]>
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 | ||
1da177e4 LT |
489 | STATIC void |
490 | xfs_map_at_offset( | |
1da177e4 | 491 | struct buffer_head *bh, |
1defeac9 | 492 | loff_t offset, |
1da177e4 | 493 | int block_bits, |
1defeac9 | 494 | xfs_iomap_t *iomapp) |
1da177e4 LT |
495 | { |
496 | xfs_daddr_t bn; | |
1da177e4 LT |
497 | int sector_shift; |
498 | ||
499 | ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE)); | |
500 | ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY)); | |
501 | ASSERT(iomapp->iomap_bn != IOMAP_DADDR_NULL); | |
502 | ||
1da177e4 | 503 | sector_shift = block_bits - BBSHIFT; |
1defeac9 CH |
504 | bn = (iomapp->iomap_bn >> sector_shift) + |
505 | ((offset - iomapp->iomap_offset) >> block_bits); | |
506 | ||
507 | ASSERT(bn || (iomapp->iomap_flags & IOMAP_REALTIME)); | |
1da177e4 LT |
508 | ASSERT((bn << sector_shift) >= iomapp->iomap_bn); |
509 | ||
510 | lock_buffer(bh); | |
511 | bh->b_blocknr = bn; | |
ce8e922c | 512 | bh->b_bdev = iomapp->iomap_target->bt_bdev; |
1da177e4 LT |
513 | set_buffer_mapped(bh); |
514 | clear_buffer_delay(bh); | |
f6d6d4fc | 515 | clear_buffer_unwritten(bh); |
1da177e4 LT |
516 | } |
517 | ||
518 | /* | |
6c4fe19f | 519 | * Look for a page at index that is suitable for clustering. |
1da177e4 LT |
520 | */ |
521 | STATIC unsigned int | |
6c4fe19f | 522 | xfs_probe_page( |
10ce4444 | 523 | struct page *page, |
6c4fe19f CH |
524 | unsigned int pg_offset, |
525 | int mapped) | |
1da177e4 | 526 | { |
1da177e4 LT |
527 | int ret = 0; |
528 | ||
1da177e4 | 529 | if (PageWriteback(page)) |
10ce4444 | 530 | return 0; |
1da177e4 LT |
531 | |
532 | if (page->mapping && PageDirty(page)) { | |
533 | if (page_has_buffers(page)) { | |
534 | struct buffer_head *bh, *head; | |
535 | ||
536 | bh = head = page_buffers(page); | |
537 | do { | |
6c4fe19f CH |
538 | if (!buffer_uptodate(bh)) |
539 | break; | |
540 | if (mapped != buffer_mapped(bh)) | |
1da177e4 LT |
541 | break; |
542 | ret += bh->b_size; | |
543 | if (ret >= pg_offset) | |
544 | break; | |
545 | } while ((bh = bh->b_this_page) != head); | |
546 | } else | |
6c4fe19f | 547 | ret = mapped ? 0 : PAGE_CACHE_SIZE; |
1da177e4 LT |
548 | } |
549 | ||
1da177e4 LT |
550 | return ret; |
551 | } | |
552 | ||
f6d6d4fc | 553 | STATIC size_t |
6c4fe19f | 554 | xfs_probe_cluster( |
1da177e4 LT |
555 | struct inode *inode, |
556 | struct page *startpage, | |
557 | struct buffer_head *bh, | |
6c4fe19f CH |
558 | struct buffer_head *head, |
559 | int mapped) | |
1da177e4 | 560 | { |
10ce4444 | 561 | struct pagevec pvec; |
1da177e4 | 562 | pgoff_t tindex, tlast, tloff; |
10ce4444 CH |
563 | size_t total = 0; |
564 | int done = 0, i; | |
1da177e4 LT |
565 | |
566 | /* First sum forwards in this page */ | |
567 | do { | |
2353e8e9 | 568 | if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh))) |
10ce4444 | 569 | return total; |
1da177e4 LT |
570 | total += bh->b_size; |
571 | } while ((bh = bh->b_this_page) != head); | |
572 | ||
10ce4444 CH |
573 | /* if we reached the end of the page, sum forwards in following pages */ |
574 | tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT; | |
575 | tindex = startpage->index + 1; | |
576 | ||
577 | /* Prune this back to avoid pathological behavior */ | |
578 | tloff = min(tlast, startpage->index + 64); | |
579 | ||
580 | pagevec_init(&pvec, 0); | |
581 | while (!done && tindex <= tloff) { | |
582 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
583 | ||
584 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
585 | break; | |
586 | ||
587 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
588 | struct page *page = pvec.pages[i]; | |
589 | size_t pg_offset, len = 0; | |
590 | ||
591 | if (tindex == tlast) { | |
592 | pg_offset = | |
593 | i_size_read(inode) & (PAGE_CACHE_SIZE - 1); | |
1defeac9 CH |
594 | if (!pg_offset) { |
595 | done = 1; | |
10ce4444 | 596 | break; |
1defeac9 | 597 | } |
10ce4444 CH |
598 | } else |
599 | pg_offset = PAGE_CACHE_SIZE; | |
600 | ||
601 | if (page->index == tindex && !TestSetPageLocked(page)) { | |
6c4fe19f | 602 | len = xfs_probe_page(page, pg_offset, mapped); |
10ce4444 CH |
603 | unlock_page(page); |
604 | } | |
605 | ||
606 | if (!len) { | |
607 | done = 1; | |
608 | break; | |
609 | } | |
610 | ||
1da177e4 | 611 | total += len; |
1defeac9 | 612 | tindex++; |
1da177e4 | 613 | } |
10ce4444 CH |
614 | |
615 | pagevec_release(&pvec); | |
616 | cond_resched(); | |
1da177e4 | 617 | } |
10ce4444 | 618 | |
1da177e4 LT |
619 | return total; |
620 | } | |
621 | ||
622 | /* | |
10ce4444 CH |
623 | * Test if a given page is suitable for writing as part of an unwritten |
624 | * or delayed allocate extent. | |
1da177e4 | 625 | */ |
10ce4444 CH |
626 | STATIC int |
627 | xfs_is_delayed_page( | |
628 | struct page *page, | |
f6d6d4fc | 629 | unsigned int type) |
1da177e4 | 630 | { |
1da177e4 | 631 | if (PageWriteback(page)) |
10ce4444 | 632 | return 0; |
1da177e4 LT |
633 | |
634 | if (page->mapping && page_has_buffers(page)) { | |
635 | struct buffer_head *bh, *head; | |
636 | int acceptable = 0; | |
637 | ||
638 | bh = head = page_buffers(page); | |
639 | do { | |
f6d6d4fc CH |
640 | if (buffer_unwritten(bh)) |
641 | acceptable = (type == IOMAP_UNWRITTEN); | |
642 | else if (buffer_delay(bh)) | |
643 | acceptable = (type == IOMAP_DELAY); | |
6c4fe19f CH |
644 | else if (buffer_mapped(bh)) |
645 | acceptable = (type == 0); | |
f6d6d4fc | 646 | else |
1da177e4 | 647 | break; |
1da177e4 LT |
648 | } while ((bh = bh->b_this_page) != head); |
649 | ||
650 | if (acceptable) | |
10ce4444 | 651 | return 1; |
1da177e4 LT |
652 | } |
653 | ||
10ce4444 | 654 | return 0; |
1da177e4 LT |
655 | } |
656 | ||
1da177e4 LT |
657 | /* |
658 | * Allocate & map buffers for page given the extent map. Write it out. | |
659 | * except for the original page of a writepage, this is called on | |
660 | * delalloc/unwritten pages only, for the original page it is possible | |
661 | * that the page has no mapping at all. | |
662 | */ | |
f6d6d4fc | 663 | STATIC int |
1da177e4 LT |
664 | xfs_convert_page( |
665 | struct inode *inode, | |
666 | struct page *page, | |
10ce4444 | 667 | loff_t tindex, |
1defeac9 | 668 | xfs_iomap_t *mp, |
f6d6d4fc | 669 | xfs_ioend_t **ioendp, |
1da177e4 | 670 | struct writeback_control *wbc, |
1da177e4 LT |
671 | int startio, |
672 | int all_bh) | |
673 | { | |
f6d6d4fc | 674 | struct buffer_head *bh, *head; |
9260dc6b CH |
675 | xfs_off_t end_offset; |
676 | unsigned long p_offset; | |
f6d6d4fc | 677 | unsigned int type; |
1da177e4 | 678 | int bbits = inode->i_blkbits; |
24e17b5f | 679 | int len, page_dirty; |
f6d6d4fc | 680 | int count = 0, done = 0, uptodate = 1; |
9260dc6b | 681 | xfs_off_t offset = page_offset(page); |
1da177e4 | 682 | |
10ce4444 CH |
683 | if (page->index != tindex) |
684 | goto fail; | |
685 | if (TestSetPageLocked(page)) | |
686 | goto fail; | |
687 | if (PageWriteback(page)) | |
688 | goto fail_unlock_page; | |
689 | if (page->mapping != inode->i_mapping) | |
690 | goto fail_unlock_page; | |
691 | if (!xfs_is_delayed_page(page, (*ioendp)->io_type)) | |
692 | goto fail_unlock_page; | |
693 | ||
24e17b5f NS |
694 | /* |
695 | * page_dirty is initially a count of buffers on the page before | |
696 | * EOF and is decrememted as we move each into a cleanable state. | |
9260dc6b CH |
697 | * |
698 | * Derivation: | |
699 | * | |
700 | * End offset is the highest offset that this page should represent. | |
701 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
702 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
703 | * hence give us the correct page_dirty count. On any other page, | |
704 | * it will be zero and in that case we need page_dirty to be the | |
705 | * count of buffers on the page. | |
24e17b5f | 706 | */ |
9260dc6b CH |
707 | end_offset = min_t(unsigned long long, |
708 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, | |
709 | i_size_read(inode)); | |
710 | ||
24e17b5f | 711 | len = 1 << inode->i_blkbits; |
9260dc6b CH |
712 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
713 | PAGE_CACHE_SIZE); | |
714 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
715 | page_dirty = p_offset / len; | |
24e17b5f | 716 | |
1da177e4 LT |
717 | bh = head = page_buffers(page); |
718 | do { | |
9260dc6b | 719 | if (offset >= end_offset) |
1da177e4 | 720 | break; |
f6d6d4fc CH |
721 | if (!buffer_uptodate(bh)) |
722 | uptodate = 0; | |
723 | if (!(PageUptodate(page) || buffer_uptodate(bh))) { | |
724 | done = 1; | |
1da177e4 | 725 | continue; |
f6d6d4fc CH |
726 | } |
727 | ||
9260dc6b CH |
728 | if (buffer_unwritten(bh) || buffer_delay(bh)) { |
729 | if (buffer_unwritten(bh)) | |
730 | type = IOMAP_UNWRITTEN; | |
731 | else | |
732 | type = IOMAP_DELAY; | |
733 | ||
734 | if (!xfs_iomap_valid(mp, offset)) { | |
f6d6d4fc | 735 | done = 1; |
9260dc6b CH |
736 | continue; |
737 | } | |
738 | ||
739 | ASSERT(!(mp->iomap_flags & IOMAP_HOLE)); | |
740 | ASSERT(!(mp->iomap_flags & IOMAP_DELAY)); | |
741 | ||
742 | xfs_map_at_offset(bh, offset, bbits, mp); | |
743 | if (startio) { | |
7336cea8 | 744 | xfs_add_to_ioend(inode, bh, offset, |
9260dc6b CH |
745 | type, ioendp, done); |
746 | } else { | |
747 | set_buffer_dirty(bh); | |
748 | unlock_buffer(bh); | |
749 | mark_buffer_dirty(bh); | |
750 | } | |
751 | page_dirty--; | |
752 | count++; | |
753 | } else { | |
754 | type = 0; | |
755 | if (buffer_mapped(bh) && all_bh && startio) { | |
1da177e4 | 756 | lock_buffer(bh); |
7336cea8 | 757 | xfs_add_to_ioend(inode, bh, offset, |
f6d6d4fc CH |
758 | type, ioendp, done); |
759 | count++; | |
24e17b5f | 760 | page_dirty--; |
9260dc6b CH |
761 | } else { |
762 | done = 1; | |
1da177e4 | 763 | } |
1da177e4 | 764 | } |
7336cea8 | 765 | } while (offset += len, (bh = bh->b_this_page) != head); |
1da177e4 | 766 | |
f6d6d4fc CH |
767 | if (uptodate && bh == head) |
768 | SetPageUptodate(page); | |
769 | ||
770 | if (startio) { | |
f5e596bb CH |
771 | if (count) { |
772 | struct backing_dev_info *bdi; | |
773 | ||
774 | bdi = inode->i_mapping->backing_dev_info; | |
9fddaca2 | 775 | wbc->nr_to_write--; |
f5e596bb CH |
776 | if (bdi_write_congested(bdi)) { |
777 | wbc->encountered_congestion = 1; | |
778 | done = 1; | |
9fddaca2 | 779 | } else if (wbc->nr_to_write <= 0) { |
f5e596bb CH |
780 | done = 1; |
781 | } | |
782 | } | |
f6d6d4fc | 783 | xfs_start_page_writeback(page, wbc, !page_dirty, count); |
1da177e4 | 784 | } |
f6d6d4fc CH |
785 | |
786 | return done; | |
10ce4444 CH |
787 | fail_unlock_page: |
788 | unlock_page(page); | |
789 | fail: | |
790 | return 1; | |
1da177e4 LT |
791 | } |
792 | ||
793 | /* | |
794 | * Convert & write out a cluster of pages in the same extent as defined | |
795 | * by mp and following the start page. | |
796 | */ | |
797 | STATIC void | |
798 | xfs_cluster_write( | |
799 | struct inode *inode, | |
800 | pgoff_t tindex, | |
801 | xfs_iomap_t *iomapp, | |
f6d6d4fc | 802 | xfs_ioend_t **ioendp, |
1da177e4 LT |
803 | struct writeback_control *wbc, |
804 | int startio, | |
805 | int all_bh, | |
806 | pgoff_t tlast) | |
807 | { | |
10ce4444 CH |
808 | struct pagevec pvec; |
809 | int done = 0, i; | |
1da177e4 | 810 | |
10ce4444 CH |
811 | pagevec_init(&pvec, 0); |
812 | while (!done && tindex <= tlast) { | |
813 | unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1); | |
814 | ||
815 | if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len)) | |
1da177e4 | 816 | break; |
10ce4444 CH |
817 | |
818 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
819 | done = xfs_convert_page(inode, pvec.pages[i], tindex++, | |
820 | iomapp, ioendp, wbc, startio, all_bh); | |
821 | if (done) | |
822 | break; | |
823 | } | |
824 | ||
825 | pagevec_release(&pvec); | |
826 | cond_resched(); | |
1da177e4 LT |
827 | } |
828 | } | |
829 | ||
830 | /* | |
831 | * Calling this without startio set means we are being asked to make a dirty | |
832 | * page ready for freeing it's buffers. When called with startio set then | |
833 | * we are coming from writepage. | |
834 | * | |
835 | * When called with startio set it is important that we write the WHOLE | |
836 | * page if possible. | |
837 | * The bh->b_state's cannot know if any of the blocks or which block for | |
838 | * that matter are dirty due to mmap writes, and therefore bh uptodate is | |
839 | * only vaild if the page itself isn't completely uptodate. Some layers | |
840 | * may clear the page dirty flag prior to calling write page, under the | |
841 | * assumption the entire page will be written out; by not writing out the | |
842 | * whole page the page can be reused before all valid dirty data is | |
843 | * written out. Note: in the case of a page that has been dirty'd by | |
844 | * mapwrite and but partially setup by block_prepare_write the | |
845 | * bh->b_states's will not agree and only ones setup by BPW/BCW will have | |
846 | * valid state, thus the whole page must be written out thing. | |
847 | */ | |
848 | ||
849 | STATIC int | |
850 | xfs_page_state_convert( | |
851 | struct inode *inode, | |
852 | struct page *page, | |
853 | struct writeback_control *wbc, | |
854 | int startio, | |
855 | int unmapped) /* also implies page uptodate */ | |
856 | { | |
f6d6d4fc | 857 | struct buffer_head *bh, *head; |
1defeac9 | 858 | xfs_iomap_t iomap; |
f6d6d4fc | 859 | xfs_ioend_t *ioend = NULL, *iohead = NULL; |
1da177e4 LT |
860 | loff_t offset; |
861 | unsigned long p_offset = 0; | |
f6d6d4fc | 862 | unsigned int type; |
1da177e4 LT |
863 | __uint64_t end_offset; |
864 | pgoff_t end_index, last_index, tlast; | |
d5cb48aa CH |
865 | ssize_t size, len; |
866 | int flags, err, iomap_valid = 0, uptodate = 1; | |
f6d6d4fc | 867 | int page_dirty, count = 0, trylock_flag = 0; |
6c4fe19f | 868 | int all_bh = unmapped; |
1da177e4 | 869 | |
3ba0815a | 870 | /* wait for other IO threads? */ |
f5e596bb | 871 | if (startio && (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)) |
f6d6d4fc | 872 | trylock_flag |= BMAPI_TRYLOCK; |
3ba0815a | 873 | |
1da177e4 LT |
874 | /* Is this page beyond the end of the file? */ |
875 | offset = i_size_read(inode); | |
876 | end_index = offset >> PAGE_CACHE_SHIFT; | |
877 | last_index = (offset - 1) >> PAGE_CACHE_SHIFT; | |
878 | if (page->index >= end_index) { | |
879 | if ((page->index >= end_index + 1) || | |
880 | !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) { | |
19d5bcf3 NS |
881 | if (startio) |
882 | unlock_page(page); | |
883 | return 0; | |
1da177e4 LT |
884 | } |
885 | } | |
886 | ||
1da177e4 | 887 | /* |
24e17b5f NS |
888 | * page_dirty is initially a count of buffers on the page before |
889 | * EOF and is decrememted as we move each into a cleanable state. | |
f6d6d4fc CH |
890 | * |
891 | * Derivation: | |
892 | * | |
893 | * End offset is the highest offset that this page should represent. | |
894 | * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1)) | |
895 | * will evaluate non-zero and be less than PAGE_CACHE_SIZE and | |
896 | * hence give us the correct page_dirty count. On any other page, | |
897 | * it will be zero and in that case we need page_dirty to be the | |
898 | * count of buffers on the page. | |
899 | */ | |
900 | end_offset = min_t(unsigned long long, | |
901 | (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset); | |
24e17b5f | 902 | len = 1 << inode->i_blkbits; |
f6d6d4fc CH |
903 | p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1), |
904 | PAGE_CACHE_SIZE); | |
905 | p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE; | |
24e17b5f NS |
906 | page_dirty = p_offset / len; |
907 | ||
24e17b5f | 908 | bh = head = page_buffers(page); |
f6d6d4fc | 909 | offset = page_offset(page); |
6c4fe19f CH |
910 | flags = -1; |
911 | type = 0; | |
f6d6d4fc | 912 | |
f6d6d4fc | 913 | /* TODO: cleanup count and page_dirty */ |
1da177e4 LT |
914 | |
915 | do { | |
916 | if (offset >= end_offset) | |
917 | break; | |
918 | if (!buffer_uptodate(bh)) | |
919 | uptodate = 0; | |
f6d6d4fc | 920 | if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) { |
1defeac9 CH |
921 | /* |
922 | * the iomap is actually still valid, but the ioend | |
923 | * isn't. shouldn't happen too often. | |
924 | */ | |
925 | iomap_valid = 0; | |
1da177e4 | 926 | continue; |
f6d6d4fc | 927 | } |
1da177e4 | 928 | |
1defeac9 CH |
929 | if (iomap_valid) |
930 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1da177e4 LT |
931 | |
932 | /* | |
933 | * First case, map an unwritten extent and prepare for | |
934 | * extent state conversion transaction on completion. | |
f6d6d4fc | 935 | * |
1da177e4 LT |
936 | * Second case, allocate space for a delalloc buffer. |
937 | * We can return EAGAIN here in the release page case. | |
d5cb48aa CH |
938 | * |
939 | * Third case, an unmapped buffer was found, and we are | |
940 | * in a path where we need to write the whole page out. | |
941 | */ | |
942 | if (buffer_unwritten(bh) || buffer_delay(bh) || | |
943 | ((buffer_uptodate(bh) || PageUptodate(page)) && | |
944 | !buffer_mapped(bh) && (unmapped || startio))) { | |
6c4fe19f CH |
945 | /* |
946 | * Make sure we don't use a read-only iomap | |
947 | */ | |
948 | if (flags == BMAPI_READ) | |
949 | iomap_valid = 0; | |
950 | ||
f6d6d4fc CH |
951 | if (buffer_unwritten(bh)) { |
952 | type = IOMAP_UNWRITTEN; | |
953 | flags = BMAPI_WRITE|BMAPI_IGNSTATE; | |
d5cb48aa | 954 | } else if (buffer_delay(bh)) { |
f6d6d4fc CH |
955 | type = IOMAP_DELAY; |
956 | flags = BMAPI_ALLOCATE; | |
957 | if (!startio) | |
958 | flags |= trylock_flag; | |
d5cb48aa | 959 | } else { |
6c4fe19f | 960 | type = IOMAP_NEW; |
d5cb48aa | 961 | flags = BMAPI_WRITE|BMAPI_MMAP; |
f6d6d4fc CH |
962 | } |
963 | ||
1defeac9 | 964 | if (!iomap_valid) { |
6c4fe19f CH |
965 | if (type == IOMAP_NEW) { |
966 | size = xfs_probe_cluster(inode, | |
967 | page, bh, head, 0); | |
d5cb48aa CH |
968 | } else { |
969 | size = len; | |
970 | } | |
971 | ||
972 | err = xfs_map_blocks(inode, offset, size, | |
973 | &iomap, flags); | |
f6d6d4fc | 974 | if (err) |
1da177e4 | 975 | goto error; |
1defeac9 | 976 | iomap_valid = xfs_iomap_valid(&iomap, offset); |
1da177e4 | 977 | } |
1defeac9 CH |
978 | if (iomap_valid) { |
979 | xfs_map_at_offset(bh, offset, | |
980 | inode->i_blkbits, &iomap); | |
1da177e4 | 981 | if (startio) { |
7336cea8 | 982 | xfs_add_to_ioend(inode, bh, offset, |
1defeac9 CH |
983 | type, &ioend, |
984 | !iomap_valid); | |
1da177e4 LT |
985 | } else { |
986 | set_buffer_dirty(bh); | |
987 | unlock_buffer(bh); | |
988 | mark_buffer_dirty(bh); | |
989 | } | |
990 | page_dirty--; | |
f6d6d4fc | 991 | count++; |
1da177e4 | 992 | } |
d5cb48aa | 993 | } else if (buffer_uptodate(bh) && startio) { |
6c4fe19f CH |
994 | /* |
995 | * we got here because the buffer is already mapped. | |
996 | * That means it must already have extents allocated | |
997 | * underneath it. Map the extent by reading it. | |
998 | */ | |
999 | if (!iomap_valid || type != 0) { | |
1000 | flags = BMAPI_READ; | |
1001 | size = xfs_probe_cluster(inode, page, bh, | |
1002 | head, 1); | |
1003 | err = xfs_map_blocks(inode, offset, size, | |
1004 | &iomap, flags); | |
1005 | if (err) | |
1006 | goto error; | |
1007 | iomap_valid = xfs_iomap_valid(&iomap, offset); | |
1008 | } | |
d5cb48aa | 1009 | |
6c4fe19f | 1010 | type = 0; |
d5cb48aa CH |
1011 | if (!test_and_set_bit(BH_Lock, &bh->b_state)) { |
1012 | ASSERT(buffer_mapped(bh)); | |
6c4fe19f CH |
1013 | if (iomap_valid) |
1014 | all_bh = 1; | |
7336cea8 | 1015 | xfs_add_to_ioend(inode, bh, offset, type, |
d5cb48aa CH |
1016 | &ioend, !iomap_valid); |
1017 | page_dirty--; | |
1018 | count++; | |
f6d6d4fc | 1019 | } else { |
1defeac9 | 1020 | iomap_valid = 0; |
1da177e4 | 1021 | } |
d5cb48aa CH |
1022 | } else if ((buffer_uptodate(bh) || PageUptodate(page)) && |
1023 | (unmapped || startio)) { | |
1024 | iomap_valid = 0; | |
1da177e4 | 1025 | } |
f6d6d4fc CH |
1026 | |
1027 | if (!iohead) | |
1028 | iohead = ioend; | |
1029 | ||
1030 | } while (offset += len, ((bh = bh->b_this_page) != head)); | |
1da177e4 LT |
1031 | |
1032 | if (uptodate && bh == head) | |
1033 | SetPageUptodate(page); | |
1034 | ||
f6d6d4fc CH |
1035 | if (startio) |
1036 | xfs_start_page_writeback(page, wbc, 1, count); | |
1da177e4 | 1037 | |
1defeac9 CH |
1038 | if (ioend && iomap_valid) { |
1039 | offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >> | |
1da177e4 | 1040 | PAGE_CACHE_SHIFT; |
775bf6c9 | 1041 | tlast = min_t(pgoff_t, offset, last_index); |
1defeac9 | 1042 | xfs_cluster_write(inode, page->index + 1, &iomap, &ioend, |
6c4fe19f | 1043 | wbc, startio, all_bh, tlast); |
1da177e4 LT |
1044 | } |
1045 | ||
f6d6d4fc CH |
1046 | if (iohead) |
1047 | xfs_submit_ioend(iohead); | |
1048 | ||
1da177e4 LT |
1049 | return page_dirty; |
1050 | ||
1051 | error: | |
f6d6d4fc CH |
1052 | if (iohead) |
1053 | xfs_cancel_ioend(iohead); | |
1da177e4 LT |
1054 | |
1055 | /* | |
1056 | * If it's delalloc and we have nowhere to put it, | |
1057 | * throw it away, unless the lower layers told | |
1058 | * us to try again. | |
1059 | */ | |
1060 | if (err != -EAGAIN) { | |
f6d6d4fc | 1061 | if (!unmapped) |
1da177e4 | 1062 | block_invalidatepage(page, 0); |
1da177e4 LT |
1063 | ClearPageUptodate(page); |
1064 | } | |
1065 | return err; | |
1066 | } | |
1067 | ||
f51623b2 NS |
1068 | /* |
1069 | * writepage: Called from one of two places: | |
1070 | * | |
1071 | * 1. we are flushing a delalloc buffer head. | |
1072 | * | |
1073 | * 2. we are writing out a dirty page. Typically the page dirty | |
1074 | * state is cleared before we get here. In this case is it | |
1075 | * conceivable we have no buffer heads. | |
1076 | * | |
1077 | * For delalloc space on the page we need to allocate space and | |
1078 | * flush it. For unmapped buffer heads on the page we should | |
1079 | * allocate space if the page is uptodate. For any other dirty | |
1080 | * buffer heads on the page we should flush them. | |
1081 | * | |
1082 | * If we detect that a transaction would be required to flush | |
1083 | * the page, we have to check the process flags first, if we | |
1084 | * are already in a transaction or disk I/O during allocations | |
1085 | * is off, we need to fail the writepage and redirty the page. | |
1086 | */ | |
1087 | ||
1088 | STATIC int | |
1089 | linvfs_writepage( | |
1090 | struct page *page, | |
1091 | struct writeback_control *wbc) | |
1092 | { | |
1093 | int error; | |
1094 | int need_trans; | |
1095 | int delalloc, unmapped, unwritten; | |
1096 | struct inode *inode = page->mapping->host; | |
1097 | ||
1098 | xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0); | |
1099 | ||
1100 | /* | |
1101 | * We need a transaction if: | |
1102 | * 1. There are delalloc buffers on the page | |
1103 | * 2. The page is uptodate and we have unmapped buffers | |
1104 | * 3. The page is uptodate and we have no buffers | |
1105 | * 4. There are unwritten buffers on the page | |
1106 | */ | |
1107 | ||
1108 | if (!page_has_buffers(page)) { | |
1109 | unmapped = 1; | |
1110 | need_trans = 1; | |
1111 | } else { | |
1112 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1113 | if (!PageUptodate(page)) | |
1114 | unmapped = 0; | |
1115 | need_trans = delalloc + unmapped + unwritten; | |
1116 | } | |
1117 | ||
1118 | /* | |
1119 | * If we need a transaction and the process flags say | |
1120 | * we are already in a transaction, or no IO is allowed | |
1121 | * then mark the page dirty again and leave the page | |
1122 | * as is. | |
1123 | */ | |
1124 | if (PFLAGS_TEST_FSTRANS() && need_trans) | |
1125 | goto out_fail; | |
1126 | ||
1127 | /* | |
1128 | * Delay hooking up buffer heads until we have | |
1129 | * made our go/no-go decision. | |
1130 | */ | |
1131 | if (!page_has_buffers(page)) | |
1132 | create_empty_buffers(page, 1 << inode->i_blkbits, 0); | |
1133 | ||
1134 | /* | |
1135 | * Convert delayed allocate, unwritten or unmapped space | |
1136 | * to real space and flush out to disk. | |
1137 | */ | |
1138 | error = xfs_page_state_convert(inode, page, wbc, 1, unmapped); | |
1139 | if (error == -EAGAIN) | |
1140 | goto out_fail; | |
1141 | if (unlikely(error < 0)) | |
1142 | goto out_unlock; | |
1143 | ||
1144 | return 0; | |
1145 | ||
1146 | out_fail: | |
1147 | redirty_page_for_writepage(wbc, page); | |
1148 | unlock_page(page); | |
1149 | return 0; | |
1150 | out_unlock: | |
1151 | unlock_page(page); | |
1152 | return error; | |
1153 | } | |
1154 | ||
1155 | /* | |
1156 | * Called to move a page into cleanable state - and from there | |
1157 | * to be released. Possibly the page is already clean. We always | |
1158 | * have buffer heads in this call. | |
1159 | * | |
1160 | * Returns 0 if the page is ok to release, 1 otherwise. | |
1161 | * | |
1162 | * Possible scenarios are: | |
1163 | * | |
1164 | * 1. We are being called to release a page which has been written | |
1165 | * to via regular I/O. buffer heads will be dirty and possibly | |
1166 | * delalloc. If no delalloc buffer heads in this case then we | |
1167 | * can just return zero. | |
1168 | * | |
1169 | * 2. We are called to release a page which has been written via | |
1170 | * mmap, all we need to do is ensure there is no delalloc | |
1171 | * state in the buffer heads, if not we can let the caller | |
1172 | * free them and we should come back later via writepage. | |
1173 | */ | |
1174 | STATIC int | |
1175 | linvfs_release_page( | |
1176 | struct page *page, | |
1177 | gfp_t gfp_mask) | |
1178 | { | |
1179 | struct inode *inode = page->mapping->host; | |
1180 | int dirty, delalloc, unmapped, unwritten; | |
1181 | struct writeback_control wbc = { | |
1182 | .sync_mode = WB_SYNC_ALL, | |
1183 | .nr_to_write = 1, | |
1184 | }; | |
1185 | ||
1186 | xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, gfp_mask); | |
1187 | ||
1188 | xfs_count_page_state(page, &delalloc, &unmapped, &unwritten); | |
1189 | if (!delalloc && !unwritten) | |
1190 | goto free_buffers; | |
1191 | ||
1192 | if (!(gfp_mask & __GFP_FS)) | |
1193 | return 0; | |
1194 | ||
1195 | /* If we are already inside a transaction or the thread cannot | |
1196 | * do I/O, we cannot release this page. | |
1197 | */ | |
1198 | if (PFLAGS_TEST_FSTRANS()) | |
1199 | return 0; | |
1200 | ||
1201 | /* | |
1202 | * Convert delalloc space to real space, do not flush the | |
1203 | * data out to disk, that will be done by the caller. | |
1204 | * Never need to allocate space here - we will always | |
1205 | * come back to writepage in that case. | |
1206 | */ | |
1207 | dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0); | |
1208 | if (dirty == 0 && !unwritten) | |
1209 | goto free_buffers; | |
1210 | return 0; | |
1211 | ||
1212 | free_buffers: | |
1213 | return try_to_free_buffers(page); | |
1214 | } | |
1215 | ||
1da177e4 LT |
1216 | STATIC int |
1217 | __linvfs_get_block( | |
1218 | struct inode *inode, | |
1219 | sector_t iblock, | |
1220 | unsigned long blocks, | |
1221 | struct buffer_head *bh_result, | |
1222 | int create, | |
1223 | int direct, | |
1224 | bmapi_flags_t flags) | |
1225 | { | |
1226 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1227 | xfs_iomap_t iomap; | |
fdc7ed75 NS |
1228 | xfs_off_t offset; |
1229 | ssize_t size; | |
1da177e4 LT |
1230 | int retpbbm = 1; |
1231 | int error; | |
1da177e4 | 1232 | |
fdc7ed75 | 1233 | offset = (xfs_off_t)iblock << inode->i_blkbits; |
a4656391 NS |
1234 | if (blocks) |
1235 | size = (ssize_t) min_t(xfs_off_t, LONG_MAX, | |
1236 | (xfs_off_t)blocks << inode->i_blkbits); | |
1237 | else | |
1238 | size = 1 << inode->i_blkbits; | |
1da177e4 LT |
1239 | |
1240 | VOP_BMAP(vp, offset, size, | |
1241 | create ? flags : BMAPI_READ, &iomap, &retpbbm, error); | |
1242 | if (error) | |
1243 | return -error; | |
1244 | ||
1245 | if (retpbbm == 0) | |
1246 | return 0; | |
1247 | ||
1248 | if (iomap.iomap_bn != IOMAP_DADDR_NULL) { | |
fdc7ed75 NS |
1249 | xfs_daddr_t bn; |
1250 | xfs_off_t delta; | |
1da177e4 LT |
1251 | |
1252 | /* For unwritten extents do not report a disk address on | |
1253 | * the read case (treat as if we're reading into a hole). | |
1254 | */ | |
1255 | if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1256 | delta = offset - iomap.iomap_offset; | |
1257 | delta >>= inode->i_blkbits; | |
1258 | ||
1259 | bn = iomap.iomap_bn >> (inode->i_blkbits - BBSHIFT); | |
1260 | bn += delta; | |
1261 | BUG_ON(!bn && !(iomap.iomap_flags & IOMAP_REALTIME)); | |
1262 | bh_result->b_blocknr = bn; | |
1263 | set_buffer_mapped(bh_result); | |
1264 | } | |
1265 | if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) { | |
1266 | if (direct) | |
1267 | bh_result->b_private = inode; | |
1268 | set_buffer_unwritten(bh_result); | |
1269 | set_buffer_delay(bh_result); | |
1270 | } | |
1271 | } | |
1272 | ||
1273 | /* If this is a realtime file, data might be on a new device */ | |
ce8e922c | 1274 | bh_result->b_bdev = iomap.iomap_target->bt_bdev; |
1da177e4 LT |
1275 | |
1276 | /* If we previously allocated a block out beyond eof and | |
1277 | * we are now coming back to use it then we will need to | |
1278 | * flag it as new even if it has a disk address. | |
1279 | */ | |
1280 | if (create && | |
1281 | ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) || | |
fdc7ed75 | 1282 | (offset >= i_size_read(inode)) || (iomap.iomap_flags & IOMAP_NEW))) |
1da177e4 | 1283 | set_buffer_new(bh_result); |
1da177e4 LT |
1284 | |
1285 | if (iomap.iomap_flags & IOMAP_DELAY) { | |
1286 | BUG_ON(direct); | |
1287 | if (create) { | |
1288 | set_buffer_uptodate(bh_result); | |
1289 | set_buffer_mapped(bh_result); | |
1290 | set_buffer_delay(bh_result); | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | if (blocks) { | |
fdc7ed75 NS |
1295 | ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0); |
1296 | offset = min_t(xfs_off_t, | |
1297 | iomap.iomap_bsize - iomap.iomap_delta, | |
a4656391 | 1298 | (xfs_off_t)blocks << inode->i_blkbits); |
fdc7ed75 | 1299 | bh_result->b_size = (u32) min_t(xfs_off_t, UINT_MAX, offset); |
1da177e4 LT |
1300 | } |
1301 | ||
1302 | return 0; | |
1303 | } | |
1304 | ||
1305 | int | |
1306 | linvfs_get_block( | |
1307 | struct inode *inode, | |
1308 | sector_t iblock, | |
1309 | struct buffer_head *bh_result, | |
1310 | int create) | |
1311 | { | |
1312 | return __linvfs_get_block(inode, iblock, 0, bh_result, | |
1313 | create, 0, BMAPI_WRITE); | |
1314 | } | |
1315 | ||
1316 | STATIC int | |
1317 | linvfs_get_blocks_direct( | |
1318 | struct inode *inode, | |
1319 | sector_t iblock, | |
1320 | unsigned long max_blocks, | |
1321 | struct buffer_head *bh_result, | |
1322 | int create) | |
1323 | { | |
1324 | return __linvfs_get_block(inode, iblock, max_blocks, bh_result, | |
1325 | create, 1, BMAPI_WRITE|BMAPI_DIRECT); | |
1326 | } | |
1327 | ||
f0973863 CH |
1328 | STATIC void |
1329 | linvfs_end_io_direct( | |
1330 | struct kiocb *iocb, | |
1331 | loff_t offset, | |
1332 | ssize_t size, | |
1333 | void *private) | |
1334 | { | |
1335 | xfs_ioend_t *ioend = iocb->private; | |
1336 | ||
1337 | /* | |
1338 | * Non-NULL private data means we need to issue a transaction to | |
1339 | * convert a range from unwritten to written extents. This needs | |
1340 | * to happen from process contect but aio+dio I/O completion | |
1341 | * happens from irq context so we need to defer it to a workqueue. | |
1342 | * This is not nessecary for synchronous direct I/O, but we do | |
1343 | * it anyway to keep the code uniform and simpler. | |
1344 | * | |
1345 | * The core direct I/O code might be changed to always call the | |
1346 | * completion handler in the future, in which case all this can | |
1347 | * go away. | |
1348 | */ | |
1349 | if (private && size > 0) { | |
1350 | ioend->io_offset = offset; | |
1351 | ioend->io_size = size; | |
1352 | xfs_finish_ioend(ioend); | |
1353 | } else { | |
1354 | ASSERT(size >= 0); | |
1355 | xfs_destroy_ioend(ioend); | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * blockdev_direct_IO can return an error even afer the I/O | |
1360 | * completion handler was called. Thus we need to protect | |
1361 | * against double-freeing. | |
1362 | */ | |
1363 | iocb->private = NULL; | |
1364 | } | |
1365 | ||
1da177e4 LT |
1366 | STATIC ssize_t |
1367 | linvfs_direct_IO( | |
1368 | int rw, | |
1369 | struct kiocb *iocb, | |
1370 | const struct iovec *iov, | |
1371 | loff_t offset, | |
1372 | unsigned long nr_segs) | |
1373 | { | |
1374 | struct file *file = iocb->ki_filp; | |
1375 | struct inode *inode = file->f_mapping->host; | |
1376 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1377 | xfs_iomap_t iomap; | |
1378 | int maps = 1; | |
1379 | int error; | |
f0973863 | 1380 | ssize_t ret; |
1da177e4 LT |
1381 | |
1382 | VOP_BMAP(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps, error); | |
1383 | if (error) | |
1384 | return -error; | |
1385 | ||
f6d6d4fc | 1386 | iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN); |
f0973863 CH |
1387 | |
1388 | ret = blockdev_direct_IO_own_locking(rw, iocb, inode, | |
ce8e922c | 1389 | iomap.iomap_target->bt_bdev, |
1da177e4 LT |
1390 | iov, offset, nr_segs, |
1391 | linvfs_get_blocks_direct, | |
f0973863 CH |
1392 | linvfs_end_io_direct); |
1393 | ||
1394 | if (unlikely(ret <= 0 && iocb->private)) | |
1395 | xfs_destroy_ioend(iocb->private); | |
1396 | return ret; | |
1da177e4 LT |
1397 | } |
1398 | ||
f51623b2 NS |
1399 | STATIC int |
1400 | linvfs_prepare_write( | |
1401 | struct file *file, | |
1402 | struct page *page, | |
1403 | unsigned int from, | |
1404 | unsigned int to) | |
1405 | { | |
1406 | return block_prepare_write(page, from, to, linvfs_get_block); | |
1407 | } | |
1da177e4 LT |
1408 | |
1409 | STATIC sector_t | |
1410 | linvfs_bmap( | |
1411 | struct address_space *mapping, | |
1412 | sector_t block) | |
1413 | { | |
1414 | struct inode *inode = (struct inode *)mapping->host; | |
1415 | vnode_t *vp = LINVFS_GET_VP(inode); | |
1416 | int error; | |
1417 | ||
1418 | vn_trace_entry(vp, "linvfs_bmap", (inst_t *)__return_address); | |
1419 | ||
1420 | VOP_RWLOCK(vp, VRWLOCK_READ); | |
1421 | VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error); | |
1422 | VOP_RWUNLOCK(vp, VRWLOCK_READ); | |
1423 | return generic_block_bmap(mapping, block, linvfs_get_block); | |
1424 | } | |
1425 | ||
1426 | STATIC int | |
1427 | linvfs_readpage( | |
1428 | struct file *unused, | |
1429 | struct page *page) | |
1430 | { | |
1431 | return mpage_readpage(page, linvfs_get_block); | |
1432 | } | |
1433 | ||
1434 | STATIC int | |
1435 | linvfs_readpages( | |
1436 | struct file *unused, | |
1437 | struct address_space *mapping, | |
1438 | struct list_head *pages, | |
1439 | unsigned nr_pages) | |
1440 | { | |
1441 | return mpage_readpages(mapping, pages, nr_pages, linvfs_get_block); | |
1442 | } | |
1443 | ||
bcec2b7f NS |
1444 | STATIC int |
1445 | linvfs_invalidate_page( | |
1446 | struct page *page, | |
1447 | unsigned long offset) | |
1448 | { | |
1449 | xfs_page_trace(XFS_INVALIDPAGE_ENTER, | |
1450 | page->mapping->host, page, offset); | |
1451 | return block_invalidatepage(page, offset); | |
1452 | } | |
1453 | ||
1da177e4 LT |
1454 | struct address_space_operations linvfs_aops = { |
1455 | .readpage = linvfs_readpage, | |
1456 | .readpages = linvfs_readpages, | |
1457 | .writepage = linvfs_writepage, | |
1458 | .sync_page = block_sync_page, | |
1459 | .releasepage = linvfs_release_page, | |
bcec2b7f | 1460 | .invalidatepage = linvfs_invalidate_page, |
1da177e4 LT |
1461 | .prepare_write = linvfs_prepare_write, |
1462 | .commit_write = generic_commit_write, | |
1463 | .bmap = linvfs_bmap, | |
1464 | .direct_IO = linvfs_direct_IO, | |
e965f963 | 1465 | .migratepage = buffer_migrate_page, |
1da177e4 | 1466 | }; |