
1521 | struct inode *inode, |
1522 | sector_t iblock, | |
1523 | struct buffer_head *bh_result, | |
1524 | int create) | |
1525 | { | |
c2536668 | 1526 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1527 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1528 | } |
1529 | ||
1530 | STATIC int | |
e4c573bb | 1531 | xfs_get_blocks_direct( |
1da177e4 LT |
1532 | struct inode *inode, |
1533 | sector_t iblock, | |
1da177e4 LT |
1534 | struct buffer_head *bh_result, |
1535 | int create) | |
1536 | { | |
c2536668 | 1537 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1538 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1539 | } |
1540 | ||
f0973863 | 1541 | STATIC void |
e4c573bb | 1542 | xfs_end_io_direct( |
f0973863 CH |
1543 | struct kiocb *iocb, |
1544 | loff_t offset, | |
1545 | ssize_t size, | |
1546 | void *private) | |
1547 | { | |
1548 | xfs_ioend_t *ioend = iocb->private; | |
1549 | ||
1550 | /* | |
1551 | * Non-NULL private data means we need to issue a transaction to | |
1552 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1553 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1554 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1555 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1556 | * it anyway to keep the code uniform and simpler. |
1557 | * | |
e927af90 DC |
1558 | * Well, if only it were that simple. Because synchronous direct I/O |
1559 | * requires extent conversion to occur *before* we return to userspace, | |
1560 | * we have to wait for extent conversion to complete. Look at the | |
1561 | * iocb that has been passed to us to determine if this is AIO or | |
1562 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1563 | * workqueue and wait for it to complete. | |
1564 | * | |
f0973863 CH |
1565 | * The core direct I/O code might be changed to always call the |
1566 | * completion handler in the future, in which case all this can | |
1567 | * go away. | |
1568 | */ | |
ba87ea69 LM |
1569 | ioend->io_offset = offset; |
1570 | ioend->io_size = size; | |
1571 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1572 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1573 | } else if (private && size > 0) { |
e927af90 | 1574 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1575 | } else { |
ba87ea69 LM |
1576 | /* |
1577 | * A direct I/O write ioend starts it's life in unwritten | |
1578 | * state in case they map an unwritten extent. This write | |
1579 | * didn't map an unwritten extent so switch it's completion | |
1580 | * handler. | |
1581 | */ | |
5ec4fabb | 1582 | ioend->io_type = IOMAP_NEW; |
e927af90 | 1583 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1584 | } |
1585 | ||
1586 | /* | |
c41564b5 | 1587 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1588 | * completion handler was called. Thus we need to protect |
1589 | * against double-freeing. | |
1590 | */ | |
1591 | iocb->private = NULL; | |
1592 | } | |
1593 | ||
1da177e4 | 1594 | STATIC ssize_t |
e4c573bb | 1595 | xfs_vm_direct_IO( |
1da177e4 LT |
1596 | int rw, |
1597 | struct kiocb *iocb, | |
1598 | const struct iovec *iov, | |
1599 | loff_t offset, | |
1600 | unsigned long nr_segs) | |
1601 | { | |
1602 | struct file *file = iocb->ki_filp; | |
1603 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1604 | struct block_device *bdev; |
f0973863 | 1605 | ssize_t ret; |
1da177e4 | 1606 | |
6214ed44 | 1607 | bdev = xfs_find_bdev_for_inode(XFS_I(inode)); |
1da177e4 | 1608 | |
5fe878ae CH |
1609 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
1610 | IOMAP_UNWRITTEN : IOMAP_READ); | |
1611 | ||
1612 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1613 | offset, nr_segs, | |
1614 | xfs_get_blocks_direct, | |
1615 | xfs_end_io_direct); | |
f0973863 | 1616 | |
8459d86a | 1617 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1618 | xfs_destroy_ioend(iocb->private); |
1619 | return ret; | |
1da177e4 LT |
1620 | } |
1621 | ||
f51623b2 | 1622 | STATIC int |
d79689c7 | 1623 | xfs_vm_write_begin( |
f51623b2 | 1624 | struct file *file, |
d79689c7 NP |
1625 | struct address_space *mapping, |
1626 | loff_t pos, | |
1627 | unsigned len, | |
1628 | unsigned flags, | |
1629 | struct page **pagep, | |
1630 | void **fsdata) | |
f51623b2 | 1631 | { |
d79689c7 NP |
1632 | *pagep = NULL; |
1633 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1634 | xfs_get_blocks); | |
f51623b2 | 1635 | } |
1da177e4 LT |
1636 | |
1637 | STATIC sector_t | |
e4c573bb | 1638 | xfs_vm_bmap( |
1da177e4 LT |
1639 | struct address_space *mapping, |
1640 | sector_t block) | |
1641 | { | |
1642 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1643 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1644 | |
cf441eeb | 1645 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1646 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1647 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1648 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1649 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1650 | } |
1651 | ||
1652 | STATIC int | |
e4c573bb | 1653 | xfs_vm_readpage( |
1da177e4 LT |
1654 | struct file *unused, |
1655 | struct page *page) | |
1656 | { | |
c2536668 | 1657 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1658 | } |
1659 | ||
1660 | STATIC int | |
e4c573bb | 1661 | xfs_vm_readpages( |
1da177e4 LT |
1662 | struct file *unused, |
1663 | struct address_space *mapping, | |
1664 | struct list_head *pages, | |
1665 | unsigned nr_pages) | |
1666 | { | |
c2536668 | 1667 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1668 | } |
1669 | ||
f5e54d6e | 1670 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1671 | .readpage = xfs_vm_readpage, |
1672 | .readpages = xfs_vm_readpages, | |
1673 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1674 | .writepages = xfs_vm_writepages, |
1da177e4 | 1675 | .sync_page = block_sync_page, |
238f4c54 NS |
1676 | .releasepage = xfs_vm_releasepage, |
1677 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1678 | .write_begin = xfs_vm_write_begin, |
1679 | .write_end = generic_write_end, | |
e4c573bb NS |
1680 | .bmap = xfs_vm_bmap, |
1681 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1682 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1683 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1684 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1685 | }; |