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