Commit | Line | Data |
---|---|---|
31143d5d DH |
1 | /* handling of writes to regular files and writing back to the server |
2 | * | |
3 | * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. | |
4 | * Written by David Howells (dhowells@redhat.com) | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | */ | |
4af3c9cc | 11 | #include <linux/backing-dev.h> |
31143d5d DH |
12 | #include <linux/slab.h> |
13 | #include <linux/fs.h> | |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/writeback.h> | |
16 | #include <linux/pagevec.h> | |
17 | #include "internal.h" | |
18 | ||
19 | static int afs_write_back_from_locked_page(struct afs_writeback *wb, | |
20 | struct page *page); | |
21 | ||
22 | /* | |
23 | * mark a page as having been made dirty and thus needing writeback | |
24 | */ | |
25 | int afs_set_page_dirty(struct page *page) | |
26 | { | |
27 | _enter(""); | |
28 | return __set_page_dirty_nobuffers(page); | |
29 | } | |
30 | ||
31 | /* | |
32 | * unlink a writeback record because its usage has reached zero | |
33 | * - must be called with the wb->vnode->writeback_lock held | |
34 | */ | |
35 | static void afs_unlink_writeback(struct afs_writeback *wb) | |
36 | { | |
37 | struct afs_writeback *front; | |
38 | struct afs_vnode *vnode = wb->vnode; | |
39 | ||
40 | list_del_init(&wb->link); | |
41 | if (!list_empty(&vnode->writebacks)) { | |
42 | /* if an fsync rises to the front of the queue then wake it | |
43 | * up */ | |
44 | front = list_entry(vnode->writebacks.next, | |
45 | struct afs_writeback, link); | |
46 | if (front->state == AFS_WBACK_SYNCING) { | |
47 | _debug("wake up sync"); | |
48 | front->state = AFS_WBACK_COMPLETE; | |
49 | wake_up(&front->waitq); | |
50 | } | |
51 | } | |
52 | } | |
53 | ||
54 | /* | |
55 | * free a writeback record | |
56 | */ | |
57 | static void afs_free_writeback(struct afs_writeback *wb) | |
58 | { | |
59 | _enter(""); | |
60 | key_put(wb->key); | |
61 | kfree(wb); | |
62 | } | |
63 | ||
64 | /* | |
65 | * dispose of a reference to a writeback record | |
66 | */ | |
67 | void afs_put_writeback(struct afs_writeback *wb) | |
68 | { | |
69 | struct afs_vnode *vnode = wb->vnode; | |
70 | ||
71 | _enter("{%d}", wb->usage); | |
72 | ||
73 | spin_lock(&vnode->writeback_lock); | |
74 | if (--wb->usage == 0) | |
75 | afs_unlink_writeback(wb); | |
76 | else | |
77 | wb = NULL; | |
78 | spin_unlock(&vnode->writeback_lock); | |
79 | if (wb) | |
80 | afs_free_writeback(wb); | |
81 | } | |
82 | ||
83 | /* | |
84 | * partly or wholly fill a page that's under preparation for writing | |
85 | */ | |
86 | static int afs_fill_page(struct afs_vnode *vnode, struct key *key, | |
15b4650e | 87 | loff_t pos, unsigned len, struct page *page) |
31143d5d | 88 | { |
15b4650e NP |
89 | loff_t i_size; |
90 | unsigned eof; | |
31143d5d DH |
91 | int ret; |
92 | ||
15b4650e | 93 | _enter(",,%llu,%u", (unsigned long long)pos, len); |
31143d5d | 94 | |
15b4650e | 95 | ASSERTCMP(len, <=, PAGE_CACHE_SIZE); |
31143d5d | 96 | |
15b4650e NP |
97 | i_size = i_size_read(&vnode->vfs_inode); |
98 | if (pos + len > i_size) | |
99 | eof = i_size; | |
100 | else | |
101 | eof = PAGE_CACHE_SIZE; | |
102 | ||
103 | ret = afs_vnode_fetch_data(vnode, key, 0, eof, page); | |
31143d5d DH |
104 | if (ret < 0) { |
105 | if (ret == -ENOENT) { | |
106 | _debug("got NOENT from server" | |
107 | " - marking file deleted and stale"); | |
108 | set_bit(AFS_VNODE_DELETED, &vnode->flags); | |
109 | ret = -ESTALE; | |
110 | } | |
111 | } | |
112 | ||
113 | _leave(" = %d", ret); | |
114 | return ret; | |
115 | } | |
116 | ||
31143d5d DH |
117 | /* |
118 | * prepare to perform part of a write to a page | |
31143d5d | 119 | */ |
15b4650e NP |
120 | int afs_write_begin(struct file *file, struct address_space *mapping, |
121 | loff_t pos, unsigned len, unsigned flags, | |
122 | struct page **pagep, void **fsdata) | |
31143d5d DH |
123 | { |
124 | struct afs_writeback *candidate, *wb; | |
125 | struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode); | |
15b4650e | 126 | struct page *page; |
31143d5d | 127 | struct key *key = file->private_data; |
15b4650e NP |
128 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); |
129 | unsigned to = from + len; | |
130 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
31143d5d DH |
131 | int ret; |
132 | ||
133 | _enter("{%x:%u},{%lx},%u,%u", | |
15b4650e | 134 | vnode->fid.vid, vnode->fid.vnode, index, from, to); |
31143d5d DH |
135 | |
136 | candidate = kzalloc(sizeof(*candidate), GFP_KERNEL); | |
137 | if (!candidate) | |
138 | return -ENOMEM; | |
139 | candidate->vnode = vnode; | |
15b4650e NP |
140 | candidate->first = candidate->last = index; |
141 | candidate->offset_first = from; | |
31143d5d DH |
142 | candidate->to_last = to; |
143 | candidate->usage = 1; | |
144 | candidate->state = AFS_WBACK_PENDING; | |
145 | init_waitqueue_head(&candidate->waitq); | |
146 | ||
54566b2c | 147 | page = grab_cache_page_write_begin(mapping, index, flags); |
15b4650e NP |
148 | if (!page) { |
149 | kfree(candidate); | |
150 | return -ENOMEM; | |
151 | } | |
152 | *pagep = page; | |
153 | /* page won't leak in error case: it eventually gets cleaned off LRU */ | |
154 | ||
31143d5d DH |
155 | if (!PageUptodate(page)) { |
156 | _debug("not up to date"); | |
15b4650e | 157 | ret = afs_fill_page(vnode, key, pos, len, page); |
31143d5d DH |
158 | if (ret < 0) { |
159 | kfree(candidate); | |
160 | _leave(" = %d [prep]", ret); | |
161 | return ret; | |
162 | } | |
15b4650e | 163 | SetPageUptodate(page); |
31143d5d DH |
164 | } |
165 | ||
166 | try_again: | |
31143d5d DH |
167 | spin_lock(&vnode->writeback_lock); |
168 | ||
169 | /* see if this page is already pending a writeback under a suitable key | |
170 | * - if so we can just join onto that one */ | |
171 | wb = (struct afs_writeback *) page_private(page); | |
172 | if (wb) { | |
173 | if (wb->key == key && wb->state == AFS_WBACK_PENDING) | |
174 | goto subsume_in_current_wb; | |
175 | goto flush_conflicting_wb; | |
176 | } | |
177 | ||
178 | if (index > 0) { | |
179 | /* see if we can find an already pending writeback that we can | |
180 | * append this page to */ | |
181 | list_for_each_entry(wb, &vnode->writebacks, link) { | |
182 | if (wb->last == index - 1 && wb->key == key && | |
183 | wb->state == AFS_WBACK_PENDING) | |
184 | goto append_to_previous_wb; | |
185 | } | |
186 | } | |
187 | ||
188 | list_add_tail(&candidate->link, &vnode->writebacks); | |
189 | candidate->key = key_get(key); | |
190 | spin_unlock(&vnode->writeback_lock); | |
191 | SetPagePrivate(page); | |
192 | set_page_private(page, (unsigned long) candidate); | |
193 | _leave(" = 0 [new]"); | |
194 | return 0; | |
195 | ||
196 | subsume_in_current_wb: | |
197 | _debug("subsume"); | |
198 | ASSERTRANGE(wb->first, <=, index, <=, wb->last); | |
15b4650e NP |
199 | if (index == wb->first && from < wb->offset_first) |
200 | wb->offset_first = from; | |
31143d5d DH |
201 | if (index == wb->last && to > wb->to_last) |
202 | wb->to_last = to; | |
203 | spin_unlock(&vnode->writeback_lock); | |
204 | kfree(candidate); | |
205 | _leave(" = 0 [sub]"); | |
206 | return 0; | |
207 | ||
208 | append_to_previous_wb: | |
209 | _debug("append into %lx-%lx", wb->first, wb->last); | |
210 | wb->usage++; | |
211 | wb->last++; | |
212 | wb->to_last = to; | |
213 | spin_unlock(&vnode->writeback_lock); | |
214 | SetPagePrivate(page); | |
215 | set_page_private(page, (unsigned long) wb); | |
216 | kfree(candidate); | |
217 | _leave(" = 0 [app]"); | |
218 | return 0; | |
219 | ||
220 | /* the page is currently bound to another context, so if it's dirty we | |
221 | * need to flush it before we can use the new context */ | |
222 | flush_conflicting_wb: | |
223 | _debug("flush conflict"); | |
224 | if (wb->state == AFS_WBACK_PENDING) | |
225 | wb->state = AFS_WBACK_CONFLICTING; | |
226 | spin_unlock(&vnode->writeback_lock); | |
227 | if (PageDirty(page)) { | |
228 | ret = afs_write_back_from_locked_page(wb, page); | |
229 | if (ret < 0) { | |
230 | afs_put_writeback(candidate); | |
231 | _leave(" = %d", ret); | |
232 | return ret; | |
233 | } | |
234 | } | |
235 | ||
236 | /* the page holds a ref on the writeback record */ | |
237 | afs_put_writeback(wb); | |
238 | set_page_private(page, 0); | |
239 | ClearPagePrivate(page); | |
240 | goto try_again; | |
241 | } | |
242 | ||
243 | /* | |
244 | * finalise part of a write to a page | |
245 | */ | |
15b4650e NP |
246 | int afs_write_end(struct file *file, struct address_space *mapping, |
247 | loff_t pos, unsigned len, unsigned copied, | |
248 | struct page *page, void *fsdata) | |
31143d5d DH |
249 | { |
250 | struct afs_vnode *vnode = AFS_FS_I(file->f_dentry->d_inode); | |
251 | loff_t i_size, maybe_i_size; | |
252 | ||
15b4650e NP |
253 | _enter("{%x:%u},{%lx}", |
254 | vnode->fid.vid, vnode->fid.vnode, page->index); | |
31143d5d | 255 | |
15b4650e | 256 | maybe_i_size = pos + copied; |
31143d5d DH |
257 | |
258 | i_size = i_size_read(&vnode->vfs_inode); | |
259 | if (maybe_i_size > i_size) { | |
260 | spin_lock(&vnode->writeback_lock); | |
261 | i_size = i_size_read(&vnode->vfs_inode); | |
262 | if (maybe_i_size > i_size) | |
263 | i_size_write(&vnode->vfs_inode, maybe_i_size); | |
264 | spin_unlock(&vnode->writeback_lock); | |
265 | } | |
266 | ||
267 | set_page_dirty(page); | |
31143d5d DH |
268 | if (PageDirty(page)) |
269 | _debug("dirtied"); | |
15b4650e NP |
270 | unlock_page(page); |
271 | page_cache_release(page); | |
31143d5d | 272 | |
15b4650e | 273 | return copied; |
31143d5d DH |
274 | } |
275 | ||
276 | /* | |
277 | * kill all the pages in the given range | |
278 | */ | |
279 | static void afs_kill_pages(struct afs_vnode *vnode, bool error, | |
280 | pgoff_t first, pgoff_t last) | |
281 | { | |
282 | struct pagevec pv; | |
283 | unsigned count, loop; | |
284 | ||
285 | _enter("{%x:%u},%lx-%lx", | |
286 | vnode->fid.vid, vnode->fid.vnode, first, last); | |
287 | ||
288 | pagevec_init(&pv, 0); | |
289 | ||
290 | do { | |
291 | _debug("kill %lx-%lx", first, last); | |
292 | ||
293 | count = last - first + 1; | |
294 | if (count > PAGEVEC_SIZE) | |
295 | count = PAGEVEC_SIZE; | |
296 | pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping, | |
297 | first, count, pv.pages); | |
298 | ASSERTCMP(pv.nr, ==, count); | |
299 | ||
300 | for (loop = 0; loop < count; loop++) { | |
301 | ClearPageUptodate(pv.pages[loop]); | |
302 | if (error) | |
303 | SetPageError(pv.pages[loop]); | |
304 | end_page_writeback(pv.pages[loop]); | |
305 | } | |
306 | ||
307 | __pagevec_release(&pv); | |
308 | } while (first < last); | |
309 | ||
310 | _leave(""); | |
311 | } | |
312 | ||
313 | /* | |
314 | * synchronously write back the locked page and any subsequent non-locked dirty | |
315 | * pages also covered by the same writeback record | |
316 | */ | |
317 | static int afs_write_back_from_locked_page(struct afs_writeback *wb, | |
318 | struct page *primary_page) | |
319 | { | |
320 | struct page *pages[8], *page; | |
321 | unsigned long count; | |
322 | unsigned n, offset, to; | |
323 | pgoff_t start, first, last; | |
324 | int loop, ret; | |
325 | ||
326 | _enter(",%lx", primary_page->index); | |
327 | ||
328 | count = 1; | |
329 | if (!clear_page_dirty_for_io(primary_page)) | |
330 | BUG(); | |
331 | if (test_set_page_writeback(primary_page)) | |
332 | BUG(); | |
333 | ||
334 | /* find all consecutive lockable dirty pages, stopping when we find a | |
335 | * page that is not immediately lockable, is not dirty or is missing, | |
336 | * or we reach the end of the range */ | |
337 | start = primary_page->index; | |
338 | if (start >= wb->last) | |
339 | goto no_more; | |
340 | start++; | |
341 | do { | |
342 | _debug("more %lx [%lx]", start, count); | |
343 | n = wb->last - start + 1; | |
344 | if (n > ARRAY_SIZE(pages)) | |
345 | n = ARRAY_SIZE(pages); | |
346 | n = find_get_pages_contig(wb->vnode->vfs_inode.i_mapping, | |
347 | start, n, pages); | |
348 | _debug("fgpc %u", n); | |
349 | if (n == 0) | |
350 | goto no_more; | |
351 | if (pages[0]->index != start) { | |
9d577b6a DH |
352 | do { |
353 | put_page(pages[--n]); | |
354 | } while (n > 0); | |
31143d5d DH |
355 | goto no_more; |
356 | } | |
357 | ||
358 | for (loop = 0; loop < n; loop++) { | |
359 | page = pages[loop]; | |
360 | if (page->index > wb->last) | |
361 | break; | |
529ae9aa | 362 | if (!trylock_page(page)) |
31143d5d DH |
363 | break; |
364 | if (!PageDirty(page) || | |
365 | page_private(page) != (unsigned long) wb) { | |
366 | unlock_page(page); | |
367 | break; | |
368 | } | |
369 | if (!clear_page_dirty_for_io(page)) | |
370 | BUG(); | |
371 | if (test_set_page_writeback(page)) | |
372 | BUG(); | |
373 | unlock_page(page); | |
374 | put_page(page); | |
375 | } | |
376 | count += loop; | |
377 | if (loop < n) { | |
378 | for (; loop < n; loop++) | |
379 | put_page(pages[loop]); | |
380 | goto no_more; | |
381 | } | |
382 | ||
383 | start += loop; | |
384 | } while (start <= wb->last && count < 65536); | |
385 | ||
386 | no_more: | |
387 | /* we now have a contiguous set of dirty pages, each with writeback set | |
388 | * and the dirty mark cleared; the first page is locked and must remain | |
389 | * so, all the rest are unlocked */ | |
390 | first = primary_page->index; | |
391 | last = first + count - 1; | |
392 | ||
393 | offset = (first == wb->first) ? wb->offset_first : 0; | |
394 | to = (last == wb->last) ? wb->to_last : PAGE_SIZE; | |
395 | ||
396 | _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to); | |
397 | ||
398 | ret = afs_vnode_store_data(wb, first, last, offset, to); | |
399 | if (ret < 0) { | |
400 | switch (ret) { | |
401 | case -EDQUOT: | |
402 | case -ENOSPC: | |
403 | set_bit(AS_ENOSPC, | |
404 | &wb->vnode->vfs_inode.i_mapping->flags); | |
405 | break; | |
406 | case -EROFS: | |
407 | case -EIO: | |
408 | case -EREMOTEIO: | |
409 | case -EFBIG: | |
410 | case -ENOENT: | |
411 | case -ENOMEDIUM: | |
412 | case -ENXIO: | |
413 | afs_kill_pages(wb->vnode, true, first, last); | |
414 | set_bit(AS_EIO, &wb->vnode->vfs_inode.i_mapping->flags); | |
415 | break; | |
416 | case -EACCES: | |
417 | case -EPERM: | |
418 | case -ENOKEY: | |
419 | case -EKEYEXPIRED: | |
420 | case -EKEYREJECTED: | |
421 | case -EKEYREVOKED: | |
422 | afs_kill_pages(wb->vnode, false, first, last); | |
423 | break; | |
424 | default: | |
425 | break; | |
426 | } | |
427 | } else { | |
428 | ret = count; | |
429 | } | |
430 | ||
431 | _leave(" = %d", ret); | |
432 | return ret; | |
433 | } | |
434 | ||
435 | /* | |
436 | * write a page back to the server | |
437 | * - the caller locked the page for us | |
438 | */ | |
439 | int afs_writepage(struct page *page, struct writeback_control *wbc) | |
440 | { | |
441 | struct backing_dev_info *bdi = page->mapping->backing_dev_info; | |
442 | struct afs_writeback *wb; | |
443 | int ret; | |
444 | ||
445 | _enter("{%lx},", page->index); | |
446 | ||
31143d5d DH |
447 | wb = (struct afs_writeback *) page_private(page); |
448 | ASSERT(wb != NULL); | |
449 | ||
450 | ret = afs_write_back_from_locked_page(wb, page); | |
451 | unlock_page(page); | |
452 | if (ret < 0) { | |
453 | _leave(" = %d", ret); | |
454 | return 0; | |
455 | } | |
456 | ||
457 | wbc->nr_to_write -= ret; | |
458 | if (wbc->nonblocking && bdi_write_congested(bdi)) | |
459 | wbc->encountered_congestion = 1; | |
460 | ||
461 | _leave(" = 0"); | |
462 | return 0; | |
463 | } | |
464 | ||
465 | /* | |
466 | * write a region of pages back to the server | |
467 | */ | |
c1206a2c AB |
468 | static int afs_writepages_region(struct address_space *mapping, |
469 | struct writeback_control *wbc, | |
470 | pgoff_t index, pgoff_t end, pgoff_t *_next) | |
31143d5d DH |
471 | { |
472 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
473 | struct afs_writeback *wb; | |
474 | struct page *page; | |
475 | int ret, n; | |
476 | ||
477 | _enter(",,%lx,%lx,", index, end); | |
478 | ||
479 | do { | |
480 | n = find_get_pages_tag(mapping, &index, PAGECACHE_TAG_DIRTY, | |
481 | 1, &page); | |
482 | if (!n) | |
483 | break; | |
484 | ||
485 | _debug("wback %lx", page->index); | |
486 | ||
487 | if (page->index > end) { | |
488 | *_next = index; | |
489 | page_cache_release(page); | |
490 | _leave(" = 0 [%lx]", *_next); | |
491 | return 0; | |
492 | } | |
493 | ||
494 | /* at this point we hold neither mapping->tree_lock nor lock on | |
495 | * the page itself: the page may be truncated or invalidated | |
496 | * (changing page->mapping to NULL), or even swizzled back from | |
497 | * swapper_space to tmpfs file mapping | |
498 | */ | |
499 | lock_page(page); | |
500 | ||
501 | if (page->mapping != mapping) { | |
502 | unlock_page(page); | |
503 | page_cache_release(page); | |
504 | continue; | |
505 | } | |
506 | ||
507 | if (wbc->sync_mode != WB_SYNC_NONE) | |
508 | wait_on_page_writeback(page); | |
509 | ||
510 | if (PageWriteback(page) || !PageDirty(page)) { | |
511 | unlock_page(page); | |
512 | continue; | |
513 | } | |
514 | ||
515 | wb = (struct afs_writeback *) page_private(page); | |
516 | ASSERT(wb != NULL); | |
517 | ||
518 | spin_lock(&wb->vnode->writeback_lock); | |
519 | wb->state = AFS_WBACK_WRITING; | |
520 | spin_unlock(&wb->vnode->writeback_lock); | |
521 | ||
522 | ret = afs_write_back_from_locked_page(wb, page); | |
523 | unlock_page(page); | |
524 | page_cache_release(page); | |
525 | if (ret < 0) { | |
526 | _leave(" = %d", ret); | |
527 | return ret; | |
528 | } | |
529 | ||
530 | wbc->nr_to_write -= ret; | |
531 | ||
532 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
533 | wbc->encountered_congestion = 1; | |
534 | break; | |
535 | } | |
536 | ||
537 | cond_resched(); | |
538 | } while (index < end && wbc->nr_to_write > 0); | |
539 | ||
540 | *_next = index; | |
541 | _leave(" = 0 [%lx]", *_next); | |
542 | return 0; | |
543 | } | |
544 | ||
545 | /* | |
546 | * write some of the pending data back to the server | |
547 | */ | |
548 | int afs_writepages(struct address_space *mapping, | |
549 | struct writeback_control *wbc) | |
550 | { | |
551 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
552 | pgoff_t start, end, next; | |
553 | int ret; | |
554 | ||
555 | _enter(""); | |
556 | ||
557 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
558 | wbc->encountered_congestion = 1; | |
559 | _leave(" = 0 [congest]"); | |
560 | return 0; | |
561 | } | |
562 | ||
563 | if (wbc->range_cyclic) { | |
564 | start = mapping->writeback_index; | |
565 | end = -1; | |
566 | ret = afs_writepages_region(mapping, wbc, start, end, &next); | |
567 | if (start > 0 && wbc->nr_to_write > 0 && ret == 0 && | |
568 | !(wbc->nonblocking && wbc->encountered_congestion)) | |
569 | ret = afs_writepages_region(mapping, wbc, 0, start, | |
570 | &next); | |
571 | mapping->writeback_index = next; | |
572 | } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) { | |
573 | end = (pgoff_t)(LLONG_MAX >> PAGE_CACHE_SHIFT); | |
574 | ret = afs_writepages_region(mapping, wbc, 0, end, &next); | |
575 | if (wbc->nr_to_write > 0) | |
576 | mapping->writeback_index = next; | |
577 | } else { | |
578 | start = wbc->range_start >> PAGE_CACHE_SHIFT; | |
579 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
580 | ret = afs_writepages_region(mapping, wbc, start, end, &next); | |
581 | } | |
582 | ||
583 | _leave(" = %d", ret); | |
584 | return ret; | |
585 | } | |
586 | ||
587 | /* | |
588 | * write an inode back | |
589 | */ | |
590 | int afs_write_inode(struct inode *inode, int sync) | |
591 | { | |
592 | struct afs_vnode *vnode = AFS_FS_I(inode); | |
593 | int ret; | |
594 | ||
595 | _enter("{%x:%u},", vnode->fid.vid, vnode->fid.vnode); | |
596 | ||
597 | ret = 0; | |
598 | if (sync) { | |
599 | ret = filemap_fdatawait(inode->i_mapping); | |
600 | if (ret < 0) | |
601 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
602 | } | |
603 | ||
604 | _leave(" = %d", ret); | |
605 | return ret; | |
606 | } | |
607 | ||
608 | /* | |
609 | * completion of write to server | |
610 | */ | |
611 | void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call) | |
612 | { | |
613 | struct afs_writeback *wb = call->wb; | |
614 | struct pagevec pv; | |
615 | unsigned count, loop; | |
616 | pgoff_t first = call->first, last = call->last; | |
617 | bool free_wb; | |
618 | ||
619 | _enter("{%x:%u},{%lx-%lx}", | |
620 | vnode->fid.vid, vnode->fid.vnode, first, last); | |
621 | ||
622 | ASSERT(wb != NULL); | |
623 | ||
624 | pagevec_init(&pv, 0); | |
625 | ||
626 | do { | |
5bbf5d39 | 627 | _debug("done %lx-%lx", first, last); |
31143d5d DH |
628 | |
629 | count = last - first + 1; | |
630 | if (count > PAGEVEC_SIZE) | |
631 | count = PAGEVEC_SIZE; | |
632 | pv.nr = find_get_pages_contig(call->mapping, first, count, | |
633 | pv.pages); | |
634 | ASSERTCMP(pv.nr, ==, count); | |
635 | ||
636 | spin_lock(&vnode->writeback_lock); | |
637 | for (loop = 0; loop < count; loop++) { | |
638 | struct page *page = pv.pages[loop]; | |
639 | end_page_writeback(page); | |
640 | if (page_private(page) == (unsigned long) wb) { | |
641 | set_page_private(page, 0); | |
642 | ClearPagePrivate(page); | |
643 | wb->usage--; | |
644 | } | |
645 | } | |
646 | free_wb = false; | |
647 | if (wb->usage == 0) { | |
648 | afs_unlink_writeback(wb); | |
649 | free_wb = true; | |
650 | } | |
651 | spin_unlock(&vnode->writeback_lock); | |
652 | first += count; | |
653 | if (free_wb) { | |
654 | afs_free_writeback(wb); | |
655 | wb = NULL; | |
656 | } | |
657 | ||
658 | __pagevec_release(&pv); | |
5bbf5d39 | 659 | } while (first <= last); |
31143d5d DH |
660 | |
661 | _leave(""); | |
662 | } | |
663 | ||
664 | /* | |
665 | * write to an AFS file | |
666 | */ | |
667 | ssize_t afs_file_write(struct kiocb *iocb, const struct iovec *iov, | |
668 | unsigned long nr_segs, loff_t pos) | |
669 | { | |
670 | struct dentry *dentry = iocb->ki_filp->f_path.dentry; | |
671 | struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode); | |
672 | ssize_t result; | |
673 | size_t count = iov_length(iov, nr_segs); | |
674 | int ret; | |
675 | ||
676 | _enter("{%x.%u},{%zu},%lu,", | |
677 | vnode->fid.vid, vnode->fid.vnode, count, nr_segs); | |
678 | ||
679 | if (IS_SWAPFILE(&vnode->vfs_inode)) { | |
680 | printk(KERN_INFO | |
681 | "AFS: Attempt to write to active swap file!\n"); | |
682 | return -EBUSY; | |
683 | } | |
684 | ||
685 | if (!count) | |
686 | return 0; | |
687 | ||
688 | result = generic_file_aio_write(iocb, iov, nr_segs, pos); | |
689 | if (IS_ERR_VALUE(result)) { | |
690 | _leave(" = %zd", result); | |
691 | return result; | |
692 | } | |
693 | ||
694 | /* return error values for O_SYNC and IS_SYNC() */ | |
695 | if (IS_SYNC(&vnode->vfs_inode) || iocb->ki_filp->f_flags & O_SYNC) { | |
696 | ret = afs_fsync(iocb->ki_filp, dentry, 1); | |
697 | if (ret < 0) | |
698 | result = ret; | |
699 | } | |
700 | ||
701 | _leave(" = %zd", result); | |
702 | return result; | |
703 | } | |
704 | ||
705 | /* | |
706 | * flush the vnode to the fileserver | |
707 | */ | |
708 | int afs_writeback_all(struct afs_vnode *vnode) | |
709 | { | |
710 | struct address_space *mapping = vnode->vfs_inode.i_mapping; | |
711 | struct writeback_control wbc = { | |
712 | .bdi = mapping->backing_dev_info, | |
713 | .sync_mode = WB_SYNC_ALL, | |
714 | .nr_to_write = LONG_MAX, | |
31143d5d DH |
715 | .range_cyclic = 1, |
716 | }; | |
717 | int ret; | |
718 | ||
719 | _enter(""); | |
720 | ||
721 | ret = mapping->a_ops->writepages(mapping, &wbc); | |
722 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
723 | ||
724 | _leave(" = %d", ret); | |
725 | return ret; | |
726 | } | |
727 | ||
728 | /* | |
729 | * flush any dirty pages for this process, and check for write errors. | |
730 | * - the return status from this call provides a reliable indication of | |
731 | * whether any write errors occurred for this process. | |
732 | */ | |
733 | int afs_fsync(struct file *file, struct dentry *dentry, int datasync) | |
734 | { | |
735 | struct afs_writeback *wb, *xwb; | |
736 | struct afs_vnode *vnode = AFS_FS_I(dentry->d_inode); | |
737 | int ret; | |
738 | ||
739 | _enter("{%x:%u},{n=%s},%d", | |
740 | vnode->fid.vid, vnode->fid.vnode, dentry->d_name.name, | |
741 | datasync); | |
742 | ||
743 | /* use a writeback record as a marker in the queue - when this reaches | |
744 | * the front of the queue, all the outstanding writes are either | |
745 | * completed or rejected */ | |
746 | wb = kzalloc(sizeof(*wb), GFP_KERNEL); | |
747 | if (!wb) | |
748 | return -ENOMEM; | |
749 | wb->vnode = vnode; | |
750 | wb->first = 0; | |
751 | wb->last = -1; | |
752 | wb->offset_first = 0; | |
753 | wb->to_last = PAGE_SIZE; | |
754 | wb->usage = 1; | |
755 | wb->state = AFS_WBACK_SYNCING; | |
756 | init_waitqueue_head(&wb->waitq); | |
757 | ||
758 | spin_lock(&vnode->writeback_lock); | |
759 | list_for_each_entry(xwb, &vnode->writebacks, link) { | |
760 | if (xwb->state == AFS_WBACK_PENDING) | |
761 | xwb->state = AFS_WBACK_CONFLICTING; | |
762 | } | |
763 | list_add_tail(&wb->link, &vnode->writebacks); | |
764 | spin_unlock(&vnode->writeback_lock); | |
765 | ||
766 | /* push all the outstanding writebacks to the server */ | |
767 | ret = afs_writeback_all(vnode); | |
768 | if (ret < 0) { | |
769 | afs_put_writeback(wb); | |
770 | _leave(" = %d [wb]", ret); | |
771 | return ret; | |
772 | } | |
773 | ||
774 | /* wait for the preceding writes to actually complete */ | |
775 | ret = wait_event_interruptible(wb->waitq, | |
776 | wb->state == AFS_WBACK_COMPLETE || | |
777 | vnode->writebacks.next == &wb->link); | |
778 | afs_put_writeback(wb); | |
779 | _leave(" = %d", ret); | |
780 | return ret; | |
781 | } | |
9b3f26c9 DH |
782 | |
783 | /* | |
784 | * notification that a previously read-only page is about to become writable | |
785 | * - if it returns an error, the caller will deliver a bus error signal | |
786 | */ | |
787 | int afs_page_mkwrite(struct vm_area_struct *vma, struct page *page) | |
788 | { | |
789 | struct afs_vnode *vnode = AFS_FS_I(vma->vm_file->f_mapping->host); | |
790 | ||
791 | _enter("{{%x:%u}},{%lx}", | |
792 | vnode->fid.vid, vnode->fid.vnode, page->index); | |
793 | ||
794 | /* wait for the page to be written to the cache before we allow it to | |
795 | * be modified */ | |
796 | #ifdef CONFIG_AFS_FSCACHE | |
797 | fscache_wait_on_page_write(vnode->cache, page); | |
798 | #endif | |
799 | ||
800 | _leave(" = 0"); | |
801 | return 0; | |
802 | } |