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 | ||
fbcc0256 DC |
39 | /* |
40 | * structure owned by writepages passed to individual writepage calls | |
41 | */ | |
42 | struct xfs_writepage_ctx { | |
43 | struct xfs_bmbt_irec imap; | |
44 | bool imap_valid; | |
45 | unsigned int io_type; | |
fbcc0256 DC |
46 | struct xfs_ioend *ioend; |
47 | sector_t last_block; | |
48 | }; | |
49 | ||
0b1b213f | 50 | void |
f51623b2 NS |
51 | xfs_count_page_state( |
52 | struct page *page, | |
53 | int *delalloc, | |
f51623b2 NS |
54 | int *unwritten) |
55 | { | |
56 | struct buffer_head *bh, *head; | |
57 | ||
20cb52eb | 58 | *delalloc = *unwritten = 0; |
f51623b2 NS |
59 | |
60 | bh = head = page_buffers(page); | |
61 | do { | |
20cb52eb | 62 | if (buffer_unwritten(bh)) |
f51623b2 NS |
63 | (*unwritten) = 1; |
64 | else if (buffer_delay(bh)) | |
65 | (*delalloc) = 1; | |
66 | } while ((bh = bh->b_this_page) != head); | |
67 | } | |
68 | ||
6214ed44 CH |
69 | STATIC struct block_device * |
70 | xfs_find_bdev_for_inode( | |
046f1685 | 71 | struct inode *inode) |
6214ed44 | 72 | { |
046f1685 | 73 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
74 | struct xfs_mount *mp = ip->i_mount; |
75 | ||
71ddabb9 | 76 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
77 | return mp->m_rtdev_targp->bt_bdev; |
78 | else | |
79 | return mp->m_ddev_targp->bt_bdev; | |
80 | } | |
81 | ||
f6d6d4fc CH |
82 | /* |
83 | * We're now finished for good with this ioend structure. | |
84 | * Update the page state via the associated buffer_heads, | |
85 | * release holds on the inode and bio, and finally free | |
86 | * up memory. Do not use the ioend after this. | |
87 | */ | |
0829c360 CH |
88 | STATIC void |
89 | xfs_destroy_ioend( | |
90 | xfs_ioend_t *ioend) | |
91 | { | |
f6d6d4fc CH |
92 | struct buffer_head *bh, *next; |
93 | ||
94 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
95 | next = bh->b_private; | |
7d04a335 | 96 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 97 | } |
583fa586 | 98 | |
0829c360 CH |
99 | mempool_free(ioend, xfs_ioend_pool); |
100 | } | |
101 | ||
fc0063c4 CH |
102 | /* |
103 | * Fast and loose check if this write could update the on-disk inode size. | |
104 | */ | |
105 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
106 | { | |
107 | return ioend->io_offset + ioend->io_size > | |
108 | XFS_I(ioend->io_inode)->i_d.di_size; | |
109 | } | |
110 | ||
281627df CH |
111 | STATIC int |
112 | xfs_setfilesize_trans_alloc( | |
113 | struct xfs_ioend *ioend) | |
114 | { | |
115 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
116 | struct xfs_trans *tp; | |
117 | int error; | |
118 | ||
119 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
120 | ||
3d3c8b52 | 121 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
281627df | 122 | if (error) { |
4906e215 | 123 | xfs_trans_cancel(tp); |
281627df CH |
124 | return error; |
125 | } | |
126 | ||
127 | ioend->io_append_trans = tp; | |
128 | ||
d9457dc0 | 129 | /* |
437a255a | 130 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
131 | * we released it. |
132 | */ | |
bee9182d | 133 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
134 | /* |
135 | * We hand off the transaction to the completion thread now, so | |
136 | * clear the flag here. | |
137 | */ | |
138 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
139 | return 0; | |
140 | } | |
141 | ||
ba87ea69 | 142 | /* |
2813d682 | 143 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 144 | */ |
281627df | 145 | STATIC int |
ba87ea69 | 146 | xfs_setfilesize( |
2ba66237 CH |
147 | struct xfs_inode *ip, |
148 | struct xfs_trans *tp, | |
149 | xfs_off_t offset, | |
150 | size_t size) | |
ba87ea69 | 151 | { |
ba87ea69 | 152 | xfs_fsize_t isize; |
ba87ea69 | 153 | |
aa6bf01d | 154 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 155 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
156 | if (!isize) { |
157 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 158 | xfs_trans_cancel(tp); |
281627df | 159 | return 0; |
ba87ea69 LM |
160 | } |
161 | ||
2ba66237 | 162 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
163 | |
164 | ip->i_d.di_size = isize; | |
165 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
166 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
167 | ||
70393313 | 168 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
169 | } |
170 | ||
2ba66237 CH |
171 | STATIC int |
172 | xfs_setfilesize_ioend( | |
173 | struct xfs_ioend *ioend) | |
174 | { | |
175 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
176 | struct xfs_trans *tp = ioend->io_append_trans; | |
177 | ||
178 | /* | |
179 | * The transaction may have been allocated in the I/O submission thread, | |
180 | * thus we need to mark ourselves as being in a transaction manually. | |
181 | * Similarly for freeze protection. | |
182 | */ | |
183 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 184 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 185 | |
5cb13dcd Z |
186 | /* we abort the update if there was an IO error */ |
187 | if (ioend->io_error) { | |
188 | xfs_trans_cancel(tp); | |
189 | return ioend->io_error; | |
190 | } | |
191 | ||
2ba66237 CH |
192 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
193 | } | |
194 | ||
77d7a0c2 | 195 | /* |
209fb87a | 196 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
197 | * |
198 | * If there is no work to do we might as well call it a day and free the | |
199 | * ioend right now. | |
77d7a0c2 DC |
200 | */ |
201 | STATIC void | |
202 | xfs_finish_ioend( | |
209fb87a | 203 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
204 | { |
205 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
206 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
207 | ||
0d882a36 | 208 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
aa6bf01d | 209 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
2ba66237 | 210 | else if (ioend->io_append_trans) |
aa6bf01d | 211 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
212 | else |
213 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 214 | } |
ba87ea69 LM |
215 | } |
216 | ||
0829c360 | 217 | /* |
5ec4fabb | 218 | * IO write completion. |
f6d6d4fc CH |
219 | */ |
220 | STATIC void | |
5ec4fabb | 221 | xfs_end_io( |
77d7a0c2 | 222 | struct work_struct *work) |
0829c360 | 223 | { |
77d7a0c2 DC |
224 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
225 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 226 | int error = 0; |
ba87ea69 | 227 | |
04f658ee | 228 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 229 | ioend->io_error = -EIO; |
04f658ee CH |
230 | goto done; |
231 | } | |
04f658ee | 232 | |
5ec4fabb CH |
233 | /* |
234 | * For unwritten extents we need to issue transactions to convert a | |
235 | * range to normal written extens after the data I/O has finished. | |
5cb13dcd Z |
236 | * Detecting and handling completion IO errors is done individually |
237 | * for each case as different cleanup operations need to be performed | |
238 | * on error. | |
5ec4fabb | 239 | */ |
0d882a36 | 240 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
5cb13dcd Z |
241 | if (ioend->io_error) |
242 | goto done; | |
437a255a DC |
243 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
244 | ioend->io_size); | |
281627df | 245 | } else if (ioend->io_append_trans) { |
2ba66237 | 246 | error = xfs_setfilesize_ioend(ioend); |
84803fb7 | 247 | } else { |
281627df | 248 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 249 | } |
ba87ea69 | 250 | |
04f658ee | 251 | done: |
437a255a | 252 | if (error) |
2451337d | 253 | ioend->io_error = error; |
aa6bf01d | 254 | xfs_destroy_ioend(ioend); |
c626d174 DC |
255 | } |
256 | ||
0829c360 CH |
257 | /* |
258 | * Allocate and initialise an IO completion structure. | |
259 | * We need to track unwritten extent write completion here initially. | |
260 | * We'll need to extend this for updating the ondisk inode size later | |
261 | * (vs. incore size). | |
262 | */ | |
263 | STATIC xfs_ioend_t * | |
264 | xfs_alloc_ioend( | |
f6d6d4fc CH |
265 | struct inode *inode, |
266 | unsigned int type) | |
0829c360 CH |
267 | { |
268 | xfs_ioend_t *ioend; | |
269 | ||
270 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
271 | ||
272 | /* | |
273 | * Set the count to 1 initially, which will prevent an I/O | |
274 | * completion callback from happening before we have started | |
275 | * all the I/O from calling the completion routine too early. | |
276 | */ | |
277 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 278 | ioend->io_error = 0; |
e10de372 | 279 | INIT_LIST_HEAD(&ioend->io_list); |
f6d6d4fc | 280 | ioend->io_type = type; |
b677c210 | 281 | ioend->io_inode = inode; |
c1a073bd | 282 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 283 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
284 | ioend->io_offset = 0; |
285 | ioend->io_size = 0; | |
281627df | 286 | ioend->io_append_trans = NULL; |
0829c360 | 287 | |
5ec4fabb | 288 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
289 | return ioend; |
290 | } | |
291 | ||
1da177e4 LT |
292 | STATIC int |
293 | xfs_map_blocks( | |
294 | struct inode *inode, | |
295 | loff_t offset, | |
207d0416 | 296 | struct xfs_bmbt_irec *imap, |
988ef927 | 297 | int type) |
1da177e4 | 298 | { |
a206c817 CH |
299 | struct xfs_inode *ip = XFS_I(inode); |
300 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 301 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
302 | xfs_fileoff_t offset_fsb, end_fsb; |
303 | int error = 0; | |
a206c817 CH |
304 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
305 | int nimaps = 1; | |
306 | ||
307 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 308 | return -EIO; |
a206c817 | 309 | |
0d882a36 | 310 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 311 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d | 312 | |
988ef927 | 313 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
8ff2957d CH |
314 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
315 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 316 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 317 | |
d2c28191 DC |
318 | if (offset + count > mp->m_super->s_maxbytes) |
319 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
320 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
321 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
322 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
323 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 324 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 325 | |
8ff2957d | 326 | if (error) |
2451337d | 327 | return error; |
a206c817 | 328 | |
0d882a36 | 329 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 330 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 331 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
332 | if (!error) |
333 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 334 | return error; |
a206c817 CH |
335 | } |
336 | ||
8ff2957d | 337 | #ifdef DEBUG |
0d882a36 | 338 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
339 | ASSERT(nimaps); |
340 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
341 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
342 | } | |
343 | #endif | |
344 | if (nimaps) | |
345 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
346 | return 0; | |
1da177e4 LT |
347 | } |
348 | ||
fbcc0256 | 349 | STATIC bool |
558e6891 | 350 | xfs_imap_valid( |
8699bb0a | 351 | struct inode *inode, |
207d0416 | 352 | struct xfs_bmbt_irec *imap, |
558e6891 | 353 | xfs_off_t offset) |
1da177e4 | 354 | { |
558e6891 | 355 | offset >>= inode->i_blkbits; |
8699bb0a | 356 | |
558e6891 CH |
357 | return offset >= imap->br_startoff && |
358 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
359 | } |
360 | ||
f6d6d4fc CH |
361 | /* |
362 | * BIO completion handler for buffered IO. | |
363 | */ | |
782e3b3b | 364 | STATIC void |
f6d6d4fc | 365 | xfs_end_bio( |
4246a0b6 | 366 | struct bio *bio) |
f6d6d4fc CH |
367 | { |
368 | xfs_ioend_t *ioend = bio->bi_private; | |
369 | ||
77a78806 LT |
370 | if (!ioend->io_error) |
371 | ioend->io_error = bio->bi_error; | |
f6d6d4fc CH |
372 | |
373 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
374 | bio->bi_private = NULL; |
375 | bio->bi_end_io = NULL; | |
f6d6d4fc | 376 | bio_put(bio); |
7d04a335 | 377 | |
209fb87a | 378 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
379 | } |
380 | ||
381 | STATIC void | |
382 | xfs_submit_ioend_bio( | |
06342cf8 CH |
383 | struct writeback_control *wbc, |
384 | xfs_ioend_t *ioend, | |
385 | struct bio *bio) | |
f6d6d4fc CH |
386 | { |
387 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
388 | bio->bi_private = ioend; |
389 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 390 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
391 | } |
392 | ||
393 | STATIC struct bio * | |
394 | xfs_alloc_ioend_bio( | |
395 | struct buffer_head *bh) | |
396 | { | |
b54ffb73 | 397 | struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); |
f6d6d4fc CH |
398 | |
399 | ASSERT(bio->bi_private == NULL); | |
4f024f37 | 400 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
f6d6d4fc | 401 | bio->bi_bdev = bh->b_bdev; |
f6d6d4fc CH |
402 | return bio; |
403 | } | |
404 | ||
405 | STATIC void | |
406 | xfs_start_buffer_writeback( | |
407 | struct buffer_head *bh) | |
408 | { | |
409 | ASSERT(buffer_mapped(bh)); | |
410 | ASSERT(buffer_locked(bh)); | |
411 | ASSERT(!buffer_delay(bh)); | |
412 | ASSERT(!buffer_unwritten(bh)); | |
413 | ||
414 | mark_buffer_async_write(bh); | |
415 | set_buffer_uptodate(bh); | |
416 | clear_buffer_dirty(bh); | |
417 | } | |
418 | ||
419 | STATIC void | |
420 | xfs_start_page_writeback( | |
421 | struct page *page, | |
e10de372 | 422 | int clear_dirty) |
f6d6d4fc CH |
423 | { |
424 | ASSERT(PageLocked(page)); | |
425 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
426 | |
427 | /* | |
428 | * if the page was not fully cleaned, we need to ensure that the higher | |
429 | * layers come back to it correctly. That means we need to keep the page | |
430 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
431 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
432 | * write this page in this writeback sweep will be made. | |
433 | */ | |
434 | if (clear_dirty) { | |
92132021 | 435 | clear_page_dirty_for_io(page); |
0d085a52 DC |
436 | set_page_writeback(page); |
437 | } else | |
438 | set_page_writeback_keepwrite(page); | |
439 | ||
f6d6d4fc | 440 | unlock_page(page); |
f6d6d4fc CH |
441 | } |
442 | ||
c7c1a7d8 | 443 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
444 | { |
445 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
446 | } | |
447 | ||
448 | /* | |
e10de372 DC |
449 | * Submit all of the bios for an ioend. We are only passed a single ioend at a |
450 | * time; the caller is responsible for chaining prior to submission. | |
7bf7f352 DC |
451 | * |
452 | * If @fail is non-zero, it means that we have a situation where some part of | |
453 | * the submission process has failed after we have marked paged for writeback | |
454 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
455 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
f6d6d4fc | 456 | */ |
e10de372 | 457 | STATIC int |
f6d6d4fc | 458 | xfs_submit_ioend( |
06342cf8 | 459 | struct writeback_control *wbc, |
7bf7f352 | 460 | xfs_ioend_t *ioend, |
e10de372 | 461 | int status) |
f6d6d4fc | 462 | { |
f6d6d4fc CH |
463 | struct buffer_head *bh; |
464 | struct bio *bio; | |
465 | sector_t lastblock = 0; | |
466 | ||
e10de372 DC |
467 | /* Reserve log space if we might write beyond the on-disk inode size. */ |
468 | if (!status && | |
469 | ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) | |
470 | status = xfs_setfilesize_trans_alloc(ioend); | |
471 | /* | |
472 | * If we are failing the IO now, just mark the ioend with an | |
473 | * error and finish it. This will run IO completion immediately | |
474 | * as there is only one reference to the ioend at this point in | |
475 | * time. | |
476 | */ | |
477 | if (status) { | |
478 | ioend->io_error = status; | |
479 | xfs_finish_ioend(ioend); | |
480 | return status; | |
481 | } | |
d88992f6 | 482 | |
e10de372 DC |
483 | bio = NULL; |
484 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc | 485 | |
e10de372 DC |
486 | if (!bio) { |
487 | retry: | |
488 | bio = xfs_alloc_ioend_bio(bh); | |
489 | } else if (bh->b_blocknr != lastblock + 1) { | |
490 | xfs_submit_ioend_bio(wbc, ioend, bio); | |
491 | goto retry; | |
7bf7f352 DC |
492 | } |
493 | ||
e10de372 | 494 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
06342cf8 | 495 | xfs_submit_ioend_bio(wbc, ioend, bio); |
e10de372 DC |
496 | goto retry; |
497 | } | |
498 | ||
499 | lastblock = bh->b_blocknr; | |
500 | } | |
501 | if (bio) | |
502 | xfs_submit_ioend_bio(wbc, ioend, bio); | |
503 | xfs_finish_ioend(ioend); | |
504 | return 0; | |
f6d6d4fc CH |
505 | } |
506 | ||
f6d6d4fc CH |
507 | /* |
508 | * Test to see if we've been building up a completion structure for | |
509 | * earlier buffers -- if so, we try to append to this ioend if we | |
510 | * can, otherwise we finish off any current ioend and start another. | |
e10de372 DC |
511 | * Return the ioend we finished off so that the caller can submit it |
512 | * once it has finished processing the dirty page. | |
f6d6d4fc CH |
513 | */ |
514 | STATIC void | |
515 | xfs_add_to_ioend( | |
516 | struct inode *inode, | |
517 | struct buffer_head *bh, | |
7336cea8 | 518 | xfs_off_t offset, |
e10de372 DC |
519 | struct xfs_writepage_ctx *wpc, |
520 | struct list_head *iolist) | |
f6d6d4fc | 521 | { |
fbcc0256 DC |
522 | if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type || |
523 | bh->b_blocknr != wpc->last_block + 1) { | |
524 | struct xfs_ioend *new; | |
525 | ||
e10de372 DC |
526 | if (wpc->ioend) |
527 | list_add(&wpc->ioend->io_list, iolist); | |
528 | ||
fbcc0256 DC |
529 | new = xfs_alloc_ioend(inode, wpc->io_type); |
530 | new->io_offset = offset; | |
531 | new->io_buffer_head = bh; | |
532 | new->io_buffer_tail = bh; | |
fbcc0256 | 533 | wpc->ioend = new; |
f6d6d4fc | 534 | } else { |
fbcc0256 DC |
535 | wpc->ioend->io_buffer_tail->b_private = bh; |
536 | wpc->ioend->io_buffer_tail = bh; | |
f6d6d4fc CH |
537 | } |
538 | ||
539 | bh->b_private = NULL; | |
fbcc0256 DC |
540 | wpc->ioend->io_size += bh->b_size; |
541 | wpc->last_block = bh->b_blocknr; | |
e10de372 | 542 | xfs_start_buffer_writeback(bh); |
f6d6d4fc CH |
543 | } |
544 | ||
87cbc49c NS |
545 | STATIC void |
546 | xfs_map_buffer( | |
046f1685 | 547 | struct inode *inode, |
87cbc49c | 548 | struct buffer_head *bh, |
207d0416 | 549 | struct xfs_bmbt_irec *imap, |
046f1685 | 550 | xfs_off_t offset) |
87cbc49c NS |
551 | { |
552 | sector_t bn; | |
8699bb0a | 553 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
554 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
555 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 556 | |
207d0416 CH |
557 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
558 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 559 | |
e513182d | 560 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 561 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 562 | |
046f1685 | 563 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
564 | |
565 | bh->b_blocknr = bn; | |
566 | set_buffer_mapped(bh); | |
567 | } | |
568 | ||
1da177e4 LT |
569 | STATIC void |
570 | xfs_map_at_offset( | |
046f1685 | 571 | struct inode *inode, |
1da177e4 | 572 | struct buffer_head *bh, |
207d0416 | 573 | struct xfs_bmbt_irec *imap, |
046f1685 | 574 | xfs_off_t offset) |
1da177e4 | 575 | { |
207d0416 CH |
576 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
577 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 578 | |
207d0416 | 579 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
580 | set_buffer_mapped(bh); |
581 | clear_buffer_delay(bh); | |
f6d6d4fc | 582 | clear_buffer_unwritten(bh); |
1da177e4 LT |
583 | } |
584 | ||
1da177e4 | 585 | /* |
a49935f2 DC |
586 | * Test if a given page contains at least one buffer of a given @type. |
587 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
588 | * try to find one of the type passed in. If it is not set, then the caller only | |
589 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 590 | */ |
a49935f2 | 591 | STATIC bool |
6ffc4db5 | 592 | xfs_check_page_type( |
10ce4444 | 593 | struct page *page, |
a49935f2 DC |
594 | unsigned int type, |
595 | bool check_all_buffers) | |
1da177e4 | 596 | { |
a49935f2 DC |
597 | struct buffer_head *bh; |
598 | struct buffer_head *head; | |
1da177e4 | 599 | |
a49935f2 DC |
600 | if (PageWriteback(page)) |
601 | return false; | |
602 | if (!page->mapping) | |
603 | return false; | |
604 | if (!page_has_buffers(page)) | |
605 | return false; | |
1da177e4 | 606 | |
a49935f2 DC |
607 | bh = head = page_buffers(page); |
608 | do { | |
609 | if (buffer_unwritten(bh)) { | |
610 | if (type == XFS_IO_UNWRITTEN) | |
611 | return true; | |
612 | } else if (buffer_delay(bh)) { | |
805eeb8e | 613 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
614 | return true; |
615 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 616 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
617 | return true; |
618 | } | |
1da177e4 | 619 | |
a49935f2 DC |
620 | /* If we are only checking the first buffer, we are done now. */ |
621 | if (!check_all_buffers) | |
622 | break; | |
623 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 624 | |
a49935f2 | 625 | return false; |
1da177e4 LT |
626 | } |
627 | ||
3ed3a434 DC |
628 | STATIC void |
629 | xfs_vm_invalidatepage( | |
630 | struct page *page, | |
d47992f8 LC |
631 | unsigned int offset, |
632 | unsigned int length) | |
3ed3a434 | 633 | { |
34097dfe LC |
634 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
635 | length); | |
636 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
637 | } |
638 | ||
639 | /* | |
640 | * If the page has delalloc buffers on it, we need to punch them out before we | |
641 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
642 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
643 | * is done on that same region - the delalloc extent is returned when none is | |
644 | * supposed to be there. | |
645 | * | |
646 | * We prevent this by truncating away the delalloc regions on the page before | |
647 | * invalidating it. Because they are delalloc, we can do this without needing a | |
648 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
649 | * truncation without a transaction as there is no space left for block | |
650 | * reservation (typically why we see a ENOSPC in writeback). | |
651 | * | |
652 | * This is not a performance critical path, so for now just do the punching a | |
653 | * buffer head at a time. | |
654 | */ | |
655 | STATIC void | |
656 | xfs_aops_discard_page( | |
657 | struct page *page) | |
658 | { | |
659 | struct inode *inode = page->mapping->host; | |
660 | struct xfs_inode *ip = XFS_I(inode); | |
661 | struct buffer_head *bh, *head; | |
662 | loff_t offset = page_offset(page); | |
3ed3a434 | 663 | |
a49935f2 | 664 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
665 | goto out_invalidate; |
666 | ||
e8c3753c DC |
667 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
668 | goto out_invalidate; | |
669 | ||
4f10700a | 670 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
671 | "page discard on page %p, inode 0x%llx, offset %llu.", |
672 | page, ip->i_ino, offset); | |
673 | ||
674 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
675 | bh = head = page_buffers(page); | |
676 | do { | |
3ed3a434 | 677 | int error; |
c726de44 | 678 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
679 | |
680 | if (!buffer_delay(bh)) | |
681 | goto next_buffer; | |
682 | ||
c726de44 DC |
683 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
684 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
685 | if (error) { |
686 | /* something screwed, just bail */ | |
e8c3753c | 687 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 688 | xfs_alert(ip->i_mount, |
3ed3a434 | 689 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 690 | } |
3ed3a434 DC |
691 | break; |
692 | } | |
693 | next_buffer: | |
c726de44 | 694 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
695 | |
696 | } while ((bh = bh->b_this_page) != head); | |
697 | ||
698 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
699 | out_invalidate: | |
d47992f8 | 700 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
3ed3a434 DC |
701 | return; |
702 | } | |
703 | ||
e10de372 DC |
704 | /* |
705 | * We implement an immediate ioend submission policy here to avoid needing to | |
706 | * chain multiple ioends and hence nest mempool allocations which can violate | |
707 | * forward progress guarantees we need to provide. The current ioend we are | |
708 | * adding buffers to is cached on the writepage context, and if the new buffer | |
709 | * does not append to the cached ioend it will create a new ioend and cache that | |
710 | * instead. | |
711 | * | |
712 | * If a new ioend is created and cached, the old ioend is returned and queued | |
713 | * locally for submission once the entire page is processed or an error has been | |
714 | * detected. While ioends are submitted immediately after they are completed, | |
715 | * batching optimisations are provided by higher level block plugging. | |
716 | * | |
717 | * At the end of a writeback pass, there will be a cached ioend remaining on the | |
718 | * writepage context that the caller will need to submit. | |
719 | */ | |
bfce7d2e DC |
720 | static int |
721 | xfs_writepage_map( | |
722 | struct xfs_writepage_ctx *wpc, | |
e10de372 | 723 | struct writeback_control *wbc, |
bfce7d2e DC |
724 | struct inode *inode, |
725 | struct page *page, | |
726 | loff_t offset, | |
727 | __uint64_t end_offset) | |
728 | { | |
e10de372 DC |
729 | LIST_HEAD(submit_list); |
730 | struct xfs_ioend *ioend, *next; | |
bfce7d2e DC |
731 | struct buffer_head *bh, *head; |
732 | ssize_t len = 1 << inode->i_blkbits; | |
733 | int error = 0; | |
bfce7d2e | 734 | int count = 0; |
e10de372 | 735 | int uptodate = 1; |
bfce7d2e DC |
736 | |
737 | bh = head = page_buffers(page); | |
738 | offset = page_offset(page); | |
bfce7d2e DC |
739 | do { |
740 | if (offset >= end_offset) | |
741 | break; | |
742 | if (!buffer_uptodate(bh)) | |
743 | uptodate = 0; | |
744 | ||
745 | /* | |
746 | * set_page_dirty dirties all buffers in a page, independent | |
747 | * of their state. The dirty state however is entirely | |
748 | * meaningless for holes (!mapped && uptodate), so skip | |
749 | * buffers covering holes here. | |
750 | */ | |
751 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
752 | wpc->imap_valid = false; | |
753 | continue; | |
754 | } | |
755 | ||
756 | if (buffer_unwritten(bh)) { | |
757 | if (wpc->io_type != XFS_IO_UNWRITTEN) { | |
758 | wpc->io_type = XFS_IO_UNWRITTEN; | |
759 | wpc->imap_valid = false; | |
760 | } | |
761 | } else if (buffer_delay(bh)) { | |
762 | if (wpc->io_type != XFS_IO_DELALLOC) { | |
763 | wpc->io_type = XFS_IO_DELALLOC; | |
764 | wpc->imap_valid = false; | |
765 | } | |
766 | } else if (buffer_uptodate(bh)) { | |
767 | if (wpc->io_type != XFS_IO_OVERWRITE) { | |
768 | wpc->io_type = XFS_IO_OVERWRITE; | |
769 | wpc->imap_valid = false; | |
770 | } | |
771 | } else { | |
772 | if (PageUptodate(page)) | |
773 | ASSERT(buffer_mapped(bh)); | |
774 | /* | |
775 | * This buffer is not uptodate and will not be | |
776 | * written to disk. Ensure that we will put any | |
777 | * subsequent writeable buffers into a new | |
778 | * ioend. | |
779 | */ | |
780 | wpc->imap_valid = false; | |
781 | continue; | |
782 | } | |
783 | ||
784 | if (wpc->imap_valid) | |
785 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, | |
786 | offset); | |
787 | if (!wpc->imap_valid) { | |
788 | error = xfs_map_blocks(inode, offset, &wpc->imap, | |
789 | wpc->io_type); | |
790 | if (error) | |
e10de372 | 791 | goto out; |
bfce7d2e DC |
792 | wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, |
793 | offset); | |
794 | } | |
795 | if (wpc->imap_valid) { | |
796 | lock_buffer(bh); | |
797 | if (wpc->io_type != XFS_IO_OVERWRITE) | |
798 | xfs_map_at_offset(inode, bh, &wpc->imap, offset); | |
e10de372 | 799 | xfs_add_to_ioend(inode, bh, offset, wpc, &submit_list); |
bfce7d2e DC |
800 | count++; |
801 | } | |
802 | ||
bfce7d2e DC |
803 | } while (offset += len, ((bh = bh->b_this_page) != head)); |
804 | ||
805 | if (uptodate && bh == head) | |
806 | SetPageUptodate(page); | |
807 | ||
e10de372 | 808 | ASSERT(wpc->ioend || list_empty(&submit_list)); |
bfce7d2e | 809 | |
e10de372 | 810 | out: |
bfce7d2e | 811 | /* |
e10de372 DC |
812 | * On error, we have to fail the ioend here because we have locked |
813 | * buffers in the ioend. If we don't do this, we'll deadlock | |
814 | * invalidating the page as that tries to lock the buffers on the page. | |
815 | * Also, because we may have set pages under writeback, we have to make | |
816 | * sure we run IO completion to mark the error state of the IO | |
817 | * appropriately, so we can't cancel the ioend directly here. That means | |
818 | * we have to mark this page as under writeback if we included any | |
819 | * buffers from it in the ioend chain so that completion treats it | |
820 | * correctly. | |
bfce7d2e | 821 | * |
e10de372 DC |
822 | * If we didn't include the page in the ioend, the on error we can |
823 | * simply discard and unlock it as there are no other users of the page | |
824 | * or it's buffers right now. The caller will still need to trigger | |
825 | * submission of outstanding ioends on the writepage context so they are | |
826 | * treated correctly on error. | |
bfce7d2e | 827 | */ |
e10de372 DC |
828 | if (count) { |
829 | xfs_start_page_writeback(page, !error); | |
830 | ||
831 | /* | |
832 | * Preserve the original error if there was one, otherwise catch | |
833 | * submission errors here and propagate into subsequent ioend | |
834 | * submissions. | |
835 | */ | |
836 | list_for_each_entry_safe(ioend, next, &submit_list, io_list) { | |
837 | int error2; | |
838 | ||
839 | list_del_init(&ioend->io_list); | |
840 | error2 = xfs_submit_ioend(wbc, ioend, error); | |
841 | if (error2 && !error) | |
842 | error = error2; | |
843 | } | |
844 | } else if (error) { | |
bfce7d2e DC |
845 | xfs_aops_discard_page(page); |
846 | ClearPageUptodate(page); | |
847 | unlock_page(page); | |
e10de372 DC |
848 | } else { |
849 | /* | |
850 | * We can end up here with no error and nothing to write if we | |
851 | * race with a partial page truncate on a sub-page block sized | |
852 | * filesystem. In that case we need to mark the page clean. | |
853 | */ | |
854 | xfs_start_page_writeback(page, 1); | |
855 | end_page_writeback(page); | |
bfce7d2e | 856 | } |
e10de372 | 857 | |
bfce7d2e DC |
858 | mapping_set_error(page->mapping, error); |
859 | return error; | |
860 | } | |
861 | ||
1da177e4 | 862 | /* |
89f3b363 CH |
863 | * Write out a dirty page. |
864 | * | |
865 | * For delalloc space on the page we need to allocate space and flush it. | |
866 | * For unwritten space on the page we need to start the conversion to | |
867 | * regular allocated space. | |
89f3b363 | 868 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 869 | */ |
1da177e4 | 870 | STATIC int |
fbcc0256 | 871 | xfs_do_writepage( |
89f3b363 | 872 | struct page *page, |
fbcc0256 DC |
873 | struct writeback_control *wbc, |
874 | void *data) | |
1da177e4 | 875 | { |
fbcc0256 | 876 | struct xfs_writepage_ctx *wpc = data; |
89f3b363 | 877 | struct inode *inode = page->mapping->host; |
1da177e4 | 878 | loff_t offset; |
1da177e4 | 879 | __uint64_t end_offset; |
ad68972a | 880 | pgoff_t end_index; |
89f3b363 | 881 | |
34097dfe | 882 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 883 | |
20cb52eb CH |
884 | ASSERT(page_has_buffers(page)); |
885 | ||
89f3b363 CH |
886 | /* |
887 | * Refuse to write the page out if we are called from reclaim context. | |
888 | * | |
d4f7a5cb CH |
889 | * This avoids stack overflows when called from deeply used stacks in |
890 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
891 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 892 | * |
94054fa3 MG |
893 | * This should never happen except in the case of a VM regression so |
894 | * warn about it. | |
89f3b363 | 895 | */ |
94054fa3 MG |
896 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
897 | PF_MEMALLOC)) | |
b5420f23 | 898 | goto redirty; |
1da177e4 | 899 | |
89f3b363 | 900 | /* |
680a647b CH |
901 | * Given that we do not allow direct reclaim to call us, we should |
902 | * never be called while in a filesystem transaction. | |
89f3b363 | 903 | */ |
448011e2 | 904 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 905 | goto redirty; |
89f3b363 | 906 | |
8695d27e | 907 | /* |
ad68972a DC |
908 | * Is this page beyond the end of the file? |
909 | * | |
8695d27e JL |
910 | * The page index is less than the end_index, adjust the end_offset |
911 | * to the highest offset that this page should represent. | |
912 | * ----------------------------------------------------- | |
913 | * | file mapping | <EOF> | | |
914 | * ----------------------------------------------------- | |
915 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
916 | * ^--------------------------------^----------|-------- | |
917 | * | desired writeback range | see else | | |
918 | * ---------------------------------^------------------| | |
919 | */ | |
ad68972a DC |
920 | offset = i_size_read(inode); |
921 | end_index = offset >> PAGE_CACHE_SHIFT; | |
8695d27e JL |
922 | if (page->index < end_index) |
923 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; | |
924 | else { | |
925 | /* | |
926 | * Check whether the page to write out is beyond or straddles | |
927 | * i_size or not. | |
928 | * ------------------------------------------------------- | |
929 | * | file mapping | <EOF> | | |
930 | * ------------------------------------------------------- | |
931 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
932 | * ^--------------------------------^-----------|--------- | |
933 | * | | Straddles | | |
934 | * ---------------------------------^-----------|--------| | |
935 | */ | |
6b7a03f0 CH |
936 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
937 | ||
938 | /* | |
ff9a28f6 JK |
939 | * Skip the page if it is fully outside i_size, e.g. due to a |
940 | * truncate operation that is in progress. We must redirty the | |
941 | * page so that reclaim stops reclaiming it. Otherwise | |
942 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
943 | * |
944 | * Note that the end_index is unsigned long, it would overflow | |
945 | * if the given offset is greater than 16TB on 32-bit system | |
946 | * and if we do check the page is fully outside i_size or not | |
947 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
948 | * will be evaluated to 0. Hence this page will be redirtied | |
949 | * and be written out repeatedly which would result in an | |
950 | * infinite loop, the user program that perform this operation | |
951 | * will hang. Instead, we can verify this situation by checking | |
952 | * if the page to write is totally beyond the i_size or if it's | |
953 | * offset is just equal to the EOF. | |
6b7a03f0 | 954 | */ |
8695d27e JL |
955 | if (page->index > end_index || |
956 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 957 | goto redirty; |
6b7a03f0 CH |
958 | |
959 | /* | |
960 | * The page straddles i_size. It must be zeroed out on each | |
961 | * and every writepage invocation because it may be mmapped. | |
962 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 963 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
964 | * memory is zeroed when mapped, and writes to that region are |
965 | * not written out to the file." | |
966 | */ | |
967 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
8695d27e JL |
968 | |
969 | /* Adjust the end_offset to the end of file */ | |
970 | end_offset = offset; | |
1da177e4 LT |
971 | } |
972 | ||
e10de372 | 973 | return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset); |
f51623b2 | 974 | |
b5420f23 | 975 | redirty: |
f51623b2 NS |
976 | redirty_page_for_writepage(wbc, page); |
977 | unlock_page(page); | |
978 | return 0; | |
f51623b2 NS |
979 | } |
980 | ||
fbcc0256 DC |
981 | STATIC int |
982 | xfs_vm_writepage( | |
983 | struct page *page, | |
984 | struct writeback_control *wbc) | |
985 | { | |
986 | struct xfs_writepage_ctx wpc = { | |
987 | .io_type = XFS_IO_INVALID, | |
988 | }; | |
989 | int ret; | |
990 | ||
991 | ret = xfs_do_writepage(page, wbc, &wpc); | |
e10de372 DC |
992 | if (wpc.ioend) |
993 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
994 | return ret; | |
fbcc0256 DC |
995 | } |
996 | ||
7d4fb40a NS |
997 | STATIC int |
998 | xfs_vm_writepages( | |
999 | struct address_space *mapping, | |
1000 | struct writeback_control *wbc) | |
1001 | { | |
fbcc0256 DC |
1002 | struct xfs_writepage_ctx wpc = { |
1003 | .io_type = XFS_IO_INVALID, | |
1004 | }; | |
1005 | int ret; | |
1006 | ||
b3aea4ed | 1007 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
fbcc0256 | 1008 | ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc); |
e10de372 DC |
1009 | if (wpc.ioend) |
1010 | ret = xfs_submit_ioend(wbc, wpc.ioend, ret); | |
1011 | return ret; | |
7d4fb40a NS |
1012 | } |
1013 | ||
f51623b2 NS |
1014 | /* |
1015 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1016 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1017 | * have buffer heads in this call. |
1018 | * | |
89f3b363 | 1019 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1020 | */ |
1021 | STATIC int | |
238f4c54 | 1022 | xfs_vm_releasepage( |
f51623b2 NS |
1023 | struct page *page, |
1024 | gfp_t gfp_mask) | |
1025 | { | |
20cb52eb | 1026 | int delalloc, unwritten; |
f51623b2 | 1027 | |
34097dfe | 1028 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1029 | |
20cb52eb | 1030 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1031 | |
448011e2 | 1032 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1033 | return 0; |
448011e2 | 1034 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1035 | return 0; |
1036 | ||
f51623b2 NS |
1037 | return try_to_free_buffers(page); |
1038 | } | |
1039 | ||
a719370b | 1040 | /* |
a06c277a DC |
1041 | * When we map a DIO buffer, we may need to attach an ioend that describes the |
1042 | * type of write IO we are doing. This passes to the completion function the | |
1043 | * operations it needs to perform. If the mapping is for an overwrite wholly | |
1044 | * within the EOF then we don't need an ioend and so we don't allocate one. | |
1045 | * This avoids the unnecessary overhead of allocating and freeing ioends for | |
1046 | * workloads that don't require transactions on IO completion. | |
d5cc2e3f DC |
1047 | * |
1048 | * If we get multiple mappings in a single IO, we might be mapping different | |
1049 | * types. But because the direct IO can only have a single private pointer, we | |
1050 | * need to ensure that: | |
1051 | * | |
a06c277a DC |
1052 | * a) i) the ioend spans the entire region of unwritten mappings; or |
1053 | * ii) the ioend spans all the mappings that cross or are beyond EOF; and | |
d5cc2e3f DC |
1054 | * b) if it contains unwritten extents, it is *permanently* marked as such |
1055 | * | |
1056 | * We could do this by chaining ioends like buffered IO does, but we only | |
1057 | * actually get one IO completion callback from the direct IO, and that spans | |
1058 | * the entire IO regardless of how many mappings and IOs are needed to complete | |
1059 | * the DIO. There is only going to be one reference to the ioend and its life | |
1060 | * cycle is constrained by the DIO completion code. hence we don't need | |
1061 | * reference counting here. | |
3e12dbbd DC |
1062 | * |
1063 | * Note that for DIO, an IO to the highest supported file block offset (i.e. | |
1064 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 | |
1065 | * bit variable. Hence if we see this overflow, we have to assume that the IO is | |
1066 | * extending the file size. We won't know for sure until IO completion is run | |
1067 | * and the actual max write offset is communicated to the IO completion | |
1068 | * routine. | |
1069 | * | |
1070 | * For DAX page faults, we are preparing to never see unwritten extents here, | |
1071 | * nor should we ever extend the inode size. Hence we will soon have nothing to | |
1072 | * do here for this case, ensuring we don't have to provide an IO completion | |
1073 | * callback to free an ioend that we don't actually need for a fault into the | |
1074 | * page at offset (2^63 - 1FSB) bytes. | |
a719370b | 1075 | */ |
3e12dbbd | 1076 | |
a719370b DC |
1077 | static void |
1078 | xfs_map_direct( | |
1079 | struct inode *inode, | |
1080 | struct buffer_head *bh_result, | |
1081 | struct xfs_bmbt_irec *imap, | |
3e12dbbd DC |
1082 | xfs_off_t offset, |
1083 | bool dax_fault) | |
a719370b | 1084 | { |
d5cc2e3f DC |
1085 | struct xfs_ioend *ioend; |
1086 | xfs_off_t size = bh_result->b_size; | |
1087 | int type; | |
1088 | ||
1089 | if (ISUNWRITTEN(imap)) | |
1090 | type = XFS_IO_UNWRITTEN; | |
1091 | else | |
1092 | type = XFS_IO_OVERWRITE; | |
1093 | ||
1094 | trace_xfs_gbmap_direct(XFS_I(inode), offset, size, type, imap); | |
1095 | ||
3e12dbbd DC |
1096 | if (dax_fault) { |
1097 | ASSERT(type == XFS_IO_OVERWRITE); | |
1098 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, | |
1099 | imap); | |
1100 | return; | |
1101 | } | |
3e12dbbd | 1102 | |
d5cc2e3f DC |
1103 | if (bh_result->b_private) { |
1104 | ioend = bh_result->b_private; | |
1105 | ASSERT(ioend->io_size > 0); | |
1106 | ASSERT(offset >= ioend->io_offset); | |
1107 | if (offset + size > ioend->io_offset + ioend->io_size) | |
1108 | ioend->io_size = offset - ioend->io_offset + size; | |
1109 | ||
1110 | if (type == XFS_IO_UNWRITTEN && type != ioend->io_type) | |
1111 | ioend->io_type = XFS_IO_UNWRITTEN; | |
1112 | ||
1113 | trace_xfs_gbmap_direct_update(XFS_I(inode), ioend->io_offset, | |
1114 | ioend->io_size, ioend->io_type, | |
1115 | imap); | |
a06c277a | 1116 | } else if (type == XFS_IO_UNWRITTEN || |
3e12dbbd DC |
1117 | offset + size > i_size_read(inode) || |
1118 | offset + size < 0) { | |
d5cc2e3f DC |
1119 | ioend = xfs_alloc_ioend(inode, type); |
1120 | ioend->io_offset = offset; | |
1121 | ioend->io_size = size; | |
a06c277a | 1122 | |
d5cc2e3f | 1123 | bh_result->b_private = ioend; |
a06c277a | 1124 | set_buffer_defer_completion(bh_result); |
d5cc2e3f DC |
1125 | |
1126 | trace_xfs_gbmap_direct_new(XFS_I(inode), offset, size, type, | |
1127 | imap); | |
a06c277a DC |
1128 | } else { |
1129 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, | |
1130 | imap); | |
a719370b DC |
1131 | } |
1132 | } | |
1133 | ||
1fdca9c2 DC |
1134 | /* |
1135 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1136 | * is, so that we can avoid repeated get_blocks calls. | |
1137 | * | |
1138 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1139 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1140 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1141 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1142 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1143 | * to and including EOF, and a second mapping for beyond EOF. | |
1144 | */ | |
1145 | static void | |
1146 | xfs_map_trim_size( | |
1147 | struct inode *inode, | |
1148 | sector_t iblock, | |
1149 | struct buffer_head *bh_result, | |
1150 | struct xfs_bmbt_irec *imap, | |
1151 | xfs_off_t offset, | |
1152 | ssize_t size) | |
1153 | { | |
1154 | xfs_off_t mapping_size; | |
1155 | ||
1156 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1157 | mapping_size <<= inode->i_blkbits; | |
1158 | ||
1159 | ASSERT(mapping_size > 0); | |
1160 | if (mapping_size > size) | |
1161 | mapping_size = size; | |
1162 | if (offset < i_size_read(inode) && | |
1163 | offset + mapping_size >= i_size_read(inode)) { | |
1164 | /* limit mapping to block that spans EOF */ | |
1165 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1166 | 1 << inode->i_blkbits); | |
1167 | } | |
1168 | if (mapping_size > LONG_MAX) | |
1169 | mapping_size = LONG_MAX; | |
1170 | ||
1171 | bh_result->b_size = mapping_size; | |
1172 | } | |
1173 | ||
1da177e4 | 1174 | STATIC int |
c2536668 | 1175 | __xfs_get_blocks( |
1da177e4 LT |
1176 | struct inode *inode, |
1177 | sector_t iblock, | |
1da177e4 LT |
1178 | struct buffer_head *bh_result, |
1179 | int create, | |
3e12dbbd DC |
1180 | bool direct, |
1181 | bool dax_fault) | |
1da177e4 | 1182 | { |
a206c817 CH |
1183 | struct xfs_inode *ip = XFS_I(inode); |
1184 | struct xfs_mount *mp = ip->i_mount; | |
1185 | xfs_fileoff_t offset_fsb, end_fsb; | |
1186 | int error = 0; | |
1187 | int lockmode = 0; | |
207d0416 | 1188 | struct xfs_bmbt_irec imap; |
a206c817 | 1189 | int nimaps = 1; |
fdc7ed75 NS |
1190 | xfs_off_t offset; |
1191 | ssize_t size; | |
207d0416 | 1192 | int new = 0; |
a206c817 CH |
1193 | |
1194 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1195 | return -EIO; |
1da177e4 | 1196 | |
fdc7ed75 | 1197 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1198 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1199 | size = bh_result->b_size; | |
364f358a LM |
1200 | |
1201 | if (!create && direct && offset >= i_size_read(inode)) | |
1202 | return 0; | |
1203 | ||
507630b2 DC |
1204 | /* |
1205 | * Direct I/O is usually done on preallocated files, so try getting | |
1206 | * a block mapping without an exclusive lock first. For buffered | |
1207 | * writes we already have the exclusive iolock anyway, so avoiding | |
1208 | * a lock roundtrip here by taking the ilock exclusive from the | |
1209 | * beginning is a useful micro optimization. | |
1210 | */ | |
1211 | if (create && !direct) { | |
a206c817 CH |
1212 | lockmode = XFS_ILOCK_EXCL; |
1213 | xfs_ilock(ip, lockmode); | |
1214 | } else { | |
309ecac8 | 1215 | lockmode = xfs_ilock_data_map_shared(ip); |
a206c817 | 1216 | } |
f2bde9b8 | 1217 | |
d2c28191 DC |
1218 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1219 | if (offset + size > mp->m_super->s_maxbytes) | |
1220 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1221 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1222 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1223 | ||
5c8ed202 DC |
1224 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1225 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1226 | if (error) |
a206c817 CH |
1227 | goto out_unlock; |
1228 | ||
1ca19157 | 1229 | /* for DAX, we convert unwritten extents directly */ |
a206c817 CH |
1230 | if (create && |
1231 | (!nimaps || | |
1232 | (imap.br_startblock == HOLESTARTBLOCK || | |
1ca19157 DC |
1233 | imap.br_startblock == DELAYSTARTBLOCK) || |
1234 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { | |
aff3a9ed | 1235 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 | 1236 | /* |
009c6e87 BF |
1237 | * xfs_iomap_write_direct() expects the shared lock. It |
1238 | * is unlocked on return. | |
507630b2 | 1239 | */ |
009c6e87 BF |
1240 | if (lockmode == XFS_ILOCK_EXCL) |
1241 | xfs_ilock_demote(ip, lockmode); | |
1242 | ||
a206c817 CH |
1243 | error = xfs_iomap_write_direct(ip, offset, size, |
1244 | &imap, nimaps); | |
507630b2 | 1245 | if (error) |
2451337d | 1246 | return error; |
d3bc815a | 1247 | new = 1; |
6b698ede | 1248 | |
a206c817 | 1249 | } else { |
507630b2 DC |
1250 | /* |
1251 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1252 | * we can go on without dropping the lock here. If we |
1253 | * are allocating a new delalloc block, make sure that | |
1254 | * we set the new flag so that we mark the buffer new so | |
1255 | * that we know that it is newly allocated if the write | |
1256 | * fails. | |
507630b2 | 1257 | */ |
d3bc815a DC |
1258 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1259 | new = 1; | |
a206c817 | 1260 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1261 | if (error) |
1262 | goto out_unlock; | |
1263 | ||
1264 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1265 | } |
d5cc2e3f DC |
1266 | trace_xfs_get_blocks_alloc(ip, offset, size, |
1267 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1268 | : XFS_IO_DELALLOC, &imap); | |
a206c817 | 1269 | } else if (nimaps) { |
d5cc2e3f DC |
1270 | trace_xfs_get_blocks_found(ip, offset, size, |
1271 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1272 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1273 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1274 | } else { |
1275 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1276 | goto out_unlock; | |
1277 | } | |
1da177e4 | 1278 | |
1ca19157 DC |
1279 | if (IS_DAX(inode) && create) { |
1280 | ASSERT(!ISUNWRITTEN(&imap)); | |
1281 | /* zeroing is not needed at a higher layer */ | |
1282 | new = 0; | |
1283 | } | |
1284 | ||
1fdca9c2 DC |
1285 | /* trim mapping down to size requested */ |
1286 | if (direct || size > (1 << inode->i_blkbits)) | |
1287 | xfs_map_trim_size(inode, iblock, bh_result, | |
1288 | &imap, offset, size); | |
1289 | ||
a719370b DC |
1290 | /* |
1291 | * For unwritten extents do not report a disk address in the buffered | |
1292 | * read case (treat as if we're reading into a hole). | |
1293 | */ | |
207d0416 | 1294 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1295 | imap.br_startblock != DELAYSTARTBLOCK && |
1296 | (create || !ISUNWRITTEN(&imap))) { | |
1297 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1298 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1299 | set_buffer_unwritten(bh_result); |
a719370b DC |
1300 | /* direct IO needs special help */ |
1301 | if (create && direct) | |
3e12dbbd DC |
1302 | xfs_map_direct(inode, bh_result, &imap, offset, |
1303 | dax_fault); | |
1da177e4 LT |
1304 | } |
1305 | ||
c2536668 NS |
1306 | /* |
1307 | * If this is a realtime file, data may be on a different device. | |
1308 | * to that pointed to from the buffer_head b_bdev currently. | |
1309 | */ | |
046f1685 | 1310 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1311 | |
c2536668 | 1312 | /* |
549054af DC |
1313 | * If we previously allocated a block out beyond eof and we are now |
1314 | * coming back to use it then we will need to flag it as new even if it | |
1315 | * has a disk address. | |
1316 | * | |
1317 | * With sub-block writes into unwritten extents we also need to mark | |
1318 | * the buffer as new so that the unwritten parts of the buffer gets | |
1319 | * correctly zeroed. | |
1da177e4 LT |
1320 | */ |
1321 | if (create && | |
1322 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1323 | (offset >= i_size_read(inode)) || |
207d0416 | 1324 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1325 | set_buffer_new(bh_result); |
1da177e4 | 1326 | |
207d0416 | 1327 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1328 | BUG_ON(direct); |
1329 | if (create) { | |
1330 | set_buffer_uptodate(bh_result); | |
1331 | set_buffer_mapped(bh_result); | |
1332 | set_buffer_delay(bh_result); | |
1333 | } | |
1334 | } | |
1335 | ||
1da177e4 | 1336 | return 0; |
a206c817 CH |
1337 | |
1338 | out_unlock: | |
1339 | xfs_iunlock(ip, lockmode); | |
2451337d | 1340 | return error; |
1da177e4 LT |
1341 | } |
1342 | ||
1343 | int | |
c2536668 | 1344 | xfs_get_blocks( |
1da177e4 LT |
1345 | struct inode *inode, |
1346 | sector_t iblock, | |
1347 | struct buffer_head *bh_result, | |
1348 | int create) | |
1349 | { | |
3e12dbbd | 1350 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
1da177e4 LT |
1351 | } |
1352 | ||
6b698ede | 1353 | int |
e4c573bb | 1354 | xfs_get_blocks_direct( |
1da177e4 LT |
1355 | struct inode *inode, |
1356 | sector_t iblock, | |
1da177e4 LT |
1357 | struct buffer_head *bh_result, |
1358 | int create) | |
1359 | { | |
3e12dbbd DC |
1360 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
1361 | } | |
1362 | ||
1363 | int | |
1364 | xfs_get_blocks_dax_fault( | |
1365 | struct inode *inode, | |
1366 | sector_t iblock, | |
1367 | struct buffer_head *bh_result, | |
1368 | int create) | |
1369 | { | |
1370 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); | |
1da177e4 LT |
1371 | } |
1372 | ||
6b698ede DC |
1373 | static void |
1374 | __xfs_end_io_direct_write( | |
1375 | struct inode *inode, | |
1376 | struct xfs_ioend *ioend, | |
209fb87a | 1377 | loff_t offset, |
6b698ede | 1378 | ssize_t size) |
f0973863 | 1379 | { |
6b698ede | 1380 | struct xfs_mount *mp = XFS_I(inode)->i_mount; |
a06c277a | 1381 | |
6b698ede | 1382 | if (XFS_FORCED_SHUTDOWN(mp) || ioend->io_error) |
6dfa1b67 | 1383 | goto out_end_io; |
f0973863 | 1384 | |
2813d682 | 1385 | /* |
d5cc2e3f DC |
1386 | * dio completion end_io functions are only called on writes if more |
1387 | * than 0 bytes was written. | |
2813d682 | 1388 | */ |
d5cc2e3f DC |
1389 | ASSERT(size > 0); |
1390 | ||
1391 | /* | |
1392 | * The ioend only maps whole blocks, while the IO may be sector aligned. | |
a06c277a DC |
1393 | * Hence the ioend offset/size may not match the IO offset/size exactly. |
1394 | * Because we don't map overwrites within EOF into the ioend, the offset | |
1395 | * may not match, but only if the endio spans EOF. Either way, write | |
1396 | * the IO sizes into the ioend so that completion processing does the | |
1397 | * right thing. | |
d5cc2e3f | 1398 | */ |
d5cc2e3f DC |
1399 | ASSERT(offset + size <= ioend->io_offset + ioend->io_size); |
1400 | ioend->io_size = size; | |
1401 | ioend->io_offset = offset; | |
f0973863 | 1402 | |
2813d682 | 1403 | /* |
6dfa1b67 DC |
1404 | * The ioend tells us whether we are doing unwritten extent conversion |
1405 | * or an append transaction that updates the on-disk file size. These | |
1406 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1407 | * to update the VFS inode size. |
6dfa1b67 DC |
1408 | * |
1409 | * We need to update the in-core inode size here so that we don't end up | |
a06c277a DC |
1410 | * with the on-disk inode size being outside the in-core inode size. We |
1411 | * have no other method of updating EOF for AIO, so always do it here | |
1412 | * if necessary. | |
b9d59846 DC |
1413 | * |
1414 | * We need to lock the test/set EOF update as we can be racing with | |
1415 | * other IO completions here to update the EOF. Failing to serialise | |
1416 | * here can result in EOF moving backwards and Bad Things Happen when | |
1417 | * that occurs. | |
2813d682 | 1418 | */ |
6b698ede | 1419 | spin_lock(&XFS_I(inode)->i_flags_lock); |
2ba66237 CH |
1420 | if (offset + size > i_size_read(inode)) |
1421 | i_size_write(inode, offset + size); | |
6b698ede | 1422 | spin_unlock(&XFS_I(inode)->i_flags_lock); |
2813d682 | 1423 | |
f0973863 | 1424 | /* |
6dfa1b67 DC |
1425 | * If we are doing an append IO that needs to update the EOF on disk, |
1426 | * do the transaction reserve now so we can use common end io | |
1427 | * processing. Stashing the error (if there is one) in the ioend will | |
1428 | * result in the ioend processing passing on the error if it is | |
1429 | * possible as we can't return it from here. | |
f0973863 | 1430 | */ |
a06c277a | 1431 | if (ioend->io_type == XFS_IO_OVERWRITE) |
6dfa1b67 | 1432 | ioend->io_error = xfs_setfilesize_trans_alloc(ioend); |
209fb87a | 1433 | |
6dfa1b67 DC |
1434 | out_end_io: |
1435 | xfs_end_io(&ioend->io_work); | |
1436 | return; | |
f0973863 CH |
1437 | } |
1438 | ||
6b698ede DC |
1439 | /* |
1440 | * Complete a direct I/O write request. | |
1441 | * | |
1442 | * The ioend structure is passed from __xfs_get_blocks() to tell us what to do. | |
1443 | * If no ioend exists (i.e. @private == NULL) then the write IO is an overwrite | |
1444 | * wholly within the EOF and so there is nothing for us to do. Note that in this | |
1445 | * case the completion can be called in interrupt context, whereas if we have an | |
1446 | * ioend we will always be called in task context (i.e. from a workqueue). | |
1447 | */ | |
1448 | STATIC void | |
1449 | xfs_end_io_direct_write( | |
1450 | struct kiocb *iocb, | |
1451 | loff_t offset, | |
1452 | ssize_t size, | |
1453 | void *private) | |
1454 | { | |
1455 | struct inode *inode = file_inode(iocb->ki_filp); | |
1456 | struct xfs_ioend *ioend = private; | |
1457 | ||
1458 | trace_xfs_gbmap_direct_endio(XFS_I(inode), offset, size, | |
1459 | ioend ? ioend->io_type : 0, NULL); | |
1460 | ||
1461 | if (!ioend) { | |
1462 | ASSERT(offset + size <= i_size_read(inode)); | |
1463 | return; | |
1464 | } | |
1465 | ||
1466 | __xfs_end_io_direct_write(inode, ioend, offset, size); | |
1467 | } | |
1468 | ||
6e1ba0bc DC |
1469 | static inline ssize_t |
1470 | xfs_vm_do_dio( | |
1471 | struct inode *inode, | |
1472 | struct kiocb *iocb, | |
1473 | struct iov_iter *iter, | |
1474 | loff_t offset, | |
1475 | void (*endio)(struct kiocb *iocb, | |
1476 | loff_t offset, | |
1477 | ssize_t size, | |
1478 | void *private), | |
1479 | int flags) | |
1480 | { | |
1481 | struct block_device *bdev; | |
1482 | ||
1483 | if (IS_DAX(inode)) | |
1484 | return dax_do_io(iocb, inode, iter, offset, | |
1485 | xfs_get_blocks_direct, endio, 0); | |
1486 | ||
1487 | bdev = xfs_find_bdev_for_inode(inode); | |
1488 | return __blockdev_direct_IO(iocb, inode, bdev, iter, offset, | |
1489 | xfs_get_blocks_direct, endio, NULL, flags); | |
1490 | } | |
1491 | ||
1da177e4 | 1492 | STATIC ssize_t |
e4c573bb | 1493 | xfs_vm_direct_IO( |
1da177e4 | 1494 | struct kiocb *iocb, |
d8d3d94b AV |
1495 | struct iov_iter *iter, |
1496 | loff_t offset) | |
1da177e4 | 1497 | { |
209fb87a | 1498 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
209fb87a | 1499 | |
6e1ba0bc DC |
1500 | if (iov_iter_rw(iter) == WRITE) |
1501 | return xfs_vm_do_dio(inode, iocb, iter, offset, | |
1502 | xfs_end_io_direct_write, DIO_ASYNC_EXTEND); | |
1503 | return xfs_vm_do_dio(inode, iocb, iter, offset, NULL, 0); | |
1da177e4 LT |
1504 | } |
1505 | ||
d3bc815a DC |
1506 | /* |
1507 | * Punch out the delalloc blocks we have already allocated. | |
1508 | * | |
1509 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1510 | * as the page is still locked at this point. | |
1511 | */ | |
1512 | STATIC void | |
1513 | xfs_vm_kill_delalloc_range( | |
1514 | struct inode *inode, | |
1515 | loff_t start, | |
1516 | loff_t end) | |
1517 | { | |
1518 | struct xfs_inode *ip = XFS_I(inode); | |
1519 | xfs_fileoff_t start_fsb; | |
1520 | xfs_fileoff_t end_fsb; | |
1521 | int error; | |
1522 | ||
1523 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1524 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1525 | if (end_fsb <= start_fsb) | |
1526 | return; | |
1527 | ||
1528 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1529 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1530 | end_fsb - start_fsb); | |
1531 | if (error) { | |
1532 | /* something screwed, just bail */ | |
1533 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1534 | xfs_alert(ip->i_mount, | |
1535 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1536 | ip->i_ino); | |
1537 | } | |
1538 | } | |
1539 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1540 | } | |
1541 | ||
fa9b227e CH |
1542 | STATIC void |
1543 | xfs_vm_write_failed( | |
d3bc815a DC |
1544 | struct inode *inode, |
1545 | struct page *page, | |
1546 | loff_t pos, | |
1547 | unsigned len) | |
fa9b227e | 1548 | { |
58e59854 | 1549 | loff_t block_offset; |
d3bc815a DC |
1550 | loff_t block_start; |
1551 | loff_t block_end; | |
1552 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1553 | loff_t to = from + len; | |
1554 | struct buffer_head *bh, *head; | |
fa9b227e | 1555 | |
58e59854 JL |
1556 | /* |
1557 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1558 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1559 | * platform, the high 32-bit will be masked off if we evaluate the | |
1560 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1561 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1562 | * which could cause the following ASSERT failed in most cases. | |
1563 | * In order to avoid this, we can evaluate the block_offset of the | |
1564 | * start of the page by using shifts rather than masks the mismatch | |
1565 | * problem. | |
1566 | */ | |
1567 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1568 | ||
d3bc815a | 1569 | ASSERT(block_offset + from == pos); |
c726de44 | 1570 | |
d3bc815a DC |
1571 | head = page_buffers(page); |
1572 | block_start = 0; | |
1573 | for (bh = head; bh != head || !block_start; | |
1574 | bh = bh->b_this_page, block_start = block_end, | |
1575 | block_offset += bh->b_size) { | |
1576 | block_end = block_start + bh->b_size; | |
c726de44 | 1577 | |
d3bc815a DC |
1578 | /* skip buffers before the write */ |
1579 | if (block_end <= from) | |
1580 | continue; | |
1581 | ||
1582 | /* if the buffer is after the write, we're done */ | |
1583 | if (block_start >= to) | |
1584 | break; | |
1585 | ||
1586 | if (!buffer_delay(bh)) | |
1587 | continue; | |
1588 | ||
1589 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1590 | continue; | |
1591 | ||
1592 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1593 | block_offset + bh->b_size); | |
4ab9ed57 DC |
1594 | |
1595 | /* | |
1596 | * This buffer does not contain data anymore. make sure anyone | |
1597 | * who finds it knows that for certain. | |
1598 | */ | |
1599 | clear_buffer_delay(bh); | |
1600 | clear_buffer_uptodate(bh); | |
1601 | clear_buffer_mapped(bh); | |
1602 | clear_buffer_new(bh); | |
1603 | clear_buffer_dirty(bh); | |
fa9b227e | 1604 | } |
d3bc815a | 1605 | |
fa9b227e CH |
1606 | } |
1607 | ||
d3bc815a DC |
1608 | /* |
1609 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1610 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1611 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1612 | * the appropriate point. | |
1613 | */ | |
f51623b2 | 1614 | STATIC int |
d79689c7 | 1615 | xfs_vm_write_begin( |
f51623b2 | 1616 | struct file *file, |
d79689c7 NP |
1617 | struct address_space *mapping, |
1618 | loff_t pos, | |
1619 | unsigned len, | |
1620 | unsigned flags, | |
1621 | struct page **pagep, | |
1622 | void **fsdata) | |
f51623b2 | 1623 | { |
d3bc815a DC |
1624 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1625 | struct page *page; | |
1626 | int status; | |
155130a4 | 1627 | |
d3bc815a DC |
1628 | ASSERT(len <= PAGE_CACHE_SIZE); |
1629 | ||
ad22c7a0 | 1630 | page = grab_cache_page_write_begin(mapping, index, flags); |
d3bc815a DC |
1631 | if (!page) |
1632 | return -ENOMEM; | |
1633 | ||
1634 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1635 | if (unlikely(status)) { | |
1636 | struct inode *inode = mapping->host; | |
72ab70a1 | 1637 | size_t isize = i_size_read(inode); |
d3bc815a DC |
1638 | |
1639 | xfs_vm_write_failed(inode, page, pos, len); | |
1640 | unlock_page(page); | |
1641 | ||
72ab70a1 DC |
1642 | /* |
1643 | * If the write is beyond EOF, we only want to kill blocks | |
1644 | * allocated in this write, not blocks that were previously | |
1645 | * written successfully. | |
1646 | */ | |
1647 | if (pos + len > isize) { | |
1648 | ssize_t start = max_t(ssize_t, pos, isize); | |
1649 | ||
1650 | truncate_pagecache_range(inode, start, pos + len); | |
1651 | } | |
d3bc815a DC |
1652 | |
1653 | page_cache_release(page); | |
1654 | page = NULL; | |
1655 | } | |
1656 | ||
1657 | *pagep = page; | |
1658 | return status; | |
fa9b227e CH |
1659 | } |
1660 | ||
d3bc815a | 1661 | /* |
aad3f375 DC |
1662 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
1663 | * this specific write because they will never be written. Previous writes | |
1664 | * beyond EOF where block allocation succeeded do not need to be trashed, so | |
1665 | * only new blocks from this write should be trashed. For blocks within | |
1666 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be | |
1667 | * written with all the other valid data. | |
d3bc815a | 1668 | */ |
fa9b227e CH |
1669 | STATIC int |
1670 | xfs_vm_write_end( | |
1671 | struct file *file, | |
1672 | struct address_space *mapping, | |
1673 | loff_t pos, | |
1674 | unsigned len, | |
1675 | unsigned copied, | |
1676 | struct page *page, | |
1677 | void *fsdata) | |
1678 | { | |
1679 | int ret; | |
155130a4 | 1680 | |
d3bc815a DC |
1681 | ASSERT(len <= PAGE_CACHE_SIZE); |
1682 | ||
fa9b227e | 1683 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1684 | if (unlikely(ret < len)) { |
1685 | struct inode *inode = mapping->host; | |
1686 | size_t isize = i_size_read(inode); | |
1687 | loff_t to = pos + len; | |
1688 | ||
1689 | if (to > isize) { | |
aad3f375 DC |
1690 | /* only kill blocks in this write beyond EOF */ |
1691 | if (pos > isize) | |
1692 | isize = pos; | |
d3bc815a | 1693 | xfs_vm_kill_delalloc_range(inode, isize, to); |
aad3f375 | 1694 | truncate_pagecache_range(inode, isize, to); |
d3bc815a DC |
1695 | } |
1696 | } | |
155130a4 | 1697 | return ret; |
f51623b2 | 1698 | } |
1da177e4 LT |
1699 | |
1700 | STATIC sector_t | |
e4c573bb | 1701 | xfs_vm_bmap( |
1da177e4 LT |
1702 | struct address_space *mapping, |
1703 | sector_t block) | |
1704 | { | |
1705 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1706 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1707 | |
cca28fb8 | 1708 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1709 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1710 | filemap_write_and_wait(mapping); |
126468b1 | 1711 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1712 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1713 | } |
1714 | ||
1715 | STATIC int | |
e4c573bb | 1716 | xfs_vm_readpage( |
1da177e4 LT |
1717 | struct file *unused, |
1718 | struct page *page) | |
1719 | { | |
121e213e | 1720 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1721 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1722 | } |
1723 | ||
1724 | STATIC int | |
e4c573bb | 1725 | xfs_vm_readpages( |
1da177e4 LT |
1726 | struct file *unused, |
1727 | struct address_space *mapping, | |
1728 | struct list_head *pages, | |
1729 | unsigned nr_pages) | |
1730 | { | |
121e213e | 1731 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1732 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1733 | } |
1734 | ||
22e757a4 DC |
1735 | /* |
1736 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1737 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1738 | * dirty, we'll never be able to clean them because we don't write buffers | |
1739 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1740 | * that have been marked dirty. Further, the dirty state can leak into | |
1741 | * the file interior if the file is extended, resulting in all sorts of | |
1742 | * bad things happening as the state does not match the underlying data. | |
1743 | * | |
1744 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1745 | * this only exist because of bufferheads and how the generic code manages them. | |
1746 | */ | |
1747 | STATIC int | |
1748 | xfs_vm_set_page_dirty( | |
1749 | struct page *page) | |
1750 | { | |
1751 | struct address_space *mapping = page->mapping; | |
1752 | struct inode *inode = mapping->host; | |
1753 | loff_t end_offset; | |
1754 | loff_t offset; | |
1755 | int newly_dirty; | |
c4843a75 | 1756 | struct mem_cgroup *memcg; |
22e757a4 DC |
1757 | |
1758 | if (unlikely(!mapping)) | |
1759 | return !TestSetPageDirty(page); | |
1760 | ||
1761 | end_offset = i_size_read(inode); | |
1762 | offset = page_offset(page); | |
1763 | ||
1764 | spin_lock(&mapping->private_lock); | |
1765 | if (page_has_buffers(page)) { | |
1766 | struct buffer_head *head = page_buffers(page); | |
1767 | struct buffer_head *bh = head; | |
1768 | ||
1769 | do { | |
1770 | if (offset < end_offset) | |
1771 | set_buffer_dirty(bh); | |
1772 | bh = bh->b_this_page; | |
1773 | offset += 1 << inode->i_blkbits; | |
1774 | } while (bh != head); | |
1775 | } | |
c4843a75 GT |
1776 | /* |
1777 | * Use mem_group_begin_page_stat() to keep PageDirty synchronized with | |
1778 | * per-memcg dirty page counters. | |
1779 | */ | |
1780 | memcg = mem_cgroup_begin_page_stat(page); | |
22e757a4 DC |
1781 | newly_dirty = !TestSetPageDirty(page); |
1782 | spin_unlock(&mapping->private_lock); | |
1783 | ||
1784 | if (newly_dirty) { | |
1785 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1786 | unsigned long flags; | |
1787 | ||
1788 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1789 | if (page->mapping) { /* Race with truncate? */ | |
1790 | WARN_ON_ONCE(!PageUptodate(page)); | |
c4843a75 | 1791 | account_page_dirtied(page, mapping, memcg); |
22e757a4 DC |
1792 | radix_tree_tag_set(&mapping->page_tree, |
1793 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1794 | } | |
1795 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1796 | } |
c4843a75 GT |
1797 | mem_cgroup_end_page_stat(memcg); |
1798 | if (newly_dirty) | |
1799 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1800 | return newly_dirty; |
1801 | } | |
1802 | ||
f5e54d6e | 1803 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1804 | .readpage = xfs_vm_readpage, |
1805 | .readpages = xfs_vm_readpages, | |
1806 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1807 | .writepages = xfs_vm_writepages, |
22e757a4 | 1808 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1809 | .releasepage = xfs_vm_releasepage, |
1810 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1811 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1812 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1813 | .bmap = xfs_vm_bmap, |
1814 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1815 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1816 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1817 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1818 | }; |