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 | 19 | #include "xfs_shared.h" |
239880ef DC |
20 | #include "xfs_format.h" |
21 | #include "xfs_log_format.h" | |
22 | #include "xfs_trans_resv.h" | |
1da177e4 | 23 | #include "xfs_sb.h" |
a844f451 | 24 | #include "xfs_ag.h" |
1da177e4 | 25 | #include "xfs_mount.h" |
1da177e4 | 26 | #include "xfs_inode.h" |
239880ef | 27 | #include "xfs_trans.h" |
281627df | 28 | #include "xfs_inode_item.h" |
a844f451 | 29 | #include "xfs_alloc.h" |
1da177e4 | 30 | #include "xfs_error.h" |
1da177e4 | 31 | #include "xfs_iomap.h" |
0b1b213f | 32 | #include "xfs_trace.h" |
3ed3a434 | 33 | #include "xfs_bmap.h" |
68988114 | 34 | #include "xfs_bmap_util.h" |
a4fbe6ab DC |
35 | #include "xfs_bmap_btree.h" |
36 | #include "xfs_dinode.h" | |
a27bb332 | 37 | #include <linux/aio.h> |
5a0e3ad6 | 38 | #include <linux/gfp.h> |
1da177e4 | 39 | #include <linux/mpage.h> |
10ce4444 | 40 | #include <linux/pagevec.h> |
1da177e4 LT |
41 | #include <linux/writeback.h> |
42 | ||
0b1b213f | 43 | void |
f51623b2 NS |
44 | xfs_count_page_state( |
45 | struct page *page, | |
46 | int *delalloc, | |
f51623b2 NS |
47 | int *unwritten) |
48 | { | |
49 | struct buffer_head *bh, *head; | |
50 | ||
20cb52eb | 51 | *delalloc = *unwritten = 0; |
f51623b2 NS |
52 | |
53 | bh = head = page_buffers(page); | |
54 | do { | |
20cb52eb | 55 | if (buffer_unwritten(bh)) |
f51623b2 NS |
56 | (*unwritten) = 1; |
57 | else if (buffer_delay(bh)) | |
58 | (*delalloc) = 1; | |
59 | } while ((bh = bh->b_this_page) != head); | |
60 | } | |
61 | ||
6214ed44 CH |
62 | STATIC struct block_device * |
63 | xfs_find_bdev_for_inode( | |
046f1685 | 64 | struct inode *inode) |
6214ed44 | 65 | { |
046f1685 | 66 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
67 | struct xfs_mount *mp = ip->i_mount; |
68 | ||
71ddabb9 | 69 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
70 | return mp->m_rtdev_targp->bt_bdev; |
71 | else | |
72 | return mp->m_ddev_targp->bt_bdev; | |
73 | } | |
74 | ||
f6d6d4fc CH |
75 | /* |
76 | * We're now finished for good with this ioend structure. | |
77 | * Update the page state via the associated buffer_heads, | |
78 | * release holds on the inode and bio, and finally free | |
79 | * up memory. Do not use the ioend after this. | |
80 | */ | |
0829c360 CH |
81 | STATIC void |
82 | xfs_destroy_ioend( | |
83 | xfs_ioend_t *ioend) | |
84 | { | |
f6d6d4fc CH |
85 | struct buffer_head *bh, *next; |
86 | ||
87 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
88 | next = bh->b_private; | |
7d04a335 | 89 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 90 | } |
583fa586 | 91 | |
0829c360 CH |
92 | mempool_free(ioend, xfs_ioend_pool); |
93 | } | |
94 | ||
fc0063c4 CH |
95 | /* |
96 | * Fast and loose check if this write could update the on-disk inode size. | |
97 | */ | |
98 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
99 | { | |
100 | return ioend->io_offset + ioend->io_size > | |
101 | XFS_I(ioend->io_inode)->i_d.di_size; | |
102 | } | |
103 | ||
281627df CH |
104 | STATIC int |
105 | xfs_setfilesize_trans_alloc( | |
106 | struct xfs_ioend *ioend) | |
107 | { | |
108 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
109 | struct xfs_trans *tp; | |
110 | int error; | |
111 | ||
112 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
113 | ||
3d3c8b52 | 114 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
281627df CH |
115 | if (error) { |
116 | xfs_trans_cancel(tp, 0); | |
117 | return error; | |
118 | } | |
119 | ||
120 | ioend->io_append_trans = tp; | |
121 | ||
d9457dc0 | 122 | /* |
437a255a | 123 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
124 | * we released it. |
125 | */ | |
126 | rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
127 | 1, _THIS_IP_); | |
281627df CH |
128 | /* |
129 | * We hand off the transaction to the completion thread now, so | |
130 | * clear the flag here. | |
131 | */ | |
132 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
133 | return 0; | |
134 | } | |
135 | ||
ba87ea69 | 136 | /* |
2813d682 | 137 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 138 | */ |
281627df | 139 | STATIC int |
ba87ea69 | 140 | xfs_setfilesize( |
aa6bf01d | 141 | struct xfs_ioend *ioend) |
ba87ea69 | 142 | { |
aa6bf01d | 143 | struct xfs_inode *ip = XFS_I(ioend->io_inode); |
281627df | 144 | struct xfs_trans *tp = ioend->io_append_trans; |
ba87ea69 | 145 | xfs_fsize_t isize; |
ba87ea69 | 146 | |
281627df | 147 | /* |
437a255a DC |
148 | * The transaction may have been allocated in the I/O submission thread, |
149 | * thus we need to mark ourselves as beeing in a transaction manually. | |
150 | * Similarly for freeze protection. | |
281627df CH |
151 | */ |
152 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
437a255a DC |
153 | rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], |
154 | 0, 1, _THIS_IP_); | |
281627df | 155 | |
aa6bf01d | 156 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
6923e686 | 157 | isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size); |
281627df CH |
158 | if (!isize) { |
159 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
160 | xfs_trans_cancel(tp, 0); | |
161 | return 0; | |
ba87ea69 LM |
162 | } |
163 | ||
281627df CH |
164 | trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size); |
165 | ||
166 | ip->i_d.di_size = isize; | |
167 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
168 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
169 | ||
170 | return xfs_trans_commit(tp, 0); | |
77d7a0c2 DC |
171 | } |
172 | ||
173 | /* | |
209fb87a | 174 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
175 | * |
176 | * If there is no work to do we might as well call it a day and free the | |
177 | * ioend right now. | |
77d7a0c2 DC |
178 | */ |
179 | STATIC void | |
180 | xfs_finish_ioend( | |
209fb87a | 181 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
182 | { |
183 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
184 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
185 | ||
0d882a36 | 186 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
aa6bf01d | 187 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
437a255a DC |
188 | else if (ioend->io_append_trans || |
189 | (ioend->io_isdirect && xfs_ioend_is_append(ioend))) | |
aa6bf01d | 190 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
191 | else |
192 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 193 | } |
ba87ea69 LM |
194 | } |
195 | ||
0829c360 | 196 | /* |
5ec4fabb | 197 | * IO write completion. |
f6d6d4fc CH |
198 | */ |
199 | STATIC void | |
5ec4fabb | 200 | xfs_end_io( |
77d7a0c2 | 201 | struct work_struct *work) |
0829c360 | 202 | { |
77d7a0c2 DC |
203 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
204 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 205 | int error = 0; |
ba87ea69 | 206 | |
04f658ee | 207 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 208 | ioend->io_error = -EIO; |
04f658ee CH |
209 | goto done; |
210 | } | |
211 | if (ioend->io_error) | |
212 | goto done; | |
213 | ||
5ec4fabb CH |
214 | /* |
215 | * For unwritten extents we need to issue transactions to convert a | |
216 | * range to normal written extens after the data I/O has finished. | |
217 | */ | |
0d882a36 | 218 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
437a255a DC |
219 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
220 | ioend->io_size); | |
221 | } else if (ioend->io_isdirect && xfs_ioend_is_append(ioend)) { | |
281627df | 222 | /* |
437a255a DC |
223 | * For direct I/O we do not know if we need to allocate blocks |
224 | * or not so we can't preallocate an append transaction as that | |
225 | * results in nested reservations and log space deadlocks. Hence | |
226 | * allocate the transaction here. While this is sub-optimal and | |
227 | * can block IO completion for some time, we're stuck with doing | |
228 | * it this way until we can pass the ioend to the direct IO | |
229 | * allocation callbacks and avoid nesting that way. | |
281627df | 230 | */ |
437a255a DC |
231 | error = xfs_setfilesize_trans_alloc(ioend); |
232 | if (error) | |
04f658ee | 233 | goto done; |
437a255a | 234 | error = xfs_setfilesize(ioend); |
281627df CH |
235 | } else if (ioend->io_append_trans) { |
236 | error = xfs_setfilesize(ioend); | |
84803fb7 | 237 | } else { |
281627df | 238 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 239 | } |
ba87ea69 | 240 | |
04f658ee | 241 | done: |
437a255a | 242 | if (error) |
2451337d | 243 | ioend->io_error = error; |
aa6bf01d | 244 | xfs_destroy_ioend(ioend); |
c626d174 DC |
245 | } |
246 | ||
209fb87a CH |
247 | /* |
248 | * Call IO completion handling in caller context on the final put of an ioend. | |
249 | */ | |
250 | STATIC void | |
251 | xfs_finish_ioend_sync( | |
252 | struct xfs_ioend *ioend) | |
253 | { | |
254 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
255 | xfs_end_io(&ioend->io_work); | |
256 | } | |
257 | ||
0829c360 CH |
258 | /* |
259 | * Allocate and initialise an IO completion structure. | |
260 | * We need to track unwritten extent write completion here initially. | |
261 | * We'll need to extend this for updating the ondisk inode size later | |
262 | * (vs. incore size). | |
263 | */ | |
264 | STATIC xfs_ioend_t * | |
265 | xfs_alloc_ioend( | |
f6d6d4fc CH |
266 | struct inode *inode, |
267 | unsigned int type) | |
0829c360 CH |
268 | { |
269 | xfs_ioend_t *ioend; | |
270 | ||
271 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
272 | ||
273 | /* | |
274 | * Set the count to 1 initially, which will prevent an I/O | |
275 | * completion callback from happening before we have started | |
276 | * all the I/O from calling the completion routine too early. | |
277 | */ | |
278 | atomic_set(&ioend->io_remaining, 1); | |
281627df | 279 | ioend->io_isdirect = 0; |
7d04a335 | 280 | ioend->io_error = 0; |
f6d6d4fc CH |
281 | ioend->io_list = NULL; |
282 | ioend->io_type = type; | |
b677c210 | 283 | ioend->io_inode = inode; |
c1a073bd | 284 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 285 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
286 | ioend->io_offset = 0; |
287 | ioend->io_size = 0; | |
281627df | 288 | ioend->io_append_trans = NULL; |
0829c360 | 289 | |
5ec4fabb | 290 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
291 | return ioend; |
292 | } | |
293 | ||
1da177e4 LT |
294 | STATIC int |
295 | xfs_map_blocks( | |
296 | struct inode *inode, | |
297 | loff_t offset, | |
207d0416 | 298 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
299 | int type, |
300 | int nonblocking) | |
1da177e4 | 301 | { |
a206c817 CH |
302 | struct xfs_inode *ip = XFS_I(inode); |
303 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 304 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
305 | xfs_fileoff_t offset_fsb, end_fsb; |
306 | int error = 0; | |
a206c817 CH |
307 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
308 | int nimaps = 1; | |
309 | ||
310 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 311 | return -EIO; |
a206c817 | 312 | |
0d882a36 | 313 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 314 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
315 | |
316 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
317 | if (nonblocking) | |
b474c7ae | 318 | return -EAGAIN; |
8ff2957d | 319 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
a206c817 CH |
320 | } |
321 | ||
8ff2957d CH |
322 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
323 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 324 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 325 | |
d2c28191 DC |
326 | if (offset + count > mp->m_super->s_maxbytes) |
327 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
328 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
329 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
330 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
331 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 332 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 333 | |
8ff2957d | 334 | if (error) |
2451337d | 335 | return error; |
a206c817 | 336 | |
0d882a36 | 337 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 338 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 339 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
340 | if (!error) |
341 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 342 | return error; |
a206c817 CH |
343 | } |
344 | ||
8ff2957d | 345 | #ifdef DEBUG |
0d882a36 | 346 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
347 | ASSERT(nimaps); |
348 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
349 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
350 | } | |
351 | #endif | |
352 | if (nimaps) | |
353 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
354 | return 0; | |
1da177e4 LT |
355 | } |
356 | ||
b8f82a4a | 357 | STATIC int |
558e6891 | 358 | xfs_imap_valid( |
8699bb0a | 359 | struct inode *inode, |
207d0416 | 360 | struct xfs_bmbt_irec *imap, |
558e6891 | 361 | xfs_off_t offset) |
1da177e4 | 362 | { |
558e6891 | 363 | offset >>= inode->i_blkbits; |
8699bb0a | 364 | |
558e6891 CH |
365 | return offset >= imap->br_startoff && |
366 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
367 | } |
368 | ||
f6d6d4fc CH |
369 | /* |
370 | * BIO completion handler for buffered IO. | |
371 | */ | |
782e3b3b | 372 | STATIC void |
f6d6d4fc CH |
373 | xfs_end_bio( |
374 | struct bio *bio, | |
f6d6d4fc CH |
375 | int error) |
376 | { | |
377 | xfs_ioend_t *ioend = bio->bi_private; | |
378 | ||
f6d6d4fc | 379 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 380 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
381 | |
382 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
383 | bio->bi_private = NULL; |
384 | bio->bi_end_io = NULL; | |
f6d6d4fc | 385 | bio_put(bio); |
7d04a335 | 386 | |
209fb87a | 387 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
388 | } |
389 | ||
390 | STATIC void | |
391 | xfs_submit_ioend_bio( | |
06342cf8 CH |
392 | struct writeback_control *wbc, |
393 | xfs_ioend_t *ioend, | |
394 | struct bio *bio) | |
f6d6d4fc CH |
395 | { |
396 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
397 | bio->bi_private = ioend; |
398 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 399 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
400 | } |
401 | ||
402 | STATIC struct bio * | |
403 | xfs_alloc_ioend_bio( | |
404 | struct buffer_head *bh) | |
405 | { | |
f6d6d4fc | 406 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 407 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
408 | |
409 | ASSERT(bio->bi_private == NULL); | |
4f024f37 | 410 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
f6d6d4fc | 411 | bio->bi_bdev = bh->b_bdev; |
f6d6d4fc CH |
412 | return bio; |
413 | } | |
414 | ||
415 | STATIC void | |
416 | xfs_start_buffer_writeback( | |
417 | struct buffer_head *bh) | |
418 | { | |
419 | ASSERT(buffer_mapped(bh)); | |
420 | ASSERT(buffer_locked(bh)); | |
421 | ASSERT(!buffer_delay(bh)); | |
422 | ASSERT(!buffer_unwritten(bh)); | |
423 | ||
424 | mark_buffer_async_write(bh); | |
425 | set_buffer_uptodate(bh); | |
426 | clear_buffer_dirty(bh); | |
427 | } | |
428 | ||
429 | STATIC void | |
430 | xfs_start_page_writeback( | |
431 | struct page *page, | |
f6d6d4fc CH |
432 | int clear_dirty, |
433 | int buffers) | |
434 | { | |
435 | ASSERT(PageLocked(page)); | |
436 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
437 | |
438 | /* | |
439 | * if the page was not fully cleaned, we need to ensure that the higher | |
440 | * layers come back to it correctly. That means we need to keep the page | |
441 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
442 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
443 | * write this page in this writeback sweep will be made. | |
444 | */ | |
445 | if (clear_dirty) { | |
92132021 | 446 | clear_page_dirty_for_io(page); |
0d085a52 DC |
447 | set_page_writeback(page); |
448 | } else | |
449 | set_page_writeback_keepwrite(page); | |
450 | ||
f6d6d4fc | 451 | unlock_page(page); |
0d085a52 | 452 | |
1f7decf6 FW |
453 | /* If no buffers on the page are to be written, finish it here */ |
454 | if (!buffers) | |
f6d6d4fc | 455 | end_page_writeback(page); |
f6d6d4fc CH |
456 | } |
457 | ||
c7c1a7d8 | 458 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
459 | { |
460 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
461 | } | |
462 | ||
463 | /* | |
d88992f6 DC |
464 | * Submit all of the bios for all of the ioends we have saved up, covering the |
465 | * initial writepage page and also any probed pages. | |
466 | * | |
467 | * Because we may have multiple ioends spanning a page, we need to start | |
468 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
469 | * buffers as we got, then we can end up with a page that only has buffers | |
470 | * marked async write and I/O complete on can occur before we mark the other | |
471 | * buffers async write. | |
472 | * | |
473 | * The end result of this is that we trip a bug in end_page_writeback() because | |
474 | * we call it twice for the one page as the code in end_buffer_async_write() | |
475 | * assumes that all buffers on the page are started at the same time. | |
476 | * | |
477 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 478 | * buffer_heads, and then submit them for I/O on the second pass. |
7bf7f352 DC |
479 | * |
480 | * If @fail is non-zero, it means that we have a situation where some part of | |
481 | * the submission process has failed after we have marked paged for writeback | |
482 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
483 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
f6d6d4fc CH |
484 | */ |
485 | STATIC void | |
486 | xfs_submit_ioend( | |
06342cf8 | 487 | struct writeback_control *wbc, |
7bf7f352 DC |
488 | xfs_ioend_t *ioend, |
489 | int fail) | |
f6d6d4fc | 490 | { |
d88992f6 | 491 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
492 | xfs_ioend_t *next; |
493 | struct buffer_head *bh; | |
494 | struct bio *bio; | |
495 | sector_t lastblock = 0; | |
496 | ||
d88992f6 DC |
497 | /* Pass 1 - start writeback */ |
498 | do { | |
499 | next = ioend->io_list; | |
221cb251 | 500 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 501 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
502 | } while ((ioend = next) != NULL); |
503 | ||
504 | /* Pass 2 - submit I/O */ | |
505 | ioend = head; | |
f6d6d4fc CH |
506 | do { |
507 | next = ioend->io_list; | |
508 | bio = NULL; | |
509 | ||
7bf7f352 DC |
510 | /* |
511 | * If we are failing the IO now, just mark the ioend with an | |
512 | * error and finish it. This will run IO completion immediately | |
513 | * as there is only one reference to the ioend at this point in | |
514 | * time. | |
515 | */ | |
516 | if (fail) { | |
2451337d | 517 | ioend->io_error = fail; |
7bf7f352 DC |
518 | xfs_finish_ioend(ioend); |
519 | continue; | |
520 | } | |
521 | ||
f6d6d4fc | 522 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
f6d6d4fc CH |
523 | |
524 | if (!bio) { | |
525 | retry: | |
526 | bio = xfs_alloc_ioend_bio(bh); | |
527 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 528 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
529 | goto retry; |
530 | } | |
531 | ||
c7c1a7d8 | 532 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
06342cf8 | 533 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
534 | goto retry; |
535 | } | |
536 | ||
537 | lastblock = bh->b_blocknr; | |
538 | } | |
539 | if (bio) | |
06342cf8 | 540 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 541 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
542 | } while ((ioend = next) != NULL); |
543 | } | |
544 | ||
545 | /* | |
546 | * Cancel submission of all buffer_heads so far in this endio. | |
547 | * Toss the endio too. Only ever called for the initial page | |
548 | * in a writepage request, so only ever one page. | |
549 | */ | |
550 | STATIC void | |
551 | xfs_cancel_ioend( | |
552 | xfs_ioend_t *ioend) | |
553 | { | |
554 | xfs_ioend_t *next; | |
555 | struct buffer_head *bh, *next_bh; | |
556 | ||
557 | do { | |
558 | next = ioend->io_list; | |
559 | bh = ioend->io_buffer_head; | |
560 | do { | |
561 | next_bh = bh->b_private; | |
562 | clear_buffer_async_write(bh); | |
07d08681 BF |
563 | /* |
564 | * The unwritten flag is cleared when added to the | |
565 | * ioend. We're not submitting for I/O so mark the | |
566 | * buffer unwritten again for next time around. | |
567 | */ | |
568 | if (ioend->io_type == XFS_IO_UNWRITTEN) | |
569 | set_buffer_unwritten(bh); | |
f6d6d4fc CH |
570 | unlock_buffer(bh); |
571 | } while ((bh = next_bh) != NULL); | |
572 | ||
f6d6d4fc CH |
573 | mempool_free(ioend, xfs_ioend_pool); |
574 | } while ((ioend = next) != NULL); | |
575 | } | |
576 | ||
577 | /* | |
578 | * Test to see if we've been building up a completion structure for | |
579 | * earlier buffers -- if so, we try to append to this ioend if we | |
580 | * can, otherwise we finish off any current ioend and start another. | |
581 | * Return true if we've finished the given ioend. | |
582 | */ | |
583 | STATIC void | |
584 | xfs_add_to_ioend( | |
585 | struct inode *inode, | |
586 | struct buffer_head *bh, | |
7336cea8 | 587 | xfs_off_t offset, |
f6d6d4fc CH |
588 | unsigned int type, |
589 | xfs_ioend_t **result, | |
590 | int need_ioend) | |
591 | { | |
592 | xfs_ioend_t *ioend = *result; | |
593 | ||
594 | if (!ioend || need_ioend || type != ioend->io_type) { | |
595 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 596 | |
f6d6d4fc CH |
597 | ioend = xfs_alloc_ioend(inode, type); |
598 | ioend->io_offset = offset; | |
599 | ioend->io_buffer_head = bh; | |
600 | ioend->io_buffer_tail = bh; | |
601 | if (previous) | |
602 | previous->io_list = ioend; | |
603 | *result = ioend; | |
604 | } else { | |
605 | ioend->io_buffer_tail->b_private = bh; | |
606 | ioend->io_buffer_tail = bh; | |
607 | } | |
608 | ||
609 | bh->b_private = NULL; | |
610 | ioend->io_size += bh->b_size; | |
611 | } | |
612 | ||
87cbc49c NS |
613 | STATIC void |
614 | xfs_map_buffer( | |
046f1685 | 615 | struct inode *inode, |
87cbc49c | 616 | struct buffer_head *bh, |
207d0416 | 617 | struct xfs_bmbt_irec *imap, |
046f1685 | 618 | xfs_off_t offset) |
87cbc49c NS |
619 | { |
620 | sector_t bn; | |
8699bb0a | 621 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
622 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
623 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 624 | |
207d0416 CH |
625 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
626 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 627 | |
e513182d | 628 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 629 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 630 | |
046f1685 | 631 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
632 | |
633 | bh->b_blocknr = bn; | |
634 | set_buffer_mapped(bh); | |
635 | } | |
636 | ||
1da177e4 LT |
637 | STATIC void |
638 | xfs_map_at_offset( | |
046f1685 | 639 | struct inode *inode, |
1da177e4 | 640 | struct buffer_head *bh, |
207d0416 | 641 | struct xfs_bmbt_irec *imap, |
046f1685 | 642 | xfs_off_t offset) |
1da177e4 | 643 | { |
207d0416 CH |
644 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
645 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 646 | |
207d0416 | 647 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
648 | set_buffer_mapped(bh); |
649 | clear_buffer_delay(bh); | |
f6d6d4fc | 650 | clear_buffer_unwritten(bh); |
1da177e4 LT |
651 | } |
652 | ||
1da177e4 | 653 | /* |
a49935f2 DC |
654 | * Test if a given page contains at least one buffer of a given @type. |
655 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
656 | * try to find one of the type passed in. If it is not set, then the caller only | |
657 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 658 | */ |
a49935f2 | 659 | STATIC bool |
6ffc4db5 | 660 | xfs_check_page_type( |
10ce4444 | 661 | struct page *page, |
a49935f2 DC |
662 | unsigned int type, |
663 | bool check_all_buffers) | |
1da177e4 | 664 | { |
a49935f2 DC |
665 | struct buffer_head *bh; |
666 | struct buffer_head *head; | |
1da177e4 | 667 | |
a49935f2 DC |
668 | if (PageWriteback(page)) |
669 | return false; | |
670 | if (!page->mapping) | |
671 | return false; | |
672 | if (!page_has_buffers(page)) | |
673 | return false; | |
1da177e4 | 674 | |
a49935f2 DC |
675 | bh = head = page_buffers(page); |
676 | do { | |
677 | if (buffer_unwritten(bh)) { | |
678 | if (type == XFS_IO_UNWRITTEN) | |
679 | return true; | |
680 | } else if (buffer_delay(bh)) { | |
805eeb8e | 681 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
682 | return true; |
683 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 684 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
685 | return true; |
686 | } | |
1da177e4 | 687 | |
a49935f2 DC |
688 | /* If we are only checking the first buffer, we are done now. */ |
689 | if (!check_all_buffers) | |
690 | break; | |
691 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 692 | |
a49935f2 | 693 | return false; |
1da177e4 LT |
694 | } |
695 | ||
1da177e4 LT |
696 | /* |
697 | * Allocate & map buffers for page given the extent map. Write it out. | |
698 | * except for the original page of a writepage, this is called on | |
699 | * delalloc/unwritten pages only, for the original page it is possible | |
700 | * that the page has no mapping at all. | |
701 | */ | |
f6d6d4fc | 702 | STATIC int |
1da177e4 LT |
703 | xfs_convert_page( |
704 | struct inode *inode, | |
705 | struct page *page, | |
10ce4444 | 706 | loff_t tindex, |
207d0416 | 707 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 708 | xfs_ioend_t **ioendp, |
2fa24f92 | 709 | struct writeback_control *wbc) |
1da177e4 | 710 | { |
f6d6d4fc | 711 | struct buffer_head *bh, *head; |
9260dc6b CH |
712 | xfs_off_t end_offset; |
713 | unsigned long p_offset; | |
f6d6d4fc | 714 | unsigned int type; |
24e17b5f | 715 | int len, page_dirty; |
f6d6d4fc | 716 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 717 | xfs_off_t offset = page_offset(page); |
1da177e4 | 718 | |
10ce4444 CH |
719 | if (page->index != tindex) |
720 | goto fail; | |
529ae9aa | 721 | if (!trylock_page(page)) |
10ce4444 CH |
722 | goto fail; |
723 | if (PageWriteback(page)) | |
724 | goto fail_unlock_page; | |
725 | if (page->mapping != inode->i_mapping) | |
726 | goto fail_unlock_page; | |
a49935f2 | 727 | if (!xfs_check_page_type(page, (*ioendp)->io_type, false)) |
10ce4444 CH |
728 | goto fail_unlock_page; |
729 | ||
24e17b5f NS |
730 | /* |
731 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 732 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
733 | * |
734 | * Derivation: | |
735 | * | |
736 | * End offset is the highest offset that this page should represent. | |
737 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
738 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
739 | * hence give us the correct page_dirty count. On any other page, | |
740 | * it will be zero and in that case we need page_dirty to be the | |
741 | * count of buffers on the page. | |
24e17b5f | 742 | */ |
9260dc6b CH |
743 | end_offset = min_t(unsigned long long, |
744 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
745 | i_size_read(inode)); | |
746 | ||
480d7467 DC |
747 | /* |
748 | * If the current map does not span the entire page we are about to try | |
749 | * to write, then give up. The only way we can write a page that spans | |
750 | * multiple mappings in a single writeback iteration is via the | |
751 | * xfs_vm_writepage() function. Data integrity writeback requires the | |
752 | * entire page to be written in a single attempt, otherwise the part of | |
753 | * the page we don't write here doesn't get written as part of the data | |
754 | * integrity sync. | |
755 | * | |
756 | * For normal writeback, we also don't attempt to write partial pages | |
757 | * here as it simply means that write_cache_pages() will see it under | |
758 | * writeback and ignore the page until some point in the future, at | |
759 | * which time this will be the only page in the file that needs | |
760 | * writeback. Hence for more optimal IO patterns, we should always | |
761 | * avoid partial page writeback due to multiple mappings on a page here. | |
762 | */ | |
763 | if (!xfs_imap_valid(inode, imap, end_offset)) | |
764 | goto fail_unlock_page; | |
765 | ||
24e17b5f | 766 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
767 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
768 | PAGE_CACHE_SIZE); | |
769 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
770 | page_dirty = p_offset / len; | |
24e17b5f | 771 | |
a49935f2 DC |
772 | /* |
773 | * The moment we find a buffer that doesn't match our current type | |
774 | * specification or can't be written, abort the loop and start | |
775 | * writeback. As per the above xfs_imap_valid() check, only | |
776 | * xfs_vm_writepage() can handle partial page writeback fully - we are | |
777 | * limited here to the buffers that are contiguous with the current | |
778 | * ioend, and hence a buffer we can't write breaks that contiguity and | |
779 | * we have to defer the rest of the IO to xfs_vm_writepage(). | |
780 | */ | |
1da177e4 LT |
781 | bh = head = page_buffers(page); |
782 | do { | |
9260dc6b | 783 | if (offset >= end_offset) |
1da177e4 | 784 | break; |
f6d6d4fc CH |
785 | if (!buffer_uptodate(bh)) |
786 | uptodate = 0; | |
787 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
788 | done = 1; | |
a49935f2 | 789 | break; |
f6d6d4fc CH |
790 | } |
791 | ||
2fa24f92 CH |
792 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
793 | buffer_mapped(bh)) { | |
9260dc6b | 794 | if (buffer_unwritten(bh)) |
0d882a36 | 795 | type = XFS_IO_UNWRITTEN; |
2fa24f92 | 796 | else if (buffer_delay(bh)) |
0d882a36 | 797 | type = XFS_IO_DELALLOC; |
2fa24f92 | 798 | else |
0d882a36 | 799 | type = XFS_IO_OVERWRITE; |
9260dc6b | 800 | |
a49935f2 DC |
801 | /* |
802 | * imap should always be valid because of the above | |
803 | * partial page end_offset check on the imap. | |
804 | */ | |
805 | ASSERT(xfs_imap_valid(inode, imap, offset)); | |
9260dc6b | 806 | |
ecff71e6 | 807 | lock_buffer(bh); |
0d882a36 | 808 | if (type != XFS_IO_OVERWRITE) |
2fa24f92 | 809 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
810 | xfs_add_to_ioend(inode, bh, offset, type, |
811 | ioendp, done); | |
812 | ||
9260dc6b CH |
813 | page_dirty--; |
814 | count++; | |
815 | } else { | |
2fa24f92 | 816 | done = 1; |
a49935f2 | 817 | break; |
1da177e4 | 818 | } |
7336cea8 | 819 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 820 | |
f6d6d4fc CH |
821 | if (uptodate && bh == head) |
822 | SetPageUptodate(page); | |
823 | ||
89f3b363 | 824 | if (count) { |
efceab1d DC |
825 | if (--wbc->nr_to_write <= 0 && |
826 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 827 | done = 1; |
1da177e4 | 828 | } |
89f3b363 | 829 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
830 | |
831 | return done; | |
10ce4444 CH |
832 | fail_unlock_page: |
833 | unlock_page(page); | |
834 | fail: | |
835 | return 1; | |
1da177e4 LT |
836 | } |
837 | ||
838 | /* | |
839 | * Convert & write out a cluster of pages in the same extent as defined | |
840 | * by mp and following the start page. | |
841 | */ | |
842 | STATIC void | |
843 | xfs_cluster_write( | |
844 | struct inode *inode, | |
845 | pgoff_t tindex, | |
207d0416 | 846 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 847 | xfs_ioend_t **ioendp, |
1da177e4 | 848 | struct writeback_control *wbc, |
1da177e4 LT |
849 | pgoff_t tlast) |
850 | { | |
10ce4444 CH |
851 | struct pagevec pvec; |
852 | int done = 0, i; | |
1da177e4 | 853 | |
10ce4444 CH |
854 | pagevec_init(&pvec, 0); |
855 | while (!done && tindex <= tlast) { | |
856 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
857 | ||
858 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 859 | break; |
10ce4444 CH |
860 | |
861 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
862 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 863 | imap, ioendp, wbc); |
10ce4444 CH |
864 | if (done) |
865 | break; | |
866 | } | |
867 | ||
868 | pagevec_release(&pvec); | |
869 | cond_resched(); | |
1da177e4 LT |
870 | } |
871 | } | |
872 | ||
3ed3a434 DC |
873 | STATIC void |
874 | xfs_vm_invalidatepage( | |
875 | struct page *page, | |
d47992f8 LC |
876 | unsigned int offset, |
877 | unsigned int length) | |
3ed3a434 | 878 | { |
34097dfe LC |
879 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
880 | length); | |
881 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
882 | } |
883 | ||
884 | /* | |
885 | * If the page has delalloc buffers on it, we need to punch them out before we | |
886 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
887 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
888 | * is done on that same region - the delalloc extent is returned when none is | |
889 | * supposed to be there. | |
890 | * | |
891 | * We prevent this by truncating away the delalloc regions on the page before | |
892 | * invalidating it. Because they are delalloc, we can do this without needing a | |
893 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
894 | * truncation without a transaction as there is no space left for block | |
895 | * reservation (typically why we see a ENOSPC in writeback). | |
896 | * | |
897 | * This is not a performance critical path, so for now just do the punching a | |
898 | * buffer head at a time. | |
899 | */ | |
900 | STATIC void | |
901 | xfs_aops_discard_page( | |
902 | struct page *page) | |
903 | { | |
904 | struct inode *inode = page->mapping->host; | |
905 | struct xfs_inode *ip = XFS_I(inode); | |
906 | struct buffer_head *bh, *head; | |
907 | loff_t offset = page_offset(page); | |
3ed3a434 | 908 | |
a49935f2 | 909 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
910 | goto out_invalidate; |
911 | ||
e8c3753c DC |
912 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
913 | goto out_invalidate; | |
914 | ||
4f10700a | 915 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
916 | "page discard on page %p, inode 0x%llx, offset %llu.", |
917 | page, ip->i_ino, offset); | |
918 | ||
919 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
920 | bh = head = page_buffers(page); | |
921 | do { | |
3ed3a434 | 922 | int error; |
c726de44 | 923 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
924 | |
925 | if (!buffer_delay(bh)) | |
926 | goto next_buffer; | |
927 | ||
c726de44 DC |
928 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
929 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
930 | if (error) { |
931 | /* something screwed, just bail */ | |
e8c3753c | 932 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 933 | xfs_alert(ip->i_mount, |
3ed3a434 | 934 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 935 | } |
3ed3a434 DC |
936 | break; |
937 | } | |
938 | next_buffer: | |
c726de44 | 939 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
940 | |
941 | } while ((bh = bh->b_this_page) != head); | |
942 | ||
943 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
944 | out_invalidate: | |
d47992f8 | 945 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
3ed3a434 DC |
946 | return; |
947 | } | |
948 | ||
1da177e4 | 949 | /* |
89f3b363 CH |
950 | * Write out a dirty page. |
951 | * | |
952 | * For delalloc space on the page we need to allocate space and flush it. | |
953 | * For unwritten space on the page we need to start the conversion to | |
954 | * regular allocated space. | |
89f3b363 | 955 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 956 | */ |
1da177e4 | 957 | STATIC int |
89f3b363 CH |
958 | xfs_vm_writepage( |
959 | struct page *page, | |
960 | struct writeback_control *wbc) | |
1da177e4 | 961 | { |
89f3b363 | 962 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 963 | struct buffer_head *bh, *head; |
207d0416 | 964 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 965 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 966 | loff_t offset; |
f6d6d4fc | 967 | unsigned int type; |
1da177e4 | 968 | __uint64_t end_offset; |
bd1556a1 | 969 | pgoff_t end_index, last_index; |
ed1e7b7e | 970 | ssize_t len; |
a206c817 | 971 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 972 | int count = 0; |
a206c817 | 973 | int nonblocking = 0; |
89f3b363 | 974 | |
34097dfe | 975 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 976 | |
20cb52eb CH |
977 | ASSERT(page_has_buffers(page)); |
978 | ||
89f3b363 CH |
979 | /* |
980 | * Refuse to write the page out if we are called from reclaim context. | |
981 | * | |
d4f7a5cb CH |
982 | * This avoids stack overflows when called from deeply used stacks in |
983 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
984 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 985 | * |
94054fa3 MG |
986 | * This should never happen except in the case of a VM regression so |
987 | * warn about it. | |
89f3b363 | 988 | */ |
94054fa3 MG |
989 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
990 | PF_MEMALLOC)) | |
b5420f23 | 991 | goto redirty; |
1da177e4 | 992 | |
89f3b363 | 993 | /* |
680a647b CH |
994 | * Given that we do not allow direct reclaim to call us, we should |
995 | * never be called while in a filesystem transaction. | |
89f3b363 | 996 | */ |
448011e2 | 997 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 998 | goto redirty; |
89f3b363 | 999 | |
1da177e4 LT |
1000 | /* Is this page beyond the end of the file? */ |
1001 | offset = i_size_read(inode); | |
1002 | end_index = offset >> PAGE_CACHE_SHIFT; | |
1003 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
8695d27e JL |
1004 | |
1005 | /* | |
1006 | * The page index is less than the end_index, adjust the end_offset | |
1007 | * to the highest offset that this page should represent. | |
1008 | * ----------------------------------------------------- | |
1009 | * | file mapping | <EOF> | | |
1010 | * ----------------------------------------------------- | |
1011 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1012 | * ^--------------------------------^----------|-------- | |
1013 | * | desired writeback range | see else | | |
1014 | * ---------------------------------^------------------| | |
1015 | */ | |
1016 | if (page->index < end_index) | |
1017 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; | |
1018 | else { | |
1019 | /* | |
1020 | * Check whether the page to write out is beyond or straddles | |
1021 | * i_size or not. | |
1022 | * ------------------------------------------------------- | |
1023 | * | file mapping | <EOF> | | |
1024 | * ------------------------------------------------------- | |
1025 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1026 | * ^--------------------------------^-----------|--------- | |
1027 | * | | Straddles | | |
1028 | * ---------------------------------^-----------|--------| | |
1029 | */ | |
6b7a03f0 CH |
1030 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
1031 | ||
1032 | /* | |
ff9a28f6 JK |
1033 | * Skip the page if it is fully outside i_size, e.g. due to a |
1034 | * truncate operation that is in progress. We must redirty the | |
1035 | * page so that reclaim stops reclaiming it. Otherwise | |
1036 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1037 | * |
1038 | * Note that the end_index is unsigned long, it would overflow | |
1039 | * if the given offset is greater than 16TB on 32-bit system | |
1040 | * and if we do check the page is fully outside i_size or not | |
1041 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1042 | * will be evaluated to 0. Hence this page will be redirtied | |
1043 | * and be written out repeatedly which would result in an | |
1044 | * infinite loop, the user program that perform this operation | |
1045 | * will hang. Instead, we can verify this situation by checking | |
1046 | * if the page to write is totally beyond the i_size or if it's | |
1047 | * offset is just equal to the EOF. | |
6b7a03f0 | 1048 | */ |
8695d27e JL |
1049 | if (page->index > end_index || |
1050 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1051 | goto redirty; |
6b7a03f0 CH |
1052 | |
1053 | /* | |
1054 | * The page straddles i_size. It must be zeroed out on each | |
1055 | * and every writepage invocation because it may be mmapped. | |
1056 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1057 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1058 | * memory is zeroed when mapped, and writes to that region are |
1059 | * not written out to the file." | |
1060 | */ | |
1061 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
8695d27e JL |
1062 | |
1063 | /* Adjust the end_offset to the end of file */ | |
1064 | end_offset = offset; | |
1da177e4 LT |
1065 | } |
1066 | ||
24e17b5f | 1067 | len = 1 << inode->i_blkbits; |
24e17b5f | 1068 | |
24e17b5f | 1069 | bh = head = page_buffers(page); |
f6d6d4fc | 1070 | offset = page_offset(page); |
0d882a36 | 1071 | type = XFS_IO_OVERWRITE; |
a206c817 | 1072 | |
dbcdde3e | 1073 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 1074 | nonblocking = 1; |
f6d6d4fc | 1075 | |
1da177e4 | 1076 | do { |
6ac7248e CH |
1077 | int new_ioend = 0; |
1078 | ||
1da177e4 LT |
1079 | if (offset >= end_offset) |
1080 | break; | |
1081 | if (!buffer_uptodate(bh)) | |
1082 | uptodate = 0; | |
1da177e4 | 1083 | |
3d9b02e3 | 1084 | /* |
ece413f5 CH |
1085 | * set_page_dirty dirties all buffers in a page, independent |
1086 | * of their state. The dirty state however is entirely | |
1087 | * meaningless for holes (!mapped && uptodate), so skip | |
1088 | * buffers covering holes here. | |
3d9b02e3 ES |
1089 | */ |
1090 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
1091 | imap_valid = 0; |
1092 | continue; | |
1093 | } | |
1094 | ||
aeea1b1f | 1095 | if (buffer_unwritten(bh)) { |
0d882a36 AR |
1096 | if (type != XFS_IO_UNWRITTEN) { |
1097 | type = XFS_IO_UNWRITTEN; | |
aeea1b1f | 1098 | imap_valid = 0; |
1da177e4 | 1099 | } |
aeea1b1f | 1100 | } else if (buffer_delay(bh)) { |
0d882a36 AR |
1101 | if (type != XFS_IO_DELALLOC) { |
1102 | type = XFS_IO_DELALLOC; | |
aeea1b1f | 1103 | imap_valid = 0; |
1da177e4 | 1104 | } |
89f3b363 | 1105 | } else if (buffer_uptodate(bh)) { |
0d882a36 AR |
1106 | if (type != XFS_IO_OVERWRITE) { |
1107 | type = XFS_IO_OVERWRITE; | |
85da94c6 CH |
1108 | imap_valid = 0; |
1109 | } | |
aeea1b1f | 1110 | } else { |
7d0fa3ec | 1111 | if (PageUptodate(page)) |
aeea1b1f | 1112 | ASSERT(buffer_mapped(bh)); |
7d0fa3ec AR |
1113 | /* |
1114 | * This buffer is not uptodate and will not be | |
1115 | * written to disk. Ensure that we will put any | |
1116 | * subsequent writeable buffers into a new | |
1117 | * ioend. | |
1118 | */ | |
1119 | imap_valid = 0; | |
aeea1b1f CH |
1120 | continue; |
1121 | } | |
d5cb48aa | 1122 | |
aeea1b1f CH |
1123 | if (imap_valid) |
1124 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1125 | if (!imap_valid) { | |
1126 | /* | |
1127 | * If we didn't have a valid mapping then we need to | |
1128 | * put the new mapping into a separate ioend structure. | |
1129 | * This ensures non-contiguous extents always have | |
1130 | * separate ioends, which is particularly important | |
1131 | * for unwritten extent conversion at I/O completion | |
1132 | * time. | |
1133 | */ | |
1134 | new_ioend = 1; | |
1135 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1136 | nonblocking); | |
1137 | if (err) | |
1138 | goto error; | |
1139 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1140 | } | |
1141 | if (imap_valid) { | |
ecff71e6 | 1142 | lock_buffer(bh); |
0d882a36 | 1143 | if (type != XFS_IO_OVERWRITE) |
aeea1b1f CH |
1144 | xfs_map_at_offset(inode, bh, &imap, offset); |
1145 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1146 | new_ioend); | |
1147 | count++; | |
1da177e4 | 1148 | } |
f6d6d4fc CH |
1149 | |
1150 | if (!iohead) | |
1151 | iohead = ioend; | |
1152 | ||
1153 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1154 | |
1155 | if (uptodate && bh == head) | |
1156 | SetPageUptodate(page); | |
1157 | ||
89f3b363 | 1158 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1159 | |
7bf7f352 DC |
1160 | /* if there is no IO to be submitted for this page, we are done */ |
1161 | if (!ioend) | |
1162 | return 0; | |
1163 | ||
1164 | ASSERT(iohead); | |
1165 | ||
1166 | /* | |
1167 | * Any errors from this point onwards need tobe reported through the IO | |
1168 | * completion path as we have marked the initial page as under writeback | |
1169 | * and unlocked it. | |
1170 | */ | |
1171 | if (imap_valid) { | |
bd1556a1 CH |
1172 | xfs_off_t end_index; |
1173 | ||
1174 | end_index = imap.br_startoff + imap.br_blockcount; | |
1175 | ||
1176 | /* to bytes */ | |
1177 | end_index <<= inode->i_blkbits; | |
1178 | ||
1179 | /* to pages */ | |
1180 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1181 | ||
1182 | /* check against file size */ | |
1183 | if (end_index > last_index) | |
1184 | end_index = last_index; | |
8699bb0a | 1185 | |
207d0416 | 1186 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1187 | wbc, end_index); |
1da177e4 LT |
1188 | } |
1189 | ||
281627df | 1190 | |
7bf7f352 DC |
1191 | /* |
1192 | * Reserve log space if we might write beyond the on-disk inode size. | |
1193 | */ | |
1194 | err = 0; | |
1195 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) | |
1196 | err = xfs_setfilesize_trans_alloc(ioend); | |
1197 | ||
1198 | xfs_submit_ioend(wbc, iohead, err); | |
f6d6d4fc | 1199 | |
89f3b363 | 1200 | return 0; |
1da177e4 LT |
1201 | |
1202 | error: | |
f6d6d4fc CH |
1203 | if (iohead) |
1204 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1205 | |
b5420f23 CH |
1206 | if (err == -EAGAIN) |
1207 | goto redirty; | |
1208 | ||
20cb52eb | 1209 | xfs_aops_discard_page(page); |
89f3b363 CH |
1210 | ClearPageUptodate(page); |
1211 | unlock_page(page); | |
1da177e4 | 1212 | return err; |
f51623b2 | 1213 | |
b5420f23 | 1214 | redirty: |
f51623b2 NS |
1215 | redirty_page_for_writepage(wbc, page); |
1216 | unlock_page(page); | |
1217 | return 0; | |
f51623b2 NS |
1218 | } |
1219 | ||
7d4fb40a NS |
1220 | STATIC int |
1221 | xfs_vm_writepages( | |
1222 | struct address_space *mapping, | |
1223 | struct writeback_control *wbc) | |
1224 | { | |
b3aea4ed | 1225 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1226 | return generic_writepages(mapping, wbc); |
1227 | } | |
1228 | ||
f51623b2 NS |
1229 | /* |
1230 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1231 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1232 | * have buffer heads in this call. |
1233 | * | |
89f3b363 | 1234 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1235 | */ |
1236 | STATIC int | |
238f4c54 | 1237 | xfs_vm_releasepage( |
f51623b2 NS |
1238 | struct page *page, |
1239 | gfp_t gfp_mask) | |
1240 | { | |
20cb52eb | 1241 | int delalloc, unwritten; |
f51623b2 | 1242 | |
34097dfe | 1243 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1244 | |
20cb52eb | 1245 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1246 | |
448011e2 | 1247 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1248 | return 0; |
448011e2 | 1249 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1250 | return 0; |
1251 | ||
f51623b2 NS |
1252 | return try_to_free_buffers(page); |
1253 | } | |
1254 | ||
1da177e4 | 1255 | STATIC int |
c2536668 | 1256 | __xfs_get_blocks( |
1da177e4 LT |
1257 | struct inode *inode, |
1258 | sector_t iblock, | |
1da177e4 LT |
1259 | struct buffer_head *bh_result, |
1260 | int create, | |
f2bde9b8 | 1261 | int direct) |
1da177e4 | 1262 | { |
a206c817 CH |
1263 | struct xfs_inode *ip = XFS_I(inode); |
1264 | struct xfs_mount *mp = ip->i_mount; | |
1265 | xfs_fileoff_t offset_fsb, end_fsb; | |
1266 | int error = 0; | |
1267 | int lockmode = 0; | |
207d0416 | 1268 | struct xfs_bmbt_irec imap; |
a206c817 | 1269 | int nimaps = 1; |
fdc7ed75 NS |
1270 | xfs_off_t offset; |
1271 | ssize_t size; | |
207d0416 | 1272 | int new = 0; |
a206c817 CH |
1273 | |
1274 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1275 | return -EIO; |
1da177e4 | 1276 | |
fdc7ed75 | 1277 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1278 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1279 | size = bh_result->b_size; | |
364f358a LM |
1280 | |
1281 | if (!create && direct && offset >= i_size_read(inode)) | |
1282 | return 0; | |
1283 | ||
507630b2 DC |
1284 | /* |
1285 | * Direct I/O is usually done on preallocated files, so try getting | |
1286 | * a block mapping without an exclusive lock first. For buffered | |
1287 | * writes we already have the exclusive iolock anyway, so avoiding | |
1288 | * a lock roundtrip here by taking the ilock exclusive from the | |
1289 | * beginning is a useful micro optimization. | |
1290 | */ | |
1291 | if (create && !direct) { | |
a206c817 CH |
1292 | lockmode = XFS_ILOCK_EXCL; |
1293 | xfs_ilock(ip, lockmode); | |
1294 | } else { | |
309ecac8 | 1295 | lockmode = xfs_ilock_data_map_shared(ip); |
a206c817 | 1296 | } |
f2bde9b8 | 1297 | |
d2c28191 DC |
1298 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1299 | if (offset + size > mp->m_super->s_maxbytes) | |
1300 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1301 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1302 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1303 | ||
5c8ed202 DC |
1304 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1305 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1306 | if (error) |
a206c817 CH |
1307 | goto out_unlock; |
1308 | ||
1309 | if (create && | |
1310 | (!nimaps || | |
1311 | (imap.br_startblock == HOLESTARTBLOCK || | |
1312 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
aff3a9ed | 1313 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 DC |
1314 | /* |
1315 | * Drop the ilock in preparation for starting the block | |
1316 | * allocation transaction. It will be retaken | |
1317 | * exclusively inside xfs_iomap_write_direct for the | |
1318 | * actual allocation. | |
1319 | */ | |
1320 | xfs_iunlock(ip, lockmode); | |
a206c817 CH |
1321 | error = xfs_iomap_write_direct(ip, offset, size, |
1322 | &imap, nimaps); | |
507630b2 | 1323 | if (error) |
2451337d | 1324 | return error; |
d3bc815a | 1325 | new = 1; |
a206c817 | 1326 | } else { |
507630b2 DC |
1327 | /* |
1328 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1329 | * we can go on without dropping the lock here. If we |
1330 | * are allocating a new delalloc block, make sure that | |
1331 | * we set the new flag so that we mark the buffer new so | |
1332 | * that we know that it is newly allocated if the write | |
1333 | * fails. | |
507630b2 | 1334 | */ |
d3bc815a DC |
1335 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1336 | new = 1; | |
a206c817 | 1337 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1338 | if (error) |
1339 | goto out_unlock; | |
1340 | ||
1341 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1342 | } |
a206c817 CH |
1343 | |
1344 | trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap); | |
1345 | } else if (nimaps) { | |
1346 | trace_xfs_get_blocks_found(ip, offset, size, 0, &imap); | |
507630b2 | 1347 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1348 | } else { |
1349 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1350 | goto out_unlock; | |
1351 | } | |
1da177e4 | 1352 | |
207d0416 CH |
1353 | if (imap.br_startblock != HOLESTARTBLOCK && |
1354 | imap.br_startblock != DELAYSTARTBLOCK) { | |
87cbc49c NS |
1355 | /* |
1356 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1357 | * the read case (treat as if we're reading into a hole). |
1358 | */ | |
207d0416 CH |
1359 | if (create || !ISUNWRITTEN(&imap)) |
1360 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1361 | if (create && ISUNWRITTEN(&imap)) { | |
7b7a8665 | 1362 | if (direct) { |
1da177e4 | 1363 | bh_result->b_private = inode; |
7b7a8665 CH |
1364 | set_buffer_defer_completion(bh_result); |
1365 | } | |
1da177e4 | 1366 | set_buffer_unwritten(bh_result); |
1da177e4 LT |
1367 | } |
1368 | } | |
1369 | ||
c2536668 NS |
1370 | /* |
1371 | * If this is a realtime file, data may be on a different device. | |
1372 | * to that pointed to from the buffer_head b_bdev currently. | |
1373 | */ | |
046f1685 | 1374 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1375 | |
c2536668 | 1376 | /* |
549054af DC |
1377 | * If we previously allocated a block out beyond eof and we are now |
1378 | * coming back to use it then we will need to flag it as new even if it | |
1379 | * has a disk address. | |
1380 | * | |
1381 | * With sub-block writes into unwritten extents we also need to mark | |
1382 | * the buffer as new so that the unwritten parts of the buffer gets | |
1383 | * correctly zeroed. | |
1da177e4 LT |
1384 | */ |
1385 | if (create && | |
1386 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1387 | (offset >= i_size_read(inode)) || |
207d0416 | 1388 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1389 | set_buffer_new(bh_result); |
1da177e4 | 1390 | |
207d0416 | 1391 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1392 | BUG_ON(direct); |
1393 | if (create) { | |
1394 | set_buffer_uptodate(bh_result); | |
1395 | set_buffer_mapped(bh_result); | |
1396 | set_buffer_delay(bh_result); | |
1397 | } | |
1398 | } | |
1399 | ||
2b8f12b7 CH |
1400 | /* |
1401 | * If this is O_DIRECT or the mpage code calling tell them how large | |
1402 | * the mapping is, so that we can avoid repeated get_blocks calls. | |
0e1f789d DC |
1403 | * |
1404 | * If the mapping spans EOF, then we have to break the mapping up as the | |
1405 | * mapping for blocks beyond EOF must be marked new so that sub block | |
1406 | * regions can be correctly zeroed. We can't do this for mappings within | |
1407 | * EOF unless the mapping was just allocated or is unwritten, otherwise | |
1408 | * the callers would overwrite existing data with zeros. Hence we have | |
1409 | * to split the mapping into a range up to and including EOF, and a | |
1410 | * second mapping for beyond EOF. | |
2b8f12b7 | 1411 | */ |
c2536668 | 1412 | if (direct || size > (1 << inode->i_blkbits)) { |
2b8f12b7 CH |
1413 | xfs_off_t mapping_size; |
1414 | ||
1415 | mapping_size = imap.br_startoff + imap.br_blockcount - iblock; | |
1416 | mapping_size <<= inode->i_blkbits; | |
1417 | ||
1418 | ASSERT(mapping_size > 0); | |
1419 | if (mapping_size > size) | |
1420 | mapping_size = size; | |
0e1f789d DC |
1421 | if (offset < i_size_read(inode) && |
1422 | offset + mapping_size >= i_size_read(inode)) { | |
1423 | /* limit mapping to block that spans EOF */ | |
1424 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1425 | 1 << inode->i_blkbits); | |
1426 | } | |
2b8f12b7 CH |
1427 | if (mapping_size > LONG_MAX) |
1428 | mapping_size = LONG_MAX; | |
1429 | ||
1430 | bh_result->b_size = mapping_size; | |
1da177e4 LT |
1431 | } |
1432 | ||
1433 | return 0; | |
a206c817 CH |
1434 | |
1435 | out_unlock: | |
1436 | xfs_iunlock(ip, lockmode); | |
2451337d | 1437 | return error; |
1da177e4 LT |
1438 | } |
1439 | ||
1440 | int | |
c2536668 | 1441 | xfs_get_blocks( |
1da177e4 LT |
1442 | struct inode *inode, |
1443 | sector_t iblock, | |
1444 | struct buffer_head *bh_result, | |
1445 | int create) | |
1446 | { | |
f2bde9b8 | 1447 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1448 | } |
1449 | ||
1450 | STATIC int | |
e4c573bb | 1451 | xfs_get_blocks_direct( |
1da177e4 LT |
1452 | struct inode *inode, |
1453 | sector_t iblock, | |
1da177e4 LT |
1454 | struct buffer_head *bh_result, |
1455 | int create) | |
1456 | { | |
f2bde9b8 | 1457 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1458 | } |
1459 | ||
209fb87a CH |
1460 | /* |
1461 | * Complete a direct I/O write request. | |
1462 | * | |
1463 | * If the private argument is non-NULL __xfs_get_blocks signals us that we | |
1464 | * need to issue a transaction to convert the range from unwritten to written | |
1465 | * extents. In case this is regular synchronous I/O we just call xfs_end_io | |
25985edc | 1466 | * to do this and we are done. But in case this was a successful AIO |
209fb87a CH |
1467 | * request this handler is called from interrupt context, from which we |
1468 | * can't start transactions. In that case offload the I/O completion to | |
1469 | * the workqueues we also use for buffered I/O completion. | |
1470 | */ | |
f0973863 | 1471 | STATIC void |
209fb87a CH |
1472 | xfs_end_io_direct_write( |
1473 | struct kiocb *iocb, | |
1474 | loff_t offset, | |
1475 | ssize_t size, | |
7b7a8665 | 1476 | void *private) |
f0973863 | 1477 | { |
209fb87a | 1478 | struct xfs_ioend *ioend = iocb->private; |
f0973863 | 1479 | |
2813d682 CH |
1480 | /* |
1481 | * While the generic direct I/O code updates the inode size, it does | |
1482 | * so only after the end_io handler is called, which means our | |
1483 | * end_io handler thinks the on-disk size is outside the in-core | |
1484 | * size. To prevent this just update it a little bit earlier here. | |
1485 | */ | |
1486 | if (offset + size > i_size_read(ioend->io_inode)) | |
1487 | i_size_write(ioend->io_inode, offset + size); | |
1488 | ||
f0973863 | 1489 | /* |
209fb87a CH |
1490 | * blockdev_direct_IO can return an error even after the I/O |
1491 | * completion handler was called. Thus we need to protect | |
1492 | * against double-freeing. | |
f0973863 | 1493 | */ |
209fb87a CH |
1494 | iocb->private = NULL; |
1495 | ||
ba87ea69 LM |
1496 | ioend->io_offset = offset; |
1497 | ioend->io_size = size; | |
209fb87a | 1498 | if (private && size > 0) |
0d882a36 | 1499 | ioend->io_type = XFS_IO_UNWRITTEN; |
209fb87a | 1500 | |
7b7a8665 | 1501 | xfs_finish_ioend_sync(ioend); |
f0973863 CH |
1502 | } |
1503 | ||
1da177e4 | 1504 | STATIC ssize_t |
e4c573bb | 1505 | xfs_vm_direct_IO( |
1da177e4 LT |
1506 | int rw, |
1507 | struct kiocb *iocb, | |
d8d3d94b AV |
1508 | struct iov_iter *iter, |
1509 | loff_t offset) | |
1da177e4 | 1510 | { |
209fb87a CH |
1511 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1512 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
281627df | 1513 | struct xfs_ioend *ioend = NULL; |
209fb87a CH |
1514 | ssize_t ret; |
1515 | ||
1516 | if (rw & WRITE) { | |
a6cbcd4a | 1517 | size_t size = iov_iter_count(iter); |
281627df CH |
1518 | |
1519 | /* | |
437a255a DC |
1520 | * We cannot preallocate a size update transaction here as we |
1521 | * don't know whether allocation is necessary or not. Hence we | |
1522 | * can only tell IO completion that one is necessary if we are | |
1523 | * not doing unwritten extent conversion. | |
281627df | 1524 | */ |
0d882a36 | 1525 | iocb->private = ioend = xfs_alloc_ioend(inode, XFS_IO_DIRECT); |
437a255a | 1526 | if (offset + size > XFS_I(inode)->i_d.di_size) |
281627df | 1527 | ioend->io_isdirect = 1; |
209fb87a | 1528 | |
31b14039 AV |
1529 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter, |
1530 | offset, xfs_get_blocks_direct, | |
9862f62f CH |
1531 | xfs_end_io_direct_write, NULL, |
1532 | DIO_ASYNC_EXTEND); | |
209fb87a | 1533 | if (ret != -EIOCBQUEUED && iocb->private) |
437a255a | 1534 | goto out_destroy_ioend; |
209fb87a | 1535 | } else { |
31b14039 AV |
1536 | ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iter, |
1537 | offset, xfs_get_blocks_direct, | |
eafdc7d1 | 1538 | NULL, NULL, 0); |
209fb87a | 1539 | } |
f0973863 | 1540 | |
f0973863 | 1541 | return ret; |
281627df | 1542 | |
281627df CH |
1543 | out_destroy_ioend: |
1544 | xfs_destroy_ioend(ioend); | |
1545 | return ret; | |
1da177e4 LT |
1546 | } |
1547 | ||
d3bc815a DC |
1548 | /* |
1549 | * Punch out the delalloc blocks we have already allocated. | |
1550 | * | |
1551 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1552 | * as the page is still locked at this point. | |
1553 | */ | |
1554 | STATIC void | |
1555 | xfs_vm_kill_delalloc_range( | |
1556 | struct inode *inode, | |
1557 | loff_t start, | |
1558 | loff_t end) | |
1559 | { | |
1560 | struct xfs_inode *ip = XFS_I(inode); | |
1561 | xfs_fileoff_t start_fsb; | |
1562 | xfs_fileoff_t end_fsb; | |
1563 | int error; | |
1564 | ||
1565 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1566 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1567 | if (end_fsb <= start_fsb) | |
1568 | return; | |
1569 | ||
1570 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1571 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1572 | end_fsb - start_fsb); | |
1573 | if (error) { | |
1574 | /* something screwed, just bail */ | |
1575 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1576 | xfs_alert(ip->i_mount, | |
1577 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1578 | ip->i_ino); | |
1579 | } | |
1580 | } | |
1581 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1582 | } | |
1583 | ||
fa9b227e CH |
1584 | STATIC void |
1585 | xfs_vm_write_failed( | |
d3bc815a DC |
1586 | struct inode *inode, |
1587 | struct page *page, | |
1588 | loff_t pos, | |
1589 | unsigned len) | |
fa9b227e | 1590 | { |
58e59854 | 1591 | loff_t block_offset; |
d3bc815a DC |
1592 | loff_t block_start; |
1593 | loff_t block_end; | |
1594 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1595 | loff_t to = from + len; | |
1596 | struct buffer_head *bh, *head; | |
fa9b227e | 1597 | |
58e59854 JL |
1598 | /* |
1599 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1600 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1601 | * platform, the high 32-bit will be masked off if we evaluate the | |
1602 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1603 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1604 | * which could cause the following ASSERT failed in most cases. | |
1605 | * In order to avoid this, we can evaluate the block_offset of the | |
1606 | * start of the page by using shifts rather than masks the mismatch | |
1607 | * problem. | |
1608 | */ | |
1609 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1610 | ||
d3bc815a | 1611 | ASSERT(block_offset + from == pos); |
c726de44 | 1612 | |
d3bc815a DC |
1613 | head = page_buffers(page); |
1614 | block_start = 0; | |
1615 | for (bh = head; bh != head || !block_start; | |
1616 | bh = bh->b_this_page, block_start = block_end, | |
1617 | block_offset += bh->b_size) { | |
1618 | block_end = block_start + bh->b_size; | |
c726de44 | 1619 | |
d3bc815a DC |
1620 | /* skip buffers before the write */ |
1621 | if (block_end <= from) | |
1622 | continue; | |
1623 | ||
1624 | /* if the buffer is after the write, we're done */ | |
1625 | if (block_start >= to) | |
1626 | break; | |
1627 | ||
1628 | if (!buffer_delay(bh)) | |
1629 | continue; | |
1630 | ||
1631 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1632 | continue; | |
1633 | ||
1634 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1635 | block_offset + bh->b_size); | |
4ab9ed57 DC |
1636 | |
1637 | /* | |
1638 | * This buffer does not contain data anymore. make sure anyone | |
1639 | * who finds it knows that for certain. | |
1640 | */ | |
1641 | clear_buffer_delay(bh); | |
1642 | clear_buffer_uptodate(bh); | |
1643 | clear_buffer_mapped(bh); | |
1644 | clear_buffer_new(bh); | |
1645 | clear_buffer_dirty(bh); | |
fa9b227e | 1646 | } |
d3bc815a | 1647 | |
fa9b227e CH |
1648 | } |
1649 | ||
d3bc815a DC |
1650 | /* |
1651 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1652 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1653 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1654 | * the appropriate point. | |
1655 | */ | |
f51623b2 | 1656 | STATIC int |
d79689c7 | 1657 | xfs_vm_write_begin( |
f51623b2 | 1658 | struct file *file, |
d79689c7 NP |
1659 | struct address_space *mapping, |
1660 | loff_t pos, | |
1661 | unsigned len, | |
1662 | unsigned flags, | |
1663 | struct page **pagep, | |
1664 | void **fsdata) | |
f51623b2 | 1665 | { |
d3bc815a DC |
1666 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1667 | struct page *page; | |
1668 | int status; | |
155130a4 | 1669 | |
d3bc815a DC |
1670 | ASSERT(len <= PAGE_CACHE_SIZE); |
1671 | ||
ad22c7a0 | 1672 | page = grab_cache_page_write_begin(mapping, index, flags); |
d3bc815a DC |
1673 | if (!page) |
1674 | return -ENOMEM; | |
1675 | ||
1676 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1677 | if (unlikely(status)) { | |
1678 | struct inode *inode = mapping->host; | |
72ab70a1 | 1679 | size_t isize = i_size_read(inode); |
d3bc815a DC |
1680 | |
1681 | xfs_vm_write_failed(inode, page, pos, len); | |
1682 | unlock_page(page); | |
1683 | ||
72ab70a1 DC |
1684 | /* |
1685 | * If the write is beyond EOF, we only want to kill blocks | |
1686 | * allocated in this write, not blocks that were previously | |
1687 | * written successfully. | |
1688 | */ | |
1689 | if (pos + len > isize) { | |
1690 | ssize_t start = max_t(ssize_t, pos, isize); | |
1691 | ||
1692 | truncate_pagecache_range(inode, start, pos + len); | |
1693 | } | |
d3bc815a DC |
1694 | |
1695 | page_cache_release(page); | |
1696 | page = NULL; | |
1697 | } | |
1698 | ||
1699 | *pagep = page; | |
1700 | return status; | |
fa9b227e CH |
1701 | } |
1702 | ||
d3bc815a | 1703 | /* |
aad3f375 DC |
1704 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
1705 | * this specific write because they will never be written. Previous writes | |
1706 | * beyond EOF where block allocation succeeded do not need to be trashed, so | |
1707 | * only new blocks from this write should be trashed. For blocks within | |
1708 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be | |
1709 | * written with all the other valid data. | |
d3bc815a | 1710 | */ |
fa9b227e CH |
1711 | STATIC int |
1712 | xfs_vm_write_end( | |
1713 | struct file *file, | |
1714 | struct address_space *mapping, | |
1715 | loff_t pos, | |
1716 | unsigned len, | |
1717 | unsigned copied, | |
1718 | struct page *page, | |
1719 | void *fsdata) | |
1720 | { | |
1721 | int ret; | |
155130a4 | 1722 | |
d3bc815a DC |
1723 | ASSERT(len <= PAGE_CACHE_SIZE); |
1724 | ||
fa9b227e | 1725 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1726 | if (unlikely(ret < len)) { |
1727 | struct inode *inode = mapping->host; | |
1728 | size_t isize = i_size_read(inode); | |
1729 | loff_t to = pos + len; | |
1730 | ||
1731 | if (to > isize) { | |
aad3f375 DC |
1732 | /* only kill blocks in this write beyond EOF */ |
1733 | if (pos > isize) | |
1734 | isize = pos; | |
d3bc815a | 1735 | xfs_vm_kill_delalloc_range(inode, isize, to); |
aad3f375 | 1736 | truncate_pagecache_range(inode, isize, to); |
d3bc815a DC |
1737 | } |
1738 | } | |
155130a4 | 1739 | return ret; |
f51623b2 | 1740 | } |
1da177e4 LT |
1741 | |
1742 | STATIC sector_t | |
e4c573bb | 1743 | xfs_vm_bmap( |
1da177e4 LT |
1744 | struct address_space *mapping, |
1745 | sector_t block) | |
1746 | { | |
1747 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1748 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1749 | |
cca28fb8 | 1750 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1751 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1752 | filemap_write_and_wait(mapping); |
126468b1 | 1753 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1754 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1755 | } |
1756 | ||
1757 | STATIC int | |
e4c573bb | 1758 | xfs_vm_readpage( |
1da177e4 LT |
1759 | struct file *unused, |
1760 | struct page *page) | |
1761 | { | |
c2536668 | 1762 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1763 | } |
1764 | ||
1765 | STATIC int | |
e4c573bb | 1766 | xfs_vm_readpages( |
1da177e4 LT |
1767 | struct file *unused, |
1768 | struct address_space *mapping, | |
1769 | struct list_head *pages, | |
1770 | unsigned nr_pages) | |
1771 | { | |
c2536668 | 1772 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1773 | } |
1774 | ||
22e757a4 DC |
1775 | /* |
1776 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1777 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1778 | * dirty, we'll never be able to clean them because we don't write buffers | |
1779 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1780 | * that have been marked dirty. Further, the dirty state can leak into | |
1781 | * the file interior if the file is extended, resulting in all sorts of | |
1782 | * bad things happening as the state does not match the underlying data. | |
1783 | * | |
1784 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1785 | * this only exist because of bufferheads and how the generic code manages them. | |
1786 | */ | |
1787 | STATIC int | |
1788 | xfs_vm_set_page_dirty( | |
1789 | struct page *page) | |
1790 | { | |
1791 | struct address_space *mapping = page->mapping; | |
1792 | struct inode *inode = mapping->host; | |
1793 | loff_t end_offset; | |
1794 | loff_t offset; | |
1795 | int newly_dirty; | |
1796 | ||
1797 | if (unlikely(!mapping)) | |
1798 | return !TestSetPageDirty(page); | |
1799 | ||
1800 | end_offset = i_size_read(inode); | |
1801 | offset = page_offset(page); | |
1802 | ||
1803 | spin_lock(&mapping->private_lock); | |
1804 | if (page_has_buffers(page)) { | |
1805 | struct buffer_head *head = page_buffers(page); | |
1806 | struct buffer_head *bh = head; | |
1807 | ||
1808 | do { | |
1809 | if (offset < end_offset) | |
1810 | set_buffer_dirty(bh); | |
1811 | bh = bh->b_this_page; | |
1812 | offset += 1 << inode->i_blkbits; | |
1813 | } while (bh != head); | |
1814 | } | |
1815 | newly_dirty = !TestSetPageDirty(page); | |
1816 | spin_unlock(&mapping->private_lock); | |
1817 | ||
1818 | if (newly_dirty) { | |
1819 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1820 | unsigned long flags; | |
1821 | ||
1822 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1823 | if (page->mapping) { /* Race with truncate? */ | |
1824 | WARN_ON_ONCE(!PageUptodate(page)); | |
1825 | account_page_dirtied(page, mapping); | |
1826 | radix_tree_tag_set(&mapping->page_tree, | |
1827 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1828 | } | |
1829 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
1830 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1831 | } | |
1832 | return newly_dirty; | |
1833 | } | |
1834 | ||
f5e54d6e | 1835 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1836 | .readpage = xfs_vm_readpage, |
1837 | .readpages = xfs_vm_readpages, | |
1838 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1839 | .writepages = xfs_vm_writepages, |
22e757a4 | 1840 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1841 | .releasepage = xfs_vm_releasepage, |
1842 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1843 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1844 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1845 | .bmap = xfs_vm_bmap, |
1846 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1847 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1848 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1849 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1850 | }; |