]>
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" |
5a0e3ad6 | 34 | #include <linux/gfp.h> |
1da177e4 | 35 | #include <linux/mpage.h> |
10ce4444 | 36 | #include <linux/pagevec.h> |
1da177e4 LT |
37 | #include <linux/writeback.h> |
38 | ||
273dda76 CH |
39 | /* flags for direct write completions */ |
40 | #define XFS_DIO_FLAG_UNWRITTEN (1 << 0) | |
41 | #define XFS_DIO_FLAG_APPEND (1 << 1) | |
42 | ||
fbcc0256 DC |
43 | /* |
44 | * structure owned by writepages passed to individual writepage calls | |
45 | */ | |
46 | struct xfs_writepage_ctx { | |
47 | struct xfs_bmbt_irec imap; | |
48 | bool imap_valid; | |
49 | unsigned int io_type; | |
fbcc0256 DC |
50 | struct xfs_ioend *ioend; |
51 | sector_t last_block; | |
52 | }; | |
53 | ||
0b1b213f | 54 | void |
f51623b2 NS |
55 | xfs_count_page_state( |
56 | struct page *page, | |
57 | int *delalloc, | |
f51623b2 NS |
58 | int *unwritten) |
59 | { | |
60 | struct buffer_head *bh, *head; | |
61 | ||
20cb52eb | 62 | *delalloc = *unwritten = 0; |
f51623b2 NS |
63 | |
64 | bh = head = page_buffers(page); | |
65 | do { | |
20cb52eb | 66 | if (buffer_unwritten(bh)) |
f51623b2 NS |
67 | (*unwritten) = 1; |
68 | else if (buffer_delay(bh)) | |
69 | (*delalloc) = 1; | |
70 | } while ((bh = bh->b_this_page) != head); | |
71 | } | |
72 | ||
20a90f58 | 73 | struct block_device * |
6214ed44 | 74 | xfs_find_bdev_for_inode( |
046f1685 | 75 | struct inode *inode) |
6214ed44 | 76 | { |
046f1685 | 77 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
78 | struct xfs_mount *mp = ip->i_mount; |
79 | ||
71ddabb9 | 80 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
81 | return mp->m_rtdev_targp->bt_bdev; |
82 | else | |
83 | return mp->m_ddev_targp->bt_bdev; | |
84 | } | |
85 | ||
f6d6d4fc | 86 | /* |
37992c18 DC |
87 | * We're now finished for good with this page. Update the page state via the |
88 | * associated buffer_heads, paying attention to the start and end offsets that | |
89 | * we need to process on the page. | |
90 | */ | |
91 | static void | |
92 | xfs_finish_page_writeback( | |
93 | struct inode *inode, | |
94 | struct bio_vec *bvec, | |
95 | int error) | |
96 | { | |
37992c18 DC |
97 | unsigned int end = bvec->bv_offset + bvec->bv_len - 1; |
98 | struct buffer_head *head, *bh; | |
99 | unsigned int off = 0; | |
100 | ||
101 | ASSERT(bvec->bv_offset < PAGE_SIZE); | |
690a7871 | 102 | ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0); |
37992c18 | 103 | ASSERT(end < PAGE_SIZE); |
690a7871 | 104 | ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0); |
37992c18 DC |
105 | |
106 | bh = head = page_buffers(bvec->bv_page); | |
107 | ||
108 | do { | |
109 | if (off < bvec->bv_offset) | |
110 | goto next_bh; | |
111 | if (off > end) | |
112 | break; | |
113 | bh->b_end_io(bh, !error); | |
114 | next_bh: | |
115 | off += bh->b_size; | |
116 | } while ((bh = bh->b_this_page) != head); | |
117 | } | |
118 | ||
119 | /* | |
120 | * We're now finished for good with this ioend structure. Update the page | |
121 | * state, release holds on bios, and finally free up memory. Do not use the | |
122 | * ioend after this. | |
f6d6d4fc | 123 | */ |
0829c360 CH |
124 | STATIC void |
125 | xfs_destroy_ioend( | |
0e51a8e1 CH |
126 | struct xfs_ioend *ioend, |
127 | int error) | |
0829c360 | 128 | { |
37992c18 | 129 | struct inode *inode = ioend->io_inode; |
0e51a8e1 | 130 | struct bio *last = ioend->io_bio; |
37992c18 | 131 | struct bio *bio, *next; |
f6d6d4fc | 132 | |
0e51a8e1 | 133 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
37992c18 DC |
134 | struct bio_vec *bvec; |
135 | int i; | |
136 | ||
0e51a8e1 CH |
137 | /* |
138 | * For the last bio, bi_private points to the ioend, so we | |
139 | * need to explicitly end the iteration here. | |
140 | */ | |
141 | if (bio == last) | |
142 | next = NULL; | |
143 | else | |
144 | next = bio->bi_private; | |
583fa586 | 145 | |
37992c18 DC |
146 | /* walk each page on bio, ending page IO on them */ |
147 | bio_for_each_segment_all(bvec, bio, i) | |
148 | xfs_finish_page_writeback(inode, bvec, error); | |
149 | ||
150 | bio_put(bio); | |
f6d6d4fc | 151 | } |
0829c360 CH |
152 | } |
153 | ||
fc0063c4 CH |
154 | /* |
155 | * Fast and loose check if this write could update the on-disk inode size. | |
156 | */ | |
157 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
158 | { | |
159 | return ioend->io_offset + ioend->io_size > | |
160 | XFS_I(ioend->io_inode)->i_d.di_size; | |
161 | } | |
162 | ||
281627df CH |
163 | STATIC int |
164 | xfs_setfilesize_trans_alloc( | |
165 | struct xfs_ioend *ioend) | |
166 | { | |
167 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
168 | struct xfs_trans *tp; | |
169 | int error; | |
170 | ||
253f4911 CH |
171 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
172 | if (error) | |
281627df | 173 | return error; |
281627df CH |
174 | |
175 | ioend->io_append_trans = tp; | |
176 | ||
d9457dc0 | 177 | /* |
437a255a | 178 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
179 | * we released it. |
180 | */ | |
bee9182d | 181 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
182 | /* |
183 | * We hand off the transaction to the completion thread now, so | |
184 | * clear the flag here. | |
185 | */ | |
186 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
187 | return 0; | |
188 | } | |
189 | ||
ba87ea69 | 190 | /* |
2813d682 | 191 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 192 | */ |
281627df | 193 | STATIC int |
ba87ea69 | 194 | xfs_setfilesize( |
2ba66237 CH |
195 | struct xfs_inode *ip, |
196 | struct xfs_trans *tp, | |
197 | xfs_off_t offset, | |
198 | size_t size) | |
ba87ea69 | 199 | { |
ba87ea69 | 200 | xfs_fsize_t isize; |
ba87ea69 | 201 | |
aa6bf01d | 202 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 203 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
204 | if (!isize) { |
205 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 206 | xfs_trans_cancel(tp); |
281627df | 207 | return 0; |
ba87ea69 LM |
208 | } |
209 | ||
2ba66237 | 210 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
211 | |
212 | ip->i_d.di_size = isize; | |
213 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
214 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
215 | ||
70393313 | 216 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
217 | } |
218 | ||
2ba66237 CH |
219 | STATIC int |
220 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
221 | struct xfs_ioend *ioend, |
222 | int error) | |
2ba66237 CH |
223 | { |
224 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
225 | struct xfs_trans *tp = ioend->io_append_trans; | |
226 | ||
227 | /* | |
228 | * The transaction may have been allocated in the I/O submission thread, | |
229 | * thus we need to mark ourselves as being in a transaction manually. | |
230 | * Similarly for freeze protection. | |
231 | */ | |
232 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 233 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 234 | |
5cb13dcd | 235 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 236 | if (error) { |
5cb13dcd | 237 | xfs_trans_cancel(tp); |
0e51a8e1 | 238 | return error; |
5cb13dcd Z |
239 | } |
240 | ||
2ba66237 CH |
241 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
242 | } | |
243 | ||
0829c360 | 244 | /* |
5ec4fabb | 245 | * IO write completion. |
f6d6d4fc CH |
246 | */ |
247 | STATIC void | |
5ec4fabb | 248 | xfs_end_io( |
77d7a0c2 | 249 | struct work_struct *work) |
0829c360 | 250 | { |
0e51a8e1 CH |
251 | struct xfs_ioend *ioend = |
252 | container_of(work, struct xfs_ioend, io_work); | |
253 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
254 | int error = ioend->io_bio->bi_error; | |
ba87ea69 | 255 | |
af055e37 BF |
256 | /* |
257 | * Set an error if the mount has shut down and proceed with end I/O | |
258 | * processing so it can perform whatever cleanups are necessary. | |
259 | */ | |
260 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
0e51a8e1 | 261 | error = -EIO; |
04f658ee | 262 | |
5ec4fabb CH |
263 | /* |
264 | * For unwritten extents we need to issue transactions to convert a | |
265 | * range to normal written extens after the data I/O has finished. | |
5cb13dcd Z |
266 | * Detecting and handling completion IO errors is done individually |
267 | * for each case as different cleanup operations need to be performed | |
268 | * on error. | |
5ec4fabb | 269 | */ |
0d882a36 | 270 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
0e51a8e1 | 271 | if (error) |
5cb13dcd | 272 | goto done; |
437a255a DC |
273 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
274 | ioend->io_size); | |
281627df | 275 | } else if (ioend->io_append_trans) { |
0e51a8e1 | 276 | error = xfs_setfilesize_ioend(ioend, error); |
84803fb7 | 277 | } else { |
281627df | 278 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 279 | } |
ba87ea69 | 280 | |
04f658ee | 281 | done: |
0e51a8e1 | 282 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
283 | } |
284 | ||
0e51a8e1 CH |
285 | STATIC void |
286 | xfs_end_bio( | |
287 | struct bio *bio) | |
0829c360 | 288 | { |
0e51a8e1 CH |
289 | struct xfs_ioend *ioend = bio->bi_private; |
290 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 291 | |
0e51a8e1 CH |
292 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
293 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); | |
294 | else if (ioend->io_append_trans) | |
295 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
296 | else | |
297 | xfs_destroy_ioend(ioend, bio->bi_error); | |
0829c360 CH |
298 | } |
299 | ||
1da177e4 LT |
300 | STATIC int |
301 | xfs_map_blocks( | |
302 | struct inode *inode, | |
303 | loff_t offset, | |
207d0416 | 304 | struct xfs_bmbt_irec *imap, |
988ef927 | 305 | int type) |
1da177e4 | 306 | { |
a206c817 CH |
307 | struct xfs_inode *ip = XFS_I(inode); |
308 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 309 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
310 | xfs_fileoff_t offset_fsb, end_fsb; |
311 | int error = 0; | |
a206c817 CH |
312 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
313 | int nimaps = 1; | |
314 | ||
315 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 316 | return -EIO; |
a206c817 | 317 | |
0d882a36 | 318 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 319 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 320 | |
988ef927 | 321 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
322 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
323 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 324 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 325 | |
d2c28191 DC |
326 | if (offset + count > mp->m_super->s_maxbytes) |
327 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
328 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
329 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
330 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
331 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 332 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 333 | |
8ff2957d | 334 | if (error) |
2451337d | 335 | return error; |
a206c817 | 336 | |
0d882a36 | 337 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 338 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 339 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
340 | if (!error) |
341 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 342 | return error; |
a206c817 CH |
343 | } |
344 | ||
8ff2957d | 345 | #ifdef DEBUG |
0d882a36 | 346 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
347 | ASSERT(nimaps); |
348 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
349 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
350 | } | |
351 | #endif | |
352 | if (nimaps) | |
353 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
354 | return 0; | |
1da177e4 LT |
355 | } |
356 | ||
fbcc0256 | 357 | STATIC bool |
558e6891 | 358 | xfs_imap_valid( |
8699bb0a | 359 | struct inode *inode, |
207d0416 | 360 | struct xfs_bmbt_irec *imap, |
558e6891 | 361 | xfs_off_t offset) |
1da177e4 | 362 | { |
558e6891 | 363 | offset >>= inode->i_blkbits; |
8699bb0a | 364 | |
558e6891 CH |
365 | return offset >= imap->br_startoff && |
366 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
367 | } |
368 | ||
f6d6d4fc CH |
369 | STATIC void |
370 | xfs_start_buffer_writeback( | |
371 | struct buffer_head *bh) | |
372 | { | |
373 | ASSERT(buffer_mapped(bh)); | |
374 | ASSERT(buffer_locked(bh)); | |
375 | ASSERT(!buffer_delay(bh)); | |
376 | ASSERT(!buffer_unwritten(bh)); | |
377 | ||
378 | mark_buffer_async_write(bh); | |
379 | set_buffer_uptodate(bh); | |
380 | clear_buffer_dirty(bh); | |
381 | } | |
382 | ||
383 | STATIC void | |
384 | xfs_start_page_writeback( | |
385 | struct page *page, | |
e10de372 | 386 | int clear_dirty) |
f6d6d4fc CH |
387 | { |
388 | ASSERT(PageLocked(page)); | |
389 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
390 | |
391 | /* | |
392 | * if the page was not fully cleaned, we need to ensure that the higher | |
393 | * layers come back to it correctly. That means we need to keep the page | |
394 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
395 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
396 | * write this page in this writeback sweep will be made. | |
397 | */ | |
398 | if (clear_dirty) { | |
92132021 | 399 | clear_page_dirty_for_io(page); |
0d085a52 DC |
400 | set_page_writeback(page); |
401 | } else | |
402 | set_page_writeback_keepwrite(page); | |
403 | ||
f6d6d4fc | 404 | unlock_page(page); |
f6d6d4fc CH |
405 | } |
406 | ||
c7c1a7d8 | 407 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
408 | { |
409 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
410 | } | |
411 | ||
412 | /* | |
bb18782a DC |
413 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
414 | * it, and we submit that bio. The ioend may be used for multiple bio | |
415 | * submissions, so we only want to allocate an append transaction for the ioend | |
416 | * once. In the case of multiple bio submission, each bio will take an IO | |
417 | * reference to the ioend to ensure that the ioend completion is only done once | |
418 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
419 | * |
420 | * If @fail is non-zero, it means that we have a situation where some part of | |
421 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
422 | * and unlocked them. In this situation, we need to fail the bio and ioend |
423 | * rather than submit it to IO. This typically only happens on a filesystem | |
424 | * shutdown. | |
f6d6d4fc | 425 | */ |
e10de372 | 426 | STATIC int |
f6d6d4fc | 427 | xfs_submit_ioend( |
06342cf8 | 428 | struct writeback_control *wbc, |
0e51a8e1 | 429 | struct xfs_ioend *ioend, |
e10de372 | 430 | int status) |
f6d6d4fc | 431 | { |
e10de372 DC |
432 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
433 | if (!status && | |
0e51a8e1 | 434 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
435 | xfs_ioend_is_append(ioend) && |
436 | !ioend->io_append_trans) | |
e10de372 | 437 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 438 | |
0e51a8e1 CH |
439 | ioend->io_bio->bi_private = ioend; |
440 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
441 | ||
e10de372 DC |
442 | /* |
443 | * If we are failing the IO now, just mark the ioend with an | |
444 | * error and finish it. This will run IO completion immediately | |
445 | * as there is only one reference to the ioend at this point in | |
446 | * time. | |
447 | */ | |
448 | if (status) { | |
0e51a8e1 CH |
449 | ioend->io_bio->bi_error = status; |
450 | bio_endio(ioend->io_bio); | |
e10de372 DC |
451 | return status; |
452 | } | |
d88992f6 | 453 | |
0e51a8e1 CH |
454 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, |
455 | ioend->io_bio); | |
e10de372 | 456 | return 0; |
f6d6d4fc | 457 | } |
f6d6d4fc | 458 | |
0e51a8e1 CH |
459 | static void |
460 | xfs_init_bio_from_bh( | |
461 | struct bio *bio, | |
462 | struct buffer_head *bh) | |
463 | { | |
464 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
465 | bio->bi_bdev = bh->b_bdev; | |
466 | } | |
7bf7f352 | 467 | |
0e51a8e1 CH |
468 | static struct xfs_ioend * |
469 | xfs_alloc_ioend( | |
470 | struct inode *inode, | |
471 | unsigned int type, | |
472 | xfs_off_t offset, | |
473 | struct buffer_head *bh) | |
474 | { | |
475 | struct xfs_ioend *ioend; | |
476 | struct bio *bio; | |
f6d6d4fc | 477 | |
0e51a8e1 CH |
478 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
479 | xfs_init_bio_from_bh(bio, bh); | |
480 | ||
481 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
482 | INIT_LIST_HEAD(&ioend->io_list); | |
483 | ioend->io_type = type; | |
484 | ioend->io_inode = inode; | |
485 | ioend->io_size = 0; | |
486 | ioend->io_offset = offset; | |
487 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
488 | ioend->io_append_trans = NULL; | |
489 | ioend->io_bio = bio; | |
490 | return ioend; | |
491 | } | |
492 | ||
493 | /* | |
494 | * Allocate a new bio, and chain the old bio to the new one. | |
495 | * | |
496 | * Note that we have to do perform the chaining in this unintuitive order | |
497 | * so that the bi_private linkage is set up in the right direction for the | |
498 | * traversal in xfs_destroy_ioend(). | |
499 | */ | |
500 | static void | |
501 | xfs_chain_bio( | |
502 | struct xfs_ioend *ioend, | |
503 | struct writeback_control *wbc, | |
504 | struct buffer_head *bh) | |
505 | { | |
506 | struct bio *new; | |
507 | ||
508 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
509 | xfs_init_bio_from_bh(new, bh); | |
510 | ||
511 | bio_chain(ioend->io_bio, new); | |
512 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
513 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, | |
514 | ioend->io_bio); | |
515 | ioend->io_bio = new; | |
f6d6d4fc CH |
516 | } |
517 | ||
518 | /* | |
519 | * Test to see if we've been building up a completion structure for | |
520 | * earlier buffers -- if so, we try to append to this ioend if we | |
521 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
522 | * Return the ioend we finished off so that the caller can submit it |
523 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
524 | */ |
525 | STATIC void | |
526 | xfs_add_to_ioend( | |
527 | struct inode *inode, | |
528 | struct buffer_head *bh, | |
7336cea8 | 529 | xfs_off_t offset, |
e10de372 | 530 | struct xfs_writepage_ctx *wpc, |
bb18782a | 531 | struct writeback_control *wbc, |
e10de372 | 532 | struct list_head *iolist) |
f6d6d4fc | 533 | { |
fbcc0256 | 534 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
535 | bh->b_blocknr != wpc->last_block + 1 || |
536 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
537 | if (wpc->ioend) |
538 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 539 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
540 | } |
541 | ||
0e51a8e1 CH |
542 | /* |
543 | * If the buffer doesn't fit into the bio we need to allocate a new | |
544 | * one. This shouldn't happen more than once for a given buffer. | |
545 | */ | |
546 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
547 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 548 | |
fbcc0256 DC |
549 | wpc->ioend->io_size += bh->b_size; |
550 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 551 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
552 | } |
553 | ||
87cbc49c NS |
554 | STATIC void |
555 | xfs_map_buffer( | |
046f1685 | 556 | struct inode *inode, |
87cbc49c | 557 | struct buffer_head *bh, |
207d0416 | 558 | struct xfs_bmbt_irec *imap, |
046f1685 | 559 | xfs_off_t offset) |
87cbc49c NS |
560 | { |
561 | sector_t bn; | |
8699bb0a | 562 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
563 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
564 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 565 | |
207d0416 CH |
566 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
567 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 568 | |
e513182d | 569 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 570 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 571 | |
046f1685 | 572 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
573 | |
574 | bh->b_blocknr = bn; | |
575 | set_buffer_mapped(bh); | |
576 | } | |
577 | ||
1da177e4 LT |
578 | STATIC void |
579 | xfs_map_at_offset( | |
046f1685 | 580 | struct inode *inode, |
1da177e4 | 581 | struct buffer_head *bh, |
207d0416 | 582 | struct xfs_bmbt_irec *imap, |
046f1685 | 583 | xfs_off_t offset) |
1da177e4 | 584 | { |
207d0416 CH |
585 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
586 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 587 | |
207d0416 | 588 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
589 | set_buffer_mapped(bh); |
590 | clear_buffer_delay(bh); | |
f6d6d4fc | 591 | clear_buffer_unwritten(bh); |
1da177e4 LT |
592 | } |
593 | ||
1da177e4 | 594 | /* |
a49935f2 DC |
595 | * Test if a given page contains at least one buffer of a given @type. |
596 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
597 | * try to find one of the type passed in. If it is not set, then the caller only | |
598 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 599 | */ |
a49935f2 | 600 | STATIC bool |
6ffc4db5 | 601 | xfs_check_page_type( |
10ce4444 | 602 | struct page *page, |
a49935f2 DC |
603 | unsigned int type, |
604 | bool check_all_buffers) | |
1da177e4 | 605 | { |
a49935f2 DC |
606 | struct buffer_head *bh; |
607 | struct buffer_head *head; | |
1da177e4 | 608 | |
a49935f2 DC |
609 | if (PageWriteback(page)) |
610 | return false; | |
611 | if (!page->mapping) | |
612 | return false; | |
613 | if (!page_has_buffers(page)) | |
614 | return false; | |
1da177e4 | 615 | |
a49935f2 DC |
616 | bh = head = page_buffers(page); |
617 | do { | |
618 | if (buffer_unwritten(bh)) { | |
619 | if (type == XFS_IO_UNWRITTEN) | |
620 | return true; | |
621 | } else if (buffer_delay(bh)) { | |
805eeb8e | 622 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
623 | return true; |
624 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 625 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
626 | return true; |
627 | } | |
1da177e4 | 628 | |
a49935f2 DC |
629 | /* If we are only checking the first buffer, we are done now. */ |
630 | if (!check_all_buffers) | |
631 | break; | |
632 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 633 | |
a49935f2 | 634 | return false; |
1da177e4 LT |
635 | } |
636 | ||
3ed3a434 DC |
637 | STATIC void |
638 | xfs_vm_invalidatepage( | |
639 | struct page *page, | |
d47992f8 LC |
640 | unsigned int offset, |
641 | unsigned int length) | |
3ed3a434 | 642 | { |
34097dfe LC |
643 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
644 | length); | |
645 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
646 | } |
647 | ||
648 | /* | |
649 | * If the page has delalloc buffers on it, we need to punch them out before we | |
650 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
651 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
652 | * is done on that same region - the delalloc extent is returned when none is | |
653 | * supposed to be there. | |
654 | * | |
655 | * We prevent this by truncating away the delalloc regions on the page before | |
656 | * invalidating it. Because they are delalloc, we can do this without needing a | |
657 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
658 | * truncation without a transaction as there is no space left for block | |
659 | * reservation (typically why we see a ENOSPC in writeback). | |
660 | * | |
661 | * This is not a performance critical path, so for now just do the punching a | |
662 | * buffer head at a time. | |
663 | */ | |
664 | STATIC void | |
665 | xfs_aops_discard_page( | |
666 | struct page *page) | |
667 | { | |
668 | struct inode *inode = page->mapping->host; | |
669 | struct xfs_inode *ip = XFS_I(inode); | |
670 | struct buffer_head *bh, *head; | |
671 | loff_t offset = page_offset(page); | |
3ed3a434 | 672 | |
a49935f2 | 673 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
674 | goto out_invalidate; |
675 | ||
e8c3753c DC |
676 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
677 | goto out_invalidate; | |
678 | ||
4f10700a | 679 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
680 | "page discard on page %p, inode 0x%llx, offset %llu.", |
681 | page, ip->i_ino, offset); | |
682 | ||
683 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
684 | bh = head = page_buffers(page); | |
685 | do { | |
3ed3a434 | 686 | int error; |
c726de44 | 687 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
688 | |
689 | if (!buffer_delay(bh)) | |
690 | goto next_buffer; | |
691 | ||
c726de44 DC |
692 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
693 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
694 | if (error) { |
695 | /* something screwed, just bail */ | |
e8c3753c | 696 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 697 | xfs_alert(ip->i_mount, |
3ed3a434 | 698 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 699 | } |
3ed3a434 DC |
700 | break; |
701 | } | |
702 | next_buffer: | |
c726de44 | 703 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
704 | |
705 | } while ((bh = bh->b_this_page) != head); | |
706 | ||
707 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
708 | out_invalidate: | |
09cbfeaf | 709 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
710 | return; |
711 | } | |
712 | ||
e10de372 DC |
713 | /* |
714 | * We implement an immediate ioend submission policy here to avoid needing to | |
715 | * chain multiple ioends and hence nest mempool allocations which can violate | |
716 | * forward progress guarantees we need to provide. The current ioend we are | |
717 | * adding buffers to is cached on the writepage context, and if the new buffer | |
718 | * does not append to the cached ioend it will create a new ioend and cache that | |
719 | * instead. | |
720 | * | |
721 | * If a new ioend is created and cached, the old ioend is returned and queued | |
722 | * locally for submission once the entire page is processed or an error has been | |
723 | * detected. While ioends are submitted immediately after they are completed, | |
724 | * batching optimisations are provided by higher level block plugging. | |
725 | * | |
726 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
727 | * writepage context that the caller will need to submit. | |
728 | */ | |
bfce7d2e DC |
729 | static int |
730 | xfs_writepage_map( | |
731 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 732 | struct writeback_control *wbc, |
bfce7d2e DC |
733 | struct inode *inode, |
734 | struct page *page, | |
735 | loff_t offset, | |
736 | __uint64_t end_offset) | |
737 | { | |
e10de372 DC |
738 | LIST_HEAD(submit_list); |
739 | struct xfs_ioend *ioend, *next; | |
bfce7d2e DC |
740 | struct buffer_head *bh, *head; |
741 | ssize_t len = 1 << inode->i_blkbits; | |
742 | int error = 0; | |
bfce7d2e | 743 | int count = 0; |
e10de372 | 744 | int uptodate = 1; |
bfce7d2e DC |
745 | |
746 | bh = head = page_buffers(page); | |
747 | offset = page_offset(page); | |
bfce7d2e DC |
748 | do { |
749 | if (offset >= end_offset) | |
750 | break; | |
751 | if (!buffer_uptodate(bh)) | |
752 | uptodate = 0; | |
753 | ||
754 | /* | |
755 | * set_page_dirty dirties all buffers in a page, independent | |
756 | * of their state. The dirty state however is entirely | |
757 | * meaningless for holes (!mapped && uptodate), so skip | |
758 | * buffers covering holes here. | |
759 | */ | |
760 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
761 | wpc->imap_valid = false; | |
762 | continue; | |
763 | } | |
764 | ||
765 | if (buffer_unwritten(bh)) { | |
766 | if (wpc->io_type != XFS_IO_UNWRITTEN) { | |
767 | wpc->io_type = XFS_IO_UNWRITTEN; | |
768 | wpc->imap_valid = false; | |
769 | } | |
770 | } else if (buffer_delay(bh)) { | |
771 | if (wpc->io_type != XFS_IO_DELALLOC) { | |
772 | wpc->io_type = XFS_IO_DELALLOC; | |
773 | wpc->imap_valid = false; | |
774 | } | |
775 | } else if (buffer_uptodate(bh)) { | |
776 | if (wpc->io_type != XFS_IO_OVERWRITE) { | |
777 | wpc->io_type = XFS_IO_OVERWRITE; | |
778 | wpc->imap_valid = false; | |
779 | } | |
780 | } else { | |
781 | if (PageUptodate(page)) | |
782 | ASSERT(buffer_mapped(bh)); | |
783 | /* | |
784 | * This buffer is not uptodate and will not be | |
785 | * written to disk. Ensure that we will put any | |
786 | * subsequent writeable buffers into a new | |
787 | * ioend. | |
788 | */ | |
789 | wpc->imap_valid = false; | |
790 | continue; | |
791 | } | |
792 | ||
793 | if (wpc->imap_valid) | |
794 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
795 | offset); | |
796 | if (!wpc->imap_valid) { | |
797 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
798 | wpc->io_type); | |
799 | if (error) | |
e10de372 | 800 | goto out; |
bfce7d2e DC |
801 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
802 | offset); | |
803 | } | |
804 | if (wpc->imap_valid) { | |
805 | lock_buffer(bh); | |
806 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
807 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 808 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
809 | count++; |
810 | } | |
811 | ||
bfce7d2e DC |
812 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
813 | ||
814 | if (uptodate && bh == head) | |
815 | SetPageUptodate(page); | |
816 | ||
e10de372 | 817 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 818 | |
e10de372 | 819 | out: |
bfce7d2e | 820 | /* |
e10de372 DC |
821 | * On error, we have to fail the ioend here because we have locked |
822 | * buffers in the ioend. If we don't do this, we'll deadlock | |
823 | * invalidating the page as that tries to lock the buffers on the page. | |
824 | * Also, because we may have set pages under writeback, we have to make | |
825 | * sure we run IO completion to mark the error state of the IO | |
826 | * appropriately, so we can't cancel the ioend directly here. That means | |
827 | * we have to mark this page as under writeback if we included any | |
828 | * buffers from it in the ioend chain so that completion treats it | |
829 | * correctly. | |
bfce7d2e | 830 | * |
e10de372 DC |
831 | * If we didn't include the page in the ioend, the on error we can |
832 | * simply discard and unlock it as there are no other users of the page | |
833 | * or it's buffers right now. The caller will still need to trigger | |
834 | * submission of outstanding ioends on the writepage context so they are | |
835 | * treated correctly on error. | |
bfce7d2e | 836 | */ |
e10de372 DC |
837 | if (count) { |
838 | xfs_start_page_writeback(page, !error); | |
839 | ||
840 | /* | |
841 | * Preserve the original error if there was one, otherwise catch | |
842 | * submission errors here and propagate into subsequent ioend | |
843 | * submissions. | |
844 | */ | |
845 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
846 | int error2; | |
847 | ||
848 | list_del_init(&ioend->io_list); | |
849 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
850 | if (error2 && !error) | |
851 | error = error2; | |
852 | } | |
853 | } else if (error) { | |
bfce7d2e DC |
854 | xfs_aops_discard_page(page); |
855 | ClearPageUptodate(page); | |
856 | unlock_page(page); | |
e10de372 DC |
857 | } else { |
858 | /* | |
859 | * We can end up here with no error and nothing to write if we | |
860 | * race with a partial page truncate on a sub-page block sized | |
861 | * filesystem. In that case we need to mark the page clean. | |
862 | */ | |
863 | xfs_start_page_writeback(page, 1); | |
864 | end_page_writeback(page); | |
bfce7d2e | 865 | } |
e10de372 | 866 | |
bfce7d2e DC |
867 | mapping_set_error(page->mapping, error); |
868 | return error; | |
869 | } | |
870 | ||
1da177e4 | 871 | /* |
89f3b363 CH |
872 | * Write out a dirty page. |
873 | * | |
874 | * For delalloc space on the page we need to allocate space and flush it. | |
875 | * For unwritten space on the page we need to start the conversion to | |
876 | * regular allocated space. | |
89f3b363 | 877 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 878 | */ |
1da177e4 | 879 | STATIC int |
fbcc0256 | 880 | xfs_do_writepage( |
89f3b363 | 881 | struct page *page, |
fbcc0256 DC |
882 | struct writeback_control *wbc, |
883 | void *data) | |
1da177e4 | 884 | { |
fbcc0256 | 885 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 886 | struct inode *inode = page->mapping->host; |
1da177e4 | 887 | loff_t offset; |
1da177e4 | 888 | __uint64_t end_offset; |
ad68972a | 889 | pgoff_t end_index; |
89f3b363 | 890 | |
34097dfe | 891 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 892 | |
20cb52eb CH |
893 | ASSERT(page_has_buffers(page)); |
894 | ||
89f3b363 CH |
895 | /* |
896 | * Refuse to write the page out if we are called from reclaim context. | |
897 | * | |
d4f7a5cb CH |
898 | * This avoids stack overflows when called from deeply used stacks in |
899 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
900 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 901 | * |
94054fa3 MG |
902 | * This should never happen except in the case of a VM regression so |
903 | * warn about it. | |
89f3b363 | 904 | */ |
94054fa3 MG |
905 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
906 | PF_MEMALLOC)) | |
b5420f23 | 907 | goto redirty; |
1da177e4 | 908 | |
89f3b363 | 909 | /* |
680a647b CH |
910 | * Given that we do not allow direct reclaim to call us, we should |
911 | * never be called while in a filesystem transaction. | |
89f3b363 | 912 | */ |
448011e2 | 913 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 914 | goto redirty; |
89f3b363 | 915 | |
8695d27e | 916 | /* |
ad68972a DC |
917 | * Is this page beyond the end of the file? |
918 | * | |
8695d27e JL |
919 | * The page index is less than the end_index, adjust the end_offset |
920 | * to the highest offset that this page should represent. | |
921 | * ----------------------------------------------------- | |
922 | * | file mapping | <EOF> | | |
923 | * ----------------------------------------------------- | |
924 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
925 | * ^--------------------------------^----------|-------- | |
926 | * | desired writeback range | see else | | |
927 | * ---------------------------------^------------------| | |
928 | */ | |
ad68972a | 929 | offset = i_size_read(inode); |
09cbfeaf | 930 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 931 | if (page->index < end_index) |
09cbfeaf | 932 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
933 | else { |
934 | /* | |
935 | * Check whether the page to write out is beyond or straddles | |
936 | * i_size or not. | |
937 | * ------------------------------------------------------- | |
938 | * | file mapping | <EOF> | | |
939 | * ------------------------------------------------------- | |
940 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
941 | * ^--------------------------------^-----------|--------- | |
942 | * | | Straddles | | |
943 | * ---------------------------------^-----------|--------| | |
944 | */ | |
09cbfeaf | 945 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
946 | |
947 | /* | |
ff9a28f6 JK |
948 | * Skip the page if it is fully outside i_size, e.g. due to a |
949 | * truncate operation that is in progress. We must redirty the | |
950 | * page so that reclaim stops reclaiming it. Otherwise | |
951 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
952 | * |
953 | * Note that the end_index is unsigned long, it would overflow | |
954 | * if the given offset is greater than 16TB on 32-bit system | |
955 | * and if we do check the page is fully outside i_size or not | |
956 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
957 | * will be evaluated to 0. Hence this page will be redirtied | |
958 | * and be written out repeatedly which would result in an | |
959 | * infinite loop, the user program that perform this operation | |
960 | * will hang. Instead, we can verify this situation by checking | |
961 | * if the page to write is totally beyond the i_size or if it's | |
962 | * offset is just equal to the EOF. | |
6b7a03f0 | 963 | */ |
8695d27e JL |
964 | if (page->index > end_index || |
965 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 966 | goto redirty; |
6b7a03f0 CH |
967 | |
968 | /* | |
969 | * The page straddles i_size. It must be zeroed out on each | |
970 | * and every writepage invocation because it may be mmapped. | |
971 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 972 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
973 | * memory is zeroed when mapped, and writes to that region are |
974 | * not written out to the file." | |
975 | */ | |
09cbfeaf | 976 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
977 | |
978 | /* Adjust the end_offset to the end of file */ | |
979 | end_offset = offset; | |
1da177e4 LT |
980 | } |
981 | ||
e10de372 | 982 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
f51623b2 | 983 | |
b5420f23 | 984 | redirty: |
f51623b2 NS |
985 | redirty_page_for_writepage(wbc, page); |
986 | unlock_page(page); | |
987 | return 0; | |
f51623b2 NS |
988 | } |
989 | ||
fbcc0256 DC |
990 | STATIC int |
991 | xfs_vm_writepage( | |
992 | struct page *page, | |
993 | struct writeback_control *wbc) | |
994 | { | |
995 | struct xfs_writepage_ctx wpc = { | |
996 | .io_type = XFS_IO_INVALID, | |
997 | }; | |
998 | int ret; | |
999 | ||
1000 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1001 | if (wpc.ioend) |
1002 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1003 | return ret; | |
fbcc0256 DC |
1004 | } |
1005 | ||
7d4fb40a NS |
1006 | STATIC int |
1007 | xfs_vm_writepages( | |
1008 | struct address_space *mapping, | |
1009 | struct writeback_control *wbc) | |
1010 | { | |
fbcc0256 DC |
1011 | struct xfs_writepage_ctx wpc = { |
1012 | .io_type = XFS_IO_INVALID, | |
1013 | }; | |
1014 | int ret; | |
1015 | ||
b3aea4ed | 1016 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7f6d5b52 RZ |
1017 | if (dax_mapping(mapping)) |
1018 | return dax_writeback_mapping_range(mapping, | |
1019 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1020 | ||
fbcc0256 | 1021 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1022 | if (wpc.ioend) |
1023 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1024 | return ret; | |
7d4fb40a NS |
1025 | } |
1026 | ||
f51623b2 NS |
1027 | /* |
1028 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1029 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1030 | * have buffer heads in this call. |
1031 | * | |
89f3b363 | 1032 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1033 | */ |
1034 | STATIC int | |
238f4c54 | 1035 | xfs_vm_releasepage( |
f51623b2 NS |
1036 | struct page *page, |
1037 | gfp_t gfp_mask) | |
1038 | { | |
20cb52eb | 1039 | int delalloc, unwritten; |
f51623b2 | 1040 | |
34097dfe | 1041 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1042 | |
20cb52eb | 1043 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1044 | |
448011e2 | 1045 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1046 | return 0; |
448011e2 | 1047 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1048 | return 0; |
1049 | ||
f51623b2 NS |
1050 | return try_to_free_buffers(page); |
1051 | } | |
1052 | ||
a719370b | 1053 | /* |
273dda76 CH |
1054 | * When we map a DIO buffer, we may need to pass flags to |
1055 | * xfs_end_io_direct_write to tell it what kind of write IO we are doing. | |
3e12dbbd DC |
1056 | * |
1057 | * Note that for DIO, an IO to the highest supported file block offset (i.e. | |
1058 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 | |
1059 | * bit variable. Hence if we see this overflow, we have to assume that the IO is | |
1060 | * extending the file size. We won't know for sure until IO completion is run | |
1061 | * and the actual max write offset is communicated to the IO completion | |
1062 | * routine. | |
a719370b DC |
1063 | */ |
1064 | static void | |
1065 | xfs_map_direct( | |
1066 | struct inode *inode, | |
1067 | struct buffer_head *bh_result, | |
1068 | struct xfs_bmbt_irec *imap, | |
273dda76 | 1069 | xfs_off_t offset) |
a719370b | 1070 | { |
273dda76 | 1071 | uintptr_t *flags = (uintptr_t *)&bh_result->b_private; |
d5cc2e3f | 1072 | xfs_off_t size = bh_result->b_size; |
d5cc2e3f | 1073 | |
273dda76 CH |
1074 | trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size, |
1075 | ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap); | |
d5cc2e3f | 1076 | |
273dda76 CH |
1077 | if (ISUNWRITTEN(imap)) { |
1078 | *flags |= XFS_DIO_FLAG_UNWRITTEN; | |
1079 | set_buffer_defer_completion(bh_result); | |
1080 | } else if (offset + size > i_size_read(inode) || offset + size < 0) { | |
1081 | *flags |= XFS_DIO_FLAG_APPEND; | |
a06c277a | 1082 | set_buffer_defer_completion(bh_result); |
a719370b DC |
1083 | } |
1084 | } | |
1085 | ||
1fdca9c2 DC |
1086 | /* |
1087 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1088 | * is, so that we can avoid repeated get_blocks calls. | |
1089 | * | |
1090 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1091 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1092 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1093 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1094 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1095 | * to and including EOF, and a second mapping for beyond EOF. | |
1096 | */ | |
1097 | static void | |
1098 | xfs_map_trim_size( | |
1099 | struct inode *inode, | |
1100 | sector_t iblock, | |
1101 | struct buffer_head *bh_result, | |
1102 | struct xfs_bmbt_irec *imap, | |
1103 | xfs_off_t offset, | |
1104 | ssize_t size) | |
1105 | { | |
1106 | xfs_off_t mapping_size; | |
1107 | ||
1108 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1109 | mapping_size <<= inode->i_blkbits; | |
1110 | ||
1111 | ASSERT(mapping_size > 0); | |
1112 | if (mapping_size > size) | |
1113 | mapping_size = size; | |
1114 | if (offset < i_size_read(inode) && | |
1115 | offset + mapping_size >= i_size_read(inode)) { | |
1116 | /* limit mapping to block that spans EOF */ | |
1117 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1118 | 1 << inode->i_blkbits); | |
1119 | } | |
1120 | if (mapping_size > LONG_MAX) | |
1121 | mapping_size = LONG_MAX; | |
1122 | ||
1123 | bh_result->b_size = mapping_size; | |
1124 | } | |
1125 | ||
1da177e4 | 1126 | STATIC int |
c2536668 | 1127 | __xfs_get_blocks( |
1da177e4 LT |
1128 | struct inode *inode, |
1129 | sector_t iblock, | |
1da177e4 LT |
1130 | struct buffer_head *bh_result, |
1131 | int create, | |
3e12dbbd DC |
1132 | bool direct, |
1133 | bool dax_fault) | |
1da177e4 | 1134 | { |
a206c817 CH |
1135 | struct xfs_inode *ip = XFS_I(inode); |
1136 | struct xfs_mount *mp = ip->i_mount; | |
1137 | xfs_fileoff_t offset_fsb, end_fsb; | |
1138 | int error = 0; | |
1139 | int lockmode = 0; | |
207d0416 | 1140 | struct xfs_bmbt_irec imap; |
a206c817 | 1141 | int nimaps = 1; |
fdc7ed75 NS |
1142 | xfs_off_t offset; |
1143 | ssize_t size; | |
207d0416 | 1144 | int new = 0; |
a206c817 CH |
1145 | |
1146 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1147 | return -EIO; |
1da177e4 | 1148 | |
fdc7ed75 | 1149 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1150 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1151 | size = bh_result->b_size; | |
364f358a LM |
1152 | |
1153 | if (!create && direct && offset >= i_size_read(inode)) | |
1154 | return 0; | |
1155 | ||
507630b2 DC |
1156 | /* |
1157 | * Direct I/O is usually done on preallocated files, so try getting | |
1158 | * a block mapping without an exclusive lock first. For buffered | |
1159 | * writes we already have the exclusive iolock anyway, so avoiding | |
1160 | * a lock roundtrip here by taking the ilock exclusive from the | |
1161 | * beginning is a useful micro optimization. | |
1162 | */ | |
1163 | if (create && !direct) { | |
a206c817 CH |
1164 | lockmode = XFS_ILOCK_EXCL; |
1165 | xfs_ilock(ip, lockmode); | |
1166 | } else { | |
309ecac8 | 1167 | lockmode = xfs_ilock_data_map_shared(ip); |
a206c817 | 1168 | } |
f2bde9b8 | 1169 | |
d2c28191 DC |
1170 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1171 | if (offset + size > mp->m_super->s_maxbytes) | |
1172 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1173 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1174 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1175 | ||
5c8ed202 DC |
1176 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1177 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1178 | if (error) |
a206c817 CH |
1179 | goto out_unlock; |
1180 | ||
1ca19157 | 1181 | /* for DAX, we convert unwritten extents directly */ |
a206c817 CH |
1182 | if (create && |
1183 | (!nimaps || | |
1184 | (imap.br_startblock == HOLESTARTBLOCK || | |
1ca19157 DC |
1185 | imap.br_startblock == DELAYSTARTBLOCK) || |
1186 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { | |
aff3a9ed | 1187 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 | 1188 | /* |
009c6e87 BF |
1189 | * xfs_iomap_write_direct() expects the shared lock. It |
1190 | * is unlocked on return. | |
507630b2 | 1191 | */ |
009c6e87 BF |
1192 | if (lockmode == XFS_ILOCK_EXCL) |
1193 | xfs_ilock_demote(ip, lockmode); | |
1194 | ||
a206c817 CH |
1195 | error = xfs_iomap_write_direct(ip, offset, size, |
1196 | &imap, nimaps); | |
507630b2 | 1197 | if (error) |
2451337d | 1198 | return error; |
d3bc815a | 1199 | new = 1; |
6b698ede | 1200 | |
a206c817 | 1201 | } else { |
507630b2 DC |
1202 | /* |
1203 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1204 | * we can go on without dropping the lock here. If we |
1205 | * are allocating a new delalloc block, make sure that | |
1206 | * we set the new flag so that we mark the buffer new so | |
1207 | * that we know that it is newly allocated if the write | |
1208 | * fails. | |
507630b2 | 1209 | */ |
d3bc815a DC |
1210 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1211 | new = 1; | |
a206c817 | 1212 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1213 | if (error) |
1214 | goto out_unlock; | |
1215 | ||
1216 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1217 | } |
d5cc2e3f DC |
1218 | trace_xfs_get_blocks_alloc(ip, offset, size, |
1219 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1220 | : XFS_IO_DELALLOC, &imap); | |
a206c817 | 1221 | } else if (nimaps) { |
d5cc2e3f DC |
1222 | trace_xfs_get_blocks_found(ip, offset, size, |
1223 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1224 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1225 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1226 | } else { |
1227 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1228 | goto out_unlock; | |
1229 | } | |
1da177e4 | 1230 | |
1ca19157 DC |
1231 | if (IS_DAX(inode) && create) { |
1232 | ASSERT(!ISUNWRITTEN(&imap)); | |
1233 | /* zeroing is not needed at a higher layer */ | |
1234 | new = 0; | |
1235 | } | |
1236 | ||
1fdca9c2 DC |
1237 | /* trim mapping down to size requested */ |
1238 | if (direct || size > (1 << inode->i_blkbits)) | |
1239 | xfs_map_trim_size(inode, iblock, bh_result, | |
1240 | &imap, offset, size); | |
1241 | ||
a719370b DC |
1242 | /* |
1243 | * For unwritten extents do not report a disk address in the buffered | |
1244 | * read case (treat as if we're reading into a hole). | |
1245 | */ | |
207d0416 | 1246 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1247 | imap.br_startblock != DELAYSTARTBLOCK && |
1248 | (create || !ISUNWRITTEN(&imap))) { | |
1249 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1250 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1251 | set_buffer_unwritten(bh_result); |
a719370b | 1252 | /* direct IO needs special help */ |
273dda76 CH |
1253 | if (create && direct) { |
1254 | if (dax_fault) | |
1255 | ASSERT(!ISUNWRITTEN(&imap)); | |
1256 | else | |
1257 | xfs_map_direct(inode, bh_result, &imap, offset); | |
1258 | } | |
1da177e4 LT |
1259 | } |
1260 | ||
c2536668 NS |
1261 | /* |
1262 | * If this is a realtime file, data may be on a different device. | |
1263 | * to that pointed to from the buffer_head b_bdev currently. | |
1264 | */ | |
046f1685 | 1265 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1266 | |
c2536668 | 1267 | /* |
549054af DC |
1268 | * If we previously allocated a block out beyond eof and we are now |
1269 | * coming back to use it then we will need to flag it as new even if it | |
1270 | * has a disk address. | |
1271 | * | |
1272 | * With sub-block writes into unwritten extents we also need to mark | |
1273 | * the buffer as new so that the unwritten parts of the buffer gets | |
1274 | * correctly zeroed. | |
1da177e4 LT |
1275 | */ |
1276 | if (create && | |
1277 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1278 | (offset >= i_size_read(inode)) || |
207d0416 | 1279 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1280 | set_buffer_new(bh_result); |
1da177e4 | 1281 | |
207d0416 | 1282 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1283 | BUG_ON(direct); |
1284 | if (create) { | |
1285 | set_buffer_uptodate(bh_result); | |
1286 | set_buffer_mapped(bh_result); | |
1287 | set_buffer_delay(bh_result); | |
1288 | } | |
1289 | } | |
1290 | ||
1da177e4 | 1291 | return 0; |
a206c817 CH |
1292 | |
1293 | out_unlock: | |
1294 | xfs_iunlock(ip, lockmode); | |
2451337d | 1295 | return error; |
1da177e4 LT |
1296 | } |
1297 | ||
1298 | int | |
c2536668 | 1299 | xfs_get_blocks( |
1da177e4 LT |
1300 | struct inode *inode, |
1301 | sector_t iblock, | |
1302 | struct buffer_head *bh_result, | |
1303 | int create) | |
1304 | { | |
3e12dbbd | 1305 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
1da177e4 LT |
1306 | } |
1307 | ||
6b698ede | 1308 | int |
e4c573bb | 1309 | xfs_get_blocks_direct( |
1da177e4 LT |
1310 | struct inode *inode, |
1311 | sector_t iblock, | |
1da177e4 LT |
1312 | struct buffer_head *bh_result, |
1313 | int create) | |
1314 | { | |
3e12dbbd DC |
1315 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
1316 | } | |
1317 | ||
1318 | int | |
1319 | xfs_get_blocks_dax_fault( | |
1320 | struct inode *inode, | |
1321 | sector_t iblock, | |
1322 | struct buffer_head *bh_result, | |
1323 | int create) | |
1324 | { | |
1325 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); | |
1da177e4 LT |
1326 | } |
1327 | ||
273dda76 CH |
1328 | /* |
1329 | * Complete a direct I/O write request. | |
1330 | * | |
1331 | * xfs_map_direct passes us some flags in the private data to tell us what to | |
1332 | * do. If no flags are set, then the write IO is an overwrite wholly within | |
1333 | * the existing allocated file size and so there is nothing for us to do. | |
1334 | * | |
1335 | * Note that in this case the completion can be called in interrupt context, | |
1336 | * whereas if we have flags set we will always be called in task context | |
1337 | * (i.e. from a workqueue). | |
1338 | */ | |
1339 | STATIC int | |
1340 | xfs_end_io_direct_write( | |
1341 | struct kiocb *iocb, | |
209fb87a | 1342 | loff_t offset, |
273dda76 CH |
1343 | ssize_t size, |
1344 | void *private) | |
f0973863 | 1345 | { |
273dda76 CH |
1346 | struct inode *inode = file_inode(iocb->ki_filp); |
1347 | struct xfs_inode *ip = XFS_I(inode); | |
1348 | struct xfs_mount *mp = ip->i_mount; | |
1349 | uintptr_t flags = (uintptr_t)private; | |
1350 | int error = 0; | |
a06c277a | 1351 | |
273dda76 | 1352 | trace_xfs_end_io_direct_write(ip, offset, size); |
f0973863 | 1353 | |
273dda76 CH |
1354 | if (XFS_FORCED_SHUTDOWN(mp)) |
1355 | return -EIO; | |
d5cc2e3f | 1356 | |
273dda76 CH |
1357 | if (size <= 0) |
1358 | return size; | |
f0973863 | 1359 | |
2813d682 | 1360 | /* |
273dda76 | 1361 | * The flags tell us whether we are doing unwritten extent conversions |
6dfa1b67 DC |
1362 | * or an append transaction that updates the on-disk file size. These |
1363 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1364 | * to update the VFS inode size. |
273dda76 CH |
1365 | */ |
1366 | if (flags == 0) { | |
1367 | ASSERT(offset + size <= i_size_read(inode)); | |
1368 | return 0; | |
1369 | } | |
1370 | ||
1371 | /* | |
6dfa1b67 | 1372 | * We need to update the in-core inode size here so that we don't end up |
a06c277a DC |
1373 | * with the on-disk inode size being outside the in-core inode size. We |
1374 | * have no other method of updating EOF for AIO, so always do it here | |
1375 | * if necessary. | |
b9d59846 DC |
1376 | * |
1377 | * We need to lock the test/set EOF update as we can be racing with | |
1378 | * other IO completions here to update the EOF. Failing to serialise | |
1379 | * here can result in EOF moving backwards and Bad Things Happen when | |
1380 | * that occurs. | |
2813d682 | 1381 | */ |
273dda76 | 1382 | spin_lock(&ip->i_flags_lock); |
2ba66237 CH |
1383 | if (offset + size > i_size_read(inode)) |
1384 | i_size_write(inode, offset + size); | |
273dda76 | 1385 | spin_unlock(&ip->i_flags_lock); |
2813d682 | 1386 | |
273dda76 CH |
1387 | if (flags & XFS_DIO_FLAG_UNWRITTEN) { |
1388 | trace_xfs_end_io_direct_write_unwritten(ip, offset, size); | |
209fb87a | 1389 | |
273dda76 CH |
1390 | error = xfs_iomap_write_unwritten(ip, offset, size); |
1391 | } else if (flags & XFS_DIO_FLAG_APPEND) { | |
1392 | struct xfs_trans *tp; | |
f0973863 | 1393 | |
273dda76 | 1394 | trace_xfs_end_io_direct_write_append(ip, offset, size); |
6b698ede | 1395 | |
253f4911 CH |
1396 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, |
1397 | &tp); | |
1398 | if (!error) | |
1399 | error = xfs_setfilesize(ip, tp, offset, size); | |
6b698ede DC |
1400 | } |
1401 | ||
273dda76 | 1402 | return error; |
6b698ede DC |
1403 | } |
1404 | ||
c19b104a CH |
1405 | STATIC ssize_t |
1406 | xfs_vm_direct_IO( | |
6e1ba0bc | 1407 | struct kiocb *iocb, |
c8b8e32d | 1408 | struct iov_iter *iter) |
6e1ba0bc | 1409 | { |
c19b104a CH |
1410 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1411 | dio_iodone_t *endio = NULL; | |
1412 | int flags = 0; | |
6e1ba0bc DC |
1413 | struct block_device *bdev; |
1414 | ||
c19b104a CH |
1415 | if (iov_iter_rw(iter) == WRITE) { |
1416 | endio = xfs_end_io_direct_write; | |
1417 | flags = DIO_ASYNC_EXTEND; | |
1418 | } | |
1419 | ||
1420 | if (IS_DAX(inode)) { | |
c8b8e32d | 1421 | return dax_do_io(iocb, inode, iter, |
6e1ba0bc | 1422 | xfs_get_blocks_direct, endio, 0); |
c19b104a | 1423 | } |
6e1ba0bc DC |
1424 | |
1425 | bdev = xfs_find_bdev_for_inode(inode); | |
c8b8e32d | 1426 | return __blockdev_direct_IO(iocb, inode, bdev, iter, |
c19b104a | 1427 | xfs_get_blocks_direct, endio, NULL, flags); |
1da177e4 LT |
1428 | } |
1429 | ||
1da177e4 | 1430 | STATIC sector_t |
e4c573bb | 1431 | xfs_vm_bmap( |
1da177e4 LT |
1432 | struct address_space *mapping, |
1433 | sector_t block) | |
1434 | { | |
1435 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1436 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1437 | |
cca28fb8 | 1438 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1439 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1440 | filemap_write_and_wait(mapping); |
126468b1 | 1441 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1442 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1443 | } |
1444 | ||
1445 | STATIC int | |
e4c573bb | 1446 | xfs_vm_readpage( |
1da177e4 LT |
1447 | struct file *unused, |
1448 | struct page *page) | |
1449 | { | |
121e213e | 1450 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1451 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1452 | } |
1453 | ||
1454 | STATIC int | |
e4c573bb | 1455 | xfs_vm_readpages( |
1da177e4 LT |
1456 | struct file *unused, |
1457 | struct address_space *mapping, | |
1458 | struct list_head *pages, | |
1459 | unsigned nr_pages) | |
1460 | { | |
121e213e | 1461 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1462 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1463 | } |
1464 | ||
22e757a4 DC |
1465 | /* |
1466 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1467 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1468 | * dirty, we'll never be able to clean them because we don't write buffers | |
1469 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1470 | * that have been marked dirty. Further, the dirty state can leak into | |
1471 | * the file interior if the file is extended, resulting in all sorts of | |
1472 | * bad things happening as the state does not match the underlying data. | |
1473 | * | |
1474 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1475 | * this only exist because of bufferheads and how the generic code manages them. | |
1476 | */ | |
1477 | STATIC int | |
1478 | xfs_vm_set_page_dirty( | |
1479 | struct page *page) | |
1480 | { | |
1481 | struct address_space *mapping = page->mapping; | |
1482 | struct inode *inode = mapping->host; | |
1483 | loff_t end_offset; | |
1484 | loff_t offset; | |
1485 | int newly_dirty; | |
1486 | ||
1487 | if (unlikely(!mapping)) | |
1488 | return !TestSetPageDirty(page); | |
1489 | ||
1490 | end_offset = i_size_read(inode); | |
1491 | offset = page_offset(page); | |
1492 | ||
1493 | spin_lock(&mapping->private_lock); | |
1494 | if (page_has_buffers(page)) { | |
1495 | struct buffer_head *head = page_buffers(page); | |
1496 | struct buffer_head *bh = head; | |
1497 | ||
1498 | do { | |
1499 | if (offset < end_offset) | |
1500 | set_buffer_dirty(bh); | |
1501 | bh = bh->b_this_page; | |
1502 | offset += 1 << inode->i_blkbits; | |
1503 | } while (bh != head); | |
1504 | } | |
c4843a75 | 1505 | /* |
81f8c3a4 JW |
1506 | * Lock out page->mem_cgroup migration to keep PageDirty |
1507 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1508 | */ |
62cccb8c | 1509 | lock_page_memcg(page); |
22e757a4 DC |
1510 | newly_dirty = !TestSetPageDirty(page); |
1511 | spin_unlock(&mapping->private_lock); | |
1512 | ||
1513 | if (newly_dirty) { | |
1514 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1515 | unsigned long flags; | |
1516 | ||
1517 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1518 | if (page->mapping) { /* Race with truncate? */ | |
1519 | WARN_ON_ONCE(!PageUptodate(page)); | |
62cccb8c | 1520 | account_page_dirtied(page, mapping); |
22e757a4 DC |
1521 | radix_tree_tag_set(&mapping->page_tree, |
1522 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1523 | } | |
1524 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1525 | } |
62cccb8c | 1526 | unlock_page_memcg(page); |
c4843a75 GT |
1527 | if (newly_dirty) |
1528 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1529 | return newly_dirty; |
1530 | } | |
1531 | ||
f5e54d6e | 1532 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1533 | .readpage = xfs_vm_readpage, |
1534 | .readpages = xfs_vm_readpages, | |
1535 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1536 | .writepages = xfs_vm_writepages, |
22e757a4 | 1537 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1538 | .releasepage = xfs_vm_releasepage, |
1539 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb NS |
1540 | .bmap = xfs_vm_bmap, |
1541 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1542 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1543 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1544 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1545 | }; |