Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
70a9883c | 7 | #include "xfs_shared.h" |
239880ef DC |
8 | #include "xfs_format.h" |
9 | #include "xfs_log_format.h" | |
10 | #include "xfs_trans_resv.h" | |
1da177e4 | 11 | #include "xfs_mount.h" |
1da177e4 | 12 | #include "xfs_inode.h" |
239880ef | 13 | #include "xfs_trans.h" |
281627df | 14 | #include "xfs_inode_item.h" |
a844f451 | 15 | #include "xfs_alloc.h" |
1da177e4 | 16 | #include "xfs_error.h" |
1da177e4 | 17 | #include "xfs_iomap.h" |
0b1b213f | 18 | #include "xfs_trace.h" |
3ed3a434 | 19 | #include "xfs_bmap.h" |
68988114 | 20 | #include "xfs_bmap_util.h" |
a4fbe6ab | 21 | #include "xfs_bmap_btree.h" |
ef473667 | 22 | #include "xfs_reflink.h" |
5a0e3ad6 | 23 | #include <linux/gfp.h> |
1da177e4 | 24 | #include <linux/mpage.h> |
10ce4444 | 25 | #include <linux/pagevec.h> |
1da177e4 LT |
26 | #include <linux/writeback.h> |
27 | ||
fbcc0256 DC |
28 | /* |
29 | * structure owned by writepages passed to individual writepage calls | |
30 | */ | |
31 | struct xfs_writepage_ctx { | |
32 | struct xfs_bmbt_irec imap; | |
33 | bool imap_valid; | |
34 | unsigned int io_type; | |
fbcc0256 DC |
35 | struct xfs_ioend *ioend; |
36 | sector_t last_block; | |
37 | }; | |
38 | ||
0b1b213f | 39 | void |
f51623b2 NS |
40 | xfs_count_page_state( |
41 | struct page *page, | |
42 | int *delalloc, | |
f51623b2 NS |
43 | int *unwritten) |
44 | { | |
45 | struct buffer_head *bh, *head; | |
46 | ||
20cb52eb | 47 | *delalloc = *unwritten = 0; |
f51623b2 NS |
48 | |
49 | bh = head = page_buffers(page); | |
50 | do { | |
20cb52eb | 51 | if (buffer_unwritten(bh)) |
f51623b2 NS |
52 | (*unwritten) = 1; |
53 | else if (buffer_delay(bh)) | |
54 | (*delalloc) = 1; | |
55 | } while ((bh = bh->b_this_page) != head); | |
56 | } | |
57 | ||
20a90f58 | 58 | struct block_device * |
6214ed44 | 59 | xfs_find_bdev_for_inode( |
046f1685 | 60 | struct inode *inode) |
6214ed44 | 61 | { |
046f1685 | 62 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
63 | struct xfs_mount *mp = ip->i_mount; |
64 | ||
71ddabb9 | 65 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
66 | return mp->m_rtdev_targp->bt_bdev; |
67 | else | |
68 | return mp->m_ddev_targp->bt_bdev; | |
69 | } | |
70 | ||
486aff5e DW |
71 | struct dax_device * |
72 | xfs_find_daxdev_for_inode( | |
73 | struct inode *inode) | |
74 | { | |
75 | struct xfs_inode *ip = XFS_I(inode); | |
76 | struct xfs_mount *mp = ip->i_mount; | |
77 | ||
78 | if (XFS_IS_REALTIME_INODE(ip)) | |
79 | return mp->m_rtdev_targp->bt_daxdev; | |
80 | else | |
81 | return mp->m_ddev_targp->bt_daxdev; | |
82 | } | |
83 | ||
f6d6d4fc | 84 | /* |
37992c18 DC |
85 | * We're now finished for good with this page. Update the page state via the |
86 | * associated buffer_heads, paying attention to the start and end offsets that | |
87 | * we need to process on the page. | |
28b783e4 | 88 | * |
8353a814 CH |
89 | * Note that we open code the action in end_buffer_async_write here so that we |
90 | * only have to iterate over the buffers attached to the page once. This is not | |
91 | * only more efficient, but also ensures that we only calls end_page_writeback | |
92 | * at the end of the iteration, and thus avoids the pitfall of having the page | |
93 | * and buffers potentially freed after every call to end_buffer_async_write. | |
37992c18 DC |
94 | */ |
95 | static void | |
96 | xfs_finish_page_writeback( | |
97 | struct inode *inode, | |
98 | struct bio_vec *bvec, | |
99 | int error) | |
100 | { | |
8353a814 CH |
101 | struct buffer_head *head = page_buffers(bvec->bv_page), *bh = head; |
102 | bool busy = false; | |
37992c18 | 103 | unsigned int off = 0; |
8353a814 | 104 | unsigned long flags; |
37992c18 DC |
105 | |
106 | ASSERT(bvec->bv_offset < PAGE_SIZE); | |
93407472 | 107 | ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0); |
8353a814 | 108 | ASSERT(bvec->bv_offset + bvec->bv_len <= PAGE_SIZE); |
93407472 | 109 | ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0); |
37992c18 | 110 | |
8353a814 CH |
111 | local_irq_save(flags); |
112 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
37992c18 | 113 | do { |
8353a814 CH |
114 | if (off >= bvec->bv_offset && |
115 | off < bvec->bv_offset + bvec->bv_len) { | |
116 | ASSERT(buffer_async_write(bh)); | |
117 | ASSERT(bh->b_end_io == NULL); | |
118 | ||
119 | if (error) { | |
120 | mark_buffer_write_io_error(bh); | |
121 | clear_buffer_uptodate(bh); | |
122 | SetPageError(bvec->bv_page); | |
123 | } else { | |
124 | set_buffer_uptodate(bh); | |
125 | } | |
126 | clear_buffer_async_write(bh); | |
127 | unlock_buffer(bh); | |
128 | } else if (buffer_async_write(bh)) { | |
129 | ASSERT(buffer_locked(bh)); | |
130 | busy = true; | |
131 | } | |
132 | off += bh->b_size; | |
133 | } while ((bh = bh->b_this_page) != head); | |
134 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
135 | local_irq_restore(flags); | |
136 | ||
137 | if (!busy) | |
138 | end_page_writeback(bvec->bv_page); | |
37992c18 DC |
139 | } |
140 | ||
141 | /* | |
142 | * We're now finished for good with this ioend structure. Update the page | |
143 | * state, release holds on bios, and finally free up memory. Do not use the | |
144 | * ioend after this. | |
f6d6d4fc | 145 | */ |
0829c360 CH |
146 | STATIC void |
147 | xfs_destroy_ioend( | |
0e51a8e1 CH |
148 | struct xfs_ioend *ioend, |
149 | int error) | |
0829c360 | 150 | { |
37992c18 | 151 | struct inode *inode = ioend->io_inode; |
8353a814 CH |
152 | struct bio *bio = &ioend->io_inline_bio; |
153 | struct bio *last = ioend->io_bio, *next; | |
154 | u64 start = bio->bi_iter.bi_sector; | |
155 | bool quiet = bio_flagged(bio, BIO_QUIET); | |
f6d6d4fc | 156 | |
0e51a8e1 | 157 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
37992c18 DC |
158 | struct bio_vec *bvec; |
159 | int i; | |
160 | ||
0e51a8e1 CH |
161 | /* |
162 | * For the last bio, bi_private points to the ioend, so we | |
163 | * need to explicitly end the iteration here. | |
164 | */ | |
165 | if (bio == last) | |
166 | next = NULL; | |
167 | else | |
168 | next = bio->bi_private; | |
583fa586 | 169 | |
37992c18 DC |
170 | /* walk each page on bio, ending page IO on them */ |
171 | bio_for_each_segment_all(bvec, bio, i) | |
172 | xfs_finish_page_writeback(inode, bvec, error); | |
173 | ||
174 | bio_put(bio); | |
f6d6d4fc | 175 | } |
8353a814 CH |
176 | |
177 | if (unlikely(error && !quiet)) { | |
178 | xfs_err_ratelimited(XFS_I(inode)->i_mount, | |
179 | "writeback error on sector %llu", start); | |
180 | } | |
0829c360 CH |
181 | } |
182 | ||
fc0063c4 CH |
183 | /* |
184 | * Fast and loose check if this write could update the on-disk inode size. | |
185 | */ | |
186 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
187 | { | |
188 | return ioend->io_offset + ioend->io_size > | |
189 | XFS_I(ioend->io_inode)->i_d.di_size; | |
190 | } | |
191 | ||
281627df CH |
192 | STATIC int |
193 | xfs_setfilesize_trans_alloc( | |
194 | struct xfs_ioend *ioend) | |
195 | { | |
196 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
197 | struct xfs_trans *tp; | |
198 | int error; | |
199 | ||
4df0f7f1 DC |
200 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, |
201 | XFS_TRANS_NOFS, &tp); | |
253f4911 | 202 | if (error) |
281627df | 203 | return error; |
281627df CH |
204 | |
205 | ioend->io_append_trans = tp; | |
206 | ||
d9457dc0 | 207 | /* |
437a255a | 208 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
209 | * we released it. |
210 | */ | |
bee9182d | 211 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
212 | /* |
213 | * We hand off the transaction to the completion thread now, so | |
214 | * clear the flag here. | |
215 | */ | |
9070733b | 216 | current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
281627df CH |
217 | return 0; |
218 | } | |
219 | ||
ba87ea69 | 220 | /* |
2813d682 | 221 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 222 | */ |
281627df | 223 | STATIC int |
e372843a | 224 | __xfs_setfilesize( |
2ba66237 CH |
225 | struct xfs_inode *ip, |
226 | struct xfs_trans *tp, | |
227 | xfs_off_t offset, | |
228 | size_t size) | |
ba87ea69 | 229 | { |
ba87ea69 | 230 | xfs_fsize_t isize; |
ba87ea69 | 231 | |
aa6bf01d | 232 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 233 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
234 | if (!isize) { |
235 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 236 | xfs_trans_cancel(tp); |
281627df | 237 | return 0; |
ba87ea69 LM |
238 | } |
239 | ||
2ba66237 | 240 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
241 | |
242 | ip->i_d.di_size = isize; | |
243 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
244 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
245 | ||
70393313 | 246 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
247 | } |
248 | ||
e372843a CH |
249 | int |
250 | xfs_setfilesize( | |
251 | struct xfs_inode *ip, | |
252 | xfs_off_t offset, | |
253 | size_t size) | |
254 | { | |
255 | struct xfs_mount *mp = ip->i_mount; | |
256 | struct xfs_trans *tp; | |
257 | int error; | |
258 | ||
259 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); | |
260 | if (error) | |
261 | return error; | |
262 | ||
263 | return __xfs_setfilesize(ip, tp, offset, size); | |
264 | } | |
265 | ||
2ba66237 CH |
266 | STATIC int |
267 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
268 | struct xfs_ioend *ioend, |
269 | int error) | |
2ba66237 CH |
270 | { |
271 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
272 | struct xfs_trans *tp = ioend->io_append_trans; | |
273 | ||
274 | /* | |
275 | * The transaction may have been allocated in the I/O submission thread, | |
276 | * thus we need to mark ourselves as being in a transaction manually. | |
277 | * Similarly for freeze protection. | |
278 | */ | |
9070733b | 279 | current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
bee9182d | 280 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 281 | |
5cb13dcd | 282 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 283 | if (error) { |
5cb13dcd | 284 | xfs_trans_cancel(tp); |
0e51a8e1 | 285 | return error; |
5cb13dcd Z |
286 | } |
287 | ||
e372843a | 288 | return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
2ba66237 CH |
289 | } |
290 | ||
0829c360 | 291 | /* |
5ec4fabb | 292 | * IO write completion. |
f6d6d4fc CH |
293 | */ |
294 | STATIC void | |
5ec4fabb | 295 | xfs_end_io( |
77d7a0c2 | 296 | struct work_struct *work) |
0829c360 | 297 | { |
0e51a8e1 CH |
298 | struct xfs_ioend *ioend = |
299 | container_of(work, struct xfs_ioend, io_work); | |
300 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
787eb485 CH |
301 | xfs_off_t offset = ioend->io_offset; |
302 | size_t size = ioend->io_size; | |
4e4cbee9 | 303 | int error; |
ba87ea69 | 304 | |
af055e37 | 305 | /* |
787eb485 | 306 | * Just clean up the in-memory strutures if the fs has been shut down. |
af055e37 | 307 | */ |
787eb485 | 308 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
0e51a8e1 | 309 | error = -EIO; |
787eb485 CH |
310 | goto done; |
311 | } | |
04f658ee | 312 | |
43caeb18 | 313 | /* |
787eb485 | 314 | * Clean up any COW blocks on an I/O error. |
43caeb18 | 315 | */ |
4e4cbee9 | 316 | error = blk_status_to_errno(ioend->io_bio->bi_status); |
787eb485 CH |
317 | if (unlikely(error)) { |
318 | switch (ioend->io_type) { | |
319 | case XFS_IO_COW: | |
320 | xfs_reflink_cancel_cow_range(ip, offset, size, true); | |
321 | break; | |
43caeb18 | 322 | } |
787eb485 CH |
323 | |
324 | goto done; | |
43caeb18 DW |
325 | } |
326 | ||
5ec4fabb | 327 | /* |
787eb485 | 328 | * Success: commit the COW or unwritten blocks if needed. |
5ec4fabb | 329 | */ |
787eb485 CH |
330 | switch (ioend->io_type) { |
331 | case XFS_IO_COW: | |
332 | error = xfs_reflink_end_cow(ip, offset, size); | |
333 | break; | |
334 | case XFS_IO_UNWRITTEN: | |
ee70daab EG |
335 | /* writeback should never update isize */ |
336 | error = xfs_iomap_write_unwritten(ip, offset, size, false); | |
787eb485 CH |
337 | break; |
338 | default: | |
339 | ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans); | |
340 | break; | |
5ec4fabb | 341 | } |
ba87ea69 | 342 | |
04f658ee | 343 | done: |
787eb485 CH |
344 | if (ioend->io_append_trans) |
345 | error = xfs_setfilesize_ioend(ioend, error); | |
0e51a8e1 | 346 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
347 | } |
348 | ||
0e51a8e1 CH |
349 | STATIC void |
350 | xfs_end_bio( | |
351 | struct bio *bio) | |
0829c360 | 352 | { |
0e51a8e1 CH |
353 | struct xfs_ioend *ioend = bio->bi_private; |
354 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 355 | |
43caeb18 | 356 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
0e51a8e1 CH |
357 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
358 | else if (ioend->io_append_trans) | |
359 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
360 | else | |
4e4cbee9 | 361 | xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status)); |
0829c360 CH |
362 | } |
363 | ||
1da177e4 LT |
364 | STATIC int |
365 | xfs_map_blocks( | |
366 | struct inode *inode, | |
367 | loff_t offset, | |
207d0416 | 368 | struct xfs_bmbt_irec *imap, |
988ef927 | 369 | int type) |
1da177e4 | 370 | { |
a206c817 CH |
371 | struct xfs_inode *ip = XFS_I(inode); |
372 | struct xfs_mount *mp = ip->i_mount; | |
93407472 | 373 | ssize_t count = i_blocksize(inode); |
a206c817 CH |
374 | xfs_fileoff_t offset_fsb, end_fsb; |
375 | int error = 0; | |
a206c817 CH |
376 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
377 | int nimaps = 1; | |
378 | ||
379 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 380 | return -EIO; |
a206c817 | 381 | |
70c57dcd DW |
382 | /* |
383 | * Truncate can race with writeback since writeback doesn't take the | |
384 | * iolock and truncate decreases the file size before it starts | |
385 | * truncating the pages between new_size and old_size. Therefore, we | |
386 | * can end up in the situation where writeback gets a CoW fork mapping | |
387 | * but the truncate makes the mapping invalid and we end up in here | |
388 | * trying to get a new mapping. Bail out here so that we simply never | |
389 | * get a valid mapping and so we drop the write altogether. The page | |
390 | * truncation will kill the contents anyway. | |
391 | */ | |
392 | if (type == XFS_IO_COW && offset > i_size_read(inode)) | |
393 | return 0; | |
394 | ||
ef473667 | 395 | ASSERT(type != XFS_IO_COW); |
0d882a36 | 396 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 397 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 398 | |
988ef927 | 399 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
400 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
401 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 402 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 403 | |
b4d8ad7f | 404 | if (offset > mp->m_super->s_maxbytes - count) |
d2c28191 | 405 | count = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
406 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
407 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
408 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
409 | imap, &nimaps, bmapi_flags); | |
ef473667 DW |
410 | /* |
411 | * Truncate an overwrite extent if there's a pending CoW | |
412 | * reservation before the end of this extent. This forces us | |
413 | * to come back to writepage to take care of the CoW. | |
414 | */ | |
415 | if (nimaps && type == XFS_IO_OVERWRITE) | |
416 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); | |
8ff2957d | 417 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 418 | |
8ff2957d | 419 | if (error) |
2451337d | 420 | return error; |
a206c817 | 421 | |
0d882a36 | 422 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 423 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
60b4984f DW |
424 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
425 | imap); | |
a206c817 | 426 | if (!error) |
ef473667 | 427 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
2451337d | 428 | return error; |
a206c817 CH |
429 | } |
430 | ||
8ff2957d | 431 | #ifdef DEBUG |
0d882a36 | 432 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
433 | ASSERT(nimaps); |
434 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
435 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
436 | } | |
437 | #endif | |
438 | if (nimaps) | |
439 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
440 | return 0; | |
1da177e4 LT |
441 | } |
442 | ||
fbcc0256 | 443 | STATIC bool |
558e6891 | 444 | xfs_imap_valid( |
8699bb0a | 445 | struct inode *inode, |
207d0416 | 446 | struct xfs_bmbt_irec *imap, |
558e6891 | 447 | xfs_off_t offset) |
1da177e4 | 448 | { |
558e6891 | 449 | offset >>= inode->i_blkbits; |
8699bb0a | 450 | |
40214d12 BF |
451 | /* |
452 | * We have to make sure the cached mapping is within EOF to protect | |
453 | * against eofblocks trimming on file release leaving us with a stale | |
454 | * mapping. Otherwise, a page for a subsequent file extending buffered | |
455 | * write could get picked up by this writeback cycle and written to the | |
456 | * wrong blocks. | |
457 | * | |
458 | * Note that what we really want here is a generic mapping invalidation | |
459 | * mechanism to protect us from arbitrary extent modifying contexts, not | |
460 | * just eofblocks. | |
461 | */ | |
462 | xfs_trim_extent_eof(imap, XFS_I(inode)); | |
463 | ||
558e6891 CH |
464 | return offset >= imap->br_startoff && |
465 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
466 | } |
467 | ||
f6d6d4fc CH |
468 | STATIC void |
469 | xfs_start_buffer_writeback( | |
470 | struct buffer_head *bh) | |
471 | { | |
472 | ASSERT(buffer_mapped(bh)); | |
473 | ASSERT(buffer_locked(bh)); | |
474 | ASSERT(!buffer_delay(bh)); | |
475 | ASSERT(!buffer_unwritten(bh)); | |
476 | ||
8353a814 CH |
477 | bh->b_end_io = NULL; |
478 | set_buffer_async_write(bh); | |
f6d6d4fc CH |
479 | set_buffer_uptodate(bh); |
480 | clear_buffer_dirty(bh); | |
481 | } | |
482 | ||
483 | STATIC void | |
484 | xfs_start_page_writeback( | |
485 | struct page *page, | |
e10de372 | 486 | int clear_dirty) |
f6d6d4fc CH |
487 | { |
488 | ASSERT(PageLocked(page)); | |
489 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
490 | |
491 | /* | |
492 | * if the page was not fully cleaned, we need to ensure that the higher | |
493 | * layers come back to it correctly. That means we need to keep the page | |
494 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
495 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
496 | * write this page in this writeback sweep will be made. | |
497 | */ | |
498 | if (clear_dirty) { | |
92132021 | 499 | clear_page_dirty_for_io(page); |
0d085a52 DC |
500 | set_page_writeback(page); |
501 | } else | |
502 | set_page_writeback_keepwrite(page); | |
503 | ||
f6d6d4fc | 504 | unlock_page(page); |
f6d6d4fc CH |
505 | } |
506 | ||
c7c1a7d8 | 507 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
508 | { |
509 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
510 | } | |
511 | ||
512 | /* | |
bb18782a DC |
513 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
514 | * it, and we submit that bio. The ioend may be used for multiple bio | |
515 | * submissions, so we only want to allocate an append transaction for the ioend | |
516 | * once. In the case of multiple bio submission, each bio will take an IO | |
517 | * reference to the ioend to ensure that the ioend completion is only done once | |
518 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
519 | * |
520 | * If @fail is non-zero, it means that we have a situation where some part of | |
521 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
522 | * and unlocked them. In this situation, we need to fail the bio and ioend |
523 | * rather than submit it to IO. This typically only happens on a filesystem | |
524 | * shutdown. | |
f6d6d4fc | 525 | */ |
e10de372 | 526 | STATIC int |
f6d6d4fc | 527 | xfs_submit_ioend( |
06342cf8 | 528 | struct writeback_control *wbc, |
0e51a8e1 | 529 | struct xfs_ioend *ioend, |
e10de372 | 530 | int status) |
f6d6d4fc | 531 | { |
5eda4300 DW |
532 | /* Convert CoW extents to regular */ |
533 | if (!status && ioend->io_type == XFS_IO_COW) { | |
4a2d01b0 DC |
534 | /* |
535 | * Yuk. This can do memory allocation, but is not a | |
536 | * transactional operation so everything is done in GFP_KERNEL | |
537 | * context. That can deadlock, because we hold pages in | |
538 | * writeback state and GFP_KERNEL allocations can block on them. | |
539 | * Hence we must operate in nofs conditions here. | |
540 | */ | |
541 | unsigned nofs_flag; | |
542 | ||
543 | nofs_flag = memalloc_nofs_save(); | |
5eda4300 DW |
544 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), |
545 | ioend->io_offset, ioend->io_size); | |
4a2d01b0 | 546 | memalloc_nofs_restore(nofs_flag); |
5eda4300 DW |
547 | } |
548 | ||
e10de372 DC |
549 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
550 | if (!status && | |
0e51a8e1 | 551 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
552 | xfs_ioend_is_append(ioend) && |
553 | !ioend->io_append_trans) | |
e10de372 | 554 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 555 | |
0e51a8e1 CH |
556 | ioend->io_bio->bi_private = ioend; |
557 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
7637241e | 558 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
70fd7614 | 559 | |
e10de372 DC |
560 | /* |
561 | * If we are failing the IO now, just mark the ioend with an | |
562 | * error and finish it. This will run IO completion immediately | |
563 | * as there is only one reference to the ioend at this point in | |
564 | * time. | |
565 | */ | |
566 | if (status) { | |
4e4cbee9 | 567 | ioend->io_bio->bi_status = errno_to_blk_status(status); |
0e51a8e1 | 568 | bio_endio(ioend->io_bio); |
e10de372 DC |
569 | return status; |
570 | } | |
d88992f6 | 571 | |
31d7d58d | 572 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 573 | submit_bio(ioend->io_bio); |
e10de372 | 574 | return 0; |
f6d6d4fc | 575 | } |
f6d6d4fc | 576 | |
0e51a8e1 CH |
577 | static void |
578 | xfs_init_bio_from_bh( | |
579 | struct bio *bio, | |
580 | struct buffer_head *bh) | |
581 | { | |
582 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
74d46992 | 583 | bio_set_dev(bio, bh->b_bdev); |
0e51a8e1 | 584 | } |
7bf7f352 | 585 | |
0e51a8e1 CH |
586 | static struct xfs_ioend * |
587 | xfs_alloc_ioend( | |
588 | struct inode *inode, | |
589 | unsigned int type, | |
590 | xfs_off_t offset, | |
591 | struct buffer_head *bh) | |
592 | { | |
593 | struct xfs_ioend *ioend; | |
594 | struct bio *bio; | |
f6d6d4fc | 595 | |
e292d7bc | 596 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, &xfs_ioend_bioset); |
0e51a8e1 CH |
597 | xfs_init_bio_from_bh(bio, bh); |
598 | ||
599 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
600 | INIT_LIST_HEAD(&ioend->io_list); | |
601 | ioend->io_type = type; | |
602 | ioend->io_inode = inode; | |
603 | ioend->io_size = 0; | |
604 | ioend->io_offset = offset; | |
605 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
606 | ioend->io_append_trans = NULL; | |
607 | ioend->io_bio = bio; | |
608 | return ioend; | |
609 | } | |
610 | ||
611 | /* | |
612 | * Allocate a new bio, and chain the old bio to the new one. | |
613 | * | |
614 | * Note that we have to do perform the chaining in this unintuitive order | |
615 | * so that the bi_private linkage is set up in the right direction for the | |
616 | * traversal in xfs_destroy_ioend(). | |
617 | */ | |
618 | static void | |
619 | xfs_chain_bio( | |
620 | struct xfs_ioend *ioend, | |
621 | struct writeback_control *wbc, | |
622 | struct buffer_head *bh) | |
623 | { | |
624 | struct bio *new; | |
625 | ||
626 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
627 | xfs_init_bio_from_bh(new, bh); | |
628 | ||
629 | bio_chain(ioend->io_bio, new); | |
630 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
7637241e | 631 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
31d7d58d | 632 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 633 | submit_bio(ioend->io_bio); |
0e51a8e1 | 634 | ioend->io_bio = new; |
f6d6d4fc CH |
635 | } |
636 | ||
637 | /* | |
638 | * Test to see if we've been building up a completion structure for | |
639 | * earlier buffers -- if so, we try to append to this ioend if we | |
640 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
641 | * Return the ioend we finished off so that the caller can submit it |
642 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
643 | */ |
644 | STATIC void | |
645 | xfs_add_to_ioend( | |
646 | struct inode *inode, | |
647 | struct buffer_head *bh, | |
7336cea8 | 648 | xfs_off_t offset, |
e10de372 | 649 | struct xfs_writepage_ctx *wpc, |
bb18782a | 650 | struct writeback_control *wbc, |
e10de372 | 651 | struct list_head *iolist) |
f6d6d4fc | 652 | { |
fbcc0256 | 653 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
654 | bh->b_blocknr != wpc->last_block + 1 || |
655 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
656 | if (wpc->ioend) |
657 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 658 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
659 | } |
660 | ||
0e51a8e1 CH |
661 | /* |
662 | * If the buffer doesn't fit into the bio we need to allocate a new | |
663 | * one. This shouldn't happen more than once for a given buffer. | |
664 | */ | |
665 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
666 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 667 | |
fbcc0256 DC |
668 | wpc->ioend->io_size += bh->b_size; |
669 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 670 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
671 | } |
672 | ||
87cbc49c NS |
673 | STATIC void |
674 | xfs_map_buffer( | |
046f1685 | 675 | struct inode *inode, |
87cbc49c | 676 | struct buffer_head *bh, |
207d0416 | 677 | struct xfs_bmbt_irec *imap, |
046f1685 | 678 | xfs_off_t offset) |
87cbc49c NS |
679 | { |
680 | sector_t bn; | |
8699bb0a | 681 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
682 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
683 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 684 | |
207d0416 CH |
685 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
686 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 687 | |
e513182d | 688 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 689 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 690 | |
046f1685 | 691 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
692 | |
693 | bh->b_blocknr = bn; | |
694 | set_buffer_mapped(bh); | |
695 | } | |
696 | ||
1da177e4 LT |
697 | STATIC void |
698 | xfs_map_at_offset( | |
046f1685 | 699 | struct inode *inode, |
1da177e4 | 700 | struct buffer_head *bh, |
207d0416 | 701 | struct xfs_bmbt_irec *imap, |
046f1685 | 702 | xfs_off_t offset) |
1da177e4 | 703 | { |
207d0416 CH |
704 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
705 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 706 | |
207d0416 | 707 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
708 | set_buffer_mapped(bh); |
709 | clear_buffer_delay(bh); | |
f6d6d4fc | 710 | clear_buffer_unwritten(bh); |
1da177e4 LT |
711 | } |
712 | ||
1da177e4 | 713 | /* |
a49935f2 DC |
714 | * Test if a given page contains at least one buffer of a given @type. |
715 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
716 | * try to find one of the type passed in. If it is not set, then the caller only | |
717 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 718 | */ |
a49935f2 | 719 | STATIC bool |
6ffc4db5 | 720 | xfs_check_page_type( |
10ce4444 | 721 | struct page *page, |
a49935f2 DC |
722 | unsigned int type, |
723 | bool check_all_buffers) | |
1da177e4 | 724 | { |
a49935f2 DC |
725 | struct buffer_head *bh; |
726 | struct buffer_head *head; | |
1da177e4 | 727 | |
a49935f2 DC |
728 | if (PageWriteback(page)) |
729 | return false; | |
730 | if (!page->mapping) | |
731 | return false; | |
732 | if (!page_has_buffers(page)) | |
733 | return false; | |
1da177e4 | 734 | |
a49935f2 DC |
735 | bh = head = page_buffers(page); |
736 | do { | |
737 | if (buffer_unwritten(bh)) { | |
738 | if (type == XFS_IO_UNWRITTEN) | |
739 | return true; | |
740 | } else if (buffer_delay(bh)) { | |
805eeb8e | 741 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
742 | return true; |
743 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 744 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
745 | return true; |
746 | } | |
1da177e4 | 747 | |
a49935f2 DC |
748 | /* If we are only checking the first buffer, we are done now. */ |
749 | if (!check_all_buffers) | |
750 | break; | |
751 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 752 | |
a49935f2 | 753 | return false; |
1da177e4 LT |
754 | } |
755 | ||
3ed3a434 DC |
756 | STATIC void |
757 | xfs_vm_invalidatepage( | |
758 | struct page *page, | |
d47992f8 LC |
759 | unsigned int offset, |
760 | unsigned int length) | |
3ed3a434 | 761 | { |
34097dfe LC |
762 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
763 | length); | |
793d7dbe DC |
764 | |
765 | /* | |
766 | * If we are invalidating the entire page, clear the dirty state from it | |
767 | * so that we can check for attempts to release dirty cached pages in | |
768 | * xfs_vm_releasepage(). | |
769 | */ | |
770 | if (offset == 0 && length >= PAGE_SIZE) | |
771 | cancel_dirty_page(page); | |
34097dfe | 772 | block_invalidatepage(page, offset, length); |
3ed3a434 DC |
773 | } |
774 | ||
775 | /* | |
776 | * If the page has delalloc buffers on it, we need to punch them out before we | |
777 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
778 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
779 | * is done on that same region - the delalloc extent is returned when none is | |
780 | * supposed to be there. | |
781 | * | |
782 | * We prevent this by truncating away the delalloc regions on the page before | |
783 | * invalidating it. Because they are delalloc, we can do this without needing a | |
784 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
785 | * truncation without a transaction as there is no space left for block | |
786 | * reservation (typically why we see a ENOSPC in writeback). | |
787 | * | |
788 | * This is not a performance critical path, so for now just do the punching a | |
789 | * buffer head at a time. | |
790 | */ | |
791 | STATIC void | |
792 | xfs_aops_discard_page( | |
793 | struct page *page) | |
794 | { | |
795 | struct inode *inode = page->mapping->host; | |
796 | struct xfs_inode *ip = XFS_I(inode); | |
797 | struct buffer_head *bh, *head; | |
798 | loff_t offset = page_offset(page); | |
3ed3a434 | 799 | |
a49935f2 | 800 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
801 | goto out_invalidate; |
802 | ||
e8c3753c DC |
803 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
804 | goto out_invalidate; | |
805 | ||
4f10700a | 806 | xfs_alert(ip->i_mount, |
c9690043 | 807 | "page discard on page "PTR_FMT", inode 0x%llx, offset %llu.", |
3ed3a434 DC |
808 | page, ip->i_ino, offset); |
809 | ||
810 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
811 | bh = head = page_buffers(page); | |
812 | do { | |
3ed3a434 | 813 | int error; |
c726de44 | 814 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
815 | |
816 | if (!buffer_delay(bh)) | |
817 | goto next_buffer; | |
818 | ||
c726de44 DC |
819 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
820 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
821 | if (error) { |
822 | /* something screwed, just bail */ | |
e8c3753c | 823 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 824 | xfs_alert(ip->i_mount, |
3ed3a434 | 825 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 826 | } |
3ed3a434 DC |
827 | break; |
828 | } | |
829 | next_buffer: | |
93407472 | 830 | offset += i_blocksize(inode); |
3ed3a434 DC |
831 | |
832 | } while ((bh = bh->b_this_page) != head); | |
833 | ||
834 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
835 | out_invalidate: | |
09cbfeaf | 836 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
837 | return; |
838 | } | |
839 | ||
ef473667 DW |
840 | static int |
841 | xfs_map_cow( | |
842 | struct xfs_writepage_ctx *wpc, | |
843 | struct inode *inode, | |
844 | loff_t offset, | |
845 | unsigned int *new_type) | |
846 | { | |
847 | struct xfs_inode *ip = XFS_I(inode); | |
848 | struct xfs_bmbt_irec imap; | |
092d5d9d | 849 | bool is_cow = false; |
ef473667 DW |
850 | int error; |
851 | ||
852 | /* | |
853 | * If we already have a valid COW mapping keep using it. | |
854 | */ | |
855 | if (wpc->io_type == XFS_IO_COW) { | |
856 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
857 | if (wpc->imap_valid) { | |
858 | *new_type = XFS_IO_COW; | |
859 | return 0; | |
860 | } | |
861 | } | |
862 | ||
863 | /* | |
864 | * Else we need to check if there is a COW mapping at this offset. | |
865 | */ | |
866 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
092d5d9d | 867 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap); |
ef473667 DW |
868 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
869 | ||
870 | if (!is_cow) | |
871 | return 0; | |
872 | ||
873 | /* | |
874 | * And if the COW mapping has a delayed extent here we need to | |
875 | * allocate real space for it now. | |
876 | */ | |
092d5d9d | 877 | if (isnullstartblock(imap.br_startblock)) { |
ef473667 DW |
878 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, |
879 | &imap); | |
880 | if (error) | |
881 | return error; | |
882 | } | |
883 | ||
884 | wpc->io_type = *new_type = XFS_IO_COW; | |
885 | wpc->imap_valid = true; | |
886 | wpc->imap = imap; | |
887 | return 0; | |
888 | } | |
889 | ||
e10de372 DC |
890 | /* |
891 | * We implement an immediate ioend submission policy here to avoid needing to | |
892 | * chain multiple ioends and hence nest mempool allocations which can violate | |
893 | * forward progress guarantees we need to provide. The current ioend we are | |
894 | * adding buffers to is cached on the writepage context, and if the new buffer | |
895 | * does not append to the cached ioend it will create a new ioend and cache that | |
896 | * instead. | |
897 | * | |
898 | * If a new ioend is created and cached, the old ioend is returned and queued | |
899 | * locally for submission once the entire page is processed or an error has been | |
900 | * detected. While ioends are submitted immediately after they are completed, | |
901 | * batching optimisations are provided by higher level block plugging. | |
902 | * | |
903 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
904 | * writepage context that the caller will need to submit. | |
905 | */ | |
bfce7d2e DC |
906 | static int |
907 | xfs_writepage_map( | |
908 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 909 | struct writeback_control *wbc, |
bfce7d2e DC |
910 | struct inode *inode, |
911 | struct page *page, | |
2d5f4b5b | 912 | uint64_t end_offset) |
bfce7d2e | 913 | { |
e10de372 DC |
914 | LIST_HEAD(submit_list); |
915 | struct xfs_ioend *ioend, *next; | |
bfce7d2e | 916 | struct buffer_head *bh, *head; |
93407472 | 917 | ssize_t len = i_blocksize(inode); |
2d5f4b5b | 918 | uint64_t offset; |
bfce7d2e | 919 | int error = 0; |
bfce7d2e | 920 | int count = 0; |
e10de372 | 921 | int uptodate = 1; |
ef473667 | 922 | unsigned int new_type; |
bfce7d2e DC |
923 | |
924 | bh = head = page_buffers(page); | |
925 | offset = page_offset(page); | |
bfce7d2e DC |
926 | do { |
927 | if (offset >= end_offset) | |
928 | break; | |
929 | if (!buffer_uptodate(bh)) | |
930 | uptodate = 0; | |
931 | ||
932 | /* | |
933 | * set_page_dirty dirties all buffers in a page, independent | |
934 | * of their state. The dirty state however is entirely | |
935 | * meaningless for holes (!mapped && uptodate), so skip | |
936 | * buffers covering holes here. | |
937 | */ | |
938 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
939 | wpc->imap_valid = false; | |
940 | continue; | |
941 | } | |
942 | ||
ef473667 DW |
943 | if (buffer_unwritten(bh)) |
944 | new_type = XFS_IO_UNWRITTEN; | |
945 | else if (buffer_delay(bh)) | |
946 | new_type = XFS_IO_DELALLOC; | |
947 | else if (buffer_uptodate(bh)) | |
948 | new_type = XFS_IO_OVERWRITE; | |
949 | else { | |
bfce7d2e DC |
950 | if (PageUptodate(page)) |
951 | ASSERT(buffer_mapped(bh)); | |
952 | /* | |
953 | * This buffer is not uptodate and will not be | |
954 | * written to disk. Ensure that we will put any | |
955 | * subsequent writeable buffers into a new | |
956 | * ioend. | |
957 | */ | |
958 | wpc->imap_valid = false; | |
959 | continue; | |
960 | } | |
961 | ||
ef473667 DW |
962 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
963 | error = xfs_map_cow(wpc, inode, offset, &new_type); | |
964 | if (error) | |
965 | goto out; | |
966 | } | |
967 | ||
968 | if (wpc->io_type != new_type) { | |
969 | wpc->io_type = new_type; | |
970 | wpc->imap_valid = false; | |
971 | } | |
972 | ||
bfce7d2e DC |
973 | if (wpc->imap_valid) |
974 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
975 | offset); | |
976 | if (!wpc->imap_valid) { | |
977 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
978 | wpc->io_type); | |
979 | if (error) | |
e10de372 | 980 | goto out; |
bfce7d2e DC |
981 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
982 | offset); | |
983 | } | |
984 | if (wpc->imap_valid) { | |
985 | lock_buffer(bh); | |
986 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
987 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 988 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
989 | count++; |
990 | } | |
991 | ||
bfce7d2e DC |
992 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
993 | ||
994 | if (uptodate && bh == head) | |
995 | SetPageUptodate(page); | |
996 | ||
e10de372 | 997 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 998 | |
e10de372 | 999 | out: |
bfce7d2e | 1000 | /* |
e10de372 DC |
1001 | * On error, we have to fail the ioend here because we have locked |
1002 | * buffers in the ioend. If we don't do this, we'll deadlock | |
1003 | * invalidating the page as that tries to lock the buffers on the page. | |
1004 | * Also, because we may have set pages under writeback, we have to make | |
1005 | * sure we run IO completion to mark the error state of the IO | |
1006 | * appropriately, so we can't cancel the ioend directly here. That means | |
1007 | * we have to mark this page as under writeback if we included any | |
1008 | * buffers from it in the ioend chain so that completion treats it | |
1009 | * correctly. | |
bfce7d2e | 1010 | * |
e10de372 DC |
1011 | * If we didn't include the page in the ioend, the on error we can |
1012 | * simply discard and unlock it as there are no other users of the page | |
1013 | * or it's buffers right now. The caller will still need to trigger | |
1014 | * submission of outstanding ioends on the writepage context so they are | |
1015 | * treated correctly on error. | |
bfce7d2e | 1016 | */ |
e10de372 DC |
1017 | if (count) { |
1018 | xfs_start_page_writeback(page, !error); | |
1019 | ||
1020 | /* | |
1021 | * Preserve the original error if there was one, otherwise catch | |
1022 | * submission errors here and propagate into subsequent ioend | |
1023 | * submissions. | |
1024 | */ | |
1025 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
1026 | int error2; | |
1027 | ||
1028 | list_del_init(&ioend->io_list); | |
1029 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
1030 | if (error2 && !error) | |
1031 | error = error2; | |
1032 | } | |
1033 | } else if (error) { | |
bfce7d2e DC |
1034 | xfs_aops_discard_page(page); |
1035 | ClearPageUptodate(page); | |
1036 | unlock_page(page); | |
e10de372 DC |
1037 | } else { |
1038 | /* | |
1039 | * We can end up here with no error and nothing to write if we | |
1040 | * race with a partial page truncate on a sub-page block sized | |
1041 | * filesystem. In that case we need to mark the page clean. | |
1042 | */ | |
1043 | xfs_start_page_writeback(page, 1); | |
1044 | end_page_writeback(page); | |
bfce7d2e | 1045 | } |
e10de372 | 1046 | |
bfce7d2e DC |
1047 | mapping_set_error(page->mapping, error); |
1048 | return error; | |
1049 | } | |
1050 | ||
1da177e4 | 1051 | /* |
89f3b363 CH |
1052 | * Write out a dirty page. |
1053 | * | |
1054 | * For delalloc space on the page we need to allocate space and flush it. | |
1055 | * For unwritten space on the page we need to start the conversion to | |
1056 | * regular allocated space. | |
89f3b363 | 1057 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1058 | */ |
1da177e4 | 1059 | STATIC int |
fbcc0256 | 1060 | xfs_do_writepage( |
89f3b363 | 1061 | struct page *page, |
fbcc0256 DC |
1062 | struct writeback_control *wbc, |
1063 | void *data) | |
1da177e4 | 1064 | { |
fbcc0256 | 1065 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 1066 | struct inode *inode = page->mapping->host; |
1da177e4 | 1067 | loff_t offset; |
c8ce540d | 1068 | uint64_t end_offset; |
ad68972a | 1069 | pgoff_t end_index; |
89f3b363 | 1070 | |
34097dfe | 1071 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 1072 | |
20cb52eb CH |
1073 | ASSERT(page_has_buffers(page)); |
1074 | ||
89f3b363 CH |
1075 | /* |
1076 | * Refuse to write the page out if we are called from reclaim context. | |
1077 | * | |
d4f7a5cb CH |
1078 | * This avoids stack overflows when called from deeply used stacks in |
1079 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1080 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 1081 | * |
94054fa3 MG |
1082 | * This should never happen except in the case of a VM regression so |
1083 | * warn about it. | |
89f3b363 | 1084 | */ |
94054fa3 MG |
1085 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
1086 | PF_MEMALLOC)) | |
b5420f23 | 1087 | goto redirty; |
1da177e4 | 1088 | |
89f3b363 | 1089 | /* |
680a647b CH |
1090 | * Given that we do not allow direct reclaim to call us, we should |
1091 | * never be called while in a filesystem transaction. | |
89f3b363 | 1092 | */ |
9070733b | 1093 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) |
b5420f23 | 1094 | goto redirty; |
89f3b363 | 1095 | |
8695d27e | 1096 | /* |
ad68972a DC |
1097 | * Is this page beyond the end of the file? |
1098 | * | |
8695d27e JL |
1099 | * The page index is less than the end_index, adjust the end_offset |
1100 | * to the highest offset that this page should represent. | |
1101 | * ----------------------------------------------------- | |
1102 | * | file mapping | <EOF> | | |
1103 | * ----------------------------------------------------- | |
1104 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1105 | * ^--------------------------------^----------|-------- | |
1106 | * | desired writeback range | see else | | |
1107 | * ---------------------------------^------------------| | |
1108 | */ | |
ad68972a | 1109 | offset = i_size_read(inode); |
09cbfeaf | 1110 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 1111 | if (page->index < end_index) |
09cbfeaf | 1112 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
1113 | else { |
1114 | /* | |
1115 | * Check whether the page to write out is beyond or straddles | |
1116 | * i_size or not. | |
1117 | * ------------------------------------------------------- | |
1118 | * | file mapping | <EOF> | | |
1119 | * ------------------------------------------------------- | |
1120 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1121 | * ^--------------------------------^-----------|--------- | |
1122 | * | | Straddles | | |
1123 | * ---------------------------------^-----------|--------| | |
1124 | */ | |
09cbfeaf | 1125 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
1126 | |
1127 | /* | |
ff9a28f6 JK |
1128 | * Skip the page if it is fully outside i_size, e.g. due to a |
1129 | * truncate operation that is in progress. We must redirty the | |
1130 | * page so that reclaim stops reclaiming it. Otherwise | |
1131 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1132 | * |
1133 | * Note that the end_index is unsigned long, it would overflow | |
1134 | * if the given offset is greater than 16TB on 32-bit system | |
1135 | * and if we do check the page is fully outside i_size or not | |
1136 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1137 | * will be evaluated to 0. Hence this page will be redirtied | |
1138 | * and be written out repeatedly which would result in an | |
1139 | * infinite loop, the user program that perform this operation | |
1140 | * will hang. Instead, we can verify this situation by checking | |
1141 | * if the page to write is totally beyond the i_size or if it's | |
1142 | * offset is just equal to the EOF. | |
6b7a03f0 | 1143 | */ |
8695d27e JL |
1144 | if (page->index > end_index || |
1145 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1146 | goto redirty; |
6b7a03f0 CH |
1147 | |
1148 | /* | |
1149 | * The page straddles i_size. It must be zeroed out on each | |
1150 | * and every writepage invocation because it may be mmapped. | |
1151 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1152 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1153 | * memory is zeroed when mapped, and writes to that region are |
1154 | * not written out to the file." | |
1155 | */ | |
09cbfeaf | 1156 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
1157 | |
1158 | /* Adjust the end_offset to the end of file */ | |
1159 | end_offset = offset; | |
1da177e4 LT |
1160 | } |
1161 | ||
2d5f4b5b | 1162 | return xfs_writepage_map(wpc, wbc, inode, page, end_offset); |
f51623b2 | 1163 | |
b5420f23 | 1164 | redirty: |
f51623b2 NS |
1165 | redirty_page_for_writepage(wbc, page); |
1166 | unlock_page(page); | |
1167 | return 0; | |
f51623b2 NS |
1168 | } |
1169 | ||
fbcc0256 DC |
1170 | STATIC int |
1171 | xfs_vm_writepage( | |
1172 | struct page *page, | |
1173 | struct writeback_control *wbc) | |
1174 | { | |
1175 | struct xfs_writepage_ctx wpc = { | |
1176 | .io_type = XFS_IO_INVALID, | |
1177 | }; | |
1178 | int ret; | |
1179 | ||
1180 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1181 | if (wpc.ioend) |
1182 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1183 | return ret; | |
fbcc0256 DC |
1184 | } |
1185 | ||
7d4fb40a NS |
1186 | STATIC int |
1187 | xfs_vm_writepages( | |
1188 | struct address_space *mapping, | |
1189 | struct writeback_control *wbc) | |
1190 | { | |
fbcc0256 DC |
1191 | struct xfs_writepage_ctx wpc = { |
1192 | .io_type = XFS_IO_INVALID, | |
1193 | }; | |
1194 | int ret; | |
1195 | ||
b3aea4ed | 1196 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
fbcc0256 | 1197 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1198 | if (wpc.ioend) |
1199 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1200 | return ret; | |
7d4fb40a NS |
1201 | } |
1202 | ||
6e2608df DW |
1203 | STATIC int |
1204 | xfs_dax_writepages( | |
1205 | struct address_space *mapping, | |
1206 | struct writeback_control *wbc) | |
1207 | { | |
1208 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); | |
1209 | return dax_writeback_mapping_range(mapping, | |
1210 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1211 | } | |
1212 | ||
f51623b2 NS |
1213 | /* |
1214 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1215 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1216 | * have buffer heads in this call. |
1217 | * | |
89f3b363 | 1218 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1219 | */ |
1220 | STATIC int | |
238f4c54 | 1221 | xfs_vm_releasepage( |
f51623b2 NS |
1222 | struct page *page, |
1223 | gfp_t gfp_mask) | |
1224 | { | |
20cb52eb | 1225 | int delalloc, unwritten; |
f51623b2 | 1226 | |
34097dfe | 1227 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1228 | |
99579cce BF |
1229 | /* |
1230 | * mm accommodates an old ext3 case where clean pages might not have had | |
1231 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1232 | * ->releasepage() via shrink_active_list(). Conversely, | |
793d7dbe DC |
1233 | * block_invalidatepage() can send pages that are still marked dirty but |
1234 | * otherwise have invalidated buffers. | |
99579cce | 1235 | * |
0a417b8d | 1236 | * We want to release the latter to avoid unnecessary buildup of the |
793d7dbe DC |
1237 | * LRU, so xfs_vm_invalidatepage() clears the page dirty flag on pages |
1238 | * that are entirely invalidated and need to be released. Hence the | |
1239 | * only time we should get dirty pages here is through | |
1240 | * shrink_active_list() and so we can simply skip those now. | |
1241 | * | |
1242 | * warn if we've left any lingering delalloc/unwritten buffers on clean | |
1243 | * or invalidated pages we are about to release. | |
99579cce | 1244 | */ |
793d7dbe DC |
1245 | if (PageDirty(page)) |
1246 | return 0; | |
1247 | ||
20cb52eb | 1248 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1249 | |
793d7dbe | 1250 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1251 | return 0; |
793d7dbe | 1252 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1253 | return 0; |
1254 | ||
f51623b2 NS |
1255 | return try_to_free_buffers(page); |
1256 | } | |
1257 | ||
1fdca9c2 DC |
1258 | /* |
1259 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1260 | * is, so that we can avoid repeated get_blocks calls. | |
1261 | * | |
1262 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1263 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1264 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1265 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1266 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1267 | * to and including EOF, and a second mapping for beyond EOF. | |
1268 | */ | |
1269 | static void | |
1270 | xfs_map_trim_size( | |
1271 | struct inode *inode, | |
1272 | sector_t iblock, | |
1273 | struct buffer_head *bh_result, | |
1274 | struct xfs_bmbt_irec *imap, | |
1275 | xfs_off_t offset, | |
1276 | ssize_t size) | |
1277 | { | |
1278 | xfs_off_t mapping_size; | |
1279 | ||
1280 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1281 | mapping_size <<= inode->i_blkbits; | |
1282 | ||
1283 | ASSERT(mapping_size > 0); | |
1284 | if (mapping_size > size) | |
1285 | mapping_size = size; | |
1286 | if (offset < i_size_read(inode) && | |
22a6c837 | 1287 | (xfs_ufsize_t)offset + mapping_size >= i_size_read(inode)) { |
1fdca9c2 DC |
1288 | /* limit mapping to block that spans EOF */ |
1289 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
93407472 | 1290 | i_blocksize(inode)); |
1fdca9c2 DC |
1291 | } |
1292 | if (mapping_size > LONG_MAX) | |
1293 | mapping_size = LONG_MAX; | |
1294 | ||
1295 | bh_result->b_size = mapping_size; | |
1296 | } | |
1297 | ||
0613f16c | 1298 | static int |
acdda3aa | 1299 | xfs_get_blocks( |
1da177e4 LT |
1300 | struct inode *inode, |
1301 | sector_t iblock, | |
1da177e4 | 1302 | struct buffer_head *bh_result, |
acdda3aa | 1303 | int create) |
1da177e4 | 1304 | { |
a206c817 CH |
1305 | struct xfs_inode *ip = XFS_I(inode); |
1306 | struct xfs_mount *mp = ip->i_mount; | |
1307 | xfs_fileoff_t offset_fsb, end_fsb; | |
1308 | int error = 0; | |
1309 | int lockmode = 0; | |
207d0416 | 1310 | struct xfs_bmbt_irec imap; |
a206c817 | 1311 | int nimaps = 1; |
fdc7ed75 NS |
1312 | xfs_off_t offset; |
1313 | ssize_t size; | |
a206c817 | 1314 | |
acdda3aa | 1315 | BUG_ON(create); |
6e8a27a8 | 1316 | |
a206c817 | 1317 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1318 | return -EIO; |
1da177e4 | 1319 | |
fdc7ed75 | 1320 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
93407472 | 1321 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
c2536668 | 1322 | size = bh_result->b_size; |
364f358a | 1323 | |
acdda3aa | 1324 | if (offset >= i_size_read(inode)) |
364f358a LM |
1325 | return 0; |
1326 | ||
507630b2 DC |
1327 | /* |
1328 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1329 | * a block mapping without an exclusive lock first. |
507630b2 | 1330 | */ |
6e8a27a8 | 1331 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1332 | |
d2c28191 | 1333 | ASSERT(offset <= mp->m_super->s_maxbytes); |
b4d8ad7f | 1334 | if (offset > mp->m_super->s_maxbytes - size) |
d2c28191 | 1335 | size = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
1336 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1337 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1338 | ||
7d9df3c1 CH |
1339 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap, |
1340 | &nimaps, 0); | |
1da177e4 | 1341 | if (error) |
a206c817 | 1342 | goto out_unlock; |
1d4352de | 1343 | if (!nimaps) { |
a206c817 CH |
1344 | trace_xfs_get_blocks_notfound(ip, offset, size); |
1345 | goto out_unlock; | |
1346 | } | |
1da177e4 | 1347 | |
1d4352de CH |
1348 | trace_xfs_get_blocks_found(ip, offset, size, |
1349 | imap.br_state == XFS_EXT_UNWRITTEN ? | |
1350 | XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap); | |
1351 | xfs_iunlock(ip, lockmode); | |
1352 | ||
1fdca9c2 | 1353 | /* trim mapping down to size requested */ |
6e8a27a8 | 1354 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1355 | |
a719370b DC |
1356 | /* |
1357 | * For unwritten extents do not report a disk address in the buffered | |
1358 | * read case (treat as if we're reading into a hole). | |
1359 | */ | |
9c4f29d3 | 1360 | if (xfs_bmap_is_real_extent(&imap)) |
a719370b | 1361 | xfs_map_buffer(inode, bh_result, &imap, offset); |
1da177e4 | 1362 | |
c2536668 NS |
1363 | /* |
1364 | * If this is a realtime file, data may be on a different device. | |
1365 | * to that pointed to from the buffer_head b_bdev currently. | |
1366 | */ | |
046f1685 | 1367 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1368 | return 0; |
a206c817 CH |
1369 | |
1370 | out_unlock: | |
1371 | xfs_iunlock(ip, lockmode); | |
2451337d | 1372 | return error; |
1da177e4 LT |
1373 | } |
1374 | ||
1da177e4 | 1375 | STATIC sector_t |
e4c573bb | 1376 | xfs_vm_bmap( |
1da177e4 LT |
1377 | struct address_space *mapping, |
1378 | sector_t block) | |
1379 | { | |
b84e7722 | 1380 | struct xfs_inode *ip = XFS_I(mapping->host); |
1da177e4 | 1381 | |
b84e7722 | 1382 | trace_xfs_vm_bmap(ip); |
db1327b1 DW |
1383 | |
1384 | /* | |
1385 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
793057e1 | 1386 | * bypasses the file system for actual I/O. We really can't allow |
db1327b1 | 1387 | * that on reflinks inodes, so we have to skip out here. And yes, |
eb5e248d DW |
1388 | * 0 is the magic code for a bmap error. |
1389 | * | |
1390 | * Since we don't pass back blockdev info, we can't return bmap | |
1391 | * information for rt files either. | |
db1327b1 | 1392 | */ |
eb5e248d | 1393 | if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) |
db1327b1 | 1394 | return 0; |
b84e7722 | 1395 | return iomap_bmap(mapping, block, &xfs_iomap_ops); |
1da177e4 LT |
1396 | } |
1397 | ||
1398 | STATIC int | |
e4c573bb | 1399 | xfs_vm_readpage( |
1da177e4 LT |
1400 | struct file *unused, |
1401 | struct page *page) | |
1402 | { | |
121e213e | 1403 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1404 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1405 | } |
1406 | ||
1407 | STATIC int | |
e4c573bb | 1408 | xfs_vm_readpages( |
1da177e4 LT |
1409 | struct file *unused, |
1410 | struct address_space *mapping, | |
1411 | struct list_head *pages, | |
1412 | unsigned nr_pages) | |
1413 | { | |
121e213e | 1414 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1415 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1416 | } |
1417 | ||
22e757a4 DC |
1418 | /* |
1419 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1420 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1421 | * dirty, we'll never be able to clean them because we don't write buffers | |
1422 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1423 | * that have been marked dirty. Further, the dirty state can leak into | |
1424 | * the file interior if the file is extended, resulting in all sorts of | |
1425 | * bad things happening as the state does not match the underlying data. | |
1426 | * | |
1427 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1428 | * this only exist because of bufferheads and how the generic code manages them. | |
1429 | */ | |
1430 | STATIC int | |
1431 | xfs_vm_set_page_dirty( | |
1432 | struct page *page) | |
1433 | { | |
1434 | struct address_space *mapping = page->mapping; | |
1435 | struct inode *inode = mapping->host; | |
1436 | loff_t end_offset; | |
1437 | loff_t offset; | |
1438 | int newly_dirty; | |
1439 | ||
1440 | if (unlikely(!mapping)) | |
1441 | return !TestSetPageDirty(page); | |
1442 | ||
1443 | end_offset = i_size_read(inode); | |
1444 | offset = page_offset(page); | |
1445 | ||
1446 | spin_lock(&mapping->private_lock); | |
1447 | if (page_has_buffers(page)) { | |
1448 | struct buffer_head *head = page_buffers(page); | |
1449 | struct buffer_head *bh = head; | |
1450 | ||
1451 | do { | |
1452 | if (offset < end_offset) | |
1453 | set_buffer_dirty(bh); | |
1454 | bh = bh->b_this_page; | |
93407472 | 1455 | offset += i_blocksize(inode); |
22e757a4 DC |
1456 | } while (bh != head); |
1457 | } | |
c4843a75 | 1458 | /* |
81f8c3a4 JW |
1459 | * Lock out page->mem_cgroup migration to keep PageDirty |
1460 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1461 | */ |
62cccb8c | 1462 | lock_page_memcg(page); |
22e757a4 DC |
1463 | newly_dirty = !TestSetPageDirty(page); |
1464 | spin_unlock(&mapping->private_lock); | |
1465 | ||
f82b3764 MW |
1466 | if (newly_dirty) |
1467 | __set_page_dirty(page, mapping, 1); | |
62cccb8c | 1468 | unlock_page_memcg(page); |
c4843a75 GT |
1469 | if (newly_dirty) |
1470 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1471 | return newly_dirty; |
1472 | } | |
1473 | ||
67482129 DW |
1474 | static int |
1475 | xfs_iomap_swapfile_activate( | |
1476 | struct swap_info_struct *sis, | |
1477 | struct file *swap_file, | |
1478 | sector_t *span) | |
1479 | { | |
1480 | sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file)); | |
1481 | return iomap_swapfile_activate(sis, swap_file, span, &xfs_iomap_ops); | |
1482 | } | |
1483 | ||
f5e54d6e | 1484 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1485 | .readpage = xfs_vm_readpage, |
1486 | .readpages = xfs_vm_readpages, | |
1487 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1488 | .writepages = xfs_vm_writepages, |
22e757a4 | 1489 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1490 | .releasepage = xfs_vm_releasepage, |
1491 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb | 1492 | .bmap = xfs_vm_bmap, |
6e2608df | 1493 | .direct_IO = noop_direct_IO, |
e965f963 | 1494 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1495 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1496 | .error_remove_page = generic_error_remove_page, |
67482129 | 1497 | .swap_activate = xfs_iomap_swapfile_activate, |
1da177e4 | 1498 | }; |
6e2608df DW |
1499 | |
1500 | const struct address_space_operations xfs_dax_aops = { | |
1501 | .writepages = xfs_dax_writepages, | |
1502 | .direct_IO = noop_direct_IO, | |
1503 | .set_page_dirty = noop_set_page_dirty, | |
1504 | .invalidatepage = noop_invalidatepage, | |
67482129 | 1505 | .swap_activate = xfs_iomap_swapfile_activate, |
6e2608df | 1506 | }; |