]>
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 { |
161f55ef EG |
114 | if (off > end) |
115 | break; | |
28b783e4 | 116 | next = bh->b_this_page; |
37992c18 DC |
117 | if (off < bvec->bv_offset) |
118 | goto next_bh; | |
37992c18 DC |
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 | */ | |
9070733b | 192 | current_restore_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
281627df CH |
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 | */ | |
9070733b | 255 | current_set_flags_nested(&tp->t_pflags, PF_MEMALLOC_NOFS); |
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); | |
787eb485 CH |
277 | xfs_off_t offset = ioend->io_offset; |
278 | size_t size = ioend->io_size; | |
4e4cbee9 | 279 | int error; |
ba87ea69 | 280 | |
af055e37 | 281 | /* |
787eb485 | 282 | * Just clean up the in-memory strutures if the fs has been shut down. |
af055e37 | 283 | */ |
787eb485 | 284 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
0e51a8e1 | 285 | error = -EIO; |
787eb485 CH |
286 | goto done; |
287 | } | |
04f658ee | 288 | |
43caeb18 | 289 | /* |
787eb485 | 290 | * Clean up any COW blocks on an I/O error. |
43caeb18 | 291 | */ |
4e4cbee9 | 292 | error = blk_status_to_errno(ioend->io_bio->bi_status); |
787eb485 CH |
293 | if (unlikely(error)) { |
294 | switch (ioend->io_type) { | |
295 | case XFS_IO_COW: | |
296 | xfs_reflink_cancel_cow_range(ip, offset, size, true); | |
297 | break; | |
43caeb18 | 298 | } |
787eb485 CH |
299 | |
300 | goto done; | |
43caeb18 DW |
301 | } |
302 | ||
5ec4fabb | 303 | /* |
787eb485 | 304 | * Success: commit the COW or unwritten blocks if needed. |
5ec4fabb | 305 | */ |
787eb485 CH |
306 | switch (ioend->io_type) { |
307 | case XFS_IO_COW: | |
308 | error = xfs_reflink_end_cow(ip, offset, size); | |
309 | break; | |
310 | case XFS_IO_UNWRITTEN: | |
311 | error = xfs_iomap_write_unwritten(ip, offset, size); | |
312 | break; | |
313 | default: | |
314 | ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans); | |
315 | break; | |
5ec4fabb | 316 | } |
ba87ea69 | 317 | |
04f658ee | 318 | done: |
787eb485 CH |
319 | if (ioend->io_append_trans) |
320 | error = xfs_setfilesize_ioend(ioend, error); | |
0e51a8e1 | 321 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
322 | } |
323 | ||
0e51a8e1 CH |
324 | STATIC void |
325 | xfs_end_bio( | |
326 | struct bio *bio) | |
0829c360 | 327 | { |
0e51a8e1 CH |
328 | struct xfs_ioend *ioend = bio->bi_private; |
329 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 330 | |
43caeb18 | 331 | if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW) |
0e51a8e1 CH |
332 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
333 | else if (ioend->io_append_trans) | |
334 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
335 | else | |
4e4cbee9 | 336 | xfs_destroy_ioend(ioend, blk_status_to_errno(bio->bi_status)); |
0829c360 CH |
337 | } |
338 | ||
1da177e4 LT |
339 | STATIC int |
340 | xfs_map_blocks( | |
341 | struct inode *inode, | |
342 | loff_t offset, | |
207d0416 | 343 | struct xfs_bmbt_irec *imap, |
988ef927 | 344 | int type) |
1da177e4 | 345 | { |
a206c817 CH |
346 | struct xfs_inode *ip = XFS_I(inode); |
347 | struct xfs_mount *mp = ip->i_mount; | |
93407472 | 348 | ssize_t count = i_blocksize(inode); |
a206c817 CH |
349 | xfs_fileoff_t offset_fsb, end_fsb; |
350 | int error = 0; | |
a206c817 CH |
351 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
352 | int nimaps = 1; | |
353 | ||
354 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 355 | return -EIO; |
a206c817 | 356 | |
ef473667 | 357 | ASSERT(type != XFS_IO_COW); |
0d882a36 | 358 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 359 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 360 | |
988ef927 | 361 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
362 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
363 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 364 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 365 | |
d2c28191 DC |
366 | if (offset + count > mp->m_super->s_maxbytes) |
367 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
368 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
369 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
370 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
371 | imap, &nimaps, bmapi_flags); | |
ef473667 DW |
372 | /* |
373 | * Truncate an overwrite extent if there's a pending CoW | |
374 | * reservation before the end of this extent. This forces us | |
375 | * to come back to writepage to take care of the CoW. | |
376 | */ | |
377 | if (nimaps && type == XFS_IO_OVERWRITE) | |
378 | xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap); | |
8ff2957d | 379 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 380 | |
8ff2957d | 381 | if (error) |
2451337d | 382 | return error; |
a206c817 | 383 | |
0d882a36 | 384 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 385 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
60b4984f DW |
386 | error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset, |
387 | imap); | |
a206c817 | 388 | if (!error) |
ef473667 | 389 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); |
2451337d | 390 | return error; |
a206c817 CH |
391 | } |
392 | ||
8ff2957d | 393 | #ifdef DEBUG |
0d882a36 | 394 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
395 | ASSERT(nimaps); |
396 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
397 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
398 | } | |
399 | #endif | |
400 | if (nimaps) | |
401 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
402 | return 0; | |
1da177e4 LT |
403 | } |
404 | ||
fbcc0256 | 405 | STATIC bool |
558e6891 | 406 | xfs_imap_valid( |
8699bb0a | 407 | struct inode *inode, |
207d0416 | 408 | struct xfs_bmbt_irec *imap, |
558e6891 | 409 | xfs_off_t offset) |
1da177e4 | 410 | { |
558e6891 | 411 | offset >>= inode->i_blkbits; |
8699bb0a | 412 | |
558e6891 CH |
413 | return offset >= imap->br_startoff && |
414 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
415 | } |
416 | ||
f6d6d4fc CH |
417 | STATIC void |
418 | xfs_start_buffer_writeback( | |
419 | struct buffer_head *bh) | |
420 | { | |
421 | ASSERT(buffer_mapped(bh)); | |
422 | ASSERT(buffer_locked(bh)); | |
423 | ASSERT(!buffer_delay(bh)); | |
424 | ASSERT(!buffer_unwritten(bh)); | |
425 | ||
426 | mark_buffer_async_write(bh); | |
427 | set_buffer_uptodate(bh); | |
428 | clear_buffer_dirty(bh); | |
429 | } | |
430 | ||
431 | STATIC void | |
432 | xfs_start_page_writeback( | |
433 | struct page *page, | |
e10de372 | 434 | int clear_dirty) |
f6d6d4fc CH |
435 | { |
436 | ASSERT(PageLocked(page)); | |
437 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
438 | |
439 | /* | |
440 | * if the page was not fully cleaned, we need to ensure that the higher | |
441 | * layers come back to it correctly. That means we need to keep the page | |
442 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
443 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
444 | * write this page in this writeback sweep will be made. | |
445 | */ | |
446 | if (clear_dirty) { | |
92132021 | 447 | clear_page_dirty_for_io(page); |
0d085a52 DC |
448 | set_page_writeback(page); |
449 | } else | |
450 | set_page_writeback_keepwrite(page); | |
451 | ||
f6d6d4fc | 452 | unlock_page(page); |
f6d6d4fc CH |
453 | } |
454 | ||
c7c1a7d8 | 455 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
456 | { |
457 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
458 | } | |
459 | ||
460 | /* | |
bb18782a DC |
461 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
462 | * it, and we submit that bio. The ioend may be used for multiple bio | |
463 | * submissions, so we only want to allocate an append transaction for the ioend | |
464 | * once. In the case of multiple bio submission, each bio will take an IO | |
465 | * reference to the ioend to ensure that the ioend completion is only done once | |
466 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
467 | * |
468 | * If @fail is non-zero, it means that we have a situation where some part of | |
469 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
470 | * and unlocked them. In this situation, we need to fail the bio and ioend |
471 | * rather than submit it to IO. This typically only happens on a filesystem | |
472 | * shutdown. | |
f6d6d4fc | 473 | */ |
e10de372 | 474 | STATIC int |
f6d6d4fc | 475 | xfs_submit_ioend( |
06342cf8 | 476 | struct writeback_control *wbc, |
0e51a8e1 | 477 | struct xfs_ioend *ioend, |
e10de372 | 478 | int status) |
f6d6d4fc | 479 | { |
5eda4300 DW |
480 | /* Convert CoW extents to regular */ |
481 | if (!status && ioend->io_type == XFS_IO_COW) { | |
482 | status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode), | |
483 | ioend->io_offset, ioend->io_size); | |
484 | } | |
485 | ||
e10de372 DC |
486 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
487 | if (!status && | |
0e51a8e1 | 488 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
489 | xfs_ioend_is_append(ioend) && |
490 | !ioend->io_append_trans) | |
e10de372 | 491 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 492 | |
0e51a8e1 CH |
493 | ioend->io_bio->bi_private = ioend; |
494 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
7637241e | 495 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
70fd7614 | 496 | |
e10de372 DC |
497 | /* |
498 | * If we are failing the IO now, just mark the ioend with an | |
499 | * error and finish it. This will run IO completion immediately | |
500 | * as there is only one reference to the ioend at this point in | |
501 | * time. | |
502 | */ | |
503 | if (status) { | |
4e4cbee9 | 504 | ioend->io_bio->bi_status = errno_to_blk_status(status); |
0e51a8e1 | 505 | bio_endio(ioend->io_bio); |
e10de372 DC |
506 | return status; |
507 | } | |
d88992f6 | 508 | |
31d7d58d | 509 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 510 | submit_bio(ioend->io_bio); |
e10de372 | 511 | return 0; |
f6d6d4fc | 512 | } |
f6d6d4fc | 513 | |
0e51a8e1 CH |
514 | static void |
515 | xfs_init_bio_from_bh( | |
516 | struct bio *bio, | |
517 | struct buffer_head *bh) | |
518 | { | |
519 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
520 | bio->bi_bdev = bh->b_bdev; | |
521 | } | |
7bf7f352 | 522 | |
0e51a8e1 CH |
523 | static struct xfs_ioend * |
524 | xfs_alloc_ioend( | |
525 | struct inode *inode, | |
526 | unsigned int type, | |
527 | xfs_off_t offset, | |
528 | struct buffer_head *bh) | |
529 | { | |
530 | struct xfs_ioend *ioend; | |
531 | struct bio *bio; | |
f6d6d4fc | 532 | |
0e51a8e1 CH |
533 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
534 | xfs_init_bio_from_bh(bio, bh); | |
535 | ||
536 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
537 | INIT_LIST_HEAD(&ioend->io_list); | |
538 | ioend->io_type = type; | |
539 | ioend->io_inode = inode; | |
540 | ioend->io_size = 0; | |
541 | ioend->io_offset = offset; | |
542 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
543 | ioend->io_append_trans = NULL; | |
544 | ioend->io_bio = bio; | |
545 | return ioend; | |
546 | } | |
547 | ||
548 | /* | |
549 | * Allocate a new bio, and chain the old bio to the new one. | |
550 | * | |
551 | * Note that we have to do perform the chaining in this unintuitive order | |
552 | * so that the bi_private linkage is set up in the right direction for the | |
553 | * traversal in xfs_destroy_ioend(). | |
554 | */ | |
555 | static void | |
556 | xfs_chain_bio( | |
557 | struct xfs_ioend *ioend, | |
558 | struct writeback_control *wbc, | |
559 | struct buffer_head *bh) | |
560 | { | |
561 | struct bio *new; | |
562 | ||
563 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
564 | xfs_init_bio_from_bh(new, bh); | |
565 | ||
566 | bio_chain(ioend->io_bio, new); | |
567 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
7637241e | 568 | ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc); |
31d7d58d | 569 | ioend->io_bio->bi_write_hint = ioend->io_inode->i_write_hint; |
4e49ea4a | 570 | submit_bio(ioend->io_bio); |
0e51a8e1 | 571 | ioend->io_bio = new; |
f6d6d4fc CH |
572 | } |
573 | ||
574 | /* | |
575 | * Test to see if we've been building up a completion structure for | |
576 | * earlier buffers -- if so, we try to append to this ioend if we | |
577 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
578 | * Return the ioend we finished off so that the caller can submit it |
579 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
580 | */ |
581 | STATIC void | |
582 | xfs_add_to_ioend( | |
583 | struct inode *inode, | |
584 | struct buffer_head *bh, | |
7336cea8 | 585 | xfs_off_t offset, |
e10de372 | 586 | struct xfs_writepage_ctx *wpc, |
bb18782a | 587 | struct writeback_control *wbc, |
e10de372 | 588 | struct list_head *iolist) |
f6d6d4fc | 589 | { |
fbcc0256 | 590 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
591 | bh->b_blocknr != wpc->last_block + 1 || |
592 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
593 | if (wpc->ioend) |
594 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 595 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
596 | } |
597 | ||
0e51a8e1 CH |
598 | /* |
599 | * If the buffer doesn't fit into the bio we need to allocate a new | |
600 | * one. This shouldn't happen more than once for a given buffer. | |
601 | */ | |
602 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
603 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 604 | |
fbcc0256 DC |
605 | wpc->ioend->io_size += bh->b_size; |
606 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 607 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
608 | } |
609 | ||
87cbc49c NS |
610 | STATIC void |
611 | xfs_map_buffer( | |
046f1685 | 612 | struct inode *inode, |
87cbc49c | 613 | struct buffer_head *bh, |
207d0416 | 614 | struct xfs_bmbt_irec *imap, |
046f1685 | 615 | xfs_off_t offset) |
87cbc49c NS |
616 | { |
617 | sector_t bn; | |
8699bb0a | 618 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
619 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
620 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 621 | |
207d0416 CH |
622 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
623 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 624 | |
e513182d | 625 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 626 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 627 | |
046f1685 | 628 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
629 | |
630 | bh->b_blocknr = bn; | |
631 | set_buffer_mapped(bh); | |
632 | } | |
633 | ||
1da177e4 LT |
634 | STATIC void |
635 | xfs_map_at_offset( | |
046f1685 | 636 | struct inode *inode, |
1da177e4 | 637 | struct buffer_head *bh, |
207d0416 | 638 | struct xfs_bmbt_irec *imap, |
046f1685 | 639 | xfs_off_t offset) |
1da177e4 | 640 | { |
207d0416 CH |
641 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
642 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 643 | |
207d0416 | 644 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
645 | set_buffer_mapped(bh); |
646 | clear_buffer_delay(bh); | |
f6d6d4fc | 647 | clear_buffer_unwritten(bh); |
1da177e4 LT |
648 | } |
649 | ||
1da177e4 | 650 | /* |
a49935f2 DC |
651 | * Test if a given page contains at least one buffer of a given @type. |
652 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
653 | * try to find one of the type passed in. If it is not set, then the caller only | |
654 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 655 | */ |
a49935f2 | 656 | STATIC bool |
6ffc4db5 | 657 | xfs_check_page_type( |
10ce4444 | 658 | struct page *page, |
a49935f2 DC |
659 | unsigned int type, |
660 | bool check_all_buffers) | |
1da177e4 | 661 | { |
a49935f2 DC |
662 | struct buffer_head *bh; |
663 | struct buffer_head *head; | |
1da177e4 | 664 | |
a49935f2 DC |
665 | if (PageWriteback(page)) |
666 | return false; | |
667 | if (!page->mapping) | |
668 | return false; | |
669 | if (!page_has_buffers(page)) | |
670 | return false; | |
1da177e4 | 671 | |
a49935f2 DC |
672 | bh = head = page_buffers(page); |
673 | do { | |
674 | if (buffer_unwritten(bh)) { | |
675 | if (type == XFS_IO_UNWRITTEN) | |
676 | return true; | |
677 | } else if (buffer_delay(bh)) { | |
805eeb8e | 678 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
679 | return true; |
680 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 681 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
682 | return true; |
683 | } | |
1da177e4 | 684 | |
a49935f2 DC |
685 | /* If we are only checking the first buffer, we are done now. */ |
686 | if (!check_all_buffers) | |
687 | break; | |
688 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 689 | |
a49935f2 | 690 | return false; |
1da177e4 LT |
691 | } |
692 | ||
3ed3a434 DC |
693 | STATIC void |
694 | xfs_vm_invalidatepage( | |
695 | struct page *page, | |
d47992f8 LC |
696 | unsigned int offset, |
697 | unsigned int length) | |
3ed3a434 | 698 | { |
34097dfe LC |
699 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
700 | length); | |
701 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
702 | } |
703 | ||
704 | /* | |
705 | * If the page has delalloc buffers on it, we need to punch them out before we | |
706 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
707 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
708 | * is done on that same region - the delalloc extent is returned when none is | |
709 | * supposed to be there. | |
710 | * | |
711 | * We prevent this by truncating away the delalloc regions on the page before | |
712 | * invalidating it. Because they are delalloc, we can do this without needing a | |
713 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
714 | * truncation without a transaction as there is no space left for block | |
715 | * reservation (typically why we see a ENOSPC in writeback). | |
716 | * | |
717 | * This is not a performance critical path, so for now just do the punching a | |
718 | * buffer head at a time. | |
719 | */ | |
720 | STATIC void | |
721 | xfs_aops_discard_page( | |
722 | struct page *page) | |
723 | { | |
724 | struct inode *inode = page->mapping->host; | |
725 | struct xfs_inode *ip = XFS_I(inode); | |
726 | struct buffer_head *bh, *head; | |
727 | loff_t offset = page_offset(page); | |
3ed3a434 | 728 | |
a49935f2 | 729 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
730 | goto out_invalidate; |
731 | ||
e8c3753c DC |
732 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
733 | goto out_invalidate; | |
734 | ||
4f10700a | 735 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
736 | "page discard on page %p, inode 0x%llx, offset %llu.", |
737 | page, ip->i_ino, offset); | |
738 | ||
739 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
740 | bh = head = page_buffers(page); | |
741 | do { | |
3ed3a434 | 742 | int error; |
c726de44 | 743 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
744 | |
745 | if (!buffer_delay(bh)) | |
746 | goto next_buffer; | |
747 | ||
c726de44 DC |
748 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
749 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
750 | if (error) { |
751 | /* something screwed, just bail */ | |
e8c3753c | 752 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 753 | xfs_alert(ip->i_mount, |
3ed3a434 | 754 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 755 | } |
3ed3a434 DC |
756 | break; |
757 | } | |
758 | next_buffer: | |
93407472 | 759 | offset += i_blocksize(inode); |
3ed3a434 DC |
760 | |
761 | } while ((bh = bh->b_this_page) != head); | |
762 | ||
763 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
764 | out_invalidate: | |
09cbfeaf | 765 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
766 | return; |
767 | } | |
768 | ||
ef473667 DW |
769 | static int |
770 | xfs_map_cow( | |
771 | struct xfs_writepage_ctx *wpc, | |
772 | struct inode *inode, | |
773 | loff_t offset, | |
774 | unsigned int *new_type) | |
775 | { | |
776 | struct xfs_inode *ip = XFS_I(inode); | |
777 | struct xfs_bmbt_irec imap; | |
092d5d9d | 778 | bool is_cow = false; |
ef473667 DW |
779 | int error; |
780 | ||
781 | /* | |
782 | * If we already have a valid COW mapping keep using it. | |
783 | */ | |
784 | if (wpc->io_type == XFS_IO_COW) { | |
785 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset); | |
786 | if (wpc->imap_valid) { | |
787 | *new_type = XFS_IO_COW; | |
788 | return 0; | |
789 | } | |
790 | } | |
791 | ||
792 | /* | |
793 | * Else we need to check if there is a COW mapping at this offset. | |
794 | */ | |
795 | xfs_ilock(ip, XFS_ILOCK_SHARED); | |
092d5d9d | 796 | is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap); |
ef473667 DW |
797 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
798 | ||
799 | if (!is_cow) | |
800 | return 0; | |
801 | ||
802 | /* | |
803 | * And if the COW mapping has a delayed extent here we need to | |
804 | * allocate real space for it now. | |
805 | */ | |
092d5d9d | 806 | if (isnullstartblock(imap.br_startblock)) { |
ef473667 DW |
807 | error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset, |
808 | &imap); | |
809 | if (error) | |
810 | return error; | |
811 | } | |
812 | ||
813 | wpc->io_type = *new_type = XFS_IO_COW; | |
814 | wpc->imap_valid = true; | |
815 | wpc->imap = imap; | |
816 | return 0; | |
817 | } | |
818 | ||
e10de372 DC |
819 | /* |
820 | * We implement an immediate ioend submission policy here to avoid needing to | |
821 | * chain multiple ioends and hence nest mempool allocations which can violate | |
822 | * forward progress guarantees we need to provide. The current ioend we are | |
823 | * adding buffers to is cached on the writepage context, and if the new buffer | |
824 | * does not append to the cached ioend it will create a new ioend and cache that | |
825 | * instead. | |
826 | * | |
827 | * If a new ioend is created and cached, the old ioend is returned and queued | |
828 | * locally for submission once the entire page is processed or an error has been | |
829 | * detected. While ioends are submitted immediately after they are completed, | |
830 | * batching optimisations are provided by higher level block plugging. | |
831 | * | |
832 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
833 | * writepage context that the caller will need to submit. | |
834 | */ | |
bfce7d2e DC |
835 | static int |
836 | xfs_writepage_map( | |
837 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 838 | struct writeback_control *wbc, |
bfce7d2e DC |
839 | struct inode *inode, |
840 | struct page *page, | |
841 | loff_t offset, | |
c8ce540d | 842 | uint64_t end_offset) |
bfce7d2e | 843 | { |
e10de372 DC |
844 | LIST_HEAD(submit_list); |
845 | struct xfs_ioend *ioend, *next; | |
bfce7d2e | 846 | struct buffer_head *bh, *head; |
93407472 | 847 | ssize_t len = i_blocksize(inode); |
bfce7d2e | 848 | int error = 0; |
bfce7d2e | 849 | int count = 0; |
e10de372 | 850 | int uptodate = 1; |
ef473667 | 851 | unsigned int new_type; |
bfce7d2e DC |
852 | |
853 | bh = head = page_buffers(page); | |
854 | offset = page_offset(page); | |
bfce7d2e DC |
855 | do { |
856 | if (offset >= end_offset) | |
857 | break; | |
858 | if (!buffer_uptodate(bh)) | |
859 | uptodate = 0; | |
860 | ||
861 | /* | |
862 | * set_page_dirty dirties all buffers in a page, independent | |
863 | * of their state. The dirty state however is entirely | |
864 | * meaningless for holes (!mapped && uptodate), so skip | |
865 | * buffers covering holes here. | |
866 | */ | |
867 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
868 | wpc->imap_valid = false; | |
869 | continue; | |
870 | } | |
871 | ||
ef473667 DW |
872 | if (buffer_unwritten(bh)) |
873 | new_type = XFS_IO_UNWRITTEN; | |
874 | else if (buffer_delay(bh)) | |
875 | new_type = XFS_IO_DELALLOC; | |
876 | else if (buffer_uptodate(bh)) | |
877 | new_type = XFS_IO_OVERWRITE; | |
878 | else { | |
bfce7d2e DC |
879 | if (PageUptodate(page)) |
880 | ASSERT(buffer_mapped(bh)); | |
881 | /* | |
882 | * This buffer is not uptodate and will not be | |
883 | * written to disk. Ensure that we will put any | |
884 | * subsequent writeable buffers into a new | |
885 | * ioend. | |
886 | */ | |
887 | wpc->imap_valid = false; | |
888 | continue; | |
889 | } | |
890 | ||
ef473667 DW |
891 | if (xfs_is_reflink_inode(XFS_I(inode))) { |
892 | error = xfs_map_cow(wpc, inode, offset, &new_type); | |
893 | if (error) | |
894 | goto out; | |
895 | } | |
896 | ||
897 | if (wpc->io_type != new_type) { | |
898 | wpc->io_type = new_type; | |
899 | wpc->imap_valid = false; | |
900 | } | |
901 | ||
bfce7d2e DC |
902 | if (wpc->imap_valid) |
903 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
904 | offset); | |
905 | if (!wpc->imap_valid) { | |
906 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
907 | wpc->io_type); | |
908 | if (error) | |
e10de372 | 909 | goto out; |
bfce7d2e DC |
910 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
911 | offset); | |
912 | } | |
913 | if (wpc->imap_valid) { | |
914 | lock_buffer(bh); | |
915 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
916 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 917 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
918 | count++; |
919 | } | |
920 | ||
bfce7d2e DC |
921 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
922 | ||
923 | if (uptodate && bh == head) | |
924 | SetPageUptodate(page); | |
925 | ||
e10de372 | 926 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 927 | |
e10de372 | 928 | out: |
bfce7d2e | 929 | /* |
e10de372 DC |
930 | * On error, we have to fail the ioend here because we have locked |
931 | * buffers in the ioend. If we don't do this, we'll deadlock | |
932 | * invalidating the page as that tries to lock the buffers on the page. | |
933 | * Also, because we may have set pages under writeback, we have to make | |
934 | * sure we run IO completion to mark the error state of the IO | |
935 | * appropriately, so we can't cancel the ioend directly here. That means | |
936 | * we have to mark this page as under writeback if we included any | |
937 | * buffers from it in the ioend chain so that completion treats it | |
938 | * correctly. | |
bfce7d2e | 939 | * |
e10de372 DC |
940 | * If we didn't include the page in the ioend, the on error we can |
941 | * simply discard and unlock it as there are no other users of the page | |
942 | * or it's buffers right now. The caller will still need to trigger | |
943 | * submission of outstanding ioends on the writepage context so they are | |
944 | * treated correctly on error. | |
bfce7d2e | 945 | */ |
e10de372 DC |
946 | if (count) { |
947 | xfs_start_page_writeback(page, !error); | |
948 | ||
949 | /* | |
950 | * Preserve the original error if there was one, otherwise catch | |
951 | * submission errors here and propagate into subsequent ioend | |
952 | * submissions. | |
953 | */ | |
954 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
955 | int error2; | |
956 | ||
957 | list_del_init(&ioend->io_list); | |
958 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
959 | if (error2 && !error) | |
960 | error = error2; | |
961 | } | |
962 | } else if (error) { | |
bfce7d2e DC |
963 | xfs_aops_discard_page(page); |
964 | ClearPageUptodate(page); | |
965 | unlock_page(page); | |
e10de372 DC |
966 | } else { |
967 | /* | |
968 | * We can end up here with no error and nothing to write if we | |
969 | * race with a partial page truncate on a sub-page block sized | |
970 | * filesystem. In that case we need to mark the page clean. | |
971 | */ | |
972 | xfs_start_page_writeback(page, 1); | |
973 | end_page_writeback(page); | |
bfce7d2e | 974 | } |
e10de372 | 975 | |
bfce7d2e DC |
976 | mapping_set_error(page->mapping, error); |
977 | return error; | |
978 | } | |
979 | ||
1da177e4 | 980 | /* |
89f3b363 CH |
981 | * Write out a dirty page. |
982 | * | |
983 | * For delalloc space on the page we need to allocate space and flush it. | |
984 | * For unwritten space on the page we need to start the conversion to | |
985 | * regular allocated space. | |
89f3b363 | 986 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 987 | */ |
1da177e4 | 988 | STATIC int |
fbcc0256 | 989 | xfs_do_writepage( |
89f3b363 | 990 | struct page *page, |
fbcc0256 DC |
991 | struct writeback_control *wbc, |
992 | void *data) | |
1da177e4 | 993 | { |
fbcc0256 | 994 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 995 | struct inode *inode = page->mapping->host; |
1da177e4 | 996 | loff_t offset; |
c8ce540d | 997 | uint64_t end_offset; |
ad68972a | 998 | pgoff_t end_index; |
89f3b363 | 999 | |
34097dfe | 1000 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 1001 | |
20cb52eb CH |
1002 | ASSERT(page_has_buffers(page)); |
1003 | ||
89f3b363 CH |
1004 | /* |
1005 | * Refuse to write the page out if we are called from reclaim context. | |
1006 | * | |
d4f7a5cb CH |
1007 | * This avoids stack overflows when called from deeply used stacks in |
1008 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
1009 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 1010 | * |
94054fa3 MG |
1011 | * This should never happen except in the case of a VM regression so |
1012 | * warn about it. | |
89f3b363 | 1013 | */ |
94054fa3 MG |
1014 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
1015 | PF_MEMALLOC)) | |
b5420f23 | 1016 | goto redirty; |
1da177e4 | 1017 | |
89f3b363 | 1018 | /* |
680a647b CH |
1019 | * Given that we do not allow direct reclaim to call us, we should |
1020 | * never be called while in a filesystem transaction. | |
89f3b363 | 1021 | */ |
9070733b | 1022 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC_NOFS)) |
b5420f23 | 1023 | goto redirty; |
89f3b363 | 1024 | |
8695d27e | 1025 | /* |
ad68972a DC |
1026 | * Is this page beyond the end of the file? |
1027 | * | |
8695d27e JL |
1028 | * The page index is less than the end_index, adjust the end_offset |
1029 | * to the highest offset that this page should represent. | |
1030 | * ----------------------------------------------------- | |
1031 | * | file mapping | <EOF> | | |
1032 | * ----------------------------------------------------- | |
1033 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1034 | * ^--------------------------------^----------|-------- | |
1035 | * | desired writeback range | see else | | |
1036 | * ---------------------------------^------------------| | |
1037 | */ | |
ad68972a | 1038 | offset = i_size_read(inode); |
09cbfeaf | 1039 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 1040 | if (page->index < end_index) |
09cbfeaf | 1041 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
1042 | else { |
1043 | /* | |
1044 | * Check whether the page to write out is beyond or straddles | |
1045 | * i_size or not. | |
1046 | * ------------------------------------------------------- | |
1047 | * | file mapping | <EOF> | | |
1048 | * ------------------------------------------------------- | |
1049 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1050 | * ^--------------------------------^-----------|--------- | |
1051 | * | | Straddles | | |
1052 | * ---------------------------------^-----------|--------| | |
1053 | */ | |
09cbfeaf | 1054 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
1055 | |
1056 | /* | |
ff9a28f6 JK |
1057 | * Skip the page if it is fully outside i_size, e.g. due to a |
1058 | * truncate operation that is in progress. We must redirty the | |
1059 | * page so that reclaim stops reclaiming it. Otherwise | |
1060 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1061 | * |
1062 | * Note that the end_index is unsigned long, it would overflow | |
1063 | * if the given offset is greater than 16TB on 32-bit system | |
1064 | * and if we do check the page is fully outside i_size or not | |
1065 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1066 | * will be evaluated to 0. Hence this page will be redirtied | |
1067 | * and be written out repeatedly which would result in an | |
1068 | * infinite loop, the user program that perform this operation | |
1069 | * will hang. Instead, we can verify this situation by checking | |
1070 | * if the page to write is totally beyond the i_size or if it's | |
1071 | * offset is just equal to the EOF. | |
6b7a03f0 | 1072 | */ |
8695d27e JL |
1073 | if (page->index > end_index || |
1074 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1075 | goto redirty; |
6b7a03f0 CH |
1076 | |
1077 | /* | |
1078 | * The page straddles i_size. It must be zeroed out on each | |
1079 | * and every writepage invocation because it may be mmapped. | |
1080 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1081 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1082 | * memory is zeroed when mapped, and writes to that region are |
1083 | * not written out to the file." | |
1084 | */ | |
09cbfeaf | 1085 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
1086 | |
1087 | /* Adjust the end_offset to the end of file */ | |
1088 | end_offset = offset; | |
1da177e4 LT |
1089 | } |
1090 | ||
e10de372 | 1091 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
f51623b2 | 1092 | |
b5420f23 | 1093 | redirty: |
f51623b2 NS |
1094 | redirty_page_for_writepage(wbc, page); |
1095 | unlock_page(page); | |
1096 | return 0; | |
f51623b2 NS |
1097 | } |
1098 | ||
fbcc0256 DC |
1099 | STATIC int |
1100 | xfs_vm_writepage( | |
1101 | struct page *page, | |
1102 | struct writeback_control *wbc) | |
1103 | { | |
1104 | struct xfs_writepage_ctx wpc = { | |
1105 | .io_type = XFS_IO_INVALID, | |
1106 | }; | |
1107 | int ret; | |
1108 | ||
1109 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1110 | if (wpc.ioend) |
1111 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1112 | return ret; | |
fbcc0256 DC |
1113 | } |
1114 | ||
7d4fb40a NS |
1115 | STATIC int |
1116 | xfs_vm_writepages( | |
1117 | struct address_space *mapping, | |
1118 | struct writeback_control *wbc) | |
1119 | { | |
fbcc0256 DC |
1120 | struct xfs_writepage_ctx wpc = { |
1121 | .io_type = XFS_IO_INVALID, | |
1122 | }; | |
1123 | int ret; | |
1124 | ||
b3aea4ed | 1125 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7f6d5b52 RZ |
1126 | if (dax_mapping(mapping)) |
1127 | return dax_writeback_mapping_range(mapping, | |
1128 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1129 | ||
fbcc0256 | 1130 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1131 | if (wpc.ioend) |
1132 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1133 | return ret; | |
7d4fb40a NS |
1134 | } |
1135 | ||
f51623b2 NS |
1136 | /* |
1137 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1138 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1139 | * have buffer heads in this call. |
1140 | * | |
89f3b363 | 1141 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1142 | */ |
1143 | STATIC int | |
238f4c54 | 1144 | xfs_vm_releasepage( |
f51623b2 NS |
1145 | struct page *page, |
1146 | gfp_t gfp_mask) | |
1147 | { | |
20cb52eb | 1148 | int delalloc, unwritten; |
f51623b2 | 1149 | |
34097dfe | 1150 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1151 | |
99579cce BF |
1152 | /* |
1153 | * mm accommodates an old ext3 case where clean pages might not have had | |
1154 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1155 | * ->releasepage() via shrink_active_list(). Conversely, | |
1156 | * block_invalidatepage() can send pages that are still marked dirty | |
1157 | * but otherwise have invalidated buffers. | |
1158 | * | |
0a417b8d JK |
1159 | * We want to release the latter to avoid unnecessary buildup of the |
1160 | * LRU, skip the former and warn if we've left any lingering | |
1161 | * delalloc/unwritten buffers on clean pages. Skip pages with delalloc | |
1162 | * or unwritten buffers and warn if the page is not dirty. Otherwise | |
1163 | * try to release the buffers. | |
99579cce | 1164 | */ |
20cb52eb | 1165 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1166 | |
0a417b8d JK |
1167 | if (delalloc) { |
1168 | WARN_ON_ONCE(!PageDirty(page)); | |
f51623b2 | 1169 | return 0; |
0a417b8d JK |
1170 | } |
1171 | if (unwritten) { | |
1172 | WARN_ON_ONCE(!PageDirty(page)); | |
f51623b2 | 1173 | return 0; |
0a417b8d | 1174 | } |
f51623b2 | 1175 | |
f51623b2 NS |
1176 | return try_to_free_buffers(page); |
1177 | } | |
1178 | ||
1fdca9c2 DC |
1179 | /* |
1180 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1181 | * is, so that we can avoid repeated get_blocks calls. | |
1182 | * | |
1183 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1184 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1185 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1186 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1187 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1188 | * to and including EOF, and a second mapping for beyond EOF. | |
1189 | */ | |
1190 | static void | |
1191 | xfs_map_trim_size( | |
1192 | struct inode *inode, | |
1193 | sector_t iblock, | |
1194 | struct buffer_head *bh_result, | |
1195 | struct xfs_bmbt_irec *imap, | |
1196 | xfs_off_t offset, | |
1197 | ssize_t size) | |
1198 | { | |
1199 | xfs_off_t mapping_size; | |
1200 | ||
1201 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1202 | mapping_size <<= inode->i_blkbits; | |
1203 | ||
1204 | ASSERT(mapping_size > 0); | |
1205 | if (mapping_size > size) | |
1206 | mapping_size = size; | |
1207 | if (offset < i_size_read(inode) && | |
1208 | offset + mapping_size >= i_size_read(inode)) { | |
1209 | /* limit mapping to block that spans EOF */ | |
1210 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
93407472 | 1211 | i_blocksize(inode)); |
1fdca9c2 DC |
1212 | } |
1213 | if (mapping_size > LONG_MAX) | |
1214 | mapping_size = LONG_MAX; | |
1215 | ||
1216 | bh_result->b_size = mapping_size; | |
1217 | } | |
1218 | ||
0613f16c | 1219 | static int |
acdda3aa | 1220 | xfs_get_blocks( |
1da177e4 LT |
1221 | struct inode *inode, |
1222 | sector_t iblock, | |
1da177e4 | 1223 | struct buffer_head *bh_result, |
acdda3aa | 1224 | int create) |
1da177e4 | 1225 | { |
a206c817 CH |
1226 | struct xfs_inode *ip = XFS_I(inode); |
1227 | struct xfs_mount *mp = ip->i_mount; | |
1228 | xfs_fileoff_t offset_fsb, end_fsb; | |
1229 | int error = 0; | |
1230 | int lockmode = 0; | |
207d0416 | 1231 | struct xfs_bmbt_irec imap; |
a206c817 | 1232 | int nimaps = 1; |
fdc7ed75 NS |
1233 | xfs_off_t offset; |
1234 | ssize_t size; | |
a206c817 | 1235 | |
acdda3aa | 1236 | BUG_ON(create); |
6e8a27a8 | 1237 | |
a206c817 | 1238 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1239 | return -EIO; |
1da177e4 | 1240 | |
fdc7ed75 | 1241 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
93407472 | 1242 | ASSERT(bh_result->b_size >= i_blocksize(inode)); |
c2536668 | 1243 | size = bh_result->b_size; |
364f358a | 1244 | |
acdda3aa | 1245 | if (offset >= i_size_read(inode)) |
364f358a LM |
1246 | return 0; |
1247 | ||
507630b2 DC |
1248 | /* |
1249 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1250 | * a block mapping without an exclusive lock first. |
507630b2 | 1251 | */ |
6e8a27a8 | 1252 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1253 | |
d2c28191 DC |
1254 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1255 | if (offset + size > mp->m_super->s_maxbytes) | |
1256 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1257 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1258 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1259 | ||
acdda3aa CH |
1260 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1261 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1262 | if (error) |
a206c817 CH |
1263 | goto out_unlock; |
1264 | ||
acdda3aa | 1265 | if (nimaps) { |
d5cc2e3f | 1266 | trace_xfs_get_blocks_found(ip, offset, size, |
63fbb4c1 CH |
1267 | imap.br_state == XFS_EXT_UNWRITTEN ? |
1268 | XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1269 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1270 | } else { |
1271 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1272 | goto out_unlock; | |
1273 | } | |
1da177e4 | 1274 | |
1fdca9c2 | 1275 | /* trim mapping down to size requested */ |
6e8a27a8 | 1276 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1277 | |
a719370b DC |
1278 | /* |
1279 | * For unwritten extents do not report a disk address in the buffered | |
1280 | * read case (treat as if we're reading into a hole). | |
1281 | */ | |
9c4f29d3 | 1282 | if (xfs_bmap_is_real_extent(&imap)) |
a719370b | 1283 | xfs_map_buffer(inode, bh_result, &imap, offset); |
1da177e4 | 1284 | |
c2536668 NS |
1285 | /* |
1286 | * If this is a realtime file, data may be on a different device. | |
1287 | * to that pointed to from the buffer_head b_bdev currently. | |
1288 | */ | |
046f1685 | 1289 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1290 | return 0; |
a206c817 CH |
1291 | |
1292 | out_unlock: | |
1293 | xfs_iunlock(ip, lockmode); | |
2451337d | 1294 | return error; |
1da177e4 LT |
1295 | } |
1296 | ||
c19b104a CH |
1297 | STATIC ssize_t |
1298 | xfs_vm_direct_IO( | |
6e1ba0bc | 1299 | struct kiocb *iocb, |
c8b8e32d | 1300 | struct iov_iter *iter) |
6e1ba0bc | 1301 | { |
58e59854 | 1302 | /* |
fa8d972d | 1303 | * We just need the method present so that open/fcntl allow direct I/O. |
58e59854 | 1304 | */ |
fa8d972d | 1305 | return -EINVAL; |
f51623b2 | 1306 | } |
1da177e4 LT |
1307 | |
1308 | STATIC sector_t | |
e4c573bb | 1309 | xfs_vm_bmap( |
1da177e4 LT |
1310 | struct address_space *mapping, |
1311 | sector_t block) | |
1312 | { | |
1313 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1314 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1315 | |
cca28fb8 | 1316 | trace_xfs_vm_bmap(XFS_I(inode)); |
db1327b1 DW |
1317 | |
1318 | /* | |
1319 | * The swap code (ab-)uses ->bmap to get a block mapping and then | |
1320 | * bypasseѕ the file system for actual I/O. We really can't allow | |
1321 | * that on reflinks inodes, so we have to skip out here. And yes, | |
eb5e248d DW |
1322 | * 0 is the magic code for a bmap error. |
1323 | * | |
1324 | * Since we don't pass back blockdev info, we can't return bmap | |
1325 | * information for rt files either. | |
db1327b1 | 1326 | */ |
eb5e248d | 1327 | if (xfs_is_reflink_inode(ip) || XFS_IS_REALTIME_INODE(ip)) |
db1327b1 | 1328 | return 0; |
65523218 | 1329 | |
4bc1ea6b | 1330 | filemap_write_and_wait(mapping); |
c2536668 | 1331 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1332 | } |
1333 | ||
1334 | STATIC int | |
e4c573bb | 1335 | xfs_vm_readpage( |
1da177e4 LT |
1336 | struct file *unused, |
1337 | struct page *page) | |
1338 | { | |
121e213e | 1339 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1340 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | STATIC int | |
e4c573bb | 1344 | xfs_vm_readpages( |
1da177e4 LT |
1345 | struct file *unused, |
1346 | struct address_space *mapping, | |
1347 | struct list_head *pages, | |
1348 | unsigned nr_pages) | |
1349 | { | |
121e213e | 1350 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1351 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1352 | } |
1353 | ||
22e757a4 DC |
1354 | /* |
1355 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1356 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1357 | * dirty, we'll never be able to clean them because we don't write buffers | |
1358 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1359 | * that have been marked dirty. Further, the dirty state can leak into | |
1360 | * the file interior if the file is extended, resulting in all sorts of | |
1361 | * bad things happening as the state does not match the underlying data. | |
1362 | * | |
1363 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1364 | * this only exist because of bufferheads and how the generic code manages them. | |
1365 | */ | |
1366 | STATIC int | |
1367 | xfs_vm_set_page_dirty( | |
1368 | struct page *page) | |
1369 | { | |
1370 | struct address_space *mapping = page->mapping; | |
1371 | struct inode *inode = mapping->host; | |
1372 | loff_t end_offset; | |
1373 | loff_t offset; | |
1374 | int newly_dirty; | |
1375 | ||
1376 | if (unlikely(!mapping)) | |
1377 | return !TestSetPageDirty(page); | |
1378 | ||
1379 | end_offset = i_size_read(inode); | |
1380 | offset = page_offset(page); | |
1381 | ||
1382 | spin_lock(&mapping->private_lock); | |
1383 | if (page_has_buffers(page)) { | |
1384 | struct buffer_head *head = page_buffers(page); | |
1385 | struct buffer_head *bh = head; | |
1386 | ||
1387 | do { | |
1388 | if (offset < end_offset) | |
1389 | set_buffer_dirty(bh); | |
1390 | bh = bh->b_this_page; | |
93407472 | 1391 | offset += i_blocksize(inode); |
22e757a4 DC |
1392 | } while (bh != head); |
1393 | } | |
c4843a75 | 1394 | /* |
81f8c3a4 JW |
1395 | * Lock out page->mem_cgroup migration to keep PageDirty |
1396 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1397 | */ |
62cccb8c | 1398 | lock_page_memcg(page); |
22e757a4 DC |
1399 | newly_dirty = !TestSetPageDirty(page); |
1400 | spin_unlock(&mapping->private_lock); | |
1401 | ||
1402 | if (newly_dirty) { | |
1403 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1404 | unsigned long flags; | |
1405 | ||
1406 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1407 | if (page->mapping) { /* Race with truncate? */ | |
1408 | WARN_ON_ONCE(!PageUptodate(page)); | |
62cccb8c | 1409 | account_page_dirtied(page, mapping); |
22e757a4 DC |
1410 | radix_tree_tag_set(&mapping->page_tree, |
1411 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1412 | } | |
1413 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1414 | } |
62cccb8c | 1415 | unlock_page_memcg(page); |
c4843a75 GT |
1416 | if (newly_dirty) |
1417 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1418 | return newly_dirty; |
1419 | } | |
1420 | ||
f5e54d6e | 1421 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1422 | .readpage = xfs_vm_readpage, |
1423 | .readpages = xfs_vm_readpages, | |
1424 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1425 | .writepages = xfs_vm_writepages, |
22e757a4 | 1426 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1427 | .releasepage = xfs_vm_releasepage, |
1428 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb NS |
1429 | .bmap = xfs_vm_bmap, |
1430 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1431 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1432 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1433 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1434 | }; |