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. | |
28b783e4 DC |
90 | * |
91 | * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last | |
92 | * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or | |
93 | * the page at all, as we may be racing with memory reclaim and it can free both | |
94 | * the bufferhead chain and the page as it will see the page as clean and | |
95 | * unused. | |
37992c18 DC |
96 | */ |
97 | static void | |
98 | xfs_finish_page_writeback( | |
99 | struct inode *inode, | |
100 | struct bio_vec *bvec, | |
101 | int error) | |
102 | { | |
37992c18 | 103 | unsigned int end = bvec->bv_offset + bvec->bv_len - 1; |
28b783e4 | 104 | struct buffer_head *head, *bh, *next; |
37992c18 | 105 | unsigned int off = 0; |
28b783e4 | 106 | unsigned int bsize; |
37992c18 DC |
107 | |
108 | ASSERT(bvec->bv_offset < PAGE_SIZE); | |
690a7871 | 109 | ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0); |
37992c18 | 110 | ASSERT(end < PAGE_SIZE); |
690a7871 | 111 | ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0); |
37992c18 DC |
112 | |
113 | bh = head = page_buffers(bvec->bv_page); | |
114 | ||
28b783e4 | 115 | bsize = bh->b_size; |
37992c18 | 116 | do { |
28b783e4 | 117 | next = bh->b_this_page; |
37992c18 DC |
118 | if (off < bvec->bv_offset) |
119 | goto next_bh; | |
120 | if (off > end) | |
121 | break; | |
122 | bh->b_end_io(bh, !error); | |
123 | next_bh: | |
28b783e4 DC |
124 | off += bsize; |
125 | } while ((bh = next) != head); | |
37992c18 DC |
126 | } |
127 | ||
128 | /* | |
129 | * We're now finished for good with this ioend structure. Update the page | |
130 | * state, release holds on bios, and finally free up memory. Do not use the | |
131 | * ioend after this. | |
f6d6d4fc | 132 | */ |
0829c360 CH |
133 | STATIC void |
134 | xfs_destroy_ioend( | |
0e51a8e1 CH |
135 | struct xfs_ioend *ioend, |
136 | int error) | |
0829c360 | 137 | { |
37992c18 | 138 | struct inode *inode = ioend->io_inode; |
0e51a8e1 | 139 | struct bio *last = ioend->io_bio; |
37992c18 | 140 | struct bio *bio, *next; |
f6d6d4fc | 141 | |
0e51a8e1 | 142 | for (bio = &ioend->io_inline_bio; bio; bio = next) { |
37992c18 DC |
143 | struct bio_vec *bvec; |
144 | int i; | |
145 | ||
0e51a8e1 CH |
146 | /* |
147 | * For the last bio, bi_private points to the ioend, so we | |
148 | * need to explicitly end the iteration here. | |
149 | */ | |
150 | if (bio == last) | |
151 | next = NULL; | |
152 | else | |
153 | next = bio->bi_private; | |
583fa586 | 154 | |
37992c18 DC |
155 | /* walk each page on bio, ending page IO on them */ |
156 | bio_for_each_segment_all(bvec, bio, i) | |
157 | xfs_finish_page_writeback(inode, bvec, error); | |
158 | ||
159 | bio_put(bio); | |
f6d6d4fc | 160 | } |
0829c360 CH |
161 | } |
162 | ||
fc0063c4 CH |
163 | /* |
164 | * Fast and loose check if this write could update the on-disk inode size. | |
165 | */ | |
166 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
167 | { | |
168 | return ioend->io_offset + ioend->io_size > | |
169 | XFS_I(ioend->io_inode)->i_d.di_size; | |
170 | } | |
171 | ||
281627df CH |
172 | STATIC int |
173 | xfs_setfilesize_trans_alloc( | |
174 | struct xfs_ioend *ioend) | |
175 | { | |
176 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
177 | struct xfs_trans *tp; | |
178 | int error; | |
179 | ||
253f4911 CH |
180 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); |
181 | if (error) | |
281627df | 182 | return error; |
281627df CH |
183 | |
184 | ioend->io_append_trans = tp; | |
185 | ||
d9457dc0 | 186 | /* |
437a255a | 187 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
188 | * we released it. |
189 | */ | |
bee9182d | 190 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
191 | /* |
192 | * We hand off the transaction to the completion thread now, so | |
193 | * clear the flag here. | |
194 | */ | |
195 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
196 | return 0; | |
197 | } | |
198 | ||
ba87ea69 | 199 | /* |
2813d682 | 200 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 201 | */ |
281627df | 202 | STATIC int |
ba87ea69 | 203 | xfs_setfilesize( |
2ba66237 CH |
204 | struct xfs_inode *ip, |
205 | struct xfs_trans *tp, | |
206 | xfs_off_t offset, | |
207 | size_t size) | |
ba87ea69 | 208 | { |
ba87ea69 | 209 | xfs_fsize_t isize; |
ba87ea69 | 210 | |
aa6bf01d | 211 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 212 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
213 | if (!isize) { |
214 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 215 | xfs_trans_cancel(tp); |
281627df | 216 | return 0; |
ba87ea69 LM |
217 | } |
218 | ||
2ba66237 | 219 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
220 | |
221 | ip->i_d.di_size = isize; | |
222 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
223 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
224 | ||
70393313 | 225 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
226 | } |
227 | ||
2ba66237 CH |
228 | STATIC int |
229 | xfs_setfilesize_ioend( | |
0e51a8e1 CH |
230 | struct xfs_ioend *ioend, |
231 | int error) | |
2ba66237 CH |
232 | { |
233 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
234 | struct xfs_trans *tp = ioend->io_append_trans; | |
235 | ||
236 | /* | |
237 | * The transaction may have been allocated in the I/O submission thread, | |
238 | * thus we need to mark ourselves as being in a transaction manually. | |
239 | * Similarly for freeze protection. | |
240 | */ | |
241 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 242 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 243 | |
5cb13dcd | 244 | /* we abort the update if there was an IO error */ |
0e51a8e1 | 245 | if (error) { |
5cb13dcd | 246 | xfs_trans_cancel(tp); |
0e51a8e1 | 247 | return error; |
5cb13dcd Z |
248 | } |
249 | ||
2ba66237 CH |
250 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
251 | } | |
252 | ||
0829c360 | 253 | /* |
5ec4fabb | 254 | * IO write completion. |
f6d6d4fc CH |
255 | */ |
256 | STATIC void | |
5ec4fabb | 257 | xfs_end_io( |
77d7a0c2 | 258 | struct work_struct *work) |
0829c360 | 259 | { |
0e51a8e1 CH |
260 | struct xfs_ioend *ioend = |
261 | container_of(work, struct xfs_ioend, io_work); | |
262 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
263 | int error = ioend->io_bio->bi_error; | |
ba87ea69 | 264 | |
af055e37 BF |
265 | /* |
266 | * Set an error if the mount has shut down and proceed with end I/O | |
267 | * processing so it can perform whatever cleanups are necessary. | |
268 | */ | |
269 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) | |
0e51a8e1 | 270 | error = -EIO; |
04f658ee | 271 | |
5ec4fabb CH |
272 | /* |
273 | * For unwritten extents we need to issue transactions to convert a | |
274 | * range to normal written extens after the data I/O has finished. | |
5cb13dcd Z |
275 | * Detecting and handling completion IO errors is done individually |
276 | * for each case as different cleanup operations need to be performed | |
277 | * on error. | |
5ec4fabb | 278 | */ |
0d882a36 | 279 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
0e51a8e1 | 280 | if (error) |
5cb13dcd | 281 | goto done; |
437a255a DC |
282 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
283 | ioend->io_size); | |
281627df | 284 | } else if (ioend->io_append_trans) { |
0e51a8e1 | 285 | error = xfs_setfilesize_ioend(ioend, error); |
84803fb7 | 286 | } else { |
281627df | 287 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 288 | } |
ba87ea69 | 289 | |
04f658ee | 290 | done: |
0e51a8e1 | 291 | xfs_destroy_ioend(ioend, error); |
c626d174 DC |
292 | } |
293 | ||
0e51a8e1 CH |
294 | STATIC void |
295 | xfs_end_bio( | |
296 | struct bio *bio) | |
0829c360 | 297 | { |
0e51a8e1 CH |
298 | struct xfs_ioend *ioend = bio->bi_private; |
299 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
0829c360 | 300 | |
0e51a8e1 CH |
301 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
302 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); | |
303 | else if (ioend->io_append_trans) | |
304 | queue_work(mp->m_data_workqueue, &ioend->io_work); | |
305 | else | |
306 | xfs_destroy_ioend(ioend, bio->bi_error); | |
0829c360 CH |
307 | } |
308 | ||
1da177e4 LT |
309 | STATIC int |
310 | xfs_map_blocks( | |
311 | struct inode *inode, | |
312 | loff_t offset, | |
207d0416 | 313 | struct xfs_bmbt_irec *imap, |
988ef927 | 314 | int type) |
1da177e4 | 315 | { |
a206c817 CH |
316 | struct xfs_inode *ip = XFS_I(inode); |
317 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 318 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
319 | xfs_fileoff_t offset_fsb, end_fsb; |
320 | int error = 0; | |
a206c817 CH |
321 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
322 | int nimaps = 1; | |
323 | ||
324 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 325 | return -EIO; |
a206c817 | 326 | |
0d882a36 | 327 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 328 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 329 | |
988ef927 | 330 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
331 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
332 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 333 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 334 | |
d2c28191 DC |
335 | if (offset + count > mp->m_super->s_maxbytes) |
336 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
337 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
338 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
339 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
340 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 341 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 342 | |
8ff2957d | 343 | if (error) |
2451337d | 344 | return error; |
a206c817 | 345 | |
0d882a36 | 346 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 347 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 348 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
349 | if (!error) |
350 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 351 | return error; |
a206c817 CH |
352 | } |
353 | ||
8ff2957d | 354 | #ifdef DEBUG |
0d882a36 | 355 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
356 | ASSERT(nimaps); |
357 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
358 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
359 | } | |
360 | #endif | |
361 | if (nimaps) | |
362 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
363 | return 0; | |
1da177e4 LT |
364 | } |
365 | ||
fbcc0256 | 366 | STATIC bool |
558e6891 | 367 | xfs_imap_valid( |
8699bb0a | 368 | struct inode *inode, |
207d0416 | 369 | struct xfs_bmbt_irec *imap, |
558e6891 | 370 | xfs_off_t offset) |
1da177e4 | 371 | { |
558e6891 | 372 | offset >>= inode->i_blkbits; |
8699bb0a | 373 | |
558e6891 CH |
374 | return offset >= imap->br_startoff && |
375 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
376 | } |
377 | ||
f6d6d4fc CH |
378 | STATIC void |
379 | xfs_start_buffer_writeback( | |
380 | struct buffer_head *bh) | |
381 | { | |
382 | ASSERT(buffer_mapped(bh)); | |
383 | ASSERT(buffer_locked(bh)); | |
384 | ASSERT(!buffer_delay(bh)); | |
385 | ASSERT(!buffer_unwritten(bh)); | |
386 | ||
387 | mark_buffer_async_write(bh); | |
388 | set_buffer_uptodate(bh); | |
389 | clear_buffer_dirty(bh); | |
390 | } | |
391 | ||
392 | STATIC void | |
393 | xfs_start_page_writeback( | |
394 | struct page *page, | |
e10de372 | 395 | int clear_dirty) |
f6d6d4fc CH |
396 | { |
397 | ASSERT(PageLocked(page)); | |
398 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
399 | |
400 | /* | |
401 | * if the page was not fully cleaned, we need to ensure that the higher | |
402 | * layers come back to it correctly. That means we need to keep the page | |
403 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
404 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
405 | * write this page in this writeback sweep will be made. | |
406 | */ | |
407 | if (clear_dirty) { | |
92132021 | 408 | clear_page_dirty_for_io(page); |
0d085a52 DC |
409 | set_page_writeback(page); |
410 | } else | |
411 | set_page_writeback_keepwrite(page); | |
412 | ||
f6d6d4fc | 413 | unlock_page(page); |
f6d6d4fc CH |
414 | } |
415 | ||
c7c1a7d8 | 416 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
417 | { |
418 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
419 | } | |
420 | ||
421 | /* | |
bb18782a DC |
422 | * Submit the bio for an ioend. We are passed an ioend with a bio attached to |
423 | * it, and we submit that bio. The ioend may be used for multiple bio | |
424 | * submissions, so we only want to allocate an append transaction for the ioend | |
425 | * once. In the case of multiple bio submission, each bio will take an IO | |
426 | * reference to the ioend to ensure that the ioend completion is only done once | |
427 | * all bios have been submitted and the ioend is really done. | |
7bf7f352 DC |
428 | * |
429 | * If @fail is non-zero, it means that we have a situation where some part of | |
430 | * the submission process has failed after we have marked paged for writeback | |
bb18782a DC |
431 | * and unlocked them. In this situation, we need to fail the bio and ioend |
432 | * rather than submit it to IO. This typically only happens on a filesystem | |
433 | * shutdown. | |
f6d6d4fc | 434 | */ |
e10de372 | 435 | STATIC int |
f6d6d4fc | 436 | xfs_submit_ioend( |
06342cf8 | 437 | struct writeback_control *wbc, |
0e51a8e1 | 438 | struct xfs_ioend *ioend, |
e10de372 | 439 | int status) |
f6d6d4fc | 440 | { |
e10de372 DC |
441 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
442 | if (!status && | |
0e51a8e1 | 443 | ioend->io_type != XFS_IO_UNWRITTEN && |
bb18782a DC |
444 | xfs_ioend_is_append(ioend) && |
445 | !ioend->io_append_trans) | |
e10de372 | 446 | status = xfs_setfilesize_trans_alloc(ioend); |
bb18782a | 447 | |
0e51a8e1 CH |
448 | ioend->io_bio->bi_private = ioend; |
449 | ioend->io_bio->bi_end_io = xfs_end_bio; | |
50bfcd0c MC |
450 | bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE, |
451 | (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0); | |
e10de372 DC |
452 | /* |
453 | * If we are failing the IO now, just mark the ioend with an | |
454 | * error and finish it. This will run IO completion immediately | |
455 | * as there is only one reference to the ioend at this point in | |
456 | * time. | |
457 | */ | |
458 | if (status) { | |
0e51a8e1 CH |
459 | ioend->io_bio->bi_error = status; |
460 | bio_endio(ioend->io_bio); | |
e10de372 DC |
461 | return status; |
462 | } | |
d88992f6 | 463 | |
4e49ea4a | 464 | submit_bio(ioend->io_bio); |
e10de372 | 465 | return 0; |
f6d6d4fc | 466 | } |
f6d6d4fc | 467 | |
0e51a8e1 CH |
468 | static void |
469 | xfs_init_bio_from_bh( | |
470 | struct bio *bio, | |
471 | struct buffer_head *bh) | |
472 | { | |
473 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
474 | bio->bi_bdev = bh->b_bdev; | |
475 | } | |
7bf7f352 | 476 | |
0e51a8e1 CH |
477 | static struct xfs_ioend * |
478 | xfs_alloc_ioend( | |
479 | struct inode *inode, | |
480 | unsigned int type, | |
481 | xfs_off_t offset, | |
482 | struct buffer_head *bh) | |
483 | { | |
484 | struct xfs_ioend *ioend; | |
485 | struct bio *bio; | |
f6d6d4fc | 486 | |
0e51a8e1 CH |
487 | bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset); |
488 | xfs_init_bio_from_bh(bio, bh); | |
489 | ||
490 | ioend = container_of(bio, struct xfs_ioend, io_inline_bio); | |
491 | INIT_LIST_HEAD(&ioend->io_list); | |
492 | ioend->io_type = type; | |
493 | ioend->io_inode = inode; | |
494 | ioend->io_size = 0; | |
495 | ioend->io_offset = offset; | |
496 | INIT_WORK(&ioend->io_work, xfs_end_io); | |
497 | ioend->io_append_trans = NULL; | |
498 | ioend->io_bio = bio; | |
499 | return ioend; | |
500 | } | |
501 | ||
502 | /* | |
503 | * Allocate a new bio, and chain the old bio to the new one. | |
504 | * | |
505 | * Note that we have to do perform the chaining in this unintuitive order | |
506 | * so that the bi_private linkage is set up in the right direction for the | |
507 | * traversal in xfs_destroy_ioend(). | |
508 | */ | |
509 | static void | |
510 | xfs_chain_bio( | |
511 | struct xfs_ioend *ioend, | |
512 | struct writeback_control *wbc, | |
513 | struct buffer_head *bh) | |
514 | { | |
515 | struct bio *new; | |
516 | ||
517 | new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES); | |
518 | xfs_init_bio_from_bh(new, bh); | |
519 | ||
520 | bio_chain(ioend->io_bio, new); | |
521 | bio_get(ioend->io_bio); /* for xfs_destroy_ioend */ | |
50bfcd0c MC |
522 | bio_set_op_attrs(ioend->io_bio, REQ_OP_WRITE, |
523 | (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0); | |
4e49ea4a | 524 | submit_bio(ioend->io_bio); |
0e51a8e1 | 525 | ioend->io_bio = new; |
f6d6d4fc CH |
526 | } |
527 | ||
528 | /* | |
529 | * Test to see if we've been building up a completion structure for | |
530 | * earlier buffers -- if so, we try to append to this ioend if we | |
531 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
532 | * Return the ioend we finished off so that the caller can submit it |
533 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
534 | */ |
535 | STATIC void | |
536 | xfs_add_to_ioend( | |
537 | struct inode *inode, | |
538 | struct buffer_head *bh, | |
7336cea8 | 539 | xfs_off_t offset, |
e10de372 | 540 | struct xfs_writepage_ctx *wpc, |
bb18782a | 541 | struct writeback_control *wbc, |
e10de372 | 542 | struct list_head *iolist) |
f6d6d4fc | 543 | { |
fbcc0256 | 544 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
0df61da8 DW |
545 | bh->b_blocknr != wpc->last_block + 1 || |
546 | offset != wpc->ioend->io_offset + wpc->ioend->io_size) { | |
e10de372 DC |
547 | if (wpc->ioend) |
548 | list_add(&wpc->ioend->io_list, iolist); | |
0e51a8e1 | 549 | wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh); |
f6d6d4fc CH |
550 | } |
551 | ||
0e51a8e1 CH |
552 | /* |
553 | * If the buffer doesn't fit into the bio we need to allocate a new | |
554 | * one. This shouldn't happen more than once for a given buffer. | |
555 | */ | |
556 | while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size) | |
557 | xfs_chain_bio(wpc->ioend, wbc, bh); | |
bb18782a | 558 | |
fbcc0256 DC |
559 | wpc->ioend->io_size += bh->b_size; |
560 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 561 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
562 | } |
563 | ||
87cbc49c NS |
564 | STATIC void |
565 | xfs_map_buffer( | |
046f1685 | 566 | struct inode *inode, |
87cbc49c | 567 | struct buffer_head *bh, |
207d0416 | 568 | struct xfs_bmbt_irec *imap, |
046f1685 | 569 | xfs_off_t offset) |
87cbc49c NS |
570 | { |
571 | sector_t bn; | |
8699bb0a | 572 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
573 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
574 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 575 | |
207d0416 CH |
576 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
577 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 578 | |
e513182d | 579 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 580 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 581 | |
046f1685 | 582 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
583 | |
584 | bh->b_blocknr = bn; | |
585 | set_buffer_mapped(bh); | |
586 | } | |
587 | ||
1da177e4 LT |
588 | STATIC void |
589 | xfs_map_at_offset( | |
046f1685 | 590 | struct inode *inode, |
1da177e4 | 591 | struct buffer_head *bh, |
207d0416 | 592 | struct xfs_bmbt_irec *imap, |
046f1685 | 593 | xfs_off_t offset) |
1da177e4 | 594 | { |
207d0416 CH |
595 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
596 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 597 | |
207d0416 | 598 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
599 | set_buffer_mapped(bh); |
600 | clear_buffer_delay(bh); | |
f6d6d4fc | 601 | clear_buffer_unwritten(bh); |
1da177e4 LT |
602 | } |
603 | ||
1da177e4 | 604 | /* |
a49935f2 DC |
605 | * Test if a given page contains at least one buffer of a given @type. |
606 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
607 | * try to find one of the type passed in. If it is not set, then the caller only | |
608 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 609 | */ |
a49935f2 | 610 | STATIC bool |
6ffc4db5 | 611 | xfs_check_page_type( |
10ce4444 | 612 | struct page *page, |
a49935f2 DC |
613 | unsigned int type, |
614 | bool check_all_buffers) | |
1da177e4 | 615 | { |
a49935f2 DC |
616 | struct buffer_head *bh; |
617 | struct buffer_head *head; | |
1da177e4 | 618 | |
a49935f2 DC |
619 | if (PageWriteback(page)) |
620 | return false; | |
621 | if (!page->mapping) | |
622 | return false; | |
623 | if (!page_has_buffers(page)) | |
624 | return false; | |
1da177e4 | 625 | |
a49935f2 DC |
626 | bh = head = page_buffers(page); |
627 | do { | |
628 | if (buffer_unwritten(bh)) { | |
629 | if (type == XFS_IO_UNWRITTEN) | |
630 | return true; | |
631 | } else if (buffer_delay(bh)) { | |
805eeb8e | 632 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
633 | return true; |
634 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 635 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
636 | return true; |
637 | } | |
1da177e4 | 638 | |
a49935f2 DC |
639 | /* If we are only checking the first buffer, we are done now. */ |
640 | if (!check_all_buffers) | |
641 | break; | |
642 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 643 | |
a49935f2 | 644 | return false; |
1da177e4 LT |
645 | } |
646 | ||
3ed3a434 DC |
647 | STATIC void |
648 | xfs_vm_invalidatepage( | |
649 | struct page *page, | |
d47992f8 LC |
650 | unsigned int offset, |
651 | unsigned int length) | |
3ed3a434 | 652 | { |
34097dfe LC |
653 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
654 | length); | |
655 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
656 | } |
657 | ||
658 | /* | |
659 | * If the page has delalloc buffers on it, we need to punch them out before we | |
660 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
661 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
662 | * is done on that same region - the delalloc extent is returned when none is | |
663 | * supposed to be there. | |
664 | * | |
665 | * We prevent this by truncating away the delalloc regions on the page before | |
666 | * invalidating it. Because they are delalloc, we can do this without needing a | |
667 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
668 | * truncation without a transaction as there is no space left for block | |
669 | * reservation (typically why we see a ENOSPC in writeback). | |
670 | * | |
671 | * This is not a performance critical path, so for now just do the punching a | |
672 | * buffer head at a time. | |
673 | */ | |
674 | STATIC void | |
675 | xfs_aops_discard_page( | |
676 | struct page *page) | |
677 | { | |
678 | struct inode *inode = page->mapping->host; | |
679 | struct xfs_inode *ip = XFS_I(inode); | |
680 | struct buffer_head *bh, *head; | |
681 | loff_t offset = page_offset(page); | |
3ed3a434 | 682 | |
a49935f2 | 683 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
684 | goto out_invalidate; |
685 | ||
e8c3753c DC |
686 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
687 | goto out_invalidate; | |
688 | ||
4f10700a | 689 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
690 | "page discard on page %p, inode 0x%llx, offset %llu.", |
691 | page, ip->i_ino, offset); | |
692 | ||
693 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
694 | bh = head = page_buffers(page); | |
695 | do { | |
3ed3a434 | 696 | int error; |
c726de44 | 697 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
698 | |
699 | if (!buffer_delay(bh)) | |
700 | goto next_buffer; | |
701 | ||
c726de44 DC |
702 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
703 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
704 | if (error) { |
705 | /* something screwed, just bail */ | |
e8c3753c | 706 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 707 | xfs_alert(ip->i_mount, |
3ed3a434 | 708 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 709 | } |
3ed3a434 DC |
710 | break; |
711 | } | |
712 | next_buffer: | |
c726de44 | 713 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
714 | |
715 | } while ((bh = bh->b_this_page) != head); | |
716 | ||
717 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
718 | out_invalidate: | |
09cbfeaf | 719 | xfs_vm_invalidatepage(page, 0, PAGE_SIZE); |
3ed3a434 DC |
720 | return; |
721 | } | |
722 | ||
e10de372 DC |
723 | /* |
724 | * We implement an immediate ioend submission policy here to avoid needing to | |
725 | * chain multiple ioends and hence nest mempool allocations which can violate | |
726 | * forward progress guarantees we need to provide. The current ioend we are | |
727 | * adding buffers to is cached on the writepage context, and if the new buffer | |
728 | * does not append to the cached ioend it will create a new ioend and cache that | |
729 | * instead. | |
730 | * | |
731 | * If a new ioend is created and cached, the old ioend is returned and queued | |
732 | * locally for submission once the entire page is processed or an error has been | |
733 | * detected. While ioends are submitted immediately after they are completed, | |
734 | * batching optimisations are provided by higher level block plugging. | |
735 | * | |
736 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
737 | * writepage context that the caller will need to submit. | |
738 | */ | |
bfce7d2e DC |
739 | static int |
740 | xfs_writepage_map( | |
741 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 742 | struct writeback_control *wbc, |
bfce7d2e DC |
743 | struct inode *inode, |
744 | struct page *page, | |
745 | loff_t offset, | |
746 | __uint64_t end_offset) | |
747 | { | |
e10de372 DC |
748 | LIST_HEAD(submit_list); |
749 | struct xfs_ioend *ioend, *next; | |
bfce7d2e DC |
750 | struct buffer_head *bh, *head; |
751 | ssize_t len = 1 << inode->i_blkbits; | |
752 | int error = 0; | |
bfce7d2e | 753 | int count = 0; |
e10de372 | 754 | int uptodate = 1; |
bfce7d2e DC |
755 | |
756 | bh = head = page_buffers(page); | |
757 | offset = page_offset(page); | |
bfce7d2e DC |
758 | do { |
759 | if (offset >= end_offset) | |
760 | break; | |
761 | if (!buffer_uptodate(bh)) | |
762 | uptodate = 0; | |
763 | ||
764 | /* | |
765 | * set_page_dirty dirties all buffers in a page, independent | |
766 | * of their state. The dirty state however is entirely | |
767 | * meaningless for holes (!mapped && uptodate), so skip | |
768 | * buffers covering holes here. | |
769 | */ | |
770 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
771 | wpc->imap_valid = false; | |
772 | continue; | |
773 | } | |
774 | ||
775 | if (buffer_unwritten(bh)) { | |
776 | if (wpc->io_type != XFS_IO_UNWRITTEN) { | |
777 | wpc->io_type = XFS_IO_UNWRITTEN; | |
778 | wpc->imap_valid = false; | |
779 | } | |
780 | } else if (buffer_delay(bh)) { | |
781 | if (wpc->io_type != XFS_IO_DELALLOC) { | |
782 | wpc->io_type = XFS_IO_DELALLOC; | |
783 | wpc->imap_valid = false; | |
784 | } | |
785 | } else if (buffer_uptodate(bh)) { | |
786 | if (wpc->io_type != XFS_IO_OVERWRITE) { | |
787 | wpc->io_type = XFS_IO_OVERWRITE; | |
788 | wpc->imap_valid = false; | |
789 | } | |
790 | } else { | |
791 | if (PageUptodate(page)) | |
792 | ASSERT(buffer_mapped(bh)); | |
793 | /* | |
794 | * This buffer is not uptodate and will not be | |
795 | * written to disk. Ensure that we will put any | |
796 | * subsequent writeable buffers into a new | |
797 | * ioend. | |
798 | */ | |
799 | wpc->imap_valid = false; | |
800 | continue; | |
801 | } | |
802 | ||
803 | if (wpc->imap_valid) | |
804 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
805 | offset); | |
806 | if (!wpc->imap_valid) { | |
807 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
808 | wpc->io_type); | |
809 | if (error) | |
e10de372 | 810 | goto out; |
bfce7d2e DC |
811 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
812 | offset); | |
813 | } | |
814 | if (wpc->imap_valid) { | |
815 | lock_buffer(bh); | |
816 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
817 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
bb18782a | 818 | xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list); |
bfce7d2e DC |
819 | count++; |
820 | } | |
821 | ||
bfce7d2e DC |
822 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
823 | ||
824 | if (uptodate && bh == head) | |
825 | SetPageUptodate(page); | |
826 | ||
e10de372 | 827 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 828 | |
e10de372 | 829 | out: |
bfce7d2e | 830 | /* |
e10de372 DC |
831 | * On error, we have to fail the ioend here because we have locked |
832 | * buffers in the ioend. If we don't do this, we'll deadlock | |
833 | * invalidating the page as that tries to lock the buffers on the page. | |
834 | * Also, because we may have set pages under writeback, we have to make | |
835 | * sure we run IO completion to mark the error state of the IO | |
836 | * appropriately, so we can't cancel the ioend directly here. That means | |
837 | * we have to mark this page as under writeback if we included any | |
838 | * buffers from it in the ioend chain so that completion treats it | |
839 | * correctly. | |
bfce7d2e | 840 | * |
e10de372 DC |
841 | * If we didn't include the page in the ioend, the on error we can |
842 | * simply discard and unlock it as there are no other users of the page | |
843 | * or it's buffers right now. The caller will still need to trigger | |
844 | * submission of outstanding ioends on the writepage context so they are | |
845 | * treated correctly on error. | |
bfce7d2e | 846 | */ |
e10de372 DC |
847 | if (count) { |
848 | xfs_start_page_writeback(page, !error); | |
849 | ||
850 | /* | |
851 | * Preserve the original error if there was one, otherwise catch | |
852 | * submission errors here and propagate into subsequent ioend | |
853 | * submissions. | |
854 | */ | |
855 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
856 | int error2; | |
857 | ||
858 | list_del_init(&ioend->io_list); | |
859 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
860 | if (error2 && !error) | |
861 | error = error2; | |
862 | } | |
863 | } else if (error) { | |
bfce7d2e DC |
864 | xfs_aops_discard_page(page); |
865 | ClearPageUptodate(page); | |
866 | unlock_page(page); | |
e10de372 DC |
867 | } else { |
868 | /* | |
869 | * We can end up here with no error and nothing to write if we | |
870 | * race with a partial page truncate on a sub-page block sized | |
871 | * filesystem. In that case we need to mark the page clean. | |
872 | */ | |
873 | xfs_start_page_writeback(page, 1); | |
874 | end_page_writeback(page); | |
bfce7d2e | 875 | } |
e10de372 | 876 | |
bfce7d2e DC |
877 | mapping_set_error(page->mapping, error); |
878 | return error; | |
879 | } | |
880 | ||
1da177e4 | 881 | /* |
89f3b363 CH |
882 | * Write out a dirty page. |
883 | * | |
884 | * For delalloc space on the page we need to allocate space and flush it. | |
885 | * For unwritten space on the page we need to start the conversion to | |
886 | * regular allocated space. | |
89f3b363 | 887 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 888 | */ |
1da177e4 | 889 | STATIC int |
fbcc0256 | 890 | xfs_do_writepage( |
89f3b363 | 891 | struct page *page, |
fbcc0256 DC |
892 | struct writeback_control *wbc, |
893 | void *data) | |
1da177e4 | 894 | { |
fbcc0256 | 895 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 896 | struct inode *inode = page->mapping->host; |
1da177e4 | 897 | loff_t offset; |
1da177e4 | 898 | __uint64_t end_offset; |
ad68972a | 899 | pgoff_t end_index; |
89f3b363 | 900 | |
34097dfe | 901 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 902 | |
20cb52eb CH |
903 | ASSERT(page_has_buffers(page)); |
904 | ||
89f3b363 CH |
905 | /* |
906 | * Refuse to write the page out if we are called from reclaim context. | |
907 | * | |
d4f7a5cb CH |
908 | * This avoids stack overflows when called from deeply used stacks in |
909 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
910 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 911 | * |
94054fa3 MG |
912 | * This should never happen except in the case of a VM regression so |
913 | * warn about it. | |
89f3b363 | 914 | */ |
94054fa3 MG |
915 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
916 | PF_MEMALLOC)) | |
b5420f23 | 917 | goto redirty; |
1da177e4 | 918 | |
89f3b363 | 919 | /* |
680a647b CH |
920 | * Given that we do not allow direct reclaim to call us, we should |
921 | * never be called while in a filesystem transaction. | |
89f3b363 | 922 | */ |
448011e2 | 923 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 924 | goto redirty; |
89f3b363 | 925 | |
8695d27e | 926 | /* |
ad68972a DC |
927 | * Is this page beyond the end of the file? |
928 | * | |
8695d27e JL |
929 | * The page index is less than the end_index, adjust the end_offset |
930 | * to the highest offset that this page should represent. | |
931 | * ----------------------------------------------------- | |
932 | * | file mapping | <EOF> | | |
933 | * ----------------------------------------------------- | |
934 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
935 | * ^--------------------------------^----------|-------- | |
936 | * | desired writeback range | see else | | |
937 | * ---------------------------------^------------------| | |
938 | */ | |
ad68972a | 939 | offset = i_size_read(inode); |
09cbfeaf | 940 | end_index = offset >> PAGE_SHIFT; |
8695d27e | 941 | if (page->index < end_index) |
09cbfeaf | 942 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT; |
8695d27e JL |
943 | else { |
944 | /* | |
945 | * Check whether the page to write out is beyond or straddles | |
946 | * i_size or not. | |
947 | * ------------------------------------------------------- | |
948 | * | file mapping | <EOF> | | |
949 | * ------------------------------------------------------- | |
950 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
951 | * ^--------------------------------^-----------|--------- | |
952 | * | | Straddles | | |
953 | * ---------------------------------^-----------|--------| | |
954 | */ | |
09cbfeaf | 955 | unsigned offset_into_page = offset & (PAGE_SIZE - 1); |
6b7a03f0 CH |
956 | |
957 | /* | |
ff9a28f6 JK |
958 | * Skip the page if it is fully outside i_size, e.g. due to a |
959 | * truncate operation that is in progress. We must redirty the | |
960 | * page so that reclaim stops reclaiming it. Otherwise | |
961 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
962 | * |
963 | * Note that the end_index is unsigned long, it would overflow | |
964 | * if the given offset is greater than 16TB on 32-bit system | |
965 | * and if we do check the page is fully outside i_size or not | |
966 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
967 | * will be evaluated to 0. Hence this page will be redirtied | |
968 | * and be written out repeatedly which would result in an | |
969 | * infinite loop, the user program that perform this operation | |
970 | * will hang. Instead, we can verify this situation by checking | |
971 | * if the page to write is totally beyond the i_size or if it's | |
972 | * offset is just equal to the EOF. | |
6b7a03f0 | 973 | */ |
8695d27e JL |
974 | if (page->index > end_index || |
975 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 976 | goto redirty; |
6b7a03f0 CH |
977 | |
978 | /* | |
979 | * The page straddles i_size. It must be zeroed out on each | |
980 | * and every writepage invocation because it may be mmapped. | |
981 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 982 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
983 | * memory is zeroed when mapped, and writes to that region are |
984 | * not written out to the file." | |
985 | */ | |
09cbfeaf | 986 | zero_user_segment(page, offset_into_page, PAGE_SIZE); |
8695d27e JL |
987 | |
988 | /* Adjust the end_offset to the end of file */ | |
989 | end_offset = offset; | |
1da177e4 LT |
990 | } |
991 | ||
e10de372 | 992 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
f51623b2 | 993 | |
b5420f23 | 994 | redirty: |
f51623b2 NS |
995 | redirty_page_for_writepage(wbc, page); |
996 | unlock_page(page); | |
997 | return 0; | |
f51623b2 NS |
998 | } |
999 | ||
fbcc0256 DC |
1000 | STATIC int |
1001 | xfs_vm_writepage( | |
1002 | struct page *page, | |
1003 | struct writeback_control *wbc) | |
1004 | { | |
1005 | struct xfs_writepage_ctx wpc = { | |
1006 | .io_type = XFS_IO_INVALID, | |
1007 | }; | |
1008 | int ret; | |
1009 | ||
1010 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
1011 | if (wpc.ioend) |
1012 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1013 | return ret; | |
fbcc0256 DC |
1014 | } |
1015 | ||
7d4fb40a NS |
1016 | STATIC int |
1017 | xfs_vm_writepages( | |
1018 | struct address_space *mapping, | |
1019 | struct writeback_control *wbc) | |
1020 | { | |
fbcc0256 DC |
1021 | struct xfs_writepage_ctx wpc = { |
1022 | .io_type = XFS_IO_INVALID, | |
1023 | }; | |
1024 | int ret; | |
1025 | ||
b3aea4ed | 1026 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7f6d5b52 RZ |
1027 | if (dax_mapping(mapping)) |
1028 | return dax_writeback_mapping_range(mapping, | |
1029 | xfs_find_bdev_for_inode(mapping->host), wbc); | |
1030 | ||
fbcc0256 | 1031 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1032 | if (wpc.ioend) |
1033 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1034 | return ret; | |
7d4fb40a NS |
1035 | } |
1036 | ||
f51623b2 NS |
1037 | /* |
1038 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1039 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1040 | * have buffer heads in this call. |
1041 | * | |
89f3b363 | 1042 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1043 | */ |
1044 | STATIC int | |
238f4c54 | 1045 | xfs_vm_releasepage( |
f51623b2 NS |
1046 | struct page *page, |
1047 | gfp_t gfp_mask) | |
1048 | { | |
20cb52eb | 1049 | int delalloc, unwritten; |
f51623b2 | 1050 | |
34097dfe | 1051 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1052 | |
99579cce BF |
1053 | /* |
1054 | * mm accommodates an old ext3 case where clean pages might not have had | |
1055 | * the dirty bit cleared. Thus, it can send actual dirty pages to | |
1056 | * ->releasepage() via shrink_active_list(). Conversely, | |
1057 | * block_invalidatepage() can send pages that are still marked dirty | |
1058 | * but otherwise have invalidated buffers. | |
1059 | * | |
1060 | * We've historically freed buffers on the latter. Instead, quietly | |
1061 | * filter out all dirty pages to avoid spurious buffer state warnings. | |
1062 | * This can likely be removed once shrink_active_list() is fixed. | |
1063 | */ | |
1064 | if (PageDirty(page)) | |
1065 | return 0; | |
1066 | ||
20cb52eb | 1067 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1068 | |
448011e2 | 1069 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1070 | return 0; |
448011e2 | 1071 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1072 | return 0; |
1073 | ||
f51623b2 NS |
1074 | return try_to_free_buffers(page); |
1075 | } | |
1076 | ||
a719370b | 1077 | /* |
273dda76 CH |
1078 | * When we map a DIO buffer, we may need to pass flags to |
1079 | * xfs_end_io_direct_write to tell it what kind of write IO we are doing. | |
3e12dbbd DC |
1080 | * |
1081 | * Note that for DIO, an IO to the highest supported file block offset (i.e. | |
1082 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 | |
1083 | * bit variable. Hence if we see this overflow, we have to assume that the IO is | |
1084 | * extending the file size. We won't know for sure until IO completion is run | |
1085 | * and the actual max write offset is communicated to the IO completion | |
1086 | * routine. | |
a719370b DC |
1087 | */ |
1088 | static void | |
1089 | xfs_map_direct( | |
1090 | struct inode *inode, | |
1091 | struct buffer_head *bh_result, | |
1092 | struct xfs_bmbt_irec *imap, | |
273dda76 | 1093 | xfs_off_t offset) |
a719370b | 1094 | { |
273dda76 | 1095 | uintptr_t *flags = (uintptr_t *)&bh_result->b_private; |
d5cc2e3f | 1096 | xfs_off_t size = bh_result->b_size; |
d5cc2e3f | 1097 | |
273dda76 CH |
1098 | trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size, |
1099 | ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap); | |
d5cc2e3f | 1100 | |
273dda76 CH |
1101 | if (ISUNWRITTEN(imap)) { |
1102 | *flags |= XFS_DIO_FLAG_UNWRITTEN; | |
1103 | set_buffer_defer_completion(bh_result); | |
1104 | } else if (offset + size > i_size_read(inode) || offset + size < 0) { | |
1105 | *flags |= XFS_DIO_FLAG_APPEND; | |
a06c277a | 1106 | set_buffer_defer_completion(bh_result); |
a719370b DC |
1107 | } |
1108 | } | |
1109 | ||
1fdca9c2 DC |
1110 | /* |
1111 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1112 | * is, so that we can avoid repeated get_blocks calls. | |
1113 | * | |
1114 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1115 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1116 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1117 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1118 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1119 | * to and including EOF, and a second mapping for beyond EOF. | |
1120 | */ | |
1121 | static void | |
1122 | xfs_map_trim_size( | |
1123 | struct inode *inode, | |
1124 | sector_t iblock, | |
1125 | struct buffer_head *bh_result, | |
1126 | struct xfs_bmbt_irec *imap, | |
1127 | xfs_off_t offset, | |
1128 | ssize_t size) | |
1129 | { | |
1130 | xfs_off_t mapping_size; | |
1131 | ||
1132 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1133 | mapping_size <<= inode->i_blkbits; | |
1134 | ||
1135 | ASSERT(mapping_size > 0); | |
1136 | if (mapping_size > size) | |
1137 | mapping_size = size; | |
1138 | if (offset < i_size_read(inode) && | |
1139 | offset + mapping_size >= i_size_read(inode)) { | |
1140 | /* limit mapping to block that spans EOF */ | |
1141 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1142 | 1 << inode->i_blkbits); | |
1143 | } | |
1144 | if (mapping_size > LONG_MAX) | |
1145 | mapping_size = LONG_MAX; | |
1146 | ||
1147 | bh_result->b_size = mapping_size; | |
1148 | } | |
1149 | ||
1da177e4 | 1150 | STATIC int |
c2536668 | 1151 | __xfs_get_blocks( |
1da177e4 LT |
1152 | struct inode *inode, |
1153 | sector_t iblock, | |
1da177e4 LT |
1154 | struct buffer_head *bh_result, |
1155 | int create, | |
3e12dbbd DC |
1156 | bool direct, |
1157 | bool dax_fault) | |
1da177e4 | 1158 | { |
a206c817 CH |
1159 | struct xfs_inode *ip = XFS_I(inode); |
1160 | struct xfs_mount *mp = ip->i_mount; | |
1161 | xfs_fileoff_t offset_fsb, end_fsb; | |
1162 | int error = 0; | |
1163 | int lockmode = 0; | |
207d0416 | 1164 | struct xfs_bmbt_irec imap; |
a206c817 | 1165 | int nimaps = 1; |
fdc7ed75 NS |
1166 | xfs_off_t offset; |
1167 | ssize_t size; | |
207d0416 | 1168 | int new = 0; |
a206c817 | 1169 | |
6e8a27a8 CH |
1170 | BUG_ON(create && !direct); |
1171 | ||
a206c817 | 1172 | if (XFS_FORCED_SHUTDOWN(mp)) |
b474c7ae | 1173 | return -EIO; |
1da177e4 | 1174 | |
fdc7ed75 | 1175 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1176 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1177 | size = bh_result->b_size; | |
364f358a | 1178 | |
6e8a27a8 | 1179 | if (!create && offset >= i_size_read(inode)) |
364f358a LM |
1180 | return 0; |
1181 | ||
507630b2 DC |
1182 | /* |
1183 | * Direct I/O is usually done on preallocated files, so try getting | |
6e8a27a8 | 1184 | * a block mapping without an exclusive lock first. |
507630b2 | 1185 | */ |
6e8a27a8 | 1186 | lockmode = xfs_ilock_data_map_shared(ip); |
f2bde9b8 | 1187 | |
d2c28191 DC |
1188 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1189 | if (offset + size > mp->m_super->s_maxbytes) | |
1190 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1191 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1192 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1193 | ||
5c8ed202 DC |
1194 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1195 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1196 | if (error) |
a206c817 CH |
1197 | goto out_unlock; |
1198 | ||
1ca19157 | 1199 | /* for DAX, we convert unwritten extents directly */ |
a206c817 CH |
1200 | if (create && |
1201 | (!nimaps || | |
1202 | (imap.br_startblock == HOLESTARTBLOCK || | |
1ca19157 DC |
1203 | imap.br_startblock == DELAYSTARTBLOCK) || |
1204 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { | |
6e8a27a8 CH |
1205 | /* |
1206 | * xfs_iomap_write_direct() expects the shared lock. It | |
1207 | * is unlocked on return. | |
1208 | */ | |
1209 | if (lockmode == XFS_ILOCK_EXCL) | |
1210 | xfs_ilock_demote(ip, lockmode); | |
6b698ede | 1211 | |
6e8a27a8 CH |
1212 | error = xfs_iomap_write_direct(ip, offset, size, |
1213 | &imap, nimaps); | |
1214 | if (error) | |
1215 | return error; | |
1216 | new = 1; | |
507630b2 | 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 | 1237 | /* trim mapping down to size requested */ |
6e8a27a8 | 1238 | xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size); |
1fdca9c2 | 1239 | |
a719370b DC |
1240 | /* |
1241 | * For unwritten extents do not report a disk address in the buffered | |
1242 | * read case (treat as if we're reading into a hole). | |
1243 | */ | |
207d0416 | 1244 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1245 | imap.br_startblock != DELAYSTARTBLOCK && |
1246 | (create || !ISUNWRITTEN(&imap))) { | |
1247 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1248 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1249 | set_buffer_unwritten(bh_result); |
a719370b | 1250 | /* direct IO needs special help */ |
6e8a27a8 | 1251 | if (create) { |
273dda76 CH |
1252 | if (dax_fault) |
1253 | ASSERT(!ISUNWRITTEN(&imap)); | |
1254 | else | |
1255 | xfs_map_direct(inode, bh_result, &imap, offset); | |
1256 | } | |
1da177e4 LT |
1257 | } |
1258 | ||
c2536668 NS |
1259 | /* |
1260 | * If this is a realtime file, data may be on a different device. | |
1261 | * to that pointed to from the buffer_head b_bdev currently. | |
1262 | */ | |
046f1685 | 1263 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1264 | |
c2536668 | 1265 | /* |
549054af DC |
1266 | * If we previously allocated a block out beyond eof and we are now |
1267 | * coming back to use it then we will need to flag it as new even if it | |
1268 | * has a disk address. | |
1269 | * | |
1270 | * With sub-block writes into unwritten extents we also need to mark | |
1271 | * the buffer as new so that the unwritten parts of the buffer gets | |
1272 | * correctly zeroed. | |
1da177e4 LT |
1273 | */ |
1274 | if (create && | |
1275 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1276 | (offset >= i_size_read(inode)) || |
207d0416 | 1277 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1278 | set_buffer_new(bh_result); |
1da177e4 | 1279 | |
6e8a27a8 | 1280 | BUG_ON(direct && imap.br_startblock == DELAYSTARTBLOCK); |
1da177e4 | 1281 | |
1da177e4 | 1282 | return 0; |
a206c817 CH |
1283 | |
1284 | out_unlock: | |
1285 | xfs_iunlock(ip, lockmode); | |
2451337d | 1286 | return error; |
1da177e4 LT |
1287 | } |
1288 | ||
1289 | int | |
c2536668 | 1290 | xfs_get_blocks( |
1da177e4 LT |
1291 | struct inode *inode, |
1292 | sector_t iblock, | |
1293 | struct buffer_head *bh_result, | |
1294 | int create) | |
1295 | { | |
3e12dbbd | 1296 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
1da177e4 LT |
1297 | } |
1298 | ||
6b698ede | 1299 | int |
e4c573bb | 1300 | xfs_get_blocks_direct( |
1da177e4 LT |
1301 | struct inode *inode, |
1302 | sector_t iblock, | |
1da177e4 LT |
1303 | struct buffer_head *bh_result, |
1304 | int create) | |
1305 | { | |
3e12dbbd DC |
1306 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
1307 | } | |
1308 | ||
1309 | int | |
1310 | xfs_get_blocks_dax_fault( | |
1311 | struct inode *inode, | |
1312 | sector_t iblock, | |
1313 | struct buffer_head *bh_result, | |
1314 | int create) | |
1315 | { | |
1316 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); | |
1da177e4 LT |
1317 | } |
1318 | ||
273dda76 CH |
1319 | /* |
1320 | * Complete a direct I/O write request. | |
1321 | * | |
1322 | * xfs_map_direct passes us some flags in the private data to tell us what to | |
1323 | * do. If no flags are set, then the write IO is an overwrite wholly within | |
1324 | * the existing allocated file size and so there is nothing for us to do. | |
1325 | * | |
1326 | * Note that in this case the completion can be called in interrupt context, | |
1327 | * whereas if we have flags set we will always be called in task context | |
1328 | * (i.e. from a workqueue). | |
1329 | */ | |
fa8d972d | 1330 | int |
273dda76 CH |
1331 | xfs_end_io_direct_write( |
1332 | struct kiocb *iocb, | |
209fb87a | 1333 | loff_t offset, |
273dda76 CH |
1334 | ssize_t size, |
1335 | void *private) | |
f0973863 | 1336 | { |
273dda76 CH |
1337 | struct inode *inode = file_inode(iocb->ki_filp); |
1338 | struct xfs_inode *ip = XFS_I(inode); | |
1339 | struct xfs_mount *mp = ip->i_mount; | |
1340 | uintptr_t flags = (uintptr_t)private; | |
1341 | int error = 0; | |
a06c277a | 1342 | |
273dda76 | 1343 | trace_xfs_end_io_direct_write(ip, offset, size); |
f0973863 | 1344 | |
273dda76 CH |
1345 | if (XFS_FORCED_SHUTDOWN(mp)) |
1346 | return -EIO; | |
d5cc2e3f | 1347 | |
273dda76 CH |
1348 | if (size <= 0) |
1349 | return size; | |
f0973863 | 1350 | |
2813d682 | 1351 | /* |
273dda76 | 1352 | * The flags tell us whether we are doing unwritten extent conversions |
6dfa1b67 DC |
1353 | * or an append transaction that updates the on-disk file size. These |
1354 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1355 | * to update the VFS inode size. |
273dda76 CH |
1356 | */ |
1357 | if (flags == 0) { | |
1358 | ASSERT(offset + size <= i_size_read(inode)); | |
1359 | return 0; | |
1360 | } | |
1361 | ||
1362 | /* | |
6dfa1b67 | 1363 | * We need to update the in-core inode size here so that we don't end up |
a06c277a DC |
1364 | * with the on-disk inode size being outside the in-core inode size. We |
1365 | * have no other method of updating EOF for AIO, so always do it here | |
1366 | * if necessary. | |
b9d59846 DC |
1367 | * |
1368 | * We need to lock the test/set EOF update as we can be racing with | |
1369 | * other IO completions here to update the EOF. Failing to serialise | |
1370 | * here can result in EOF moving backwards and Bad Things Happen when | |
1371 | * that occurs. | |
2813d682 | 1372 | */ |
273dda76 | 1373 | spin_lock(&ip->i_flags_lock); |
2ba66237 CH |
1374 | if (offset + size > i_size_read(inode)) |
1375 | i_size_write(inode, offset + size); | |
273dda76 | 1376 | spin_unlock(&ip->i_flags_lock); |
2813d682 | 1377 | |
273dda76 CH |
1378 | if (flags & XFS_DIO_FLAG_UNWRITTEN) { |
1379 | trace_xfs_end_io_direct_write_unwritten(ip, offset, size); | |
209fb87a | 1380 | |
273dda76 CH |
1381 | error = xfs_iomap_write_unwritten(ip, offset, size); |
1382 | } else if (flags & XFS_DIO_FLAG_APPEND) { | |
1383 | struct xfs_trans *tp; | |
f0973863 | 1384 | |
273dda76 | 1385 | trace_xfs_end_io_direct_write_append(ip, offset, size); |
6b698ede | 1386 | |
253f4911 CH |
1387 | error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, |
1388 | &tp); | |
1389 | if (!error) | |
1390 | error = xfs_setfilesize(ip, tp, offset, size); | |
6b698ede DC |
1391 | } |
1392 | ||
273dda76 | 1393 | return error; |
6b698ede DC |
1394 | } |
1395 | ||
c19b104a CH |
1396 | STATIC ssize_t |
1397 | xfs_vm_direct_IO( | |
6e1ba0bc | 1398 | struct kiocb *iocb, |
c8b8e32d | 1399 | struct iov_iter *iter) |
6e1ba0bc | 1400 | { |
58e59854 | 1401 | /* |
fa8d972d | 1402 | * We just need the method present so that open/fcntl allow direct I/O. |
58e59854 | 1403 | */ |
fa8d972d | 1404 | return -EINVAL; |
f51623b2 | 1405 | } |
1da177e4 LT |
1406 | |
1407 | STATIC sector_t | |
e4c573bb | 1408 | xfs_vm_bmap( |
1da177e4 LT |
1409 | struct address_space *mapping, |
1410 | sector_t block) | |
1411 | { | |
1412 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1413 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1414 | |
cca28fb8 | 1415 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1416 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1417 | filemap_write_and_wait(mapping); |
126468b1 | 1418 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1419 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1420 | } |
1421 | ||
1422 | STATIC int | |
e4c573bb | 1423 | xfs_vm_readpage( |
1da177e4 LT |
1424 | struct file *unused, |
1425 | struct page *page) | |
1426 | { | |
121e213e | 1427 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1428 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1429 | } |
1430 | ||
1431 | STATIC int | |
e4c573bb | 1432 | xfs_vm_readpages( |
1da177e4 LT |
1433 | struct file *unused, |
1434 | struct address_space *mapping, | |
1435 | struct list_head *pages, | |
1436 | unsigned nr_pages) | |
1437 | { | |
121e213e | 1438 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1439 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1440 | } |
1441 | ||
22e757a4 DC |
1442 | /* |
1443 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1444 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1445 | * dirty, we'll never be able to clean them because we don't write buffers | |
1446 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1447 | * that have been marked dirty. Further, the dirty state can leak into | |
1448 | * the file interior if the file is extended, resulting in all sorts of | |
1449 | * bad things happening as the state does not match the underlying data. | |
1450 | * | |
1451 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1452 | * this only exist because of bufferheads and how the generic code manages them. | |
1453 | */ | |
1454 | STATIC int | |
1455 | xfs_vm_set_page_dirty( | |
1456 | struct page *page) | |
1457 | { | |
1458 | struct address_space *mapping = page->mapping; | |
1459 | struct inode *inode = mapping->host; | |
1460 | loff_t end_offset; | |
1461 | loff_t offset; | |
1462 | int newly_dirty; | |
1463 | ||
1464 | if (unlikely(!mapping)) | |
1465 | return !TestSetPageDirty(page); | |
1466 | ||
1467 | end_offset = i_size_read(inode); | |
1468 | offset = page_offset(page); | |
1469 | ||
1470 | spin_lock(&mapping->private_lock); | |
1471 | if (page_has_buffers(page)) { | |
1472 | struct buffer_head *head = page_buffers(page); | |
1473 | struct buffer_head *bh = head; | |
1474 | ||
1475 | do { | |
1476 | if (offset < end_offset) | |
1477 | set_buffer_dirty(bh); | |
1478 | bh = bh->b_this_page; | |
1479 | offset += 1 << inode->i_blkbits; | |
1480 | } while (bh != head); | |
1481 | } | |
c4843a75 | 1482 | /* |
81f8c3a4 JW |
1483 | * Lock out page->mem_cgroup migration to keep PageDirty |
1484 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 1485 | */ |
62cccb8c | 1486 | lock_page_memcg(page); |
22e757a4 DC |
1487 | newly_dirty = !TestSetPageDirty(page); |
1488 | spin_unlock(&mapping->private_lock); | |
1489 | ||
1490 | if (newly_dirty) { | |
1491 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1492 | unsigned long flags; | |
1493 | ||
1494 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1495 | if (page->mapping) { /* Race with truncate? */ | |
1496 | WARN_ON_ONCE(!PageUptodate(page)); | |
62cccb8c | 1497 | account_page_dirtied(page, mapping); |
22e757a4 DC |
1498 | radix_tree_tag_set(&mapping->page_tree, |
1499 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1500 | } | |
1501 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1502 | } |
62cccb8c | 1503 | unlock_page_memcg(page); |
c4843a75 GT |
1504 | if (newly_dirty) |
1505 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1506 | return newly_dirty; |
1507 | } | |
1508 | ||
f5e54d6e | 1509 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1510 | .readpage = xfs_vm_readpage, |
1511 | .readpages = xfs_vm_readpages, | |
1512 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1513 | .writepages = xfs_vm_writepages, |
22e757a4 | 1514 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1515 | .releasepage = xfs_vm_releasepage, |
1516 | .invalidatepage = xfs_vm_invalidatepage, | |
e4c573bb NS |
1517 | .bmap = xfs_vm_bmap, |
1518 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1519 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1520 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1521 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1522 | }; |