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" |
70a9883c DC |
19 | #include "xfs_format.h" |
20 | #include "xfs_shared.h" | |
1da177e4 | 21 | #include "xfs_sb.h" |
a844f451 | 22 | #include "xfs_ag.h" |
70a9883c | 23 | #include "xfs_log.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 | 31 | #include "xfs_error.h" |
1da177e4 | 32 | #include "xfs_iomap.h" |
0b1b213f | 33 | #include "xfs_trace.h" |
3ed3a434 | 34 | #include "xfs_bmap.h" |
68988114 | 35 | #include "xfs_bmap_util.h" |
a27bb332 | 36 | #include <linux/aio.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 | |
0829c360 CH |
91 | mempool_free(ioend, xfs_ioend_pool); |
92 | } | |
93 | ||
fc0063c4 CH |
94 | /* |
95 | * Fast and loose check if this write could update the on-disk inode size. | |
96 | */ | |
97 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
98 | { | |
99 | return ioend->io_offset + ioend->io_size > | |
100 | XFS_I(ioend->io_inode)->i_d.di_size; | |
101 | } | |
102 | ||
281627df CH |
103 | STATIC int |
104 | xfs_setfilesize_trans_alloc( | |
105 | struct xfs_ioend *ioend) | |
106 | { | |
107 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
108 | struct xfs_trans *tp; | |
109 | int error; | |
110 | ||
111 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
112 | ||
3d3c8b52 | 113 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
281627df CH |
114 | if (error) { |
115 | xfs_trans_cancel(tp, 0); | |
116 | return error; | |
117 | } | |
118 | ||
119 | ioend->io_append_trans = tp; | |
120 | ||
d9457dc0 | 121 | /* |
437a255a | 122 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
123 | * we released it. |
124 | */ | |
125 | rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
126 | 1, _THIS_IP_); | |
281627df CH |
127 | /* |
128 | * We hand off the transaction to the completion thread now, so | |
129 | * clear the flag here. | |
130 | */ | |
131 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
132 | return 0; | |
133 | } | |
134 | ||
ba87ea69 | 135 | /* |
2813d682 | 136 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 137 | */ |
281627df | 138 | STATIC int |
ba87ea69 | 139 | xfs_setfilesize( |
aa6bf01d | 140 | struct xfs_ioend *ioend) |
ba87ea69 | 141 | { |
aa6bf01d | 142 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
281627df | 143 | struct xfs_trans *tp = ioend->io_append_trans; |
ba87ea69 | 144 | xfs_fsize_t isize; |
ba87ea69 | 145 | |
281627df | 146 | /* |
437a255a DC |
147 | * The transaction may have been allocated in the I/O submission thread, |
148 | * thus we need to mark ourselves as beeing in a transaction manually. | |
149 | * Similarly for freeze protection. | |
281627df CH |
150 | */ |
151 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
437a255a DC |
152 | rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], |
153 | 0, 1, _THIS_IP_); | |
281627df | 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 | ||
0d882a36 | 185 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
aa6bf01d | 186 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
437a255a DC |
187 | else if (ioend->io_append_trans || |
188 | (ioend->io_isdirect && xfs_ioend_is_append(ioend))) | |
aa6bf01d | 189 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
190 | else |
191 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 192 | } |
ba87ea69 LM |
193 | } |
194 | ||
0829c360 | 195 | /* |
5ec4fabb | 196 | * IO write completion. |
f6d6d4fc CH |
197 | */ |
198 | STATIC void | |
5ec4fabb | 199 | xfs_end_io( |
77d7a0c2 | 200 | struct work_struct *work) |
0829c360 | 201 | { |
77d7a0c2 DC |
202 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
203 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 204 | int error = 0; |
ba87ea69 | 205 | |
04f658ee | 206 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 207 | ioend->io_error = -EIO; |
04f658ee CH |
208 | goto done; |
209 | } | |
210 | if (ioend->io_error) | |
211 | goto done; | |
212 | ||
5ec4fabb CH |
213 | /* |
214 | * For unwritten extents we need to issue transactions to convert a | |
215 | * range to normal written extens after the data I/O has finished. | |
216 | */ | |
0d882a36 | 217 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
437a255a DC |
218 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
219 | ioend->io_size); | |
220 | } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) { | |
281627df | 221 | /* |
437a255a DC |
222 | * For direct I/O we do not know if we need to allocate blocks |
223 | * or not so we can't preallocate an append transaction as that | |
224 | * results in nested reservations and log space deadlocks. Hence | |
225 | * allocate the transaction here. While this is sub-optimal and | |
226 | * can block IO completion for some time, we're stuck with doing | |
227 | * it this way until we can pass the ioend to the direct IO | |
228 | * allocation callbacks and avoid nesting that way. | |
281627df | 229 | */ |
437a255a DC |
230 | error = xfs_setfilesize_trans_alloc(ioend); |
231 | if (error) | |
04f658ee | 232 | goto done; |
437a255a | 233 | error = xfs_setfilesize(ioend); |
281627df CH |
234 | } else if (ioend->io_append_trans) { |
235 | error = xfs_setfilesize(ioend); | |
84803fb7 | 236 | } else { |
281627df | 237 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 238 | } |
ba87ea69 | 239 | |
04f658ee | 240 | done: |
437a255a DC |
241 | if (error) |
242 | ioend->io_error = -error; | |
aa6bf01d | 243 | xfs_destroy_ioend(ioend); |
c626d174 DC |
244 | } |
245 | ||
209fb87a CH |
246 | /* |
247 | * Call IO completion handling in caller context on the final put of an ioend. | |
248 | */ | |
249 | STATIC void | |
250 | xfs_finish_ioend_sync( | |
251 | struct xfs_ioend *ioend) | |
252 | { | |
253 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
254 | xfs_end_io(&ioend->io_work); | |
255 | } | |
256 | ||
0829c360 CH |
257 | /* |
258 | * Allocate and initialise an IO completion structure. | |
259 | * We need to track unwritten extent write completion here initially. | |
260 | * We'll need to extend this for updating the ondisk inode size later | |
261 | * (vs. incore size). | |
262 | */ | |
263 | STATIC xfs_ioend_t * | |
264 | xfs_alloc_ioend( | |
f6d6d4fc CH |
265 | struct inode *inode, |
266 | unsigned int type) | |
0829c360 CH |
267 | { |
268 | xfs_ioend_t *ioend; | |
269 | ||
270 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
271 | ||
272 | /* | |
273 | * Set the count to 1 initially, which will prevent an I/O | |
274 | * completion callback from happening before we have started | |
275 | * all the I/O from calling the completion routine too early. | |
276 | */ | |
277 | atomic_set(&ioend->io_remaining, 1); | |
281627df | 278 | ioend->io_isdirect = 0; |
7d04a335 | 279 | ioend->io_error = 0; |
f6d6d4fc CH |
280 | ioend->io_list = NULL; |
281 | ioend->io_type = type; | |
b677c210 | 282 | ioend->io_inode = inode; |
c1a073bd | 283 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 284 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
285 | ioend->io_offset = 0; |
286 | ioend->io_size = 0; | |
281627df | 287 | ioend->io_append_trans = NULL; |
0829c360 | 288 | |
5ec4fabb | 289 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
290 | return ioend; |
291 | } | |
292 | ||
1da177e4 LT |
293 | STATIC int |
294 | xfs_map_blocks( | |
295 | struct inode *inode, | |
296 | loff_t offset, | |
207d0416 | 297 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
298 | int type, |
299 | int nonblocking) | |
1da177e4 | 300 | { |
a206c817 CH |
301 | struct xfs_inode *ip = XFS_I(inode); |
302 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 303 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
304 | xfs_fileoff_t offset_fsb, end_fsb; |
305 | int error = 0; | |
a206c817 CH |
306 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
307 | int nimaps = 1; | |
308 | ||
309 | if (XFS_FORCED_SHUTDOWN(mp)) | |
310 | return -XFS_ERROR(EIO); | |
311 | ||
0d882a36 | 312 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 313 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
314 | |
315 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
316 | if (nonblocking) | |
317 | return -XFS_ERROR(EAGAIN); | |
318 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
a206c817 CH |
319 | } |
320 | ||
8ff2957d CH |
321 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
322 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 323 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 324 | |
d2c28191 DC |
325 | if (offset + count > mp->m_super->s_maxbytes) |
326 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
327 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
328 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
329 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
330 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 331 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 332 | |
8ff2957d CH |
333 | if (error) |
334 | return -XFS_ERROR(error); | |
a206c817 | 335 | |
0d882a36 | 336 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 337 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 338 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
339 | if (!error) |
340 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
8ff2957d | 341 | return -XFS_ERROR(error); |
a206c817 CH |
342 | } |
343 | ||
8ff2957d | 344 | #ifdef DEBUG |
0d882a36 | 345 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
346 | ASSERT(nimaps); |
347 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
348 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
349 | } | |
350 | #endif | |
351 | if (nimaps) | |
352 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
353 | return 0; | |
1da177e4 LT |
354 | } |
355 | ||
b8f82a4a | 356 | STATIC int |
558e6891 | 357 | xfs_imap_valid( |
8699bb0a | 358 | struct inode *inode, |
207d0416 | 359 | struct xfs_bmbt_irec *imap, |
558e6891 | 360 | xfs_off_t offset) |
1da177e4 | 361 | { |
558e6891 | 362 | offset >>= inode->i_blkbits; |
8699bb0a | 363 | |
558e6891 CH |
364 | return offset >= imap->br_startoff && |
365 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
366 | } |
367 | ||
f6d6d4fc CH |
368 | /* |
369 | * BIO completion handler for buffered IO. | |
370 | */ | |
782e3b3b | 371 | STATIC void |
f6d6d4fc CH |
372 | xfs_end_bio( |
373 | struct bio *bio, | |
f6d6d4fc CH |
374 | int error) |
375 | { | |
376 | xfs_ioend_t *ioend = bio->bi_private; | |
377 | ||
f6d6d4fc | 378 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 379 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
380 | |
381 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
382 | bio->bi_private = NULL; |
383 | bio->bi_end_io = NULL; | |
f6d6d4fc | 384 | bio_put(bio); |
7d04a335 | 385 | |
209fb87a | 386 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
387 | } |
388 | ||
389 | STATIC void | |
390 | xfs_submit_ioend_bio( | |
06342cf8 CH |
391 | struct writeback_control *wbc, |
392 | xfs_ioend_t *ioend, | |
393 | struct bio *bio) | |
f6d6d4fc CH |
394 | { |
395 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
396 | bio->bi_private = ioend; |
397 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 398 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
399 | } |
400 | ||
401 | STATIC struct bio * | |
402 | xfs_alloc_ioend_bio( | |
403 | struct buffer_head *bh) | |
404 | { | |
f6d6d4fc | 405 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 406 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
407 | |
408 | ASSERT(bio->bi_private == NULL); | |
409 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
410 | bio->bi_bdev = bh->b_bdev; | |
f6d6d4fc CH |
411 | return bio; |
412 | } | |
413 | ||
414 | STATIC void | |
415 | xfs_start_buffer_writeback( | |
416 | struct buffer_head *bh) | |
417 | { | |
418 | ASSERT(buffer_mapped(bh)); | |
419 | ASSERT(buffer_locked(bh)); | |
420 | ASSERT(!buffer_delay(bh)); | |
421 | ASSERT(!buffer_unwritten(bh)); | |
422 | ||
423 | mark_buffer_async_write(bh); | |
424 | set_buffer_uptodate(bh); | |
425 | clear_buffer_dirty(bh); | |
426 | } | |
427 | ||
428 | STATIC void | |
429 | xfs_start_page_writeback( | |
430 | struct page *page, | |
f6d6d4fc CH |
431 | int clear_dirty, |
432 | int buffers) | |
433 | { | |
434 | ASSERT(PageLocked(page)); | |
435 | ASSERT(!PageWriteback(page)); | |
f6d6d4fc | 436 | if (clear_dirty) |
92132021 DC |
437 | clear_page_dirty_for_io(page); |
438 | set_page_writeback(page); | |
f6d6d4fc | 439 | unlock_page(page); |
1f7decf6 FW |
440 | /* If no buffers on the page are to be written, finish it here */ |
441 | if (!buffers) | |
f6d6d4fc | 442 | end_page_writeback(page); |
f6d6d4fc CH |
443 | } |
444 | ||
c7c1a7d8 | 445 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
446 | { |
447 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
448 | } | |
449 | ||
450 | /* | |
d88992f6 DC |
451 | * Submit all of the bios for all of the ioends we have saved up, covering the |
452 | * initial writepage page and also any probed pages. | |
453 | * | |
454 | * Because we may have multiple ioends spanning a page, we need to start | |
455 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
456 | * buffers as we got, then we can end up with a page that only has buffers | |
457 | * marked async write and I/O complete on can occur before we mark the other | |
458 | * buffers async write. | |
459 | * | |
460 | * The end result of this is that we trip a bug in end_page_writeback() because | |
461 | * we call it twice for the one page as the code in end_buffer_async_write() | |
462 | * assumes that all buffers on the page are started at the same time. | |
463 | * | |
464 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 465 | * buffer_heads, and then submit them for I/O on the second pass. |
7bf7f352 DC |
466 | * |
467 | * If @fail is non-zero, it means that we have a situation where some part of | |
468 | * the submission process has failed after we have marked paged for writeback | |
469 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
470 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
f6d6d4fc CH |
471 | */ |
472 | STATIC void | |
473 | xfs_submit_ioend( | |
06342cf8 | 474 | struct writeback_control *wbc, |
7bf7f352 DC |
475 | xfs_ioend_t *ioend, |
476 | int fail) | |
f6d6d4fc | 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 | ||
7bf7f352 DC |
497 | /* |
498 | * If we are failing the IO now, just mark the ioend with an | |
499 | * error and finish it. This will run IO completion immediately | |
500 | * as there is only one reference to the ioend at this point in | |
501 | * time. | |
502 | */ | |
503 | if (fail) { | |
504 | ioend->io_error = -fail; | |
505 | xfs_finish_ioend(ioend); | |
506 | continue; | |
507 | } | |
508 | ||
f6d6d4fc | 509 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
f6d6d4fc CH |
510 | |
511 | if (!bio) { | |
512 | retry: | |
513 | bio = xfs_alloc_ioend_bio(bh); | |
514 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 515 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
516 | goto retry; |
517 | } | |
518 | ||
c7c1a7d8 | 519 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
06342cf8 | 520 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
521 | goto retry; |
522 | } | |
523 | ||
524 | lastblock = bh->b_blocknr; | |
525 | } | |
526 | if (bio) | |
06342cf8 | 527 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 528 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
529 | } while ((ioend = next) != NULL); |
530 | } | |
531 | ||
532 | /* | |
533 | * Cancel submission of all buffer_heads so far in this endio. | |
534 | * Toss the endio too. Only ever called for the initial page | |
535 | * in a writepage request, so only ever one page. | |
536 | */ | |
537 | STATIC void | |
538 | xfs_cancel_ioend( | |
539 | xfs_ioend_t *ioend) | |
540 | { | |
541 | xfs_ioend_t *next; | |
542 | struct buffer_head *bh, *next_bh; | |
543 | ||
544 | do { | |
545 | next = ioend->io_list; | |
546 | bh = ioend->io_buffer_head; | |
547 | do { | |
548 | next_bh = bh->b_private; | |
549 | clear_buffer_async_write(bh); | |
550 | unlock_buffer(bh); | |
551 | } while ((bh = next_bh) != NULL); | |
552 | ||
f6d6d4fc CH |
553 | mempool_free(ioend, xfs_ioend_pool); |
554 | } while ((ioend = next) != NULL); | |
555 | } | |
556 | ||
557 | /* | |
558 | * Test to see if we've been building up a completion structure for | |
559 | * earlier buffers -- if so, we try to append to this ioend if we | |
560 | * can, otherwise we finish off any current ioend and start another. | |
561 | * Return true if we've finished the given ioend. | |
562 | */ | |
563 | STATIC void | |
564 | xfs_add_to_ioend( | |
565 | struct inode *inode, | |
566 | struct buffer_head *bh, | |
7336cea8 | 567 | xfs_off_t offset, |
f6d6d4fc CH |
568 | unsigned int type, |
569 | xfs_ioend_t **result, | |
570 | int need_ioend) | |
571 | { | |
572 | xfs_ioend_t *ioend = *result; | |
573 | ||
574 | if (!ioend || need_ioend || type != ioend->io_type) { | |
575 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 576 | |
f6d6d4fc CH |
577 | ioend = xfs_alloc_ioend(inode, type); |
578 | ioend->io_offset = offset; | |
579 | ioend->io_buffer_head = bh; | |
580 | ioend->io_buffer_tail = bh; | |
581 | if (previous) | |
582 | previous->io_list = ioend; | |
583 | *result = ioend; | |
584 | } else { | |
585 | ioend->io_buffer_tail->b_private = bh; | |
586 | ioend->io_buffer_tail = bh; | |
587 | } | |
588 | ||
589 | bh->b_private = NULL; | |
590 | ioend->io_size += bh->b_size; | |
591 | } | |
592 | ||
87cbc49c NS |
593 | STATIC void |
594 | xfs_map_buffer( | |
046f1685 | 595 | struct inode *inode, |
87cbc49c | 596 | struct buffer_head *bh, |
207d0416 | 597 | struct xfs_bmbt_irec *imap, |
046f1685 | 598 | xfs_off_t offset) |
87cbc49c NS |
599 | { |
600 | sector_t bn; | |
8699bb0a | 601 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
602 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
603 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 604 | |
207d0416 CH |
605 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
606 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 607 | |
e513182d | 608 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 609 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 610 | |
046f1685 | 611 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
612 | |
613 | bh->b_blocknr = bn; | |
614 | set_buffer_mapped(bh); | |
615 | } | |
616 | ||
1da177e4 LT |
617 | STATIC void |
618 | xfs_map_at_offset( | |
046f1685 | 619 | struct inode *inode, |
1da177e4 | 620 | struct buffer_head *bh, |
207d0416 | 621 | struct xfs_bmbt_irec *imap, |
046f1685 | 622 | xfs_off_t offset) |
1da177e4 | 623 | { |
207d0416 CH |
624 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
625 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 626 | |
207d0416 | 627 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
628 | set_buffer_mapped(bh); |
629 | clear_buffer_delay(bh); | |
f6d6d4fc | 630 | clear_buffer_unwritten(bh); |
1da177e4 LT |
631 | } |
632 | ||
1da177e4 | 633 | /* |
10ce4444 CH |
634 | * Test if a given page is suitable for writing as part of an unwritten |
635 | * or delayed allocate extent. | |
1da177e4 | 636 | */ |
10ce4444 | 637 | STATIC int |
6ffc4db5 | 638 | xfs_check_page_type( |
10ce4444 | 639 | struct page *page, |
f6d6d4fc | 640 | unsigned int type) |
1da177e4 | 641 | { |
1da177e4 | 642 | if (PageWriteback(page)) |
10ce4444 | 643 | return 0; |
1da177e4 LT |
644 | |
645 | if (page->mapping && page_has_buffers(page)) { | |
646 | struct buffer_head *bh, *head; | |
647 | int acceptable = 0; | |
648 | ||
649 | bh = head = page_buffers(page); | |
650 | do { | |
f6d6d4fc | 651 | if (buffer_unwritten(bh)) |
0d882a36 | 652 | acceptable += (type == XFS_IO_UNWRITTEN); |
f6d6d4fc | 653 | else if (buffer_delay(bh)) |
0d882a36 | 654 | acceptable += (type == XFS_IO_DELALLOC); |
2ddee844 | 655 | else if (buffer_dirty(bh) && buffer_mapped(bh)) |
0d882a36 | 656 | acceptable += (type == XFS_IO_OVERWRITE); |
f6d6d4fc | 657 | else |
1da177e4 | 658 | break; |
1da177e4 LT |
659 | } while ((bh = bh->b_this_page) != head); |
660 | ||
661 | if (acceptable) | |
10ce4444 | 662 | return 1; |
1da177e4 LT |
663 | } |
664 | ||
10ce4444 | 665 | return 0; |
1da177e4 LT |
666 | } |
667 | ||
1da177e4 LT |
668 | /* |
669 | * Allocate & map buffers for page given the extent map. Write it out. | |
670 | * except for the original page of a writepage, this is called on | |
671 | * delalloc/unwritten pages only, for the original page it is possible | |
672 | * that the page has no mapping at all. | |
673 | */ | |
f6d6d4fc | 674 | STATIC int |
1da177e4 LT |
675 | xfs_convert_page( |
676 | struct inode *inode, | |
677 | struct page *page, | |
10ce4444 | 678 | loff_t tindex, |
207d0416 | 679 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 680 | xfs_ioend_t **ioendp, |
2fa24f92 | 681 | struct writeback_control *wbc) |
1da177e4 | 682 | { |
f6d6d4fc | 683 | struct buffer_head *bh, *head; |
9260dc6b CH |
684 | xfs_off_t end_offset; |
685 | unsigned long p_offset; | |
f6d6d4fc | 686 | unsigned int type; |
24e17b5f | 687 | int len, page_dirty; |
f6d6d4fc | 688 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 689 | xfs_off_t offset = page_offset(page); |
1da177e4 | 690 | |
10ce4444 CH |
691 | if (page->index != tindex) |
692 | goto fail; | |
529ae9aa | 693 | if (!trylock_page(page)) |
10ce4444 CH |
694 | goto fail; |
695 | if (PageWriteback(page)) | |
696 | goto fail_unlock_page; | |
697 | if (page->mapping != inode->i_mapping) | |
698 | goto fail_unlock_page; | |
6ffc4db5 | 699 | if (!xfs_check_page_type(page, (*ioendp)->io_type)) |
10ce4444 CH |
700 | goto fail_unlock_page; |
701 | ||
24e17b5f NS |
702 | /* |
703 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 704 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
705 | * |
706 | * Derivation: | |
707 | * | |
708 | * End offset is the highest offset that this page should represent. | |
709 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
710 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
711 | * hence give us the correct page_dirty count. On any other page, | |
712 | * it will be zero and in that case we need page_dirty to be the | |
713 | * count of buffers on the page. | |
24e17b5f | 714 | */ |
9260dc6b CH |
715 | end_offset = min_t(unsigned long long, |
716 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
717 | i_size_read(inode)); | |
718 | ||
480d7467 DC |
719 | /* |
720 | * If the current map does not span the entire page we are about to try | |
721 | * to write, then give up. The only way we can write a page that spans | |
722 | * multiple mappings in a single writeback iteration is via the | |
723 | * xfs_vm_writepage() function. Data integrity writeback requires the | |
724 | * entire page to be written in a single attempt, otherwise the part of | |
725 | * the page we don't write here doesn't get written as part of the data | |
726 | * integrity sync. | |
727 | * | |
728 | * For normal writeback, we also don't attempt to write partial pages | |
729 | * here as it simply means that write_cache_pages() will see it under | |
730 | * writeback and ignore the page until some point in the future, at | |
731 | * which time this will be the only page in the file that needs | |
732 | * writeback. Hence for more optimal IO patterns, we should always | |
733 | * avoid partial page writeback due to multiple mappings on a page here. | |
734 | */ | |
735 | if (!xfs_imap_valid(inode, imap, end_offset)) | |
736 | goto fail_unlock_page; | |
737 | ||
24e17b5f | 738 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
739 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
740 | PAGE_CACHE_SIZE); | |
741 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
742 | page_dirty = p_offset / len; | |
24e17b5f | 743 | |
1da177e4 LT |
744 | bh = head = page_buffers(page); |
745 | do { | |
9260dc6b | 746 | if (offset >= end_offset) |
1da177e4 | 747 | break; |
f6d6d4fc CH |
748 | if (!buffer_uptodate(bh)) |
749 | uptodate = 0; | |
750 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
751 | done = 1; | |
1da177e4 | 752 | continue; |
f6d6d4fc CH |
753 | } |
754 | ||
2fa24f92 CH |
755 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
756 | buffer_mapped(bh)) { | |
9260dc6b | 757 | if (buffer_unwritten(bh)) |
0d882a36 | 758 | type = XFS_IO_UNWRITTEN; |
2fa24f92 | 759 | else if (buffer_delay(bh)) |
0d882a36 | 760 | type = XFS_IO_DELALLOC; |
2fa24f92 | 761 | else |
0d882a36 | 762 | type = XFS_IO_OVERWRITE; |
9260dc6b | 763 | |
558e6891 | 764 | if (!xfs_imap_valid(inode, imap, offset)) { |
f6d6d4fc | 765 | done = 1; |
9260dc6b CH |
766 | continue; |
767 | } | |
768 | ||
ecff71e6 | 769 | lock_buffer(bh); |
0d882a36 | 770 | if (type != XFS_IO_OVERWRITE) |
2fa24f92 | 771 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
772 | xfs_add_to_ioend(inode, bh, offset, type, |
773 | ioendp, done); | |
774 | ||
9260dc6b CH |
775 | page_dirty--; |
776 | count++; | |
777 | } else { | |
2fa24f92 | 778 | done = 1; |
1da177e4 | 779 | } |
7336cea8 | 780 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 781 | |
f6d6d4fc CH |
782 | if (uptodate && bh == head) |
783 | SetPageUptodate(page); | |
784 | ||
89f3b363 | 785 | if (count) { |
efceab1d DC |
786 | if (--wbc->nr_to_write <= 0 && |
787 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 788 | done = 1; |
1da177e4 | 789 | } |
89f3b363 | 790 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
791 | |
792 | return done; | |
10ce4444 CH |
793 | fail_unlock_page: |
794 | unlock_page(page); | |
795 | fail: | |
796 | return 1; | |
1da177e4 LT |
797 | } |
798 | ||
799 | /* | |
800 | * Convert & write out a cluster of pages in the same extent as defined | |
801 | * by mp and following the start page. | |
802 | */ | |
803 | STATIC void | |
804 | xfs_cluster_write( | |
805 | struct inode *inode, | |
806 | pgoff_t tindex, | |
207d0416 | 807 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 808 | xfs_ioend_t **ioendp, |
1da177e4 | 809 | struct writeback_control *wbc, |
1da177e4 LT |
810 | pgoff_t tlast) |
811 | { | |
10ce4444 CH |
812 | struct pagevec pvec; |
813 | int done = 0, i; | |
1da177e4 | 814 | |
10ce4444 CH |
815 | pagevec_init(&pvec, 0); |
816 | while (!done && tindex <= tlast) { | |
817 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
818 | ||
819 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 820 | break; |
10ce4444 CH |
821 | |
822 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
823 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 824 | imap, ioendp, wbc); |
10ce4444 CH |
825 | if (done) |
826 | break; | |
827 | } | |
828 | ||
829 | pagevec_release(&pvec); | |
830 | cond_resched(); | |
1da177e4 LT |
831 | } |
832 | } | |
833 | ||
3ed3a434 DC |
834 | STATIC void |
835 | xfs_vm_invalidatepage( | |
836 | struct page *page, | |
d47992f8 LC |
837 | unsigned int offset, |
838 | unsigned int length) | |
3ed3a434 | 839 | { |
34097dfe LC |
840 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
841 | length); | |
842 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
843 | } |
844 | ||
845 | /* | |
846 | * If the page has delalloc buffers on it, we need to punch them out before we | |
847 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
848 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
849 | * is done on that same region - the delalloc extent is returned when none is | |
850 | * supposed to be there. | |
851 | * | |
852 | * We prevent this by truncating away the delalloc regions on the page before | |
853 | * invalidating it. Because they are delalloc, we can do this without needing a | |
854 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
855 | * truncation without a transaction as there is no space left for block | |
856 | * reservation (typically why we see a ENOSPC in writeback). | |
857 | * | |
858 | * This is not a performance critical path, so for now just do the punching a | |
859 | * buffer head at a time. | |
860 | */ | |
861 | STATIC void | |
862 | xfs_aops_discard_page( | |
863 | struct page *page) | |
864 | { | |
865 | struct inode *inode = page->mapping->host; | |
866 | struct xfs_inode *ip = XFS_I(inode); | |
867 | struct buffer_head *bh, *head; | |
868 | loff_t offset = page_offset(page); | |
3ed3a434 | 869 | |
0d882a36 | 870 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC)) |
3ed3a434 DC |
871 | goto out_invalidate; |
872 | ||
e8c3753c DC |
873 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
874 | goto out_invalidate; | |
875 | ||
4f10700a | 876 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
877 | "page discard on page %p, inode 0x%llx, offset %llu.", |
878 | page, ip->i_ino, offset); | |
879 | ||
880 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
881 | bh = head = page_buffers(page); | |
882 | do { | |
3ed3a434 | 883 | int error; |
c726de44 | 884 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
885 | |
886 | if (!buffer_delay(bh)) | |
887 | goto next_buffer; | |
888 | ||
c726de44 DC |
889 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
890 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
891 | if (error) { |
892 | /* something screwed, just bail */ | |
e8c3753c | 893 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 894 | xfs_alert(ip->i_mount, |
3ed3a434 | 895 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 896 | } |
3ed3a434 DC |
897 | break; |
898 | } | |
899 | next_buffer: | |
c726de44 | 900 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
901 | |
902 | } while ((bh = bh->b_this_page) != head); | |
903 | ||
904 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
905 | out_invalidate: | |
d47992f8 | 906 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
3ed3a434 DC |
907 | return; |
908 | } | |
909 | ||
1da177e4 | 910 | /* |
89f3b363 CH |
911 | * Write out a dirty page. |
912 | * | |
913 | * For delalloc space on the page we need to allocate space and flush it. | |
914 | * For unwritten space on the page we need to start the conversion to | |
915 | * regular allocated space. | |
89f3b363 | 916 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 917 | */ |
1da177e4 | 918 | STATIC int |
89f3b363 CH |
919 | xfs_vm_writepage( |
920 | struct page *page, | |
921 | struct writeback_control *wbc) | |
1da177e4 | 922 | { |
89f3b363 | 923 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 924 | struct buffer_head *bh, *head; |
207d0416 | 925 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 926 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 927 | loff_t offset; |
f6d6d4fc | 928 | unsigned int type; |
1da177e4 | 929 | __uint64_t end_offset; |
bd1556a1 | 930 | pgoff_t end_index, last_index; |
ed1e7b7e | 931 | ssize_t len; |
a206c817 | 932 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 933 | int count = 0; |
a206c817 | 934 | int nonblocking = 0; |
89f3b363 | 935 | |
34097dfe | 936 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 937 | |
20cb52eb CH |
938 | ASSERT(page_has_buffers(page)); |
939 | ||
89f3b363 CH |
940 | /* |
941 | * Refuse to write the page out if we are called from reclaim context. | |
942 | * | |
d4f7a5cb CH |
943 | * This avoids stack overflows when called from deeply used stacks in |
944 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
945 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 946 | * |
94054fa3 MG |
947 | * This should never happen except in the case of a VM regression so |
948 | * warn about it. | |
89f3b363 | 949 | */ |
94054fa3 MG |
950 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
951 | PF_MEMALLOC)) | |
b5420f23 | 952 | goto redirty; |
1da177e4 | 953 | |
89f3b363 | 954 | /* |
680a647b CH |
955 | * Given that we do not allow direct reclaim to call us, we should |
956 | * never be called while in a filesystem transaction. | |
89f3b363 | 957 | */ |
680a647b | 958 | if (WARN_ON(current->flags & PF_FSTRANS)) |
b5420f23 | 959 | goto redirty; |
89f3b363 | 960 | |
1da177e4 LT |
961 | /* Is this page beyond the end of the file? */ |
962 | offset = i_size_read(inode); | |
963 | end_index = offset >> PAGE_CACHE_SHIFT; | |
964 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
965 | if (page->index >= end_index) { | |
6b7a03f0 CH |
966 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
967 | ||
968 | /* | |
ff9a28f6 JK |
969 | * Skip the page if it is fully outside i_size, e.g. due to a |
970 | * truncate operation that is in progress. We must redirty the | |
971 | * page so that reclaim stops reclaiming it. Otherwise | |
972 | * xfs_vm_releasepage() is called on it and gets confused. | |
6b7a03f0 | 973 | */ |
ff9a28f6 JK |
974 | if (page->index >= end_index + 1 || offset_into_page == 0) |
975 | goto redirty; | |
6b7a03f0 CH |
976 | |
977 | /* | |
978 | * The page straddles i_size. It must be zeroed out on each | |
979 | * and every writepage invocation because it may be mmapped. | |
980 | * "A file is mapped in multiples of the page size. For a file | |
981 | * that is not a multiple of the page size, the remaining | |
982 | * memory is zeroed when mapped, and writes to that region are | |
983 | * not written out to the file." | |
984 | */ | |
985 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
1da177e4 LT |
986 | } |
987 | ||
f6d6d4fc | 988 | end_offset = min_t(unsigned long long, |
20cb52eb CH |
989 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, |
990 | offset); | |
24e17b5f | 991 | len = 1 << inode->i_blkbits; |
24e17b5f | 992 | |
24e17b5f | 993 | bh = head = page_buffers(page); |
f6d6d4fc | 994 | offset = page_offset(page); |
0d882a36 | 995 | type = XFS_IO_OVERWRITE; |
a206c817 | 996 | |
dbcdde3e | 997 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 998 | nonblocking = 1; |
f6d6d4fc | 999 | |
1da177e4 | 1000 | do { |
6ac7248e CH |
1001 | int new_ioend = 0; |
1002 | ||
1da177e4 LT |
1003 | if (offset >= end_offset) |
1004 | break; | |
1005 | if (!buffer_uptodate(bh)) | |
1006 | uptodate = 0; | |
1da177e4 | 1007 | |
3d9b02e3 | 1008 | /* |
ece413f5 CH |
1009 | * set_page_dirty dirties all buffers in a page, independent |
1010 | * of their state. The dirty state however is entirely | |
1011 | * meaningless for holes (!mapped && uptodate), so skip | |
1012 | * buffers covering holes here. | |
3d9b02e3 ES |
1013 | */ |
1014 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
1015 | imap_valid = 0; |
1016 | continue; | |
1017 | } | |
1018 | ||
aeea1b1f | 1019 | if (buffer_unwritten(bh)) { |
0d882a36 AR |
1020 | if (type != XFS_IO_UNWRITTEN) { |
1021 | type = XFS_IO_UNWRITTEN; | |
aeea1b1f | 1022 | imap_valid = 0; |
1da177e4 | 1023 | } |
aeea1b1f | 1024 | } else if (buffer_delay(bh)) { |
0d882a36 AR |
1025 | if (type != XFS_IO_DELALLOC) { |
1026 | type = XFS_IO_DELALLOC; | |
aeea1b1f | 1027 | imap_valid = 0; |
1da177e4 | 1028 | } |
89f3b363 | 1029 | } else if (buffer_uptodate(bh)) { |
0d882a36 AR |
1030 | if (type != XFS_IO_OVERWRITE) { |
1031 | type = XFS_IO_OVERWRITE; | |
85da94c6 CH |
1032 | imap_valid = 0; |
1033 | } | |
aeea1b1f | 1034 | } else { |
7d0fa3ec | 1035 | if (PageUptodate(page)) |
aeea1b1f | 1036 | ASSERT(buffer_mapped(bh)); |
7d0fa3ec AR |
1037 | /* |
1038 | * This buffer is not uptodate and will not be | |
1039 | * written to disk. Ensure that we will put any | |
1040 | * subsequent writeable buffers into a new | |
1041 | * ioend. | |
1042 | */ | |
1043 | imap_valid = 0; | |
aeea1b1f CH |
1044 | continue; |
1045 | } | |
d5cb48aa | 1046 | |
aeea1b1f CH |
1047 | if (imap_valid) |
1048 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1049 | if (!imap_valid) { | |
1050 | /* | |
1051 | * If we didn't have a valid mapping then we need to | |
1052 | * put the new mapping into a separate ioend structure. | |
1053 | * This ensures non-contiguous extents always have | |
1054 | * separate ioends, which is particularly important | |
1055 | * for unwritten extent conversion at I/O completion | |
1056 | * time. | |
1057 | */ | |
1058 | new_ioend = 1; | |
1059 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1060 | nonblocking); | |
1061 | if (err) | |
1062 | goto error; | |
1063 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1064 | } | |
1065 | if (imap_valid) { | |
ecff71e6 | 1066 | lock_buffer(bh); |
0d882a36 | 1067 | if (type != XFS_IO_OVERWRITE) |
aeea1b1f CH |
1068 | xfs_map_at_offset(inode, bh, &imap, offset); |
1069 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1070 | new_ioend); | |
1071 | count++; | |
1da177e4 | 1072 | } |
f6d6d4fc CH |
1073 | |
1074 | if (!iohead) | |
1075 | iohead = ioend; | |
1076 | ||
1077 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1078 | |
1079 | if (uptodate && bh == head) | |
1080 | SetPageUptodate(page); | |
1081 | ||
89f3b363 | 1082 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1083 | |
7bf7f352 DC |
1084 | /* if there is no IO to be submitted for this page, we are done */ |
1085 | if (!ioend) | |
1086 | return 0; | |
1087 | ||
1088 | ASSERT(iohead); | |
1089 | ||
1090 | /* | |
1091 | * Any errors from this point onwards need tobe reported through the IO | |
1092 | * completion path as we have marked the initial page as under writeback | |
1093 | * and unlocked it. | |
1094 | */ | |
1095 | if (imap_valid) { | |
bd1556a1 CH |
1096 | xfs_off_t end_index; |
1097 | ||
1098 | end_index = imap.br_startoff + imap.br_blockcount; | |
1099 | ||
1100 | /* to bytes */ | |
1101 | end_index <<= inode->i_blkbits; | |
1102 | ||
1103 | /* to pages */ | |
1104 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1105 | ||
1106 | /* check against file size */ | |
1107 | if (end_index > last_index) | |
1108 | end_index = last_index; | |
8699bb0a | 1109 | |
207d0416 | 1110 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1111 | wbc, end_index); |
1da177e4 LT |
1112 | } |
1113 | ||
281627df | 1114 | |
7bf7f352 DC |
1115 | /* |
1116 | * Reserve log space if we might write beyond the on-disk inode size. | |
1117 | */ | |
1118 | err = 0; | |
1119 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) | |
1120 | err = xfs_setfilesize_trans_alloc(ioend); | |
1121 | ||
1122 | xfs_submit_ioend(wbc, iohead, err); | |
f6d6d4fc | 1123 | |
89f3b363 | 1124 | return 0; |
1da177e4 LT |
1125 | |
1126 | error: | |
f6d6d4fc CH |
1127 | if (iohead) |
1128 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1129 | |
b5420f23 CH |
1130 | if (err == -EAGAIN) |
1131 | goto redirty; | |
1132 | ||
20cb52eb | 1133 | xfs_aops_discard_page(page); |
89f3b363 CH |
1134 | ClearPageUptodate(page); |
1135 | unlock_page(page); | |
1da177e4 | 1136 | return err; |
f51623b2 | 1137 | |
b5420f23 | 1138 | redirty: |
f51623b2 NS |
1139 | redirty_page_for_writepage(wbc, page); |
1140 | unlock_page(page); | |
1141 | return 0; | |
f51623b2 NS |
1142 | } |
1143 | ||
7d4fb40a NS |
1144 | STATIC int |
1145 | xfs_vm_writepages( | |
1146 | struct address_space *mapping, | |
1147 | struct writeback_control *wbc) | |
1148 | { | |
b3aea4ed | 1149 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1150 | return generic_writepages(mapping, wbc); |
1151 | } | |
1152 | ||
f51623b2 NS |
1153 | /* |
1154 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1155 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1156 | * have buffer heads in this call. |
1157 | * | |
89f3b363 | 1158 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1159 | */ |
1160 | STATIC int | |
238f4c54 | 1161 | xfs_vm_releasepage( |
f51623b2 NS |
1162 | struct page *page, |
1163 | gfp_t gfp_mask) | |
1164 | { | |
20cb52eb | 1165 | int delalloc, unwritten; |
f51623b2 | 1166 | |
34097dfe | 1167 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1168 | |
20cb52eb | 1169 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1170 | |
89f3b363 | 1171 | if (WARN_ON(delalloc)) |
f51623b2 | 1172 | return 0; |
89f3b363 | 1173 | if (WARN_ON(unwritten)) |
f51623b2 NS |
1174 | return 0; |
1175 | ||
f51623b2 NS |
1176 | return try_to_free_buffers(page); |
1177 | } | |
1178 | ||
1da177e4 | 1179 | STATIC int |
c2536668 | 1180 | __xfs_get_blocks( |
1da177e4 LT |
1181 | struct inode *inode, |
1182 | sector_t iblock, | |
1da177e4 LT |
1183 | struct buffer_head *bh_result, |
1184 | int create, | |
f2bde9b8 | 1185 | int direct) |
1da177e4 | 1186 | { |
a206c817 CH |
1187 | struct xfs_inode *ip = XFS_I(inode); |
1188 | struct xfs_mount *mp = ip->i_mount; | |
1189 | xfs_fileoff_t offset_fsb, end_fsb; | |
1190 | int error = 0; | |
1191 | int lockmode = 0; | |
207d0416 | 1192 | struct xfs_bmbt_irec imap; |
a206c817 | 1193 | int nimaps = 1; |
fdc7ed75 NS |
1194 | xfs_off_t offset; |
1195 | ssize_t size; | |
207d0416 | 1196 | int new = 0; |
a206c817 CH |
1197 | |
1198 | if (XFS_FORCED_SHUTDOWN(mp)) | |
1199 | return -XFS_ERROR(EIO); | |
1da177e4 | 1200 | |
fdc7ed75 | 1201 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1202 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1203 | size = bh_result->b_size; | |
364f358a LM |
1204 | |
1205 | if (!create && direct && offset >= i_size_read(inode)) | |
1206 | return 0; | |
1207 | ||
507630b2 DC |
1208 | /* |
1209 | * Direct I/O is usually done on preallocated files, so try getting | |
1210 | * a block mapping without an exclusive lock first. For buffered | |
1211 | * writes we already have the exclusive iolock anyway, so avoiding | |
1212 | * a lock roundtrip here by taking the ilock exclusive from the | |
1213 | * beginning is a useful micro optimization. | |
1214 | */ | |
1215 | if (create && !direct) { | |
a206c817 CH |
1216 | lockmode = XFS_ILOCK_EXCL; |
1217 | xfs_ilock(ip, lockmode); | |
1218 | } else { | |
1219 | lockmode = xfs_ilock_map_shared(ip); | |
1220 | } | |
f2bde9b8 | 1221 | |
d2c28191 DC |
1222 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1223 | if (offset + size > mp->m_super->s_maxbytes) | |
1224 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1225 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1226 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1227 | ||
5c8ed202 DC |
1228 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1229 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1230 | if (error) |
a206c817 CH |
1231 | goto out_unlock; |
1232 | ||
1233 | if (create && | |
1234 | (!nimaps || | |
1235 | (imap.br_startblock == HOLESTARTBLOCK || | |
1236 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
aff3a9ed | 1237 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 DC |
1238 | /* |
1239 | * Drop the ilock in preparation for starting the block | |
1240 | * allocation transaction. It will be retaken | |
1241 | * exclusively inside xfs_iomap_write_direct for the | |
1242 | * actual allocation. | |
1243 | */ | |
1244 | xfs_iunlock(ip, lockmode); | |
a206c817 CH |
1245 | error = xfs_iomap_write_direct(ip, offset, size, |
1246 | &imap, nimaps); | |
507630b2 DC |
1247 | if (error) |
1248 | return -error; | |
d3bc815a | 1249 | new = 1; |
a206c817 | 1250 | } else { |
507630b2 DC |
1251 | /* |
1252 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1253 | * we can go on without dropping the lock here. If we |
1254 | * are allocating a new delalloc block, make sure that | |
1255 | * we set the new flag so that we mark the buffer new so | |
1256 | * that we know that it is newly allocated if the write | |
1257 | * fails. | |
507630b2 | 1258 | */ |
d3bc815a DC |
1259 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1260 | new = 1; | |
a206c817 | 1261 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1262 | if (error) |
1263 | goto out_unlock; | |
1264 | ||
1265 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1266 | } |
a206c817 CH |
1267 | |
1268 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1269 | } else if (nimaps) { | |
1270 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
507630b2 | 1271 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1272 | } else { |
1273 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1274 | goto out_unlock; | |
1275 | } | |
1da177e4 | 1276 | |
207d0416 CH |
1277 | if (imap.br_startblock != HOLESTARTBLOCK && |
1278 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1279 | /* |
1280 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1281 | * the read case (treat as if we're reading into a hole). |
1282 | */ | |
207d0416 CH |
1283 | if (create || !ISUNWRITTEN(&imap)) |
1284 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1285 | if (create && ISUNWRITTEN(&imap)) { | |
7b7a8665 | 1286 | if (direct) { |
1da177e4 | 1287 | bh_result->b_private = inode; |
7b7a8665 CH |
1288 | set_buffer_defer_completion(bh_result); |
1289 | } | |
1da177e4 | 1290 | set_buffer_unwritten(bh_result); |
1da177e4 LT |
1291 | } |
1292 | } | |
1293 | ||
c2536668 NS |
1294 | /* |
1295 | * If this is a realtime file, data may be on a different device. | |
1296 | * to that pointed to from the buffer_head b_bdev currently. | |
1297 | */ | |
046f1685 | 1298 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1299 | |
c2536668 | 1300 | /* |
549054af DC |
1301 | * If we previously allocated a block out beyond eof and we are now |
1302 | * coming back to use it then we will need to flag it as new even if it | |
1303 | * has a disk address. | |
1304 | * | |
1305 | * With sub-block writes into unwritten extents we also need to mark | |
1306 | * the buffer as new so that the unwritten parts of the buffer gets | |
1307 | * correctly zeroed. | |
1da177e4 LT |
1308 | */ |
1309 | if (create && | |
1310 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1311 | (offset >= i_size_read(inode)) || |
207d0416 | 1312 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1313 | set_buffer_new(bh_result); |
1da177e4 | 1314 | |
207d0416 | 1315 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1316 | BUG_ON(direct); |
1317 | if (create) { | |
1318 | set_buffer_uptodate(bh_result); | |
1319 | set_buffer_mapped(bh_result); | |
1320 | set_buffer_delay(bh_result); | |
1321 | } | |
1322 | } | |
1323 | ||
2b8f12b7 CH |
1324 | /* |
1325 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1326 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
1327 | */ | |
c2536668 | 1328 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1329 | xfs_off_t mapping_size; |
1330 | ||
1331 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1332 | mapping_size <<= inode->i_blkbits; | |
1333 | ||
1334 | ASSERT(mapping_size > 0); | |
1335 | if (mapping_size > size) | |
1336 | mapping_size = size; | |
1337 | if (mapping_size > LONG_MAX) | |
1338 | mapping_size = LONG_MAX; | |
1339 | ||
1340 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | return 0; | |
a206c817 CH |
1344 | |
1345 | out_unlock: | |
1346 | xfs_iunlock(ip, lockmode); | |
1347 | return -error; | |
1da177e4 LT |
1348 | } |
1349 | ||
1350 | int | |
c2536668 | 1351 | xfs_get_blocks( |
1da177e4 LT |
1352 | struct inode *inode, |
1353 | sector_t iblock, | |
1354 | struct buffer_head *bh_result, | |
1355 | int create) | |
1356 | { | |
f2bde9b8 | 1357 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1358 | } |
1359 | ||
1360 | STATIC int | |
e4c573bb | 1361 | xfs_get_blocks_direct( |
1da177e4 LT |
1362 | struct inode *inode, |
1363 | sector_t iblock, | |
1da177e4 LT |
1364 | struct buffer_head *bh_result, |
1365 | int create) | |
1366 | { | |
f2bde9b8 | 1367 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1368 | } |
1369 | ||
209fb87a CH |
1370 | /* |
1371 | * Complete a direct I/O write request. | |
1372 | * | |
1373 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1374 | * need to issue a transaction to convert the range from unwritten to written | |
1375 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1376 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1377 | * request this handler is called from interrupt context, from which we |
1378 | * can't start transactions. In that case offload the I/O completion to | |
1379 | * the workqueues we also use for buffered I/O completion. | |
1380 | */ | |
f0973863 | 1381 | STATIC void |
209fb87a CH |
1382 | xfs_end_io_direct_write( |
1383 | struct kiocb *iocb, | |
1384 | loff_t offset, | |
1385 | ssize_t size, | |
7b7a8665 | 1386 | void *private) |
f0973863 | 1387 | { |
209fb87a | 1388 | struct xfs_ioend *ioend = iocb->private; |
f0973863 | 1389 | |
2813d682 CH |
1390 | /* |
1391 | * While the generic direct I/O code updates the inode size, it does | |
1392 | * so only after the end_io handler is called, which means our | |
1393 | * end_io handler thinks the on-disk size is outside the in-core | |
1394 | * size. To prevent this just update it a little bit earlier here. | |
1395 | */ | |
1396 | if (offset + size > i_size_read(ioend->io_inode)) | |
1397 | i_size_write(ioend->io_inode, offset + size); | |
1398 | ||
f0973863 | 1399 | /* |
209fb87a CH |
1400 | * blockdev_direct_IO can return an error even after the I/O |
1401 | * completion handler was called. Thus we need to protect | |
1402 | * against double-freeing. | |
f0973863 | 1403 | */ |
209fb87a CH |
1404 | iocb->private = NULL; |
1405 | ||
ba87ea69 LM |
1406 | ioend->io_offset = offset; |
1407 | ioend->io_size = size; | |
209fb87a | 1408 | if (private && size > 0) |
0d882a36 | 1409 | ioend->io_type = XFS_IO_UNWRITTEN; |
209fb87a | 1410 | |
7b7a8665 | 1411 | xfs_finish_ioend_sync(ioend); |
f0973863 CH |
1412 | } |
1413 | ||
1da177e4 | 1414 | STATIC ssize_t |
e4c573bb | 1415 | xfs_vm_direct_IO( |
1da177e4 LT |
1416 | int rw, |
1417 | struct kiocb *iocb, | |
1418 | const struct iovec *iov, | |
1419 | loff_t offset, | |
1420 | unsigned long nr_segs) | |
1421 | { | |
209fb87a CH |
1422 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1423 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
281627df | 1424 | struct xfs_ioend *ioend = NULL; |
209fb87a CH |
1425 | ssize_t ret; |
1426 | ||
1427 | if (rw & WRITE) { | |
281627df CH |
1428 | size_t size = iov_length(iov, nr_segs); |
1429 | ||
1430 | /* | |
437a255a DC |
1431 | * We cannot preallocate a size update transaction here as we |
1432 | * don't know whether allocation is necessary or not. Hence we | |
1433 | * can only tell IO completion that one is necessary if we are | |
1434 | * not doing unwritten extent conversion. | |
281627df | 1435 | */ |
0d882a36 | 1436 | iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT); |
437a255a | 1437 | if (offset + size > XFS_I(inode)->i_d.di_size) |
281627df | 1438 | ioend->io_isdirect = 1; |
209fb87a | 1439 | |
eafdc7d1 CH |
1440 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1441 | offset, nr_segs, | |
1442 | xfs_get_blocks_direct, | |
1443 | xfs_end_io_direct_write, NULL, 0); | |
209fb87a | 1444 | if (ret != -EIOCBQUEUED && iocb->private) |
437a255a | 1445 | goto out_destroy_ioend; |
209fb87a | 1446 | } else { |
eafdc7d1 CH |
1447 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov, |
1448 | offset, nr_segs, | |
1449 | xfs_get_blocks_direct, | |
1450 | NULL, NULL, 0); | |
209fb87a | 1451 | } |
f0973863 | 1452 | |
f0973863 | 1453 | return ret; |
281627df | 1454 | |
281627df CH |
1455 | out_destroy_ioend: |
1456 | xfs_destroy_ioend(ioend); | |
1457 | return ret; | |
1da177e4 LT |
1458 | } |
1459 | ||
d3bc815a DC |
1460 | /* |
1461 | * Punch out the delalloc blocks we have already allocated. | |
1462 | * | |
1463 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1464 | * as the page is still locked at this point. | |
1465 | */ | |
1466 | STATIC void | |
1467 | xfs_vm_kill_delalloc_range( | |
1468 | struct inode *inode, | |
1469 | loff_t start, | |
1470 | loff_t end) | |
1471 | { | |
1472 | struct xfs_inode *ip = XFS_I(inode); | |
1473 | xfs_fileoff_t start_fsb; | |
1474 | xfs_fileoff_t end_fsb; | |
1475 | int error; | |
1476 | ||
1477 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1478 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1479 | if (end_fsb <= start_fsb) | |
1480 | return; | |
1481 | ||
1482 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1483 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1484 | end_fsb - start_fsb); | |
1485 | if (error) { | |
1486 | /* something screwed, just bail */ | |
1487 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1488 | xfs_alert(ip->i_mount, | |
1489 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1490 | ip->i_ino); | |
1491 | } | |
1492 | } | |
1493 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1494 | } | |
1495 | ||
fa9b227e CH |
1496 | STATIC void |
1497 | xfs_vm_write_failed( | |
d3bc815a DC |
1498 | struct inode *inode, |
1499 | struct page *page, | |
1500 | loff_t pos, | |
1501 | unsigned len) | |
fa9b227e | 1502 | { |
58e59854 | 1503 | loff_t block_offset; |
d3bc815a DC |
1504 | loff_t block_start; |
1505 | loff_t block_end; | |
1506 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1507 | loff_t to = from + len; | |
1508 | struct buffer_head *bh, *head; | |
fa9b227e | 1509 | |
58e59854 JL |
1510 | /* |
1511 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1512 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1513 | * platform, the high 32-bit will be masked off if we evaluate the | |
1514 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1515 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1516 | * which could cause the following ASSERT failed in most cases. | |
1517 | * In order to avoid this, we can evaluate the block_offset of the | |
1518 | * start of the page by using shifts rather than masks the mismatch | |
1519 | * problem. | |
1520 | */ | |
1521 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1522 | ||
d3bc815a | 1523 | ASSERT(block_offset + from == pos); |
c726de44 | 1524 | |
d3bc815a DC |
1525 | head = page_buffers(page); |
1526 | block_start = 0; | |
1527 | for (bh = head; bh != head || !block_start; | |
1528 | bh = bh->b_this_page, block_start = block_end, | |
1529 | block_offset += bh->b_size) { | |
1530 | block_end = block_start + bh->b_size; | |
c726de44 | 1531 | |
d3bc815a DC |
1532 | /* skip buffers before the write */ |
1533 | if (block_end <= from) | |
1534 | continue; | |
1535 | ||
1536 | /* if the buffer is after the write, we're done */ | |
1537 | if (block_start >= to) | |
1538 | break; | |
1539 | ||
1540 | if (!buffer_delay(bh)) | |
1541 | continue; | |
1542 | ||
1543 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1544 | continue; | |
1545 | ||
1546 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1547 | block_offset + bh->b_size); | |
fa9b227e | 1548 | } |
d3bc815a | 1549 | |
fa9b227e CH |
1550 | } |
1551 | ||
d3bc815a DC |
1552 | /* |
1553 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1554 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1555 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1556 | * the appropriate point. | |
1557 | */ | |
f51623b2 | 1558 | STATIC int |
d79689c7 | 1559 | xfs_vm_write_begin( |
f51623b2 | 1560 | struct file *file, |
d79689c7 NP |
1561 | struct address_space *mapping, |
1562 | loff_t pos, | |
1563 | unsigned len, | |
1564 | unsigned flags, | |
1565 | struct page **pagep, | |
1566 | void **fsdata) | |
f51623b2 | 1567 | { |
d3bc815a DC |
1568 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1569 | struct page *page; | |
1570 | int status; | |
155130a4 | 1571 | |
d3bc815a DC |
1572 | ASSERT(len <= PAGE_CACHE_SIZE); |
1573 | ||
1574 | page = grab_cache_page_write_begin(mapping, index, | |
1575 | flags | AOP_FLAG_NOFS); | |
1576 | if (!page) | |
1577 | return -ENOMEM; | |
1578 | ||
1579 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1580 | if (unlikely(status)) { | |
1581 | struct inode *inode = mapping->host; | |
1582 | ||
1583 | xfs_vm_write_failed(inode, page, pos, len); | |
1584 | unlock_page(page); | |
1585 | ||
1586 | if (pos + len > i_size_read(inode)) | |
7caef267 | 1587 | truncate_pagecache(inode, i_size_read(inode)); |
d3bc815a DC |
1588 | |
1589 | page_cache_release(page); | |
1590 | page = NULL; | |
1591 | } | |
1592 | ||
1593 | *pagep = page; | |
1594 | return status; | |
fa9b227e CH |
1595 | } |
1596 | ||
d3bc815a DC |
1597 | /* |
1598 | * On failure, we only need to kill delalloc blocks beyond EOF because they | |
1599 | * will never be written. For blocks within EOF, generic_write_end() zeros them | |
1600 | * so they are safe to leave alone and be written with all the other valid data. | |
1601 | */ | |
fa9b227e CH |
1602 | STATIC int |
1603 | xfs_vm_write_end( | |
1604 | struct file *file, | |
1605 | struct address_space *mapping, | |
1606 | loff_t pos, | |
1607 | unsigned len, | |
1608 | unsigned copied, | |
1609 | struct page *page, | |
1610 | void *fsdata) | |
1611 | { | |
1612 | int ret; | |
155130a4 | 1613 | |
d3bc815a DC |
1614 | ASSERT(len <= PAGE_CACHE_SIZE); |
1615 | ||
fa9b227e | 1616 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1617 | if (unlikely(ret < len)) { |
1618 | struct inode *inode = mapping->host; | |
1619 | size_t isize = i_size_read(inode); | |
1620 | loff_t to = pos + len; | |
1621 | ||
1622 | if (to > isize) { | |
7caef267 | 1623 | truncate_pagecache(inode, isize); |
d3bc815a DC |
1624 | xfs_vm_kill_delalloc_range(inode, isize, to); |
1625 | } | |
1626 | } | |
155130a4 | 1627 | return ret; |
f51623b2 | 1628 | } |
1da177e4 LT |
1629 | |
1630 | STATIC sector_t | |
e4c573bb | 1631 | xfs_vm_bmap( |
1da177e4 LT |
1632 | struct address_space *mapping, |
1633 | sector_t block) | |
1634 | { | |
1635 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1636 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1637 | |
cca28fb8 | 1638 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1639 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1640 | filemap_write_and_wait(mapping); |
126468b1 | 1641 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1642 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1643 | } |
1644 | ||
1645 | STATIC int | |
e4c573bb | 1646 | xfs_vm_readpage( |
1da177e4 LT |
1647 | struct file *unused, |
1648 | struct page *page) | |
1649 | { | |
c2536668 | 1650 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1651 | } |
1652 | ||
1653 | STATIC int | |
e4c573bb | 1654 | xfs_vm_readpages( |
1da177e4 LT |
1655 | struct file *unused, |
1656 | struct address_space *mapping, | |
1657 | struct list_head *pages, | |
1658 | unsigned nr_pages) | |
1659 | { | |
c2536668 | 1660 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1661 | } |
1662 | ||
f5e54d6e | 1663 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1664 | .readpage = xfs_vm_readpage, |
1665 | .readpages = xfs_vm_readpages, | |
1666 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1667 | .writepages = xfs_vm_writepages, |
238f4c54 NS |
1668 | .releasepage = xfs_vm_releasepage, |
1669 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1670 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1671 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1672 | .bmap = xfs_vm_bmap, |
1673 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1674 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1675 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1676 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1677 | }; |