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 | ||
0b1b213f | 41 | void |
f51623b2 NS |
42 | xfs_count_page_state( |
43 | struct page *page, | |
44 | int *delalloc, | |
f51623b2 NS |
45 | int *unwritten) |
46 | { | |
47 | struct buffer_head *bh, *head; | |
48 | ||
20cb52eb | 49 | *delalloc = *unwritten = 0; |
f51623b2 NS |
50 | |
51 | bh = head = page_buffers(page); | |
52 | do { | |
20cb52eb | 53 | if (buffer_unwritten(bh)) |
f51623b2 NS |
54 | (*unwritten) = 1; |
55 | else if (buffer_delay(bh)) | |
56 | (*delalloc) = 1; | |
57 | } while ((bh = bh->b_this_page) != head); | |
58 | } | |
59 | ||
6214ed44 CH |
60 | STATIC struct block_device * |
61 | xfs_find_bdev_for_inode( | |
046f1685 | 62 | struct inode *inode) |
6214ed44 | 63 | { |
046f1685 | 64 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
65 | struct xfs_mount *mp = ip->i_mount; |
66 | ||
71ddabb9 | 67 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
68 | return mp->m_rtdev_targp->bt_bdev; |
69 | else | |
70 | return mp->m_ddev_targp->bt_bdev; | |
71 | } | |
72 | ||
f6d6d4fc CH |
73 | /* |
74 | * We're now finished for good with this ioend structure. | |
75 | * Update the page state via the associated buffer_heads, | |
76 | * release holds on the inode and bio, and finally free | |
77 | * up memory. Do not use the ioend after this. | |
78 | */ | |
0829c360 CH |
79 | STATIC void |
80 | xfs_destroy_ioend( | |
81 | xfs_ioend_t *ioend) | |
82 | { | |
f6d6d4fc CH |
83 | struct buffer_head *bh, *next; |
84 | ||
85 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
86 | next = bh->b_private; | |
7d04a335 | 87 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 88 | } |
583fa586 | 89 | |
c859cdd1 | 90 | if (ioend->io_iocb) { |
04f658ee CH |
91 | if (ioend->io_isasync) { |
92 | aio_complete(ioend->io_iocb, ioend->io_error ? | |
93 | ioend->io_error : ioend->io_result, 0); | |
94 | } | |
c859cdd1 CH |
95 | inode_dio_done(ioend->io_inode); |
96 | } | |
4a06fd26 | 97 | |
0829c360 CH |
98 | mempool_free(ioend, xfs_ioend_pool); |
99 | } | |
100 | ||
932640e8 DC |
101 | /* |
102 | * If the end of the current ioend is beyond the current EOF, | |
103 | * return the new EOF value, otherwise zero. | |
104 | */ | |
105 | STATIC xfs_fsize_t | |
106 | xfs_ioend_new_eof( | |
107 | xfs_ioend_t *ioend) | |
108 | { | |
109 | xfs_inode_t *ip = XFS_I(ioend->io_inode); | |
110 | xfs_fsize_t isize; | |
111 | xfs_fsize_t bsize; | |
112 | ||
113 | bsize = ioend->io_offset + ioend->io_size; | |
114 | isize = MAX(ip->i_size, ip->i_new_size); | |
115 | isize = MIN(isize, bsize); | |
116 | return isize > ip->i_d.di_size ? isize : 0; | |
117 | } | |
118 | ||
fc0063c4 CH |
119 | /* |
120 | * Fast and loose check if this write could update the on-disk inode size. | |
121 | */ | |
122 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
123 | { | |
124 | return ioend->io_offset + ioend->io_size > | |
125 | XFS_I(ioend->io_inode)->i_d.di_size; | |
126 | } | |
127 | ||
ba87ea69 | 128 | /* |
77d7a0c2 DC |
129 | * Update on-disk file size now that data has been written to disk. The |
130 | * current in-memory file size is i_size. If a write is beyond eof i_new_size | |
131 | * will be the intended file size until i_size is updated. If this write does | |
132 | * not extend all the way to the valid file size then restrict this update to | |
133 | * the end of the write. | |
134 | * | |
135 | * This function does not block as blocking on the inode lock in IO completion | |
136 | * can lead to IO completion order dependency deadlocks.. If it can't get the | |
137 | * inode ilock it will return EAGAIN. Callers must handle this. | |
ba87ea69 | 138 | */ |
77d7a0c2 | 139 | STATIC int |
ba87ea69 LM |
140 | xfs_setfilesize( |
141 | xfs_ioend_t *ioend) | |
142 | { | |
b677c210 | 143 | xfs_inode_t *ip = XFS_I(ioend->io_inode); |
ba87ea69 | 144 | xfs_fsize_t isize; |
ba87ea69 | 145 | |
77d7a0c2 DC |
146 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL)) |
147 | return EAGAIN; | |
ba87ea69 | 148 | |
932640e8 DC |
149 | isize = xfs_ioend_new_eof(ioend); |
150 | if (isize) { | |
55fb25d5 | 151 | trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); |
ba87ea69 | 152 | ip->i_d.di_size = isize; |
66d834ea | 153 | xfs_mark_inode_dirty(ip); |
ba87ea69 LM |
154 | } |
155 | ||
156 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
77d7a0c2 DC |
157 | return 0; |
158 | } | |
159 | ||
160 | /* | |
209fb87a | 161 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
162 | * |
163 | * If there is no work to do we might as well call it a day and free the | |
164 | * ioend right now. | |
77d7a0c2 DC |
165 | */ |
166 | STATIC void | |
167 | xfs_finish_ioend( | |
209fb87a | 168 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
169 | { |
170 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
209fb87a CH |
171 | if (ioend->io_type == IO_UNWRITTEN) |
172 | queue_work(xfsconvertd_workqueue, &ioend->io_work); | |
fc0063c4 | 173 | else if (xfs_ioend_is_append(ioend)) |
209fb87a | 174 | queue_work(xfsdatad_workqueue, &ioend->io_work); |
fc0063c4 CH |
175 | else |
176 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 177 | } |
ba87ea69 LM |
178 | } |
179 | ||
0829c360 | 180 | /* |
5ec4fabb | 181 | * IO write completion. |
f6d6d4fc CH |
182 | */ |
183 | STATIC void | |
5ec4fabb | 184 | xfs_end_io( |
77d7a0c2 | 185 | struct work_struct *work) |
0829c360 | 186 | { |
77d7a0c2 DC |
187 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
188 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 189 | int error = 0; |
ba87ea69 | 190 | |
04f658ee | 191 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 192 | ioend->io_error = -EIO; |
04f658ee CH |
193 | goto done; |
194 | } | |
195 | if (ioend->io_error) | |
196 | goto done; | |
197 | ||
5ec4fabb CH |
198 | /* |
199 | * For unwritten extents we need to issue transactions to convert a | |
200 | * range to normal written extens after the data I/O has finished. | |
201 | */ | |
04f658ee | 202 | if (ioend->io_type == IO_UNWRITTEN) { |
5ec4fabb CH |
203 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
204 | ioend->io_size); | |
04f658ee CH |
205 | if (error) { |
206 | ioend->io_error = -error; | |
207 | goto done; | |
208 | } | |
5ec4fabb | 209 | } |
ba87ea69 | 210 | |
5ec4fabb CH |
211 | /* |
212 | * We might have to update the on-disk file size after extending | |
213 | * writes. | |
214 | */ | |
a206c817 CH |
215 | error = xfs_setfilesize(ioend); |
216 | ASSERT(!error || error == EAGAIN); | |
77d7a0c2 | 217 | |
04f658ee | 218 | done: |
77d7a0c2 DC |
219 | /* |
220 | * If we didn't complete processing of the ioend, requeue it to the | |
221 | * tail of the workqueue for another attempt later. Otherwise destroy | |
222 | * it. | |
223 | */ | |
224 | if (error == EAGAIN) { | |
225 | atomic_inc(&ioend->io_remaining); | |
209fb87a | 226 | xfs_finish_ioend(ioend); |
77d7a0c2 DC |
227 | /* ensure we don't spin on blocked ioends */ |
228 | delay(1); | |
fb511f21 | 229 | } else { |
77d7a0c2 | 230 | xfs_destroy_ioend(ioend); |
fb511f21 | 231 | } |
c626d174 DC |
232 | } |
233 | ||
209fb87a CH |
234 | /* |
235 | * Call IO completion handling in caller context on the final put of an ioend. | |
236 | */ | |
237 | STATIC void | |
238 | xfs_finish_ioend_sync( | |
239 | struct xfs_ioend *ioend) | |
240 | { | |
241 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
242 | xfs_end_io(&ioend->io_work); | |
243 | } | |
244 | ||
0829c360 CH |
245 | /* |
246 | * Allocate and initialise an IO completion structure. | |
247 | * We need to track unwritten extent write completion here initially. | |
248 | * We'll need to extend this for updating the ondisk inode size later | |
249 | * (vs. incore size). | |
250 | */ | |
251 | STATIC xfs_ioend_t * | |
252 | xfs_alloc_ioend( | |
f6d6d4fc CH |
253 | struct inode *inode, |
254 | unsigned int type) | |
0829c360 CH |
255 | { |
256 | xfs_ioend_t *ioend; | |
257 | ||
258 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
259 | ||
260 | /* | |
261 | * Set the count to 1 initially, which will prevent an I/O | |
262 | * completion callback from happening before we have started | |
263 | * all the I/O from calling the completion routine too early. | |
264 | */ | |
265 | atomic_set(&ioend->io_remaining, 1); | |
c859cdd1 | 266 | ioend->io_isasync = 0; |
7d04a335 | 267 | ioend->io_error = 0; |
f6d6d4fc CH |
268 | ioend->io_list = NULL; |
269 | ioend->io_type = type; | |
b677c210 | 270 | ioend->io_inode = inode; |
c1a073bd | 271 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 272 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
273 | ioend->io_offset = 0; |
274 | ioend->io_size = 0; | |
fb511f21 CH |
275 | ioend->io_iocb = NULL; |
276 | ioend->io_result = 0; | |
0829c360 | 277 | |
5ec4fabb | 278 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
279 | return ioend; |
280 | } | |
281 | ||
1da177e4 LT |
282 | STATIC int |
283 | xfs_map_blocks( | |
284 | struct inode *inode, | |
285 | loff_t offset, | |
207d0416 | 286 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
287 | int type, |
288 | int nonblocking) | |
1da177e4 | 289 | { |
a206c817 CH |
290 | struct xfs_inode *ip = XFS_I(inode); |
291 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 292 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
293 | xfs_fileoff_t offset_fsb, end_fsb; |
294 | int error = 0; | |
a206c817 CH |
295 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
296 | int nimaps = 1; | |
297 | ||
298 | if (XFS_FORCED_SHUTDOWN(mp)) | |
299 | return -XFS_ERROR(EIO); | |
300 | ||
8ff2957d | 301 | if (type == IO_UNWRITTEN) |
a206c817 | 302 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
303 | |
304 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
305 | if (nonblocking) | |
306 | return -XFS_ERROR(EAGAIN); | |
307 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
a206c817 CH |
308 | } |
309 | ||
8ff2957d CH |
310 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
311 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
a206c817 | 312 | ASSERT(offset <= mp->m_maxioffset); |
8ff2957d | 313 | |
a206c817 CH |
314 | if (offset + count > mp->m_maxioffset) |
315 | count = mp->m_maxioffset - offset; | |
316 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
317 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
318 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
319 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 320 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 321 | |
8ff2957d CH |
322 | if (error) |
323 | return -XFS_ERROR(error); | |
a206c817 | 324 | |
8ff2957d CH |
325 | if (type == IO_DELALLOC && |
326 | (!nimaps || isnullstartblock(imap->br_startblock))) { | |
a206c817 CH |
327 | error = xfs_iomap_write_allocate(ip, offset, count, imap); |
328 | if (!error) | |
329 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
8ff2957d | 330 | return -XFS_ERROR(error); |
a206c817 CH |
331 | } |
332 | ||
8ff2957d CH |
333 | #ifdef DEBUG |
334 | if (type == IO_UNWRITTEN) { | |
335 | ASSERT(nimaps); | |
336 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
337 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
338 | } | |
339 | #endif | |
340 | if (nimaps) | |
341 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
342 | return 0; | |
1da177e4 LT |
343 | } |
344 | ||
b8f82a4a | 345 | STATIC int |
558e6891 | 346 | xfs_imap_valid( |
8699bb0a | 347 | struct inode *inode, |
207d0416 | 348 | struct xfs_bmbt_irec *imap, |
558e6891 | 349 | xfs_off_t offset) |
1da177e4 | 350 | { |
558e6891 | 351 | offset >>= inode->i_blkbits; |
8699bb0a | 352 | |
558e6891 CH |
353 | return offset >= imap->br_startoff && |
354 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
355 | } |
356 | ||
f6d6d4fc CH |
357 | /* |
358 | * BIO completion handler for buffered IO. | |
359 | */ | |
782e3b3b | 360 | STATIC void |
f6d6d4fc CH |
361 | xfs_end_bio( |
362 | struct bio *bio, | |
f6d6d4fc CH |
363 | int error) |
364 | { | |
365 | xfs_ioend_t *ioend = bio->bi_private; | |
366 | ||
f6d6d4fc | 367 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 368 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
369 | |
370 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
371 | bio->bi_private = NULL; |
372 | bio->bi_end_io = NULL; | |
f6d6d4fc | 373 | bio_put(bio); |
7d04a335 | 374 | |
209fb87a | 375 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
376 | } |
377 | ||
378 | STATIC void | |
379 | xfs_submit_ioend_bio( | |
06342cf8 CH |
380 | struct writeback_control *wbc, |
381 | xfs_ioend_t *ioend, | |
382 | struct bio *bio) | |
f6d6d4fc CH |
383 | { |
384 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
385 | bio->bi_private = ioend; |
386 | bio->bi_end_io = xfs_end_bio; | |
387 | ||
932640e8 DC |
388 | /* |
389 | * If the I/O is beyond EOF we mark the inode dirty immediately | |
390 | * but don't update the inode size until I/O completion. | |
391 | */ | |
392 | if (xfs_ioend_new_eof(ioend)) | |
66d834ea | 393 | xfs_mark_inode_dirty(XFS_I(ioend->io_inode)); |
932640e8 | 394 | |
721a9602 | 395 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
396 | } |
397 | ||
398 | STATIC struct bio * | |
399 | xfs_alloc_ioend_bio( | |
400 | struct buffer_head *bh) | |
401 | { | |
f6d6d4fc | 402 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 403 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
404 | |
405 | ASSERT(bio->bi_private == NULL); | |
406 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
407 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
408 | return bio; |
409 | } | |
410 | ||
411 | STATIC void | |
412 | xfs_start_buffer_writeback( | |
413 | struct buffer_head *bh) | |
414 | { | |
415 | ASSERT(buffer_mapped(bh)); | |
416 | ASSERT(buffer_locked(bh)); | |
417 | ASSERT(!buffer_delay(bh)); | |
418 | ASSERT(!buffer_unwritten(bh)); | |
419 | ||
420 | mark_buffer_async_write(bh); | |
421 | set_buffer_uptodate(bh); | |
422 | clear_buffer_dirty(bh); | |
423 | } | |
424 | ||
425 | STATIC void | |
426 | xfs_start_page_writeback( | |
427 | struct page *page, | |
f6d6d4fc CH |
428 | int clear_dirty, |
429 | int buffers) | |
430 | { | |
431 | ASSERT(PageLocked(page)); | |
432 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 433 | if (clear_dirty) |
92132021 DC |
434 | clear_page_dirty_for_io(page); |
435 | set_page_writeback(page); | |
f6d6d4fc | 436 | unlock_page(page); |
1f7decf6 FW |
437 | /* If no buffers on the page are to be written, finish it here */ |
438 | if (!buffers) | |
f6d6d4fc | 439 | end_page_writeback(page); |
f6d6d4fc CH |
440 | } |
441 | ||
442 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
443 | { | |
444 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
445 | } | |
446 | ||
447 | /* | |
d88992f6 DC |
448 | * Submit all of the bios for all of the ioends we have saved up, covering the |
449 | * initial writepage page and also any probed pages. | |
450 | * | |
451 | * Because we may have multiple ioends spanning a page, we need to start | |
452 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
453 | * buffers as we got, then we can end up with a page that only has buffers | |
454 | * marked async write and I/O complete on can occur before we mark the other | |
455 | * buffers async write. | |
456 | * | |
457 | * The end result of this is that we trip a bug in end_page_writeback() because | |
458 | * we call it twice for the one page as the code in end_buffer_async_write() | |
459 | * assumes that all buffers on the page are started at the same time. | |
460 | * | |
461 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 462 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
463 | */ |
464 | STATIC void | |
465 | xfs_submit_ioend( | |
06342cf8 | 466 | struct writeback_control *wbc, |
f6d6d4fc CH |
467 | xfs_ioend_t *ioend) |
468 | { | |
d88992f6 | 469 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
470 | xfs_ioend_t *next; |
471 | struct buffer_head *bh; | |
472 | struct bio *bio; | |
473 | sector_t lastblock = 0; | |
474 | ||
d88992f6 DC |
475 | /* Pass 1 - start writeback */ |
476 | do { | |
477 | next = ioend->io_list; | |
221cb251 | 478 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 479 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
480 | } while ((ioend = next) != NULL); |
481 | ||
482 | /* Pass 2 - submit I/O */ | |
483 | ioend = head; | |
f6d6d4fc CH |
484 | do { |
485 | next = ioend->io_list; | |
486 | bio = NULL; | |
487 | ||
488 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
489 | |
490 | if (!bio) { | |
491 | retry: | |
492 | bio = xfs_alloc_ioend_bio(bh); | |
493 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 494 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
495 | goto retry; |
496 | } | |
497 | ||
498 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 499 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
500 | goto retry; |
501 | } | |
502 | ||
503 | lastblock = bh->b_blocknr; | |
504 | } | |
505 | if (bio) | |
06342cf8 | 506 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 507 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
508 | } while ((ioend = next) != NULL); |
509 | } | |
510 | ||
511 | /* | |
512 | * Cancel submission of all buffer_heads so far in this endio. | |
513 | * Toss the endio too. Only ever called for the initial page | |
514 | * in a writepage request, so only ever one page. | |
515 | */ | |
516 | STATIC void | |
517 | xfs_cancel_ioend( | |
518 | xfs_ioend_t *ioend) | |
519 | { | |
520 | xfs_ioend_t *next; | |
521 | struct buffer_head *bh, *next_bh; | |
522 | ||
523 | do { | |
524 | next = ioend->io_list; | |
525 | bh = ioend->io_buffer_head; | |
526 | do { | |
527 | next_bh = bh->b_private; | |
528 | clear_buffer_async_write(bh); | |
529 | unlock_buffer(bh); | |
530 | } while ((bh = next_bh) != NULL); | |
531 | ||
f6d6d4fc CH |
532 | mempool_free(ioend, xfs_ioend_pool); |
533 | } while ((ioend = next) != NULL); | |
534 | } | |
535 | ||
536 | /* | |
537 | * Test to see if we've been building up a completion structure for | |
538 | * earlier buffers -- if so, we try to append to this ioend if we | |
539 | * can, otherwise we finish off any current ioend and start another. | |
540 | * Return true if we've finished the given ioend. | |
541 | */ | |
542 | STATIC void | |
543 | xfs_add_to_ioend( | |
544 | struct inode *inode, | |
545 | struct buffer_head *bh, | |
7336cea8 | 546 | xfs_off_t offset, |
f6d6d4fc CH |
547 | unsigned int type, |
548 | xfs_ioend_t **result, | |
549 | int need_ioend) | |
550 | { | |
551 | xfs_ioend_t *ioend = *result; | |
552 | ||
553 | if (!ioend || need_ioend || type != ioend->io_type) { | |
554 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 555 | |
f6d6d4fc CH |
556 | ioend = xfs_alloc_ioend(inode, type); |
557 | ioend->io_offset = offset; | |
558 | ioend->io_buffer_head = bh; | |
559 | ioend->io_buffer_tail = bh; | |
560 | if (previous) | |
561 | previous->io_list = ioend; | |
562 | *result = ioend; | |
563 | } else { | |
564 | ioend->io_buffer_tail->b_private = bh; | |
565 | ioend->io_buffer_tail = bh; | |
566 | } | |
567 | ||
568 | bh->b_private = NULL; | |
569 | ioend->io_size += bh->b_size; | |
570 | } | |
571 | ||
87cbc49c NS |
572 | STATIC void |
573 | xfs_map_buffer( | |
046f1685 | 574 | struct inode *inode, |
87cbc49c | 575 | struct buffer_head *bh, |
207d0416 | 576 | struct xfs_bmbt_irec *imap, |
046f1685 | 577 | xfs_off_t offset) |
87cbc49c NS |
578 | { |
579 | sector_t bn; | |
8699bb0a | 580 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
581 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
582 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 583 | |
207d0416 CH |
584 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
585 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 586 | |
e513182d | 587 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 588 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 589 | |
046f1685 | 590 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
591 | |
592 | bh->b_blocknr = bn; | |
593 | set_buffer_mapped(bh); | |
594 | } | |
595 | ||
1da177e4 LT |
596 | STATIC void |
597 | xfs_map_at_offset( | |
046f1685 | 598 | struct inode *inode, |
1da177e4 | 599 | struct buffer_head *bh, |
207d0416 | 600 | struct xfs_bmbt_irec *imap, |
046f1685 | 601 | xfs_off_t offset) |
1da177e4 | 602 | { |
207d0416 CH |
603 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
604 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 605 | |
207d0416 | 606 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
607 | set_buffer_mapped(bh); |
608 | clear_buffer_delay(bh); | |
f6d6d4fc | 609 | clear_buffer_unwritten(bh); |
1da177e4 LT |
610 | } |
611 | ||
1da177e4 | 612 | /* |
10ce4444 CH |
613 | * Test if a given page is suitable for writing as part of an unwritten |
614 | * or delayed allocate extent. | |
1da177e4 | 615 | */ |
10ce4444 CH |
616 | STATIC int |
617 | xfs_is_delayed_page( | |
618 | struct page *page, | |
f6d6d4fc | 619 | unsigned int type) |
1da177e4 | 620 | { |
1da177e4 | 621 | if (PageWriteback(page)) |
10ce4444 | 622 | return 0; |
1da177e4 LT |
623 | |
624 | if (page->mapping && page_has_buffers(page)) { | |
625 | struct buffer_head *bh, *head; | |
626 | int acceptable = 0; | |
627 | ||
628 | bh = head = page_buffers(page); | |
629 | do { | |
f6d6d4fc | 630 | if (buffer_unwritten(bh)) |
34a52c6c | 631 | acceptable = (type == IO_UNWRITTEN); |
f6d6d4fc | 632 | else if (buffer_delay(bh)) |
a206c817 | 633 | acceptable = (type == IO_DELALLOC); |
2ddee844 | 634 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
a206c817 | 635 | acceptable = (type == IO_OVERWRITE); |
f6d6d4fc | 636 | else |
1da177e4 | 637 | break; |
1da177e4 LT |
638 | } while ((bh = bh->b_this_page) != head); |
639 | ||
640 | if (acceptable) | |
10ce4444 | 641 | return 1; |
1da177e4 LT |
642 | } |
643 | ||
10ce4444 | 644 | return 0; |
1da177e4 LT |
645 | } |
646 | ||
1da177e4 LT |
647 | /* |
648 | * Allocate & map buffers for page given the extent map. Write it out. | |
649 | * except for the original page of a writepage, this is called on | |
650 | * delalloc/unwritten pages only, for the original page it is possible | |
651 | * that the page has no mapping at all. | |
652 | */ | |
f6d6d4fc | 653 | STATIC int |
1da177e4 LT |
654 | xfs_convert_page( |
655 | struct inode *inode, | |
656 | struct page *page, | |
10ce4444 | 657 | loff_t tindex, |
207d0416 | 658 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 659 | xfs_ioend_t **ioendp, |
2fa24f92 | 660 | struct writeback_control *wbc) |
1da177e4 | 661 | { |
f6d6d4fc | 662 | struct buffer_head *bh, *head; |
9260dc6b CH |
663 | xfs_off_t end_offset; |
664 | unsigned long p_offset; | |
f6d6d4fc | 665 | unsigned int type; |
24e17b5f | 666 | int len, page_dirty; |
f6d6d4fc | 667 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 668 | xfs_off_t offset = page_offset(page); |
1da177e4 | 669 | |
10ce4444 CH |
670 | if (page->index != tindex) |
671 | goto fail; | |
529ae9aa | 672 | if (!trylock_page(page)) |
10ce4444 CH |
673 | goto fail; |
674 | if (PageWriteback(page)) | |
675 | goto fail_unlock_page; | |
676 | if (page->mapping != inode->i_mapping) | |
677 | goto fail_unlock_page; | |
678 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
679 | goto fail_unlock_page; | |
680 | ||
24e17b5f NS |
681 | /* |
682 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 683 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
684 | * |
685 | * Derivation: | |
686 | * | |
687 | * End offset is the highest offset that this page should represent. | |
688 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
689 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
690 | * hence give us the correct page_dirty count. On any other page, | |
691 | * it will be zero and in that case we need page_dirty to be the | |
692 | * count of buffers on the page. | |
24e17b5f | 693 | */ |
9260dc6b CH |
694 | end_offset = min_t(unsigned long long, |
695 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
696 | i_size_read(inode)); | |
697 | ||
24e17b5f | 698 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
699 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
700 | PAGE_CACHE_SIZE); | |
701 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
702 | page_dirty = p_offset / len; | |
24e17b5f | 703 | |
1da177e4 LT |
704 | bh = head = page_buffers(page); |
705 | do { | |
9260dc6b | 706 | if (offset >= end_offset) |
1da177e4 | 707 | break; |
f6d6d4fc CH |
708 | if (!buffer_uptodate(bh)) |
709 | uptodate = 0; | |
710 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
711 | done = 1; | |
1da177e4 | 712 | continue; |
f6d6d4fc CH |
713 | } |
714 | ||
2fa24f92 CH |
715 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
716 | buffer_mapped(bh)) { | |
9260dc6b | 717 | if (buffer_unwritten(bh)) |
34a52c6c | 718 | type = IO_UNWRITTEN; |
2fa24f92 | 719 | else if (buffer_delay(bh)) |
a206c817 | 720 | type = IO_DELALLOC; |
2fa24f92 CH |
721 | else |
722 | type = IO_OVERWRITE; | |
9260dc6b | 723 | |
558e6891 | 724 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 725 | done = 1; |
9260dc6b CH |
726 | continue; |
727 | } | |
728 | ||
ecff71e6 CH |
729 | lock_buffer(bh); |
730 | if (type != IO_OVERWRITE) | |
2fa24f92 | 731 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
732 | xfs_add_to_ioend(inode, bh, offset, type, |
733 | ioendp, done); | |
734 | ||
9260dc6b CH |
735 | page_dirty--; |
736 | count++; | |
737 | } else { | |
2fa24f92 | 738 | done = 1; |
1da177e4 | 739 | } |
7336cea8 | 740 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 741 | |
f6d6d4fc CH |
742 | if (uptodate && bh == head) |
743 | SetPageUptodate(page); | |
744 | ||
89f3b363 | 745 | if (count) { |
efceab1d DC |
746 | if (--wbc->nr_to_write <= 0 && |
747 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 748 | done = 1; |
1da177e4 | 749 | } |
89f3b363 | 750 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
751 | |
752 | return done; | |
10ce4444 CH |
753 | fail_unlock_page: |
754 | unlock_page(page); | |
755 | fail: | |
756 | return 1; | |
1da177e4 LT |
757 | } |
758 | ||
759 | /* | |
760 | * Convert & write out a cluster of pages in the same extent as defined | |
761 | * by mp and following the start page. | |
762 | */ | |
763 | STATIC void | |
764 | xfs_cluster_write( | |
765 | struct inode *inode, | |
766 | pgoff_t tindex, | |
207d0416 | 767 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 768 | xfs_ioend_t **ioendp, |
1da177e4 | 769 | struct writeback_control *wbc, |
1da177e4 LT |
770 | pgoff_t tlast) |
771 | { | |
10ce4444 CH |
772 | struct pagevec pvec; |
773 | int done = 0, i; | |
1da177e4 | 774 | |
10ce4444 CH |
775 | pagevec_init(&pvec, 0); |
776 | while (!done && tindex <= tlast) { | |
777 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
778 | ||
779 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 780 | break; |
10ce4444 CH |
781 | |
782 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
783 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 784 | imap, ioendp, wbc); |
10ce4444 CH |
785 | if (done) |
786 | break; | |
787 | } | |
788 | ||
789 | pagevec_release(&pvec); | |
790 | cond_resched(); | |
1da177e4 LT |
791 | } |
792 | } | |
793 | ||
3ed3a434 DC |
794 | STATIC void |
795 | xfs_vm_invalidatepage( | |
796 | struct page *page, | |
797 | unsigned long offset) | |
798 | { | |
799 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
800 | block_invalidatepage(page, offset); | |
801 | } | |
802 | ||
803 | /* | |
804 | * If the page has delalloc buffers on it, we need to punch them out before we | |
805 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
806 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
807 | * is done on that same region - the delalloc extent is returned when none is | |
808 | * supposed to be there. | |
809 | * | |
810 | * We prevent this by truncating away the delalloc regions on the page before | |
811 | * invalidating it. Because they are delalloc, we can do this without needing a | |
812 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
813 | * truncation without a transaction as there is no space left for block | |
814 | * reservation (typically why we see a ENOSPC in writeback). | |
815 | * | |
816 | * This is not a performance critical path, so for now just do the punching a | |
817 | * buffer head at a time. | |
818 | */ | |
819 | STATIC void | |
820 | xfs_aops_discard_page( | |
821 | struct page *page) | |
822 | { | |
823 | struct inode *inode = page->mapping->host; | |
824 | struct xfs_inode *ip = XFS_I(inode); | |
825 | struct buffer_head *bh, *head; | |
826 | loff_t offset = page_offset(page); | |
3ed3a434 | 827 | |
a206c817 | 828 | if (!xfs_is_delayed_page(page, IO_DELALLOC)) |
3ed3a434 DC |
829 | goto out_invalidate; |
830 | ||
e8c3753c DC |
831 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
832 | goto out_invalidate; | |
833 | ||
4f10700a | 834 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
835 | "page discard on page %p, inode 0x%llx, offset %llu.", |
836 | page, ip->i_ino, offset); | |
837 | ||
838 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
839 | bh = head = page_buffers(page); | |
840 | do { | |
3ed3a434 | 841 | int error; |
c726de44 | 842 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
843 | |
844 | if (!buffer_delay(bh)) | |
845 | goto next_buffer; | |
846 | ||
c726de44 DC |
847 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
848 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
849 | if (error) { |
850 | /* something screwed, just bail */ | |
e8c3753c | 851 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 852 | xfs_alert(ip->i_mount, |
3ed3a434 | 853 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 854 | } |
3ed3a434 DC |
855 | break; |
856 | } | |
857 | next_buffer: | |
c726de44 | 858 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
859 | |
860 | } while ((bh = bh->b_this_page) != head); | |
861 | ||
862 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
863 | out_invalidate: | |
864 | xfs_vm_invalidatepage(page, 0); | |
865 | return; | |
866 | } | |
867 | ||
1da177e4 | 868 | /* |
89f3b363 CH |
869 | * Write out a dirty page. |
870 | * | |
871 | * For delalloc space on the page we need to allocate space and flush it. | |
872 | * For unwritten space on the page we need to start the conversion to | |
873 | * regular allocated space. | |
89f3b363 | 874 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 875 | */ |
1da177e4 | 876 | STATIC int |
89f3b363 CH |
877 | xfs_vm_writepage( |
878 | struct page *page, | |
879 | struct writeback_control *wbc) | |
1da177e4 | 880 | { |
89f3b363 | 881 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 882 | struct buffer_head *bh, *head; |
207d0416 | 883 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 884 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 885 | loff_t offset; |
f6d6d4fc | 886 | unsigned int type; |
1da177e4 | 887 | __uint64_t end_offset; |
bd1556a1 | 888 | pgoff_t end_index, last_index; |
ed1e7b7e | 889 | ssize_t len; |
a206c817 | 890 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 891 | int count = 0; |
a206c817 | 892 | int nonblocking = 0; |
89f3b363 CH |
893 | |
894 | trace_xfs_writepage(inode, page, 0); | |
895 | ||
20cb52eb CH |
896 | ASSERT(page_has_buffers(page)); |
897 | ||
89f3b363 CH |
898 | /* |
899 | * Refuse to write the page out if we are called from reclaim context. | |
900 | * | |
d4f7a5cb CH |
901 | * This avoids stack overflows when called from deeply used stacks in |
902 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
903 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 904 | * |
94054fa3 MG |
905 | * This should never happen except in the case of a VM regression so |
906 | * warn about it. | |
89f3b363 | 907 | */ |
94054fa3 MG |
908 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
909 | PF_MEMALLOC)) | |
b5420f23 | 910 | goto redirty; |
1da177e4 | 911 | |
89f3b363 | 912 | /* |
680a647b CH |
913 | * Given that we do not allow direct reclaim to call us, we should |
914 | * never be called while in a filesystem transaction. | |
89f3b363 | 915 | */ |
680a647b | 916 | if (WARN_ON(current->flags & PF_FSTRANS)) |
b5420f23 | 917 | goto redirty; |
89f3b363 | 918 | |
1da177e4 LT |
919 | /* Is this page beyond the end of the file? */ |
920 | offset = i_size_read(inode); | |
921 | end_index = offset >> PAGE_CACHE_SHIFT; | |
922 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
923 | if (page->index >= end_index) { | |
924 | if ((page->index >= end_index + 1) || | |
925 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 926 | unlock_page(page); |
19d5bcf3 | 927 | return 0; |
1da177e4 LT |
928 | } |
929 | } | |
930 | ||
f6d6d4fc | 931 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
932 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
933 | offset); | |
24e17b5f | 934 | len = 1 << inode->i_blkbits; |
24e17b5f | 935 | |
24e17b5f | 936 | bh = head = page_buffers(page); |
f6d6d4fc | 937 | offset = page_offset(page); |
a206c817 CH |
938 | type = IO_OVERWRITE; |
939 | ||
dbcdde3e | 940 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 941 | nonblocking = 1; |
f6d6d4fc | 942 | |
1da177e4 | 943 | do { |
6ac7248e CH |
944 | int new_ioend = 0; |
945 | ||
1da177e4 LT |
946 | if (offset >= end_offset) |
947 | break; | |
948 | if (!buffer_uptodate(bh)) | |
949 | uptodate = 0; | |
1da177e4 | 950 | |
3d9b02e3 | 951 | /* |
ece413f5 CH |
952 | * set_page_dirty dirties all buffers in a page, independent |
953 | * of their state. The dirty state however is entirely | |
954 | * meaningless for holes (!mapped && uptodate), so skip | |
955 | * buffers covering holes here. | |
3d9b02e3 ES |
956 | */ |
957 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
958 | imap_valid = 0; |
959 | continue; | |
960 | } | |
961 | ||
aeea1b1f CH |
962 | if (buffer_unwritten(bh)) { |
963 | if (type != IO_UNWRITTEN) { | |
964 | type = IO_UNWRITTEN; | |
965 | imap_valid = 0; | |
1da177e4 | 966 | } |
aeea1b1f CH |
967 | } else if (buffer_delay(bh)) { |
968 | if (type != IO_DELALLOC) { | |
969 | type = IO_DELALLOC; | |
970 | imap_valid = 0; | |
1da177e4 | 971 | } |
89f3b363 | 972 | } else if (buffer_uptodate(bh)) { |
a206c817 CH |
973 | if (type != IO_OVERWRITE) { |
974 | type = IO_OVERWRITE; | |
85da94c6 CH |
975 | imap_valid = 0; |
976 | } | |
aeea1b1f CH |
977 | } else { |
978 | if (PageUptodate(page)) { | |
979 | ASSERT(buffer_mapped(bh)); | |
980 | imap_valid = 0; | |
6c4fe19f | 981 | } |
aeea1b1f CH |
982 | continue; |
983 | } | |
d5cb48aa | 984 | |
aeea1b1f CH |
985 | if (imap_valid) |
986 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
987 | if (!imap_valid) { | |
988 | /* | |
989 | * If we didn't have a valid mapping then we need to | |
990 | * put the new mapping into a separate ioend structure. | |
991 | * This ensures non-contiguous extents always have | |
992 | * separate ioends, which is particularly important | |
993 | * for unwritten extent conversion at I/O completion | |
994 | * time. | |
995 | */ | |
996 | new_ioend = 1; | |
997 | err = xfs_map_blocks(inode, offset, &imap, type, | |
998 | nonblocking); | |
999 | if (err) | |
1000 | goto error; | |
1001 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1002 | } | |
1003 | if (imap_valid) { | |
ecff71e6 CH |
1004 | lock_buffer(bh); |
1005 | if (type != IO_OVERWRITE) | |
aeea1b1f CH |
1006 | xfs_map_at_offset(inode, bh, &imap, offset); |
1007 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1008 | new_ioend); | |
1009 | count++; | |
1da177e4 | 1010 | } |
f6d6d4fc CH |
1011 | |
1012 | if (!iohead) | |
1013 | iohead = ioend; | |
1014 | ||
1015 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1016 | |
1017 | if (uptodate && bh == head) | |
1018 | SetPageUptodate(page); | |
1019 | ||
89f3b363 | 1020 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1021 | |
558e6891 | 1022 | if (ioend && imap_valid) { |
bd1556a1 CH |
1023 | xfs_off_t end_index; |
1024 | ||
1025 | end_index = imap.br_startoff + imap.br_blockcount; | |
1026 | ||
1027 | /* to bytes */ | |
1028 | end_index <<= inode->i_blkbits; | |
1029 | ||
1030 | /* to pages */ | |
1031 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1032 | ||
1033 | /* check against file size */ | |
1034 | if (end_index > last_index) | |
1035 | end_index = last_index; | |
8699bb0a | 1036 | |
207d0416 | 1037 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1038 | wbc, end_index); |
1da177e4 LT |
1039 | } |
1040 | ||
f6d6d4fc | 1041 | if (iohead) |
06342cf8 | 1042 | xfs_submit_ioend(wbc, iohead); |
f6d6d4fc | 1043 | |
89f3b363 | 1044 | return 0; |
1da177e4 LT |
1045 | |
1046 | error: | |
f6d6d4fc CH |
1047 | if (iohead) |
1048 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1049 | |
b5420f23 CH |
1050 | if (err == -EAGAIN) |
1051 | goto redirty; | |
1052 | ||
20cb52eb | 1053 | xfs_aops_discard_page(page); |
89f3b363 CH |
1054 | ClearPageUptodate(page); |
1055 | unlock_page(page); | |
1da177e4 | 1056 | return err; |
f51623b2 | 1057 | |
b5420f23 | 1058 | redirty: |
f51623b2 NS |
1059 | redirty_page_for_writepage(wbc, page); |
1060 | unlock_page(page); | |
1061 | return 0; | |
f51623b2 NS |
1062 | } |
1063 | ||
7d4fb40a NS |
1064 | STATIC int |
1065 | xfs_vm_writepages( | |
1066 | struct address_space *mapping, | |
1067 | struct writeback_control *wbc) | |
1068 | { | |
b3aea4ed | 1069 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1070 | return generic_writepages(mapping, wbc); |
1071 | } | |
1072 | ||
f51623b2 NS |
1073 | /* |
1074 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1075 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1076 | * have buffer heads in this call. |
1077 | * | |
89f3b363 | 1078 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1079 | */ |
1080 | STATIC int | |
238f4c54 | 1081 | xfs_vm_releasepage( |
f51623b2 NS |
1082 | struct page *page, |
1083 | gfp_t gfp_mask) | |
1084 | { | |
20cb52eb | 1085 | int delalloc, unwritten; |
f51623b2 | 1086 | |
89f3b363 | 1087 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1088 | |
20cb52eb | 1089 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1090 | |
89f3b363 | 1091 | if (WARN_ON(delalloc)) |
f51623b2 | 1092 | return 0; |
89f3b363 | 1093 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1094 | return 0; |
1095 | ||
f51623b2 NS |
1096 | return try_to_free_buffers(page); |
1097 | } | |
1098 | ||
1da177e4 | 1099 | STATIC int |
c2536668 | 1100 | __xfs_get_blocks( |
1da177e4 LT |
1101 | struct inode *inode, |
1102 | sector_t iblock, | |
1da177e4 LT |
1103 | struct buffer_head *bh_result, |
1104 | int create, | |
f2bde9b8 | 1105 | int direct) |
1da177e4 | 1106 | { |
a206c817 CH |
1107 | struct xfs_inode *ip = XFS_I(inode); |
1108 | struct xfs_mount *mp = ip->i_mount; | |
1109 | xfs_fileoff_t offset_fsb, end_fsb; | |
1110 | int error = 0; | |
1111 | int lockmode = 0; | |
207d0416 | 1112 | struct xfs_bmbt_irec imap; |
a206c817 | 1113 | int nimaps = 1; |
fdc7ed75 NS |
1114 | xfs_off_t offset; |
1115 | ssize_t size; | |
207d0416 | 1116 | int new = 0; |
a206c817 CH |
1117 | |
1118 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1119 | return -XFS_ERROR(EIO); | |
1da177e4 | 1120 | |
fdc7ed75 | 1121 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1122 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1123 | size = bh_result->b_size; | |
364f358a LM |
1124 | |
1125 | if (!create && direct && offset >= i_size_read(inode)) | |
1126 | return 0; | |
1127 | ||
a206c817 CH |
1128 | if (create) { |
1129 | lockmode = XFS_ILOCK_EXCL; | |
1130 | xfs_ilock(ip, lockmode); | |
1131 | } else { | |
1132 | lockmode = xfs_ilock_map_shared(ip); | |
1133 | } | |
f2bde9b8 | 1134 | |
a206c817 CH |
1135 | ASSERT(offset <= mp->m_maxioffset); |
1136 | if (offset + size > mp->m_maxioffset) | |
1137 | size = mp->m_maxioffset - offset; | |
1138 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); | |
1139 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1140 | ||
5c8ed202 DC |
1141 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1142 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1143 | if (error) |
a206c817 CH |
1144 | goto out_unlock; |
1145 | ||
1146 | if (create && | |
1147 | (!nimaps || | |
1148 | (imap.br_startblock == HOLESTARTBLOCK || | |
1149 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
1150 | if (direct) { | |
1151 | error = xfs_iomap_write_direct(ip, offset, size, | |
1152 | &imap, nimaps); | |
1153 | } else { | |
1154 | error = xfs_iomap_write_delay(ip, offset, size, &imap); | |
1155 | } | |
1156 | if (error) | |
1157 | goto out_unlock; | |
1158 | ||
1159 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1160 | } else if (nimaps) { | |
1161 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
1162 | } else { | |
1163 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1164 | goto out_unlock; | |
1165 | } | |
1166 | xfs_iunlock(ip, lockmode); | |
1da177e4 | 1167 | |
207d0416 CH |
1168 | if (imap.br_startblock != HOLESTARTBLOCK && |
1169 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1170 | /* |
1171 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1172 | * the read case (treat as if we're reading into a hole). |
1173 | */ | |
207d0416 CH |
1174 | if (create || !ISUNWRITTEN(&imap)) |
1175 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1176 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1177 | if (direct) |
1178 | bh_result->b_private = inode; | |
1179 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1180 | } |
1181 | } | |
1182 | ||
c2536668 NS |
1183 | /* |
1184 | * If this is a realtime file, data may be on a different device. | |
1185 | * to that pointed to from the buffer_head b_bdev currently. | |
1186 | */ | |
046f1685 | 1187 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1188 | |
c2536668 | 1189 | /* |
549054af DC |
1190 | * If we previously allocated a block out beyond eof and we are now |
1191 | * coming back to use it then we will need to flag it as new even if it | |
1192 | * has a disk address. | |
1193 | * | |
1194 | * With sub-block writes into unwritten extents we also need to mark | |
1195 | * the buffer as new so that the unwritten parts of the buffer gets | |
1196 | * correctly zeroed. | |
1da177e4 LT |
1197 | */ |
1198 | if (create && | |
1199 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1200 | (offset >= i_size_read(inode)) || |
207d0416 | 1201 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1202 | set_buffer_new(bh_result); |
1da177e4 | 1203 | |
207d0416 | 1204 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1205 | BUG_ON(direct); |
1206 | if (create) { | |
1207 | set_buffer_uptodate(bh_result); | |
1208 | set_buffer_mapped(bh_result); | |
1209 | set_buffer_delay(bh_result); | |
1210 | } | |
1211 | } | |
1212 | ||
2b8f12b7 CH |
1213 | /* |
1214 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1215 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1216 | */ | |
c2536668 | 1217 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1218 | xfs_off_t mapping_size; |
1219 | ||
1220 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1221 | mapping_size <<= inode->i_blkbits; | |
1222 | ||
1223 | ASSERT(mapping_size > 0); | |
1224 | if (mapping_size > size) | |
1225 | mapping_size = size; | |
1226 | if (mapping_size > LONG_MAX) | |
1227 | mapping_size = LONG_MAX; | |
1228 | ||
1229 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1230 | } |
1231 | ||
1232 | return 0; | |
a206c817 CH |
1233 | |
1234 | out_unlock: | |
1235 | xfs_iunlock(ip, lockmode); | |
1236 | return -error; | |
1da177e4 LT |
1237 | } |
1238 | ||
1239 | int | |
c2536668 | 1240 | xfs_get_blocks( |
1da177e4 LT |
1241 | struct inode *inode, |
1242 | sector_t iblock, | |
1243 | struct buffer_head *bh_result, | |
1244 | int create) | |
1245 | { | |
f2bde9b8 | 1246 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1247 | } |
1248 | ||
1249 | STATIC int | |
e4c573bb | 1250 | xfs_get_blocks_direct( |
1da177e4 LT |
1251 | struct inode *inode, |
1252 | sector_t iblock, | |
1da177e4 LT |
1253 | struct buffer_head *bh_result, |
1254 | int create) | |
1255 | { | |
f2bde9b8 | 1256 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1257 | } |
1258 | ||
209fb87a CH |
1259 | /* |
1260 | * Complete a direct I/O write request. | |
1261 | * | |
1262 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1263 | * need to issue a transaction to convert the range from unwritten to written | |
1264 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1265 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1266 | * request this handler is called from interrupt context, from which we |
1267 | * can't start transactions. In that case offload the I/O completion to | |
1268 | * the workqueues we also use for buffered I/O completion. | |
1269 | */ | |
f0973863 | 1270 | STATIC void |
209fb87a CH |
1271 | xfs_end_io_direct_write( |
1272 | struct kiocb *iocb, | |
1273 | loff_t offset, | |
1274 | ssize_t size, | |
1275 | void *private, | |
1276 | int ret, | |
1277 | bool is_async) | |
f0973863 | 1278 | { |
209fb87a | 1279 | struct xfs_ioend *ioend = iocb->private; |
f0973863 CH |
1280 | |
1281 | /* | |
209fb87a CH |
1282 | * blockdev_direct_IO can return an error even after the I/O |
1283 | * completion handler was called. Thus we need to protect | |
1284 | * against double-freeing. | |
f0973863 | 1285 | */ |
209fb87a CH |
1286 | iocb->private = NULL; |
1287 | ||
ba87ea69 LM |
1288 | ioend->io_offset = offset; |
1289 | ioend->io_size = size; | |
c859cdd1 CH |
1290 | ioend->io_iocb = iocb; |
1291 | ioend->io_result = ret; | |
209fb87a CH |
1292 | if (private && size > 0) |
1293 | ioend->io_type = IO_UNWRITTEN; | |
1294 | ||
1295 | if (is_async) { | |
c859cdd1 | 1296 | ioend->io_isasync = 1; |
209fb87a | 1297 | xfs_finish_ioend(ioend); |
f0973863 | 1298 | } else { |
209fb87a | 1299 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1300 | } |
f0973863 CH |
1301 | } |
1302 | ||
1da177e4 | 1303 | STATIC ssize_t |
e4c573bb | 1304 | xfs_vm_direct_IO( |
1da177e4 LT |
1305 | int rw, |
1306 | struct kiocb *iocb, | |
1307 | const struct iovec *iov, | |
1308 | loff_t offset, | |
1309 | unsigned long nr_segs) | |
1310 | { | |
209fb87a CH |
1311 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1312 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
1313 | ssize_t ret; | |
1314 | ||
1315 | if (rw & WRITE) { | |
a206c817 | 1316 | iocb->private = xfs_alloc_ioend(inode, IO_DIRECT); |
209fb87a | 1317 | |
eafdc7d1 CH |
1318 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1319 | offset, nr_segs, | |
1320 | xfs_get_blocks_direct, | |
1321 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a CH |
1322 | if (ret != -EIOCBQUEUED && iocb->private) |
1323 | xfs_destroy_ioend(iocb->private); | |
1324 | } else { | |
eafdc7d1 CH |
1325 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1326 | offset, nr_segs, | |
1327 | xfs_get_blocks_direct, | |
1328 | NULL, NULL, 0); | |
209fb87a | 1329 | } |
f0973863 | 1330 | |
f0973863 | 1331 | return ret; |
1da177e4 LT |
1332 | } |
1333 | ||
fa9b227e CH |
1334 | STATIC void |
1335 | xfs_vm_write_failed( | |
1336 | struct address_space *mapping, | |
1337 | loff_t to) | |
1338 | { | |
1339 | struct inode *inode = mapping->host; | |
1340 | ||
1341 | if (to > inode->i_size) { | |
c726de44 DC |
1342 | /* |
1343 | * punch out the delalloc blocks we have already allocated. We | |
1344 | * don't call xfs_setattr() to do this as we may be in the | |
1345 | * middle of a multi-iovec write and so the vfs inode->i_size | |
1346 | * will not match the xfs ip->i_size and so it will zero too | |
1347 | * much. Hence we jus truncate the page cache to zero what is | |
1348 | * necessary and punch the delalloc blocks directly. | |
1349 | */ | |
1350 | struct xfs_inode *ip = XFS_I(inode); | |
1351 | xfs_fileoff_t start_fsb; | |
1352 | xfs_fileoff_t end_fsb; | |
1353 | int error; | |
1354 | ||
1355 | truncate_pagecache(inode, to, inode->i_size); | |
1356 | ||
1357 | /* | |
1358 | * Check if there are any blocks that are outside of i_size | |
1359 | * that need to be trimmed back. | |
1360 | */ | |
1361 | start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1; | |
1362 | end_fsb = XFS_B_TO_FSB(ip->i_mount, to); | |
1363 | if (end_fsb <= start_fsb) | |
1364 | return; | |
1365 | ||
1366 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1367 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1368 | end_fsb - start_fsb); | |
1369 | if (error) { | |
1370 | /* something screwed, just bail */ | |
1371 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
4f10700a | 1372 | xfs_alert(ip->i_mount, |
c726de44 DC |
1373 | "xfs_vm_write_failed: unable to clean up ino %lld", |
1374 | ip->i_ino); | |
1375 | } | |
1376 | } | |
1377 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
fa9b227e CH |
1378 | } |
1379 | } | |
1380 | ||
f51623b2 | 1381 | STATIC int |
d79689c7 | 1382 | xfs_vm_write_begin( |
f51623b2 | 1383 | struct file *file, |
d79689c7 NP |
1384 | struct address_space *mapping, |
1385 | loff_t pos, | |
1386 | unsigned len, | |
1387 | unsigned flags, | |
1388 | struct page **pagep, | |
1389 | void **fsdata) | |
f51623b2 | 1390 | { |
155130a4 CH |
1391 | int ret; |
1392 | ||
1393 | ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS, | |
1394 | pagep, xfs_get_blocks); | |
fa9b227e CH |
1395 | if (unlikely(ret)) |
1396 | xfs_vm_write_failed(mapping, pos + len); | |
1397 | return ret; | |
1398 | } | |
1399 | ||
1400 | STATIC int | |
1401 | xfs_vm_write_end( | |
1402 | struct file *file, | |
1403 | struct address_space *mapping, | |
1404 | loff_t pos, | |
1405 | unsigned len, | |
1406 | unsigned copied, | |
1407 | struct page *page, | |
1408 | void *fsdata) | |
1409 | { | |
1410 | int ret; | |
155130a4 | 1411 | |
fa9b227e CH |
1412 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
1413 | if (unlikely(ret < len)) | |
1414 | xfs_vm_write_failed(mapping, pos + len); | |
155130a4 | 1415 | return ret; |
f51623b2 | 1416 | } |
1da177e4 LT |
1417 | |
1418 | STATIC sector_t | |
e4c573bb | 1419 | xfs_vm_bmap( |
1da177e4 LT |
1420 | struct address_space *mapping, |
1421 | sector_t block) | |
1422 | { | |
1423 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1424 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1425 | |
cca28fb8 | 1426 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1427 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1428 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1429 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1430 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1431 | } |
1432 | ||
1433 | STATIC int | |
e4c573bb | 1434 | xfs_vm_readpage( |
1da177e4 LT |
1435 | struct file *unused, |
1436 | struct page *page) | |
1437 | { | |
c2536668 | 1438 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1439 | } |
1440 | ||
1441 | STATIC int | |
e4c573bb | 1442 | xfs_vm_readpages( |
1da177e4 LT |
1443 | struct file *unused, |
1444 | struct address_space *mapping, | |
1445 | struct list_head *pages, | |
1446 | unsigned nr_pages) | |
1447 | { | |
c2536668 | 1448 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1449 | } |
1450 | ||
f5e54d6e | 1451 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1452 | .readpage = xfs_vm_readpage, |
1453 | .readpages = xfs_vm_readpages, | |
1454 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1455 | .writepages = xfs_vm_writepages, |
238f4c54 NS |
1456 | .releasepage = xfs_vm_releasepage, |
1457 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1458 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1459 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1460 | .bmap = xfs_vm_bmap, |
1461 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1462 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1463 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1464 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1465 | }; |