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" | |
281627df | 29 | #include "xfs_inode_item.h" |
a844f451 | 30 | #include "xfs_alloc.h" |
1da177e4 LT |
31 | #include "xfs_error.h" |
32 | #include "xfs_rw.h" | |
33 | #include "xfs_iomap.h" | |
739bfb2a | 34 | #include "xfs_vnodeops.h" |
0b1b213f | 35 | #include "xfs_trace.h" |
3ed3a434 | 36 | #include "xfs_bmap.h" |
5a0e3ad6 | 37 | #include <linux/gfp.h> |
1da177e4 | 38 | #include <linux/mpage.h> |
10ce4444 | 39 | #include <linux/pagevec.h> |
1da177e4 LT |
40 | #include <linux/writeback.h> |
41 | ||
0b1b213f | 42 | void |
f51623b2 NS |
43 | xfs_count_page_state( |
44 | struct page *page, | |
45 | int *delalloc, | |
f51623b2 NS |
46 | int *unwritten) |
47 | { | |
48 | struct buffer_head *bh, *head; | |
49 | ||
20cb52eb | 50 | *delalloc = *unwritten = 0; |
f51623b2 NS |
51 | |
52 | bh = head = page_buffers(page); | |
53 | do { | |
20cb52eb | 54 | if (buffer_unwritten(bh)) |
f51623b2 NS |
55 | (*unwritten) = 1; |
56 | else if (buffer_delay(bh)) | |
57 | (*delalloc) = 1; | |
58 | } while ((bh = bh->b_this_page) != head); | |
59 | } | |
60 | ||
6214ed44 CH |
61 | STATIC struct block_device * |
62 | xfs_find_bdev_for_inode( | |
046f1685 | 63 | struct inode *inode) |
6214ed44 | 64 | { |
046f1685 | 65 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
66 | struct xfs_mount *mp = ip->i_mount; |
67 | ||
71ddabb9 | 68 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
69 | return mp->m_rtdev_targp->bt_bdev; |
70 | else | |
71 | return mp->m_ddev_targp->bt_bdev; | |
72 | } | |
73 | ||
f6d6d4fc CH |
74 | /* |
75 | * We're now finished for good with this ioend structure. | |
76 | * Update the page state via the associated buffer_heads, | |
77 | * release holds on the inode and bio, and finally free | |
78 | * up memory. Do not use the ioend after this. | |
79 | */ | |
0829c360 CH |
80 | STATIC void |
81 | xfs_destroy_ioend( | |
82 | xfs_ioend_t *ioend) | |
83 | { | |
f6d6d4fc CH |
84 | struct buffer_head *bh, *next; |
85 | ||
86 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
87 | next = bh->b_private; | |
7d04a335 | 88 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 89 | } |
583fa586 | 90 | |
c859cdd1 | 91 | if (ioend->io_iocb) { |
04f658ee CH |
92 | if (ioend->io_isasync) { |
93 | aio_complete(ioend->io_iocb, ioend->io_error ? | |
94 | ioend->io_error : ioend->io_result, 0); | |
95 | } | |
c859cdd1 CH |
96 | inode_dio_done(ioend->io_inode); |
97 | } | |
4a06fd26 | 98 | |
0829c360 CH |
99 | mempool_free(ioend, xfs_ioend_pool); |
100 | } | |
101 | ||
fc0063c4 CH |
102 | /* |
103 | * Fast and loose check if this write could update the on-disk inode size. | |
104 | */ | |
105 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
106 | { | |
107 | return ioend->io_offset + ioend->io_size > | |
108 | XFS_I(ioend->io_inode)->i_d.di_size; | |
109 | } | |
110 | ||
281627df CH |
111 | STATIC int |
112 | xfs_setfilesize_trans_alloc( | |
113 | struct xfs_ioend *ioend) | |
114 | { | |
115 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
116 | struct xfs_trans *tp; | |
117 | int error; | |
118 | ||
119 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
120 | ||
121 | error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0); | |
122 | if (error) { | |
123 | xfs_trans_cancel(tp, 0); | |
124 | return error; | |
125 | } | |
126 | ||
127 | ioend->io_append_trans = tp; | |
128 | ||
129 | /* | |
130 | * We hand off the transaction to the completion thread now, so | |
131 | * clear the flag here. | |
132 | */ | |
133 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
134 | return 0; | |
135 | } | |
136 | ||
ba87ea69 | 137 | /* |
2813d682 | 138 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 139 | */ |
281627df | 140 | STATIC int |
ba87ea69 | 141 | xfs_setfilesize( |
aa6bf01d | 142 | struct xfs_ioend *ioend) |
ba87ea69 | 143 | { |
aa6bf01d | 144 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
281627df | 145 | struct xfs_trans *tp = ioend->io_append_trans; |
ba87ea69 | 146 | xfs_fsize_t isize; |
ba87ea69 | 147 | |
281627df CH |
148 | /* |
149 | * The transaction was allocated in the I/O submission thread, | |
150 | * thus we need to mark ourselves as beeing in a transaction | |
151 | * manually. | |
152 | */ | |
153 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
154 | ||
aa6bf01d | 155 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
6923e686 | 156 | isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size); |
281627df CH |
157 | if (!isize) { |
158 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
159 | xfs_trans_cancel(tp, 0); | |
160 | return 0; | |
ba87ea69 LM |
161 | } |
162 | ||
281627df CH |
163 | trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); |
164 | ||
165 | ip->i_d.di_size = isize; | |
166 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
167 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
168 | ||
169 | return xfs_trans_commit(tp, 0); | |
77d7a0c2 DC |
170 | } |
171 | ||
172 | /* | |
209fb87a | 173 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
174 | * |
175 | * If there is no work to do we might as well call it a day and free the | |
176 | * ioend right now. | |
77d7a0c2 DC |
177 | */ |
178 | STATIC void | |
179 | xfs_finish_ioend( | |
209fb87a | 180 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
181 | { |
182 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
183 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
184 | ||
209fb87a | 185 | if (ioend->io_type == IO_UNWRITTEN) |
aa6bf01d | 186 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
281627df | 187 | else if (ioend->io_append_trans) |
aa6bf01d | 188 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
189 | else |
190 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 191 | } |
ba87ea69 LM |
192 | } |
193 | ||
0829c360 | 194 | /* |
5ec4fabb | 195 | * IO write completion. |
f6d6d4fc CH |
196 | */ |
197 | STATIC void | |
5ec4fabb | 198 | xfs_end_io( |
77d7a0c2 | 199 | struct work_struct *work) |
0829c360 | 200 | { |
77d7a0c2 DC |
201 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
202 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 203 | int error = 0; |
ba87ea69 | 204 | |
04f658ee | 205 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 206 | ioend->io_error = -EIO; |
04f658ee CH |
207 | goto done; |
208 | } | |
209 | if (ioend->io_error) | |
210 | goto done; | |
211 | ||
5ec4fabb CH |
212 | /* |
213 | * For unwritten extents we need to issue transactions to convert a | |
214 | * range to normal written extens after the data I/O has finished. | |
215 | */ | |
04f658ee | 216 | if (ioend->io_type == IO_UNWRITTEN) { |
281627df CH |
217 | /* |
218 | * For buffered I/O we never preallocate a transaction when | |
219 | * doing the unwritten extent conversion, but for direct I/O | |
220 | * we do not know if we are converting an unwritten extent | |
221 | * or not at the point where we preallocate the transaction. | |
222 | */ | |
223 | if (ioend->io_append_trans) { | |
224 | ASSERT(ioend->io_isdirect); | |
225 | ||
226 | current_set_flags_nested( | |
227 | &ioend->io_append_trans->t_pflags, PF_FSTRANS); | |
228 | xfs_trans_cancel(ioend->io_append_trans, 0); | |
229 | } | |
230 | ||
5ec4fabb CH |
231 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
232 | ioend->io_size); | |
04f658ee CH |
233 | if (error) { |
234 | ioend->io_error = -error; | |
235 | goto done; | |
236 | } | |
281627df CH |
237 | } else if (ioend->io_append_trans) { |
238 | error = xfs_setfilesize(ioend); | |
239 | if (error) | |
240 | ioend->io_error = -error; | |
84803fb7 | 241 | } else { |
281627df | 242 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 243 | } |
ba87ea69 | 244 | |
04f658ee | 245 | done: |
aa6bf01d | 246 | xfs_destroy_ioend(ioend); |
c626d174 DC |
247 | } |
248 | ||
209fb87a CH |
249 | /* |
250 | * Call IO completion handling in caller context on the final put of an ioend. | |
251 | */ | |
252 | STATIC void | |
253 | xfs_finish_ioend_sync( | |
254 | struct xfs_ioend *ioend) | |
255 | { | |
256 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
257 | xfs_end_io(&ioend->io_work); | |
258 | } | |
259 | ||
0829c360 CH |
260 | /* |
261 | * Allocate and initialise an IO completion structure. | |
262 | * We need to track unwritten extent write completion here initially. | |
263 | * We'll need to extend this for updating the ondisk inode size later | |
264 | * (vs. incore size). | |
265 | */ | |
266 | STATIC xfs_ioend_t * | |
267 | xfs_alloc_ioend( | |
f6d6d4fc CH |
268 | struct inode *inode, |
269 | unsigned int type) | |
0829c360 CH |
270 | { |
271 | xfs_ioend_t *ioend; | |
272 | ||
273 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
274 | ||
275 | /* | |
276 | * Set the count to 1 initially, which will prevent an I/O | |
277 | * completion callback from happening before we have started | |
278 | * all the I/O from calling the completion routine too early. | |
279 | */ | |
280 | atomic_set(&ioend->io_remaining, 1); | |
c859cdd1 | 281 | ioend->io_isasync = 0; |
281627df | 282 | ioend->io_isdirect = 0; |
7d04a335 | 283 | ioend->io_error = 0; |
f6d6d4fc CH |
284 | ioend->io_list = NULL; |
285 | ioend->io_type = type; | |
b677c210 | 286 | ioend->io_inode = inode; |
c1a073bd | 287 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 288 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
289 | ioend->io_offset = 0; |
290 | ioend->io_size = 0; | |
fb511f21 CH |
291 | ioend->io_iocb = NULL; |
292 | ioend->io_result = 0; | |
281627df | 293 | ioend->io_append_trans = NULL; |
0829c360 | 294 | |
5ec4fabb | 295 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
296 | return ioend; |
297 | } | |
298 | ||
1da177e4 LT |
299 | STATIC int |
300 | xfs_map_blocks( | |
301 | struct inode *inode, | |
302 | loff_t offset, | |
207d0416 | 303 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
304 | int type, |
305 | int nonblocking) | |
1da177e4 | 306 | { |
a206c817 CH |
307 | struct xfs_inode *ip = XFS_I(inode); |
308 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 309 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
310 | xfs_fileoff_t offset_fsb, end_fsb; |
311 | int error = 0; | |
a206c817 CH |
312 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
313 | int nimaps = 1; | |
314 | ||
315 | if (XFS_FORCED_SHUTDOWN(mp)) | |
316 | return -XFS_ERROR(EIO); | |
317 | ||
8ff2957d | 318 | if (type == IO_UNWRITTEN) |
a206c817 | 319 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
320 | |
321 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
322 | if (nonblocking) | |
323 | return -XFS_ERROR(EAGAIN); | |
324 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
a206c817 CH |
325 | } |
326 | ||
8ff2957d CH |
327 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
328 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
a206c817 | 329 | ASSERT(offset <= mp->m_maxioffset); |
8ff2957d | 330 | |
a206c817 CH |
331 | if (offset + count > mp->m_maxioffset) |
332 | count = mp->m_maxioffset - offset; | |
333 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); | |
334 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
335 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
336 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 337 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 338 | |
8ff2957d CH |
339 | if (error) |
340 | return -XFS_ERROR(error); | |
a206c817 | 341 | |
8ff2957d CH |
342 | if (type == IO_DELALLOC && |
343 | (!nimaps || isnullstartblock(imap->br_startblock))) { | |
a206c817 CH |
344 | error = xfs_iomap_write_allocate(ip, offset, count, imap); |
345 | if (!error) | |
346 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
8ff2957d | 347 | return -XFS_ERROR(error); |
a206c817 CH |
348 | } |
349 | ||
8ff2957d CH |
350 | #ifdef DEBUG |
351 | if (type == IO_UNWRITTEN) { | |
352 | ASSERT(nimaps); | |
353 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
354 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
355 | } | |
356 | #endif | |
357 | if (nimaps) | |
358 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
359 | return 0; | |
1da177e4 LT |
360 | } |
361 | ||
b8f82a4a | 362 | STATIC int |
558e6891 | 363 | xfs_imap_valid( |
8699bb0a | 364 | struct inode *inode, |
207d0416 | 365 | struct xfs_bmbt_irec *imap, |
558e6891 | 366 | xfs_off_t offset) |
1da177e4 | 367 | { |
558e6891 | 368 | offset >>= inode->i_blkbits; |
8699bb0a | 369 | |
558e6891 CH |
370 | return offset >= imap->br_startoff && |
371 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
372 | } |
373 | ||
f6d6d4fc CH |
374 | /* |
375 | * BIO completion handler for buffered IO. | |
376 | */ | |
782e3b3b | 377 | STATIC void |
f6d6d4fc CH |
378 | xfs_end_bio( |
379 | struct bio *bio, | |
f6d6d4fc CH |
380 | int error) |
381 | { | |
382 | xfs_ioend_t *ioend = bio->bi_private; | |
383 | ||
f6d6d4fc | 384 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 385 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
386 | |
387 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
388 | bio->bi_private = NULL; |
389 | bio->bi_end_io = NULL; | |
f6d6d4fc | 390 | bio_put(bio); |
7d04a335 | 391 | |
209fb87a | 392 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
393 | } |
394 | ||
395 | STATIC void | |
396 | xfs_submit_ioend_bio( | |
06342cf8 CH |
397 | struct writeback_control *wbc, |
398 | xfs_ioend_t *ioend, | |
399 | struct bio *bio) | |
f6d6d4fc CH |
400 | { |
401 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
402 | bio->bi_private = ioend; |
403 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 404 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
405 | } |
406 | ||
407 | STATIC struct bio * | |
408 | xfs_alloc_ioend_bio( | |
409 | struct buffer_head *bh) | |
410 | { | |
f6d6d4fc | 411 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 412 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
413 | |
414 | ASSERT(bio->bi_private == NULL); | |
415 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
416 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
417 | return bio; |
418 | } | |
419 | ||
420 | STATIC void | |
421 | xfs_start_buffer_writeback( | |
422 | struct buffer_head *bh) | |
423 | { | |
424 | ASSERT(buffer_mapped(bh)); | |
425 | ASSERT(buffer_locked(bh)); | |
426 | ASSERT(!buffer_delay(bh)); | |
427 | ASSERT(!buffer_unwritten(bh)); | |
428 | ||
429 | mark_buffer_async_write(bh); | |
430 | set_buffer_uptodate(bh); | |
431 | clear_buffer_dirty(bh); | |
432 | } | |
433 | ||
434 | STATIC void | |
435 | xfs_start_page_writeback( | |
436 | struct page *page, | |
f6d6d4fc CH |
437 | int clear_dirty, |
438 | int buffers) | |
439 | { | |
440 | ASSERT(PageLocked(page)); | |
441 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 442 | if (clear_dirty) |
92132021 DC |
443 | clear_page_dirty_for_io(page); |
444 | set_page_writeback(page); | |
f6d6d4fc | 445 | unlock_page(page); |
1f7decf6 FW |
446 | /* If no buffers on the page are to be written, finish it here */ |
447 | if (!buffers) | |
f6d6d4fc | 448 | end_page_writeback(page); |
f6d6d4fc CH |
449 | } |
450 | ||
451 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
452 | { | |
453 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
454 | } | |
455 | ||
456 | /* | |
d88992f6 DC |
457 | * Submit all of the bios for all of the ioends we have saved up, covering the |
458 | * initial writepage page and also any probed pages. | |
459 | * | |
460 | * Because we may have multiple ioends spanning a page, we need to start | |
461 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
462 | * buffers as we got, then we can end up with a page that only has buffers | |
463 | * marked async write and I/O complete on can occur before we mark the other | |
464 | * buffers async write. | |
465 | * | |
466 | * The end result of this is that we trip a bug in end_page_writeback() because | |
467 | * we call it twice for the one page as the code in end_buffer_async_write() | |
468 | * assumes that all buffers on the page are started at the same time. | |
469 | * | |
470 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 471 | * buffer_heads, and then submit them for I/O on the second pass. |
f6d6d4fc CH |
472 | */ |
473 | STATIC void | |
474 | xfs_submit_ioend( | |
06342cf8 | 475 | struct writeback_control *wbc, |
f6d6d4fc CH |
476 | xfs_ioend_t *ioend) |
477 | { | |
d88992f6 | 478 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
479 | xfs_ioend_t *next; |
480 | struct buffer_head *bh; | |
481 | struct bio *bio; | |
482 | sector_t lastblock = 0; | |
483 | ||
d88992f6 DC |
484 | /* Pass 1 - start writeback */ |
485 | do { | |
486 | next = ioend->io_list; | |
221cb251 | 487 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 488 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
489 | } while ((ioend = next) != NULL); |
490 | ||
491 | /* Pass 2 - submit I/O */ | |
492 | ioend = head; | |
f6d6d4fc CH |
493 | do { |
494 | next = ioend->io_list; | |
495 | bio = NULL; | |
496 | ||
497 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
498 | |
499 | if (!bio) { | |
500 | retry: | |
501 | bio = xfs_alloc_ioend_bio(bh); | |
502 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 503 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
504 | goto retry; |
505 | } | |
506 | ||
507 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
06342cf8 | 508 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
509 | goto retry; |
510 | } | |
511 | ||
512 | lastblock = bh->b_blocknr; | |
513 | } | |
514 | if (bio) | |
06342cf8 | 515 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 516 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
517 | } while ((ioend = next) != NULL); |
518 | } | |
519 | ||
520 | /* | |
521 | * Cancel submission of all buffer_heads so far in this endio. | |
522 | * Toss the endio too. Only ever called for the initial page | |
523 | * in a writepage request, so only ever one page. | |
524 | */ | |
525 | STATIC void | |
526 | xfs_cancel_ioend( | |
527 | xfs_ioend_t *ioend) | |
528 | { | |
529 | xfs_ioend_t *next; | |
530 | struct buffer_head *bh, *next_bh; | |
531 | ||
532 | do { | |
533 | next = ioend->io_list; | |
534 | bh = ioend->io_buffer_head; | |
535 | do { | |
536 | next_bh = bh->b_private; | |
537 | clear_buffer_async_write(bh); | |
538 | unlock_buffer(bh); | |
539 | } while ((bh = next_bh) != NULL); | |
540 | ||
f6d6d4fc CH |
541 | mempool_free(ioend, xfs_ioend_pool); |
542 | } while ((ioend = next) != NULL); | |
543 | } | |
544 | ||
545 | /* | |
546 | * Test to see if we've been building up a completion structure for | |
547 | * earlier buffers -- if so, we try to append to this ioend if we | |
548 | * can, otherwise we finish off any current ioend and start another. | |
549 | * Return true if we've finished the given ioend. | |
550 | */ | |
551 | STATIC void | |
552 | xfs_add_to_ioend( | |
553 | struct inode *inode, | |
554 | struct buffer_head *bh, | |
7336cea8 | 555 | xfs_off_t offset, |
f6d6d4fc CH |
556 | unsigned int type, |
557 | xfs_ioend_t **result, | |
558 | int need_ioend) | |
559 | { | |
560 | xfs_ioend_t *ioend = *result; | |
561 | ||
562 | if (!ioend || need_ioend || type != ioend->io_type) { | |
563 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 564 | |
f6d6d4fc CH |
565 | ioend = xfs_alloc_ioend(inode, type); |
566 | ioend->io_offset = offset; | |
567 | ioend->io_buffer_head = bh; | |
568 | ioend->io_buffer_tail = bh; | |
569 | if (previous) | |
570 | previous->io_list = ioend; | |
571 | *result = ioend; | |
572 | } else { | |
573 | ioend->io_buffer_tail->b_private = bh; | |
574 | ioend->io_buffer_tail = bh; | |
575 | } | |
576 | ||
577 | bh->b_private = NULL; | |
578 | ioend->io_size += bh->b_size; | |
579 | } | |
580 | ||
87cbc49c NS |
581 | STATIC void |
582 | xfs_map_buffer( | |
046f1685 | 583 | struct inode *inode, |
87cbc49c | 584 | struct buffer_head *bh, |
207d0416 | 585 | struct xfs_bmbt_irec *imap, |
046f1685 | 586 | xfs_off_t offset) |
87cbc49c NS |
587 | { |
588 | sector_t bn; | |
8699bb0a | 589 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
590 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
591 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 592 | |
207d0416 CH |
593 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
594 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 595 | |
e513182d | 596 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 597 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 598 | |
046f1685 | 599 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
600 | |
601 | bh->b_blocknr = bn; | |
602 | set_buffer_mapped(bh); | |
603 | } | |
604 | ||
1da177e4 LT |
605 | STATIC void |
606 | xfs_map_at_offset( | |
046f1685 | 607 | struct inode *inode, |
1da177e4 | 608 | struct buffer_head *bh, |
207d0416 | 609 | struct xfs_bmbt_irec *imap, |
046f1685 | 610 | xfs_off_t offset) |
1da177e4 | 611 | { |
207d0416 CH |
612 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
613 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 614 | |
207d0416 | 615 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
616 | set_buffer_mapped(bh); |
617 | clear_buffer_delay(bh); | |
f6d6d4fc | 618 | clear_buffer_unwritten(bh); |
1da177e4 LT |
619 | } |
620 | ||
1da177e4 | 621 | /* |
10ce4444 CH |
622 | * Test if a given page is suitable for writing as part of an unwritten |
623 | * or delayed allocate extent. | |
1da177e4 | 624 | */ |
10ce4444 | 625 | STATIC int |
6ffc4db5 | 626 | xfs_check_page_type( |
10ce4444 | 627 | struct page *page, |
f6d6d4fc | 628 | unsigned int type) |
1da177e4 | 629 | { |
1da177e4 | 630 | if (PageWriteback(page)) |
10ce4444 | 631 | return 0; |
1da177e4 LT |
632 | |
633 | if (page->mapping && page_has_buffers(page)) { | |
634 | struct buffer_head *bh, *head; | |
635 | int acceptable = 0; | |
636 | ||
637 | bh = head = page_buffers(page); | |
638 | do { | |
f6d6d4fc | 639 | if (buffer_unwritten(bh)) |
6ffc4db5 | 640 | acceptable += (type == IO_UNWRITTEN); |
f6d6d4fc | 641 | else if (buffer_delay(bh)) |
6ffc4db5 | 642 | acceptable += (type == IO_DELALLOC); |
2ddee844 | 643 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
6ffc4db5 | 644 | acceptable += (type == IO_OVERWRITE); |
f6d6d4fc | 645 | else |
1da177e4 | 646 | break; |
1da177e4 LT |
647 | } while ((bh = bh->b_this_page) != head); |
648 | ||
649 | if (acceptable) | |
10ce4444 | 650 | return 1; |
1da177e4 LT |
651 | } |
652 | ||
10ce4444 | 653 | return 0; |
1da177e4 LT |
654 | } |
655 | ||
1da177e4 LT |
656 | /* |
657 | * Allocate & map buffers for page given the extent map. Write it out. | |
658 | * except for the original page of a writepage, this is called on | |
659 | * delalloc/unwritten pages only, for the original page it is possible | |
660 | * that the page has no mapping at all. | |
661 | */ | |
f6d6d4fc | 662 | STATIC int |
1da177e4 LT |
663 | xfs_convert_page( |
664 | struct inode *inode, | |
665 | struct page *page, | |
10ce4444 | 666 | loff_t tindex, |
207d0416 | 667 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 668 | xfs_ioend_t **ioendp, |
2fa24f92 | 669 | struct writeback_control *wbc) |
1da177e4 | 670 | { |
f6d6d4fc | 671 | struct buffer_head *bh, *head; |
9260dc6b CH |
672 | xfs_off_t end_offset; |
673 | unsigned long p_offset; | |
f6d6d4fc | 674 | unsigned int type; |
24e17b5f | 675 | int len, page_dirty; |
f6d6d4fc | 676 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 677 | xfs_off_t offset = page_offset(page); |
1da177e4 | 678 | |
10ce4444 CH |
679 | if (page->index != tindex) |
680 | goto fail; | |
529ae9aa | 681 | if (!trylock_page(page)) |
10ce4444 CH |
682 | goto fail; |
683 | if (PageWriteback(page)) | |
684 | goto fail_unlock_page; | |
685 | if (page->mapping != inode->i_mapping) | |
686 | goto fail_unlock_page; | |
6ffc4db5 | 687 | if (!xfs_check_page_type(page, (*ioendp)->io_type)) |
10ce4444 CH |
688 | goto fail_unlock_page; |
689 | ||
24e17b5f NS |
690 | /* |
691 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 692 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
693 | * |
694 | * Derivation: | |
695 | * | |
696 | * End offset is the highest offset that this page should represent. | |
697 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
698 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
699 | * hence give us the correct page_dirty count. On any other page, | |
700 | * it will be zero and in that case we need page_dirty to be the | |
701 | * count of buffers on the page. | |
24e17b5f | 702 | */ |
9260dc6b CH |
703 | end_offset = min_t(unsigned long long, |
704 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
705 | i_size_read(inode)); | |
706 | ||
24e17b5f | 707 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
708 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
709 | PAGE_CACHE_SIZE); | |
710 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
711 | page_dirty = p_offset / len; | |
24e17b5f | 712 | |
1da177e4 LT |
713 | bh = head = page_buffers(page); |
714 | do { | |
9260dc6b | 715 | if (offset >= end_offset) |
1da177e4 | 716 | break; |
f6d6d4fc CH |
717 | if (!buffer_uptodate(bh)) |
718 | uptodate = 0; | |
719 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
720 | done = 1; | |
1da177e4 | 721 | continue; |
f6d6d4fc CH |
722 | } |
723 | ||
2fa24f92 CH |
724 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
725 | buffer_mapped(bh)) { | |
9260dc6b | 726 | if (buffer_unwritten(bh)) |
34a52c6c | 727 | type = IO_UNWRITTEN; |
2fa24f92 | 728 | else if (buffer_delay(bh)) |
a206c817 | 729 | type = IO_DELALLOC; |
2fa24f92 CH |
730 | else |
731 | type = IO_OVERWRITE; | |
9260dc6b | 732 | |
558e6891 | 733 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 734 | done = 1; |
9260dc6b CH |
735 | continue; |
736 | } | |
737 | ||
ecff71e6 CH |
738 | lock_buffer(bh); |
739 | if (type != IO_OVERWRITE) | |
2fa24f92 | 740 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
741 | xfs_add_to_ioend(inode, bh, offset, type, |
742 | ioendp, done); | |
743 | ||
9260dc6b CH |
744 | page_dirty--; |
745 | count++; | |
746 | } else { | |
2fa24f92 | 747 | done = 1; |
1da177e4 | 748 | } |
7336cea8 | 749 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 750 | |
f6d6d4fc CH |
751 | if (uptodate && bh == head) |
752 | SetPageUptodate(page); | |
753 | ||
89f3b363 | 754 | if (count) { |
efceab1d DC |
755 | if (--wbc->nr_to_write <= 0 && |
756 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 757 | done = 1; |
1da177e4 | 758 | } |
89f3b363 | 759 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
760 | |
761 | return done; | |
10ce4444 CH |
762 | fail_unlock_page: |
763 | unlock_page(page); | |
764 | fail: | |
765 | return 1; | |
1da177e4 LT |
766 | } |
767 | ||
768 | /* | |
769 | * Convert & write out a cluster of pages in the same extent as defined | |
770 | * by mp and following the start page. | |
771 | */ | |
772 | STATIC void | |
773 | xfs_cluster_write( | |
774 | struct inode *inode, | |
775 | pgoff_t tindex, | |
207d0416 | 776 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 777 | xfs_ioend_t **ioendp, |
1da177e4 | 778 | struct writeback_control *wbc, |
1da177e4 LT |
779 | pgoff_t tlast) |
780 | { | |
10ce4444 CH |
781 | struct pagevec pvec; |
782 | int done = 0, i; | |
1da177e4 | 783 | |
10ce4444 CH |
784 | pagevec_init(&pvec, 0); |
785 | while (!done && tindex <= tlast) { | |
786 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
787 | ||
788 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 789 | break; |
10ce4444 CH |
790 | |
791 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
792 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 793 | imap, ioendp, wbc); |
10ce4444 CH |
794 | if (done) |
795 | break; | |
796 | } | |
797 | ||
798 | pagevec_release(&pvec); | |
799 | cond_resched(); | |
1da177e4 LT |
800 | } |
801 | } | |
802 | ||
3ed3a434 DC |
803 | STATIC void |
804 | xfs_vm_invalidatepage( | |
805 | struct page *page, | |
806 | unsigned long offset) | |
807 | { | |
808 | trace_xfs_invalidatepage(page->mapping->host, page, offset); | |
809 | block_invalidatepage(page, offset); | |
810 | } | |
811 | ||
812 | /* | |
813 | * If the page has delalloc buffers on it, we need to punch them out before we | |
814 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
815 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
816 | * is done on that same region - the delalloc extent is returned when none is | |
817 | * supposed to be there. | |
818 | * | |
819 | * We prevent this by truncating away the delalloc regions on the page before | |
820 | * invalidating it. Because they are delalloc, we can do this without needing a | |
821 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
822 | * truncation without a transaction as there is no space left for block | |
823 | * reservation (typically why we see a ENOSPC in writeback). | |
824 | * | |
825 | * This is not a performance critical path, so for now just do the punching a | |
826 | * buffer head at a time. | |
827 | */ | |
828 | STATIC void | |
829 | xfs_aops_discard_page( | |
830 | struct page *page) | |
831 | { | |
832 | struct inode *inode = page->mapping->host; | |
833 | struct xfs_inode *ip = XFS_I(inode); | |
834 | struct buffer_head *bh, *head; | |
835 | loff_t offset = page_offset(page); | |
3ed3a434 | 836 | |
6ffc4db5 | 837 | if (!xfs_check_page_type(page, IO_DELALLOC)) |
3ed3a434 DC |
838 | goto out_invalidate; |
839 | ||
e8c3753c DC |
840 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
841 | goto out_invalidate; | |
842 | ||
4f10700a | 843 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
844 | "page discard on page %p, inode 0x%llx, offset %llu.", |
845 | page, ip->i_ino, offset); | |
846 | ||
847 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
848 | bh = head = page_buffers(page); | |
849 | do { | |
3ed3a434 | 850 | int error; |
c726de44 | 851 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
852 | |
853 | if (!buffer_delay(bh)) | |
854 | goto next_buffer; | |
855 | ||
c726de44 DC |
856 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
857 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
858 | if (error) { |
859 | /* something screwed, just bail */ | |
e8c3753c | 860 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 861 | xfs_alert(ip->i_mount, |
3ed3a434 | 862 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 863 | } |
3ed3a434 DC |
864 | break; |
865 | } | |
866 | next_buffer: | |
c726de44 | 867 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
868 | |
869 | } while ((bh = bh->b_this_page) != head); | |
870 | ||
871 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
872 | out_invalidate: | |
873 | xfs_vm_invalidatepage(page, 0); | |
874 | return; | |
875 | } | |
876 | ||
1da177e4 | 877 | /* |
89f3b363 CH |
878 | * Write out a dirty page. |
879 | * | |
880 | * For delalloc space on the page we need to allocate space and flush it. | |
881 | * For unwritten space on the page we need to start the conversion to | |
882 | * regular allocated space. | |
89f3b363 | 883 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 884 | */ |
1da177e4 | 885 | STATIC int |
89f3b363 CH |
886 | xfs_vm_writepage( |
887 | struct page *page, | |
888 | struct writeback_control *wbc) | |
1da177e4 | 889 | { |
89f3b363 | 890 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 891 | struct buffer_head *bh, *head; |
207d0416 | 892 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 893 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 894 | loff_t offset; |
f6d6d4fc | 895 | unsigned int type; |
1da177e4 | 896 | __uint64_t end_offset; |
bd1556a1 | 897 | pgoff_t end_index, last_index; |
ed1e7b7e | 898 | ssize_t len; |
a206c817 | 899 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 900 | int count = 0; |
a206c817 | 901 | int nonblocking = 0; |
89f3b363 CH |
902 | |
903 | trace_xfs_writepage(inode, page, 0); | |
904 | ||
20cb52eb CH |
905 | ASSERT(page_has_buffers(page)); |
906 | ||
89f3b363 CH |
907 | /* |
908 | * Refuse to write the page out if we are called from reclaim context. | |
909 | * | |
d4f7a5cb CH |
910 | * This avoids stack overflows when called from deeply used stacks in |
911 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
912 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 913 | * |
94054fa3 MG |
914 | * This should never happen except in the case of a VM regression so |
915 | * warn about it. | |
89f3b363 | 916 | */ |
94054fa3 MG |
917 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
918 | PF_MEMALLOC)) | |
b5420f23 | 919 | goto redirty; |
1da177e4 | 920 | |
89f3b363 | 921 | /* |
680a647b CH |
922 | * Given that we do not allow direct reclaim to call us, we should |
923 | * never be called while in a filesystem transaction. | |
89f3b363 | 924 | */ |
680a647b | 925 | if (WARN_ON(current->flags & PF_FSTRANS)) |
b5420f23 | 926 | goto redirty; |
89f3b363 | 927 | |
1da177e4 LT |
928 | /* Is this page beyond the end of the file? */ |
929 | offset = i_size_read(inode); | |
930 | end_index = offset >> PAGE_CACHE_SHIFT; | |
931 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
932 | if (page->index >= end_index) { | |
933 | if ((page->index >= end_index + 1) || | |
934 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
89f3b363 | 935 | unlock_page(page); |
19d5bcf3 | 936 | return 0; |
1da177e4 LT |
937 | } |
938 | } | |
939 | ||
f6d6d4fc | 940 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
941 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
942 | offset); | |
24e17b5f | 943 | len = 1 << inode->i_blkbits; |
24e17b5f | 944 | |
24e17b5f | 945 | bh = head = page_buffers(page); |
f6d6d4fc | 946 | offset = page_offset(page); |
a206c817 CH |
947 | type = IO_OVERWRITE; |
948 | ||
dbcdde3e | 949 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 950 | nonblocking = 1; |
f6d6d4fc | 951 | |
1da177e4 | 952 | do { |
6ac7248e CH |
953 | int new_ioend = 0; |
954 | ||
1da177e4 LT |
955 | if (offset >= end_offset) |
956 | break; | |
957 | if (!buffer_uptodate(bh)) | |
958 | uptodate = 0; | |
1da177e4 | 959 | |
3d9b02e3 | 960 | /* |
ece413f5 CH |
961 | * set_page_dirty dirties all buffers in a page, independent |
962 | * of their state. The dirty state however is entirely | |
963 | * meaningless for holes (!mapped && uptodate), so skip | |
964 | * buffers covering holes here. | |
3d9b02e3 ES |
965 | */ |
966 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
967 | imap_valid = 0; |
968 | continue; | |
969 | } | |
970 | ||
aeea1b1f CH |
971 | if (buffer_unwritten(bh)) { |
972 | if (type != IO_UNWRITTEN) { | |
973 | type = IO_UNWRITTEN; | |
974 | imap_valid = 0; | |
1da177e4 | 975 | } |
aeea1b1f CH |
976 | } else if (buffer_delay(bh)) { |
977 | if (type != IO_DELALLOC) { | |
978 | type = IO_DELALLOC; | |
979 | imap_valid = 0; | |
1da177e4 | 980 | } |
89f3b363 | 981 | } else if (buffer_uptodate(bh)) { |
a206c817 CH |
982 | if (type != IO_OVERWRITE) { |
983 | type = IO_OVERWRITE; | |
85da94c6 CH |
984 | imap_valid = 0; |
985 | } | |
aeea1b1f CH |
986 | } else { |
987 | if (PageUptodate(page)) { | |
988 | ASSERT(buffer_mapped(bh)); | |
989 | imap_valid = 0; | |
6c4fe19f | 990 | } |
aeea1b1f CH |
991 | continue; |
992 | } | |
d5cb48aa | 993 | |
aeea1b1f CH |
994 | if (imap_valid) |
995 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
996 | if (!imap_valid) { | |
997 | /* | |
998 | * If we didn't have a valid mapping then we need to | |
999 | * put the new mapping into a separate ioend structure. | |
1000 | * This ensures non-contiguous extents always have | |
1001 | * separate ioends, which is particularly important | |
1002 | * for unwritten extent conversion at I/O completion | |
1003 | * time. | |
1004 | */ | |
1005 | new_ioend = 1; | |
1006 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1007 | nonblocking); | |
1008 | if (err) | |
1009 | goto error; | |
1010 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1011 | } | |
1012 | if (imap_valid) { | |
ecff71e6 CH |
1013 | lock_buffer(bh); |
1014 | if (type != IO_OVERWRITE) | |
aeea1b1f CH |
1015 | xfs_map_at_offset(inode, bh, &imap, offset); |
1016 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1017 | new_ioend); | |
1018 | count++; | |
1da177e4 | 1019 | } |
f6d6d4fc CH |
1020 | |
1021 | if (!iohead) | |
1022 | iohead = ioend; | |
1023 | ||
1024 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1025 | |
1026 | if (uptodate && bh == head) | |
1027 | SetPageUptodate(page); | |
1028 | ||
89f3b363 | 1029 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1030 | |
558e6891 | 1031 | if (ioend && imap_valid) { |
bd1556a1 CH |
1032 | xfs_off_t end_index; |
1033 | ||
1034 | end_index = imap.br_startoff + imap.br_blockcount; | |
1035 | ||
1036 | /* to bytes */ | |
1037 | end_index <<= inode->i_blkbits; | |
1038 | ||
1039 | /* to pages */ | |
1040 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1041 | ||
1042 | /* check against file size */ | |
1043 | if (end_index > last_index) | |
1044 | end_index = last_index; | |
8699bb0a | 1045 | |
207d0416 | 1046 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1047 | wbc, end_index); |
1da177e4 LT |
1048 | } |
1049 | ||
281627df CH |
1050 | if (iohead) { |
1051 | /* | |
1052 | * Reserve log space if we might write beyond the on-disk | |
1053 | * inode size. | |
1054 | */ | |
1055 | if (ioend->io_type != IO_UNWRITTEN && | |
1056 | xfs_ioend_is_append(ioend)) { | |
1057 | err = xfs_setfilesize_trans_alloc(ioend); | |
1058 | if (err) | |
1059 | goto error; | |
1060 | } | |
1061 | ||
06342cf8 | 1062 | xfs_submit_ioend(wbc, iohead); |
281627df | 1063 | } |
f6d6d4fc | 1064 | |
89f3b363 | 1065 | return 0; |
1da177e4 LT |
1066 | |
1067 | error: | |
f6d6d4fc CH |
1068 | if (iohead) |
1069 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1070 | |
b5420f23 CH |
1071 | if (err == -EAGAIN) |
1072 | goto redirty; | |
1073 | ||
20cb52eb | 1074 | xfs_aops_discard_page(page); |
89f3b363 CH |
1075 | ClearPageUptodate(page); |
1076 | unlock_page(page); | |
1da177e4 | 1077 | return err; |
f51623b2 | 1078 | |
b5420f23 | 1079 | redirty: |
f51623b2 NS |
1080 | redirty_page_for_writepage(wbc, page); |
1081 | unlock_page(page); | |
1082 | return 0; | |
f51623b2 NS |
1083 | } |
1084 | ||
7d4fb40a NS |
1085 | STATIC int |
1086 | xfs_vm_writepages( | |
1087 | struct address_space *mapping, | |
1088 | struct writeback_control *wbc) | |
1089 | { | |
b3aea4ed | 1090 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1091 | return generic_writepages(mapping, wbc); |
1092 | } | |
1093 | ||
f51623b2 NS |
1094 | /* |
1095 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1096 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1097 | * have buffer heads in this call. |
1098 | * | |
89f3b363 | 1099 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1100 | */ |
1101 | STATIC int | |
238f4c54 | 1102 | xfs_vm_releasepage( |
f51623b2 NS |
1103 | struct page *page, |
1104 | gfp_t gfp_mask) | |
1105 | { | |
20cb52eb | 1106 | int delalloc, unwritten; |
f51623b2 | 1107 | |
89f3b363 | 1108 | trace_xfs_releasepage(page->mapping->host, page, 0); |
238f4c54 | 1109 | |
20cb52eb | 1110 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1111 | |
89f3b363 | 1112 | if (WARN_ON(delalloc)) |
f51623b2 | 1113 | return 0; |
89f3b363 | 1114 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1115 | return 0; |
1116 | ||
f51623b2 NS |
1117 | return try_to_free_buffers(page); |
1118 | } | |
1119 | ||
1da177e4 | 1120 | STATIC int |
c2536668 | 1121 | __xfs_get_blocks( |
1da177e4 LT |
1122 | struct inode *inode, |
1123 | sector_t iblock, | |
1da177e4 LT |
1124 | struct buffer_head *bh_result, |
1125 | int create, | |
f2bde9b8 | 1126 | int direct) |
1da177e4 | 1127 | { |
a206c817 CH |
1128 | struct xfs_inode *ip = XFS_I(inode); |
1129 | struct xfs_mount *mp = ip->i_mount; | |
1130 | xfs_fileoff_t offset_fsb, end_fsb; | |
1131 | int error = 0; | |
1132 | int lockmode = 0; | |
207d0416 | 1133 | struct xfs_bmbt_irec imap; |
a206c817 | 1134 | int nimaps = 1; |
fdc7ed75 NS |
1135 | xfs_off_t offset; |
1136 | ssize_t size; | |
207d0416 | 1137 | int new = 0; |
a206c817 CH |
1138 | |
1139 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1140 | return -XFS_ERROR(EIO); | |
1da177e4 | 1141 | |
fdc7ed75 | 1142 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1143 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1144 | size = bh_result->b_size; | |
364f358a LM |
1145 | |
1146 | if (!create && direct && offset >= i_size_read(inode)) | |
1147 | return 0; | |
1148 | ||
507630b2 DC |
1149 | /* |
1150 | * Direct I/O is usually done on preallocated files, so try getting | |
1151 | * a block mapping without an exclusive lock first. For buffered | |
1152 | * writes we already have the exclusive iolock anyway, so avoiding | |
1153 | * a lock roundtrip here by taking the ilock exclusive from the | |
1154 | * beginning is a useful micro optimization. | |
1155 | */ | |
1156 | if (create && !direct) { | |
a206c817 CH |
1157 | lockmode = XFS_ILOCK_EXCL; |
1158 | xfs_ilock(ip, lockmode); | |
1159 | } else { | |
1160 | lockmode = xfs_ilock_map_shared(ip); | |
1161 | } | |
f2bde9b8 | 1162 | |
a206c817 CH |
1163 | ASSERT(offset <= mp->m_maxioffset); |
1164 | if (offset + size > mp->m_maxioffset) | |
1165 | size = mp->m_maxioffset - offset; | |
1166 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); | |
1167 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1168 | ||
5c8ed202 DC |
1169 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1170 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1171 | if (error) |
a206c817 CH |
1172 | goto out_unlock; |
1173 | ||
1174 | if (create && | |
1175 | (!nimaps || | |
1176 | (imap.br_startblock == HOLESTARTBLOCK || | |
1177 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
1178 | if (direct) { | |
507630b2 DC |
1179 | /* |
1180 | * Drop the ilock in preparation for starting the block | |
1181 | * allocation transaction. It will be retaken | |
1182 | * exclusively inside xfs_iomap_write_direct for the | |
1183 | * actual allocation. | |
1184 | */ | |
1185 | xfs_iunlock(ip, lockmode); | |
a206c817 CH |
1186 | error = xfs_iomap_write_direct(ip, offset, size, |
1187 | &imap, nimaps); | |
507630b2 DC |
1188 | if (error) |
1189 | return -error; | |
d3bc815a | 1190 | new = 1; |
a206c817 | 1191 | } else { |
507630b2 DC |
1192 | /* |
1193 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1194 | * we can go on without dropping the lock here. If we |
1195 | * are allocating a new delalloc block, make sure that | |
1196 | * we set the new flag so that we mark the buffer new so | |
1197 | * that we know that it is newly allocated if the write | |
1198 | * fails. | |
507630b2 | 1199 | */ |
d3bc815a DC |
1200 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1201 | new = 1; | |
a206c817 | 1202 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1203 | if (error) |
1204 | goto out_unlock; | |
1205 | ||
1206 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1207 | } |
a206c817 CH |
1208 | |
1209 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1210 | } else if (nimaps) { | |
1211 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
507630b2 | 1212 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1213 | } else { |
1214 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1215 | goto out_unlock; | |
1216 | } | |
1da177e4 | 1217 | |
207d0416 CH |
1218 | if (imap.br_startblock != HOLESTARTBLOCK && |
1219 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1220 | /* |
1221 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1222 | * the read case (treat as if we're reading into a hole). |
1223 | */ | |
207d0416 CH |
1224 | if (create || !ISUNWRITTEN(&imap)) |
1225 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1226 | if (create && ISUNWRITTEN(&imap)) { | |
1da177e4 LT |
1227 | if (direct) |
1228 | bh_result->b_private = inode; | |
1229 | set_buffer_unwritten(bh_result); | |
1da177e4 LT |
1230 | } |
1231 | } | |
1232 | ||
c2536668 NS |
1233 | /* |
1234 | * If this is a realtime file, data may be on a different device. | |
1235 | * to that pointed to from the buffer_head b_bdev currently. | |
1236 | */ | |
046f1685 | 1237 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1238 | |
c2536668 | 1239 | /* |
549054af DC |
1240 | * If we previously allocated a block out beyond eof and we are now |
1241 | * coming back to use it then we will need to flag it as new even if it | |
1242 | * has a disk address. | |
1243 | * | |
1244 | * With sub-block writes into unwritten extents we also need to mark | |
1245 | * the buffer as new so that the unwritten parts of the buffer gets | |
1246 | * correctly zeroed. | |
1da177e4 LT |
1247 | */ |
1248 | if (create && | |
1249 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1250 | (offset >= i_size_read(inode)) || |
207d0416 | 1251 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1252 | set_buffer_new(bh_result); |
1da177e4 | 1253 | |
207d0416 | 1254 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1255 | BUG_ON(direct); |
1256 | if (create) { | |
1257 | set_buffer_uptodate(bh_result); | |
1258 | set_buffer_mapped(bh_result); | |
1259 | set_buffer_delay(bh_result); | |
1260 | } | |
1261 | } | |
1262 | ||
2b8f12b7 CH |
1263 | /* |
1264 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1265 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1266 | */ | |
c2536668 | 1267 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1268 | xfs_off_t mapping_size; |
1269 | ||
1270 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1271 | mapping_size <<= inode->i_blkbits; | |
1272 | ||
1273 | ASSERT(mapping_size > 0); | |
1274 | if (mapping_size > size) | |
1275 | mapping_size = size; | |
1276 | if (mapping_size > LONG_MAX) | |
1277 | mapping_size = LONG_MAX; | |
1278 | ||
1279 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1280 | } |
1281 | ||
1282 | return 0; | |
a206c817 CH |
1283 | |
1284 | out_unlock: | |
1285 | xfs_iunlock(ip, lockmode); | |
1286 | return -error; | |
1da177e4 LT |
1287 | } |
1288 | ||
1289 | int | |
c2536668 | 1290 | xfs_get_blocks( |
1da177e4 LT |
1291 | struct inode *inode, |
1292 | sector_t iblock, | |
1293 | struct buffer_head *bh_result, | |
1294 | int create) | |
1295 | { | |
f2bde9b8 | 1296 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1297 | } |
1298 | ||
1299 | STATIC int | |
e4c573bb | 1300 | xfs_get_blocks_direct( |
1da177e4 LT |
1301 | struct inode *inode, |
1302 | sector_t iblock, | |
1da177e4 LT |
1303 | struct buffer_head *bh_result, |
1304 | int create) | |
1305 | { | |
f2bde9b8 | 1306 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1307 | } |
1308 | ||
209fb87a CH |
1309 | /* |
1310 | * Complete a direct I/O write request. | |
1311 | * | |
1312 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1313 | * need to issue a transaction to convert the range from unwritten to written | |
1314 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1315 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1316 | * request this handler is called from interrupt context, from which we |
1317 | * can't start transactions. In that case offload the I/O completion to | |
1318 | * the workqueues we also use for buffered I/O completion. | |
1319 | */ | |
f0973863 | 1320 | STATIC void |
209fb87a CH |
1321 | xfs_end_io_direct_write( |
1322 | struct kiocb *iocb, | |
1323 | loff_t offset, | |
1324 | ssize_t size, | |
1325 | void *private, | |
1326 | int ret, | |
1327 | bool is_async) | |
f0973863 | 1328 | { |
209fb87a | 1329 | struct xfs_ioend *ioend = iocb->private; |
f0973863 | 1330 | |
2813d682 CH |
1331 | /* |
1332 | * While the generic direct I/O code updates the inode size, it does | |
1333 | * so only after the end_io handler is called, which means our | |
1334 | * end_io handler thinks the on-disk size is outside the in-core | |
1335 | * size. To prevent this just update it a little bit earlier here. | |
1336 | */ | |
1337 | if (offset + size > i_size_read(ioend->io_inode)) | |
1338 | i_size_write(ioend->io_inode, offset + size); | |
1339 | ||
f0973863 | 1340 | /* |
209fb87a CH |
1341 | * blockdev_direct_IO can return an error even after the I/O |
1342 | * completion handler was called. Thus we need to protect | |
1343 | * against double-freeing. | |
f0973863 | 1344 | */ |
209fb87a CH |
1345 | iocb->private = NULL; |
1346 | ||
ba87ea69 LM |
1347 | ioend->io_offset = offset; |
1348 | ioend->io_size = size; | |
c859cdd1 CH |
1349 | ioend->io_iocb = iocb; |
1350 | ioend->io_result = ret; | |
209fb87a CH |
1351 | if (private && size > 0) |
1352 | ioend->io_type = IO_UNWRITTEN; | |
1353 | ||
1354 | if (is_async) { | |
c859cdd1 | 1355 | ioend->io_isasync = 1; |
209fb87a | 1356 | xfs_finish_ioend(ioend); |
f0973863 | 1357 | } else { |
209fb87a | 1358 | xfs_finish_ioend_sync(ioend); |
f0973863 | 1359 | } |
f0973863 CH |
1360 | } |
1361 | ||
1da177e4 | 1362 | STATIC ssize_t |
e4c573bb | 1363 | xfs_vm_direct_IO( |
1da177e4 LT |
1364 | int rw, |
1365 | struct kiocb *iocb, | |
1366 | const struct iovec *iov, | |
1367 | loff_t offset, | |
1368 | unsigned long nr_segs) | |
1369 | { | |
209fb87a CH |
1370 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1371 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
281627df | 1372 | struct xfs_ioend *ioend = NULL; |
209fb87a CH |
1373 | ssize_t ret; |
1374 | ||
1375 | if (rw & WRITE) { | |
281627df CH |
1376 | size_t size = iov_length(iov, nr_segs); |
1377 | ||
1378 | /* | |
1379 | * We need to preallocate a transaction for a size update | |
1380 | * here. In the case that this write both updates the size | |
1381 | * and converts at least on unwritten extent we will cancel | |
1382 | * the still clean transaction after the I/O has finished. | |
1383 | */ | |
1384 | iocb->private = ioend = xfs_alloc_ioend(inode, IO_DIRECT); | |
1385 | if (offset + size > XFS_I(inode)->i_d.di_size) { | |
1386 | ret = xfs_setfilesize_trans_alloc(ioend); | |
1387 | if (ret) | |
1388 | goto out_destroy_ioend; | |
1389 | ioend->io_isdirect = 1; | |
1390 | } | |
209fb87a | 1391 | |
eafdc7d1 CH |
1392 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1393 | offset, nr_segs, | |
1394 | xfs_get_blocks_direct, | |
1395 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a | 1396 | if (ret != -EIOCBQUEUED && iocb->private) |
281627df | 1397 | goto out_trans_cancel; |
209fb87a | 1398 | } else { |
eafdc7d1 CH |
1399 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1400 | offset, nr_segs, | |
1401 | xfs_get_blocks_direct, | |
1402 | NULL, NULL, 0); | |
209fb87a | 1403 | } |
f0973863 | 1404 | |
f0973863 | 1405 | return ret; |
281627df CH |
1406 | |
1407 | out_trans_cancel: | |
1408 | if (ioend->io_append_trans) { | |
1409 | current_set_flags_nested(&ioend->io_append_trans->t_pflags, | |
1410 | PF_FSTRANS); | |
1411 | xfs_trans_cancel(ioend->io_append_trans, 0); | |
1412 | } | |
1413 | out_destroy_ioend: | |
1414 | xfs_destroy_ioend(ioend); | |
1415 | return ret; | |
1da177e4 LT |
1416 | } |
1417 | ||
d3bc815a DC |
1418 | /* |
1419 | * Punch out the delalloc blocks we have already allocated. | |
1420 | * | |
1421 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1422 | * as the page is still locked at this point. | |
1423 | */ | |
1424 | STATIC void | |
1425 | xfs_vm_kill_delalloc_range( | |
1426 | struct inode *inode, | |
1427 | loff_t start, | |
1428 | loff_t end) | |
1429 | { | |
1430 | struct xfs_inode *ip = XFS_I(inode); | |
1431 | xfs_fileoff_t start_fsb; | |
1432 | xfs_fileoff_t end_fsb; | |
1433 | int error; | |
1434 | ||
1435 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1436 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1437 | if (end_fsb <= start_fsb) | |
1438 | return; | |
1439 | ||
1440 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1441 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1442 | end_fsb - start_fsb); | |
1443 | if (error) { | |
1444 | /* something screwed, just bail */ | |
1445 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1446 | xfs_alert(ip->i_mount, | |
1447 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1448 | ip->i_ino); | |
1449 | } | |
1450 | } | |
1451 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1452 | } | |
1453 | ||
fa9b227e CH |
1454 | STATIC void |
1455 | xfs_vm_write_failed( | |
d3bc815a DC |
1456 | struct inode *inode, |
1457 | struct page *page, | |
1458 | loff_t pos, | |
1459 | unsigned len) | |
fa9b227e | 1460 | { |
d3bc815a DC |
1461 | loff_t block_offset = pos & PAGE_MASK; |
1462 | loff_t block_start; | |
1463 | loff_t block_end; | |
1464 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1465 | loff_t to = from + len; | |
1466 | struct buffer_head *bh, *head; | |
fa9b227e | 1467 | |
d3bc815a | 1468 | ASSERT(block_offset + from == pos); |
c726de44 | 1469 | |
d3bc815a DC |
1470 | head = page_buffers(page); |
1471 | block_start = 0; | |
1472 | for (bh = head; bh != head || !block_start; | |
1473 | bh = bh->b_this_page, block_start = block_end, | |
1474 | block_offset += bh->b_size) { | |
1475 | block_end = block_start + bh->b_size; | |
c726de44 | 1476 | |
d3bc815a DC |
1477 | /* skip buffers before the write */ |
1478 | if (block_end <= from) | |
1479 | continue; | |
1480 | ||
1481 | /* if the buffer is after the write, we're done */ | |
1482 | if (block_start >= to) | |
1483 | break; | |
1484 | ||
1485 | if (!buffer_delay(bh)) | |
1486 | continue; | |
1487 | ||
1488 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1489 | continue; | |
1490 | ||
1491 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1492 | block_offset + bh->b_size); | |
fa9b227e | 1493 | } |
d3bc815a | 1494 | |
fa9b227e CH |
1495 | } |
1496 | ||
d3bc815a DC |
1497 | /* |
1498 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1499 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1500 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1501 | * the appropriate point. | |
1502 | */ | |
f51623b2 | 1503 | STATIC int |
d79689c7 | 1504 | xfs_vm_write_begin( |
f51623b2 | 1505 | struct file *file, |
d79689c7 NP |
1506 | struct address_space *mapping, |
1507 | loff_t pos, | |
1508 | unsigned len, | |
1509 | unsigned flags, | |
1510 | struct page **pagep, | |
1511 | void **fsdata) | |
f51623b2 | 1512 | { |
d3bc815a DC |
1513 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1514 | struct page *page; | |
1515 | int status; | |
155130a4 | 1516 | |
d3bc815a DC |
1517 | ASSERT(len <= PAGE_CACHE_SIZE); |
1518 | ||
1519 | page = grab_cache_page_write_begin(mapping, index, | |
1520 | flags | AOP_FLAG_NOFS); | |
1521 | if (!page) | |
1522 | return -ENOMEM; | |
1523 | ||
1524 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1525 | if (unlikely(status)) { | |
1526 | struct inode *inode = mapping->host; | |
1527 | ||
1528 | xfs_vm_write_failed(inode, page, pos, len); | |
1529 | unlock_page(page); | |
1530 | ||
1531 | if (pos + len > i_size_read(inode)) | |
1532 | truncate_pagecache(inode, pos + len, i_size_read(inode)); | |
1533 | ||
1534 | page_cache_release(page); | |
1535 | page = NULL; | |
1536 | } | |
1537 | ||
1538 | *pagep = page; | |
1539 | return status; | |
fa9b227e CH |
1540 | } |
1541 | ||
d3bc815a DC |
1542 | /* |
1543 | * On failure, we only need to kill delalloc blocks beyond EOF because they | |
1544 | * will never be written. For blocks within EOF, generic_write_end() zeros them | |
1545 | * so they are safe to leave alone and be written with all the other valid data. | |
1546 | */ | |
fa9b227e CH |
1547 | STATIC int |
1548 | xfs_vm_write_end( | |
1549 | struct file *file, | |
1550 | struct address_space *mapping, | |
1551 | loff_t pos, | |
1552 | unsigned len, | |
1553 | unsigned copied, | |
1554 | struct page *page, | |
1555 | void *fsdata) | |
1556 | { | |
1557 | int ret; | |
155130a4 | 1558 | |
d3bc815a DC |
1559 | ASSERT(len <= PAGE_CACHE_SIZE); |
1560 | ||
fa9b227e | 1561 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1562 | if (unlikely(ret < len)) { |
1563 | struct inode *inode = mapping->host; | |
1564 | size_t isize = i_size_read(inode); | |
1565 | loff_t to = pos + len; | |
1566 | ||
1567 | if (to > isize) { | |
1568 | truncate_pagecache(inode, to, isize); | |
1569 | xfs_vm_kill_delalloc_range(inode, isize, to); | |
1570 | } | |
1571 | } | |
155130a4 | 1572 | return ret; |
f51623b2 | 1573 | } |
1da177e4 LT |
1574 | |
1575 | STATIC sector_t | |
e4c573bb | 1576 | xfs_vm_bmap( |
1da177e4 LT |
1577 | struct address_space *mapping, |
1578 | sector_t block) | |
1579 | { | |
1580 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1581 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1582 | |
cca28fb8 | 1583 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1584 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
739bfb2a | 1585 | xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF); |
126468b1 | 1586 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1587 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1588 | } |
1589 | ||
1590 | STATIC int | |
e4c573bb | 1591 | xfs_vm_readpage( |
1da177e4 LT |
1592 | struct file *unused, |
1593 | struct page *page) | |
1594 | { | |
c2536668 | 1595 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1596 | } |
1597 | ||
1598 | STATIC int | |
e4c573bb | 1599 | xfs_vm_readpages( |
1da177e4 LT |
1600 | struct file *unused, |
1601 | struct address_space *mapping, | |
1602 | struct list_head *pages, | |
1603 | unsigned nr_pages) | |
1604 | { | |
c2536668 | 1605 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1606 | } |
1607 | ||
f5e54d6e | 1608 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1609 | .readpage = xfs_vm_readpage, |
1610 | .readpages = xfs_vm_readpages, | |
1611 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1612 | .writepages = xfs_vm_writepages, |
238f4c54 NS |
1613 | .releasepage = xfs_vm_releasepage, |
1614 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1615 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1616 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1617 | .bmap = xfs_vm_bmap, |
1618 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1619 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1620 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1621 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1622 | }; |