]>
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" |
ef473667 | 34 | #include "xfs_reflink.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 | ||
fbcc0256 DC |
40 | /* |
41 | * structure owned by writepages passed to individual writepage calls | |
42 | */ | |
43 | struct xfs_writepage_ctx { | |
44 | struct xfs_bmbt_irec imap; | |
45 | bool imap_valid; | |
46 | unsigned int io_type; | |
fbcc0256 DC |
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 | ||
20a90f58 | 70 | struct block_device * |
6214ed44 | 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 | ||
486aff5e DW |
83 | struct dax_device * |
84 | xfs_find_daxdev_for_inode( | |
85 | struct inode *inode) | |
86 | { | |
87 | struct xfs_inode *ip = XFS_I(inode); | |
88 | struct xfs_mount *mp = ip->i_mount; | |
89 | ||
90 | if (XFS_IS_REALTIME_INODE(ip)) | |
91 | return mp->m_rtdev_targp->bt_daxdev; | |
92 | else | |
93 | return mp->m_ddev_targp->bt_daxdev; | |
94 | } | |
95 | ||
f6d6d4fc | 96 | /* |
37992c18 DC |
97 | * We're now finished for good with this page. Update the page state via the |
98 | * associated buffer_heads, paying attention to the start and end offsets that | |
99 | * we need to process on the page. | |
28b783e4 | 100 | * |
8353a814 CH |
101 | * Note that we open code the action in end_buffer_async_write here so that we |
102 | * only have to iterate over the buffers attached to the page once. This is not | |
103 | * only more efficient, but also ensures that we only calls end_page_writeback | |
104 | * at the end of the iteration, and thus avoids the pitfall of having the page | |
105 | * and buffers potentially freed after every call to end_buffer_async_write. | |
37992c18 DC |
106 | */ |
107 | static void | |
108 | xfs_finish_page_writeback( | |
109 | struct inode *inode, | |
110 | struct bio_vec *bvec, | |
111 | int error) | |
112 | { | |
8353a814 CH |
113 | struct buffer_head *head = page_buffers(bvec->bv_page), *bh = head; |
114 | bool busy = false; | |
37992c18 | 115 | unsigned int off = 0; |
8353a814 | 116 | unsigned long flags; |
37992c18 DC |
117 | |
118 | ASSERT(bvec->bv_offset < PAGE_SIZE); | |
93407472 | 119 | ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0); |
8353a814 | 120 | ASSERT(bvec->bv_offset + bvec->bv_len <= PAGE_SIZE); |
93407472 | 121 | ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0); |
37992c18 | 122 | |
8353a814 CH |
123 | local_irq_save(flags); |
124 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
37992c18 | 125 | do { |
8353a814 CH |
126 | if (off >= bvec->bv_offset && |
127 | off < bvec->bv_offset + bvec->bv_len) { | |
128 | ASSERT(buffer_async_write(bh)); | |
129 | ASSERT(bh->b_end_io == NULL); | |
130 | ||
131 | if (error) { | |
132 | mark_buffer_write_io_error(bh); | |
133 | clear_buffer_uptodate(bh); | |
134 | SetPageError(bvec->bv_page); | |
135 | } else { | |
136 | set_buffer_uptodate(bh); | |
137 | } | |
138 | clear_buffer_async_write(bh); | |
139 | unlock_buffer(bh); | |
140 | } else if (buffer_async_write(bh)) { | |
141 | ASSERT(buffer_locked(bh)); | |
142 | busy = true; | |
143 | } | |
144 | off += bh->b_size; | |
145 | } while ((bh = bh->b_this_page) != head); | |
146 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
147 | local_irq_restore(flags); | |
148 | ||
149 | if (!busy) | |
150 | end_page_writeback(bvec->bv_page); | |
37992c18 DC |
151 | } |
152 | ||
153 | /* | |
154 | * We're now finished for good with this ioend structure. Update the page | |
155 | * state, release holds on bios, and finally free up memory. Do not use the | |
156 | * ioend after this. | |
f6d6d4fc | 157 | */ |
0829c360 CH |
158 | STATIC void |
159 | xfs_destroy_ioend( | |
0e51a8e1 CH |
160 | struct xfs_ioend *ioend, |
161 | int error) | |
0829c360 | 162 | { |
37992c18 | 163 | struct inode *inode = ioend->io_inode; |
8353a814 CH |
164 | struct bio *bio = &ioend->io_inline_bio; |
165 | struct bio *last = ioend->io_bio, *next; | |
166 | u64 start = bio->bi_iter.bi_sector; | |
167 | bool quiet = bio_flagged(bio, BIO_QUIET); | |
f6d6d4fc | 168 | |
0e51a8e1 | 169 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
37992c18 DC |
170 | struct bio_vec *bvec; |
171 | int i; | |
172 | ||
0e51a8e1 CH |
173 | /* |
174 | * For the last bio, bi_private points to the ioend, so we | |
175 | * need to explicitly end the iteration here. | |
176 | */ | |
177 | if (bio == last) | |
178 | next = NULL; | |
179 | else | |
180 | next = bio->bi_private; | |
583fa586 | 181 | |
37992c18 DC |
182 | /* walk each page on bio, ending page IO on them */ |
183 | bio_for_each_segment_all(bvec, bio, i) | |
184 | xfs_finish_page_writeback(inode, bvec, error); | |
185 | ||
186 | bio_put(bio); | |
f6d6d4fc | 187 | } |
8353a814 CH |
188 | |
189 | if (unlikely(error && !quiet)) { | |
190 | xfs_err_ratelimited(XFS_I(inode)->i_mount, | |
191 | "writeback error on sector %llu", start); | |
192 | } | |
0829c360 CH |
193 | } |
194 | ||
fc0063c4 CH |
195 | /* |
196 | * Fast and loose check if this write could update the on-disk inode size. | |
197 | */ | |
198 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
199 | { | |
200 | return ioend->io_offset + ioend->io_size > | |
201 | XFS_I(ioend->io_inode)->i_d.di_size; | |
202 | } | |
203 | ||
281627df CH |
204 | STATIC int |
205 | xfs_setfilesize_trans_alloc( | |
206 | struct xfs_ioend *ioend) | |
207 | { | |
208 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
209 | struct xfs_trans *tp; | |
210 | int error; | |
211 | ||
4df0f7f1 DC |
212 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, |
213 | XFS_TRANS_NOFS, &tp); | |
253f4911 | 214 | if (error) |
281627df | 215 | return error; |
281627df CH |
216 | |
217 | ioend->io_append_trans = tp; | |
218 | ||
d9457dc0 | 219 | /* |
437a255a | 220 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
221 | * we released it. |
222 | */ | |
bee9182d | 223 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
224 | /* |
225 | * We hand off the transaction to the completion thread now, so | |
226 | * clear the flag here. | |
227 | */ | |
9070733b | 228 | current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
281627df CH |
229 | return 0; |
230 | } | |
231 | ||
ba87ea69 | 232 | /* |
2813d682 | 233 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 234 | */ |
281627df | 235 | STATIC int |
e372843a | 236 | __xfs_setfilesize( |
2ba66237 CH |
237 | struct xfs_inode *ip, |
238 | struct xfs_trans *tp, | |
239 | xfs_off_t offset, | |
240 | size_t size) | |
ba87ea69 | 241 | { |
ba87ea69 | 242 | xfs_fsize_t isize; |
ba87ea69 | 243 | |
aa6bf01d | 244 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 245 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
246 | if (!isize) { |
247 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 248 | xfs_trans_cancel(tp); |
281627df | 249 | return 0; |
ba87ea69 LM |
250 | } |
251 | ||
2ba66237 | 252 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
253 | |
254 | ip->i_d.di_size = isize; | |
255 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
256 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
257 | ||
70393313 | 258 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
259 | } |
260 | ||
e372843a CH |
261 | int |
262 | xfs_setfilesize( | |
263 | struct xfs_inode *ip, | |
264 | xfs_off_t offset, | |
265 | size_t size) | |
266 | { | |
267 | struct xfs_mount *mp = ip->i_mount; | |
268 | struct xfs_trans *tp; | |
269 | int error; | |
270 | ||
271 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); | |
272 | if (error) | |
273 | return error; | |
274 | ||
275 | return __xfs_setfilesize(ip, tp, offset, size); | |
276 | } | |
277 | ||
2ba66237 CH |
278 | STATIC int |
279 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
280 | struct xfs_ioend *ioend, |
281 | int error) | |
2ba66237 CH |
282 | { |
283 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
284 | struct xfs_trans *tp = ioend->io_append_trans; | |
285 | ||
286 | /* | |
287 | * The transaction may have been allocated in the I/O submission thread, | |
288 | * thus we need to mark ourselves as being in a transaction manually. | |
289 | * Similarly for freeze protection. | |
290 | */ | |
9070733b | 291 | current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
bee9182d | 292 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 293 | |
5cb13dcd | 294 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 295 | if (error) { |
5cb13dcd | 296 | xfs_trans_cancel(tp); |
0e51a8e1 | 297 | return error; |
5cb13dcd Z |
298 | } |
299 | ||
e372843a | 300 | return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
2ba66237 CH |
301 | } |
302 | ||
0829c360 | 303 | /* |
5ec4fabb | 304 | * IO write completion. |
f6d6d4fc CH |
305 | */ |
306 | STATIC void | |
5ec4fabb | 307 | xfs_end_io( |
77d7a0c2 | 308 | struct work_struct *work) |
0829c360 | 309 | { |
0e51a8e1 CH |
310 | struct xfs_ioend *ioend = |
311 | container_of(work, struct xfs_ioend, io_work); | |
312 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
787eb485 CH |
313 | xfs_off_t offset = ioend->io_offset; |
314 | size_t size = ioend->io_size; | |
4e4cbee9 | 315 | int error; |
ba87ea69 | 316 | |
af055e37 | 317 | /* |
787eb485 | 318 | * Just clean up the in-memory strutures if the fs has been shut down. |
af055e37 | 319 | */ |
787eb485 | 320 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
0e51a8e1 | 321 | error = -EIO; |
787eb485 CH |
322 | goto done; |
323 | } | |
04f658ee | 324 | |
43caeb18 | 325 | /* |
787eb485 | 326 | * Clean up any COW blocks on an I/O error. |
43caeb18 | 327 | */ |
4e4cbee9 | 328 | error = blk_status_to_errno(ioend->io_bio->bi_status); |
787eb485 CH |
329 | if (unlikely(error)) { |
330 | switch (ioend->io_type) { | |
331 | case XFS_IO_COW: | |
332 | xfs_reflink_cancel_cow_range(ip, offset, size, true); | |
333 | break; | |
43caeb18 | 334 | } |
787eb485 CH |
335 | |
336 | goto done; | |
43caeb18 DW |
337 | } |
338 | ||
5ec4fabb | 339 | /* |
787eb485 | 340 | * Success: commit the COW or unwritten blocks if needed. |
5ec4fabb | 341 | */ |
787eb485 CH |
342 | switch (ioend->io_type) { |
343 | case XFS_IO_COW: | |
344 | error = xfs_reflink_end_cow(ip, offset, size); | |
345 | break; | |
346 | case XFS_IO_UNWRITTEN: | |
ee70daab EG |
347 | /* writeback should never update isize */ |
348 | error = xfs_iomap_write_unwritten(ip, offset, size, false); | |
787eb485 CH |
349 | break; |
350 | default: | |
351 | ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans); | |
352 | break; | |
5ec4fabb | 353 | } |
ba87ea69 | 354 | |
04f658ee | 355 | done: |
787eb485 CH |
356 | if (ioend->io_append_trans) |
357 | error = xfs_setfilesize_ioend(ioend, error); | |
0e51a8e1 | 358 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
359 | } |
360 | ||
0e51a8e1 CH |
361 | STATIC void |
362 | xfs_end_bio( | |
363 | struct bio *bio) | |
0829c360 | 364 | { |
0e51a8e1 CH |
365 | struct xfs_ioend *ioend = bio->bi_private; |
366 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 367 | |
43caeb18 | 368 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
0e51a8e1 CH |
369 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
370 | else if (ioend->io_append_trans) | |
371 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
372 | else | |
4e4cbee9 | 373 | xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status)); |
0829c360 CH |
374 | } |
375 | ||
1da177e4 LT |
376 | STATIC int |
377 | xfs_map_blocks( | |
378 | struct inode *inode, | |
379 | loff_t offset, | |
207d0416 | 380 | struct xfs_bmbt_irec *imap, |
988ef927 | 381 | int type) |
1da177e4 | 382 | { |
a206c817 CH |
383 | struct xfs_inode *ip = XFS_I(inode); |
384 | struct xfs_mount *mp = ip->i_mount; | |
93407472 | 385 | ssize_t count = i_blocksize(inode); |
a206c817 CH |
386 | xfs_fileoff_t offset_fsb, end_fsb; |
387 | int error = 0; | |
a206c817 CH |
388 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
389 | int nimaps = 1; | |
390 | ||
391 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 392 | return -EIO; |
a206c817 | 393 | |
70c57dcd DW |
394 | /* |
395 | * Truncate can race with writeback since writeback doesn't take the | |
396 | * iolock and truncate decreases the file size before it starts | |
397 | * truncating the pages between new_size and old_size. Therefore, we | |
398 | * can end up in the situation where writeback gets a CoW fork mapping | |
399 | * but the truncate makes the mapping invalid and we end up in here | |
400 | * trying to get a new mapping. Bail out here so that we simply never | |
401 | * get a valid mapping and so we drop the write altogether. The page | |
402 | * truncation will kill the contents anyway. | |
403 | */ | |
404 | if (type == XFS_IO_COW && offset > i_size_read(inode)) | |
405 | return 0; | |
406 | ||
ef473667 | 407 | ASSERT(type != XFS_IO_COW); |
0d882a36 | 408 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 409 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 410 | |
988ef927 | 411 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
412 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
413 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 414 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 415 | |
b4d8ad7f | 416 | if (offset > mp->m_super->s_maxbytes - count) |
d2c28191 | 417 | count = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
418 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
419 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
420 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
421 | imap, &nimaps, bmapi_flags); | |
ef473667 DW |
422 | /* |
423 | * Truncate an overwrite extent if there's a pending CoW | |
424 | * reservation before the end of this extent. This forces us | |
425 | * to come back to writepage to take care of the CoW. | |
426 | */ | |
427 | if (nimaps && type == XFS_IO_OVERWRITE) | |
428 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); | |
8ff2957d | 429 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 430 | |
8ff2957d | 431 | if (error) |
2451337d | 432 | return error; |
a206c817 | 433 | |
0d882a36 | 434 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 435 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
60b4984f DW |
436 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
437 | imap); | |
a206c817 | 438 | if (!error) |
ef473667 | 439 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
2451337d | 440 | return error; |
a206c817 CH |
441 | } |
442 | ||
8ff2957d | 443 | #ifdef DEBUG |
0d882a36 | 444 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
445 | ASSERT(nimaps); |
446 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
447 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
448 | } | |
449 | #endif | |
450 | if (nimaps) | |
451 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
452 | return 0; | |
1da177e4 LT |
453 | } |
454 | ||
fbcc0256 | 455 | STATIC bool |
558e6891 | 456 | xfs_imap_valid( |
8699bb0a | 457 | struct inode *inode, |
207d0416 | 458 | struct xfs_bmbt_irec *imap, |
558e6891 | 459 | xfs_off_t offset) |
1da177e4 | 460 | { |
558e6891 | 461 | offset >>= inode->i_blkbits; |
8699bb0a | 462 | |
40214d12 BF |
463 | /* |
464 | * We have to make sure the cached mapping is within EOF to protect | |
465 | * against eofblocks trimming on file release leaving us with a stale | |
466 | * mapping. Otherwise, a page for a subsequent file extending buffered | |
467 | * write could get picked up by this writeback cycle and written to the | |
468 | * wrong blocks. | |
469 | * | |
470 | * Note that what we really want here is a generic mapping invalidation | |
471 | * mechanism to protect us from arbitrary extent modifying contexts, not | |
472 | * just eofblocks. | |
473 | */ | |
474 | xfs_trim_extent_eof(imap, XFS_I(inode)); | |
475 | ||
558e6891 CH |
476 | return offset >= imap->br_startoff && |
477 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
478 | } |
479 | ||
f6d6d4fc CH |
480 | STATIC void |
481 | xfs_start_buffer_writeback( | |
482 | struct buffer_head *bh) | |
483 | { | |
484 | ASSERT(buffer_mapped(bh)); | |
485 | ASSERT(buffer_locked(bh)); | |
486 | ASSERT(!buffer_delay(bh)); | |
487 | ASSERT(!buffer_unwritten(bh)); | |
488 | ||
8353a814 CH |
489 | bh->b_end_io = NULL; |
490 | set_buffer_async_write(bh); | |
f6d6d4fc CH |
491 | set_buffer_uptodate(bh); |
492 | clear_buffer_dirty(bh); | |
493 | } | |
494 | ||
495 | STATIC void | |
496 | xfs_start_page_writeback( | |
497 | struct page *page, | |
e10de372 | 498 | int clear_dirty) |
f6d6d4fc CH |
499 | { |
500 | ASSERT(PageLocked(page)); | |
501 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
502 | |
503 | /* | |
504 | * if the page was not fully cleaned, we need to ensure that the higher | |
505 | * layers come back to it correctly. That means we need to keep the page | |
506 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
507 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
508 | * write this page in this writeback sweep will be made. | |
509 | */ | |
510 | if (clear_dirty) { | |
92132021 | 511 | clear_page_dirty_for_io(page); |
0d085a52 DC |
512 | set_page_writeback(page); |
513 | } else | |
514 | set_page_writeback_keepwrite(page); | |
515 | ||
f6d6d4fc | 516 | unlock_page(page); |
f6d6d4fc CH |
517 | } |
518 | ||
c7c1a7d8 | 519 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
520 | { |
521 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
522 | } | |
523 | ||
524 | /* | |
bb18782a DC |
525 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
526 | * it, and we submit that bio. The ioend may be used for multiple bio | |
527 | * submissions, so we only want to allocate an append transaction for the ioend | |
528 | * once. In the case of multiple bio submission, each bio will take an IO | |
529 | * reference to the ioend to ensure that the ioend completion is only done once | |
530 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
531 | * |
532 | * If @fail is non-zero, it means that we have a situation where some part of | |
533 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
534 | * and unlocked them. In this situation, we need to fail the bio and ioend |
535 | * rather than submit it to IO. This typically only happens on a filesystem | |
536 | * shutdown. | |
f6d6d4fc | 537 | */ |
e10de372 | 538 | STATIC int |
f6d6d4fc | 539 | xfs_submit_ioend( |
06342cf8 | 540 | struct writeback_control *wbc, |
0e51a8e1 | 541 | struct xfs_ioend *ioend, |
e10de372 | 542 | int status) |
f6d6d4fc | 543 | { |
5eda4300 DW |
544 | /* Convert CoW extents to regular */ |
545 | if (!status && ioend->io_type == XFS_IO_COW) { | |
546 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), | |
547 | ioend->io_offset, ioend->io_size); | |
548 | } | |
549 | ||
e10de372 DC |
550 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
551 | if (!status && | |
0e51a8e1 | 552 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
553 | xfs_ioend_is_append(ioend) && |
554 | !ioend->io_append_trans) | |
e10de372 | 555 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 556 | |
0e51a8e1 CH |
557 | ioend->io_bio->bi_private = ioend; |
558 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
7637241e | 559 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
70fd7614 | 560 | |
e10de372 DC |
561 | /* |
562 | * If we are failing the IO now, just mark the ioend with an | |
563 | * error and finish it. This will run IO completion immediately | |
564 | * as there is only one reference to the ioend at this point in | |
565 | * time. | |
566 | */ | |
567 | if (status) { | |
4e4cbee9 | 568 | ioend->io_bio->bi_status = errno_to_blk_status(status); |
0e51a8e1 | 569 | bio_endio(ioend->io_bio); |
e10de372 DC |
570 | return status; |
571 | } | |
d88992f6 | 572 | |
31d7d58d | 573 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 574 | submit_bio(ioend->io_bio); |
e10de372 | 575 | return 0; |
f6d6d4fc | 576 | } |
f6d6d4fc | 577 | |
0e51a8e1 CH |
578 | static void |
579 | xfs_init_bio_from_bh( | |
580 | struct bio *bio, | |
581 | struct buffer_head *bh) | |
582 | { | |
583 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
74d46992 | 584 | bio_set_dev(bio, bh->b_bdev); |
0e51a8e1 | 585 | } |
7bf7f352 | 586 | |
0e51a8e1 CH |
587 | static struct xfs_ioend * |
588 | xfs_alloc_ioend( | |
589 | struct inode *inode, | |
590 | unsigned int type, | |
591 | xfs_off_t offset, | |
592 | struct buffer_head *bh) | |
593 | { | |
594 | struct xfs_ioend *ioend; | |
595 | struct bio *bio; | |
f6d6d4fc | 596 | |
0e51a8e1 CH |
597 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
598 | xfs_init_bio_from_bh(bio, bh); | |
599 | ||
600 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
601 | INIT_LIST_HEAD(&ioend->io_list); | |
602 | ioend->io_type = type; | |
603 | ioend->io_inode = inode; | |
604 | ioend->io_size = 0; | |
605 | ioend->io_offset = offset; | |
606 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
607 | ioend->io_append_trans = NULL; | |
608 | ioend->io_bio = bio; | |
609 | return ioend; | |
610 | } | |
611 | ||
612 | /* | |
613 | * Allocate a new bio, and chain the old bio to the new one. | |
614 | * | |
615 | * Note that we have to do perform the chaining in this unintuitive order | |
616 | * so that the bi_private linkage is set up in the right direction for the | |
617 | * traversal in xfs_destroy_ioend(). | |
618 | */ | |
619 | static void | |
620 | xfs_chain_bio( | |
621 | struct xfs_ioend *ioend, | |
622 | struct writeback_control *wbc, | |
623 | struct buffer_head *bh) | |
624 | { | |
625 | struct bio *new; | |
626 | ||
627 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
628 | xfs_init_bio_from_bh(new, bh); | |
629 | ||
630 | bio_chain(ioend->io_bio, new); | |
631 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
7637241e | 632 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
31d7d58d | 633 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 634 | submit_bio(ioend->io_bio); |
0e51a8e1 | 635 | ioend->io_bio = new; |
f6d6d4fc CH |
636 | } |
637 | ||
638 | /* | |
639 | * Test to see if we've been building up a completion structure for | |
640 | * earlier buffers -- if so, we try to append to this ioend if we | |
641 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
642 | * Return the ioend we finished off so that the caller can submit it |
643 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
644 | */ |
645 | STATIC void | |
646 | xfs_add_to_ioend( | |
647 | struct inode *inode, | |
648 | struct buffer_head *bh, | |
7336cea8 | 649 | xfs_off_t offset, |
e10de372 | 650 | struct xfs_writepage_ctx *wpc, |
bb18782a | 651 | struct writeback_control *wbc, |
e10de372 | 652 | struct list_head *iolist) |
f6d6d4fc | 653 | { |
fbcc0256 | 654 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
655 | bh->b_blocknr != wpc->last_block + 1 || |
656 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
657 | if (wpc->ioend) |
658 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 659 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
660 | } |
661 | ||
0e51a8e1 CH |
662 | /* |
663 | * If the buffer doesn't fit into the bio we need to allocate a new | |
664 | * one. This shouldn't happen more than once for a given buffer. | |
665 | */ | |
666 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
667 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 668 | |
fbcc0256 DC |
669 | wpc->ioend->io_size += bh->b_size; |
670 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 671 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
672 | } |
673 | ||
87cbc49c NS |
674 | STATIC void |
675 | xfs_map_buffer( | |
046f1685 | 676 | struct inode *inode, |
87cbc49c | 677 | struct buffer_head *bh, |
207d0416 | 678 | struct xfs_bmbt_irec *imap, |
046f1685 | 679 | xfs_off_t offset) |
87cbc49c NS |
680 | { |
681 | sector_t bn; | |
8699bb0a | 682 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
683 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
684 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 685 | |
207d0416 CH |
686 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
687 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 688 | |
e513182d | 689 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 690 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 691 | |
046f1685 | 692 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
693 | |
694 | bh->b_blocknr = bn; | |
695 | set_buffer_mapped(bh); | |
696 | } | |
697 | ||
1da177e4 LT |
698 | STATIC void |
699 | xfs_map_at_offset( | |
046f1685 | 700 | struct inode *inode, |
1da177e4 | 701 | struct buffer_head *bh, |
207d0416 | 702 | struct xfs_bmbt_irec *imap, |
046f1685 | 703 | xfs_off_t offset) |
1da177e4 | 704 | { |
207d0416 CH |
705 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
706 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 707 | |
207d0416 | 708 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
709 | set_buffer_mapped(bh); |
710 | clear_buffer_delay(bh); | |
f6d6d4fc | 711 | clear_buffer_unwritten(bh); |
1da177e4 LT |
712 | } |
713 | ||
1da177e4 | 714 | /* |
a49935f2 DC |
715 | * Test if a given page contains at least one buffer of a given @type. |
716 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
717 | * try to find one of the type passed in. If it is not set, then the caller only | |
718 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 719 | */ |
a49935f2 | 720 | STATIC bool |
6ffc4db5 | 721 | xfs_check_page_type( |
10ce4444 | 722 | struct page *page, |
a49935f2 DC |
723 | unsigned int type, |
724 | bool check_all_buffers) | |
1da177e4 | 725 | { |
a49935f2 DC |
726 | struct buffer_head *bh; |
727 | struct buffer_head *head; | |
1da177e4 | 728 | |
a49935f2 DC |
729 | if (PageWriteback(page)) |
730 | return false; | |
731 | if (!page->mapping) | |
732 | return false; | |
733 | if (!page_has_buffers(page)) | |
734 | return false; | |
1da177e4 | 735 | |
a49935f2 DC |
736 | bh = head = page_buffers(page); |
737 | do { | |
738 | if (buffer_unwritten(bh)) { | |
739 | if (type == XFS_IO_UNWRITTEN) | |
740 | return true; | |
741 | } else if (buffer_delay(bh)) { | |
805eeb8e | 742 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
743 | return true; |
744 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 745 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
746 | return true; |
747 | } | |
1da177e4 | 748 | |
a49935f2 DC |
749 | /* If we are only checking the first buffer, we are done now. */ |
750 | if (!check_all_buffers) | |
751 | break; | |
752 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 753 | |
a49935f2 | 754 | return false; |
1da177e4 LT |
755 | } |
756 | ||
3ed3a434 DC |
757 | STATIC void |
758 | xfs_vm_invalidatepage( | |
759 | struct page *page, | |
d47992f8 LC |
760 | unsigned int offset, |
761 | unsigned int length) | |
3ed3a434 | 762 | { |
34097dfe LC |
763 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
764 | length); | |
793d7dbe DC |
765 | |
766 | /* | |
767 | * If we are invalidating the entire page, clear the dirty state from it | |
768 | * so that we can check for attempts to release dirty cached pages in | |
769 | * xfs_vm_releasepage(). | |
770 | */ | |
771 | if (offset == 0 && length >= PAGE_SIZE) | |
772 | cancel_dirty_page(page); | |
34097dfe | 773 | block_invalidatepage(page, offset, length); |
3ed3a434 DC |
774 | } |
775 | ||
776 | /* | |
777 | * If the page has delalloc buffers on it, we need to punch them out before we | |
778 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
779 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
780 | * is done on that same region - the delalloc extent is returned when none is | |
781 | * supposed to be there. | |
782 | * | |
783 | * We prevent this by truncating away the delalloc regions on the page before | |
784 | * invalidating it. Because they are delalloc, we can do this without needing a | |
785 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
786 | * truncation without a transaction as there is no space left for block | |
787 | * reservation (typically why we see a ENOSPC in writeback). | |
788 | * | |
789 | * This is not a performance critical path, so for now just do the punching a | |
790 | * buffer head at a time. | |
791 | */ | |
792 | STATIC void | |
793 | xfs_aops_discard_page( | |
794 | struct page *page) | |
795 | { | |
796 | struct inode *inode = page->mapping->host; | |
797 | struct xfs_inode *ip = XFS_I(inode); | |
798 | struct buffer_head *bh, *head; | |
799 | loff_t offset = page_offset(page); | |
3ed3a434 | 800 | |
a49935f2 | 801 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
802 | goto out_invalidate; |
803 | ||
e8c3753c DC |
804 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
805 | goto out_invalidate; | |
806 | ||
4f10700a | 807 | xfs_alert(ip->i_mount, |
c9690043 | 808 | "page discard on page "PTR_FMT", inode 0x%llx, offset %llu.", |
3ed3a434 DC |
809 | page, ip->i_ino, offset); |
810 | ||
811 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
812 | bh = head = page_buffers(page); | |
813 | do { | |
3ed3a434 | 814 | int error; |
c726de44 | 815 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
816 | |
817 | if (!buffer_delay(bh)) | |
818 | goto next_buffer; | |
819 | ||
c726de44 DC |
820 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
821 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
822 | if (error) { |
823 | /* something screwed, just bail */ | |
e8c3753c | 824 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 825 | xfs_alert(ip->i_mount, |
3ed3a434 | 826 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 827 | } |
3ed3a434 DC |
828 | break; |
829 | } | |
830 | next_buffer: | |
93407472 | 831 | offset += i_blocksize(inode); |
3ed3a434 DC |
832 | |
833 | } while ((bh = bh->b_this_page) != head); | |
834 | ||
835 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
836 | out_invalidate: | |
09cbfeaf | 837 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
838 | return; |
839 | } | |
840 | ||
ef473667 DW |
841 | static int |
842 | xfs_map_cow( | |
843 | struct xfs_writepage_ctx *wpc, | |
844 | struct inode *inode, | |
845 | loff_t offset, | |
846 | unsigned int *new_type) | |
847 | { | |
848 | struct xfs_inode *ip = XFS_I(inode); | |
849 | struct xfs_bmbt_irec imap; | |
092d5d9d | 850 | bool is_cow = false; |
ef473667 DW |
851 | int error; |
852 | ||
853 | /* | |
854 | * If we already have a valid COW mapping keep using it. | |
855 | */ | |
856 | if (wpc->io_type == XFS_IO_COW) { | |
857 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
858 | if (wpc->imap_valid) { | |
859 | *new_type = XFS_IO_COW; | |
860 | return 0; | |
861 | } | |
862 | } | |
863 | ||
864 | /* | |
865 | * Else we need to check if there is a COW mapping at this offset. | |
866 | */ | |
867 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
092d5d9d | 868 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap); |
ef473667 DW |
869 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
870 | ||
871 | if (!is_cow) | |
872 | return 0; | |
873 | ||
874 | /* | |
875 | * And if the COW mapping has a delayed extent here we need to | |
876 | * allocate real space for it now. | |
877 | */ | |
092d5d9d | 878 | if (isnullstartblock(imap.br_startblock)) { |
ef473667 DW |
879 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, |
880 | &imap); | |
881 | if (error) | |
882 | return error; | |
883 | } | |
884 | ||
885 | wpc->io_type = *new_type = XFS_IO_COW; | |
886 | wpc->imap_valid = true; | |
887 | wpc->imap = imap; | |
888 | return 0; | |
889 | } | |
890 | ||
e10de372 DC |
891 | /* |
892 | * We implement an immediate ioend submission policy here to avoid needing to | |
893 | * chain multiple ioends and hence nest mempool allocations which can violate | |
894 | * forward progress guarantees we need to provide. The current ioend we are | |
895 | * adding buffers to is cached on the writepage context, and if the new buffer | |
896 | * does not append to the cached ioend it will create a new ioend and cache that | |
897 | * instead. | |
898 | * | |
899 | * If a new ioend is created and cached, the old ioend is returned and queued | |
900 | * locally for submission once the entire page is processed or an error has been | |
901 | * detected. While ioends are submitted immediately after they are completed, | |
902 | * batching optimisations are provided by higher level block plugging. | |
903 | * | |
904 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
905 | * writepage context that the caller will need to submit. | |
906 | */ | |
bfce7d2e DC |
907 | static int |
908 | xfs_writepage_map( | |
909 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 910 | struct writeback_control *wbc, |
bfce7d2e DC |
911 | struct inode *inode, |
912 | struct page *page, | |
2d5f4b5b | 913 | uint64_t end_offset) |
bfce7d2e | 914 | { |
e10de372 DC |
915 | LIST_HEAD(submit_list); |
916 | struct xfs_ioend *ioend, *next; | |
bfce7d2e | 917 | struct buffer_head *bh, *head; |
93407472 | 918 | ssize_t len = i_blocksize(inode); |
2d5f4b5b | 919 | uint64_t offset; |
bfce7d2e | 920 | int error = 0; |
bfce7d2e | 921 | int count = 0; |
e10de372 | 922 | int uptodate = 1; |
ef473667 | 923 | unsigned int new_type; |
bfce7d2e DC |
924 | |
925 | bh = head = page_buffers(page); | |
926 | offset = page_offset(page); | |
bfce7d2e DC |
927 | do { |
928 | if (offset >= end_offset) | |
929 | break; | |
930 | if (!buffer_uptodate(bh)) | |
931 | uptodate = 0; | |
932 | ||
933 | /* | |
934 | * set_page_dirty dirties all buffers in a page, independent | |
935 | * of their state. The dirty state however is entirely | |
936 | * meaningless for holes (!mapped && uptodate), so skip | |
937 | * buffers covering holes here. | |
938 | */ | |
939 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
940 | wpc->imap_valid = false; | |
941 | continue; | |
942 | } | |
943 | ||
ef473667 DW |
944 | if (buffer_unwritten(bh)) |
945 | new_type = XFS_IO_UNWRITTEN; | |
946 | else if (buffer_delay(bh)) | |
947 | new_type = XFS_IO_DELALLOC; | |
948 | else if (buffer_uptodate(bh)) | |
949 | new_type = XFS_IO_OVERWRITE; | |
950 | else { | |
bfce7d2e DC |
951 | if (PageUptodate(page)) |
952 | ASSERT(buffer_mapped(bh)); | |
953 | /* | |
954 | * This buffer is not uptodate and will not be | |
955 | * written to disk. Ensure that we will put any | |
956 | * subsequent writeable buffers into a new | |
957 | * ioend. | |
958 | */ | |
959 | wpc->imap_valid = false; | |
960 | continue; | |
961 | } | |
962 | ||
ef473667 DW |
963 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
964 | error = xfs_map_cow(wpc, inode, offset, &new_type); | |
965 | if (error) | |
966 | goto out; | |
967 | } | |
968 | ||
969 | if (wpc->io_type != new_type) { | |
970 | wpc->io_type = new_type; | |
971 | wpc->imap_valid = false; | |
972 | } | |
973 | ||
bfce7d2e DC |
974 | if (wpc->imap_valid) |
975 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
976 | offset); | |
977 | if (!wpc->imap_valid) { | |
978 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
979 | wpc->io_type); | |
980 | if (error) | |
e10de372 | 981 | goto out; |
bfce7d2e DC |
982 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
983 | offset); | |
984 | } | |
985 | if (wpc->imap_valid) { | |
986 | lock_buffer(bh); | |
987 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
988 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 989 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
990 | count++; |
991 | } | |
992 | ||
bfce7d2e DC |
993 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
994 | ||
995 | if (uptodate && bh == head) | |
996 | SetPageUptodate(page); | |
997 | ||
e10de372 | 998 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 999 | |
e10de372 | 1000 | out: |
bfce7d2e | 1001 | /* |
e10de372 DC |
1002 | * On error, we have to fail the ioend here because we have locked |
1003 | * buffers in the ioend. If we don't do this, we'll deadlock | |
1004 | * invalidating the page as that tries to lock the buffers on the page. | |
1005 | * Also, because we may have set pages under writeback, we have to make | |
1006 | * sure we run IO completion to mark the error state of the IO | |
1007 | * appropriately, so we can't cancel the ioend directly here. That means | |
1008 | * we have to mark this page as under writeback if we included any | |
1009 | * buffers from it in the ioend chain so that completion treats it | |
1010 | * correctly. | |
bfce7d2e | 1011 | * |
e10de372 DC |
1012 | * If we didn't include the page in the ioend, the on error we can |
1013 | * simply discard and unlock it as there are no other users of the page | |
1014 | * or it's buffers right now. The caller will still need to trigger | |
1015 | * submission of outstanding ioends on the writepage context so they are | |
1016 | * treated correctly on error. | |
bfce7d2e | 1017 | */ |
e10de372 DC |
1018 | if (count) { |
1019 | xfs_start_page_writeback(page, !error); | |
1020 | ||
1021 | /* | |
1022 | * Preserve the original error if there was one, otherwise catch | |
1023 | * submission errors here and propagate into subsequent ioend | |
1024 | * submissions. | |
1025 | */ | |
1026 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
1027 | int error2; | |
1028 | ||
1029 | list_del_init(&ioend->io_list); | |
1030 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
1031 | if (error2 && !error) | |
1032 | error = error2; | |
1033 | } | |
1034 | } else if (error) { | |
bfce7d2e DC |
1035 | xfs_aops_discard_page(page); |
1036 | ClearPageUptodate(page); | |
1037 | unlock_page(page); | |
e10de372 DC |
1038 | } else { |
1039 | /* | |
1040 | * We can end up here with no error and nothing to write if we | |
1041 | * race with a partial page truncate on a sub-page block sized | |
1042 | * filesystem. In that case we need to mark the page clean. | |
1043 | */ | |
1044 | xfs_start_page_writeback(page, 1); | |
1045 | end_page_writeback(page); | |
bfce7d2e | 1046 | } |
e10de372 | 1047 | |
bfce7d2e DC |
1048 | mapping_set_error(page->mapping, error); |
1049 | return error; | |
1050 | } | |
1051 | ||
1da177e4 | 1052 | /* |
89f3b363 CH |
1053 | * Write out a dirty page. |
1054 | * | |
1055 | * For delalloc space on the page we need to allocate space and flush it. | |
1056 | * For unwritten space on the page we need to start the conversion to | |
1057 | * regular allocated space. | |
89f3b363 | 1058 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1059 | */ |
1da177e4 | 1060 | STATIC int |
fbcc0256 | 1061 | xfs_do_writepage( |
89f3b363 | 1062 | struct page *page, |
fbcc0256 DC |
1063 | struct writeback_control *wbc, |
1064 | void *data) | |
1da177e4 | 1065 | { |
fbcc0256 | 1066 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 1067 | struct inode *inode = page->mapping->host; |
1da177e4 | 1068 | loff_t offset; |
c8ce540d | 1069 | uint64_t end_offset; |
ad68972a | 1070 | pgoff_t end_index; |
89f3b363 | 1071 | |
34097dfe | 1072 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 1073 | |
20cb52eb CH |
1074 | ASSERT(page_has_buffers(page)); |
1075 | ||
89f3b363 CH |
1076 | /* |
1077 | * Refuse to write the page out if we are called from reclaim context. | |
1078 | * | |
d4f7a5cb CH |
1079 | * This avoids stack overflows when called from deeply used stacks in |
1080 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1081 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 1082 | * |
94054fa3 MG |
1083 | * This should never happen except in the case of a VM regression so |
1084 | * warn about it. | |
89f3b363 | 1085 | */ |
94054fa3 MG |
1086 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
1087 | PF_MEMALLOC)) | |
b5420f23 | 1088 | goto redirty; |
1da177e4 | 1089 | |
89f3b363 | 1090 | /* |
680a647b CH |
1091 | * Given that we do not allow direct reclaim to call us, we should |
1092 | * never be called while in a filesystem transaction. | |
89f3b363 | 1093 | */ |
9070733b | 1094 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) |
b5420f23 | 1095 | goto redirty; |
89f3b363 | 1096 | |
8695d27e | 1097 | /* |
ad68972a DC |
1098 | * Is this page beyond the end of the file? |
1099 | * | |
8695d27e JL |
1100 | * The page index is less than the end_index, adjust the end_offset |
1101 | * to the highest offset that this page should represent. | |
1102 | * ----------------------------------------------------- | |
1103 | * | file mapping | <EOF> | | |
1104 | * ----------------------------------------------------- | |
1105 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1106 | * ^--------------------------------^----------|-------- | |
1107 | * | desired writeback range | see else | | |
1108 | * ---------------------------------^------------------| | |
1109 | */ | |
ad68972a | 1110 | offset = i_size_read(inode); |
09cbfeaf | 1111 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 1112 | if (page->index < end_index) |
09cbfeaf | 1113 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
1114 | else { |
1115 | /* | |
1116 | * Check whether the page to write out is beyond or straddles | |
1117 | * i_size or not. | |
1118 | * ------------------------------------------------------- | |
1119 | * | file mapping | <EOF> | | |
1120 | * ------------------------------------------------------- | |
1121 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1122 | * ^--------------------------------^-----------|--------- | |
1123 | * | | Straddles | | |
1124 | * ---------------------------------^-----------|--------| | |
1125 | */ | |
09cbfeaf | 1126 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
1127 | |
1128 | /* | |
ff9a28f6 JK |
1129 | * Skip the page if it is fully outside i_size, e.g. due to a |
1130 | * truncate operation that is in progress. We must redirty the | |
1131 | * page so that reclaim stops reclaiming it. Otherwise | |
1132 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1133 | * |
1134 | * Note that the end_index is unsigned long, it would overflow | |
1135 | * if the given offset is greater than 16TB on 32-bit system | |
1136 | * and if we do check the page is fully outside i_size or not | |
1137 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1138 | * will be evaluated to 0. Hence this page will be redirtied | |
1139 | * and be written out repeatedly which would result in an | |
1140 | * infinite loop, the user program that perform this operation | |
1141 | * will hang. Instead, we can verify this situation by checking | |
1142 | * if the page to write is totally beyond the i_size or if it's | |
1143 | * offset is just equal to the EOF. | |
6b7a03f0 | 1144 | */ |
8695d27e JL |
1145 | if (page->index > end_index || |
1146 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1147 | goto redirty; |
6b7a03f0 CH |
1148 | |
1149 | /* | |
1150 | * The page straddles i_size. It must be zeroed out on each | |
1151 | * and every writepage invocation because it may be mmapped. | |
1152 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1153 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1154 | * memory is zeroed when mapped, and writes to that region are |
1155 | * not written out to the file." | |
1156 | */ | |
09cbfeaf | 1157 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
1158 | |
1159 | /* Adjust the end_offset to the end of file */ | |
1160 | end_offset = offset; | |
1da177e4 LT |
1161 | } |
1162 | ||
2d5f4b5b | 1163 | return xfs_writepage_map(wpc, wbc, inode, page, end_offset); |
f51623b2 | 1164 | |
b5420f23 | 1165 | redirty: |
f51623b2 NS |
1166 | redirty_page_for_writepage(wbc, page); |
1167 | unlock_page(page); | |
1168 | return 0; | |
f51623b2 NS |
1169 | } |
1170 | ||
fbcc0256 DC |
1171 | STATIC int |
1172 | xfs_vm_writepage( | |
1173 | struct page *page, | |
1174 | struct writeback_control *wbc) | |
1175 | { | |
1176 | struct xfs_writepage_ctx wpc = { | |
1177 | .io_type = XFS_IO_INVALID, | |
1178 | }; | |
1179 | int ret; | |
1180 | ||
1181 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1182 | if (wpc.ioend) |
1183 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1184 | return ret; | |
fbcc0256 DC |
1185 | } |
1186 | ||
7d4fb40a NS |
1187 | STATIC int |
1188 | xfs_vm_writepages( | |
1189 | struct address_space *mapping, | |
1190 | struct writeback_control *wbc) | |
1191 | { | |
fbcc0256 DC |
1192 | struct xfs_writepage_ctx wpc = { |
1193 | .io_type = XFS_IO_INVALID, | |
1194 | }; | |
1195 | int ret; | |
1196 | ||
b3aea4ed | 1197 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
fbcc0256 | 1198 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1199 | if (wpc.ioend) |
1200 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1201 | return ret; | |
7d4fb40a NS |
1202 | } |
1203 | ||
6e2608df DW |
1204 | STATIC int |
1205 | xfs_dax_writepages( | |
1206 | struct address_space *mapping, | |
1207 | struct writeback_control *wbc) | |
1208 | { | |
1209 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); | |
1210 | return dax_writeback_mapping_range(mapping, | |
1211 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1212 | } | |
1213 | ||
f51623b2 NS |
1214 | /* |
1215 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1216 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1217 | * have buffer heads in this call. |
1218 | * | |
89f3b363 | 1219 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1220 | */ |
1221 | STATIC int | |
238f4c54 | 1222 | xfs_vm_releasepage( |
f51623b2 NS |
1223 | struct page *page, |
1224 | gfp_t gfp_mask) | |
1225 | { | |
20cb52eb | 1226 | int delalloc, unwritten; |
f51623b2 | 1227 | |
34097dfe | 1228 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1229 | |
99579cce BF |
1230 | /* |
1231 | * mm accommodates an old ext3 case where clean pages might not have had | |
1232 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1233 | * ->releasepage() via shrink_active_list(). Conversely, | |
793d7dbe DC |
1234 | * block_invalidatepage() can send pages that are still marked dirty but |
1235 | * otherwise have invalidated buffers. | |
99579cce | 1236 | * |
0a417b8d | 1237 | * We want to release the latter to avoid unnecessary buildup of the |
793d7dbe DC |
1238 | * LRU, so xfs_vm_invalidatepage() clears the page dirty flag on pages |
1239 | * that are entirely invalidated and need to be released. Hence the | |
1240 | * only time we should get dirty pages here is through | |
1241 | * shrink_active_list() and so we can simply skip those now. | |
1242 | * | |
1243 | * warn if we've left any lingering delalloc/unwritten buffers on clean | |
1244 | * or invalidated pages we are about to release. | |
99579cce | 1245 | */ |
793d7dbe DC |
1246 | if (PageDirty(page)) |
1247 | return 0; | |
1248 | ||
20cb52eb | 1249 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1250 | |
793d7dbe | 1251 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1252 | return 0; |
793d7dbe | 1253 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1254 | return 0; |
1255 | ||
f51623b2 NS |
1256 | return try_to_free_buffers(page); |
1257 | } | |
1258 | ||
1fdca9c2 DC |
1259 | /* |
1260 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1261 | * is, so that we can avoid repeated get_blocks calls. | |
1262 | * | |
1263 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1264 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1265 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1266 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1267 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1268 | * to and including EOF, and a second mapping for beyond EOF. | |
1269 | */ | |
1270 | static void | |
1271 | xfs_map_trim_size( | |
1272 | struct inode *inode, | |
1273 | sector_t iblock, | |
1274 | struct buffer_head *bh_result, | |
1275 | struct xfs_bmbt_irec *imap, | |
1276 | xfs_off_t offset, | |
1277 | ssize_t size) | |
1278 | { | |
1279 | xfs_off_t mapping_size; | |
1280 | ||
1281 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1282 | mapping_size <<= inode->i_blkbits; | |
1283 | ||
1284 | ASSERT(mapping_size > 0); | |
1285 | if (mapping_size > size) | |
1286 | mapping_size = size; | |
1287 | if (offset < i_size_read(inode) && | |
22a6c837 | 1288 | (xfs_ufsize_t)offset + mapping_size >= i_size_read(inode)) { |
1fdca9c2 DC |
1289 | /* limit mapping to block that spans EOF */ |
1290 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
93407472 | 1291 | i_blocksize(inode)); |
1fdca9c2 DC |
1292 | } |
1293 | if (mapping_size > LONG_MAX) | |
1294 | mapping_size = LONG_MAX; | |
1295 | ||
1296 | bh_result->b_size = mapping_size; | |
1297 | } | |
1298 | ||
0613f16c | 1299 | static int |
acdda3aa | 1300 | xfs_get_blocks( |
1da177e4 LT |
1301 | struct inode *inode, |
1302 | sector_t iblock, | |
1da177e4 | 1303 | struct buffer_head *bh_result, |
acdda3aa | 1304 | int create) |
1da177e4 | 1305 | { |
a206c817 CH |
1306 | struct xfs_inode *ip = XFS_I(inode); |
1307 | struct xfs_mount *mp = ip->i_mount; | |
1308 | xfs_fileoff_t offset_fsb, end_fsb; | |
1309 | int error = 0; | |
1310 | int lockmode = 0; | |
207d0416 | 1311 | struct xfs_bmbt_irec imap; |
a206c817 | 1312 | int nimaps = 1; |
fdc7ed75 NS |
1313 | xfs_off_t offset; |
1314 | ssize_t size; | |
a206c817 | 1315 | |
acdda3aa | 1316 | BUG_ON(create); |
6e8a27a8 | 1317 | |
a206c817 | 1318 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1319 | return -EIO; |
1da177e4 | 1320 | |
fdc7ed75 | 1321 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
93407472 | 1322 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
c2536668 | 1323 | size = bh_result->b_size; |
364f358a | 1324 | |
acdda3aa | 1325 | if (offset >= i_size_read(inode)) |
364f358a LM |
1326 | return 0; |
1327 | ||
507630b2 DC |
1328 | /* |
1329 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1330 | * a block mapping without an exclusive lock first. |
507630b2 | 1331 | */ |
6e8a27a8 | 1332 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1333 | |
d2c28191 | 1334 | ASSERT(offset <= mp->m_super->s_maxbytes); |
b4d8ad7f | 1335 | if (offset > mp->m_super->s_maxbytes - size) |
d2c28191 | 1336 | size = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
1337 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1338 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1339 | ||
7d9df3c1 CH |
1340 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, |
1341 | &nimaps, 0); | |
1da177e4 | 1342 | if (error) |
a206c817 | 1343 | goto out_unlock; |
1d4352de | 1344 | if (!nimaps) { |
a206c817 CH |
1345 | trace_xfs_get_blocks_notfound(ip, offset, size); |
1346 | goto out_unlock; | |
1347 | } | |
1da177e4 | 1348 | |
1d4352de CH |
1349 | trace_xfs_get_blocks_found(ip, offset, size, |
1350 | imap.br_state == XFS_EXT_UNWRITTEN ? | |
1351 | XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap); | |
1352 | xfs_iunlock(ip, lockmode); | |
1353 | ||
1fdca9c2 | 1354 | /* trim mapping down to size requested */ |
6e8a27a8 | 1355 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1356 | |
a719370b DC |
1357 | /* |
1358 | * For unwritten extents do not report a disk address in the buffered | |
1359 | * read case (treat as if we're reading into a hole). | |
1360 | */ | |
9c4f29d3 | 1361 | if (xfs_bmap_is_real_extent(&imap)) |
a719370b | 1362 | xfs_map_buffer(inode, bh_result, &imap, offset); |
1da177e4 | 1363 | |
c2536668 NS |
1364 | /* |
1365 | * If this is a realtime file, data may be on a different device. | |
1366 | * to that pointed to from the buffer_head b_bdev currently. | |
1367 | */ | |
046f1685 | 1368 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1369 | return 0; |
a206c817 CH |
1370 | |
1371 | out_unlock: | |
1372 | xfs_iunlock(ip, lockmode); | |
2451337d | 1373 | return error; |
1da177e4 LT |
1374 | } |
1375 | ||
1da177e4 | 1376 | STATIC sector_t |
e4c573bb | 1377 | xfs_vm_bmap( |
1da177e4 LT |
1378 | struct address_space *mapping, |
1379 | sector_t block) | |
1380 | { | |
1381 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1382 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1383 | |
cca28fb8 | 1384 | trace_xfs_vm_bmap(XFS_I(inode)); |
db1327b1 DW |
1385 | |
1386 | /* | |
1387 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
793057e1 | 1388 | * bypasses the file system for actual I/O. We really can't allow |
db1327b1 | 1389 | * that on reflinks inodes, so we have to skip out here. And yes, |
eb5e248d DW |
1390 | * 0 is the magic code for a bmap error. |
1391 | * | |
1392 | * Since we don't pass back blockdev info, we can't return bmap | |
1393 | * information for rt files either. | |
db1327b1 | 1394 | */ |
eb5e248d | 1395 | if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) |
db1327b1 | 1396 | return 0; |
65523218 | 1397 | |
4bc1ea6b | 1398 | filemap_write_and_wait(mapping); |
c2536668 | 1399 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1400 | } |
1401 | ||
1402 | STATIC int | |
e4c573bb | 1403 | xfs_vm_readpage( |
1da177e4 LT |
1404 | struct file *unused, |
1405 | struct page *page) | |
1406 | { | |
121e213e | 1407 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1408 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | STATIC int | |
e4c573bb | 1412 | xfs_vm_readpages( |
1da177e4 LT |
1413 | struct file *unused, |
1414 | struct address_space *mapping, | |
1415 | struct list_head *pages, | |
1416 | unsigned nr_pages) | |
1417 | { | |
121e213e | 1418 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1419 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1420 | } |
1421 | ||
22e757a4 DC |
1422 | /* |
1423 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1424 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1425 | * dirty, we'll never be able to clean them because we don't write buffers | |
1426 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1427 | * that have been marked dirty. Further, the dirty state can leak into | |
1428 | * the file interior if the file is extended, resulting in all sorts of | |
1429 | * bad things happening as the state does not match the underlying data. | |
1430 | * | |
1431 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1432 | * this only exist because of bufferheads and how the generic code manages them. | |
1433 | */ | |
1434 | STATIC int | |
1435 | xfs_vm_set_page_dirty( | |
1436 | struct page *page) | |
1437 | { | |
1438 | struct address_space *mapping = page->mapping; | |
1439 | struct inode *inode = mapping->host; | |
1440 | loff_t end_offset; | |
1441 | loff_t offset; | |
1442 | int newly_dirty; | |
1443 | ||
1444 | if (unlikely(!mapping)) | |
1445 | return !TestSetPageDirty(page); | |
1446 | ||
1447 | end_offset = i_size_read(inode); | |
1448 | offset = page_offset(page); | |
1449 | ||
1450 | spin_lock(&mapping->private_lock); | |
1451 | if (page_has_buffers(page)) { | |
1452 | struct buffer_head *head = page_buffers(page); | |
1453 | struct buffer_head *bh = head; | |
1454 | ||
1455 | do { | |
1456 | if (offset < end_offset) | |
1457 | set_buffer_dirty(bh); | |
1458 | bh = bh->b_this_page; | |
93407472 | 1459 | offset += i_blocksize(inode); |
22e757a4 DC |
1460 | } while (bh != head); |
1461 | } | |
c4843a75 | 1462 | /* |
81f8c3a4 JW |
1463 | * Lock out page->mem_cgroup migration to keep PageDirty |
1464 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1465 | */ |
62cccb8c | 1466 | lock_page_memcg(page); |
22e757a4 DC |
1467 | newly_dirty = !TestSetPageDirty(page); |
1468 | spin_unlock(&mapping->private_lock); | |
1469 | ||
f82b3764 MW |
1470 | if (newly_dirty) |
1471 | __set_page_dirty(page, mapping, 1); | |
62cccb8c | 1472 | unlock_page_memcg(page); |
c4843a75 GT |
1473 | if (newly_dirty) |
1474 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1475 | return newly_dirty; |
1476 | } | |
1477 | ||
f5e54d6e | 1478 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1479 | .readpage = xfs_vm_readpage, |
1480 | .readpages = xfs_vm_readpages, | |
1481 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1482 | .writepages = xfs_vm_writepages, |
22e757a4 | 1483 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1484 | .releasepage = xfs_vm_releasepage, |
1485 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb | 1486 | .bmap = xfs_vm_bmap, |
6e2608df | 1487 | .direct_IO = noop_direct_IO, |
e965f963 | 1488 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1489 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1490 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1491 | }; |
6e2608df DW |
1492 | |
1493 | const struct address_space_operations xfs_dax_aops = { | |
1494 | .writepages = xfs_dax_writepages, | |
1495 | .direct_IO = noop_direct_IO, | |
1496 | .set_page_dirty = noop_set_page_dirty, | |
1497 | .invalidatepage = noop_invalidatepage, | |
1498 | }; |