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 | ||
0b1b213f | 43 | void |
f51623b2 NS |
44 | xfs_count_page_state( |
45 | struct page *page, | |
46 | int *delalloc, | |
f51623b2 NS |
47 | int *unwritten) |
48 | { | |
49 | struct buffer_head *bh, *head; | |
50 | ||
20cb52eb | 51 | *delalloc = *unwritten = 0; |
f51623b2 NS |
52 | |
53 | bh = head = page_buffers(page); | |
54 | do { | |
20cb52eb | 55 | if (buffer_unwritten(bh)) |
f51623b2 NS |
56 | (*unwritten) = 1; |
57 | else if (buffer_delay(bh)) | |
58 | (*delalloc) = 1; | |
59 | } while ((bh = bh->b_this_page) != head); | |
60 | } | |
61 | ||
6214ed44 CH |
62 | STATIC struct block_device * |
63 | xfs_find_bdev_for_inode( | |
046f1685 | 64 | struct inode *inode) |
6214ed44 | 65 | { |
046f1685 | 66 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
67 | struct xfs_mount *mp = ip->i_mount; |
68 | ||
71ddabb9 | 69 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
70 | return mp->m_rtdev_targp->bt_bdev; |
71 | else | |
72 | return mp->m_ddev_targp->bt_bdev; | |
73 | } | |
74 | ||
f6d6d4fc CH |
75 | /* |
76 | * We're now finished for good with this ioend structure. | |
77 | * Update the page state via the associated buffer_heads, | |
78 | * release holds on the inode and bio, and finally free | |
79 | * up memory. Do not use the ioend after this. | |
80 | */ | |
0829c360 CH |
81 | STATIC void |
82 | xfs_destroy_ioend( | |
83 | xfs_ioend_t *ioend) | |
84 | { | |
f6d6d4fc CH |
85 | struct buffer_head *bh, *next; |
86 | ||
87 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
88 | next = bh->b_private; | |
7d04a335 | 89 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 90 | } |
583fa586 | 91 | |
0829c360 CH |
92 | mempool_free(ioend, xfs_ioend_pool); |
93 | } | |
94 | ||
fc0063c4 CH |
95 | /* |
96 | * Fast and loose check if this write could update the on-disk inode size. | |
97 | */ | |
98 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
99 | { | |
100 | return ioend->io_offset + ioend->io_size > | |
101 | XFS_I(ioend->io_inode)->i_d.di_size; | |
102 | } | |
103 | ||
281627df CH |
104 | STATIC int |
105 | xfs_setfilesize_trans_alloc( | |
106 | struct xfs_ioend *ioend) | |
107 | { | |
108 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
109 | struct xfs_trans *tp; | |
110 | int error; | |
111 | ||
112 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
113 | ||
3d3c8b52 | 114 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
281627df | 115 | if (error) { |
4906e215 | 116 | xfs_trans_cancel(tp); |
281627df CH |
117 | return error; |
118 | } | |
119 | ||
120 | ioend->io_append_trans = tp; | |
121 | ||
d9457dc0 | 122 | /* |
437a255a | 123 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
124 | * we released it. |
125 | */ | |
bee9182d | 126 | __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS); |
281627df CH |
127 | /* |
128 | * We hand off the transaction to the completion thread now, so | |
129 | * clear the flag here. | |
130 | */ | |
131 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
132 | return 0; | |
133 | } | |
134 | ||
ba87ea69 | 135 | /* |
2813d682 | 136 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 137 | */ |
281627df | 138 | STATIC int |
ba87ea69 | 139 | xfs_setfilesize( |
2ba66237 CH |
140 | struct xfs_inode *ip, |
141 | struct xfs_trans *tp, | |
142 | xfs_off_t offset, | |
143 | size_t size) | |
ba87ea69 | 144 | { |
ba87ea69 | 145 | xfs_fsize_t isize; |
ba87ea69 | 146 | |
aa6bf01d | 147 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 148 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
149 | if (!isize) { |
150 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
4906e215 | 151 | xfs_trans_cancel(tp); |
281627df | 152 | return 0; |
ba87ea69 LM |
153 | } |
154 | ||
2ba66237 | 155 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
156 | |
157 | ip->i_d.di_size = isize; | |
158 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
159 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
160 | ||
70393313 | 161 | return xfs_trans_commit(tp); |
77d7a0c2 DC |
162 | } |
163 | ||
2ba66237 CH |
164 | STATIC int |
165 | xfs_setfilesize_ioend( | |
166 | struct xfs_ioend *ioend) | |
167 | { | |
168 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
169 | struct xfs_trans *tp = ioend->io_append_trans; | |
170 | ||
171 | /* | |
172 | * The transaction may have been allocated in the I/O submission thread, | |
173 | * thus we need to mark ourselves as being in a transaction manually. | |
174 | * Similarly for freeze protection. | |
175 | */ | |
176 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
bee9182d | 177 | __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS); |
2ba66237 | 178 | |
5cb13dcd Z |
179 | /* we abort the update if there was an IO error */ |
180 | if (ioend->io_error) { | |
181 | xfs_trans_cancel(tp); | |
182 | return ioend->io_error; | |
183 | } | |
184 | ||
2ba66237 CH |
185 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); |
186 | } | |
187 | ||
77d7a0c2 | 188 | /* |
209fb87a | 189 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
190 | * |
191 | * If there is no work to do we might as well call it a day and free the | |
192 | * ioend right now. | |
77d7a0c2 DC |
193 | */ |
194 | STATIC void | |
195 | xfs_finish_ioend( | |
209fb87a | 196 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
197 | { |
198 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
199 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
200 | ||
0d882a36 | 201 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
aa6bf01d | 202 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
2ba66237 | 203 | else if (ioend->io_append_trans) |
aa6bf01d | 204 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
205 | else |
206 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 207 | } |
ba87ea69 LM |
208 | } |
209 | ||
0829c360 | 210 | /* |
5ec4fabb | 211 | * IO write completion. |
f6d6d4fc CH |
212 | */ |
213 | STATIC void | |
5ec4fabb | 214 | xfs_end_io( |
77d7a0c2 | 215 | struct work_struct *work) |
0829c360 | 216 | { |
77d7a0c2 DC |
217 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
218 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 219 | int error = 0; |
ba87ea69 | 220 | |
04f658ee | 221 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 222 | ioend->io_error = -EIO; |
04f658ee CH |
223 | goto done; |
224 | } | |
04f658ee | 225 | |
5ec4fabb CH |
226 | /* |
227 | * For unwritten extents we need to issue transactions to convert a | |
228 | * range to normal written extens after the data I/O has finished. | |
5cb13dcd Z |
229 | * Detecting and handling completion IO errors is done individually |
230 | * for each case as different cleanup operations need to be performed | |
231 | * on error. | |
5ec4fabb | 232 | */ |
0d882a36 | 233 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
5cb13dcd Z |
234 | if (ioend->io_error) |
235 | goto done; | |
437a255a DC |
236 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
237 | ioend->io_size); | |
281627df | 238 | } else if (ioend->io_append_trans) { |
2ba66237 | 239 | error = xfs_setfilesize_ioend(ioend); |
84803fb7 | 240 | } else { |
281627df | 241 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 242 | } |
ba87ea69 | 243 | |
04f658ee | 244 | done: |
437a255a | 245 | if (error) |
2451337d | 246 | ioend->io_error = error; |
aa6bf01d | 247 | xfs_destroy_ioend(ioend); |
c626d174 DC |
248 | } |
249 | ||
0829c360 CH |
250 | /* |
251 | * Allocate and initialise an IO completion structure. | |
252 | * We need to track unwritten extent write completion here initially. | |
253 | * We'll need to extend this for updating the ondisk inode size later | |
254 | * (vs. incore size). | |
255 | */ | |
256 | STATIC xfs_ioend_t * | |
257 | xfs_alloc_ioend( | |
f6d6d4fc CH |
258 | struct inode *inode, |
259 | unsigned int type) | |
0829c360 CH |
260 | { |
261 | xfs_ioend_t *ioend; | |
262 | ||
263 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
264 | ||
265 | /* | |
266 | * Set the count to 1 initially, which will prevent an I/O | |
267 | * completion callback from happening before we have started | |
268 | * all the I/O from calling the completion routine too early. | |
269 | */ | |
270 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 271 | ioend->io_error = 0; |
f6d6d4fc CH |
272 | ioend->io_list = NULL; |
273 | ioend->io_type = type; | |
b677c210 | 274 | ioend->io_inode = inode; |
c1a073bd | 275 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 276 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
277 | ioend->io_offset = 0; |
278 | ioend->io_size = 0; | |
281627df | 279 | ioend->io_append_trans = NULL; |
0829c360 | 280 | |
5ec4fabb | 281 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
282 | return ioend; |
283 | } | |
284 | ||
1da177e4 LT |
285 | STATIC int |
286 | xfs_map_blocks( | |
287 | struct inode *inode, | |
288 | loff_t offset, | |
207d0416 | 289 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
290 | int type, |
291 | int nonblocking) | |
1da177e4 | 292 | { |
a206c817 CH |
293 | struct xfs_inode *ip = XFS_I(inode); |
294 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 295 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
296 | xfs_fileoff_t offset_fsb, end_fsb; |
297 | int error = 0; | |
a206c817 CH |
298 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
299 | int nimaps = 1; | |
300 | ||
301 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 302 | return -EIO; |
a206c817 | 303 | |
0d882a36 | 304 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 305 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
306 | |
307 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
308 | if (nonblocking) | |
b474c7ae | 309 | return -EAGAIN; |
8ff2957d | 310 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
a206c817 CH |
311 | } |
312 | ||
8ff2957d CH |
313 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
314 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 315 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 316 | |
d2c28191 DC |
317 | if (offset + count > mp->m_super->s_maxbytes) |
318 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
319 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
320 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
321 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
322 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 323 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 324 | |
8ff2957d | 325 | if (error) |
2451337d | 326 | return error; |
a206c817 | 327 | |
0d882a36 | 328 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 329 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 330 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
331 | if (!error) |
332 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 333 | return error; |
a206c817 CH |
334 | } |
335 | ||
8ff2957d | 336 | #ifdef DEBUG |
0d882a36 | 337 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
338 | ASSERT(nimaps); |
339 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
340 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
341 | } | |
342 | #endif | |
343 | if (nimaps) | |
344 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
345 | return 0; | |
1da177e4 LT |
346 | } |
347 | ||
b8f82a4a | 348 | STATIC int |
558e6891 | 349 | xfs_imap_valid( |
8699bb0a | 350 | struct inode *inode, |
207d0416 | 351 | struct xfs_bmbt_irec *imap, |
558e6891 | 352 | xfs_off_t offset) |
1da177e4 | 353 | { |
558e6891 | 354 | offset >>= inode->i_blkbits; |
8699bb0a | 355 | |
558e6891 CH |
356 | return offset >= imap->br_startoff && |
357 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
358 | } |
359 | ||
f6d6d4fc CH |
360 | /* |
361 | * BIO completion handler for buffered IO. | |
362 | */ | |
782e3b3b | 363 | STATIC void |
f6d6d4fc | 364 | xfs_end_bio( |
4246a0b6 | 365 | struct bio *bio) |
f6d6d4fc CH |
366 | { |
367 | xfs_ioend_t *ioend = bio->bi_private; | |
368 | ||
77a78806 LT |
369 | if (!ioend->io_error) |
370 | ioend->io_error = bio->bi_error; | |
f6d6d4fc CH |
371 | |
372 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
373 | bio->bi_private = NULL; |
374 | bio->bi_end_io = NULL; | |
f6d6d4fc | 375 | bio_put(bio); |
7d04a335 | 376 | |
209fb87a | 377 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
378 | } |
379 | ||
380 | STATIC void | |
381 | xfs_submit_ioend_bio( | |
06342cf8 CH |
382 | struct writeback_control *wbc, |
383 | xfs_ioend_t *ioend, | |
384 | struct bio *bio) | |
f6d6d4fc CH |
385 | { |
386 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
387 | bio->bi_private = ioend; |
388 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 389 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
390 | } |
391 | ||
392 | STATIC struct bio * | |
393 | xfs_alloc_ioend_bio( | |
394 | struct buffer_head *bh) | |
395 | { | |
b54ffb73 | 396 | struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES); |
f6d6d4fc CH |
397 | |
398 | ASSERT(bio->bi_private == NULL); | |
4f024f37 | 399 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
f6d6d4fc | 400 | bio->bi_bdev = bh->b_bdev; |
f6d6d4fc CH |
401 | return bio; |
402 | } | |
403 | ||
404 | STATIC void | |
405 | xfs_start_buffer_writeback( | |
406 | struct buffer_head *bh) | |
407 | { | |
408 | ASSERT(buffer_mapped(bh)); | |
409 | ASSERT(buffer_locked(bh)); | |
410 | ASSERT(!buffer_delay(bh)); | |
411 | ASSERT(!buffer_unwritten(bh)); | |
412 | ||
413 | mark_buffer_async_write(bh); | |
414 | set_buffer_uptodate(bh); | |
415 | clear_buffer_dirty(bh); | |
416 | } | |
417 | ||
418 | STATIC void | |
419 | xfs_start_page_writeback( | |
420 | struct page *page, | |
f6d6d4fc CH |
421 | int clear_dirty, |
422 | int buffers) | |
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); |
0d085a52 | 441 | |
1f7decf6 FW |
442 | /* If no buffers on the page are to be written, finish it here */ |
443 | if (!buffers) | |
f6d6d4fc | 444 | end_page_writeback(page); |
f6d6d4fc CH |
445 | } |
446 | ||
c7c1a7d8 | 447 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
448 | { |
449 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
450 | } | |
451 | ||
452 | /* | |
d88992f6 DC |
453 | * Submit all of the bios for all of the ioends we have saved up, covering the |
454 | * initial writepage page and also any probed pages. | |
455 | * | |
456 | * Because we may have multiple ioends spanning a page, we need to start | |
457 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
458 | * buffers as we got, then we can end up with a page that only has buffers | |
459 | * marked async write and I/O complete on can occur before we mark the other | |
460 | * buffers async write. | |
461 | * | |
462 | * The end result of this is that we trip a bug in end_page_writeback() because | |
463 | * we call it twice for the one page as the code in end_buffer_async_write() | |
464 | * assumes that all buffers on the page are started at the same time. | |
465 | * | |
466 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 467 | * buffer_heads, and then submit them for I/O on the second pass. |
7bf7f352 DC |
468 | * |
469 | * If @fail is non-zero, it means that we have a situation where some part of | |
470 | * the submission process has failed after we have marked paged for writeback | |
471 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
472 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
f6d6d4fc CH |
473 | */ |
474 | STATIC void | |
475 | xfs_submit_ioend( | |
06342cf8 | 476 | struct writeback_control *wbc, |
7bf7f352 DC |
477 | xfs_ioend_t *ioend, |
478 | int fail) | |
f6d6d4fc | 479 | { |
d88992f6 | 480 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
481 | xfs_ioend_t *next; |
482 | struct buffer_head *bh; | |
483 | struct bio *bio; | |
484 | sector_t lastblock = 0; | |
485 | ||
d88992f6 DC |
486 | /* Pass 1 - start writeback */ |
487 | do { | |
488 | next = ioend->io_list; | |
221cb251 | 489 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 490 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
491 | } while ((ioend = next) != NULL); |
492 | ||
493 | /* Pass 2 - submit I/O */ | |
494 | ioend = head; | |
f6d6d4fc CH |
495 | do { |
496 | next = ioend->io_list; | |
497 | bio = NULL; | |
498 | ||
7bf7f352 DC |
499 | /* |
500 | * If we are failing the IO now, just mark the ioend with an | |
501 | * error and finish it. This will run IO completion immediately | |
502 | * as there is only one reference to the ioend at this point in | |
503 | * time. | |
504 | */ | |
505 | if (fail) { | |
2451337d | 506 | ioend->io_error = fail; |
7bf7f352 DC |
507 | xfs_finish_ioend(ioend); |
508 | continue; | |
509 | } | |
510 | ||
f6d6d4fc | 511 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
f6d6d4fc CH |
512 | |
513 | if (!bio) { | |
514 | retry: | |
515 | bio = xfs_alloc_ioend_bio(bh); | |
516 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 517 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
518 | goto retry; |
519 | } | |
520 | ||
c7c1a7d8 | 521 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
06342cf8 | 522 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
523 | goto retry; |
524 | } | |
525 | ||
526 | lastblock = bh->b_blocknr; | |
527 | } | |
528 | if (bio) | |
06342cf8 | 529 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 530 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
531 | } while ((ioend = next) != NULL); |
532 | } | |
533 | ||
534 | /* | |
535 | * Cancel submission of all buffer_heads so far in this endio. | |
536 | * Toss the endio too. Only ever called for the initial page | |
537 | * in a writepage request, so only ever one page. | |
538 | */ | |
539 | STATIC void | |
540 | xfs_cancel_ioend( | |
541 | xfs_ioend_t *ioend) | |
542 | { | |
543 | xfs_ioend_t *next; | |
544 | struct buffer_head *bh, *next_bh; | |
545 | ||
546 | do { | |
547 | next = ioend->io_list; | |
548 | bh = ioend->io_buffer_head; | |
549 | do { | |
550 | next_bh = bh->b_private; | |
551 | clear_buffer_async_write(bh); | |
07d08681 BF |
552 | /* |
553 | * The unwritten flag is cleared when added to the | |
554 | * ioend. We're not submitting for I/O so mark the | |
555 | * buffer unwritten again for next time around. | |
556 | */ | |
557 | if (ioend->io_type == XFS_IO_UNWRITTEN) | |
558 | set_buffer_unwritten(bh); | |
f6d6d4fc CH |
559 | unlock_buffer(bh); |
560 | } while ((bh = next_bh) != NULL); | |
561 | ||
f6d6d4fc CH |
562 | mempool_free(ioend, xfs_ioend_pool); |
563 | } while ((ioend = next) != NULL); | |
564 | } | |
565 | ||
566 | /* | |
567 | * Test to see if we've been building up a completion structure for | |
568 | * earlier buffers -- if so, we try to append to this ioend if we | |
569 | * can, otherwise we finish off any current ioend and start another. | |
570 | * Return true if we've finished the given ioend. | |
571 | */ | |
572 | STATIC void | |
573 | xfs_add_to_ioend( | |
574 | struct inode *inode, | |
575 | struct buffer_head *bh, | |
7336cea8 | 576 | xfs_off_t offset, |
f6d6d4fc CH |
577 | unsigned int type, |
578 | xfs_ioend_t **result, | |
579 | int need_ioend) | |
580 | { | |
581 | xfs_ioend_t *ioend = *result; | |
582 | ||
583 | if (!ioend || need_ioend || type != ioend->io_type) { | |
584 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 585 | |
f6d6d4fc CH |
586 | ioend = xfs_alloc_ioend(inode, type); |
587 | ioend->io_offset = offset; | |
588 | ioend->io_buffer_head = bh; | |
589 | ioend->io_buffer_tail = bh; | |
590 | if (previous) | |
591 | previous->io_list = ioend; | |
592 | *result = ioend; | |
593 | } else { | |
594 | ioend->io_buffer_tail->b_private = bh; | |
595 | ioend->io_buffer_tail = bh; | |
596 | } | |
597 | ||
598 | bh->b_private = NULL; | |
599 | ioend->io_size += bh->b_size; | |
600 | } | |
601 | ||
87cbc49c NS |
602 | STATIC void |
603 | xfs_map_buffer( | |
046f1685 | 604 | struct inode *inode, |
87cbc49c | 605 | struct buffer_head *bh, |
207d0416 | 606 | struct xfs_bmbt_irec *imap, |
046f1685 | 607 | xfs_off_t offset) |
87cbc49c NS |
608 | { |
609 | sector_t bn; | |
8699bb0a | 610 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
611 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
612 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 613 | |
207d0416 CH |
614 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
615 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 616 | |
e513182d | 617 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 618 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 619 | |
046f1685 | 620 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
621 | |
622 | bh->b_blocknr = bn; | |
623 | set_buffer_mapped(bh); | |
624 | } | |
625 | ||
1da177e4 LT |
626 | STATIC void |
627 | xfs_map_at_offset( | |
046f1685 | 628 | struct inode *inode, |
1da177e4 | 629 | struct buffer_head *bh, |
207d0416 | 630 | struct xfs_bmbt_irec *imap, |
046f1685 | 631 | xfs_off_t offset) |
1da177e4 | 632 | { |
207d0416 CH |
633 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
634 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 635 | |
207d0416 | 636 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
637 | set_buffer_mapped(bh); |
638 | clear_buffer_delay(bh); | |
f6d6d4fc | 639 | clear_buffer_unwritten(bh); |
1da177e4 LT |
640 | } |
641 | ||
1da177e4 | 642 | /* |
a49935f2 DC |
643 | * Test if a given page contains at least one buffer of a given @type. |
644 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
645 | * try to find one of the type passed in. If it is not set, then the caller only | |
646 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 647 | */ |
a49935f2 | 648 | STATIC bool |
6ffc4db5 | 649 | xfs_check_page_type( |
10ce4444 | 650 | struct page *page, |
a49935f2 DC |
651 | unsigned int type, |
652 | bool check_all_buffers) | |
1da177e4 | 653 | { |
a49935f2 DC |
654 | struct buffer_head *bh; |
655 | struct buffer_head *head; | |
1da177e4 | 656 | |
a49935f2 DC |
657 | if (PageWriteback(page)) |
658 | return false; | |
659 | if (!page->mapping) | |
660 | return false; | |
661 | if (!page_has_buffers(page)) | |
662 | return false; | |
1da177e4 | 663 | |
a49935f2 DC |
664 | bh = head = page_buffers(page); |
665 | do { | |
666 | if (buffer_unwritten(bh)) { | |
667 | if (type == XFS_IO_UNWRITTEN) | |
668 | return true; | |
669 | } else if (buffer_delay(bh)) { | |
805eeb8e | 670 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
671 | return true; |
672 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 673 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
674 | return true; |
675 | } | |
1da177e4 | 676 | |
a49935f2 DC |
677 | /* If we are only checking the first buffer, we are done now. */ |
678 | if (!check_all_buffers) | |
679 | break; | |
680 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 681 | |
a49935f2 | 682 | return false; |
1da177e4 LT |
683 | } |
684 | ||
1da177e4 LT |
685 | /* |
686 | * Allocate & map buffers for page given the extent map. Write it out. | |
687 | * except for the original page of a writepage, this is called on | |
688 | * delalloc/unwritten pages only, for the original page it is possible | |
689 | * that the page has no mapping at all. | |
690 | */ | |
f6d6d4fc | 691 | STATIC int |
1da177e4 LT |
692 | xfs_convert_page( |
693 | struct inode *inode, | |
694 | struct page *page, | |
10ce4444 | 695 | loff_t tindex, |
207d0416 | 696 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 697 | xfs_ioend_t **ioendp, |
2fa24f92 | 698 | struct writeback_control *wbc) |
1da177e4 | 699 | { |
f6d6d4fc | 700 | struct buffer_head *bh, *head; |
9260dc6b CH |
701 | xfs_off_t end_offset; |
702 | unsigned long p_offset; | |
f6d6d4fc | 703 | unsigned int type; |
24e17b5f | 704 | int len, page_dirty; |
f6d6d4fc | 705 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 706 | xfs_off_t offset = page_offset(page); |
1da177e4 | 707 | |
10ce4444 CH |
708 | if (page->index != tindex) |
709 | goto fail; | |
529ae9aa | 710 | if (!trylock_page(page)) |
10ce4444 CH |
711 | goto fail; |
712 | if (PageWriteback(page)) | |
713 | goto fail_unlock_page; | |
714 | if (page->mapping != inode->i_mapping) | |
715 | goto fail_unlock_page; | |
a49935f2 | 716 | if (!xfs_check_page_type(page, (*ioendp)->io_type, false)) |
10ce4444 CH |
717 | goto fail_unlock_page; |
718 | ||
24e17b5f NS |
719 | /* |
720 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 721 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
722 | * |
723 | * Derivation: | |
724 | * | |
725 | * End offset is the highest offset that this page should represent. | |
726 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
727 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
728 | * hence give us the correct page_dirty count. On any other page, | |
729 | * it will be zero and in that case we need page_dirty to be the | |
730 | * count of buffers on the page. | |
24e17b5f | 731 | */ |
9260dc6b CH |
732 | end_offset = min_t(unsigned long long, |
733 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
734 | i_size_read(inode)); | |
735 | ||
480d7467 DC |
736 | /* |
737 | * If the current map does not span the entire page we are about to try | |
738 | * to write, then give up. The only way we can write a page that spans | |
739 | * multiple mappings in a single writeback iteration is via the | |
740 | * xfs_vm_writepage() function. Data integrity writeback requires the | |
741 | * entire page to be written in a single attempt, otherwise the part of | |
742 | * the page we don't write here doesn't get written as part of the data | |
743 | * integrity sync. | |
744 | * | |
745 | * For normal writeback, we also don't attempt to write partial pages | |
746 | * here as it simply means that write_cache_pages() will see it under | |
747 | * writeback and ignore the page until some point in the future, at | |
748 | * which time this will be the only page in the file that needs | |
749 | * writeback. Hence for more optimal IO patterns, we should always | |
750 | * avoid partial page writeback due to multiple mappings on a page here. | |
751 | */ | |
752 | if (!xfs_imap_valid(inode, imap, end_offset)) | |
753 | goto fail_unlock_page; | |
754 | ||
24e17b5f | 755 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
756 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
757 | PAGE_CACHE_SIZE); | |
758 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
759 | page_dirty = p_offset / len; | |
24e17b5f | 760 | |
a49935f2 DC |
761 | /* |
762 | * The moment we find a buffer that doesn't match our current type | |
763 | * specification or can't be written, abort the loop and start | |
764 | * writeback. As per the above xfs_imap_valid() check, only | |
765 | * xfs_vm_writepage() can handle partial page writeback fully - we are | |
766 | * limited here to the buffers that are contiguous with the current | |
767 | * ioend, and hence a buffer we can't write breaks that contiguity and | |
768 | * we have to defer the rest of the IO to xfs_vm_writepage(). | |
769 | */ | |
1da177e4 LT |
770 | bh = head = page_buffers(page); |
771 | do { | |
9260dc6b | 772 | if (offset >= end_offset) |
1da177e4 | 773 | break; |
f6d6d4fc CH |
774 | if (!buffer_uptodate(bh)) |
775 | uptodate = 0; | |
776 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
777 | done = 1; | |
a49935f2 | 778 | break; |
f6d6d4fc CH |
779 | } |
780 | ||
2fa24f92 CH |
781 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
782 | buffer_mapped(bh)) { | |
9260dc6b | 783 | if (buffer_unwritten(bh)) |
0d882a36 | 784 | type = XFS_IO_UNWRITTEN; |
2fa24f92 | 785 | else if (buffer_delay(bh)) |
0d882a36 | 786 | type = XFS_IO_DELALLOC; |
2fa24f92 | 787 | else |
0d882a36 | 788 | type = XFS_IO_OVERWRITE; |
9260dc6b | 789 | |
a49935f2 DC |
790 | /* |
791 | * imap should always be valid because of the above | |
792 | * partial page end_offset check on the imap. | |
793 | */ | |
794 | ASSERT(xfs_imap_valid(inode, imap, offset)); | |
9260dc6b | 795 | |
ecff71e6 | 796 | lock_buffer(bh); |
0d882a36 | 797 | if (type != XFS_IO_OVERWRITE) |
2fa24f92 | 798 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
799 | xfs_add_to_ioend(inode, bh, offset, type, |
800 | ioendp, done); | |
801 | ||
9260dc6b CH |
802 | page_dirty--; |
803 | count++; | |
804 | } else { | |
2fa24f92 | 805 | done = 1; |
a49935f2 | 806 | break; |
1da177e4 | 807 | } |
7336cea8 | 808 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 809 | |
f6d6d4fc CH |
810 | if (uptodate && bh == head) |
811 | SetPageUptodate(page); | |
812 | ||
89f3b363 | 813 | if (count) { |
efceab1d DC |
814 | if (--wbc->nr_to_write <= 0 && |
815 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 816 | done = 1; |
1da177e4 | 817 | } |
89f3b363 | 818 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
819 | |
820 | return done; | |
10ce4444 CH |
821 | fail_unlock_page: |
822 | unlock_page(page); | |
823 | fail: | |
824 | return 1; | |
1da177e4 LT |
825 | } |
826 | ||
827 | /* | |
828 | * Convert & write out a cluster of pages in the same extent as defined | |
829 | * by mp and following the start page. | |
830 | */ | |
831 | STATIC void | |
832 | xfs_cluster_write( | |
833 | struct inode *inode, | |
834 | pgoff_t tindex, | |
207d0416 | 835 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 836 | xfs_ioend_t **ioendp, |
1da177e4 | 837 | struct writeback_control *wbc, |
1da177e4 LT |
838 | pgoff_t tlast) |
839 | { | |
10ce4444 CH |
840 | struct pagevec pvec; |
841 | int done = 0, i; | |
1da177e4 | 842 | |
10ce4444 CH |
843 | pagevec_init(&pvec, 0); |
844 | while (!done && tindex <= tlast) { | |
845 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
846 | ||
847 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 848 | break; |
10ce4444 CH |
849 | |
850 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
851 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 852 | imap, ioendp, wbc); |
10ce4444 CH |
853 | if (done) |
854 | break; | |
855 | } | |
856 | ||
857 | pagevec_release(&pvec); | |
858 | cond_resched(); | |
1da177e4 LT |
859 | } |
860 | } | |
861 | ||
3ed3a434 DC |
862 | STATIC void |
863 | xfs_vm_invalidatepage( | |
864 | struct page *page, | |
d47992f8 LC |
865 | unsigned int offset, |
866 | unsigned int length) | |
3ed3a434 | 867 | { |
34097dfe LC |
868 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
869 | length); | |
870 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
871 | } |
872 | ||
873 | /* | |
874 | * If the page has delalloc buffers on it, we need to punch them out before we | |
875 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
876 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
877 | * is done on that same region - the delalloc extent is returned when none is | |
878 | * supposed to be there. | |
879 | * | |
880 | * We prevent this by truncating away the delalloc regions on the page before | |
881 | * invalidating it. Because they are delalloc, we can do this without needing a | |
882 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
883 | * truncation without a transaction as there is no space left for block | |
884 | * reservation (typically why we see a ENOSPC in writeback). | |
885 | * | |
886 | * This is not a performance critical path, so for now just do the punching a | |
887 | * buffer head at a time. | |
888 | */ | |
889 | STATIC void | |
890 | xfs_aops_discard_page( | |
891 | struct page *page) | |
892 | { | |
893 | struct inode *inode = page->mapping->host; | |
894 | struct xfs_inode *ip = XFS_I(inode); | |
895 | struct buffer_head *bh, *head; | |
896 | loff_t offset = page_offset(page); | |
3ed3a434 | 897 | |
a49935f2 | 898 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
899 | goto out_invalidate; |
900 | ||
e8c3753c DC |
901 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
902 | goto out_invalidate; | |
903 | ||
4f10700a | 904 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
905 | "page discard on page %p, inode 0x%llx, offset %llu.", |
906 | page, ip->i_ino, offset); | |
907 | ||
908 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
909 | bh = head = page_buffers(page); | |
910 | do { | |
3ed3a434 | 911 | int error; |
c726de44 | 912 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
913 | |
914 | if (!buffer_delay(bh)) | |
915 | goto next_buffer; | |
916 | ||
c726de44 DC |
917 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
918 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
919 | if (error) { |
920 | /* something screwed, just bail */ | |
e8c3753c | 921 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 922 | xfs_alert(ip->i_mount, |
3ed3a434 | 923 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 924 | } |
3ed3a434 DC |
925 | break; |
926 | } | |
927 | next_buffer: | |
c726de44 | 928 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
929 | |
930 | } while ((bh = bh->b_this_page) != head); | |
931 | ||
932 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
933 | out_invalidate: | |
d47992f8 | 934 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
3ed3a434 DC |
935 | return; |
936 | } | |
937 | ||
1da177e4 | 938 | /* |
89f3b363 CH |
939 | * Write out a dirty page. |
940 | * | |
941 | * For delalloc space on the page we need to allocate space and flush it. | |
942 | * For unwritten space on the page we need to start the conversion to | |
943 | * regular allocated space. | |
89f3b363 | 944 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 945 | */ |
1da177e4 | 946 | STATIC int |
89f3b363 CH |
947 | xfs_vm_writepage( |
948 | struct page *page, | |
949 | struct writeback_control *wbc) | |
1da177e4 | 950 | { |
89f3b363 | 951 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 952 | struct buffer_head *bh, *head; |
207d0416 | 953 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 954 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 955 | loff_t offset; |
f6d6d4fc | 956 | unsigned int type; |
1da177e4 | 957 | __uint64_t end_offset; |
bd1556a1 | 958 | pgoff_t end_index, last_index; |
ed1e7b7e | 959 | ssize_t len; |
a206c817 | 960 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 961 | int count = 0; |
a206c817 | 962 | int nonblocking = 0; |
89f3b363 | 963 | |
34097dfe | 964 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 965 | |
20cb52eb CH |
966 | ASSERT(page_has_buffers(page)); |
967 | ||
89f3b363 CH |
968 | /* |
969 | * Refuse to write the page out if we are called from reclaim context. | |
970 | * | |
d4f7a5cb CH |
971 | * This avoids stack overflows when called from deeply used stacks in |
972 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
973 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 974 | * |
94054fa3 MG |
975 | * This should never happen except in the case of a VM regression so |
976 | * warn about it. | |
89f3b363 | 977 | */ |
94054fa3 MG |
978 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
979 | PF_MEMALLOC)) | |
b5420f23 | 980 | goto redirty; |
1da177e4 | 981 | |
89f3b363 | 982 | /* |
680a647b CH |
983 | * Given that we do not allow direct reclaim to call us, we should |
984 | * never be called while in a filesystem transaction. | |
89f3b363 | 985 | */ |
448011e2 | 986 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 987 | goto redirty; |
89f3b363 | 988 | |
1da177e4 LT |
989 | /* Is this page beyond the end of the file? */ |
990 | offset = i_size_read(inode); | |
991 | end_index = offset >> PAGE_CACHE_SHIFT; | |
992 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
8695d27e JL |
993 | |
994 | /* | |
995 | * The page index is less than the end_index, adjust the end_offset | |
996 | * to the highest offset that this page should represent. | |
997 | * ----------------------------------------------------- | |
998 | * | file mapping | <EOF> | | |
999 | * ----------------------------------------------------- | |
1000 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
1001 | * ^--------------------------------^----------|-------- | |
1002 | * | desired writeback range | see else | | |
1003 | * ---------------------------------^------------------| | |
1004 | */ | |
1005 | if (page->index < end_index) | |
1006 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; | |
1007 | else { | |
1008 | /* | |
1009 | * Check whether the page to write out is beyond or straddles | |
1010 | * i_size or not. | |
1011 | * ------------------------------------------------------- | |
1012 | * | file mapping | <EOF> | | |
1013 | * ------------------------------------------------------- | |
1014 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1015 | * ^--------------------------------^-----------|--------- | |
1016 | * | | Straddles | | |
1017 | * ---------------------------------^-----------|--------| | |
1018 | */ | |
6b7a03f0 CH |
1019 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
1020 | ||
1021 | /* | |
ff9a28f6 JK |
1022 | * Skip the page if it is fully outside i_size, e.g. due to a |
1023 | * truncate operation that is in progress. We must redirty the | |
1024 | * page so that reclaim stops reclaiming it. Otherwise | |
1025 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1026 | * |
1027 | * Note that the end_index is unsigned long, it would overflow | |
1028 | * if the given offset is greater than 16TB on 32-bit system | |
1029 | * and if we do check the page is fully outside i_size or not | |
1030 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1031 | * will be evaluated to 0. Hence this page will be redirtied | |
1032 | * and be written out repeatedly which would result in an | |
1033 | * infinite loop, the user program that perform this operation | |
1034 | * will hang. Instead, we can verify this situation by checking | |
1035 | * if the page to write is totally beyond the i_size or if it's | |
1036 | * offset is just equal to the EOF. | |
6b7a03f0 | 1037 | */ |
8695d27e JL |
1038 | if (page->index > end_index || |
1039 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1040 | goto redirty; |
6b7a03f0 CH |
1041 | |
1042 | /* | |
1043 | * The page straddles i_size. It must be zeroed out on each | |
1044 | * and every writepage invocation because it may be mmapped. | |
1045 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1046 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1047 | * memory is zeroed when mapped, and writes to that region are |
1048 | * not written out to the file." | |
1049 | */ | |
1050 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
8695d27e JL |
1051 | |
1052 | /* Adjust the end_offset to the end of file */ | |
1053 | end_offset = offset; | |
1da177e4 LT |
1054 | } |
1055 | ||
24e17b5f | 1056 | len = 1 << inode->i_blkbits; |
24e17b5f | 1057 | |
24e17b5f | 1058 | bh = head = page_buffers(page); |
f6d6d4fc | 1059 | offset = page_offset(page); |
0d882a36 | 1060 | type = XFS_IO_OVERWRITE; |
a206c817 | 1061 | |
dbcdde3e | 1062 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 1063 | nonblocking = 1; |
f6d6d4fc | 1064 | |
1da177e4 | 1065 | do { |
6ac7248e CH |
1066 | int new_ioend = 0; |
1067 | ||
1da177e4 LT |
1068 | if (offset >= end_offset) |
1069 | break; | |
1070 | if (!buffer_uptodate(bh)) | |
1071 | uptodate = 0; | |
1da177e4 | 1072 | |
3d9b02e3 | 1073 | /* |
ece413f5 CH |
1074 | * set_page_dirty dirties all buffers in a page, independent |
1075 | * of their state. The dirty state however is entirely | |
1076 | * meaningless for holes (!mapped && uptodate), so skip | |
1077 | * buffers covering holes here. | |
3d9b02e3 ES |
1078 | */ |
1079 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
1080 | imap_valid = 0; |
1081 | continue; | |
1082 | } | |
1083 | ||
aeea1b1f | 1084 | if (buffer_unwritten(bh)) { |
0d882a36 AR |
1085 | if (type != XFS_IO_UNWRITTEN) { |
1086 | type = XFS_IO_UNWRITTEN; | |
aeea1b1f | 1087 | imap_valid = 0; |
1da177e4 | 1088 | } |
aeea1b1f | 1089 | } else if (buffer_delay(bh)) { |
0d882a36 AR |
1090 | if (type != XFS_IO_DELALLOC) { |
1091 | type = XFS_IO_DELALLOC; | |
aeea1b1f | 1092 | imap_valid = 0; |
1da177e4 | 1093 | } |
89f3b363 | 1094 | } else if (buffer_uptodate(bh)) { |
0d882a36 AR |
1095 | if (type != XFS_IO_OVERWRITE) { |
1096 | type = XFS_IO_OVERWRITE; | |
85da94c6 CH |
1097 | imap_valid = 0; |
1098 | } | |
aeea1b1f | 1099 | } else { |
7d0fa3ec | 1100 | if (PageUptodate(page)) |
aeea1b1f | 1101 | ASSERT(buffer_mapped(bh)); |
7d0fa3ec AR |
1102 | /* |
1103 | * This buffer is not uptodate and will not be | |
1104 | * written to disk. Ensure that we will put any | |
1105 | * subsequent writeable buffers into a new | |
1106 | * ioend. | |
1107 | */ | |
1108 | imap_valid = 0; | |
aeea1b1f CH |
1109 | continue; |
1110 | } | |
d5cb48aa | 1111 | |
aeea1b1f CH |
1112 | if (imap_valid) |
1113 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1114 | if (!imap_valid) { | |
1115 | /* | |
1116 | * If we didn't have a valid mapping then we need to | |
1117 | * put the new mapping into a separate ioend structure. | |
1118 | * This ensures non-contiguous extents always have | |
1119 | * separate ioends, which is particularly important | |
1120 | * for unwritten extent conversion at I/O completion | |
1121 | * time. | |
1122 | */ | |
1123 | new_ioend = 1; | |
1124 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1125 | nonblocking); | |
1126 | if (err) | |
1127 | goto error; | |
1128 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1129 | } | |
1130 | if (imap_valid) { | |
ecff71e6 | 1131 | lock_buffer(bh); |
0d882a36 | 1132 | if (type != XFS_IO_OVERWRITE) |
aeea1b1f CH |
1133 | xfs_map_at_offset(inode, bh, &imap, offset); |
1134 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1135 | new_ioend); | |
1136 | count++; | |
1da177e4 | 1137 | } |
f6d6d4fc CH |
1138 | |
1139 | if (!iohead) | |
1140 | iohead = ioend; | |
1141 | ||
1142 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1143 | |
1144 | if (uptodate && bh == head) | |
1145 | SetPageUptodate(page); | |
1146 | ||
89f3b363 | 1147 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1148 | |
7bf7f352 DC |
1149 | /* if there is no IO to be submitted for this page, we are done */ |
1150 | if (!ioend) | |
1151 | return 0; | |
1152 | ||
1153 | ASSERT(iohead); | |
1154 | ||
1155 | /* | |
1156 | * Any errors from this point onwards need tobe reported through the IO | |
1157 | * completion path as we have marked the initial page as under writeback | |
1158 | * and unlocked it. | |
1159 | */ | |
1160 | if (imap_valid) { | |
bd1556a1 CH |
1161 | xfs_off_t end_index; |
1162 | ||
1163 | end_index = imap.br_startoff + imap.br_blockcount; | |
1164 | ||
1165 | /* to bytes */ | |
1166 | end_index <<= inode->i_blkbits; | |
1167 | ||
1168 | /* to pages */ | |
1169 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1170 | ||
1171 | /* check against file size */ | |
1172 | if (end_index > last_index) | |
1173 | end_index = last_index; | |
8699bb0a | 1174 | |
207d0416 | 1175 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1176 | wbc, end_index); |
1da177e4 LT |
1177 | } |
1178 | ||
281627df | 1179 | |
7bf7f352 DC |
1180 | /* |
1181 | * Reserve log space if we might write beyond the on-disk inode size. | |
1182 | */ | |
1183 | err = 0; | |
1184 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) | |
1185 | err = xfs_setfilesize_trans_alloc(ioend); | |
1186 | ||
1187 | xfs_submit_ioend(wbc, iohead, err); | |
f6d6d4fc | 1188 | |
89f3b363 | 1189 | return 0; |
1da177e4 LT |
1190 | |
1191 | error: | |
f6d6d4fc CH |
1192 | if (iohead) |
1193 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1194 | |
b5420f23 CH |
1195 | if (err == -EAGAIN) |
1196 | goto redirty; | |
1197 | ||
20cb52eb | 1198 | xfs_aops_discard_page(page); |
89f3b363 CH |
1199 | ClearPageUptodate(page); |
1200 | unlock_page(page); | |
1da177e4 | 1201 | return err; |
f51623b2 | 1202 | |
b5420f23 | 1203 | redirty: |
f51623b2 NS |
1204 | redirty_page_for_writepage(wbc, page); |
1205 | unlock_page(page); | |
1206 | return 0; | |
f51623b2 NS |
1207 | } |
1208 | ||
7d4fb40a NS |
1209 | STATIC int |
1210 | xfs_vm_writepages( | |
1211 | struct address_space *mapping, | |
1212 | struct writeback_control *wbc) | |
1213 | { | |
b3aea4ed | 1214 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1215 | return generic_writepages(mapping, wbc); |
1216 | } | |
1217 | ||
f51623b2 NS |
1218 | /* |
1219 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1220 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1221 | * have buffer heads in this call. |
1222 | * | |
89f3b363 | 1223 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1224 | */ |
1225 | STATIC int | |
238f4c54 | 1226 | xfs_vm_releasepage( |
f51623b2 NS |
1227 | struct page *page, |
1228 | gfp_t gfp_mask) | |
1229 | { | |
20cb52eb | 1230 | int delalloc, unwritten; |
f51623b2 | 1231 | |
34097dfe | 1232 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1233 | |
20cb52eb | 1234 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1235 | |
448011e2 | 1236 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1237 | return 0; |
448011e2 | 1238 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1239 | return 0; |
1240 | ||
f51623b2 NS |
1241 | return try_to_free_buffers(page); |
1242 | } | |
1243 | ||
a719370b | 1244 | /* |
273dda76 CH |
1245 | * When we map a DIO buffer, we may need to pass flags to |
1246 | * xfs_end_io_direct_write to tell it what kind of write IO we are doing. | |
3e12dbbd DC |
1247 | * |
1248 | * Note that for DIO, an IO to the highest supported file block offset (i.e. | |
1249 | * 2^63 - 1FSB bytes) will result in the offset + count overflowing a signed 64 | |
1250 | * bit variable. Hence if we see this overflow, we have to assume that the IO is | |
1251 | * extending the file size. We won't know for sure until IO completion is run | |
1252 | * and the actual max write offset is communicated to the IO completion | |
1253 | * routine. | |
a719370b DC |
1254 | */ |
1255 | static void | |
1256 | xfs_map_direct( | |
1257 | struct inode *inode, | |
1258 | struct buffer_head *bh_result, | |
1259 | struct xfs_bmbt_irec *imap, | |
273dda76 | 1260 | xfs_off_t offset) |
a719370b | 1261 | { |
273dda76 | 1262 | uintptr_t *flags = (uintptr_t *)&bh_result->b_private; |
d5cc2e3f | 1263 | xfs_off_t size = bh_result->b_size; |
d5cc2e3f | 1264 | |
273dda76 CH |
1265 | trace_xfs_get_blocks_map_direct(XFS_I(inode), offset, size, |
1266 | ISUNWRITTEN(imap) ? XFS_IO_UNWRITTEN : XFS_IO_OVERWRITE, imap); | |
a06c277a | 1267 | |
273dda76 CH |
1268 | if (ISUNWRITTEN(imap)) { |
1269 | *flags |= XFS_DIO_FLAG_UNWRITTEN; | |
1270 | set_buffer_defer_completion(bh_result); | |
1271 | } else if (offset + size > i_size_read(inode) || offset + size < 0) { | |
1272 | *flags |= XFS_DIO_FLAG_APPEND; | |
a06c277a | 1273 | set_buffer_defer_completion(bh_result); |
a719370b DC |
1274 | } |
1275 | } | |
1276 | ||
1fdca9c2 DC |
1277 | /* |
1278 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1279 | * is, so that we can avoid repeated get_blocks calls. | |
1280 | * | |
1281 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1282 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1283 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1284 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1285 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1286 | * to and including EOF, and a second mapping for beyond EOF. | |
1287 | */ | |
1288 | static void | |
1289 | xfs_map_trim_size( | |
1290 | struct inode *inode, | |
1291 | sector_t iblock, | |
1292 | struct buffer_head *bh_result, | |
1293 | struct xfs_bmbt_irec *imap, | |
1294 | xfs_off_t offset, | |
1295 | ssize_t size) | |
1296 | { | |
1297 | xfs_off_t mapping_size; | |
1298 | ||
1299 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1300 | mapping_size <<= inode->i_blkbits; | |
1301 | ||
1302 | ASSERT(mapping_size > 0); | |
1303 | if (mapping_size > size) | |
1304 | mapping_size = size; | |
1305 | if (offset < i_size_read(inode) && | |
1306 | offset + mapping_size >= i_size_read(inode)) { | |
1307 | /* limit mapping to block that spans EOF */ | |
1308 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1309 | 1 << inode->i_blkbits); | |
1310 | } | |
1311 | if (mapping_size > LONG_MAX) | |
1312 | mapping_size = LONG_MAX; | |
1313 | ||
1314 | bh_result->b_size = mapping_size; | |
1315 | } | |
1316 | ||
1da177e4 | 1317 | STATIC int |
c2536668 | 1318 | __xfs_get_blocks( |
1da177e4 LT |
1319 | struct inode *inode, |
1320 | sector_t iblock, | |
1da177e4 LT |
1321 | struct buffer_head *bh_result, |
1322 | int create, | |
3e12dbbd DC |
1323 | bool direct, |
1324 | bool dax_fault) | |
1da177e4 | 1325 | { |
a206c817 CH |
1326 | struct xfs_inode *ip = XFS_I(inode); |
1327 | struct xfs_mount *mp = ip->i_mount; | |
1328 | xfs_fileoff_t offset_fsb, end_fsb; | |
1329 | int error = 0; | |
1330 | int lockmode = 0; | |
207d0416 | 1331 | struct xfs_bmbt_irec imap; |
a206c817 | 1332 | int nimaps = 1; |
fdc7ed75 NS |
1333 | xfs_off_t offset; |
1334 | ssize_t size; | |
207d0416 | 1335 | int new = 0; |
a206c817 CH |
1336 | |
1337 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1338 | return -EIO; |
1da177e4 | 1339 | |
fdc7ed75 | 1340 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1341 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1342 | size = bh_result->b_size; | |
364f358a LM |
1343 | |
1344 | if (!create && direct && offset >= i_size_read(inode)) | |
1345 | return 0; | |
1346 | ||
507630b2 DC |
1347 | /* |
1348 | * Direct I/O is usually done on preallocated files, so try getting | |
1349 | * a block mapping without an exclusive lock first. For buffered | |
1350 | * writes we already have the exclusive iolock anyway, so avoiding | |
1351 | * a lock roundtrip here by taking the ilock exclusive from the | |
1352 | * beginning is a useful micro optimization. | |
1353 | */ | |
1354 | if (create && !direct) { | |
a206c817 CH |
1355 | lockmode = XFS_ILOCK_EXCL; |
1356 | xfs_ilock(ip, lockmode); | |
1357 | } else { | |
309ecac8 | 1358 | lockmode = xfs_ilock_data_map_shared(ip); |
a206c817 | 1359 | } |
f2bde9b8 | 1360 | |
d2c28191 DC |
1361 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1362 | if (offset + size > mp->m_super->s_maxbytes) | |
1363 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1364 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1365 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1366 | ||
5c8ed202 DC |
1367 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1368 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1369 | if (error) |
a206c817 CH |
1370 | goto out_unlock; |
1371 | ||
1ca19157 | 1372 | /* for DAX, we convert unwritten extents directly */ |
a206c817 CH |
1373 | if (create && |
1374 | (!nimaps || | |
1375 | (imap.br_startblock == HOLESTARTBLOCK || | |
1ca19157 DC |
1376 | imap.br_startblock == DELAYSTARTBLOCK) || |
1377 | (IS_DAX(inode) && ISUNWRITTEN(&imap)))) { | |
aff3a9ed | 1378 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 | 1379 | /* |
009c6e87 BF |
1380 | * xfs_iomap_write_direct() expects the shared lock. It |
1381 | * is unlocked on return. | |
507630b2 | 1382 | */ |
009c6e87 BF |
1383 | if (lockmode == XFS_ILOCK_EXCL) |
1384 | xfs_ilock_demote(ip, lockmode); | |
1385 | ||
a206c817 CH |
1386 | error = xfs_iomap_write_direct(ip, offset, size, |
1387 | &imap, nimaps); | |
507630b2 | 1388 | if (error) |
2451337d | 1389 | return error; |
d3bc815a | 1390 | new = 1; |
6b698ede | 1391 | |
a206c817 | 1392 | } else { |
507630b2 DC |
1393 | /* |
1394 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1395 | * we can go on without dropping the lock here. If we |
1396 | * are allocating a new delalloc block, make sure that | |
1397 | * we set the new flag so that we mark the buffer new so | |
1398 | * that we know that it is newly allocated if the write | |
1399 | * fails. | |
507630b2 | 1400 | */ |
d3bc815a DC |
1401 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1402 | new = 1; | |
a206c817 | 1403 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1404 | if (error) |
1405 | goto out_unlock; | |
1406 | ||
1407 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1408 | } |
d5cc2e3f DC |
1409 | trace_xfs_get_blocks_alloc(ip, offset, size, |
1410 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1411 | : XFS_IO_DELALLOC, &imap); | |
a206c817 | 1412 | } else if (nimaps) { |
d5cc2e3f DC |
1413 | trace_xfs_get_blocks_found(ip, offset, size, |
1414 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1415 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1416 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1417 | } else { |
1418 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1419 | goto out_unlock; | |
1420 | } | |
1da177e4 | 1421 | |
1ca19157 DC |
1422 | if (IS_DAX(inode) && create) { |
1423 | ASSERT(!ISUNWRITTEN(&imap)); | |
1424 | /* zeroing is not needed at a higher layer */ | |
1425 | new = 0; | |
1426 | } | |
1427 | ||
1fdca9c2 DC |
1428 | /* trim mapping down to size requested */ |
1429 | if (direct || size > (1 << inode->i_blkbits)) | |
1430 | xfs_map_trim_size(inode, iblock, bh_result, | |
1431 | &imap, offset, size); | |
1432 | ||
a719370b DC |
1433 | /* |
1434 | * For unwritten extents do not report a disk address in the buffered | |
1435 | * read case (treat as if we're reading into a hole). | |
1436 | */ | |
207d0416 | 1437 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1438 | imap.br_startblock != DELAYSTARTBLOCK && |
1439 | (create || !ISUNWRITTEN(&imap))) { | |
1440 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1441 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1442 | set_buffer_unwritten(bh_result); |
a719370b | 1443 | /* direct IO needs special help */ |
273dda76 CH |
1444 | if (create && direct) { |
1445 | if (dax_fault) | |
1446 | ASSERT(!ISUNWRITTEN(&imap)); | |
1447 | else | |
1448 | xfs_map_direct(inode, bh_result, &imap, offset); | |
1449 | } | |
1da177e4 LT |
1450 | } |
1451 | ||
c2536668 NS |
1452 | /* |
1453 | * If this is a realtime file, data may be on a different device. | |
1454 | * to that pointed to from the buffer_head b_bdev currently. | |
1455 | */ | |
046f1685 | 1456 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1457 | |
c2536668 | 1458 | /* |
549054af DC |
1459 | * If we previously allocated a block out beyond eof and we are now |
1460 | * coming back to use it then we will need to flag it as new even if it | |
1461 | * has a disk address. | |
1462 | * | |
1463 | * With sub-block writes into unwritten extents we also need to mark | |
1464 | * the buffer as new so that the unwritten parts of the buffer gets | |
1465 | * correctly zeroed. | |
1da177e4 LT |
1466 | */ |
1467 | if (create && | |
1468 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1469 | (offset >= i_size_read(inode)) || |
207d0416 | 1470 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1471 | set_buffer_new(bh_result); |
1da177e4 | 1472 | |
207d0416 | 1473 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1474 | BUG_ON(direct); |
1475 | if (create) { | |
1476 | set_buffer_uptodate(bh_result); | |
1477 | set_buffer_mapped(bh_result); | |
1478 | set_buffer_delay(bh_result); | |
1479 | } | |
1480 | } | |
1481 | ||
1da177e4 | 1482 | return 0; |
a206c817 CH |
1483 | |
1484 | out_unlock: | |
1485 | xfs_iunlock(ip, lockmode); | |
2451337d | 1486 | return error; |
1da177e4 LT |
1487 | } |
1488 | ||
1489 | int | |
c2536668 | 1490 | xfs_get_blocks( |
1da177e4 LT |
1491 | struct inode *inode, |
1492 | sector_t iblock, | |
1493 | struct buffer_head *bh_result, | |
1494 | int create) | |
1495 | { | |
3e12dbbd | 1496 | return __xfs_get_blocks(inode, iblock, bh_result, create, false, false); |
1da177e4 LT |
1497 | } |
1498 | ||
6b698ede | 1499 | int |
e4c573bb | 1500 | xfs_get_blocks_direct( |
1da177e4 LT |
1501 | struct inode *inode, |
1502 | sector_t iblock, | |
1da177e4 LT |
1503 | struct buffer_head *bh_result, |
1504 | int create) | |
1505 | { | |
3e12dbbd DC |
1506 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, false); |
1507 | } | |
1508 | ||
1509 | int | |
1510 | xfs_get_blocks_dax_fault( | |
1511 | struct inode *inode, | |
1512 | sector_t iblock, | |
1513 | struct buffer_head *bh_result, | |
1514 | int create) | |
1515 | { | |
1516 | return __xfs_get_blocks(inode, iblock, bh_result, create, true, true); | |
1da177e4 LT |
1517 | } |
1518 | ||
273dda76 CH |
1519 | /* |
1520 | * Complete a direct I/O write request. | |
1521 | * | |
1522 | * xfs_map_direct passes us some flags in the private data to tell us what to | |
1523 | * do. If no flags are set, then the write IO is an overwrite wholly within | |
1524 | * the existing allocated file size and so there is nothing for us to do. | |
1525 | * | |
1526 | * Note that in this case the completion can be called in interrupt context, | |
1527 | * whereas if we have flags set we will always be called in task context | |
1528 | * (i.e. from a workqueue). | |
1529 | */ | |
1530 | STATIC int | |
1531 | xfs_end_io_direct_write( | |
1532 | struct kiocb *iocb, | |
209fb87a | 1533 | loff_t offset, |
273dda76 CH |
1534 | ssize_t size, |
1535 | void *private) | |
f0973863 | 1536 | { |
273dda76 CH |
1537 | struct inode *inode = file_inode(iocb->ki_filp); |
1538 | struct xfs_inode *ip = XFS_I(inode); | |
1539 | struct xfs_mount *mp = ip->i_mount; | |
1540 | uintptr_t flags = (uintptr_t)private; | |
1541 | int error = 0; | |
a06c277a | 1542 | |
273dda76 | 1543 | trace_xfs_end_io_direct_write(ip, offset, size); |
f0973863 | 1544 | |
273dda76 CH |
1545 | if (XFS_FORCED_SHUTDOWN(mp)) |
1546 | return -EIO; | |
d5cc2e3f | 1547 | |
273dda76 CH |
1548 | if (size <= 0) |
1549 | return size; | |
f0973863 | 1550 | |
2813d682 | 1551 | /* |
273dda76 | 1552 | * The flags tell us whether we are doing unwritten extent conversions |
6dfa1b67 DC |
1553 | * or an append transaction that updates the on-disk file size. These |
1554 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1555 | * to update the VFS inode size. |
273dda76 CH |
1556 | */ |
1557 | if (flags == 0) { | |
1558 | ASSERT(offset + size <= i_size_read(inode)); | |
1559 | return 0; | |
1560 | } | |
1561 | ||
1562 | /* | |
6dfa1b67 | 1563 | * We need to update the in-core inode size here so that we don't end up |
a06c277a DC |
1564 | * with the on-disk inode size being outside the in-core inode size. We |
1565 | * have no other method of updating EOF for AIO, so always do it here | |
1566 | * if necessary. | |
b9d59846 DC |
1567 | * |
1568 | * We need to lock the test/set EOF update as we can be racing with | |
1569 | * other IO completions here to update the EOF. Failing to serialise | |
1570 | * here can result in EOF moving backwards and Bad Things Happen when | |
1571 | * that occurs. | |
2813d682 | 1572 | */ |
273dda76 | 1573 | spin_lock(&ip->i_flags_lock); |
2ba66237 CH |
1574 | if (offset + size > i_size_read(inode)) |
1575 | i_size_write(inode, offset + size); | |
273dda76 | 1576 | spin_unlock(&ip->i_flags_lock); |
2813d682 | 1577 | |
273dda76 CH |
1578 | if (flags & XFS_DIO_FLAG_UNWRITTEN) { |
1579 | trace_xfs_end_io_direct_write_unwritten(ip, offset, size); | |
209fb87a | 1580 | |
273dda76 CH |
1581 | error = xfs_iomap_write_unwritten(ip, offset, size); |
1582 | } else if (flags & XFS_DIO_FLAG_APPEND) { | |
1583 | struct xfs_trans *tp; | |
f0973863 | 1584 | |
273dda76 | 1585 | trace_xfs_end_io_direct_write_append(ip, offset, size); |
6b698ede | 1586 | |
273dda76 CH |
1587 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); |
1588 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); | |
1589 | if (error) { | |
1590 | xfs_trans_cancel(tp); | |
1591 | return error; | |
1592 | } | |
1593 | error = xfs_setfilesize(ip, tp, offset, size); | |
6b698ede DC |
1594 | } |
1595 | ||
273dda76 | 1596 | return error; |
6b698ede DC |
1597 | } |
1598 | ||
6e1ba0bc DC |
1599 | static inline ssize_t |
1600 | xfs_vm_do_dio( | |
1601 | struct inode *inode, | |
1602 | struct kiocb *iocb, | |
1603 | struct iov_iter *iter, | |
1604 | loff_t offset, | |
187372a3 | 1605 | dio_iodone_t endio, |
6e1ba0bc DC |
1606 | int flags) |
1607 | { | |
1608 | struct block_device *bdev; | |
1609 | ||
1610 | if (IS_DAX(inode)) | |
1611 | return dax_do_io(iocb, inode, iter, offset, | |
1612 | xfs_get_blocks_direct, endio, 0); | |
1613 | ||
1614 | bdev = xfs_find_bdev_for_inode(inode); | |
1615 | return __blockdev_direct_IO(iocb, inode, bdev, iter, offset, | |
1616 | xfs_get_blocks_direct, endio, NULL, flags); | |
1617 | } | |
1618 | ||
1da177e4 | 1619 | STATIC ssize_t |
e4c573bb | 1620 | xfs_vm_direct_IO( |
1da177e4 | 1621 | struct kiocb *iocb, |
d8d3d94b AV |
1622 | struct iov_iter *iter, |
1623 | loff_t offset) | |
1da177e4 | 1624 | { |
209fb87a | 1625 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
209fb87a | 1626 | |
6e1ba0bc DC |
1627 | if (iov_iter_rw(iter) == WRITE) |
1628 | return xfs_vm_do_dio(inode, iocb, iter, offset, | |
1629 | xfs_end_io_direct_write, DIO_ASYNC_EXTEND); | |
1630 | return xfs_vm_do_dio(inode, iocb, iter, offset, NULL, 0); | |
1da177e4 LT |
1631 | } |
1632 | ||
d3bc815a DC |
1633 | /* |
1634 | * Punch out the delalloc blocks we have already allocated. | |
1635 | * | |
1636 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1637 | * as the page is still locked at this point. | |
1638 | */ | |
1639 | STATIC void | |
1640 | xfs_vm_kill_delalloc_range( | |
1641 | struct inode *inode, | |
1642 | loff_t start, | |
1643 | loff_t end) | |
1644 | { | |
1645 | struct xfs_inode *ip = XFS_I(inode); | |
1646 | xfs_fileoff_t start_fsb; | |
1647 | xfs_fileoff_t end_fsb; | |
1648 | int error; | |
1649 | ||
1650 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1651 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1652 | if (end_fsb <= start_fsb) | |
1653 | return; | |
1654 | ||
1655 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1656 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1657 | end_fsb - start_fsb); | |
1658 | if (error) { | |
1659 | /* something screwed, just bail */ | |
1660 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1661 | xfs_alert(ip->i_mount, | |
1662 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1663 | ip->i_ino); | |
1664 | } | |
1665 | } | |
1666 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1667 | } | |
1668 | ||
fa9b227e CH |
1669 | STATIC void |
1670 | xfs_vm_write_failed( | |
d3bc815a DC |
1671 | struct inode *inode, |
1672 | struct page *page, | |
1673 | loff_t pos, | |
1674 | unsigned len) | |
fa9b227e | 1675 | { |
58e59854 | 1676 | loff_t block_offset; |
d3bc815a DC |
1677 | loff_t block_start; |
1678 | loff_t block_end; | |
1679 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1680 | loff_t to = from + len; | |
1681 | struct buffer_head *bh, *head; | |
fa9b227e | 1682 | |
58e59854 JL |
1683 | /* |
1684 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1685 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1686 | * platform, the high 32-bit will be masked off if we evaluate the | |
1687 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1688 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1689 | * which could cause the following ASSERT failed in most cases. | |
1690 | * In order to avoid this, we can evaluate the block_offset of the | |
1691 | * start of the page by using shifts rather than masks the mismatch | |
1692 | * problem. | |
1693 | */ | |
1694 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1695 | ||
d3bc815a | 1696 | ASSERT(block_offset + from == pos); |
c726de44 | 1697 | |
d3bc815a DC |
1698 | head = page_buffers(page); |
1699 | block_start = 0; | |
1700 | for (bh = head; bh != head || !block_start; | |
1701 | bh = bh->b_this_page, block_start = block_end, | |
1702 | block_offset += bh->b_size) { | |
1703 | block_end = block_start + bh->b_size; | |
c726de44 | 1704 | |
d3bc815a DC |
1705 | /* skip buffers before the write */ |
1706 | if (block_end <= from) | |
1707 | continue; | |
1708 | ||
1709 | /* if the buffer is after the write, we're done */ | |
1710 | if (block_start >= to) | |
1711 | break; | |
1712 | ||
1713 | if (!buffer_delay(bh)) | |
1714 | continue; | |
1715 | ||
1716 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1717 | continue; | |
1718 | ||
1719 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1720 | block_offset + bh->b_size); | |
4ab9ed57 DC |
1721 | |
1722 | /* | |
1723 | * This buffer does not contain data anymore. make sure anyone | |
1724 | * who finds it knows that for certain. | |
1725 | */ | |
1726 | clear_buffer_delay(bh); | |
1727 | clear_buffer_uptodate(bh); | |
1728 | clear_buffer_mapped(bh); | |
1729 | clear_buffer_new(bh); | |
1730 | clear_buffer_dirty(bh); | |
fa9b227e | 1731 | } |
d3bc815a | 1732 | |
fa9b227e CH |
1733 | } |
1734 | ||
d3bc815a DC |
1735 | /* |
1736 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1737 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1738 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1739 | * the appropriate point. | |
1740 | */ | |
f51623b2 | 1741 | STATIC int |
d79689c7 | 1742 | xfs_vm_write_begin( |
f51623b2 | 1743 | struct file *file, |
d79689c7 NP |
1744 | struct address_space *mapping, |
1745 | loff_t pos, | |
1746 | unsigned len, | |
1747 | unsigned flags, | |
1748 | struct page **pagep, | |
1749 | void **fsdata) | |
f51623b2 | 1750 | { |
d3bc815a DC |
1751 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1752 | struct page *page; | |
1753 | int status; | |
155130a4 | 1754 | |
d3bc815a DC |
1755 | ASSERT(len <= PAGE_CACHE_SIZE); |
1756 | ||
ad22c7a0 | 1757 | page = grab_cache_page_write_begin(mapping, index, flags); |
d3bc815a DC |
1758 | if (!page) |
1759 | return -ENOMEM; | |
1760 | ||
1761 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1762 | if (unlikely(status)) { | |
1763 | struct inode *inode = mapping->host; | |
72ab70a1 | 1764 | size_t isize = i_size_read(inode); |
d3bc815a DC |
1765 | |
1766 | xfs_vm_write_failed(inode, page, pos, len); | |
1767 | unlock_page(page); | |
1768 | ||
72ab70a1 DC |
1769 | /* |
1770 | * If the write is beyond EOF, we only want to kill blocks | |
1771 | * allocated in this write, not blocks that were previously | |
1772 | * written successfully. | |
1773 | */ | |
1774 | if (pos + len > isize) { | |
1775 | ssize_t start = max_t(ssize_t, pos, isize); | |
1776 | ||
1777 | truncate_pagecache_range(inode, start, pos + len); | |
1778 | } | |
d3bc815a DC |
1779 | |
1780 | page_cache_release(page); | |
1781 | page = NULL; | |
1782 | } | |
1783 | ||
1784 | *pagep = page; | |
1785 | return status; | |
fa9b227e CH |
1786 | } |
1787 | ||
d3bc815a | 1788 | /* |
aad3f375 DC |
1789 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
1790 | * this specific write because they will never be written. Previous writes | |
1791 | * beyond EOF where block allocation succeeded do not need to be trashed, so | |
1792 | * only new blocks from this write should be trashed. For blocks within | |
1793 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be | |
1794 | * written with all the other valid data. | |
d3bc815a | 1795 | */ |
fa9b227e CH |
1796 | STATIC int |
1797 | xfs_vm_write_end( | |
1798 | struct file *file, | |
1799 | struct address_space *mapping, | |
1800 | loff_t pos, | |
1801 | unsigned len, | |
1802 | unsigned copied, | |
1803 | struct page *page, | |
1804 | void *fsdata) | |
1805 | { | |
1806 | int ret; | |
155130a4 | 1807 | |
d3bc815a DC |
1808 | ASSERT(len <= PAGE_CACHE_SIZE); |
1809 | ||
fa9b227e | 1810 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1811 | if (unlikely(ret < len)) { |
1812 | struct inode *inode = mapping->host; | |
1813 | size_t isize = i_size_read(inode); | |
1814 | loff_t to = pos + len; | |
1815 | ||
1816 | if (to > isize) { | |
aad3f375 DC |
1817 | /* only kill blocks in this write beyond EOF */ |
1818 | if (pos > isize) | |
1819 | isize = pos; | |
d3bc815a | 1820 | xfs_vm_kill_delalloc_range(inode, isize, to); |
aad3f375 | 1821 | truncate_pagecache_range(inode, isize, to); |
d3bc815a DC |
1822 | } |
1823 | } | |
155130a4 | 1824 | return ret; |
f51623b2 | 1825 | } |
1da177e4 LT |
1826 | |
1827 | STATIC sector_t | |
e4c573bb | 1828 | xfs_vm_bmap( |
1da177e4 LT |
1829 | struct address_space *mapping, |
1830 | sector_t block) | |
1831 | { | |
1832 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1833 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1834 | |
cca28fb8 | 1835 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1836 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1837 | filemap_write_and_wait(mapping); |
126468b1 | 1838 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1839 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1840 | } |
1841 | ||
1842 | STATIC int | |
e4c573bb | 1843 | xfs_vm_readpage( |
1da177e4 LT |
1844 | struct file *unused, |
1845 | struct page *page) | |
1846 | { | |
121e213e | 1847 | trace_xfs_vm_readpage(page->mapping->host, 1); |
c2536668 | 1848 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1849 | } |
1850 | ||
1851 | STATIC int | |
e4c573bb | 1852 | xfs_vm_readpages( |
1da177e4 LT |
1853 | struct file *unused, |
1854 | struct address_space *mapping, | |
1855 | struct list_head *pages, | |
1856 | unsigned nr_pages) | |
1857 | { | |
121e213e | 1858 | trace_xfs_vm_readpages(mapping->host, nr_pages); |
c2536668 | 1859 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1860 | } |
1861 | ||
22e757a4 DC |
1862 | /* |
1863 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1864 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1865 | * dirty, we'll never be able to clean them because we don't write buffers | |
1866 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1867 | * that have been marked dirty. Further, the dirty state can leak into | |
1868 | * the file interior if the file is extended, resulting in all sorts of | |
1869 | * bad things happening as the state does not match the underlying data. | |
1870 | * | |
1871 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1872 | * this only exist because of bufferheads and how the generic code manages them. | |
1873 | */ | |
1874 | STATIC int | |
1875 | xfs_vm_set_page_dirty( | |
1876 | struct page *page) | |
1877 | { | |
1878 | struct address_space *mapping = page->mapping; | |
1879 | struct inode *inode = mapping->host; | |
1880 | loff_t end_offset; | |
1881 | loff_t offset; | |
1882 | int newly_dirty; | |
c4843a75 | 1883 | struct mem_cgroup *memcg; |
22e757a4 DC |
1884 | |
1885 | if (unlikely(!mapping)) | |
1886 | return !TestSetPageDirty(page); | |
1887 | ||
1888 | end_offset = i_size_read(inode); | |
1889 | offset = page_offset(page); | |
1890 | ||
1891 | spin_lock(&mapping->private_lock); | |
1892 | if (page_has_buffers(page)) { | |
1893 | struct buffer_head *head = page_buffers(page); | |
1894 | struct buffer_head *bh = head; | |
1895 | ||
1896 | do { | |
1897 | if (offset < end_offset) | |
1898 | set_buffer_dirty(bh); | |
1899 | bh = bh->b_this_page; | |
1900 | offset += 1 << inode->i_blkbits; | |
1901 | } while (bh != head); | |
1902 | } | |
c4843a75 GT |
1903 | /* |
1904 | * Use mem_group_begin_page_stat() to keep PageDirty synchronized with | |
1905 | * per-memcg dirty page counters. | |
1906 | */ | |
1907 | memcg = mem_cgroup_begin_page_stat(page); | |
22e757a4 DC |
1908 | newly_dirty = !TestSetPageDirty(page); |
1909 | spin_unlock(&mapping->private_lock); | |
1910 | ||
1911 | if (newly_dirty) { | |
1912 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1913 | unsigned long flags; | |
1914 | ||
1915 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1916 | if (page->mapping) { /* Race with truncate? */ | |
1917 | WARN_ON_ONCE(!PageUptodate(page)); | |
c4843a75 | 1918 | account_page_dirtied(page, mapping, memcg); |
22e757a4 DC |
1919 | radix_tree_tag_set(&mapping->page_tree, |
1920 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1921 | } | |
1922 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
22e757a4 | 1923 | } |
c4843a75 GT |
1924 | mem_cgroup_end_page_stat(memcg); |
1925 | if (newly_dirty) | |
1926 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
22e757a4 DC |
1927 | return newly_dirty; |
1928 | } | |
1929 | ||
f5e54d6e | 1930 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1931 | .readpage = xfs_vm_readpage, |
1932 | .readpages = xfs_vm_readpages, | |
1933 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1934 | .writepages = xfs_vm_writepages, |
22e757a4 | 1935 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1936 | .releasepage = xfs_vm_releasepage, |
1937 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1938 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1939 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1940 | .bmap = xfs_vm_bmap, |
1941 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1942 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1943 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1944 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1945 | }; |