Loading...
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}
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 "internal.h"
15
16/*
17 * mark a page as having been made dirty and thus needing writeback
18 */
19int afs_set_page_dirty(struct page *page)
20{
21 _enter("");
22 return __set_page_dirty_nobuffers(page);
23}
24
25/*
26 * partly or wholly fill a page that's under preparation for writing
27 */
28static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
29 loff_t pos, unsigned int len, struct page *page)
30{
31 struct afs_read *req;
32 size_t p;
33 void *data;
34 int ret;
35
36 _enter(",,%llu", (unsigned long long)pos);
37
38 if (pos >= vnode->vfs_inode.i_size) {
39 p = pos & ~PAGE_MASK;
40 ASSERTCMP(p + len, <=, PAGE_SIZE);
41 data = kmap(page);
42 memset(data + p, 0, len);
43 kunmap(page);
44 return 0;
45 }
46
47 req = kzalloc(struct_size(req, array, 1), GFP_KERNEL);
48 if (!req)
49 return -ENOMEM;
50
51 refcount_set(&req->usage, 1);
52 req->pos = pos;
53 req->len = len;
54 req->nr_pages = 1;
55 req->pages = req->array;
56 req->pages[0] = page;
57 get_page(page);
58
59 ret = afs_fetch_data(vnode, key, req);
60 afs_put_read(req);
61 if (ret < 0) {
62 if (ret == -ENOENT) {
63 _debug("got NOENT from server"
64 " - marking file deleted and stale");
65 set_bit(AFS_VNODE_DELETED, &vnode->flags);
66 ret = -ESTALE;
67 }
68 }
69
70 _leave(" = %d", ret);
71 return ret;
72}
73
74/*
75 * prepare to perform part of a write to a page
76 */
77int afs_write_begin(struct file *file, struct address_space *mapping,
78 loff_t pos, unsigned len, unsigned flags,
79 struct page **pagep, void **fsdata)
80{
81 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
82 struct page *page;
83 struct key *key = afs_file_key(file);
84 unsigned long priv;
85 unsigned f, from = pos & (PAGE_SIZE - 1);
86 unsigned t, to = from + len;
87 pgoff_t index = pos >> PAGE_SHIFT;
88 int ret;
89
90 _enter("{%llx:%llu},{%lx},%u,%u",
91 vnode->fid.vid, vnode->fid.vnode, index, from, to);
92
93 /* We want to store information about how much of a page is altered in
94 * page->private.
95 */
96 BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
97
98 page = grab_cache_page_write_begin(mapping, index, flags);
99 if (!page)
100 return -ENOMEM;
101
102 if (!PageUptodate(page) && len != PAGE_SIZE) {
103 ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
104 if (ret < 0) {
105 unlock_page(page);
106 put_page(page);
107 _leave(" = %d [prep]", ret);
108 return ret;
109 }
110 SetPageUptodate(page);
111 }
112
113 /* page won't leak in error case: it eventually gets cleaned off LRU */
114 *pagep = page;
115
116try_again:
117 /* See if this page is already partially written in a way that we can
118 * merge the new write with.
119 */
120 t = f = 0;
121 if (PagePrivate(page)) {
122 priv = page_private(page);
123 f = priv & AFS_PRIV_MAX;
124 t = priv >> AFS_PRIV_SHIFT;
125 ASSERTCMP(f, <=, t);
126 }
127
128 if (f != t) {
129 if (PageWriteback(page)) {
130 trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
131 page->index, priv);
132 goto flush_conflicting_write;
133 }
134 /* If the file is being filled locally, allow inter-write
135 * spaces to be merged into writes. If it's not, only write
136 * back what the user gives us.
137 */
138 if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
139 (to < f || from > t))
140 goto flush_conflicting_write;
141 if (from < f)
142 f = from;
143 if (to > t)
144 t = to;
145 } else {
146 f = from;
147 t = to;
148 }
149
150 priv = (unsigned long)t << AFS_PRIV_SHIFT;
151 priv |= f;
152 trace_afs_page_dirty(vnode, tracepoint_string("begin"),
153 page->index, priv);
154 SetPagePrivate(page);
155 set_page_private(page, priv);
156 _leave(" = 0");
157 return 0;
158
159 /* The previous write and this write aren't adjacent or overlapping, so
160 * flush the page out.
161 */
162flush_conflicting_write:
163 _debug("flush conflict");
164 ret = write_one_page(page);
165 if (ret < 0) {
166 _leave(" = %d", ret);
167 return ret;
168 }
169
170 ret = lock_page_killable(page);
171 if (ret < 0) {
172 _leave(" = %d", ret);
173 return ret;
174 }
175 goto try_again;
176}
177
178/*
179 * finalise part of a write to a page
180 */
181int afs_write_end(struct file *file, struct address_space *mapping,
182 loff_t pos, unsigned len, unsigned copied,
183 struct page *page, void *fsdata)
184{
185 struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
186 struct key *key = afs_file_key(file);
187 loff_t i_size, maybe_i_size;
188 int ret;
189
190 _enter("{%llx:%llu},{%lx}",
191 vnode->fid.vid, vnode->fid.vnode, page->index);
192
193 maybe_i_size = pos + copied;
194
195 i_size = i_size_read(&vnode->vfs_inode);
196 if (maybe_i_size > i_size) {
197 write_seqlock(&vnode->cb_lock);
198 i_size = i_size_read(&vnode->vfs_inode);
199 if (maybe_i_size > i_size)
200 i_size_write(&vnode->vfs_inode, maybe_i_size);
201 write_sequnlock(&vnode->cb_lock);
202 }
203
204 if (!PageUptodate(page)) {
205 if (copied < len) {
206 /* Try and load any missing data from the server. The
207 * unmarshalling routine will take care of clearing any
208 * bits that are beyond the EOF.
209 */
210 ret = afs_fill_page(vnode, key, pos + copied,
211 len - copied, page);
212 if (ret < 0)
213 goto out;
214 }
215 SetPageUptodate(page);
216 }
217
218 set_page_dirty(page);
219 if (PageDirty(page))
220 _debug("dirtied");
221 ret = copied;
222
223out:
224 unlock_page(page);
225 put_page(page);
226 return ret;
227}
228
229/*
230 * kill all the pages in the given range
231 */
232static void afs_kill_pages(struct address_space *mapping,
233 pgoff_t first, pgoff_t last)
234{
235 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
236 struct pagevec pv;
237 unsigned count, loop;
238
239 _enter("{%llx:%llu},%lx-%lx",
240 vnode->fid.vid, vnode->fid.vnode, first, last);
241
242 pagevec_init(&pv);
243
244 do {
245 _debug("kill %lx-%lx", first, last);
246
247 count = last - first + 1;
248 if (count > PAGEVEC_SIZE)
249 count = PAGEVEC_SIZE;
250 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
251 ASSERTCMP(pv.nr, ==, count);
252
253 for (loop = 0; loop < count; loop++) {
254 struct page *page = pv.pages[loop];
255 ClearPageUptodate(page);
256 SetPageError(page);
257 end_page_writeback(page);
258 if (page->index >= first)
259 first = page->index + 1;
260 lock_page(page);
261 generic_error_remove_page(mapping, page);
262 unlock_page(page);
263 }
264
265 __pagevec_release(&pv);
266 } while (first <= last);
267
268 _leave("");
269}
270
271/*
272 * Redirty all the pages in a given range.
273 */
274static void afs_redirty_pages(struct writeback_control *wbc,
275 struct address_space *mapping,
276 pgoff_t first, pgoff_t last)
277{
278 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
279 struct pagevec pv;
280 unsigned count, loop;
281
282 _enter("{%llx:%llu},%lx-%lx",
283 vnode->fid.vid, vnode->fid.vnode, first, last);
284
285 pagevec_init(&pv);
286
287 do {
288 _debug("redirty %lx-%lx", first, last);
289
290 count = last - first + 1;
291 if (count > PAGEVEC_SIZE)
292 count = PAGEVEC_SIZE;
293 pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
294 ASSERTCMP(pv.nr, ==, count);
295
296 for (loop = 0; loop < count; loop++) {
297 struct page *page = pv.pages[loop];
298
299 redirty_page_for_writepage(wbc, page);
300 end_page_writeback(page);
301 if (page->index >= first)
302 first = page->index + 1;
303 }
304
305 __pagevec_release(&pv);
306 } while (first <= last);
307
308 _leave("");
309}
310
311/*
312 * completion of write to server
313 */
314static void afs_pages_written_back(struct afs_vnode *vnode,
315 pgoff_t first, pgoff_t last)
316{
317 struct pagevec pv;
318 unsigned long priv;
319 unsigned count, loop;
320
321 _enter("{%llx:%llu},{%lx-%lx}",
322 vnode->fid.vid, vnode->fid.vnode, first, last);
323
324 pagevec_init(&pv);
325
326 do {
327 _debug("done %lx-%lx", first, last);
328
329 count = last - first + 1;
330 if (count > PAGEVEC_SIZE)
331 count = PAGEVEC_SIZE;
332 pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
333 first, count, pv.pages);
334 ASSERTCMP(pv.nr, ==, count);
335
336 for (loop = 0; loop < count; loop++) {
337 priv = page_private(pv.pages[loop]);
338 trace_afs_page_dirty(vnode, tracepoint_string("clear"),
339 pv.pages[loop]->index, priv);
340 set_page_private(pv.pages[loop], 0);
341 end_page_writeback(pv.pages[loop]);
342 }
343 first += count;
344 __pagevec_release(&pv);
345 } while (first <= last);
346
347 afs_prune_wb_keys(vnode);
348 _leave("");
349}
350
351/*
352 * Find a key to use for the writeback. We cached the keys used to author the
353 * writes on the vnode. *_wbk will contain the last writeback key used or NULL
354 * and we need to start from there if it's set.
355 */
356static int afs_get_writeback_key(struct afs_vnode *vnode,
357 struct afs_wb_key **_wbk)
358{
359 struct afs_wb_key *wbk = NULL;
360 struct list_head *p;
361 int ret = -ENOKEY, ret2;
362
363 spin_lock(&vnode->wb_lock);
364 if (*_wbk)
365 p = (*_wbk)->vnode_link.next;
366 else
367 p = vnode->wb_keys.next;
368
369 while (p != &vnode->wb_keys) {
370 wbk = list_entry(p, struct afs_wb_key, vnode_link);
371 _debug("wbk %u", key_serial(wbk->key));
372 ret2 = key_validate(wbk->key);
373 if (ret2 == 0) {
374 refcount_inc(&wbk->usage);
375 _debug("USE WB KEY %u", key_serial(wbk->key));
376 break;
377 }
378
379 wbk = NULL;
380 if (ret == -ENOKEY)
381 ret = ret2;
382 p = p->next;
383 }
384
385 spin_unlock(&vnode->wb_lock);
386 if (*_wbk)
387 afs_put_wb_key(*_wbk);
388 *_wbk = wbk;
389 return 0;
390}
391
392static void afs_store_data_success(struct afs_operation *op)
393{
394 struct afs_vnode *vnode = op->file[0].vnode;
395
396 op->ctime = op->file[0].scb.status.mtime_client;
397 afs_vnode_commit_status(op, &op->file[0]);
398 if (op->error == 0) {
399 afs_pages_written_back(vnode, op->store.first, op->store.last);
400 afs_stat_v(vnode, n_stores);
401 atomic_long_add((op->store.last * PAGE_SIZE + op->store.last_to) -
402 (op->store.first * PAGE_SIZE + op->store.first_offset),
403 &afs_v2net(vnode)->n_store_bytes);
404 }
405}
406
407static const struct afs_operation_ops afs_store_data_operation = {
408 .issue_afs_rpc = afs_fs_store_data,
409 .issue_yfs_rpc = yfs_fs_store_data,
410 .success = afs_store_data_success,
411};
412
413/*
414 * write to a file
415 */
416static int afs_store_data(struct address_space *mapping,
417 pgoff_t first, pgoff_t last,
418 unsigned offset, unsigned to)
419{
420 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
421 struct afs_operation *op;
422 struct afs_wb_key *wbk = NULL;
423 int ret;
424
425 _enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
426 vnode->volume->name,
427 vnode->fid.vid,
428 vnode->fid.vnode,
429 vnode->fid.unique,
430 first, last, offset, to);
431
432 ret = afs_get_writeback_key(vnode, &wbk);
433 if (ret) {
434 _leave(" = %d [no keys]", ret);
435 return ret;
436 }
437
438 op = afs_alloc_operation(wbk->key, vnode->volume);
439 if (IS_ERR(op)) {
440 afs_put_wb_key(wbk);
441 return -ENOMEM;
442 }
443
444 afs_op_set_vnode(op, 0, vnode);
445 op->file[0].dv_delta = 1;
446 op->store.mapping = mapping;
447 op->store.first = first;
448 op->store.last = last;
449 op->store.first_offset = offset;
450 op->store.last_to = to;
451 op->mtime = vnode->vfs_inode.i_mtime;
452 op->flags |= AFS_OPERATION_UNINTR;
453 op->ops = &afs_store_data_operation;
454
455try_next_key:
456 afs_begin_vnode_operation(op);
457 afs_wait_for_operation(op);
458
459 switch (op->error) {
460 case -EACCES:
461 case -EPERM:
462 case -ENOKEY:
463 case -EKEYEXPIRED:
464 case -EKEYREJECTED:
465 case -EKEYREVOKED:
466 _debug("next");
467
468 ret = afs_get_writeback_key(vnode, &wbk);
469 if (ret == 0) {
470 key_put(op->key);
471 op->key = key_get(wbk->key);
472 goto try_next_key;
473 }
474 break;
475 }
476
477 afs_put_wb_key(wbk);
478 _leave(" = %d", op->error);
479 return afs_put_operation(op);
480}
481
482/*
483 * Synchronously write back the locked page and any subsequent non-locked dirty
484 * pages.
485 */
486static int afs_write_back_from_locked_page(struct address_space *mapping,
487 struct writeback_control *wbc,
488 struct page *primary_page,
489 pgoff_t final_page)
490{
491 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
492 struct page *pages[8], *page;
493 unsigned long count, priv;
494 unsigned n, offset, to, f, t;
495 pgoff_t start, first, last;
496 loff_t i_size, end;
497 int loop, ret;
498
499 _enter(",%lx", primary_page->index);
500
501 count = 1;
502 if (test_set_page_writeback(primary_page))
503 BUG();
504
505 /* Find all consecutive lockable dirty pages that have contiguous
506 * written regions, stopping when we find a page that is not
507 * immediately lockable, is not dirty or is missing, or we reach the
508 * end of the range.
509 */
510 start = primary_page->index;
511 priv = page_private(primary_page);
512 offset = priv & AFS_PRIV_MAX;
513 to = priv >> AFS_PRIV_SHIFT;
514 trace_afs_page_dirty(vnode, tracepoint_string("store"),
515 primary_page->index, priv);
516
517 WARN_ON(offset == to);
518 if (offset == to)
519 trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
520 primary_page->index, priv);
521
522 if (start >= final_page ||
523 (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
524 goto no_more;
525
526 start++;
527 do {
528 _debug("more %lx [%lx]", start, count);
529 n = final_page - start + 1;
530 if (n > ARRAY_SIZE(pages))
531 n = ARRAY_SIZE(pages);
532 n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
533 _debug("fgpc %u", n);
534 if (n == 0)
535 goto no_more;
536 if (pages[0]->index != start) {
537 do {
538 put_page(pages[--n]);
539 } while (n > 0);
540 goto no_more;
541 }
542
543 for (loop = 0; loop < n; loop++) {
544 page = pages[loop];
545 if (to != PAGE_SIZE &&
546 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
547 break;
548 if (page->index > final_page)
549 break;
550 if (!trylock_page(page))
551 break;
552 if (!PageDirty(page) || PageWriteback(page)) {
553 unlock_page(page);
554 break;
555 }
556
557 priv = page_private(page);
558 f = priv & AFS_PRIV_MAX;
559 t = priv >> AFS_PRIV_SHIFT;
560 if (f != 0 &&
561 !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
562 unlock_page(page);
563 break;
564 }
565 to = t;
566
567 trace_afs_page_dirty(vnode, tracepoint_string("store+"),
568 page->index, priv);
569
570 if (!clear_page_dirty_for_io(page))
571 BUG();
572 if (test_set_page_writeback(page))
573 BUG();
574 unlock_page(page);
575 put_page(page);
576 }
577 count += loop;
578 if (loop < n) {
579 for (; loop < n; loop++)
580 put_page(pages[loop]);
581 goto no_more;
582 }
583
584 start += loop;
585 } while (start <= final_page && count < 65536);
586
587no_more:
588 /* We now have a contiguous set of dirty pages, each with writeback
589 * set; the first page is still locked at this point, but all the rest
590 * have been unlocked.
591 */
592 unlock_page(primary_page);
593
594 first = primary_page->index;
595 last = first + count - 1;
596
597 end = (loff_t)last * PAGE_SIZE + to;
598 i_size = i_size_read(&vnode->vfs_inode);
599
600 _debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
601 if (end > i_size)
602 to = i_size & ~PAGE_MASK;
603
604 ret = afs_store_data(mapping, first, last, offset, to);
605 switch (ret) {
606 case 0:
607 ret = count;
608 break;
609
610 default:
611 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
612 fallthrough;
613 case -EACCES:
614 case -EPERM:
615 case -ENOKEY:
616 case -EKEYEXPIRED:
617 case -EKEYREJECTED:
618 case -EKEYREVOKED:
619 afs_redirty_pages(wbc, mapping, first, last);
620 mapping_set_error(mapping, ret);
621 break;
622
623 case -EDQUOT:
624 case -ENOSPC:
625 afs_redirty_pages(wbc, mapping, first, last);
626 mapping_set_error(mapping, -ENOSPC);
627 break;
628
629 case -EROFS:
630 case -EIO:
631 case -EREMOTEIO:
632 case -EFBIG:
633 case -ENOENT:
634 case -ENOMEDIUM:
635 case -ENXIO:
636 trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
637 afs_kill_pages(mapping, first, last);
638 mapping_set_error(mapping, ret);
639 break;
640 }
641
642 _leave(" = %d", ret);
643 return ret;
644}
645
646/*
647 * write a page back to the server
648 * - the caller locked the page for us
649 */
650int afs_writepage(struct page *page, struct writeback_control *wbc)
651{
652 int ret;
653
654 _enter("{%lx},", page->index);
655
656 ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
657 wbc->range_end >> PAGE_SHIFT);
658 if (ret < 0) {
659 _leave(" = %d", ret);
660 return 0;
661 }
662
663 wbc->nr_to_write -= ret;
664
665 _leave(" = 0");
666 return 0;
667}
668
669/*
670 * write a region of pages back to the server
671 */
672static int afs_writepages_region(struct address_space *mapping,
673 struct writeback_control *wbc,
674 pgoff_t index, pgoff_t end, pgoff_t *_next)
675{
676 struct page *page;
677 int ret, n;
678
679 _enter(",,%lx,%lx,", index, end);
680
681 do {
682 n = find_get_pages_range_tag(mapping, &index, end,
683 PAGECACHE_TAG_DIRTY, 1, &page);
684 if (!n)
685 break;
686
687 _debug("wback %lx", page->index);
688
689 /*
690 * at this point we hold neither the i_pages lock nor the
691 * page lock: the page may be truncated or invalidated
692 * (changing page->mapping to NULL), or even swizzled
693 * back from swapper_space to tmpfs file mapping
694 */
695 ret = lock_page_killable(page);
696 if (ret < 0) {
697 put_page(page);
698 _leave(" = %d", ret);
699 return ret;
700 }
701
702 if (page->mapping != mapping || !PageDirty(page)) {
703 unlock_page(page);
704 put_page(page);
705 continue;
706 }
707
708 if (PageWriteback(page)) {
709 unlock_page(page);
710 if (wbc->sync_mode != WB_SYNC_NONE)
711 wait_on_page_writeback(page);
712 put_page(page);
713 continue;
714 }
715
716 if (!clear_page_dirty_for_io(page))
717 BUG();
718 ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
719 put_page(page);
720 if (ret < 0) {
721 _leave(" = %d", ret);
722 return ret;
723 }
724
725 wbc->nr_to_write -= ret;
726
727 cond_resched();
728 } while (index < end && wbc->nr_to_write > 0);
729
730 *_next = index;
731 _leave(" = 0 [%lx]", *_next);
732 return 0;
733}
734
735/*
736 * write some of the pending data back to the server
737 */
738int afs_writepages(struct address_space *mapping,
739 struct writeback_control *wbc)
740{
741 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
742 pgoff_t start, end, next;
743 int ret;
744
745 _enter("");
746
747 /* We have to be careful as we can end up racing with setattr()
748 * truncating the pagecache since the caller doesn't take a lock here
749 * to prevent it.
750 */
751 if (wbc->sync_mode == WB_SYNC_ALL)
752 down_read(&vnode->validate_lock);
753 else if (!down_read_trylock(&vnode->validate_lock))
754 return 0;
755
756 if (wbc->range_cyclic) {
757 start = mapping->writeback_index;
758 end = -1;
759 ret = afs_writepages_region(mapping, wbc, start, end, &next);
760 if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
761 ret = afs_writepages_region(mapping, wbc, 0, start,
762 &next);
763 mapping->writeback_index = next;
764 } else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
765 end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
766 ret = afs_writepages_region(mapping, wbc, 0, end, &next);
767 if (wbc->nr_to_write > 0)
768 mapping->writeback_index = next;
769 } else {
770 start = wbc->range_start >> PAGE_SHIFT;
771 end = wbc->range_end >> PAGE_SHIFT;
772 ret = afs_writepages_region(mapping, wbc, start, end, &next);
773 }
774
775 up_read(&vnode->validate_lock);
776 _leave(" = %d", ret);
777 return ret;
778}
779
780/*
781 * write to an AFS file
782 */
783ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
784{
785 struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
786 ssize_t result;
787 size_t count = iov_iter_count(from);
788
789 _enter("{%llx:%llu},{%zu},",
790 vnode->fid.vid, vnode->fid.vnode, count);
791
792 if (IS_SWAPFILE(&vnode->vfs_inode)) {
793 printk(KERN_INFO
794 "AFS: Attempt to write to active swap file!\n");
795 return -EBUSY;
796 }
797
798 if (!count)
799 return 0;
800
801 result = generic_file_write_iter(iocb, from);
802
803 _leave(" = %zd", result);
804 return result;
805}
806
807/*
808 * flush any dirty pages for this process, and check for write errors.
809 * - the return status from this call provides a reliable indication of
810 * whether any write errors occurred for this process.
811 */
812int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
813{
814 struct inode *inode = file_inode(file);
815 struct afs_vnode *vnode = AFS_FS_I(inode);
816
817 _enter("{%llx:%llu},{n=%pD},%d",
818 vnode->fid.vid, vnode->fid.vnode, file,
819 datasync);
820
821 return file_write_and_wait_range(file, start, end);
822}
823
824/*
825 * notification that a previously read-only page is about to become writable
826 * - if it returns an error, the caller will deliver a bus error signal
827 */
828vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
829{
830 struct file *file = vmf->vma->vm_file;
831 struct inode *inode = file_inode(file);
832 struct afs_vnode *vnode = AFS_FS_I(inode);
833 unsigned long priv;
834
835 _enter("{{%llx:%llu}},{%lx}",
836 vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
837
838 sb_start_pagefault(inode->i_sb);
839
840 /* Wait for the page to be written to the cache before we allow it to
841 * be modified. We then assume the entire page will need writing back.
842 */
843#ifdef CONFIG_AFS_FSCACHE
844 fscache_wait_on_page_write(vnode->cache, vmf->page);
845#endif
846
847 if (PageWriteback(vmf->page) &&
848 wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
849 return VM_FAULT_RETRY;
850
851 if (lock_page_killable(vmf->page) < 0)
852 return VM_FAULT_RETRY;
853
854 /* We mustn't change page->private until writeback is complete as that
855 * details the portion of the page we need to write back and we might
856 * need to redirty the page if there's a problem.
857 */
858 wait_on_page_writeback(vmf->page);
859
860 priv = (unsigned long)PAGE_SIZE << AFS_PRIV_SHIFT; /* To */
861 priv |= 0; /* From */
862 trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
863 vmf->page->index, priv);
864 SetPagePrivate(vmf->page);
865 set_page_private(vmf->page, priv);
866 file_update_time(file);
867
868 sb_end_pagefault(inode->i_sb);
869 return VM_FAULT_LOCKED;
870}
871
872/*
873 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
874 */
875void afs_prune_wb_keys(struct afs_vnode *vnode)
876{
877 LIST_HEAD(graveyard);
878 struct afs_wb_key *wbk, *tmp;
879
880 /* Discard unused keys */
881 spin_lock(&vnode->wb_lock);
882
883 if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
884 !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
885 list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
886 if (refcount_read(&wbk->usage) == 1)
887 list_move(&wbk->vnode_link, &graveyard);
888 }
889 }
890
891 spin_unlock(&vnode->wb_lock);
892
893 while (!list_empty(&graveyard)) {
894 wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
895 list_del(&wbk->vnode_link);
896 afs_put_wb_key(wbk);
897 }
898}
899
900/*
901 * Clean up a page during invalidation.
902 */
903int afs_launder_page(struct page *page)
904{
905 struct address_space *mapping = page->mapping;
906 struct afs_vnode *vnode = AFS_FS_I(mapping->host);
907 unsigned long priv;
908 unsigned int f, t;
909 int ret = 0;
910
911 _enter("{%lx}", page->index);
912
913 priv = page_private(page);
914 if (clear_page_dirty_for_io(page)) {
915 f = 0;
916 t = PAGE_SIZE;
917 if (PagePrivate(page)) {
918 f = priv & AFS_PRIV_MAX;
919 t = priv >> AFS_PRIV_SHIFT;
920 }
921
922 trace_afs_page_dirty(vnode, tracepoint_string("launder"),
923 page->index, priv);
924 ret = afs_store_data(mapping, page->index, page->index, t, f);
925 }
926
927 trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
928 page->index, priv);
929 set_page_private(page, 0);
930 ClearPagePrivate(page);
931
932#ifdef CONFIG_AFS_FSCACHE
933 if (PageFsCache(page)) {
934 fscache_wait_on_page_write(vnode->cache, page);
935 fscache_uncache_page(vnode->cache, page);
936 }
937#endif
938 return ret;
939}