]>
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" |
a844f451 | 19 | #include "xfs_bit.h" |
1da177e4 | 20 | #include "xfs_log.h" |
a844f451 | 21 | #include "xfs_inum.h" |
1da177e4 | 22 | #include "xfs_sb.h" |
a844f451 | 23 | #include "xfs_ag.h" |
1da177e4 LT |
24 | #include "xfs_dir.h" |
25 | #include "xfs_dir2.h" | |
26 | #include "xfs_trans.h" | |
27 | #include "xfs_dmapi.h" | |
28 | #include "xfs_mount.h" | |
29 | #include "xfs_bmap_btree.h" | |
30 | #include "xfs_alloc_btree.h" | |
31 | #include "xfs_ialloc_btree.h" | |
1da177e4 LT |
32 | #include "xfs_dir_sf.h" |
33 | #include "xfs_dir2_sf.h" | |
a844f451 | 34 | #include "xfs_attr_sf.h" |
1da177e4 LT |
35 | #include "xfs_dinode.h" |
36 | #include "xfs_inode.h" | |
a844f451 NS |
37 | #include "xfs_alloc.h" |
38 | #include "xfs_btree.h" | |
1da177e4 LT |
39 | #include "xfs_error.h" |
40 | #include "xfs_rw.h" | |
41 | #include "xfs_iomap.h" | |
42 | #include <linux/mpage.h> | |
10ce4444 | 43 | #include <linux/pagevec.h> |
1da177e4 LT |
44 | #include <linux/writeback.h> |
45 | ||
46 | STATIC void xfs_count_page_state(struct page *, int *, int *, int *); | |
1da177e4 LT |
47 | |
48 | #if defined(XFS_RW_TRACE) | |
49 | void | |
50 | xfs_page_trace( | |
51 | int tag, | |
52 | struct inode *inode, | |
53 | struct page *page, | |
54 | int mask) | |
55 | { | |
56 | xfs_inode_t *ip; | |
1da177e4 LT |
57 | vnode_t *vp = LINVFS_GET_VP(inode); |
58 | loff_t isize = i_size_read(inode); | |
f6d6d4fc | 59 | loff_t offset = page_offset(page); |
1da177e4 LT |
60 | int delalloc = -1, unmapped = -1, unwritten = -1; |
61 | ||
62 | if (page_has_buffers(page)) | |
63 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
64 | ||
75e17b3c | 65 | ip = xfs_vtoi(vp); |
1da177e4 LT |
66 | if (!ip->i_rwtrace) |
67 | return; | |
68 | ||
69 | ktrace_enter(ip->i_rwtrace, | |
70 | (void *)((unsigned long)tag), | |
71 | (void *)ip, | |
72 | (void *)inode, | |
73 | (void *)page, | |
74 | (void *)((unsigned long)mask), | |
75 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), | |
76 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
77 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
78 | (void *)((unsigned long)(isize & 0xffffffff)), | |
79 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
80 | (void *)((unsigned long)(offset & 0xffffffff)), | |
81 | (void *)((unsigned long)delalloc), | |
82 | (void *)((unsigned long)unmapped), | |
83 | (void *)((unsigned long)unwritten), | |
84 | (void *)NULL, | |
85 | (void *)NULL); | |
86 | } | |
87 | #else | |
88 | #define xfs_page_trace(tag, inode, page, mask) | |
89 | #endif | |
90 | ||
0829c360 CH |
91 | /* |
92 | * Schedule IO completion handling on a xfsdatad if this was | |
93 | * the final hold on this ioend. | |
94 | */ | |
95 | STATIC void | |
96 | xfs_finish_ioend( | |
97 | xfs_ioend_t *ioend) | |
98 | { | |
99 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
100 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
101 | } | |
102 | ||
f6d6d4fc CH |
103 | /* |
104 | * We're now finished for good with this ioend structure. | |
105 | * Update the page state via the associated buffer_heads, | |
106 | * release holds on the inode and bio, and finally free | |
107 | * up memory. Do not use the ioend after this. | |
108 | */ | |
0829c360 CH |
109 | STATIC void |
110 | xfs_destroy_ioend( | |
111 | xfs_ioend_t *ioend) | |
112 | { | |
f6d6d4fc CH |
113 | struct buffer_head *bh, *next; |
114 | ||
115 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
116 | next = bh->b_private; | |
117 | bh->b_end_io(bh, ioend->io_uptodate); | |
118 | } | |
119 | ||
0829c360 CH |
120 | vn_iowake(ioend->io_vnode); |
121 | mempool_free(ioend, xfs_ioend_pool); | |
122 | } | |
123 | ||
124 | /* | |
f6d6d4fc CH |
125 | * Buffered IO write completion for delayed allocate extents. |
126 | * TODO: Update ondisk isize now that we know the file data | |
127 | * has been flushed (i.e. the notorious "NULL file" problem). | |
128 | */ | |
129 | STATIC void | |
130 | xfs_end_bio_delalloc( | |
131 | void *data) | |
132 | { | |
133 | xfs_ioend_t *ioend = data; | |
134 | ||
135 | xfs_destroy_ioend(ioend); | |
136 | } | |
137 | ||
138 | /* | |
139 | * Buffered IO write completion for regular, written extents. | |
140 | */ | |
141 | STATIC void | |
142 | xfs_end_bio_written( | |
143 | void *data) | |
144 | { | |
145 | xfs_ioend_t *ioend = data; | |
146 | ||
147 | xfs_destroy_ioend(ioend); | |
148 | } | |
149 | ||
150 | /* | |
151 | * IO write completion for unwritten extents. | |
152 | * | |
0829c360 | 153 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 154 | * to written extents. |
0829c360 CH |
155 | */ |
156 | STATIC void | |
157 | xfs_end_bio_unwritten( | |
158 | void *data) | |
159 | { | |
160 | xfs_ioend_t *ioend = data; | |
161 | vnode_t *vp = ioend->io_vnode; | |
162 | xfs_off_t offset = ioend->io_offset; | |
163 | size_t size = ioend->io_size; | |
164 | int error; | |
165 | ||
166 | if (ioend->io_uptodate) | |
167 | VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error); | |
168 | xfs_destroy_ioend(ioend); | |
169 | } | |
170 | ||
171 | /* | |
172 | * Allocate and initialise an IO completion structure. | |
173 | * We need to track unwritten extent write completion here initially. | |
174 | * We'll need to extend this for updating the ondisk inode size later | |
175 | * (vs. incore size). | |
176 | */ | |
177 | STATIC xfs_ioend_t * | |
178 | xfs_alloc_ioend( | |
f6d6d4fc CH |
179 | struct inode *inode, |
180 | unsigned int type) | |
0829c360 CH |
181 | { |
182 | xfs_ioend_t *ioend; | |
183 | ||
184 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
185 | ||
186 | /* | |
187 | * Set the count to 1 initially, which will prevent an I/O | |
188 | * completion callback from happening before we have started | |
189 | * all the I/O from calling the completion routine too early. | |
190 | */ | |
191 | atomic_set(&ioend->io_remaining, 1); | |
192 | ioend->io_uptodate = 1; /* cleared if any I/O fails */ | |
f6d6d4fc CH |
193 | ioend->io_list = NULL; |
194 | ioend->io_type = type; | |
0829c360 | 195 | ioend->io_vnode = LINVFS_GET_VP(inode); |
c1a073bd | 196 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 197 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
198 | atomic_inc(&ioend->io_vnode->v_iocount); |
199 | ioend->io_offset = 0; | |
200 | ioend->io_size = 0; | |
201 | ||
f6d6d4fc CH |
202 | if (type == IOMAP_UNWRITTEN) |
203 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend); | |
204 | else if (type == IOMAP_DELAY) | |
205 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend); | |
206 | else | |
207 | INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend); | |
0829c360 CH |
208 | |
209 | return ioend; | |
210 | } | |
211 | ||
1da177e4 LT |
212 | STATIC int |
213 | xfs_map_blocks( | |
214 | struct inode *inode, | |
215 | loff_t offset, | |
216 | ssize_t count, | |
217 | xfs_iomap_t *mapp, | |
218 | int flags) | |
219 | { | |
220 | vnode_t *vp = LINVFS_GET_VP(inode); | |
221 | int error, nmaps = 1; | |
222 | ||
223 | VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error); | |
224 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) | |
225 | VMODIFY(vp); | |
226 | return -error; | |
227 | } | |
228 | ||
1defeac9 CH |
229 | STATIC inline int |
230 | xfs_iomap_valid( | |
1da177e4 | 231 | xfs_iomap_t *iomapp, |
1defeac9 | 232 | loff_t offset) |
1da177e4 | 233 | { |
1defeac9 CH |
234 | return offset >= iomapp->iomap_offset && |
235 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
236 | } |
237 | ||
f6d6d4fc CH |
238 | /* |
239 | * BIO completion handler for buffered IO. | |
240 | */ | |
241 | STATIC int | |
242 | xfs_end_bio( | |
243 | struct bio *bio, | |
244 | unsigned int bytes_done, | |
245 | int error) | |
246 | { | |
247 | xfs_ioend_t *ioend = bio->bi_private; | |
248 | ||
249 | if (bio->bi_size) | |
250 | return 1; | |
251 | ||
252 | ASSERT(ioend); | |
253 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); | |
254 | ||
255 | /* Toss bio and pass work off to an xfsdatad thread */ | |
256 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
257 | ioend->io_uptodate = 0; | |
258 | bio->bi_private = NULL; | |
259 | bio->bi_end_io = NULL; | |
260 | ||
261 | bio_put(bio); | |
262 | xfs_finish_ioend(ioend); | |
263 | return 0; | |
264 | } | |
265 | ||
266 | STATIC void | |
267 | xfs_submit_ioend_bio( | |
268 | xfs_ioend_t *ioend, | |
269 | struct bio *bio) | |
270 | { | |
271 | atomic_inc(&ioend->io_remaining); | |
272 | ||
273 | bio->bi_private = ioend; | |
274 | bio->bi_end_io = xfs_end_bio; | |
275 | ||
276 | submit_bio(WRITE, bio); | |
277 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
278 | bio_put(bio); | |
279 | } | |
280 | ||
281 | STATIC struct bio * | |
282 | xfs_alloc_ioend_bio( | |
283 | struct buffer_head *bh) | |
284 | { | |
285 | struct bio *bio; | |
286 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
287 | ||
288 | do { | |
289 | bio = bio_alloc(GFP_NOIO, nvecs); | |
290 | nvecs >>= 1; | |
291 | } while (!bio); | |
292 | ||
293 | ASSERT(bio->bi_private == NULL); | |
294 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
295 | bio->bi_bdev = bh->b_bdev; | |
296 | bio_get(bio); | |
297 | return bio; | |
298 | } | |
299 | ||
300 | STATIC void | |
301 | xfs_start_buffer_writeback( | |
302 | struct buffer_head *bh) | |
303 | { | |
304 | ASSERT(buffer_mapped(bh)); | |
305 | ASSERT(buffer_locked(bh)); | |
306 | ASSERT(!buffer_delay(bh)); | |
307 | ASSERT(!buffer_unwritten(bh)); | |
308 | ||
309 | mark_buffer_async_write(bh); | |
310 | set_buffer_uptodate(bh); | |
311 | clear_buffer_dirty(bh); | |
312 | } | |
313 | ||
314 | STATIC void | |
315 | xfs_start_page_writeback( | |
316 | struct page *page, | |
317 | struct writeback_control *wbc, | |
318 | int clear_dirty, | |
319 | int buffers) | |
320 | { | |
321 | ASSERT(PageLocked(page)); | |
322 | ASSERT(!PageWriteback(page)); | |
323 | set_page_writeback(page); | |
324 | if (clear_dirty) | |
325 | clear_page_dirty(page); | |
326 | unlock_page(page); | |
327 | if (!buffers) { | |
328 | end_page_writeback(page); | |
329 | wbc->pages_skipped++; /* We didn't write this page */ | |
330 | } | |
331 | } | |
332 | ||
333 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
334 | { | |
335 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
336 | } | |
337 | ||
338 | /* | |
d88992f6 DC |
339 | * Submit all of the bios for all of the ioends we have saved up, covering the |
340 | * initial writepage page and also any probed pages. | |
341 | * | |
342 | * Because we may have multiple ioends spanning a page, we need to start | |
343 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
344 | * buffers as we got, then we can end up with a page that only has buffers | |
345 | * marked async write and I/O complete on can occur before we mark the other | |
346 | * buffers async write. | |
347 | * | |
348 | * The end result of this is that we trip a bug in end_page_writeback() because | |
349 | * we call it twice for the one page as the code in end_buffer_async_write() | |
350 | * assumes that all buffers on the page are started at the same time. | |
351 | * | |
352 | * The fix is two passes across the ioend list - one to start writeback on the | |
353 | * bufferheads, and then the second one submit them for I/O. | |
f6d6d4fc CH |
354 | */ |
355 | STATIC void | |
356 | xfs_submit_ioend( | |
357 | xfs_ioend_t *ioend) | |
358 | { | |
d88992f6 | 359 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
360 | xfs_ioend_t *next; |
361 | struct buffer_head *bh; | |
362 | struct bio *bio; | |
363 | sector_t lastblock = 0; | |
364 | ||
d88992f6 DC |
365 | /* Pass 1 - start writeback */ |
366 | do { | |
367 | next = ioend->io_list; | |
368 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
369 | xfs_start_buffer_writeback(bh); | |
370 | } | |
371 | } while ((ioend = next) != NULL); | |
372 | ||
373 | /* Pass 2 - submit I/O */ | |
374 | ioend = head; | |
f6d6d4fc CH |
375 | do { |
376 | next = ioend->io_list; | |
377 | bio = NULL; | |
378 | ||
379 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
380 | |
381 | if (!bio) { | |
382 | retry: | |
383 | bio = xfs_alloc_ioend_bio(bh); | |
384 | } else if (bh->b_blocknr != lastblock + 1) { | |
385 | xfs_submit_ioend_bio(ioend, bio); | |
386 | goto retry; | |
387 | } | |
388 | ||
389 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
390 | xfs_submit_ioend_bio(ioend, bio); | |
391 | goto retry; | |
392 | } | |
393 | ||
394 | lastblock = bh->b_blocknr; | |
395 | } | |
396 | if (bio) | |
397 | xfs_submit_ioend_bio(ioend, bio); | |
398 | xfs_finish_ioend(ioend); | |
399 | } while ((ioend = next) != NULL); | |
400 | } | |
401 | ||
402 | /* | |
403 | * Cancel submission of all buffer_heads so far in this endio. | |
404 | * Toss the endio too. Only ever called for the initial page | |
405 | * in a writepage request, so only ever one page. | |
406 | */ | |
407 | STATIC void | |
408 | xfs_cancel_ioend( | |
409 | xfs_ioend_t *ioend) | |
410 | { | |
411 | xfs_ioend_t *next; | |
412 | struct buffer_head *bh, *next_bh; | |
413 | ||
414 | do { | |
415 | next = ioend->io_list; | |
416 | bh = ioend->io_buffer_head; | |
417 | do { | |
418 | next_bh = bh->b_private; | |
419 | clear_buffer_async_write(bh); | |
420 | unlock_buffer(bh); | |
421 | } while ((bh = next_bh) != NULL); | |
422 | ||
423 | vn_iowake(ioend->io_vnode); | |
424 | mempool_free(ioend, xfs_ioend_pool); | |
425 | } while ((ioend = next) != NULL); | |
426 | } | |
427 | ||
428 | /* | |
429 | * Test to see if we've been building up a completion structure for | |
430 | * earlier buffers -- if so, we try to append to this ioend if we | |
431 | * can, otherwise we finish off any current ioend and start another. | |
432 | * Return true if we've finished the given ioend. | |
433 | */ | |
434 | STATIC void | |
435 | xfs_add_to_ioend( | |
436 | struct inode *inode, | |
437 | struct buffer_head *bh, | |
7336cea8 | 438 | xfs_off_t offset, |
f6d6d4fc CH |
439 | unsigned int type, |
440 | xfs_ioend_t **result, | |
441 | int need_ioend) | |
442 | { | |
443 | xfs_ioend_t *ioend = *result; | |
444 | ||
445 | if (!ioend || need_ioend || type != ioend->io_type) { | |
446 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 447 | |
f6d6d4fc CH |
448 | ioend = xfs_alloc_ioend(inode, type); |
449 | ioend->io_offset = offset; | |
450 | ioend->io_buffer_head = bh; | |
451 | ioend->io_buffer_tail = bh; | |
452 | if (previous) | |
453 | previous->io_list = ioend; | |
454 | *result = ioend; | |
455 | } else { | |
456 | ioend->io_buffer_tail->b_private = bh; | |
457 | ioend->io_buffer_tail = bh; | |
458 | } | |
459 | ||
460 | bh->b_private = NULL; | |
461 | ioend->io_size += bh->b_size; | |
462 | } | |
463 | ||
1da177e4 LT |
464 | STATIC void |
465 | xfs_map_at_offset( | |
1da177e4 | 466 | struct buffer_head *bh, |
1defeac9 | 467 | loff_t offset, |
1da177e4 | 468 | int block_bits, |
1defeac9 | 469 | xfs_iomap_t *iomapp) |
1da177e4 LT |
470 | { |
471 | xfs_daddr_t bn; | |
1da177e4 LT |
472 | int sector_shift; |
473 | ||
474 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); | |
475 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
476 | ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL); | |
477 | ||
1da177e4 | 478 | sector_shift = block_bits - BBSHIFT; |
1defeac9 CH |
479 | bn = (iomapp->iomap_bn >> sector_shift) + |
480 | ((offset - iomapp->iomap_offset) >> block_bits); | |
481 | ||
482 | ASSERT(bn || (iomapp->iomap_flags & IOMAP_REALTIME)); | |
1da177e4 LT |
483 | ASSERT((bn << sector_shift) >= iomapp->iomap_bn); |
484 | ||
485 | lock_buffer(bh); | |
486 | bh->b_blocknr = bn; | |
ce8e922c | 487 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
488 | set_buffer_mapped(bh); |
489 | clear_buffer_delay(bh); | |
f6d6d4fc | 490 | clear_buffer_unwritten(bh); |
1da177e4 LT |
491 | } |
492 | ||
493 | /* | |
6c4fe19f | 494 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
495 | */ |
496 | STATIC unsigned int | |
6c4fe19f | 497 | xfs_probe_page( |
10ce4444 | 498 | struct page *page, |
6c4fe19f CH |
499 | unsigned int pg_offset, |
500 | int mapped) | |
1da177e4 | 501 | { |
1da177e4 LT |
502 | int ret = 0; |
503 | ||
1da177e4 | 504 | if (PageWriteback(page)) |
10ce4444 | 505 | return 0; |
1da177e4 LT |
506 | |
507 | if (page->mapping && PageDirty(page)) { | |
508 | if (page_has_buffers(page)) { | |
509 | struct buffer_head *bh, *head; | |
510 | ||
511 | bh = head = page_buffers(page); | |
512 | do { | |
6c4fe19f CH |
513 | if (!buffer_uptodate(bh)) |
514 | break; | |
515 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
516 | break; |
517 | ret += bh->b_size; | |
518 | if (ret >= pg_offset) | |
519 | break; | |
520 | } while ((bh = bh->b_this_page) != head); | |
521 | } else | |
6c4fe19f | 522 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
523 | } |
524 | ||
1da177e4 LT |
525 | return ret; |
526 | } | |
527 | ||
f6d6d4fc | 528 | STATIC size_t |
6c4fe19f | 529 | xfs_probe_cluster( |
1da177e4 LT |
530 | struct inode *inode, |
531 | struct page *startpage, | |
532 | struct buffer_head *bh, | |
6c4fe19f CH |
533 | struct buffer_head *head, |
534 | int mapped) | |
1da177e4 | 535 | { |
10ce4444 | 536 | struct pagevec pvec; |
1da177e4 | 537 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
538 | size_t total = 0; |
539 | int done = 0, i; | |
1da177e4 LT |
540 | |
541 | /* First sum forwards in this page */ | |
542 | do { | |
6c4fe19f | 543 | if (mapped != buffer_mapped(bh)) |
10ce4444 | 544 | return total; |
1da177e4 LT |
545 | total += bh->b_size; |
546 | } while ((bh = bh->b_this_page) != head); | |
547 | ||
10ce4444 CH |
548 | /* if we reached the end of the page, sum forwards in following pages */ |
549 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
550 | tindex = startpage->index + 1; | |
551 | ||
552 | /* Prune this back to avoid pathological behavior */ | |
553 | tloff = min(tlast, startpage->index + 64); | |
554 | ||
555 | pagevec_init(&pvec, 0); | |
556 | while (!done && tindex <= tloff) { | |
557 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
558 | ||
559 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
560 | break; | |
561 | ||
562 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
563 | struct page *page = pvec.pages[i]; | |
564 | size_t pg_offset, len = 0; | |
565 | ||
566 | if (tindex == tlast) { | |
567 | pg_offset = | |
568 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
569 | if (!pg_offset) { |
570 | done = 1; | |
10ce4444 | 571 | break; |
1defeac9 | 572 | } |
10ce4444 CH |
573 | } else |
574 | pg_offset = PAGE_CACHE_SIZE; | |
575 | ||
576 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
6c4fe19f | 577 | len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
578 | unlock_page(page); |
579 | } | |
580 | ||
581 | if (!len) { | |
582 | done = 1; | |
583 | break; | |
584 | } | |
585 | ||
1da177e4 | 586 | total += len; |
1defeac9 | 587 | tindex++; |
1da177e4 | 588 | } |
10ce4444 CH |
589 | |
590 | pagevec_release(&pvec); | |
591 | cond_resched(); | |
1da177e4 | 592 | } |
10ce4444 | 593 | |
1da177e4 LT |
594 | return total; |
595 | } | |
596 | ||
597 | /* | |
10ce4444 CH |
598 | * Test if a given page is suitable for writing as part of an unwritten |
599 | * or delayed allocate extent. | |
1da177e4 | 600 | */ |
10ce4444 CH |
601 | STATIC int |
602 | xfs_is_delayed_page( | |
603 | struct page *page, | |
f6d6d4fc | 604 | unsigned int type) |
1da177e4 | 605 | { |
1da177e4 | 606 | if (PageWriteback(page)) |
10ce4444 | 607 | return 0; |
1da177e4 LT |
608 | |
609 | if (page->mapping && page_has_buffers(page)) { | |
610 | struct buffer_head *bh, *head; | |
611 | int acceptable = 0; | |
612 | ||
613 | bh = head = page_buffers(page); | |
614 | do { | |
f6d6d4fc CH |
615 | if (buffer_unwritten(bh)) |
616 | acceptable = (type == IOMAP_UNWRITTEN); | |
617 | else if (buffer_delay(bh)) | |
618 | acceptable = (type == IOMAP_DELAY); | |
6c4fe19f CH |
619 | else if (buffer_mapped(bh)) |
620 | acceptable = (type == 0); | |
f6d6d4fc | 621 | else |
1da177e4 | 622 | break; |
1da177e4 LT |
623 | } while ((bh = bh->b_this_page) != head); |
624 | ||
625 | if (acceptable) | |
10ce4444 | 626 | return 1; |
1da177e4 LT |
627 | } |
628 | ||
10ce4444 | 629 | return 0; |
1da177e4 LT |
630 | } |
631 | ||
1da177e4 LT |
632 | /* |
633 | * Allocate & map buffers for page given the extent map. Write it out. | |
634 | * except for the original page of a writepage, this is called on | |
635 | * delalloc/unwritten pages only, for the original page it is possible | |
636 | * that the page has no mapping at all. | |
637 | */ | |
f6d6d4fc | 638 | STATIC int |
1da177e4 LT |
639 | xfs_convert_page( |
640 | struct inode *inode, | |
641 | struct page *page, | |
10ce4444 | 642 | loff_t tindex, |
1defeac9 | 643 | xfs_iomap_t *mp, |
f6d6d4fc | 644 | xfs_ioend_t **ioendp, |
1da177e4 | 645 | struct writeback_control *wbc, |
1da177e4 LT |
646 | int startio, |
647 | int all_bh) | |
648 | { | |
f6d6d4fc | 649 | struct buffer_head *bh, *head; |
9260dc6b CH |
650 | xfs_off_t end_offset; |
651 | unsigned long p_offset; | |
f6d6d4fc | 652 | unsigned int type; |
1da177e4 | 653 | int bbits = inode->i_blkbits; |
24e17b5f | 654 | int len, page_dirty; |
f6d6d4fc | 655 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 656 | xfs_off_t offset = page_offset(page); |
1da177e4 | 657 | |
10ce4444 CH |
658 | if (page->index != tindex) |
659 | goto fail; | |
660 | if (TestSetPageLocked(page)) | |
661 | goto fail; | |
662 | if (PageWriteback(page)) | |
663 | goto fail_unlock_page; | |
664 | if (page->mapping != inode->i_mapping) | |
665 | goto fail_unlock_page; | |
666 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
667 | goto fail_unlock_page; | |
668 | ||
24e17b5f NS |
669 | /* |
670 | * page_dirty is initially a count of buffers on the page before | |
671 | * EOF and is decrememted as we move each into a cleanable state. | |
9260dc6b CH |
672 | * |
673 | * Derivation: | |
674 | * | |
675 | * End offset is the highest offset that this page should represent. | |
676 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
677 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
678 | * hence give us the correct page_dirty count. On any other page, | |
679 | * it will be zero and in that case we need page_dirty to be the | |
680 | * count of buffers on the page. | |
24e17b5f | 681 | */ |
9260dc6b CH |
682 | end_offset = min_t(unsigned long long, |
683 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
684 | i_size_read(inode)); | |
685 | ||
24e17b5f | 686 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
687 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
688 | PAGE_CACHE_SIZE); | |
689 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
690 | page_dirty = p_offset / len; | |
24e17b5f | 691 | |
1da177e4 LT |
692 | bh = head = page_buffers(page); |
693 | do { | |
9260dc6b | 694 | if (offset >= end_offset) |
1da177e4 | 695 | break; |
f6d6d4fc CH |
696 | if (!buffer_uptodate(bh)) |
697 | uptodate = 0; | |
698 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
699 | done = 1; | |
1da177e4 | 700 | continue; |
f6d6d4fc CH |
701 | } |
702 | ||
9260dc6b CH |
703 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
704 | if (buffer_unwritten(bh)) | |
705 | type = IOMAP_UNWRITTEN; | |
706 | else | |
707 | type = IOMAP_DELAY; | |
708 | ||
709 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 710 | done = 1; |
9260dc6b CH |
711 | continue; |
712 | } | |
713 | ||
714 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
715 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
716 | ||
717 | xfs_map_at_offset(bh, offset, bbits, mp); | |
718 | if (startio) { | |
7336cea8 | 719 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
720 | type, ioendp, done); |
721 | } else { | |
722 | set_buffer_dirty(bh); | |
723 | unlock_buffer(bh); | |
724 | mark_buffer_dirty(bh); | |
725 | } | |
726 | page_dirty--; | |
727 | count++; | |
728 | } else { | |
729 | type = 0; | |
730 | if (buffer_mapped(bh) && all_bh && startio) { | |
1da177e4 | 731 | lock_buffer(bh); |
7336cea8 | 732 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
733 | type, ioendp, done); |
734 | count++; | |
24e17b5f | 735 | page_dirty--; |
9260dc6b CH |
736 | } else { |
737 | done = 1; | |
1da177e4 | 738 | } |
1da177e4 | 739 | } |
7336cea8 | 740 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 741 | |
f6d6d4fc CH |
742 | if (uptodate && bh == head) |
743 | SetPageUptodate(page); | |
744 | ||
745 | if (startio) { | |
f5e596bb CH |
746 | if (count) { |
747 | struct backing_dev_info *bdi; | |
748 | ||
749 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 750 | wbc->nr_to_write--; |
f5e596bb CH |
751 | if (bdi_write_congested(bdi)) { |
752 | wbc->encountered_congestion = 1; | |
753 | done = 1; | |
9fddaca2 | 754 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
755 | done = 1; |
756 | } | |
757 | } | |
f6d6d4fc | 758 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 759 | } |
f6d6d4fc CH |
760 | |
761 | return done; | |
10ce4444 CH |
762 | fail_unlock_page: |
763 | unlock_page(page); | |
764 | fail: | |
765 | return 1; | |
1da177e4 LT |
766 | } |
767 | ||
768 | /* | |
769 | * Convert & write out a cluster of pages in the same extent as defined | |
770 | * by mp and following the start page. | |
771 | */ | |
772 | STATIC void | |
773 | xfs_cluster_write( | |
774 | struct inode *inode, | |
775 | pgoff_t tindex, | |
776 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 777 | xfs_ioend_t **ioendp, |
1da177e4 LT |
778 | struct writeback_control *wbc, |
779 | int startio, | |
780 | int all_bh, | |
781 | pgoff_t tlast) | |
782 | { | |
10ce4444 CH |
783 | struct pagevec pvec; |
784 | int done = 0, i; | |
1da177e4 | 785 | |
10ce4444 CH |
786 | pagevec_init(&pvec, 0); |
787 | while (!done && tindex <= tlast) { | |
788 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
789 | ||
790 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 791 | break; |
10ce4444 CH |
792 | |
793 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
794 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
795 | iomapp, ioendp, wbc, startio, all_bh); | |
796 | if (done) | |
797 | break; | |
798 | } | |
799 | ||
800 | pagevec_release(&pvec); | |
801 | cond_resched(); | |
1da177e4 LT |
802 | } |
803 | } | |
804 | ||
805 | /* | |
806 | * Calling this without startio set means we are being asked to make a dirty | |
807 | * page ready for freeing it's buffers. When called with startio set then | |
808 | * we are coming from writepage. | |
809 | * | |
810 | * When called with startio set it is important that we write the WHOLE | |
811 | * page if possible. | |
812 | * The bh->b_state's cannot know if any of the blocks or which block for | |
813 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
814 | * only vaild if the page itself isn't completely uptodate. Some layers | |
815 | * may clear the page dirty flag prior to calling write page, under the | |
816 | * assumption the entire page will be written out; by not writing out the | |
817 | * whole page the page can be reused before all valid dirty data is | |
818 | * written out. Note: in the case of a page that has been dirty'd by | |
819 | * mapwrite and but partially setup by block_prepare_write the | |
820 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
821 | * valid state, thus the whole page must be written out thing. | |
822 | */ | |
823 | ||
824 | STATIC int | |
825 | xfs_page_state_convert( | |
826 | struct inode *inode, | |
827 | struct page *page, | |
828 | struct writeback_control *wbc, | |
829 | int startio, | |
830 | int unmapped) /* also implies page uptodate */ | |
831 | { | |
f6d6d4fc | 832 | struct buffer_head *bh, *head; |
1defeac9 | 833 | xfs_iomap_t iomap; |
f6d6d4fc | 834 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
835 | loff_t offset; |
836 | unsigned long p_offset = 0; | |
f6d6d4fc | 837 | unsigned int type; |
1da177e4 LT |
838 | __uint64_t end_offset; |
839 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
840 | ssize_t size, len; |
841 | int flags, err, iomap_valid = 0, uptodate = 1; | |
f6d6d4fc | 842 | int page_dirty, count = 0, trylock_flag = 0; |
6c4fe19f | 843 | int all_bh = unmapped; |
1da177e4 | 844 | |
3ba0815a | 845 | /* wait for other IO threads? */ |
f5e596bb | 846 | if (startio && (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)) |
f6d6d4fc | 847 | trylock_flag |= BMAPI_TRYLOCK; |
3ba0815a | 848 | |
1da177e4 LT |
849 | /* Is this page beyond the end of the file? */ |
850 | offset = i_size_read(inode); | |
851 | end_index = offset >> PAGE_CACHE_SHIFT; | |
852 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
853 | if (page->index >= end_index) { | |
854 | if ((page->index >= end_index + 1) || | |
855 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
856 | if (startio) |
857 | unlock_page(page); | |
858 | return 0; | |
1da177e4 LT |
859 | } |
860 | } | |
861 | ||
1da177e4 | 862 | /* |
24e17b5f NS |
863 | * page_dirty is initially a count of buffers on the page before |
864 | * EOF and is decrememted as we move each into a cleanable state. | |
f6d6d4fc CH |
865 | * |
866 | * Derivation: | |
867 | * | |
868 | * End offset is the highest offset that this page should represent. | |
869 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
870 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
871 | * hence give us the correct page_dirty count. On any other page, | |
872 | * it will be zero and in that case we need page_dirty to be the | |
873 | * count of buffers on the page. | |
874 | */ | |
875 | end_offset = min_t(unsigned long long, | |
876 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 877 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
878 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
879 | PAGE_CACHE_SIZE); | |
880 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
881 | page_dirty = p_offset / len; |
882 | ||
24e17b5f | 883 | bh = head = page_buffers(page); |
f6d6d4fc | 884 | offset = page_offset(page); |
6c4fe19f CH |
885 | flags = -1; |
886 | type = 0; | |
f6d6d4fc | 887 | |
f6d6d4fc | 888 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
889 | |
890 | do { | |
891 | if (offset >= end_offset) | |
892 | break; | |
893 | if (!buffer_uptodate(bh)) | |
894 | uptodate = 0; | |
f6d6d4fc | 895 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
896 | /* |
897 | * the iomap is actually still valid, but the ioend | |
898 | * isn't. shouldn't happen too often. | |
899 | */ | |
900 | iomap_valid = 0; | |
1da177e4 | 901 | continue; |
f6d6d4fc | 902 | } |
1da177e4 | 903 | |
1defeac9 CH |
904 | if (iomap_valid) |
905 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
906 | |
907 | /* | |
908 | * First case, map an unwritten extent and prepare for | |
909 | * extent state conversion transaction on completion. | |
f6d6d4fc | 910 | * |
1da177e4 LT |
911 | * Second case, allocate space for a delalloc buffer. |
912 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
913 | * |
914 | * Third case, an unmapped buffer was found, and we are | |
915 | * in a path where we need to write the whole page out. | |
916 | */ | |
917 | if (buffer_unwritten(bh) || buffer_delay(bh) || | |
918 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
919 | !buffer_mapped(bh) && (unmapped || startio))) { | |
6c4fe19f CH |
920 | /* |
921 | * Make sure we don't use a read-only iomap | |
922 | */ | |
923 | if (flags == BMAPI_READ) | |
924 | iomap_valid = 0; | |
925 | ||
f6d6d4fc CH |
926 | if (buffer_unwritten(bh)) { |
927 | type = IOMAP_UNWRITTEN; | |
928 | flags = BMAPI_WRITE|BMAPI_IGNSTATE; | |
d5cb48aa | 929 | } else if (buffer_delay(bh)) { |
f6d6d4fc CH |
930 | type = IOMAP_DELAY; |
931 | flags = BMAPI_ALLOCATE; | |
932 | if (!startio) | |
933 | flags |= trylock_flag; | |
d5cb48aa | 934 | } else { |
6c4fe19f | 935 | type = IOMAP_NEW; |
d5cb48aa | 936 | flags = BMAPI_WRITE|BMAPI_MMAP; |
f6d6d4fc CH |
937 | } |
938 | ||
1defeac9 | 939 | if (!iomap_valid) { |
6c4fe19f CH |
940 | if (type == IOMAP_NEW) { |
941 | size = xfs_probe_cluster(inode, | |
942 | page, bh, head, 0); | |
d5cb48aa CH |
943 | } else { |
944 | size = len; | |
945 | } | |
946 | ||
947 | err = xfs_map_blocks(inode, offset, size, | |
948 | &iomap, flags); | |
f6d6d4fc | 949 | if (err) |
1da177e4 | 950 | goto error; |
1defeac9 | 951 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 952 | } |
1defeac9 CH |
953 | if (iomap_valid) { |
954 | xfs_map_at_offset(bh, offset, | |
955 | inode->i_blkbits, &iomap); | |
1da177e4 | 956 | if (startio) { |
7336cea8 | 957 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 CH |
958 | type, &ioend, |
959 | !iomap_valid); | |
1da177e4 LT |
960 | } else { |
961 | set_buffer_dirty(bh); | |
962 | unlock_buffer(bh); | |
963 | mark_buffer_dirty(bh); | |
964 | } | |
965 | page_dirty--; | |
f6d6d4fc | 966 | count++; |
1da177e4 | 967 | } |
d5cb48aa | 968 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
969 | /* |
970 | * we got here because the buffer is already mapped. | |
971 | * That means it must already have extents allocated | |
972 | * underneath it. Map the extent by reading it. | |
973 | */ | |
974 | if (!iomap_valid || type != 0) { | |
975 | flags = BMAPI_READ; | |
976 | size = xfs_probe_cluster(inode, page, bh, | |
977 | head, 1); | |
978 | err = xfs_map_blocks(inode, offset, size, | |
979 | &iomap, flags); | |
980 | if (err) | |
981 | goto error; | |
982 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
983 | } | |
d5cb48aa | 984 | |
6c4fe19f | 985 | type = 0; |
d5cb48aa CH |
986 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
987 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
988 | if (iomap_valid) |
989 | all_bh = 1; | |
7336cea8 | 990 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
991 | &ioend, !iomap_valid); |
992 | page_dirty--; | |
993 | count++; | |
f6d6d4fc | 994 | } else { |
1defeac9 | 995 | iomap_valid = 0; |
1da177e4 | 996 | } |
d5cb48aa CH |
997 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
998 | (unmapped || startio)) { | |
999 | iomap_valid = 0; | |
1da177e4 | 1000 | } |
f6d6d4fc CH |
1001 | |
1002 | if (!iohead) | |
1003 | iohead = ioend; | |
1004 | ||
1005 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1006 | |
1007 | if (uptodate && bh == head) | |
1008 | SetPageUptodate(page); | |
1009 | ||
f6d6d4fc CH |
1010 | if (startio) |
1011 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1012 | |
1defeac9 CH |
1013 | if (ioend && iomap_valid) { |
1014 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1015 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1016 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1017 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1018 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1019 | } |
1020 | ||
f6d6d4fc CH |
1021 | if (iohead) |
1022 | xfs_submit_ioend(iohead); | |
1023 | ||
1da177e4 LT |
1024 | return page_dirty; |
1025 | ||
1026 | error: | |
f6d6d4fc CH |
1027 | if (iohead) |
1028 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1029 | |
1030 | /* | |
1031 | * If it's delalloc and we have nowhere to put it, | |
1032 | * throw it away, unless the lower layers told | |
1033 | * us to try again. | |
1034 | */ | |
1035 | if (err != -EAGAIN) { | |
f6d6d4fc | 1036 | if (!unmapped) |
1da177e4 | 1037 | block_invalidatepage(page, 0); |
1da177e4 LT |
1038 | ClearPageUptodate(page); |
1039 | } | |
1040 | return err; | |
1041 | } | |
1042 | ||
1043 | STATIC int | |
1044 | __linvfs_get_block( | |
1045 | struct inode *inode, | |
1046 | sector_t iblock, | |
1047 | unsigned long blocks, | |
1048 | struct buffer_head *bh_result, | |
1049 | int create, | |
1050 | int direct, | |
1051 | bmapi_flags_t flags) | |
1052 | { | |
1053 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1054 | xfs_iomap_t iomap; | |
fdc7ed75 NS |
1055 | xfs_off_t offset; |
1056 | ssize_t size; | |
1da177e4 LT |
1057 | int retpbbm = 1; |
1058 | int error; | |
1da177e4 | 1059 | |
fdc7ed75 | 1060 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
a4656391 NS |
1061 | if (blocks) |
1062 | size = (ssize_t) min_t(xfs_off_t, LONG_MAX, | |
1063 | (xfs_off_t)blocks << inode->i_blkbits); | |
1064 | else | |
1065 | size = 1 << inode->i_blkbits; | |
1da177e4 LT |
1066 | |
1067 | VOP_BMAP(vp, offset, size, | |
1068 | create ? flags : BMAPI_READ, &iomap, &retpbbm, error); | |
1069 | if (error) | |
1070 | return -error; | |
1071 | ||
1072 | if (retpbbm == 0) | |
1073 | return 0; | |
1074 | ||
1075 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
fdc7ed75 NS |
1076 | xfs_daddr_t bn; |
1077 | xfs_off_t delta; | |
1da177e4 LT |
1078 | |
1079 | /* For unwritten extents do not report a disk address on | |
1080 | * the read case (treat as if we're reading into a hole). | |
1081 | */ | |
1082 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1083 | delta = offset - iomap.iomap_offset; | |
1084 | delta >>= inode->i_blkbits; | |
1085 | ||
1086 | bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT); | |
1087 | bn += delta; | |
1088 | BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME)); | |
1089 | bh_result->b_blocknr = bn; | |
1090 | set_buffer_mapped(bh_result); | |
1091 | } | |
1092 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1093 | if (direct) | |
1094 | bh_result->b_private = inode; | |
1095 | set_buffer_unwritten(bh_result); | |
1096 | set_buffer_delay(bh_result); | |
1097 | } | |
1098 | } | |
1099 | ||
1100 | /* If this is a realtime file, data might be on a new device */ | |
ce8e922c | 1101 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 LT |
1102 | |
1103 | /* If we previously allocated a block out beyond eof and | |
1104 | * we are now coming back to use it then we will need to | |
1105 | * flag it as new even if it has a disk address. | |
1106 | */ | |
1107 | if (create && | |
1108 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
fdc7ed75 | 1109 | (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) |
1da177e4 | 1110 | set_buffer_new(bh_result); |
1da177e4 LT |
1111 | |
1112 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1113 | BUG_ON(direct); | |
1114 | if (create) { | |
1115 | set_buffer_uptodate(bh_result); | |
1116 | set_buffer_mapped(bh_result); | |
1117 | set_buffer_delay(bh_result); | |
1118 | } | |
1119 | } | |
1120 | ||
1121 | if (blocks) { | |
fdc7ed75 NS |
1122 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1123 | offset = min_t(xfs_off_t, | |
1124 | iomap.iomap_bsize - iomap.iomap_delta, | |
a4656391 | 1125 | (xfs_off_t)blocks << inode->i_blkbits); |
fdc7ed75 | 1126 | bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset); |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | return 0; | |
1130 | } | |
1131 | ||
1132 | int | |
1133 | linvfs_get_block( | |
1134 | struct inode *inode, | |
1135 | sector_t iblock, | |
1136 | struct buffer_head *bh_result, | |
1137 | int create) | |
1138 | { | |
1139 | return __linvfs_get_block(inode, iblock, 0, bh_result, | |
1140 | create, 0, BMAPI_WRITE); | |
1141 | } | |
1142 | ||
1143 | STATIC int | |
1144 | linvfs_get_blocks_direct( | |
1145 | struct inode *inode, | |
1146 | sector_t iblock, | |
1147 | unsigned long max_blocks, | |
1148 | struct buffer_head *bh_result, | |
1149 | int create) | |
1150 | { | |
1151 | return __linvfs_get_block(inode, iblock, max_blocks, bh_result, | |
1152 | create, 1, BMAPI_WRITE|BMAPI_DIRECT); | |
1153 | } | |
1154 | ||
f0973863 CH |
1155 | STATIC void |
1156 | linvfs_end_io_direct( | |
1157 | struct kiocb *iocb, | |
1158 | loff_t offset, | |
1159 | ssize_t size, | |
1160 | void *private) | |
1161 | { | |
1162 | xfs_ioend_t *ioend = iocb->private; | |
1163 | ||
1164 | /* | |
1165 | * Non-NULL private data means we need to issue a transaction to | |
1166 | * convert a range from unwritten to written extents. This needs | |
1167 | * to happen from process contect but aio+dio I/O completion | |
1168 | * happens from irq context so we need to defer it to a workqueue. | |
1169 | * This is not nessecary for synchronous direct I/O, but we do | |
1170 | * it anyway to keep the code uniform and simpler. | |
1171 | * | |
1172 | * The core direct I/O code might be changed to always call the | |
1173 | * completion handler in the future, in which case all this can | |
1174 | * go away. | |
1175 | */ | |
1176 | if (private && size > 0) { | |
1177 | ioend->io_offset = offset; | |
1178 | ioend->io_size = size; | |
1179 | xfs_finish_ioend(ioend); | |
1180 | } else { | |
1181 | ASSERT(size >= 0); | |
1182 | xfs_destroy_ioend(ioend); | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * blockdev_direct_IO can return an error even afer the I/O | |
1187 | * completion handler was called. Thus we need to protect | |
1188 | * against double-freeing. | |
1189 | */ | |
1190 | iocb->private = NULL; | |
1191 | } | |
1192 | ||
1da177e4 LT |
1193 | STATIC ssize_t |
1194 | linvfs_direct_IO( | |
1195 | int rw, | |
1196 | struct kiocb *iocb, | |
1197 | const struct iovec *iov, | |
1198 | loff_t offset, | |
1199 | unsigned long nr_segs) | |
1200 | { | |
1201 | struct file *file = iocb->ki_filp; | |
1202 | struct inode *inode = file->f_mapping->host; | |
1203 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1204 | xfs_iomap_t iomap; | |
1205 | int maps = 1; | |
1206 | int error; | |
f0973863 | 1207 | ssize_t ret; |
1da177e4 LT |
1208 | |
1209 | VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error); | |
1210 | if (error) | |
1211 | return -error; | |
1212 | ||
f6d6d4fc | 1213 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
f0973863 CH |
1214 | |
1215 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, | |
ce8e922c | 1216 | iomap.iomap_target->bt_bdev, |
1da177e4 LT |
1217 | iov, offset, nr_segs, |
1218 | linvfs_get_blocks_direct, | |
f0973863 CH |
1219 | linvfs_end_io_direct); |
1220 | ||
1221 | if (unlikely(ret <= 0 && iocb->private)) | |
1222 | xfs_destroy_ioend(iocb->private); | |
1223 | return ret; | |
1da177e4 LT |
1224 | } |
1225 | ||
1226 | ||
1227 | STATIC sector_t | |
1228 | linvfs_bmap( | |
1229 | struct address_space *mapping, | |
1230 | sector_t block) | |
1231 | { | |
1232 | struct inode *inode = (struct inode *)mapping->host; | |
1233 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1234 | int error; | |
1235 | ||
1236 | vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address); | |
1237 | ||
1238 | VOP_RWLOCK(vp, VRWLOCK_READ); | |
1239 | VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error); | |
1240 | VOP_RWUNLOCK(vp, VRWLOCK_READ); | |
1241 | return generic_block_bmap(mapping, block, linvfs_get_block); | |
1242 | } | |
1243 | ||
1244 | STATIC int | |
1245 | linvfs_readpage( | |
1246 | struct file *unused, | |
1247 | struct page *page) | |
1248 | { | |
1249 | return mpage_readpage(page, linvfs_get_block); | |
1250 | } | |
1251 | ||
1252 | STATIC int | |
1253 | linvfs_readpages( | |
1254 | struct file *unused, | |
1255 | struct address_space *mapping, | |
1256 | struct list_head *pages, | |
1257 | unsigned nr_pages) | |
1258 | { | |
1259 | return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block); | |
1260 | } | |
1261 | ||
1262 | STATIC void | |
1263 | xfs_count_page_state( | |
1264 | struct page *page, | |
1265 | int *delalloc, | |
1266 | int *unmapped, | |
1267 | int *unwritten) | |
1268 | { | |
1269 | struct buffer_head *bh, *head; | |
1270 | ||
1271 | *delalloc = *unmapped = *unwritten = 0; | |
1272 | ||
1273 | bh = head = page_buffers(page); | |
1274 | do { | |
1275 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
1276 | (*unmapped) = 1; | |
1277 | else if (buffer_unwritten(bh) && !buffer_delay(bh)) | |
1278 | clear_buffer_unwritten(bh); | |
1279 | else if (buffer_unwritten(bh)) | |
1280 | (*unwritten) = 1; | |
1281 | else if (buffer_delay(bh)) | |
1282 | (*delalloc) = 1; | |
1283 | } while ((bh = bh->b_this_page) != head); | |
1284 | } | |
1285 | ||
1286 | ||
1287 | /* | |
1288 | * writepage: Called from one of two places: | |
1289 | * | |
1290 | * 1. we are flushing a delalloc buffer head. | |
1291 | * | |
1292 | * 2. we are writing out a dirty page. Typically the page dirty | |
1293 | * state is cleared before we get here. In this case is it | |
1294 | * conceivable we have no buffer heads. | |
1295 | * | |
1296 | * For delalloc space on the page we need to allocate space and | |
1297 | * flush it. For unmapped buffer heads on the page we should | |
1298 | * allocate space if the page is uptodate. For any other dirty | |
1299 | * buffer heads on the page we should flush them. | |
1300 | * | |
1301 | * If we detect that a transaction would be required to flush | |
1302 | * the page, we have to check the process flags first, if we | |
1303 | * are already in a transaction or disk I/O during allocations | |
1304 | * is off, we need to fail the writepage and redirty the page. | |
1305 | */ | |
1306 | ||
1307 | STATIC int | |
1308 | linvfs_writepage( | |
1309 | struct page *page, | |
1310 | struct writeback_control *wbc) | |
1311 | { | |
1312 | int error; | |
1313 | int need_trans; | |
1314 | int delalloc, unmapped, unwritten; | |
1315 | struct inode *inode = page->mapping->host; | |
1316 | ||
1317 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1318 | ||
1319 | /* | |
1320 | * We need a transaction if: | |
1321 | * 1. There are delalloc buffers on the page | |
1322 | * 2. The page is uptodate and we have unmapped buffers | |
1323 | * 3. The page is uptodate and we have no buffers | |
1324 | * 4. There are unwritten buffers on the page | |
1325 | */ | |
1326 | ||
1327 | if (!page_has_buffers(page)) { | |
1328 | unmapped = 1; | |
1329 | need_trans = 1; | |
1330 | } else { | |
1331 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1332 | if (!PageUptodate(page)) | |
1333 | unmapped = 0; | |
1334 | need_trans = delalloc + unmapped + unwritten; | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * If we need a transaction and the process flags say | |
1339 | * we are already in a transaction, or no IO is allowed | |
1340 | * then mark the page dirty again and leave the page | |
1341 | * as is. | |
1342 | */ | |
1343 | if (PFLAGS_TEST_FSTRANS() && need_trans) | |
1344 | goto out_fail; | |
1345 | ||
1346 | /* | |
1347 | * Delay hooking up buffer heads until we have | |
1348 | * made our go/no-go decision. | |
1349 | */ | |
1350 | if (!page_has_buffers(page)) | |
1351 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1352 | ||
1353 | /* | |
1354 | * Convert delayed allocate, unwritten or unmapped space | |
1355 | * to real space and flush out to disk. | |
1356 | */ | |
1357 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1358 | if (error == -EAGAIN) | |
1359 | goto out_fail; | |
1360 | if (unlikely(error < 0)) | |
1361 | goto out_unlock; | |
1362 | ||
1363 | return 0; | |
1364 | ||
1365 | out_fail: | |
1366 | redirty_page_for_writepage(wbc, page); | |
1367 | unlock_page(page); | |
1368 | return 0; | |
1369 | out_unlock: | |
1370 | unlock_page(page); | |
1371 | return error; | |
1372 | } | |
1373 | ||
bcec2b7f NS |
1374 | STATIC int |
1375 | linvfs_invalidate_page( | |
1376 | struct page *page, | |
1377 | unsigned long offset) | |
1378 | { | |
1379 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1380 | page->mapping->host, page, offset); | |
1381 | return block_invalidatepage(page, offset); | |
1382 | } | |
1383 | ||
1da177e4 LT |
1384 | /* |
1385 | * Called to move a page into cleanable state - and from there | |
1386 | * to be released. Possibly the page is already clean. We always | |
1387 | * have buffer heads in this call. | |
1388 | * | |
1389 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1390 | * | |
1391 | * Possible scenarios are: | |
1392 | * | |
1393 | * 1. We are being called to release a page which has been written | |
1394 | * to via regular I/O. buffer heads will be dirty and possibly | |
1395 | * delalloc. If no delalloc buffer heads in this case then we | |
1396 | * can just return zero. | |
1397 | * | |
1398 | * 2. We are called to release a page which has been written via | |
1399 | * mmap, all we need to do is ensure there is no delalloc | |
1400 | * state in the buffer heads, if not we can let the caller | |
1401 | * free them and we should come back later via writepage. | |
1402 | */ | |
1403 | STATIC int | |
1404 | linvfs_release_page( | |
1405 | struct page *page, | |
27496a8c | 1406 | gfp_t gfp_mask) |
1da177e4 LT |
1407 | { |
1408 | struct inode *inode = page->mapping->host; | |
1409 | int dirty, delalloc, unmapped, unwritten; | |
1410 | struct writeback_control wbc = { | |
1411 | .sync_mode = WB_SYNC_ALL, | |
1412 | .nr_to_write = 1, | |
1413 | }; | |
1414 | ||
1415 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask); | |
1416 | ||
1417 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1418 | if (!delalloc && !unwritten) | |
1419 | goto free_buffers; | |
1420 | ||
1421 | if (!(gfp_mask & __GFP_FS)) | |
1422 | return 0; | |
1423 | ||
1424 | /* If we are already inside a transaction or the thread cannot | |
1425 | * do I/O, we cannot release this page. | |
1426 | */ | |
1427 | if (PFLAGS_TEST_FSTRANS()) | |
1428 | return 0; | |
1429 | ||
1430 | /* | |
1431 | * Convert delalloc space to real space, do not flush the | |
1432 | * data out to disk, that will be done by the caller. | |
1433 | * Never need to allocate space here - we will always | |
1434 | * come back to writepage in that case. | |
1435 | */ | |
1436 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1437 | if (dirty == 0 && !unwritten) | |
1438 | goto free_buffers; | |
1439 | return 0; | |
1440 | ||
1441 | free_buffers: | |
1442 | return try_to_free_buffers(page); | |
1443 | } | |
1444 | ||
1445 | STATIC int | |
1446 | linvfs_prepare_write( | |
1447 | struct file *file, | |
1448 | struct page *page, | |
1449 | unsigned int from, | |
1450 | unsigned int to) | |
1451 | { | |
1452 | return block_prepare_write(page, from, to, linvfs_get_block); | |
1453 | } | |
1454 | ||
1455 | struct address_space_operations linvfs_aops = { | |
1456 | .readpage = linvfs_readpage, | |
1457 | .readpages = linvfs_readpages, | |
1458 | .writepage = linvfs_writepage, | |
1459 | .sync_page = block_sync_page, | |
1460 | .releasepage = linvfs_release_page, | |
bcec2b7f | 1461 | .invalidatepage = linvfs_invalidate_page, |
1da177e4 LT |
1462 | .prepare_write = linvfs_prepare_write, |
1463 | .commit_write = generic_commit_write, | |
1464 | .bmap = linvfs_bmap, | |
1465 | .direct_IO = linvfs_direct_IO, | |
e965f963 | 1466 | .migratepage = buffer_migrate_page, |
1da177e4 | 1467 | }; |