]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
70a9883c | 19 | #include "xfs_shared.h" |
239880ef DC |
20 | #include "xfs_format.h" |
21 | #include "xfs_log_format.h" | |
22 | #include "xfs_trans_resv.h" | |
1da177e4 | 23 | #include "xfs_mount.h" |
1da177e4 | 24 | #include "xfs_inode.h" |
239880ef | 25 | #include "xfs_trans.h" |
281627df | 26 | #include "xfs_inode_item.h" |
a844f451 | 27 | #include "xfs_alloc.h" |
1da177e4 | 28 | #include "xfs_error.h" |
1da177e4 | 29 | #include "xfs_iomap.h" |
0b1b213f | 30 | #include "xfs_trace.h" |
3ed3a434 | 31 | #include "xfs_bmap.h" |
68988114 | 32 | #include "xfs_bmap_util.h" |
a4fbe6ab | 33 | #include "xfs_bmap_btree.h" |
ef473667 | 34 | #include "xfs_reflink.h" |
5a0e3ad6 | 35 | #include <linux/gfp.h> |
1da177e4 | 36 | #include <linux/mpage.h> |
10ce4444 | 37 | #include <linux/pagevec.h> |
1da177e4 LT |
38 | #include <linux/writeback.h> |
39 | ||
fbcc0256 DC |
40 | /* |
41 | * structure owned by writepages passed to individual writepage calls | |
42 | */ | |
43 | struct xfs_writepage_ctx { | |
44 | struct xfs_bmbt_irec imap; | |
45 | bool imap_valid; | |
46 | unsigned int io_type; | |
fbcc0256 DC |
47 | struct xfs_ioend *ioend; |
48 | sector_t last_block; | |
49 | }; | |
50 | ||
0b1b213f | 51 | void |
f51623b2 NS |
52 | xfs_count_page_state( |
53 | struct page *page, | |
54 | int *delalloc, | |
f51623b2 NS |
55 | int *unwritten) |
56 | { | |
57 | struct buffer_head *bh, *head; | |
58 | ||
20cb52eb | 59 | *delalloc = *unwritten = 0; |
f51623b2 NS |
60 | |
61 | bh = head = page_buffers(page); | |
62 | do { | |
20cb52eb | 63 | if (buffer_unwritten(bh)) |
f51623b2 NS |
64 | (*unwritten) = 1; |
65 | else if (buffer_delay(bh)) | |
66 | (*delalloc) = 1; | |
67 | } while ((bh = bh->b_this_page) != head); | |
68 | } | |
69 | ||
20a90f58 | 70 | struct block_device * |
6214ed44 | 71 | xfs_find_bdev_for_inode( |
046f1685 | 72 | struct inode *inode) |
6214ed44 | 73 | { |
046f1685 | 74 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
75 | struct xfs_mount *mp = ip->i_mount; |
76 | ||
71ddabb9 | 77 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
78 | return mp->m_rtdev_targp->bt_bdev; |
79 | else | |
80 | return mp->m_ddev_targp->bt_bdev; | |
81 | } | |
82 | ||
f6d6d4fc | 83 | /* |
37992c18 DC |
84 | * We're now finished for good with this page. Update the page state via the |
85 | * associated buffer_heads, paying attention to the start and end offsets that | |
86 | * we need to process on the page. | |
28b783e4 DC |
87 | * |
88 | * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last | |
89 | * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or | |
90 | * the page at all, as we may be racing with memory reclaim and it can free both | |
91 | * the bufferhead chain and the page as it will see the page as clean and | |
92 | * unused. | |
37992c18 DC |
93 | */ |
94 | static void | |
95 | xfs_finish_page_writeback( | |
96 | struct inode *inode, | |
97 | struct bio_vec *bvec, | |
98 | int error) | |
99 | { | |
37992c18 | 100 | unsigned int end = bvec->bv_offset + bvec->bv_len - 1; |
28b783e4 | 101 | struct buffer_head *head, *bh, *next; |
37992c18 | 102 | unsigned int off = 0; |
28b783e4 | 103 | unsigned int bsize; |
37992c18 DC |
104 | |
105 | ASSERT(bvec->bv_offset < PAGE_SIZE); | |
93407472 | 106 | ASSERT((bvec->bv_offset & (i_blocksize(inode) - 1)) == 0); |
37992c18 | 107 | ASSERT(end < PAGE_SIZE); |
93407472 | 108 | ASSERT((bvec->bv_len & (i_blocksize(inode) - 1)) == 0); |
37992c18 DC |
109 | |
110 | bh = head = page_buffers(bvec->bv_page); | |
111 | ||
28b783e4 | 112 | bsize = bh->b_size; |
37992c18 | 113 | do { |
28b783e4 | 114 | next = bh->b_this_page; |
37992c18 DC |
115 | if (off < bvec->bv_offset) |
116 | goto next_bh; | |
117 | if (off > end) | |
118 | break; | |
119 | bh->b_end_io(bh, !error); | |
120 | next_bh: | |
28b783e4 DC |
121 | off += bsize; |
122 | } while ((bh = next) != head); | |
37992c18 DC |
123 | } |
124 | ||
125 | /* | |
126 | * We're now finished for good with this ioend structure. Update the page | |
127 | * state, release holds on bios, and finally free up memory. Do not use the | |
128 | * ioend after this. | |
f6d6d4fc | 129 | */ |
0829c360 CH |
130 | STATIC void |
131 | xfs_destroy_ioend( | |
0e51a8e1 CH |
132 | struct xfs_ioend *ioend, |
133 | int error) | |
0829c360 | 134 | { |
37992c18 | 135 | struct inode *inode = ioend->io_inode; |
0e51a8e1 | 136 | struct bio *last = ioend->io_bio; |
37992c18 | 137 | struct bio *bio, *next; |
f6d6d4fc | 138 | |
0e51a8e1 | 139 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
37992c18 DC |
140 | struct bio_vec *bvec; |
141 | int i; | |
142 | ||
0e51a8e1 CH |
143 | /* |
144 | * For the last bio, bi_private points to the ioend, so we | |
145 | * need to explicitly end the iteration here. | |
146 | */ | |
147 | if (bio == last) | |
148 | next = NULL; | |
149 | else | |
150 | next = bio->bi_private; | |
583fa586 | 151 | |
37992c18 DC |
152 | /* walk each page on bio, ending page IO on them */ |
153 | bio_for_each_segment_all(bvec, bio, i) | |
154 | xfs_finish_page_writeback(inode, bvec, error); | |
155 | ||
156 | bio_put(bio); | |
f6d6d4fc | 157 | } |
0829c360 CH |
158 | } |
159 | ||
fc0063c4 CH |
160 | /* |
161 | * Fast and loose check if this write could update the on-disk inode size. | |
162 | */ | |
163 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
164 | { | |
165 | return ioend->io_offset + ioend->io_size > | |
166 | XFS_I(ioend->io_inode)->i_d.di_size; | |
167 | } | |
168 | ||
281627df CH |
169 | STATIC int |
170 | xfs_setfilesize_trans_alloc( | |
171 | struct xfs_ioend *ioend) | |
172 | { | |
173 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
174 | struct xfs_trans *tp; | |
175 | int error; | |
176 | ||
253f4911 CH |
177 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
178 | if (error) | |
281627df | 179 | return error; |
281627df CH |
180 | |
181 | ioend->io_append_trans = tp; | |
182 | ||
d9457dc0 | 183 | /* |
437a255a | 184 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
185 | * we released it. |
186 | */ | |
bee9182d | 187 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
188 | /* |
189 | * We hand off the transaction to the completion thread now, so | |
190 | * clear the flag here. | |
191 | */ | |
192 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
193 | return 0; | |
194 | } | |
195 | ||
ba87ea69 | 196 | /* |
2813d682 | 197 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 198 | */ |
281627df | 199 | STATIC int |
e372843a | 200 | __xfs_setfilesize( |
2ba66237 CH |
201 | struct xfs_inode *ip, |
202 | struct xfs_trans *tp, | |
203 | xfs_off_t offset, | |
204 | size_t size) | |
ba87ea69 | 205 | { |
ba87ea69 | 206 | xfs_fsize_t isize; |
ba87ea69 | 207 | |
aa6bf01d | 208 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 209 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
210 | if (!isize) { |
211 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 212 | xfs_trans_cancel(tp); |
281627df | 213 | return 0; |
ba87ea69 LM |
214 | } |
215 | ||
2ba66237 | 216 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
217 | |
218 | ip->i_d.di_size = isize; | |
219 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
220 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
221 | ||
70393313 | 222 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
223 | } |
224 | ||
e372843a CH |
225 | int |
226 | xfs_setfilesize( | |
227 | struct xfs_inode *ip, | |
228 | xfs_off_t offset, | |
229 | size_t size) | |
230 | { | |
231 | struct xfs_mount *mp = ip->i_mount; | |
232 | struct xfs_trans *tp; | |
233 | int error; | |
234 | ||
235 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); | |
236 | if (error) | |
237 | return error; | |
238 | ||
239 | return __xfs_setfilesize(ip, tp, offset, size); | |
240 | } | |
241 | ||
2ba66237 CH |
242 | STATIC int |
243 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
244 | struct xfs_ioend *ioend, |
245 | int error) | |
2ba66237 CH |
246 | { |
247 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
248 | struct xfs_trans *tp = ioend->io_append_trans; | |
249 | ||
250 | /* | |
251 | * The transaction may have been allocated in the I/O submission thread, | |
252 | * thus we need to mark ourselves as being in a transaction manually. | |
253 | * Similarly for freeze protection. | |
254 | */ | |
255 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 256 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 257 | |
5cb13dcd | 258 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 259 | if (error) { |
5cb13dcd | 260 | xfs_trans_cancel(tp); |
0e51a8e1 | 261 | return error; |
5cb13dcd Z |
262 | } |
263 | ||
e372843a | 264 | return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
2ba66237 CH |
265 | } |
266 | ||
0829c360 | 267 | /* |
5ec4fabb | 268 | * IO write completion. |
f6d6d4fc CH |
269 | */ |
270 | STATIC void | |
5ec4fabb | 271 | xfs_end_io( |
77d7a0c2 | 272 | struct work_struct *work) |
0829c360 | 273 | { |
0e51a8e1 CH |
274 | struct xfs_ioend *ioend = |
275 | container_of(work, struct xfs_ioend, io_work); | |
276 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
277 | int error = ioend->io_bio->bi_error; | |
ba87ea69 | 278 | |
af055e37 BF |
279 | /* |
280 | * Set an error if the mount has shut down and proceed with end I/O | |
281 | * processing so it can perform whatever cleanups are necessary. | |
282 | */ | |
283 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
0e51a8e1 | 284 | error = -EIO; |
04f658ee | 285 | |
43caeb18 DW |
286 | /* |
287 | * For a CoW extent, we need to move the mapping from the CoW fork | |
288 | * to the data fork. If instead an error happened, just dump the | |
289 | * new blocks. | |
290 | */ | |
291 | if (ioend->io_type == XFS_IO_COW) { | |
292 | if (error) | |
293 | goto done; | |
294 | if (ioend->io_bio->bi_error) { | |
295 | error = xfs_reflink_cancel_cow_range(ip, | |
296 | ioend->io_offset, ioend->io_size); | |
297 | goto done; | |
298 | } | |
299 | error = xfs_reflink_end_cow(ip, ioend->io_offset, | |
300 | ioend->io_size); | |
301 | if (error) | |
302 | goto done; | |
303 | } | |
304 | ||
5ec4fabb CH |
305 | /* |
306 | * For unwritten extents we need to issue transactions to convert a | |
307 | * range to normal written extens after the data I/O has finished. | |
5cb13dcd Z |
308 | * Detecting and handling completion IO errors is done individually |
309 | * for each case as different cleanup operations need to be performed | |
310 | * on error. | |
5ec4fabb | 311 | */ |
0d882a36 | 312 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
0e51a8e1 | 313 | if (error) |
5cb13dcd | 314 | goto done; |
437a255a DC |
315 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
316 | ioend->io_size); | |
281627df | 317 | } else if (ioend->io_append_trans) { |
0e51a8e1 | 318 | error = xfs_setfilesize_ioend(ioend, error); |
84803fb7 | 319 | } else { |
43caeb18 DW |
320 | ASSERT(!xfs_ioend_is_append(ioend) || |
321 | ioend->io_type == XFS_IO_COW); | |
5ec4fabb | 322 | } |
ba87ea69 | 323 | |
04f658ee | 324 | done: |
0e51a8e1 | 325 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
326 | } |
327 | ||
0e51a8e1 CH |
328 | STATIC void |
329 | xfs_end_bio( | |
330 | struct bio *bio) | |
0829c360 | 331 | { |
0e51a8e1 CH |
332 | struct xfs_ioend *ioend = bio->bi_private; |
333 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 334 | |
43caeb18 | 335 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
0e51a8e1 CH |
336 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
337 | else if (ioend->io_append_trans) | |
338 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
339 | else | |
340 | xfs_destroy_ioend(ioend, bio->bi_error); | |
0829c360 CH |
341 | } |
342 | ||
1da177e4 LT |
343 | STATIC int |
344 | xfs_map_blocks( | |
345 | struct inode *inode, | |
346 | loff_t offset, | |
207d0416 | 347 | struct xfs_bmbt_irec *imap, |
988ef927 | 348 | int type) |
1da177e4 | 349 | { |
a206c817 CH |
350 | struct xfs_inode *ip = XFS_I(inode); |
351 | struct xfs_mount *mp = ip->i_mount; | |
93407472 | 352 | ssize_t count = i_blocksize(inode); |
a206c817 CH |
353 | xfs_fileoff_t offset_fsb, end_fsb; |
354 | int error = 0; | |
a206c817 CH |
355 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
356 | int nimaps = 1; | |
357 | ||
358 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 359 | return -EIO; |
a206c817 | 360 | |
ef473667 | 361 | ASSERT(type != XFS_IO_COW); |
0d882a36 | 362 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 363 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 364 | |
988ef927 | 365 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
366 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
367 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 368 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 369 | |
d2c28191 DC |
370 | if (offset + count > mp->m_super->s_maxbytes) |
371 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
372 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
373 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
374 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
375 | imap, &nimaps, bmapi_flags); | |
ef473667 DW |
376 | /* |
377 | * Truncate an overwrite extent if there's a pending CoW | |
378 | * reservation before the end of this extent. This forces us | |
379 | * to come back to writepage to take care of the CoW. | |
380 | */ | |
381 | if (nimaps && type == XFS_IO_OVERWRITE) | |
382 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); | |
8ff2957d | 383 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 384 | |
8ff2957d | 385 | if (error) |
2451337d | 386 | return error; |
a206c817 | 387 | |
0d882a36 | 388 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 389 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
60b4984f DW |
390 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
391 | imap); | |
a206c817 | 392 | if (!error) |
ef473667 | 393 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
2451337d | 394 | return error; |
a206c817 CH |
395 | } |
396 | ||
8ff2957d | 397 | #ifdef DEBUG |
0d882a36 | 398 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
399 | ASSERT(nimaps); |
400 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
401 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
402 | } | |
403 | #endif | |
404 | if (nimaps) | |
405 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
406 | return 0; | |
1da177e4 LT |
407 | } |
408 | ||
fbcc0256 | 409 | STATIC bool |
558e6891 | 410 | xfs_imap_valid( |
8699bb0a | 411 | struct inode *inode, |
207d0416 | 412 | struct xfs_bmbt_irec *imap, |
558e6891 | 413 | xfs_off_t offset) |
1da177e4 | 414 | { |
558e6891 | 415 | offset >>= inode->i_blkbits; |
8699bb0a | 416 | |
558e6891 CH |
417 | return offset >= imap->br_startoff && |
418 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
419 | } |
420 | ||
f6d6d4fc CH |
421 | STATIC void |
422 | xfs_start_buffer_writeback( | |
423 | struct buffer_head *bh) | |
424 | { | |
425 | ASSERT(buffer_mapped(bh)); | |
426 | ASSERT(buffer_locked(bh)); | |
427 | ASSERT(!buffer_delay(bh)); | |
428 | ASSERT(!buffer_unwritten(bh)); | |
429 | ||
430 | mark_buffer_async_write(bh); | |
431 | set_buffer_uptodate(bh); | |
432 | clear_buffer_dirty(bh); | |
433 | } | |
434 | ||
435 | STATIC void | |
436 | xfs_start_page_writeback( | |
437 | struct page *page, | |
e10de372 | 438 | int clear_dirty) |
f6d6d4fc CH |
439 | { |
440 | ASSERT(PageLocked(page)); | |
441 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
442 | |
443 | /* | |
444 | * if the page was not fully cleaned, we need to ensure that the higher | |
445 | * layers come back to it correctly. That means we need to keep the page | |
446 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
447 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
448 | * write this page in this writeback sweep will be made. | |
449 | */ | |
450 | if (clear_dirty) { | |
92132021 | 451 | clear_page_dirty_for_io(page); |
0d085a52 DC |
452 | set_page_writeback(page); |
453 | } else | |
454 | set_page_writeback_keepwrite(page); | |
455 | ||
f6d6d4fc | 456 | unlock_page(page); |
f6d6d4fc CH |
457 | } |
458 | ||
c7c1a7d8 | 459 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
460 | { |
461 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
462 | } | |
463 | ||
464 | /* | |
bb18782a DC |
465 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
466 | * it, and we submit that bio. The ioend may be used for multiple bio | |
467 | * submissions, so we only want to allocate an append transaction for the ioend | |
468 | * once. In the case of multiple bio submission, each bio will take an IO | |
469 | * reference to the ioend to ensure that the ioend completion is only done once | |
470 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
471 | * |
472 | * If @fail is non-zero, it means that we have a situation where some part of | |
473 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
474 | * and unlocked them. In this situation, we need to fail the bio and ioend |
475 | * rather than submit it to IO. This typically only happens on a filesystem | |
476 | * shutdown. | |
f6d6d4fc | 477 | */ |
e10de372 | 478 | STATIC int |
f6d6d4fc | 479 | xfs_submit_ioend( |
06342cf8 | 480 | struct writeback_control *wbc, |
0e51a8e1 | 481 | struct xfs_ioend *ioend, |
e10de372 | 482 | int status) |
f6d6d4fc | 483 | { |
5eda4300 DW |
484 | /* Convert CoW extents to regular */ |
485 | if (!status && ioend->io_type == XFS_IO_COW) { | |
486 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), | |
487 | ioend->io_offset, ioend->io_size); | |
488 | } | |
489 | ||
e10de372 DC |
490 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
491 | if (!status && | |
0e51a8e1 | 492 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
493 | xfs_ioend_is_append(ioend) && |
494 | !ioend->io_append_trans) | |
e10de372 | 495 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 496 | |
0e51a8e1 CH |
497 | ioend->io_bio->bi_private = ioend; |
498 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
7637241e | 499 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
70fd7614 | 500 | |
e10de372 DC |
501 | /* |
502 | * If we are failing the IO now, just mark the ioend with an | |
503 | * error and finish it. This will run IO completion immediately | |
504 | * as there is only one reference to the ioend at this point in | |
505 | * time. | |
506 | */ | |
507 | if (status) { | |
0e51a8e1 CH |
508 | ioend->io_bio->bi_error = status; |
509 | bio_endio(ioend->io_bio); | |
e10de372 DC |
510 | return status; |
511 | } | |
d88992f6 | 512 | |
4e49ea4a | 513 | submit_bio(ioend->io_bio); |
e10de372 | 514 | return 0; |
f6d6d4fc | 515 | } |
f6d6d4fc | 516 | |
0e51a8e1 CH |
517 | static void |
518 | xfs_init_bio_from_bh( | |
519 | struct bio *bio, | |
520 | struct buffer_head *bh) | |
521 | { | |
522 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
523 | bio->bi_bdev = bh->b_bdev; | |
524 | } | |
7bf7f352 | 525 | |
0e51a8e1 CH |
526 | static struct xfs_ioend * |
527 | xfs_alloc_ioend( | |
528 | struct inode *inode, | |
529 | unsigned int type, | |
530 | xfs_off_t offset, | |
531 | struct buffer_head *bh) | |
532 | { | |
533 | struct xfs_ioend *ioend; | |
534 | struct bio *bio; | |
f6d6d4fc | 535 | |
0e51a8e1 CH |
536 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
537 | xfs_init_bio_from_bh(bio, bh); | |
538 | ||
539 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
540 | INIT_LIST_HEAD(&ioend->io_list); | |
541 | ioend->io_type = type; | |
542 | ioend->io_inode = inode; | |
543 | ioend->io_size = 0; | |
544 | ioend->io_offset = offset; | |
545 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
546 | ioend->io_append_trans = NULL; | |
547 | ioend->io_bio = bio; | |
548 | return ioend; | |
549 | } | |
550 | ||
551 | /* | |
552 | * Allocate a new bio, and chain the old bio to the new one. | |
553 | * | |
554 | * Note that we have to do perform the chaining in this unintuitive order | |
555 | * so that the bi_private linkage is set up in the right direction for the | |
556 | * traversal in xfs_destroy_ioend(). | |
557 | */ | |
558 | static void | |
559 | xfs_chain_bio( | |
560 | struct xfs_ioend *ioend, | |
561 | struct writeback_control *wbc, | |
562 | struct buffer_head *bh) | |
563 | { | |
564 | struct bio *new; | |
565 | ||
566 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
567 | xfs_init_bio_from_bh(new, bh); | |
568 | ||
569 | bio_chain(ioend->io_bio, new); | |
570 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
7637241e | 571 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
4e49ea4a | 572 | submit_bio(ioend->io_bio); |
0e51a8e1 | 573 | ioend->io_bio = new; |
f6d6d4fc CH |
574 | } |
575 | ||
576 | /* | |
577 | * Test to see if we've been building up a completion structure for | |
578 | * earlier buffers -- if so, we try to append to this ioend if we | |
579 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
580 | * Return the ioend we finished off so that the caller can submit it |
581 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
582 | */ |
583 | STATIC void | |
584 | xfs_add_to_ioend( | |
585 | struct inode *inode, | |
586 | struct buffer_head *bh, | |
7336cea8 | 587 | xfs_off_t offset, |
e10de372 | 588 | struct xfs_writepage_ctx *wpc, |
bb18782a | 589 | struct writeback_control *wbc, |
e10de372 | 590 | struct list_head *iolist) |
f6d6d4fc | 591 | { |
fbcc0256 | 592 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
593 | bh->b_blocknr != wpc->last_block + 1 || |
594 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
595 | if (wpc->ioend) |
596 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 597 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
598 | } |
599 | ||
0e51a8e1 CH |
600 | /* |
601 | * If the buffer doesn't fit into the bio we need to allocate a new | |
602 | * one. This shouldn't happen more than once for a given buffer. | |
603 | */ | |
604 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
605 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 606 | |
fbcc0256 DC |
607 | wpc->ioend->io_size += bh->b_size; |
608 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 609 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
610 | } |
611 | ||
87cbc49c NS |
612 | STATIC void |
613 | xfs_map_buffer( | |
046f1685 | 614 | struct inode *inode, |
87cbc49c | 615 | struct buffer_head *bh, |
207d0416 | 616 | struct xfs_bmbt_irec *imap, |
046f1685 | 617 | xfs_off_t offset) |
87cbc49c NS |
618 | { |
619 | sector_t bn; | |
8699bb0a | 620 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
621 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
622 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 623 | |
207d0416 CH |
624 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
625 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 626 | |
e513182d | 627 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 628 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 629 | |
046f1685 | 630 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
631 | |
632 | bh->b_blocknr = bn; | |
633 | set_buffer_mapped(bh); | |
634 | } | |
635 | ||
1da177e4 LT |
636 | STATIC void |
637 | xfs_map_at_offset( | |
046f1685 | 638 | struct inode *inode, |
1da177e4 | 639 | struct buffer_head *bh, |
207d0416 | 640 | struct xfs_bmbt_irec *imap, |
046f1685 | 641 | xfs_off_t offset) |
1da177e4 | 642 | { |
207d0416 CH |
643 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
644 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 645 | |
207d0416 | 646 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
647 | set_buffer_mapped(bh); |
648 | clear_buffer_delay(bh); | |
f6d6d4fc | 649 | clear_buffer_unwritten(bh); |
1da177e4 LT |
650 | } |
651 | ||
1da177e4 | 652 | /* |
a49935f2 DC |
653 | * Test if a given page contains at least one buffer of a given @type. |
654 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
655 | * try to find one of the type passed in. If it is not set, then the caller only | |
656 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 657 | */ |
a49935f2 | 658 | STATIC bool |
6ffc4db5 | 659 | xfs_check_page_type( |
10ce4444 | 660 | struct page *page, |
a49935f2 DC |
661 | unsigned int type, |
662 | bool check_all_buffers) | |
1da177e4 | 663 | { |
a49935f2 DC |
664 | struct buffer_head *bh; |
665 | struct buffer_head *head; | |
1da177e4 | 666 | |
a49935f2 DC |
667 | if (PageWriteback(page)) |
668 | return false; | |
669 | if (!page->mapping) | |
670 | return false; | |
671 | if (!page_has_buffers(page)) | |
672 | return false; | |
1da177e4 | 673 | |
a49935f2 DC |
674 | bh = head = page_buffers(page); |
675 | do { | |
676 | if (buffer_unwritten(bh)) { | |
677 | if (type == XFS_IO_UNWRITTEN) | |
678 | return true; | |
679 | } else if (buffer_delay(bh)) { | |
805eeb8e | 680 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
681 | return true; |
682 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 683 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
684 | return true; |
685 | } | |
1da177e4 | 686 | |
a49935f2 DC |
687 | /* If we are only checking the first buffer, we are done now. */ |
688 | if (!check_all_buffers) | |
689 | break; | |
690 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 691 | |
a49935f2 | 692 | return false; |
1da177e4 LT |
693 | } |
694 | ||
3ed3a434 DC |
695 | STATIC void |
696 | xfs_vm_invalidatepage( | |
697 | struct page *page, | |
d47992f8 LC |
698 | unsigned int offset, |
699 | unsigned int length) | |
3ed3a434 | 700 | { |
34097dfe LC |
701 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
702 | length); | |
703 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
704 | } |
705 | ||
706 | /* | |
707 | * If the page has delalloc buffers on it, we need to punch them out before we | |
708 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
709 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
710 | * is done on that same region - the delalloc extent is returned when none is | |
711 | * supposed to be there. | |
712 | * | |
713 | * We prevent this by truncating away the delalloc regions on the page before | |
714 | * invalidating it. Because they are delalloc, we can do this without needing a | |
715 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
716 | * truncation without a transaction as there is no space left for block | |
717 | * reservation (typically why we see a ENOSPC in writeback). | |
718 | * | |
719 | * This is not a performance critical path, so for now just do the punching a | |
720 | * buffer head at a time. | |
721 | */ | |
722 | STATIC void | |
723 | xfs_aops_discard_page( | |
724 | struct page *page) | |
725 | { | |
726 | struct inode *inode = page->mapping->host; | |
727 | struct xfs_inode *ip = XFS_I(inode); | |
728 | struct buffer_head *bh, *head; | |
729 | loff_t offset = page_offset(page); | |
3ed3a434 | 730 | |
a49935f2 | 731 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
732 | goto out_invalidate; |
733 | ||
e8c3753c DC |
734 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
735 | goto out_invalidate; | |
736 | ||
4f10700a | 737 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
738 | "page discard on page %p, inode 0x%llx, offset %llu.", |
739 | page, ip->i_ino, offset); | |
740 | ||
741 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
742 | bh = head = page_buffers(page); | |
743 | do { | |
3ed3a434 | 744 | int error; |
c726de44 | 745 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
746 | |
747 | if (!buffer_delay(bh)) | |
748 | goto next_buffer; | |
749 | ||
c726de44 DC |
750 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
751 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
752 | if (error) { |
753 | /* something screwed, just bail */ | |
e8c3753c | 754 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 755 | xfs_alert(ip->i_mount, |
3ed3a434 | 756 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 757 | } |
3ed3a434 DC |
758 | break; |
759 | } | |
760 | next_buffer: | |
93407472 | 761 | offset += i_blocksize(inode); |
3ed3a434 DC |
762 | |
763 | } while ((bh = bh->b_this_page) != head); | |
764 | ||
765 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
766 | out_invalidate: | |
09cbfeaf | 767 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
768 | return; |
769 | } | |
770 | ||
ef473667 DW |
771 | static int |
772 | xfs_map_cow( | |
773 | struct xfs_writepage_ctx *wpc, | |
774 | struct inode *inode, | |
775 | loff_t offset, | |
776 | unsigned int *new_type) | |
777 | { | |
778 | struct xfs_inode *ip = XFS_I(inode); | |
779 | struct xfs_bmbt_irec imap; | |
092d5d9d | 780 | bool is_cow = false; |
ef473667 DW |
781 | int error; |
782 | ||
783 | /* | |
784 | * If we already have a valid COW mapping keep using it. | |
785 | */ | |
786 | if (wpc->io_type == XFS_IO_COW) { | |
787 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
788 | if (wpc->imap_valid) { | |
789 | *new_type = XFS_IO_COW; | |
790 | return 0; | |
791 | } | |
792 | } | |
793 | ||
794 | /* | |
795 | * Else we need to check if there is a COW mapping at this offset. | |
796 | */ | |
797 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
092d5d9d | 798 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap); |
ef473667 DW |
799 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
800 | ||
801 | if (!is_cow) | |
802 | return 0; | |
803 | ||
804 | /* | |
805 | * And if the COW mapping has a delayed extent here we need to | |
806 | * allocate real space for it now. | |
807 | */ | |
092d5d9d | 808 | if (isnullstartblock(imap.br_startblock)) { |
ef473667 DW |
809 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, |
810 | &imap); | |
811 | if (error) | |
812 | return error; | |
813 | } | |
814 | ||
815 | wpc->io_type = *new_type = XFS_IO_COW; | |
816 | wpc->imap_valid = true; | |
817 | wpc->imap = imap; | |
818 | return 0; | |
819 | } | |
820 | ||
e10de372 DC |
821 | /* |
822 | * We implement an immediate ioend submission policy here to avoid needing to | |
823 | * chain multiple ioends and hence nest mempool allocations which can violate | |
824 | * forward progress guarantees we need to provide. The current ioend we are | |
825 | * adding buffers to is cached on the writepage context, and if the new buffer | |
826 | * does not append to the cached ioend it will create a new ioend and cache that | |
827 | * instead. | |
828 | * | |
829 | * If a new ioend is created and cached, the old ioend is returned and queued | |
830 | * locally for submission once the entire page is processed or an error has been | |
831 | * detected. While ioends are submitted immediately after they are completed, | |
832 | * batching optimisations are provided by higher level block plugging. | |
833 | * | |
834 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
835 | * writepage context that the caller will need to submit. | |
836 | */ | |
bfce7d2e DC |
837 | static int |
838 | xfs_writepage_map( | |
839 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 840 | struct writeback_control *wbc, |
bfce7d2e DC |
841 | struct inode *inode, |
842 | struct page *page, | |
843 | loff_t offset, | |
844 | __uint64_t end_offset) | |
845 | { | |
e10de372 DC |
846 | LIST_HEAD(submit_list); |
847 | struct xfs_ioend *ioend, *next; | |
bfce7d2e | 848 | struct buffer_head *bh, *head; |
93407472 | 849 | ssize_t len = i_blocksize(inode); |
bfce7d2e | 850 | int error = 0; |
bfce7d2e | 851 | int count = 0; |
e10de372 | 852 | int uptodate = 1; |
ef473667 | 853 | unsigned int new_type; |
bfce7d2e DC |
854 | |
855 | bh = head = page_buffers(page); | |
856 | offset = page_offset(page); | |
bfce7d2e DC |
857 | do { |
858 | if (offset >= end_offset) | |
859 | break; | |
860 | if (!buffer_uptodate(bh)) | |
861 | uptodate = 0; | |
862 | ||
863 | /* | |
864 | * set_page_dirty dirties all buffers in a page, independent | |
865 | * of their state. The dirty state however is entirely | |
866 | * meaningless for holes (!mapped && uptodate), so skip | |
867 | * buffers covering holes here. | |
868 | */ | |
869 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
870 | wpc->imap_valid = false; | |
871 | continue; | |
872 | } | |
873 | ||
ef473667 DW |
874 | if (buffer_unwritten(bh)) |
875 | new_type = XFS_IO_UNWRITTEN; | |
876 | else if (buffer_delay(bh)) | |
877 | new_type = XFS_IO_DELALLOC; | |
878 | else if (buffer_uptodate(bh)) | |
879 | new_type = XFS_IO_OVERWRITE; | |
880 | else { | |
bfce7d2e DC |
881 | if (PageUptodate(page)) |
882 | ASSERT(buffer_mapped(bh)); | |
883 | /* | |
884 | * This buffer is not uptodate and will not be | |
885 | * written to disk. Ensure that we will put any | |
886 | * subsequent writeable buffers into a new | |
887 | * ioend. | |
888 | */ | |
889 | wpc->imap_valid = false; | |
890 | continue; | |
891 | } | |
892 | ||
ef473667 DW |
893 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
894 | error = xfs_map_cow(wpc, inode, offset, &new_type); | |
895 | if (error) | |
896 | goto out; | |
897 | } | |
898 | ||
899 | if (wpc->io_type != new_type) { | |
900 | wpc->io_type = new_type; | |
901 | wpc->imap_valid = false; | |
902 | } | |
903 | ||
bfce7d2e DC |
904 | if (wpc->imap_valid) |
905 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
906 | offset); | |
907 | if (!wpc->imap_valid) { | |
908 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
909 | wpc->io_type); | |
910 | if (error) | |
e10de372 | 911 | goto out; |
bfce7d2e DC |
912 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
913 | offset); | |
914 | } | |
915 | if (wpc->imap_valid) { | |
916 | lock_buffer(bh); | |
917 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
918 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 919 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
920 | count++; |
921 | } | |
922 | ||
bfce7d2e DC |
923 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
924 | ||
925 | if (uptodate && bh == head) | |
926 | SetPageUptodate(page); | |
927 | ||
e10de372 | 928 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 929 | |
e10de372 | 930 | out: |
bfce7d2e | 931 | /* |
e10de372 DC |
932 | * On error, we have to fail the ioend here because we have locked |
933 | * buffers in the ioend. If we don't do this, we'll deadlock | |
934 | * invalidating the page as that tries to lock the buffers on the page. | |
935 | * Also, because we may have set pages under writeback, we have to make | |
936 | * sure we run IO completion to mark the error state of the IO | |
937 | * appropriately, so we can't cancel the ioend directly here. That means | |
938 | * we have to mark this page as under writeback if we included any | |
939 | * buffers from it in the ioend chain so that completion treats it | |
940 | * correctly. | |
bfce7d2e | 941 | * |
e10de372 DC |
942 | * If we didn't include the page in the ioend, the on error we can |
943 | * simply discard and unlock it as there are no other users of the page | |
944 | * or it's buffers right now. The caller will still need to trigger | |
945 | * submission of outstanding ioends on the writepage context so they are | |
946 | * treated correctly on error. | |
bfce7d2e | 947 | */ |
e10de372 DC |
948 | if (count) { |
949 | xfs_start_page_writeback(page, !error); | |
950 | ||
951 | /* | |
952 | * Preserve the original error if there was one, otherwise catch | |
953 | * submission errors here and propagate into subsequent ioend | |
954 | * submissions. | |
955 | */ | |
956 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
957 | int error2; | |
958 | ||
959 | list_del_init(&ioend->io_list); | |
960 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
961 | if (error2 && !error) | |
962 | error = error2; | |
963 | } | |
964 | } else if (error) { | |
bfce7d2e DC |
965 | xfs_aops_discard_page(page); |
966 | ClearPageUptodate(page); | |
967 | unlock_page(page); | |
e10de372 DC |
968 | } else { |
969 | /* | |
970 | * We can end up here with no error and nothing to write if we | |
971 | * race with a partial page truncate on a sub-page block sized | |
972 | * filesystem. In that case we need to mark the page clean. | |
973 | */ | |
974 | xfs_start_page_writeback(page, 1); | |
975 | end_page_writeback(page); | |
bfce7d2e | 976 | } |
e10de372 | 977 | |
bfce7d2e DC |
978 | mapping_set_error(page->mapping, error); |
979 | return error; | |
980 | } | |
981 | ||
1da177e4 | 982 | /* |
89f3b363 CH |
983 | * Write out a dirty page. |
984 | * | |
985 | * For delalloc space on the page we need to allocate space and flush it. | |
986 | * For unwritten space on the page we need to start the conversion to | |
987 | * regular allocated space. | |
89f3b363 | 988 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 989 | */ |
1da177e4 | 990 | STATIC int |
fbcc0256 | 991 | xfs_do_writepage( |
89f3b363 | 992 | struct page *page, |
fbcc0256 DC |
993 | struct writeback_control *wbc, |
994 | void *data) | |
1da177e4 | 995 | { |
fbcc0256 | 996 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 997 | struct inode *inode = page->mapping->host; |
1da177e4 | 998 | loff_t offset; |
1da177e4 | 999 | __uint64_t end_offset; |
ad68972a | 1000 | pgoff_t end_index; |
89f3b363 | 1001 | |
34097dfe | 1002 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 1003 | |
20cb52eb CH |
1004 | ASSERT(page_has_buffers(page)); |
1005 | ||
89f3b363 CH |
1006 | /* |
1007 | * Refuse to write the page out if we are called from reclaim context. | |
1008 | * | |
d4f7a5cb CH |
1009 | * This avoids stack overflows when called from deeply used stacks in |
1010 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1011 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 1012 | * |
94054fa3 MG |
1013 | * This should never happen except in the case of a VM regression so |
1014 | * warn about it. | |
89f3b363 | 1015 | */ |
94054fa3 MG |
1016 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
1017 | PF_MEMALLOC)) | |
b5420f23 | 1018 | goto redirty; |
1da177e4 | 1019 | |
89f3b363 | 1020 | /* |
680a647b CH |
1021 | * Given that we do not allow direct reclaim to call us, we should |
1022 | * never be called while in a filesystem transaction. | |
89f3b363 | 1023 | */ |
448011e2 | 1024 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 1025 | goto redirty; |
89f3b363 | 1026 | |
8695d27e | 1027 | /* |
ad68972a DC |
1028 | * Is this page beyond the end of the file? |
1029 | * | |
8695d27e JL |
1030 | * The page index is less than the end_index, adjust the end_offset |
1031 | * to the highest offset that this page should represent. | |
1032 | * ----------------------------------------------------- | |
1033 | * | file mapping | <EOF> | | |
1034 | * ----------------------------------------------------- | |
1035 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1036 | * ^--------------------------------^----------|-------- | |
1037 | * | desired writeback range | see else | | |
1038 | * ---------------------------------^------------------| | |
1039 | */ | |
ad68972a | 1040 | offset = i_size_read(inode); |
09cbfeaf | 1041 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 1042 | if (page->index < end_index) |
09cbfeaf | 1043 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
1044 | else { |
1045 | /* | |
1046 | * Check whether the page to write out is beyond or straddles | |
1047 | * i_size or not. | |
1048 | * ------------------------------------------------------- | |
1049 | * | file mapping | <EOF> | | |
1050 | * ------------------------------------------------------- | |
1051 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1052 | * ^--------------------------------^-----------|--------- | |
1053 | * | | Straddles | | |
1054 | * ---------------------------------^-----------|--------| | |
1055 | */ | |
09cbfeaf | 1056 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
1057 | |
1058 | /* | |
ff9a28f6 JK |
1059 | * Skip the page if it is fully outside i_size, e.g. due to a |
1060 | * truncate operation that is in progress. We must redirty the | |
1061 | * page so that reclaim stops reclaiming it. Otherwise | |
1062 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1063 | * |
1064 | * Note that the end_index is unsigned long, it would overflow | |
1065 | * if the given offset is greater than 16TB on 32-bit system | |
1066 | * and if we do check the page is fully outside i_size or not | |
1067 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1068 | * will be evaluated to 0. Hence this page will be redirtied | |
1069 | * and be written out repeatedly which would result in an | |
1070 | * infinite loop, the user program that perform this operation | |
1071 | * will hang. Instead, we can verify this situation by checking | |
1072 | * if the page to write is totally beyond the i_size or if it's | |
1073 | * offset is just equal to the EOF. | |
6b7a03f0 | 1074 | */ |
8695d27e JL |
1075 | if (page->index > end_index || |
1076 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1077 | goto redirty; |
6b7a03f0 CH |
1078 | |
1079 | /* | |
1080 | * The page straddles i_size. It must be zeroed out on each | |
1081 | * and every writepage invocation because it may be mmapped. | |
1082 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1083 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1084 | * memory is zeroed when mapped, and writes to that region are |
1085 | * not written out to the file." | |
1086 | */ | |
09cbfeaf | 1087 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
1088 | |
1089 | /* Adjust the end_offset to the end of file */ | |
1090 | end_offset = offset; | |
1da177e4 LT |
1091 | } |
1092 | ||
e10de372 | 1093 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
f51623b2 | 1094 | |
b5420f23 | 1095 | redirty: |
f51623b2 NS |
1096 | redirty_page_for_writepage(wbc, page); |
1097 | unlock_page(page); | |
1098 | return 0; | |
f51623b2 NS |
1099 | } |
1100 | ||
fbcc0256 DC |
1101 | STATIC int |
1102 | xfs_vm_writepage( | |
1103 | struct page *page, | |
1104 | struct writeback_control *wbc) | |
1105 | { | |
1106 | struct xfs_writepage_ctx wpc = { | |
1107 | .io_type = XFS_IO_INVALID, | |
1108 | }; | |
1109 | int ret; | |
1110 | ||
1111 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1112 | if (wpc.ioend) |
1113 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1114 | return ret; | |
fbcc0256 DC |
1115 | } |
1116 | ||
7d4fb40a NS |
1117 | STATIC int |
1118 | xfs_vm_writepages( | |
1119 | struct address_space *mapping, | |
1120 | struct writeback_control *wbc) | |
1121 | { | |
fbcc0256 DC |
1122 | struct xfs_writepage_ctx wpc = { |
1123 | .io_type = XFS_IO_INVALID, | |
1124 | }; | |
1125 | int ret; | |
1126 | ||
b3aea4ed | 1127 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7f6d5b52 RZ |
1128 | if (dax_mapping(mapping)) |
1129 | return dax_writeback_mapping_range(mapping, | |
1130 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1131 | ||
fbcc0256 | 1132 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1133 | if (wpc.ioend) |
1134 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1135 | return ret; | |
7d4fb40a NS |
1136 | } |
1137 | ||
f51623b2 NS |
1138 | /* |
1139 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1140 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1141 | * have buffer heads in this call. |
1142 | * | |
89f3b363 | 1143 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1144 | */ |
1145 | STATIC int | |
238f4c54 | 1146 | xfs_vm_releasepage( |
f51623b2 NS |
1147 | struct page *page, |
1148 | gfp_t gfp_mask) | |
1149 | { | |
20cb52eb | 1150 | int delalloc, unwritten; |
f51623b2 | 1151 | |
34097dfe | 1152 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1153 | |
99579cce BF |
1154 | /* |
1155 | * mm accommodates an old ext3 case where clean pages might not have had | |
1156 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1157 | * ->releasepage() via shrink_active_list(). Conversely, | |
1158 | * block_invalidatepage() can send pages that are still marked dirty | |
1159 | * but otherwise have invalidated buffers. | |
1160 | * | |
0a417b8d JK |
1161 | * We want to release the latter to avoid unnecessary buildup of the |
1162 | * LRU, skip the former and warn if we've left any lingering | |
1163 | * delalloc/unwritten buffers on clean pages. Skip pages with delalloc | |
1164 | * or unwritten buffers and warn if the page is not dirty. Otherwise | |
1165 | * try to release the buffers. | |
99579cce | 1166 | */ |
20cb52eb | 1167 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1168 | |
0a417b8d JK |
1169 | if (delalloc) { |
1170 | WARN_ON_ONCE(!PageDirty(page)); | |
f51623b2 | 1171 | return 0; |
0a417b8d JK |
1172 | } |
1173 | if (unwritten) { | |
1174 | WARN_ON_ONCE(!PageDirty(page)); | |
f51623b2 | 1175 | return 0; |
0a417b8d | 1176 | } |
f51623b2 | 1177 | |
f51623b2 NS |
1178 | return try_to_free_buffers(page); |
1179 | } | |
1180 | ||
1fdca9c2 DC |
1181 | /* |
1182 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1183 | * is, so that we can avoid repeated get_blocks calls. | |
1184 | * | |
1185 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1186 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1187 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1188 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1189 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1190 | * to and including EOF, and a second mapping for beyond EOF. | |
1191 | */ | |
1192 | static void | |
1193 | xfs_map_trim_size( | |
1194 | struct inode *inode, | |
1195 | sector_t iblock, | |
1196 | struct buffer_head *bh_result, | |
1197 | struct xfs_bmbt_irec *imap, | |
1198 | xfs_off_t offset, | |
1199 | ssize_t size) | |
1200 | { | |
1201 | xfs_off_t mapping_size; | |
1202 | ||
1203 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1204 | mapping_size <<= inode->i_blkbits; | |
1205 | ||
1206 | ASSERT(mapping_size > 0); | |
1207 | if (mapping_size > size) | |
1208 | mapping_size = size; | |
1209 | if (offset < i_size_read(inode) && | |
1210 | offset + mapping_size >= i_size_read(inode)) { | |
1211 | /* limit mapping to block that spans EOF */ | |
1212 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
93407472 | 1213 | i_blocksize(inode)); |
1fdca9c2 DC |
1214 | } |
1215 | if (mapping_size > LONG_MAX) | |
1216 | mapping_size = LONG_MAX; | |
1217 | ||
1218 | bh_result->b_size = mapping_size; | |
1219 | } | |
1220 | ||
0613f16c | 1221 | static int |
acdda3aa | 1222 | xfs_get_blocks( |
1da177e4 LT |
1223 | struct inode *inode, |
1224 | sector_t iblock, | |
1da177e4 | 1225 | struct buffer_head *bh_result, |
acdda3aa | 1226 | int create) |
1da177e4 | 1227 | { |
a206c817 CH |
1228 | struct xfs_inode *ip = XFS_I(inode); |
1229 | struct xfs_mount *mp = ip->i_mount; | |
1230 | xfs_fileoff_t offset_fsb, end_fsb; | |
1231 | int error = 0; | |
1232 | int lockmode = 0; | |
207d0416 | 1233 | struct xfs_bmbt_irec imap; |
a206c817 | 1234 | int nimaps = 1; |
fdc7ed75 NS |
1235 | xfs_off_t offset; |
1236 | ssize_t size; | |
a206c817 | 1237 | |
acdda3aa | 1238 | BUG_ON(create); |
6e8a27a8 | 1239 | |
a206c817 | 1240 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1241 | return -EIO; |
1da177e4 | 1242 | |
fdc7ed75 | 1243 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
93407472 | 1244 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
c2536668 | 1245 | size = bh_result->b_size; |
364f358a | 1246 | |
acdda3aa | 1247 | if (offset >= i_size_read(inode)) |
364f358a LM |
1248 | return 0; |
1249 | ||
507630b2 DC |
1250 | /* |
1251 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1252 | * a block mapping without an exclusive lock first. |
507630b2 | 1253 | */ |
6e8a27a8 | 1254 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1255 | |
d2c28191 DC |
1256 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1257 | if (offset + size > mp->m_super->s_maxbytes) | |
1258 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1259 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1260 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1261 | ||
acdda3aa CH |
1262 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1263 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1264 | if (error) |
a206c817 CH |
1265 | goto out_unlock; |
1266 | ||
acdda3aa | 1267 | if (nimaps) { |
d5cc2e3f DC |
1268 | trace_xfs_get_blocks_found(ip, offset, size, |
1269 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1270 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1271 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1272 | } else { |
1273 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1274 | goto out_unlock; | |
1275 | } | |
1da177e4 | 1276 | |
1fdca9c2 | 1277 | /* trim mapping down to size requested */ |
6e8a27a8 | 1278 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1279 | |
a719370b DC |
1280 | /* |
1281 | * For unwritten extents do not report a disk address in the buffered | |
1282 | * read case (treat as if we're reading into a hole). | |
1283 | */ | |
207d0416 | 1284 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b | 1285 | imap.br_startblock != DELAYSTARTBLOCK && |
acdda3aa | 1286 | !ISUNWRITTEN(&imap)) |
a719370b | 1287 | xfs_map_buffer(inode, bh_result, &imap, offset); |
1da177e4 | 1288 | |
c2536668 NS |
1289 | /* |
1290 | * If this is a realtime file, data may be on a different device. | |
1291 | * to that pointed to from the buffer_head b_bdev currently. | |
1292 | */ | |
046f1685 | 1293 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1294 | return 0; |
a206c817 CH |
1295 | |
1296 | out_unlock: | |
1297 | xfs_iunlock(ip, lockmode); | |
2451337d | 1298 | return error; |
1da177e4 LT |
1299 | } |
1300 | ||
c19b104a CH |
1301 | STATIC ssize_t |
1302 | xfs_vm_direct_IO( | |
6e1ba0bc | 1303 | struct kiocb *iocb, |
c8b8e32d | 1304 | struct iov_iter *iter) |
6e1ba0bc | 1305 | { |
58e59854 | 1306 | /* |
fa8d972d | 1307 | * We just need the method present so that open/fcntl allow direct I/O. |
58e59854 | 1308 | */ |
fa8d972d | 1309 | return -EINVAL; |
f51623b2 | 1310 | } |
1da177e4 LT |
1311 | |
1312 | STATIC sector_t | |
e4c573bb | 1313 | xfs_vm_bmap( |
1da177e4 LT |
1314 | struct address_space *mapping, |
1315 | sector_t block) | |
1316 | { | |
1317 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1318 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1319 | |
cca28fb8 | 1320 | trace_xfs_vm_bmap(XFS_I(inode)); |
db1327b1 DW |
1321 | |
1322 | /* | |
1323 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
1324 | * bypasseѕ the file system for actual I/O. We really can't allow | |
1325 | * that on reflinks inodes, so we have to skip out here. And yes, | |
1326 | * 0 is the magic code for a bmap error.. | |
1327 | */ | |
65523218 | 1328 | if (xfs_is_reflink_inode(ip)) |
db1327b1 | 1329 | return 0; |
65523218 | 1330 | |
4bc1ea6b | 1331 | filemap_write_and_wait(mapping); |
c2536668 | 1332 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1333 | } |
1334 | ||
1335 | STATIC int | |
e4c573bb | 1336 | xfs_vm_readpage( |
1da177e4 LT |
1337 | struct file *unused, |
1338 | struct page *page) | |
1339 | { | |
121e213e | 1340 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1341 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1342 | } |
1343 | ||
1344 | STATIC int | |
e4c573bb | 1345 | xfs_vm_readpages( |
1da177e4 LT |
1346 | struct file *unused, |
1347 | struct address_space *mapping, | |
1348 | struct list_head *pages, | |
1349 | unsigned nr_pages) | |
1350 | { | |
121e213e | 1351 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1352 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1353 | } |
1354 | ||
22e757a4 DC |
1355 | /* |
1356 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1357 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1358 | * dirty, we'll never be able to clean them because we don't write buffers | |
1359 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1360 | * that have been marked dirty. Further, the dirty state can leak into | |
1361 | * the file interior if the file is extended, resulting in all sorts of | |
1362 | * bad things happening as the state does not match the underlying data. | |
1363 | * | |
1364 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1365 | * this only exist because of bufferheads and how the generic code manages them. | |
1366 | */ | |
1367 | STATIC int | |
1368 | xfs_vm_set_page_dirty( | |
1369 | struct page *page) | |
1370 | { | |
1371 | struct address_space *mapping = page->mapping; | |
1372 | struct inode *inode = mapping->host; | |
1373 | loff_t end_offset; | |
1374 | loff_t offset; | |
1375 | int newly_dirty; | |
1376 | ||
1377 | if (unlikely(!mapping)) | |
1378 | return !TestSetPageDirty(page); | |
1379 | ||
1380 | end_offset = i_size_read(inode); | |
1381 | offset = page_offset(page); | |
1382 | ||
1383 | spin_lock(&mapping->private_lock); | |
1384 | if (page_has_buffers(page)) { | |
1385 | struct buffer_head *head = page_buffers(page); | |
1386 | struct buffer_head *bh = head; | |
1387 | ||
1388 | do { | |
1389 | if (offset < end_offset) | |
1390 | set_buffer_dirty(bh); | |
1391 | bh = bh->b_this_page; | |
93407472 | 1392 | offset += i_blocksize(inode); |
22e757a4 DC |
1393 | } while (bh != head); |
1394 | } | |
c4843a75 | 1395 | /* |
81f8c3a4 JW |
1396 | * Lock out page->mem_cgroup migration to keep PageDirty |
1397 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1398 | */ |
62cccb8c | 1399 | lock_page_memcg(page); |
22e757a4 DC |
1400 | newly_dirty = !TestSetPageDirty(page); |
1401 | spin_unlock(&mapping->private_lock); | |
1402 | ||
1403 | if (newly_dirty) { | |
1404 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1405 | unsigned long flags; | |
1406 | ||
1407 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1408 | if (page->mapping) { /* Race with truncate? */ | |
1409 | WARN_ON_ONCE(!PageUptodate(page)); | |
62cccb8c | 1410 | account_page_dirtied(page, mapping); |
22e757a4 DC |
1411 | radix_tree_tag_set(&mapping->page_tree, |
1412 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1413 | } | |
1414 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1415 | } |
62cccb8c | 1416 | unlock_page_memcg(page); |
c4843a75 GT |
1417 | if (newly_dirty) |
1418 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1419 | return newly_dirty; |
1420 | } | |
1421 | ||
f5e54d6e | 1422 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1423 | .readpage = xfs_vm_readpage, |
1424 | .readpages = xfs_vm_readpages, | |
1425 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1426 | .writepages = xfs_vm_writepages, |
22e757a4 | 1427 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1428 | .releasepage = xfs_vm_releasepage, |
1429 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb NS |
1430 | .bmap = xfs_vm_bmap, |
1431 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1432 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1433 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1434 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1435 | }; |