]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_bit.h" |
1da177e4 | 20 | #include "xfs_log.h" |
a844f451 | 21 | #include "xfs_inum.h" |
1da177e4 | 22 | #include "xfs_sb.h" |
a844f451 | 23 | #include "xfs_ag.h" |
1da177e4 | 24 | #include "xfs_trans.h" |
1da177e4 LT |
25 | #include "xfs_mount.h" |
26 | #include "xfs_bmap_btree.h" | |
1da177e4 LT |
27 | #include "xfs_dinode.h" |
28 | #include "xfs_inode.h" | |
a844f451 | 29 | #include "xfs_alloc.h" |
1da177e4 LT |
30 | #include "xfs_error.h" |
31 | #include "xfs_rw.h" | |
32 | #include "xfs_iomap.h" | |
739bfb2a | 33 | #include "xfs_vnodeops.h" |
0b1b213f | 34 | #include "xfs_trace.h" |
3ed3a434 | 35 | #include "xfs_bmap.h" |
5a0e3ad6 | 36 | #include <linux/gfp.h> |
1da177e4 | 37 | #include <linux/mpage.h> |
10ce4444 | 38 | #include <linux/pagevec.h> |
1da177e4 LT |
39 | #include <linux/writeback.h> |
40 | ||
25e41b3d CH |
41 | |
42 | /* | |
43 | * Prime number of hash buckets since address is used as the key. | |
44 | */ | |
45 | #define NVSYNC 37 | |
46 | #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC]) | |
47 | static wait_queue_head_t xfs_ioend_wq[NVSYNC]; | |
48 | ||
49 | void __init | |
50 | xfs_ioend_init(void) | |
51 | { | |
52 | int i; | |
53 | ||
54 | for (i = 0; i < NVSYNC; i++) | |
55 | init_waitqueue_head(&xfs_ioend_wq[i]); | |
56 | } | |
57 | ||
58 | void | |
59 | xfs_ioend_wait( | |
60 | xfs_inode_t *ip) | |
61 | { | |
62 | wait_queue_head_t *wq = to_ioend_wq(ip); | |
63 | ||
64 | wait_event(*wq, (atomic_read(&ip->i_iocount) == 0)); | |
65 | } | |
66 | ||
67 | STATIC void | |
68 | xfs_ioend_wake( | |
69 | xfs_inode_t *ip) | |
70 | { | |
71 | if (atomic_dec_and_test(&ip->i_iocount)) | |
72 | wake_up(to_ioend_wq(ip)); | |
73 | } | |
74 | ||
0b1b213f | 75 | void |
f51623b2 NS |
76 | xfs_count_page_state( |
77 | struct page *page, | |
78 | int *delalloc, | |
f51623b2 NS |
79 | int *unwritten) |
80 | { | |
81 | struct buffer_head *bh, *head; | |
82 | ||
20cb52eb | 83 | *delalloc = *unwritten = 0; |
f51623b2 NS |
84 | |
85 | bh = head = page_buffers(page); | |
86 | do { | |
20cb52eb | 87 | if (buffer_unwritten(bh)) |
f51623b2 NS |
88 | (*unwritten) = 1; |
89 | else if (buffer_delay(bh)) | |
90 | (*delalloc) = 1; | |
91 | } while ((bh = bh->b_this_page) != head); | |
92 | } | |
93 | ||
6214ed44 CH |
94 | STATIC struct block_device * |
95 | xfs_find_bdev_for_inode( | |
046f1685 | 96 | struct inode *inode) |
6214ed44 | 97 | { |
046f1685 | 98 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
99 | struct xfs_mount *mp = ip->i_mount; |
100 | ||
71ddabb9 | 101 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
102 | return mp->m_rtdev_targp->bt_bdev; |
103 | else | |
104 | return mp->m_ddev_targp->bt_bdev; | |
105 | } | |
106 | ||
f6d6d4fc CH |
107 | /* |
108 | * We're now finished for good with this ioend structure. | |
109 | * Update the page state via the associated buffer_heads, | |
110 | * release holds on the inode and bio, and finally free | |
111 | * up memory. Do not use the ioend after this. | |
112 | */ | |
0829c360 CH |
113 | STATIC void |
114 | xfs_destroy_ioend( | |
115 | xfs_ioend_t *ioend) | |
116 | { | |
f6d6d4fc | 117 | struct buffer_head *bh, *next; |
583fa586 | 118 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
f6d6d4fc CH |
119 | |
120 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
121 | next = bh->b_private; | |
7d04a335 | 122 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 123 | } |
583fa586 CH |
124 | |
125 | /* | |
126 | * Volume managers supporting multiple paths can send back ENODEV | |
127 | * when the final path disappears. In this case continuing to fill | |
128 | * the page cache with dirty data which cannot be written out is | |
129 | * evil, so prevent that. | |
130 | */ | |
131 | if (unlikely(ioend->io_error == -ENODEV)) { | |
132 | xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ, | |
133 | __FILE__, __LINE__); | |
b677c210 | 134 | } |
583fa586 | 135 | |
25e41b3d | 136 | xfs_ioend_wake(ip); |
0829c360 CH |
137 | mempool_free(ioend, xfs_ioend_pool); |
138 | } | |
139 | ||
932640e8 DC |
140 | /* |
141 | * If the end of the current ioend is beyond the current EOF, | |
142 | * return the new EOF value, otherwise zero. | |
143 | */ | |
144 | STATIC xfs_fsize_t | |
145 | xfs_ioend_new_eof( | |
146 | xfs_ioend_t *ioend) | |
147 | { | |
148 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
149 | xfs_fsize_t isize; | |
150 | xfs_fsize_t bsize; | |
151 | ||
152 | bsize = ioend->io_offset + ioend->io_size; | |
153 | isize = MAX(ip->i_size, ip->i_new_size); | |
154 | isize = MIN(isize, bsize); | |
155 | return isize > ip->i_d.di_size ? isize : 0; | |
156 | } | |
157 | ||
ba87ea69 | 158 | /* |
77d7a0c2 DC |
159 | * Update on-disk file size now that data has been written to disk. The |
160 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
161 | * will be the intended file size until i_size is updated. If this write does | |
162 | * not extend all the way to the valid file size then restrict this update to | |
163 | * the end of the write. | |
164 | * | |
165 | * This function does not block as blocking on the inode lock in IO completion | |
166 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
167 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 168 | */ |
77d7a0c2 | 169 | STATIC int |
ba87ea69 LM |
170 | xfs_setfilesize( |
171 | xfs_ioend_t *ioend) | |
172 | { | |
b677c210 | 173 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 174 | xfs_fsize_t isize; |
ba87ea69 | 175 | |
ba87ea69 | 176 | if (unlikely(ioend->io_error)) |
77d7a0c2 DC |
177 | return 0; |
178 | ||
179 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
180 | return EAGAIN; | |
ba87ea69 | 181 | |
932640e8 DC |
182 | isize = xfs_ioend_new_eof(ioend); |
183 | if (isize) { | |
ba87ea69 | 184 | ip->i_d.di_size = isize; |
66d834ea | 185 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
186 | } |
187 | ||
188 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
189 | return 0; |
190 | } | |
191 | ||
192 | /* | |
209fb87a | 193 | * Schedule IO completion handling on the final put of an ioend. |
77d7a0c2 DC |
194 | */ |
195 | STATIC void | |
196 | xfs_finish_ioend( | |
209fb87a | 197 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
198 | { |
199 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
209fb87a CH |
200 | if (ioend->io_type == IO_UNWRITTEN) |
201 | queue_work(xfsconvertd_workqueue, &ioend->io_work); | |
202 | else | |
203 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
77d7a0c2 | 204 | } |
ba87ea69 LM |
205 | } |
206 | ||
0829c360 | 207 | /* |
5ec4fabb | 208 | * IO write completion. |
f6d6d4fc CH |
209 | */ |
210 | STATIC void | |
5ec4fabb | 211 | xfs_end_io( |
77d7a0c2 | 212 | struct work_struct *work) |
0829c360 | 213 | { |
77d7a0c2 DC |
214 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
215 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 216 | int error = 0; |
ba87ea69 | 217 | |
5ec4fabb CH |
218 | /* |
219 | * For unwritten extents we need to issue transactions to convert a | |
220 | * range to normal written extens after the data I/O has finished. | |
221 | */ | |
34a52c6c | 222 | if (ioend->io_type == IO_UNWRITTEN && |
5ec4fabb | 223 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { |
5ec4fabb CH |
224 | |
225 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
226 | ioend->io_size); | |
227 | if (error) | |
228 | ioend->io_error = error; | |
229 | } | |
ba87ea69 | 230 | |
5ec4fabb CH |
231 | /* |
232 | * We might have to update the on-disk file size after extending | |
233 | * writes. | |
234 | */ | |
a206c817 CH |
235 | error = xfs_setfilesize(ioend); |
236 | ASSERT(!error || error == EAGAIN); | |
77d7a0c2 DC |
237 | |
238 | /* | |
239 | * If we didn't complete processing of the ioend, requeue it to the | |
240 | * tail of the workqueue for another attempt later. Otherwise destroy | |
241 | * it. | |
242 | */ | |
243 | if (error == EAGAIN) { | |
244 | atomic_inc(&ioend->io_remaining); | |
209fb87a | 245 | xfs_finish_ioend(ioend); |
77d7a0c2 DC |
246 | /* ensure we don't spin on blocked ioends */ |
247 | delay(1); | |
fb511f21 CH |
248 | } else { |
249 | if (ioend->io_iocb) | |
250 | aio_complete(ioend->io_iocb, ioend->io_result, 0); | |
77d7a0c2 | 251 | xfs_destroy_ioend(ioend); |
fb511f21 | 252 | } |
c626d174 DC |
253 | } |
254 | ||
209fb87a CH |
255 | /* |
256 | * Call IO completion handling in caller context on the final put of an ioend. | |
257 | */ | |
258 | STATIC void | |
259 | xfs_finish_ioend_sync( | |
260 | struct xfs_ioend *ioend) | |
261 | { | |
262 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
263 | xfs_end_io(&ioend->io_work); | |
264 | } | |
265 | ||
0829c360 CH |
266 | /* |
267 | * Allocate and initialise an IO completion structure. | |
268 | * We need to track unwritten extent write completion here initially. | |
269 | * We'll need to extend this for updating the ondisk inode size later | |
270 | * (vs. incore size). | |
271 | */ | |
272 | STATIC xfs_ioend_t * | |
273 | xfs_alloc_ioend( | |
f6d6d4fc CH |
274 | struct inode *inode, |
275 | unsigned int type) | |
0829c360 CH |
276 | { |
277 | xfs_ioend_t *ioend; | |
278 | ||
279 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
280 | ||
281 | /* | |
282 | * Set the count to 1 initially, which will prevent an I/O | |
283 | * completion callback from happening before we have started | |
284 | * all the I/O from calling the completion routine too early. | |
285 | */ | |
286 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 287 | ioend->io_error = 0; |
f6d6d4fc CH |
288 | ioend->io_list = NULL; |
289 | ioend->io_type = type; | |
b677c210 | 290 | ioend->io_inode = inode; |
c1a073bd | 291 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 292 | ioend->io_buffer_tail = NULL; |
b677c210 | 293 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
294 | ioend->io_offset = 0; |
295 | ioend->io_size = 0; | |
fb511f21 CH |
296 | ioend->io_iocb = NULL; |
297 | ioend->io_result = 0; | |
0829c360 | 298 | |
5ec4fabb | 299 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
300 | return ioend; |
301 | } | |
302 | ||
1da177e4 LT |
303 | STATIC int |
304 | xfs_map_blocks( | |
305 | struct inode *inode, | |
306 | loff_t offset, | |
207d0416 | 307 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
308 | int type, |
309 | int nonblocking) | |
1da177e4 | 310 | { |
a206c817 CH |
311 | struct xfs_inode *ip = XFS_I(inode); |
312 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 313 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
314 | xfs_fileoff_t offset_fsb, end_fsb; |
315 | int error = 0; | |
a206c817 CH |
316 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
317 | int nimaps = 1; | |
318 | ||
319 | if (XFS_FORCED_SHUTDOWN(mp)) | |
320 | return -XFS_ERROR(EIO); | |
321 | ||
8ff2957d | 322 | if (type == IO_UNWRITTEN) |
a206c817 | 323 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
324 | |
325 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
326 | if (nonblocking) | |
327 | return -XFS_ERROR(EAGAIN); | |
328 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
a206c817 CH |
329 | } |
330 | ||
8ff2957d CH |
331 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
332 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
a206c817 | 333 | ASSERT(offset <= mp->m_maxioffset); |
8ff2957d | 334 | |
a206c817 CH |
335 | if (offset + count > mp->m_maxioffset) |
336 | count = mp->m_maxioffset - offset; | |
337 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
338 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
a206c817 CH |
339 | error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb, |
340 | bmapi_flags, NULL, 0, imap, &nimaps, NULL); | |
8ff2957d | 341 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 342 | |
8ff2957d CH |
343 | if (error) |
344 | return -XFS_ERROR(error); | |
a206c817 | 345 | |
8ff2957d CH |
346 | if (type == IO_DELALLOC && |
347 | (!nimaps || isnullstartblock(imap->br_startblock))) { | |
a206c817 CH |
348 | error = xfs_iomap_write_allocate(ip, offset, count, imap); |
349 | if (!error) | |
350 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
8ff2957d | 351 | return -XFS_ERROR(error); |
a206c817 CH |
352 | } |
353 | ||
8ff2957d CH |
354 | #ifdef DEBUG |
355 | if (type == IO_UNWRITTEN) { | |
356 | ASSERT(nimaps); | |
357 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
358 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
359 | } | |
360 | #endif | |
361 | if (nimaps) | |
362 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
363 | return 0; | |
1da177e4 LT |
364 | } |
365 | ||
b8f82a4a | 366 | STATIC int |
558e6891 | 367 | xfs_imap_valid( |
8699bb0a | 368 | struct inode *inode, |
207d0416 | 369 | struct xfs_bmbt_irec *imap, |
558e6891 | 370 | xfs_off_t offset) |
1da177e4 | 371 | { |
558e6891 | 372 | offset >>= inode->i_blkbits; |
8699bb0a | 373 | |
558e6891 CH |
374 | return offset >= imap->br_startoff && |
375 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
376 | } |
377 | ||
f6d6d4fc CH |
378 | /* |
379 | * BIO completion handler for buffered IO. | |
380 | */ | |
782e3b3b | 381 | STATIC void |
f6d6d4fc CH |
382 | xfs_end_bio( |
383 | struct bio *bio, | |
f6d6d4fc CH |
384 | int error) |
385 | { | |
386 | xfs_ioend_t *ioend = bio->bi_private; | |
387 | ||
f6d6d4fc | 388 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 389 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
390 | |
391 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
392 | bio->bi_private = NULL; |
393 | bio->bi_end_io = NULL; | |
f6d6d4fc | 394 | bio_put(bio); |
7d04a335 | 395 | |
209fb87a | 396 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
397 | } |
398 | ||
399 | STATIC void | |
400 | xfs_submit_ioend_bio( | |
06342cf8 CH |
401 | struct writeback_control *wbc, |
402 | xfs_ioend_t *ioend, | |
403 | struct bio *bio) | |
f6d6d4fc CH |
404 | { |
405 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
406 | bio->bi_private = ioend; |
407 | bio->bi_end_io = xfs_end_bio; | |
408 | ||
932640e8 DC |
409 | /* |
410 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
411 | * but don't update the inode size until I/O completion. | |
412 | */ | |
413 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 414 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 415 | |
721a9602 | 416 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
417 | } |
418 | ||
419 | STATIC struct bio * | |
420 | xfs_alloc_ioend_bio( | |
421 | struct buffer_head *bh) | |
422 | { | |
f6d6d4fc | 423 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 424 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
425 | |
426 | ASSERT(bio->bi_private == NULL); | |
427 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
428 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
429 | return bio; |
430 | } | |
431 | ||
432 | STATIC void | |
433 | xfs_start_buffer_writeback( | |
434 | struct buffer_head *bh) | |
435 | { | |
436 | ASSERT(buffer_mapped(bh)); | |
437 | ASSERT(buffer_locked(bh)); | |
438 | ASSERT(!buffer_delay(bh)); | |
439 | ASSERT(!buffer_unwritten(bh)); | |
440 | ||
441 | mark_buffer_async_write(bh); | |
442 | set_buffer_uptodate(bh); | |
443 | clear_buffer_dirty(bh); | |
444 | } | |
445 | ||
446 | STATIC void | |
447 | xfs_start_page_writeback( | |
448 | struct page *page, | |
f6d6d4fc CH |
449 | int clear_dirty, |
450 | int buffers) | |
451 | { | |
452 | ASSERT(PageLocked(page)); | |
453 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 454 | if (clear_dirty) |
92132021 DC |
455 | clear_page_dirty_for_io(page); |
456 | set_page_writeback(page); | |
f6d6d4fc | 457 | unlock_page(page); |
1f7decf6 FW |
458 | /* If no buffers on the page are to be written, finish it here */ |
459 | if (!buffers) | |
f6d6d4fc | 460 | end_page_writeback(page); |
f6d6d4fc CH |
461 | } |
462 | ||
463 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
464 | { | |
465 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
466 | } | |
467 | ||
468 | /* | |
d88992f6 DC |
469 | * Submit all of the bios for all of the ioends we have saved up, covering the |
470 | * initial writepage page and also any probed pages. | |
471 | * | |
472 | * Because we may have multiple ioends spanning a page, we need to start | |
473 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
474 | * buffers as we got, then we can end up with a page that only has buffers | |
475 | * marked async write and I/O complete on can occur before we mark the other | |
476 | * buffers async write. | |
477 | * | |
478 | * The end result of this is that we trip a bug in end_page_writeback() because | |
479 | * we call it twice for the one page as the code in end_buffer_async_write() | |
480 | * assumes that all buffers on the page are started at the same time. | |
481 | * | |
482 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 483 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
484 | */ |
485 | STATIC void | |
486 | xfs_submit_ioend( | |
06342cf8 | 487 | struct writeback_control *wbc, |
f6d6d4fc CH |
488 | xfs_ioend_t *ioend) |
489 | { | |
d88992f6 | 490 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
491 | xfs_ioend_t *next; |
492 | struct buffer_head *bh; | |
493 | struct bio *bio; | |
494 | sector_t lastblock = 0; | |
495 | ||
d88992f6 DC |
496 | /* Pass 1 - start writeback */ |
497 | do { | |
498 | next = ioend->io_list; | |
221cb251 | 499 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 500 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
501 | } while ((ioend = next) != NULL); |
502 | ||
503 | /* Pass 2 - submit I/O */ | |
504 | ioend = head; | |
f6d6d4fc CH |
505 | do { |
506 | next = ioend->io_list; | |
507 | bio = NULL; | |
508 | ||
509 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
510 | |
511 | if (!bio) { | |
512 | retry: | |
513 | bio = xfs_alloc_ioend_bio(bh); | |
514 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 515 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
516 | goto retry; |
517 | } | |
518 | ||
519 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 520 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
521 | goto retry; |
522 | } | |
523 | ||
524 | lastblock = bh->b_blocknr; | |
525 | } | |
526 | if (bio) | |
06342cf8 | 527 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 528 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
529 | } while ((ioend = next) != NULL); |
530 | } | |
531 | ||
532 | /* | |
533 | * Cancel submission of all buffer_heads so far in this endio. | |
534 | * Toss the endio too. Only ever called for the initial page | |
535 | * in a writepage request, so only ever one page. | |
536 | */ | |
537 | STATIC void | |
538 | xfs_cancel_ioend( | |
539 | xfs_ioend_t *ioend) | |
540 | { | |
541 | xfs_ioend_t *next; | |
542 | struct buffer_head *bh, *next_bh; | |
543 | ||
544 | do { | |
545 | next = ioend->io_list; | |
546 | bh = ioend->io_buffer_head; | |
547 | do { | |
548 | next_bh = bh->b_private; | |
549 | clear_buffer_async_write(bh); | |
550 | unlock_buffer(bh); | |
551 | } while ((bh = next_bh) != NULL); | |
552 | ||
25e41b3d | 553 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
554 | mempool_free(ioend, xfs_ioend_pool); |
555 | } while ((ioend = next) != NULL); | |
556 | } | |
557 | ||
558 | /* | |
559 | * Test to see if we've been building up a completion structure for | |
560 | * earlier buffers -- if so, we try to append to this ioend if we | |
561 | * can, otherwise we finish off any current ioend and start another. | |
562 | * Return true if we've finished the given ioend. | |
563 | */ | |
564 | STATIC void | |
565 | xfs_add_to_ioend( | |
566 | struct inode *inode, | |
567 | struct buffer_head *bh, | |
7336cea8 | 568 | xfs_off_t offset, |
f6d6d4fc CH |
569 | unsigned int type, |
570 | xfs_ioend_t **result, | |
571 | int need_ioend) | |
572 | { | |
573 | xfs_ioend_t *ioend = *result; | |
574 | ||
575 | if (!ioend || need_ioend || type != ioend->io_type) { | |
576 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 577 | |
f6d6d4fc CH |
578 | ioend = xfs_alloc_ioend(inode, type); |
579 | ioend->io_offset = offset; | |
580 | ioend->io_buffer_head = bh; | |
581 | ioend->io_buffer_tail = bh; | |
582 | if (previous) | |
583 | previous->io_list = ioend; | |
584 | *result = ioend; | |
585 | } else { | |
586 | ioend->io_buffer_tail->b_private = bh; | |
587 | ioend->io_buffer_tail = bh; | |
588 | } | |
589 | ||
590 | bh->b_private = NULL; | |
591 | ioend->io_size += bh->b_size; | |
592 | } | |
593 | ||
87cbc49c NS |
594 | STATIC void |
595 | xfs_map_buffer( | |
046f1685 | 596 | struct inode *inode, |
87cbc49c | 597 | struct buffer_head *bh, |
207d0416 | 598 | struct xfs_bmbt_irec *imap, |
046f1685 | 599 | xfs_off_t offset) |
87cbc49c NS |
600 | { |
601 | sector_t bn; | |
8699bb0a | 602 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
603 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
604 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 605 | |
207d0416 CH |
606 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
607 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 608 | |
e513182d | 609 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 610 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 611 | |
046f1685 | 612 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
613 | |
614 | bh->b_blocknr = bn; | |
615 | set_buffer_mapped(bh); | |
616 | } | |
617 | ||
1da177e4 LT |
618 | STATIC void |
619 | xfs_map_at_offset( | |
046f1685 | 620 | struct inode *inode, |
1da177e4 | 621 | struct buffer_head *bh, |
207d0416 | 622 | struct xfs_bmbt_irec *imap, |
046f1685 | 623 | xfs_off_t offset) |
1da177e4 | 624 | { |
207d0416 CH |
625 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
626 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 627 | |
207d0416 | 628 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
629 | set_buffer_mapped(bh); |
630 | clear_buffer_delay(bh); | |
f6d6d4fc | 631 | clear_buffer_unwritten(bh); |
1da177e4 LT |
632 | } |
633 | ||
1da177e4 | 634 | /* |
10ce4444 CH |
635 | * Test if a given page is suitable for writing as part of an unwritten |
636 | * or delayed allocate extent. | |
1da177e4 | 637 | */ |
10ce4444 CH |
638 | STATIC int |
639 | xfs_is_delayed_page( | |
640 | struct page *page, | |
f6d6d4fc | 641 | unsigned int type) |
1da177e4 | 642 | { |
1da177e4 | 643 | if (PageWriteback(page)) |
10ce4444 | 644 | return 0; |
1da177e4 LT |
645 | |
646 | if (page->mapping && page_has_buffers(page)) { | |
647 | struct buffer_head *bh, *head; | |
648 | int acceptable = 0; | |
649 | ||
650 | bh = head = page_buffers(page); | |
651 | do { | |
f6d6d4fc | 652 | if (buffer_unwritten(bh)) |
34a52c6c | 653 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 654 | else if (buffer_delay(bh)) |
a206c817 | 655 | acceptable = (type == IO_DELALLOC); |
2ddee844 | 656 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
a206c817 | 657 | acceptable = (type == IO_OVERWRITE); |
f6d6d4fc | 658 | else |
1da177e4 | 659 | break; |
1da177e4 LT |
660 | } while ((bh = bh->b_this_page) != head); |
661 | ||
662 | if (acceptable) | |
10ce4444 | 663 | return 1; |
1da177e4 LT |
664 | } |
665 | ||
10ce4444 | 666 | return 0; |
1da177e4 LT |
667 | } |
668 | ||
1da177e4 LT |
669 | /* |
670 | * Allocate & map buffers for page given the extent map. Write it out. | |
671 | * except for the original page of a writepage, this is called on | |
672 | * delalloc/unwritten pages only, for the original page it is possible | |
673 | * that the page has no mapping at all. | |
674 | */ | |
f6d6d4fc | 675 | STATIC int |
1da177e4 LT |
676 | xfs_convert_page( |
677 | struct inode *inode, | |
678 | struct page *page, | |
10ce4444 | 679 | loff_t tindex, |
207d0416 | 680 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 681 | xfs_ioend_t **ioendp, |
2fa24f92 | 682 | struct writeback_control *wbc) |
1da177e4 | 683 | { |
f6d6d4fc | 684 | struct buffer_head *bh, *head; |
9260dc6b CH |
685 | xfs_off_t end_offset; |
686 | unsigned long p_offset; | |
f6d6d4fc | 687 | unsigned int type; |
24e17b5f | 688 | int len, page_dirty; |
f6d6d4fc | 689 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 690 | xfs_off_t offset = page_offset(page); |
1da177e4 | 691 | |
10ce4444 CH |
692 | if (page->index != tindex) |
693 | goto fail; | |
529ae9aa | 694 | if (!trylock_page(page)) |
10ce4444 CH |
695 | goto fail; |
696 | if (PageWriteback(page)) | |
697 | goto fail_unlock_page; | |
698 | if (page->mapping != inode->i_mapping) | |
699 | goto fail_unlock_page; | |
700 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
701 | goto fail_unlock_page; | |
702 | ||
24e17b5f NS |
703 | /* |
704 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 705 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
706 | * |
707 | * Derivation: | |
708 | * | |
709 | * End offset is the highest offset that this page should represent. | |
710 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
711 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
712 | * hence give us the correct page_dirty count. On any other page, | |
713 | * it will be zero and in that case we need page_dirty to be the | |
714 | * count of buffers on the page. | |
24e17b5f | 715 | */ |
9260dc6b CH |
716 | end_offset = min_t(unsigned long long, |
717 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
718 | i_size_read(inode)); | |
719 | ||
24e17b5f | 720 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
721 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
722 | PAGE_CACHE_SIZE); | |
723 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
724 | page_dirty = p_offset / len; | |
24e17b5f | 725 | |
1da177e4 LT |
726 | bh = head = page_buffers(page); |
727 | do { | |
9260dc6b | 728 | if (offset >= end_offset) |
1da177e4 | 729 | break; |
f6d6d4fc CH |
730 | if (!buffer_uptodate(bh)) |
731 | uptodate = 0; | |
732 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
733 | done = 1; | |
1da177e4 | 734 | continue; |
f6d6d4fc CH |
735 | } |
736 | ||
2fa24f92 CH |
737 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
738 | buffer_mapped(bh)) { | |
9260dc6b | 739 | if (buffer_unwritten(bh)) |
34a52c6c | 740 | type = IO_UNWRITTEN; |
2fa24f92 | 741 | else if (buffer_delay(bh)) |
a206c817 | 742 | type = IO_DELALLOC; |
2fa24f92 CH |
743 | else |
744 | type = IO_OVERWRITE; | |
9260dc6b | 745 | |
558e6891 | 746 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 747 | done = 1; |
9260dc6b CH |
748 | continue; |
749 | } | |
750 | ||
ecff71e6 CH |
751 | lock_buffer(bh); |
752 | if (type != IO_OVERWRITE) | |
2fa24f92 | 753 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
754 | xfs_add_to_ioend(inode, bh, offset, type, |
755 | ioendp, done); | |
756 | ||
9260dc6b CH |
757 | page_dirty--; |
758 | count++; | |
759 | } else { | |
2fa24f92 | 760 | done = 1; |
1da177e4 | 761 | } |
7336cea8 | 762 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 763 | |
f6d6d4fc CH |
764 | if (uptodate && bh == head) |
765 | SetPageUptodate(page); | |
766 | ||
89f3b363 | 767 | if (count) { |
efceab1d DC |
768 | if (--wbc->nr_to_write <= 0 && |
769 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 770 | done = 1; |
1da177e4 | 771 | } |
89f3b363 | 772 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
773 | |
774 | return done; | |
10ce4444 CH |
775 | fail_unlock_page: |
776 | unlock_page(page); | |
777 | fail: | |
778 | return 1; | |
1da177e4 LT |
779 | } |
780 | ||
781 | /* | |
782 | * Convert & write out a cluster of pages in the same extent as defined | |
783 | * by mp and following the start page. | |
784 | */ | |
785 | STATIC void | |
786 | xfs_cluster_write( | |
787 | struct inode *inode, | |
788 | pgoff_t tindex, | |
207d0416 | 789 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 790 | xfs_ioend_t **ioendp, |
1da177e4 | 791 | struct writeback_control *wbc, |
1da177e4 LT |
792 | pgoff_t tlast) |
793 | { | |
10ce4444 CH |
794 | struct pagevec pvec; |
795 | int done = 0, i; | |
1da177e4 | 796 | |
10ce4444 CH |
797 | pagevec_init(&pvec, 0); |
798 | while (!done && tindex <= tlast) { | |
799 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
800 | ||
801 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 802 | break; |
10ce4444 CH |
803 | |
804 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
805 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 806 | imap, ioendp, wbc); |
10ce4444 CH |
807 | if (done) |
808 | break; | |
809 | } | |
810 | ||
811 | pagevec_release(&pvec); | |
812 | cond_resched(); | |
1da177e4 LT |
813 | } |
814 | } | |
815 | ||
3ed3a434 DC |
816 | STATIC void |
817 | xfs_vm_invalidatepage( | |
818 | struct page *page, | |
819 | unsigned long offset) | |
820 | { | |
821 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
822 | block_invalidatepage(page, offset); | |
823 | } | |
824 | ||
825 | /* | |
826 | * If the page has delalloc buffers on it, we need to punch them out before we | |
827 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
828 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
829 | * is done on that same region - the delalloc extent is returned when none is | |
830 | * supposed to be there. | |
831 | * | |
832 | * We prevent this by truncating away the delalloc regions on the page before | |
833 | * invalidating it. Because they are delalloc, we can do this without needing a | |
834 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
835 | * truncation without a transaction as there is no space left for block | |
836 | * reservation (typically why we see a ENOSPC in writeback). | |
837 | * | |
838 | * This is not a performance critical path, so for now just do the punching a | |
839 | * buffer head at a time. | |
840 | */ | |
841 | STATIC void | |
842 | xfs_aops_discard_page( | |
843 | struct page *page) | |
844 | { | |
845 | struct inode *inode = page->mapping->host; | |
846 | struct xfs_inode *ip = XFS_I(inode); | |
847 | struct buffer_head *bh, *head; | |
848 | loff_t offset = page_offset(page); | |
3ed3a434 | 849 | |
a206c817 | 850 | if (!xfs_is_delayed_page(page, IO_DELALLOC)) |
3ed3a434 DC |
851 | goto out_invalidate; |
852 | ||
e8c3753c DC |
853 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
854 | goto out_invalidate; | |
855 | ||
4f10700a | 856 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
857 | "page discard on page %p, inode 0x%llx, offset %llu.", |
858 | page, ip->i_ino, offset); | |
859 | ||
860 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
861 | bh = head = page_buffers(page); | |
862 | do { | |
3ed3a434 | 863 | int error; |
c726de44 | 864 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
865 | |
866 | if (!buffer_delay(bh)) | |
867 | goto next_buffer; | |
868 | ||
c726de44 DC |
869 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
870 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
871 | if (error) { |
872 | /* something screwed, just bail */ | |
e8c3753c | 873 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 874 | xfs_alert(ip->i_mount, |
3ed3a434 | 875 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 876 | } |
3ed3a434 DC |
877 | break; |
878 | } | |
879 | next_buffer: | |
c726de44 | 880 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
881 | |
882 | } while ((bh = bh->b_this_page) != head); | |
883 | ||
884 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
885 | out_invalidate: | |
886 | xfs_vm_invalidatepage(page, 0); | |
887 | return; | |
888 | } | |
889 | ||
1da177e4 | 890 | /* |
89f3b363 CH |
891 | * Write out a dirty page. |
892 | * | |
893 | * For delalloc space on the page we need to allocate space and flush it. | |
894 | * For unwritten space on the page we need to start the conversion to | |
895 | * regular allocated space. | |
89f3b363 | 896 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 897 | */ |
1da177e4 | 898 | STATIC int |
89f3b363 CH |
899 | xfs_vm_writepage( |
900 | struct page *page, | |
901 | struct writeback_control *wbc) | |
1da177e4 | 902 | { |
89f3b363 | 903 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 904 | struct buffer_head *bh, *head; |
207d0416 | 905 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 906 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 907 | loff_t offset; |
f6d6d4fc | 908 | unsigned int type; |
1da177e4 | 909 | __uint64_t end_offset; |
bd1556a1 | 910 | pgoff_t end_index, last_index; |
ed1e7b7e | 911 | ssize_t len; |
a206c817 | 912 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 913 | int count = 0; |
a206c817 | 914 | int nonblocking = 0; |
89f3b363 CH |
915 | |
916 | trace_xfs_writepage(inode, page, 0); | |
917 | ||
20cb52eb CH |
918 | ASSERT(page_has_buffers(page)); |
919 | ||
89f3b363 CH |
920 | /* |
921 | * Refuse to write the page out if we are called from reclaim context. | |
922 | * | |
d4f7a5cb CH |
923 | * This avoids stack overflows when called from deeply used stacks in |
924 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
925 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 CH |
926 | * |
927 | * This should really be done by the core VM, but until that happens | |
928 | * filesystems like XFS, btrfs and ext4 have to take care of this | |
929 | * by themselves. | |
930 | */ | |
d4f7a5cb | 931 | if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC) |
b5420f23 | 932 | goto redirty; |
1da177e4 | 933 | |
89f3b363 | 934 | /* |
680a647b CH |
935 | * Given that we do not allow direct reclaim to call us, we should |
936 | * never be called while in a filesystem transaction. | |
89f3b363 | 937 | */ |
680a647b | 938 | if (WARN_ON(current->flags & PF_FSTRANS)) |
b5420f23 | 939 | goto redirty; |
89f3b363 | 940 | |
1da177e4 LT |
941 | /* Is this page beyond the end of the file? */ |
942 | offset = i_size_read(inode); | |
943 | end_index = offset >> PAGE_CACHE_SHIFT; | |
944 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
945 | if (page->index >= end_index) { | |
946 | if ((page->index >= end_index + 1) || | |
947 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 948 | unlock_page(page); |
19d5bcf3 | 949 | return 0; |
1da177e4 LT |
950 | } |
951 | } | |
952 | ||
f6d6d4fc | 953 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
954 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
955 | offset); | |
24e17b5f | 956 | len = 1 << inode->i_blkbits; |
24e17b5f | 957 | |
24e17b5f | 958 | bh = head = page_buffers(page); |
f6d6d4fc | 959 | offset = page_offset(page); |
a206c817 CH |
960 | type = IO_OVERWRITE; |
961 | ||
dbcdde3e | 962 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 963 | nonblocking = 1; |
f6d6d4fc | 964 | |
1da177e4 | 965 | do { |
6ac7248e CH |
966 | int new_ioend = 0; |
967 | ||
1da177e4 LT |
968 | if (offset >= end_offset) |
969 | break; | |
970 | if (!buffer_uptodate(bh)) | |
971 | uptodate = 0; | |
1da177e4 | 972 | |
3d9b02e3 | 973 | /* |
ece413f5 CH |
974 | * set_page_dirty dirties all buffers in a page, independent |
975 | * of their state. The dirty state however is entirely | |
976 | * meaningless for holes (!mapped && uptodate), so skip | |
977 | * buffers covering holes here. | |
3d9b02e3 ES |
978 | */ |
979 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
980 | imap_valid = 0; |
981 | continue; | |
982 | } | |
983 | ||
aeea1b1f CH |
984 | if (buffer_unwritten(bh)) { |
985 | if (type != IO_UNWRITTEN) { | |
986 | type = IO_UNWRITTEN; | |
987 | imap_valid = 0; | |
1da177e4 | 988 | } |
aeea1b1f CH |
989 | } else if (buffer_delay(bh)) { |
990 | if (type != IO_DELALLOC) { | |
991 | type = IO_DELALLOC; | |
992 | imap_valid = 0; | |
1da177e4 | 993 | } |
89f3b363 | 994 | } else if (buffer_uptodate(bh)) { |
a206c817 CH |
995 | if (type != IO_OVERWRITE) { |
996 | type = IO_OVERWRITE; | |
85da94c6 CH |
997 | imap_valid = 0; |
998 | } | |
aeea1b1f CH |
999 | } else { |
1000 | if (PageUptodate(page)) { | |
1001 | ASSERT(buffer_mapped(bh)); | |
1002 | imap_valid = 0; | |
6c4fe19f | 1003 | } |
aeea1b1f CH |
1004 | continue; |
1005 | } | |
d5cb48aa | 1006 | |
aeea1b1f CH |
1007 | if (imap_valid) |
1008 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1009 | if (!imap_valid) { | |
1010 | /* | |
1011 | * If we didn't have a valid mapping then we need to | |
1012 | * put the new mapping into a separate ioend structure. | |
1013 | * This ensures non-contiguous extents always have | |
1014 | * separate ioends, which is particularly important | |
1015 | * for unwritten extent conversion at I/O completion | |
1016 | * time. | |
1017 | */ | |
1018 | new_ioend = 1; | |
1019 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1020 | nonblocking); | |
1021 | if (err) | |
1022 | goto error; | |
1023 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1024 | } | |
1025 | if (imap_valid) { | |
ecff71e6 CH |
1026 | lock_buffer(bh); |
1027 | if (type != IO_OVERWRITE) | |
aeea1b1f CH |
1028 | xfs_map_at_offset(inode, bh, &imap, offset); |
1029 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1030 | new_ioend); | |
1031 | count++; | |
1da177e4 | 1032 | } |
f6d6d4fc CH |
1033 | |
1034 | if (!iohead) | |
1035 | iohead = ioend; | |
1036 | ||
1037 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1038 | |
1039 | if (uptodate && bh == head) | |
1040 | SetPageUptodate(page); | |
1041 | ||
89f3b363 | 1042 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1043 | |
558e6891 | 1044 | if (ioend && imap_valid) { |
bd1556a1 CH |
1045 | xfs_off_t end_index; |
1046 | ||
1047 | end_index = imap.br_startoff + imap.br_blockcount; | |
1048 | ||
1049 | /* to bytes */ | |
1050 | end_index <<= inode->i_blkbits; | |
1051 | ||
1052 | /* to pages */ | |
1053 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1054 | ||
1055 | /* check against file size */ | |
1056 | if (end_index > last_index) | |
1057 | end_index = last_index; | |
8699bb0a | 1058 | |
207d0416 | 1059 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1060 | wbc, end_index); |
1da177e4 LT |
1061 | } |
1062 | ||
f6d6d4fc | 1063 | if (iohead) |
06342cf8 | 1064 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1065 | |
89f3b363 | 1066 | return 0; |
1da177e4 LT |
1067 | |
1068 | error: | |
f6d6d4fc CH |
1069 | if (iohead) |
1070 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1071 | |
b5420f23 CH |
1072 | if (err == -EAGAIN) |
1073 | goto redirty; | |
1074 | ||
20cb52eb | 1075 | xfs_aops_discard_page(page); |
89f3b363 CH |
1076 | ClearPageUptodate(page); |
1077 | unlock_page(page); | |
1da177e4 | 1078 | return err; |
f51623b2 | 1079 | |
b5420f23 | 1080 | redirty: |
f51623b2 NS |
1081 | redirty_page_for_writepage(wbc, page); |
1082 | unlock_page(page); | |
1083 | return 0; | |
f51623b2 NS |
1084 | } |
1085 | ||
7d4fb40a NS |
1086 | STATIC int |
1087 | xfs_vm_writepages( | |
1088 | struct address_space *mapping, | |
1089 | struct writeback_control *wbc) | |
1090 | { | |
b3aea4ed | 1091 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1092 | return generic_writepages(mapping, wbc); |
1093 | } | |
1094 | ||
f51623b2 NS |
1095 | /* |
1096 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1097 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1098 | * have buffer heads in this call. |
1099 | * | |
89f3b363 | 1100 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1101 | */ |
1102 | STATIC int | |
238f4c54 | 1103 | xfs_vm_releasepage( |
f51623b2 NS |
1104 | struct page *page, |
1105 | gfp_t gfp_mask) | |
1106 | { | |
20cb52eb | 1107 | int delalloc, unwritten; |
f51623b2 | 1108 | |
89f3b363 | 1109 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1110 | |
20cb52eb | 1111 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1112 | |
89f3b363 | 1113 | if (WARN_ON(delalloc)) |
f51623b2 | 1114 | return 0; |
89f3b363 | 1115 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1116 | return 0; |
1117 | ||
f51623b2 NS |
1118 | return try_to_free_buffers(page); |
1119 | } | |
1120 | ||
1da177e4 | 1121 | STATIC int |
c2536668 | 1122 | __xfs_get_blocks( |
1da177e4 LT |
1123 | struct inode *inode, |
1124 | sector_t iblock, | |
1da177e4 LT |
1125 | struct buffer_head *bh_result, |
1126 | int create, | |
f2bde9b8 | 1127 | int direct) |
1da177e4 | 1128 | { |
a206c817 CH |
1129 | struct xfs_inode *ip = XFS_I(inode); |
1130 | struct xfs_mount *mp = ip->i_mount; | |
1131 | xfs_fileoff_t offset_fsb, end_fsb; | |
1132 | int error = 0; | |
1133 | int lockmode = 0; | |
207d0416 | 1134 | struct xfs_bmbt_irec imap; |
a206c817 | 1135 | int nimaps = 1; |
fdc7ed75 NS |
1136 | xfs_off_t offset; |
1137 | ssize_t size; | |
207d0416 | 1138 | int new = 0; |
a206c817 CH |
1139 | |
1140 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1141 | return -XFS_ERROR(EIO); | |
1da177e4 | 1142 | |
fdc7ed75 | 1143 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1144 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1145 | size = bh_result->b_size; | |
364f358a LM |
1146 | |
1147 | if (!create && direct && offset >= i_size_read(inode)) | |
1148 | return 0; | |
1149 | ||
a206c817 CH |
1150 | if (create) { |
1151 | lockmode = XFS_ILOCK_EXCL; | |
1152 | xfs_ilock(ip, lockmode); | |
1153 | } else { | |
1154 | lockmode = xfs_ilock_map_shared(ip); | |
1155 | } | |
f2bde9b8 | 1156 | |
a206c817 CH |
1157 | ASSERT(offset <= mp->m_maxioffset); |
1158 | if (offset + size > mp->m_maxioffset) | |
1159 | size = mp->m_maxioffset - offset; | |
1160 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); | |
1161 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1162 | ||
1163 | error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb, | |
1164 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, &nimaps, NULL); | |
1da177e4 | 1165 | if (error) |
a206c817 CH |
1166 | goto out_unlock; |
1167 | ||
1168 | if (create && | |
1169 | (!nimaps || | |
1170 | (imap.br_startblock == HOLESTARTBLOCK || | |
1171 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
1172 | if (direct) { | |
1173 | error = xfs_iomap_write_direct(ip, offset, size, | |
1174 | &imap, nimaps); | |
1175 | } else { | |
1176 | error = xfs_iomap_write_delay(ip, offset, size, &imap); | |
1177 | } | |
1178 | if (error) | |
1179 | goto out_unlock; | |
1180 | ||
1181 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1182 | } else if (nimaps) { | |
1183 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
1184 | } else { | |
1185 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1186 | goto out_unlock; | |
1187 | } | |
1188 | xfs_iunlock(ip, lockmode); | |
1da177e4 | 1189 | |
207d0416 CH |
1190 | if (imap.br_startblock != HOLESTARTBLOCK && |
1191 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1192 | /* |
1193 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1194 | * the read case (treat as if we're reading into a hole). |
1195 | */ | |
207d0416 CH |
1196 | if (create || !ISUNWRITTEN(&imap)) |
1197 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1198 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1199 | if (direct) |
1200 | bh_result->b_private = inode; | |
1201 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1202 | } |
1203 | } | |
1204 | ||
c2536668 NS |
1205 | /* |
1206 | * If this is a realtime file, data may be on a different device. | |
1207 | * to that pointed to from the buffer_head b_bdev currently. | |
1208 | */ | |
046f1685 | 1209 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1210 | |
c2536668 | 1211 | /* |
549054af DC |
1212 | * If we previously allocated a block out beyond eof and we are now |
1213 | * coming back to use it then we will need to flag it as new even if it | |
1214 | * has a disk address. | |
1215 | * | |
1216 | * With sub-block writes into unwritten extents we also need to mark | |
1217 | * the buffer as new so that the unwritten parts of the buffer gets | |
1218 | * correctly zeroed. | |
1da177e4 LT |
1219 | */ |
1220 | if (create && | |
1221 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1222 | (offset >= i_size_read(inode)) || |
207d0416 | 1223 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1224 | set_buffer_new(bh_result); |
1da177e4 | 1225 | |
207d0416 | 1226 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1227 | BUG_ON(direct); |
1228 | if (create) { | |
1229 | set_buffer_uptodate(bh_result); | |
1230 | set_buffer_mapped(bh_result); | |
1231 | set_buffer_delay(bh_result); | |
1232 | } | |
1233 | } | |
1234 | ||
2b8f12b7 CH |
1235 | /* |
1236 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1237 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1238 | */ | |
c2536668 | 1239 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1240 | xfs_off_t mapping_size; |
1241 | ||
1242 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1243 | mapping_size <<= inode->i_blkbits; | |
1244 | ||
1245 | ASSERT(mapping_size > 0); | |
1246 | if (mapping_size > size) | |
1247 | mapping_size = size; | |
1248 | if (mapping_size > LONG_MAX) | |
1249 | mapping_size = LONG_MAX; | |
1250 | ||
1251 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1252 | } |
1253 | ||
1254 | return 0; | |
a206c817 CH |
1255 | |
1256 | out_unlock: | |
1257 | xfs_iunlock(ip, lockmode); | |
1258 | return -error; | |
1da177e4 LT |
1259 | } |
1260 | ||
1261 | int | |
c2536668 | 1262 | xfs_get_blocks( |
1da177e4 LT |
1263 | struct inode *inode, |
1264 | sector_t iblock, | |
1265 | struct buffer_head *bh_result, | |
1266 | int create) | |
1267 | { | |
f2bde9b8 | 1268 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1269 | } |
1270 | ||
1271 | STATIC int | |
e4c573bb | 1272 | xfs_get_blocks_direct( |
1da177e4 LT |
1273 | struct inode *inode, |
1274 | sector_t iblock, | |
1da177e4 LT |
1275 | struct buffer_head *bh_result, |
1276 | int create) | |
1277 | { | |
f2bde9b8 | 1278 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1279 | } |
1280 | ||
209fb87a CH |
1281 | /* |
1282 | * Complete a direct I/O write request. | |
1283 | * | |
1284 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1285 | * need to issue a transaction to convert the range from unwritten to written | |
1286 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1287 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1288 | * request this handler is called from interrupt context, from which we |
1289 | * can't start transactions. In that case offload the I/O completion to | |
1290 | * the workqueues we also use for buffered I/O completion. | |
1291 | */ | |
f0973863 | 1292 | STATIC void |
209fb87a CH |
1293 | xfs_end_io_direct_write( |
1294 | struct kiocb *iocb, | |
1295 | loff_t offset, | |
1296 | ssize_t size, | |
1297 | void *private, | |
1298 | int ret, | |
1299 | bool is_async) | |
f0973863 | 1300 | { |
209fb87a | 1301 | struct xfs_ioend *ioend = iocb->private; |
f0973863 CH |
1302 | |
1303 | /* | |
209fb87a CH |
1304 | * blockdev_direct_IO can return an error even after the I/O |
1305 | * completion handler was called. Thus we need to protect | |
1306 | * against double-freeing. | |
f0973863 | 1307 | */ |
209fb87a CH |
1308 | iocb->private = NULL; |
1309 | ||
ba87ea69 LM |
1310 | ioend->io_offset = offset; |
1311 | ioend->io_size = size; | |
209fb87a CH |
1312 | if (private && size > 0) |
1313 | ioend->io_type = IO_UNWRITTEN; | |
1314 | ||
1315 | if (is_async) { | |
1316 | /* | |
1317 | * If we are converting an unwritten extent we need to delay | |
1318 | * the AIO completion until after the unwrittent extent | |
1319 | * conversion has completed, otherwise do it ASAP. | |
1320 | */ | |
1321 | if (ioend->io_type == IO_UNWRITTEN) { | |
fb511f21 CH |
1322 | ioend->io_iocb = iocb; |
1323 | ioend->io_result = ret; | |
fb511f21 | 1324 | } else { |
209fb87a | 1325 | aio_complete(iocb, ret, 0); |
fb511f21 | 1326 | } |
209fb87a | 1327 | xfs_finish_ioend(ioend); |
f0973863 | 1328 | } else { |
209fb87a | 1329 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1330 | } |
f0973863 CH |
1331 | } |
1332 | ||
1da177e4 | 1333 | STATIC ssize_t |
e4c573bb | 1334 | xfs_vm_direct_IO( |
1da177e4 LT |
1335 | int rw, |
1336 | struct kiocb *iocb, | |
1337 | const struct iovec *iov, | |
1338 | loff_t offset, | |
1339 | unsigned long nr_segs) | |
1340 | { | |
209fb87a CH |
1341 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1342 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1343 | ssize_t ret; | |
1344 | ||
1345 | if (rw & WRITE) { | |
a206c817 | 1346 | iocb->private = xfs_alloc_ioend(inode, IO_DIRECT); |
209fb87a | 1347 | |
eafdc7d1 CH |
1348 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1349 | offset, nr_segs, | |
1350 | xfs_get_blocks_direct, | |
1351 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a CH |
1352 | if (ret != -EIOCBQUEUED && iocb->private) |
1353 | xfs_destroy_ioend(iocb->private); | |
1354 | } else { | |
eafdc7d1 CH |
1355 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1356 | offset, nr_segs, | |
1357 | xfs_get_blocks_direct, | |
1358 | NULL, NULL, 0); | |
209fb87a | 1359 | } |
f0973863 | 1360 | |
f0973863 | 1361 | return ret; |
1da177e4 LT |
1362 | } |
1363 | ||
fa9b227e CH |
1364 | STATIC void |
1365 | xfs_vm_write_failed( | |
1366 | struct address_space *mapping, | |
1367 | loff_t to) | |
1368 | { | |
1369 | struct inode *inode = mapping->host; | |
1370 | ||
1371 | if (to > inode->i_size) { | |
c726de44 DC |
1372 | /* |
1373 | * punch out the delalloc blocks we have already allocated. We | |
1374 | * don't call xfs_setattr() to do this as we may be in the | |
1375 | * middle of a multi-iovec write and so the vfs inode->i_size | |
1376 | * will not match the xfs ip->i_size and so it will zero too | |
1377 | * much. Hence we jus truncate the page cache to zero what is | |
1378 | * necessary and punch the delalloc blocks directly. | |
1379 | */ | |
1380 | struct xfs_inode *ip = XFS_I(inode); | |
1381 | xfs_fileoff_t start_fsb; | |
1382 | xfs_fileoff_t end_fsb; | |
1383 | int error; | |
1384 | ||
1385 | truncate_pagecache(inode, to, inode->i_size); | |
1386 | ||
1387 | /* | |
1388 | * Check if there are any blocks that are outside of i_size | |
1389 | * that need to be trimmed back. | |
1390 | */ | |
1391 | start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1; | |
1392 | end_fsb = XFS_B_TO_FSB(ip->i_mount, to); | |
1393 | if (end_fsb <= start_fsb) | |
1394 | return; | |
1395 | ||
1396 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1397 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1398 | end_fsb - start_fsb); | |
1399 | if (error) { | |
1400 | /* something screwed, just bail */ | |
1401 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
4f10700a | 1402 | xfs_alert(ip->i_mount, |
c726de44 DC |
1403 | "xfs_vm_write_failed: unable to clean up ino %lld", |
1404 | ip->i_ino); | |
1405 | } | |
1406 | } | |
1407 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
fa9b227e CH |
1408 | } |
1409 | } | |
1410 | ||
f51623b2 | 1411 | STATIC int |
d79689c7 | 1412 | xfs_vm_write_begin( |
f51623b2 | 1413 | struct file *file, |
d79689c7 NP |
1414 | struct address_space *mapping, |
1415 | loff_t pos, | |
1416 | unsigned len, | |
1417 | unsigned flags, | |
1418 | struct page **pagep, | |
1419 | void **fsdata) | |
f51623b2 | 1420 | { |
155130a4 CH |
1421 | int ret; |
1422 | ||
1423 | ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS, | |
1424 | pagep, xfs_get_blocks); | |
fa9b227e CH |
1425 | if (unlikely(ret)) |
1426 | xfs_vm_write_failed(mapping, pos + len); | |
1427 | return ret; | |
1428 | } | |
1429 | ||
1430 | STATIC int | |
1431 | xfs_vm_write_end( | |
1432 | struct file *file, | |
1433 | struct address_space *mapping, | |
1434 | loff_t pos, | |
1435 | unsigned len, | |
1436 | unsigned copied, | |
1437 | struct page *page, | |
1438 | void *fsdata) | |
1439 | { | |
1440 | int ret; | |
155130a4 | 1441 | |
fa9b227e CH |
1442 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
1443 | if (unlikely(ret < len)) | |
1444 | xfs_vm_write_failed(mapping, pos + len); | |
155130a4 | 1445 | return ret; |
f51623b2 | 1446 | } |
1da177e4 LT |
1447 | |
1448 | STATIC sector_t | |
e4c573bb | 1449 | xfs_vm_bmap( |
1da177e4 LT |
1450 | struct address_space *mapping, |
1451 | sector_t block) | |
1452 | { | |
1453 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1454 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1455 | |
cca28fb8 | 1456 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1457 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1458 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1459 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1460 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1461 | } |
1462 | ||
1463 | STATIC int | |
e4c573bb | 1464 | xfs_vm_readpage( |
1da177e4 LT |
1465 | struct file *unused, |
1466 | struct page *page) | |
1467 | { | |
c2536668 | 1468 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1469 | } |
1470 | ||
1471 | STATIC int | |
e4c573bb | 1472 | xfs_vm_readpages( |
1da177e4 LT |
1473 | struct file *unused, |
1474 | struct address_space *mapping, | |
1475 | struct list_head *pages, | |
1476 | unsigned nr_pages) | |
1477 | { | |
c2536668 | 1478 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1479 | } |
1480 | ||
f5e54d6e | 1481 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1482 | .readpage = xfs_vm_readpage, |
1483 | .readpages = xfs_vm_readpages, | |
1484 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1485 | .writepages = xfs_vm_writepages, |
238f4c54 NS |
1486 | .releasepage = xfs_vm_releasepage, |
1487 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1488 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1489 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1490 | .bmap = xfs_vm_bmap, |
1491 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1492 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1493 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1494 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1495 | }; |