]>
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 | ||
253f4911 CH |
212 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
213 | if (error) | |
281627df | 214 | return error; |
281627df CH |
215 | |
216 | ioend->io_append_trans = tp; | |
217 | ||
d9457dc0 | 218 | /* |
437a255a | 219 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
220 | * we released it. |
221 | */ | |
bee9182d | 222 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
223 | /* |
224 | * We hand off the transaction to the completion thread now, so | |
225 | * clear the flag here. | |
226 | */ | |
9070733b | 227 | current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
281627df CH |
228 | return 0; |
229 | } | |
230 | ||
ba87ea69 | 231 | /* |
2813d682 | 232 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 233 | */ |
281627df | 234 | STATIC int |
e372843a | 235 | __xfs_setfilesize( |
2ba66237 CH |
236 | struct xfs_inode *ip, |
237 | struct xfs_trans *tp, | |
238 | xfs_off_t offset, | |
239 | size_t size) | |
ba87ea69 | 240 | { |
ba87ea69 | 241 | xfs_fsize_t isize; |
ba87ea69 | 242 | |
aa6bf01d | 243 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 244 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
245 | if (!isize) { |
246 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 247 | xfs_trans_cancel(tp); |
281627df | 248 | return 0; |
ba87ea69 LM |
249 | } |
250 | ||
2ba66237 | 251 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
252 | |
253 | ip->i_d.di_size = isize; | |
254 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
255 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
256 | ||
70393313 | 257 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
258 | } |
259 | ||
e372843a CH |
260 | int |
261 | xfs_setfilesize( | |
262 | struct xfs_inode *ip, | |
263 | xfs_off_t offset, | |
264 | size_t size) | |
265 | { | |
266 | struct xfs_mount *mp = ip->i_mount; | |
267 | struct xfs_trans *tp; | |
268 | int error; | |
269 | ||
270 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); | |
271 | if (error) | |
272 | return error; | |
273 | ||
274 | return __xfs_setfilesize(ip, tp, offset, size); | |
275 | } | |
276 | ||
2ba66237 CH |
277 | STATIC int |
278 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
279 | struct xfs_ioend *ioend, |
280 | int error) | |
2ba66237 CH |
281 | { |
282 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
283 | struct xfs_trans *tp = ioend->io_append_trans; | |
284 | ||
285 | /* | |
286 | * The transaction may have been allocated in the I/O submission thread, | |
287 | * thus we need to mark ourselves as being in a transaction manually. | |
288 | * Similarly for freeze protection. | |
289 | */ | |
9070733b | 290 | current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
bee9182d | 291 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 292 | |
5cb13dcd | 293 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 294 | if (error) { |
5cb13dcd | 295 | xfs_trans_cancel(tp); |
0e51a8e1 | 296 | return error; |
5cb13dcd Z |
297 | } |
298 | ||
e372843a | 299 | return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
2ba66237 CH |
300 | } |
301 | ||
0829c360 | 302 | /* |
5ec4fabb | 303 | * IO write completion. |
f6d6d4fc CH |
304 | */ |
305 | STATIC void | |
5ec4fabb | 306 | xfs_end_io( |
77d7a0c2 | 307 | struct work_struct *work) |
0829c360 | 308 | { |
0e51a8e1 CH |
309 | struct xfs_ioend *ioend = |
310 | container_of(work, struct xfs_ioend, io_work); | |
311 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
787eb485 CH |
312 | xfs_off_t offset = ioend->io_offset; |
313 | size_t size = ioend->io_size; | |
4e4cbee9 | 314 | int error; |
ba87ea69 | 315 | |
af055e37 | 316 | /* |
787eb485 | 317 | * Just clean up the in-memory strutures if the fs has been shut down. |
af055e37 | 318 | */ |
787eb485 | 319 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
0e51a8e1 | 320 | error = -EIO; |
787eb485 CH |
321 | goto done; |
322 | } | |
04f658ee | 323 | |
43caeb18 | 324 | /* |
787eb485 | 325 | * Clean up any COW blocks on an I/O error. |
43caeb18 | 326 | */ |
4e4cbee9 | 327 | error = blk_status_to_errno(ioend->io_bio->bi_status); |
787eb485 CH |
328 | if (unlikely(error)) { |
329 | switch (ioend->io_type) { | |
330 | case XFS_IO_COW: | |
331 | xfs_reflink_cancel_cow_range(ip, offset, size, true); | |
332 | break; | |
43caeb18 | 333 | } |
787eb485 CH |
334 | |
335 | goto done; | |
43caeb18 DW |
336 | } |
337 | ||
5ec4fabb | 338 | /* |
787eb485 | 339 | * Success: commit the COW or unwritten blocks if needed. |
5ec4fabb | 340 | */ |
787eb485 CH |
341 | switch (ioend->io_type) { |
342 | case XFS_IO_COW: | |
343 | error = xfs_reflink_end_cow(ip, offset, size); | |
344 | break; | |
345 | case XFS_IO_UNWRITTEN: | |
ee70daab EG |
346 | /* writeback should never update isize */ |
347 | error = xfs_iomap_write_unwritten(ip, offset, size, false); | |
787eb485 CH |
348 | break; |
349 | default: | |
350 | ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans); | |
351 | break; | |
5ec4fabb | 352 | } |
ba87ea69 | 353 | |
04f658ee | 354 | done: |
787eb485 CH |
355 | if (ioend->io_append_trans) |
356 | error = xfs_setfilesize_ioend(ioend, error); | |
0e51a8e1 | 357 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
358 | } |
359 | ||
0e51a8e1 CH |
360 | STATIC void |
361 | xfs_end_bio( | |
362 | struct bio *bio) | |
0829c360 | 363 | { |
0e51a8e1 CH |
364 | struct xfs_ioend *ioend = bio->bi_private; |
365 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 366 | |
43caeb18 | 367 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
0e51a8e1 CH |
368 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
369 | else if (ioend->io_append_trans) | |
370 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
371 | else | |
4e4cbee9 | 372 | xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status)); |
0829c360 CH |
373 | } |
374 | ||
1da177e4 LT |
375 | STATIC int |
376 | xfs_map_blocks( | |
377 | struct inode *inode, | |
378 | loff_t offset, | |
207d0416 | 379 | struct xfs_bmbt_irec *imap, |
988ef927 | 380 | int type) |
1da177e4 | 381 | { |
a206c817 CH |
382 | struct xfs_inode *ip = XFS_I(inode); |
383 | struct xfs_mount *mp = ip->i_mount; | |
93407472 | 384 | ssize_t count = i_blocksize(inode); |
a206c817 CH |
385 | xfs_fileoff_t offset_fsb, end_fsb; |
386 | int error = 0; | |
a206c817 CH |
387 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
388 | int nimaps = 1; | |
389 | ||
390 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 391 | return -EIO; |
a206c817 | 392 | |
ef473667 | 393 | ASSERT(type != XFS_IO_COW); |
0d882a36 | 394 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 395 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 396 | |
988ef927 | 397 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
398 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
399 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 400 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 401 | |
b4d8ad7f | 402 | if (offset > mp->m_super->s_maxbytes - count) |
d2c28191 | 403 | count = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
404 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
405 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
406 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
407 | imap, &nimaps, bmapi_flags); | |
ef473667 DW |
408 | /* |
409 | * Truncate an overwrite extent if there's a pending CoW | |
410 | * reservation before the end of this extent. This forces us | |
411 | * to come back to writepage to take care of the CoW. | |
412 | */ | |
413 | if (nimaps && type == XFS_IO_OVERWRITE) | |
414 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); | |
8ff2957d | 415 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 416 | |
8ff2957d | 417 | if (error) |
2451337d | 418 | return error; |
a206c817 | 419 | |
0d882a36 | 420 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 421 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
60b4984f DW |
422 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
423 | imap); | |
a206c817 | 424 | if (!error) |
ef473667 | 425 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
2451337d | 426 | return error; |
a206c817 CH |
427 | } |
428 | ||
8ff2957d | 429 | #ifdef DEBUG |
0d882a36 | 430 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
431 | ASSERT(nimaps); |
432 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
433 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
434 | } | |
435 | #endif | |
436 | if (nimaps) | |
437 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
438 | return 0; | |
1da177e4 LT |
439 | } |
440 | ||
fbcc0256 | 441 | STATIC bool |
558e6891 | 442 | xfs_imap_valid( |
8699bb0a | 443 | struct inode *inode, |
207d0416 | 444 | struct xfs_bmbt_irec *imap, |
558e6891 | 445 | xfs_off_t offset) |
1da177e4 | 446 | { |
558e6891 | 447 | offset >>= inode->i_blkbits; |
8699bb0a | 448 | |
40214d12 BF |
449 | /* |
450 | * We have to make sure the cached mapping is within EOF to protect | |
451 | * against eofblocks trimming on file release leaving us with a stale | |
452 | * mapping. Otherwise, a page for a subsequent file extending buffered | |
453 | * write could get picked up by this writeback cycle and written to the | |
454 | * wrong blocks. | |
455 | * | |
456 | * Note that what we really want here is a generic mapping invalidation | |
457 | * mechanism to protect us from arbitrary extent modifying contexts, not | |
458 | * just eofblocks. | |
459 | */ | |
460 | xfs_trim_extent_eof(imap, XFS_I(inode)); | |
461 | ||
558e6891 CH |
462 | return offset >= imap->br_startoff && |
463 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
464 | } |
465 | ||
f6d6d4fc CH |
466 | STATIC void |
467 | xfs_start_buffer_writeback( | |
468 | struct buffer_head *bh) | |
469 | { | |
470 | ASSERT(buffer_mapped(bh)); | |
471 | ASSERT(buffer_locked(bh)); | |
472 | ASSERT(!buffer_delay(bh)); | |
473 | ASSERT(!buffer_unwritten(bh)); | |
474 | ||
8353a814 CH |
475 | bh->b_end_io = NULL; |
476 | set_buffer_async_write(bh); | |
f6d6d4fc CH |
477 | set_buffer_uptodate(bh); |
478 | clear_buffer_dirty(bh); | |
479 | } | |
480 | ||
481 | STATIC void | |
482 | xfs_start_page_writeback( | |
483 | struct page *page, | |
e10de372 | 484 | int clear_dirty) |
f6d6d4fc CH |
485 | { |
486 | ASSERT(PageLocked(page)); | |
487 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
488 | |
489 | /* | |
490 | * if the page was not fully cleaned, we need to ensure that the higher | |
491 | * layers come back to it correctly. That means we need to keep the page | |
492 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
493 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
494 | * write this page in this writeback sweep will be made. | |
495 | */ | |
496 | if (clear_dirty) { | |
92132021 | 497 | clear_page_dirty_for_io(page); |
0d085a52 DC |
498 | set_page_writeback(page); |
499 | } else | |
500 | set_page_writeback_keepwrite(page); | |
501 | ||
f6d6d4fc | 502 | unlock_page(page); |
f6d6d4fc CH |
503 | } |
504 | ||
c7c1a7d8 | 505 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
506 | { |
507 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
508 | } | |
509 | ||
510 | /* | |
bb18782a DC |
511 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
512 | * it, and we submit that bio. The ioend may be used for multiple bio | |
513 | * submissions, so we only want to allocate an append transaction for the ioend | |
514 | * once. In the case of multiple bio submission, each bio will take an IO | |
515 | * reference to the ioend to ensure that the ioend completion is only done once | |
516 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
517 | * |
518 | * If @fail is non-zero, it means that we have a situation where some part of | |
519 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
520 | * and unlocked them. In this situation, we need to fail the bio and ioend |
521 | * rather than submit it to IO. This typically only happens on a filesystem | |
522 | * shutdown. | |
f6d6d4fc | 523 | */ |
e10de372 | 524 | STATIC int |
f6d6d4fc | 525 | xfs_submit_ioend( |
06342cf8 | 526 | struct writeback_control *wbc, |
0e51a8e1 | 527 | struct xfs_ioend *ioend, |
e10de372 | 528 | int status) |
f6d6d4fc | 529 | { |
5eda4300 DW |
530 | /* Convert CoW extents to regular */ |
531 | if (!status && ioend->io_type == XFS_IO_COW) { | |
532 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), | |
533 | ioend->io_offset, ioend->io_size); | |
534 | } | |
535 | ||
e10de372 DC |
536 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
537 | if (!status && | |
0e51a8e1 | 538 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
539 | xfs_ioend_is_append(ioend) && |
540 | !ioend->io_append_trans) | |
e10de372 | 541 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 542 | |
0e51a8e1 CH |
543 | ioend->io_bio->bi_private = ioend; |
544 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
7637241e | 545 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
70fd7614 | 546 | |
e10de372 DC |
547 | /* |
548 | * If we are failing the IO now, just mark the ioend with an | |
549 | * error and finish it. This will run IO completion immediately | |
550 | * as there is only one reference to the ioend at this point in | |
551 | * time. | |
552 | */ | |
553 | if (status) { | |
4e4cbee9 | 554 | ioend->io_bio->bi_status = errno_to_blk_status(status); |
0e51a8e1 | 555 | bio_endio(ioend->io_bio); |
e10de372 DC |
556 | return status; |
557 | } | |
d88992f6 | 558 | |
31d7d58d | 559 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 560 | submit_bio(ioend->io_bio); |
e10de372 | 561 | return 0; |
f6d6d4fc | 562 | } |
f6d6d4fc | 563 | |
0e51a8e1 CH |
564 | static void |
565 | xfs_init_bio_from_bh( | |
566 | struct bio *bio, | |
567 | struct buffer_head *bh) | |
568 | { | |
569 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
74d46992 | 570 | bio_set_dev(bio, bh->b_bdev); |
0e51a8e1 | 571 | } |
7bf7f352 | 572 | |
0e51a8e1 CH |
573 | static struct xfs_ioend * |
574 | xfs_alloc_ioend( | |
575 | struct inode *inode, | |
576 | unsigned int type, | |
577 | xfs_off_t offset, | |
578 | struct buffer_head *bh) | |
579 | { | |
580 | struct xfs_ioend *ioend; | |
581 | struct bio *bio; | |
f6d6d4fc | 582 | |
0e51a8e1 CH |
583 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
584 | xfs_init_bio_from_bh(bio, bh); | |
585 | ||
586 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
587 | INIT_LIST_HEAD(&ioend->io_list); | |
588 | ioend->io_type = type; | |
589 | ioend->io_inode = inode; | |
590 | ioend->io_size = 0; | |
591 | ioend->io_offset = offset; | |
592 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
593 | ioend->io_append_trans = NULL; | |
594 | ioend->io_bio = bio; | |
595 | return ioend; | |
596 | } | |
597 | ||
598 | /* | |
599 | * Allocate a new bio, and chain the old bio to the new one. | |
600 | * | |
601 | * Note that we have to do perform the chaining in this unintuitive order | |
602 | * so that the bi_private linkage is set up in the right direction for the | |
603 | * traversal in xfs_destroy_ioend(). | |
604 | */ | |
605 | static void | |
606 | xfs_chain_bio( | |
607 | struct xfs_ioend *ioend, | |
608 | struct writeback_control *wbc, | |
609 | struct buffer_head *bh) | |
610 | { | |
611 | struct bio *new; | |
612 | ||
613 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
614 | xfs_init_bio_from_bh(new, bh); | |
615 | ||
616 | bio_chain(ioend->io_bio, new); | |
617 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
7637241e | 618 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
31d7d58d | 619 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 620 | submit_bio(ioend->io_bio); |
0e51a8e1 | 621 | ioend->io_bio = new; |
f6d6d4fc CH |
622 | } |
623 | ||
624 | /* | |
625 | * Test to see if we've been building up a completion structure for | |
626 | * earlier buffers -- if so, we try to append to this ioend if we | |
627 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
628 | * Return the ioend we finished off so that the caller can submit it |
629 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
630 | */ |
631 | STATIC void | |
632 | xfs_add_to_ioend( | |
633 | struct inode *inode, | |
634 | struct buffer_head *bh, | |
7336cea8 | 635 | xfs_off_t offset, |
e10de372 | 636 | struct xfs_writepage_ctx *wpc, |
bb18782a | 637 | struct writeback_control *wbc, |
e10de372 | 638 | struct list_head *iolist) |
f6d6d4fc | 639 | { |
fbcc0256 | 640 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
641 | bh->b_blocknr != wpc->last_block + 1 || |
642 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
643 | if (wpc->ioend) |
644 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 645 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
646 | } |
647 | ||
0e51a8e1 CH |
648 | /* |
649 | * If the buffer doesn't fit into the bio we need to allocate a new | |
650 | * one. This shouldn't happen more than once for a given buffer. | |
651 | */ | |
652 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
653 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 654 | |
fbcc0256 DC |
655 | wpc->ioend->io_size += bh->b_size; |
656 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 657 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
658 | } |
659 | ||
87cbc49c NS |
660 | STATIC void |
661 | xfs_map_buffer( | |
046f1685 | 662 | struct inode *inode, |
87cbc49c | 663 | struct buffer_head *bh, |
207d0416 | 664 | struct xfs_bmbt_irec *imap, |
046f1685 | 665 | xfs_off_t offset) |
87cbc49c NS |
666 | { |
667 | sector_t bn; | |
8699bb0a | 668 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
669 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
670 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 671 | |
207d0416 CH |
672 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
673 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 674 | |
e513182d | 675 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 676 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 677 | |
046f1685 | 678 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
679 | |
680 | bh->b_blocknr = bn; | |
681 | set_buffer_mapped(bh); | |
682 | } | |
683 | ||
1da177e4 LT |
684 | STATIC void |
685 | xfs_map_at_offset( | |
046f1685 | 686 | struct inode *inode, |
1da177e4 | 687 | struct buffer_head *bh, |
207d0416 | 688 | struct xfs_bmbt_irec *imap, |
046f1685 | 689 | xfs_off_t offset) |
1da177e4 | 690 | { |
207d0416 CH |
691 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
692 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 693 | |
207d0416 | 694 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
695 | set_buffer_mapped(bh); |
696 | clear_buffer_delay(bh); | |
f6d6d4fc | 697 | clear_buffer_unwritten(bh); |
1da177e4 LT |
698 | } |
699 | ||
1da177e4 | 700 | /* |
a49935f2 DC |
701 | * Test if a given page contains at least one buffer of a given @type. |
702 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
703 | * try to find one of the type passed in. If it is not set, then the caller only | |
704 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 705 | */ |
a49935f2 | 706 | STATIC bool |
6ffc4db5 | 707 | xfs_check_page_type( |
10ce4444 | 708 | struct page *page, |
a49935f2 DC |
709 | unsigned int type, |
710 | bool check_all_buffers) | |
1da177e4 | 711 | { |
a49935f2 DC |
712 | struct buffer_head *bh; |
713 | struct buffer_head *head; | |
1da177e4 | 714 | |
a49935f2 DC |
715 | if (PageWriteback(page)) |
716 | return false; | |
717 | if (!page->mapping) | |
718 | return false; | |
719 | if (!page_has_buffers(page)) | |
720 | return false; | |
1da177e4 | 721 | |
a49935f2 DC |
722 | bh = head = page_buffers(page); |
723 | do { | |
724 | if (buffer_unwritten(bh)) { | |
725 | if (type == XFS_IO_UNWRITTEN) | |
726 | return true; | |
727 | } else if (buffer_delay(bh)) { | |
805eeb8e | 728 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
729 | return true; |
730 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 731 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
732 | return true; |
733 | } | |
1da177e4 | 734 | |
a49935f2 DC |
735 | /* If we are only checking the first buffer, we are done now. */ |
736 | if (!check_all_buffers) | |
737 | break; | |
738 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 739 | |
a49935f2 | 740 | return false; |
1da177e4 LT |
741 | } |
742 | ||
3ed3a434 DC |
743 | STATIC void |
744 | xfs_vm_invalidatepage( | |
745 | struct page *page, | |
d47992f8 LC |
746 | unsigned int offset, |
747 | unsigned int length) | |
3ed3a434 | 748 | { |
34097dfe LC |
749 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
750 | length); | |
793d7dbe DC |
751 | |
752 | /* | |
753 | * If we are invalidating the entire page, clear the dirty state from it | |
754 | * so that we can check for attempts to release dirty cached pages in | |
755 | * xfs_vm_releasepage(). | |
756 | */ | |
757 | if (offset == 0 && length >= PAGE_SIZE) | |
758 | cancel_dirty_page(page); | |
34097dfe | 759 | block_invalidatepage(page, offset, length); |
3ed3a434 DC |
760 | } |
761 | ||
762 | /* | |
763 | * If the page has delalloc buffers on it, we need to punch them out before we | |
764 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
765 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
766 | * is done on that same region - the delalloc extent is returned when none is | |
767 | * supposed to be there. | |
768 | * | |
769 | * We prevent this by truncating away the delalloc regions on the page before | |
770 | * invalidating it. Because they are delalloc, we can do this without needing a | |
771 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
772 | * truncation without a transaction as there is no space left for block | |
773 | * reservation (typically why we see a ENOSPC in writeback). | |
774 | * | |
775 | * This is not a performance critical path, so for now just do the punching a | |
776 | * buffer head at a time. | |
777 | */ | |
778 | STATIC void | |
779 | xfs_aops_discard_page( | |
780 | struct page *page) | |
781 | { | |
782 | struct inode *inode = page->mapping->host; | |
783 | struct xfs_inode *ip = XFS_I(inode); | |
784 | struct buffer_head *bh, *head; | |
785 | loff_t offset = page_offset(page); | |
3ed3a434 | 786 | |
a49935f2 | 787 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
788 | goto out_invalidate; |
789 | ||
e8c3753c DC |
790 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
791 | goto out_invalidate; | |
792 | ||
4f10700a | 793 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
794 | "page discard on page %p, inode 0x%llx, offset %llu.", |
795 | page, ip->i_ino, offset); | |
796 | ||
797 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
798 | bh = head = page_buffers(page); | |
799 | do { | |
3ed3a434 | 800 | int error; |
c726de44 | 801 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
802 | |
803 | if (!buffer_delay(bh)) | |
804 | goto next_buffer; | |
805 | ||
c726de44 DC |
806 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
807 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
808 | if (error) { |
809 | /* something screwed, just bail */ | |
e8c3753c | 810 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 811 | xfs_alert(ip->i_mount, |
3ed3a434 | 812 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 813 | } |
3ed3a434 DC |
814 | break; |
815 | } | |
816 | next_buffer: | |
93407472 | 817 | offset += i_blocksize(inode); |
3ed3a434 DC |
818 | |
819 | } while ((bh = bh->b_this_page) != head); | |
820 | ||
821 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
822 | out_invalidate: | |
09cbfeaf | 823 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
824 | return; |
825 | } | |
826 | ||
ef473667 DW |
827 | static int |
828 | xfs_map_cow( | |
829 | struct xfs_writepage_ctx *wpc, | |
830 | struct inode *inode, | |
831 | loff_t offset, | |
832 | unsigned int *new_type) | |
833 | { | |
834 | struct xfs_inode *ip = XFS_I(inode); | |
835 | struct xfs_bmbt_irec imap; | |
092d5d9d | 836 | bool is_cow = false; |
ef473667 DW |
837 | int error; |
838 | ||
839 | /* | |
840 | * If we already have a valid COW mapping keep using it. | |
841 | */ | |
842 | if (wpc->io_type == XFS_IO_COW) { | |
843 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
844 | if (wpc->imap_valid) { | |
845 | *new_type = XFS_IO_COW; | |
846 | return 0; | |
847 | } | |
848 | } | |
849 | ||
850 | /* | |
851 | * Else we need to check if there is a COW mapping at this offset. | |
852 | */ | |
853 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
092d5d9d | 854 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap); |
ef473667 DW |
855 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
856 | ||
857 | if (!is_cow) | |
858 | return 0; | |
859 | ||
860 | /* | |
861 | * And if the COW mapping has a delayed extent here we need to | |
862 | * allocate real space for it now. | |
863 | */ | |
092d5d9d | 864 | if (isnullstartblock(imap.br_startblock)) { |
ef473667 DW |
865 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, |
866 | &imap); | |
867 | if (error) | |
868 | return error; | |
869 | } | |
870 | ||
871 | wpc->io_type = *new_type = XFS_IO_COW; | |
872 | wpc->imap_valid = true; | |
873 | wpc->imap = imap; | |
874 | return 0; | |
875 | } | |
876 | ||
e10de372 DC |
877 | /* |
878 | * We implement an immediate ioend submission policy here to avoid needing to | |
879 | * chain multiple ioends and hence nest mempool allocations which can violate | |
880 | * forward progress guarantees we need to provide. The current ioend we are | |
881 | * adding buffers to is cached on the writepage context, and if the new buffer | |
882 | * does not append to the cached ioend it will create a new ioend and cache that | |
883 | * instead. | |
884 | * | |
885 | * If a new ioend is created and cached, the old ioend is returned and queued | |
886 | * locally for submission once the entire page is processed or an error has been | |
887 | * detected. While ioends are submitted immediately after they are completed, | |
888 | * batching optimisations are provided by higher level block plugging. | |
889 | * | |
890 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
891 | * writepage context that the caller will need to submit. | |
892 | */ | |
bfce7d2e DC |
893 | static int |
894 | xfs_writepage_map( | |
895 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 896 | struct writeback_control *wbc, |
bfce7d2e DC |
897 | struct inode *inode, |
898 | struct page *page, | |
2d5f4b5b | 899 | uint64_t end_offset) |
bfce7d2e | 900 | { |
e10de372 DC |
901 | LIST_HEAD(submit_list); |
902 | struct xfs_ioend *ioend, *next; | |
bfce7d2e | 903 | struct buffer_head *bh, *head; |
93407472 | 904 | ssize_t len = i_blocksize(inode); |
2d5f4b5b | 905 | uint64_t offset; |
bfce7d2e | 906 | int error = 0; |
bfce7d2e | 907 | int count = 0; |
e10de372 | 908 | int uptodate = 1; |
ef473667 | 909 | unsigned int new_type; |
bfce7d2e DC |
910 | |
911 | bh = head = page_buffers(page); | |
912 | offset = page_offset(page); | |
bfce7d2e DC |
913 | do { |
914 | if (offset >= end_offset) | |
915 | break; | |
916 | if (!buffer_uptodate(bh)) | |
917 | uptodate = 0; | |
918 | ||
919 | /* | |
920 | * set_page_dirty dirties all buffers in a page, independent | |
921 | * of their state. The dirty state however is entirely | |
922 | * meaningless for holes (!mapped && uptodate), so skip | |
923 | * buffers covering holes here. | |
924 | */ | |
925 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
926 | wpc->imap_valid = false; | |
927 | continue; | |
928 | } | |
929 | ||
ef473667 DW |
930 | if (buffer_unwritten(bh)) |
931 | new_type = XFS_IO_UNWRITTEN; | |
932 | else if (buffer_delay(bh)) | |
933 | new_type = XFS_IO_DELALLOC; | |
934 | else if (buffer_uptodate(bh)) | |
935 | new_type = XFS_IO_OVERWRITE; | |
936 | else { | |
bfce7d2e DC |
937 | if (PageUptodate(page)) |
938 | ASSERT(buffer_mapped(bh)); | |
939 | /* | |
940 | * This buffer is not uptodate and will not be | |
941 | * written to disk. Ensure that we will put any | |
942 | * subsequent writeable buffers into a new | |
943 | * ioend. | |
944 | */ | |
945 | wpc->imap_valid = false; | |
946 | continue; | |
947 | } | |
948 | ||
ef473667 DW |
949 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
950 | error = xfs_map_cow(wpc, inode, offset, &new_type); | |
951 | if (error) | |
952 | goto out; | |
953 | } | |
954 | ||
955 | if (wpc->io_type != new_type) { | |
956 | wpc->io_type = new_type; | |
957 | wpc->imap_valid = false; | |
958 | } | |
959 | ||
bfce7d2e DC |
960 | if (wpc->imap_valid) |
961 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
962 | offset); | |
963 | if (!wpc->imap_valid) { | |
964 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
965 | wpc->io_type); | |
966 | if (error) | |
e10de372 | 967 | goto out; |
bfce7d2e DC |
968 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
969 | offset); | |
970 | } | |
971 | if (wpc->imap_valid) { | |
972 | lock_buffer(bh); | |
973 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
974 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 975 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
976 | count++; |
977 | } | |
978 | ||
bfce7d2e DC |
979 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
980 | ||
981 | if (uptodate && bh == head) | |
982 | SetPageUptodate(page); | |
983 | ||
e10de372 | 984 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 985 | |
e10de372 | 986 | out: |
bfce7d2e | 987 | /* |
e10de372 DC |
988 | * On error, we have to fail the ioend here because we have locked |
989 | * buffers in the ioend. If we don't do this, we'll deadlock | |
990 | * invalidating the page as that tries to lock the buffers on the page. | |
991 | * Also, because we may have set pages under writeback, we have to make | |
992 | * sure we run IO completion to mark the error state of the IO | |
993 | * appropriately, so we can't cancel the ioend directly here. That means | |
994 | * we have to mark this page as under writeback if we included any | |
995 | * buffers from it in the ioend chain so that completion treats it | |
996 | * correctly. | |
bfce7d2e | 997 | * |
e10de372 DC |
998 | * If we didn't include the page in the ioend, the on error we can |
999 | * simply discard and unlock it as there are no other users of the page | |
1000 | * or it's buffers right now. The caller will still need to trigger | |
1001 | * submission of outstanding ioends on the writepage context so they are | |
1002 | * treated correctly on error. | |
bfce7d2e | 1003 | */ |
e10de372 DC |
1004 | if (count) { |
1005 | xfs_start_page_writeback(page, !error); | |
1006 | ||
1007 | /* | |
1008 | * Preserve the original error if there was one, otherwise catch | |
1009 | * submission errors here and propagate into subsequent ioend | |
1010 | * submissions. | |
1011 | */ | |
1012 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
1013 | int error2; | |
1014 | ||
1015 | list_del_init(&ioend->io_list); | |
1016 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
1017 | if (error2 && !error) | |
1018 | error = error2; | |
1019 | } | |
1020 | } else if (error) { | |
bfce7d2e DC |
1021 | xfs_aops_discard_page(page); |
1022 | ClearPageUptodate(page); | |
1023 | unlock_page(page); | |
e10de372 DC |
1024 | } else { |
1025 | /* | |
1026 | * We can end up here with no error and nothing to write if we | |
1027 | * race with a partial page truncate on a sub-page block sized | |
1028 | * filesystem. In that case we need to mark the page clean. | |
1029 | */ | |
1030 | xfs_start_page_writeback(page, 1); | |
1031 | end_page_writeback(page); | |
bfce7d2e | 1032 | } |
e10de372 | 1033 | |
bfce7d2e DC |
1034 | mapping_set_error(page->mapping, error); |
1035 | return error; | |
1036 | } | |
1037 | ||
1da177e4 | 1038 | /* |
89f3b363 CH |
1039 | * Write out a dirty page. |
1040 | * | |
1041 | * For delalloc space on the page we need to allocate space and flush it. | |
1042 | * For unwritten space on the page we need to start the conversion to | |
1043 | * regular allocated space. | |
89f3b363 | 1044 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1045 | */ |
1da177e4 | 1046 | STATIC int |
fbcc0256 | 1047 | xfs_do_writepage( |
89f3b363 | 1048 | struct page *page, |
fbcc0256 DC |
1049 | struct writeback_control *wbc, |
1050 | void *data) | |
1da177e4 | 1051 | { |
fbcc0256 | 1052 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 1053 | struct inode *inode = page->mapping->host; |
1da177e4 | 1054 | loff_t offset; |
c8ce540d | 1055 | uint64_t end_offset; |
ad68972a | 1056 | pgoff_t end_index; |
89f3b363 | 1057 | |
34097dfe | 1058 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 1059 | |
20cb52eb CH |
1060 | ASSERT(page_has_buffers(page)); |
1061 | ||
89f3b363 CH |
1062 | /* |
1063 | * Refuse to write the page out if we are called from reclaim context. | |
1064 | * | |
d4f7a5cb CH |
1065 | * This avoids stack overflows when called from deeply used stacks in |
1066 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1067 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 1068 | * |
94054fa3 MG |
1069 | * This should never happen except in the case of a VM regression so |
1070 | * warn about it. | |
89f3b363 | 1071 | */ |
94054fa3 MG |
1072 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
1073 | PF_MEMALLOC)) | |
b5420f23 | 1074 | goto redirty; |
1da177e4 | 1075 | |
89f3b363 | 1076 | /* |
680a647b CH |
1077 | * Given that we do not allow direct reclaim to call us, we should |
1078 | * never be called while in a filesystem transaction. | |
89f3b363 | 1079 | */ |
9070733b | 1080 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) |
b5420f23 | 1081 | goto redirty; |
89f3b363 | 1082 | |
8695d27e | 1083 | /* |
ad68972a DC |
1084 | * Is this page beyond the end of the file? |
1085 | * | |
8695d27e JL |
1086 | * The page index is less than the end_index, adjust the end_offset |
1087 | * to the highest offset that this page should represent. | |
1088 | * ----------------------------------------------------- | |
1089 | * | file mapping | <EOF> | | |
1090 | * ----------------------------------------------------- | |
1091 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1092 | * ^--------------------------------^----------|-------- | |
1093 | * | desired writeback range | see else | | |
1094 | * ---------------------------------^------------------| | |
1095 | */ | |
ad68972a | 1096 | offset = i_size_read(inode); |
09cbfeaf | 1097 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 1098 | if (page->index < end_index) |
09cbfeaf | 1099 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
1100 | else { |
1101 | /* | |
1102 | * Check whether the page to write out is beyond or straddles | |
1103 | * i_size or not. | |
1104 | * ------------------------------------------------------- | |
1105 | * | file mapping | <EOF> | | |
1106 | * ------------------------------------------------------- | |
1107 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1108 | * ^--------------------------------^-----------|--------- | |
1109 | * | | Straddles | | |
1110 | * ---------------------------------^-----------|--------| | |
1111 | */ | |
09cbfeaf | 1112 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
1113 | |
1114 | /* | |
ff9a28f6 JK |
1115 | * Skip the page if it is fully outside i_size, e.g. due to a |
1116 | * truncate operation that is in progress. We must redirty the | |
1117 | * page so that reclaim stops reclaiming it. Otherwise | |
1118 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1119 | * |
1120 | * Note that the end_index is unsigned long, it would overflow | |
1121 | * if the given offset is greater than 16TB on 32-bit system | |
1122 | * and if we do check the page is fully outside i_size or not | |
1123 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1124 | * will be evaluated to 0. Hence this page will be redirtied | |
1125 | * and be written out repeatedly which would result in an | |
1126 | * infinite loop, the user program that perform this operation | |
1127 | * will hang. Instead, we can verify this situation by checking | |
1128 | * if the page to write is totally beyond the i_size or if it's | |
1129 | * offset is just equal to the EOF. | |
6b7a03f0 | 1130 | */ |
8695d27e JL |
1131 | if (page->index > end_index || |
1132 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1133 | goto redirty; |
6b7a03f0 CH |
1134 | |
1135 | /* | |
1136 | * The page straddles i_size. It must be zeroed out on each | |
1137 | * and every writepage invocation because it may be mmapped. | |
1138 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1139 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1140 | * memory is zeroed when mapped, and writes to that region are |
1141 | * not written out to the file." | |
1142 | */ | |
09cbfeaf | 1143 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
1144 | |
1145 | /* Adjust the end_offset to the end of file */ | |
1146 | end_offset = offset; | |
1da177e4 LT |
1147 | } |
1148 | ||
2d5f4b5b | 1149 | return xfs_writepage_map(wpc, wbc, inode, page, end_offset); |
f51623b2 | 1150 | |
b5420f23 | 1151 | redirty: |
f51623b2 NS |
1152 | redirty_page_for_writepage(wbc, page); |
1153 | unlock_page(page); | |
1154 | return 0; | |
f51623b2 NS |
1155 | } |
1156 | ||
fbcc0256 DC |
1157 | STATIC int |
1158 | xfs_vm_writepage( | |
1159 | struct page *page, | |
1160 | struct writeback_control *wbc) | |
1161 | { | |
1162 | struct xfs_writepage_ctx wpc = { | |
1163 | .io_type = XFS_IO_INVALID, | |
1164 | }; | |
1165 | int ret; | |
1166 | ||
1167 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1168 | if (wpc.ioend) |
1169 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1170 | return ret; | |
fbcc0256 DC |
1171 | } |
1172 | ||
7d4fb40a NS |
1173 | STATIC int |
1174 | xfs_vm_writepages( | |
1175 | struct address_space *mapping, | |
1176 | struct writeback_control *wbc) | |
1177 | { | |
fbcc0256 DC |
1178 | struct xfs_writepage_ctx wpc = { |
1179 | .io_type = XFS_IO_INVALID, | |
1180 | }; | |
1181 | int ret; | |
1182 | ||
b3aea4ed | 1183 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7f6d5b52 RZ |
1184 | if (dax_mapping(mapping)) |
1185 | return dax_writeback_mapping_range(mapping, | |
1186 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1187 | ||
fbcc0256 | 1188 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1189 | if (wpc.ioend) |
1190 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1191 | return ret; | |
7d4fb40a NS |
1192 | } |
1193 | ||
f51623b2 NS |
1194 | /* |
1195 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1196 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1197 | * have buffer heads in this call. |
1198 | * | |
89f3b363 | 1199 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1200 | */ |
1201 | STATIC int | |
238f4c54 | 1202 | xfs_vm_releasepage( |
f51623b2 NS |
1203 | struct page *page, |
1204 | gfp_t gfp_mask) | |
1205 | { | |
20cb52eb | 1206 | int delalloc, unwritten; |
f51623b2 | 1207 | |
34097dfe | 1208 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1209 | |
99579cce BF |
1210 | /* |
1211 | * mm accommodates an old ext3 case where clean pages might not have had | |
1212 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1213 | * ->releasepage() via shrink_active_list(). Conversely, | |
793d7dbe DC |
1214 | * block_invalidatepage() can send pages that are still marked dirty but |
1215 | * otherwise have invalidated buffers. | |
99579cce | 1216 | * |
0a417b8d | 1217 | * We want to release the latter to avoid unnecessary buildup of the |
793d7dbe DC |
1218 | * LRU, so xfs_vm_invalidatepage() clears the page dirty flag on pages |
1219 | * that are entirely invalidated and need to be released. Hence the | |
1220 | * only time we should get dirty pages here is through | |
1221 | * shrink_active_list() and so we can simply skip those now. | |
1222 | * | |
1223 | * warn if we've left any lingering delalloc/unwritten buffers on clean | |
1224 | * or invalidated pages we are about to release. | |
99579cce | 1225 | */ |
793d7dbe DC |
1226 | if (PageDirty(page)) |
1227 | return 0; | |
1228 | ||
20cb52eb | 1229 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1230 | |
793d7dbe | 1231 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1232 | return 0; |
793d7dbe | 1233 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1234 | return 0; |
1235 | ||
f51623b2 NS |
1236 | return try_to_free_buffers(page); |
1237 | } | |
1238 | ||
1fdca9c2 DC |
1239 | /* |
1240 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1241 | * is, so that we can avoid repeated get_blocks calls. | |
1242 | * | |
1243 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1244 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1245 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1246 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1247 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1248 | * to and including EOF, and a second mapping for beyond EOF. | |
1249 | */ | |
1250 | static void | |
1251 | xfs_map_trim_size( | |
1252 | struct inode *inode, | |
1253 | sector_t iblock, | |
1254 | struct buffer_head *bh_result, | |
1255 | struct xfs_bmbt_irec *imap, | |
1256 | xfs_off_t offset, | |
1257 | ssize_t size) | |
1258 | { | |
1259 | xfs_off_t mapping_size; | |
1260 | ||
1261 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1262 | mapping_size <<= inode->i_blkbits; | |
1263 | ||
1264 | ASSERT(mapping_size > 0); | |
1265 | if (mapping_size > size) | |
1266 | mapping_size = size; | |
1267 | if (offset < i_size_read(inode) && | |
22a6c837 | 1268 | (xfs_ufsize_t)offset + mapping_size >= i_size_read(inode)) { |
1fdca9c2 DC |
1269 | /* limit mapping to block that spans EOF */ |
1270 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
93407472 | 1271 | i_blocksize(inode)); |
1fdca9c2 DC |
1272 | } |
1273 | if (mapping_size > LONG_MAX) | |
1274 | mapping_size = LONG_MAX; | |
1275 | ||
1276 | bh_result->b_size = mapping_size; | |
1277 | } | |
1278 | ||
0613f16c | 1279 | static int |
acdda3aa | 1280 | xfs_get_blocks( |
1da177e4 LT |
1281 | struct inode *inode, |
1282 | sector_t iblock, | |
1da177e4 | 1283 | struct buffer_head *bh_result, |
acdda3aa | 1284 | int create) |
1da177e4 | 1285 | { |
a206c817 CH |
1286 | struct xfs_inode *ip = XFS_I(inode); |
1287 | struct xfs_mount *mp = ip->i_mount; | |
1288 | xfs_fileoff_t offset_fsb, end_fsb; | |
1289 | int error = 0; | |
1290 | int lockmode = 0; | |
207d0416 | 1291 | struct xfs_bmbt_irec imap; |
a206c817 | 1292 | int nimaps = 1; |
fdc7ed75 NS |
1293 | xfs_off_t offset; |
1294 | ssize_t size; | |
a206c817 | 1295 | |
acdda3aa | 1296 | BUG_ON(create); |
6e8a27a8 | 1297 | |
a206c817 | 1298 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1299 | return -EIO; |
1da177e4 | 1300 | |
fdc7ed75 | 1301 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
93407472 | 1302 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
c2536668 | 1303 | size = bh_result->b_size; |
364f358a | 1304 | |
acdda3aa | 1305 | if (offset >= i_size_read(inode)) |
364f358a LM |
1306 | return 0; |
1307 | ||
507630b2 DC |
1308 | /* |
1309 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1310 | * a block mapping without an exclusive lock first. |
507630b2 | 1311 | */ |
6e8a27a8 | 1312 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1313 | |
d2c28191 | 1314 | ASSERT(offset <= mp->m_super->s_maxbytes); |
b4d8ad7f | 1315 | if (offset > mp->m_super->s_maxbytes - size) |
d2c28191 | 1316 | size = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
1317 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1318 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1319 | ||
acdda3aa CH |
1320 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1321 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1322 | if (error) |
a206c817 CH |
1323 | goto out_unlock; |
1324 | ||
acdda3aa | 1325 | if (nimaps) { |
d5cc2e3f | 1326 | trace_xfs_get_blocks_found(ip, offset, size, |
63fbb4c1 CH |
1327 | imap.br_state == XFS_EXT_UNWRITTEN ? |
1328 | XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1329 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1330 | } else { |
1331 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1332 | goto out_unlock; | |
1333 | } | |
1da177e4 | 1334 | |
1fdca9c2 | 1335 | /* trim mapping down to size requested */ |
6e8a27a8 | 1336 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1337 | |
a719370b DC |
1338 | /* |
1339 | * For unwritten extents do not report a disk address in the buffered | |
1340 | * read case (treat as if we're reading into a hole). | |
1341 | */ | |
9c4f29d3 | 1342 | if (xfs_bmap_is_real_extent(&imap)) |
a719370b | 1343 | xfs_map_buffer(inode, bh_result, &imap, offset); |
1da177e4 | 1344 | |
c2536668 NS |
1345 | /* |
1346 | * If this is a realtime file, data may be on a different device. | |
1347 | * to that pointed to from the buffer_head b_bdev currently. | |
1348 | */ | |
046f1685 | 1349 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1350 | return 0; |
a206c817 CH |
1351 | |
1352 | out_unlock: | |
1353 | xfs_iunlock(ip, lockmode); | |
2451337d | 1354 | return error; |
1da177e4 LT |
1355 | } |
1356 | ||
c19b104a CH |
1357 | STATIC ssize_t |
1358 | xfs_vm_direct_IO( | |
6e1ba0bc | 1359 | struct kiocb *iocb, |
c8b8e32d | 1360 | struct iov_iter *iter) |
6e1ba0bc | 1361 | { |
58e59854 | 1362 | /* |
fa8d972d | 1363 | * We just need the method present so that open/fcntl allow direct I/O. |
58e59854 | 1364 | */ |
fa8d972d | 1365 | return -EINVAL; |
f51623b2 | 1366 | } |
1da177e4 LT |
1367 | |
1368 | STATIC sector_t | |
e4c573bb | 1369 | xfs_vm_bmap( |
1da177e4 LT |
1370 | struct address_space *mapping, |
1371 | sector_t block) | |
1372 | { | |
1373 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1374 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1375 | |
cca28fb8 | 1376 | trace_xfs_vm_bmap(XFS_I(inode)); |
db1327b1 DW |
1377 | |
1378 | /* | |
1379 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
1380 | * bypasseѕ the file system for actual I/O. We really can't allow | |
1381 | * that on reflinks inodes, so we have to skip out here. And yes, | |
eb5e248d DW |
1382 | * 0 is the magic code for a bmap error. |
1383 | * | |
1384 | * Since we don't pass back blockdev info, we can't return bmap | |
1385 | * information for rt files either. | |
db1327b1 | 1386 | */ |
eb5e248d | 1387 | if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) |
db1327b1 | 1388 | return 0; |
65523218 | 1389 | |
4bc1ea6b | 1390 | filemap_write_and_wait(mapping); |
c2536668 | 1391 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1392 | } |
1393 | ||
1394 | STATIC int | |
e4c573bb | 1395 | xfs_vm_readpage( |
1da177e4 LT |
1396 | struct file *unused, |
1397 | struct page *page) | |
1398 | { | |
121e213e | 1399 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1400 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1401 | } |
1402 | ||
1403 | STATIC int | |
e4c573bb | 1404 | xfs_vm_readpages( |
1da177e4 LT |
1405 | struct file *unused, |
1406 | struct address_space *mapping, | |
1407 | struct list_head *pages, | |
1408 | unsigned nr_pages) | |
1409 | { | |
121e213e | 1410 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1411 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1412 | } |
1413 | ||
22e757a4 DC |
1414 | /* |
1415 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1416 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1417 | * dirty, we'll never be able to clean them because we don't write buffers | |
1418 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1419 | * that have been marked dirty. Further, the dirty state can leak into | |
1420 | * the file interior if the file is extended, resulting in all sorts of | |
1421 | * bad things happening as the state does not match the underlying data. | |
1422 | * | |
1423 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1424 | * this only exist because of bufferheads and how the generic code manages them. | |
1425 | */ | |
1426 | STATIC int | |
1427 | xfs_vm_set_page_dirty( | |
1428 | struct page *page) | |
1429 | { | |
1430 | struct address_space *mapping = page->mapping; | |
1431 | struct inode *inode = mapping->host; | |
1432 | loff_t end_offset; | |
1433 | loff_t offset; | |
1434 | int newly_dirty; | |
1435 | ||
1436 | if (unlikely(!mapping)) | |
1437 | return !TestSetPageDirty(page); | |
1438 | ||
1439 | end_offset = i_size_read(inode); | |
1440 | offset = page_offset(page); | |
1441 | ||
1442 | spin_lock(&mapping->private_lock); | |
1443 | if (page_has_buffers(page)) { | |
1444 | struct buffer_head *head = page_buffers(page); | |
1445 | struct buffer_head *bh = head; | |
1446 | ||
1447 | do { | |
1448 | if (offset < end_offset) | |
1449 | set_buffer_dirty(bh); | |
1450 | bh = bh->b_this_page; | |
93407472 | 1451 | offset += i_blocksize(inode); |
22e757a4 DC |
1452 | } while (bh != head); |
1453 | } | |
c4843a75 | 1454 | /* |
81f8c3a4 JW |
1455 | * Lock out page->mem_cgroup migration to keep PageDirty |
1456 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1457 | */ |
62cccb8c | 1458 | lock_page_memcg(page); |
22e757a4 DC |
1459 | newly_dirty = !TestSetPageDirty(page); |
1460 | spin_unlock(&mapping->private_lock); | |
1461 | ||
1462 | if (newly_dirty) { | |
1463 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1464 | unsigned long flags; | |
1465 | ||
1466 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1467 | if (page->mapping) { /* Race with truncate? */ | |
1468 | WARN_ON_ONCE(!PageUptodate(page)); | |
62cccb8c | 1469 | account_page_dirtied(page, mapping); |
22e757a4 DC |
1470 | radix_tree_tag_set(&mapping->page_tree, |
1471 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1472 | } | |
1473 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1474 | } |
62cccb8c | 1475 | unlock_page_memcg(page); |
c4843a75 GT |
1476 | if (newly_dirty) |
1477 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1478 | return newly_dirty; |
1479 | } | |
1480 | ||
f5e54d6e | 1481 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1482 | .readpage = xfs_vm_readpage, |
1483 | .readpages = xfs_vm_readpages, | |
1484 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1485 | .writepages = xfs_vm_writepages, |
22e757a4 | 1486 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1487 | .releasepage = xfs_vm_releasepage, |
1488 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb NS |
1489 | .bmap = xfs_vm_bmap, |
1490 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1491 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1492 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1493 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1494 | }; |