]>
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" |
5a0e3ad6 | 34 | #include <linux/gfp.h> |
1da177e4 | 35 | #include <linux/mpage.h> |
10ce4444 | 36 | #include <linux/pagevec.h> |
1da177e4 LT |
37 | #include <linux/writeback.h> |
38 | ||
fbcc0256 DC |
39 | /* |
40 | * structure owned by writepages passed to individual writepage calls | |
41 | */ | |
42 | struct xfs_writepage_ctx { | |
43 | struct xfs_bmbt_irec imap; | |
44 | bool imap_valid; | |
45 | unsigned int io_type; | |
46 | struct xfs_ioend *iohead; | |
47 | struct xfs_ioend *ioend; | |
48 | sector_t last_block; | |
49 | }; | |
50 | ||
0b1b213f | 51 | void |
f51623b2 NS |
52 | xfs_count_page_state( |
53 | struct page *page, | |
54 | int *delalloc, | |
f51623b2 NS |
55 | int *unwritten) |
56 | { | |
57 | struct buffer_head *bh, *head; | |
58 | ||
20cb52eb | 59 | *delalloc = *unwritten = 0; |
f51623b2 NS |
60 | |
61 | bh = head = page_buffers(page); | |
62 | do { | |
20cb52eb | 63 | if (buffer_unwritten(bh)) |
f51623b2 NS |
64 | (*unwritten) = 1; |
65 | else if (buffer_delay(bh)) | |
66 | (*delalloc) = 1; | |
67 | } while ((bh = bh->b_this_page) != head); | |
68 | } | |
69 | ||
6214ed44 CH |
70 | STATIC struct block_device * |
71 | xfs_find_bdev_for_inode( | |
046f1685 | 72 | struct inode *inode) |
6214ed44 | 73 | { |
046f1685 | 74 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
75 | struct xfs_mount *mp = ip->i_mount; |
76 | ||
71ddabb9 | 77 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
78 | return mp->m_rtdev_targp->bt_bdev; |
79 | else | |
80 | return mp->m_ddev_targp->bt_bdev; | |
81 | } | |
82 | ||
f6d6d4fc CH |
83 | /* |
84 | * We're now finished for good with this ioend structure. | |
85 | * Update the page state via the associated buffer_heads, | |
86 | * release holds on the inode and bio, and finally free | |
87 | * up memory. Do not use the ioend after this. | |
88 | */ | |
0829c360 CH |
89 | STATIC void |
90 | xfs_destroy_ioend( | |
91 | xfs_ioend_t *ioend) | |
92 | { | |
f6d6d4fc CH |
93 | struct buffer_head *bh, *next; |
94 | ||
95 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
96 | next = bh->b_private; | |
7d04a335 | 97 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 98 | } |
583fa586 | 99 | |
0829c360 CH |
100 | mempool_free(ioend, xfs_ioend_pool); |
101 | } | |
102 | ||
fc0063c4 CH |
103 | /* |
104 | * Fast and loose check if this write could update the on-disk inode size. | |
105 | */ | |
106 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
107 | { | |
108 | return ioend->io_offset + ioend->io_size > | |
109 | XFS_I(ioend->io_inode)->i_d.di_size; | |
110 | } | |
111 | ||
281627df CH |
112 | STATIC int |
113 | xfs_setfilesize_trans_alloc( | |
114 | struct xfs_ioend *ioend) | |
115 | { | |
116 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
117 | struct xfs_trans *tp; | |
118 | int error; | |
119 | ||
120 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
121 | ||
3d3c8b52 | 122 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
281627df | 123 | if (error) { |
4906e215 | 124 | xfs_trans_cancel(tp); |
281627df CH |
125 | return error; |
126 | } | |
127 | ||
128 | ioend->io_append_trans = tp; | |
129 | ||
d9457dc0 | 130 | /* |
437a255a | 131 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
132 | * we released it. |
133 | */ | |
bee9182d | 134 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
135 | /* |
136 | * We hand off the transaction to the completion thread now, so | |
137 | * clear the flag here. | |
138 | */ | |
139 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
140 | return 0; | |
141 | } | |
142 | ||
ba87ea69 | 143 | /* |
2813d682 | 144 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 145 | */ |
281627df | 146 | STATIC int |
ba87ea69 | 147 | xfs_setfilesize( |
2ba66237 CH |
148 | struct xfs_inode *ip, |
149 | struct xfs_trans *tp, | |
150 | xfs_off_t offset, | |
151 | size_t size) | |
ba87ea69 | 152 | { |
ba87ea69 | 153 | xfs_fsize_t isize; |
ba87ea69 | 154 | |
aa6bf01d | 155 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 156 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
157 | if (!isize) { |
158 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 159 | xfs_trans_cancel(tp); |
281627df | 160 | return 0; |
ba87ea69 LM |
161 | } |
162 | ||
2ba66237 | 163 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
164 | |
165 | ip->i_d.di_size = isize; | |
166 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
167 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
168 | ||
70393313 | 169 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
170 | } |
171 | ||
2ba66237 CH |
172 | STATIC int |
173 | xfs_setfilesize_ioend( | |
174 | struct xfs_ioend *ioend) | |
175 | { | |
176 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
177 | struct xfs_trans *tp = ioend->io_append_trans; | |
178 | ||
179 | /* | |
180 | * The transaction may have been allocated in the I/O submission thread, | |
181 | * thus we need to mark ourselves as being in a transaction manually. | |
182 | * Similarly for freeze protection. | |
183 | */ | |
184 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 185 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 186 | |
5cb13dcd Z |
187 | /* we abort the update if there was an IO error */ |
188 | if (ioend->io_error) { | |
189 | xfs_trans_cancel(tp); | |
190 | return ioend->io_error; | |
191 | } | |
192 | ||
2ba66237 CH |
193 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
194 | } | |
195 | ||
77d7a0c2 | 196 | /* |
209fb87a | 197 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
198 | * |
199 | * If there is no work to do we might as well call it a day and free the | |
200 | * ioend right now. | |
77d7a0c2 DC |
201 | */ |
202 | STATIC void | |
203 | xfs_finish_ioend( | |
209fb87a | 204 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
205 | { |
206 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
207 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
208 | ||
0d882a36 | 209 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
aa6bf01d | 210 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
2ba66237 | 211 | else if (ioend->io_append_trans) |
aa6bf01d | 212 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
213 | else |
214 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 215 | } |
ba87ea69 LM |
216 | } |
217 | ||
0829c360 | 218 | /* |
5ec4fabb | 219 | * IO write completion. |
f6d6d4fc CH |
220 | */ |
221 | STATIC void | |
5ec4fabb | 222 | xfs_end_io( |
77d7a0c2 | 223 | struct work_struct *work) |
0829c360 | 224 | { |
77d7a0c2 DC |
225 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
226 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 227 | int error = 0; |
ba87ea69 | 228 | |
04f658ee | 229 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 230 | ioend->io_error = -EIO; |
04f658ee CH |
231 | goto done; |
232 | } | |
04f658ee | 233 | |
5ec4fabb CH |
234 | /* |
235 | * For unwritten extents we need to issue transactions to convert a | |
236 | * range to normal written extens after the data I/O has finished. | |
5cb13dcd Z |
237 | * Detecting and handling completion IO errors is done individually |
238 | * for each case as different cleanup operations need to be performed | |
239 | * on error. | |
5ec4fabb | 240 | */ |
0d882a36 | 241 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
5cb13dcd Z |
242 | if (ioend->io_error) |
243 | goto done; | |
437a255a DC |
244 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
245 | ioend->io_size); | |
281627df | 246 | } else if (ioend->io_append_trans) { |
2ba66237 | 247 | error = xfs_setfilesize_ioend(ioend); |
84803fb7 | 248 | } else { |
281627df | 249 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 250 | } |
ba87ea69 | 251 | |
04f658ee | 252 | done: |
437a255a | 253 | if (error) |
2451337d | 254 | ioend->io_error = error; |
aa6bf01d | 255 | xfs_destroy_ioend(ioend); |
c626d174 DC |
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); | |
7d04a335 | 279 | ioend->io_error = 0; |
f6d6d4fc CH |
280 | ioend->io_list = NULL; |
281 | ioend->io_type = type; | |
b677c210 | 282 | ioend->io_inode = inode; |
c1a073bd | 283 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 284 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
285 | ioend->io_offset = 0; |
286 | ioend->io_size = 0; | |
281627df | 287 | ioend->io_append_trans = NULL; |
0829c360 | 288 | |
5ec4fabb | 289 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
290 | return ioend; |
291 | } | |
292 | ||
1da177e4 LT |
293 | STATIC int |
294 | xfs_map_blocks( | |
295 | struct inode *inode, | |
296 | loff_t offset, | |
207d0416 | 297 | struct xfs_bmbt_irec *imap, |
988ef927 | 298 | int type) |
1da177e4 | 299 | { |
a206c817 CH |
300 | struct xfs_inode *ip = XFS_I(inode); |
301 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 302 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
303 | xfs_fileoff_t offset_fsb, end_fsb; |
304 | int error = 0; | |
a206c817 CH |
305 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
306 | int nimaps = 1; | |
307 | ||
308 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 309 | return -EIO; |
a206c817 | 310 | |
0d882a36 | 311 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 312 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 313 | |
988ef927 | 314 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
315 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
316 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 317 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 318 | |
d2c28191 DC |
319 | if (offset + count > mp->m_super->s_maxbytes) |
320 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
321 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
322 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
323 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
324 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 325 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 326 | |
8ff2957d | 327 | if (error) |
2451337d | 328 | return error; |
a206c817 | 329 | |
0d882a36 | 330 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 331 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 332 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
333 | if (!error) |
334 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 335 | return error; |
a206c817 CH |
336 | } |
337 | ||
8ff2957d | 338 | #ifdef DEBUG |
0d882a36 | 339 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
340 | ASSERT(nimaps); |
341 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
342 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
343 | } | |
344 | #endif | |
345 | if (nimaps) | |
346 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
347 | return 0; | |
1da177e4 LT |
348 | } |
349 | ||
fbcc0256 | 350 | STATIC bool |
558e6891 | 351 | xfs_imap_valid( |
8699bb0a | 352 | struct inode *inode, |
207d0416 | 353 | struct xfs_bmbt_irec *imap, |
558e6891 | 354 | xfs_off_t offset) |
1da177e4 | 355 | { |
558e6891 | 356 | offset >>= inode->i_blkbits; |
8699bb0a | 357 | |
558e6891 CH |
358 | return offset >= imap->br_startoff && |
359 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
360 | } |
361 | ||
f6d6d4fc CH |
362 | /* |
363 | * BIO completion handler for buffered IO. | |
364 | */ | |
782e3b3b | 365 | STATIC void |
f6d6d4fc | 366 | xfs_end_bio( |
4246a0b6 | 367 | struct bio *bio) |
f6d6d4fc CH |
368 | { |
369 | xfs_ioend_t *ioend = bio->bi_private; | |
370 | ||
77a78806 LT |
371 | if (!ioend->io_error) |
372 | ioend->io_error = bio->bi_error; | |
f6d6d4fc CH |
373 | |
374 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
375 | bio->bi_private = NULL; |
376 | bio->bi_end_io = NULL; | |
f6d6d4fc | 377 | bio_put(bio); |
7d04a335 | 378 | |
209fb87a | 379 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
380 | } |
381 | ||
382 | STATIC void | |
383 | xfs_submit_ioend_bio( | |
06342cf8 CH |
384 | struct writeback_control *wbc, |
385 | xfs_ioend_t *ioend, | |
386 | struct bio *bio) | |
f6d6d4fc CH |
387 | { |
388 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
389 | bio->bi_private = ioend; |
390 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 391 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
392 | } |
393 | ||
394 | STATIC struct bio * | |
395 | xfs_alloc_ioend_bio( | |
396 | struct buffer_head *bh) | |
397 | { | |
b54ffb73 | 398 | struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); |
f6d6d4fc CH |
399 | |
400 | ASSERT(bio->bi_private == NULL); | |
4f024f37 | 401 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
f6d6d4fc | 402 | bio->bi_bdev = bh->b_bdev; |
f6d6d4fc CH |
403 | return bio; |
404 | } | |
405 | ||
406 | STATIC void | |
407 | xfs_start_buffer_writeback( | |
408 | struct buffer_head *bh) | |
409 | { | |
410 | ASSERT(buffer_mapped(bh)); | |
411 | ASSERT(buffer_locked(bh)); | |
412 | ASSERT(!buffer_delay(bh)); | |
413 | ASSERT(!buffer_unwritten(bh)); | |
414 | ||
415 | mark_buffer_async_write(bh); | |
416 | set_buffer_uptodate(bh); | |
417 | clear_buffer_dirty(bh); | |
418 | } | |
419 | ||
420 | STATIC void | |
421 | xfs_start_page_writeback( | |
422 | struct page *page, | |
f6d6d4fc CH |
423 | int clear_dirty, |
424 | int buffers) | |
425 | { | |
426 | ASSERT(PageLocked(page)); | |
427 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
428 | |
429 | /* | |
430 | * if the page was not fully cleaned, we need to ensure that the higher | |
431 | * layers come back to it correctly. That means we need to keep the page | |
432 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
433 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
434 | * write this page in this writeback sweep will be made. | |
435 | */ | |
436 | if (clear_dirty) { | |
92132021 | 437 | clear_page_dirty_for_io(page); |
0d085a52 DC |
438 | set_page_writeback(page); |
439 | } else | |
440 | set_page_writeback_keepwrite(page); | |
441 | ||
f6d6d4fc | 442 | unlock_page(page); |
0d085a52 | 443 | |
1f7decf6 FW |
444 | /* If no buffers on the page are to be written, finish it here */ |
445 | if (!buffers) | |
f6d6d4fc | 446 | end_page_writeback(page); |
f6d6d4fc CH |
447 | } |
448 | ||
c7c1a7d8 | 449 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
450 | { |
451 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
452 | } | |
453 | ||
454 | /* | |
d88992f6 DC |
455 | * Submit all of the bios for all of the ioends we have saved up, covering the |
456 | * initial writepage page and also any probed pages. | |
457 | * | |
458 | * Because we may have multiple ioends spanning a page, we need to start | |
459 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
460 | * buffers as we got, then we can end up with a page that only has buffers | |
461 | * marked async write and I/O complete on can occur before we mark the other | |
462 | * buffers async write. | |
463 | * | |
464 | * The end result of this is that we trip a bug in end_page_writeback() because | |
465 | * we call it twice for the one page as the code in end_buffer_async_write() | |
466 | * assumes that all buffers on the page are started at the same time. | |
467 | * | |
468 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 469 | * buffer_heads, and then submit them for I/O on the second pass. |
7bf7f352 DC |
470 | * |
471 | * If @fail is non-zero, it means that we have a situation where some part of | |
472 | * the submission process has failed after we have marked paged for writeback | |
473 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
474 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
f6d6d4fc CH |
475 | */ |
476 | STATIC void | |
477 | xfs_submit_ioend( | |
06342cf8 | 478 | struct writeback_control *wbc, |
7bf7f352 DC |
479 | xfs_ioend_t *ioend, |
480 | int fail) | |
f6d6d4fc | 481 | { |
d88992f6 | 482 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
483 | xfs_ioend_t *next; |
484 | struct buffer_head *bh; | |
485 | struct bio *bio; | |
486 | sector_t lastblock = 0; | |
487 | ||
d88992f6 DC |
488 | /* Pass 1 - start writeback */ |
489 | do { | |
490 | next = ioend->io_list; | |
221cb251 | 491 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 492 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
493 | } while ((ioend = next) != NULL); |
494 | ||
495 | /* Pass 2 - submit I/O */ | |
496 | ioend = head; | |
f6d6d4fc CH |
497 | do { |
498 | next = ioend->io_list; | |
499 | bio = NULL; | |
500 | ||
7bf7f352 DC |
501 | /* |
502 | * If we are failing the IO now, just mark the ioend with an | |
503 | * error and finish it. This will run IO completion immediately | |
504 | * as there is only one reference to the ioend at this point in | |
505 | * time. | |
506 | */ | |
507 | if (fail) { | |
2451337d | 508 | ioend->io_error = fail; |
7bf7f352 DC |
509 | xfs_finish_ioend(ioend); |
510 | continue; | |
511 | } | |
512 | ||
f6d6d4fc | 513 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
f6d6d4fc CH |
514 | |
515 | if (!bio) { | |
516 | retry: | |
517 | bio = xfs_alloc_ioend_bio(bh); | |
518 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 519 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
520 | goto retry; |
521 | } | |
522 | ||
c7c1a7d8 | 523 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
06342cf8 | 524 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
525 | goto retry; |
526 | } | |
527 | ||
528 | lastblock = bh->b_blocknr; | |
529 | } | |
530 | if (bio) | |
06342cf8 | 531 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 532 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
533 | } while ((ioend = next) != NULL); |
534 | } | |
535 | ||
f6d6d4fc CH |
536 | /* |
537 | * Test to see if we've been building up a completion structure for | |
538 | * earlier buffers -- if so, we try to append to this ioend if we | |
539 | * can, otherwise we finish off any current ioend and start another. | |
540 | * Return true if we've finished the given ioend. | |
541 | */ | |
542 | STATIC void | |
543 | xfs_add_to_ioend( | |
544 | struct inode *inode, | |
545 | struct buffer_head *bh, | |
7336cea8 | 546 | xfs_off_t offset, |
fbcc0256 | 547 | struct xfs_writepage_ctx *wpc) |
f6d6d4fc | 548 | { |
fbcc0256 DC |
549 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
550 | bh->b_blocknr != wpc->last_block + 1) { | |
551 | struct xfs_ioend *new; | |
552 | ||
553 | new = xfs_alloc_ioend(inode, wpc->io_type); | |
554 | new->io_offset = offset; | |
555 | new->io_buffer_head = bh; | |
556 | new->io_buffer_tail = bh; | |
557 | if (wpc->ioend) | |
558 | wpc->ioend->io_list = new; | |
559 | wpc->ioend = new; | |
f6d6d4fc | 560 | } else { |
fbcc0256 DC |
561 | wpc->ioend->io_buffer_tail->b_private = bh; |
562 | wpc->ioend->io_buffer_tail = bh; | |
f6d6d4fc CH |
563 | } |
564 | ||
565 | bh->b_private = NULL; | |
fbcc0256 DC |
566 | wpc->ioend->io_size += bh->b_size; |
567 | wpc->last_block = bh->b_blocknr; | |
f6d6d4fc CH |
568 | } |
569 | ||
87cbc49c NS |
570 | STATIC void |
571 | xfs_map_buffer( | |
046f1685 | 572 | struct inode *inode, |
87cbc49c | 573 | struct buffer_head *bh, |
207d0416 | 574 | struct xfs_bmbt_irec *imap, |
046f1685 | 575 | xfs_off_t offset) |
87cbc49c NS |
576 | { |
577 | sector_t bn; | |
8699bb0a | 578 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
579 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
580 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 581 | |
207d0416 CH |
582 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
583 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 584 | |
e513182d | 585 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 586 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 587 | |
046f1685 | 588 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
589 | |
590 | bh->b_blocknr = bn; | |
591 | set_buffer_mapped(bh); | |
592 | } | |
593 | ||
1da177e4 LT |
594 | STATIC void |
595 | xfs_map_at_offset( | |
046f1685 | 596 | struct inode *inode, |
1da177e4 | 597 | struct buffer_head *bh, |
207d0416 | 598 | struct xfs_bmbt_irec *imap, |
046f1685 | 599 | xfs_off_t offset) |
1da177e4 | 600 | { |
207d0416 CH |
601 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
602 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 603 | |
207d0416 | 604 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
605 | set_buffer_mapped(bh); |
606 | clear_buffer_delay(bh); | |
f6d6d4fc | 607 | clear_buffer_unwritten(bh); |
1da177e4 LT |
608 | } |
609 | ||
1da177e4 | 610 | /* |
a49935f2 DC |
611 | * Test if a given page contains at least one buffer of a given @type. |
612 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
613 | * try to find one of the type passed in. If it is not set, then the caller only | |
614 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 615 | */ |
a49935f2 | 616 | STATIC bool |
6ffc4db5 | 617 | xfs_check_page_type( |
10ce4444 | 618 | struct page *page, |
a49935f2 DC |
619 | unsigned int type, |
620 | bool check_all_buffers) | |
1da177e4 | 621 | { |
a49935f2 DC |
622 | struct buffer_head *bh; |
623 | struct buffer_head *head; | |
1da177e4 | 624 | |
a49935f2 DC |
625 | if (PageWriteback(page)) |
626 | return false; | |
627 | if (!page->mapping) | |
628 | return false; | |
629 | if (!page_has_buffers(page)) | |
630 | return false; | |
1da177e4 | 631 | |
a49935f2 DC |
632 | bh = head = page_buffers(page); |
633 | do { | |
634 | if (buffer_unwritten(bh)) { | |
635 | if (type == XFS_IO_UNWRITTEN) | |
636 | return true; | |
637 | } else if (buffer_delay(bh)) { | |
805eeb8e | 638 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
639 | return true; |
640 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 641 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
642 | return true; |
643 | } | |
1da177e4 | 644 | |
a49935f2 DC |
645 | /* If we are only checking the first buffer, we are done now. */ |
646 | if (!check_all_buffers) | |
647 | break; | |
648 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 649 | |
a49935f2 | 650 | return false; |
1da177e4 LT |
651 | } |
652 | ||
3ed3a434 DC |
653 | STATIC void |
654 | xfs_vm_invalidatepage( | |
655 | struct page *page, | |
d47992f8 LC |
656 | unsigned int offset, |
657 | unsigned int length) | |
3ed3a434 | 658 | { |
34097dfe LC |
659 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
660 | length); | |
661 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
662 | } |
663 | ||
664 | /* | |
665 | * If the page has delalloc buffers on it, we need to punch them out before we | |
666 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
667 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
668 | * is done on that same region - the delalloc extent is returned when none is | |
669 | * supposed to be there. | |
670 | * | |
671 | * We prevent this by truncating away the delalloc regions on the page before | |
672 | * invalidating it. Because they are delalloc, we can do this without needing a | |
673 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
674 | * truncation without a transaction as there is no space left for block | |
675 | * reservation (typically why we see a ENOSPC in writeback). | |
676 | * | |
677 | * This is not a performance critical path, so for now just do the punching a | |
678 | * buffer head at a time. | |
679 | */ | |
680 | STATIC void | |
681 | xfs_aops_discard_page( | |
682 | struct page *page) | |
683 | { | |
684 | struct inode *inode = page->mapping->host; | |
685 | struct xfs_inode *ip = XFS_I(inode); | |
686 | struct buffer_head *bh, *head; | |
687 | loff_t offset = page_offset(page); | |
3ed3a434 | 688 | |
a49935f2 | 689 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
690 | goto out_invalidate; |
691 | ||
e8c3753c DC |
692 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
693 | goto out_invalidate; | |
694 | ||
4f10700a | 695 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
696 | "page discard on page %p, inode 0x%llx, offset %llu.", |
697 | page, ip->i_ino, offset); | |
698 | ||
699 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
700 | bh = head = page_buffers(page); | |
701 | do { | |
3ed3a434 | 702 | int error; |
c726de44 | 703 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
704 | |
705 | if (!buffer_delay(bh)) | |
706 | goto next_buffer; | |
707 | ||
c726de44 DC |
708 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
709 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
710 | if (error) { |
711 | /* something screwed, just bail */ | |
e8c3753c | 712 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 713 | xfs_alert(ip->i_mount, |
3ed3a434 | 714 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 715 | } |
3ed3a434 DC |
716 | break; |
717 | } | |
718 | next_buffer: | |
c726de44 | 719 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
720 | |
721 | } while ((bh = bh->b_this_page) != head); | |
722 | ||
723 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
724 | out_invalidate: | |
d47992f8 | 725 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
3ed3a434 DC |
726 | return; |
727 | } | |
728 | ||
150d5be0 DC |
729 | static int |
730 | xfs_writepage_submit( | |
fbcc0256 | 731 | struct xfs_writepage_ctx *wpc, |
150d5be0 DC |
732 | struct writeback_control *wbc, |
733 | int status) | |
734 | { | |
735 | struct blk_plug plug; | |
736 | ||
737 | /* Reserve log space if we might write beyond the on-disk inode size. */ | |
fbcc0256 DC |
738 | if (!status && wpc->ioend && wpc->ioend->io_type != XFS_IO_UNWRITTEN && |
739 | xfs_ioend_is_append(wpc->ioend)) | |
740 | status = xfs_setfilesize_trans_alloc(wpc->ioend); | |
150d5be0 | 741 | |
fbcc0256 | 742 | if (wpc->iohead) { |
150d5be0 | 743 | blk_start_plug(&plug); |
fbcc0256 | 744 | xfs_submit_ioend(wbc, wpc->iohead, status); |
150d5be0 DC |
745 | blk_finish_plug(&plug); |
746 | } | |
747 | return status; | |
748 | } | |
749 | ||
1da177e4 | 750 | /* |
89f3b363 CH |
751 | * Write out a dirty page. |
752 | * | |
753 | * For delalloc space on the page we need to allocate space and flush it. | |
754 | * For unwritten space on the page we need to start the conversion to | |
755 | * regular allocated space. | |
89f3b363 | 756 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 757 | */ |
1da177e4 | 758 | STATIC int |
fbcc0256 | 759 | xfs_do_writepage( |
89f3b363 | 760 | struct page *page, |
fbcc0256 DC |
761 | struct writeback_control *wbc, |
762 | void *data) | |
1da177e4 | 763 | { |
fbcc0256 | 764 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 765 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 766 | struct buffer_head *bh, *head; |
1da177e4 | 767 | loff_t offset; |
1da177e4 | 768 | __uint64_t end_offset; |
ad68972a | 769 | pgoff_t end_index; |
ed1e7b7e | 770 | ssize_t len; |
fbcc0256 | 771 | int err, uptodate = 1; |
89f3b363 | 772 | int count = 0; |
89f3b363 | 773 | |
34097dfe | 774 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 775 | |
20cb52eb CH |
776 | ASSERT(page_has_buffers(page)); |
777 | ||
89f3b363 CH |
778 | /* |
779 | * Refuse to write the page out if we are called from reclaim context. | |
780 | * | |
d4f7a5cb CH |
781 | * This avoids stack overflows when called from deeply used stacks in |
782 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
783 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 784 | * |
94054fa3 MG |
785 | * This should never happen except in the case of a VM regression so |
786 | * warn about it. | |
89f3b363 | 787 | */ |
94054fa3 MG |
788 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
789 | PF_MEMALLOC)) | |
b5420f23 | 790 | goto redirty; |
1da177e4 | 791 | |
89f3b363 | 792 | /* |
680a647b CH |
793 | * Given that we do not allow direct reclaim to call us, we should |
794 | * never be called while in a filesystem transaction. | |
89f3b363 | 795 | */ |
448011e2 | 796 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 797 | goto redirty; |
89f3b363 | 798 | |
8695d27e | 799 | /* |
ad68972a DC |
800 | * Is this page beyond the end of the file? |
801 | * | |
8695d27e JL |
802 | * The page index is less than the end_index, adjust the end_offset |
803 | * to the highest offset that this page should represent. | |
804 | * ----------------------------------------------------- | |
805 | * | file mapping | <EOF> | | |
806 | * ----------------------------------------------------- | |
807 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
808 | * ^--------------------------------^----------|-------- | |
809 | * | desired writeback range | see else | | |
810 | * ---------------------------------^------------------| | |
811 | */ | |
ad68972a DC |
812 | offset = i_size_read(inode); |
813 | end_index = offset >> PAGE_CACHE_SHIFT; | |
8695d27e JL |
814 | if (page->index < end_index) |
815 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; | |
816 | else { | |
817 | /* | |
818 | * Check whether the page to write out is beyond or straddles | |
819 | * i_size or not. | |
820 | * ------------------------------------------------------- | |
821 | * | file mapping | <EOF> | | |
822 | * ------------------------------------------------------- | |
823 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
824 | * ^--------------------------------^-----------|--------- | |
825 | * | | Straddles | | |
826 | * ---------------------------------^-----------|--------| | |
827 | */ | |
6b7a03f0 CH |
828 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
829 | ||
830 | /* | |
ff9a28f6 JK |
831 | * Skip the page if it is fully outside i_size, e.g. due to a |
832 | * truncate operation that is in progress. We must redirty the | |
833 | * page so that reclaim stops reclaiming it. Otherwise | |
834 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
835 | * |
836 | * Note that the end_index is unsigned long, it would overflow | |
837 | * if the given offset is greater than 16TB on 32-bit system | |
838 | * and if we do check the page is fully outside i_size or not | |
839 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
840 | * will be evaluated to 0. Hence this page will be redirtied | |
841 | * and be written out repeatedly which would result in an | |
842 | * infinite loop, the user program that perform this operation | |
843 | * will hang. Instead, we can verify this situation by checking | |
844 | * if the page to write is totally beyond the i_size or if it's | |
845 | * offset is just equal to the EOF. | |
6b7a03f0 | 846 | */ |
8695d27e JL |
847 | if (page->index > end_index || |
848 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 849 | goto redirty; |
6b7a03f0 CH |
850 | |
851 | /* | |
852 | * The page straddles i_size. It must be zeroed out on each | |
853 | * and every writepage invocation because it may be mmapped. | |
854 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 855 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
856 | * memory is zeroed when mapped, and writes to that region are |
857 | * not written out to the file." | |
858 | */ | |
859 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
8695d27e JL |
860 | |
861 | /* Adjust the end_offset to the end of file */ | |
862 | end_offset = offset; | |
1da177e4 LT |
863 | } |
864 | ||
24e17b5f | 865 | len = 1 << inode->i_blkbits; |
24e17b5f | 866 | |
24e17b5f | 867 | bh = head = page_buffers(page); |
f6d6d4fc | 868 | offset = page_offset(page); |
a206c817 | 869 | |
1da177e4 LT |
870 | do { |
871 | if (offset >= end_offset) | |
872 | break; | |
873 | if (!buffer_uptodate(bh)) | |
874 | uptodate = 0; | |
1da177e4 | 875 | |
3d9b02e3 | 876 | /* |
ece413f5 CH |
877 | * set_page_dirty dirties all buffers in a page, independent |
878 | * of their state. The dirty state however is entirely | |
879 | * meaningless for holes (!mapped && uptodate), so skip | |
880 | * buffers covering holes here. | |
3d9b02e3 ES |
881 | */ |
882 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
fbcc0256 | 883 | wpc->imap_valid = false; |
3d9b02e3 ES |
884 | continue; |
885 | } | |
886 | ||
aeea1b1f | 887 | if (buffer_unwritten(bh)) { |
fbcc0256 DC |
888 | if (wpc->io_type != XFS_IO_UNWRITTEN) { |
889 | wpc->io_type = XFS_IO_UNWRITTEN; | |
890 | wpc->imap_valid = false; | |
1da177e4 | 891 | } |
aeea1b1f | 892 | } else if (buffer_delay(bh)) { |
fbcc0256 DC |
893 | if (wpc->io_type != XFS_IO_DELALLOC) { |
894 | wpc->io_type = XFS_IO_DELALLOC; | |
895 | wpc->imap_valid = false; | |
1da177e4 | 896 | } |
89f3b363 | 897 | } else if (buffer_uptodate(bh)) { |
fbcc0256 DC |
898 | if (wpc->io_type != XFS_IO_OVERWRITE) { |
899 | wpc->io_type = XFS_IO_OVERWRITE; | |
900 | wpc->imap_valid = false; | |
85da94c6 | 901 | } |
aeea1b1f | 902 | } else { |
7d0fa3ec | 903 | if (PageUptodate(page)) |
aeea1b1f | 904 | ASSERT(buffer_mapped(bh)); |
7d0fa3ec AR |
905 | /* |
906 | * This buffer is not uptodate and will not be | |
907 | * written to disk. Ensure that we will put any | |
908 | * subsequent writeable buffers into a new | |
909 | * ioend. | |
910 | */ | |
fbcc0256 | 911 | wpc->imap_valid = 0; |
aeea1b1f CH |
912 | continue; |
913 | } | |
d5cb48aa | 914 | |
fbcc0256 DC |
915 | if (wpc->imap_valid) |
916 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
917 | if (!wpc->imap_valid) { | |
918 | err = xfs_map_blocks(inode, offset, &wpc->imap, | |
919 | wpc->io_type); | |
aeea1b1f CH |
920 | if (err) |
921 | goto error; | |
fbcc0256 | 922 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); |
aeea1b1f | 923 | } |
fbcc0256 | 924 | if (wpc->imap_valid) { |
ecff71e6 | 925 | lock_buffer(bh); |
fbcc0256 DC |
926 | if (wpc->io_type != XFS_IO_OVERWRITE) |
927 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
928 | xfs_add_to_ioend(inode, bh, offset, wpc); | |
aeea1b1f | 929 | count++; |
1da177e4 | 930 | } |
f6d6d4fc | 931 | |
fbcc0256 DC |
932 | if (!wpc->iohead) |
933 | wpc->iohead = wpc->ioend; | |
f6d6d4fc CH |
934 | |
935 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
936 | |
937 | if (uptodate && bh == head) | |
938 | SetPageUptodate(page); | |
939 | ||
89f3b363 | 940 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 941 | |
ad68972a | 942 | ASSERT(wpc->iohead || !count); |
fbcc0256 | 943 | return 0; |
281627df | 944 | |
150d5be0 | 945 | error: |
7bf7f352 | 946 | /* |
150d5be0 DC |
947 | * On error, we have to fail the iohead here because we buffers locked |
948 | * in the ioend chain. If we don't do this, we'll deadlock invalidating | |
949 | * the page as that tries to lock the buffers on the page. Also, because | |
fbcc0256 DC |
950 | * we may have set pages under writeback, we have to make sure we run |
951 | * IO completion to mark the error state of the IO appropriately, so we | |
952 | * can't cancel the ioend directly here. That means we have to mark this | |
953 | * page as under writeback if we included any buffers from it in the | |
954 | * ioend chain so that completion treats it correctly. | |
955 | * | |
956 | * If we didn't include the page in the ioend, then we can simply | |
957 | * discard and unlock it as there are no other users of the page or it's | |
958 | * buffers right now. The caller will still need to trigger submission | |
959 | * of outstanding ioends on the writepage context so they are treated | |
960 | * correctly on error. | |
7bf7f352 | 961 | */ |
150d5be0 DC |
962 | if (count) |
963 | xfs_start_page_writeback(page, 0, count); | |
fbcc0256 | 964 | else { |
150d5be0 DC |
965 | xfs_aops_discard_page(page); |
966 | ClearPageUptodate(page); | |
967 | unlock_page(page); | |
968 | } | |
969 | mapping_set_error(page->mapping, err); | |
1da177e4 | 970 | return err; |
f51623b2 | 971 | |
b5420f23 | 972 | redirty: |
f51623b2 NS |
973 | redirty_page_for_writepage(wbc, page); |
974 | unlock_page(page); | |
975 | return 0; | |
f51623b2 NS |
976 | } |
977 | ||
fbcc0256 DC |
978 | STATIC int |
979 | xfs_vm_writepage( | |
980 | struct page *page, | |
981 | struct writeback_control *wbc) | |
982 | { | |
983 | struct xfs_writepage_ctx wpc = { | |
984 | .io_type = XFS_IO_INVALID, | |
985 | }; | |
986 | int ret; | |
987 | ||
988 | ret = xfs_do_writepage(page, wbc, &wpc); | |
989 | return xfs_writepage_submit(&wpc, wbc, ret); | |
990 | } | |
991 | ||
7d4fb40a NS |
992 | STATIC int |
993 | xfs_vm_writepages( | |
994 | struct address_space *mapping, | |
995 | struct writeback_control *wbc) | |
996 | { | |
fbcc0256 DC |
997 | struct xfs_writepage_ctx wpc = { |
998 | .io_type = XFS_IO_INVALID, | |
999 | }; | |
1000 | int ret; | |
1001 | ||
b3aea4ed | 1002 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
fbcc0256 DC |
1003 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
1004 | return xfs_writepage_submit(&wpc, wbc, ret); | |
7d4fb40a NS |
1005 | } |
1006 | ||
f51623b2 NS |
1007 | /* |
1008 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1009 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1010 | * have buffer heads in this call. |
1011 | * | |
89f3b363 | 1012 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1013 | */ |
1014 | STATIC int | |
238f4c54 | 1015 | xfs_vm_releasepage( |
f51623b2 NS |
1016 | struct page *page, |
1017 | gfp_t gfp_mask) | |
1018 | { | |
20cb52eb | 1019 | int delalloc, unwritten; |
f51623b2 | 1020 | |
34097dfe | 1021 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1022 | |
20cb52eb | 1023 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1024 | |
448011e2 | 1025 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1026 | return 0; |
448011e2 | 1027 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1028 | return 0; |
1029 | ||
f51623b2 NS |
1030 | return try_to_free_buffers(page); |
1031 | } | |
1032 | ||
a719370b | 1033 | /* |
a06c277a DC |
1034 | * When we map a DIO buffer, we may need to attach an ioend that describes the |
1035 | * type of write IO we are doing. This passes to the completion function the | |
1036 | * operations it needs to perform. If the mapping is for an overwrite wholly | |
1037 | * within the EOF then we don't need an ioend and so we don't allocate one. | |
1038 | * This avoids the unnecessary overhead of allocating and freeing ioends for | |
1039 | * workloads that don't require transactions on IO completion. | |
d5cc2e3f DC |
1040 | * |
1041 | * If we get multiple mappings in a single IO, we might be mapping different | |
1042 | * types. But because the direct IO can only have a single private pointer, we | |
1043 | * need to ensure that: | |
1044 | * | |
a06c277a DC |
1045 | * a) i) the ioend spans the entire region of unwritten mappings; or |
1046 | * ii) the ioend spans all the mappings that cross or are beyond EOF; and | |
d5cc2e3f DC |
1047 | * b) if it contains unwritten extents, it is *permanently* marked as such |
1048 | * | |
1049 | * We could do this by chaining ioends like buffered IO does, but we only | |
1050 | * actually get one IO completion callback from the direct IO, and that spans | |
1051 | * the entire IO regardless of how many mappings and IOs are needed to complete | |
1052 | * the DIO. There is only going to be one reference to the ioend and its life | |
1053 | * cycle is constrained by the DIO completion code. hence we don't need | |
1054 | * reference counting here. | |
3e12dbbd DC |
1055 | * |
1056 | * Note that for DIO, an IO to the highest supported file block offset (i.e. | |
1057 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 | |
1058 | * bit variable. Hence if we see this overflow, we have to assume that the IO is | |
1059 | * extending the file size. We won't know for sure until IO completion is run | |
1060 | * and the actual max write offset is communicated to the IO completion | |
1061 | * routine. | |
1062 | * | |
1063 | * For DAX page faults, we are preparing to never see unwritten extents here, | |
1064 | * nor should we ever extend the inode size. Hence we will soon have nothing to | |
1065 | * do here for this case, ensuring we don't have to provide an IO completion | |
1066 | * callback to free an ioend that we don't actually need for a fault into the | |
1067 | * page at offset (2^63 - 1FSB) bytes. | |
a719370b | 1068 | */ |
3e12dbbd | 1069 | |
a719370b DC |
1070 | static void |
1071 | xfs_map_direct( | |
1072 | struct inode *inode, | |
1073 | struct buffer_head *bh_result, | |
1074 | struct xfs_bmbt_irec *imap, | |
3e12dbbd DC |
1075 | xfs_off_t offset, |
1076 | bool dax_fault) | |
a719370b | 1077 | { |
d5cc2e3f DC |
1078 | struct xfs_ioend *ioend; |
1079 | xfs_off_t size = bh_result->b_size; | |
1080 | int type; | |
1081 | ||
1082 | if (ISUNWRITTEN(imap)) | |
1083 | type = XFS_IO_UNWRITTEN; | |
1084 | else | |
1085 | type = XFS_IO_OVERWRITE; | |
1086 | ||
1087 | trace_xfs_gbmap_direct(XFS_I(inode), offset, size, type, imap); | |
1088 | ||
3e12dbbd DC |
1089 | if (dax_fault) { |
1090 | ASSERT(type == XFS_IO_OVERWRITE); | |
1091 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, | |
1092 | imap); | |
1093 | return; | |
1094 | } | |
3e12dbbd | 1095 | |
d5cc2e3f DC |
1096 | if (bh_result->b_private) { |
1097 | ioend = bh_result->b_private; | |
1098 | ASSERT(ioend->io_size > 0); | |
1099 | ASSERT(offset >= ioend->io_offset); | |
1100 | if (offset + size > ioend->io_offset + ioend->io_size) | |
1101 | ioend->io_size = offset - ioend->io_offset + size; | |
1102 | ||
1103 | if (type == XFS_IO_UNWRITTEN && type != ioend->io_type) | |
1104 | ioend->io_type = XFS_IO_UNWRITTEN; | |
1105 | ||
1106 | trace_xfs_gbmap_direct_update(XFS_I(inode), ioend->io_offset, | |
1107 | ioend->io_size, ioend->io_type, | |
1108 | imap); | |
a06c277a | 1109 | } else if (type == XFS_IO_UNWRITTEN || |
3e12dbbd DC |
1110 | offset + size > i_size_read(inode) || |
1111 | offset + size < 0) { | |
d5cc2e3f DC |
1112 | ioend = xfs_alloc_ioend(inode, type); |
1113 | ioend->io_offset = offset; | |
1114 | ioend->io_size = size; | |
a06c277a | 1115 | |
d5cc2e3f | 1116 | bh_result->b_private = ioend; |
a06c277a | 1117 | set_buffer_defer_completion(bh_result); |
d5cc2e3f DC |
1118 | |
1119 | trace_xfs_gbmap_direct_new(XFS_I(inode), offset, size, type, | |
1120 | imap); | |
a06c277a DC |
1121 | } else { |
1122 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, | |
1123 | imap); | |
a719370b DC |
1124 | } |
1125 | } | |
1126 | ||
1fdca9c2 DC |
1127 | /* |
1128 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1129 | * is, so that we can avoid repeated get_blocks calls. | |
1130 | * | |
1131 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1132 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1133 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1134 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1135 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1136 | * to and including EOF, and a second mapping for beyond EOF. | |
1137 | */ | |
1138 | static void | |
1139 | xfs_map_trim_size( | |
1140 | struct inode *inode, | |
1141 | sector_t iblock, | |
1142 | struct buffer_head *bh_result, | |
1143 | struct xfs_bmbt_irec *imap, | |
1144 | xfs_off_t offset, | |
1145 | ssize_t size) | |
1146 | { | |
1147 | xfs_off_t mapping_size; | |
1148 | ||
1149 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1150 | mapping_size <<= inode->i_blkbits; | |
1151 | ||
1152 | ASSERT(mapping_size > 0); | |
1153 | if (mapping_size > size) | |
1154 | mapping_size = size; | |
1155 | if (offset < i_size_read(inode) && | |
1156 | offset + mapping_size >= i_size_read(inode)) { | |
1157 | /* limit mapping to block that spans EOF */ | |
1158 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1159 | 1 << inode->i_blkbits); | |
1160 | } | |
1161 | if (mapping_size > LONG_MAX) | |
1162 | mapping_size = LONG_MAX; | |
1163 | ||
1164 | bh_result->b_size = mapping_size; | |
1165 | } | |
1166 | ||
1da177e4 | 1167 | STATIC int |
c2536668 | 1168 | __xfs_get_blocks( |
1da177e4 LT |
1169 | struct inode *inode, |
1170 | sector_t iblock, | |
1da177e4 LT |
1171 | struct buffer_head *bh_result, |
1172 | int create, | |
3e12dbbd DC |
1173 | bool direct, |
1174 | bool dax_fault) | |
1da177e4 | 1175 | { |
a206c817 CH |
1176 | struct xfs_inode *ip = XFS_I(inode); |
1177 | struct xfs_mount *mp = ip->i_mount; | |
1178 | xfs_fileoff_t offset_fsb, end_fsb; | |
1179 | int error = 0; | |
1180 | int lockmode = 0; | |
207d0416 | 1181 | struct xfs_bmbt_irec imap; |
a206c817 | 1182 | int nimaps = 1; |
fdc7ed75 NS |
1183 | xfs_off_t offset; |
1184 | ssize_t size; | |
207d0416 | 1185 | int new = 0; |
a206c817 CH |
1186 | |
1187 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1188 | return -EIO; |
1da177e4 | 1189 | |
fdc7ed75 | 1190 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1191 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1192 | size = bh_result->b_size; | |
364f358a LM |
1193 | |
1194 | if (!create && direct && offset >= i_size_read(inode)) | |
1195 | return 0; | |
1196 | ||
507630b2 DC |
1197 | /* |
1198 | * Direct I/O is usually done on preallocated files, so try getting | |
1199 | * a block mapping without an exclusive lock first. For buffered | |
1200 | * writes we already have the exclusive iolock anyway, so avoiding | |
1201 | * a lock roundtrip here by taking the ilock exclusive from the | |
1202 | * beginning is a useful micro optimization. | |
1203 | */ | |
1204 | if (create && !direct) { | |
a206c817 CH |
1205 | lockmode = XFS_ILOCK_EXCL; |
1206 | xfs_ilock(ip, lockmode); | |
1207 | } else { | |
309ecac8 | 1208 | lockmode = xfs_ilock_data_map_shared(ip); |
a206c817 | 1209 | } |
f2bde9b8 | 1210 | |
d2c28191 DC |
1211 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1212 | if (offset + size > mp->m_super->s_maxbytes) | |
1213 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1214 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1215 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1216 | ||
5c8ed202 DC |
1217 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1218 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1219 | if (error) |
a206c817 CH |
1220 | goto out_unlock; |
1221 | ||
1ca19157 | 1222 | /* for DAX, we convert unwritten extents directly */ |
a206c817 CH |
1223 | if (create && |
1224 | (!nimaps || | |
1225 | (imap.br_startblock == HOLESTARTBLOCK || | |
1ca19157 DC |
1226 | imap.br_startblock == DELAYSTARTBLOCK) || |
1227 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { | |
aff3a9ed | 1228 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 | 1229 | /* |
009c6e87 BF |
1230 | * xfs_iomap_write_direct() expects the shared lock. It |
1231 | * is unlocked on return. | |
507630b2 | 1232 | */ |
009c6e87 BF |
1233 | if (lockmode == XFS_ILOCK_EXCL) |
1234 | xfs_ilock_demote(ip, lockmode); | |
1235 | ||
a206c817 CH |
1236 | error = xfs_iomap_write_direct(ip, offset, size, |
1237 | &imap, nimaps); | |
507630b2 | 1238 | if (error) |
2451337d | 1239 | return error; |
d3bc815a | 1240 | new = 1; |
6b698ede | 1241 | |
a206c817 | 1242 | } else { |
507630b2 DC |
1243 | /* |
1244 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1245 | * we can go on without dropping the lock here. If we |
1246 | * are allocating a new delalloc block, make sure that | |
1247 | * we set the new flag so that we mark the buffer new so | |
1248 | * that we know that it is newly allocated if the write | |
1249 | * fails. | |
507630b2 | 1250 | */ |
d3bc815a DC |
1251 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1252 | new = 1; | |
a206c817 | 1253 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1254 | if (error) |
1255 | goto out_unlock; | |
1256 | ||
1257 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1258 | } |
d5cc2e3f DC |
1259 | trace_xfs_get_blocks_alloc(ip, offset, size, |
1260 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1261 | : XFS_IO_DELALLOC, &imap); | |
a206c817 | 1262 | } else if (nimaps) { |
d5cc2e3f DC |
1263 | trace_xfs_get_blocks_found(ip, offset, size, |
1264 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1265 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1266 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1267 | } else { |
1268 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1269 | goto out_unlock; | |
1270 | } | |
1da177e4 | 1271 | |
1ca19157 DC |
1272 | if (IS_DAX(inode) && create) { |
1273 | ASSERT(!ISUNWRITTEN(&imap)); | |
1274 | /* zeroing is not needed at a higher layer */ | |
1275 | new = 0; | |
1276 | } | |
1277 | ||
1fdca9c2 DC |
1278 | /* trim mapping down to size requested */ |
1279 | if (direct || size > (1 << inode->i_blkbits)) | |
1280 | xfs_map_trim_size(inode, iblock, bh_result, | |
1281 | &imap, offset, size); | |
1282 | ||
a719370b DC |
1283 | /* |
1284 | * For unwritten extents do not report a disk address in the buffered | |
1285 | * read case (treat as if we're reading into a hole). | |
1286 | */ | |
207d0416 | 1287 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1288 | imap.br_startblock != DELAYSTARTBLOCK && |
1289 | (create || !ISUNWRITTEN(&imap))) { | |
1290 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1291 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1292 | set_buffer_unwritten(bh_result); |
a719370b DC |
1293 | /* direct IO needs special help */ |
1294 | if (create && direct) | |
3e12dbbd DC |
1295 | xfs_map_direct(inode, bh_result, &imap, offset, |
1296 | dax_fault); | |
1da177e4 LT |
1297 | } |
1298 | ||
c2536668 NS |
1299 | /* |
1300 | * If this is a realtime file, data may be on a different device. | |
1301 | * to that pointed to from the buffer_head b_bdev currently. | |
1302 | */ | |
046f1685 | 1303 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1304 | |
c2536668 | 1305 | /* |
549054af DC |
1306 | * If we previously allocated a block out beyond eof and we are now |
1307 | * coming back to use it then we will need to flag it as new even if it | |
1308 | * has a disk address. | |
1309 | * | |
1310 | * With sub-block writes into unwritten extents we also need to mark | |
1311 | * the buffer as new so that the unwritten parts of the buffer gets | |
1312 | * correctly zeroed. | |
1da177e4 LT |
1313 | */ |
1314 | if (create && | |
1315 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1316 | (offset >= i_size_read(inode)) || |
207d0416 | 1317 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1318 | set_buffer_new(bh_result); |
1da177e4 | 1319 | |
207d0416 | 1320 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1321 | BUG_ON(direct); |
1322 | if (create) { | |
1323 | set_buffer_uptodate(bh_result); | |
1324 | set_buffer_mapped(bh_result); | |
1325 | set_buffer_delay(bh_result); | |
1326 | } | |
1327 | } | |
1328 | ||
1da177e4 | 1329 | return 0; |
a206c817 CH |
1330 | |
1331 | out_unlock: | |
1332 | xfs_iunlock(ip, lockmode); | |
2451337d | 1333 | return error; |
1da177e4 LT |
1334 | } |
1335 | ||
1336 | int | |
c2536668 | 1337 | xfs_get_blocks( |
1da177e4 LT |
1338 | struct inode *inode, |
1339 | sector_t iblock, | |
1340 | struct buffer_head *bh_result, | |
1341 | int create) | |
1342 | { | |
3e12dbbd | 1343 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
1da177e4 LT |
1344 | } |
1345 | ||
6b698ede | 1346 | int |
e4c573bb | 1347 | xfs_get_blocks_direct( |
1da177e4 LT |
1348 | struct inode *inode, |
1349 | sector_t iblock, | |
1da177e4 LT |
1350 | struct buffer_head *bh_result, |
1351 | int create) | |
1352 | { | |
3e12dbbd DC |
1353 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
1354 | } | |
1355 | ||
1356 | int | |
1357 | xfs_get_blocks_dax_fault( | |
1358 | struct inode *inode, | |
1359 | sector_t iblock, | |
1360 | struct buffer_head *bh_result, | |
1361 | int create) | |
1362 | { | |
1363 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); | |
1da177e4 LT |
1364 | } |
1365 | ||
6b698ede DC |
1366 | static void |
1367 | __xfs_end_io_direct_write( | |
1368 | struct inode *inode, | |
1369 | struct xfs_ioend *ioend, | |
209fb87a | 1370 | loff_t offset, |
6b698ede | 1371 | ssize_t size) |
f0973863 | 1372 | { |
6b698ede | 1373 | struct xfs_mount *mp = XFS_I(inode)->i_mount; |
a06c277a | 1374 | |
6b698ede | 1375 | if (XFS_FORCED_SHUTDOWN(mp) || ioend->io_error) |
6dfa1b67 | 1376 | goto out_end_io; |
f0973863 | 1377 | |
2813d682 | 1378 | /* |
d5cc2e3f DC |
1379 | * dio completion end_io functions are only called on writes if more |
1380 | * than 0 bytes was written. | |
2813d682 | 1381 | */ |
d5cc2e3f DC |
1382 | ASSERT(size > 0); |
1383 | ||
1384 | /* | |
1385 | * The ioend only maps whole blocks, while the IO may be sector aligned. | |
a06c277a DC |
1386 | * Hence the ioend offset/size may not match the IO offset/size exactly. |
1387 | * Because we don't map overwrites within EOF into the ioend, the offset | |
1388 | * may not match, but only if the endio spans EOF. Either way, write | |
1389 | * the IO sizes into the ioend so that completion processing does the | |
1390 | * right thing. | |
d5cc2e3f | 1391 | */ |
d5cc2e3f DC |
1392 | ASSERT(offset + size <= ioend->io_offset + ioend->io_size); |
1393 | ioend->io_size = size; | |
1394 | ioend->io_offset = offset; | |
f0973863 | 1395 | |
2813d682 | 1396 | /* |
6dfa1b67 DC |
1397 | * The ioend tells us whether we are doing unwritten extent conversion |
1398 | * or an append transaction that updates the on-disk file size. These | |
1399 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1400 | * to update the VFS inode size. |
6dfa1b67 DC |
1401 | * |
1402 | * We need to update the in-core inode size here so that we don't end up | |
a06c277a DC |
1403 | * with the on-disk inode size being outside the in-core inode size. We |
1404 | * have no other method of updating EOF for AIO, so always do it here | |
1405 | * if necessary. | |
b9d59846 DC |
1406 | * |
1407 | * We need to lock the test/set EOF update as we can be racing with | |
1408 | * other IO completions here to update the EOF. Failing to serialise | |
1409 | * here can result in EOF moving backwards and Bad Things Happen when | |
1410 | * that occurs. | |
2813d682 | 1411 | */ |
6b698ede | 1412 | spin_lock(&XFS_I(inode)->i_flags_lock); |
2ba66237 CH |
1413 | if (offset + size > i_size_read(inode)) |
1414 | i_size_write(inode, offset + size); | |
6b698ede | 1415 | spin_unlock(&XFS_I(inode)->i_flags_lock); |
2813d682 | 1416 | |
f0973863 | 1417 | /* |
6dfa1b67 DC |
1418 | * If we are doing an append IO that needs to update the EOF on disk, |
1419 | * do the transaction reserve now so we can use common end io | |
1420 | * processing. Stashing the error (if there is one) in the ioend will | |
1421 | * result in the ioend processing passing on the error if it is | |
1422 | * possible as we can't return it from here. | |
f0973863 | 1423 | */ |
a06c277a | 1424 | if (ioend->io_type == XFS_IO_OVERWRITE) |
6dfa1b67 | 1425 | ioend->io_error = xfs_setfilesize_trans_alloc(ioend); |
209fb87a | 1426 | |
6dfa1b67 DC |
1427 | out_end_io: |
1428 | xfs_end_io(&ioend->io_work); | |
1429 | return; | |
f0973863 CH |
1430 | } |
1431 | ||
6b698ede DC |
1432 | /* |
1433 | * Complete a direct I/O write request. | |
1434 | * | |
1435 | * The ioend structure is passed from __xfs_get_blocks() to tell us what to do. | |
1436 | * If no ioend exists (i.e. @private == NULL) then the write IO is an overwrite | |
1437 | * wholly within the EOF and so there is nothing for us to do. Note that in this | |
1438 | * case the completion can be called in interrupt context, whereas if we have an | |
1439 | * ioend we will always be called in task context (i.e. from a workqueue). | |
1440 | */ | |
1441 | STATIC void | |
1442 | xfs_end_io_direct_write( | |
1443 | struct kiocb *iocb, | |
1444 | loff_t offset, | |
1445 | ssize_t size, | |
1446 | void *private) | |
1447 | { | |
1448 | struct inode *inode = file_inode(iocb->ki_filp); | |
1449 | struct xfs_ioend *ioend = private; | |
1450 | ||
1451 | trace_xfs_gbmap_direct_endio(XFS_I(inode), offset, size, | |
1452 | ioend ? ioend->io_type : 0, NULL); | |
1453 | ||
1454 | if (!ioend) { | |
1455 | ASSERT(offset + size <= i_size_read(inode)); | |
1456 | return; | |
1457 | } | |
1458 | ||
1459 | __xfs_end_io_direct_write(inode, ioend, offset, size); | |
1460 | } | |
1461 | ||
6e1ba0bc DC |
1462 | static inline ssize_t |
1463 | xfs_vm_do_dio( | |
1464 | struct inode *inode, | |
1465 | struct kiocb *iocb, | |
1466 | struct iov_iter *iter, | |
1467 | loff_t offset, | |
1468 | void (*endio)(struct kiocb *iocb, | |
1469 | loff_t offset, | |
1470 | ssize_t size, | |
1471 | void *private), | |
1472 | int flags) | |
1473 | { | |
1474 | struct block_device *bdev; | |
1475 | ||
1476 | if (IS_DAX(inode)) | |
1477 | return dax_do_io(iocb, inode, iter, offset, | |
1478 | xfs_get_blocks_direct, endio, 0); | |
1479 | ||
1480 | bdev = xfs_find_bdev_for_inode(inode); | |
1481 | return __blockdev_direct_IO(iocb, inode, bdev, iter, offset, | |
1482 | xfs_get_blocks_direct, endio, NULL, flags); | |
1483 | } | |
1484 | ||
1da177e4 | 1485 | STATIC ssize_t |
e4c573bb | 1486 | xfs_vm_direct_IO( |
1da177e4 | 1487 | struct kiocb *iocb, |
d8d3d94b AV |
1488 | struct iov_iter *iter, |
1489 | loff_t offset) | |
1da177e4 | 1490 | { |
209fb87a | 1491 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
209fb87a | 1492 | |
6e1ba0bc DC |
1493 | if (iov_iter_rw(iter) == WRITE) |
1494 | return xfs_vm_do_dio(inode, iocb, iter, offset, | |
1495 | xfs_end_io_direct_write, DIO_ASYNC_EXTEND); | |
1496 | return xfs_vm_do_dio(inode, iocb, iter, offset, NULL, 0); | |
1da177e4 LT |
1497 | } |
1498 | ||
d3bc815a DC |
1499 | /* |
1500 | * Punch out the delalloc blocks we have already allocated. | |
1501 | * | |
1502 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1503 | * as the page is still locked at this point. | |
1504 | */ | |
1505 | STATIC void | |
1506 | xfs_vm_kill_delalloc_range( | |
1507 | struct inode *inode, | |
1508 | loff_t start, | |
1509 | loff_t end) | |
1510 | { | |
1511 | struct xfs_inode *ip = XFS_I(inode); | |
1512 | xfs_fileoff_t start_fsb; | |
1513 | xfs_fileoff_t end_fsb; | |
1514 | int error; | |
1515 | ||
1516 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1517 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1518 | if (end_fsb <= start_fsb) | |
1519 | return; | |
1520 | ||
1521 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1522 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1523 | end_fsb - start_fsb); | |
1524 | if (error) { | |
1525 | /* something screwed, just bail */ | |
1526 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1527 | xfs_alert(ip->i_mount, | |
1528 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1529 | ip->i_ino); | |
1530 | } | |
1531 | } | |
1532 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1533 | } | |
1534 | ||
fa9b227e CH |
1535 | STATIC void |
1536 | xfs_vm_write_failed( | |
d3bc815a DC |
1537 | struct inode *inode, |
1538 | struct page *page, | |
1539 | loff_t pos, | |
1540 | unsigned len) | |
fa9b227e | 1541 | { |
58e59854 | 1542 | loff_t block_offset; |
d3bc815a DC |
1543 | loff_t block_start; |
1544 | loff_t block_end; | |
1545 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1546 | loff_t to = from + len; | |
1547 | struct buffer_head *bh, *head; | |
fa9b227e | 1548 | |
58e59854 JL |
1549 | /* |
1550 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1551 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1552 | * platform, the high 32-bit will be masked off if we evaluate the | |
1553 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1554 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1555 | * which could cause the following ASSERT failed in most cases. | |
1556 | * In order to avoid this, we can evaluate the block_offset of the | |
1557 | * start of the page by using shifts rather than masks the mismatch | |
1558 | * problem. | |
1559 | */ | |
1560 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1561 | ||
d3bc815a | 1562 | ASSERT(block_offset + from == pos); |
c726de44 | 1563 | |
d3bc815a DC |
1564 | head = page_buffers(page); |
1565 | block_start = 0; | |
1566 | for (bh = head; bh != head || !block_start; | |
1567 | bh = bh->b_this_page, block_start = block_end, | |
1568 | block_offset += bh->b_size) { | |
1569 | block_end = block_start + bh->b_size; | |
c726de44 | 1570 | |
d3bc815a DC |
1571 | /* skip buffers before the write */ |
1572 | if (block_end <= from) | |
1573 | continue; | |
1574 | ||
1575 | /* if the buffer is after the write, we're done */ | |
1576 | if (block_start >= to) | |
1577 | break; | |
1578 | ||
1579 | if (!buffer_delay(bh)) | |
1580 | continue; | |
1581 | ||
1582 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1583 | continue; | |
1584 | ||
1585 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1586 | block_offset + bh->b_size); | |
4ab9ed57 DC |
1587 | |
1588 | /* | |
1589 | * This buffer does not contain data anymore. make sure anyone | |
1590 | * who finds it knows that for certain. | |
1591 | */ | |
1592 | clear_buffer_delay(bh); | |
1593 | clear_buffer_uptodate(bh); | |
1594 | clear_buffer_mapped(bh); | |
1595 | clear_buffer_new(bh); | |
1596 | clear_buffer_dirty(bh); | |
fa9b227e | 1597 | } |
d3bc815a | 1598 | |
fa9b227e CH |
1599 | } |
1600 | ||
d3bc815a DC |
1601 | /* |
1602 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1603 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1604 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1605 | * the appropriate point. | |
1606 | */ | |
f51623b2 | 1607 | STATIC int |
d79689c7 | 1608 | xfs_vm_write_begin( |
f51623b2 | 1609 | struct file *file, |
d79689c7 NP |
1610 | struct address_space *mapping, |
1611 | loff_t pos, | |
1612 | unsigned len, | |
1613 | unsigned flags, | |
1614 | struct page **pagep, | |
1615 | void **fsdata) | |
f51623b2 | 1616 | { |
d3bc815a DC |
1617 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1618 | struct page *page; | |
1619 | int status; | |
155130a4 | 1620 | |
d3bc815a DC |
1621 | ASSERT(len <= PAGE_CACHE_SIZE); |
1622 | ||
ad22c7a0 | 1623 | page = grab_cache_page_write_begin(mapping, index, flags); |
d3bc815a DC |
1624 | if (!page) |
1625 | return -ENOMEM; | |
1626 | ||
1627 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1628 | if (unlikely(status)) { | |
1629 | struct inode *inode = mapping->host; | |
72ab70a1 | 1630 | size_t isize = i_size_read(inode); |
d3bc815a DC |
1631 | |
1632 | xfs_vm_write_failed(inode, page, pos, len); | |
1633 | unlock_page(page); | |
1634 | ||
72ab70a1 DC |
1635 | /* |
1636 | * If the write is beyond EOF, we only want to kill blocks | |
1637 | * allocated in this write, not blocks that were previously | |
1638 | * written successfully. | |
1639 | */ | |
1640 | if (pos + len > isize) { | |
1641 | ssize_t start = max_t(ssize_t, pos, isize); | |
1642 | ||
1643 | truncate_pagecache_range(inode, start, pos + len); | |
1644 | } | |
d3bc815a DC |
1645 | |
1646 | page_cache_release(page); | |
1647 | page = NULL; | |
1648 | } | |
1649 | ||
1650 | *pagep = page; | |
1651 | return status; | |
fa9b227e CH |
1652 | } |
1653 | ||
d3bc815a | 1654 | /* |
aad3f375 DC |
1655 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
1656 | * this specific write because they will never be written. Previous writes | |
1657 | * beyond EOF where block allocation succeeded do not need to be trashed, so | |
1658 | * only new blocks from this write should be trashed. For blocks within | |
1659 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be | |
1660 | * written with all the other valid data. | |
d3bc815a | 1661 | */ |
fa9b227e CH |
1662 | STATIC int |
1663 | xfs_vm_write_end( | |
1664 | struct file *file, | |
1665 | struct address_space *mapping, | |
1666 | loff_t pos, | |
1667 | unsigned len, | |
1668 | unsigned copied, | |
1669 | struct page *page, | |
1670 | void *fsdata) | |
1671 | { | |
1672 | int ret; | |
155130a4 | 1673 | |
d3bc815a DC |
1674 | ASSERT(len <= PAGE_CACHE_SIZE); |
1675 | ||
fa9b227e | 1676 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1677 | if (unlikely(ret < len)) { |
1678 | struct inode *inode = mapping->host; | |
1679 | size_t isize = i_size_read(inode); | |
1680 | loff_t to = pos + len; | |
1681 | ||
1682 | if (to > isize) { | |
aad3f375 DC |
1683 | /* only kill blocks in this write beyond EOF */ |
1684 | if (pos > isize) | |
1685 | isize = pos; | |
d3bc815a | 1686 | xfs_vm_kill_delalloc_range(inode, isize, to); |
aad3f375 | 1687 | truncate_pagecache_range(inode, isize, to); |
d3bc815a DC |
1688 | } |
1689 | } | |
155130a4 | 1690 | return ret; |
f51623b2 | 1691 | } |
1da177e4 LT |
1692 | |
1693 | STATIC sector_t | |
e4c573bb | 1694 | xfs_vm_bmap( |
1da177e4 LT |
1695 | struct address_space *mapping, |
1696 | sector_t block) | |
1697 | { | |
1698 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1699 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1700 | |
cca28fb8 | 1701 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1702 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1703 | filemap_write_and_wait(mapping); |
126468b1 | 1704 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1705 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1706 | } |
1707 | ||
1708 | STATIC int | |
e4c573bb | 1709 | xfs_vm_readpage( |
1da177e4 LT |
1710 | struct file *unused, |
1711 | struct page *page) | |
1712 | { | |
121e213e | 1713 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1714 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1715 | } |
1716 | ||
1717 | STATIC int | |
e4c573bb | 1718 | xfs_vm_readpages( |
1da177e4 LT |
1719 | struct file *unused, |
1720 | struct address_space *mapping, | |
1721 | struct list_head *pages, | |
1722 | unsigned nr_pages) | |
1723 | { | |
121e213e | 1724 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1725 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1726 | } |
1727 | ||
22e757a4 DC |
1728 | /* |
1729 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1730 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1731 | * dirty, we'll never be able to clean them because we don't write buffers | |
1732 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1733 | * that have been marked dirty. Further, the dirty state can leak into | |
1734 | * the file interior if the file is extended, resulting in all sorts of | |
1735 | * bad things happening as the state does not match the underlying data. | |
1736 | * | |
1737 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1738 | * this only exist because of bufferheads and how the generic code manages them. | |
1739 | */ | |
1740 | STATIC int | |
1741 | xfs_vm_set_page_dirty( | |
1742 | struct page *page) | |
1743 | { | |
1744 | struct address_space *mapping = page->mapping; | |
1745 | struct inode *inode = mapping->host; | |
1746 | loff_t end_offset; | |
1747 | loff_t offset; | |
1748 | int newly_dirty; | |
c4843a75 | 1749 | struct mem_cgroup *memcg; |
22e757a4 DC |
1750 | |
1751 | if (unlikely(!mapping)) | |
1752 | return !TestSetPageDirty(page); | |
1753 | ||
1754 | end_offset = i_size_read(inode); | |
1755 | offset = page_offset(page); | |
1756 | ||
1757 | spin_lock(&mapping->private_lock); | |
1758 | if (page_has_buffers(page)) { | |
1759 | struct buffer_head *head = page_buffers(page); | |
1760 | struct buffer_head *bh = head; | |
1761 | ||
1762 | do { | |
1763 | if (offset < end_offset) | |
1764 | set_buffer_dirty(bh); | |
1765 | bh = bh->b_this_page; | |
1766 | offset += 1 << inode->i_blkbits; | |
1767 | } while (bh != head); | |
1768 | } | |
c4843a75 GT |
1769 | /* |
1770 | * Use mem_group_begin_page_stat() to keep PageDirty synchronized with | |
1771 | * per-memcg dirty page counters. | |
1772 | */ | |
1773 | memcg = mem_cgroup_begin_page_stat(page); | |
22e757a4 DC |
1774 | newly_dirty = !TestSetPageDirty(page); |
1775 | spin_unlock(&mapping->private_lock); | |
1776 | ||
1777 | if (newly_dirty) { | |
1778 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1779 | unsigned long flags; | |
1780 | ||
1781 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1782 | if (page->mapping) { /* Race with truncate? */ | |
1783 | WARN_ON_ONCE(!PageUptodate(page)); | |
c4843a75 | 1784 | account_page_dirtied(page, mapping, memcg); |
22e757a4 DC |
1785 | radix_tree_tag_set(&mapping->page_tree, |
1786 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1787 | } | |
1788 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1789 | } |
c4843a75 GT |
1790 | mem_cgroup_end_page_stat(memcg); |
1791 | if (newly_dirty) | |
1792 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1793 | return newly_dirty; |
1794 | } | |
1795 | ||
f5e54d6e | 1796 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1797 | .readpage = xfs_vm_readpage, |
1798 | .readpages = xfs_vm_readpages, | |
1799 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1800 | .writepages = xfs_vm_writepages, |
22e757a4 | 1801 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1802 | .releasepage = xfs_vm_releasepage, |
1803 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1804 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1805 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1806 | .bmap = xfs_vm_bmap, |
1807 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1808 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1809 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1810 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1811 | }; |