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( | |
046f1685 | 106 | struct inode *inode) |
6214ed44 | 107 | { |
046f1685 | 108 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
109 | struct xfs_mount *mp = ip->i_mount; |
110 | ||
71ddabb9 | 111 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
112 | return mp->m_rtdev_targp->bt_bdev; |
113 | else | |
114 | return mp->m_ddev_targp->bt_bdev; | |
115 | } | |
116 | ||
f6d6d4fc CH |
117 | /* |
118 | * We're now finished for good with this ioend structure. | |
119 | * Update the page state via the associated buffer_heads, | |
120 | * release holds on the inode and bio, and finally free | |
121 | * up memory. Do not use the ioend after this. | |
122 | */ | |
0829c360 CH |
123 | STATIC void |
124 | xfs_destroy_ioend( | |
125 | xfs_ioend_t *ioend) | |
126 | { | |
f6d6d4fc | 127 | struct buffer_head *bh, *next; |
583fa586 | 128 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
129 | |
130 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
131 | next = bh->b_private; | |
7d04a335 | 132 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 133 | } |
583fa586 CH |
134 | |
135 | /* | |
136 | * Volume managers supporting multiple paths can send back ENODEV | |
137 | * when the final path disappears. In this case continuing to fill | |
138 | * the page cache with dirty data which cannot be written out is | |
139 | * evil, so prevent that. | |
140 | */ | |
141 | if (unlikely(ioend->io_error == -ENODEV)) { | |
142 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
143 | __FILE__, __LINE__); | |
b677c210 | 144 | } |
583fa586 | 145 | |
25e41b3d | 146 | xfs_ioend_wake(ip); |
0829c360 CH |
147 | mempool_free(ioend, xfs_ioend_pool); |
148 | } | |
149 | ||
932640e8 DC |
150 | /* |
151 | * If the end of the current ioend is beyond the current EOF, | |
152 | * return the new EOF value, otherwise zero. | |
153 | */ | |
154 | STATIC xfs_fsize_t | |
155 | xfs_ioend_new_eof( | |
156 | xfs_ioend_t *ioend) | |
157 | { | |
158 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
159 | xfs_fsize_t isize; | |
160 | xfs_fsize_t bsize; | |
161 | ||
162 | bsize = ioend->io_offset + ioend->io_size; | |
163 | isize = MAX(ip->i_size, ip->i_new_size); | |
164 | isize = MIN(isize, bsize); | |
165 | return isize > ip->i_d.di_size ? isize : 0; | |
166 | } | |
167 | ||
ba87ea69 | 168 | /* |
77d7a0c2 DC |
169 | * Update on-disk file size now that data has been written to disk. The |
170 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
171 | * will be the intended file size until i_size is updated. If this write does | |
172 | * not extend all the way to the valid file size then restrict this update to | |
173 | * the end of the write. | |
174 | * | |
175 | * This function does not block as blocking on the inode lock in IO completion | |
176 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
177 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 178 | */ |
77d7a0c2 | 179 | STATIC int |
ba87ea69 LM |
180 | xfs_setfilesize( |
181 | xfs_ioend_t *ioend) | |
182 | { | |
b677c210 | 183 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 184 | xfs_fsize_t isize; |
ba87ea69 | 185 | |
ba87ea69 LM |
186 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
187 | ASSERT(ioend->io_type != IOMAP_READ); | |
188 | ||
189 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
190 | return 0; |
191 | ||
192 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
193 | return EAGAIN; | |
ba87ea69 | 194 | |
932640e8 DC |
195 | isize = xfs_ioend_new_eof(ioend); |
196 | if (isize) { | |
ba87ea69 | 197 | ip->i_d.di_size = isize; |
66d834ea | 198 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
199 | } |
200 | ||
201 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
202 | return 0; |
203 | } | |
204 | ||
205 | /* | |
206 | * Schedule IO completion handling on a xfsdatad if this was | |
207 | * the final hold on this ioend. If we are asked to wait, | |
208 | * flush the workqueue. | |
209 | */ | |
210 | STATIC void | |
211 | xfs_finish_ioend( | |
212 | xfs_ioend_t *ioend, | |
213 | int wait) | |
214 | { | |
215 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
216 | struct workqueue_struct *wq; | |
217 | ||
218 | wq = (ioend->io_type == IOMAP_UNWRITTEN) ? | |
219 | xfsconvertd_workqueue : xfsdatad_workqueue; | |
220 | queue_work(wq, &ioend->io_work); | |
221 | if (wait) | |
222 | flush_workqueue(wq); | |
223 | } | |
ba87ea69 LM |
224 | } |
225 | ||
0829c360 | 226 | /* |
5ec4fabb | 227 | * IO write completion. |
f6d6d4fc CH |
228 | */ |
229 | STATIC void | |
5ec4fabb | 230 | xfs_end_io( |
77d7a0c2 | 231 | struct work_struct *work) |
0829c360 | 232 | { |
77d7a0c2 DC |
233 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
234 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 235 | int error = 0; |
ba87ea69 | 236 | |
5ec4fabb CH |
237 | /* |
238 | * For unwritten extents we need to issue transactions to convert a | |
239 | * range to normal written extens after the data I/O has finished. | |
240 | */ | |
241 | if (ioend->io_type == IOMAP_UNWRITTEN && | |
242 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { | |
5ec4fabb CH |
243 | |
244 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
245 | ioend->io_size); | |
246 | if (error) | |
247 | ioend->io_error = error; | |
248 | } | |
ba87ea69 | 249 | |
5ec4fabb CH |
250 | /* |
251 | * We might have to update the on-disk file size after extending | |
252 | * writes. | |
253 | */ | |
77d7a0c2 DC |
254 | if (ioend->io_type != IOMAP_READ) { |
255 | error = xfs_setfilesize(ioend); | |
256 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 257 | } |
77d7a0c2 DC |
258 | |
259 | /* | |
260 | * If we didn't complete processing of the ioend, requeue it to the | |
261 | * tail of the workqueue for another attempt later. Otherwise destroy | |
262 | * it. | |
263 | */ | |
264 | if (error == EAGAIN) { | |
265 | atomic_inc(&ioend->io_remaining); | |
266 | xfs_finish_ioend(ioend, 0); | |
267 | /* ensure we don't spin on blocked ioends */ | |
268 | delay(1); | |
269 | } else | |
270 | xfs_destroy_ioend(ioend); | |
c626d174 DC |
271 | } |
272 | ||
0829c360 CH |
273 | /* |
274 | * Allocate and initialise an IO completion structure. | |
275 | * We need to track unwritten extent write completion here initially. | |
276 | * We'll need to extend this for updating the ondisk inode size later | |
277 | * (vs. incore size). | |
278 | */ | |
279 | STATIC xfs_ioend_t * | |
280 | xfs_alloc_ioend( | |
f6d6d4fc CH |
281 | struct inode *inode, |
282 | unsigned int type) | |
0829c360 CH |
283 | { |
284 | xfs_ioend_t *ioend; | |
285 | ||
286 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
287 | ||
288 | /* | |
289 | * Set the count to 1 initially, which will prevent an I/O | |
290 | * completion callback from happening before we have started | |
291 | * all the I/O from calling the completion routine too early. | |
292 | */ | |
293 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 294 | ioend->io_error = 0; |
f6d6d4fc CH |
295 | ioend->io_list = NULL; |
296 | ioend->io_type = type; | |
b677c210 | 297 | ioend->io_inode = inode; |
c1a073bd | 298 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 299 | ioend->io_buffer_tail = NULL; |
b677c210 | 300 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
301 | ioend->io_offset = 0; |
302 | ioend->io_size = 0; | |
303 | ||
5ec4fabb | 304 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
305 | return ioend; |
306 | } | |
307 | ||
1da177e4 LT |
308 | STATIC int |
309 | xfs_map_blocks( | |
310 | struct inode *inode, | |
311 | loff_t offset, | |
312 | ssize_t count, | |
207d0416 | 313 | struct xfs_bmbt_irec *imap, |
1da177e4 LT |
314 | int flags) |
315 | { | |
6bd16ff2 | 316 | int nmaps = 1; |
207d0416 | 317 | int new = 0; |
6bd16ff2 | 318 | |
207d0416 | 319 | return -xfs_iomap(XFS_I(inode), offset, count, flags, imap, &nmaps, &new); |
1da177e4 LT |
320 | } |
321 | ||
b8f82a4a | 322 | STATIC int |
1defeac9 | 323 | xfs_iomap_valid( |
8699bb0a | 324 | struct inode *inode, |
207d0416 | 325 | struct xfs_bmbt_irec *imap, |
1defeac9 | 326 | loff_t offset) |
1da177e4 | 327 | { |
8699bb0a | 328 | struct xfs_mount *mp = XFS_I(inode)->i_mount; |
207d0416 CH |
329 | xfs_off_t iomap_offset = XFS_FSB_TO_B(mp, imap->br_startoff); |
330 | xfs_off_t iomap_bsize = XFS_FSB_TO_B(mp, imap->br_blockcount); | |
8699bb0a CH |
331 | |
332 | return offset >= iomap_offset && | |
333 | offset < iomap_offset + iomap_bsize; | |
1da177e4 LT |
334 | } |
335 | ||
f6d6d4fc CH |
336 | /* |
337 | * BIO completion handler for buffered IO. | |
338 | */ | |
782e3b3b | 339 | STATIC void |
f6d6d4fc CH |
340 | xfs_end_bio( |
341 | struct bio *bio, | |
f6d6d4fc CH |
342 | int error) |
343 | { | |
344 | xfs_ioend_t *ioend = bio->bi_private; | |
345 | ||
f6d6d4fc | 346 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 347 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
348 | |
349 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
350 | bio->bi_private = NULL; |
351 | bio->bi_end_io = NULL; | |
f6d6d4fc | 352 | bio_put(bio); |
7d04a335 | 353 | |
e927af90 | 354 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
355 | } |
356 | ||
357 | STATIC void | |
358 | xfs_submit_ioend_bio( | |
06342cf8 CH |
359 | struct writeback_control *wbc, |
360 | xfs_ioend_t *ioend, | |
361 | struct bio *bio) | |
f6d6d4fc CH |
362 | { |
363 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
364 | bio->bi_private = ioend; |
365 | bio->bi_end_io = xfs_end_bio; | |
366 | ||
932640e8 DC |
367 | /* |
368 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
369 | * but don't update the inode size until I/O completion. | |
370 | */ | |
371 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 372 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 373 | |
06342cf8 CH |
374 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
375 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
376 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
377 | bio_put(bio); | |
378 | } | |
379 | ||
380 | STATIC struct bio * | |
381 | xfs_alloc_ioend_bio( | |
382 | struct buffer_head *bh) | |
383 | { | |
384 | struct bio *bio; | |
385 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
386 | ||
387 | do { | |
388 | bio = bio_alloc(GFP_NOIO, nvecs); | |
389 | nvecs >>= 1; | |
390 | } while (!bio); | |
391 | ||
392 | ASSERT(bio->bi_private == NULL); | |
393 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
394 | bio->bi_bdev = bh->b_bdev; | |
395 | bio_get(bio); | |
396 | return bio; | |
397 | } | |
398 | ||
399 | STATIC void | |
400 | xfs_start_buffer_writeback( | |
401 | struct buffer_head *bh) | |
402 | { | |
403 | ASSERT(buffer_mapped(bh)); | |
404 | ASSERT(buffer_locked(bh)); | |
405 | ASSERT(!buffer_delay(bh)); | |
406 | ASSERT(!buffer_unwritten(bh)); | |
407 | ||
408 | mark_buffer_async_write(bh); | |
409 | set_buffer_uptodate(bh); | |
410 | clear_buffer_dirty(bh); | |
411 | } | |
412 | ||
413 | STATIC void | |
414 | xfs_start_page_writeback( | |
415 | struct page *page, | |
f6d6d4fc CH |
416 | int clear_dirty, |
417 | int buffers) | |
418 | { | |
419 | ASSERT(PageLocked(page)); | |
420 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 421 | if (clear_dirty) |
92132021 DC |
422 | clear_page_dirty_for_io(page); |
423 | set_page_writeback(page); | |
f6d6d4fc | 424 | unlock_page(page); |
1f7decf6 FW |
425 | /* If no buffers on the page are to be written, finish it here */ |
426 | if (!buffers) | |
f6d6d4fc | 427 | end_page_writeback(page); |
f6d6d4fc CH |
428 | } |
429 | ||
430 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
431 | { | |
432 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
433 | } | |
434 | ||
435 | /* | |
d88992f6 DC |
436 | * Submit all of the bios for all of the ioends we have saved up, covering the |
437 | * initial writepage page and also any probed pages. | |
438 | * | |
439 | * Because we may have multiple ioends spanning a page, we need to start | |
440 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
441 | * buffers as we got, then we can end up with a page that only has buffers | |
442 | * marked async write and I/O complete on can occur before we mark the other | |
443 | * buffers async write. | |
444 | * | |
445 | * The end result of this is that we trip a bug in end_page_writeback() because | |
446 | * we call it twice for the one page as the code in end_buffer_async_write() | |
447 | * assumes that all buffers on the page are started at the same time. | |
448 | * | |
449 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 450 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
451 | */ |
452 | STATIC void | |
453 | xfs_submit_ioend( | |
06342cf8 | 454 | struct writeback_control *wbc, |
f6d6d4fc CH |
455 | xfs_ioend_t *ioend) |
456 | { | |
d88992f6 | 457 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
458 | xfs_ioend_t *next; |
459 | struct buffer_head *bh; | |
460 | struct bio *bio; | |
461 | sector_t lastblock = 0; | |
462 | ||
d88992f6 DC |
463 | /* Pass 1 - start writeback */ |
464 | do { | |
465 | next = ioend->io_list; | |
466 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
467 | xfs_start_buffer_writeback(bh); | |
468 | } | |
469 | } while ((ioend = next) != NULL); | |
470 | ||
471 | /* Pass 2 - submit I/O */ | |
472 | ioend = head; | |
f6d6d4fc CH |
473 | do { |
474 | next = ioend->io_list; | |
475 | bio = NULL; | |
476 | ||
477 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
478 | |
479 | if (!bio) { | |
480 | retry: | |
481 | bio = xfs_alloc_ioend_bio(bh); | |
482 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 483 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
484 | goto retry; |
485 | } | |
486 | ||
487 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 488 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
489 | goto retry; |
490 | } | |
491 | ||
492 | lastblock = bh->b_blocknr; | |
493 | } | |
494 | if (bio) | |
06342cf8 | 495 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 496 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
497 | } while ((ioend = next) != NULL); |
498 | } | |
499 | ||
500 | /* | |
501 | * Cancel submission of all buffer_heads so far in this endio. | |
502 | * Toss the endio too. Only ever called for the initial page | |
503 | * in a writepage request, so only ever one page. | |
504 | */ | |
505 | STATIC void | |
506 | xfs_cancel_ioend( | |
507 | xfs_ioend_t *ioend) | |
508 | { | |
509 | xfs_ioend_t *next; | |
510 | struct buffer_head *bh, *next_bh; | |
511 | ||
512 | do { | |
513 | next = ioend->io_list; | |
514 | bh = ioend->io_buffer_head; | |
515 | do { | |
516 | next_bh = bh->b_private; | |
517 | clear_buffer_async_write(bh); | |
518 | unlock_buffer(bh); | |
519 | } while ((bh = next_bh) != NULL); | |
520 | ||
25e41b3d | 521 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
522 | mempool_free(ioend, xfs_ioend_pool); |
523 | } while ((ioend = next) != NULL); | |
524 | } | |
525 | ||
526 | /* | |
527 | * Test to see if we've been building up a completion structure for | |
528 | * earlier buffers -- if so, we try to append to this ioend if we | |
529 | * can, otherwise we finish off any current ioend and start another. | |
530 | * Return true if we've finished the given ioend. | |
531 | */ | |
532 | STATIC void | |
533 | xfs_add_to_ioend( | |
534 | struct inode *inode, | |
535 | struct buffer_head *bh, | |
7336cea8 | 536 | xfs_off_t offset, |
f6d6d4fc CH |
537 | unsigned int type, |
538 | xfs_ioend_t **result, | |
539 | int need_ioend) | |
540 | { | |
541 | xfs_ioend_t *ioend = *result; | |
542 | ||
543 | if (!ioend || need_ioend || type != ioend->io_type) { | |
544 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 545 | |
f6d6d4fc CH |
546 | ioend = xfs_alloc_ioend(inode, type); |
547 | ioend->io_offset = offset; | |
548 | ioend->io_buffer_head = bh; | |
549 | ioend->io_buffer_tail = bh; | |
550 | if (previous) | |
551 | previous->io_list = ioend; | |
552 | *result = ioend; | |
553 | } else { | |
554 | ioend->io_buffer_tail->b_private = bh; | |
555 | ioend->io_buffer_tail = bh; | |
556 | } | |
557 | ||
558 | bh->b_private = NULL; | |
559 | ioend->io_size += bh->b_size; | |
560 | } | |
561 | ||
87cbc49c NS |
562 | STATIC void |
563 | xfs_map_buffer( | |
046f1685 | 564 | struct inode *inode, |
87cbc49c | 565 | struct buffer_head *bh, |
207d0416 | 566 | struct xfs_bmbt_irec *imap, |
046f1685 | 567 | xfs_off_t offset) |
87cbc49c NS |
568 | { |
569 | sector_t bn; | |
8699bb0a | 570 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
571 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
572 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 573 | |
207d0416 CH |
574 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
575 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 576 | |
e513182d | 577 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 578 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 579 | |
046f1685 | 580 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
581 | |
582 | bh->b_blocknr = bn; | |
583 | set_buffer_mapped(bh); | |
584 | } | |
585 | ||
1da177e4 LT |
586 | STATIC void |
587 | xfs_map_at_offset( | |
046f1685 | 588 | struct inode *inode, |
1da177e4 | 589 | struct buffer_head *bh, |
207d0416 | 590 | struct xfs_bmbt_irec *imap, |
046f1685 | 591 | xfs_off_t offset) |
1da177e4 | 592 | { |
207d0416 CH |
593 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
594 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 LT |
595 | |
596 | lock_buffer(bh); | |
207d0416 | 597 | xfs_map_buffer(inode, bh, imap, offset); |
046f1685 | 598 | bh->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 LT |
599 | set_buffer_mapped(bh); |
600 | clear_buffer_delay(bh); | |
f6d6d4fc | 601 | clear_buffer_unwritten(bh); |
1da177e4 LT |
602 | } |
603 | ||
604 | /* | |
6c4fe19f | 605 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
606 | */ |
607 | STATIC unsigned int | |
6c4fe19f | 608 | xfs_probe_page( |
10ce4444 | 609 | struct page *page, |
6c4fe19f CH |
610 | unsigned int pg_offset, |
611 | int mapped) | |
1da177e4 | 612 | { |
1da177e4 LT |
613 | int ret = 0; |
614 | ||
1da177e4 | 615 | if (PageWriteback(page)) |
10ce4444 | 616 | return 0; |
1da177e4 LT |
617 | |
618 | if (page->mapping && PageDirty(page)) { | |
619 | if (page_has_buffers(page)) { | |
620 | struct buffer_head *bh, *head; | |
621 | ||
622 | bh = head = page_buffers(page); | |
623 | do { | |
6c4fe19f CH |
624 | if (!buffer_uptodate(bh)) |
625 | break; | |
626 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
627 | break; |
628 | ret += bh->b_size; | |
629 | if (ret >= pg_offset) | |
630 | break; | |
631 | } while ((bh = bh->b_this_page) != head); | |
632 | } else | |
6c4fe19f | 633 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
634 | } |
635 | ||
1da177e4 LT |
636 | return ret; |
637 | } | |
638 | ||
f6d6d4fc | 639 | STATIC size_t |
6c4fe19f | 640 | xfs_probe_cluster( |
1da177e4 LT |
641 | struct inode *inode, |
642 | struct page *startpage, | |
643 | struct buffer_head *bh, | |
6c4fe19f CH |
644 | struct buffer_head *head, |
645 | int mapped) | |
1da177e4 | 646 | { |
10ce4444 | 647 | struct pagevec pvec; |
1da177e4 | 648 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
649 | size_t total = 0; |
650 | int done = 0, i; | |
1da177e4 LT |
651 | |
652 | /* First sum forwards in this page */ | |
653 | do { | |
2353e8e9 | 654 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 655 | return total; |
1da177e4 LT |
656 | total += bh->b_size; |
657 | } while ((bh = bh->b_this_page) != head); | |
658 | ||
10ce4444 CH |
659 | /* if we reached the end of the page, sum forwards in following pages */ |
660 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
661 | tindex = startpage->index + 1; | |
662 | ||
663 | /* Prune this back to avoid pathological behavior */ | |
664 | tloff = min(tlast, startpage->index + 64); | |
665 | ||
666 | pagevec_init(&pvec, 0); | |
667 | while (!done && tindex <= tloff) { | |
668 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
669 | ||
670 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
671 | break; | |
672 | ||
673 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
674 | struct page *page = pvec.pages[i]; | |
265c1fac | 675 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
676 | |
677 | if (tindex == tlast) { | |
678 | pg_offset = | |
679 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
680 | if (!pg_offset) { |
681 | done = 1; | |
10ce4444 | 682 | break; |
1defeac9 | 683 | } |
10ce4444 CH |
684 | } else |
685 | pg_offset = PAGE_CACHE_SIZE; | |
686 | ||
529ae9aa | 687 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 688 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
689 | unlock_page(page); |
690 | } | |
691 | ||
265c1fac | 692 | if (!pg_len) { |
10ce4444 CH |
693 | done = 1; |
694 | break; | |
695 | } | |
696 | ||
265c1fac | 697 | total += pg_len; |
1defeac9 | 698 | tindex++; |
1da177e4 | 699 | } |
10ce4444 CH |
700 | |
701 | pagevec_release(&pvec); | |
702 | cond_resched(); | |
1da177e4 | 703 | } |
10ce4444 | 704 | |
1da177e4 LT |
705 | return total; |
706 | } | |
707 | ||
708 | /* | |
10ce4444 CH |
709 | * Test if a given page is suitable for writing as part of an unwritten |
710 | * or delayed allocate extent. | |
1da177e4 | 711 | */ |
10ce4444 CH |
712 | STATIC int |
713 | xfs_is_delayed_page( | |
714 | struct page *page, | |
f6d6d4fc | 715 | unsigned int type) |
1da177e4 | 716 | { |
1da177e4 | 717 | if (PageWriteback(page)) |
10ce4444 | 718 | return 0; |
1da177e4 LT |
719 | |
720 | if (page->mapping && page_has_buffers(page)) { | |
721 | struct buffer_head *bh, *head; | |
722 | int acceptable = 0; | |
723 | ||
724 | bh = head = page_buffers(page); | |
725 | do { | |
f6d6d4fc CH |
726 | if (buffer_unwritten(bh)) |
727 | acceptable = (type == IOMAP_UNWRITTEN); | |
728 | else if (buffer_delay(bh)) | |
729 | acceptable = (type == IOMAP_DELAY); | |
2ddee844 | 730 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
df3c7244 | 731 | acceptable = (type == IOMAP_NEW); |
f6d6d4fc | 732 | else |
1da177e4 | 733 | break; |
1da177e4 LT |
734 | } while ((bh = bh->b_this_page) != head); |
735 | ||
736 | if (acceptable) | |
10ce4444 | 737 | return 1; |
1da177e4 LT |
738 | } |
739 | ||
10ce4444 | 740 | return 0; |
1da177e4 LT |
741 | } |
742 | ||
1da177e4 LT |
743 | /* |
744 | * Allocate & map buffers for page given the extent map. Write it out. | |
745 | * except for the original page of a writepage, this is called on | |
746 | * delalloc/unwritten pages only, for the original page it is possible | |
747 | * that the page has no mapping at all. | |
748 | */ | |
f6d6d4fc | 749 | STATIC int |
1da177e4 LT |
750 | xfs_convert_page( |
751 | struct inode *inode, | |
752 | struct page *page, | |
10ce4444 | 753 | loff_t tindex, |
207d0416 | 754 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 755 | xfs_ioend_t **ioendp, |
1da177e4 | 756 | struct writeback_control *wbc, |
1da177e4 LT |
757 | int startio, |
758 | int all_bh) | |
759 | { | |
f6d6d4fc | 760 | struct buffer_head *bh, *head; |
9260dc6b CH |
761 | xfs_off_t end_offset; |
762 | unsigned long p_offset; | |
f6d6d4fc | 763 | unsigned int type; |
24e17b5f | 764 | int len, page_dirty; |
f6d6d4fc | 765 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 766 | xfs_off_t offset = page_offset(page); |
1da177e4 | 767 | |
10ce4444 CH |
768 | if (page->index != tindex) |
769 | goto fail; | |
529ae9aa | 770 | if (!trylock_page(page)) |
10ce4444 CH |
771 | goto fail; |
772 | if (PageWriteback(page)) | |
773 | goto fail_unlock_page; | |
774 | if (page->mapping != inode->i_mapping) | |
775 | goto fail_unlock_page; | |
776 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
777 | goto fail_unlock_page; | |
778 | ||
24e17b5f NS |
779 | /* |
780 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 781 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
782 | * |
783 | * Derivation: | |
784 | * | |
785 | * End offset is the highest offset that this page should represent. | |
786 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
787 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
788 | * hence give us the correct page_dirty count. On any other page, | |
789 | * it will be zero and in that case we need page_dirty to be the | |
790 | * count of buffers on the page. | |
24e17b5f | 791 | */ |
9260dc6b CH |
792 | end_offset = min_t(unsigned long long, |
793 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
794 | i_size_read(inode)); | |
795 | ||
24e17b5f | 796 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
797 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
798 | PAGE_CACHE_SIZE); | |
799 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
800 | page_dirty = p_offset / len; | |
24e17b5f | 801 | |
1da177e4 LT |
802 | bh = head = page_buffers(page); |
803 | do { | |
9260dc6b | 804 | if (offset >= end_offset) |
1da177e4 | 805 | break; |
f6d6d4fc CH |
806 | if (!buffer_uptodate(bh)) |
807 | uptodate = 0; | |
808 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
809 | done = 1; | |
1da177e4 | 810 | continue; |
f6d6d4fc CH |
811 | } |
812 | ||
9260dc6b CH |
813 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
814 | if (buffer_unwritten(bh)) | |
815 | type = IOMAP_UNWRITTEN; | |
816 | else | |
817 | type = IOMAP_DELAY; | |
818 | ||
207d0416 | 819 | if (!xfs_iomap_valid(inode, imap, offset)) { |
f6d6d4fc | 820 | done = 1; |
9260dc6b CH |
821 | continue; |
822 | } | |
823 | ||
207d0416 CH |
824 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
825 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
9260dc6b | 826 | |
207d0416 | 827 | xfs_map_at_offset(inode, bh, imap, offset); |
9260dc6b | 828 | if (startio) { |
7336cea8 | 829 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
830 | type, ioendp, done); |
831 | } else { | |
832 | set_buffer_dirty(bh); | |
833 | unlock_buffer(bh); | |
834 | mark_buffer_dirty(bh); | |
835 | } | |
836 | page_dirty--; | |
837 | count++; | |
838 | } else { | |
df3c7244 | 839 | type = IOMAP_NEW; |
9260dc6b | 840 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 841 | lock_buffer(bh); |
7336cea8 | 842 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
843 | type, ioendp, done); |
844 | count++; | |
24e17b5f | 845 | page_dirty--; |
9260dc6b CH |
846 | } else { |
847 | done = 1; | |
1da177e4 | 848 | } |
1da177e4 | 849 | } |
7336cea8 | 850 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 851 | |
f6d6d4fc CH |
852 | if (uptodate && bh == head) |
853 | SetPageUptodate(page); | |
854 | ||
855 | if (startio) { | |
f5e596bb | 856 | if (count) { |
9fddaca2 | 857 | wbc->nr_to_write--; |
0d99519e | 858 | if (wbc->nr_to_write <= 0) |
f5e596bb | 859 | done = 1; |
f5e596bb | 860 | } |
b41759cf | 861 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 862 | } |
f6d6d4fc CH |
863 | |
864 | return done; | |
10ce4444 CH |
865 | fail_unlock_page: |
866 | unlock_page(page); | |
867 | fail: | |
868 | return 1; | |
1da177e4 LT |
869 | } |
870 | ||
871 | /* | |
872 | * Convert & write out a cluster of pages in the same extent as defined | |
873 | * by mp and following the start page. | |
874 | */ | |
875 | STATIC void | |
876 | xfs_cluster_write( | |
877 | struct inode *inode, | |
878 | pgoff_t tindex, | |
207d0416 | 879 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 880 | xfs_ioend_t **ioendp, |
1da177e4 LT |
881 | struct writeback_control *wbc, |
882 | int startio, | |
883 | int all_bh, | |
884 | pgoff_t tlast) | |
885 | { | |
10ce4444 CH |
886 | struct pagevec pvec; |
887 | int done = 0, i; | |
1da177e4 | 888 | |
10ce4444 CH |
889 | pagevec_init(&pvec, 0); |
890 | while (!done && tindex <= tlast) { | |
891 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
892 | ||
893 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 894 | break; |
10ce4444 CH |
895 | |
896 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
897 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
207d0416 | 898 | imap, ioendp, wbc, startio, all_bh); |
10ce4444 CH |
899 | if (done) |
900 | break; | |
901 | } | |
902 | ||
903 | pagevec_release(&pvec); | |
904 | cond_resched(); | |
1da177e4 LT |
905 | } |
906 | } | |
907 | ||
3ed3a434 DC |
908 | STATIC void |
909 | xfs_vm_invalidatepage( | |
910 | struct page *page, | |
911 | unsigned long offset) | |
912 | { | |
913 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
914 | block_invalidatepage(page, offset); | |
915 | } | |
916 | ||
917 | /* | |
918 | * If the page has delalloc buffers on it, we need to punch them out before we | |
919 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
920 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
921 | * is done on that same region - the delalloc extent is returned when none is | |
922 | * supposed to be there. | |
923 | * | |
924 | * We prevent this by truncating away the delalloc regions on the page before | |
925 | * invalidating it. Because they are delalloc, we can do this without needing a | |
926 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
927 | * truncation without a transaction as there is no space left for block | |
928 | * reservation (typically why we see a ENOSPC in writeback). | |
929 | * | |
930 | * This is not a performance critical path, so for now just do the punching a | |
931 | * buffer head at a time. | |
932 | */ | |
933 | STATIC void | |
934 | xfs_aops_discard_page( | |
935 | struct page *page) | |
936 | { | |
937 | struct inode *inode = page->mapping->host; | |
938 | struct xfs_inode *ip = XFS_I(inode); | |
939 | struct buffer_head *bh, *head; | |
940 | loff_t offset = page_offset(page); | |
941 | ssize_t len = 1 << inode->i_blkbits; | |
942 | ||
943 | if (!xfs_is_delayed_page(page, IOMAP_DELAY)) | |
944 | goto out_invalidate; | |
945 | ||
e8c3753c DC |
946 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
947 | goto out_invalidate; | |
948 | ||
3ed3a434 DC |
949 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
950 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
951 | page, ip->i_ino, offset); | |
952 | ||
953 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
954 | bh = head = page_buffers(page); | |
955 | do { | |
956 | int done; | |
957 | xfs_fileoff_t offset_fsb; | |
958 | xfs_bmbt_irec_t imap; | |
959 | int nimaps = 1; | |
960 | int error; | |
961 | xfs_fsblock_t firstblock; | |
962 | xfs_bmap_free_t flist; | |
963 | ||
964 | if (!buffer_delay(bh)) | |
965 | goto next_buffer; | |
966 | ||
967 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
968 | ||
969 | /* | |
970 | * Map the range first and check that it is a delalloc extent | |
971 | * before trying to unmap the range. Otherwise we will be | |
972 | * trying to remove a real extent (which requires a | |
973 | * transaction) or a hole, which is probably a bad idea... | |
974 | */ | |
975 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
976 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
977 | &nimaps, NULL, NULL); | |
978 | ||
979 | if (error) { | |
980 | /* something screwed, just bail */ | |
e8c3753c DC |
981 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
982 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
983 | "page discard failed delalloc mapping lookup."); | |
984 | } | |
3ed3a434 DC |
985 | break; |
986 | } | |
987 | if (!nimaps) { | |
988 | /* nothing there */ | |
989 | goto next_buffer; | |
990 | } | |
991 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
992 | /* been converted, ignore */ | |
993 | goto next_buffer; | |
994 | } | |
995 | WARN_ON(imap.br_blockcount == 0); | |
996 | ||
997 | /* | |
998 | * Note: while we initialise the firstblock/flist pair, they | |
999 | * should never be used because blocks should never be | |
1000 | * allocated or freed for a delalloc extent and hence we need | |
1001 | * don't cancel or finish them after the xfs_bunmapi() call. | |
1002 | */ | |
1003 | xfs_bmap_init(&flist, &firstblock); | |
1004 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
1005 | &flist, NULL, &done); | |
1006 | ||
1007 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
1008 | if (error) { | |
1009 | /* something screwed, just bail */ | |
e8c3753c DC |
1010 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1011 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 1012 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 1013 | } |
3ed3a434 DC |
1014 | break; |
1015 | } | |
1016 | next_buffer: | |
1017 | offset += len; | |
1018 | ||
1019 | } while ((bh = bh->b_this_page) != head); | |
1020 | ||
1021 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1022 | out_invalidate: | |
1023 | xfs_vm_invalidatepage(page, 0); | |
1024 | return; | |
1025 | } | |
1026 | ||
1da177e4 LT |
1027 | /* |
1028 | * Calling this without startio set means we are being asked to make a dirty | |
1029 | * page ready for freeing it's buffers. When called with startio set then | |
1030 | * we are coming from writepage. | |
1031 | * | |
1032 | * When called with startio set it is important that we write the WHOLE | |
1033 | * page if possible. | |
1034 | * The bh->b_state's cannot know if any of the blocks or which block for | |
1035 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 1036 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
1037 | * may clear the page dirty flag prior to calling write page, under the |
1038 | * assumption the entire page will be written out; by not writing out the | |
1039 | * whole page the page can be reused before all valid dirty data is | |
1040 | * written out. Note: in the case of a page that has been dirty'd by | |
1041 | * mapwrite and but partially setup by block_prepare_write the | |
1042 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
1043 | * valid state, thus the whole page must be written out thing. | |
1044 | */ | |
1045 | ||
1046 | STATIC int | |
1047 | xfs_page_state_convert( | |
1048 | struct inode *inode, | |
1049 | struct page *page, | |
1050 | struct writeback_control *wbc, | |
1051 | int startio, | |
1052 | int unmapped) /* also implies page uptodate */ | |
1053 | { | |
f6d6d4fc | 1054 | struct buffer_head *bh, *head; |
207d0416 | 1055 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 1056 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
1057 | loff_t offset; |
1058 | unsigned long p_offset = 0; | |
f6d6d4fc | 1059 | unsigned int type; |
1da177e4 LT |
1060 | __uint64_t end_offset; |
1061 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
1062 | ssize_t size, len; |
1063 | int flags, err, iomap_valid = 0, uptodate = 1; | |
8272145c NS |
1064 | int page_dirty, count = 0; |
1065 | int trylock = 0; | |
6c4fe19f | 1066 | int all_bh = unmapped; |
1da177e4 | 1067 | |
8272145c NS |
1068 | if (startio) { |
1069 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
1070 | trylock |= BMAPI_TRYLOCK; | |
1071 | } | |
3ba0815a | 1072 | |
1da177e4 LT |
1073 | /* Is this page beyond the end of the file? */ |
1074 | offset = i_size_read(inode); | |
1075 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1076 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1077 | if (page->index >= end_index) { | |
1078 | if ((page->index >= end_index + 1) || | |
1079 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
1080 | if (startio) |
1081 | unlock_page(page); | |
1082 | return 0; | |
1da177e4 LT |
1083 | } |
1084 | } | |
1085 | ||
1da177e4 | 1086 | /* |
24e17b5f | 1087 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 1088 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
1089 | * |
1090 | * Derivation: | |
1091 | * | |
1092 | * End offset is the highest offset that this page should represent. | |
1093 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
1094 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
1095 | * hence give us the correct page_dirty count. On any other page, | |
1096 | * it will be zero and in that case we need page_dirty to be the | |
1097 | * count of buffers on the page. | |
1098 | */ | |
1099 | end_offset = min_t(unsigned long long, | |
1100 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 1101 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
1102 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
1103 | PAGE_CACHE_SIZE); | |
1104 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
1105 | page_dirty = p_offset / len; |
1106 | ||
24e17b5f | 1107 | bh = head = page_buffers(page); |
f6d6d4fc | 1108 | offset = page_offset(page); |
df3c7244 DC |
1109 | flags = BMAPI_READ; |
1110 | type = IOMAP_NEW; | |
f6d6d4fc | 1111 | |
f6d6d4fc | 1112 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
1113 | |
1114 | do { | |
1115 | if (offset >= end_offset) | |
1116 | break; | |
1117 | if (!buffer_uptodate(bh)) | |
1118 | uptodate = 0; | |
f6d6d4fc | 1119 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
1120 | /* |
1121 | * the iomap is actually still valid, but the ioend | |
1122 | * isn't. shouldn't happen too often. | |
1123 | */ | |
1124 | iomap_valid = 0; | |
1da177e4 | 1125 | continue; |
f6d6d4fc | 1126 | } |
1da177e4 | 1127 | |
1defeac9 | 1128 | if (iomap_valid) |
207d0416 | 1129 | iomap_valid = xfs_iomap_valid(inode, &imap, offset); |
1da177e4 LT |
1130 | |
1131 | /* | |
1132 | * First case, map an unwritten extent and prepare for | |
1133 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1134 | * |
1da177e4 LT |
1135 | * Second case, allocate space for a delalloc buffer. |
1136 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1137 | * |
1138 | * Third case, an unmapped buffer was found, and we are | |
1139 | * in a path where we need to write the whole page out. | |
df3c7244 | 1140 | */ |
d5cb48aa CH |
1141 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1142 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1143 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1144 | int new_ioend = 0; |
1145 | ||
df3c7244 | 1146 | /* |
6c4fe19f CH |
1147 | * Make sure we don't use a read-only iomap |
1148 | */ | |
df3c7244 | 1149 | if (flags == BMAPI_READ) |
6c4fe19f CH |
1150 | iomap_valid = 0; |
1151 | ||
f6d6d4fc CH |
1152 | if (buffer_unwritten(bh)) { |
1153 | type = IOMAP_UNWRITTEN; | |
8272145c | 1154 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1155 | } else if (buffer_delay(bh)) { |
f6d6d4fc | 1156 | type = IOMAP_DELAY; |
8272145c | 1157 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1158 | } else { |
6c4fe19f | 1159 | type = IOMAP_NEW; |
8272145c | 1160 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1161 | } |
1162 | ||
1defeac9 | 1163 | if (!iomap_valid) { |
effd120e DC |
1164 | /* |
1165 | * if we didn't have a valid mapping then we | |
1166 | * need to ensure that we put the new mapping | |
1167 | * in a new ioend structure. This needs to be | |
1168 | * done to ensure that the ioends correctly | |
1169 | * reflect the block mappings at io completion | |
1170 | * for unwritten extent conversion. | |
1171 | */ | |
1172 | new_ioend = 1; | |
6c4fe19f CH |
1173 | if (type == IOMAP_NEW) { |
1174 | size = xfs_probe_cluster(inode, | |
1175 | page, bh, head, 0); | |
d5cb48aa CH |
1176 | } else { |
1177 | size = len; | |
1178 | } | |
1179 | ||
1180 | err = xfs_map_blocks(inode, offset, size, | |
207d0416 | 1181 | &imap, flags); |
f6d6d4fc | 1182 | if (err) |
1da177e4 | 1183 | goto error; |
207d0416 | 1184 | iomap_valid = xfs_iomap_valid(inode, &imap, offset); |
1da177e4 | 1185 | } |
1defeac9 | 1186 | if (iomap_valid) { |
207d0416 | 1187 | xfs_map_at_offset(inode, bh, &imap, offset); |
1da177e4 | 1188 | if (startio) { |
7336cea8 | 1189 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1190 | type, &ioend, |
effd120e | 1191 | new_ioend); |
1da177e4 LT |
1192 | } else { |
1193 | set_buffer_dirty(bh); | |
1194 | unlock_buffer(bh); | |
1195 | mark_buffer_dirty(bh); | |
1196 | } | |
1197 | page_dirty--; | |
f6d6d4fc | 1198 | count++; |
1da177e4 | 1199 | } |
d5cb48aa | 1200 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1201 | /* |
1202 | * we got here because the buffer is already mapped. | |
1203 | * That means it must already have extents allocated | |
1204 | * underneath it. Map the extent by reading it. | |
1205 | */ | |
df3c7244 | 1206 | if (!iomap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1207 | flags = BMAPI_READ; |
1208 | size = xfs_probe_cluster(inode, page, bh, | |
1209 | head, 1); | |
1210 | err = xfs_map_blocks(inode, offset, size, | |
207d0416 | 1211 | &imap, flags); |
6c4fe19f CH |
1212 | if (err) |
1213 | goto error; | |
207d0416 | 1214 | iomap_valid = xfs_iomap_valid(inode, &imap, offset); |
6c4fe19f | 1215 | } |
d5cb48aa | 1216 | |
df3c7244 DC |
1217 | /* |
1218 | * We set the type to IOMAP_NEW in case we are doing a | |
1219 | * small write at EOF that is extending the file but | |
1220 | * without needing an allocation. We need to update the | |
1221 | * file size on I/O completion in this case so it is | |
1222 | * the same case as having just allocated a new extent | |
1223 | * that we are writing into for the first time. | |
1224 | */ | |
1225 | type = IOMAP_NEW; | |
ca5de404 | 1226 | if (trylock_buffer(bh)) { |
d5cb48aa | 1227 | ASSERT(buffer_mapped(bh)); |
6c4fe19f CH |
1228 | if (iomap_valid) |
1229 | all_bh = 1; | |
7336cea8 | 1230 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1231 | &ioend, !iomap_valid); |
1232 | page_dirty--; | |
1233 | count++; | |
f6d6d4fc | 1234 | } else { |
1defeac9 | 1235 | iomap_valid = 0; |
1da177e4 | 1236 | } |
d5cb48aa CH |
1237 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1238 | (unmapped || startio)) { | |
1239 | iomap_valid = 0; | |
1da177e4 | 1240 | } |
f6d6d4fc CH |
1241 | |
1242 | if (!iohead) | |
1243 | iohead = ioend; | |
1244 | ||
1245 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1246 | |
1247 | if (uptodate && bh == head) | |
1248 | SetPageUptodate(page); | |
1249 | ||
f6d6d4fc | 1250 | if (startio) |
b41759cf | 1251 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1252 | |
1defeac9 | 1253 | if (ioend && iomap_valid) { |
8699bb0a | 1254 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
1255 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap.br_startoff); |
1256 | xfs_off_t iomap_bsize = XFS_FSB_TO_B(m, imap.br_blockcount); | |
8699bb0a CH |
1257 | |
1258 | offset = (iomap_offset + iomap_bsize - 1) >> | |
1da177e4 | 1259 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1260 | tlast = min_t(pgoff_t, offset, last_index); |
207d0416 | 1261 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
6c4fe19f | 1262 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1263 | } |
1264 | ||
f6d6d4fc | 1265 | if (iohead) |
06342cf8 | 1266 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1267 | |
1da177e4 LT |
1268 | return page_dirty; |
1269 | ||
1270 | error: | |
f6d6d4fc CH |
1271 | if (iohead) |
1272 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1273 | |
1274 | /* | |
1275 | * If it's delalloc and we have nowhere to put it, | |
1276 | * throw it away, unless the lower layers told | |
1277 | * us to try again. | |
1278 | */ | |
1279 | if (err != -EAGAIN) { | |
f6d6d4fc | 1280 | if (!unmapped) |
3ed3a434 | 1281 | xfs_aops_discard_page(page); |
1da177e4 LT |
1282 | ClearPageUptodate(page); |
1283 | } | |
1284 | return err; | |
1285 | } | |
1286 | ||
f51623b2 NS |
1287 | /* |
1288 | * writepage: Called from one of two places: | |
1289 | * | |
1290 | * 1. we are flushing a delalloc buffer head. | |
1291 | * | |
1292 | * 2. we are writing out a dirty page. Typically the page dirty | |
1293 | * state is cleared before we get here. In this case is it | |
1294 | * conceivable we have no buffer heads. | |
1295 | * | |
1296 | * For delalloc space on the page we need to allocate space and | |
1297 | * flush it. For unmapped buffer heads on the page we should | |
1298 | * allocate space if the page is uptodate. For any other dirty | |
1299 | * buffer heads on the page we should flush them. | |
1300 | * | |
1301 | * If we detect that a transaction would be required to flush | |
1302 | * the page, we have to check the process flags first, if we | |
1303 | * are already in a transaction or disk I/O during allocations | |
1304 | * is off, we need to fail the writepage and redirty the page. | |
1305 | */ | |
1306 | ||
1307 | STATIC int | |
e4c573bb | 1308 | xfs_vm_writepage( |
f51623b2 NS |
1309 | struct page *page, |
1310 | struct writeback_control *wbc) | |
1311 | { | |
1312 | int error; | |
1313 | int need_trans; | |
1314 | int delalloc, unmapped, unwritten; | |
1315 | struct inode *inode = page->mapping->host; | |
1316 | ||
0b1b213f | 1317 | trace_xfs_writepage(inode, page, 0); |
f51623b2 NS |
1318 | |
1319 | /* | |
1320 | * We need a transaction if: | |
1321 | * 1. There are delalloc buffers on the page | |
1322 | * 2. The page is uptodate and we have unmapped buffers | |
1323 | * 3. The page is uptodate and we have no buffers | |
1324 | * 4. There are unwritten buffers on the page | |
1325 | */ | |
1326 | ||
1327 | if (!page_has_buffers(page)) { | |
1328 | unmapped = 1; | |
1329 | need_trans = 1; | |
1330 | } else { | |
1331 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1332 | if (!PageUptodate(page)) | |
1333 | unmapped = 0; | |
1334 | need_trans = delalloc + unmapped + unwritten; | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * If we need a transaction and the process flags say | |
1339 | * we are already in a transaction, or no IO is allowed | |
1340 | * then mark the page dirty again and leave the page | |
1341 | * as is. | |
1342 | */ | |
59c1b082 | 1343 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1344 | goto out_fail; |
1345 | ||
1346 | /* | |
1347 | * Delay hooking up buffer heads until we have | |
1348 | * made our go/no-go decision. | |
1349 | */ | |
1350 | if (!page_has_buffers(page)) | |
1351 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1352 | ||
c8a4051c ES |
1353 | |
1354 | /* | |
1355 | * VM calculation for nr_to_write seems off. Bump it way | |
1356 | * up, this gets simple streaming writes zippy again. | |
1357 | * To be reviewed again after Jens' writeback changes. | |
1358 | */ | |
1359 | wbc->nr_to_write *= 4; | |
1360 | ||
f51623b2 NS |
1361 | /* |
1362 | * Convert delayed allocate, unwritten or unmapped space | |
1363 | * to real space and flush out to disk. | |
1364 | */ | |
1365 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1366 | if (error == -EAGAIN) | |
1367 | goto out_fail; | |
1368 | if (unlikely(error < 0)) | |
1369 | goto out_unlock; | |
1370 | ||
1371 | return 0; | |
1372 | ||
1373 | out_fail: | |
1374 | redirty_page_for_writepage(wbc, page); | |
1375 | unlock_page(page); | |
1376 | return 0; | |
1377 | out_unlock: | |
1378 | unlock_page(page); | |
1379 | return error; | |
1380 | } | |
1381 | ||
7d4fb40a NS |
1382 | STATIC int |
1383 | xfs_vm_writepages( | |
1384 | struct address_space *mapping, | |
1385 | struct writeback_control *wbc) | |
1386 | { | |
b3aea4ed | 1387 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1388 | return generic_writepages(mapping, wbc); |
1389 | } | |
1390 | ||
f51623b2 NS |
1391 | /* |
1392 | * Called to move a page into cleanable state - and from there | |
1393 | * to be released. Possibly the page is already clean. We always | |
1394 | * have buffer heads in this call. | |
1395 | * | |
1396 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1397 | * | |
1398 | * Possible scenarios are: | |
1399 | * | |
1400 | * 1. We are being called to release a page which has been written | |
1401 | * to via regular I/O. buffer heads will be dirty and possibly | |
1402 | * delalloc. If no delalloc buffer heads in this case then we | |
1403 | * can just return zero. | |
1404 | * | |
1405 | * 2. We are called to release a page which has been written via | |
1406 | * mmap, all we need to do is ensure there is no delalloc | |
1407 | * state in the buffer heads, if not we can let the caller | |
1408 | * free them and we should come back later via writepage. | |
1409 | */ | |
1410 | STATIC int | |
238f4c54 | 1411 | xfs_vm_releasepage( |
f51623b2 NS |
1412 | struct page *page, |
1413 | gfp_t gfp_mask) | |
1414 | { | |
1415 | struct inode *inode = page->mapping->host; | |
1416 | int dirty, delalloc, unmapped, unwritten; | |
1417 | struct writeback_control wbc = { | |
1418 | .sync_mode = WB_SYNC_ALL, | |
1419 | .nr_to_write = 1, | |
1420 | }; | |
1421 | ||
0b1b213f | 1422 | trace_xfs_releasepage(inode, page, 0); |
f51623b2 | 1423 | |
238f4c54 NS |
1424 | if (!page_has_buffers(page)) |
1425 | return 0; | |
1426 | ||
f51623b2 NS |
1427 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1428 | if (!delalloc && !unwritten) | |
1429 | goto free_buffers; | |
1430 | ||
1431 | if (!(gfp_mask & __GFP_FS)) | |
1432 | return 0; | |
1433 | ||
1434 | /* If we are already inside a transaction or the thread cannot | |
1435 | * do I/O, we cannot release this page. | |
1436 | */ | |
59c1b082 | 1437 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1438 | return 0; |
1439 | ||
1440 | /* | |
1441 | * Convert delalloc space to real space, do not flush the | |
1442 | * data out to disk, that will be done by the caller. | |
1443 | * Never need to allocate space here - we will always | |
1444 | * come back to writepage in that case. | |
1445 | */ | |
1446 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1447 | if (dirty == 0 && !unwritten) | |
1448 | goto free_buffers; | |
1449 | return 0; | |
1450 | ||
1451 | free_buffers: | |
1452 | return try_to_free_buffers(page); | |
1453 | } | |
1454 | ||
1da177e4 | 1455 | STATIC int |
c2536668 | 1456 | __xfs_get_blocks( |
1da177e4 LT |
1457 | struct inode *inode, |
1458 | sector_t iblock, | |
1da177e4 LT |
1459 | struct buffer_head *bh_result, |
1460 | int create, | |
1461 | int direct, | |
1462 | bmapi_flags_t flags) | |
1463 | { | |
207d0416 | 1464 | struct xfs_bmbt_irec imap; |
fdc7ed75 NS |
1465 | xfs_off_t offset; |
1466 | ssize_t size; | |
207d0416 CH |
1467 | int nimap = 1; |
1468 | int new = 0; | |
1da177e4 | 1469 | int error; |
1da177e4 | 1470 | |
fdc7ed75 | 1471 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1472 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1473 | size = bh_result->b_size; | |
364f358a LM |
1474 | |
1475 | if (!create && direct && offset >= i_size_read(inode)) | |
1476 | return 0; | |
1477 | ||
541d7d3c | 1478 | error = xfs_iomap(XFS_I(inode), offset, size, |
207d0416 | 1479 | create ? flags : BMAPI_READ, &imap, &nimap, &new); |
1da177e4 LT |
1480 | if (error) |
1481 | return -error; | |
207d0416 | 1482 | if (nimap == 0) |
1da177e4 LT |
1483 | return 0; |
1484 | ||
207d0416 CH |
1485 | if (imap.br_startblock != HOLESTARTBLOCK && |
1486 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1487 | /* |
1488 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1489 | * the read case (treat as if we're reading into a hole). |
1490 | */ | |
207d0416 CH |
1491 | if (create || !ISUNWRITTEN(&imap)) |
1492 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1493 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1494 | if (direct) |
1495 | bh_result->b_private = inode; | |
1496 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1497 | } |
1498 | } | |
1499 | ||
c2536668 NS |
1500 | /* |
1501 | * If this is a realtime file, data may be on a different device. | |
1502 | * to that pointed to from the buffer_head b_bdev currently. | |
1503 | */ | |
046f1685 | 1504 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1505 | |
c2536668 | 1506 | /* |
549054af DC |
1507 | * If we previously allocated a block out beyond eof and we are now |
1508 | * coming back to use it then we will need to flag it as new even if it | |
1509 | * has a disk address. | |
1510 | * | |
1511 | * With sub-block writes into unwritten extents we also need to mark | |
1512 | * the buffer as new so that the unwritten parts of the buffer gets | |
1513 | * correctly zeroed. | |
1da177e4 LT |
1514 | */ |
1515 | if (create && | |
1516 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1517 | (offset >= i_size_read(inode)) || |
207d0416 | 1518 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1519 | set_buffer_new(bh_result); |
1da177e4 | 1520 | |
207d0416 | 1521 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1522 | BUG_ON(direct); |
1523 | if (create) { | |
1524 | set_buffer_uptodate(bh_result); | |
1525 | set_buffer_mapped(bh_result); | |
1526 | set_buffer_delay(bh_result); | |
1527 | } | |
1528 | } | |
1529 | ||
c2536668 | 1530 | if (direct || size > (1 << inode->i_blkbits)) { |
8699bb0a | 1531 | struct xfs_mount *mp = XFS_I(inode)->i_mount; |
207d0416 | 1532 | xfs_off_t iomap_offset = XFS_FSB_TO_B(mp, imap.br_startoff); |
8699bb0a | 1533 | xfs_off_t iomap_delta = offset - iomap_offset; |
207d0416 | 1534 | xfs_off_t iomap_bsize = XFS_FSB_TO_B(mp, imap.br_blockcount); |
9563b3d8 | 1535 | |
8699bb0a | 1536 | ASSERT(iomap_bsize - iomap_delta > 0); |
fdc7ed75 | 1537 | offset = min_t(xfs_off_t, |
8699bb0a | 1538 | iomap_bsize - iomap_delta, size); |
c2536668 | 1539 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); |
1da177e4 LT |
1540 | } |
1541 | ||
1542 | return 0; | |
1543 | } | |
1544 | ||
1545 | int | |
c2536668 | 1546 | xfs_get_blocks( |
1da177e4 LT |
1547 | struct inode *inode, |
1548 | sector_t iblock, | |
1549 | struct buffer_head *bh_result, | |
1550 | int create) | |
1551 | { | |
c2536668 | 1552 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1553 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1554 | } |
1555 | ||
1556 | STATIC int | |
e4c573bb | 1557 | xfs_get_blocks_direct( |
1da177e4 LT |
1558 | struct inode *inode, |
1559 | sector_t iblock, | |
1da177e4 LT |
1560 | struct buffer_head *bh_result, |
1561 | int create) | |
1562 | { | |
c2536668 | 1563 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1564 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1565 | } |
1566 | ||
f0973863 | 1567 | STATIC void |
e4c573bb | 1568 | xfs_end_io_direct( |
f0973863 CH |
1569 | struct kiocb *iocb, |
1570 | loff_t offset, | |
1571 | ssize_t size, | |
1572 | void *private) | |
1573 | { | |
1574 | xfs_ioend_t *ioend = iocb->private; | |
1575 | ||
1576 | /* | |
1577 | * Non-NULL private data means we need to issue a transaction to | |
1578 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1579 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1580 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1581 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1582 | * it anyway to keep the code uniform and simpler. |
1583 | * | |
e927af90 DC |
1584 | * Well, if only it were that simple. Because synchronous direct I/O |
1585 | * requires extent conversion to occur *before* we return to userspace, | |
1586 | * we have to wait for extent conversion to complete. Look at the | |
1587 | * iocb that has been passed to us to determine if this is AIO or | |
1588 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1589 | * workqueue and wait for it to complete. | |
1590 | * | |
f0973863 CH |
1591 | * The core direct I/O code might be changed to always call the |
1592 | * completion handler in the future, in which case all this can | |
1593 | * go away. | |
1594 | */ | |
ba87ea69 LM |
1595 | ioend->io_offset = offset; |
1596 | ioend->io_size = size; | |
1597 | if (ioend->io_type == IOMAP_READ) { | |
e927af90 | 1598 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1599 | } else if (private && size > 0) { |
e927af90 | 1600 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1601 | } else { |
ba87ea69 LM |
1602 | /* |
1603 | * A direct I/O write ioend starts it's life in unwritten | |
1604 | * state in case they map an unwritten extent. This write | |
1605 | * didn't map an unwritten extent so switch it's completion | |
1606 | * handler. | |
1607 | */ | |
5ec4fabb | 1608 | ioend->io_type = IOMAP_NEW; |
e927af90 | 1609 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1610 | } |
1611 | ||
1612 | /* | |
c41564b5 | 1613 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1614 | * completion handler was called. Thus we need to protect |
1615 | * against double-freeing. | |
1616 | */ | |
1617 | iocb->private = NULL; | |
1618 | } | |
1619 | ||
1da177e4 | 1620 | STATIC ssize_t |
e4c573bb | 1621 | xfs_vm_direct_IO( |
1da177e4 LT |
1622 | int rw, |
1623 | struct kiocb *iocb, | |
1624 | const struct iovec *iov, | |
1625 | loff_t offset, | |
1626 | unsigned long nr_segs) | |
1627 | { | |
1628 | struct file *file = iocb->ki_filp; | |
1629 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1630 | struct block_device *bdev; |
f0973863 | 1631 | ssize_t ret; |
1da177e4 | 1632 | |
046f1685 | 1633 | bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1634 | |
5fe878ae CH |
1635 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
1636 | IOMAP_UNWRITTEN : IOMAP_READ); | |
1637 | ||
1638 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1639 | offset, nr_segs, | |
1640 | xfs_get_blocks_direct, | |
1641 | xfs_end_io_direct); | |
f0973863 | 1642 | |
8459d86a | 1643 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1644 | xfs_destroy_ioend(iocb->private); |
1645 | return ret; | |
1da177e4 LT |
1646 | } |
1647 | ||
f51623b2 | 1648 | STATIC int |
d79689c7 | 1649 | xfs_vm_write_begin( |
f51623b2 | 1650 | struct file *file, |
d79689c7 NP |
1651 | struct address_space *mapping, |
1652 | loff_t pos, | |
1653 | unsigned len, | |
1654 | unsigned flags, | |
1655 | struct page **pagep, | |
1656 | void **fsdata) | |
f51623b2 | 1657 | { |
d79689c7 NP |
1658 | *pagep = NULL; |
1659 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1660 | xfs_get_blocks); | |
f51623b2 | 1661 | } |
1da177e4 LT |
1662 | |
1663 | STATIC sector_t | |
e4c573bb | 1664 | xfs_vm_bmap( |
1da177e4 LT |
1665 | struct address_space *mapping, |
1666 | sector_t block) | |
1667 | { | |
1668 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1669 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1670 | |
cf441eeb | 1671 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1672 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1673 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1674 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1675 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1676 | } |
1677 | ||
1678 | STATIC int | |
e4c573bb | 1679 | xfs_vm_readpage( |
1da177e4 LT |
1680 | struct file *unused, |
1681 | struct page *page) | |
1682 | { | |
c2536668 | 1683 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1684 | } |
1685 | ||
1686 | STATIC int | |
e4c573bb | 1687 | xfs_vm_readpages( |
1da177e4 LT |
1688 | struct file *unused, |
1689 | struct address_space *mapping, | |
1690 | struct list_head *pages, | |
1691 | unsigned nr_pages) | |
1692 | { | |
c2536668 | 1693 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1694 | } |
1695 | ||
f5e54d6e | 1696 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1697 | .readpage = xfs_vm_readpage, |
1698 | .readpages = xfs_vm_readpages, | |
1699 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1700 | .writepages = xfs_vm_writepages, |
1da177e4 | 1701 | .sync_page = block_sync_page, |
238f4c54 NS |
1702 | .releasepage = xfs_vm_releasepage, |
1703 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1704 | .write_begin = xfs_vm_write_begin, |
1705 | .write_end = generic_write_end, | |
e4c573bb NS |
1706 | .bmap = xfs_vm_bmap, |
1707 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1708 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1709 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1710 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1711 | }; |