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 | STATIC void |
47 | xfs_count_page_state( | |
48 | struct page *page, | |
49 | int *delalloc, | |
50 | int *unmapped, | |
51 | int *unwritten) | |
52 | { | |
53 | struct buffer_head *bh, *head; | |
54 | ||
55 | *delalloc = *unmapped = *unwritten = 0; | |
56 | ||
57 | bh = head = page_buffers(page); | |
58 | do { | |
59 | if (buffer_uptodate(bh) && !buffer_mapped(bh)) | |
60 | (*unmapped) = 1; | |
61 | else if (buffer_unwritten(bh) && !buffer_delay(bh)) | |
62 | clear_buffer_unwritten(bh); | |
63 | else if (buffer_unwritten(bh)) | |
64 | (*unwritten) = 1; | |
65 | else if (buffer_delay(bh)) | |
66 | (*delalloc) = 1; | |
67 | } while ((bh = bh->b_this_page) != head); | |
68 | } | |
69 | ||
1da177e4 LT |
70 | #if defined(XFS_RW_TRACE) |
71 | void | |
72 | xfs_page_trace( | |
73 | int tag, | |
74 | struct inode *inode, | |
75 | struct page *page, | |
76 | int mask) | |
77 | { | |
78 | xfs_inode_t *ip; | |
1da177e4 LT |
79 | vnode_t *vp = LINVFS_GET_VP(inode); |
80 | loff_t isize = i_size_read(inode); | |
f6d6d4fc | 81 | loff_t offset = page_offset(page); |
1da177e4 LT |
82 | int delalloc = -1, unmapped = -1, unwritten = -1; |
83 | ||
84 | if (page_has_buffers(page)) | |
85 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
86 | ||
75e17b3c | 87 | ip = xfs_vtoi(vp); |
1da177e4 LT |
88 | if (!ip->i_rwtrace) |
89 | return; | |
90 | ||
91 | ktrace_enter(ip->i_rwtrace, | |
92 | (void *)((unsigned long)tag), | |
93 | (void *)ip, | |
94 | (void *)inode, | |
95 | (void *)page, | |
96 | (void *)((unsigned long)mask), | |
97 | (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)), | |
98 | (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)), | |
99 | (void *)((unsigned long)((isize >> 32) & 0xffffffff)), | |
100 | (void *)((unsigned long)(isize & 0xffffffff)), | |
101 | (void *)((unsigned long)((offset >> 32) & 0xffffffff)), | |
102 | (void *)((unsigned long)(offset & 0xffffffff)), | |
103 | (void *)((unsigned long)delalloc), | |
104 | (void *)((unsigned long)unmapped), | |
105 | (void *)((unsigned long)unwritten), | |
106 | (void *)NULL, | |
107 | (void *)NULL); | |
108 | } | |
109 | #else | |
110 | #define xfs_page_trace(tag, inode, page, mask) | |
111 | #endif | |
112 | ||
0829c360 CH |
113 | /* |
114 | * Schedule IO completion handling on a xfsdatad if this was | |
115 | * the final hold on this ioend. | |
116 | */ | |
117 | STATIC void | |
118 | xfs_finish_ioend( | |
119 | xfs_ioend_t *ioend) | |
120 | { | |
121 | if (atomic_dec_and_test(&ioend->io_remaining)) | |
122 | queue_work(xfsdatad_workqueue, &ioend->io_work); | |
123 | } | |
124 | ||
f6d6d4fc CH |
125 | /* |
126 | * We're now finished for good with this ioend structure. | |
127 | * Update the page state via the associated buffer_heads, | |
128 | * release holds on the inode and bio, and finally free | |
129 | * up memory. Do not use the ioend after this. | |
130 | */ | |
0829c360 CH |
131 | STATIC void |
132 | xfs_destroy_ioend( | |
133 | xfs_ioend_t *ioend) | |
134 | { | |
f6d6d4fc CH |
135 | struct buffer_head *bh, *next; |
136 | ||
137 | for (bh = ioend->io_buffer_head; bh; bh = next) { | |
138 | next = bh->b_private; | |
139 | bh->b_end_io(bh, ioend->io_uptodate); | |
140 | } | |
141 | ||
0829c360 CH |
142 | vn_iowake(ioend->io_vnode); |
143 | mempool_free(ioend, xfs_ioend_pool); | |
144 | } | |
145 | ||
146 | /* | |
f6d6d4fc CH |
147 | * Buffered IO write completion for delayed allocate extents. |
148 | * TODO: Update ondisk isize now that we know the file data | |
149 | * has been flushed (i.e. the notorious "NULL file" problem). | |
150 | */ | |
151 | STATIC void | |
152 | xfs_end_bio_delalloc( | |
153 | void *data) | |
154 | { | |
155 | xfs_ioend_t *ioend = data; | |
156 | ||
157 | xfs_destroy_ioend(ioend); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Buffered IO write completion for regular, written extents. | |
162 | */ | |
163 | STATIC void | |
164 | xfs_end_bio_written( | |
165 | void *data) | |
166 | { | |
167 | xfs_ioend_t *ioend = data; | |
168 | ||
169 | xfs_destroy_ioend(ioend); | |
170 | } | |
171 | ||
172 | /* | |
173 | * IO write completion for unwritten extents. | |
174 | * | |
0829c360 | 175 | * Issue transactions to convert a buffer range from unwritten |
f0973863 | 176 | * to written extents. |
0829c360 CH |
177 | */ |
178 | STATIC void | |
179 | xfs_end_bio_unwritten( | |
180 | void *data) | |
181 | { | |
182 | xfs_ioend_t *ioend = data; | |
183 | vnode_t *vp = ioend->io_vnode; | |
184 | xfs_off_t offset = ioend->io_offset; | |
185 | size_t size = ioend->io_size; | |
186 | int error; | |
187 | ||
188 | if (ioend->io_uptodate) | |
189 | VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error); | |
190 | xfs_destroy_ioend(ioend); | |
191 | } | |
192 | ||
193 | /* | |
194 | * Allocate and initialise an IO completion structure. | |
195 | * We need to track unwritten extent write completion here initially. | |
196 | * We'll need to extend this for updating the ondisk inode size later | |
197 | * (vs. incore size). | |
198 | */ | |
199 | STATIC xfs_ioend_t * | |
200 | xfs_alloc_ioend( | |
f6d6d4fc CH |
201 | struct inode *inode, |
202 | unsigned int type) | |
0829c360 CH |
203 | { |
204 | xfs_ioend_t *ioend; | |
205 | ||
206 | ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS); | |
207 | ||
208 | /* | |
209 | * Set the count to 1 initially, which will prevent an I/O | |
210 | * completion callback from happening before we have started | |
211 | * all the I/O from calling the completion routine too early. | |
212 | */ | |
213 | atomic_set(&ioend->io_remaining, 1); | |
214 | ioend->io_uptodate = 1; /* cleared if any I/O fails */ | |
f6d6d4fc CH |
215 | ioend->io_list = NULL; |
216 | ioend->io_type = type; | |
0829c360 | 217 | ioend->io_vnode = LINVFS_GET_VP(inode); |
c1a073bd | 218 | ioend->io_buffer_head = NULL; |
f6d6d4fc | 219 | ioend->io_buffer_tail = NULL; |
0829c360 CH |
220 | atomic_inc(&ioend->io_vnode->v_iocount); |
221 | ioend->io_offset = 0; | |
222 | ioend->io_size = 0; | |
223 | ||
f6d6d4fc CH |
224 | if (type == IOMAP_UNWRITTEN) |
225 | INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten, ioend); | |
226 | else if (type == IOMAP_DELAY) | |
227 | INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc, ioend); | |
228 | else | |
229 | INIT_WORK(&ioend->io_work, xfs_end_bio_written, ioend); | |
0829c360 CH |
230 | |
231 | return ioend; | |
232 | } | |
233 | ||
1da177e4 LT |
234 | STATIC int |
235 | xfs_map_blocks( | |
236 | struct inode *inode, | |
237 | loff_t offset, | |
238 | ssize_t count, | |
239 | xfs_iomap_t *mapp, | |
240 | int flags) | |
241 | { | |
242 | vnode_t *vp = LINVFS_GET_VP(inode); | |
243 | int error, nmaps = 1; | |
244 | ||
245 | VOP_BMAP(vp, offset, count, flags, mapp, &nmaps, error); | |
246 | if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE))) | |
247 | VMODIFY(vp); | |
248 | return -error; | |
249 | } | |
250 | ||
1defeac9 CH |
251 | STATIC inline int |
252 | xfs_iomap_valid( | |
1da177e4 | 253 | xfs_iomap_t *iomapp, |
1defeac9 | 254 | loff_t offset) |
1da177e4 | 255 | { |
1defeac9 CH |
256 | return offset >= iomapp->iomap_offset && |
257 | offset < iomapp->iomap_offset + iomapp->iomap_bsize; | |
1da177e4 LT |
258 | } |
259 | ||
f6d6d4fc CH |
260 | /* |
261 | * BIO completion handler for buffered IO. | |
262 | */ | |
263 | STATIC int | |
264 | xfs_end_bio( | |
265 | struct bio *bio, | |
266 | unsigned int bytes_done, | |
267 | int error) | |
268 | { | |
269 | xfs_ioend_t *ioend = bio->bi_private; | |
270 | ||
271 | if (bio->bi_size) | |
272 | return 1; | |
273 | ||
274 | ASSERT(ioend); | |
275 | ASSERT(atomic_read(&bio->bi_cnt) >= 1); | |
276 | ||
277 | /* Toss bio and pass work off to an xfsdatad thread */ | |
278 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
279 | ioend->io_uptodate = 0; | |
280 | bio->bi_private = NULL; | |
281 | bio->bi_end_io = NULL; | |
282 | ||
283 | bio_put(bio); | |
284 | xfs_finish_ioend(ioend); | |
285 | return 0; | |
286 | } | |
287 | ||
288 | STATIC void | |
289 | xfs_submit_ioend_bio( | |
290 | xfs_ioend_t *ioend, | |
291 | struct bio *bio) | |
292 | { | |
293 | atomic_inc(&ioend->io_remaining); | |
294 | ||
295 | bio->bi_private = ioend; | |
296 | bio->bi_end_io = xfs_end_bio; | |
297 | ||
298 | submit_bio(WRITE, bio); | |
299 | ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP)); | |
300 | bio_put(bio); | |
301 | } | |
302 | ||
303 | STATIC struct bio * | |
304 | xfs_alloc_ioend_bio( | |
305 | struct buffer_head *bh) | |
306 | { | |
307 | struct bio *bio; | |
308 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
309 | ||
310 | do { | |
311 | bio = bio_alloc(GFP_NOIO, nvecs); | |
312 | nvecs >>= 1; | |
313 | } while (!bio); | |
314 | ||
315 | ASSERT(bio->bi_private == NULL); | |
316 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
317 | bio->bi_bdev = bh->b_bdev; | |
318 | bio_get(bio); | |
319 | return bio; | |
320 | } | |
321 | ||
322 | STATIC void | |
323 | xfs_start_buffer_writeback( | |
324 | struct buffer_head *bh) | |
325 | { | |
326 | ASSERT(buffer_mapped(bh)); | |
327 | ASSERT(buffer_locked(bh)); | |
328 | ASSERT(!buffer_delay(bh)); | |
329 | ASSERT(!buffer_unwritten(bh)); | |
330 | ||
331 | mark_buffer_async_write(bh); | |
332 | set_buffer_uptodate(bh); | |
333 | clear_buffer_dirty(bh); | |
334 | } | |
335 | ||
336 | STATIC void | |
337 | xfs_start_page_writeback( | |
338 | struct page *page, | |
339 | struct writeback_control *wbc, | |
340 | int clear_dirty, | |
341 | int buffers) | |
342 | { | |
343 | ASSERT(PageLocked(page)); | |
344 | ASSERT(!PageWriteback(page)); | |
345 | set_page_writeback(page); | |
346 | if (clear_dirty) | |
347 | clear_page_dirty(page); | |
348 | unlock_page(page); | |
349 | if (!buffers) { | |
350 | end_page_writeback(page); | |
351 | wbc->pages_skipped++; /* We didn't write this page */ | |
352 | } | |
353 | } | |
354 | ||
355 | static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh) | |
356 | { | |
357 | return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
358 | } | |
359 | ||
360 | /* | |
d88992f6 DC |
361 | * Submit all of the bios for all of the ioends we have saved up, covering the |
362 | * initial writepage page and also any probed pages. | |
363 | * | |
364 | * Because we may have multiple ioends spanning a page, we need to start | |
365 | * writeback on all the buffers before we submit them for I/O. If we mark the | |
366 | * buffers as we got, then we can end up with a page that only has buffers | |
367 | * marked async write and I/O complete on can occur before we mark the other | |
368 | * buffers async write. | |
369 | * | |
370 | * The end result of this is that we trip a bug in end_page_writeback() because | |
371 | * we call it twice for the one page as the code in end_buffer_async_write() | |
372 | * assumes that all buffers on the page are started at the same time. | |
373 | * | |
374 | * The fix is two passes across the ioend list - one to start writeback on the | |
375 | * bufferheads, and then the second one submit them for I/O. | |
f6d6d4fc CH |
376 | */ |
377 | STATIC void | |
378 | xfs_submit_ioend( | |
379 | xfs_ioend_t *ioend) | |
380 | { | |
d88992f6 | 381 | xfs_ioend_t *head = ioend; |
f6d6d4fc CH |
382 | xfs_ioend_t *next; |
383 | struct buffer_head *bh; | |
384 | struct bio *bio; | |
385 | sector_t lastblock = 0; | |
386 | ||
d88992f6 DC |
387 | /* Pass 1 - start writeback */ |
388 | do { | |
389 | next = ioend->io_list; | |
390 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
391 | xfs_start_buffer_writeback(bh); | |
392 | } | |
393 | } while ((ioend = next) != NULL); | |
394 | ||
395 | /* Pass 2 - submit I/O */ | |
396 | ioend = head; | |
f6d6d4fc CH |
397 | do { |
398 | next = ioend->io_list; | |
399 | bio = NULL; | |
400 | ||
401 | for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) { | |
f6d6d4fc CH |
402 | |
403 | if (!bio) { | |
404 | retry: | |
405 | bio = xfs_alloc_ioend_bio(bh); | |
406 | } else if (bh->b_blocknr != lastblock + 1) { | |
407 | xfs_submit_ioend_bio(ioend, bio); | |
408 | goto retry; | |
409 | } | |
410 | ||
411 | if (bio_add_buffer(bio, bh) != bh->b_size) { | |
412 | xfs_submit_ioend_bio(ioend, bio); | |
413 | goto retry; | |
414 | } | |
415 | ||
416 | lastblock = bh->b_blocknr; | |
417 | } | |
418 | if (bio) | |
419 | xfs_submit_ioend_bio(ioend, bio); | |
420 | xfs_finish_ioend(ioend); | |
421 | } while ((ioend = next) != NULL); | |
422 | } | |
423 | ||
424 | /* | |
425 | * Cancel submission of all buffer_heads so far in this endio. | |
426 | * Toss the endio too. Only ever called for the initial page | |
427 | * in a writepage request, so only ever one page. | |
428 | */ | |
429 | STATIC void | |
430 | xfs_cancel_ioend( | |
431 | xfs_ioend_t *ioend) | |
432 | { | |
433 | xfs_ioend_t *next; | |
434 | struct buffer_head *bh, *next_bh; | |
435 | ||
436 | do { | |
437 | next = ioend->io_list; | |
438 | bh = ioend->io_buffer_head; | |
439 | do { | |
440 | next_bh = bh->b_private; | |
441 | clear_buffer_async_write(bh); | |
442 | unlock_buffer(bh); | |
443 | } while ((bh = next_bh) != NULL); | |
444 | ||
445 | vn_iowake(ioend->io_vnode); | |
446 | mempool_free(ioend, xfs_ioend_pool); | |
447 | } while ((ioend = next) != NULL); | |
448 | } | |
449 | ||
450 | /* | |
451 | * Test to see if we've been building up a completion structure for | |
452 | * earlier buffers -- if so, we try to append to this ioend if we | |
453 | * can, otherwise we finish off any current ioend and start another. | |
454 | * Return true if we've finished the given ioend. | |
455 | */ | |
456 | STATIC void | |
457 | xfs_add_to_ioend( | |
458 | struct inode *inode, | |
459 | struct buffer_head *bh, | |
7336cea8 | 460 | xfs_off_t offset, |
f6d6d4fc CH |
461 | unsigned int type, |
462 | xfs_ioend_t **result, | |
463 | int need_ioend) | |
464 | { | |
465 | xfs_ioend_t *ioend = *result; | |
466 | ||
467 | if (!ioend || need_ioend || type != ioend->io_type) { | |
468 | xfs_ioend_t *previous = *result; | |
f6d6d4fc | 469 | |
f6d6d4fc CH |
470 | ioend = xfs_alloc_ioend(inode, type); |
471 | ioend->io_offset = offset; | |
472 | ioend->io_buffer_head = bh; | |
473 | ioend->io_buffer_tail = bh; | |
474 | if (previous) | |
475 | previous->io_list = ioend; | |
476 | *result = ioend; | |
477 | } else { | |
478 | ioend->io_buffer_tail->b_private = bh; | |
479 | ioend->io_buffer_tail = bh; | |
480 | } | |
481 | ||
482 | bh->b_private = NULL; | |
483 | ioend->io_size += bh->b_size; | |
484 | } | |
485 | ||
87cbc49c NS |
486 | STATIC void |
487 | xfs_map_buffer( | |
488 | struct buffer_head *bh, | |
489 | xfs_iomap_t *mp, | |
490 | xfs_off_t offset, | |
491 | uint block_bits) | |
492 | { | |
493 | sector_t bn; | |
494 | ||
495 | ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL); | |
496 | ||
497 | bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) + | |
498 | ((offset - mp->iomap_offset) >> block_bits); | |
499 | ||
500 | ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME)); | |
501 | ||
502 | bh->b_blocknr = bn; | |
503 | set_buffer_mapped(bh); | |
504 | } | |
505 | ||
1da177e4 LT |
506 | STATIC void |
507 | xfs_map_at_offset( | |
1da177e4 | 508 | struct buffer_head *bh, |
1defeac9 | 509 | loff_t offset, |
1da177e4 | 510 | int block_bits, |
1defeac9 | 511 | xfs_iomap_t *iomapp) |
1da177e4 | 512 | { |
1da177e4 LT |
513 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); |
514 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
1da177e4 LT |
515 | |
516 | lock_buffer(bh); | |
87cbc49c | 517 | xfs_map_buffer(bh, iomapp, offset, block_bits); |
ce8e922c | 518 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
519 | set_buffer_mapped(bh); |
520 | clear_buffer_delay(bh); | |
f6d6d4fc | 521 | clear_buffer_unwritten(bh); |
1da177e4 LT |
522 | } |
523 | ||
524 | /* | |
6c4fe19f | 525 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
526 | */ |
527 | STATIC unsigned int | |
6c4fe19f | 528 | xfs_probe_page( |
10ce4444 | 529 | struct page *page, |
6c4fe19f CH |
530 | unsigned int pg_offset, |
531 | int mapped) | |
1da177e4 | 532 | { |
1da177e4 LT |
533 | int ret = 0; |
534 | ||
1da177e4 | 535 | if (PageWriteback(page)) |
10ce4444 | 536 | return 0; |
1da177e4 LT |
537 | |
538 | if (page->mapping && PageDirty(page)) { | |
539 | if (page_has_buffers(page)) { | |
540 | struct buffer_head *bh, *head; | |
541 | ||
542 | bh = head = page_buffers(page); | |
543 | do { | |
6c4fe19f CH |
544 | if (!buffer_uptodate(bh)) |
545 | break; | |
546 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
547 | break; |
548 | ret += bh->b_size; | |
549 | if (ret >= pg_offset) | |
550 | break; | |
551 | } while ((bh = bh->b_this_page) != head); | |
552 | } else | |
6c4fe19f | 553 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
554 | } |
555 | ||
1da177e4 LT |
556 | return ret; |
557 | } | |
558 | ||
f6d6d4fc | 559 | STATIC size_t |
6c4fe19f | 560 | xfs_probe_cluster( |
1da177e4 LT |
561 | struct inode *inode, |
562 | struct page *startpage, | |
563 | struct buffer_head *bh, | |
6c4fe19f CH |
564 | struct buffer_head *head, |
565 | int mapped) | |
1da177e4 | 566 | { |
10ce4444 | 567 | struct pagevec pvec; |
1da177e4 | 568 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
569 | size_t total = 0; |
570 | int done = 0, i; | |
1da177e4 LT |
571 | |
572 | /* First sum forwards in this page */ | |
573 | do { | |
2353e8e9 | 574 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 575 | return total; |
1da177e4 LT |
576 | total += bh->b_size; |
577 | } while ((bh = bh->b_this_page) != head); | |
578 | ||
10ce4444 CH |
579 | /* if we reached the end of the page, sum forwards in following pages */ |
580 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
581 | tindex = startpage->index + 1; | |
582 | ||
583 | /* Prune this back to avoid pathological behavior */ | |
584 | tloff = min(tlast, startpage->index + 64); | |
585 | ||
586 | pagevec_init(&pvec, 0); | |
587 | while (!done && tindex <= tloff) { | |
588 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
589 | ||
590 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
591 | break; | |
592 | ||
593 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
594 | struct page *page = pvec.pages[i]; | |
595 | size_t pg_offset, len = 0; | |
596 | ||
597 | if (tindex == tlast) { | |
598 | pg_offset = | |
599 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
600 | if (!pg_offset) { |
601 | done = 1; | |
10ce4444 | 602 | break; |
1defeac9 | 603 | } |
10ce4444 CH |
604 | } else |
605 | pg_offset = PAGE_CACHE_SIZE; | |
606 | ||
607 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
6c4fe19f | 608 | len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
609 | unlock_page(page); |
610 | } | |
611 | ||
612 | if (!len) { | |
613 | done = 1; | |
614 | break; | |
615 | } | |
616 | ||
1da177e4 | 617 | total += len; |
1defeac9 | 618 | tindex++; |
1da177e4 | 619 | } |
10ce4444 CH |
620 | |
621 | pagevec_release(&pvec); | |
622 | cond_resched(); | |
1da177e4 | 623 | } |
10ce4444 | 624 | |
1da177e4 LT |
625 | return total; |
626 | } | |
627 | ||
628 | /* | |
10ce4444 CH |
629 | * Test if a given page is suitable for writing as part of an unwritten |
630 | * or delayed allocate extent. | |
1da177e4 | 631 | */ |
10ce4444 CH |
632 | STATIC int |
633 | xfs_is_delayed_page( | |
634 | struct page *page, | |
f6d6d4fc | 635 | unsigned int type) |
1da177e4 | 636 | { |
1da177e4 | 637 | if (PageWriteback(page)) |
10ce4444 | 638 | return 0; |
1da177e4 LT |
639 | |
640 | if (page->mapping && page_has_buffers(page)) { | |
641 | struct buffer_head *bh, *head; | |
642 | int acceptable = 0; | |
643 | ||
644 | bh = head = page_buffers(page); | |
645 | do { | |
f6d6d4fc CH |
646 | if (buffer_unwritten(bh)) |
647 | acceptable = (type == IOMAP_UNWRITTEN); | |
648 | else if (buffer_delay(bh)) | |
649 | acceptable = (type == IOMAP_DELAY); | |
6c4fe19f CH |
650 | else if (buffer_mapped(bh)) |
651 | acceptable = (type == 0); | |
f6d6d4fc | 652 | else |
1da177e4 | 653 | break; |
1da177e4 LT |
654 | } while ((bh = bh->b_this_page) != head); |
655 | ||
656 | if (acceptable) | |
10ce4444 | 657 | return 1; |
1da177e4 LT |
658 | } |
659 | ||
10ce4444 | 660 | return 0; |
1da177e4 LT |
661 | } |
662 | ||
1da177e4 LT |
663 | /* |
664 | * Allocate & map buffers for page given the extent map. Write it out. | |
665 | * except for the original page of a writepage, this is called on | |
666 | * delalloc/unwritten pages only, for the original page it is possible | |
667 | * that the page has no mapping at all. | |
668 | */ | |
f6d6d4fc | 669 | STATIC int |
1da177e4 LT |
670 | xfs_convert_page( |
671 | struct inode *inode, | |
672 | struct page *page, | |
10ce4444 | 673 | loff_t tindex, |
1defeac9 | 674 | xfs_iomap_t *mp, |
f6d6d4fc | 675 | xfs_ioend_t **ioendp, |
1da177e4 | 676 | struct writeback_control *wbc, |
1da177e4 LT |
677 | int startio, |
678 | int all_bh) | |
679 | { | |
f6d6d4fc | 680 | struct buffer_head *bh, *head; |
9260dc6b CH |
681 | xfs_off_t end_offset; |
682 | unsigned long p_offset; | |
f6d6d4fc | 683 | unsigned int type; |
1da177e4 | 684 | int bbits = inode->i_blkbits; |
24e17b5f | 685 | int len, page_dirty; |
f6d6d4fc | 686 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 687 | xfs_off_t offset = page_offset(page); |
1da177e4 | 688 | |
10ce4444 CH |
689 | if (page->index != tindex) |
690 | goto fail; | |
691 | if (TestSetPageLocked(page)) | |
692 | goto fail; | |
693 | if (PageWriteback(page)) | |
694 | goto fail_unlock_page; | |
695 | if (page->mapping != inode->i_mapping) | |
696 | goto fail_unlock_page; | |
697 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
698 | goto fail_unlock_page; | |
699 | ||
24e17b5f NS |
700 | /* |
701 | * page_dirty is initially a count of buffers on the page before | |
702 | * EOF and is decrememted as we move each into a cleanable state. | |
9260dc6b CH |
703 | * |
704 | * Derivation: | |
705 | * | |
706 | * End offset is the highest offset that this page should represent. | |
707 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
708 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
709 | * hence give us the correct page_dirty count. On any other page, | |
710 | * it will be zero and in that case we need page_dirty to be the | |
711 | * count of buffers on the page. | |
24e17b5f | 712 | */ |
9260dc6b CH |
713 | end_offset = min_t(unsigned long long, |
714 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
715 | i_size_read(inode)); | |
716 | ||
24e17b5f | 717 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
718 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
719 | PAGE_CACHE_SIZE); | |
720 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
721 | page_dirty = p_offset / len; | |
24e17b5f | 722 | |
1da177e4 LT |
723 | bh = head = page_buffers(page); |
724 | do { | |
9260dc6b | 725 | if (offset >= end_offset) |
1da177e4 | 726 | break; |
f6d6d4fc CH |
727 | if (!buffer_uptodate(bh)) |
728 | uptodate = 0; | |
729 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
730 | done = 1; | |
1da177e4 | 731 | continue; |
f6d6d4fc CH |
732 | } |
733 | ||
9260dc6b CH |
734 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
735 | if (buffer_unwritten(bh)) | |
736 | type = IOMAP_UNWRITTEN; | |
737 | else | |
738 | type = IOMAP_DELAY; | |
739 | ||
740 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 741 | done = 1; |
9260dc6b CH |
742 | continue; |
743 | } | |
744 | ||
745 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
746 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
747 | ||
748 | xfs_map_at_offset(bh, offset, bbits, mp); | |
749 | if (startio) { | |
7336cea8 | 750 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
751 | type, ioendp, done); |
752 | } else { | |
753 | set_buffer_dirty(bh); | |
754 | unlock_buffer(bh); | |
755 | mark_buffer_dirty(bh); | |
756 | } | |
757 | page_dirty--; | |
758 | count++; | |
759 | } else { | |
760 | type = 0; | |
761 | if (buffer_mapped(bh) && all_bh && startio) { | |
1da177e4 | 762 | lock_buffer(bh); |
7336cea8 | 763 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
764 | type, ioendp, done); |
765 | count++; | |
24e17b5f | 766 | page_dirty--; |
9260dc6b CH |
767 | } else { |
768 | done = 1; | |
1da177e4 | 769 | } |
1da177e4 | 770 | } |
7336cea8 | 771 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 772 | |
f6d6d4fc CH |
773 | if (uptodate && bh == head) |
774 | SetPageUptodate(page); | |
775 | ||
776 | if (startio) { | |
f5e596bb CH |
777 | if (count) { |
778 | struct backing_dev_info *bdi; | |
779 | ||
780 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 781 | wbc->nr_to_write--; |
f5e596bb CH |
782 | if (bdi_write_congested(bdi)) { |
783 | wbc->encountered_congestion = 1; | |
784 | done = 1; | |
9fddaca2 | 785 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
786 | done = 1; |
787 | } | |
788 | } | |
f6d6d4fc | 789 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 790 | } |
f6d6d4fc CH |
791 | |
792 | return done; | |
10ce4444 CH |
793 | fail_unlock_page: |
794 | unlock_page(page); | |
795 | fail: | |
796 | return 1; | |
1da177e4 LT |
797 | } |
798 | ||
799 | /* | |
800 | * Convert & write out a cluster of pages in the same extent as defined | |
801 | * by mp and following the start page. | |
802 | */ | |
803 | STATIC void | |
804 | xfs_cluster_write( | |
805 | struct inode *inode, | |
806 | pgoff_t tindex, | |
807 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 808 | xfs_ioend_t **ioendp, |
1da177e4 LT |
809 | struct writeback_control *wbc, |
810 | int startio, | |
811 | int all_bh, | |
812 | pgoff_t tlast) | |
813 | { | |
10ce4444 CH |
814 | struct pagevec pvec; |
815 | int done = 0, i; | |
1da177e4 | 816 | |
10ce4444 CH |
817 | pagevec_init(&pvec, 0); |
818 | while (!done && tindex <= tlast) { | |
819 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
820 | ||
821 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 822 | break; |
10ce4444 CH |
823 | |
824 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
825 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
826 | iomapp, ioendp, wbc, startio, all_bh); | |
827 | if (done) | |
828 | break; | |
829 | } | |
830 | ||
831 | pagevec_release(&pvec); | |
832 | cond_resched(); | |
1da177e4 LT |
833 | } |
834 | } | |
835 | ||
836 | /* | |
837 | * Calling this without startio set means we are being asked to make a dirty | |
838 | * page ready for freeing it's buffers. When called with startio set then | |
839 | * we are coming from writepage. | |
840 | * | |
841 | * When called with startio set it is important that we write the WHOLE | |
842 | * page if possible. | |
843 | * The bh->b_state's cannot know if any of the blocks or which block for | |
844 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
845 | * only vaild if the page itself isn't completely uptodate. Some layers | |
846 | * may clear the page dirty flag prior to calling write page, under the | |
847 | * assumption the entire page will be written out; by not writing out the | |
848 | * whole page the page can be reused before all valid dirty data is | |
849 | * written out. Note: in the case of a page that has been dirty'd by | |
850 | * mapwrite and but partially setup by block_prepare_write the | |
851 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
852 | * valid state, thus the whole page must be written out thing. | |
853 | */ | |
854 | ||
855 | STATIC int | |
856 | xfs_page_state_convert( | |
857 | struct inode *inode, | |
858 | struct page *page, | |
859 | struct writeback_control *wbc, | |
860 | int startio, | |
861 | int unmapped) /* also implies page uptodate */ | |
862 | { | |
f6d6d4fc | 863 | struct buffer_head *bh, *head; |
1defeac9 | 864 | xfs_iomap_t iomap; |
f6d6d4fc | 865 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
866 | loff_t offset; |
867 | unsigned long p_offset = 0; | |
f6d6d4fc | 868 | unsigned int type; |
1da177e4 LT |
869 | __uint64_t end_offset; |
870 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
871 | ssize_t size, len; |
872 | int flags, err, iomap_valid = 0, uptodate = 1; | |
f6d6d4fc | 873 | int page_dirty, count = 0, trylock_flag = 0; |
6c4fe19f | 874 | int all_bh = unmapped; |
1da177e4 | 875 | |
3ba0815a | 876 | /* wait for other IO threads? */ |
f5e596bb | 877 | if (startio && (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)) |
f6d6d4fc | 878 | trylock_flag |= BMAPI_TRYLOCK; |
3ba0815a | 879 | |
1da177e4 LT |
880 | /* Is this page beyond the end of the file? */ |
881 | offset = i_size_read(inode); | |
882 | end_index = offset >> PAGE_CACHE_SHIFT; | |
883 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
884 | if (page->index >= end_index) { | |
885 | if ((page->index >= end_index + 1) || | |
886 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
887 | if (startio) |
888 | unlock_page(page); | |
889 | return 0; | |
1da177e4 LT |
890 | } |
891 | } | |
892 | ||
1da177e4 | 893 | /* |
24e17b5f NS |
894 | * page_dirty is initially a count of buffers on the page before |
895 | * EOF and is decrememted as we move each into a cleanable state. | |
f6d6d4fc CH |
896 | * |
897 | * Derivation: | |
898 | * | |
899 | * End offset is the highest offset that this page should represent. | |
900 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
901 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
902 | * hence give us the correct page_dirty count. On any other page, | |
903 | * it will be zero and in that case we need page_dirty to be the | |
904 | * count of buffers on the page. | |
905 | */ | |
906 | end_offset = min_t(unsigned long long, | |
907 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 908 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
909 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
910 | PAGE_CACHE_SIZE); | |
911 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
912 | page_dirty = p_offset / len; |
913 | ||
24e17b5f | 914 | bh = head = page_buffers(page); |
f6d6d4fc | 915 | offset = page_offset(page); |
6c4fe19f CH |
916 | flags = -1; |
917 | type = 0; | |
f6d6d4fc | 918 | |
f6d6d4fc | 919 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
920 | |
921 | do { | |
922 | if (offset >= end_offset) | |
923 | break; | |
924 | if (!buffer_uptodate(bh)) | |
925 | uptodate = 0; | |
f6d6d4fc | 926 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
927 | /* |
928 | * the iomap is actually still valid, but the ioend | |
929 | * isn't. shouldn't happen too often. | |
930 | */ | |
931 | iomap_valid = 0; | |
1da177e4 | 932 | continue; |
f6d6d4fc | 933 | } |
1da177e4 | 934 | |
1defeac9 CH |
935 | if (iomap_valid) |
936 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
937 | |
938 | /* | |
939 | * First case, map an unwritten extent and prepare for | |
940 | * extent state conversion transaction on completion. | |
f6d6d4fc | 941 | * |
1da177e4 LT |
942 | * Second case, allocate space for a delalloc buffer. |
943 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
944 | * |
945 | * Third case, an unmapped buffer was found, and we are | |
946 | * in a path where we need to write the whole page out. | |
947 | */ | |
948 | if (buffer_unwritten(bh) || buffer_delay(bh) || | |
949 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
950 | !buffer_mapped(bh) && (unmapped || startio))) { | |
6c4fe19f CH |
951 | /* |
952 | * Make sure we don't use a read-only iomap | |
953 | */ | |
954 | if (flags == BMAPI_READ) | |
955 | iomap_valid = 0; | |
956 | ||
f6d6d4fc CH |
957 | if (buffer_unwritten(bh)) { |
958 | type = IOMAP_UNWRITTEN; | |
959 | flags = BMAPI_WRITE|BMAPI_IGNSTATE; | |
d5cb48aa | 960 | } else if (buffer_delay(bh)) { |
f6d6d4fc CH |
961 | type = IOMAP_DELAY; |
962 | flags = BMAPI_ALLOCATE; | |
963 | if (!startio) | |
964 | flags |= trylock_flag; | |
d5cb48aa | 965 | } else { |
6c4fe19f | 966 | type = IOMAP_NEW; |
d5cb48aa | 967 | flags = BMAPI_WRITE|BMAPI_MMAP; |
f6d6d4fc CH |
968 | } |
969 | ||
1defeac9 | 970 | if (!iomap_valid) { |
6c4fe19f CH |
971 | if (type == IOMAP_NEW) { |
972 | size = xfs_probe_cluster(inode, | |
973 | page, bh, head, 0); | |
d5cb48aa CH |
974 | } else { |
975 | size = len; | |
976 | } | |
977 | ||
978 | err = xfs_map_blocks(inode, offset, size, | |
979 | &iomap, flags); | |
f6d6d4fc | 980 | if (err) |
1da177e4 | 981 | goto error; |
1defeac9 | 982 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 983 | } |
1defeac9 CH |
984 | if (iomap_valid) { |
985 | xfs_map_at_offset(bh, offset, | |
986 | inode->i_blkbits, &iomap); | |
1da177e4 | 987 | if (startio) { |
7336cea8 | 988 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 CH |
989 | type, &ioend, |
990 | !iomap_valid); | |
1da177e4 LT |
991 | } else { |
992 | set_buffer_dirty(bh); | |
993 | unlock_buffer(bh); | |
994 | mark_buffer_dirty(bh); | |
995 | } | |
996 | page_dirty--; | |
f6d6d4fc | 997 | count++; |
1da177e4 | 998 | } |
d5cb48aa | 999 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
1000 | /* |
1001 | * we got here because the buffer is already mapped. | |
1002 | * That means it must already have extents allocated | |
1003 | * underneath it. Map the extent by reading it. | |
1004 | */ | |
1005 | if (!iomap_valid || type != 0) { | |
1006 | flags = BMAPI_READ; | |
1007 | size = xfs_probe_cluster(inode, page, bh, | |
1008 | head, 1); | |
1009 | err = xfs_map_blocks(inode, offset, size, | |
1010 | &iomap, flags); | |
1011 | if (err) | |
1012 | goto error; | |
1013 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1014 | } | |
d5cb48aa | 1015 | |
6c4fe19f | 1016 | type = 0; |
d5cb48aa CH |
1017 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
1018 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
1019 | if (iomap_valid) |
1020 | all_bh = 1; | |
7336cea8 | 1021 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1022 | &ioend, !iomap_valid); |
1023 | page_dirty--; | |
1024 | count++; | |
f6d6d4fc | 1025 | } else { |
1defeac9 | 1026 | iomap_valid = 0; |
1da177e4 | 1027 | } |
d5cb48aa CH |
1028 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1029 | (unmapped || startio)) { | |
1030 | iomap_valid = 0; | |
1da177e4 | 1031 | } |
f6d6d4fc CH |
1032 | |
1033 | if (!iohead) | |
1034 | iohead = ioend; | |
1035 | ||
1036 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1037 | |
1038 | if (uptodate && bh == head) | |
1039 | SetPageUptodate(page); | |
1040 | ||
f6d6d4fc CH |
1041 | if (startio) |
1042 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1043 | |
1defeac9 CH |
1044 | if (ioend && iomap_valid) { |
1045 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1046 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1047 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1048 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1049 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1050 | } |
1051 | ||
f6d6d4fc CH |
1052 | if (iohead) |
1053 | xfs_submit_ioend(iohead); | |
1054 | ||
1da177e4 LT |
1055 | return page_dirty; |
1056 | ||
1057 | error: | |
f6d6d4fc CH |
1058 | if (iohead) |
1059 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1060 | |
1061 | /* | |
1062 | * If it's delalloc and we have nowhere to put it, | |
1063 | * throw it away, unless the lower layers told | |
1064 | * us to try again. | |
1065 | */ | |
1066 | if (err != -EAGAIN) { | |
f6d6d4fc | 1067 | if (!unmapped) |
1da177e4 | 1068 | block_invalidatepage(page, 0); |
1da177e4 LT |
1069 | ClearPageUptodate(page); |
1070 | } | |
1071 | return err; | |
1072 | } | |
1073 | ||
f51623b2 NS |
1074 | /* |
1075 | * writepage: Called from one of two places: | |
1076 | * | |
1077 | * 1. we are flushing a delalloc buffer head. | |
1078 | * | |
1079 | * 2. we are writing out a dirty page. Typically the page dirty | |
1080 | * state is cleared before we get here. In this case is it | |
1081 | * conceivable we have no buffer heads. | |
1082 | * | |
1083 | * For delalloc space on the page we need to allocate space and | |
1084 | * flush it. For unmapped buffer heads on the page we should | |
1085 | * allocate space if the page is uptodate. For any other dirty | |
1086 | * buffer heads on the page we should flush them. | |
1087 | * | |
1088 | * If we detect that a transaction would be required to flush | |
1089 | * the page, we have to check the process flags first, if we | |
1090 | * are already in a transaction or disk I/O during allocations | |
1091 | * is off, we need to fail the writepage and redirty the page. | |
1092 | */ | |
1093 | ||
1094 | STATIC int | |
e4c573bb | 1095 | xfs_vm_writepage( |
f51623b2 NS |
1096 | struct page *page, |
1097 | struct writeback_control *wbc) | |
1098 | { | |
1099 | int error; | |
1100 | int need_trans; | |
1101 | int delalloc, unmapped, unwritten; | |
1102 | struct inode *inode = page->mapping->host; | |
1103 | ||
1104 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1105 | ||
1106 | /* | |
1107 | * We need a transaction if: | |
1108 | * 1. There are delalloc buffers on the page | |
1109 | * 2. The page is uptodate and we have unmapped buffers | |
1110 | * 3. The page is uptodate and we have no buffers | |
1111 | * 4. There are unwritten buffers on the page | |
1112 | */ | |
1113 | ||
1114 | if (!page_has_buffers(page)) { | |
1115 | unmapped = 1; | |
1116 | need_trans = 1; | |
1117 | } else { | |
1118 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1119 | if (!PageUptodate(page)) | |
1120 | unmapped = 0; | |
1121 | need_trans = delalloc + unmapped + unwritten; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * If we need a transaction and the process flags say | |
1126 | * we are already in a transaction, or no IO is allowed | |
1127 | * then mark the page dirty again and leave the page | |
1128 | * as is. | |
1129 | */ | |
1130 | if (PFLAGS_TEST_FSTRANS() && need_trans) | |
1131 | goto out_fail; | |
1132 | ||
1133 | /* | |
1134 | * Delay hooking up buffer heads until we have | |
1135 | * made our go/no-go decision. | |
1136 | */ | |
1137 | if (!page_has_buffers(page)) | |
1138 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1139 | ||
1140 | /* | |
1141 | * Convert delayed allocate, unwritten or unmapped space | |
1142 | * to real space and flush out to disk. | |
1143 | */ | |
1144 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1145 | if (error == -EAGAIN) | |
1146 | goto out_fail; | |
1147 | if (unlikely(error < 0)) | |
1148 | goto out_unlock; | |
1149 | ||
1150 | return 0; | |
1151 | ||
1152 | out_fail: | |
1153 | redirty_page_for_writepage(wbc, page); | |
1154 | unlock_page(page); | |
1155 | return 0; | |
1156 | out_unlock: | |
1157 | unlock_page(page); | |
1158 | return error; | |
1159 | } | |
1160 | ||
1161 | /* | |
1162 | * Called to move a page into cleanable state - and from there | |
1163 | * to be released. Possibly the page is already clean. We always | |
1164 | * have buffer heads in this call. | |
1165 | * | |
1166 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1167 | * | |
1168 | * Possible scenarios are: | |
1169 | * | |
1170 | * 1. We are being called to release a page which has been written | |
1171 | * to via regular I/O. buffer heads will be dirty and possibly | |
1172 | * delalloc. If no delalloc buffer heads in this case then we | |
1173 | * can just return zero. | |
1174 | * | |
1175 | * 2. We are called to release a page which has been written via | |
1176 | * mmap, all we need to do is ensure there is no delalloc | |
1177 | * state in the buffer heads, if not we can let the caller | |
1178 | * free them and we should come back later via writepage. | |
1179 | */ | |
1180 | STATIC int | |
e4c573bb | 1181 | xfs_vm_release_page( |
f51623b2 NS |
1182 | struct page *page, |
1183 | gfp_t gfp_mask) | |
1184 | { | |
1185 | struct inode *inode = page->mapping->host; | |
1186 | int dirty, delalloc, unmapped, unwritten; | |
1187 | struct writeback_control wbc = { | |
1188 | .sync_mode = WB_SYNC_ALL, | |
1189 | .nr_to_write = 1, | |
1190 | }; | |
1191 | ||
1192 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask); | |
1193 | ||
1194 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1195 | if (!delalloc && !unwritten) | |
1196 | goto free_buffers; | |
1197 | ||
1198 | if (!(gfp_mask & __GFP_FS)) | |
1199 | return 0; | |
1200 | ||
1201 | /* If we are already inside a transaction or the thread cannot | |
1202 | * do I/O, we cannot release this page. | |
1203 | */ | |
1204 | if (PFLAGS_TEST_FSTRANS()) | |
1205 | return 0; | |
1206 | ||
1207 | /* | |
1208 | * Convert delalloc space to real space, do not flush the | |
1209 | * data out to disk, that will be done by the caller. | |
1210 | * Never need to allocate space here - we will always | |
1211 | * come back to writepage in that case. | |
1212 | */ | |
1213 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1214 | if (dirty == 0 && !unwritten) | |
1215 | goto free_buffers; | |
1216 | return 0; | |
1217 | ||
1218 | free_buffers: | |
1219 | return try_to_free_buffers(page); | |
1220 | } | |
1221 | ||
1da177e4 | 1222 | STATIC int |
e4c573bb | 1223 | __xfs_get_block( |
1da177e4 LT |
1224 | struct inode *inode, |
1225 | sector_t iblock, | |
1226 | unsigned long blocks, | |
1227 | struct buffer_head *bh_result, | |
1228 | int create, | |
1229 | int direct, | |
1230 | bmapi_flags_t flags) | |
1231 | { | |
1232 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1233 | xfs_iomap_t iomap; | |
fdc7ed75 NS |
1234 | xfs_off_t offset; |
1235 | ssize_t size; | |
1da177e4 LT |
1236 | int retpbbm = 1; |
1237 | int error; | |
1da177e4 | 1238 | |
fdc7ed75 | 1239 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
a4656391 NS |
1240 | if (blocks) |
1241 | size = (ssize_t) min_t(xfs_off_t, LONG_MAX, | |
1242 | (xfs_off_t)blocks << inode->i_blkbits); | |
1243 | else | |
1244 | size = 1 << inode->i_blkbits; | |
1da177e4 LT |
1245 | |
1246 | VOP_BMAP(vp, offset, size, | |
1247 | create ? flags : BMAPI_READ, &iomap, &retpbbm, error); | |
1248 | if (error) | |
1249 | return -error; | |
1250 | ||
1251 | if (retpbbm == 0) | |
1252 | return 0; | |
1253 | ||
1254 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
87cbc49c NS |
1255 | /* |
1256 | * For unwritten extents do not report a disk address on | |
1da177e4 LT |
1257 | * the read case (treat as if we're reading into a hole). |
1258 | */ | |
1259 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
87cbc49c NS |
1260 | xfs_map_buffer(bh_result, &iomap, offset, |
1261 | inode->i_blkbits); | |
1da177e4 LT |
1262 | } |
1263 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1264 | if (direct) | |
1265 | bh_result->b_private = inode; | |
1266 | set_buffer_unwritten(bh_result); | |
1267 | set_buffer_delay(bh_result); | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | /* If this is a realtime file, data might be on a new device */ | |
ce8e922c | 1272 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 LT |
1273 | |
1274 | /* If we previously allocated a block out beyond eof and | |
1275 | * we are now coming back to use it then we will need to | |
1276 | * flag it as new even if it has a disk address. | |
1277 | */ | |
1278 | if (create && | |
1279 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
fdc7ed75 | 1280 | (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) |
1da177e4 | 1281 | set_buffer_new(bh_result); |
1da177e4 LT |
1282 | |
1283 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1284 | BUG_ON(direct); | |
1285 | if (create) { | |
1286 | set_buffer_uptodate(bh_result); | |
1287 | set_buffer_mapped(bh_result); | |
1288 | set_buffer_delay(bh_result); | |
1289 | } | |
1290 | } | |
1291 | ||
1292 | if (blocks) { | |
fdc7ed75 NS |
1293 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1294 | offset = min_t(xfs_off_t, | |
1295 | iomap.iomap_bsize - iomap.iomap_delta, | |
a4656391 | 1296 | (xfs_off_t)blocks << inode->i_blkbits); |
fdc7ed75 | 1297 | bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset); |
1da177e4 LT |
1298 | } |
1299 | ||
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | int | |
e4c573bb | 1304 | xfs_get_block( |
1da177e4 LT |
1305 | struct inode *inode, |
1306 | sector_t iblock, | |
1307 | struct buffer_head *bh_result, | |
1308 | int create) | |
1309 | { | |
e4c573bb | 1310 | return __xfs_get_block(inode, iblock, 0, bh_result, |
1da177e4 LT |
1311 | create, 0, BMAPI_WRITE); |
1312 | } | |
1313 | ||
1314 | STATIC int | |
e4c573bb | 1315 | xfs_get_blocks_direct( |
1da177e4 LT |
1316 | struct inode *inode, |
1317 | sector_t iblock, | |
1318 | unsigned long max_blocks, | |
1319 | struct buffer_head *bh_result, | |
1320 | int create) | |
1321 | { | |
e4c573bb | 1322 | return __xfs_get_block(inode, iblock, max_blocks, bh_result, |
1da177e4 LT |
1323 | create, 1, BMAPI_WRITE|BMAPI_DIRECT); |
1324 | } | |
1325 | ||
f0973863 | 1326 | STATIC void |
e4c573bb | 1327 | xfs_end_io_direct( |
f0973863 CH |
1328 | struct kiocb *iocb, |
1329 | loff_t offset, | |
1330 | ssize_t size, | |
1331 | void *private) | |
1332 | { | |
1333 | xfs_ioend_t *ioend = iocb->private; | |
1334 | ||
1335 | /* | |
1336 | * Non-NULL private data means we need to issue a transaction to | |
1337 | * convert a range from unwritten to written extents. This needs | |
1338 | * to happen from process contect but aio+dio I/O completion | |
1339 | * happens from irq context so we need to defer it to a workqueue. | |
1340 | * This is not nessecary for synchronous direct I/O, but we do | |
1341 | * it anyway to keep the code uniform and simpler. | |
1342 | * | |
1343 | * The core direct I/O code might be changed to always call the | |
1344 | * completion handler in the future, in which case all this can | |
1345 | * go away. | |
1346 | */ | |
1347 | if (private && size > 0) { | |
1348 | ioend->io_offset = offset; | |
1349 | ioend->io_size = size; | |
1350 | xfs_finish_ioend(ioend); | |
1351 | } else { | |
1352 | ASSERT(size >= 0); | |
1353 | xfs_destroy_ioend(ioend); | |
1354 | } | |
1355 | ||
1356 | /* | |
1357 | * blockdev_direct_IO can return an error even afer the I/O | |
1358 | * completion handler was called. Thus we need to protect | |
1359 | * against double-freeing. | |
1360 | */ | |
1361 | iocb->private = NULL; | |
1362 | } | |
1363 | ||
1da177e4 | 1364 | STATIC ssize_t |
e4c573bb | 1365 | xfs_vm_direct_IO( |
1da177e4 LT |
1366 | int rw, |
1367 | struct kiocb *iocb, | |
1368 | const struct iovec *iov, | |
1369 | loff_t offset, | |
1370 | unsigned long nr_segs) | |
1371 | { | |
1372 | struct file *file = iocb->ki_filp; | |
1373 | struct inode *inode = file->f_mapping->host; | |
1374 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1375 | xfs_iomap_t iomap; | |
1376 | int maps = 1; | |
1377 | int error; | |
f0973863 | 1378 | ssize_t ret; |
1da177e4 LT |
1379 | |
1380 | VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error); | |
1381 | if (error) | |
1382 | return -error; | |
1383 | ||
f6d6d4fc | 1384 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
f0973863 CH |
1385 | |
1386 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, | |
ce8e922c | 1387 | iomap.iomap_target->bt_bdev, |
1da177e4 | 1388 | iov, offset, nr_segs, |
e4c573bb NS |
1389 | xfs_get_blocks_direct, |
1390 | xfs_end_io_direct); | |
f0973863 CH |
1391 | |
1392 | if (unlikely(ret <= 0 && iocb->private)) | |
1393 | xfs_destroy_ioend(iocb->private); | |
1394 | return ret; | |
1da177e4 LT |
1395 | } |
1396 | ||
f51623b2 | 1397 | STATIC int |
e4c573bb | 1398 | xfs_vm_prepare_write( |
f51623b2 NS |
1399 | struct file *file, |
1400 | struct page *page, | |
1401 | unsigned int from, | |
1402 | unsigned int to) | |
1403 | { | |
e4c573bb | 1404 | return block_prepare_write(page, from, to, xfs_get_block); |
f51623b2 | 1405 | } |
1da177e4 LT |
1406 | |
1407 | STATIC sector_t | |
e4c573bb | 1408 | xfs_vm_bmap( |
1da177e4 LT |
1409 | struct address_space *mapping, |
1410 | sector_t block) | |
1411 | { | |
1412 | struct inode *inode = (struct inode *)mapping->host; | |
1413 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1414 | int error; | |
1415 | ||
e4c573bb | 1416 | vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address); |
1da177e4 LT |
1417 | |
1418 | VOP_RWLOCK(vp, VRWLOCK_READ); | |
1419 | VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error); | |
1420 | VOP_RWUNLOCK(vp, VRWLOCK_READ); | |
e4c573bb | 1421 | return generic_block_bmap(mapping, block, xfs_get_block); |
1da177e4 LT |
1422 | } |
1423 | ||
1424 | STATIC int | |
e4c573bb | 1425 | xfs_vm_readpage( |
1da177e4 LT |
1426 | struct file *unused, |
1427 | struct page *page) | |
1428 | { | |
e4c573bb | 1429 | return mpage_readpage(page, xfs_get_block); |
1da177e4 LT |
1430 | } |
1431 | ||
1432 | STATIC int | |
e4c573bb | 1433 | xfs_vm_readpages( |
1da177e4 LT |
1434 | struct file *unused, |
1435 | struct address_space *mapping, | |
1436 | struct list_head *pages, | |
1437 | unsigned nr_pages) | |
1438 | { | |
e4c573bb | 1439 | return mpage_readpages(mapping, pages, nr_pages, xfs_get_block); |
1da177e4 LT |
1440 | } |
1441 | ||
bcec2b7f | 1442 | STATIC int |
e4c573bb | 1443 | xfs_vm_invalidate_page( |
bcec2b7f NS |
1444 | struct page *page, |
1445 | unsigned long offset) | |
1446 | { | |
1447 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1448 | page->mapping->host, page, offset); | |
1449 | return block_invalidatepage(page, offset); | |
1450 | } | |
1451 | ||
e4c573bb NS |
1452 | struct address_space_operations xfs_address_space_operations = { |
1453 | .readpage = xfs_vm_readpage, | |
1454 | .readpages = xfs_vm_readpages, | |
1455 | .writepage = xfs_vm_writepage, | |
1da177e4 | 1456 | .sync_page = block_sync_page, |
e4c573bb NS |
1457 | .releasepage = xfs_vm_release_page, |
1458 | .invalidatepage = xfs_vm_invalidate_page, | |
1459 | .prepare_write = xfs_vm_prepare_write, | |
1da177e4 | 1460 | .commit_write = generic_commit_write, |
e4c573bb NS |
1461 | .bmap = xfs_vm_bmap, |
1462 | .direct_IO = xfs_vm_direct_IO, | |
e965f963 | 1463 | .migratepage = buffer_migrate_page, |
1da177e4 | 1464 | }; |