]>
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 | ||
f51623b2 NS |
46 | |
47 | STATIC void | |
48 | xfs_count_page_state( | |
49 | struct page *page, | |
50 | int *delalloc, | |
51 | int *unmapped, | |
52 | int *unwritten) | |
53 | { | |
54 | struct buffer_head *bh, *head; | |
55 | ||
56 | *delalloc = *unmapped = *unwritten = 0; | |
57 | ||
58 | bh = head = page_buffers(page); | |
59 | do { | |
60 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
61 | (*unmapped) = 1; | |
62 | else if (buffer_unwritten(bh) && !buffer_delay(bh)) | |
63 | clear_buffer_unwritten(bh); | |
64 | else if (buffer_unwritten(bh)) | |
65 | (*unwritten) = 1; | |
66 | else if (buffer_delay(bh)) | |
67 | (*delalloc) = 1; | |
68 | } while ((bh = bh->b_this_page) != head); | |
69 | } | |
70 | ||
71 | ||
1da177e4 LT |
72 | |
73 | #if defined(XFS_RW_TRACE) | |
74 | void | |
75 | xfs_page_trace( | |
76 | int tag, | |
77 | struct inode *inode, | |
78 | struct page *page, | |
79 | int mask) | |
80 | { | |
81 | xfs_inode_t *ip; | |
1da177e4 LT |
82 | vnode_t *vp = LINVFS_GET_VP(inode); |
83 | loff_t isize = i_size_read(inode); | |
f6d6d4fc | 84 | loff_t offset = page_offset(page); |
1da177e4 LT |
85 | int delalloc = -1, unmapped = -1, unwritten = -1; |
86 | ||
87 | if (page_has_buffers(page)) | |
88 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
89 | ||
75e17b3c | 90 | ip = xfs_vtoi(vp); |
1da177e4 LT |
91 | if (!ip->i_rwtrace) |
92 | return; | |
93 | ||
94 | ktrace_enter(ip->i_rwtrace, | |
95 | (void *)((unsigned long)tag), | |
96 | (void *)ip, | |
97 | (void *)inode, | |
98 | (void *)page, | |
99 | (void *)((unsigned long)mask), | |
100 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), | |
101 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
102 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
103 | (void *)((unsigned long)(isize & 0xffffffff)), | |
104 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
105 | (void *)((unsigned long)(offset & 0xffffffff)), | |
106 | (void *)((unsigned long)delalloc), | |
107 | (void *)((unsigned long)unmapped), | |
108 | (void *)((unsigned long)unwritten), | |
109 | (void *)NULL, | |
110 | (void *)NULL); | |
111 | } | |
112 | #else | |
113 | #define xfs_page_trace(tag, inode, page, mask) | |
114 | #endif | |
115 | ||
0829c360 CH |
116 | /* |
117 | * Schedule IO completion handling on a xfsdatad if this was | |
118 | * the final hold on this ioend. | |
119 | */ | |
120 | STATIC void | |
121 | xfs_finish_ioend( | |
122 | xfs_ioend_t *ioend) | |
123 | { | |
124 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
125 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
126 | } | |
127 | ||
f6d6d4fc CH |
128 | /* |
129 | * We're now finished for good with this ioend structure. | |
130 | * Update the page state via the associated buffer_heads, | |
131 | * release holds on the inode and bio, and finally free | |
132 | * up memory. Do not use the ioend after this. | |
133 | */ | |
0829c360 CH |
134 | STATIC void |
135 | xfs_destroy_ioend( | |
136 | xfs_ioend_t *ioend) | |
137 | { | |
f6d6d4fc CH |
138 | struct buffer_head *bh, *next; |
139 | ||
140 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
141 | next = bh->b_private; | |
142 | bh->b_end_io(bh, ioend->io_uptodate); | |
143 | } | |
144 | ||
0829c360 CH |
145 | vn_iowake(ioend->io_vnode); |
146 | mempool_free(ioend, xfs_ioend_pool); | |
147 | } | |
148 | ||
149 | /* | |
f6d6d4fc CH |
150 | * Buffered IO write completion for delayed allocate extents. |
151 | * TODO: Update ondisk isize now that we know the file data | |
152 | * has been flushed (i.e. the notorious "NULL file" problem). | |
153 | */ | |
154 | STATIC void | |
155 | xfs_end_bio_delalloc( | |
156 | void *data) | |
157 | { | |
158 | xfs_ioend_t *ioend = data; | |
159 | ||
160 | xfs_destroy_ioend(ioend); | |
161 | } | |
162 | ||
163 | /* | |
164 | * Buffered IO write completion for regular, written extents. | |
165 | */ | |
166 | STATIC void | |
167 | xfs_end_bio_written( | |
168 | void *data) | |
169 | { | |
170 | xfs_ioend_t *ioend = data; | |
171 | ||
172 | xfs_destroy_ioend(ioend); | |
173 | } | |
174 | ||
175 | /* | |
176 | * IO write completion for unwritten extents. | |
177 | * | |
0829c360 | 178 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 179 | * to written extents. |
0829c360 CH |
180 | */ |
181 | STATIC void | |
182 | xfs_end_bio_unwritten( | |
183 | void *data) | |
184 | { | |
185 | xfs_ioend_t *ioend = data; | |
186 | vnode_t *vp = ioend->io_vnode; | |
187 | xfs_off_t offset = ioend->io_offset; | |
188 | size_t size = ioend->io_size; | |
189 | int error; | |
190 | ||
191 | if (ioend->io_uptodate) | |
192 | VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error); | |
193 | xfs_destroy_ioend(ioend); | |
194 | } | |
195 | ||
196 | /* | |
197 | * Allocate and initialise an IO completion structure. | |
198 | * We need to track unwritten extent write completion here initially. | |
199 | * We'll need to extend this for updating the ondisk inode size later | |
200 | * (vs. incore size). | |
201 | */ | |
202 | STATIC xfs_ioend_t * | |
203 | xfs_alloc_ioend( | |
f6d6d4fc CH |
204 | struct inode *inode, |
205 | unsigned int type) | |
0829c360 CH |
206 | { |
207 | xfs_ioend_t *ioend; | |
208 | ||
209 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
210 | ||
211 | /* | |
212 | * Set the count to 1 initially, which will prevent an I/O | |
213 | * completion callback from happening before we have started | |
214 | * all the I/O from calling the completion routine too early. | |
215 | */ | |
216 | atomic_set(&ioend->io_remaining, 1); | |
217 | ioend->io_uptodate = 1; /* cleared if any I/O fails */ | |
f6d6d4fc CH |
218 | ioend->io_list = NULL; |
219 | ioend->io_type = type; | |
0829c360 | 220 | ioend->io_vnode = LINVFS_GET_VP(inode); |
c1a073bd | 221 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 222 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
223 | atomic_inc(&ioend->io_vnode->v_iocount); |
224 | ioend->io_offset = 0; | |
225 | ioend->io_size = 0; | |
226 | ||
f6d6d4fc CH |
227 | if (type == IOMAP_UNWRITTEN) |
228 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend); | |
229 | else if (type == IOMAP_DELAY) | |
230 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend); | |
231 | else | |
232 | INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend); | |
0829c360 CH |
233 | |
234 | return ioend; | |
235 | } | |
236 | ||
1da177e4 LT |
237 | STATIC int |
238 | xfs_map_blocks( | |
239 | struct inode *inode, | |
240 | loff_t offset, | |
241 | ssize_t count, | |
242 | xfs_iomap_t *mapp, | |
243 | int flags) | |
244 | { | |
245 | vnode_t *vp = LINVFS_GET_VP(inode); | |
246 | int error, nmaps = 1; | |
247 | ||
248 | VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error); | |
249 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) | |
250 | VMODIFY(vp); | |
251 | return -error; | |
252 | } | |
253 | ||
1defeac9 CH |
254 | STATIC inline int |
255 | xfs_iomap_valid( | |
1da177e4 | 256 | xfs_iomap_t *iomapp, |
1defeac9 | 257 | loff_t offset) |
1da177e4 | 258 | { |
1defeac9 CH |
259 | return offset >= iomapp->iomap_offset && |
260 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
261 | } |
262 | ||
f6d6d4fc CH |
263 | /* |
264 | * BIO completion handler for buffered IO. | |
265 | */ | |
266 | STATIC int | |
267 | xfs_end_bio( | |
268 | struct bio *bio, | |
269 | unsigned int bytes_done, | |
270 | int error) | |
271 | { | |
272 | xfs_ioend_t *ioend = bio->bi_private; | |
273 | ||
274 | if (bio->bi_size) | |
275 | return 1; | |
276 | ||
277 | ASSERT(ioend); | |
278 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); | |
279 | ||
280 | /* Toss bio and pass work off to an xfsdatad thread */ | |
281 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
282 | ioend->io_uptodate = 0; | |
283 | bio->bi_private = NULL; | |
284 | bio->bi_end_io = NULL; | |
285 | ||
286 | bio_put(bio); | |
287 | xfs_finish_ioend(ioend); | |
288 | return 0; | |
289 | } | |
290 | ||
291 | STATIC void | |
292 | xfs_submit_ioend_bio( | |
293 | xfs_ioend_t *ioend, | |
294 | struct bio *bio) | |
295 | { | |
296 | atomic_inc(&ioend->io_remaining); | |
297 | ||
298 | bio->bi_private = ioend; | |
299 | bio->bi_end_io = xfs_end_bio; | |
300 | ||
301 | submit_bio(WRITE, bio); | |
302 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
303 | bio_put(bio); | |
304 | } | |
305 | ||
306 | STATIC struct bio * | |
307 | xfs_alloc_ioend_bio( | |
308 | struct buffer_head *bh) | |
309 | { | |
310 | struct bio *bio; | |
311 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
312 | ||
313 | do { | |
314 | bio = bio_alloc(GFP_NOIO, nvecs); | |
315 | nvecs >>= 1; | |
316 | } while (!bio); | |
317 | ||
318 | ASSERT(bio->bi_private == NULL); | |
319 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
320 | bio->bi_bdev = bh->b_bdev; | |
321 | bio_get(bio); | |
322 | return bio; | |
323 | } | |
324 | ||
325 | STATIC void | |
326 | xfs_start_buffer_writeback( | |
327 | struct buffer_head *bh) | |
328 | { | |
329 | ASSERT(buffer_mapped(bh)); | |
330 | ASSERT(buffer_locked(bh)); | |
331 | ASSERT(!buffer_delay(bh)); | |
332 | ASSERT(!buffer_unwritten(bh)); | |
333 | ||
334 | mark_buffer_async_write(bh); | |
335 | set_buffer_uptodate(bh); | |
336 | clear_buffer_dirty(bh); | |
337 | } | |
338 | ||
339 | STATIC void | |
340 | xfs_start_page_writeback( | |
341 | struct page *page, | |
342 | struct writeback_control *wbc, | |
343 | int clear_dirty, | |
344 | int buffers) | |
345 | { | |
346 | ASSERT(PageLocked(page)); | |
347 | ASSERT(!PageWriteback(page)); | |
348 | set_page_writeback(page); | |
349 | if (clear_dirty) | |
350 | clear_page_dirty(page); | |
351 | unlock_page(page); | |
352 | if (!buffers) { | |
353 | end_page_writeback(page); | |
354 | wbc->pages_skipped++; /* We didn't write this page */ | |
355 | } | |
356 | } | |
357 | ||
358 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
359 | { | |
360 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
361 | } | |
362 | ||
363 | /* | |
d88992f6 DC |
364 | * Submit all of the bios for all of the ioends we have saved up, covering the |
365 | * initial writepage page and also any probed pages. | |
366 | * | |
367 | * Because we may have multiple ioends spanning a page, we need to start | |
368 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
369 | * buffers as we got, then we can end up with a page that only has buffers | |
370 | * marked async write and I/O complete on can occur before we mark the other | |
371 | * buffers async write. | |
372 | * | |
373 | * The end result of this is that we trip a bug in end_page_writeback() because | |
374 | * we call it twice for the one page as the code in end_buffer_async_write() | |
375 | * assumes that all buffers on the page are started at the same time. | |
376 | * | |
377 | * The fix is two passes across the ioend list - one to start writeback on the | |
378 | * bufferheads, and then the second one submit them for I/O. | |
f6d6d4fc CH |
379 | */ |
380 | STATIC void | |
381 | xfs_submit_ioend( | |
382 | xfs_ioend_t *ioend) | |
383 | { | |
d88992f6 | 384 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
385 | xfs_ioend_t *next; |
386 | struct buffer_head *bh; | |
387 | struct bio *bio; | |
388 | sector_t lastblock = 0; | |
389 | ||
d88992f6 DC |
390 | /* Pass 1 - start writeback */ |
391 | do { | |
392 | next = ioend->io_list; | |
393 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
394 | xfs_start_buffer_writeback(bh); | |
395 | } | |
396 | } while ((ioend = next) != NULL); | |
397 | ||
398 | /* Pass 2 - submit I/O */ | |
399 | ioend = head; | |
f6d6d4fc CH |
400 | do { |
401 | next = ioend->io_list; | |
402 | bio = NULL; | |
403 | ||
404 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
405 | |
406 | if (!bio) { | |
407 | retry: | |
408 | bio = xfs_alloc_ioend_bio(bh); | |
409 | } else if (bh->b_blocknr != lastblock + 1) { | |
410 | xfs_submit_ioend_bio(ioend, bio); | |
411 | goto retry; | |
412 | } | |
413 | ||
414 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
415 | xfs_submit_ioend_bio(ioend, bio); | |
416 | goto retry; | |
417 | } | |
418 | ||
419 | lastblock = bh->b_blocknr; | |
420 | } | |
421 | if (bio) | |
422 | xfs_submit_ioend_bio(ioend, bio); | |
423 | xfs_finish_ioend(ioend); | |
424 | } while ((ioend = next) != NULL); | |
425 | } | |
426 | ||
427 | /* | |
428 | * Cancel submission of all buffer_heads so far in this endio. | |
429 | * Toss the endio too. Only ever called for the initial page | |
430 | * in a writepage request, so only ever one page. | |
431 | */ | |
432 | STATIC void | |
433 | xfs_cancel_ioend( | |
434 | xfs_ioend_t *ioend) | |
435 | { | |
436 | xfs_ioend_t *next; | |
437 | struct buffer_head *bh, *next_bh; | |
438 | ||
439 | do { | |
440 | next = ioend->io_list; | |
441 | bh = ioend->io_buffer_head; | |
442 | do { | |
443 | next_bh = bh->b_private; | |
444 | clear_buffer_async_write(bh); | |
445 | unlock_buffer(bh); | |
446 | } while ((bh = next_bh) != NULL); | |
447 | ||
448 | vn_iowake(ioend->io_vnode); | |
449 | mempool_free(ioend, xfs_ioend_pool); | |
450 | } while ((ioend = next) != NULL); | |
451 | } | |
452 | ||
453 | /* | |
454 | * Test to see if we've been building up a completion structure for | |
455 | * earlier buffers -- if so, we try to append to this ioend if we | |
456 | * can, otherwise we finish off any current ioend and start another. | |
457 | * Return true if we've finished the given ioend. | |
458 | */ | |
459 | STATIC void | |
460 | xfs_add_to_ioend( | |
461 | struct inode *inode, | |
462 | struct buffer_head *bh, | |
7336cea8 | 463 | xfs_off_t offset, |
f6d6d4fc CH |
464 | unsigned int type, |
465 | xfs_ioend_t **result, | |
466 | int need_ioend) | |
467 | { | |
468 | xfs_ioend_t *ioend = *result; | |
469 | ||
470 | if (!ioend || need_ioend || type != ioend->io_type) { | |
471 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 472 | |
f6d6d4fc CH |
473 | ioend = xfs_alloc_ioend(inode, type); |
474 | ioend->io_offset = offset; | |
475 | ioend->io_buffer_head = bh; | |
476 | ioend->io_buffer_tail = bh; | |
477 | if (previous) | |
478 | previous->io_list = ioend; | |
479 | *result = ioend; | |
480 | } else { | |
481 | ioend->io_buffer_tail->b_private = bh; | |
482 | ioend->io_buffer_tail = bh; | |
483 | } | |
484 | ||
485 | bh->b_private = NULL; | |
486 | ioend->io_size += bh->b_size; | |
487 | } | |
488 | ||
87cbc49c NS |
489 | STATIC void |
490 | xfs_map_buffer( | |
491 | struct buffer_head *bh, | |
492 | xfs_iomap_t *mp, | |
493 | xfs_off_t offset, | |
494 | uint block_bits) | |
495 | { | |
496 | sector_t bn; | |
497 | ||
498 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
499 | ||
500 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
501 | ((offset - mp->iomap_offset) >> block_bits); | |
502 | ||
503 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
504 | ||
505 | bh->b_blocknr = bn; | |
506 | set_buffer_mapped(bh); | |
507 | } | |
508 | ||
1da177e4 LT |
509 | STATIC void |
510 | xfs_map_at_offset( | |
1da177e4 | 511 | struct buffer_head *bh, |
1defeac9 | 512 | loff_t offset, |
1da177e4 | 513 | int block_bits, |
1defeac9 | 514 | xfs_iomap_t *iomapp) |
1da177e4 | 515 | { |
1da177e4 LT |
516 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
517 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
518 | |
519 | lock_buffer(bh); | |
87cbc49c | 520 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 521 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
522 | set_buffer_mapped(bh); |
523 | clear_buffer_delay(bh); | |
f6d6d4fc | 524 | clear_buffer_unwritten(bh); |
1da177e4 LT |
525 | } |
526 | ||
527 | /* | |
6c4fe19f | 528 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
529 | */ |
530 | STATIC unsigned int | |
6c4fe19f | 531 | xfs_probe_page( |
10ce4444 | 532 | struct page *page, |
6c4fe19f CH |
533 | unsigned int pg_offset, |
534 | int mapped) | |
1da177e4 | 535 | { |
1da177e4 LT |
536 | int ret = 0; |
537 | ||
1da177e4 | 538 | if (PageWriteback(page)) |
10ce4444 | 539 | return 0; |
1da177e4 LT |
540 | |
541 | if (page->mapping && PageDirty(page)) { | |
542 | if (page_has_buffers(page)) { | |
543 | struct buffer_head *bh, *head; | |
544 | ||
545 | bh = head = page_buffers(page); | |
546 | do { | |
6c4fe19f CH |
547 | if (!buffer_uptodate(bh)) |
548 | break; | |
549 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
550 | break; |
551 | ret += bh->b_size; | |
552 | if (ret >= pg_offset) | |
553 | break; | |
554 | } while ((bh = bh->b_this_page) != head); | |
555 | } else | |
6c4fe19f | 556 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
557 | } |
558 | ||
1da177e4 LT |
559 | return ret; |
560 | } | |
561 | ||
f6d6d4fc | 562 | STATIC size_t |
6c4fe19f | 563 | xfs_probe_cluster( |
1da177e4 LT |
564 | struct inode *inode, |
565 | struct page *startpage, | |
566 | struct buffer_head *bh, | |
6c4fe19f CH |
567 | struct buffer_head *head, |
568 | int mapped) | |
1da177e4 | 569 | { |
10ce4444 | 570 | struct pagevec pvec; |
1da177e4 | 571 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
572 | size_t total = 0; |
573 | int done = 0, i; | |
1da177e4 LT |
574 | |
575 | /* First sum forwards in this page */ | |
576 | do { | |
2353e8e9 | 577 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 578 | return total; |
1da177e4 LT |
579 | total += bh->b_size; |
580 | } while ((bh = bh->b_this_page) != head); | |
581 | ||
10ce4444 CH |
582 | /* if we reached the end of the page, sum forwards in following pages */ |
583 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
584 | tindex = startpage->index + 1; | |
585 | ||
586 | /* Prune this back to avoid pathological behavior */ | |
587 | tloff = min(tlast, startpage->index + 64); | |
588 | ||
589 | pagevec_init(&pvec, 0); | |
590 | while (!done && tindex <= tloff) { | |
591 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
592 | ||
593 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
594 | break; | |
595 | ||
596 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
597 | struct page *page = pvec.pages[i]; | |
598 | size_t pg_offset, len = 0; | |
599 | ||
600 | if (tindex == tlast) { | |
601 | pg_offset = | |
602 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
603 | if (!pg_offset) { |
604 | done = 1; | |
10ce4444 | 605 | break; |
1defeac9 | 606 | } |
10ce4444 CH |
607 | } else |
608 | pg_offset = PAGE_CACHE_SIZE; | |
609 | ||
610 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
6c4fe19f | 611 | len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
612 | unlock_page(page); |
613 | } | |
614 | ||
615 | if (!len) { | |
616 | done = 1; | |
617 | break; | |
618 | } | |
619 | ||
1da177e4 | 620 | total += len; |
1defeac9 | 621 | tindex++; |
1da177e4 | 622 | } |
10ce4444 CH |
623 | |
624 | pagevec_release(&pvec); | |
625 | cond_resched(); | |
1da177e4 | 626 | } |
10ce4444 | 627 | |
1da177e4 LT |
628 | return total; |
629 | } | |
630 | ||
631 | /* | |
10ce4444 CH |
632 | * Test if a given page is suitable for writing as part of an unwritten |
633 | * or delayed allocate extent. | |
1da177e4 | 634 | */ |
10ce4444 CH |
635 | STATIC int |
636 | xfs_is_delayed_page( | |
637 | struct page *page, | |
f6d6d4fc | 638 | unsigned int type) |
1da177e4 | 639 | { |
1da177e4 | 640 | if (PageWriteback(page)) |
10ce4444 | 641 | return 0; |
1da177e4 LT |
642 | |
643 | if (page->mapping && page_has_buffers(page)) { | |
644 | struct buffer_head *bh, *head; | |
645 | int acceptable = 0; | |
646 | ||
647 | bh = head = page_buffers(page); | |
648 | do { | |
f6d6d4fc CH |
649 | if (buffer_unwritten(bh)) |
650 | acceptable = (type == IOMAP_UNWRITTEN); | |
651 | else if (buffer_delay(bh)) | |
652 | acceptable = (type == IOMAP_DELAY); | |
6c4fe19f CH |
653 | else if (buffer_mapped(bh)) |
654 | acceptable = (type == 0); | |
f6d6d4fc | 655 | else |
1da177e4 | 656 | break; |
1da177e4 LT |
657 | } while ((bh = bh->b_this_page) != head); |
658 | ||
659 | if (acceptable) | |
10ce4444 | 660 | return 1; |
1da177e4 LT |
661 | } |
662 | ||
10ce4444 | 663 | return 0; |
1da177e4 LT |
664 | } |
665 | ||
1da177e4 LT |
666 | /* |
667 | * Allocate & map buffers for page given the extent map. Write it out. | |
668 | * except for the original page of a writepage, this is called on | |
669 | * delalloc/unwritten pages only, for the original page it is possible | |
670 | * that the page has no mapping at all. | |
671 | */ | |
f6d6d4fc | 672 | STATIC int |
1da177e4 LT |
673 | xfs_convert_page( |
674 | struct inode *inode, | |
675 | struct page *page, | |
10ce4444 | 676 | loff_t tindex, |
1defeac9 | 677 | xfs_iomap_t *mp, |
f6d6d4fc | 678 | xfs_ioend_t **ioendp, |
1da177e4 | 679 | struct writeback_control *wbc, |
1da177e4 LT |
680 | int startio, |
681 | int all_bh) | |
682 | { | |
f6d6d4fc | 683 | struct buffer_head *bh, *head; |
9260dc6b CH |
684 | xfs_off_t end_offset; |
685 | unsigned long p_offset; | |
f6d6d4fc | 686 | unsigned int type; |
1da177e4 | 687 | int bbits = inode->i_blkbits; |
24e17b5f | 688 | int len, page_dirty; |
f6d6d4fc | 689 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 690 | xfs_off_t offset = page_offset(page); |
1da177e4 | 691 | |
10ce4444 CH |
692 | if (page->index != tindex) |
693 | goto fail; | |
694 | if (TestSetPageLocked(page)) | |
695 | goto fail; | |
696 | if (PageWriteback(page)) | |
697 | goto fail_unlock_page; | |
698 | if (page->mapping != inode->i_mapping) | |
699 | goto fail_unlock_page; | |
700 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
701 | goto fail_unlock_page; | |
702 | ||
24e17b5f NS |
703 | /* |
704 | * page_dirty is initially a count of buffers on the page before | |
705 | * EOF and is decrememted as we move each into a cleanable state. | |
9260dc6b CH |
706 | * |
707 | * Derivation: | |
708 | * | |
709 | * End offset is the highest offset that this page should represent. | |
710 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
711 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
712 | * hence give us the correct page_dirty count. On any other page, | |
713 | * it will be zero and in that case we need page_dirty to be the | |
714 | * count of buffers on the page. | |
24e17b5f | 715 | */ |
9260dc6b CH |
716 | end_offset = min_t(unsigned long long, |
717 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
718 | i_size_read(inode)); | |
719 | ||
24e17b5f | 720 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
721 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
722 | PAGE_CACHE_SIZE); | |
723 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
724 | page_dirty = p_offset / len; | |
24e17b5f | 725 | |
1da177e4 LT |
726 | bh = head = page_buffers(page); |
727 | do { | |
9260dc6b | 728 | if (offset >= end_offset) |
1da177e4 | 729 | break; |
f6d6d4fc CH |
730 | if (!buffer_uptodate(bh)) |
731 | uptodate = 0; | |
732 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
733 | done = 1; | |
1da177e4 | 734 | continue; |
f6d6d4fc CH |
735 | } |
736 | ||
9260dc6b CH |
737 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
738 | if (buffer_unwritten(bh)) | |
739 | type = IOMAP_UNWRITTEN; | |
740 | else | |
741 | type = IOMAP_DELAY; | |
742 | ||
743 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 744 | done = 1; |
9260dc6b CH |
745 | continue; |
746 | } | |
747 | ||
748 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
749 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
750 | ||
751 | xfs_map_at_offset(bh, offset, bbits, mp); | |
752 | if (startio) { | |
7336cea8 | 753 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
754 | type, ioendp, done); |
755 | } else { | |
756 | set_buffer_dirty(bh); | |
757 | unlock_buffer(bh); | |
758 | mark_buffer_dirty(bh); | |
759 | } | |
760 | page_dirty--; | |
761 | count++; | |
762 | } else { | |
763 | type = 0; | |
764 | if (buffer_mapped(bh) && all_bh && startio) { | |
1da177e4 | 765 | lock_buffer(bh); |
7336cea8 | 766 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
767 | type, ioendp, done); |
768 | count++; | |
24e17b5f | 769 | page_dirty--; |
9260dc6b CH |
770 | } else { |
771 | done = 1; | |
1da177e4 | 772 | } |
1da177e4 | 773 | } |
7336cea8 | 774 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 775 | |
f6d6d4fc CH |
776 | if (uptodate && bh == head) |
777 | SetPageUptodate(page); | |
778 | ||
779 | if (startio) { | |
f5e596bb CH |
780 | if (count) { |
781 | struct backing_dev_info *bdi; | |
782 | ||
783 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 784 | wbc->nr_to_write--; |
f5e596bb CH |
785 | if (bdi_write_congested(bdi)) { |
786 | wbc->encountered_congestion = 1; | |
787 | done = 1; | |
9fddaca2 | 788 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
789 | done = 1; |
790 | } | |
791 | } | |
f6d6d4fc | 792 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 793 | } |
f6d6d4fc CH |
794 | |
795 | return done; | |
10ce4444 CH |
796 | fail_unlock_page: |
797 | unlock_page(page); | |
798 | fail: | |
799 | return 1; | |
1da177e4 LT |
800 | } |
801 | ||
802 | /* | |
803 | * Convert & write out a cluster of pages in the same extent as defined | |
804 | * by mp and following the start page. | |
805 | */ | |
806 | STATIC void | |
807 | xfs_cluster_write( | |
808 | struct inode *inode, | |
809 | pgoff_t tindex, | |
810 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 811 | xfs_ioend_t **ioendp, |
1da177e4 LT |
812 | struct writeback_control *wbc, |
813 | int startio, | |
814 | int all_bh, | |
815 | pgoff_t tlast) | |
816 | { | |
10ce4444 CH |
817 | struct pagevec pvec; |
818 | int done = 0, i; | |
1da177e4 | 819 | |
10ce4444 CH |
820 | pagevec_init(&pvec, 0); |
821 | while (!done && tindex <= tlast) { | |
822 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
823 | ||
824 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 825 | break; |
10ce4444 CH |
826 | |
827 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
828 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
829 | iomapp, ioendp, wbc, startio, all_bh); | |
830 | if (done) | |
831 | break; | |
832 | } | |
833 | ||
834 | pagevec_release(&pvec); | |
835 | cond_resched(); | |
1da177e4 LT |
836 | } |
837 | } | |
838 | ||
839 | /* | |
840 | * Calling this without startio set means we are being asked to make a dirty | |
841 | * page ready for freeing it's buffers. When called with startio set then | |
842 | * we are coming from writepage. | |
843 | * | |
844 | * When called with startio set it is important that we write the WHOLE | |
845 | * page if possible. | |
846 | * The bh->b_state's cannot know if any of the blocks or which block for | |
847 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
848 | * only vaild if the page itself isn't completely uptodate. Some layers | |
849 | * may clear the page dirty flag prior to calling write page, under the | |
850 | * assumption the entire page will be written out; by not writing out the | |
851 | * whole page the page can be reused before all valid dirty data is | |
852 | * written out. Note: in the case of a page that has been dirty'd by | |
853 | * mapwrite and but partially setup by block_prepare_write the | |
854 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
855 | * valid state, thus the whole page must be written out thing. | |
856 | */ | |
857 | ||
858 | STATIC int | |
859 | xfs_page_state_convert( | |
860 | struct inode *inode, | |
861 | struct page *page, | |
862 | struct writeback_control *wbc, | |
863 | int startio, | |
864 | int unmapped) /* also implies page uptodate */ | |
865 | { | |
f6d6d4fc | 866 | struct buffer_head *bh, *head; |
1defeac9 | 867 | xfs_iomap_t iomap; |
f6d6d4fc | 868 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
869 | loff_t offset; |
870 | unsigned long p_offset = 0; | |
f6d6d4fc | 871 | unsigned int type; |
1da177e4 LT |
872 | __uint64_t end_offset; |
873 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
874 | ssize_t size, len; |
875 | int flags, err, iomap_valid = 0, uptodate = 1; | |
f6d6d4fc | 876 | int page_dirty, count = 0, trylock_flag = 0; |
6c4fe19f | 877 | int all_bh = unmapped; |
1da177e4 | 878 | |
3ba0815a | 879 | /* wait for other IO threads? */ |
f5e596bb | 880 | if (startio && (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)) |
f6d6d4fc | 881 | trylock_flag |= BMAPI_TRYLOCK; |
3ba0815a | 882 | |
1da177e4 LT |
883 | /* Is this page beyond the end of the file? */ |
884 | offset = i_size_read(inode); | |
885 | end_index = offset >> PAGE_CACHE_SHIFT; | |
886 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
887 | if (page->index >= end_index) { | |
888 | if ((page->index >= end_index + 1) || | |
889 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
890 | if (startio) |
891 | unlock_page(page); | |
892 | return 0; | |
1da177e4 LT |
893 | } |
894 | } | |
895 | ||
1da177e4 | 896 | /* |
24e17b5f NS |
897 | * page_dirty is initially a count of buffers on the page before |
898 | * EOF and is decrememted as we move each into a cleanable state. | |
f6d6d4fc CH |
899 | * |
900 | * Derivation: | |
901 | * | |
902 | * End offset is the highest offset that this page should represent. | |
903 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
904 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
905 | * hence give us the correct page_dirty count. On any other page, | |
906 | * it will be zero and in that case we need page_dirty to be the | |
907 | * count of buffers on the page. | |
908 | */ | |
909 | end_offset = min_t(unsigned long long, | |
910 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 911 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
912 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
913 | PAGE_CACHE_SIZE); | |
914 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
915 | page_dirty = p_offset / len; |
916 | ||
24e17b5f | 917 | bh = head = page_buffers(page); |
f6d6d4fc | 918 | offset = page_offset(page); |
6c4fe19f CH |
919 | flags = -1; |
920 | type = 0; | |
f6d6d4fc | 921 | |
f6d6d4fc | 922 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
923 | |
924 | do { | |
925 | if (offset >= end_offset) | |
926 | break; | |
927 | if (!buffer_uptodate(bh)) | |
928 | uptodate = 0; | |
f6d6d4fc | 929 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
930 | /* |
931 | * the iomap is actually still valid, but the ioend | |
932 | * isn't. shouldn't happen too often. | |
933 | */ | |
934 | iomap_valid = 0; | |
1da177e4 | 935 | continue; |
f6d6d4fc | 936 | } |
1da177e4 | 937 | |
1defeac9 CH |
938 | if (iomap_valid) |
939 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
940 | |
941 | /* | |
942 | * First case, map an unwritten extent and prepare for | |
943 | * extent state conversion transaction on completion. | |
f6d6d4fc | 944 | * |
1da177e4 LT |
945 | * Second case, allocate space for a delalloc buffer. |
946 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
947 | * |
948 | * Third case, an unmapped buffer was found, and we are | |
949 | * in a path where we need to write the whole page out. | |
950 | */ | |
951 | if (buffer_unwritten(bh) || buffer_delay(bh) || | |
952 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
953 | !buffer_mapped(bh) && (unmapped || startio))) { | |
6c4fe19f CH |
954 | /* |
955 | * Make sure we don't use a read-only iomap | |
956 | */ | |
957 | if (flags == BMAPI_READ) | |
958 | iomap_valid = 0; | |
959 | ||
f6d6d4fc CH |
960 | if (buffer_unwritten(bh)) { |
961 | type = IOMAP_UNWRITTEN; | |
962 | flags = BMAPI_WRITE|BMAPI_IGNSTATE; | |
d5cb48aa | 963 | } else if (buffer_delay(bh)) { |
f6d6d4fc CH |
964 | type = IOMAP_DELAY; |
965 | flags = BMAPI_ALLOCATE; | |
966 | if (!startio) | |
967 | flags |= trylock_flag; | |
d5cb48aa | 968 | } else { |
6c4fe19f | 969 | type = IOMAP_NEW; |
d5cb48aa | 970 | flags = BMAPI_WRITE|BMAPI_MMAP; |
f6d6d4fc CH |
971 | } |
972 | ||
1defeac9 | 973 | if (!iomap_valid) { |
6c4fe19f CH |
974 | if (type == IOMAP_NEW) { |
975 | size = xfs_probe_cluster(inode, | |
976 | page, bh, head, 0); | |
d5cb48aa CH |
977 | } else { |
978 | size = len; | |
979 | } | |
980 | ||
981 | err = xfs_map_blocks(inode, offset, size, | |
982 | &iomap, flags); | |
f6d6d4fc | 983 | if (err) |
1da177e4 | 984 | goto error; |
1defeac9 | 985 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 986 | } |
1defeac9 CH |
987 | if (iomap_valid) { |
988 | xfs_map_at_offset(bh, offset, | |
989 | inode->i_blkbits, &iomap); | |
1da177e4 | 990 | if (startio) { |
7336cea8 | 991 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 CH |
992 | type, &ioend, |
993 | !iomap_valid); | |
1da177e4 LT |
994 | } else { |
995 | set_buffer_dirty(bh); | |
996 | unlock_buffer(bh); | |
997 | mark_buffer_dirty(bh); | |
998 | } | |
999 | page_dirty--; | |
f6d6d4fc | 1000 | count++; |
1da177e4 | 1001 | } |
d5cb48aa | 1002 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1003 | /* |
1004 | * we got here because the buffer is already mapped. | |
1005 | * That means it must already have extents allocated | |
1006 | * underneath it. Map the extent by reading it. | |
1007 | */ | |
1008 | if (!iomap_valid || type != 0) { | |
1009 | flags = BMAPI_READ; | |
1010 | size = xfs_probe_cluster(inode, page, bh, | |
1011 | head, 1); | |
1012 | err = xfs_map_blocks(inode, offset, size, | |
1013 | &iomap, flags); | |
1014 | if (err) | |
1015 | goto error; | |
1016 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1017 | } | |
d5cb48aa | 1018 | |
6c4fe19f | 1019 | type = 0; |
d5cb48aa CH |
1020 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
1021 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
1022 | if (iomap_valid) |
1023 | all_bh = 1; | |
7336cea8 | 1024 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1025 | &ioend, !iomap_valid); |
1026 | page_dirty--; | |
1027 | count++; | |
f6d6d4fc | 1028 | } else { |
1defeac9 | 1029 | iomap_valid = 0; |
1da177e4 | 1030 | } |
d5cb48aa CH |
1031 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1032 | (unmapped || startio)) { | |
1033 | iomap_valid = 0; | |
1da177e4 | 1034 | } |
f6d6d4fc CH |
1035 | |
1036 | if (!iohead) | |
1037 | iohead = ioend; | |
1038 | ||
1039 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1040 | |
1041 | if (uptodate && bh == head) | |
1042 | SetPageUptodate(page); | |
1043 | ||
f6d6d4fc CH |
1044 | if (startio) |
1045 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1046 | |
1defeac9 CH |
1047 | if (ioend && iomap_valid) { |
1048 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1049 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1050 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1051 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1052 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1053 | } |
1054 | ||
f6d6d4fc CH |
1055 | if (iohead) |
1056 | xfs_submit_ioend(iohead); | |
1057 | ||
1da177e4 LT |
1058 | return page_dirty; |
1059 | ||
1060 | error: | |
f6d6d4fc CH |
1061 | if (iohead) |
1062 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1063 | |
1064 | /* | |
1065 | * If it's delalloc and we have nowhere to put it, | |
1066 | * throw it away, unless the lower layers told | |
1067 | * us to try again. | |
1068 | */ | |
1069 | if (err != -EAGAIN) { | |
f6d6d4fc | 1070 | if (!unmapped) |
1da177e4 | 1071 | block_invalidatepage(page, 0); |
1da177e4 LT |
1072 | ClearPageUptodate(page); |
1073 | } | |
1074 | return err; | |
1075 | } | |
1076 | ||
f51623b2 NS |
1077 | /* |
1078 | * writepage: Called from one of two places: | |
1079 | * | |
1080 | * 1. we are flushing a delalloc buffer head. | |
1081 | * | |
1082 | * 2. we are writing out a dirty page. Typically the page dirty | |
1083 | * state is cleared before we get here. In this case is it | |
1084 | * conceivable we have no buffer heads. | |
1085 | * | |
1086 | * For delalloc space on the page we need to allocate space and | |
1087 | * flush it. For unmapped buffer heads on the page we should | |
1088 | * allocate space if the page is uptodate. For any other dirty | |
1089 | * buffer heads on the page we should flush them. | |
1090 | * | |
1091 | * If we detect that a transaction would be required to flush | |
1092 | * the page, we have to check the process flags first, if we | |
1093 | * are already in a transaction or disk I/O during allocations | |
1094 | * is off, we need to fail the writepage and redirty the page. | |
1095 | */ | |
1096 | ||
1097 | STATIC int | |
1098 | linvfs_writepage( | |
1099 | struct page *page, | |
1100 | struct writeback_control *wbc) | |
1101 | { | |
1102 | int error; | |
1103 | int need_trans; | |
1104 | int delalloc, unmapped, unwritten; | |
1105 | struct inode *inode = page->mapping->host; | |
1106 | ||
1107 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1108 | ||
1109 | /* | |
1110 | * We need a transaction if: | |
1111 | * 1. There are delalloc buffers on the page | |
1112 | * 2. The page is uptodate and we have unmapped buffers | |
1113 | * 3. The page is uptodate and we have no buffers | |
1114 | * 4. There are unwritten buffers on the page | |
1115 | */ | |
1116 | ||
1117 | if (!page_has_buffers(page)) { | |
1118 | unmapped = 1; | |
1119 | need_trans = 1; | |
1120 | } else { | |
1121 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1122 | if (!PageUptodate(page)) | |
1123 | unmapped = 0; | |
1124 | need_trans = delalloc + unmapped + unwritten; | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * If we need a transaction and the process flags say | |
1129 | * we are already in a transaction, or no IO is allowed | |
1130 | * then mark the page dirty again and leave the page | |
1131 | * as is. | |
1132 | */ | |
1133 | if (PFLAGS_TEST_FSTRANS() && need_trans) | |
1134 | goto out_fail; | |
1135 | ||
1136 | /* | |
1137 | * Delay hooking up buffer heads until we have | |
1138 | * made our go/no-go decision. | |
1139 | */ | |
1140 | if (!page_has_buffers(page)) | |
1141 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1142 | ||
1143 | /* | |
1144 | * Convert delayed allocate, unwritten or unmapped space | |
1145 | * to real space and flush out to disk. | |
1146 | */ | |
1147 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1148 | if (error == -EAGAIN) | |
1149 | goto out_fail; | |
1150 | if (unlikely(error < 0)) | |
1151 | goto out_unlock; | |
1152 | ||
1153 | return 0; | |
1154 | ||
1155 | out_fail: | |
1156 | redirty_page_for_writepage(wbc, page); | |
1157 | unlock_page(page); | |
1158 | return 0; | |
1159 | out_unlock: | |
1160 | unlock_page(page); | |
1161 | return error; | |
1162 | } | |
1163 | ||
1164 | /* | |
1165 | * Called to move a page into cleanable state - and from there | |
1166 | * to be released. Possibly the page is already clean. We always | |
1167 | * have buffer heads in this call. | |
1168 | * | |
1169 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1170 | * | |
1171 | * Possible scenarios are: | |
1172 | * | |
1173 | * 1. We are being called to release a page which has been written | |
1174 | * to via regular I/O. buffer heads will be dirty and possibly | |
1175 | * delalloc. If no delalloc buffer heads in this case then we | |
1176 | * can just return zero. | |
1177 | * | |
1178 | * 2. We are called to release a page which has been written via | |
1179 | * mmap, all we need to do is ensure there is no delalloc | |
1180 | * state in the buffer heads, if not we can let the caller | |
1181 | * free them and we should come back later via writepage. | |
1182 | */ | |
1183 | STATIC int | |
1184 | linvfs_release_page( | |
1185 | struct page *page, | |
1186 | gfp_t gfp_mask) | |
1187 | { | |
1188 | struct inode *inode = page->mapping->host; | |
1189 | int dirty, delalloc, unmapped, unwritten; | |
1190 | struct writeback_control wbc = { | |
1191 | .sync_mode = WB_SYNC_ALL, | |
1192 | .nr_to_write = 1, | |
1193 | }; | |
1194 | ||
1195 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask); | |
1196 | ||
1197 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1198 | if (!delalloc && !unwritten) | |
1199 | goto free_buffers; | |
1200 | ||
1201 | if (!(gfp_mask & __GFP_FS)) | |
1202 | return 0; | |
1203 | ||
1204 | /* If we are already inside a transaction or the thread cannot | |
1205 | * do I/O, we cannot release this page. | |
1206 | */ | |
1207 | if (PFLAGS_TEST_FSTRANS()) | |
1208 | return 0; | |
1209 | ||
1210 | /* | |
1211 | * Convert delalloc space to real space, do not flush the | |
1212 | * data out to disk, that will be done by the caller. | |
1213 | * Never need to allocate space here - we will always | |
1214 | * come back to writepage in that case. | |
1215 | */ | |
1216 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1217 | if (dirty == 0 && !unwritten) | |
1218 | goto free_buffers; | |
1219 | return 0; | |
1220 | ||
1221 | free_buffers: | |
1222 | return try_to_free_buffers(page); | |
1223 | } | |
1224 | ||
1da177e4 LT |
1225 | STATIC int |
1226 | __linvfs_get_block( | |
1227 | struct inode *inode, | |
1228 | sector_t iblock, | |
1229 | unsigned long blocks, | |
1230 | struct buffer_head *bh_result, | |
1231 | int create, | |
1232 | int direct, | |
1233 | bmapi_flags_t flags) | |
1234 | { | |
1235 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1236 | xfs_iomap_t iomap; | |
fdc7ed75 NS |
1237 | xfs_off_t offset; |
1238 | ssize_t size; | |
1da177e4 LT |
1239 | int retpbbm = 1; |
1240 | int error; | |
1da177e4 | 1241 | |
fdc7ed75 | 1242 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
a4656391 NS |
1243 | if (blocks) |
1244 | size = (ssize_t) min_t(xfs_off_t, LONG_MAX, | |
1245 | (xfs_off_t)blocks << inode->i_blkbits); | |
1246 | else | |
1247 | size = 1 << inode->i_blkbits; | |
1da177e4 LT |
1248 | |
1249 | VOP_BMAP(vp, offset, size, | |
1250 | create ? flags : BMAPI_READ, &iomap, &retpbbm, error); | |
1251 | if (error) | |
1252 | return -error; | |
1253 | ||
1254 | if (retpbbm == 0) | |
1255 | return 0; | |
1256 | ||
1257 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1258 | /* |
1259 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1260 | * the read case (treat as if we're reading into a hole). |
1261 | */ | |
1262 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1263 | xfs_map_buffer(bh_result, &iomap, offset, |
1264 | inode->i_blkbits); | |
1da177e4 LT |
1265 | } |
1266 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1267 | if (direct) | |
1268 | bh_result->b_private = inode; | |
1269 | set_buffer_unwritten(bh_result); | |
1270 | set_buffer_delay(bh_result); | |
1271 | } | |
1272 | } | |
1273 | ||
1274 | /* If this is a realtime file, data might be on a new device */ | |
ce8e922c | 1275 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 LT |
1276 | |
1277 | /* If we previously allocated a block out beyond eof and | |
1278 | * we are now coming back to use it then we will need to | |
1279 | * flag it as new even if it has a disk address. | |
1280 | */ | |
1281 | if (create && | |
1282 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
fdc7ed75 | 1283 | (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) |
1da177e4 | 1284 | set_buffer_new(bh_result); |
1da177e4 LT |
1285 | |
1286 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1287 | BUG_ON(direct); | |
1288 | if (create) { | |
1289 | set_buffer_uptodate(bh_result); | |
1290 | set_buffer_mapped(bh_result); | |
1291 | set_buffer_delay(bh_result); | |
1292 | } | |
1293 | } | |
1294 | ||
1295 | if (blocks) { | |
fdc7ed75 NS |
1296 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1297 | offset = min_t(xfs_off_t, | |
1298 | iomap.iomap_bsize - iomap.iomap_delta, | |
a4656391 | 1299 | (xfs_off_t)blocks << inode->i_blkbits); |
fdc7ed75 | 1300 | bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset); |
1da177e4 LT |
1301 | } |
1302 | ||
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | int | |
1307 | linvfs_get_block( | |
1308 | struct inode *inode, | |
1309 | sector_t iblock, | |
1310 | struct buffer_head *bh_result, | |
1311 | int create) | |
1312 | { | |
1313 | return __linvfs_get_block(inode, iblock, 0, bh_result, | |
1314 | create, 0, BMAPI_WRITE); | |
1315 | } | |
1316 | ||
1317 | STATIC int | |
1318 | linvfs_get_blocks_direct( | |
1319 | struct inode *inode, | |
1320 | sector_t iblock, | |
1321 | unsigned long max_blocks, | |
1322 | struct buffer_head *bh_result, | |
1323 | int create) | |
1324 | { | |
1325 | return __linvfs_get_block(inode, iblock, max_blocks, bh_result, | |
1326 | create, 1, BMAPI_WRITE|BMAPI_DIRECT); | |
1327 | } | |
1328 | ||
f0973863 CH |
1329 | STATIC void |
1330 | linvfs_end_io_direct( | |
1331 | struct kiocb *iocb, | |
1332 | loff_t offset, | |
1333 | ssize_t size, | |
1334 | void *private) | |
1335 | { | |
1336 | xfs_ioend_t *ioend = iocb->private; | |
1337 | ||
1338 | /* | |
1339 | * Non-NULL private data means we need to issue a transaction to | |
1340 | * convert a range from unwritten to written extents. This needs | |
1341 | * to happen from process contect but aio+dio I/O completion | |
1342 | * happens from irq context so we need to defer it to a workqueue. | |
1343 | * This is not nessecary for synchronous direct I/O, but we do | |
1344 | * it anyway to keep the code uniform and simpler. | |
1345 | * | |
1346 | * The core direct I/O code might be changed to always call the | |
1347 | * completion handler in the future, in which case all this can | |
1348 | * go away. | |
1349 | */ | |
1350 | if (private && size > 0) { | |
1351 | ioend->io_offset = offset; | |
1352 | ioend->io_size = size; | |
1353 | xfs_finish_ioend(ioend); | |
1354 | } else { | |
1355 | ASSERT(size >= 0); | |
1356 | xfs_destroy_ioend(ioend); | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * blockdev_direct_IO can return an error even afer the I/O | |
1361 | * completion handler was called. Thus we need to protect | |
1362 | * against double-freeing. | |
1363 | */ | |
1364 | iocb->private = NULL; | |
1365 | } | |
1366 | ||
1da177e4 LT |
1367 | STATIC ssize_t |
1368 | linvfs_direct_IO( | |
1369 | int rw, | |
1370 | struct kiocb *iocb, | |
1371 | const struct iovec *iov, | |
1372 | loff_t offset, | |
1373 | unsigned long nr_segs) | |
1374 | { | |
1375 | struct file *file = iocb->ki_filp; | |
1376 | struct inode *inode = file->f_mapping->host; | |
1377 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1378 | xfs_iomap_t iomap; | |
1379 | int maps = 1; | |
1380 | int error; | |
f0973863 | 1381 | ssize_t ret; |
1da177e4 LT |
1382 | |
1383 | VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error); | |
1384 | if (error) | |
1385 | return -error; | |
1386 | ||
f6d6d4fc | 1387 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
f0973863 CH |
1388 | |
1389 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, | |
ce8e922c | 1390 | iomap.iomap_target->bt_bdev, |
1da177e4 LT |
1391 | iov, offset, nr_segs, |
1392 | linvfs_get_blocks_direct, | |
f0973863 CH |
1393 | linvfs_end_io_direct); |
1394 | ||
1395 | if (unlikely(ret <= 0 && iocb->private)) | |
1396 | xfs_destroy_ioend(iocb->private); | |
1397 | return ret; | |
1da177e4 LT |
1398 | } |
1399 | ||
f51623b2 NS |
1400 | STATIC int |
1401 | linvfs_prepare_write( | |
1402 | struct file *file, | |
1403 | struct page *page, | |
1404 | unsigned int from, | |
1405 | unsigned int to) | |
1406 | { | |
1407 | return block_prepare_write(page, from, to, linvfs_get_block); | |
1408 | } | |
1da177e4 LT |
1409 | |
1410 | STATIC sector_t | |
1411 | linvfs_bmap( | |
1412 | struct address_space *mapping, | |
1413 | sector_t block) | |
1414 | { | |
1415 | struct inode *inode = (struct inode *)mapping->host; | |
1416 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1417 | int error; | |
1418 | ||
1419 | vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address); | |
1420 | ||
1421 | VOP_RWLOCK(vp, VRWLOCK_READ); | |
1422 | VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error); | |
1423 | VOP_RWUNLOCK(vp, VRWLOCK_READ); | |
1424 | return generic_block_bmap(mapping, block, linvfs_get_block); | |
1425 | } | |
1426 | ||
1427 | STATIC int | |
1428 | linvfs_readpage( | |
1429 | struct file *unused, | |
1430 | struct page *page) | |
1431 | { | |
1432 | return mpage_readpage(page, linvfs_get_block); | |
1433 | } | |
1434 | ||
1435 | STATIC int | |
1436 | linvfs_readpages( | |
1437 | struct file *unused, | |
1438 | struct address_space *mapping, | |
1439 | struct list_head *pages, | |
1440 | unsigned nr_pages) | |
1441 | { | |
1442 | return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block); | |
1443 | } | |
1444 | ||
bcec2b7f NS |
1445 | STATIC int |
1446 | linvfs_invalidate_page( | |
1447 | struct page *page, | |
1448 | unsigned long offset) | |
1449 | { | |
1450 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1451 | page->mapping->host, page, offset); | |
1452 | return block_invalidatepage(page, offset); | |
1453 | } | |
1454 | ||
1da177e4 LT |
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 | }; |