]>
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 | ||
273dda76 CH |
40 | /* flags for direct write completions */ |
41 | #define XFS_DIO_FLAG_UNWRITTEN (1 << 0) | |
42 | #define XFS_DIO_FLAG_APPEND (1 << 1) | |
0613f16c | 43 | #define XFS_DIO_FLAG_COW (1 << 2) |
273dda76 | 44 | |
fbcc0256 DC |
45 | /* |
46 | * structure owned by writepages passed to individual writepage calls | |
47 | */ | |
48 | struct xfs_writepage_ctx { | |
49 | struct xfs_bmbt_irec imap; | |
50 | bool imap_valid; | |
51 | unsigned int io_type; | |
fbcc0256 DC |
52 | struct xfs_ioend *ioend; |
53 | sector_t last_block; | |
54 | }; | |
55 | ||
0b1b213f | 56 | void |
f51623b2 NS |
57 | xfs_count_page_state( |
58 | struct page *page, | |
59 | int *delalloc, | |
f51623b2 NS |
60 | int *unwritten) |
61 | { | |
62 | struct buffer_head *bh, *head; | |
63 | ||
20cb52eb | 64 | *delalloc = *unwritten = 0; |
f51623b2 NS |
65 | |
66 | bh = head = page_buffers(page); | |
67 | do { | |
20cb52eb | 68 | if (buffer_unwritten(bh)) |
f51623b2 NS |
69 | (*unwritten) = 1; |
70 | else if (buffer_delay(bh)) | |
71 | (*delalloc) = 1; | |
72 | } while ((bh = bh->b_this_page) != head); | |
73 | } | |
74 | ||
20a90f58 | 75 | struct block_device * |
6214ed44 | 76 | xfs_find_bdev_for_inode( |
046f1685 | 77 | struct inode *inode) |
6214ed44 | 78 | { |
046f1685 | 79 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
80 | struct xfs_mount *mp = ip->i_mount; |
81 | ||
71ddabb9 | 82 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
83 | return mp->m_rtdev_targp->bt_bdev; |
84 | else | |
85 | return mp->m_ddev_targp->bt_bdev; | |
86 | } | |
87 | ||
f6d6d4fc | 88 | /* |
37992c18 DC |
89 | * We're now finished for good with this page. Update the page state via the |
90 | * associated buffer_heads, paying attention to the start and end offsets that | |
91 | * we need to process on the page. | |
28b783e4 | 92 | * |
772003c6 CH |
93 | * Note that we open code the action in end_buffer_async_write here so that we |
94 | * only have to iterate over the buffers attached to the page once. This is not | |
95 | * only more efficient, but also ensures that we only calls end_page_writeback | |
96 | * at the end of the iteration, and thus avoids the pitfall of having the page | |
97 | * and buffers potentially freed after every call to end_buffer_async_write. | |
37992c18 DC |
98 | */ |
99 | static void | |
100 | xfs_finish_page_writeback( | |
101 | struct inode *inode, | |
102 | struct bio_vec *bvec, | |
103 | int error) | |
104 | { | |
772003c6 CH |
105 | struct buffer_head *head = page_buffers(bvec->bv_page), *bh = head; |
106 | bool busy = false; | |
37992c18 | 107 | unsigned int off = 0; |
772003c6 | 108 | unsigned long flags; |
37992c18 DC |
109 | |
110 | ASSERT(bvec->bv_offset < PAGE_SIZE); | |
61604a26 | 111 | ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0); |
772003c6 | 112 | ASSERT(bvec->bv_offset + bvec->bv_len <= PAGE_SIZE); |
61604a26 | 113 | ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0); |
37992c18 | 114 | |
772003c6 CH |
115 | local_irq_save(flags); |
116 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
37992c18 | 117 | do { |
772003c6 CH |
118 | if (off >= bvec->bv_offset && |
119 | off < bvec->bv_offset + bvec->bv_len) { | |
120 | ASSERT(buffer_async_write(bh)); | |
121 | ASSERT(bh->b_end_io == NULL); | |
122 | ||
123 | if (error) { | |
124 | mapping_set_error(bvec->bv_page->mapping, -EIO); | |
125 | set_buffer_write_io_error(bh); | |
126 | clear_buffer_uptodate(bh); | |
127 | SetPageError(bvec->bv_page); | |
128 | } else { | |
129 | set_buffer_uptodate(bh); | |
130 | } | |
131 | clear_buffer_async_write(bh); | |
132 | unlock_buffer(bh); | |
133 | } else if (buffer_async_write(bh)) { | |
134 | ASSERT(buffer_locked(bh)); | |
135 | busy = true; | |
136 | } | |
137 | off += bh->b_size; | |
138 | } while ((bh = bh->b_this_page) != head); | |
139 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
140 | local_irq_restore(flags); | |
141 | ||
142 | if (!busy) | |
143 | end_page_writeback(bvec->bv_page); | |
37992c18 DC |
144 | } |
145 | ||
146 | /* | |
147 | * We're now finished for good with this ioend structure. Update the page | |
148 | * state, release holds on bios, and finally free up memory. Do not use the | |
149 | * ioend after this. | |
f6d6d4fc | 150 | */ |
0829c360 CH |
151 | STATIC void |
152 | xfs_destroy_ioend( | |
0e51a8e1 CH |
153 | struct xfs_ioend *ioend, |
154 | int error) | |
0829c360 | 155 | { |
37992c18 | 156 | struct inode *inode = ioend->io_inode; |
772003c6 CH |
157 | struct bio *bio = &ioend->io_inline_bio; |
158 | struct bio *last = ioend->io_bio, *next; | |
159 | u64 start = bio->bi_iter.bi_sector; | |
160 | bool quiet = bio_flagged(bio, BIO_QUIET); | |
f6d6d4fc | 161 | |
0e51a8e1 | 162 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
37992c18 DC |
163 | struct bio_vec *bvec; |
164 | int i; | |
165 | ||
0e51a8e1 CH |
166 | /* |
167 | * For the last bio, bi_private points to the ioend, so we | |
168 | * need to explicitly end the iteration here. | |
169 | */ | |
170 | if (bio == last) | |
171 | next = NULL; | |
172 | else | |
173 | next = bio->bi_private; | |
583fa586 | 174 | |
37992c18 DC |
175 | /* walk each page on bio, ending page IO on them */ |
176 | bio_for_each_segment_all(bvec, bio, i) | |
177 | xfs_finish_page_writeback(inode, bvec, error); | |
178 | ||
179 | bio_put(bio); | |
f6d6d4fc | 180 | } |
772003c6 CH |
181 | |
182 | if (unlikely(error && !quiet)) { | |
183 | xfs_err_ratelimited(XFS_I(inode)->i_mount, | |
184 | "writeback error on sector %llu", start); | |
185 | } | |
0829c360 CH |
186 | } |
187 | ||
fc0063c4 CH |
188 | /* |
189 | * Fast and loose check if this write could update the on-disk inode size. | |
190 | */ | |
191 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
192 | { | |
193 | return ioend->io_offset + ioend->io_size > | |
194 | XFS_I(ioend->io_inode)->i_d.di_size; | |
195 | } | |
196 | ||
281627df CH |
197 | STATIC int |
198 | xfs_setfilesize_trans_alloc( | |
199 | struct xfs_ioend *ioend) | |
200 | { | |
201 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
202 | struct xfs_trans *tp; | |
203 | int error; | |
204 | ||
253f4911 CH |
205 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
206 | if (error) | |
281627df | 207 | return error; |
281627df CH |
208 | |
209 | ioend->io_append_trans = tp; | |
210 | ||
d9457dc0 | 211 | /* |
437a255a | 212 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
213 | * we released it. |
214 | */ | |
bee9182d | 215 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
216 | /* |
217 | * We hand off the transaction to the completion thread now, so | |
218 | * clear the flag here. | |
219 | */ | |
220 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
221 | return 0; | |
222 | } | |
223 | ||
ba87ea69 | 224 | /* |
2813d682 | 225 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 226 | */ |
281627df | 227 | STATIC int |
e372843a | 228 | __xfs_setfilesize( |
2ba66237 CH |
229 | struct xfs_inode *ip, |
230 | struct xfs_trans *tp, | |
231 | xfs_off_t offset, | |
232 | size_t size) | |
ba87ea69 | 233 | { |
ba87ea69 | 234 | xfs_fsize_t isize; |
ba87ea69 | 235 | |
aa6bf01d | 236 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 237 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
238 | if (!isize) { |
239 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 240 | xfs_trans_cancel(tp); |
281627df | 241 | return 0; |
ba87ea69 LM |
242 | } |
243 | ||
2ba66237 | 244 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
245 | |
246 | ip->i_d.di_size = isize; | |
247 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
248 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
249 | ||
70393313 | 250 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
251 | } |
252 | ||
e372843a CH |
253 | int |
254 | xfs_setfilesize( | |
255 | struct xfs_inode *ip, | |
256 | xfs_off_t offset, | |
257 | size_t size) | |
258 | { | |
259 | struct xfs_mount *mp = ip->i_mount; | |
260 | struct xfs_trans *tp; | |
261 | int error; | |
262 | ||
263 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); | |
264 | if (error) | |
265 | return error; | |
266 | ||
267 | return __xfs_setfilesize(ip, tp, offset, size); | |
268 | } | |
269 | ||
2ba66237 CH |
270 | STATIC int |
271 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
272 | struct xfs_ioend *ioend, |
273 | int error) | |
2ba66237 CH |
274 | { |
275 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
276 | struct xfs_trans *tp = ioend->io_append_trans; | |
277 | ||
278 | /* | |
279 | * The transaction may have been allocated in the I/O submission thread, | |
280 | * thus we need to mark ourselves as being in a transaction manually. | |
281 | * Similarly for freeze protection. | |
282 | */ | |
283 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 284 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 285 | |
5cb13dcd | 286 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 287 | if (error) { |
5cb13dcd | 288 | xfs_trans_cancel(tp); |
0e51a8e1 | 289 | return error; |
5cb13dcd Z |
290 | } |
291 | ||
e372843a | 292 | return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
2ba66237 CH |
293 | } |
294 | ||
0829c360 | 295 | /* |
5ec4fabb | 296 | * IO write completion. |
f6d6d4fc CH |
297 | */ |
298 | STATIC void | |
5ec4fabb | 299 | xfs_end_io( |
77d7a0c2 | 300 | struct work_struct *work) |
0829c360 | 301 | { |
0e51a8e1 CH |
302 | struct xfs_ioend *ioend = |
303 | container_of(work, struct xfs_ioend, io_work); | |
304 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
d07b5855 CH |
305 | xfs_off_t offset = ioend->io_offset; |
306 | size_t size = ioend->io_size; | |
0e51a8e1 | 307 | int error = ioend->io_bio->bi_error; |
ba87ea69 | 308 | |
af055e37 | 309 | /* |
d07b5855 | 310 | * Just clean up the in-memory strutures if the fs has been shut down. |
af055e37 | 311 | */ |
d07b5855 | 312 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
0e51a8e1 | 313 | error = -EIO; |
d07b5855 CH |
314 | goto done; |
315 | } | |
04f658ee | 316 | |
43caeb18 | 317 | /* |
d07b5855 | 318 | * Clean up any COW blocks on an I/O error. |
43caeb18 | 319 | */ |
d07b5855 CH |
320 | if (unlikely(error)) { |
321 | switch (ioend->io_type) { | |
322 | case XFS_IO_COW: | |
323 | xfs_reflink_cancel_cow_range(ip, offset, size, true); | |
324 | break; | |
43caeb18 | 325 | } |
d07b5855 CH |
326 | |
327 | goto done; | |
43caeb18 DW |
328 | } |
329 | ||
5ec4fabb | 330 | /* |
d07b5855 | 331 | * Success: commit the COW or unwritten blocks if needed. |
5ec4fabb | 332 | */ |
d07b5855 CH |
333 | switch (ioend->io_type) { |
334 | case XFS_IO_COW: | |
335 | error = xfs_reflink_end_cow(ip, offset, size); | |
336 | break; | |
337 | case XFS_IO_UNWRITTEN: | |
0eebfede EG |
338 | /* writeback should never update isize */ |
339 | error = xfs_iomap_write_unwritten(ip, offset, size, false); | |
d07b5855 CH |
340 | break; |
341 | default: | |
342 | ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans); | |
343 | break; | |
5ec4fabb | 344 | } |
ba87ea69 | 345 | |
04f658ee | 346 | done: |
d07b5855 CH |
347 | if (ioend->io_append_trans) |
348 | error = xfs_setfilesize_ioend(ioend, error); | |
0e51a8e1 | 349 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
350 | } |
351 | ||
0e51a8e1 CH |
352 | STATIC void |
353 | xfs_end_bio( | |
354 | struct bio *bio) | |
0829c360 | 355 | { |
0e51a8e1 CH |
356 | struct xfs_ioend *ioend = bio->bi_private; |
357 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 358 | |
43caeb18 | 359 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
0e51a8e1 CH |
360 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
361 | else if (ioend->io_append_trans) | |
362 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
363 | else | |
364 | xfs_destroy_ioend(ioend, bio->bi_error); | |
0829c360 CH |
365 | } |
366 | ||
1da177e4 LT |
367 | STATIC int |
368 | xfs_map_blocks( | |
369 | struct inode *inode, | |
370 | loff_t offset, | |
207d0416 | 371 | struct xfs_bmbt_irec *imap, |
988ef927 | 372 | int type) |
1da177e4 | 373 | { |
a206c817 CH |
374 | struct xfs_inode *ip = XFS_I(inode); |
375 | struct xfs_mount *mp = ip->i_mount; | |
61604a26 | 376 | ssize_t count = i_blocksize(inode); |
a206c817 CH |
377 | xfs_fileoff_t offset_fsb, end_fsb; |
378 | int error = 0; | |
a206c817 CH |
379 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
380 | int nimaps = 1; | |
381 | ||
382 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 383 | return -EIO; |
a206c817 | 384 | |
ef473667 | 385 | ASSERT(type != XFS_IO_COW); |
0d882a36 | 386 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 387 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 388 | |
988ef927 | 389 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
390 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
391 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 392 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 393 | |
d9602444 | 394 | if ((xfs_ufsize_t)offset + count > mp->m_super->s_maxbytes) |
d2c28191 | 395 | count = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
396 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
397 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
398 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
399 | imap, &nimaps, bmapi_flags); | |
ef473667 DW |
400 | /* |
401 | * Truncate an overwrite extent if there's a pending CoW | |
402 | * reservation before the end of this extent. This forces us | |
403 | * to come back to writepage to take care of the CoW. | |
404 | */ | |
405 | if (nimaps && type == XFS_IO_OVERWRITE) | |
406 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); | |
8ff2957d | 407 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 408 | |
8ff2957d | 409 | if (error) |
2451337d | 410 | return error; |
a206c817 | 411 | |
0d882a36 | 412 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 413 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
60b4984f DW |
414 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
415 | imap); | |
a206c817 | 416 | if (!error) |
ef473667 | 417 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
2451337d | 418 | return error; |
a206c817 CH |
419 | } |
420 | ||
8ff2957d | 421 | #ifdef DEBUG |
0d882a36 | 422 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
423 | ASSERT(nimaps); |
424 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
425 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
426 | } | |
427 | #endif | |
428 | if (nimaps) | |
429 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
430 | return 0; | |
1da177e4 LT |
431 | } |
432 | ||
fbcc0256 | 433 | STATIC bool |
558e6891 | 434 | xfs_imap_valid( |
8699bb0a | 435 | struct inode *inode, |
207d0416 | 436 | struct xfs_bmbt_irec *imap, |
558e6891 | 437 | xfs_off_t offset) |
1da177e4 | 438 | { |
558e6891 | 439 | offset >>= inode->i_blkbits; |
8699bb0a | 440 | |
f374505b BF |
441 | /* |
442 | * We have to make sure the cached mapping is within EOF to protect | |
443 | * against eofblocks trimming on file release leaving us with a stale | |
444 | * mapping. Otherwise, a page for a subsequent file extending buffered | |
445 | * write could get picked up by this writeback cycle and written to the | |
446 | * wrong blocks. | |
447 | * | |
448 | * Note that what we really want here is a generic mapping invalidation | |
449 | * mechanism to protect us from arbitrary extent modifying contexts, not | |
450 | * just eofblocks. | |
451 | */ | |
452 | xfs_trim_extent_eof(imap, XFS_I(inode)); | |
453 | ||
558e6891 CH |
454 | return offset >= imap->br_startoff && |
455 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
456 | } |
457 | ||
f6d6d4fc CH |
458 | STATIC void |
459 | xfs_start_buffer_writeback( | |
460 | struct buffer_head *bh) | |
461 | { | |
462 | ASSERT(buffer_mapped(bh)); | |
463 | ASSERT(buffer_locked(bh)); | |
464 | ASSERT(!buffer_delay(bh)); | |
465 | ASSERT(!buffer_unwritten(bh)); | |
466 | ||
772003c6 CH |
467 | bh->b_end_io = NULL; |
468 | set_buffer_async_write(bh); | |
f6d6d4fc CH |
469 | set_buffer_uptodate(bh); |
470 | clear_buffer_dirty(bh); | |
471 | } | |
472 | ||
473 | STATIC void | |
474 | xfs_start_page_writeback( | |
475 | struct page *page, | |
e10de372 | 476 | int clear_dirty) |
f6d6d4fc CH |
477 | { |
478 | ASSERT(PageLocked(page)); | |
479 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
480 | |
481 | /* | |
482 | * if the page was not fully cleaned, we need to ensure that the higher | |
483 | * layers come back to it correctly. That means we need to keep the page | |
484 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
485 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
486 | * write this page in this writeback sweep will be made. | |
487 | */ | |
488 | if (clear_dirty) { | |
92132021 | 489 | clear_page_dirty_for_io(page); |
0d085a52 DC |
490 | set_page_writeback(page); |
491 | } else | |
492 | set_page_writeback_keepwrite(page); | |
493 | ||
f6d6d4fc | 494 | unlock_page(page); |
f6d6d4fc CH |
495 | } |
496 | ||
c7c1a7d8 | 497 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
498 | { |
499 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
500 | } | |
501 | ||
502 | /* | |
bb18782a DC |
503 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
504 | * it, and we submit that bio. The ioend may be used for multiple bio | |
505 | * submissions, so we only want to allocate an append transaction for the ioend | |
506 | * once. In the case of multiple bio submission, each bio will take an IO | |
507 | * reference to the ioend to ensure that the ioend completion is only done once | |
508 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
509 | * |
510 | * If @fail is non-zero, it means that we have a situation where some part of | |
511 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
512 | * and unlocked them. In this situation, we need to fail the bio and ioend |
513 | * rather than submit it to IO. This typically only happens on a filesystem | |
514 | * shutdown. | |
f6d6d4fc | 515 | */ |
e10de372 | 516 | STATIC int |
f6d6d4fc | 517 | xfs_submit_ioend( |
06342cf8 | 518 | struct writeback_control *wbc, |
0e51a8e1 | 519 | struct xfs_ioend *ioend, |
e10de372 | 520 | int status) |
f6d6d4fc | 521 | { |
e02f0ff2 DW |
522 | /* Convert CoW extents to regular */ |
523 | if (!status && ioend->io_type == XFS_IO_COW) { | |
524 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), | |
525 | ioend->io_offset, ioend->io_size); | |
526 | } | |
527 | ||
e10de372 DC |
528 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
529 | if (!status && | |
0e51a8e1 | 530 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
531 | xfs_ioend_is_append(ioend) && |
532 | !ioend->io_append_trans) | |
e10de372 | 533 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 534 | |
0e51a8e1 CH |
535 | ioend->io_bio->bi_private = ioend; |
536 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
50bfcd0c MC |
537 | bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE, |
538 | (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0); | |
e10de372 DC |
539 | /* |
540 | * If we are failing the IO now, just mark the ioend with an | |
541 | * error and finish it. This will run IO completion immediately | |
542 | * as there is only one reference to the ioend at this point in | |
543 | * time. | |
544 | */ | |
545 | if (status) { | |
0e51a8e1 CH |
546 | ioend->io_bio->bi_error = status; |
547 | bio_endio(ioend->io_bio); | |
e10de372 DC |
548 | return status; |
549 | } | |
d88992f6 | 550 | |
4e49ea4a | 551 | submit_bio(ioend->io_bio); |
e10de372 | 552 | return 0; |
f6d6d4fc | 553 | } |
f6d6d4fc | 554 | |
0e51a8e1 CH |
555 | static void |
556 | xfs_init_bio_from_bh( | |
557 | struct bio *bio, | |
558 | struct buffer_head *bh) | |
559 | { | |
560 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
561 | bio->bi_bdev = bh->b_bdev; | |
562 | } | |
7bf7f352 | 563 | |
0e51a8e1 CH |
564 | static struct xfs_ioend * |
565 | xfs_alloc_ioend( | |
566 | struct inode *inode, | |
567 | unsigned int type, | |
568 | xfs_off_t offset, | |
569 | struct buffer_head *bh) | |
570 | { | |
571 | struct xfs_ioend *ioend; | |
572 | struct bio *bio; | |
f6d6d4fc | 573 | |
0e51a8e1 CH |
574 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
575 | xfs_init_bio_from_bh(bio, bh); | |
576 | ||
577 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
578 | INIT_LIST_HEAD(&ioend->io_list); | |
579 | ioend->io_type = type; | |
580 | ioend->io_inode = inode; | |
581 | ioend->io_size = 0; | |
582 | ioend->io_offset = offset; | |
583 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
584 | ioend->io_append_trans = NULL; | |
585 | ioend->io_bio = bio; | |
586 | return ioend; | |
587 | } | |
588 | ||
589 | /* | |
590 | * Allocate a new bio, and chain the old bio to the new one. | |
591 | * | |
592 | * Note that we have to do perform the chaining in this unintuitive order | |
593 | * so that the bi_private linkage is set up in the right direction for the | |
594 | * traversal in xfs_destroy_ioend(). | |
595 | */ | |
596 | static void | |
597 | xfs_chain_bio( | |
598 | struct xfs_ioend *ioend, | |
599 | struct writeback_control *wbc, | |
600 | struct buffer_head *bh) | |
601 | { | |
602 | struct bio *new; | |
603 | ||
604 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
605 | xfs_init_bio_from_bh(new, bh); | |
606 | ||
607 | bio_chain(ioend->io_bio, new); | |
608 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
50bfcd0c MC |
609 | bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE, |
610 | (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0); | |
4e49ea4a | 611 | submit_bio(ioend->io_bio); |
0e51a8e1 | 612 | ioend->io_bio = new; |
f6d6d4fc CH |
613 | } |
614 | ||
615 | /* | |
616 | * Test to see if we've been building up a completion structure for | |
617 | * earlier buffers -- if so, we try to append to this ioend if we | |
618 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
619 | * Return the ioend we finished off so that the caller can submit it |
620 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
621 | */ |
622 | STATIC void | |
623 | xfs_add_to_ioend( | |
624 | struct inode *inode, | |
625 | struct buffer_head *bh, | |
7336cea8 | 626 | xfs_off_t offset, |
e10de372 | 627 | struct xfs_writepage_ctx *wpc, |
bb18782a | 628 | struct writeback_control *wbc, |
e10de372 | 629 | struct list_head *iolist) |
f6d6d4fc | 630 | { |
fbcc0256 | 631 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
632 | bh->b_blocknr != wpc->last_block + 1 || |
633 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
634 | if (wpc->ioend) |
635 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 636 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
637 | } |
638 | ||
0e51a8e1 CH |
639 | /* |
640 | * If the buffer doesn't fit into the bio we need to allocate a new | |
641 | * one. This shouldn't happen more than once for a given buffer. | |
642 | */ | |
643 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
644 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 645 | |
fbcc0256 DC |
646 | wpc->ioend->io_size += bh->b_size; |
647 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 648 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
649 | } |
650 | ||
87cbc49c NS |
651 | STATIC void |
652 | xfs_map_buffer( | |
046f1685 | 653 | struct inode *inode, |
87cbc49c | 654 | struct buffer_head *bh, |
207d0416 | 655 | struct xfs_bmbt_irec *imap, |
046f1685 | 656 | xfs_off_t offset) |
87cbc49c NS |
657 | { |
658 | sector_t bn; | |
8699bb0a | 659 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
660 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
661 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 662 | |
207d0416 CH |
663 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
664 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 665 | |
e513182d | 666 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 667 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 668 | |
046f1685 | 669 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
670 | |
671 | bh->b_blocknr = bn; | |
672 | set_buffer_mapped(bh); | |
673 | } | |
674 | ||
1da177e4 LT |
675 | STATIC void |
676 | xfs_map_at_offset( | |
046f1685 | 677 | struct inode *inode, |
1da177e4 | 678 | struct buffer_head *bh, |
207d0416 | 679 | struct xfs_bmbt_irec *imap, |
046f1685 | 680 | xfs_off_t offset) |
1da177e4 | 681 | { |
207d0416 CH |
682 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
683 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 684 | |
207d0416 | 685 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
686 | set_buffer_mapped(bh); |
687 | clear_buffer_delay(bh); | |
f6d6d4fc | 688 | clear_buffer_unwritten(bh); |
1da177e4 LT |
689 | } |
690 | ||
1da177e4 | 691 | /* |
a49935f2 DC |
692 | * Test if a given page contains at least one buffer of a given @type. |
693 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
694 | * try to find one of the type passed in. If it is not set, then the caller only | |
695 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 696 | */ |
a49935f2 | 697 | STATIC bool |
6ffc4db5 | 698 | xfs_check_page_type( |
10ce4444 | 699 | struct page *page, |
a49935f2 DC |
700 | unsigned int type, |
701 | bool check_all_buffers) | |
1da177e4 | 702 | { |
a49935f2 DC |
703 | struct buffer_head *bh; |
704 | struct buffer_head *head; | |
1da177e4 | 705 | |
a49935f2 DC |
706 | if (PageWriteback(page)) |
707 | return false; | |
708 | if (!page->mapping) | |
709 | return false; | |
710 | if (!page_has_buffers(page)) | |
711 | return false; | |
1da177e4 | 712 | |
a49935f2 DC |
713 | bh = head = page_buffers(page); |
714 | do { | |
715 | if (buffer_unwritten(bh)) { | |
716 | if (type == XFS_IO_UNWRITTEN) | |
717 | return true; | |
718 | } else if (buffer_delay(bh)) { | |
805eeb8e | 719 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
720 | return true; |
721 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 722 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
723 | return true; |
724 | } | |
1da177e4 | 725 | |
a49935f2 DC |
726 | /* If we are only checking the first buffer, we are done now. */ |
727 | if (!check_all_buffers) | |
728 | break; | |
729 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 730 | |
a49935f2 | 731 | return false; |
1da177e4 LT |
732 | } |
733 | ||
3ed3a434 DC |
734 | STATIC void |
735 | xfs_vm_invalidatepage( | |
736 | struct page *page, | |
d47992f8 LC |
737 | unsigned int offset, |
738 | unsigned int length) | |
3ed3a434 | 739 | { |
34097dfe LC |
740 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
741 | length); | |
245262c6 DC |
742 | |
743 | /* | |
744 | * If we are invalidating the entire page, clear the dirty state from it | |
745 | * so that we can check for attempts to release dirty cached pages in | |
746 | * xfs_vm_releasepage(). | |
747 | */ | |
748 | if (offset == 0 && length >= PAGE_SIZE) | |
749 | cancel_dirty_page(page); | |
34097dfe | 750 | block_invalidatepage(page, offset, length); |
3ed3a434 DC |
751 | } |
752 | ||
753 | /* | |
754 | * If the page has delalloc buffers on it, we need to punch them out before we | |
755 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
756 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
757 | * is done on that same region - the delalloc extent is returned when none is | |
758 | * supposed to be there. | |
759 | * | |
760 | * We prevent this by truncating away the delalloc regions on the page before | |
761 | * invalidating it. Because they are delalloc, we can do this without needing a | |
762 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
763 | * truncation without a transaction as there is no space left for block | |
764 | * reservation (typically why we see a ENOSPC in writeback). | |
765 | * | |
766 | * This is not a performance critical path, so for now just do the punching a | |
767 | * buffer head at a time. | |
768 | */ | |
769 | STATIC void | |
770 | xfs_aops_discard_page( | |
771 | struct page *page) | |
772 | { | |
773 | struct inode *inode = page->mapping->host; | |
774 | struct xfs_inode *ip = XFS_I(inode); | |
775 | struct buffer_head *bh, *head; | |
776 | loff_t offset = page_offset(page); | |
3ed3a434 | 777 | |
a49935f2 | 778 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
779 | goto out_invalidate; |
780 | ||
e8c3753c DC |
781 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
782 | goto out_invalidate; | |
783 | ||
4f10700a | 784 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
785 | "page discard on page %p, inode 0x%llx, offset %llu.", |
786 | page, ip->i_ino, offset); | |
787 | ||
788 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
789 | bh = head = page_buffers(page); | |
790 | do { | |
3ed3a434 | 791 | int error; |
c726de44 | 792 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
793 | |
794 | if (!buffer_delay(bh)) | |
795 | goto next_buffer; | |
796 | ||
c726de44 DC |
797 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
798 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
799 | if (error) { |
800 | /* something screwed, just bail */ | |
e8c3753c | 801 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 802 | xfs_alert(ip->i_mount, |
3ed3a434 | 803 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 804 | } |
3ed3a434 DC |
805 | break; |
806 | } | |
807 | next_buffer: | |
61604a26 | 808 | offset += i_blocksize(inode); |
3ed3a434 DC |
809 | |
810 | } while ((bh = bh->b_this_page) != head); | |
811 | ||
812 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
813 | out_invalidate: | |
09cbfeaf | 814 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
815 | return; |
816 | } | |
817 | ||
ef473667 DW |
818 | static int |
819 | xfs_map_cow( | |
820 | struct xfs_writepage_ctx *wpc, | |
821 | struct inode *inode, | |
822 | loff_t offset, | |
823 | unsigned int *new_type) | |
824 | { | |
825 | struct xfs_inode *ip = XFS_I(inode); | |
826 | struct xfs_bmbt_irec imap; | |
827 | bool is_cow = false, need_alloc = false; | |
828 | int error; | |
829 | ||
830 | /* | |
831 | * If we already have a valid COW mapping keep using it. | |
832 | */ | |
833 | if (wpc->io_type == XFS_IO_COW) { | |
834 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
835 | if (wpc->imap_valid) { | |
836 | *new_type = XFS_IO_COW; | |
837 | return 0; | |
838 | } | |
839 | } | |
840 | ||
841 | /* | |
842 | * Else we need to check if there is a COW mapping at this offset. | |
843 | */ | |
844 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
845 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap, &need_alloc); | |
846 | xfs_iunlock(ip, XFS_ILOCK_SHARED); | |
847 | ||
848 | if (!is_cow) | |
849 | return 0; | |
850 | ||
851 | /* | |
852 | * And if the COW mapping has a delayed extent here we need to | |
853 | * allocate real space for it now. | |
854 | */ | |
855 | if (need_alloc) { | |
856 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, | |
857 | &imap); | |
858 | if (error) | |
859 | return error; | |
860 | } | |
861 | ||
862 | wpc->io_type = *new_type = XFS_IO_COW; | |
863 | wpc->imap_valid = true; | |
864 | wpc->imap = imap; | |
865 | return 0; | |
866 | } | |
867 | ||
e10de372 DC |
868 | /* |
869 | * We implement an immediate ioend submission policy here to avoid needing to | |
870 | * chain multiple ioends and hence nest mempool allocations which can violate | |
871 | * forward progress guarantees we need to provide. The current ioend we are | |
872 | * adding buffers to is cached on the writepage context, and if the new buffer | |
873 | * does not append to the cached ioend it will create a new ioend and cache that | |
874 | * instead. | |
875 | * | |
876 | * If a new ioend is created and cached, the old ioend is returned and queued | |
877 | * locally for submission once the entire page is processed or an error has been | |
878 | * detected. While ioends are submitted immediately after they are completed, | |
879 | * batching optimisations are provided by higher level block plugging. | |
880 | * | |
881 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
882 | * writepage context that the caller will need to submit. | |
883 | */ | |
bfce7d2e DC |
884 | static int |
885 | xfs_writepage_map( | |
886 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 887 | struct writeback_control *wbc, |
bfce7d2e DC |
888 | struct inode *inode, |
889 | struct page *page, | |
890 | loff_t offset, | |
891 | __uint64_t end_offset) | |
892 | { | |
e10de372 DC |
893 | LIST_HEAD(submit_list); |
894 | struct xfs_ioend *ioend, *next; | |
bfce7d2e | 895 | struct buffer_head *bh, *head; |
61604a26 | 896 | ssize_t len = i_blocksize(inode); |
bfce7d2e | 897 | int error = 0; |
bfce7d2e | 898 | int count = 0; |
e10de372 | 899 | int uptodate = 1; |
ef473667 | 900 | unsigned int new_type; |
bfce7d2e DC |
901 | |
902 | bh = head = page_buffers(page); | |
903 | offset = page_offset(page); | |
bfce7d2e DC |
904 | do { |
905 | if (offset >= end_offset) | |
906 | break; | |
907 | if (!buffer_uptodate(bh)) | |
908 | uptodate = 0; | |
909 | ||
910 | /* | |
911 | * set_page_dirty dirties all buffers in a page, independent | |
912 | * of their state. The dirty state however is entirely | |
913 | * meaningless for holes (!mapped && uptodate), so skip | |
914 | * buffers covering holes here. | |
915 | */ | |
916 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
917 | wpc->imap_valid = false; | |
918 | continue; | |
919 | } | |
920 | ||
ef473667 DW |
921 | if (buffer_unwritten(bh)) |
922 | new_type = XFS_IO_UNWRITTEN; | |
923 | else if (buffer_delay(bh)) | |
924 | new_type = XFS_IO_DELALLOC; | |
925 | else if (buffer_uptodate(bh)) | |
926 | new_type = XFS_IO_OVERWRITE; | |
927 | else { | |
bfce7d2e DC |
928 | if (PageUptodate(page)) |
929 | ASSERT(buffer_mapped(bh)); | |
930 | /* | |
931 | * This buffer is not uptodate and will not be | |
932 | * written to disk. Ensure that we will put any | |
933 | * subsequent writeable buffers into a new | |
934 | * ioend. | |
935 | */ | |
936 | wpc->imap_valid = false; | |
937 | continue; | |
938 | } | |
939 | ||
ef473667 DW |
940 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
941 | error = xfs_map_cow(wpc, inode, offset, &new_type); | |
942 | if (error) | |
943 | goto out; | |
944 | } | |
945 | ||
946 | if (wpc->io_type != new_type) { | |
947 | wpc->io_type = new_type; | |
948 | wpc->imap_valid = false; | |
949 | } | |
950 | ||
bfce7d2e DC |
951 | if (wpc->imap_valid) |
952 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
953 | offset); | |
954 | if (!wpc->imap_valid) { | |
955 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
956 | wpc->io_type); | |
957 | if (error) | |
e10de372 | 958 | goto out; |
bfce7d2e DC |
959 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
960 | offset); | |
961 | } | |
962 | if (wpc->imap_valid) { | |
963 | lock_buffer(bh); | |
964 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
965 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 966 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
967 | count++; |
968 | } | |
969 | ||
bfce7d2e DC |
970 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
971 | ||
972 | if (uptodate && bh == head) | |
973 | SetPageUptodate(page); | |
974 | ||
e10de372 | 975 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 976 | |
e10de372 | 977 | out: |
bfce7d2e | 978 | /* |
e10de372 DC |
979 | * On error, we have to fail the ioend here because we have locked |
980 | * buffers in the ioend. If we don't do this, we'll deadlock | |
981 | * invalidating the page as that tries to lock the buffers on the page. | |
982 | * Also, because we may have set pages under writeback, we have to make | |
983 | * sure we run IO completion to mark the error state of the IO | |
984 | * appropriately, so we can't cancel the ioend directly here. That means | |
985 | * we have to mark this page as under writeback if we included any | |
986 | * buffers from it in the ioend chain so that completion treats it | |
987 | * correctly. | |
bfce7d2e | 988 | * |
e10de372 DC |
989 | * If we didn't include the page in the ioend, the on error we can |
990 | * simply discard and unlock it as there are no other users of the page | |
991 | * or it's buffers right now. The caller will still need to trigger | |
992 | * submission of outstanding ioends on the writepage context so they are | |
993 | * treated correctly on error. | |
bfce7d2e | 994 | */ |
e10de372 DC |
995 | if (count) { |
996 | xfs_start_page_writeback(page, !error); | |
997 | ||
998 | /* | |
999 | * Preserve the original error if there was one, otherwise catch | |
1000 | * submission errors here and propagate into subsequent ioend | |
1001 | * submissions. | |
1002 | */ | |
1003 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
1004 | int error2; | |
1005 | ||
1006 | list_del_init(&ioend->io_list); | |
1007 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
1008 | if (error2 && !error) | |
1009 | error = error2; | |
1010 | } | |
1011 | } else if (error) { | |
bfce7d2e DC |
1012 | xfs_aops_discard_page(page); |
1013 | ClearPageUptodate(page); | |
1014 | unlock_page(page); | |
e10de372 DC |
1015 | } else { |
1016 | /* | |
1017 | * We can end up here with no error and nothing to write if we | |
1018 | * race with a partial page truncate on a sub-page block sized | |
1019 | * filesystem. In that case we need to mark the page clean. | |
1020 | */ | |
1021 | xfs_start_page_writeback(page, 1); | |
1022 | end_page_writeback(page); | |
bfce7d2e | 1023 | } |
e10de372 | 1024 | |
bfce7d2e DC |
1025 | mapping_set_error(page->mapping, error); |
1026 | return error; | |
1027 | } | |
1028 | ||
1da177e4 | 1029 | /* |
89f3b363 CH |
1030 | * Write out a dirty page. |
1031 | * | |
1032 | * For delalloc space on the page we need to allocate space and flush it. | |
1033 | * For unwritten space on the page we need to start the conversion to | |
1034 | * regular allocated space. | |
89f3b363 | 1035 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 1036 | */ |
1da177e4 | 1037 | STATIC int |
fbcc0256 | 1038 | xfs_do_writepage( |
89f3b363 | 1039 | struct page *page, |
fbcc0256 DC |
1040 | struct writeback_control *wbc, |
1041 | void *data) | |
1da177e4 | 1042 | { |
fbcc0256 | 1043 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 1044 | struct inode *inode = page->mapping->host; |
1da177e4 | 1045 | loff_t offset; |
1da177e4 | 1046 | __uint64_t end_offset; |
ad68972a | 1047 | pgoff_t end_index; |
89f3b363 | 1048 | |
34097dfe | 1049 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 1050 | |
20cb52eb CH |
1051 | ASSERT(page_has_buffers(page)); |
1052 | ||
89f3b363 CH |
1053 | /* |
1054 | * Refuse to write the page out if we are called from reclaim context. | |
1055 | * | |
d4f7a5cb CH |
1056 | * This avoids stack overflows when called from deeply used stacks in |
1057 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1058 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 1059 | * |
94054fa3 MG |
1060 | * This should never happen except in the case of a VM regression so |
1061 | * warn about it. | |
89f3b363 | 1062 | */ |
94054fa3 MG |
1063 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
1064 | PF_MEMALLOC)) | |
b5420f23 | 1065 | goto redirty; |
1da177e4 | 1066 | |
89f3b363 | 1067 | /* |
680a647b CH |
1068 | * Given that we do not allow direct reclaim to call us, we should |
1069 | * never be called while in a filesystem transaction. | |
89f3b363 | 1070 | */ |
448011e2 | 1071 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 1072 | goto redirty; |
89f3b363 | 1073 | |
8695d27e | 1074 | /* |
ad68972a DC |
1075 | * Is this page beyond the end of the file? |
1076 | * | |
8695d27e JL |
1077 | * The page index is less than the end_index, adjust the end_offset |
1078 | * to the highest offset that this page should represent. | |
1079 | * ----------------------------------------------------- | |
1080 | * | file mapping | <EOF> | | |
1081 | * ----------------------------------------------------- | |
1082 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1083 | * ^--------------------------------^----------|-------- | |
1084 | * | desired writeback range | see else | | |
1085 | * ---------------------------------^------------------| | |
1086 | */ | |
ad68972a | 1087 | offset = i_size_read(inode); |
09cbfeaf | 1088 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 1089 | if (page->index < end_index) |
09cbfeaf | 1090 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
1091 | else { |
1092 | /* | |
1093 | * Check whether the page to write out is beyond or straddles | |
1094 | * i_size or not. | |
1095 | * ------------------------------------------------------- | |
1096 | * | file mapping | <EOF> | | |
1097 | * ------------------------------------------------------- | |
1098 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1099 | * ^--------------------------------^-----------|--------- | |
1100 | * | | Straddles | | |
1101 | * ---------------------------------^-----------|--------| | |
1102 | */ | |
09cbfeaf | 1103 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
1104 | |
1105 | /* | |
ff9a28f6 JK |
1106 | * Skip the page if it is fully outside i_size, e.g. due to a |
1107 | * truncate operation that is in progress. We must redirty the | |
1108 | * page so that reclaim stops reclaiming it. Otherwise | |
1109 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1110 | * |
1111 | * Note that the end_index is unsigned long, it would overflow | |
1112 | * if the given offset is greater than 16TB on 32-bit system | |
1113 | * and if we do check the page is fully outside i_size or not | |
1114 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1115 | * will be evaluated to 0. Hence this page will be redirtied | |
1116 | * and be written out repeatedly which would result in an | |
1117 | * infinite loop, the user program that perform this operation | |
1118 | * will hang. Instead, we can verify this situation by checking | |
1119 | * if the page to write is totally beyond the i_size or if it's | |
1120 | * offset is just equal to the EOF. | |
6b7a03f0 | 1121 | */ |
8695d27e JL |
1122 | if (page->index > end_index || |
1123 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1124 | goto redirty; |
6b7a03f0 CH |
1125 | |
1126 | /* | |
1127 | * The page straddles i_size. It must be zeroed out on each | |
1128 | * and every writepage invocation because it may be mmapped. | |
1129 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1130 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1131 | * memory is zeroed when mapped, and writes to that region are |
1132 | * not written out to the file." | |
1133 | */ | |
09cbfeaf | 1134 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
1135 | |
1136 | /* Adjust the end_offset to the end of file */ | |
1137 | end_offset = offset; | |
1da177e4 LT |
1138 | } |
1139 | ||
e10de372 | 1140 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
f51623b2 | 1141 | |
b5420f23 | 1142 | redirty: |
f51623b2 NS |
1143 | redirty_page_for_writepage(wbc, page); |
1144 | unlock_page(page); | |
1145 | return 0; | |
f51623b2 NS |
1146 | } |
1147 | ||
fbcc0256 DC |
1148 | STATIC int |
1149 | xfs_vm_writepage( | |
1150 | struct page *page, | |
1151 | struct writeback_control *wbc) | |
1152 | { | |
1153 | struct xfs_writepage_ctx wpc = { | |
1154 | .io_type = XFS_IO_INVALID, | |
1155 | }; | |
1156 | int ret; | |
1157 | ||
1158 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1159 | if (wpc.ioend) |
1160 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1161 | return ret; | |
fbcc0256 DC |
1162 | } |
1163 | ||
7d4fb40a NS |
1164 | STATIC int |
1165 | xfs_vm_writepages( | |
1166 | struct address_space *mapping, | |
1167 | struct writeback_control *wbc) | |
1168 | { | |
fbcc0256 DC |
1169 | struct xfs_writepage_ctx wpc = { |
1170 | .io_type = XFS_IO_INVALID, | |
1171 | }; | |
1172 | int ret; | |
1173 | ||
b3aea4ed | 1174 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7f6d5b52 RZ |
1175 | if (dax_mapping(mapping)) |
1176 | return dax_writeback_mapping_range(mapping, | |
1177 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1178 | ||
fbcc0256 | 1179 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1180 | if (wpc.ioend) |
1181 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1182 | return ret; | |
7d4fb40a NS |
1183 | } |
1184 | ||
f51623b2 NS |
1185 | /* |
1186 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1187 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1188 | * have buffer heads in this call. |
1189 | * | |
89f3b363 | 1190 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1191 | */ |
1192 | STATIC int | |
238f4c54 | 1193 | xfs_vm_releasepage( |
f51623b2 NS |
1194 | struct page *page, |
1195 | gfp_t gfp_mask) | |
1196 | { | |
20cb52eb | 1197 | int delalloc, unwritten; |
f51623b2 | 1198 | |
34097dfe | 1199 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1200 | |
99579cce BF |
1201 | /* |
1202 | * mm accommodates an old ext3 case where clean pages might not have had | |
1203 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1204 | * ->releasepage() via shrink_active_list(). Conversely, | |
245262c6 DC |
1205 | * block_invalidatepage() can send pages that are still marked dirty but |
1206 | * otherwise have invalidated buffers. | |
99579cce | 1207 | * |
6ba35da6 | 1208 | * We want to release the latter to avoid unnecessary buildup of the |
245262c6 DC |
1209 | * LRU, so xfs_vm_invalidatepage() clears the page dirty flag on pages |
1210 | * that are entirely invalidated and need to be released. Hence the | |
1211 | * only time we should get dirty pages here is through | |
1212 | * shrink_active_list() and so we can simply skip those now. | |
1213 | * | |
1214 | * warn if we've left any lingering delalloc/unwritten buffers on clean | |
1215 | * or invalidated pages we are about to release. | |
99579cce | 1216 | */ |
245262c6 DC |
1217 | if (PageDirty(page)) |
1218 | return 0; | |
1219 | ||
20cb52eb | 1220 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1221 | |
245262c6 | 1222 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1223 | return 0; |
245262c6 | 1224 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1225 | return 0; |
1226 | ||
f51623b2 NS |
1227 | return try_to_free_buffers(page); |
1228 | } | |
1229 | ||
a719370b | 1230 | /* |
273dda76 CH |
1231 | * When we map a DIO buffer, we may need to pass flags to |
1232 | * xfs_end_io_direct_write to tell it what kind of write IO we are doing. | |
3e12dbbd DC |
1233 | * |
1234 | * Note that for DIO, an IO to the highest supported file block offset (i.e. | |
1235 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 | |
1236 | * bit variable. Hence if we see this overflow, we have to assume that the IO is | |
1237 | * extending the file size. We won't know for sure until IO completion is run | |
1238 | * and the actual max write offset is communicated to the IO completion | |
1239 | * routine. | |
a719370b DC |
1240 | */ |
1241 | static void | |
1242 | xfs_map_direct( | |
1243 | struct inode *inode, | |
1244 | struct buffer_head *bh_result, | |
1245 | struct xfs_bmbt_irec *imap, | |
0613f16c DW |
1246 | xfs_off_t offset, |
1247 | bool is_cow) | |
a719370b | 1248 | { |
273dda76 | 1249 | uintptr_t *flags = (uintptr_t *)&bh_result->b_private; |
d5cc2e3f | 1250 | xfs_off_t size = bh_result->b_size; |
d5cc2e3f | 1251 | |
273dda76 | 1252 | trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size, |
0613f16c DW |
1253 | ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : is_cow ? XFS_IO_COW : |
1254 | XFS_IO_OVERWRITE, imap); | |
d5cc2e3f | 1255 | |
273dda76 CH |
1256 | if (ISUNWRITTEN(imap)) { |
1257 | *flags |= XFS_DIO_FLAG_UNWRITTEN; | |
1258 | set_buffer_defer_completion(bh_result); | |
0613f16c DW |
1259 | } else if (is_cow) { |
1260 | *flags |= XFS_DIO_FLAG_COW; | |
1261 | set_buffer_defer_completion(bh_result); | |
1262 | } | |
1263 | if (offset + size > i_size_read(inode) || offset + size < 0) { | |
273dda76 | 1264 | *flags |= XFS_DIO_FLAG_APPEND; |
a06c277a | 1265 | set_buffer_defer_completion(bh_result); |
a719370b DC |
1266 | } |
1267 | } | |
1268 | ||
1fdca9c2 DC |
1269 | /* |
1270 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1271 | * is, so that we can avoid repeated get_blocks calls. | |
1272 | * | |
1273 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1274 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1275 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1276 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1277 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1278 | * to and including EOF, and a second mapping for beyond EOF. | |
1279 | */ | |
1280 | static void | |
1281 | xfs_map_trim_size( | |
1282 | struct inode *inode, | |
1283 | sector_t iblock, | |
1284 | struct buffer_head *bh_result, | |
1285 | struct xfs_bmbt_irec *imap, | |
1286 | xfs_off_t offset, | |
1287 | ssize_t size) | |
1288 | { | |
1289 | xfs_off_t mapping_size; | |
1290 | ||
1291 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1292 | mapping_size <<= inode->i_blkbits; | |
1293 | ||
1294 | ASSERT(mapping_size > 0); | |
1295 | if (mapping_size > size) | |
1296 | mapping_size = size; | |
1297 | if (offset < i_size_read(inode) && | |
d9602444 | 1298 | (xfs_ufsize_t)offset + mapping_size >= i_size_read(inode)) { |
1fdca9c2 DC |
1299 | /* limit mapping to block that spans EOF */ |
1300 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
61604a26 | 1301 | i_blocksize(inode)); |
1fdca9c2 DC |
1302 | } |
1303 | if (mapping_size > LONG_MAX) | |
1304 | mapping_size = LONG_MAX; | |
1305 | ||
1306 | bh_result->b_size = mapping_size; | |
1307 | } | |
1308 | ||
1da177e4 | 1309 | STATIC int |
c2536668 | 1310 | __xfs_get_blocks( |
1da177e4 LT |
1311 | struct inode *inode, |
1312 | sector_t iblock, | |
1da177e4 LT |
1313 | struct buffer_head *bh_result, |
1314 | int create, | |
3e12dbbd DC |
1315 | bool direct, |
1316 | bool dax_fault) | |
1da177e4 | 1317 | { |
a206c817 CH |
1318 | struct xfs_inode *ip = XFS_I(inode); |
1319 | struct xfs_mount *mp = ip->i_mount; | |
1320 | xfs_fileoff_t offset_fsb, end_fsb; | |
1321 | int error = 0; | |
1322 | int lockmode = 0; | |
207d0416 | 1323 | struct xfs_bmbt_irec imap; |
a206c817 | 1324 | int nimaps = 1; |
fdc7ed75 NS |
1325 | xfs_off_t offset; |
1326 | ssize_t size; | |
207d0416 | 1327 | int new = 0; |
0613f16c DW |
1328 | bool is_cow = false; |
1329 | bool need_alloc = false; | |
a206c817 | 1330 | |
6e8a27a8 CH |
1331 | BUG_ON(create && !direct); |
1332 | ||
a206c817 | 1333 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1334 | return -EIO; |
1da177e4 | 1335 | |
fdc7ed75 | 1336 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
61604a26 | 1337 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
c2536668 | 1338 | size = bh_result->b_size; |
364f358a | 1339 | |
6e8a27a8 | 1340 | if (!create && offset >= i_size_read(inode)) |
364f358a LM |
1341 | return 0; |
1342 | ||
507630b2 DC |
1343 | /* |
1344 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1345 | * a block mapping without an exclusive lock first. |
507630b2 | 1346 | */ |
6e8a27a8 | 1347 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1348 | |
d2c28191 | 1349 | ASSERT(offset <= mp->m_super->s_maxbytes); |
d9602444 | 1350 | if ((xfs_ufsize_t)offset + size > mp->m_super->s_maxbytes) |
d2c28191 | 1351 | size = mp->m_super->s_maxbytes - offset; |
a206c817 CH |
1352 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1353 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1354 | ||
0613f16c DW |
1355 | if (create && direct && xfs_is_reflink_inode(ip)) |
1356 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap, | |
1357 | &need_alloc); | |
1358 | if (!is_cow) { | |
1359 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, | |
1360 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1361 | /* | |
1362 | * Truncate an overwrite extent if there's a pending CoW | |
1363 | * reservation before the end of this extent. This | |
1364 | * forces us to come back to get_blocks to take care of | |
1365 | * the CoW. | |
1366 | */ | |
1367 | if (create && direct && nimaps && | |
1368 | imap.br_startblock != HOLESTARTBLOCK && | |
1369 | imap.br_startblock != DELAYSTARTBLOCK && | |
1370 | !ISUNWRITTEN(&imap)) | |
1371 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, | |
1372 | &imap); | |
1373 | } | |
1374 | ASSERT(!need_alloc); | |
1da177e4 | 1375 | if (error) |
a206c817 CH |
1376 | goto out_unlock; |
1377 | ||
3978c5bb BF |
1378 | /* |
1379 | * The only time we can ever safely find delalloc blocks on direct I/O | |
1380 | * is a dio write to post-eof speculative preallocation. All other | |
1381 | * scenarios are indicative of a problem or misuse (such as mixing | |
1382 | * direct and mapped I/O). | |
1383 | * | |
1384 | * The file may be unmapped by the time we get here so we cannot | |
1385 | * reliably fail the I/O based on mapping. Instead, fail the I/O if this | |
1386 | * is a read or a write within eof. Otherwise, carry on but warn as a | |
1387 | * precuation if the file happens to be mapped. | |
1388 | */ | |
1389 | if (direct && imap.br_startblock == DELAYSTARTBLOCK) { | |
1390 | if (!create || offset < i_size_read(VFS_I(ip))) { | |
1391 | WARN_ON_ONCE(1); | |
1392 | error = -EIO; | |
1393 | goto out_unlock; | |
1394 | } | |
1395 | WARN_ON_ONCE(mapping_mapped(VFS_I(ip)->i_mapping)); | |
1396 | } | |
1397 | ||
1ca19157 | 1398 | /* for DAX, we convert unwritten extents directly */ |
a206c817 CH |
1399 | if (create && |
1400 | (!nimaps || | |
1401 | (imap.br_startblock == HOLESTARTBLOCK || | |
1ca19157 DC |
1402 | imap.br_startblock == DELAYSTARTBLOCK) || |
1403 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { | |
6e8a27a8 CH |
1404 | /* |
1405 | * xfs_iomap_write_direct() expects the shared lock. It | |
1406 | * is unlocked on return. | |
1407 | */ | |
1408 | if (lockmode == XFS_ILOCK_EXCL) | |
1409 | xfs_ilock_demote(ip, lockmode); | |
6b698ede | 1410 | |
6e8a27a8 CH |
1411 | error = xfs_iomap_write_direct(ip, offset, size, |
1412 | &imap, nimaps); | |
1413 | if (error) | |
1414 | return error; | |
1415 | new = 1; | |
507630b2 | 1416 | |
d5cc2e3f DC |
1417 | trace_xfs_get_blocks_alloc(ip, offset, size, |
1418 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1419 | : XFS_IO_DELALLOC, &imap); | |
a206c817 | 1420 | } else if (nimaps) { |
d5cc2e3f DC |
1421 | trace_xfs_get_blocks_found(ip, offset, size, |
1422 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1423 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1424 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1425 | } else { |
1426 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1427 | goto out_unlock; | |
1428 | } | |
1da177e4 | 1429 | |
1ca19157 DC |
1430 | if (IS_DAX(inode) && create) { |
1431 | ASSERT(!ISUNWRITTEN(&imap)); | |
1432 | /* zeroing is not needed at a higher layer */ | |
1433 | new = 0; | |
1434 | } | |
1435 | ||
1fdca9c2 | 1436 | /* trim mapping down to size requested */ |
6e8a27a8 | 1437 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1438 | |
a719370b DC |
1439 | /* |
1440 | * For unwritten extents do not report a disk address in the buffered | |
1441 | * read case (treat as if we're reading into a hole). | |
1442 | */ | |
207d0416 | 1443 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1444 | imap.br_startblock != DELAYSTARTBLOCK && |
1445 | (create || !ISUNWRITTEN(&imap))) { | |
1446 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1447 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1448 | set_buffer_unwritten(bh_result); |
a719370b | 1449 | /* direct IO needs special help */ |
6e8a27a8 | 1450 | if (create) { |
273dda76 CH |
1451 | if (dax_fault) |
1452 | ASSERT(!ISUNWRITTEN(&imap)); | |
1453 | else | |
0613f16c DW |
1454 | xfs_map_direct(inode, bh_result, &imap, offset, |
1455 | is_cow); | |
273dda76 | 1456 | } |
1da177e4 LT |
1457 | } |
1458 | ||
c2536668 NS |
1459 | /* |
1460 | * If this is a realtime file, data may be on a different device. | |
1461 | * to that pointed to from the buffer_head b_bdev currently. | |
1462 | */ | |
046f1685 | 1463 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1464 | |
c2536668 | 1465 | /* |
549054af DC |
1466 | * If we previously allocated a block out beyond eof and we are now |
1467 | * coming back to use it then we will need to flag it as new even if it | |
1468 | * has a disk address. | |
1469 | * | |
1470 | * With sub-block writes into unwritten extents we also need to mark | |
1471 | * the buffer as new so that the unwritten parts of the buffer gets | |
1472 | * correctly zeroed. | |
1da177e4 LT |
1473 | */ |
1474 | if (create && | |
1475 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1476 | (offset >= i_size_read(inode)) || |
207d0416 | 1477 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1478 | set_buffer_new(bh_result); |
1da177e4 | 1479 | |
1da177e4 | 1480 | return 0; |
a206c817 CH |
1481 | |
1482 | out_unlock: | |
1483 | xfs_iunlock(ip, lockmode); | |
2451337d | 1484 | return error; |
1da177e4 LT |
1485 | } |
1486 | ||
1487 | int | |
c2536668 | 1488 | xfs_get_blocks( |
1da177e4 LT |
1489 | struct inode *inode, |
1490 | sector_t iblock, | |
1491 | struct buffer_head *bh_result, | |
1492 | int create) | |
1493 | { | |
3e12dbbd | 1494 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
1da177e4 LT |
1495 | } |
1496 | ||
6b698ede | 1497 | int |
e4c573bb | 1498 | xfs_get_blocks_direct( |
1da177e4 LT |
1499 | struct inode *inode, |
1500 | sector_t iblock, | |
1da177e4 LT |
1501 | struct buffer_head *bh_result, |
1502 | int create) | |
1503 | { | |
3e12dbbd DC |
1504 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
1505 | } | |
1506 | ||
1507 | int | |
1508 | xfs_get_blocks_dax_fault( | |
1509 | struct inode *inode, | |
1510 | sector_t iblock, | |
1511 | struct buffer_head *bh_result, | |
1512 | int create) | |
1513 | { | |
1514 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); | |
1da177e4 LT |
1515 | } |
1516 | ||
273dda76 CH |
1517 | /* |
1518 | * Complete a direct I/O write request. | |
1519 | * | |
1520 | * xfs_map_direct passes us some flags in the private data to tell us what to | |
1521 | * do. If no flags are set, then the write IO is an overwrite wholly within | |
1522 | * the existing allocated file size and so there is nothing for us to do. | |
1523 | * | |
1524 | * Note that in this case the completion can be called in interrupt context, | |
1525 | * whereas if we have flags set we will always be called in task context | |
1526 | * (i.e. from a workqueue). | |
1527 | */ | |
fa8d972d | 1528 | int |
273dda76 CH |
1529 | xfs_end_io_direct_write( |
1530 | struct kiocb *iocb, | |
209fb87a | 1531 | loff_t offset, |
273dda76 CH |
1532 | ssize_t size, |
1533 | void *private) | |
f0973863 | 1534 | { |
273dda76 CH |
1535 | struct inode *inode = file_inode(iocb->ki_filp); |
1536 | struct xfs_inode *ip = XFS_I(inode); | |
273dda76 CH |
1537 | uintptr_t flags = (uintptr_t)private; |
1538 | int error = 0; | |
a06c277a | 1539 | |
273dda76 | 1540 | trace_xfs_end_io_direct_write(ip, offset, size); |
f0973863 | 1541 | |
e372843a | 1542 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
273dda76 | 1543 | return -EIO; |
d5cc2e3f | 1544 | |
273dda76 CH |
1545 | if (size <= 0) |
1546 | return size; | |
f0973863 | 1547 | |
2813d682 | 1548 | /* |
273dda76 | 1549 | * The flags tell us whether we are doing unwritten extent conversions |
6dfa1b67 DC |
1550 | * or an append transaction that updates the on-disk file size. These |
1551 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1552 | * to update the VFS inode size. |
273dda76 CH |
1553 | */ |
1554 | if (flags == 0) { | |
1555 | ASSERT(offset + size <= i_size_read(inode)); | |
1556 | return 0; | |
1557 | } | |
1558 | ||
0eebfede EG |
1559 | if (flags & XFS_DIO_FLAG_COW) |
1560 | error = xfs_reflink_end_cow(ip, offset, size); | |
1561 | ||
1562 | /* | |
1563 | * Unwritten conversion updates the in-core isize after extent | |
1564 | * conversion but before updating the on-disk size. Updating isize any | |
1565 | * earlier allows a racing dio read to find unwritten extents before | |
1566 | * they are converted. | |
1567 | */ | |
1568 | if (flags & XFS_DIO_FLAG_UNWRITTEN) { | |
1569 | trace_xfs_end_io_direct_write_unwritten(ip, offset, size); | |
1570 | ||
1571 | return xfs_iomap_write_unwritten(ip, offset, size, true); | |
1572 | } | |
1573 | ||
273dda76 | 1574 | /* |
6dfa1b67 | 1575 | * We need to update the in-core inode size here so that we don't end up |
a06c277a DC |
1576 | * with the on-disk inode size being outside the in-core inode size. We |
1577 | * have no other method of updating EOF for AIO, so always do it here | |
1578 | * if necessary. | |
b9d59846 DC |
1579 | * |
1580 | * We need to lock the test/set EOF update as we can be racing with | |
1581 | * other IO completions here to update the EOF. Failing to serialise | |
1582 | * here can result in EOF moving backwards and Bad Things Happen when | |
1583 | * that occurs. | |
2813d682 | 1584 | */ |
273dda76 | 1585 | spin_lock(&ip->i_flags_lock); |
2ba66237 CH |
1586 | if (offset + size > i_size_read(inode)) |
1587 | i_size_write(inode, offset + size); | |
273dda76 | 1588 | spin_unlock(&ip->i_flags_lock); |
2813d682 | 1589 | |
0613f16c | 1590 | if (flags & XFS_DIO_FLAG_APPEND) { |
273dda76 | 1591 | trace_xfs_end_io_direct_write_append(ip, offset, size); |
6b698ede | 1592 | |
e372843a | 1593 | error = xfs_setfilesize(ip, offset, size); |
6b698ede DC |
1594 | } |
1595 | ||
273dda76 | 1596 | return error; |
6b698ede DC |
1597 | } |
1598 | ||
c19b104a CH |
1599 | STATIC ssize_t |
1600 | xfs_vm_direct_IO( | |
6e1ba0bc | 1601 | struct kiocb *iocb, |
c8b8e32d | 1602 | struct iov_iter *iter) |
6e1ba0bc | 1603 | { |
58e59854 | 1604 | /* |
fa8d972d | 1605 | * We just need the method present so that open/fcntl allow direct I/O. |
58e59854 | 1606 | */ |
fa8d972d | 1607 | return -EINVAL; |
f51623b2 | 1608 | } |
1da177e4 LT |
1609 | |
1610 | STATIC sector_t | |
e4c573bb | 1611 | xfs_vm_bmap( |
1da177e4 LT |
1612 | struct address_space *mapping, |
1613 | sector_t block) | |
1614 | { | |
1615 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1616 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1617 | |
cca28fb8 | 1618 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1619 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
db1327b1 DW |
1620 | |
1621 | /* | |
1622 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
1623 | * bypasseѕ the file system for actual I/O. We really can't allow | |
1624 | * that on reflinks inodes, so we have to skip out here. And yes, | |
621d0b75 DW |
1625 | * 0 is the magic code for a bmap error. |
1626 | * | |
1627 | * Since we don't pass back blockdev info, we can't return bmap | |
1628 | * information for rt files either. | |
db1327b1 | 1629 | */ |
621d0b75 | 1630 | if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) { |
db1327b1 DW |
1631 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
1632 | return 0; | |
1633 | } | |
4bc1ea6b | 1634 | filemap_write_and_wait(mapping); |
126468b1 | 1635 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1636 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1637 | } |
1638 | ||
1639 | STATIC int | |
e4c573bb | 1640 | xfs_vm_readpage( |
1da177e4 LT |
1641 | struct file *unused, |
1642 | struct page *page) | |
1643 | { | |
121e213e | 1644 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1645 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1646 | } |
1647 | ||
1648 | STATIC int | |
e4c573bb | 1649 | xfs_vm_readpages( |
1da177e4 LT |
1650 | struct file *unused, |
1651 | struct address_space *mapping, | |
1652 | struct list_head *pages, | |
1653 | unsigned nr_pages) | |
1654 | { | |
121e213e | 1655 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1656 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1657 | } |
1658 | ||
22e757a4 DC |
1659 | /* |
1660 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1661 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1662 | * dirty, we'll never be able to clean them because we don't write buffers | |
1663 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1664 | * that have been marked dirty. Further, the dirty state can leak into | |
1665 | * the file interior if the file is extended, resulting in all sorts of | |
1666 | * bad things happening as the state does not match the underlying data. | |
1667 | * | |
1668 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1669 | * this only exist because of bufferheads and how the generic code manages them. | |
1670 | */ | |
1671 | STATIC int | |
1672 | xfs_vm_set_page_dirty( | |
1673 | struct page *page) | |
1674 | { | |
1675 | struct address_space *mapping = page->mapping; | |
1676 | struct inode *inode = mapping->host; | |
1677 | loff_t end_offset; | |
1678 | loff_t offset; | |
1679 | int newly_dirty; | |
1680 | ||
1681 | if (unlikely(!mapping)) | |
1682 | return !TestSetPageDirty(page); | |
1683 | ||
1684 | end_offset = i_size_read(inode); | |
1685 | offset = page_offset(page); | |
1686 | ||
1687 | spin_lock(&mapping->private_lock); | |
1688 | if (page_has_buffers(page)) { | |
1689 | struct buffer_head *head = page_buffers(page); | |
1690 | struct buffer_head *bh = head; | |
1691 | ||
1692 | do { | |
1693 | if (offset < end_offset) | |
1694 | set_buffer_dirty(bh); | |
1695 | bh = bh->b_this_page; | |
61604a26 | 1696 | offset += i_blocksize(inode); |
22e757a4 DC |
1697 | } while (bh != head); |
1698 | } | |
c4843a75 | 1699 | /* |
81f8c3a4 JW |
1700 | * Lock out page->mem_cgroup migration to keep PageDirty |
1701 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1702 | */ |
62cccb8c | 1703 | lock_page_memcg(page); |
22e757a4 DC |
1704 | newly_dirty = !TestSetPageDirty(page); |
1705 | spin_unlock(&mapping->private_lock); | |
1706 | ||
1707 | if (newly_dirty) { | |
1708 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1709 | unsigned long flags; | |
1710 | ||
1711 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1712 | if (page->mapping) { /* Race with truncate? */ | |
1713 | WARN_ON_ONCE(!PageUptodate(page)); | |
62cccb8c | 1714 | account_page_dirtied(page, mapping); |
22e757a4 DC |
1715 | radix_tree_tag_set(&mapping->page_tree, |
1716 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1717 | } | |
1718 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1719 | } |
62cccb8c | 1720 | unlock_page_memcg(page); |
c4843a75 GT |
1721 | if (newly_dirty) |
1722 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1723 | return newly_dirty; |
1724 | } | |
1725 | ||
f5e54d6e | 1726 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1727 | .readpage = xfs_vm_readpage, |
1728 | .readpages = xfs_vm_readpages, | |
1729 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1730 | .writepages = xfs_vm_writepages, |
22e757a4 | 1731 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1732 | .releasepage = xfs_vm_releasepage, |
1733 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb NS |
1734 | .bmap = xfs_vm_bmap, |
1735 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1736 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1737 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1738 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1739 | }; |