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