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 | ||
0b1b213f | 39 | void |
f51623b2 NS |
40 | xfs_count_page_state( |
41 | struct page *page, | |
42 | int *delalloc, | |
f51623b2 NS |
43 | int *unwritten) |
44 | { | |
45 | struct buffer_head *bh, *head; | |
46 | ||
20cb52eb | 47 | *delalloc = *unwritten = 0; |
f51623b2 NS |
48 | |
49 | bh = head = page_buffers(page); | |
50 | do { | |
20cb52eb | 51 | if (buffer_unwritten(bh)) |
f51623b2 NS |
52 | (*unwritten) = 1; |
53 | else if (buffer_delay(bh)) | |
54 | (*delalloc) = 1; | |
55 | } while ((bh = bh->b_this_page) != head); | |
56 | } | |
57 | ||
6214ed44 CH |
58 | STATIC struct block_device * |
59 | xfs_find_bdev_for_inode( | |
046f1685 | 60 | struct inode *inode) |
6214ed44 | 61 | { |
046f1685 | 62 | struct xfs_inode *ip = XFS_I(inode); |
6214ed44 CH |
63 | struct xfs_mount *mp = ip->i_mount; |
64 | ||
71ddabb9 | 65 | if (XFS_IS_REALTIME_INODE(ip)) |
6214ed44 CH |
66 | return mp->m_rtdev_targp->bt_bdev; |
67 | else | |
68 | return mp->m_ddev_targp->bt_bdev; | |
69 | } | |
70 | ||
f6d6d4fc CH |
71 | /* |
72 | * We're now finished for good with this ioend structure. | |
73 | * Update the page state via the associated buffer_heads, | |
74 | * release holds on the inode and bio, and finally free | |
75 | * up memory. Do not use the ioend after this. | |
76 | */ | |
0829c360 CH |
77 | STATIC void |
78 | xfs_destroy_ioend( | |
79 | xfs_ioend_t *ioend) | |
80 | { | |
f6d6d4fc CH |
81 | struct buffer_head *bh, *next; |
82 | ||
83 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
84 | next = bh->b_private; | |
7d04a335 | 85 | bh->b_end_io(bh, !ioend->io_error); |
f6d6d4fc | 86 | } |
583fa586 | 87 | |
0829c360 CH |
88 | mempool_free(ioend, xfs_ioend_pool); |
89 | } | |
90 | ||
fc0063c4 CH |
91 | /* |
92 | * Fast and loose check if this write could update the on-disk inode size. | |
93 | */ | |
94 | static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend) | |
95 | { | |
96 | return ioend->io_offset + ioend->io_size > | |
97 | XFS_I(ioend->io_inode)->i_d.di_size; | |
98 | } | |
99 | ||
281627df CH |
100 | STATIC int |
101 | xfs_setfilesize_trans_alloc( | |
102 | struct xfs_ioend *ioend) | |
103 | { | |
104 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; | |
105 | struct xfs_trans *tp; | |
106 | int error; | |
107 | ||
108 | tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); | |
109 | ||
3d3c8b52 | 110 | error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); |
281627df CH |
111 | if (error) { |
112 | xfs_trans_cancel(tp, 0); | |
113 | return error; | |
114 | } | |
115 | ||
116 | ioend->io_append_trans = tp; | |
117 | ||
d9457dc0 | 118 | /* |
437a255a | 119 | * We may pass freeze protection with a transaction. So tell lockdep |
d9457dc0 JK |
120 | * we released it. |
121 | */ | |
122 | rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
123 | 1, _THIS_IP_); | |
281627df CH |
124 | /* |
125 | * We hand off the transaction to the completion thread now, so | |
126 | * clear the flag here. | |
127 | */ | |
128 | current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
129 | return 0; | |
130 | } | |
131 | ||
ba87ea69 | 132 | /* |
2813d682 | 133 | * Update on-disk file size now that data has been written to disk. |
ba87ea69 | 134 | */ |
281627df | 135 | STATIC int |
ba87ea69 | 136 | xfs_setfilesize( |
2ba66237 CH |
137 | struct xfs_inode *ip, |
138 | struct xfs_trans *tp, | |
139 | xfs_off_t offset, | |
140 | size_t size) | |
ba87ea69 | 141 | { |
ba87ea69 | 142 | xfs_fsize_t isize; |
ba87ea69 | 143 | |
aa6bf01d | 144 | xfs_ilock(ip, XFS_ILOCK_EXCL); |
2ba66237 | 145 | isize = xfs_new_eof(ip, offset + size); |
281627df CH |
146 | if (!isize) { |
147 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
148 | xfs_trans_cancel(tp, 0); | |
149 | return 0; | |
ba87ea69 LM |
150 | } |
151 | ||
2ba66237 | 152 | trace_xfs_setfilesize(ip, offset, size); |
281627df CH |
153 | |
154 | ip->i_d.di_size = isize; | |
155 | xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); | |
156 | xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); | |
157 | ||
158 | return xfs_trans_commit(tp, 0); | |
77d7a0c2 DC |
159 | } |
160 | ||
2ba66237 CH |
161 | STATIC int |
162 | xfs_setfilesize_ioend( | |
163 | struct xfs_ioend *ioend) | |
164 | { | |
165 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
166 | struct xfs_trans *tp = ioend->io_append_trans; | |
167 | ||
168 | /* | |
169 | * The transaction may have been allocated in the I/O submission thread, | |
170 | * thus we need to mark ourselves as being in a transaction manually. | |
171 | * Similarly for freeze protection. | |
172 | */ | |
173 | current_set_flags_nested(&tp->t_pflags, PF_FSTRANS); | |
174 | rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1], | |
175 | 0, 1, _THIS_IP_); | |
176 | ||
177 | return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size); | |
178 | } | |
179 | ||
77d7a0c2 | 180 | /* |
209fb87a | 181 | * Schedule IO completion handling on the final put of an ioend. |
fc0063c4 CH |
182 | * |
183 | * If there is no work to do we might as well call it a day and free the | |
184 | * ioend right now. | |
77d7a0c2 DC |
185 | */ |
186 | STATIC void | |
187 | xfs_finish_ioend( | |
209fb87a | 188 | struct xfs_ioend *ioend) |
77d7a0c2 DC |
189 | { |
190 | if (atomic_dec_and_test(&ioend->io_remaining)) { | |
aa6bf01d CH |
191 | struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount; |
192 | ||
0d882a36 | 193 | if (ioend->io_type == XFS_IO_UNWRITTEN) |
aa6bf01d | 194 | queue_work(mp->m_unwritten_workqueue, &ioend->io_work); |
2ba66237 | 195 | else if (ioend->io_append_trans) |
aa6bf01d | 196 | queue_work(mp->m_data_workqueue, &ioend->io_work); |
fc0063c4 CH |
197 | else |
198 | xfs_destroy_ioend(ioend); | |
77d7a0c2 | 199 | } |
ba87ea69 LM |
200 | } |
201 | ||
0829c360 | 202 | /* |
5ec4fabb | 203 | * IO write completion. |
f6d6d4fc CH |
204 | */ |
205 | STATIC void | |
5ec4fabb | 206 | xfs_end_io( |
77d7a0c2 | 207 | struct work_struct *work) |
0829c360 | 208 | { |
77d7a0c2 DC |
209 | xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work); |
210 | struct xfs_inode *ip = XFS_I(ioend->io_inode); | |
69418932 | 211 | int error = 0; |
ba87ea69 | 212 | |
04f658ee | 213 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
810627d9 | 214 | ioend->io_error = -EIO; |
04f658ee CH |
215 | goto done; |
216 | } | |
217 | if (ioend->io_error) | |
218 | goto done; | |
219 | ||
5ec4fabb CH |
220 | /* |
221 | * For unwritten extents we need to issue transactions to convert a | |
222 | * range to normal written extens after the data I/O has finished. | |
223 | */ | |
0d882a36 | 224 | if (ioend->io_type == XFS_IO_UNWRITTEN) { |
437a255a DC |
225 | error = xfs_iomap_write_unwritten(ip, ioend->io_offset, |
226 | ioend->io_size); | |
281627df | 227 | } else if (ioend->io_append_trans) { |
2ba66237 | 228 | error = xfs_setfilesize_ioend(ioend); |
84803fb7 | 229 | } else { |
281627df | 230 | ASSERT(!xfs_ioend_is_append(ioend)); |
5ec4fabb | 231 | } |
ba87ea69 | 232 | |
04f658ee | 233 | done: |
437a255a | 234 | if (error) |
2451337d | 235 | ioend->io_error = error; |
aa6bf01d | 236 | xfs_destroy_ioend(ioend); |
c626d174 DC |
237 | } |
238 | ||
0829c360 CH |
239 | /* |
240 | * Allocate and initialise an IO completion structure. | |
241 | * We need to track unwritten extent write completion here initially. | |
242 | * We'll need to extend this for updating the ondisk inode size later | |
243 | * (vs. incore size). | |
244 | */ | |
245 | STATIC xfs_ioend_t * | |
246 | xfs_alloc_ioend( | |
f6d6d4fc CH |
247 | struct inode *inode, |
248 | unsigned int type) | |
0829c360 CH |
249 | { |
250 | xfs_ioend_t *ioend; | |
251 | ||
252 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
253 | ||
254 | /* | |
255 | * Set the count to 1 initially, which will prevent an I/O | |
256 | * completion callback from happening before we have started | |
257 | * all the I/O from calling the completion routine too early. | |
258 | */ | |
259 | atomic_set(&ioend->io_remaining, 1); | |
7d04a335 | 260 | ioend->io_error = 0; |
f6d6d4fc CH |
261 | ioend->io_list = NULL; |
262 | ioend->io_type = type; | |
b677c210 | 263 | ioend->io_inode = inode; |
c1a073bd | 264 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 265 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
266 | ioend->io_offset = 0; |
267 | ioend->io_size = 0; | |
281627df | 268 | ioend->io_append_trans = NULL; |
0829c360 | 269 | |
5ec4fabb | 270 | INIT_WORK(&ioend->io_work, xfs_end_io); |
0829c360 CH |
271 | return ioend; |
272 | } | |
273 | ||
1da177e4 LT |
274 | STATIC int |
275 | xfs_map_blocks( | |
276 | struct inode *inode, | |
277 | loff_t offset, | |
207d0416 | 278 | struct xfs_bmbt_irec *imap, |
a206c817 CH |
279 | int type, |
280 | int nonblocking) | |
1da177e4 | 281 | { |
a206c817 CH |
282 | struct xfs_inode *ip = XFS_I(inode); |
283 | struct xfs_mount *mp = ip->i_mount; | |
ed1e7b7e | 284 | ssize_t count = 1 << inode->i_blkbits; |
a206c817 CH |
285 | xfs_fileoff_t offset_fsb, end_fsb; |
286 | int error = 0; | |
a206c817 CH |
287 | int bmapi_flags = XFS_BMAPI_ENTIRE; |
288 | int nimaps = 1; | |
289 | ||
290 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 291 | return -EIO; |
a206c817 | 292 | |
0d882a36 | 293 | if (type == XFS_IO_UNWRITTEN) |
a206c817 | 294 | bmapi_flags |= XFS_BMAPI_IGSTATE; |
8ff2957d CH |
295 | |
296 | if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) { | |
297 | if (nonblocking) | |
b474c7ae | 298 | return -EAGAIN; |
8ff2957d | 299 | xfs_ilock(ip, XFS_ILOCK_SHARED); |
a206c817 CH |
300 | } |
301 | ||
8ff2957d CH |
302 | ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE || |
303 | (ip->i_df.if_flags & XFS_IFEXTENTS)); | |
d2c28191 | 304 | ASSERT(offset <= mp->m_super->s_maxbytes); |
8ff2957d | 305 | |
d2c28191 DC |
306 | if (offset + count > mp->m_super->s_maxbytes) |
307 | count = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
308 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count); |
309 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
5c8ed202 DC |
310 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
311 | imap, &nimaps, bmapi_flags); | |
8ff2957d | 312 | xfs_iunlock(ip, XFS_ILOCK_SHARED); |
a206c817 | 313 | |
8ff2957d | 314 | if (error) |
2451337d | 315 | return error; |
a206c817 | 316 | |
0d882a36 | 317 | if (type == XFS_IO_DELALLOC && |
8ff2957d | 318 | (!nimaps || isnullstartblock(imap->br_startblock))) { |
0799a3e8 | 319 | error = xfs_iomap_write_allocate(ip, offset, imap); |
a206c817 CH |
320 | if (!error) |
321 | trace_xfs_map_blocks_alloc(ip, offset, count, type, imap); | |
2451337d | 322 | return error; |
a206c817 CH |
323 | } |
324 | ||
8ff2957d | 325 | #ifdef DEBUG |
0d882a36 | 326 | if (type == XFS_IO_UNWRITTEN) { |
8ff2957d CH |
327 | ASSERT(nimaps); |
328 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); | |
329 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
330 | } | |
331 | #endif | |
332 | if (nimaps) | |
333 | trace_xfs_map_blocks_found(ip, offset, count, type, imap); | |
334 | return 0; | |
1da177e4 LT |
335 | } |
336 | ||
b8f82a4a | 337 | STATIC int |
558e6891 | 338 | xfs_imap_valid( |
8699bb0a | 339 | struct inode *inode, |
207d0416 | 340 | struct xfs_bmbt_irec *imap, |
558e6891 | 341 | xfs_off_t offset) |
1da177e4 | 342 | { |
558e6891 | 343 | offset >>= inode->i_blkbits; |
8699bb0a | 344 | |
558e6891 CH |
345 | return offset >= imap->br_startoff && |
346 | offset < imap->br_startoff + imap->br_blockcount; | |
1da177e4 LT |
347 | } |
348 | ||
f6d6d4fc CH |
349 | /* |
350 | * BIO completion handler for buffered IO. | |
351 | */ | |
782e3b3b | 352 | STATIC void |
f6d6d4fc CH |
353 | xfs_end_bio( |
354 | struct bio *bio, | |
f6d6d4fc CH |
355 | int error) |
356 | { | |
357 | xfs_ioend_t *ioend = bio->bi_private; | |
358 | ||
f6d6d4fc | 359 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); |
7d04a335 | 360 | ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error; |
f6d6d4fc CH |
361 | |
362 | /* Toss bio and pass work off to an xfsdatad thread */ | |
f6d6d4fc CH |
363 | bio->bi_private = NULL; |
364 | bio->bi_end_io = NULL; | |
f6d6d4fc | 365 | bio_put(bio); |
7d04a335 | 366 | |
209fb87a | 367 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
368 | } |
369 | ||
370 | STATIC void | |
371 | xfs_submit_ioend_bio( | |
06342cf8 CH |
372 | struct writeback_control *wbc, |
373 | xfs_ioend_t *ioend, | |
374 | struct bio *bio) | |
f6d6d4fc CH |
375 | { |
376 | atomic_inc(&ioend->io_remaining); | |
f6d6d4fc CH |
377 | bio->bi_private = ioend; |
378 | bio->bi_end_io = xfs_end_bio; | |
721a9602 | 379 | submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio); |
f6d6d4fc CH |
380 | } |
381 | ||
382 | STATIC struct bio * | |
383 | xfs_alloc_ioend_bio( | |
384 | struct buffer_head *bh) | |
385 | { | |
f6d6d4fc | 386 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
221cb251 | 387 | struct bio *bio = bio_alloc(GFP_NOIO, nvecs); |
f6d6d4fc CH |
388 | |
389 | ASSERT(bio->bi_private == NULL); | |
4f024f37 | 390 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
f6d6d4fc | 391 | bio->bi_bdev = bh->b_bdev; |
f6d6d4fc CH |
392 | return bio; |
393 | } | |
394 | ||
395 | STATIC void | |
396 | xfs_start_buffer_writeback( | |
397 | struct buffer_head *bh) | |
398 | { | |
399 | ASSERT(buffer_mapped(bh)); | |
400 | ASSERT(buffer_locked(bh)); | |
401 | ASSERT(!buffer_delay(bh)); | |
402 | ASSERT(!buffer_unwritten(bh)); | |
403 | ||
404 | mark_buffer_async_write(bh); | |
405 | set_buffer_uptodate(bh); | |
406 | clear_buffer_dirty(bh); | |
407 | } | |
408 | ||
409 | STATIC void | |
410 | xfs_start_page_writeback( | |
411 | struct page *page, | |
f6d6d4fc CH |
412 | int clear_dirty, |
413 | int buffers) | |
414 | { | |
415 | ASSERT(PageLocked(page)); | |
416 | ASSERT(!PageWriteback(page)); | |
0d085a52 DC |
417 | |
418 | /* | |
419 | * if the page was not fully cleaned, we need to ensure that the higher | |
420 | * layers come back to it correctly. That means we need to keep the page | |
421 | * dirty, and for WB_SYNC_ALL writeback we need to ensure the | |
422 | * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to | |
423 | * write this page in this writeback sweep will be made. | |
424 | */ | |
425 | if (clear_dirty) { | |
92132021 | 426 | clear_page_dirty_for_io(page); |
0d085a52 DC |
427 | set_page_writeback(page); |
428 | } else | |
429 | set_page_writeback_keepwrite(page); | |
430 | ||
f6d6d4fc | 431 | unlock_page(page); |
0d085a52 | 432 | |
1f7decf6 FW |
433 | /* If no buffers on the page are to be written, finish it here */ |
434 | if (!buffers) | |
f6d6d4fc | 435 | end_page_writeback(page); |
f6d6d4fc CH |
436 | } |
437 | ||
c7c1a7d8 | 438 | static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh) |
f6d6d4fc CH |
439 | { |
440 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
441 | } | |
442 | ||
443 | /* | |
d88992f6 DC |
444 | * Submit all of the bios for all of the ioends we have saved up, covering the |
445 | * initial writepage page and also any probed pages. | |
446 | * | |
447 | * Because we may have multiple ioends spanning a page, we need to start | |
448 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
449 | * buffers as we got, then we can end up with a page that only has buffers | |
450 | * marked async write and I/O complete on can occur before we mark the other | |
451 | * buffers async write. | |
452 | * | |
453 | * The end result of this is that we trip a bug in end_page_writeback() because | |
454 | * we call it twice for the one page as the code in end_buffer_async_write() | |
455 | * assumes that all buffers on the page are started at the same time. | |
456 | * | |
457 | * The fix is two passes across the ioend list - one to start writeback on the | |
c41564b5 | 458 | * buffer_heads, and then submit them for I/O on the second pass. |
7bf7f352 DC |
459 | * |
460 | * If @fail is non-zero, it means that we have a situation where some part of | |
461 | * the submission process has failed after we have marked paged for writeback | |
462 | * and unlocked them. In this situation, we need to fail the ioend chain rather | |
463 | * than submit it to IO. This typically only happens on a filesystem shutdown. | |
f6d6d4fc CH |
464 | */ |
465 | STATIC void | |
466 | xfs_submit_ioend( | |
06342cf8 | 467 | struct writeback_control *wbc, |
7bf7f352 DC |
468 | xfs_ioend_t *ioend, |
469 | int fail) | |
f6d6d4fc | 470 | { |
d88992f6 | 471 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
472 | xfs_ioend_t *next; |
473 | struct buffer_head *bh; | |
474 | struct bio *bio; | |
475 | sector_t lastblock = 0; | |
476 | ||
d88992f6 DC |
477 | /* Pass 1 - start writeback */ |
478 | do { | |
479 | next = ioend->io_list; | |
221cb251 | 480 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) |
d88992f6 | 481 | xfs_start_buffer_writeback(bh); |
d88992f6 DC |
482 | } while ((ioend = next) != NULL); |
483 | ||
484 | /* Pass 2 - submit I/O */ | |
485 | ioend = head; | |
f6d6d4fc CH |
486 | do { |
487 | next = ioend->io_list; | |
488 | bio = NULL; | |
489 | ||
7bf7f352 DC |
490 | /* |
491 | * If we are failing the IO now, just mark the ioend with an | |
492 | * error and finish it. This will run IO completion immediately | |
493 | * as there is only one reference to the ioend at this point in | |
494 | * time. | |
495 | */ | |
496 | if (fail) { | |
2451337d | 497 | ioend->io_error = fail; |
7bf7f352 DC |
498 | xfs_finish_ioend(ioend); |
499 | continue; | |
500 | } | |
501 | ||
f6d6d4fc | 502 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { |
f6d6d4fc CH |
503 | |
504 | if (!bio) { | |
505 | retry: | |
506 | bio = xfs_alloc_ioend_bio(bh); | |
507 | } else if (bh->b_blocknr != lastblock + 1) { | |
06342cf8 | 508 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
509 | goto retry; |
510 | } | |
511 | ||
c7c1a7d8 | 512 | if (xfs_bio_add_buffer(bio, bh) != bh->b_size) { |
06342cf8 | 513 | xfs_submit_ioend_bio(wbc, ioend, bio); |
f6d6d4fc CH |
514 | goto retry; |
515 | } | |
516 | ||
517 | lastblock = bh->b_blocknr; | |
518 | } | |
519 | if (bio) | |
06342cf8 | 520 | xfs_submit_ioend_bio(wbc, ioend, bio); |
209fb87a | 521 | xfs_finish_ioend(ioend); |
f6d6d4fc CH |
522 | } while ((ioend = next) != NULL); |
523 | } | |
524 | ||
525 | /* | |
526 | * Cancel submission of all buffer_heads so far in this endio. | |
527 | * Toss the endio too. Only ever called for the initial page | |
528 | * in a writepage request, so only ever one page. | |
529 | */ | |
530 | STATIC void | |
531 | xfs_cancel_ioend( | |
532 | xfs_ioend_t *ioend) | |
533 | { | |
534 | xfs_ioend_t *next; | |
535 | struct buffer_head *bh, *next_bh; | |
536 | ||
537 | do { | |
538 | next = ioend->io_list; | |
539 | bh = ioend->io_buffer_head; | |
540 | do { | |
541 | next_bh = bh->b_private; | |
542 | clear_buffer_async_write(bh); | |
07d08681 BF |
543 | /* |
544 | * The unwritten flag is cleared when added to the | |
545 | * ioend. We're not submitting for I/O so mark the | |
546 | * buffer unwritten again for next time around. | |
547 | */ | |
548 | if (ioend->io_type == XFS_IO_UNWRITTEN) | |
549 | set_buffer_unwritten(bh); | |
f6d6d4fc CH |
550 | unlock_buffer(bh); |
551 | } while ((bh = next_bh) != NULL); | |
552 | ||
f6d6d4fc CH |
553 | mempool_free(ioend, xfs_ioend_pool); |
554 | } while ((ioend = next) != NULL); | |
555 | } | |
556 | ||
557 | /* | |
558 | * Test to see if we've been building up a completion structure for | |
559 | * earlier buffers -- if so, we try to append to this ioend if we | |
560 | * can, otherwise we finish off any current ioend and start another. | |
561 | * Return true if we've finished the given ioend. | |
562 | */ | |
563 | STATIC void | |
564 | xfs_add_to_ioend( | |
565 | struct inode *inode, | |
566 | struct buffer_head *bh, | |
7336cea8 | 567 | xfs_off_t offset, |
f6d6d4fc CH |
568 | unsigned int type, |
569 | xfs_ioend_t **result, | |
570 | int need_ioend) | |
571 | { | |
572 | xfs_ioend_t *ioend = *result; | |
573 | ||
574 | if (!ioend || need_ioend || type != ioend->io_type) { | |
575 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 576 | |
f6d6d4fc CH |
577 | ioend = xfs_alloc_ioend(inode, type); |
578 | ioend->io_offset = offset; | |
579 | ioend->io_buffer_head = bh; | |
580 | ioend->io_buffer_tail = bh; | |
581 | if (previous) | |
582 | previous->io_list = ioend; | |
583 | *result = ioend; | |
584 | } else { | |
585 | ioend->io_buffer_tail->b_private = bh; | |
586 | ioend->io_buffer_tail = bh; | |
587 | } | |
588 | ||
589 | bh->b_private = NULL; | |
590 | ioend->io_size += bh->b_size; | |
591 | } | |
592 | ||
87cbc49c NS |
593 | STATIC void |
594 | xfs_map_buffer( | |
046f1685 | 595 | struct inode *inode, |
87cbc49c | 596 | struct buffer_head *bh, |
207d0416 | 597 | struct xfs_bmbt_irec *imap, |
046f1685 | 598 | xfs_off_t offset) |
87cbc49c NS |
599 | { |
600 | sector_t bn; | |
8699bb0a | 601 | struct xfs_mount *m = XFS_I(inode)->i_mount; |
207d0416 CH |
602 | xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff); |
603 | xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock); | |
87cbc49c | 604 | |
207d0416 CH |
605 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
606 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
87cbc49c | 607 | |
e513182d | 608 | bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) + |
8699bb0a | 609 | ((offset - iomap_offset) >> inode->i_blkbits); |
87cbc49c | 610 | |
046f1685 | 611 | ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode))); |
87cbc49c NS |
612 | |
613 | bh->b_blocknr = bn; | |
614 | set_buffer_mapped(bh); | |
615 | } | |
616 | ||
1da177e4 LT |
617 | STATIC void |
618 | xfs_map_at_offset( | |
046f1685 | 619 | struct inode *inode, |
1da177e4 | 620 | struct buffer_head *bh, |
207d0416 | 621 | struct xfs_bmbt_irec *imap, |
046f1685 | 622 | xfs_off_t offset) |
1da177e4 | 623 | { |
207d0416 CH |
624 | ASSERT(imap->br_startblock != HOLESTARTBLOCK); |
625 | ASSERT(imap->br_startblock != DELAYSTARTBLOCK); | |
1da177e4 | 626 | |
207d0416 | 627 | xfs_map_buffer(inode, bh, imap, offset); |
1da177e4 LT |
628 | set_buffer_mapped(bh); |
629 | clear_buffer_delay(bh); | |
f6d6d4fc | 630 | clear_buffer_unwritten(bh); |
1da177e4 LT |
631 | } |
632 | ||
1da177e4 | 633 | /* |
a49935f2 DC |
634 | * Test if a given page contains at least one buffer of a given @type. |
635 | * If @check_all_buffers is true, then we walk all the buffers in the page to | |
636 | * try to find one of the type passed in. If it is not set, then the caller only | |
637 | * needs to check the first buffer on the page for a match. | |
1da177e4 | 638 | */ |
a49935f2 | 639 | STATIC bool |
6ffc4db5 | 640 | xfs_check_page_type( |
10ce4444 | 641 | struct page *page, |
a49935f2 DC |
642 | unsigned int type, |
643 | bool check_all_buffers) | |
1da177e4 | 644 | { |
a49935f2 DC |
645 | struct buffer_head *bh; |
646 | struct buffer_head *head; | |
1da177e4 | 647 | |
a49935f2 DC |
648 | if (PageWriteback(page)) |
649 | return false; | |
650 | if (!page->mapping) | |
651 | return false; | |
652 | if (!page_has_buffers(page)) | |
653 | return false; | |
1da177e4 | 654 | |
a49935f2 DC |
655 | bh = head = page_buffers(page); |
656 | do { | |
657 | if (buffer_unwritten(bh)) { | |
658 | if (type == XFS_IO_UNWRITTEN) | |
659 | return true; | |
660 | } else if (buffer_delay(bh)) { | |
805eeb8e | 661 | if (type == XFS_IO_DELALLOC) |
a49935f2 DC |
662 | return true; |
663 | } else if (buffer_dirty(bh) && buffer_mapped(bh)) { | |
805eeb8e | 664 | if (type == XFS_IO_OVERWRITE) |
a49935f2 DC |
665 | return true; |
666 | } | |
1da177e4 | 667 | |
a49935f2 DC |
668 | /* If we are only checking the first buffer, we are done now. */ |
669 | if (!check_all_buffers) | |
670 | break; | |
671 | } while ((bh = bh->b_this_page) != head); | |
1da177e4 | 672 | |
a49935f2 | 673 | return false; |
1da177e4 LT |
674 | } |
675 | ||
1da177e4 LT |
676 | /* |
677 | * Allocate & map buffers for page given the extent map. Write it out. | |
678 | * except for the original page of a writepage, this is called on | |
679 | * delalloc/unwritten pages only, for the original page it is possible | |
680 | * that the page has no mapping at all. | |
681 | */ | |
f6d6d4fc | 682 | STATIC int |
1da177e4 LT |
683 | xfs_convert_page( |
684 | struct inode *inode, | |
685 | struct page *page, | |
10ce4444 | 686 | loff_t tindex, |
207d0416 | 687 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 688 | xfs_ioend_t **ioendp, |
2fa24f92 | 689 | struct writeback_control *wbc) |
1da177e4 | 690 | { |
f6d6d4fc | 691 | struct buffer_head *bh, *head; |
9260dc6b CH |
692 | xfs_off_t end_offset; |
693 | unsigned long p_offset; | |
f6d6d4fc | 694 | unsigned int type; |
24e17b5f | 695 | int len, page_dirty; |
f6d6d4fc | 696 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 697 | xfs_off_t offset = page_offset(page); |
1da177e4 | 698 | |
10ce4444 CH |
699 | if (page->index != tindex) |
700 | goto fail; | |
529ae9aa | 701 | if (!trylock_page(page)) |
10ce4444 CH |
702 | goto fail; |
703 | if (PageWriteback(page)) | |
704 | goto fail_unlock_page; | |
705 | if (page->mapping != inode->i_mapping) | |
706 | goto fail_unlock_page; | |
a49935f2 | 707 | if (!xfs_check_page_type(page, (*ioendp)->io_type, false)) |
10ce4444 CH |
708 | goto fail_unlock_page; |
709 | ||
24e17b5f NS |
710 | /* |
711 | * page_dirty is initially a count of buffers on the page before | |
c41564b5 | 712 | * EOF and is decremented as we move each into a cleanable state. |
9260dc6b CH |
713 | * |
714 | * Derivation: | |
715 | * | |
716 | * End offset is the highest offset that this page should represent. | |
717 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
718 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
719 | * hence give us the correct page_dirty count. On any other page, | |
720 | * it will be zero and in that case we need page_dirty to be the | |
721 | * count of buffers on the page. | |
24e17b5f | 722 | */ |
9260dc6b CH |
723 | end_offset = min_t(unsigned long long, |
724 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
725 | i_size_read(inode)); | |
726 | ||
480d7467 DC |
727 | /* |
728 | * If the current map does not span the entire page we are about to try | |
729 | * to write, then give up. The only way we can write a page that spans | |
730 | * multiple mappings in a single writeback iteration is via the | |
731 | * xfs_vm_writepage() function. Data integrity writeback requires the | |
732 | * entire page to be written in a single attempt, otherwise the part of | |
733 | * the page we don't write here doesn't get written as part of the data | |
734 | * integrity sync. | |
735 | * | |
736 | * For normal writeback, we also don't attempt to write partial pages | |
737 | * here as it simply means that write_cache_pages() will see it under | |
738 | * writeback and ignore the page until some point in the future, at | |
739 | * which time this will be the only page in the file that needs | |
740 | * writeback. Hence for more optimal IO patterns, we should always | |
741 | * avoid partial page writeback due to multiple mappings on a page here. | |
742 | */ | |
743 | if (!xfs_imap_valid(inode, imap, end_offset)) | |
744 | goto fail_unlock_page; | |
745 | ||
24e17b5f | 746 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
747 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
748 | PAGE_CACHE_SIZE); | |
749 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
750 | page_dirty = p_offset / len; | |
24e17b5f | 751 | |
a49935f2 DC |
752 | /* |
753 | * The moment we find a buffer that doesn't match our current type | |
754 | * specification or can't be written, abort the loop and start | |
755 | * writeback. As per the above xfs_imap_valid() check, only | |
756 | * xfs_vm_writepage() can handle partial page writeback fully - we are | |
757 | * limited here to the buffers that are contiguous with the current | |
758 | * ioend, and hence a buffer we can't write breaks that contiguity and | |
759 | * we have to defer the rest of the IO to xfs_vm_writepage(). | |
760 | */ | |
1da177e4 LT |
761 | bh = head = page_buffers(page); |
762 | do { | |
9260dc6b | 763 | if (offset >= end_offset) |
1da177e4 | 764 | break; |
f6d6d4fc CH |
765 | if (!buffer_uptodate(bh)) |
766 | uptodate = 0; | |
767 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
768 | done = 1; | |
a49935f2 | 769 | break; |
f6d6d4fc CH |
770 | } |
771 | ||
2fa24f92 CH |
772 | if (buffer_unwritten(bh) || buffer_delay(bh) || |
773 | buffer_mapped(bh)) { | |
9260dc6b | 774 | if (buffer_unwritten(bh)) |
0d882a36 | 775 | type = XFS_IO_UNWRITTEN; |
2fa24f92 | 776 | else if (buffer_delay(bh)) |
0d882a36 | 777 | type = XFS_IO_DELALLOC; |
2fa24f92 | 778 | else |
0d882a36 | 779 | type = XFS_IO_OVERWRITE; |
9260dc6b | 780 | |
a49935f2 DC |
781 | /* |
782 | * imap should always be valid because of the above | |
783 | * partial page end_offset check on the imap. | |
784 | */ | |
785 | ASSERT(xfs_imap_valid(inode, imap, offset)); | |
9260dc6b | 786 | |
ecff71e6 | 787 | lock_buffer(bh); |
0d882a36 | 788 | if (type != XFS_IO_OVERWRITE) |
2fa24f92 | 789 | xfs_map_at_offset(inode, bh, imap, offset); |
89f3b363 CH |
790 | xfs_add_to_ioend(inode, bh, offset, type, |
791 | ioendp, done); | |
792 | ||
9260dc6b CH |
793 | page_dirty--; |
794 | count++; | |
795 | } else { | |
2fa24f92 | 796 | done = 1; |
a49935f2 | 797 | break; |
1da177e4 | 798 | } |
7336cea8 | 799 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 800 | |
f6d6d4fc CH |
801 | if (uptodate && bh == head) |
802 | SetPageUptodate(page); | |
803 | ||
89f3b363 | 804 | if (count) { |
efceab1d DC |
805 | if (--wbc->nr_to_write <= 0 && |
806 | wbc->sync_mode == WB_SYNC_NONE) | |
89f3b363 | 807 | done = 1; |
1da177e4 | 808 | } |
89f3b363 | 809 | xfs_start_page_writeback(page, !page_dirty, count); |
f6d6d4fc CH |
810 | |
811 | return done; | |
10ce4444 CH |
812 | fail_unlock_page: |
813 | unlock_page(page); | |
814 | fail: | |
815 | return 1; | |
1da177e4 LT |
816 | } |
817 | ||
818 | /* | |
819 | * Convert & write out a cluster of pages in the same extent as defined | |
820 | * by mp and following the start page. | |
821 | */ | |
822 | STATIC void | |
823 | xfs_cluster_write( | |
824 | struct inode *inode, | |
825 | pgoff_t tindex, | |
207d0416 | 826 | struct xfs_bmbt_irec *imap, |
f6d6d4fc | 827 | xfs_ioend_t **ioendp, |
1da177e4 | 828 | struct writeback_control *wbc, |
1da177e4 LT |
829 | pgoff_t tlast) |
830 | { | |
10ce4444 CH |
831 | struct pagevec pvec; |
832 | int done = 0, i; | |
1da177e4 | 833 | |
10ce4444 CH |
834 | pagevec_init(&pvec, 0); |
835 | while (!done && tindex <= tlast) { | |
836 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
837 | ||
838 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 839 | break; |
10ce4444 CH |
840 | |
841 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
842 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
2fa24f92 | 843 | imap, ioendp, wbc); |
10ce4444 CH |
844 | if (done) |
845 | break; | |
846 | } | |
847 | ||
848 | pagevec_release(&pvec); | |
849 | cond_resched(); | |
1da177e4 LT |
850 | } |
851 | } | |
852 | ||
3ed3a434 DC |
853 | STATIC void |
854 | xfs_vm_invalidatepage( | |
855 | struct page *page, | |
d47992f8 LC |
856 | unsigned int offset, |
857 | unsigned int length) | |
3ed3a434 | 858 | { |
34097dfe LC |
859 | trace_xfs_invalidatepage(page->mapping->host, page, offset, |
860 | length); | |
861 | block_invalidatepage(page, offset, length); | |
3ed3a434 DC |
862 | } |
863 | ||
864 | /* | |
865 | * If the page has delalloc buffers on it, we need to punch them out before we | |
866 | * invalidate the page. If we don't, we leave a stale delalloc mapping on the | |
867 | * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read | |
868 | * is done on that same region - the delalloc extent is returned when none is | |
869 | * supposed to be there. | |
870 | * | |
871 | * We prevent this by truncating away the delalloc regions on the page before | |
872 | * invalidating it. Because they are delalloc, we can do this without needing a | |
873 | * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this | |
874 | * truncation without a transaction as there is no space left for block | |
875 | * reservation (typically why we see a ENOSPC in writeback). | |
876 | * | |
877 | * This is not a performance critical path, so for now just do the punching a | |
878 | * buffer head at a time. | |
879 | */ | |
880 | STATIC void | |
881 | xfs_aops_discard_page( | |
882 | struct page *page) | |
883 | { | |
884 | struct inode *inode = page->mapping->host; | |
885 | struct xfs_inode *ip = XFS_I(inode); | |
886 | struct buffer_head *bh, *head; | |
887 | loff_t offset = page_offset(page); | |
3ed3a434 | 888 | |
a49935f2 | 889 | if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true)) |
3ed3a434 DC |
890 | goto out_invalidate; |
891 | ||
e8c3753c DC |
892 | if (XFS_FORCED_SHUTDOWN(ip->i_mount)) |
893 | goto out_invalidate; | |
894 | ||
4f10700a | 895 | xfs_alert(ip->i_mount, |
3ed3a434 DC |
896 | "page discard on page %p, inode 0x%llx, offset %llu.", |
897 | page, ip->i_ino, offset); | |
898 | ||
899 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
900 | bh = head = page_buffers(page); | |
901 | do { | |
3ed3a434 | 902 | int error; |
c726de44 | 903 | xfs_fileoff_t start_fsb; |
3ed3a434 DC |
904 | |
905 | if (!buffer_delay(bh)) | |
906 | goto next_buffer; | |
907 | ||
c726de44 DC |
908 | start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset); |
909 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1); | |
3ed3a434 DC |
910 | if (error) { |
911 | /* something screwed, just bail */ | |
e8c3753c | 912 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { |
4f10700a | 913 | xfs_alert(ip->i_mount, |
3ed3a434 | 914 | "page discard unable to remove delalloc mapping."); |
e8c3753c | 915 | } |
3ed3a434 DC |
916 | break; |
917 | } | |
918 | next_buffer: | |
c726de44 | 919 | offset += 1 << inode->i_blkbits; |
3ed3a434 DC |
920 | |
921 | } while ((bh = bh->b_this_page) != head); | |
922 | ||
923 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
924 | out_invalidate: | |
d47992f8 | 925 | xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
3ed3a434 DC |
926 | return; |
927 | } | |
928 | ||
1da177e4 | 929 | /* |
89f3b363 CH |
930 | * Write out a dirty page. |
931 | * | |
932 | * For delalloc space on the page we need to allocate space and flush it. | |
933 | * For unwritten space on the page we need to start the conversion to | |
934 | * regular allocated space. | |
89f3b363 | 935 | * For any other dirty buffer heads on the page we should flush them. |
1da177e4 | 936 | */ |
1da177e4 | 937 | STATIC int |
89f3b363 CH |
938 | xfs_vm_writepage( |
939 | struct page *page, | |
940 | struct writeback_control *wbc) | |
1da177e4 | 941 | { |
89f3b363 | 942 | struct inode *inode = page->mapping->host; |
f6d6d4fc | 943 | struct buffer_head *bh, *head; |
207d0416 | 944 | struct xfs_bmbt_irec imap; |
f6d6d4fc | 945 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 | 946 | loff_t offset; |
f6d6d4fc | 947 | unsigned int type; |
1da177e4 | 948 | __uint64_t end_offset; |
bd1556a1 | 949 | pgoff_t end_index, last_index; |
ed1e7b7e | 950 | ssize_t len; |
a206c817 | 951 | int err, imap_valid = 0, uptodate = 1; |
89f3b363 | 952 | int count = 0; |
a206c817 | 953 | int nonblocking = 0; |
89f3b363 | 954 | |
34097dfe | 955 | trace_xfs_writepage(inode, page, 0, 0); |
89f3b363 | 956 | |
20cb52eb CH |
957 | ASSERT(page_has_buffers(page)); |
958 | ||
89f3b363 CH |
959 | /* |
960 | * Refuse to write the page out if we are called from reclaim context. | |
961 | * | |
d4f7a5cb CH |
962 | * This avoids stack overflows when called from deeply used stacks in |
963 | * random callers for direct reclaim or memcg reclaim. We explicitly | |
964 | * allow reclaim from kswapd as the stack usage there is relatively low. | |
89f3b363 | 965 | * |
94054fa3 MG |
966 | * This should never happen except in the case of a VM regression so |
967 | * warn about it. | |
89f3b363 | 968 | */ |
94054fa3 MG |
969 | if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
970 | PF_MEMALLOC)) | |
b5420f23 | 971 | goto redirty; |
1da177e4 | 972 | |
89f3b363 | 973 | /* |
680a647b CH |
974 | * Given that we do not allow direct reclaim to call us, we should |
975 | * never be called while in a filesystem transaction. | |
89f3b363 | 976 | */ |
448011e2 | 977 | if (WARN_ON_ONCE(current->flags & PF_FSTRANS)) |
b5420f23 | 978 | goto redirty; |
89f3b363 | 979 | |
1da177e4 LT |
980 | /* Is this page beyond the end of the file? */ |
981 | offset = i_size_read(inode); | |
982 | end_index = offset >> PAGE_CACHE_SHIFT; | |
983 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
8695d27e JL |
984 | |
985 | /* | |
986 | * The page index is less than the end_index, adjust the end_offset | |
987 | * to the highest offset that this page should represent. | |
988 | * ----------------------------------------------------- | |
989 | * | file mapping | <EOF> | | |
990 | * ----------------------------------------------------- | |
991 | * | Page ... | Page N-2 | Page N-1 | Page N | | | |
992 | * ^--------------------------------^----------|-------- | |
993 | * | desired writeback range | see else | | |
994 | * ---------------------------------^------------------| | |
995 | */ | |
996 | if (page->index < end_index) | |
997 | end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT; | |
998 | else { | |
999 | /* | |
1000 | * Check whether the page to write out is beyond or straddles | |
1001 | * i_size or not. | |
1002 | * ------------------------------------------------------- | |
1003 | * | file mapping | <EOF> | | |
1004 | * ------------------------------------------------------- | |
1005 | * | Page ... | Page N-2 | Page N-1 | Page N | Beyond | | |
1006 | * ^--------------------------------^-----------|--------- | |
1007 | * | | Straddles | | |
1008 | * ---------------------------------^-----------|--------| | |
1009 | */ | |
6b7a03f0 CH |
1010 | unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1); |
1011 | ||
1012 | /* | |
ff9a28f6 JK |
1013 | * Skip the page if it is fully outside i_size, e.g. due to a |
1014 | * truncate operation that is in progress. We must redirty the | |
1015 | * page so that reclaim stops reclaiming it. Otherwise | |
1016 | * xfs_vm_releasepage() is called on it and gets confused. | |
8695d27e JL |
1017 | * |
1018 | * Note that the end_index is unsigned long, it would overflow | |
1019 | * if the given offset is greater than 16TB on 32-bit system | |
1020 | * and if we do check the page is fully outside i_size or not | |
1021 | * via "if (page->index >= end_index + 1)" as "end_index + 1" | |
1022 | * will be evaluated to 0. Hence this page will be redirtied | |
1023 | * and be written out repeatedly which would result in an | |
1024 | * infinite loop, the user program that perform this operation | |
1025 | * will hang. Instead, we can verify this situation by checking | |
1026 | * if the page to write is totally beyond the i_size or if it's | |
1027 | * offset is just equal to the EOF. | |
6b7a03f0 | 1028 | */ |
8695d27e JL |
1029 | if (page->index > end_index || |
1030 | (page->index == end_index && offset_into_page == 0)) | |
ff9a28f6 | 1031 | goto redirty; |
6b7a03f0 CH |
1032 | |
1033 | /* | |
1034 | * The page straddles i_size. It must be zeroed out on each | |
1035 | * and every writepage invocation because it may be mmapped. | |
1036 | * "A file is mapped in multiples of the page size. For a file | |
8695d27e | 1037 | * that is not a multiple of the page size, the remaining |
6b7a03f0 CH |
1038 | * memory is zeroed when mapped, and writes to that region are |
1039 | * not written out to the file." | |
1040 | */ | |
1041 | zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE); | |
8695d27e JL |
1042 | |
1043 | /* Adjust the end_offset to the end of file */ | |
1044 | end_offset = offset; | |
1da177e4 LT |
1045 | } |
1046 | ||
24e17b5f | 1047 | len = 1 << inode->i_blkbits; |
24e17b5f | 1048 | |
24e17b5f | 1049 | bh = head = page_buffers(page); |
f6d6d4fc | 1050 | offset = page_offset(page); |
0d882a36 | 1051 | type = XFS_IO_OVERWRITE; |
a206c817 | 1052 | |
dbcdde3e | 1053 | if (wbc->sync_mode == WB_SYNC_NONE) |
a206c817 | 1054 | nonblocking = 1; |
f6d6d4fc | 1055 | |
1da177e4 | 1056 | do { |
6ac7248e CH |
1057 | int new_ioend = 0; |
1058 | ||
1da177e4 LT |
1059 | if (offset >= end_offset) |
1060 | break; | |
1061 | if (!buffer_uptodate(bh)) | |
1062 | uptodate = 0; | |
1da177e4 | 1063 | |
3d9b02e3 | 1064 | /* |
ece413f5 CH |
1065 | * set_page_dirty dirties all buffers in a page, independent |
1066 | * of their state. The dirty state however is entirely | |
1067 | * meaningless for holes (!mapped && uptodate), so skip | |
1068 | * buffers covering holes here. | |
3d9b02e3 ES |
1069 | */ |
1070 | if (!buffer_mapped(bh) && buffer_uptodate(bh)) { | |
3d9b02e3 ES |
1071 | imap_valid = 0; |
1072 | continue; | |
1073 | } | |
1074 | ||
aeea1b1f | 1075 | if (buffer_unwritten(bh)) { |
0d882a36 AR |
1076 | if (type != XFS_IO_UNWRITTEN) { |
1077 | type = XFS_IO_UNWRITTEN; | |
aeea1b1f | 1078 | imap_valid = 0; |
1da177e4 | 1079 | } |
aeea1b1f | 1080 | } else if (buffer_delay(bh)) { |
0d882a36 AR |
1081 | if (type != XFS_IO_DELALLOC) { |
1082 | type = XFS_IO_DELALLOC; | |
aeea1b1f | 1083 | imap_valid = 0; |
1da177e4 | 1084 | } |
89f3b363 | 1085 | } else if (buffer_uptodate(bh)) { |
0d882a36 AR |
1086 | if (type != XFS_IO_OVERWRITE) { |
1087 | type = XFS_IO_OVERWRITE; | |
85da94c6 CH |
1088 | imap_valid = 0; |
1089 | } | |
aeea1b1f | 1090 | } else { |
7d0fa3ec | 1091 | if (PageUptodate(page)) |
aeea1b1f | 1092 | ASSERT(buffer_mapped(bh)); |
7d0fa3ec AR |
1093 | /* |
1094 | * This buffer is not uptodate and will not be | |
1095 | * written to disk. Ensure that we will put any | |
1096 | * subsequent writeable buffers into a new | |
1097 | * ioend. | |
1098 | */ | |
1099 | imap_valid = 0; | |
aeea1b1f CH |
1100 | continue; |
1101 | } | |
d5cb48aa | 1102 | |
aeea1b1f CH |
1103 | if (imap_valid) |
1104 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1105 | if (!imap_valid) { | |
1106 | /* | |
1107 | * If we didn't have a valid mapping then we need to | |
1108 | * put the new mapping into a separate ioend structure. | |
1109 | * This ensures non-contiguous extents always have | |
1110 | * separate ioends, which is particularly important | |
1111 | * for unwritten extent conversion at I/O completion | |
1112 | * time. | |
1113 | */ | |
1114 | new_ioend = 1; | |
1115 | err = xfs_map_blocks(inode, offset, &imap, type, | |
1116 | nonblocking); | |
1117 | if (err) | |
1118 | goto error; | |
1119 | imap_valid = xfs_imap_valid(inode, &imap, offset); | |
1120 | } | |
1121 | if (imap_valid) { | |
ecff71e6 | 1122 | lock_buffer(bh); |
0d882a36 | 1123 | if (type != XFS_IO_OVERWRITE) |
aeea1b1f CH |
1124 | xfs_map_at_offset(inode, bh, &imap, offset); |
1125 | xfs_add_to_ioend(inode, bh, offset, type, &ioend, | |
1126 | new_ioend); | |
1127 | count++; | |
1da177e4 | 1128 | } |
f6d6d4fc CH |
1129 | |
1130 | if (!iohead) | |
1131 | iohead = ioend; | |
1132 | ||
1133 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1134 | |
1135 | if (uptodate && bh == head) | |
1136 | SetPageUptodate(page); | |
1137 | ||
89f3b363 | 1138 | xfs_start_page_writeback(page, 1, count); |
1da177e4 | 1139 | |
7bf7f352 DC |
1140 | /* if there is no IO to be submitted for this page, we are done */ |
1141 | if (!ioend) | |
1142 | return 0; | |
1143 | ||
1144 | ASSERT(iohead); | |
1145 | ||
1146 | /* | |
1147 | * Any errors from this point onwards need tobe reported through the IO | |
1148 | * completion path as we have marked the initial page as under writeback | |
1149 | * and unlocked it. | |
1150 | */ | |
1151 | if (imap_valid) { | |
bd1556a1 CH |
1152 | xfs_off_t end_index; |
1153 | ||
1154 | end_index = imap.br_startoff + imap.br_blockcount; | |
1155 | ||
1156 | /* to bytes */ | |
1157 | end_index <<= inode->i_blkbits; | |
1158 | ||
1159 | /* to pages */ | |
1160 | end_index = (end_index - 1) >> PAGE_CACHE_SHIFT; | |
1161 | ||
1162 | /* check against file size */ | |
1163 | if (end_index > last_index) | |
1164 | end_index = last_index; | |
8699bb0a | 1165 | |
207d0416 | 1166 | xfs_cluster_write(inode, page->index + 1, &imap, &ioend, |
2fa24f92 | 1167 | wbc, end_index); |
1da177e4 LT |
1168 | } |
1169 | ||
281627df | 1170 | |
7bf7f352 DC |
1171 | /* |
1172 | * Reserve log space if we might write beyond the on-disk inode size. | |
1173 | */ | |
1174 | err = 0; | |
1175 | if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend)) | |
1176 | err = xfs_setfilesize_trans_alloc(ioend); | |
1177 | ||
1178 | xfs_submit_ioend(wbc, iohead, err); | |
f6d6d4fc | 1179 | |
89f3b363 | 1180 | return 0; |
1da177e4 LT |
1181 | |
1182 | error: | |
f6d6d4fc CH |
1183 | if (iohead) |
1184 | xfs_cancel_ioend(iohead); | |
1da177e4 | 1185 | |
b5420f23 CH |
1186 | if (err == -EAGAIN) |
1187 | goto redirty; | |
1188 | ||
20cb52eb | 1189 | xfs_aops_discard_page(page); |
89f3b363 CH |
1190 | ClearPageUptodate(page); |
1191 | unlock_page(page); | |
1da177e4 | 1192 | return err; |
f51623b2 | 1193 | |
b5420f23 | 1194 | redirty: |
f51623b2 NS |
1195 | redirty_page_for_writepage(wbc, page); |
1196 | unlock_page(page); | |
1197 | return 0; | |
f51623b2 NS |
1198 | } |
1199 | ||
7d4fb40a NS |
1200 | STATIC int |
1201 | xfs_vm_writepages( | |
1202 | struct address_space *mapping, | |
1203 | struct writeback_control *wbc) | |
1204 | { | |
b3aea4ed | 1205 | xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED); |
7d4fb40a NS |
1206 | return generic_writepages(mapping, wbc); |
1207 | } | |
1208 | ||
f51623b2 NS |
1209 | /* |
1210 | * Called to move a page into cleanable state - and from there | |
89f3b363 | 1211 | * to be released. The page should already be clean. We always |
f51623b2 NS |
1212 | * have buffer heads in this call. |
1213 | * | |
89f3b363 | 1214 | * Returns 1 if the page is ok to release, 0 otherwise. |
f51623b2 NS |
1215 | */ |
1216 | STATIC int | |
238f4c54 | 1217 | xfs_vm_releasepage( |
f51623b2 NS |
1218 | struct page *page, |
1219 | gfp_t gfp_mask) | |
1220 | { | |
20cb52eb | 1221 | int delalloc, unwritten; |
f51623b2 | 1222 | |
34097dfe | 1223 | trace_xfs_releasepage(page->mapping->host, page, 0, 0); |
238f4c54 | 1224 | |
20cb52eb | 1225 | xfs_count_page_state(page, &delalloc, &unwritten); |
f51623b2 | 1226 | |
448011e2 | 1227 | if (WARN_ON_ONCE(delalloc)) |
f51623b2 | 1228 | return 0; |
448011e2 | 1229 | if (WARN_ON_ONCE(unwritten)) |
f51623b2 NS |
1230 | return 0; |
1231 | ||
f51623b2 NS |
1232 | return try_to_free_buffers(page); |
1233 | } | |
1234 | ||
a719370b | 1235 | /* |
a06c277a DC |
1236 | * When we map a DIO buffer, we may need to attach an ioend that describes the |
1237 | * type of write IO we are doing. This passes to the completion function the | |
1238 | * operations it needs to perform. If the mapping is for an overwrite wholly | |
1239 | * within the EOF then we don't need an ioend and so we don't allocate one. | |
1240 | * This avoids the unnecessary overhead of allocating and freeing ioends for | |
1241 | * workloads that don't require transactions on IO completion. | |
d5cc2e3f DC |
1242 | * |
1243 | * If we get multiple mappings in a single IO, we might be mapping different | |
1244 | * types. But because the direct IO can only have a single private pointer, we | |
1245 | * need to ensure that: | |
1246 | * | |
a06c277a DC |
1247 | * a) i) the ioend spans the entire region of unwritten mappings; or |
1248 | * ii) the ioend spans all the mappings that cross or are beyond EOF; and | |
d5cc2e3f DC |
1249 | * b) if it contains unwritten extents, it is *permanently* marked as such |
1250 | * | |
1251 | * We could do this by chaining ioends like buffered IO does, but we only | |
1252 | * actually get one IO completion callback from the direct IO, and that spans | |
1253 | * the entire IO regardless of how many mappings and IOs are needed to complete | |
1254 | * the DIO. There is only going to be one reference to the ioend and its life | |
1255 | * cycle is constrained by the DIO completion code. hence we don't need | |
1256 | * reference counting here. | |
a719370b DC |
1257 | */ |
1258 | static void | |
1259 | xfs_map_direct( | |
1260 | struct inode *inode, | |
1261 | struct buffer_head *bh_result, | |
1262 | struct xfs_bmbt_irec *imap, | |
1263 | xfs_off_t offset) | |
1264 | { | |
d5cc2e3f DC |
1265 | struct xfs_ioend *ioend; |
1266 | xfs_off_t size = bh_result->b_size; | |
1267 | int type; | |
1268 | ||
1269 | if (ISUNWRITTEN(imap)) | |
1270 | type = XFS_IO_UNWRITTEN; | |
1271 | else | |
1272 | type = XFS_IO_OVERWRITE; | |
1273 | ||
1274 | trace_xfs_gbmap_direct(XFS_I(inode), offset, size, type, imap); | |
1275 | ||
1276 | if (bh_result->b_private) { | |
1277 | ioend = bh_result->b_private; | |
1278 | ASSERT(ioend->io_size > 0); | |
1279 | ASSERT(offset >= ioend->io_offset); | |
1280 | if (offset + size > ioend->io_offset + ioend->io_size) | |
1281 | ioend->io_size = offset - ioend->io_offset + size; | |
1282 | ||
1283 | if (type == XFS_IO_UNWRITTEN && type != ioend->io_type) | |
1284 | ioend->io_type = XFS_IO_UNWRITTEN; | |
1285 | ||
1286 | trace_xfs_gbmap_direct_update(XFS_I(inode), ioend->io_offset, | |
1287 | ioend->io_size, ioend->io_type, | |
1288 | imap); | |
a06c277a DC |
1289 | } else if (type == XFS_IO_UNWRITTEN || |
1290 | offset + size > i_size_read(inode)) { | |
d5cc2e3f DC |
1291 | ioend = xfs_alloc_ioend(inode, type); |
1292 | ioend->io_offset = offset; | |
1293 | ioend->io_size = size; | |
a06c277a | 1294 | |
d5cc2e3f | 1295 | bh_result->b_private = ioend; |
a06c277a | 1296 | set_buffer_defer_completion(bh_result); |
d5cc2e3f DC |
1297 | |
1298 | trace_xfs_gbmap_direct_new(XFS_I(inode), offset, size, type, | |
1299 | imap); | |
a06c277a DC |
1300 | } else { |
1301 | trace_xfs_gbmap_direct_none(XFS_I(inode), offset, size, type, | |
1302 | imap); | |
a719370b DC |
1303 | } |
1304 | } | |
1305 | ||
1fdca9c2 DC |
1306 | /* |
1307 | * If this is O_DIRECT or the mpage code calling tell them how large the mapping | |
1308 | * is, so that we can avoid repeated get_blocks calls. | |
1309 | * | |
1310 | * If the mapping spans EOF, then we have to break the mapping up as the mapping | |
1311 | * for blocks beyond EOF must be marked new so that sub block regions can be | |
1312 | * correctly zeroed. We can't do this for mappings within EOF unless the mapping | |
1313 | * was just allocated or is unwritten, otherwise the callers would overwrite | |
1314 | * existing data with zeros. Hence we have to split the mapping into a range up | |
1315 | * to and including EOF, and a second mapping for beyond EOF. | |
1316 | */ | |
1317 | static void | |
1318 | xfs_map_trim_size( | |
1319 | struct inode *inode, | |
1320 | sector_t iblock, | |
1321 | struct buffer_head *bh_result, | |
1322 | struct xfs_bmbt_irec *imap, | |
1323 | xfs_off_t offset, | |
1324 | ssize_t size) | |
1325 | { | |
1326 | xfs_off_t mapping_size; | |
1327 | ||
1328 | mapping_size = imap->br_startoff + imap->br_blockcount - iblock; | |
1329 | mapping_size <<= inode->i_blkbits; | |
1330 | ||
1331 | ASSERT(mapping_size > 0); | |
1332 | if (mapping_size > size) | |
1333 | mapping_size = size; | |
1334 | if (offset < i_size_read(inode) && | |
1335 | offset + mapping_size >= i_size_read(inode)) { | |
1336 | /* limit mapping to block that spans EOF */ | |
1337 | mapping_size = roundup_64(i_size_read(inode) - offset, | |
1338 | 1 << inode->i_blkbits); | |
1339 | } | |
1340 | if (mapping_size > LONG_MAX) | |
1341 | mapping_size = LONG_MAX; | |
1342 | ||
1343 | bh_result->b_size = mapping_size; | |
1344 | } | |
1345 | ||
1da177e4 | 1346 | STATIC int |
c2536668 | 1347 | __xfs_get_blocks( |
1da177e4 LT |
1348 | struct inode *inode, |
1349 | sector_t iblock, | |
1da177e4 LT |
1350 | struct buffer_head *bh_result, |
1351 | int create, | |
f2bde9b8 | 1352 | int direct) |
1da177e4 | 1353 | { |
a206c817 CH |
1354 | struct xfs_inode *ip = XFS_I(inode); |
1355 | struct xfs_mount *mp = ip->i_mount; | |
1356 | xfs_fileoff_t offset_fsb, end_fsb; | |
1357 | int error = 0; | |
1358 | int lockmode = 0; | |
207d0416 | 1359 | struct xfs_bmbt_irec imap; |
a206c817 | 1360 | int nimaps = 1; |
fdc7ed75 NS |
1361 | xfs_off_t offset; |
1362 | ssize_t size; | |
207d0416 | 1363 | int new = 0; |
a206c817 CH |
1364 | |
1365 | if (XFS_FORCED_SHUTDOWN(mp)) | |
b474c7ae | 1366 | return -EIO; |
1da177e4 | 1367 | |
fdc7ed75 | 1368 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
c2536668 NS |
1369 | ASSERT(bh_result->b_size >= (1 << inode->i_blkbits)); |
1370 | size = bh_result->b_size; | |
364f358a LM |
1371 | |
1372 | if (!create && direct && offset >= i_size_read(inode)) | |
1373 | return 0; | |
1374 | ||
507630b2 DC |
1375 | /* |
1376 | * Direct I/O is usually done on preallocated files, so try getting | |
1377 | * a block mapping without an exclusive lock first. For buffered | |
1378 | * writes we already have the exclusive iolock anyway, so avoiding | |
1379 | * a lock roundtrip here by taking the ilock exclusive from the | |
1380 | * beginning is a useful micro optimization. | |
1381 | */ | |
1382 | if (create && !direct) { | |
a206c817 CH |
1383 | lockmode = XFS_ILOCK_EXCL; |
1384 | xfs_ilock(ip, lockmode); | |
1385 | } else { | |
309ecac8 | 1386 | lockmode = xfs_ilock_data_map_shared(ip); |
a206c817 | 1387 | } |
f2bde9b8 | 1388 | |
d2c28191 DC |
1389 | ASSERT(offset <= mp->m_super->s_maxbytes); |
1390 | if (offset + size > mp->m_super->s_maxbytes) | |
1391 | size = mp->m_super->s_maxbytes - offset; | |
a206c817 CH |
1392 | end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size); |
1393 | offset_fsb = XFS_B_TO_FSBT(mp, offset); | |
1394 | ||
5c8ed202 DC |
1395 | error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, |
1396 | &imap, &nimaps, XFS_BMAPI_ENTIRE); | |
1da177e4 | 1397 | if (error) |
a206c817 CH |
1398 | goto out_unlock; |
1399 | ||
1400 | if (create && | |
1401 | (!nimaps || | |
1402 | (imap.br_startblock == HOLESTARTBLOCK || | |
1403 | imap.br_startblock == DELAYSTARTBLOCK))) { | |
aff3a9ed | 1404 | if (direct || xfs_get_extsz_hint(ip)) { |
507630b2 DC |
1405 | /* |
1406 | * Drop the ilock in preparation for starting the block | |
1407 | * allocation transaction. It will be retaken | |
1408 | * exclusively inside xfs_iomap_write_direct for the | |
1409 | * actual allocation. | |
1410 | */ | |
1411 | xfs_iunlock(ip, lockmode); | |
a206c817 CH |
1412 | error = xfs_iomap_write_direct(ip, offset, size, |
1413 | &imap, nimaps); | |
507630b2 | 1414 | if (error) |
2451337d | 1415 | return error; |
d3bc815a | 1416 | new = 1; |
a206c817 | 1417 | } else { |
507630b2 DC |
1418 | /* |
1419 | * Delalloc reservations do not require a transaction, | |
d3bc815a DC |
1420 | * we can go on without dropping the lock here. If we |
1421 | * are allocating a new delalloc block, make sure that | |
1422 | * we set the new flag so that we mark the buffer new so | |
1423 | * that we know that it is newly allocated if the write | |
1424 | * fails. | |
507630b2 | 1425 | */ |
d3bc815a DC |
1426 | if (nimaps && imap.br_startblock == HOLESTARTBLOCK) |
1427 | new = 1; | |
a206c817 | 1428 | error = xfs_iomap_write_delay(ip, offset, size, &imap); |
507630b2 DC |
1429 | if (error) |
1430 | goto out_unlock; | |
1431 | ||
1432 | xfs_iunlock(ip, lockmode); | |
a206c817 | 1433 | } |
d5cc2e3f DC |
1434 | trace_xfs_get_blocks_alloc(ip, offset, size, |
1435 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1436 | : XFS_IO_DELALLOC, &imap); | |
a206c817 | 1437 | } else if (nimaps) { |
d5cc2e3f DC |
1438 | trace_xfs_get_blocks_found(ip, offset, size, |
1439 | ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN | |
1440 | : XFS_IO_OVERWRITE, &imap); | |
507630b2 | 1441 | xfs_iunlock(ip, lockmode); |
a206c817 CH |
1442 | } else { |
1443 | trace_xfs_get_blocks_notfound(ip, offset, size); | |
1444 | goto out_unlock; | |
1445 | } | |
1da177e4 | 1446 | |
1fdca9c2 DC |
1447 | /* trim mapping down to size requested */ |
1448 | if (direct || size > (1 << inode->i_blkbits)) | |
1449 | xfs_map_trim_size(inode, iblock, bh_result, | |
1450 | &imap, offset, size); | |
1451 | ||
a719370b DC |
1452 | /* |
1453 | * For unwritten extents do not report a disk address in the buffered | |
1454 | * read case (treat as if we're reading into a hole). | |
1455 | */ | |
207d0416 | 1456 | if (imap.br_startblock != HOLESTARTBLOCK && |
a719370b DC |
1457 | imap.br_startblock != DELAYSTARTBLOCK && |
1458 | (create || !ISUNWRITTEN(&imap))) { | |
1459 | xfs_map_buffer(inode, bh_result, &imap, offset); | |
1460 | if (ISUNWRITTEN(&imap)) | |
1da177e4 | 1461 | set_buffer_unwritten(bh_result); |
a719370b DC |
1462 | /* direct IO needs special help */ |
1463 | if (create && direct) | |
1464 | xfs_map_direct(inode, bh_result, &imap, offset); | |
1da177e4 LT |
1465 | } |
1466 | ||
c2536668 NS |
1467 | /* |
1468 | * If this is a realtime file, data may be on a different device. | |
1469 | * to that pointed to from the buffer_head b_bdev currently. | |
1470 | */ | |
046f1685 | 1471 | bh_result->b_bdev = xfs_find_bdev_for_inode(inode); |
1da177e4 | 1472 | |
c2536668 | 1473 | /* |
549054af DC |
1474 | * If we previously allocated a block out beyond eof and we are now |
1475 | * coming back to use it then we will need to flag it as new even if it | |
1476 | * has a disk address. | |
1477 | * | |
1478 | * With sub-block writes into unwritten extents we also need to mark | |
1479 | * the buffer as new so that the unwritten parts of the buffer gets | |
1480 | * correctly zeroed. | |
1da177e4 LT |
1481 | */ |
1482 | if (create && | |
1483 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
549054af | 1484 | (offset >= i_size_read(inode)) || |
207d0416 | 1485 | (new || ISUNWRITTEN(&imap)))) |
1da177e4 | 1486 | set_buffer_new(bh_result); |
1da177e4 | 1487 | |
207d0416 | 1488 | if (imap.br_startblock == DELAYSTARTBLOCK) { |
1da177e4 LT |
1489 | BUG_ON(direct); |
1490 | if (create) { | |
1491 | set_buffer_uptodate(bh_result); | |
1492 | set_buffer_mapped(bh_result); | |
1493 | set_buffer_delay(bh_result); | |
1494 | } | |
1495 | } | |
1496 | ||
1da177e4 | 1497 | return 0; |
a206c817 CH |
1498 | |
1499 | out_unlock: | |
1500 | xfs_iunlock(ip, lockmode); | |
2451337d | 1501 | return error; |
1da177e4 LT |
1502 | } |
1503 | ||
1504 | int | |
c2536668 | 1505 | xfs_get_blocks( |
1da177e4 LT |
1506 | struct inode *inode, |
1507 | sector_t iblock, | |
1508 | struct buffer_head *bh_result, | |
1509 | int create) | |
1510 | { | |
f2bde9b8 | 1511 | return __xfs_get_blocks(inode, iblock, bh_result, create, 0); |
1da177e4 LT |
1512 | } |
1513 | ||
1514 | STATIC int | |
e4c573bb | 1515 | xfs_get_blocks_direct( |
1da177e4 LT |
1516 | struct inode *inode, |
1517 | sector_t iblock, | |
1da177e4 LT |
1518 | struct buffer_head *bh_result, |
1519 | int create) | |
1520 | { | |
f2bde9b8 | 1521 | return __xfs_get_blocks(inode, iblock, bh_result, create, 1); |
1da177e4 LT |
1522 | } |
1523 | ||
209fb87a CH |
1524 | /* |
1525 | * Complete a direct I/O write request. | |
1526 | * | |
a06c277a DC |
1527 | * The ioend structure is passed from __xfs_get_blocks() to tell us what to do. |
1528 | * If no ioend exists (i.e. @private == NULL) then the write IO is an overwrite | |
1529 | * wholly within the EOF and so there is nothing for us to do. Note that in this | |
1530 | * case the completion can be called in interrupt context, whereas if we have an | |
1531 | * ioend we will always be called in task context (i.e. from a workqueue). | |
209fb87a | 1532 | */ |
f0973863 | 1533 | STATIC void |
209fb87a CH |
1534 | xfs_end_io_direct_write( |
1535 | struct kiocb *iocb, | |
1536 | loff_t offset, | |
1537 | ssize_t size, | |
7b7a8665 | 1538 | void *private) |
f0973863 | 1539 | { |
2ba66237 CH |
1540 | struct inode *inode = file_inode(iocb->ki_filp); |
1541 | struct xfs_inode *ip = XFS_I(inode); | |
1542 | struct xfs_mount *mp = ip->i_mount; | |
d5cc2e3f | 1543 | struct xfs_ioend *ioend = private; |
2ba66237 | 1544 | |
a06c277a DC |
1545 | trace_xfs_gbmap_direct_endio(ip, offset, size, |
1546 | ioend ? ioend->io_type : 0, NULL); | |
1547 | ||
1548 | if (!ioend) { | |
1549 | ASSERT(offset + size <= i_size_read(inode)); | |
2ba66237 | 1550 | return; |
a06c277a | 1551 | } |
6dfa1b67 | 1552 | |
2ba66237 | 1553 | if (XFS_FORCED_SHUTDOWN(mp)) |
6dfa1b67 | 1554 | goto out_end_io; |
f0973863 | 1555 | |
2813d682 | 1556 | /* |
d5cc2e3f DC |
1557 | * dio completion end_io functions are only called on writes if more |
1558 | * than 0 bytes was written. | |
2813d682 | 1559 | */ |
d5cc2e3f DC |
1560 | ASSERT(size > 0); |
1561 | ||
1562 | /* | |
1563 | * The ioend only maps whole blocks, while the IO may be sector aligned. | |
a06c277a DC |
1564 | * Hence the ioend offset/size may not match the IO offset/size exactly. |
1565 | * Because we don't map overwrites within EOF into the ioend, the offset | |
1566 | * may not match, but only if the endio spans EOF. Either way, write | |
1567 | * the IO sizes into the ioend so that completion processing does the | |
1568 | * right thing. | |
d5cc2e3f | 1569 | */ |
d5cc2e3f DC |
1570 | ASSERT(offset + size <= ioend->io_offset + ioend->io_size); |
1571 | ioend->io_size = size; | |
1572 | ioend->io_offset = offset; | |
f0973863 | 1573 | |
2813d682 | 1574 | /* |
6dfa1b67 DC |
1575 | * The ioend tells us whether we are doing unwritten extent conversion |
1576 | * or an append transaction that updates the on-disk file size. These | |
1577 | * cases are the only cases where we should *potentially* be needing | |
a06c277a | 1578 | * to update the VFS inode size. |
6dfa1b67 DC |
1579 | * |
1580 | * We need to update the in-core inode size here so that we don't end up | |
a06c277a DC |
1581 | * with the on-disk inode size being outside the in-core inode size. We |
1582 | * have no other method of updating EOF for AIO, so always do it here | |
1583 | * if necessary. | |
b9d59846 DC |
1584 | * |
1585 | * We need to lock the test/set EOF update as we can be racing with | |
1586 | * other IO completions here to update the EOF. Failing to serialise | |
1587 | * here can result in EOF moving backwards and Bad Things Happen when | |
1588 | * that occurs. | |
2813d682 | 1589 | */ |
b9d59846 | 1590 | spin_lock(&ip->i_flags_lock); |
2ba66237 CH |
1591 | if (offset + size > i_size_read(inode)) |
1592 | i_size_write(inode, offset + size); | |
b9d59846 | 1593 | spin_unlock(&ip->i_flags_lock); |
2813d682 | 1594 | |
f0973863 | 1595 | /* |
6dfa1b67 DC |
1596 | * If we are doing an append IO that needs to update the EOF on disk, |
1597 | * do the transaction reserve now so we can use common end io | |
1598 | * processing. Stashing the error (if there is one) in the ioend will | |
1599 | * result in the ioend processing passing on the error if it is | |
1600 | * possible as we can't return it from here. | |
f0973863 | 1601 | */ |
a06c277a | 1602 | if (ioend->io_type == XFS_IO_OVERWRITE) |
6dfa1b67 | 1603 | ioend->io_error = xfs_setfilesize_trans_alloc(ioend); |
209fb87a | 1604 | |
6dfa1b67 DC |
1605 | out_end_io: |
1606 | xfs_end_io(&ioend->io_work); | |
1607 | return; | |
f0973863 CH |
1608 | } |
1609 | ||
1da177e4 | 1610 | STATIC ssize_t |
e4c573bb | 1611 | xfs_vm_direct_IO( |
1da177e4 | 1612 | struct kiocb *iocb, |
d8d3d94b AV |
1613 | struct iov_iter *iter, |
1614 | loff_t offset) | |
1da177e4 | 1615 | { |
209fb87a CH |
1616 | struct inode *inode = iocb->ki_filp->f_mapping->host; |
1617 | struct block_device *bdev = xfs_find_bdev_for_inode(inode); | |
209fb87a | 1618 | |
6f673763 | 1619 | if (iov_iter_rw(iter) == WRITE) { |
17f8c842 OS |
1620 | return __blockdev_direct_IO(iocb, inode, bdev, iter, offset, |
1621 | xfs_get_blocks_direct, | |
9862f62f CH |
1622 | xfs_end_io_direct_write, NULL, |
1623 | DIO_ASYNC_EXTEND); | |
209fb87a | 1624 | } |
17f8c842 OS |
1625 | return __blockdev_direct_IO(iocb, inode, bdev, iter, offset, |
1626 | xfs_get_blocks_direct, NULL, NULL, 0); | |
1da177e4 LT |
1627 | } |
1628 | ||
d3bc815a DC |
1629 | /* |
1630 | * Punch out the delalloc blocks we have already allocated. | |
1631 | * | |
1632 | * Don't bother with xfs_setattr given that nothing can have made it to disk yet | |
1633 | * as the page is still locked at this point. | |
1634 | */ | |
1635 | STATIC void | |
1636 | xfs_vm_kill_delalloc_range( | |
1637 | struct inode *inode, | |
1638 | loff_t start, | |
1639 | loff_t end) | |
1640 | { | |
1641 | struct xfs_inode *ip = XFS_I(inode); | |
1642 | xfs_fileoff_t start_fsb; | |
1643 | xfs_fileoff_t end_fsb; | |
1644 | int error; | |
1645 | ||
1646 | start_fsb = XFS_B_TO_FSB(ip->i_mount, start); | |
1647 | end_fsb = XFS_B_TO_FSB(ip->i_mount, end); | |
1648 | if (end_fsb <= start_fsb) | |
1649 | return; | |
1650 | ||
1651 | xfs_ilock(ip, XFS_ILOCK_EXCL); | |
1652 | error = xfs_bmap_punch_delalloc_range(ip, start_fsb, | |
1653 | end_fsb - start_fsb); | |
1654 | if (error) { | |
1655 | /* something screwed, just bail */ | |
1656 | if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) { | |
1657 | xfs_alert(ip->i_mount, | |
1658 | "xfs_vm_write_failed: unable to clean up ino %lld", | |
1659 | ip->i_ino); | |
1660 | } | |
1661 | } | |
1662 | xfs_iunlock(ip, XFS_ILOCK_EXCL); | |
1663 | } | |
1664 | ||
fa9b227e CH |
1665 | STATIC void |
1666 | xfs_vm_write_failed( | |
d3bc815a DC |
1667 | struct inode *inode, |
1668 | struct page *page, | |
1669 | loff_t pos, | |
1670 | unsigned len) | |
fa9b227e | 1671 | { |
58e59854 | 1672 | loff_t block_offset; |
d3bc815a DC |
1673 | loff_t block_start; |
1674 | loff_t block_end; | |
1675 | loff_t from = pos & (PAGE_CACHE_SIZE - 1); | |
1676 | loff_t to = from + len; | |
1677 | struct buffer_head *bh, *head; | |
fa9b227e | 1678 | |
58e59854 JL |
1679 | /* |
1680 | * The request pos offset might be 32 or 64 bit, this is all fine | |
1681 | * on 64-bit platform. However, for 64-bit pos request on 32-bit | |
1682 | * platform, the high 32-bit will be masked off if we evaluate the | |
1683 | * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is | |
1684 | * 0xfffff000 as an unsigned long, hence the result is incorrect | |
1685 | * which could cause the following ASSERT failed in most cases. | |
1686 | * In order to avoid this, we can evaluate the block_offset of the | |
1687 | * start of the page by using shifts rather than masks the mismatch | |
1688 | * problem. | |
1689 | */ | |
1690 | block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT; | |
1691 | ||
d3bc815a | 1692 | ASSERT(block_offset + from == pos); |
c726de44 | 1693 | |
d3bc815a DC |
1694 | head = page_buffers(page); |
1695 | block_start = 0; | |
1696 | for (bh = head; bh != head || !block_start; | |
1697 | bh = bh->b_this_page, block_start = block_end, | |
1698 | block_offset += bh->b_size) { | |
1699 | block_end = block_start + bh->b_size; | |
c726de44 | 1700 | |
d3bc815a DC |
1701 | /* skip buffers before the write */ |
1702 | if (block_end <= from) | |
1703 | continue; | |
1704 | ||
1705 | /* if the buffer is after the write, we're done */ | |
1706 | if (block_start >= to) | |
1707 | break; | |
1708 | ||
1709 | if (!buffer_delay(bh)) | |
1710 | continue; | |
1711 | ||
1712 | if (!buffer_new(bh) && block_offset < i_size_read(inode)) | |
1713 | continue; | |
1714 | ||
1715 | xfs_vm_kill_delalloc_range(inode, block_offset, | |
1716 | block_offset + bh->b_size); | |
4ab9ed57 DC |
1717 | |
1718 | /* | |
1719 | * This buffer does not contain data anymore. make sure anyone | |
1720 | * who finds it knows that for certain. | |
1721 | */ | |
1722 | clear_buffer_delay(bh); | |
1723 | clear_buffer_uptodate(bh); | |
1724 | clear_buffer_mapped(bh); | |
1725 | clear_buffer_new(bh); | |
1726 | clear_buffer_dirty(bh); | |
fa9b227e | 1727 | } |
d3bc815a | 1728 | |
fa9b227e CH |
1729 | } |
1730 | ||
d3bc815a DC |
1731 | /* |
1732 | * This used to call block_write_begin(), but it unlocks and releases the page | |
1733 | * on error, and we need that page to be able to punch stale delalloc blocks out | |
1734 | * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at | |
1735 | * the appropriate point. | |
1736 | */ | |
f51623b2 | 1737 | STATIC int |
d79689c7 | 1738 | xfs_vm_write_begin( |
f51623b2 | 1739 | struct file *file, |
d79689c7 NP |
1740 | struct address_space *mapping, |
1741 | loff_t pos, | |
1742 | unsigned len, | |
1743 | unsigned flags, | |
1744 | struct page **pagep, | |
1745 | void **fsdata) | |
f51623b2 | 1746 | { |
d3bc815a DC |
1747 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
1748 | struct page *page; | |
1749 | int status; | |
155130a4 | 1750 | |
d3bc815a DC |
1751 | ASSERT(len <= PAGE_CACHE_SIZE); |
1752 | ||
ad22c7a0 | 1753 | page = grab_cache_page_write_begin(mapping, index, flags); |
d3bc815a DC |
1754 | if (!page) |
1755 | return -ENOMEM; | |
1756 | ||
1757 | status = __block_write_begin(page, pos, len, xfs_get_blocks); | |
1758 | if (unlikely(status)) { | |
1759 | struct inode *inode = mapping->host; | |
72ab70a1 | 1760 | size_t isize = i_size_read(inode); |
d3bc815a DC |
1761 | |
1762 | xfs_vm_write_failed(inode, page, pos, len); | |
1763 | unlock_page(page); | |
1764 | ||
72ab70a1 DC |
1765 | /* |
1766 | * If the write is beyond EOF, we only want to kill blocks | |
1767 | * allocated in this write, not blocks that were previously | |
1768 | * written successfully. | |
1769 | */ | |
1770 | if (pos + len > isize) { | |
1771 | ssize_t start = max_t(ssize_t, pos, isize); | |
1772 | ||
1773 | truncate_pagecache_range(inode, start, pos + len); | |
1774 | } | |
d3bc815a DC |
1775 | |
1776 | page_cache_release(page); | |
1777 | page = NULL; | |
1778 | } | |
1779 | ||
1780 | *pagep = page; | |
1781 | return status; | |
fa9b227e CH |
1782 | } |
1783 | ||
d3bc815a | 1784 | /* |
aad3f375 DC |
1785 | * On failure, we only need to kill delalloc blocks beyond EOF in the range of |
1786 | * this specific write because they will never be written. Previous writes | |
1787 | * beyond EOF where block allocation succeeded do not need to be trashed, so | |
1788 | * only new blocks from this write should be trashed. For blocks within | |
1789 | * EOF, generic_write_end() zeros them so they are safe to leave alone and be | |
1790 | * written with all the other valid data. | |
d3bc815a | 1791 | */ |
fa9b227e CH |
1792 | STATIC int |
1793 | xfs_vm_write_end( | |
1794 | struct file *file, | |
1795 | struct address_space *mapping, | |
1796 | loff_t pos, | |
1797 | unsigned len, | |
1798 | unsigned copied, | |
1799 | struct page *page, | |
1800 | void *fsdata) | |
1801 | { | |
1802 | int ret; | |
155130a4 | 1803 | |
d3bc815a DC |
1804 | ASSERT(len <= PAGE_CACHE_SIZE); |
1805 | ||
fa9b227e | 1806 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
d3bc815a DC |
1807 | if (unlikely(ret < len)) { |
1808 | struct inode *inode = mapping->host; | |
1809 | size_t isize = i_size_read(inode); | |
1810 | loff_t to = pos + len; | |
1811 | ||
1812 | if (to > isize) { | |
aad3f375 DC |
1813 | /* only kill blocks in this write beyond EOF */ |
1814 | if (pos > isize) | |
1815 | isize = pos; | |
d3bc815a | 1816 | xfs_vm_kill_delalloc_range(inode, isize, to); |
aad3f375 | 1817 | truncate_pagecache_range(inode, isize, to); |
d3bc815a DC |
1818 | } |
1819 | } | |
155130a4 | 1820 | return ret; |
f51623b2 | 1821 | } |
1da177e4 LT |
1822 | |
1823 | STATIC sector_t | |
e4c573bb | 1824 | xfs_vm_bmap( |
1da177e4 LT |
1825 | struct address_space *mapping, |
1826 | sector_t block) | |
1827 | { | |
1828 | struct inode *inode = (struct inode *)mapping->host; | |
739bfb2a | 1829 | struct xfs_inode *ip = XFS_I(inode); |
1da177e4 | 1830 | |
cca28fb8 | 1831 | trace_xfs_vm_bmap(XFS_I(inode)); |
126468b1 | 1832 | xfs_ilock(ip, XFS_IOLOCK_SHARED); |
4bc1ea6b | 1833 | filemap_write_and_wait(mapping); |
126468b1 | 1834 | xfs_iunlock(ip, XFS_IOLOCK_SHARED); |
c2536668 | 1835 | return generic_block_bmap(mapping, block, xfs_get_blocks); |
1da177e4 LT |
1836 | } |
1837 | ||
1838 | STATIC int | |
e4c573bb | 1839 | xfs_vm_readpage( |
1da177e4 LT |
1840 | struct file *unused, |
1841 | struct page *page) | |
1842 | { | |
c2536668 | 1843 | return mpage_readpage(page, xfs_get_blocks); |
1da177e4 LT |
1844 | } |
1845 | ||
1846 | STATIC int | |
e4c573bb | 1847 | xfs_vm_readpages( |
1da177e4 LT |
1848 | struct file *unused, |
1849 | struct address_space *mapping, | |
1850 | struct list_head *pages, | |
1851 | unsigned nr_pages) | |
1852 | { | |
c2536668 | 1853 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks); |
1da177e4 LT |
1854 | } |
1855 | ||
22e757a4 DC |
1856 | /* |
1857 | * This is basically a copy of __set_page_dirty_buffers() with one | |
1858 | * small tweak: buffers beyond EOF do not get marked dirty. If we mark them | |
1859 | * dirty, we'll never be able to clean them because we don't write buffers | |
1860 | * beyond EOF, and that means we can't invalidate pages that span EOF | |
1861 | * that have been marked dirty. Further, the dirty state can leak into | |
1862 | * the file interior if the file is extended, resulting in all sorts of | |
1863 | * bad things happening as the state does not match the underlying data. | |
1864 | * | |
1865 | * XXX: this really indicates that bufferheads in XFS need to die. Warts like | |
1866 | * this only exist because of bufferheads and how the generic code manages them. | |
1867 | */ | |
1868 | STATIC int | |
1869 | xfs_vm_set_page_dirty( | |
1870 | struct page *page) | |
1871 | { | |
1872 | struct address_space *mapping = page->mapping; | |
1873 | struct inode *inode = mapping->host; | |
1874 | loff_t end_offset; | |
1875 | loff_t offset; | |
1876 | int newly_dirty; | |
1877 | ||
1878 | if (unlikely(!mapping)) | |
1879 | return !TestSetPageDirty(page); | |
1880 | ||
1881 | end_offset = i_size_read(inode); | |
1882 | offset = page_offset(page); | |
1883 | ||
1884 | spin_lock(&mapping->private_lock); | |
1885 | if (page_has_buffers(page)) { | |
1886 | struct buffer_head *head = page_buffers(page); | |
1887 | struct buffer_head *bh = head; | |
1888 | ||
1889 | do { | |
1890 | if (offset < end_offset) | |
1891 | set_buffer_dirty(bh); | |
1892 | bh = bh->b_this_page; | |
1893 | offset += 1 << inode->i_blkbits; | |
1894 | } while (bh != head); | |
1895 | } | |
1896 | newly_dirty = !TestSetPageDirty(page); | |
1897 | spin_unlock(&mapping->private_lock); | |
1898 | ||
1899 | if (newly_dirty) { | |
1900 | /* sigh - __set_page_dirty() is static, so copy it here, too */ | |
1901 | unsigned long flags; | |
1902 | ||
1903 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
1904 | if (page->mapping) { /* Race with truncate? */ | |
1905 | WARN_ON_ONCE(!PageUptodate(page)); | |
1906 | account_page_dirtied(page, mapping); | |
1907 | radix_tree_tag_set(&mapping->page_tree, | |
1908 | page_index(page), PAGECACHE_TAG_DIRTY); | |
1909 | } | |
1910 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
1911 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1912 | } | |
1913 | return newly_dirty; | |
1914 | } | |
1915 | ||
f5e54d6e | 1916 | const struct address_space_operations xfs_address_space_operations = { |
e4c573bb NS |
1917 | .readpage = xfs_vm_readpage, |
1918 | .readpages = xfs_vm_readpages, | |
1919 | .writepage = xfs_vm_writepage, | |
7d4fb40a | 1920 | .writepages = xfs_vm_writepages, |
22e757a4 | 1921 | .set_page_dirty = xfs_vm_set_page_dirty, |
238f4c54 NS |
1922 | .releasepage = xfs_vm_releasepage, |
1923 | .invalidatepage = xfs_vm_invalidatepage, | |
d79689c7 | 1924 | .write_begin = xfs_vm_write_begin, |
fa9b227e | 1925 | .write_end = xfs_vm_write_end, |
e4c573bb NS |
1926 | .bmap = xfs_vm_bmap, |
1927 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1928 | .migratepage = buffer_migrate_page, |
bddaafa1 | 1929 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1930 | .error_remove_page = generic_error_remove_page, |
1da177e4 | 1931 | }; |