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 | ||
34a52c6c CH |
48 | /* |
49 | * Types of I/O for bmap clustering and I/O completion tracking. | |
50 | */ | |
51 | enum { | |
52 | IO_READ, /* mapping for a read */ | |
53 | IO_DELAY, /* mapping covers delalloc region */ | |
54 | IO_UNWRITTEN, /* mapping covers allocated but uninitialized data */ | |
55 | IO_NEW /* just allocated */ | |
56 | }; | |
25e41b3d CH |
57 | |
58 | /* | |
59 | * Prime number of hash buckets since address is used as the key. | |
60 | */ | |
61 | #define NVSYNC 37 | |
62 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
63 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
64 | ||
65 | void __init | |
66 | xfs_ioend_init(void) | |
67 | { | |
68 | int i; | |
69 | ||
70 | for (i = 0; i < NVSYNC; i++) | |
71 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
72 | } | |
73 | ||
74 | void | |
75 | xfs_ioend_wait( | |
76 | xfs_inode_t *ip) | |
77 | { | |
78 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
79 | ||
80 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
81 | } | |
82 | ||
83 | STATIC void | |
84 | xfs_ioend_wake( | |
85 | xfs_inode_t *ip) | |
86 | { | |
87 | if (atomic_dec_and_test(&ip->i_iocount)) | |
88 | wake_up(to_ioend_wq(ip)); | |
89 | } | |
90 | ||
0b1b213f | 91 | void |
f51623b2 NS |
92 | xfs_count_page_state( |
93 | struct page *page, | |
94 | int *delalloc, | |
95 | int *unmapped, | |
96 | int *unwritten) | |
97 | { | |
98 | struct buffer_head *bh, *head; | |
99 | ||
100 | *delalloc = *unmapped = *unwritten = 0; | |
101 | ||
102 | bh = head = page_buffers(page); | |
103 | do { | |
104 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
105 | (*unmapped) = 1; | |
f51623b2 NS |
106 | else if (buffer_unwritten(bh)) |
107 | (*unwritten) = 1; | |
108 | else if (buffer_delay(bh)) | |
109 | (*delalloc) = 1; | |
110 | } while ((bh = bh->b_this_page) != head); | |
111 | } | |
112 | ||
6214ed44 CH |
113 | STATIC struct block_device * |
114 | xfs_find_bdev_for_inode( | |
046f1685 | 115 | struct inode *inode) |
6214ed44 | 116 | { |
046f1685 | 117 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
118 | struct xfs_mount *mp = ip->i_mount; |
119 | ||
71ddabb9 | 120 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
121 | return mp->m_rtdev_targp->bt_bdev; |
122 | else | |
123 | return mp->m_ddev_targp->bt_bdev; | |
124 | } | |
125 | ||
f6d6d4fc CH |
126 | /* |
127 | * We're now finished for good with this ioend structure. | |
128 | * Update the page state via the associated buffer_heads, | |
129 | * release holds on the inode and bio, and finally free | |
130 | * up memory. Do not use the ioend after this. | |
131 | */ | |
0829c360 CH |
132 | STATIC void |
133 | xfs_destroy_ioend( | |
134 | xfs_ioend_t *ioend) | |
135 | { | |
f6d6d4fc | 136 | struct buffer_head *bh, *next; |
583fa586 | 137 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
138 | |
139 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
140 | next = bh->b_private; | |
7d04a335 | 141 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 142 | } |
583fa586 CH |
143 | |
144 | /* | |
145 | * Volume managers supporting multiple paths can send back ENODEV | |
146 | * when the final path disappears. In this case continuing to fill | |
147 | * the page cache with dirty data which cannot be written out is | |
148 | * evil, so prevent that. | |
149 | */ | |
150 | if (unlikely(ioend->io_error == -ENODEV)) { | |
151 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
152 | __FILE__, __LINE__); | |
b677c210 | 153 | } |
583fa586 | 154 | |
25e41b3d | 155 | xfs_ioend_wake(ip); |
0829c360 CH |
156 | mempool_free(ioend, xfs_ioend_pool); |
157 | } | |
158 | ||
932640e8 DC |
159 | /* |
160 | * If the end of the current ioend is beyond the current EOF, | |
161 | * return the new EOF value, otherwise zero. | |
162 | */ | |
163 | STATIC xfs_fsize_t | |
164 | xfs_ioend_new_eof( | |
165 | xfs_ioend_t *ioend) | |
166 | { | |
167 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
168 | xfs_fsize_t isize; | |
169 | xfs_fsize_t bsize; | |
170 | ||
171 | bsize = ioend->io_offset + ioend->io_size; | |
172 | isize = MAX(ip->i_size, ip->i_new_size); | |
173 | isize = MIN(isize, bsize); | |
174 | return isize > ip->i_d.di_size ? isize : 0; | |
175 | } | |
176 | ||
ba87ea69 | 177 | /* |
77d7a0c2 DC |
178 | * Update on-disk file size now that data has been written to disk. The |
179 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
180 | * will be the intended file size until i_size is updated. If this write does | |
181 | * not extend all the way to the valid file size then restrict this update to | |
182 | * the end of the write. | |
183 | * | |
184 | * This function does not block as blocking on the inode lock in IO completion | |
185 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
186 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 187 | */ |
77d7a0c2 | 188 | STATIC int |
ba87ea69 LM |
189 | xfs_setfilesize( |
190 | xfs_ioend_t *ioend) | |
191 | { | |
b677c210 | 192 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 193 | xfs_fsize_t isize; |
ba87ea69 | 194 | |
ba87ea69 | 195 | ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG); |
34a52c6c | 196 | ASSERT(ioend->io_type != IO_READ); |
ba87ea69 LM |
197 | |
198 | if (unlikely(ioend->io_error)) | |
77d7a0c2 DC |
199 | return 0; |
200 | ||
201 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
202 | return EAGAIN; | |
ba87ea69 | 203 | |
932640e8 DC |
204 | isize = xfs_ioend_new_eof(ioend); |
205 | if (isize) { | |
ba87ea69 | 206 | ip->i_d.di_size = isize; |
66d834ea | 207 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
208 | } |
209 | ||
210 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
211 | return 0; |
212 | } | |
213 | ||
214 | /* | |
215 | * Schedule IO completion handling on a xfsdatad if this was | |
216 | * the final hold on this ioend. If we are asked to wait, | |
217 | * flush the workqueue. | |
218 | */ | |
219 | STATIC void | |
220 | xfs_finish_ioend( | |
221 | xfs_ioend_t *ioend, | |
222 | int wait) | |
223 | { | |
224 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
225 | struct workqueue_struct *wq; | |
226 | ||
34a52c6c | 227 | wq = (ioend->io_type == IO_UNWRITTEN) ? |
77d7a0c2 DC |
228 | xfsconvertd_workqueue : xfsdatad_workqueue; |
229 | queue_work(wq, &ioend->io_work); | |
230 | if (wait) | |
231 | flush_workqueue(wq); | |
232 | } | |
ba87ea69 LM |
233 | } |
234 | ||
0829c360 | 235 | /* |
5ec4fabb | 236 | * IO write completion. |
f6d6d4fc CH |
237 | */ |
238 | STATIC void | |
5ec4fabb | 239 | xfs_end_io( |
77d7a0c2 | 240 | struct work_struct *work) |
0829c360 | 241 | { |
77d7a0c2 DC |
242 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
243 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 244 | int error = 0; |
ba87ea69 | 245 | |
5ec4fabb CH |
246 | /* |
247 | * For unwritten extents we need to issue transactions to convert a | |
248 | * range to normal written extens after the data I/O has finished. | |
249 | */ | |
34a52c6c | 250 | if (ioend->io_type == IO_UNWRITTEN && |
5ec4fabb | 251 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { |
5ec4fabb CH |
252 | |
253 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
254 | ioend->io_size); | |
255 | if (error) | |
256 | ioend->io_error = error; | |
257 | } | |
ba87ea69 | 258 | |
5ec4fabb CH |
259 | /* |
260 | * We might have to update the on-disk file size after extending | |
261 | * writes. | |
262 | */ | |
34a52c6c | 263 | if (ioend->io_type != IO_READ) { |
77d7a0c2 DC |
264 | error = xfs_setfilesize(ioend); |
265 | ASSERT(!error || error == EAGAIN); | |
c626d174 | 266 | } |
77d7a0c2 DC |
267 | |
268 | /* | |
269 | * If we didn't complete processing of the ioend, requeue it to the | |
270 | * tail of the workqueue for another attempt later. Otherwise destroy | |
271 | * it. | |
272 | */ | |
273 | if (error == EAGAIN) { | |
274 | atomic_inc(&ioend->io_remaining); | |
275 | xfs_finish_ioend(ioend, 0); | |
276 | /* ensure we don't spin on blocked ioends */ | |
277 | delay(1); | |
278 | } else | |
279 | xfs_destroy_ioend(ioend); | |
c626d174 DC |
280 | } |
281 | ||
0829c360 CH |
282 | /* |
283 | * Allocate and initialise an IO completion structure. | |
284 | * We need to track unwritten extent write completion here initially. | |
285 | * We'll need to extend this for updating the ondisk inode size later | |
286 | * (vs. incore size). | |
287 | */ | |
288 | STATIC xfs_ioend_t * | |
289 | xfs_alloc_ioend( | |
f6d6d4fc CH |
290 | struct inode *inode, |
291 | unsigned int type) | |
0829c360 CH |
292 | { |
293 | xfs_ioend_t *ioend; | |
294 | ||
295 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
296 | ||
297 | /* | |
298 | * Set the count to 1 initially, which will prevent an I/O | |
299 | * completion callback from happening before we have started | |
300 | * all the I/O from calling the completion routine too early. | |
301 | */ | |
302 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 303 | ioend->io_error = 0; |
f6d6d4fc CH |
304 | ioend->io_list = NULL; |
305 | ioend->io_type = type; | |
b677c210 | 306 | ioend->io_inode = inode; |
c1a073bd | 307 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 308 | ioend->io_buffer_tail = NULL; |
b677c210 | 309 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
310 | ioend->io_offset = 0; |
311 | ioend->io_size = 0; | |
312 | ||
5ec4fabb | 313 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
314 | return ioend; |
315 | } | |
316 | ||
1da177e4 LT |
317 | STATIC int |
318 | xfs_map_blocks( | |
319 | struct inode *inode, | |
320 | loff_t offset, | |
321 | ssize_t count, | |
207d0416 | 322 | struct xfs_bmbt_irec *imap, |
1da177e4 LT |
323 | int flags) |
324 | { | |
6bd16ff2 | 325 | int nmaps = 1; |
207d0416 | 326 | int new = 0; |
6bd16ff2 | 327 | |
207d0416 | 328 | return -xfs_iomap(XFS_I(inode), offset, count, flags, imap, &nmaps, &new); |
1da177e4 LT |
329 | } |
330 | ||
b8f82a4a | 331 | STATIC int |
558e6891 | 332 | xfs_imap_valid( |
8699bb0a | 333 | struct inode *inode, |
207d0416 | 334 | struct xfs_bmbt_irec *imap, |
558e6891 | 335 | xfs_off_t offset) |
1da177e4 | 336 | { |
558e6891 | 337 | offset >>= inode->i_blkbits; |
8699bb0a | 338 | |
558e6891 CH |
339 | return offset >= imap->br_startoff && |
340 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
341 | } |
342 | ||
f6d6d4fc CH |
343 | /* |
344 | * BIO completion handler for buffered IO. | |
345 | */ | |
782e3b3b | 346 | STATIC void |
f6d6d4fc CH |
347 | xfs_end_bio( |
348 | struct bio *bio, | |
f6d6d4fc CH |
349 | int error) |
350 | { | |
351 | xfs_ioend_t *ioend = bio->bi_private; | |
352 | ||
f6d6d4fc | 353 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 354 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
355 | |
356 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
357 | bio->bi_private = NULL; |
358 | bio->bi_end_io = NULL; | |
f6d6d4fc | 359 | bio_put(bio); |
7d04a335 | 360 | |
e927af90 | 361 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
362 | } |
363 | ||
364 | STATIC void | |
365 | xfs_submit_ioend_bio( | |
06342cf8 CH |
366 | struct writeback_control *wbc, |
367 | xfs_ioend_t *ioend, | |
368 | struct bio *bio) | |
f6d6d4fc CH |
369 | { |
370 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
371 | bio->bi_private = ioend; |
372 | bio->bi_end_io = xfs_end_bio; | |
373 | ||
932640e8 DC |
374 | /* |
375 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
376 | * but don't update the inode size until I/O completion. | |
377 | */ | |
378 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 379 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 380 | |
06342cf8 CH |
381 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? |
382 | WRITE_SYNC_PLUG : WRITE, bio); | |
f6d6d4fc CH |
383 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); |
384 | bio_put(bio); | |
385 | } | |
386 | ||
387 | STATIC struct bio * | |
388 | xfs_alloc_ioend_bio( | |
389 | struct buffer_head *bh) | |
390 | { | |
391 | struct bio *bio; | |
392 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
393 | ||
394 | do { | |
395 | bio = bio_alloc(GFP_NOIO, nvecs); | |
396 | nvecs >>= 1; | |
397 | } while (!bio); | |
398 | ||
399 | ASSERT(bio->bi_private == NULL); | |
400 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
401 | bio->bi_bdev = bh->b_bdev; | |
402 | bio_get(bio); | |
403 | return bio; | |
404 | } | |
405 | ||
406 | STATIC void | |
407 | xfs_start_buffer_writeback( | |
408 | struct buffer_head *bh) | |
409 | { | |
410 | ASSERT(buffer_mapped(bh)); | |
411 | ASSERT(buffer_locked(bh)); | |
412 | ASSERT(!buffer_delay(bh)); | |
413 | ASSERT(!buffer_unwritten(bh)); | |
414 | ||
415 | mark_buffer_async_write(bh); | |
416 | set_buffer_uptodate(bh); | |
417 | clear_buffer_dirty(bh); | |
418 | } | |
419 | ||
420 | STATIC void | |
421 | xfs_start_page_writeback( | |
422 | struct page *page, | |
f6d6d4fc CH |
423 | int clear_dirty, |
424 | int buffers) | |
425 | { | |
426 | ASSERT(PageLocked(page)); | |
427 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 428 | if (clear_dirty) |
92132021 DC |
429 | clear_page_dirty_for_io(page); |
430 | set_page_writeback(page); | |
f6d6d4fc | 431 | unlock_page(page); |
1f7decf6 FW |
432 | /* If no buffers on the page are to be written, finish it here */ |
433 | if (!buffers) | |
f6d6d4fc | 434 | end_page_writeback(page); |
f6d6d4fc CH |
435 | } |
436 | ||
437 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
438 | { | |
439 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
440 | } | |
441 | ||
442 | /* | |
d88992f6 DC |
443 | * Submit all of the bios for all of the ioends we have saved up, covering the |
444 | * initial writepage page and also any probed pages. | |
445 | * | |
446 | * Because we may have multiple ioends spanning a page, we need to start | |
447 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
448 | * buffers as we got, then we can end up with a page that only has buffers | |
449 | * marked async write and I/O complete on can occur before we mark the other | |
450 | * buffers async write. | |
451 | * | |
452 | * The end result of this is that we trip a bug in end_page_writeback() because | |
453 | * we call it twice for the one page as the code in end_buffer_async_write() | |
454 | * assumes that all buffers on the page are started at the same time. | |
455 | * | |
456 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 457 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
458 | */ |
459 | STATIC void | |
460 | xfs_submit_ioend( | |
06342cf8 | 461 | struct writeback_control *wbc, |
f6d6d4fc CH |
462 | xfs_ioend_t *ioend) |
463 | { | |
d88992f6 | 464 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
465 | xfs_ioend_t *next; |
466 | struct buffer_head *bh; | |
467 | struct bio *bio; | |
468 | sector_t lastblock = 0; | |
469 | ||
d88992f6 DC |
470 | /* Pass 1 - start writeback */ |
471 | do { | |
472 | next = ioend->io_list; | |
473 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
474 | xfs_start_buffer_writeback(bh); | |
475 | } | |
476 | } while ((ioend = next) != NULL); | |
477 | ||
478 | /* Pass 2 - submit I/O */ | |
479 | ioend = head; | |
f6d6d4fc CH |
480 | do { |
481 | next = ioend->io_list; | |
482 | bio = NULL; | |
483 | ||
484 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
485 | |
486 | if (!bio) { | |
487 | retry: | |
488 | bio = xfs_alloc_ioend_bio(bh); | |
489 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 490 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
491 | goto retry; |
492 | } | |
493 | ||
494 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 495 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
496 | goto retry; |
497 | } | |
498 | ||
499 | lastblock = bh->b_blocknr; | |
500 | } | |
501 | if (bio) | |
06342cf8 | 502 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e927af90 | 503 | xfs_finish_ioend(ioend, 0); |
f6d6d4fc CH |
504 | } while ((ioend = next) != NULL); |
505 | } | |
506 | ||
507 | /* | |
508 | * Cancel submission of all buffer_heads so far in this endio. | |
509 | * Toss the endio too. Only ever called for the initial page | |
510 | * in a writepage request, so only ever one page. | |
511 | */ | |
512 | STATIC void | |
513 | xfs_cancel_ioend( | |
514 | xfs_ioend_t *ioend) | |
515 | { | |
516 | xfs_ioend_t *next; | |
517 | struct buffer_head *bh, *next_bh; | |
518 | ||
519 | do { | |
520 | next = ioend->io_list; | |
521 | bh = ioend->io_buffer_head; | |
522 | do { | |
523 | next_bh = bh->b_private; | |
524 | clear_buffer_async_write(bh); | |
525 | unlock_buffer(bh); | |
526 | } while ((bh = next_bh) != NULL); | |
527 | ||
25e41b3d | 528 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
529 | mempool_free(ioend, xfs_ioend_pool); |
530 | } while ((ioend = next) != NULL); | |
531 | } | |
532 | ||
533 | /* | |
534 | * Test to see if we've been building up a completion structure for | |
535 | * earlier buffers -- if so, we try to append to this ioend if we | |
536 | * can, otherwise we finish off any current ioend and start another. | |
537 | * Return true if we've finished the given ioend. | |
538 | */ | |
539 | STATIC void | |
540 | xfs_add_to_ioend( | |
541 | struct inode *inode, | |
542 | struct buffer_head *bh, | |
7336cea8 | 543 | xfs_off_t offset, |
f6d6d4fc CH |
544 | unsigned int type, |
545 | xfs_ioend_t **result, | |
546 | int need_ioend) | |
547 | { | |
548 | xfs_ioend_t *ioend = *result; | |
549 | ||
550 | if (!ioend || need_ioend || type != ioend->io_type) { | |
551 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 552 | |
f6d6d4fc CH |
553 | ioend = xfs_alloc_ioend(inode, type); |
554 | ioend->io_offset = offset; | |
555 | ioend->io_buffer_head = bh; | |
556 | ioend->io_buffer_tail = bh; | |
557 | if (previous) | |
558 | previous->io_list = ioend; | |
559 | *result = ioend; | |
560 | } else { | |
561 | ioend->io_buffer_tail->b_private = bh; | |
562 | ioend->io_buffer_tail = bh; | |
563 | } | |
564 | ||
565 | bh->b_private = NULL; | |
566 | ioend->io_size += bh->b_size; | |
567 | } | |
568 | ||
87cbc49c NS |
569 | STATIC void |
570 | xfs_map_buffer( | |
046f1685 | 571 | struct inode *inode, |
87cbc49c | 572 | struct buffer_head *bh, |
207d0416 | 573 | struct xfs_bmbt_irec *imap, |
046f1685 | 574 | xfs_off_t offset) |
87cbc49c NS |
575 | { |
576 | sector_t bn; | |
8699bb0a | 577 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
578 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
579 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 580 | |
207d0416 CH |
581 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
582 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 583 | |
e513182d | 584 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 585 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 586 | |
046f1685 | 587 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
588 | |
589 | bh->b_blocknr = bn; | |
590 | set_buffer_mapped(bh); | |
591 | } | |
592 | ||
1da177e4 LT |
593 | STATIC void |
594 | xfs_map_at_offset( | |
046f1685 | 595 | struct inode *inode, |
1da177e4 | 596 | struct buffer_head *bh, |
207d0416 | 597 | struct xfs_bmbt_irec *imap, |
046f1685 | 598 | xfs_off_t offset) |
1da177e4 | 599 | { |
207d0416 CH |
600 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
601 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 LT |
602 | |
603 | lock_buffer(bh); | |
207d0416 | 604 | xfs_map_buffer(inode, bh, imap, offset); |
046f1685 | 605 | bh->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 LT |
606 | set_buffer_mapped(bh); |
607 | clear_buffer_delay(bh); | |
f6d6d4fc | 608 | clear_buffer_unwritten(bh); |
1da177e4 LT |
609 | } |
610 | ||
611 | /* | |
6c4fe19f | 612 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
613 | */ |
614 | STATIC unsigned int | |
6c4fe19f | 615 | xfs_probe_page( |
10ce4444 | 616 | struct page *page, |
6c4fe19f CH |
617 | unsigned int pg_offset, |
618 | int mapped) | |
1da177e4 | 619 | { |
1da177e4 LT |
620 | int ret = 0; |
621 | ||
1da177e4 | 622 | if (PageWriteback(page)) |
10ce4444 | 623 | return 0; |
1da177e4 LT |
624 | |
625 | if (page->mapping && PageDirty(page)) { | |
626 | if (page_has_buffers(page)) { | |
627 | struct buffer_head *bh, *head; | |
628 | ||
629 | bh = head = page_buffers(page); | |
630 | do { | |
6c4fe19f CH |
631 | if (!buffer_uptodate(bh)) |
632 | break; | |
633 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
634 | break; |
635 | ret += bh->b_size; | |
636 | if (ret >= pg_offset) | |
637 | break; | |
638 | } while ((bh = bh->b_this_page) != head); | |
639 | } else | |
6c4fe19f | 640 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
641 | } |
642 | ||
1da177e4 LT |
643 | return ret; |
644 | } | |
645 | ||
f6d6d4fc | 646 | STATIC size_t |
6c4fe19f | 647 | xfs_probe_cluster( |
1da177e4 LT |
648 | struct inode *inode, |
649 | struct page *startpage, | |
650 | struct buffer_head *bh, | |
6c4fe19f CH |
651 | struct buffer_head *head, |
652 | int mapped) | |
1da177e4 | 653 | { |
10ce4444 | 654 | struct pagevec pvec; |
1da177e4 | 655 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
656 | size_t total = 0; |
657 | int done = 0, i; | |
1da177e4 LT |
658 | |
659 | /* First sum forwards in this page */ | |
660 | do { | |
2353e8e9 | 661 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 662 | return total; |
1da177e4 LT |
663 | total += bh->b_size; |
664 | } while ((bh = bh->b_this_page) != head); | |
665 | ||
10ce4444 CH |
666 | /* if we reached the end of the page, sum forwards in following pages */ |
667 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
668 | tindex = startpage->index + 1; | |
669 | ||
670 | /* Prune this back to avoid pathological behavior */ | |
671 | tloff = min(tlast, startpage->index + 64); | |
672 | ||
673 | pagevec_init(&pvec, 0); | |
674 | while (!done && tindex <= tloff) { | |
675 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
676 | ||
677 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
678 | break; | |
679 | ||
680 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
681 | struct page *page = pvec.pages[i]; | |
265c1fac | 682 | size_t pg_offset, pg_len = 0; |
10ce4444 CH |
683 | |
684 | if (tindex == tlast) { | |
685 | pg_offset = | |
686 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
687 | if (!pg_offset) { |
688 | done = 1; | |
10ce4444 | 689 | break; |
1defeac9 | 690 | } |
10ce4444 CH |
691 | } else |
692 | pg_offset = PAGE_CACHE_SIZE; | |
693 | ||
529ae9aa | 694 | if (page->index == tindex && trylock_page(page)) { |
265c1fac | 695 | pg_len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
696 | unlock_page(page); |
697 | } | |
698 | ||
265c1fac | 699 | if (!pg_len) { |
10ce4444 CH |
700 | done = 1; |
701 | break; | |
702 | } | |
703 | ||
265c1fac | 704 | total += pg_len; |
1defeac9 | 705 | tindex++; |
1da177e4 | 706 | } |
10ce4444 CH |
707 | |
708 | pagevec_release(&pvec); | |
709 | cond_resched(); | |
1da177e4 | 710 | } |
10ce4444 | 711 | |
1da177e4 LT |
712 | return total; |
713 | } | |
714 | ||
715 | /* | |
10ce4444 CH |
716 | * Test if a given page is suitable for writing as part of an unwritten |
717 | * or delayed allocate extent. | |
1da177e4 | 718 | */ |
10ce4444 CH |
719 | STATIC int |
720 | xfs_is_delayed_page( | |
721 | struct page *page, | |
f6d6d4fc | 722 | unsigned int type) |
1da177e4 | 723 | { |
1da177e4 | 724 | if (PageWriteback(page)) |
10ce4444 | 725 | return 0; |
1da177e4 LT |
726 | |
727 | if (page->mapping && page_has_buffers(page)) { | |
728 | struct buffer_head *bh, *head; | |
729 | int acceptable = 0; | |
730 | ||
731 | bh = head = page_buffers(page); | |
732 | do { | |
f6d6d4fc | 733 | if (buffer_unwritten(bh)) |
34a52c6c | 734 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 735 | else if (buffer_delay(bh)) |
34a52c6c | 736 | acceptable = (type == IO_DELAY); |
2ddee844 | 737 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
34a52c6c | 738 | acceptable = (type == IO_NEW); |
f6d6d4fc | 739 | else |
1da177e4 | 740 | break; |
1da177e4 LT |
741 | } while ((bh = bh->b_this_page) != head); |
742 | ||
743 | if (acceptable) | |
10ce4444 | 744 | return 1; |
1da177e4 LT |
745 | } |
746 | ||
10ce4444 | 747 | return 0; |
1da177e4 LT |
748 | } |
749 | ||
1da177e4 LT |
750 | /* |
751 | * Allocate & map buffers for page given the extent map. Write it out. | |
752 | * except for the original page of a writepage, this is called on | |
753 | * delalloc/unwritten pages only, for the original page it is possible | |
754 | * that the page has no mapping at all. | |
755 | */ | |
f6d6d4fc | 756 | STATIC int |
1da177e4 LT |
757 | xfs_convert_page( |
758 | struct inode *inode, | |
759 | struct page *page, | |
10ce4444 | 760 | loff_t tindex, |
207d0416 | 761 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 762 | xfs_ioend_t **ioendp, |
1da177e4 | 763 | struct writeback_control *wbc, |
1da177e4 LT |
764 | int startio, |
765 | int all_bh) | |
766 | { | |
f6d6d4fc | 767 | struct buffer_head *bh, *head; |
9260dc6b CH |
768 | xfs_off_t end_offset; |
769 | unsigned long p_offset; | |
f6d6d4fc | 770 | unsigned int type; |
24e17b5f | 771 | int len, page_dirty; |
f6d6d4fc | 772 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 773 | xfs_off_t offset = page_offset(page); |
1da177e4 | 774 | |
10ce4444 CH |
775 | if (page->index != tindex) |
776 | goto fail; | |
529ae9aa | 777 | if (!trylock_page(page)) |
10ce4444 CH |
778 | goto fail; |
779 | if (PageWriteback(page)) | |
780 | goto fail_unlock_page; | |
781 | if (page->mapping != inode->i_mapping) | |
782 | goto fail_unlock_page; | |
783 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
784 | goto fail_unlock_page; | |
785 | ||
24e17b5f NS |
786 | /* |
787 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 788 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
789 | * |
790 | * Derivation: | |
791 | * | |
792 | * End offset is the highest offset that this page should represent. | |
793 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
794 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
795 | * hence give us the correct page_dirty count. On any other page, | |
796 | * it will be zero and in that case we need page_dirty to be the | |
797 | * count of buffers on the page. | |
24e17b5f | 798 | */ |
9260dc6b CH |
799 | end_offset = min_t(unsigned long long, |
800 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
801 | i_size_read(inode)); | |
802 | ||
24e17b5f | 803 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
804 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
805 | PAGE_CACHE_SIZE); | |
806 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
807 | page_dirty = p_offset / len; | |
24e17b5f | 808 | |
1da177e4 LT |
809 | bh = head = page_buffers(page); |
810 | do { | |
9260dc6b | 811 | if (offset >= end_offset) |
1da177e4 | 812 | break; |
f6d6d4fc CH |
813 | if (!buffer_uptodate(bh)) |
814 | uptodate = 0; | |
815 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
816 | done = 1; | |
1da177e4 | 817 | continue; |
f6d6d4fc CH |
818 | } |
819 | ||
9260dc6b CH |
820 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
821 | if (buffer_unwritten(bh)) | |
34a52c6c | 822 | type = IO_UNWRITTEN; |
9260dc6b | 823 | else |
34a52c6c | 824 | type = IO_DELAY; |
9260dc6b | 825 | |
558e6891 | 826 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 827 | done = 1; |
9260dc6b CH |
828 | continue; |
829 | } | |
830 | ||
207d0416 CH |
831 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
832 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
9260dc6b | 833 | |
207d0416 | 834 | xfs_map_at_offset(inode, bh, imap, offset); |
9260dc6b | 835 | if (startio) { |
7336cea8 | 836 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
837 | type, ioendp, done); |
838 | } else { | |
839 | set_buffer_dirty(bh); | |
840 | unlock_buffer(bh); | |
841 | mark_buffer_dirty(bh); | |
842 | } | |
843 | page_dirty--; | |
844 | count++; | |
845 | } else { | |
34a52c6c | 846 | type = IO_NEW; |
9260dc6b | 847 | if (buffer_mapped(bh) && all_bh && startio) { |
1da177e4 | 848 | lock_buffer(bh); |
7336cea8 | 849 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
850 | type, ioendp, done); |
851 | count++; | |
24e17b5f | 852 | page_dirty--; |
9260dc6b CH |
853 | } else { |
854 | done = 1; | |
1da177e4 | 855 | } |
1da177e4 | 856 | } |
7336cea8 | 857 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 858 | |
f6d6d4fc CH |
859 | if (uptodate && bh == head) |
860 | SetPageUptodate(page); | |
861 | ||
862 | if (startio) { | |
f5e596bb | 863 | if (count) { |
9fddaca2 | 864 | wbc->nr_to_write--; |
0d99519e | 865 | if (wbc->nr_to_write <= 0) |
f5e596bb | 866 | done = 1; |
f5e596bb | 867 | } |
b41759cf | 868 | xfs_start_page_writeback(page, !page_dirty, count); |
1da177e4 | 869 | } |
f6d6d4fc CH |
870 | |
871 | return done; | |
10ce4444 CH |
872 | fail_unlock_page: |
873 | unlock_page(page); | |
874 | fail: | |
875 | return 1; | |
1da177e4 LT |
876 | } |
877 | ||
878 | /* | |
879 | * Convert & write out a cluster of pages in the same extent as defined | |
880 | * by mp and following the start page. | |
881 | */ | |
882 | STATIC void | |
883 | xfs_cluster_write( | |
884 | struct inode *inode, | |
885 | pgoff_t tindex, | |
207d0416 | 886 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 887 | xfs_ioend_t **ioendp, |
1da177e4 LT |
888 | struct writeback_control *wbc, |
889 | int startio, | |
890 | int all_bh, | |
891 | pgoff_t tlast) | |
892 | { | |
10ce4444 CH |
893 | struct pagevec pvec; |
894 | int done = 0, i; | |
1da177e4 | 895 | |
10ce4444 CH |
896 | pagevec_init(&pvec, 0); |
897 | while (!done && tindex <= tlast) { | |
898 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
899 | ||
900 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 901 | break; |
10ce4444 CH |
902 | |
903 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
904 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
207d0416 | 905 | imap, ioendp, wbc, startio, all_bh); |
10ce4444 CH |
906 | if (done) |
907 | break; | |
908 | } | |
909 | ||
910 | pagevec_release(&pvec); | |
911 | cond_resched(); | |
1da177e4 LT |
912 | } |
913 | } | |
914 | ||
3ed3a434 DC |
915 | STATIC void |
916 | xfs_vm_invalidatepage( | |
917 | struct page *page, | |
918 | unsigned long offset) | |
919 | { | |
920 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
921 | block_invalidatepage(page, offset); | |
922 | } | |
923 | ||
924 | /* | |
925 | * If the page has delalloc buffers on it, we need to punch them out before we | |
926 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
927 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
928 | * is done on that same region - the delalloc extent is returned when none is | |
929 | * supposed to be there. | |
930 | * | |
931 | * We prevent this by truncating away the delalloc regions on the page before | |
932 | * invalidating it. Because they are delalloc, we can do this without needing a | |
933 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
934 | * truncation without a transaction as there is no space left for block | |
935 | * reservation (typically why we see a ENOSPC in writeback). | |
936 | * | |
937 | * This is not a performance critical path, so for now just do the punching a | |
938 | * buffer head at a time. | |
939 | */ | |
940 | STATIC void | |
941 | xfs_aops_discard_page( | |
942 | struct page *page) | |
943 | { | |
944 | struct inode *inode = page->mapping->host; | |
945 | struct xfs_inode *ip = XFS_I(inode); | |
946 | struct buffer_head *bh, *head; | |
947 | loff_t offset = page_offset(page); | |
948 | ssize_t len = 1 << inode->i_blkbits; | |
949 | ||
34a52c6c | 950 | if (!xfs_is_delayed_page(page, IO_DELAY)) |
3ed3a434 DC |
951 | goto out_invalidate; |
952 | ||
e8c3753c DC |
953 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
954 | goto out_invalidate; | |
955 | ||
3ed3a434 DC |
956 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, |
957 | "page discard on page %p, inode 0x%llx, offset %llu.", | |
958 | page, ip->i_ino, offset); | |
959 | ||
960 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
961 | bh = head = page_buffers(page); | |
962 | do { | |
963 | int done; | |
964 | xfs_fileoff_t offset_fsb; | |
965 | xfs_bmbt_irec_t imap; | |
966 | int nimaps = 1; | |
967 | int error; | |
968 | xfs_fsblock_t firstblock; | |
969 | xfs_bmap_free_t flist; | |
970 | ||
971 | if (!buffer_delay(bh)) | |
972 | goto next_buffer; | |
973 | ||
974 | offset_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); | |
975 | ||
976 | /* | |
977 | * Map the range first and check that it is a delalloc extent | |
978 | * before trying to unmap the range. Otherwise we will be | |
979 | * trying to remove a real extent (which requires a | |
980 | * transaction) or a hole, which is probably a bad idea... | |
981 | */ | |
982 | error = xfs_bmapi(NULL, ip, offset_fsb, 1, | |
983 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, | |
984 | &nimaps, NULL, NULL); | |
985 | ||
986 | if (error) { | |
987 | /* something screwed, just bail */ | |
e8c3753c DC |
988 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
989 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
990 | "page discard failed delalloc mapping lookup."); | |
991 | } | |
3ed3a434 DC |
992 | break; |
993 | } | |
994 | if (!nimaps) { | |
995 | /* nothing there */ | |
996 | goto next_buffer; | |
997 | } | |
998 | if (imap.br_startblock != DELAYSTARTBLOCK) { | |
999 | /* been converted, ignore */ | |
1000 | goto next_buffer; | |
1001 | } | |
1002 | WARN_ON(imap.br_blockcount == 0); | |
1003 | ||
1004 | /* | |
1005 | * Note: while we initialise the firstblock/flist pair, they | |
1006 | * should never be used because blocks should never be | |
1007 | * allocated or freed for a delalloc extent and hence we need | |
1008 | * don't cancel or finish them after the xfs_bunmapi() call. | |
1009 | */ | |
1010 | xfs_bmap_init(&flist, &firstblock); | |
1011 | error = xfs_bunmapi(NULL, ip, offset_fsb, 1, 0, 1, &firstblock, | |
1012 | &flist, NULL, &done); | |
1013 | ||
1014 | ASSERT(!flist.xbf_count && !flist.xbf_first); | |
1015 | if (error) { | |
1016 | /* something screwed, just bail */ | |
e8c3753c DC |
1017 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
1018 | xfs_fs_cmn_err(CE_ALERT, ip->i_mount, | |
3ed3a434 | 1019 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 1020 | } |
3ed3a434 DC |
1021 | break; |
1022 | } | |
1023 | next_buffer: | |
1024 | offset += len; | |
1025 | ||
1026 | } while ((bh = bh->b_this_page) != head); | |
1027 | ||
1028 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1029 | out_invalidate: | |
1030 | xfs_vm_invalidatepage(page, 0); | |
1031 | return; | |
1032 | } | |
1033 | ||
1da177e4 LT |
1034 | /* |
1035 | * Calling this without startio set means we are being asked to make a dirty | |
1036 | * page ready for freeing it's buffers. When called with startio set then | |
1037 | * we are coming from writepage. | |
1038 | * | |
1039 | * When called with startio set it is important that we write the WHOLE | |
1040 | * page if possible. | |
1041 | * The bh->b_state's cannot know if any of the blocks or which block for | |
1042 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
c41564b5 | 1043 | * only valid if the page itself isn't completely uptodate. Some layers |
1da177e4 LT |
1044 | * may clear the page dirty flag prior to calling write page, under the |
1045 | * assumption the entire page will be written out; by not writing out the | |
1046 | * whole page the page can be reused before all valid dirty data is | |
1047 | * written out. Note: in the case of a page that has been dirty'd by | |
1048 | * mapwrite and but partially setup by block_prepare_write the | |
1049 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
1050 | * valid state, thus the whole page must be written out thing. | |
1051 | */ | |
1052 | ||
1053 | STATIC int | |
1054 | xfs_page_state_convert( | |
1055 | struct inode *inode, | |
1056 | struct page *page, | |
1057 | struct writeback_control *wbc, | |
1058 | int startio, | |
1059 | int unmapped) /* also implies page uptodate */ | |
1060 | { | |
f6d6d4fc | 1061 | struct buffer_head *bh, *head; |
207d0416 | 1062 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 1063 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
1064 | loff_t offset; |
1065 | unsigned long p_offset = 0; | |
f6d6d4fc | 1066 | unsigned int type; |
1da177e4 LT |
1067 | __uint64_t end_offset; |
1068 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa | 1069 | ssize_t size, len; |
558e6891 | 1070 | int flags, err, imap_valid = 0, uptodate = 1; |
8272145c NS |
1071 | int page_dirty, count = 0; |
1072 | int trylock = 0; | |
6c4fe19f | 1073 | int all_bh = unmapped; |
1da177e4 | 1074 | |
8272145c NS |
1075 | if (startio) { |
1076 | if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking) | |
1077 | trylock |= BMAPI_TRYLOCK; | |
1078 | } | |
3ba0815a | 1079 | |
1da177e4 LT |
1080 | /* Is this page beyond the end of the file? */ |
1081 | offset = i_size_read(inode); | |
1082 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1083 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
1084 | if (page->index >= end_index) { | |
1085 | if ((page->index >= end_index + 1) || | |
1086 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
1087 | if (startio) |
1088 | unlock_page(page); | |
1089 | return 0; | |
1da177e4 LT |
1090 | } |
1091 | } | |
1092 | ||
1da177e4 | 1093 | /* |
24e17b5f | 1094 | * page_dirty is initially a count of buffers on the page before |
c41564b5 | 1095 | * EOF and is decremented as we move each into a cleanable state. |
f6d6d4fc CH |
1096 | * |
1097 | * Derivation: | |
1098 | * | |
1099 | * End offset is the highest offset that this page should represent. | |
1100 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
1101 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
1102 | * hence give us the correct page_dirty count. On any other page, | |
1103 | * it will be zero and in that case we need page_dirty to be the | |
1104 | * count of buffers on the page. | |
1105 | */ | |
1106 | end_offset = min_t(unsigned long long, | |
1107 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 1108 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
1109 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
1110 | PAGE_CACHE_SIZE); | |
1111 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
1112 | page_dirty = p_offset / len; |
1113 | ||
24e17b5f | 1114 | bh = head = page_buffers(page); |
f6d6d4fc | 1115 | offset = page_offset(page); |
df3c7244 | 1116 | flags = BMAPI_READ; |
34a52c6c | 1117 | type = IO_NEW; |
f6d6d4fc | 1118 | |
f6d6d4fc | 1119 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
1120 | |
1121 | do { | |
1122 | if (offset >= end_offset) | |
1123 | break; | |
1124 | if (!buffer_uptodate(bh)) | |
1125 | uptodate = 0; | |
f6d6d4fc | 1126 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
1127 | /* |
1128 | * the iomap is actually still valid, but the ioend | |
1129 | * isn't. shouldn't happen too often. | |
1130 | */ | |
558e6891 | 1131 | imap_valid = 0; |
1da177e4 | 1132 | continue; |
f6d6d4fc | 1133 | } |
1da177e4 | 1134 | |
558e6891 CH |
1135 | if (imap_valid) |
1136 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1da177e4 LT |
1137 | |
1138 | /* | |
1139 | * First case, map an unwritten extent and prepare for | |
1140 | * extent state conversion transaction on completion. | |
f6d6d4fc | 1141 | * |
1da177e4 LT |
1142 | * Second case, allocate space for a delalloc buffer. |
1143 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
1144 | * |
1145 | * Third case, an unmapped buffer was found, and we are | |
1146 | * in a path where we need to write the whole page out. | |
df3c7244 | 1147 | */ |
d5cb48aa CH |
1148 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
1149 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
1150 | !buffer_mapped(bh) && (unmapped || startio))) { | |
effd120e DC |
1151 | int new_ioend = 0; |
1152 | ||
df3c7244 | 1153 | /* |
6c4fe19f CH |
1154 | * Make sure we don't use a read-only iomap |
1155 | */ | |
df3c7244 | 1156 | if (flags == BMAPI_READ) |
558e6891 | 1157 | imap_valid = 0; |
6c4fe19f | 1158 | |
f6d6d4fc | 1159 | if (buffer_unwritten(bh)) { |
34a52c6c | 1160 | type = IO_UNWRITTEN; |
8272145c | 1161 | flags = BMAPI_WRITE | BMAPI_IGNSTATE; |
d5cb48aa | 1162 | } else if (buffer_delay(bh)) { |
34a52c6c | 1163 | type = IO_DELAY; |
8272145c | 1164 | flags = BMAPI_ALLOCATE | trylock; |
d5cb48aa | 1165 | } else { |
34a52c6c | 1166 | type = IO_NEW; |
8272145c | 1167 | flags = BMAPI_WRITE | BMAPI_MMAP; |
f6d6d4fc CH |
1168 | } |
1169 | ||
558e6891 | 1170 | if (!imap_valid) { |
effd120e DC |
1171 | /* |
1172 | * if we didn't have a valid mapping then we | |
1173 | * need to ensure that we put the new mapping | |
1174 | * in a new ioend structure. This needs to be | |
1175 | * done to ensure that the ioends correctly | |
1176 | * reflect the block mappings at io completion | |
1177 | * for unwritten extent conversion. | |
1178 | */ | |
1179 | new_ioend = 1; | |
34a52c6c | 1180 | if (type == IO_NEW) { |
6c4fe19f CH |
1181 | size = xfs_probe_cluster(inode, |
1182 | page, bh, head, 0); | |
d5cb48aa CH |
1183 | } else { |
1184 | size = len; | |
1185 | } | |
1186 | ||
1187 | err = xfs_map_blocks(inode, offset, size, | |
207d0416 | 1188 | &imap, flags); |
f6d6d4fc | 1189 | if (err) |
1da177e4 | 1190 | goto error; |
558e6891 CH |
1191 | imap_valid = xfs_imap_valid(inode, &imap, |
1192 | offset); | |
1da177e4 | 1193 | } |
558e6891 | 1194 | if (imap_valid) { |
207d0416 | 1195 | xfs_map_at_offset(inode, bh, &imap, offset); |
1da177e4 | 1196 | if (startio) { |
7336cea8 | 1197 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 | 1198 | type, &ioend, |
effd120e | 1199 | new_ioend); |
1da177e4 LT |
1200 | } else { |
1201 | set_buffer_dirty(bh); | |
1202 | unlock_buffer(bh); | |
1203 | mark_buffer_dirty(bh); | |
1204 | } | |
1205 | page_dirty--; | |
f6d6d4fc | 1206 | count++; |
1da177e4 | 1207 | } |
d5cb48aa | 1208 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1209 | /* |
1210 | * we got here because the buffer is already mapped. | |
1211 | * That means it must already have extents allocated | |
1212 | * underneath it. Map the extent by reading it. | |
1213 | */ | |
558e6891 | 1214 | if (!imap_valid || flags != BMAPI_READ) { |
6c4fe19f CH |
1215 | flags = BMAPI_READ; |
1216 | size = xfs_probe_cluster(inode, page, bh, | |
1217 | head, 1); | |
1218 | err = xfs_map_blocks(inode, offset, size, | |
207d0416 | 1219 | &imap, flags); |
6c4fe19f CH |
1220 | if (err) |
1221 | goto error; | |
558e6891 CH |
1222 | imap_valid = xfs_imap_valid(inode, &imap, |
1223 | offset); | |
6c4fe19f | 1224 | } |
d5cb48aa | 1225 | |
df3c7244 | 1226 | /* |
34a52c6c | 1227 | * We set the type to IO_NEW in case we are doing a |
df3c7244 DC |
1228 | * small write at EOF that is extending the file but |
1229 | * without needing an allocation. We need to update the | |
1230 | * file size on I/O completion in this case so it is | |
1231 | * the same case as having just allocated a new extent | |
1232 | * that we are writing into for the first time. | |
1233 | */ | |
34a52c6c | 1234 | type = IO_NEW; |
ca5de404 | 1235 | if (trylock_buffer(bh)) { |
d5cb48aa | 1236 | ASSERT(buffer_mapped(bh)); |
558e6891 | 1237 | if (imap_valid) |
6c4fe19f | 1238 | all_bh = 1; |
7336cea8 | 1239 | xfs_add_to_ioend(inode, bh, offset, type, |
558e6891 | 1240 | &ioend, !imap_valid); |
d5cb48aa CH |
1241 | page_dirty--; |
1242 | count++; | |
f6d6d4fc | 1243 | } else { |
558e6891 | 1244 | imap_valid = 0; |
1da177e4 | 1245 | } |
d5cb48aa CH |
1246 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1247 | (unmapped || startio)) { | |
558e6891 | 1248 | imap_valid = 0; |
1da177e4 | 1249 | } |
f6d6d4fc CH |
1250 | |
1251 | if (!iohead) | |
1252 | iohead = ioend; | |
1253 | ||
1254 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1255 | |
1256 | if (uptodate && bh == head) | |
1257 | SetPageUptodate(page); | |
1258 | ||
f6d6d4fc | 1259 | if (startio) |
b41759cf | 1260 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1261 | |
558e6891 | 1262 | if (ioend && imap_valid) { |
8699bb0a | 1263 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
1264 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap.br_startoff); |
1265 | xfs_off_t iomap_bsize = XFS_FSB_TO_B(m, imap.br_blockcount); | |
8699bb0a CH |
1266 | |
1267 | offset = (iomap_offset + iomap_bsize - 1) >> | |
1da177e4 | 1268 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1269 | tlast = min_t(pgoff_t, offset, last_index); |
207d0416 | 1270 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
6c4fe19f | 1271 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1272 | } |
1273 | ||
f6d6d4fc | 1274 | if (iohead) |
06342cf8 | 1275 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1276 | |
1da177e4 LT |
1277 | return page_dirty; |
1278 | ||
1279 | error: | |
f6d6d4fc CH |
1280 | if (iohead) |
1281 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1282 | |
1283 | /* | |
1284 | * If it's delalloc and we have nowhere to put it, | |
1285 | * throw it away, unless the lower layers told | |
1286 | * us to try again. | |
1287 | */ | |
1288 | if (err != -EAGAIN) { | |
f6d6d4fc | 1289 | if (!unmapped) |
3ed3a434 | 1290 | xfs_aops_discard_page(page); |
1da177e4 LT |
1291 | ClearPageUptodate(page); |
1292 | } | |
1293 | return err; | |
1294 | } | |
1295 | ||
f51623b2 NS |
1296 | /* |
1297 | * writepage: Called from one of two places: | |
1298 | * | |
1299 | * 1. we are flushing a delalloc buffer head. | |
1300 | * | |
1301 | * 2. we are writing out a dirty page. Typically the page dirty | |
1302 | * state is cleared before we get here. In this case is it | |
1303 | * conceivable we have no buffer heads. | |
1304 | * | |
1305 | * For delalloc space on the page we need to allocate space and | |
1306 | * flush it. For unmapped buffer heads on the page we should | |
1307 | * allocate space if the page is uptodate. For any other dirty | |
1308 | * buffer heads on the page we should flush them. | |
1309 | * | |
1310 | * If we detect that a transaction would be required to flush | |
1311 | * the page, we have to check the process flags first, if we | |
1312 | * are already in a transaction or disk I/O during allocations | |
1313 | * is off, we need to fail the writepage and redirty the page. | |
1314 | */ | |
1315 | ||
1316 | STATIC int | |
e4c573bb | 1317 | xfs_vm_writepage( |
f51623b2 NS |
1318 | struct page *page, |
1319 | struct writeback_control *wbc) | |
1320 | { | |
1321 | int error; | |
1322 | int need_trans; | |
1323 | int delalloc, unmapped, unwritten; | |
1324 | struct inode *inode = page->mapping->host; | |
1325 | ||
0b1b213f | 1326 | trace_xfs_writepage(inode, page, 0); |
f51623b2 NS |
1327 | |
1328 | /* | |
1329 | * We need a transaction if: | |
1330 | * 1. There are delalloc buffers on the page | |
1331 | * 2. The page is uptodate and we have unmapped buffers | |
1332 | * 3. The page is uptodate and we have no buffers | |
1333 | * 4. There are unwritten buffers on the page | |
1334 | */ | |
1335 | ||
1336 | if (!page_has_buffers(page)) { | |
1337 | unmapped = 1; | |
1338 | need_trans = 1; | |
1339 | } else { | |
1340 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1341 | if (!PageUptodate(page)) | |
1342 | unmapped = 0; | |
1343 | need_trans = delalloc + unmapped + unwritten; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * If we need a transaction and the process flags say | |
1348 | * we are already in a transaction, or no IO is allowed | |
1349 | * then mark the page dirty again and leave the page | |
1350 | * as is. | |
1351 | */ | |
59c1b082 | 1352 | if (current_test_flags(PF_FSTRANS) && need_trans) |
f51623b2 NS |
1353 | goto out_fail; |
1354 | ||
1355 | /* | |
1356 | * Delay hooking up buffer heads until we have | |
1357 | * made our go/no-go decision. | |
1358 | */ | |
1359 | if (!page_has_buffers(page)) | |
1360 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1361 | ||
c8a4051c ES |
1362 | |
1363 | /* | |
1364 | * VM calculation for nr_to_write seems off. Bump it way | |
1365 | * up, this gets simple streaming writes zippy again. | |
1366 | * To be reviewed again after Jens' writeback changes. | |
1367 | */ | |
1368 | wbc->nr_to_write *= 4; | |
1369 | ||
f51623b2 NS |
1370 | /* |
1371 | * Convert delayed allocate, unwritten or unmapped space | |
1372 | * to real space and flush out to disk. | |
1373 | */ | |
1374 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1375 | if (error == -EAGAIN) | |
1376 | goto out_fail; | |
1377 | if (unlikely(error < 0)) | |
1378 | goto out_unlock; | |
1379 | ||
1380 | return 0; | |
1381 | ||
1382 | out_fail: | |
1383 | redirty_page_for_writepage(wbc, page); | |
1384 | unlock_page(page); | |
1385 | return 0; | |
1386 | out_unlock: | |
1387 | unlock_page(page); | |
1388 | return error; | |
1389 | } | |
1390 | ||
7d4fb40a NS |
1391 | STATIC int |
1392 | xfs_vm_writepages( | |
1393 | struct address_space *mapping, | |
1394 | struct writeback_control *wbc) | |
1395 | { | |
b3aea4ed | 1396 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1397 | return generic_writepages(mapping, wbc); |
1398 | } | |
1399 | ||
f51623b2 NS |
1400 | /* |
1401 | * Called to move a page into cleanable state - and from there | |
1402 | * to be released. Possibly the page is already clean. We always | |
1403 | * have buffer heads in this call. | |
1404 | * | |
1405 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1406 | * | |
1407 | * Possible scenarios are: | |
1408 | * | |
1409 | * 1. We are being called to release a page which has been written | |
1410 | * to via regular I/O. buffer heads will be dirty and possibly | |
1411 | * delalloc. If no delalloc buffer heads in this case then we | |
1412 | * can just return zero. | |
1413 | * | |
1414 | * 2. We are called to release a page which has been written via | |
1415 | * mmap, all we need to do is ensure there is no delalloc | |
1416 | * state in the buffer heads, if not we can let the caller | |
1417 | * free them and we should come back later via writepage. | |
1418 | */ | |
1419 | STATIC int | |
238f4c54 | 1420 | xfs_vm_releasepage( |
f51623b2 NS |
1421 | struct page *page, |
1422 | gfp_t gfp_mask) | |
1423 | { | |
1424 | struct inode *inode = page->mapping->host; | |
1425 | int dirty, delalloc, unmapped, unwritten; | |
1426 | struct writeback_control wbc = { | |
1427 | .sync_mode = WB_SYNC_ALL, | |
1428 | .nr_to_write = 1, | |
1429 | }; | |
1430 | ||
0b1b213f | 1431 | trace_xfs_releasepage(inode, page, 0); |
f51623b2 | 1432 | |
238f4c54 NS |
1433 | if (!page_has_buffers(page)) |
1434 | return 0; | |
1435 | ||
f51623b2 NS |
1436 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); |
1437 | if (!delalloc && !unwritten) | |
1438 | goto free_buffers; | |
1439 | ||
1440 | if (!(gfp_mask & __GFP_FS)) | |
1441 | return 0; | |
1442 | ||
1443 | /* If we are already inside a transaction or the thread cannot | |
1444 | * do I/O, we cannot release this page. | |
1445 | */ | |
59c1b082 | 1446 | if (current_test_flags(PF_FSTRANS)) |
f51623b2 NS |
1447 | return 0; |
1448 | ||
1449 | /* | |
1450 | * Convert delalloc space to real space, do not flush the | |
1451 | * data out to disk, that will be done by the caller. | |
1452 | * Never need to allocate space here - we will always | |
1453 | * come back to writepage in that case. | |
1454 | */ | |
1455 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1456 | if (dirty == 0 && !unwritten) | |
1457 | goto free_buffers; | |
1458 | return 0; | |
1459 | ||
1460 | free_buffers: | |
1461 | return try_to_free_buffers(page); | |
1462 | } | |
1463 | ||
1da177e4 | 1464 | STATIC int |
c2536668 | 1465 | __xfs_get_blocks( |
1da177e4 LT |
1466 | struct inode *inode, |
1467 | sector_t iblock, | |
1da177e4 LT |
1468 | struct buffer_head *bh_result, |
1469 | int create, | |
1470 | int direct, | |
1471 | bmapi_flags_t flags) | |
1472 | { | |
207d0416 | 1473 | struct xfs_bmbt_irec imap; |
fdc7ed75 NS |
1474 | xfs_off_t offset; |
1475 | ssize_t size; | |
207d0416 CH |
1476 | int nimap = 1; |
1477 | int new = 0; | |
1da177e4 | 1478 | int error; |
1da177e4 | 1479 | |
fdc7ed75 | 1480 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1481 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1482 | size = bh_result->b_size; | |
364f358a LM |
1483 | |
1484 | if (!create && direct && offset >= i_size_read(inode)) | |
1485 | return 0; | |
1486 | ||
541d7d3c | 1487 | error = xfs_iomap(XFS_I(inode), offset, size, |
207d0416 | 1488 | create ? flags : BMAPI_READ, &imap, &nimap, &new); |
1da177e4 LT |
1489 | if (error) |
1490 | return -error; | |
207d0416 | 1491 | if (nimap == 0) |
1da177e4 LT |
1492 | return 0; |
1493 | ||
207d0416 CH |
1494 | if (imap.br_startblock != HOLESTARTBLOCK && |
1495 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1496 | /* |
1497 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1498 | * the read case (treat as if we're reading into a hole). |
1499 | */ | |
207d0416 CH |
1500 | if (create || !ISUNWRITTEN(&imap)) |
1501 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1502 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1503 | if (direct) |
1504 | bh_result->b_private = inode; | |
1505 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1506 | } |
1507 | } | |
1508 | ||
c2536668 NS |
1509 | /* |
1510 | * If this is a realtime file, data may be on a different device. | |
1511 | * to that pointed to from the buffer_head b_bdev currently. | |
1512 | */ | |
046f1685 | 1513 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1514 | |
c2536668 | 1515 | /* |
549054af DC |
1516 | * If we previously allocated a block out beyond eof and we are now |
1517 | * coming back to use it then we will need to flag it as new even if it | |
1518 | * has a disk address. | |
1519 | * | |
1520 | * With sub-block writes into unwritten extents we also need to mark | |
1521 | * the buffer as new so that the unwritten parts of the buffer gets | |
1522 | * correctly zeroed. | |
1da177e4 LT |
1523 | */ |
1524 | if (create && | |
1525 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1526 | (offset >= i_size_read(inode)) || |
207d0416 | 1527 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1528 | set_buffer_new(bh_result); |
1da177e4 | 1529 | |
207d0416 | 1530 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1531 | BUG_ON(direct); |
1532 | if (create) { | |
1533 | set_buffer_uptodate(bh_result); | |
1534 | set_buffer_mapped(bh_result); | |
1535 | set_buffer_delay(bh_result); | |
1536 | } | |
1537 | } | |
1538 | ||
c2536668 | 1539 | if (direct || size > (1 << inode->i_blkbits)) { |
8699bb0a | 1540 | struct xfs_mount *mp = XFS_I(inode)->i_mount; |
207d0416 | 1541 | xfs_off_t iomap_offset = XFS_FSB_TO_B(mp, imap.br_startoff); |
8699bb0a | 1542 | xfs_off_t iomap_delta = offset - iomap_offset; |
207d0416 | 1543 | xfs_off_t iomap_bsize = XFS_FSB_TO_B(mp, imap.br_blockcount); |
9563b3d8 | 1544 | |
8699bb0a | 1545 | ASSERT(iomap_bsize - iomap_delta > 0); |
fdc7ed75 | 1546 | offset = min_t(xfs_off_t, |
8699bb0a | 1547 | iomap_bsize - iomap_delta, size); |
c2536668 | 1548 | bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset); |
1da177e4 LT |
1549 | } |
1550 | ||
1551 | return 0; | |
1552 | } | |
1553 | ||
1554 | int | |
c2536668 | 1555 | xfs_get_blocks( |
1da177e4 LT |
1556 | struct inode *inode, |
1557 | sector_t iblock, | |
1558 | struct buffer_head *bh_result, | |
1559 | int create) | |
1560 | { | |
c2536668 | 1561 | return __xfs_get_blocks(inode, iblock, |
fa30bd05 | 1562 | bh_result, create, 0, BMAPI_WRITE); |
1da177e4 LT |
1563 | } |
1564 | ||
1565 | STATIC int | |
e4c573bb | 1566 | xfs_get_blocks_direct( |
1da177e4 LT |
1567 | struct inode *inode, |
1568 | sector_t iblock, | |
1da177e4 LT |
1569 | struct buffer_head *bh_result, |
1570 | int create) | |
1571 | { | |
c2536668 | 1572 | return __xfs_get_blocks(inode, iblock, |
1d8fa7a2 | 1573 | bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1da177e4 LT |
1574 | } |
1575 | ||
f0973863 | 1576 | STATIC void |
e4c573bb | 1577 | xfs_end_io_direct( |
f0973863 CH |
1578 | struct kiocb *iocb, |
1579 | loff_t offset, | |
1580 | ssize_t size, | |
1581 | void *private) | |
1582 | { | |
1583 | xfs_ioend_t *ioend = iocb->private; | |
1584 | ||
1585 | /* | |
1586 | * Non-NULL private data means we need to issue a transaction to | |
1587 | * convert a range from unwritten to written extents. This needs | |
c41564b5 | 1588 | * to happen from process context but aio+dio I/O completion |
f0973863 | 1589 | * happens from irq context so we need to defer it to a workqueue. |
c41564b5 | 1590 | * This is not necessary for synchronous direct I/O, but we do |
f0973863 CH |
1591 | * it anyway to keep the code uniform and simpler. |
1592 | * | |
e927af90 DC |
1593 | * Well, if only it were that simple. Because synchronous direct I/O |
1594 | * requires extent conversion to occur *before* we return to userspace, | |
1595 | * we have to wait for extent conversion to complete. Look at the | |
1596 | * iocb that has been passed to us to determine if this is AIO or | |
1597 | * not. If it is synchronous, tell xfs_finish_ioend() to kick the | |
1598 | * workqueue and wait for it to complete. | |
1599 | * | |
f0973863 CH |
1600 | * The core direct I/O code might be changed to always call the |
1601 | * completion handler in the future, in which case all this can | |
1602 | * go away. | |
1603 | */ | |
ba87ea69 LM |
1604 | ioend->io_offset = offset; |
1605 | ioend->io_size = size; | |
34a52c6c | 1606 | if (ioend->io_type == IO_READ) { |
e927af90 | 1607 | xfs_finish_ioend(ioend, 0); |
ba87ea69 | 1608 | } else if (private && size > 0) { |
e927af90 | 1609 | xfs_finish_ioend(ioend, is_sync_kiocb(iocb)); |
f0973863 | 1610 | } else { |
ba87ea69 LM |
1611 | /* |
1612 | * A direct I/O write ioend starts it's life in unwritten | |
1613 | * state in case they map an unwritten extent. This write | |
1614 | * didn't map an unwritten extent so switch it's completion | |
1615 | * handler. | |
1616 | */ | |
34a52c6c | 1617 | ioend->io_type = IO_NEW; |
e927af90 | 1618 | xfs_finish_ioend(ioend, 0); |
f0973863 CH |
1619 | } |
1620 | ||
1621 | /* | |
c41564b5 | 1622 | * blockdev_direct_IO can return an error even after the I/O |
f0973863 CH |
1623 | * completion handler was called. Thus we need to protect |
1624 | * against double-freeing. | |
1625 | */ | |
1626 | iocb->private = NULL; | |
1627 | } | |
1628 | ||
1da177e4 | 1629 | STATIC ssize_t |
e4c573bb | 1630 | xfs_vm_direct_IO( |
1da177e4 LT |
1631 | int rw, |
1632 | struct kiocb *iocb, | |
1633 | const struct iovec *iov, | |
1634 | loff_t offset, | |
1635 | unsigned long nr_segs) | |
1636 | { | |
1637 | struct file *file = iocb->ki_filp; | |
1638 | struct inode *inode = file->f_mapping->host; | |
6214ed44 | 1639 | struct block_device *bdev; |
f0973863 | 1640 | ssize_t ret; |
1da177e4 | 1641 | |
046f1685 | 1642 | bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1643 | |
5fe878ae | 1644 | iocb->private = xfs_alloc_ioend(inode, rw == WRITE ? |
34a52c6c | 1645 | IO_UNWRITTEN : IO_READ); |
5fe878ae CH |
1646 | |
1647 | ret = blockdev_direct_IO_no_locking(rw, iocb, inode, bdev, iov, | |
1648 | offset, nr_segs, | |
1649 | xfs_get_blocks_direct, | |
1650 | xfs_end_io_direct); | |
f0973863 | 1651 | |
8459d86a | 1652 | if (unlikely(ret != -EIOCBQUEUED && iocb->private)) |
f0973863 CH |
1653 | xfs_destroy_ioend(iocb->private); |
1654 | return ret; | |
1da177e4 LT |
1655 | } |
1656 | ||
f51623b2 | 1657 | STATIC int |
d79689c7 | 1658 | xfs_vm_write_begin( |
f51623b2 | 1659 | struct file *file, |
d79689c7 NP |
1660 | struct address_space *mapping, |
1661 | loff_t pos, | |
1662 | unsigned len, | |
1663 | unsigned flags, | |
1664 | struct page **pagep, | |
1665 | void **fsdata) | |
f51623b2 | 1666 | { |
d79689c7 NP |
1667 | *pagep = NULL; |
1668 | return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata, | |
1669 | xfs_get_blocks); | |
f51623b2 | 1670 | } |
1da177e4 LT |
1671 | |
1672 | STATIC sector_t | |
e4c573bb | 1673 | xfs_vm_bmap( |
1da177e4 LT |
1674 | struct address_space *mapping, |
1675 | sector_t block) | |
1676 | { | |
1677 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1678 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1679 | |
cf441eeb | 1680 | xfs_itrace_entry(XFS_I(inode)); |
126468b1 | 1681 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1682 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1683 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1684 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1685 | } |
1686 | ||
1687 | STATIC int | |
e4c573bb | 1688 | xfs_vm_readpage( |
1da177e4 LT |
1689 | struct file *unused, |
1690 | struct page *page) | |
1691 | { | |
c2536668 | 1692 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1693 | } |
1694 | ||
1695 | STATIC int | |
e4c573bb | 1696 | xfs_vm_readpages( |
1da177e4 LT |
1697 | struct file *unused, |
1698 | struct address_space *mapping, | |
1699 | struct list_head *pages, | |
1700 | unsigned nr_pages) | |
1701 | { | |
c2536668 | 1702 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1703 | } |
1704 | ||
f5e54d6e | 1705 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1706 | .readpage = xfs_vm_readpage, |
1707 | .readpages = xfs_vm_readpages, | |
1708 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1709 | .writepages = xfs_vm_writepages, |
1da177e4 | 1710 | .sync_page = block_sync_page, |
238f4c54 NS |
1711 | .releasepage = xfs_vm_releasepage, |
1712 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 NP |
1713 | .write_begin = xfs_vm_write_begin, |
1714 | .write_end = generic_write_end, | |
e4c573bb NS |
1715 | .bmap = xfs_vm_bmap, |
1716 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1717 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1718 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1719 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1720 | }; |