]>
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" |
5a0e3ad6 | 43 | #include <linux/gfp.h> |
1da177e4 | 44 | #include <linux/mpage.h> |
10ce4444 | 45 | #include <linux/pagevec.h> |
1da177e4 LT |
46 | #include <linux/writeback.h> |
47 | ||
25e41b3d CH |
48 | |
49 | /* | |
50 | * Prime number of hash buckets since address is used as the key. | |
51 | */ | |
52 | #define NVSYNC 37 | |
53 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
54 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
55 | ||
56 | void __init | |
57 | xfs_ioend_init(void) | |
58 | { | |
59 | int i; | |
60 | ||
61 | for (i = 0; i < NVSYNC; i++) | |
62 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
63 | } | |
64 | ||
65 | void | |
66 | xfs_ioend_wait( | |
67 | xfs_inode_t *ip) | |
68 | { | |
69 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
70 | ||
71 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
72 | } | |
73 | ||
74 | STATIC void | |
75 | xfs_ioend_wake( | |
76 | xfs_inode_t *ip) | |
77 | { | |
78 | if (atomic_dec_and_test(&ip->i_iocount)) | |
79 | wake_up(to_ioend_wq(ip)); | |
80 | } | |
81 | ||
0b1b213f | 82 | void |
f51623b2 NS |
83 | xfs_count_page_state( |
84 | struct page *page, | |
85 | int *delalloc, | |
86 | int *unmapped, | |
87 | int *unwritten) | |
88 | { | |
89 | struct buffer_head *bh, *head; | |
90 | ||
91 | *delalloc = *unmapped = *unwritten = 0; | |
92 | ||
93 | bh = head = page_buffers(page); | |
94 | do { | |
95 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
96 | (*unmapped) = 1; | |
f51623b2 NS |
97 | else if (buffer_unwritten(bh)) |
98 | (*unwritten) = 1; | |
99 | else if (buffer_delay(bh)) | |
100 | (*delalloc) = 1; | |
101 | } while ((bh = bh->b_this_page) != head); | |
102 | } | |
103 | ||
6214ed44 CH |
104 | STATIC struct block_device * |
105 | xfs_find_bdev_for_inode( | |
106 | struct xfs_inode *ip) | |
107 | { | |
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 LM |
185 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
186 | ASSERT(ioend->io_type != IOMAP_READ); | |
187 | ||
188 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
189 | return 0; |
190 | ||
191 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
192 | return EAGAIN; | |
ba87ea69 | 193 | |
932640e8 DC |
194 | isize = xfs_ioend_new_eof(ioend); |
195 | if (isize) { | |
ba87ea69 | 196 | ip->i_d.di_size = isize; |
66d834ea | 197 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
198 | } |
199 | ||
200 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
201 | return 0; |
202 | } | |
203 | ||
204 | /* | |
205 | * Schedule IO completion handling on a xfsdatad if this was | |
206 | * the final hold on this ioend. If we are asked to wait, | |
207 | * flush the workqueue. | |
208 | */ | |
209 | STATIC void | |
210 | xfs_finish_ioend( | |
211 | xfs_ioend_t *ioend, | |
212 | int wait) | |
213 | { | |
214 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
215 | struct workqueue_struct *wq; | |
216 | ||
217 | wq = (ioend->io_type == IOMAP_UNWRITTEN) ? | |
218 | xfsconvertd_workqueue : xfsdatad_workqueue; | |
219 | queue_work(wq, &ioend->io_work); | |
220 | if (wait) | |
221 | flush_workqueue(wq); | |
222 | } | |
ba87ea69 LM |
223 | } |
224 | ||
0829c360 | 225 | /* |
5ec4fabb | 226 | * IO write completion. |
f6d6d4fc CH |
227 | */ |
228 | STATIC void | |
5ec4fabb | 229 | xfs_end_io( |
77d7a0c2 | 230 | struct work_struct *work) |
0829c360 | 231 | { |
77d7a0c2 DC |
232 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
233 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 234 | int error = 0; |
ba87ea69 | 235 | |
5ec4fabb CH |
236 | /* |
237 | * For unwritten extents we need to issue transactions to convert a | |
238 | * range to normal written extens after the data I/O has finished. | |
239 | */ | |
240 | if (ioend->io_type == IOMAP_UNWRITTEN && | |
241 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { | |
5ec4fabb CH |
242 | |
243 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
244 | ioend->io_size); | |
245 | if (error) | |
246 | ioend->io_error = error; | |
247 | } | |
ba87ea69 | 248 | |
5ec4fabb CH |
249 | /* |
250 | * We might have to update the on-disk file size after extending | |
251 | * writes. | |
252 | */ | |
77d7a0c2 DC |
253 | if (ioend->io_type != IOMAP_READ) { |
254 | error = xfs_setfilesize(ioend); | |
255 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 256 | } |
77d7a0c2 DC |
257 | |
258 | /* | |
259 | * If we didn't complete processing of the ioend, requeue it to the | |
260 | * tail of the workqueue for another attempt later. Otherwise destroy | |
261 | * it. | |
262 | */ | |
263 | if (error == EAGAIN) { | |
264 | atomic_inc(&ioend->io_remaining); | |
265 | xfs_finish_ioend(ioend, 0); | |
266 | /* ensure we don't spin on blocked ioends */ | |
267 | delay(1); | |
268 | } else | |
269 | xfs_destroy_ioend(ioend); | |
c626d174 DC |
270 | } |
271 | ||
0829c360 CH |
272 | /* |
273 | * Allocate and initialise an IO completion structure. | |
274 | * We need to track unwritten extent write completion here initially. | |
275 | * We'll need to extend this for updating the ondisk inode size later | |
276 | * (vs. incore size). | |
277 | */ | |
278 | STATIC xfs_ioend_t * | |
279 | xfs_alloc_ioend( | |
f6d6d4fc CH |
280 | struct inode *inode, |
281 | unsigned int type) | |
0829c360 CH |
282 | { |
283 | xfs_ioend_t *ioend; | |
284 | ||
285 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
286 | ||
287 | /* | |
288 | * Set the count to 1 initially, which will prevent an I/O | |
289 | * completion callback from happening before we have started | |
290 | * all the I/O from calling the completion routine too early. | |
291 | */ | |
292 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 293 | ioend->io_error = 0; |
f6d6d4fc CH |
294 | ioend->io_list = NULL; |
295 | ioend->io_type = type; | |
b677c210 | 296 | ioend->io_inode = inode; |
c1a073bd | 297 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 298 | ioend->io_buffer_tail = NULL; |
b677c210 | 299 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
300 | ioend->io_offset = 0; |
301 | ioend->io_size = 0; | |
302 | ||
5ec4fabb | 303 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
304 | return ioend; |
305 | } | |
306 | ||
1da177e4 LT |
307 | STATIC int |
308 | xfs_map_blocks( | |
309 | struct inode *inode, | |
310 | loff_t offset, | |
311 | ssize_t count, | |
312 | xfs_iomap_t *mapp, | |
313 | int flags) | |
314 | { | |
6bd16ff2 CH |
315 | int nmaps = 1; |
316 | ||
317 | return -xfs_iomap(XFS_I(inode), offset, count, flags, mapp, &nmaps); | |
1da177e4 LT |
318 | } |
319 | ||
b8f82a4a | 320 | STATIC int |
1defeac9 | 321 | xfs_iomap_valid( |
1da177e4 | 322 | xfs_iomap_t *iomapp, |
1defeac9 | 323 | loff_t offset) |
1da177e4 | 324 | { |
1defeac9 CH |
325 | return offset >= iomapp->iomap_offset && |
326 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
327 | } |
328 | ||
f6d6d4fc CH |
329 | /* |
330 | * BIO completion handler for buffered IO. | |
331 | */ | |
782e3b3b | 332 | STATIC void |
f6d6d4fc CH |
333 | xfs_end_bio( |
334 | struct bio *bio, | |
f6d6d4fc CH |
335 | int error) |
336 | { | |
337 | xfs_ioend_t *ioend = bio->bi_private; | |
338 | ||
f6d6d4fc | 339 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 340 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
341 | |
342 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
343 | bio->bi_private = NULL; |
344 | bio->bi_end_io = NULL; | |
f6d6d4fc | 345 | bio_put(bio); |
7d04a335 | 346 | |
e927af90 | 347 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
348 | } |
349 | ||
350 | STATIC void | |
351 | xfs_submit_ioend_bio( | |
06342cf8 CH |
352 | struct writeback_control *wbc, |
353 | xfs_ioend_t *ioend, | |
354 | struct bio *bio) | |
f6d6d4fc CH |
355 | { |
356 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
357 | bio->bi_private = ioend; |
358 | bio->bi_end_io = xfs_end_bio; | |
359 | ||
932640e8 DC |
360 | /* |
361 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
362 | * but don't update the inode size until I/O completion. | |
363 | */ | |
364 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 365 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 366 | |
06342cf8 CH |
367 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
368 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
369 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
370 | bio_put(bio); | |
371 | } | |
372 | ||
373 | STATIC struct bio * | |
374 | xfs_alloc_ioend_bio( | |
375 | struct buffer_head *bh) | |
376 | { | |
377 | struct bio *bio; | |
378 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
379 | ||
380 | do { | |
381 | bio = bio_alloc(GFP_NOIO, nvecs); | |
382 | nvecs >>= 1; | |
383 | } while (!bio); | |
384 | ||
385 | ASSERT(bio->bi_private == NULL); | |
386 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
387 | bio->bi_bdev = bh->b_bdev; | |
388 | bio_get(bio); | |
389 | return bio; | |
390 | } | |
391 | ||
392 | STATIC void | |
393 | xfs_start_buffer_writeback( | |
394 | struct buffer_head *bh) | |
395 | { | |
396 | ASSERT(buffer_mapped(bh)); | |
397 | ASSERT(buffer_locked(bh)); | |
398 | ASSERT(!buffer_delay(bh)); | |
399 | ASSERT(!buffer_unwritten(bh)); | |
400 | ||
401 | mark_buffer_async_write(bh); | |
402 | set_buffer_uptodate(bh); | |
403 | clear_buffer_dirty(bh); | |
404 | } | |
405 | ||
406 | STATIC void | |
407 | xfs_start_page_writeback( | |
408 | struct page *page, | |
f6d6d4fc CH |
409 | int clear_dirty, |
410 | int buffers) | |
411 | { | |
412 | ASSERT(PageLocked(page)); | |
413 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 414 | if (clear_dirty) |
92132021 DC |
415 | clear_page_dirty_for_io(page); |
416 | set_page_writeback(page); | |
f6d6d4fc | 417 | unlock_page(page); |
1f7decf6 FW |
418 | /* If no buffers on the page are to be written, finish it here */ |
419 | if (!buffers) | |
f6d6d4fc | 420 | end_page_writeback(page); |
f6d6d4fc CH |
421 | } |
422 | ||
423 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
424 | { | |
425 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
426 | } | |
427 | ||
428 | /* | |
d88992f6 DC |
429 | * Submit all of the bios for all of the ioends we have saved up, covering the |
430 | * initial writepage page and also any probed pages. | |
431 | * | |
432 | * Because we may have multiple ioends spanning a page, we need to start | |
433 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
434 | * buffers as we got, then we can end up with a page that only has buffers | |
435 | * marked async write and I/O complete on can occur before we mark the other | |
436 | * buffers async write. | |
437 | * | |
438 | * The end result of this is that we trip a bug in end_page_writeback() because | |
439 | * we call it twice for the one page as the code in end_buffer_async_write() | |
440 | * assumes that all buffers on the page are started at the same time. | |
441 | * | |
442 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 443 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
444 | */ |
445 | STATIC void | |
446 | xfs_submit_ioend( | |
06342cf8 | 447 | struct writeback_control *wbc, |
f6d6d4fc CH |
448 | xfs_ioend_t *ioend) |
449 | { | |
d88992f6 | 450 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
451 | xfs_ioend_t *next; |
452 | struct buffer_head *bh; | |
453 | struct bio *bio; | |
454 | sector_t lastblock = 0; | |
455 | ||
d88992f6 DC |
456 | /* Pass 1 - start writeback */ |
457 | do { | |
458 | next = ioend->io_list; | |
459 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
460 | xfs_start_buffer_writeback(bh); | |
461 | } | |
462 | } while ((ioend = next) != NULL); | |
463 | ||
464 | /* Pass 2 - submit I/O */ | |
465 | ioend = head; | |
f6d6d4fc CH |
466 | do { |
467 | next = ioend->io_list; | |
468 | bio = NULL; | |
469 | ||
470 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
471 | |
472 | if (!bio) { | |
473 | retry: | |
474 | bio = xfs_alloc_ioend_bio(bh); | |
475 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 476 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
477 | goto retry; |
478 | } | |
479 | ||
480 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 481 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
482 | goto retry; |
483 | } | |
484 | ||
485 | lastblock = bh->b_blocknr; | |
486 | } | |
487 | if (bio) | |
06342cf8 | 488 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 489 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
490 | } while ((ioend = next) != NULL); |
491 | } | |
492 | ||
493 | /* | |
494 | * Cancel submission of all buffer_heads so far in this endio. | |
495 | * Toss the endio too. Only ever called for the initial page | |
496 | * in a writepage request, so only ever one page. | |
497 | */ | |
498 | STATIC void | |
499 | xfs_cancel_ioend( | |
500 | xfs_ioend_t *ioend) | |
501 | { | |
502 | xfs_ioend_t *next; | |
503 | struct buffer_head *bh, *next_bh; | |
504 | ||
505 | do { | |
506 | next = ioend->io_list; | |
507 | bh = ioend->io_buffer_head; | |
508 | do { | |
509 | next_bh = bh->b_private; | |
510 | clear_buffer_async_write(bh); | |
511 | unlock_buffer(bh); | |
512 | } while ((bh = next_bh) != NULL); | |
513 | ||
25e41b3d | 514 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
515 | mempool_free(ioend, xfs_ioend_pool); |
516 | } while ((ioend = next) != NULL); | |
517 | } | |
518 | ||
519 | /* | |
520 | * Test to see if we've been building up a completion structure for | |
521 | * earlier buffers -- if so, we try to append to this ioend if we | |
522 | * can, otherwise we finish off any current ioend and start another. | |
523 | * Return true if we've finished the given ioend. | |
524 | */ | |
525 | STATIC void | |
526 | xfs_add_to_ioend( | |
527 | struct inode *inode, | |
528 | struct buffer_head *bh, | |
7336cea8 | 529 | xfs_off_t offset, |
f6d6d4fc CH |
530 | unsigned int type, |
531 | xfs_ioend_t **result, | |
532 | int need_ioend) | |
533 | { | |
534 | xfs_ioend_t *ioend = *result; | |
535 | ||
536 | if (!ioend || need_ioend || type != ioend->io_type) { | |
537 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 538 | |
f6d6d4fc CH |
539 | ioend = xfs_alloc_ioend(inode, type); |
540 | ioend->io_offset = offset; | |
541 | ioend->io_buffer_head = bh; | |
542 | ioend->io_buffer_tail = bh; | |
543 | if (previous) | |
544 | previous->io_list = ioend; | |
545 | *result = ioend; | |
546 | } else { | |
547 | ioend->io_buffer_tail->b_private = bh; | |
548 | ioend->io_buffer_tail = bh; | |
549 | } | |
550 | ||
551 | bh->b_private = NULL; | |
552 | ioend->io_size += bh->b_size; | |
553 | } | |
554 | ||
87cbc49c NS |
555 | STATIC void |
556 | xfs_map_buffer( | |
557 | struct buffer_head *bh, | |
558 | xfs_iomap_t *mp, | |
559 | xfs_off_t offset, | |
560 | uint block_bits) | |
561 | { | |
562 | sector_t bn; | |
563 | ||
564 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
565 | ||
566 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
567 | ((offset - mp->iomap_offset) >> block_bits); | |
568 | ||
569 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
570 | ||
571 | bh->b_blocknr = bn; | |
572 | set_buffer_mapped(bh); | |
573 | } | |
574 | ||
1da177e4 LT |
575 | STATIC void |
576 | xfs_map_at_offset( | |
1da177e4 | 577 | struct buffer_head *bh, |
1defeac9 | 578 | loff_t offset, |
1da177e4 | 579 | int block_bits, |
1defeac9 | 580 | xfs_iomap_t *iomapp) |
1da177e4 | 581 | { |
1da177e4 LT |
582 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
583 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
584 | |
585 | lock_buffer(bh); | |
87cbc49c | 586 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 587 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
588 | set_buffer_mapped(bh); |
589 | clear_buffer_delay(bh); | |
f6d6d4fc | 590 | clear_buffer_unwritten(bh); |
1da177e4 LT |
591 | } |
592 | ||
593 | /* | |
6c4fe19f | 594 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
595 | */ |
596 | STATIC unsigned int | |
6c4fe19f | 597 | xfs_probe_page( |
10ce4444 | 598 | struct page *page, |
6c4fe19f CH |
599 | unsigned int pg_offset, |
600 | int mapped) | |
1da177e4 | 601 | { |
1da177e4 LT |
602 | int ret = 0; |
603 | ||
1da177e4 | 604 | if (PageWriteback(page)) |
10ce4444 | 605 | return 0; |
1da177e4 LT |
606 | |
607 | if (page->mapping && PageDirty(page)) { | |
608 | if (page_has_buffers(page)) { | |
609 | struct buffer_head *bh, *head; | |
610 | ||
611 | bh = head = page_buffers(page); | |
612 | do { | |
6c4fe19f CH |
613 | if (!buffer_uptodate(bh)) |
614 | break; | |
615 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
616 | break; |
617 | ret += bh->b_size; | |
618 | if (ret >= pg_offset) | |
619 | break; | |
620 | } while ((bh = bh->b_this_page) != head); | |
621 | } else | |
6c4fe19f | 622 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
623 | } |
624 | ||
1da177e4 LT |
625 | return ret; |
626 | } | |
627 | ||
f6d6d4fc | 628 | STATIC size_t |
6c4fe19f | 629 | xfs_probe_cluster( |
1da177e4 LT |
630 | struct inode *inode, |
631 | struct page *startpage, | |
632 | struct buffer_head *bh, | |
6c4fe19f CH |
633 | struct buffer_head *head, |
634 | int mapped) | |
1da177e4 | 635 | { |
10ce4444 | 636 | struct pagevec pvec; |
1da177e4 | 637 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
638 | size_t total = 0; |
639 | int done = 0, i; | |
1da177e4 LT |
640 | |
641 | /* First sum forwards in this page */ | |
642 | do { | |
2353e8e9 | 643 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 644 | return total; |
1da177e4 LT |
645 | total += bh->b_size; |
646 | } while ((bh = bh->b_this_page) != head); | |
647 | ||
10ce4444 CH |
648 | /* if we reached the end of the page, sum forwards in following pages */ |
649 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
650 | tindex = startpage->index + 1; | |
651 | ||
652 | /* Prune this back to avoid pathological behavior */ | |
653 | tloff = min(tlast, startpage->index + 64); | |
654 | ||
655 | pagevec_init(&pvec, 0); | |
656 | while (!done && tindex <= tloff) { | |
657 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
658 | ||
659 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
660 | break; | |
661 | ||
662 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
663 | struct page *page = pvec.pages[i]; | |
265c1fac | 664 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
665 | |
666 | if (tindex == tlast) { | |
667 | pg_offset = | |
668 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
669 | if (!pg_offset) { |
670 | done = 1; | |
10ce4444 | 671 | break; |
1defeac9 | 672 | } |
10ce4444 CH |
673 | } else |
674 | pg_offset = PAGE_CACHE_SIZE; | |
675 | ||
529ae9aa | 676 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 677 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
678 | unlock_page(page); |
679 | } | |
680 | ||
265c1fac | 681 | if (!pg_len) { |
10ce4444 CH |
682 | done = 1; |
683 | break; | |
684 | } | |
685 | ||
265c1fac | 686 | total += pg_len; |
1defeac9 | 687 | tindex++; |
1da177e4 | 688 | } |
10ce4444 CH |
689 | |
690 | pagevec_release(&pvec); | |
691 | cond_resched(); | |
1da177e4 | 692 | } |
10ce4444 | 693 | |
1da177e4 LT |
694 | return total; |
695 | } | |
696 | ||
697 | /* | |
10ce4444 CH |
698 | * Test if a given page is suitable for writing as part of an unwritten |
699 | * or delayed allocate extent. | |
1da177e4 | 700 | */ |
10ce4444 CH |
701 | STATIC int |
702 | xfs_is_delayed_page( | |
703 | struct page *page, | |
f6d6d4fc | 704 | unsigned int type) |
1da177e4 | 705 | { |
1da177e4 | 706 | if (PageWriteback(page)) |
10ce4444 | 707 | return 0; |
1da177e4 LT |
708 | |
709 | if (page->mapping && page_has_buffers(page)) { | |
710 | struct buffer_head *bh, *head; | |
711 | int acceptable = 0; | |
712 | ||
713 | bh = head = page_buffers(page); | |
714 | do { | |
f6d6d4fc CH |
715 | if (buffer_unwritten(bh)) |
716 | acceptable = (type == IOMAP_UNWRITTEN); | |
717 | else if (buffer_delay(bh)) | |
718 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 719 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 720 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 721 | else |
1da177e4 | 722 | break; |
1da177e4 LT |
723 | } while ((bh = bh->b_this_page) != head); |
724 | ||
725 | if (acceptable) | |
10ce4444 | 726 | return 1; |
1da177e4 LT |
727 | } |
728 | ||
10ce4444 | 729 | return 0; |
1da177e4 LT |
730 | } |
731 | ||
1da177e4 LT |
732 | /* |
733 | * Allocate & map buffers for page given the extent map. Write it out. | |
734 | * except for the original page of a writepage, this is called on | |
735 | * delalloc/unwritten pages only, for the original page it is possible | |
736 | * that the page has no mapping at all. | |
737 | */ | |
f6d6d4fc | 738 | STATIC int |
1da177e4 LT |
739 | xfs_convert_page( |
740 | struct inode *inode, | |
741 | struct page *page, | |
10ce4444 | 742 | loff_t tindex, |
1defeac9 | 743 | xfs_iomap_t *mp, |
f6d6d4fc | 744 | xfs_ioend_t **ioendp, |
1da177e4 | 745 | struct writeback_control *wbc, |
1da177e4 LT |
746 | int startio, |
747 | int all_bh) | |
748 | { | |
f6d6d4fc | 749 | struct buffer_head *bh, *head; |
9260dc6b CH |
750 | xfs_off_t end_offset; |
751 | unsigned long p_offset; | |
f6d6d4fc | 752 | unsigned int type; |
1da177e4 | 753 | int bbits = inode->i_blkbits; |
24e17b5f | 754 | int len, page_dirty; |
f6d6d4fc | 755 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 756 | xfs_off_t offset = page_offset(page); |
1da177e4 | 757 | |
10ce4444 CH |
758 | if (page->index != tindex) |
759 | goto fail; | |
529ae9aa | 760 | if (!trylock_page(page)) |
10ce4444 CH |
761 | goto fail; |
762 | if (PageWriteback(page)) | |
763 | goto fail_unlock_page; | |
764 | if (page->mapping != inode->i_mapping) | |
765 | goto fail_unlock_page; | |
766 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
767 | goto fail_unlock_page; | |
768 | ||
24e17b5f NS |
769 | /* |
770 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 771 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
772 | * |
773 | * Derivation: | |
774 | * | |
775 | * End offset is the highest offset that this page should represent. | |
776 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
777 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
778 | * hence give us the correct page_dirty count. On any other page, | |
779 | * it will be zero and in that case we need page_dirty to be the | |
780 | * count of buffers on the page. | |
24e17b5f | 781 | */ |
9260dc6b CH |
782 | end_offset = min_t(unsigned long long, |
783 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
784 | i_size_read(inode)); | |
785 | ||
24e17b5f | 786 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
787 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
788 | PAGE_CACHE_SIZE); | |
789 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
790 | page_dirty = p_offset / len; | |
24e17b5f | 791 | |
1da177e4 LT |
792 | bh = head = page_buffers(page); |
793 | do { | |
9260dc6b | 794 | if (offset >= end_offset) |
1da177e4 | 795 | break; |
f6d6d4fc CH |
796 | if (!buffer_uptodate(bh)) |
797 | uptodate = 0; | |
798 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
799 | done = 1; | |
1da177e4 | 800 | continue; |
f6d6d4fc CH |
801 | } |
802 | ||
9260dc6b CH |
803 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
804 | if (buffer_unwritten(bh)) | |
805 | type = IOMAP_UNWRITTEN; | |
806 | else | |
807 | type = IOMAP_DELAY; | |
808 | ||
809 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 810 | done = 1; |
9260dc6b CH |
811 | continue; |
812 | } | |
813 | ||
814 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
815 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
816 | ||
817 | xfs_map_at_offset(bh, offset, bbits, mp); | |
818 | if (startio) { | |
7336cea8 | 819 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
820 | type, ioendp, done); |
821 | } else { | |
822 | set_buffer_dirty(bh); | |
823 | unlock_buffer(bh); | |
824 | mark_buffer_dirty(bh); | |
825 | } | |
826 | page_dirty--; | |
827 | count++; | |
828 | } else { | |
df3c7244 | 829 | type = IOMAP_NEW; |
9260dc6b | 830 | if (buffer_mapped(bh) && all_bh && startio) { |
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 | ||
845 | if (startio) { | |
f5e596bb | 846 | if (count) { |
9fddaca2 | 847 | wbc->nr_to_write--; |
0d99519e | 848 | if (wbc->nr_to_write <= 0) |
f5e596bb | 849 | done = 1; |
f5e596bb | 850 | } |
b41759cf | 851 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 852 | } |
f6d6d4fc CH |
853 | |
854 | return done; | |
10ce4444 CH |
855 | fail_unlock_page: |
856 | unlock_page(page); | |
857 | fail: | |
858 | return 1; | |
1da177e4 LT |
859 | } |
860 | ||
861 | /* | |
862 | * Convert & write out a cluster of pages in the same extent as defined | |
863 | * by mp and following the start page. | |
864 | */ | |
865 | STATIC void | |
866 | xfs_cluster_write( | |
867 | struct inode *inode, | |
868 | pgoff_t tindex, | |
869 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 870 | xfs_ioend_t **ioendp, |
1da177e4 LT |
871 | struct writeback_control *wbc, |
872 | int startio, | |
873 | int all_bh, | |
874 | pgoff_t tlast) | |
875 | { | |
10ce4444 CH |
876 | struct pagevec pvec; |
877 | int done = 0, i; | |
1da177e4 | 878 | |
10ce4444 CH |
879 | pagevec_init(&pvec, 0); |
880 | while (!done && tindex <= tlast) { | |
881 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
882 | ||
883 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 884 | break; |
10ce4444 CH |
885 | |
886 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
887 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
888 | iomapp, ioendp, wbc, startio, all_bh); | |
889 | if (done) | |
890 | break; | |
891 | } | |
892 | ||
893 | pagevec_release(&pvec); | |
894 | cond_resched(); | |
1da177e4 LT |
895 | } |
896 | } | |
897 | ||
3ed3a434 DC |
898 | STATIC void |
899 | xfs_vm_invalidatepage( | |
900 | struct page *page, | |
901 | unsigned long offset) | |
902 | { | |
903 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
904 | block_invalidatepage(page, offset); | |
905 | } | |
906 | ||
907 | /* | |
908 | * If the page has delalloc buffers on it, we need to punch them out before we | |
909 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
910 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
911 | * is done on that same region - the delalloc extent is returned when none is | |
912 | * supposed to be there. | |
913 | * | |
914 | * We prevent this by truncating away the delalloc regions on the page before | |
915 | * invalidating it. Because they are delalloc, we can do this without needing a | |
916 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
917 | * truncation without a transaction as there is no space left for block | |
918 | * reservation (typically why we see a ENOSPC in writeback). | |
919 | * | |
920 | * This is not a performance critical path, so for now just do the punching a | |
921 | * buffer head at a time. | |
922 | */ | |
923 | STATIC void | |
924 | xfs_aops_discard_page( | |
925 | struct page *page) | |
926 | { | |
927 | struct inode *inode = page->mapping->host; | |
928 | struct xfs_inode *ip = XFS_I(inode); | |
929 | struct buffer_head *bh, *head; | |
930 | loff_t offset = page_offset(page); | |
931 | ssize_t len = 1 << inode->i_blkbits; | |
932 | ||
933 | if (!xfs_is_delayed_page(page, IOMAP_DELAY)) | |
934 | goto out_invalidate; | |
935 | ||
e8c3753c DC |
936 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
937 | goto out_invalidate; | |
938 | ||
3ed3a434 DC |
939 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
940 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
941 | page, ip->i_ino, offset); | |
942 | ||
943 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
944 | bh = head = page_buffers(page); | |
945 | do { | |
946 | int done; | |
947 | xfs_fileoff_t offset_fsb; | |
948 | xfs_bmbt_irec_t imap; | |
949 | int nimaps = 1; | |
950 | int error; | |
951 | xfs_fsblock_t firstblock; | |
952 | xfs_bmap_free_t flist; | |
953 | ||
954 | if (!buffer_delay(bh)) | |
955 | goto next_buffer; | |
956 | ||
957 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
958 | ||
959 | /* | |
960 | * Map the range first and check that it is a delalloc extent | |
961 | * before trying to unmap the range. Otherwise we will be | |
962 | * trying to remove a real extent (which requires a | |
963 | * transaction) or a hole, which is probably a bad idea... | |
964 | */ | |
965 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
966 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
967 | &nimaps, NULL, NULL); | |
968 | ||
969 | if (error) { | |
970 | /* something screwed, just bail */ | |
e8c3753c DC |
971 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
972 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
973 | "page discard failed delalloc mapping lookup."); | |
974 | } | |
3ed3a434 DC |
975 | break; |
976 | } | |
977 | if (!nimaps) { | |
978 | /* nothing there */ | |
979 | goto next_buffer; | |
980 | } | |
981 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
982 | /* been converted, ignore */ | |
983 | goto next_buffer; | |
984 | } | |
985 | WARN_ON(imap.br_blockcount == 0); | |
986 | ||
987 | /* | |
988 | * Note: while we initialise the firstblock/flist pair, they | |
989 | * should never be used because blocks should never be | |
990 | * allocated or freed for a delalloc extent and hence we need | |
991 | * don't cancel or finish them after the xfs_bunmapi() call. | |
992 | */ | |
993 | xfs_bmap_init(&flist, &firstblock); | |
994 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
995 | &flist, NULL, &done); | |
996 | ||
997 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
998 | if (error) { | |
999 | /* something screwed, just bail */ | |
e8c3753c DC |
1000 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1001 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 1002 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 1003 | } |
3ed3a434 DC |
1004 | break; |
1005 | } | |
1006 | next_buffer: | |
1007 | offset += len; | |
1008 | ||
1009 | } while ((bh = bh->b_this_page) != head); | |
1010 | ||
1011 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1012 | out_invalidate: | |
1013 | xfs_vm_invalidatepage(page, 0); | |
1014 | return; | |
1015 | } | |
1016 | ||
1da177e4 LT |
1017 | /* |
1018 | * Calling this without startio set means we are being asked to make a dirty | |
1019 | * page ready for freeing it's buffers. When called with startio set then | |
1020 | * we are coming from writepage. | |
1021 | * | |
1022 | * When called with startio set it is important that we write the WHOLE | |
1023 | * page if possible. | |
1024 | * The bh->b_state's cannot know if any of the blocks or which block for | |
1025 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 1026 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
1027 | * may clear the page dirty flag prior to calling write page, under the |
1028 | * assumption the entire page will be written out; by not writing out the | |
1029 | * whole page the page can be reused before all valid dirty data is | |
1030 | * written out. Note: in the case of a page that has been dirty'd by | |
1031 | * mapwrite and but partially setup by block_prepare_write the | |
1032 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
1033 | * valid state, thus the whole page must be written out thing. | |
1034 | */ | |
1035 | ||
1036 | STATIC int | |
1037 | xfs_page_state_convert( | |
1038 | struct inode *inode, | |
1039 | struct page *page, | |
1040 | struct writeback_control *wbc, | |
1041 | int startio, | |
1042 | int unmapped) /* also implies page uptodate */ | |
1043 | { | |
f6d6d4fc | 1044 | struct buffer_head *bh, *head; |
1defeac9 | 1045 | xfs_iomap_t iomap; |
f6d6d4fc | 1046 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
1047 | loff_t offset; |
1048 | unsigned long p_offset = 0; | |
f6d6d4fc | 1049 | unsigned int type; |
1da177e4 LT |
1050 | __uint64_t end_offset; |
1051 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
1052 | ssize_t size, len; |
1053 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
1054 | int page_dirty, count = 0; |
1055 | int trylock = 0; | |
6c4fe19f | 1056 | int all_bh = unmapped; |
1da177e4 | 1057 | |
8272145c NS |
1058 | if (startio) { |
1059 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
1060 | trylock |= BMAPI_TRYLOCK; | |
1061 | } | |
3ba0815a | 1062 | |
1da177e4 LT |
1063 | /* Is this page beyond the end of the file? */ |
1064 | offset = i_size_read(inode); | |
1065 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1066 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1067 | if (page->index >= end_index) { | |
1068 | if ((page->index >= end_index + 1) || | |
1069 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
1070 | if (startio) |
1071 | unlock_page(page); | |
1072 | return 0; | |
1da177e4 LT |
1073 | } |
1074 | } | |
1075 | ||
1da177e4 | 1076 | /* |
24e17b5f | 1077 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 1078 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
1079 | * |
1080 | * Derivation: | |
1081 | * | |
1082 | * End offset is the highest offset that this page should represent. | |
1083 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
1084 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
1085 | * hence give us the correct page_dirty count. On any other page, | |
1086 | * it will be zero and in that case we need page_dirty to be the | |
1087 | * count of buffers on the page. | |
1088 | */ | |
1089 | end_offset = min_t(unsigned long long, | |
1090 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 1091 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
1092 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
1093 | PAGE_CACHE_SIZE); | |
1094 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
1095 | page_dirty = p_offset / len; |
1096 | ||
24e17b5f | 1097 | bh = head = page_buffers(page); |
f6d6d4fc | 1098 | offset = page_offset(page); |
df3c7244 DC |
1099 | flags = BMAPI_READ; |
1100 | type = IOMAP_NEW; | |
f6d6d4fc | 1101 | |
f6d6d4fc | 1102 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
1103 | |
1104 | do { | |
1105 | if (offset >= end_offset) | |
1106 | break; | |
1107 | if (!buffer_uptodate(bh)) | |
1108 | uptodate = 0; | |
f6d6d4fc | 1109 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
1110 | /* |
1111 | * the iomap is actually still valid, but the ioend | |
1112 | * isn't. shouldn't happen too often. | |
1113 | */ | |
1114 | iomap_valid = 0; | |
1da177e4 | 1115 | continue; |
f6d6d4fc | 1116 | } |
1da177e4 | 1117 | |
1defeac9 CH |
1118 | if (iomap_valid) |
1119 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
1120 | |
1121 | /* | |
1122 | * First case, map an unwritten extent and prepare for | |
1123 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1124 | * |
1da177e4 LT |
1125 | * Second case, allocate space for a delalloc buffer. |
1126 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1127 | * |
1128 | * Third case, an unmapped buffer was found, and we are | |
1129 | * in a path where we need to write the whole page out. | |
df3c7244 | 1130 | */ |
d5cb48aa CH |
1131 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1132 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1133 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1134 | int new_ioend = 0; |
1135 | ||
df3c7244 | 1136 | /* |
6c4fe19f CH |
1137 | * Make sure we don't use a read-only iomap |
1138 | */ | |
df3c7244 | 1139 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1140 | iomap_valid = 0; |
1141 | ||
f6d6d4fc CH |
1142 | if (buffer_unwritten(bh)) { |
1143 | type = IOMAP_UNWRITTEN; | |
8272145c | 1144 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1145 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1146 | type = IOMAP_DELAY; |
8272145c | 1147 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1148 | } else { |
6c4fe19f | 1149 | type = IOMAP_NEW; |
8272145c | 1150 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1151 | } |
1152 | ||
1defeac9 | 1153 | if (!iomap_valid) { |
effd120e DC |
1154 | /* |
1155 | * if we didn't have a valid mapping then we | |
1156 | * need to ensure that we put the new mapping | |
1157 | * in a new ioend structure. This needs to be | |
1158 | * done to ensure that the ioends correctly | |
1159 | * reflect the block mappings at io completion | |
1160 | * for unwritten extent conversion. | |
1161 | */ | |
1162 | new_ioend = 1; | |
6c4fe19f CH |
1163 | if (type == IOMAP_NEW) { |
1164 | size = xfs_probe_cluster(inode, | |
1165 | page, bh, head, 0); | |
d5cb48aa CH |
1166 | } else { |
1167 | size = len; | |
1168 | } | |
1169 | ||
1170 | err = xfs_map_blocks(inode, offset, size, | |
1171 | &iomap, flags); | |
f6d6d4fc | 1172 | if (err) |
1da177e4 | 1173 | goto error; |
1defeac9 | 1174 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 1175 | } |
1defeac9 CH |
1176 | if (iomap_valid) { |
1177 | xfs_map_at_offset(bh, offset, | |
1178 | inode->i_blkbits, &iomap); | |
1da177e4 | 1179 | if (startio) { |
7336cea8 | 1180 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1181 | type, &ioend, |
effd120e | 1182 | new_ioend); |
1da177e4 LT |
1183 | } else { |
1184 | set_buffer_dirty(bh); | |
1185 | unlock_buffer(bh); | |
1186 | mark_buffer_dirty(bh); | |
1187 | } | |
1188 | page_dirty--; | |
f6d6d4fc | 1189 | count++; |
1da177e4 | 1190 | } |
d5cb48aa | 1191 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1192 | /* |
1193 | * we got here because the buffer is already mapped. | |
1194 | * That means it must already have extents allocated | |
1195 | * underneath it. Map the extent by reading it. | |
1196 | */ | |
df3c7244 | 1197 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1198 | flags = BMAPI_READ; |
1199 | size = xfs_probe_cluster(inode, page, bh, | |
1200 | head, 1); | |
1201 | err = xfs_map_blocks(inode, offset, size, | |
1202 | &iomap, flags); | |
1203 | if (err) | |
1204 | goto error; | |
1205 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1206 | } | |
d5cb48aa | 1207 | |
df3c7244 DC |
1208 | /* |
1209 | * We set the type to IOMAP_NEW in case we are doing a | |
1210 | * small write at EOF that is extending the file but | |
1211 | * without needing an allocation. We need to update the | |
1212 | * file size on I/O completion in this case so it is | |
1213 | * the same case as having just allocated a new extent | |
1214 | * that we are writing into for the first time. | |
1215 | */ | |
1216 | type = IOMAP_NEW; | |
ca5de404 | 1217 | if (trylock_buffer(bh)) { |
d5cb48aa | 1218 | ASSERT(buffer_mapped(bh)); |
6c4fe19f CH |
1219 | if (iomap_valid) |
1220 | all_bh = 1; | |
7336cea8 | 1221 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1222 | &ioend, !iomap_valid); |
1223 | page_dirty--; | |
1224 | count++; | |
f6d6d4fc | 1225 | } else { |
1defeac9 | 1226 | iomap_valid = 0; |
1da177e4 | 1227 | } |
d5cb48aa CH |
1228 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1229 | (unmapped || startio)) { | |
1230 | iomap_valid = 0; | |
1da177e4 | 1231 | } |
f6d6d4fc CH |
1232 | |
1233 | if (!iohead) | |
1234 | iohead = ioend; | |
1235 | ||
1236 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1237 | |
1238 | if (uptodate && bh == head) | |
1239 | SetPageUptodate(page); | |
1240 | ||
f6d6d4fc | 1241 | if (startio) |
b41759cf | 1242 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1243 | |
1defeac9 CH |
1244 | if (ioend && iomap_valid) { |
1245 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1246 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1247 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1248 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1249 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1250 | } |
1251 | ||
f6d6d4fc | 1252 | if (iohead) |
06342cf8 | 1253 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1254 | |
1da177e4 LT |
1255 | return page_dirty; |
1256 | ||
1257 | error: | |
f6d6d4fc CH |
1258 | if (iohead) |
1259 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1260 | |
1261 | /* | |
1262 | * If it's delalloc and we have nowhere to put it, | |
1263 | * throw it away, unless the lower layers told | |
1264 | * us to try again. | |
1265 | */ | |
1266 | if (err != -EAGAIN) { | |
f6d6d4fc | 1267 | if (!unmapped) |
3ed3a434 | 1268 | xfs_aops_discard_page(page); |
1da177e4 LT |
1269 | ClearPageUptodate(page); |
1270 | } | |
1271 | return err; | |
1272 | } | |
1273 | ||
f51623b2 NS |
1274 | /* |
1275 | * writepage: Called from one of two places: | |
1276 | * | |
1277 | * 1. we are flushing a delalloc buffer head. | |
1278 | * | |
1279 | * 2. we are writing out a dirty page. Typically the page dirty | |
1280 | * state is cleared before we get here. In this case is it | |
1281 | * conceivable we have no buffer heads. | |
1282 | * | |
1283 | * For delalloc space on the page we need to allocate space and | |
1284 | * flush it. For unmapped buffer heads on the page we should | |
1285 | * allocate space if the page is uptodate. For any other dirty | |
1286 | * buffer heads on the page we should flush them. | |
1287 | * | |
1288 | * If we detect that a transaction would be required to flush | |
1289 | * the page, we have to check the process flags first, if we | |
1290 | * are already in a transaction or disk I/O during allocations | |
1291 | * is off, we need to fail the writepage and redirty the page. | |
1292 | */ | |
1293 | ||
1294 | STATIC int | |
e4c573bb | 1295 | xfs_vm_writepage( |
f51623b2 NS |
1296 | struct page *page, |
1297 | struct writeback_control *wbc) | |
1298 | { | |
1299 | int error; | |
1300 | int need_trans; | |
1301 | int delalloc, unmapped, unwritten; | |
1302 | struct inode *inode = page->mapping->host; | |
1303 | ||
0b1b213f | 1304 | trace_xfs_writepage(inode, page, 0); |
f51623b2 NS |
1305 | |
1306 | /* | |
1307 | * We need a transaction if: | |
1308 | * 1. There are delalloc buffers on the page | |
1309 | * 2. The page is uptodate and we have unmapped buffers | |
1310 | * 3. The page is uptodate and we have no buffers | |
1311 | * 4. There are unwritten buffers on the page | |
1312 | */ | |
1313 | ||
1314 | if (!page_has_buffers(page)) { | |
1315 | unmapped = 1; | |
1316 | need_trans = 1; | |
1317 | } else { | |
1318 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1319 | if (!PageUptodate(page)) | |
1320 | unmapped = 0; | |
1321 | need_trans = delalloc + unmapped + unwritten; | |
1322 | } | |
1323 | ||
1324 | /* | |
1325 | * If we need a transaction and the process flags say | |
1326 | * we are already in a transaction, or no IO is allowed | |
1327 | * then mark the page dirty again and leave the page | |
1328 | * as is. | |
1329 | */ | |
59c1b082 | 1330 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1331 | goto out_fail; |
1332 | ||
1333 | /* | |
1334 | * Delay hooking up buffer heads until we have | |
1335 | * made our go/no-go decision. | |
1336 | */ | |
1337 | if (!page_has_buffers(page)) | |
1338 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1339 | ||
c8a4051c ES |
1340 | |
1341 | /* | |
1342 | * VM calculation for nr_to_write seems off. Bump it way | |
1343 | * up, this gets simple streaming writes zippy again. | |
1344 | * To be reviewed again after Jens' writeback changes. | |
1345 | */ | |
1346 | wbc->nr_to_write *= 4; | |
1347 | ||
f51623b2 NS |
1348 | /* |
1349 | * Convert delayed allocate, unwritten or unmapped space | |
1350 | * to real space and flush out to disk. | |
1351 | */ | |
1352 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1353 | if (error == -EAGAIN) | |
1354 | goto out_fail; | |
1355 | if (unlikely(error < 0)) | |
1356 | goto out_unlock; | |
1357 | ||
1358 | return 0; | |
1359 | ||
1360 | out_fail: | |
1361 | redirty_page_for_writepage(wbc, page); | |
1362 | unlock_page(page); | |
1363 | return 0; | |
1364 | out_unlock: | |
1365 | unlock_page(page); | |
1366 | return error; | |
1367 | } | |
1368 | ||
7d4fb40a NS |
1369 | STATIC int |
1370 | xfs_vm_writepages( | |
1371 | struct address_space *mapping, | |
1372 | struct writeback_control *wbc) | |
1373 | { | |
b3aea4ed | 1374 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1375 | return generic_writepages(mapping, wbc); |
1376 | } | |
1377 | ||
f51623b2 NS |
1378 | /* |
1379 | * Called to move a page into cleanable state - and from there | |
1380 | * to be released. Possibly the page is already clean. We always | |
1381 | * have buffer heads in this call. | |
1382 | * | |
1383 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1384 | * | |
1385 | * Possible scenarios are: | |
1386 | * | |
1387 | * 1. We are being called to release a page which has been written | |
1388 | * to via regular I/O. buffer heads will be dirty and possibly | |
1389 | * delalloc. If no delalloc buffer heads in this case then we | |
1390 | * can just return zero. | |
1391 | * | |
1392 | * 2. We are called to release a page which has been written via | |
1393 | * mmap, all we need to do is ensure there is no delalloc | |
1394 | * state in the buffer heads, if not we can let the caller | |
1395 | * free them and we should come back later via writepage. | |
1396 | */ | |
1397 | STATIC int | |
238f4c54 | 1398 | xfs_vm_releasepage( |
f51623b2 NS |
1399 | struct page *page, |
1400 | gfp_t gfp_mask) | |
1401 | { | |
1402 | struct inode *inode = page->mapping->host; | |
1403 | int dirty, delalloc, unmapped, unwritten; | |
1404 | struct writeback_control wbc = { | |
1405 | .sync_mode = WB_SYNC_ALL, | |
1406 | .nr_to_write = 1, | |
1407 | }; | |
1408 | ||
0b1b213f | 1409 | trace_xfs_releasepage(inode, page, 0); |
f51623b2 | 1410 | |
238f4c54 NS |
1411 | if (!page_has_buffers(page)) |
1412 | return 0; | |
1413 | ||
f51623b2 NS |
1414 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1415 | if (!delalloc && !unwritten) | |
1416 | goto free_buffers; | |
1417 | ||
1418 | if (!(gfp_mask & __GFP_FS)) | |
1419 | return 0; | |
1420 | ||
1421 | /* If we are already inside a transaction or the thread cannot | |
1422 | * do I/O, we cannot release this page. | |
1423 | */ | |
59c1b082 | 1424 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1425 | return 0; |
1426 | ||
1427 | /* | |
1428 | * Convert delalloc space to real space, do not flush the | |
1429 | * data out to disk, that will be done by the caller. | |
1430 | * Never need to allocate space here - we will always | |
1431 | * come back to writepage in that case. | |
1432 | */ | |
1433 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1434 | if (dirty == 0 && !unwritten) | |
1435 | goto free_buffers; | |
1436 | return 0; | |
1437 | ||
1438 | free_buffers: | |
1439 | return try_to_free_buffers(page); | |
1440 | } | |
1441 | ||
1da177e4 | 1442 | STATIC int |
c2536668 | 1443 | __xfs_get_blocks( |
1da177e4 LT |
1444 | struct inode *inode, |
1445 | sector_t iblock, | |
1da177e4 LT |
1446 | struct buffer_head *bh_result, |
1447 | int create, | |
1448 | int direct, | |
1449 | bmapi_flags_t flags) | |
1450 | { | |
1da177e4 | 1451 | xfs_iomap_t iomap; |
fdc7ed75 NS |
1452 | xfs_off_t offset; |
1453 | ssize_t size; | |
c2536668 | 1454 | int niomap = 1; |
1da177e4 | 1455 | int error; |
1da177e4 | 1456 | |
fdc7ed75 | 1457 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1458 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1459 | size = bh_result->b_size; | |
364f358a LM |
1460 | |
1461 | if (!create && direct && offset >= i_size_read(inode)) | |
1462 | return 0; | |
1463 | ||
541d7d3c | 1464 | error = xfs_iomap(XFS_I(inode), offset, size, |
67fcaa73 | 1465 | create ? flags : BMAPI_READ, &iomap, &niomap); |
1da177e4 LT |
1466 | if (error) |
1467 | return -error; | |
c2536668 | 1468 | if (niomap == 0) |
1da177e4 LT |
1469 | return 0; |
1470 | ||
1471 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1472 | /* |
1473 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1474 | * the read case (treat as if we're reading into a hole). |
1475 | */ | |
1476 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1477 | xfs_map_buffer(bh_result, &iomap, offset, |
1478 | inode->i_blkbits); | |
1da177e4 LT |
1479 | } |
1480 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1481 | if (direct) | |
1482 | bh_result->b_private = inode; | |
1483 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1484 | } |
1485 | } | |
1486 | ||
c2536668 NS |
1487 | /* |
1488 | * If this is a realtime file, data may be on a different device. | |
1489 | * to that pointed to from the buffer_head b_bdev currently. | |
1490 | */ | |
ce8e922c | 1491 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 | 1492 | |
c2536668 | 1493 | /* |
549054af DC |
1494 | * If we previously allocated a block out beyond eof and we are now |
1495 | * coming back to use it then we will need to flag it as new even if it | |
1496 | * has a disk address. | |
1497 | * | |
1498 | * With sub-block writes into unwritten extents we also need to mark | |
1499 | * the buffer as new so that the unwritten parts of the buffer gets | |
1500 | * correctly zeroed. | |
1da177e4 LT |
1501 | */ |
1502 | if (create && | |
1503 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af DC |
1504 | (offset >= i_size_read(inode)) || |
1505 | (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN)))) | |
1da177e4 | 1506 | set_buffer_new(bh_result); |
1da177e4 LT |
1507 | |
1508 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1509 | BUG_ON(direct); | |
1510 | if (create) { | |
1511 | set_buffer_uptodate(bh_result); | |
1512 | set_buffer_mapped(bh_result); | |
1513 | set_buffer_delay(bh_result); | |
1514 | } | |
1515 | } | |
1516 | ||
c2536668 | 1517 | if (direct || size > (1 << inode->i_blkbits)) { |
fdc7ed75 NS |
1518 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1519 | offset = min_t(xfs_off_t, | |
c2536668 NS |
1520 | iomap.iomap_bsize - iomap.iomap_delta, size); |
1521 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); | |
1da177e4 LT |
1522 | } |
1523 | ||
1524 | return 0; | |
1525 | } | |
1526 | ||
1527 | int | |
c2536668 | 1528 | xfs_get_blocks( |
1da177e4 LT |
1529 | struct inode *inode, |
1530 | sector_t iblock, | |
1531 | struct buffer_head *bh_result, | |
1532 | int create) | |
1533 | { | |
c2536668 | 1534 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1535 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1536 | } |
1537 | ||
1538 | STATIC int | |
e4c573bb | 1539 | xfs_get_blocks_direct( |
1da177e4 LT |
1540 | struct inode *inode, |
1541 | sector_t iblock, | |
1da177e4 LT |
1542 | struct buffer_head *bh_result, |
1543 | int create) | |
1544 | { | |
c2536668 | 1545 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1546 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1547 | } |
1548 | ||
f0973863 | 1549 | STATIC void |
e4c573bb | 1550 | xfs_end_io_direct( |
f0973863 CH |
1551 | struct kiocb *iocb, |
1552 | loff_t offset, | |
1553 | ssize_t size, | |
1554 | void *private) | |
1555 | { | |
1556 | xfs_ioend_t *ioend = iocb->private; | |
1557 | ||
1558 | /* | |
1559 | * Non-NULL private data means we need to issue a transaction to | |
1560 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1561 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1562 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1563 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1564 | * it anyway to keep the code uniform and simpler. |
1565 | * | |
e927af90 DC |
1566 | * Well, if only it were that simple. Because synchronous direct I/O |
1567 | * requires extent conversion to occur *before* we return to userspace, | |
1568 | * we have to wait for extent conversion to complete. Look at the | |
1569 | * iocb that has been passed to us to determine if this is AIO or | |
1570 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1571 | * workqueue and wait for it to complete. | |
1572 | * | |
f0973863 CH |
1573 | * The core direct I/O code might be changed to always call the |
1574 | * completion handler in the future, in which case all this can | |
1575 | * go away. | |
1576 | */ | |
ba87ea69 LM |
1577 | ioend->io_offset = offset; |
1578 | ioend->io_size = size; | |
1579 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1580 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1581 | } else if (private && size > 0) { |
e927af90 | 1582 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1583 | } else { |
ba87ea69 LM |
1584 | /* |
1585 | * A direct I/O write ioend starts it's life in unwritten | |
1586 | * state in case they map an unwritten extent. This write | |
1587 | * didn't map an unwritten extent so switch it's completion | |
1588 | * handler. | |
1589 | */ | |
5ec4fabb | 1590 | ioend->io_type = IOMAP_NEW; |
e927af90 | 1591 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1592 | } |
1593 | ||
1594 | /* | |
c41564b5 | 1595 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1596 | * completion handler was called. Thus we need to protect |
1597 | * against double-freeing. | |
1598 | */ | |
1599 | iocb->private = NULL; | |
1600 | } | |
1601 | ||
1da177e4 | 1602 | STATIC ssize_t |
e4c573bb | 1603 | xfs_vm_direct_IO( |
1da177e4 LT |
1604 | int rw, |
1605 | struct kiocb *iocb, | |
1606 | const struct iovec *iov, | |
1607 | loff_t offset, | |
1608 | unsigned long nr_segs) | |
1609 | { | |
1610 | struct file *file = iocb->ki_filp; | |
1611 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1612 | struct block_device *bdev; |
f0973863 | 1613 | ssize_t ret; |
1da177e4 | 1614 | |
6214ed44 | 1615 | bdev = xfs_find_bdev_for_inode(XFS_I(inode)); |
1da177e4 | 1616 | |
5fe878ae CH |
1617 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
1618 | IOMAP_UNWRITTEN : IOMAP_READ); | |
1619 | ||
1620 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1621 | offset, nr_segs, | |
1622 | xfs_get_blocks_direct, | |
1623 | xfs_end_io_direct); | |
f0973863 | 1624 | |
8459d86a | 1625 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1626 | xfs_destroy_ioend(iocb->private); |
1627 | return ret; | |
1da177e4 LT |
1628 | } |
1629 | ||
f51623b2 | 1630 | STATIC int |
d79689c7 | 1631 | xfs_vm_write_begin( |
f51623b2 | 1632 | struct file *file, |
d79689c7 NP |
1633 | struct address_space *mapping, |
1634 | loff_t pos, | |
1635 | unsigned len, | |
1636 | unsigned flags, | |
1637 | struct page **pagep, | |
1638 | void **fsdata) | |
f51623b2 | 1639 | { |
d79689c7 NP |
1640 | *pagep = NULL; |
1641 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1642 | xfs_get_blocks); | |
f51623b2 | 1643 | } |
1da177e4 LT |
1644 | |
1645 | STATIC sector_t | |
e4c573bb | 1646 | xfs_vm_bmap( |
1da177e4 LT |
1647 | struct address_space *mapping, |
1648 | sector_t block) | |
1649 | { | |
1650 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1651 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1652 | |
cf441eeb | 1653 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1654 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1655 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1656 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1657 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1658 | } |
1659 | ||
1660 | STATIC int | |
e4c573bb | 1661 | xfs_vm_readpage( |
1da177e4 LT |
1662 | struct file *unused, |
1663 | struct page *page) | |
1664 | { | |
c2536668 | 1665 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1666 | } |
1667 | ||
1668 | STATIC int | |
e4c573bb | 1669 | xfs_vm_readpages( |
1da177e4 LT |
1670 | struct file *unused, |
1671 | struct address_space *mapping, | |
1672 | struct list_head *pages, | |
1673 | unsigned nr_pages) | |
1674 | { | |
c2536668 | 1675 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1676 | } |
1677 | ||
f5e54d6e | 1678 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1679 | .readpage = xfs_vm_readpage, |
1680 | .readpages = xfs_vm_readpages, | |
1681 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1682 | .writepages = xfs_vm_writepages, |
1da177e4 | 1683 | .sync_page = block_sync_page, |
238f4c54 NS |
1684 | .releasepage = xfs_vm_releasepage, |
1685 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1686 | .write_begin = xfs_vm_write_begin, |
1687 | .write_end = generic_write_end, | |
e4c573bb NS |
1688 | .bmap = xfs_vm_bmap, |
1689 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1690 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1691 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1692 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1693 | }; |