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 <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}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* handling of writes to regular files and writing back to the server
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#include <linux/backing-dev.h>
9#include <linux/slab.h>
10#include <linux/fs.h>
11#include <linux/pagemap.h>
12#include <linux/writeback.h>
13#include <linux/pagevec.h>
14#include <linux/netfs.h>
15#include <linux/fscache.h>
16#include "internal.h"
17
18/*
19 * mark a page as having been made dirty and thus needing writeback
20 */
21int afs_set_page_dirty(struct page *page)
22{
23 _enter("");
24 return __set_page_dirty_nobuffers(page);
25}
26
27/*
28 * prepare to perform part of a write to a page
29 */
30int afs_write_begin(struct file *file, struct address_space *mapping,
31 loff_t pos, unsigned len, unsigned flags,
32 struct page **_page, void **fsdata)
33{
34 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
35 struct page *page;
36 unsigned long priv;
37 unsigned f, from;
38 unsigned t, to;
39 pgoff_t index;
40 int ret;
41
42 _enter("{%llx:%llu},%llx,%x",
43 vnode->fid.vid, vnode->fid.vnode, pos, len);
44
45 /* Prefetch area to be written into the cache if we're caching this
46 * file. We need to do this before we get a lock on the page in case
47 * there's more than one writer competing for the same cache block.
48 */
49 ret = netfs_write_begin(file, mapping, pos, len, flags, &page, fsdata,
50 &afs_req_ops, NULL);
51 if (ret < 0)
52 return ret;
53
54 index = page->index;
55 from = pos - index * PAGE_SIZE;
56 to = from + len;
57
58try_again:
59 /* See if this page is already partially written in a way that we can
60 * merge the new write with.
61 */
62 if (PagePrivate(page)) {
63 priv = page_private(page);
64 f = afs_page_dirty_from(page, priv);
65 t = afs_page_dirty_to(page, priv);
66 ASSERTCMP(f, <=, t);
67
68 if (PageWriteback(page)) {
69 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"), page);
70 goto flush_conflicting_write;
71 }
72 /* If the file is being filled locally, allow inter-write
73 * spaces to be merged into writes. If it's not, only write
74 * back what the user gives us.
75 */
76 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
77 (to < f || from > t))
78 goto flush_conflicting_write;
79 }
80
81 *_page = page;
82 _leave(" = 0");
83 return 0;
84
85 /* The previous write and this write aren't adjacent or overlapping, so
86 * flush the page out.
87 */
88flush_conflicting_write:
89 _debug("flush conflict");
90 ret = write_one_page(page);
91 if (ret < 0)
92 goto error;
93
94 ret = lock_page_killable(page);
95 if (ret < 0)
96 goto error;
97 goto try_again;
98
99error:
100 put_page(page);
101 _leave(" = %d", ret);
102 return ret;
103}
104
105/*
106 * finalise part of a write to a page
107 */
108int afs_write_end(struct file *file, struct address_space *mapping,
109 loff_t pos, unsigned len, unsigned copied,
110 struct page *page, void *fsdata)
111{
112 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
113 unsigned long priv;
114 unsigned int f, from = pos & (thp_size(page) - 1);
115 unsigned int t, to = from + copied;
116 loff_t i_size, maybe_i_size;
117
118 _enter("{%llx:%llu},{%lx}",
119 vnode->fid.vid, vnode->fid.vnode, page->index);
120
121 if (!PageUptodate(page)) {
122 if (copied < len) {
123 copied = 0;
124 goto out;
125 }
126
127 SetPageUptodate(page);
128 }
129
130 if (copied == 0)
131 goto out;
132
133 maybe_i_size = pos + copied;
134
135 i_size = i_size_read(&vnode->vfs_inode);
136 if (maybe_i_size > i_size) {
137 write_seqlock(&vnode->cb_lock);
138 i_size = i_size_read(&vnode->vfs_inode);
139 if (maybe_i_size > i_size)
140 afs_set_i_size(vnode, maybe_i_size);
141 write_sequnlock(&vnode->cb_lock);
142 }
143
144 if (PagePrivate(page)) {
145 priv = page_private(page);
146 f = afs_page_dirty_from(page, priv);
147 t = afs_page_dirty_to(page, priv);
148 if (from < f)
149 f = from;
150 if (to > t)
151 t = to;
152 priv = afs_page_dirty(page, f, t);
153 set_page_private(page, priv);
154 trace_afs_page_dirty(vnode, tracepoint_string("dirty+"), page);
155 } else {
156 priv = afs_page_dirty(page, from, to);
157 attach_page_private(page, (void *)priv);
158 trace_afs_page_dirty(vnode, tracepoint_string("dirty"), page);
159 }
160
161 if (set_page_dirty(page))
162 _debug("dirtied %lx", page->index);
163
164out:
165 unlock_page(page);
166 put_page(page);
167 return copied;
168}
169
170/*
171 * kill all the pages in the given range
172 */
173static void afs_kill_pages(struct address_space *mapping,
174 loff_t start, loff_t len)
175{
176 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
177 struct pagevec pv;
178 unsigned int loop, psize;
179
180 _enter("{%llx:%llu},%llx @%llx",
181 vnode->fid.vid, vnode->fid.vnode, len, start);
182
183 pagevec_init(&pv);
184
185 do {
186 _debug("kill %llx @%llx", len, start);
187
188 pv.nr = find_get_pages_contig(mapping, start / PAGE_SIZE,
189 PAGEVEC_SIZE, pv.pages);
190 if (pv.nr == 0)
191 break;
192
193 for (loop = 0; loop < pv.nr; loop++) {
194 struct page *page = pv.pages[loop];
195
196 if (page->index * PAGE_SIZE >= start + len)
197 break;
198
199 psize = thp_size(page);
200 start += psize;
201 len -= psize;
202 ClearPageUptodate(page);
203 end_page_writeback(page);
204 lock_page(page);
205 generic_error_remove_page(mapping, page);
206 unlock_page(page);
207 }
208
209 __pagevec_release(&pv);
210 } while (len > 0);
211
212 _leave("");
213}
214
215/*
216 * Redirty all the pages in a given range.
217 */
218static void afs_redirty_pages(struct writeback_control *wbc,
219 struct address_space *mapping,
220 loff_t start, loff_t len)
221{
222 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
223 struct pagevec pv;
224 unsigned int loop, psize;
225
226 _enter("{%llx:%llu},%llx @%llx",
227 vnode->fid.vid, vnode->fid.vnode, len, start);
228
229 pagevec_init(&pv);
230
231 do {
232 _debug("redirty %llx @%llx", len, start);
233
234 pv.nr = find_get_pages_contig(mapping, start / PAGE_SIZE,
235 PAGEVEC_SIZE, pv.pages);
236 if (pv.nr == 0)
237 break;
238
239 for (loop = 0; loop < pv.nr; loop++) {
240 struct page *page = pv.pages[loop];
241
242 if (page->index * PAGE_SIZE >= start + len)
243 break;
244
245 psize = thp_size(page);
246 start += psize;
247 len -= psize;
248 redirty_page_for_writepage(wbc, page);
249 end_page_writeback(page);
250 }
251
252 __pagevec_release(&pv);
253 } while (len > 0);
254
255 _leave("");
256}
257
258/*
259 * completion of write to server
260 */
261static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
262{
263 struct address_space *mapping = vnode->vfs_inode.i_mapping;
264 struct page *page;
265 pgoff_t end;
266
267 XA_STATE(xas, &mapping->i_pages, start / PAGE_SIZE);
268
269 _enter("{%llx:%llu},{%x @%llx}",
270 vnode->fid.vid, vnode->fid.vnode, len, start);
271
272 rcu_read_lock();
273
274 end = (start + len - 1) / PAGE_SIZE;
275 xas_for_each(&xas, page, end) {
276 if (!PageWriteback(page)) {
277 kdebug("bad %x @%llx page %lx %lx", len, start, page->index, end);
278 ASSERT(PageWriteback(page));
279 }
280
281 trace_afs_page_dirty(vnode, tracepoint_string("clear"), page);
282 detach_page_private(page);
283 page_endio(page, true, 0);
284 }
285
286 rcu_read_unlock();
287
288 afs_prune_wb_keys(vnode);
289 _leave("");
290}
291
292/*
293 * Find a key to use for the writeback. We cached the keys used to author the
294 * writes on the vnode. *_wbk will contain the last writeback key used or NULL
295 * and we need to start from there if it's set.
296 */
297static int afs_get_writeback_key(struct afs_vnode *vnode,
298 struct afs_wb_key **_wbk)
299{
300 struct afs_wb_key *wbk = NULL;
301 struct list_head *p;
302 int ret = -ENOKEY, ret2;
303
304 spin_lock(&vnode->wb_lock);
305 if (*_wbk)
306 p = (*_wbk)->vnode_link.next;
307 else
308 p = vnode->wb_keys.next;
309
310 while (p != &vnode->wb_keys) {
311 wbk = list_entry(p, struct afs_wb_key, vnode_link);
312 _debug("wbk %u", key_serial(wbk->key));
313 ret2 = key_validate(wbk->key);
314 if (ret2 == 0) {
315 refcount_inc(&wbk->usage);
316 _debug("USE WB KEY %u", key_serial(wbk->key));
317 break;
318 }
319
320 wbk = NULL;
321 if (ret == -ENOKEY)
322 ret = ret2;
323 p = p->next;
324 }
325
326 spin_unlock(&vnode->wb_lock);
327 if (*_wbk)
328 afs_put_wb_key(*_wbk);
329 *_wbk = wbk;
330 return 0;
331}
332
333static void afs_store_data_success(struct afs_operation *op)
334{
335 struct afs_vnode *vnode = op->file[0].vnode;
336
337 op->ctime = op->file[0].scb.status.mtime_client;
338 afs_vnode_commit_status(op, &op->file[0]);
339 if (op->error == 0) {
340 if (!op->store.laundering)
341 afs_pages_written_back(vnode, op->store.pos, op->store.size);
342 afs_stat_v(vnode, n_stores);
343 atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
344 }
345}
346
347static const struct afs_operation_ops afs_store_data_operation = {
348 .issue_afs_rpc = afs_fs_store_data,
349 .issue_yfs_rpc = yfs_fs_store_data,
350 .success = afs_store_data_success,
351};
352
353/*
354 * write to a file
355 */
356static int afs_store_data(struct afs_vnode *vnode, struct iov_iter *iter, loff_t pos,
357 bool laundering)
358{
359 struct afs_operation *op;
360 struct afs_wb_key *wbk = NULL;
361 loff_t size = iov_iter_count(iter), i_size;
362 int ret = -ENOKEY;
363
364 _enter("%s{%llx:%llu.%u},%llx,%llx",
365 vnode->volume->name,
366 vnode->fid.vid,
367 vnode->fid.vnode,
368 vnode->fid.unique,
369 size, pos);
370
371 ret = afs_get_writeback_key(vnode, &wbk);
372 if (ret) {
373 _leave(" = %d [no keys]", ret);
374 return ret;
375 }
376
377 op = afs_alloc_operation(wbk->key, vnode->volume);
378 if (IS_ERR(op)) {
379 afs_put_wb_key(wbk);
380 return -ENOMEM;
381 }
382
383 i_size = i_size_read(&vnode->vfs_inode);
384
385 afs_op_set_vnode(op, 0, vnode);
386 op->file[0].dv_delta = 1;
387 op->file[0].modification = true;
388 op->store.write_iter = iter;
389 op->store.pos = pos;
390 op->store.size = size;
391 op->store.i_size = max(pos + size, i_size);
392 op->store.laundering = laundering;
393 op->mtime = vnode->vfs_inode.i_mtime;
394 op->flags |= AFS_OPERATION_UNINTR;
395 op->ops = &afs_store_data_operation;
396
397try_next_key:
398 afs_begin_vnode_operation(op);
399 afs_wait_for_operation(op);
400
401 switch (op->error) {
402 case -EACCES:
403 case -EPERM:
404 case -ENOKEY:
405 case -EKEYEXPIRED:
406 case -EKEYREJECTED:
407 case -EKEYREVOKED:
408 _debug("next");
409
410 ret = afs_get_writeback_key(vnode, &wbk);
411 if (ret == 0) {
412 key_put(op->key);
413 op->key = key_get(wbk->key);
414 goto try_next_key;
415 }
416 break;
417 }
418
419 afs_put_wb_key(wbk);
420 _leave(" = %d", op->error);
421 return afs_put_operation(op);
422}
423
424/*
425 * Extend the region to be written back to include subsequent contiguously
426 * dirty pages if possible, but don't sleep while doing so.
427 *
428 * If this page holds new content, then we can include filler zeros in the
429 * writeback.
430 */
431static void afs_extend_writeback(struct address_space *mapping,
432 struct afs_vnode *vnode,
433 long *_count,
434 loff_t start,
435 loff_t max_len,
436 bool new_content,
437 unsigned int *_len)
438{
439 struct pagevec pvec;
440 struct page *page;
441 unsigned long priv;
442 unsigned int psize, filler = 0;
443 unsigned int f, t;
444 loff_t len = *_len;
445 pgoff_t index = (start + len) / PAGE_SIZE;
446 bool stop = true;
447 unsigned int i;
448
449 XA_STATE(xas, &mapping->i_pages, index);
450 pagevec_init(&pvec);
451
452 do {
453 /* Firstly, we gather up a batch of contiguous dirty pages
454 * under the RCU read lock - but we can't clear the dirty flags
455 * there if any of those pages are mapped.
456 */
457 rcu_read_lock();
458
459 xas_for_each(&xas, page, ULONG_MAX) {
460 stop = true;
461 if (xas_retry(&xas, page))
462 continue;
463 if (xa_is_value(page))
464 break;
465 if (page->index != index)
466 break;
467
468 if (!page_cache_get_speculative(page)) {
469 xas_reset(&xas);
470 continue;
471 }
472
473 /* Has the page moved or been split? */
474 if (unlikely(page != xas_reload(&xas))) {
475 put_page(page);
476 break;
477 }
478
479 if (!trylock_page(page)) {
480 put_page(page);
481 break;
482 }
483 if (!PageDirty(page) || PageWriteback(page)) {
484 unlock_page(page);
485 put_page(page);
486 break;
487 }
488
489 psize = thp_size(page);
490 priv = page_private(page);
491 f = afs_page_dirty_from(page, priv);
492 t = afs_page_dirty_to(page, priv);
493 if (f != 0 && !new_content) {
494 unlock_page(page);
495 put_page(page);
496 break;
497 }
498
499 len += filler + t;
500 filler = psize - t;
501 if (len >= max_len || *_count <= 0)
502 stop = true;
503 else if (t == psize || new_content)
504 stop = false;
505
506 index += thp_nr_pages(page);
507 if (!pagevec_add(&pvec, page))
508 break;
509 if (stop)
510 break;
511 }
512
513 if (!stop)
514 xas_pause(&xas);
515 rcu_read_unlock();
516
517 /* Now, if we obtained any pages, we can shift them to being
518 * writable and mark them for caching.
519 */
520 if (!pagevec_count(&pvec))
521 break;
522
523 for (i = 0; i < pagevec_count(&pvec); i++) {
524 page = pvec.pages[i];
525 trace_afs_page_dirty(vnode, tracepoint_string("store+"), page);
526
527 if (!clear_page_dirty_for_io(page))
528 BUG();
529 if (test_set_page_writeback(page))
530 BUG();
531
532 *_count -= thp_nr_pages(page);
533 unlock_page(page);
534 }
535
536 pagevec_release(&pvec);
537 cond_resched();
538 } while (!stop);
539
540 *_len = len;
541}
542
543/*
544 * Synchronously write back the locked page and any subsequent non-locked dirty
545 * pages.
546 */
547static ssize_t afs_write_back_from_locked_page(struct address_space *mapping,
548 struct writeback_control *wbc,
549 struct page *page,
550 loff_t start, loff_t end)
551{
552 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
553 struct iov_iter iter;
554 unsigned long priv;
555 unsigned int offset, to, len, max_len;
556 loff_t i_size = i_size_read(&vnode->vfs_inode);
557 bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
558 long count = wbc->nr_to_write;
559 int ret;
560
561 _enter(",%lx,%llx-%llx", page->index, start, end);
562
563 if (test_set_page_writeback(page))
564 BUG();
565
566 count -= thp_nr_pages(page);
567
568 /* Find all consecutive lockable dirty pages that have contiguous
569 * written regions, stopping when we find a page that is not
570 * immediately lockable, is not dirty or is missing, or we reach the
571 * end of the range.
572 */
573 priv = page_private(page);
574 offset = afs_page_dirty_from(page, priv);
575 to = afs_page_dirty_to(page, priv);
576 trace_afs_page_dirty(vnode, tracepoint_string("store"), page);
577
578 len = to - offset;
579 start += offset;
580 if (start < i_size) {
581 /* Trim the write to the EOF; the extra data is ignored. Also
582 * put an upper limit on the size of a single storedata op.
583 */
584 max_len = 65536 * 4096;
585 max_len = min_t(unsigned long long, max_len, end - start + 1);
586 max_len = min_t(unsigned long long, max_len, i_size - start);
587
588 if (len < max_len &&
589 (to == thp_size(page) || new_content))
590 afs_extend_writeback(mapping, vnode, &count,
591 start, max_len, new_content, &len);
592 len = min_t(loff_t, len, max_len);
593 }
594
595 /* We now have a contiguous set of dirty pages, each with writeback
596 * set; the first page is still locked at this point, but all the rest
597 * have been unlocked.
598 */
599 unlock_page(page);
600
601 if (start < i_size) {
602 _debug("write back %x @%llx [%llx]", len, start, i_size);
603
604 iov_iter_xarray(&iter, WRITE, &mapping->i_pages, start, len);
605 ret = afs_store_data(vnode, &iter, start, false);
606 } else {
607 _debug("write discard %x @%llx [%llx]", len, start, i_size);
608
609 /* The dirty region was entirely beyond the EOF. */
610 afs_pages_written_back(vnode, start, len);
611 ret = 0;
612 }
613
614 switch (ret) {
615 case 0:
616 wbc->nr_to_write = count;
617 ret = len;
618 break;
619
620 default:
621 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
622 fallthrough;
623 case -EACCES:
624 case -EPERM:
625 case -ENOKEY:
626 case -EKEYEXPIRED:
627 case -EKEYREJECTED:
628 case -EKEYREVOKED:
629 afs_redirty_pages(wbc, mapping, start, len);
630 mapping_set_error(mapping, ret);
631 break;
632
633 case -EDQUOT:
634 case -ENOSPC:
635 afs_redirty_pages(wbc, mapping, start, len);
636 mapping_set_error(mapping, -ENOSPC);
637 break;
638
639 case -EROFS:
640 case -EIO:
641 case -EREMOTEIO:
642 case -EFBIG:
643 case -ENOENT:
644 case -ENOMEDIUM:
645 case -ENXIO:
646 trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
647 afs_kill_pages(mapping, start, len);
648 mapping_set_error(mapping, ret);
649 break;
650 }
651
652 _leave(" = %d", ret);
653 return ret;
654}
655
656/*
657 * write a page back to the server
658 * - the caller locked the page for us
659 */
660int afs_writepage(struct page *page, struct writeback_control *wbc)
661{
662 ssize_t ret;
663 loff_t start;
664
665 _enter("{%lx},", page->index);
666
667 start = page->index * PAGE_SIZE;
668 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
669 start, LLONG_MAX - start);
670 if (ret < 0) {
671 _leave(" = %zd", ret);
672 return ret;
673 }
674
675 _leave(" = 0");
676 return 0;
677}
678
679/*
680 * write a region of pages back to the server
681 */
682static int afs_writepages_region(struct address_space *mapping,
683 struct writeback_control *wbc,
684 loff_t start, loff_t end, loff_t *_next)
685{
686 struct page *page;
687 ssize_t ret;
688 int n;
689
690 _enter("%llx,%llx,", start, end);
691
692 do {
693 pgoff_t index = start / PAGE_SIZE;
694
695 n = find_get_pages_range_tag(mapping, &index, end / PAGE_SIZE,
696 PAGECACHE_TAG_DIRTY, 1, &page);
697 if (!n)
698 break;
699
700 start = (loff_t)page->index * PAGE_SIZE; /* May regress with THPs */
701
702 _debug("wback %lx", page->index);
703
704 /* At this point we hold neither the i_pages lock nor the
705 * page lock: the page may be truncated or invalidated
706 * (changing page->mapping to NULL), or even swizzled
707 * back from swapper_space to tmpfs file mapping
708 */
709 if (wbc->sync_mode != WB_SYNC_NONE) {
710 ret = lock_page_killable(page);
711 if (ret < 0) {
712 put_page(page);
713 return ret;
714 }
715 } else {
716 if (!trylock_page(page)) {
717 put_page(page);
718 return 0;
719 }
720 }
721
722 if (page->mapping != mapping || !PageDirty(page)) {
723 start += thp_size(page);
724 unlock_page(page);
725 put_page(page);
726 continue;
727 }
728
729 if (PageWriteback(page)) {
730 unlock_page(page);
731 if (wbc->sync_mode != WB_SYNC_NONE)
732 wait_on_page_writeback(page);
733 put_page(page);
734 continue;
735 }
736
737 if (!clear_page_dirty_for_io(page))
738 BUG();
739 ret = afs_write_back_from_locked_page(mapping, wbc, page, start, end);
740 put_page(page);
741 if (ret < 0) {
742 _leave(" = %zd", ret);
743 return ret;
744 }
745
746 start += ret;
747
748 cond_resched();
749 } while (wbc->nr_to_write > 0);
750
751 *_next = start;
752 _leave(" = 0 [%llx]", *_next);
753 return 0;
754}
755
756/*
757 * write some of the pending data back to the server
758 */
759int afs_writepages(struct address_space *mapping,
760 struct writeback_control *wbc)
761{
762 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
763 loff_t start, next;
764 int ret;
765
766 _enter("");
767
768 /* We have to be careful as we can end up racing with setattr()
769 * truncating the pagecache since the caller doesn't take a lock here
770 * to prevent it.
771 */
772 if (wbc->sync_mode == WB_SYNC_ALL)
773 down_read(&vnode->validate_lock);
774 else if (!down_read_trylock(&vnode->validate_lock))
775 return 0;
776
777 if (wbc->range_cyclic) {
778 start = mapping->writeback_index * PAGE_SIZE;
779 ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX, &next);
780 if (ret == 0) {
781 mapping->writeback_index = next / PAGE_SIZE;
782 if (start > 0 && wbc->nr_to_write > 0) {
783 ret = afs_writepages_region(mapping, wbc, 0,
784 start, &next);
785 if (ret == 0)
786 mapping->writeback_index =
787 next / PAGE_SIZE;
788 }
789 }
790 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
791 ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX, &next);
792 if (wbc->nr_to_write > 0 && ret == 0)
793 mapping->writeback_index = next / PAGE_SIZE;
794 } else {
795 ret = afs_writepages_region(mapping, wbc,
796 wbc->range_start, wbc->range_end, &next);
797 }
798
799 up_read(&vnode->validate_lock);
800 _leave(" = %d", ret);
801 return ret;
802}
803
804/*
805 * write to an AFS file
806 */
807ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
808{
809 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
810 struct afs_file *af = iocb->ki_filp->private_data;
811 ssize_t result;
812 size_t count = iov_iter_count(from);
813
814 _enter("{%llx:%llu},{%zu},",
815 vnode->fid.vid, vnode->fid.vnode, count);
816
817 if (IS_SWAPFILE(&vnode->vfs_inode)) {
818 printk(KERN_INFO
819 "AFS: Attempt to write to active swap file!\n");
820 return -EBUSY;
821 }
822
823 if (!count)
824 return 0;
825
826 result = afs_validate(vnode, af->key);
827 if (result < 0)
828 return result;
829
830 result = generic_file_write_iter(iocb, from);
831
832 _leave(" = %zd", result);
833 return result;
834}
835
836/*
837 * flush any dirty pages for this process, and check for write errors.
838 * - the return status from this call provides a reliable indication of
839 * whether any write errors occurred for this process.
840 */
841int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
842{
843 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
844 struct afs_file *af = file->private_data;
845 int ret;
846
847 _enter("{%llx:%llu},{n=%pD},%d",
848 vnode->fid.vid, vnode->fid.vnode, file,
849 datasync);
850
851 ret = afs_validate(vnode, af->key);
852 if (ret < 0)
853 return ret;
854
855 return file_write_and_wait_range(file, start, end);
856}
857
858/*
859 * notification that a previously read-only page is about to become writable
860 * - if it returns an error, the caller will deliver a bus error signal
861 */
862vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
863{
864 struct page *page = thp_head(vmf->page);
865 struct file *file = vmf->vma->vm_file;
866 struct inode *inode = file_inode(file);
867 struct afs_vnode *vnode = AFS_FS_I(inode);
868 struct afs_file *af = file->private_data;
869 unsigned long priv;
870 vm_fault_t ret = VM_FAULT_RETRY;
871
872 _enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, page->index);
873
874 afs_validate(vnode, af->key);
875
876 sb_start_pagefault(inode->i_sb);
877
878 /* Wait for the page to be written to the cache before we allow it to
879 * be modified. We then assume the entire page will need writing back.
880 */
881#ifdef CONFIG_AFS_FSCACHE
882 if (PageFsCache(page) &&
883 wait_on_page_fscache_killable(page) < 0)
884 goto out;
885#endif
886
887 if (wait_on_page_writeback_killable(page))
888 goto out;
889
890 if (lock_page_killable(page) < 0)
891 goto out;
892
893 /* We mustn't change page->private until writeback is complete as that
894 * details the portion of the page we need to write back and we might
895 * need to redirty the page if there's a problem.
896 */
897 if (wait_on_page_writeback_killable(page) < 0) {
898 unlock_page(page);
899 goto out;
900 }
901
902 priv = afs_page_dirty(page, 0, thp_size(page));
903 priv = afs_page_dirty_mmapped(priv);
904 if (PagePrivate(page)) {
905 set_page_private(page, priv);
906 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite+"), page);
907 } else {
908 attach_page_private(page, (void *)priv);
909 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"), page);
910 }
911 file_update_time(file);
912
913 ret = VM_FAULT_LOCKED;
914out:
915 sb_end_pagefault(inode->i_sb);
916 return ret;
917}
918
919/*
920 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
921 */
922void afs_prune_wb_keys(struct afs_vnode *vnode)
923{
924 LIST_HEAD(graveyard);
925 struct afs_wb_key *wbk, *tmp;
926
927 /* Discard unused keys */
928 spin_lock(&vnode->wb_lock);
929
930 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
931 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
932 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
933 if (refcount_read(&wbk->usage) == 1)
934 list_move(&wbk->vnode_link, &graveyard);
935 }
936 }
937
938 spin_unlock(&vnode->wb_lock);
939
940 while (!list_empty(&graveyard)) {
941 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
942 list_del(&wbk->vnode_link);
943 afs_put_wb_key(wbk);
944 }
945}
946
947/*
948 * Clean up a page during invalidation.
949 */
950int afs_launder_page(struct page *page)
951{
952 struct address_space *mapping = page->mapping;
953 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
954 struct iov_iter iter;
955 struct bio_vec bv[1];
956 unsigned long priv;
957 unsigned int f, t;
958 int ret = 0;
959
960 _enter("{%lx}", page->index);
961
962 priv = page_private(page);
963 if (clear_page_dirty_for_io(page)) {
964 f = 0;
965 t = thp_size(page);
966 if (PagePrivate(page)) {
967 f = afs_page_dirty_from(page, priv);
968 t = afs_page_dirty_to(page, priv);
969 }
970
971 bv[0].bv_page = page;
972 bv[0].bv_offset = f;
973 bv[0].bv_len = t - f;
974 iov_iter_bvec(&iter, WRITE, bv, 1, bv[0].bv_len);
975
976 trace_afs_page_dirty(vnode, tracepoint_string("launder"), page);
977 ret = afs_store_data(vnode, &iter, page_offset(page) + f, true);
978 }
979
980 trace_afs_page_dirty(vnode, tracepoint_string("laundered"), page);
981 detach_page_private(page);
982 wait_on_page_fscache(page);
983 return ret;
984}