]>
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 | 124 | |
c859cdd1 CH |
125 | if (ioend->io_iocb) { |
126 | if (ioend->io_isasync) | |
127 | aio_complete(ioend->io_iocb, ioend->io_result, 0); | |
128 | inode_dio_done(ioend->io_inode); | |
129 | } | |
25e41b3d | 130 | xfs_ioend_wake(ip); |
0829c360 CH |
131 | mempool_free(ioend, xfs_ioend_pool); |
132 | } | |
133 | ||
932640e8 DC |
134 | /* |
135 | * If the end of the current ioend is beyond the current EOF, | |
136 | * return the new EOF value, otherwise zero. | |
137 | */ | |
138 | STATIC xfs_fsize_t | |
139 | xfs_ioend_new_eof( | |
140 | xfs_ioend_t *ioend) | |
141 | { | |
142 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
143 | xfs_fsize_t isize; | |
144 | xfs_fsize_t bsize; | |
145 | ||
146 | bsize = ioend->io_offset + ioend->io_size; | |
147 | isize = MAX(ip->i_size, ip->i_new_size); | |
148 | isize = MIN(isize, bsize); | |
149 | return isize > ip->i_d.di_size ? isize : 0; | |
150 | } | |
151 | ||
ba87ea69 | 152 | /* |
77d7a0c2 DC |
153 | * Update on-disk file size now that data has been written to disk. The |
154 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
155 | * will be the intended file size until i_size is updated. If this write does | |
156 | * not extend all the way to the valid file size then restrict this update to | |
157 | * the end of the write. | |
158 | * | |
159 | * This function does not block as blocking on the inode lock in IO completion | |
160 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
161 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 162 | */ |
77d7a0c2 | 163 | STATIC int |
ba87ea69 LM |
164 | xfs_setfilesize( |
165 | xfs_ioend_t *ioend) | |
166 | { | |
b677c210 | 167 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 168 | xfs_fsize_t isize; |
ba87ea69 | 169 | |
ba87ea69 | 170 | if (unlikely(ioend->io_error)) |
77d7a0c2 DC |
171 | return 0; |
172 | ||
173 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) | |
174 | return EAGAIN; | |
ba87ea69 | 175 | |
932640e8 DC |
176 | isize = xfs_ioend_new_eof(ioend); |
177 | if (isize) { | |
55fb25d5 | 178 | trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); |
ba87ea69 | 179 | ip->i_d.di_size = isize; |
66d834ea | 180 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
181 | } |
182 | ||
183 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
184 | return 0; |
185 | } | |
186 | ||
187 | /* | |
209fb87a | 188 | * Schedule IO completion handling on the final put of an ioend. |
77d7a0c2 DC |
189 | */ |
190 | STATIC void | |
191 | xfs_finish_ioend( | |
209fb87a | 192 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
193 | { |
194 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
209fb87a CH |
195 | if (ioend->io_type == IO_UNWRITTEN) |
196 | queue_work(xfsconvertd_workqueue, &ioend->io_work); | |
197 | else | |
198 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
77d7a0c2 | 199 | } |
ba87ea69 LM |
200 | } |
201 | ||
0829c360 | 202 | /* |
5ec4fabb | 203 | * IO write completion. |
f6d6d4fc CH |
204 | */ |
205 | STATIC void | |
5ec4fabb | 206 | xfs_end_io( |
77d7a0c2 | 207 | struct work_struct *work) |
0829c360 | 208 | { |
77d7a0c2 DC |
209 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
210 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 211 | int error = 0; |
ba87ea69 | 212 | |
5ec4fabb CH |
213 | /* |
214 | * For unwritten extents we need to issue transactions to convert a | |
215 | * range to normal written extens after the data I/O has finished. | |
216 | */ | |
34a52c6c | 217 | if (ioend->io_type == IO_UNWRITTEN && |
5ec4fabb | 218 | likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) { |
5ec4fabb CH |
219 | |
220 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, | |
221 | ioend->io_size); | |
222 | if (error) | |
223 | ioend->io_error = error; | |
224 | } | |
ba87ea69 | 225 | |
5ec4fabb CH |
226 | /* |
227 | * We might have to update the on-disk file size after extending | |
228 | * writes. | |
229 | */ | |
a206c817 CH |
230 | error = xfs_setfilesize(ioend); |
231 | ASSERT(!error || error == EAGAIN); | |
77d7a0c2 DC |
232 | |
233 | /* | |
234 | * If we didn't complete processing of the ioend, requeue it to the | |
235 | * tail of the workqueue for another attempt later. Otherwise destroy | |
236 | * it. | |
237 | */ | |
238 | if (error == EAGAIN) { | |
239 | atomic_inc(&ioend->io_remaining); | |
209fb87a | 240 | xfs_finish_ioend(ioend); |
77d7a0c2 DC |
241 | /* ensure we don't spin on blocked ioends */ |
242 | delay(1); | |
fb511f21 | 243 | } else { |
77d7a0c2 | 244 | xfs_destroy_ioend(ioend); |
fb511f21 | 245 | } |
c626d174 DC |
246 | } |
247 | ||
209fb87a CH |
248 | /* |
249 | * Call IO completion handling in caller context on the final put of an ioend. | |
250 | */ | |
251 | STATIC void | |
252 | xfs_finish_ioend_sync( | |
253 | struct xfs_ioend *ioend) | |
254 | { | |
255 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
256 | xfs_end_io(&ioend->io_work); | |
257 | } | |
258 | ||
0829c360 CH |
259 | /* |
260 | * Allocate and initialise an IO completion structure. | |
261 | * We need to track unwritten extent write completion here initially. | |
262 | * We'll need to extend this for updating the ondisk inode size later | |
263 | * (vs. incore size). | |
264 | */ | |
265 | STATIC xfs_ioend_t * | |
266 | xfs_alloc_ioend( | |
f6d6d4fc CH |
267 | struct inode *inode, |
268 | unsigned int type) | |
0829c360 CH |
269 | { |
270 | xfs_ioend_t *ioend; | |
271 | ||
272 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
273 | ||
274 | /* | |
275 | * Set the count to 1 initially, which will prevent an I/O | |
276 | * completion callback from happening before we have started | |
277 | * all the I/O from calling the completion routine too early. | |
278 | */ | |
279 | atomic_set(&ioend->io_remaining, 1); | |
c859cdd1 | 280 | ioend->io_isasync = 0; |
7d04a335 | 281 | ioend->io_error = 0; |
f6d6d4fc CH |
282 | ioend->io_list = NULL; |
283 | ioend->io_type = type; | |
b677c210 | 284 | ioend->io_inode = inode; |
c1a073bd | 285 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 286 | ioend->io_buffer_tail = NULL; |
b677c210 | 287 | atomic_inc(&XFS_I(ioend->io_inode)->i_iocount); |
0829c360 CH |
288 | ioend->io_offset = 0; |
289 | ioend->io_size = 0; | |
fb511f21 CH |
290 | ioend->io_iocb = NULL; |
291 | ioend->io_result = 0; | |
0829c360 | 292 | |
5ec4fabb | 293 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
294 | return ioend; |
295 | } | |
296 | ||
1da177e4 LT |
297 | STATIC int |
298 | xfs_map_blocks( | |
299 | struct inode *inode, | |
300 | loff_t offset, | |
207d0416 | 301 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
302 | int type, |
303 | int nonblocking) | |
1da177e4 | 304 | { |
a206c817 CH |
305 | struct xfs_inode *ip = XFS_I(inode); |
306 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 307 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
308 | xfs_fileoff_t offset_fsb, end_fsb; |
309 | int error = 0; | |
a206c817 CH |
310 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
311 | int nimaps = 1; | |
312 | ||
313 | if (XFS_FORCED_SHUTDOWN(mp)) | |
314 | return -XFS_ERROR(EIO); | |
315 | ||
8ff2957d | 316 | if (type == IO_UNWRITTEN) |
a206c817 | 317 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
318 | |
319 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
320 | if (nonblocking) | |
321 | return -XFS_ERROR(EAGAIN); | |
322 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
a206c817 CH |
323 | } |
324 | ||
8ff2957d CH |
325 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
326 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
a206c817 | 327 | ASSERT(offset <= mp->m_maxioffset); |
8ff2957d | 328 | |
a206c817 CH |
329 | if (offset + count > mp->m_maxioffset) |
330 | count = mp->m_maxioffset - offset; | |
331 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
332 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
a206c817 CH |
333 | error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb, |
334 | bmapi_flags, NULL, 0, imap, &nimaps, NULL); | |
8ff2957d | 335 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 336 | |
8ff2957d CH |
337 | if (error) |
338 | return -XFS_ERROR(error); | |
a206c817 | 339 | |
8ff2957d CH |
340 | if (type == IO_DELALLOC && |
341 | (!nimaps || isnullstartblock(imap->br_startblock))) { | |
a206c817 CH |
342 | error = xfs_iomap_write_allocate(ip, offset, count, imap); |
343 | if (!error) | |
344 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
8ff2957d | 345 | return -XFS_ERROR(error); |
a206c817 CH |
346 | } |
347 | ||
8ff2957d CH |
348 | #ifdef DEBUG |
349 | if (type == IO_UNWRITTEN) { | |
350 | ASSERT(nimaps); | |
351 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
352 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
353 | } | |
354 | #endif | |
355 | if (nimaps) | |
356 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
357 | return 0; | |
1da177e4 LT |
358 | } |
359 | ||
b8f82a4a | 360 | STATIC int |
558e6891 | 361 | xfs_imap_valid( |
8699bb0a | 362 | struct inode *inode, |
207d0416 | 363 | struct xfs_bmbt_irec *imap, |
558e6891 | 364 | xfs_off_t offset) |
1da177e4 | 365 | { |
558e6891 | 366 | offset >>= inode->i_blkbits; |
8699bb0a | 367 | |
558e6891 CH |
368 | return offset >= imap->br_startoff && |
369 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
370 | } |
371 | ||
f6d6d4fc CH |
372 | /* |
373 | * BIO completion handler for buffered IO. | |
374 | */ | |
782e3b3b | 375 | STATIC void |
f6d6d4fc CH |
376 | xfs_end_bio( |
377 | struct bio *bio, | |
f6d6d4fc CH |
378 | int error) |
379 | { | |
380 | xfs_ioend_t *ioend = bio->bi_private; | |
381 | ||
f6d6d4fc | 382 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 383 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
384 | |
385 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
386 | bio->bi_private = NULL; |
387 | bio->bi_end_io = NULL; | |
f6d6d4fc | 388 | bio_put(bio); |
7d04a335 | 389 | |
209fb87a | 390 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
391 | } |
392 | ||
393 | STATIC void | |
394 | xfs_submit_ioend_bio( | |
06342cf8 CH |
395 | struct writeback_control *wbc, |
396 | xfs_ioend_t *ioend, | |
397 | struct bio *bio) | |
f6d6d4fc CH |
398 | { |
399 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
400 | bio->bi_private = ioend; |
401 | bio->bi_end_io = xfs_end_bio; | |
402 | ||
932640e8 DC |
403 | /* |
404 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
405 | * but don't update the inode size until I/O completion. | |
406 | */ | |
407 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 408 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 409 | |
721a9602 | 410 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
411 | } |
412 | ||
413 | STATIC struct bio * | |
414 | xfs_alloc_ioend_bio( | |
415 | struct buffer_head *bh) | |
416 | { | |
f6d6d4fc | 417 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 418 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
419 | |
420 | ASSERT(bio->bi_private == NULL); | |
421 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
422 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
423 | return bio; |
424 | } | |
425 | ||
426 | STATIC void | |
427 | xfs_start_buffer_writeback( | |
428 | struct buffer_head *bh) | |
429 | { | |
430 | ASSERT(buffer_mapped(bh)); | |
431 | ASSERT(buffer_locked(bh)); | |
432 | ASSERT(!buffer_delay(bh)); | |
433 | ASSERT(!buffer_unwritten(bh)); | |
434 | ||
435 | mark_buffer_async_write(bh); | |
436 | set_buffer_uptodate(bh); | |
437 | clear_buffer_dirty(bh); | |
438 | } | |
439 | ||
440 | STATIC void | |
441 | xfs_start_page_writeback( | |
442 | struct page *page, | |
f6d6d4fc CH |
443 | int clear_dirty, |
444 | int buffers) | |
445 | { | |
446 | ASSERT(PageLocked(page)); | |
447 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 448 | if (clear_dirty) |
92132021 DC |
449 | clear_page_dirty_for_io(page); |
450 | set_page_writeback(page); | |
f6d6d4fc | 451 | unlock_page(page); |
1f7decf6 FW |
452 | /* If no buffers on the page are to be written, finish it here */ |
453 | if (!buffers) | |
f6d6d4fc | 454 | end_page_writeback(page); |
f6d6d4fc CH |
455 | } |
456 | ||
457 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
458 | { | |
459 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
460 | } | |
461 | ||
462 | /* | |
d88992f6 DC |
463 | * Submit all of the bios for all of the ioends we have saved up, covering the |
464 | * initial writepage page and also any probed pages. | |
465 | * | |
466 | * Because we may have multiple ioends spanning a page, we need to start | |
467 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
468 | * buffers as we got, then we can end up with a page that only has buffers | |
469 | * marked async write and I/O complete on can occur before we mark the other | |
470 | * buffers async write. | |
471 | * | |
472 | * The end result of this is that we trip a bug in end_page_writeback() because | |
473 | * we call it twice for the one page as the code in end_buffer_async_write() | |
474 | * assumes that all buffers on the page are started at the same time. | |
475 | * | |
476 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 477 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
478 | */ |
479 | STATIC void | |
480 | xfs_submit_ioend( | |
06342cf8 | 481 | struct writeback_control *wbc, |
f6d6d4fc CH |
482 | xfs_ioend_t *ioend) |
483 | { | |
d88992f6 | 484 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
485 | xfs_ioend_t *next; |
486 | struct buffer_head *bh; | |
487 | struct bio *bio; | |
488 | sector_t lastblock = 0; | |
489 | ||
d88992f6 DC |
490 | /* Pass 1 - start writeback */ |
491 | do { | |
492 | next = ioend->io_list; | |
221cb251 | 493 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 494 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
495 | } while ((ioend = next) != NULL); |
496 | ||
497 | /* Pass 2 - submit I/O */ | |
498 | ioend = head; | |
f6d6d4fc CH |
499 | do { |
500 | next = ioend->io_list; | |
501 | bio = NULL; | |
502 | ||
503 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
504 | |
505 | if (!bio) { | |
506 | retry: | |
507 | bio = xfs_alloc_ioend_bio(bh); | |
508 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 509 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
510 | goto retry; |
511 | } | |
512 | ||
513 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 514 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
515 | goto retry; |
516 | } | |
517 | ||
518 | lastblock = bh->b_blocknr; | |
519 | } | |
520 | if (bio) | |
06342cf8 | 521 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 522 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
523 | } while ((ioend = next) != NULL); |
524 | } | |
525 | ||
526 | /* | |
527 | * Cancel submission of all buffer_heads so far in this endio. | |
528 | * Toss the endio too. Only ever called for the initial page | |
529 | * in a writepage request, so only ever one page. | |
530 | */ | |
531 | STATIC void | |
532 | xfs_cancel_ioend( | |
533 | xfs_ioend_t *ioend) | |
534 | { | |
535 | xfs_ioend_t *next; | |
536 | struct buffer_head *bh, *next_bh; | |
537 | ||
538 | do { | |
539 | next = ioend->io_list; | |
540 | bh = ioend->io_buffer_head; | |
541 | do { | |
542 | next_bh = bh->b_private; | |
543 | clear_buffer_async_write(bh); | |
544 | unlock_buffer(bh); | |
545 | } while ((bh = next_bh) != NULL); | |
546 | ||
25e41b3d | 547 | xfs_ioend_wake(XFS_I(ioend->io_inode)); |
f6d6d4fc CH |
548 | mempool_free(ioend, xfs_ioend_pool); |
549 | } while ((ioend = next) != NULL); | |
550 | } | |
551 | ||
552 | /* | |
553 | * Test to see if we've been building up a completion structure for | |
554 | * earlier buffers -- if so, we try to append to this ioend if we | |
555 | * can, otherwise we finish off any current ioend and start another. | |
556 | * Return true if we've finished the given ioend. | |
557 | */ | |
558 | STATIC void | |
559 | xfs_add_to_ioend( | |
560 | struct inode *inode, | |
561 | struct buffer_head *bh, | |
7336cea8 | 562 | xfs_off_t offset, |
f6d6d4fc CH |
563 | unsigned int type, |
564 | xfs_ioend_t **result, | |
565 | int need_ioend) | |
566 | { | |
567 | xfs_ioend_t *ioend = *result; | |
568 | ||
569 | if (!ioend || need_ioend || type != ioend->io_type) { | |
570 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 571 | |
f6d6d4fc CH |
572 | ioend = xfs_alloc_ioend(inode, type); |
573 | ioend->io_offset = offset; | |
574 | ioend->io_buffer_head = bh; | |
575 | ioend->io_buffer_tail = bh; | |
576 | if (previous) | |
577 | previous->io_list = ioend; | |
578 | *result = ioend; | |
579 | } else { | |
580 | ioend->io_buffer_tail->b_private = bh; | |
581 | ioend->io_buffer_tail = bh; | |
582 | } | |
583 | ||
584 | bh->b_private = NULL; | |
585 | ioend->io_size += bh->b_size; | |
586 | } | |
587 | ||
87cbc49c NS |
588 | STATIC void |
589 | xfs_map_buffer( | |
046f1685 | 590 | struct inode *inode, |
87cbc49c | 591 | struct buffer_head *bh, |
207d0416 | 592 | struct xfs_bmbt_irec *imap, |
046f1685 | 593 | xfs_off_t offset) |
87cbc49c NS |
594 | { |
595 | sector_t bn; | |
8699bb0a | 596 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
597 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
598 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 599 | |
207d0416 CH |
600 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
601 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 602 | |
e513182d | 603 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 604 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 605 | |
046f1685 | 606 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
607 | |
608 | bh->b_blocknr = bn; | |
609 | set_buffer_mapped(bh); | |
610 | } | |
611 | ||
1da177e4 LT |
612 | STATIC void |
613 | xfs_map_at_offset( | |
046f1685 | 614 | struct inode *inode, |
1da177e4 | 615 | struct buffer_head *bh, |
207d0416 | 616 | struct xfs_bmbt_irec *imap, |
046f1685 | 617 | xfs_off_t offset) |
1da177e4 | 618 | { |
207d0416 CH |
619 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
620 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 621 | |
207d0416 | 622 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
623 | set_buffer_mapped(bh); |
624 | clear_buffer_delay(bh); | |
f6d6d4fc | 625 | clear_buffer_unwritten(bh); |
1da177e4 LT |
626 | } |
627 | ||
1da177e4 | 628 | /* |
10ce4444 CH |
629 | * Test if a given page is suitable for writing as part of an unwritten |
630 | * or delayed allocate extent. | |
1da177e4 | 631 | */ |
10ce4444 CH |
632 | STATIC int |
633 | xfs_is_delayed_page( | |
634 | struct page *page, | |
f6d6d4fc | 635 | unsigned int type) |
1da177e4 | 636 | { |
1da177e4 | 637 | if (PageWriteback(page)) |
10ce4444 | 638 | return 0; |
1da177e4 LT |
639 | |
640 | if (page->mapping && page_has_buffers(page)) { | |
641 | struct buffer_head *bh, *head; | |
642 | int acceptable = 0; | |
643 | ||
644 | bh = head = page_buffers(page); | |
645 | do { | |
f6d6d4fc | 646 | if (buffer_unwritten(bh)) |
34a52c6c | 647 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 648 | else if (buffer_delay(bh)) |
a206c817 | 649 | acceptable = (type == IO_DELALLOC); |
2ddee844 | 650 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
a206c817 | 651 | acceptable = (type == IO_OVERWRITE); |
f6d6d4fc | 652 | else |
1da177e4 | 653 | break; |
1da177e4 LT |
654 | } while ((bh = bh->b_this_page) != head); |
655 | ||
656 | if (acceptable) | |
10ce4444 | 657 | return 1; |
1da177e4 LT |
658 | } |
659 | ||
10ce4444 | 660 | return 0; |
1da177e4 LT |
661 | } |
662 | ||
1da177e4 LT |
663 | /* |
664 | * Allocate & map buffers for page given the extent map. Write it out. | |
665 | * except for the original page of a writepage, this is called on | |
666 | * delalloc/unwritten pages only, for the original page it is possible | |
667 | * that the page has no mapping at all. | |
668 | */ | |
f6d6d4fc | 669 | STATIC int |
1da177e4 LT |
670 | xfs_convert_page( |
671 | struct inode *inode, | |
672 | struct page *page, | |
10ce4444 | 673 | loff_t tindex, |
207d0416 | 674 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 675 | xfs_ioend_t **ioendp, |
2fa24f92 | 676 | struct writeback_control *wbc) |
1da177e4 | 677 | { |
f6d6d4fc | 678 | struct buffer_head *bh, *head; |
9260dc6b CH |
679 | xfs_off_t end_offset; |
680 | unsigned long p_offset; | |
f6d6d4fc | 681 | unsigned int type; |
24e17b5f | 682 | int len, page_dirty; |
f6d6d4fc | 683 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 684 | xfs_off_t offset = page_offset(page); |
1da177e4 | 685 | |
10ce4444 CH |
686 | if (page->index != tindex) |
687 | goto fail; | |
529ae9aa | 688 | if (!trylock_page(page)) |
10ce4444 CH |
689 | goto fail; |
690 | if (PageWriteback(page)) | |
691 | goto fail_unlock_page; | |
692 | if (page->mapping != inode->i_mapping) | |
693 | goto fail_unlock_page; | |
694 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
695 | goto fail_unlock_page; | |
696 | ||
24e17b5f NS |
697 | /* |
698 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 699 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
700 | * |
701 | * Derivation: | |
702 | * | |
703 | * End offset is the highest offset that this page should represent. | |
704 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
705 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
706 | * hence give us the correct page_dirty count. On any other page, | |
707 | * it will be zero and in that case we need page_dirty to be the | |
708 | * count of buffers on the page. | |
24e17b5f | 709 | */ |
9260dc6b CH |
710 | end_offset = min_t(unsigned long long, |
711 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
712 | i_size_read(inode)); | |
713 | ||
24e17b5f | 714 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
715 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
716 | PAGE_CACHE_SIZE); | |
717 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
718 | page_dirty = p_offset / len; | |
24e17b5f | 719 | |
1da177e4 LT |
720 | bh = head = page_buffers(page); |
721 | do { | |
9260dc6b | 722 | if (offset >= end_offset) |
1da177e4 | 723 | break; |
f6d6d4fc CH |
724 | if (!buffer_uptodate(bh)) |
725 | uptodate = 0; | |
726 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
727 | done = 1; | |
1da177e4 | 728 | continue; |
f6d6d4fc CH |
729 | } |
730 | ||
2fa24f92 CH |
731 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
732 | buffer_mapped(bh)) { | |
9260dc6b | 733 | if (buffer_unwritten(bh)) |
34a52c6c | 734 | type = IO_UNWRITTEN; |
2fa24f92 | 735 | else if (buffer_delay(bh)) |
a206c817 | 736 | type = IO_DELALLOC; |
2fa24f92 CH |
737 | else |
738 | type = IO_OVERWRITE; | |
9260dc6b | 739 | |
558e6891 | 740 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 741 | done = 1; |
9260dc6b CH |
742 | continue; |
743 | } | |
744 | ||
ecff71e6 CH |
745 | lock_buffer(bh); |
746 | if (type != IO_OVERWRITE) | |
2fa24f92 | 747 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
748 | xfs_add_to_ioend(inode, bh, offset, type, |
749 | ioendp, done); | |
750 | ||
9260dc6b CH |
751 | page_dirty--; |
752 | count++; | |
753 | } else { | |
2fa24f92 | 754 | done = 1; |
1da177e4 | 755 | } |
7336cea8 | 756 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 757 | |
f6d6d4fc CH |
758 | if (uptodate && bh == head) |
759 | SetPageUptodate(page); | |
760 | ||
89f3b363 | 761 | if (count) { |
efceab1d DC |
762 | if (--wbc->nr_to_write <= 0 && |
763 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 764 | done = 1; |
1da177e4 | 765 | } |
89f3b363 | 766 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
767 | |
768 | return done; | |
10ce4444 CH |
769 | fail_unlock_page: |
770 | unlock_page(page); | |
771 | fail: | |
772 | return 1; | |
1da177e4 LT |
773 | } |
774 | ||
775 | /* | |
776 | * Convert & write out a cluster of pages in the same extent as defined | |
777 | * by mp and following the start page. | |
778 | */ | |
779 | STATIC void | |
780 | xfs_cluster_write( | |
781 | struct inode *inode, | |
782 | pgoff_t tindex, | |
207d0416 | 783 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 784 | xfs_ioend_t **ioendp, |
1da177e4 | 785 | struct writeback_control *wbc, |
1da177e4 LT |
786 | pgoff_t tlast) |
787 | { | |
10ce4444 CH |
788 | struct pagevec pvec; |
789 | int done = 0, i; | |
1da177e4 | 790 | |
10ce4444 CH |
791 | pagevec_init(&pvec, 0); |
792 | while (!done && tindex <= tlast) { | |
793 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
794 | ||
795 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 796 | break; |
10ce4444 CH |
797 | |
798 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
799 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 800 | imap, ioendp, wbc); |
10ce4444 CH |
801 | if (done) |
802 | break; | |
803 | } | |
804 | ||
805 | pagevec_release(&pvec); | |
806 | cond_resched(); | |
1da177e4 LT |
807 | } |
808 | } | |
809 | ||
3ed3a434 DC |
810 | STATIC void |
811 | xfs_vm_invalidatepage( | |
812 | struct page *page, | |
813 | unsigned long offset) | |
814 | { | |
815 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
816 | block_invalidatepage(page, offset); | |
817 | } | |
818 | ||
819 | /* | |
820 | * If the page has delalloc buffers on it, we need to punch them out before we | |
821 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
822 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
823 | * is done on that same region - the delalloc extent is returned when none is | |
824 | * supposed to be there. | |
825 | * | |
826 | * We prevent this by truncating away the delalloc regions on the page before | |
827 | * invalidating it. Because they are delalloc, we can do this without needing a | |
828 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
829 | * truncation without a transaction as there is no space left for block | |
830 | * reservation (typically why we see a ENOSPC in writeback). | |
831 | * | |
832 | * This is not a performance critical path, so for now just do the punching a | |
833 | * buffer head at a time. | |
834 | */ | |
835 | STATIC void | |
836 | xfs_aops_discard_page( | |
837 | struct page *page) | |
838 | { | |
839 | struct inode *inode = page->mapping->host; | |
840 | struct xfs_inode *ip = XFS_I(inode); | |
841 | struct buffer_head *bh, *head; | |
842 | loff_t offset = page_offset(page); | |
3ed3a434 | 843 | |
a206c817 | 844 | if (!xfs_is_delayed_page(page, IO_DELALLOC)) |
3ed3a434 DC |
845 | goto out_invalidate; |
846 | ||
e8c3753c DC |
847 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
848 | goto out_invalidate; | |
849 | ||
4f10700a | 850 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
851 | "page discard on page %p, inode 0x%llx, offset %llu.", |
852 | page, ip->i_ino, offset); | |
853 | ||
854 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
855 | bh = head = page_buffers(page); | |
856 | do { | |
3ed3a434 | 857 | int error; |
c726de44 | 858 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
859 | |
860 | if (!buffer_delay(bh)) | |
861 | goto next_buffer; | |
862 | ||
c726de44 DC |
863 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
864 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
865 | if (error) { |
866 | /* something screwed, just bail */ | |
e8c3753c | 867 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 868 | xfs_alert(ip->i_mount, |
3ed3a434 | 869 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 870 | } |
3ed3a434 DC |
871 | break; |
872 | } | |
873 | next_buffer: | |
c726de44 | 874 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
875 | |
876 | } while ((bh = bh->b_this_page) != head); | |
877 | ||
878 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
879 | out_invalidate: | |
880 | xfs_vm_invalidatepage(page, 0); | |
881 | return; | |
882 | } | |
883 | ||
1da177e4 | 884 | /* |
89f3b363 CH |
885 | * Write out a dirty page. |
886 | * | |
887 | * For delalloc space on the page we need to allocate space and flush it. | |
888 | * For unwritten space on the page we need to start the conversion to | |
889 | * regular allocated space. | |
89f3b363 | 890 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 891 | */ |
1da177e4 | 892 | STATIC int |
89f3b363 CH |
893 | xfs_vm_writepage( |
894 | struct page *page, | |
895 | struct writeback_control *wbc) | |
1da177e4 | 896 | { |
89f3b363 | 897 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 898 | struct buffer_head *bh, *head; |
207d0416 | 899 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 900 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 901 | loff_t offset; |
f6d6d4fc | 902 | unsigned int type; |
1da177e4 | 903 | __uint64_t end_offset; |
bd1556a1 | 904 | pgoff_t end_index, last_index; |
ed1e7b7e | 905 | ssize_t len; |
a206c817 | 906 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 907 | int count = 0; |
a206c817 | 908 | int nonblocking = 0; |
89f3b363 CH |
909 | |
910 | trace_xfs_writepage(inode, page, 0); | |
911 | ||
20cb52eb CH |
912 | ASSERT(page_has_buffers(page)); |
913 | ||
89f3b363 CH |
914 | /* |
915 | * Refuse to write the page out if we are called from reclaim context. | |
916 | * | |
d4f7a5cb CH |
917 | * This avoids stack overflows when called from deeply used stacks in |
918 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
919 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 CH |
920 | * |
921 | * This should really be done by the core VM, but until that happens | |
922 | * filesystems like XFS, btrfs and ext4 have to take care of this | |
923 | * by themselves. | |
924 | */ | |
d4f7a5cb | 925 | if ((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == PF_MEMALLOC) |
b5420f23 | 926 | goto redirty; |
1da177e4 | 927 | |
89f3b363 | 928 | /* |
680a647b CH |
929 | * Given that we do not allow direct reclaim to call us, we should |
930 | * never be called while in a filesystem transaction. | |
89f3b363 | 931 | */ |
680a647b | 932 | if (WARN_ON(current->flags & PF_FSTRANS)) |
b5420f23 | 933 | goto redirty; |
89f3b363 | 934 | |
1da177e4 LT |
935 | /* Is this page beyond the end of the file? */ |
936 | offset = i_size_read(inode); | |
937 | end_index = offset >> PAGE_CACHE_SHIFT; | |
938 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
939 | if (page->index >= end_index) { | |
940 | if ((page->index >= end_index + 1) || | |
941 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 942 | unlock_page(page); |
19d5bcf3 | 943 | return 0; |
1da177e4 LT |
944 | } |
945 | } | |
946 | ||
f6d6d4fc | 947 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
948 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
949 | offset); | |
24e17b5f | 950 | len = 1 << inode->i_blkbits; |
24e17b5f | 951 | |
24e17b5f | 952 | bh = head = page_buffers(page); |
f6d6d4fc | 953 | offset = page_offset(page); |
a206c817 CH |
954 | type = IO_OVERWRITE; |
955 | ||
dbcdde3e | 956 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 957 | nonblocking = 1; |
f6d6d4fc | 958 | |
1da177e4 | 959 | do { |
6ac7248e CH |
960 | int new_ioend = 0; |
961 | ||
1da177e4 LT |
962 | if (offset >= end_offset) |
963 | break; | |
964 | if (!buffer_uptodate(bh)) | |
965 | uptodate = 0; | |
1da177e4 | 966 | |
3d9b02e3 | 967 | /* |
ece413f5 CH |
968 | * set_page_dirty dirties all buffers in a page, independent |
969 | * of their state. The dirty state however is entirely | |
970 | * meaningless for holes (!mapped && uptodate), so skip | |
971 | * buffers covering holes here. | |
3d9b02e3 ES |
972 | */ |
973 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
974 | imap_valid = 0; |
975 | continue; | |
976 | } | |
977 | ||
aeea1b1f CH |
978 | if (buffer_unwritten(bh)) { |
979 | if (type != IO_UNWRITTEN) { | |
980 | type = IO_UNWRITTEN; | |
981 | imap_valid = 0; | |
1da177e4 | 982 | } |
aeea1b1f CH |
983 | } else if (buffer_delay(bh)) { |
984 | if (type != IO_DELALLOC) { | |
985 | type = IO_DELALLOC; | |
986 | imap_valid = 0; | |
1da177e4 | 987 | } |
89f3b363 | 988 | } else if (buffer_uptodate(bh)) { |
a206c817 CH |
989 | if (type != IO_OVERWRITE) { |
990 | type = IO_OVERWRITE; | |
85da94c6 CH |
991 | imap_valid = 0; |
992 | } | |
aeea1b1f CH |
993 | } else { |
994 | if (PageUptodate(page)) { | |
995 | ASSERT(buffer_mapped(bh)); | |
996 | imap_valid = 0; | |
6c4fe19f | 997 | } |
aeea1b1f CH |
998 | continue; |
999 | } | |
d5cb48aa | 1000 | |
aeea1b1f CH |
1001 | if (imap_valid) |
1002 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1003 | if (!imap_valid) { | |
1004 | /* | |
1005 | * If we didn't have a valid mapping then we need to | |
1006 | * put the new mapping into a separate ioend structure. | |
1007 | * This ensures non-contiguous extents always have | |
1008 | * separate ioends, which is particularly important | |
1009 | * for unwritten extent conversion at I/O completion | |
1010 | * time. | |
1011 | */ | |
1012 | new_ioend = 1; | |
1013 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1014 | nonblocking); | |
1015 | if (err) | |
1016 | goto error; | |
1017 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1018 | } | |
1019 | if (imap_valid) { | |
ecff71e6 CH |
1020 | lock_buffer(bh); |
1021 | if (type != IO_OVERWRITE) | |
aeea1b1f CH |
1022 | xfs_map_at_offset(inode, bh, &imap, offset); |
1023 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1024 | new_ioend); | |
1025 | count++; | |
1da177e4 | 1026 | } |
f6d6d4fc CH |
1027 | |
1028 | if (!iohead) | |
1029 | iohead = ioend; | |
1030 | ||
1031 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1032 | |
1033 | if (uptodate && bh == head) | |
1034 | SetPageUptodate(page); | |
1035 | ||
89f3b363 | 1036 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1037 | |
558e6891 | 1038 | if (ioend && imap_valid) { |
bd1556a1 CH |
1039 | xfs_off_t end_index; |
1040 | ||
1041 | end_index = imap.br_startoff + imap.br_blockcount; | |
1042 | ||
1043 | /* to bytes */ | |
1044 | end_index <<= inode->i_blkbits; | |
1045 | ||
1046 | /* to pages */ | |
1047 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1048 | ||
1049 | /* check against file size */ | |
1050 | if (end_index > last_index) | |
1051 | end_index = last_index; | |
8699bb0a | 1052 | |
207d0416 | 1053 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1054 | wbc, end_index); |
1da177e4 LT |
1055 | } |
1056 | ||
f6d6d4fc | 1057 | if (iohead) |
06342cf8 | 1058 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1059 | |
89f3b363 | 1060 | return 0; |
1da177e4 LT |
1061 | |
1062 | error: | |
f6d6d4fc CH |
1063 | if (iohead) |
1064 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1065 | |
b5420f23 CH |
1066 | if (err == -EAGAIN) |
1067 | goto redirty; | |
1068 | ||
20cb52eb | 1069 | xfs_aops_discard_page(page); |
89f3b363 CH |
1070 | ClearPageUptodate(page); |
1071 | unlock_page(page); | |
1da177e4 | 1072 | return err; |
f51623b2 | 1073 | |
b5420f23 | 1074 | redirty: |
f51623b2 NS |
1075 | redirty_page_for_writepage(wbc, page); |
1076 | unlock_page(page); | |
1077 | return 0; | |
f51623b2 NS |
1078 | } |
1079 | ||
7d4fb40a NS |
1080 | STATIC int |
1081 | xfs_vm_writepages( | |
1082 | struct address_space *mapping, | |
1083 | struct writeback_control *wbc) | |
1084 | { | |
b3aea4ed | 1085 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1086 | return generic_writepages(mapping, wbc); |
1087 | } | |
1088 | ||
f51623b2 NS |
1089 | /* |
1090 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1091 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1092 | * have buffer heads in this call. |
1093 | * | |
89f3b363 | 1094 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1095 | */ |
1096 | STATIC int | |
238f4c54 | 1097 | xfs_vm_releasepage( |
f51623b2 NS |
1098 | struct page *page, |
1099 | gfp_t gfp_mask) | |
1100 | { | |
20cb52eb | 1101 | int delalloc, unwritten; |
f51623b2 | 1102 | |
89f3b363 | 1103 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1104 | |
20cb52eb | 1105 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1106 | |
89f3b363 | 1107 | if (WARN_ON(delalloc)) |
f51623b2 | 1108 | return 0; |
89f3b363 | 1109 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1110 | return 0; |
1111 | ||
f51623b2 NS |
1112 | return try_to_free_buffers(page); |
1113 | } | |
1114 | ||
1da177e4 | 1115 | STATIC int |
c2536668 | 1116 | __xfs_get_blocks( |
1da177e4 LT |
1117 | struct inode *inode, |
1118 | sector_t iblock, | |
1da177e4 LT |
1119 | struct buffer_head *bh_result, |
1120 | int create, | |
f2bde9b8 | 1121 | int direct) |
1da177e4 | 1122 | { |
a206c817 CH |
1123 | struct xfs_inode *ip = XFS_I(inode); |
1124 | struct xfs_mount *mp = ip->i_mount; | |
1125 | xfs_fileoff_t offset_fsb, end_fsb; | |
1126 | int error = 0; | |
1127 | int lockmode = 0; | |
207d0416 | 1128 | struct xfs_bmbt_irec imap; |
a206c817 | 1129 | int nimaps = 1; |
fdc7ed75 NS |
1130 | xfs_off_t offset; |
1131 | ssize_t size; | |
207d0416 | 1132 | int new = 0; |
a206c817 CH |
1133 | |
1134 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1135 | return -XFS_ERROR(EIO); | |
1da177e4 | 1136 | |
fdc7ed75 | 1137 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1138 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1139 | size = bh_result->b_size; | |
364f358a LM |
1140 | |
1141 | if (!create && direct && offset >= i_size_read(inode)) | |
1142 | return 0; | |
1143 | ||
a206c817 CH |
1144 | if (create) { |
1145 | lockmode = XFS_ILOCK_EXCL; | |
1146 | xfs_ilock(ip, lockmode); | |
1147 | } else { | |
1148 | lockmode = xfs_ilock_map_shared(ip); | |
1149 | } | |
f2bde9b8 | 1150 | |
a206c817 CH |
1151 | ASSERT(offset <= mp->m_maxioffset); |
1152 | if (offset + size > mp->m_maxioffset) | |
1153 | size = mp->m_maxioffset - offset; | |
1154 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); | |
1155 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1156 | ||
1157 | error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb, | |
1158 | XFS_BMAPI_ENTIRE, NULL, 0, &imap, &nimaps, NULL); | |
1da177e4 | 1159 | if (error) |
a206c817 CH |
1160 | goto out_unlock; |
1161 | ||
1162 | if (create && | |
1163 | (!nimaps || | |
1164 | (imap.br_startblock == HOLESTARTBLOCK || | |
1165 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
1166 | if (direct) { | |
1167 | error = xfs_iomap_write_direct(ip, offset, size, | |
1168 | &imap, nimaps); | |
1169 | } else { | |
1170 | error = xfs_iomap_write_delay(ip, offset, size, &imap); | |
1171 | } | |
1172 | if (error) | |
1173 | goto out_unlock; | |
1174 | ||
1175 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1176 | } else if (nimaps) { | |
1177 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
1178 | } else { | |
1179 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1180 | goto out_unlock; | |
1181 | } | |
1182 | xfs_iunlock(ip, lockmode); | |
1da177e4 | 1183 | |
207d0416 CH |
1184 | if (imap.br_startblock != HOLESTARTBLOCK && |
1185 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1186 | /* |
1187 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1188 | * the read case (treat as if we're reading into a hole). |
1189 | */ | |
207d0416 CH |
1190 | if (create || !ISUNWRITTEN(&imap)) |
1191 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1192 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1193 | if (direct) |
1194 | bh_result->b_private = inode; | |
1195 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1196 | } |
1197 | } | |
1198 | ||
c2536668 NS |
1199 | /* |
1200 | * If this is a realtime file, data may be on a different device. | |
1201 | * to that pointed to from the buffer_head b_bdev currently. | |
1202 | */ | |
046f1685 | 1203 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1204 | |
c2536668 | 1205 | /* |
549054af DC |
1206 | * If we previously allocated a block out beyond eof and we are now |
1207 | * coming back to use it then we will need to flag it as new even if it | |
1208 | * has a disk address. | |
1209 | * | |
1210 | * With sub-block writes into unwritten extents we also need to mark | |
1211 | * the buffer as new so that the unwritten parts of the buffer gets | |
1212 | * correctly zeroed. | |
1da177e4 LT |
1213 | */ |
1214 | if (create && | |
1215 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1216 | (offset >= i_size_read(inode)) || |
207d0416 | 1217 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1218 | set_buffer_new(bh_result); |
1da177e4 | 1219 | |
207d0416 | 1220 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1221 | BUG_ON(direct); |
1222 | if (create) { | |
1223 | set_buffer_uptodate(bh_result); | |
1224 | set_buffer_mapped(bh_result); | |
1225 | set_buffer_delay(bh_result); | |
1226 | } | |
1227 | } | |
1228 | ||
2b8f12b7 CH |
1229 | /* |
1230 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1231 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1232 | */ | |
c2536668 | 1233 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1234 | xfs_off_t mapping_size; |
1235 | ||
1236 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1237 | mapping_size <<= inode->i_blkbits; | |
1238 | ||
1239 | ASSERT(mapping_size > 0); | |
1240 | if (mapping_size > size) | |
1241 | mapping_size = size; | |
1242 | if (mapping_size > LONG_MAX) | |
1243 | mapping_size = LONG_MAX; | |
1244 | ||
1245 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1246 | } |
1247 | ||
1248 | return 0; | |
a206c817 CH |
1249 | |
1250 | out_unlock: | |
1251 | xfs_iunlock(ip, lockmode); | |
1252 | return -error; | |
1da177e4 LT |
1253 | } |
1254 | ||
1255 | int | |
c2536668 | 1256 | xfs_get_blocks( |
1da177e4 LT |
1257 | struct inode *inode, |
1258 | sector_t iblock, | |
1259 | struct buffer_head *bh_result, | |
1260 | int create) | |
1261 | { | |
f2bde9b8 | 1262 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1263 | } |
1264 | ||
1265 | STATIC int | |
e4c573bb | 1266 | xfs_get_blocks_direct( |
1da177e4 LT |
1267 | struct inode *inode, |
1268 | sector_t iblock, | |
1da177e4 LT |
1269 | struct buffer_head *bh_result, |
1270 | int create) | |
1271 | { | |
f2bde9b8 | 1272 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1273 | } |
1274 | ||
209fb87a CH |
1275 | /* |
1276 | * Complete a direct I/O write request. | |
1277 | * | |
1278 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1279 | * need to issue a transaction to convert the range from unwritten to written | |
1280 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1281 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1282 | * request this handler is called from interrupt context, from which we |
1283 | * can't start transactions. In that case offload the I/O completion to | |
1284 | * the workqueues we also use for buffered I/O completion. | |
1285 | */ | |
f0973863 | 1286 | STATIC void |
209fb87a CH |
1287 | xfs_end_io_direct_write( |
1288 | struct kiocb *iocb, | |
1289 | loff_t offset, | |
1290 | ssize_t size, | |
1291 | void *private, | |
1292 | int ret, | |
1293 | bool is_async) | |
f0973863 | 1294 | { |
209fb87a | 1295 | struct xfs_ioend *ioend = iocb->private; |
f0973863 CH |
1296 | |
1297 | /* | |
209fb87a CH |
1298 | * blockdev_direct_IO can return an error even after the I/O |
1299 | * completion handler was called. Thus we need to protect | |
1300 | * against double-freeing. | |
f0973863 | 1301 | */ |
209fb87a CH |
1302 | iocb->private = NULL; |
1303 | ||
ba87ea69 LM |
1304 | ioend->io_offset = offset; |
1305 | ioend->io_size = size; | |
c859cdd1 CH |
1306 | ioend->io_iocb = iocb; |
1307 | ioend->io_result = ret; | |
209fb87a CH |
1308 | if (private && size > 0) |
1309 | ioend->io_type = IO_UNWRITTEN; | |
1310 | ||
1311 | if (is_async) { | |
c859cdd1 | 1312 | ioend->io_isasync = 1; |
209fb87a | 1313 | xfs_finish_ioend(ioend); |
f0973863 | 1314 | } else { |
209fb87a | 1315 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1316 | } |
f0973863 CH |
1317 | } |
1318 | ||
1da177e4 | 1319 | STATIC ssize_t |
e4c573bb | 1320 | xfs_vm_direct_IO( |
1da177e4 LT |
1321 | int rw, |
1322 | struct kiocb *iocb, | |
1323 | const struct iovec *iov, | |
1324 | loff_t offset, | |
1325 | unsigned long nr_segs) | |
1326 | { | |
209fb87a CH |
1327 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1328 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1329 | ssize_t ret; | |
1330 | ||
1331 | if (rw & WRITE) { | |
a206c817 | 1332 | iocb->private = xfs_alloc_ioend(inode, IO_DIRECT); |
209fb87a | 1333 | |
eafdc7d1 CH |
1334 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1335 | offset, nr_segs, | |
1336 | xfs_get_blocks_direct, | |
1337 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a CH |
1338 | if (ret != -EIOCBQUEUED && iocb->private) |
1339 | xfs_destroy_ioend(iocb->private); | |
1340 | } else { | |
eafdc7d1 CH |
1341 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1342 | offset, nr_segs, | |
1343 | xfs_get_blocks_direct, | |
1344 | NULL, NULL, 0); | |
209fb87a | 1345 | } |
f0973863 | 1346 | |
f0973863 | 1347 | return ret; |
1da177e4 LT |
1348 | } |
1349 | ||
fa9b227e CH |
1350 | STATIC void |
1351 | xfs_vm_write_failed( | |
1352 | struct address_space *mapping, | |
1353 | loff_t to) | |
1354 | { | |
1355 | struct inode *inode = mapping->host; | |
1356 | ||
1357 | if (to > inode->i_size) { | |
c726de44 DC |
1358 | /* |
1359 | * punch out the delalloc blocks we have already allocated. We | |
1360 | * don't call xfs_setattr() to do this as we may be in the | |
1361 | * middle of a multi-iovec write and so the vfs inode->i_size | |
1362 | * will not match the xfs ip->i_size and so it will zero too | |
1363 | * much. Hence we jus truncate the page cache to zero what is | |
1364 | * necessary and punch the delalloc blocks directly. | |
1365 | */ | |
1366 | struct xfs_inode *ip = XFS_I(inode); | |
1367 | xfs_fileoff_t start_fsb; | |
1368 | xfs_fileoff_t end_fsb; | |
1369 | int error; | |
1370 | ||
1371 | truncate_pagecache(inode, to, inode->i_size); | |
1372 | ||
1373 | /* | |
1374 | * Check if there are any blocks that are outside of i_size | |
1375 | * that need to be trimmed back. | |
1376 | */ | |
1377 | start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1; | |
1378 | end_fsb = XFS_B_TO_FSB(ip->i_mount, to); | |
1379 | if (end_fsb <= start_fsb) | |
1380 | return; | |
1381 | ||
1382 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1383 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1384 | end_fsb - start_fsb); | |
1385 | if (error) { | |
1386 | /* something screwed, just bail */ | |
1387 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
4f10700a | 1388 | xfs_alert(ip->i_mount, |
c726de44 DC |
1389 | "xfs_vm_write_failed: unable to clean up ino %lld", |
1390 | ip->i_ino); | |
1391 | } | |
1392 | } | |
1393 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
fa9b227e CH |
1394 | } |
1395 | } | |
1396 | ||
f51623b2 | 1397 | STATIC int |
d79689c7 | 1398 | xfs_vm_write_begin( |
f51623b2 | 1399 | struct file *file, |
d79689c7 NP |
1400 | struct address_space *mapping, |
1401 | loff_t pos, | |
1402 | unsigned len, | |
1403 | unsigned flags, | |
1404 | struct page **pagep, | |
1405 | void **fsdata) | |
f51623b2 | 1406 | { |
155130a4 CH |
1407 | int ret; |
1408 | ||
1409 | ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS, | |
1410 | pagep, xfs_get_blocks); | |
fa9b227e CH |
1411 | if (unlikely(ret)) |
1412 | xfs_vm_write_failed(mapping, pos + len); | |
1413 | return ret; | |
1414 | } | |
1415 | ||
1416 | STATIC int | |
1417 | xfs_vm_write_end( | |
1418 | struct file *file, | |
1419 | struct address_space *mapping, | |
1420 | loff_t pos, | |
1421 | unsigned len, | |
1422 | unsigned copied, | |
1423 | struct page *page, | |
1424 | void *fsdata) | |
1425 | { | |
1426 | int ret; | |
155130a4 | 1427 | |
fa9b227e CH |
1428 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
1429 | if (unlikely(ret < len)) | |
1430 | xfs_vm_write_failed(mapping, pos + len); | |
155130a4 | 1431 | return ret; |
f51623b2 | 1432 | } |
1da177e4 LT |
1433 | |
1434 | STATIC sector_t | |
e4c573bb | 1435 | xfs_vm_bmap( |
1da177e4 LT |
1436 | struct address_space *mapping, |
1437 | sector_t block) | |
1438 | { | |
1439 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1440 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1441 | |
cca28fb8 | 1442 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1443 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1444 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1445 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1446 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1447 | } |
1448 | ||
1449 | STATIC int | |
e4c573bb | 1450 | xfs_vm_readpage( |
1da177e4 LT |
1451 | struct file *unused, |
1452 | struct page *page) | |
1453 | { | |
c2536668 | 1454 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1455 | } |
1456 | ||
1457 | STATIC int | |
e4c573bb | 1458 | xfs_vm_readpages( |
1da177e4 LT |
1459 | struct file *unused, |
1460 | struct address_space *mapping, | |
1461 | struct list_head *pages, | |
1462 | unsigned nr_pages) | |
1463 | { | |
c2536668 | 1464 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1465 | } |
1466 | ||
f5e54d6e | 1467 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1468 | .readpage = xfs_vm_readpage, |
1469 | .readpages = xfs_vm_readpages, | |
1470 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1471 | .writepages = xfs_vm_writepages, |
238f4c54 NS |
1472 | .releasepage = xfs_vm_releasepage, |
1473 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1474 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1475 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1476 | .bmap = xfs_vm_bmap, |
1477 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1478 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1479 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1480 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1481 | }; |