xref: /openbmc/linux/fs/afs/write.c (revision 87875c10)
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  */
21 int 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  */
30 int 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 
58 try_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 	 */
88 flush_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 
99 error:
100 	put_page(page);
101 	_leave(" = %d", ret);
102 	return ret;
103 }
104 
105 /*
106  * finalise part of a write to a page
107  */
108 int 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 			i_size_write(&vnode->vfs_inode, 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 
164 out:
165 	unlock_page(page);
166 	put_page(page);
167 	return copied;
168 }
169 
170 /*
171  * kill all the pages in the given range
172  */
173 static 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  */
218 static 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  */
261 static 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  */
297 static 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 
333 static 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 
347 static 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  */
356 static 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 
397 try_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  */
431 static 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 				break;
476 
477 			if (!trylock_page(page))
478 				break;
479 			if (!PageDirty(page) || PageWriteback(page)) {
480 				unlock_page(page);
481 				break;
482 			}
483 
484 			psize = thp_size(page);
485 			priv = page_private(page);
486 			f = afs_page_dirty_from(page, priv);
487 			t = afs_page_dirty_to(page, priv);
488 			if (f != 0 && !new_content) {
489 				unlock_page(page);
490 				break;
491 			}
492 
493 			len += filler + t;
494 			filler = psize - t;
495 			if (len >= max_len || *_count <= 0)
496 				stop = true;
497 			else if (t == psize || new_content)
498 				stop = false;
499 
500 			index += thp_nr_pages(page);
501 			if (!pagevec_add(&pvec, page))
502 				break;
503 			if (stop)
504 				break;
505 		}
506 
507 		if (!stop)
508 			xas_pause(&xas);
509 		rcu_read_unlock();
510 
511 		/* Now, if we obtained any pages, we can shift them to being
512 		 * writable and mark them for caching.
513 		 */
514 		if (!pagevec_count(&pvec))
515 			break;
516 
517 		for (i = 0; i < pagevec_count(&pvec); i++) {
518 			page = pvec.pages[i];
519 			trace_afs_page_dirty(vnode, tracepoint_string("store+"), page);
520 
521 			if (!clear_page_dirty_for_io(page))
522 				BUG();
523 			if (test_set_page_writeback(page))
524 				BUG();
525 
526 			*_count -= thp_nr_pages(page);
527 			unlock_page(page);
528 		}
529 
530 		pagevec_release(&pvec);
531 		cond_resched();
532 	} while (!stop);
533 
534 	*_len = len;
535 }
536 
537 /*
538  * Synchronously write back the locked page and any subsequent non-locked dirty
539  * pages.
540  */
541 static ssize_t afs_write_back_from_locked_page(struct address_space *mapping,
542 					       struct writeback_control *wbc,
543 					       struct page *page,
544 					       loff_t start, loff_t end)
545 {
546 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
547 	struct iov_iter iter;
548 	unsigned long priv;
549 	unsigned int offset, to, len, max_len;
550 	loff_t i_size = i_size_read(&vnode->vfs_inode);
551 	bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
552 	long count = wbc->nr_to_write;
553 	int ret;
554 
555 	_enter(",%lx,%llx-%llx", page->index, start, end);
556 
557 	if (test_set_page_writeback(page))
558 		BUG();
559 
560 	count -= thp_nr_pages(page);
561 
562 	/* Find all consecutive lockable dirty pages that have contiguous
563 	 * written regions, stopping when we find a page that is not
564 	 * immediately lockable, is not dirty or is missing, or we reach the
565 	 * end of the range.
566 	 */
567 	priv = page_private(page);
568 	offset = afs_page_dirty_from(page, priv);
569 	to = afs_page_dirty_to(page, priv);
570 	trace_afs_page_dirty(vnode, tracepoint_string("store"), page);
571 
572 	len = to - offset;
573 	start += offset;
574 	if (start < i_size) {
575 		/* Trim the write to the EOF; the extra data is ignored.  Also
576 		 * put an upper limit on the size of a single storedata op.
577 		 */
578 		max_len = 65536 * 4096;
579 		max_len = min_t(unsigned long long, max_len, end - start + 1);
580 		max_len = min_t(unsigned long long, max_len, i_size - start);
581 
582 		if (len < max_len &&
583 		    (to == thp_size(page) || new_content))
584 			afs_extend_writeback(mapping, vnode, &count,
585 					     start, max_len, new_content, &len);
586 		len = min_t(loff_t, len, max_len);
587 	}
588 
589 	/* We now have a contiguous set of dirty pages, each with writeback
590 	 * set; the first page is still locked at this point, but all the rest
591 	 * have been unlocked.
592 	 */
593 	unlock_page(page);
594 
595 	if (start < i_size) {
596 		_debug("write back %x @%llx [%llx]", len, start, i_size);
597 
598 		iov_iter_xarray(&iter, WRITE, &mapping->i_pages, start, len);
599 		ret = afs_store_data(vnode, &iter, start, false);
600 	} else {
601 		_debug("write discard %x @%llx [%llx]", len, start, i_size);
602 
603 		/* The dirty region was entirely beyond the EOF. */
604 		afs_pages_written_back(vnode, start, len);
605 		ret = 0;
606 	}
607 
608 	switch (ret) {
609 	case 0:
610 		wbc->nr_to_write = count;
611 		ret = len;
612 		break;
613 
614 	default:
615 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
616 		fallthrough;
617 	case -EACCES:
618 	case -EPERM:
619 	case -ENOKEY:
620 	case -EKEYEXPIRED:
621 	case -EKEYREJECTED:
622 	case -EKEYREVOKED:
623 		afs_redirty_pages(wbc, mapping, start, len);
624 		mapping_set_error(mapping, ret);
625 		break;
626 
627 	case -EDQUOT:
628 	case -ENOSPC:
629 		afs_redirty_pages(wbc, mapping, start, len);
630 		mapping_set_error(mapping, -ENOSPC);
631 		break;
632 
633 	case -EROFS:
634 	case -EIO:
635 	case -EREMOTEIO:
636 	case -EFBIG:
637 	case -ENOENT:
638 	case -ENOMEDIUM:
639 	case -ENXIO:
640 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
641 		afs_kill_pages(mapping, start, len);
642 		mapping_set_error(mapping, ret);
643 		break;
644 	}
645 
646 	_leave(" = %d", ret);
647 	return ret;
648 }
649 
650 /*
651  * write a page back to the server
652  * - the caller locked the page for us
653  */
654 int afs_writepage(struct page *page, struct writeback_control *wbc)
655 {
656 	ssize_t ret;
657 	loff_t start;
658 
659 	_enter("{%lx},", page->index);
660 
661 	start = page->index * PAGE_SIZE;
662 	ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
663 					      start, LLONG_MAX - start);
664 	if (ret < 0) {
665 		_leave(" = %zd", ret);
666 		return ret;
667 	}
668 
669 	_leave(" = 0");
670 	return 0;
671 }
672 
673 /*
674  * write a region of pages back to the server
675  */
676 static int afs_writepages_region(struct address_space *mapping,
677 				 struct writeback_control *wbc,
678 				 loff_t start, loff_t end, loff_t *_next)
679 {
680 	struct page *page;
681 	ssize_t ret;
682 	int n;
683 
684 	_enter("%llx,%llx,", start, end);
685 
686 	do {
687 		pgoff_t index = start / PAGE_SIZE;
688 
689 		n = find_get_pages_range_tag(mapping, &index, end / PAGE_SIZE,
690 					     PAGECACHE_TAG_DIRTY, 1, &page);
691 		if (!n)
692 			break;
693 
694 		start = (loff_t)page->index * PAGE_SIZE; /* May regress with THPs */
695 
696 		_debug("wback %lx", page->index);
697 
698 		/* At this point we hold neither the i_pages lock nor the
699 		 * page lock: the page may be truncated or invalidated
700 		 * (changing page->mapping to NULL), or even swizzled
701 		 * back from swapper_space to tmpfs file mapping
702 		 */
703 		if (wbc->sync_mode != WB_SYNC_NONE) {
704 			ret = lock_page_killable(page);
705 			if (ret < 0) {
706 				put_page(page);
707 				return ret;
708 			}
709 		} else {
710 			if (!trylock_page(page)) {
711 				put_page(page);
712 				return 0;
713 			}
714 		}
715 
716 		if (page->mapping != mapping || !PageDirty(page)) {
717 			start += thp_size(page);
718 			unlock_page(page);
719 			put_page(page);
720 			continue;
721 		}
722 
723 		if (PageWriteback(page)) {
724 			unlock_page(page);
725 			if (wbc->sync_mode != WB_SYNC_NONE)
726 				wait_on_page_writeback(page);
727 			put_page(page);
728 			continue;
729 		}
730 
731 		if (!clear_page_dirty_for_io(page))
732 			BUG();
733 		ret = afs_write_back_from_locked_page(mapping, wbc, page, start, end);
734 		put_page(page);
735 		if (ret < 0) {
736 			_leave(" = %zd", ret);
737 			return ret;
738 		}
739 
740 		start += ret;
741 
742 		cond_resched();
743 	} while (wbc->nr_to_write > 0);
744 
745 	*_next = start;
746 	_leave(" = 0 [%llx]", *_next);
747 	return 0;
748 }
749 
750 /*
751  * write some of the pending data back to the server
752  */
753 int afs_writepages(struct address_space *mapping,
754 		   struct writeback_control *wbc)
755 {
756 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
757 	loff_t start, next;
758 	int ret;
759 
760 	_enter("");
761 
762 	/* We have to be careful as we can end up racing with setattr()
763 	 * truncating the pagecache since the caller doesn't take a lock here
764 	 * to prevent it.
765 	 */
766 	if (wbc->sync_mode == WB_SYNC_ALL)
767 		down_read(&vnode->validate_lock);
768 	else if (!down_read_trylock(&vnode->validate_lock))
769 		return 0;
770 
771 	if (wbc->range_cyclic) {
772 		start = mapping->writeback_index * PAGE_SIZE;
773 		ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX, &next);
774 		if (ret == 0) {
775 			mapping->writeback_index = next / PAGE_SIZE;
776 			if (start > 0 && wbc->nr_to_write > 0) {
777 				ret = afs_writepages_region(mapping, wbc, 0,
778 							    start, &next);
779 				if (ret == 0)
780 					mapping->writeback_index =
781 						next / PAGE_SIZE;
782 			}
783 		}
784 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
785 		ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX, &next);
786 		if (wbc->nr_to_write > 0 && ret == 0)
787 			mapping->writeback_index = next / PAGE_SIZE;
788 	} else {
789 		ret = afs_writepages_region(mapping, wbc,
790 					    wbc->range_start, wbc->range_end, &next);
791 	}
792 
793 	up_read(&vnode->validate_lock);
794 	_leave(" = %d", ret);
795 	return ret;
796 }
797 
798 /*
799  * write to an AFS file
800  */
801 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
802 {
803 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
804 	ssize_t result;
805 	size_t count = iov_iter_count(from);
806 
807 	_enter("{%llx:%llu},{%zu},",
808 	       vnode->fid.vid, vnode->fid.vnode, count);
809 
810 	if (IS_SWAPFILE(&vnode->vfs_inode)) {
811 		printk(KERN_INFO
812 		       "AFS: Attempt to write to active swap file!\n");
813 		return -EBUSY;
814 	}
815 
816 	if (!count)
817 		return 0;
818 
819 	result = generic_file_write_iter(iocb, from);
820 
821 	_leave(" = %zd", result);
822 	return result;
823 }
824 
825 /*
826  * flush any dirty pages for this process, and check for write errors.
827  * - the return status from this call provides a reliable indication of
828  *   whether any write errors occurred for this process.
829  */
830 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
831 {
832 	struct inode *inode = file_inode(file);
833 	struct afs_vnode *vnode = AFS_FS_I(inode);
834 
835 	_enter("{%llx:%llu},{n=%pD},%d",
836 	       vnode->fid.vid, vnode->fid.vnode, file,
837 	       datasync);
838 
839 	return file_write_and_wait_range(file, start, end);
840 }
841 
842 /*
843  * notification that a previously read-only page is about to become writable
844  * - if it returns an error, the caller will deliver a bus error signal
845  */
846 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
847 {
848 	struct page *page = thp_head(vmf->page);
849 	struct file *file = vmf->vma->vm_file;
850 	struct inode *inode = file_inode(file);
851 	struct afs_vnode *vnode = AFS_FS_I(inode);
852 	unsigned long priv;
853 	vm_fault_t ret = VM_FAULT_RETRY;
854 
855 	_enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, page->index);
856 
857 	sb_start_pagefault(inode->i_sb);
858 
859 	/* Wait for the page to be written to the cache before we allow it to
860 	 * be modified.  We then assume the entire page will need writing back.
861 	 */
862 #ifdef CONFIG_AFS_FSCACHE
863 	if (PageFsCache(page) &&
864 	    wait_on_page_fscache_killable(page) < 0)
865 		goto out;
866 #endif
867 
868 	if (wait_on_page_writeback_killable(page))
869 		goto out;
870 
871 	if (lock_page_killable(page) < 0)
872 		goto out;
873 
874 	/* We mustn't change page->private until writeback is complete as that
875 	 * details the portion of the page we need to write back and we might
876 	 * need to redirty the page if there's a problem.
877 	 */
878 	if (wait_on_page_writeback_killable(page) < 0) {
879 		unlock_page(page);
880 		goto out;
881 	}
882 
883 	priv = afs_page_dirty(page, 0, thp_size(page));
884 	priv = afs_page_dirty_mmapped(priv);
885 	if (PagePrivate(page)) {
886 		set_page_private(page, priv);
887 		trace_afs_page_dirty(vnode, tracepoint_string("mkwrite+"), page);
888 	} else {
889 		attach_page_private(page, (void *)priv);
890 		trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"), page);
891 	}
892 	file_update_time(file);
893 
894 	ret = VM_FAULT_LOCKED;
895 out:
896 	sb_end_pagefault(inode->i_sb);
897 	return ret;
898 }
899 
900 /*
901  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
902  */
903 void afs_prune_wb_keys(struct afs_vnode *vnode)
904 {
905 	LIST_HEAD(graveyard);
906 	struct afs_wb_key *wbk, *tmp;
907 
908 	/* Discard unused keys */
909 	spin_lock(&vnode->wb_lock);
910 
911 	if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
912 	    !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
913 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
914 			if (refcount_read(&wbk->usage) == 1)
915 				list_move(&wbk->vnode_link, &graveyard);
916 		}
917 	}
918 
919 	spin_unlock(&vnode->wb_lock);
920 
921 	while (!list_empty(&graveyard)) {
922 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
923 		list_del(&wbk->vnode_link);
924 		afs_put_wb_key(wbk);
925 	}
926 }
927 
928 /*
929  * Clean up a page during invalidation.
930  */
931 int afs_launder_page(struct page *page)
932 {
933 	struct address_space *mapping = page->mapping;
934 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
935 	struct iov_iter iter;
936 	struct bio_vec bv[1];
937 	unsigned long priv;
938 	unsigned int f, t;
939 	int ret = 0;
940 
941 	_enter("{%lx}", page->index);
942 
943 	priv = page_private(page);
944 	if (clear_page_dirty_for_io(page)) {
945 		f = 0;
946 		t = thp_size(page);
947 		if (PagePrivate(page)) {
948 			f = afs_page_dirty_from(page, priv);
949 			t = afs_page_dirty_to(page, priv);
950 		}
951 
952 		bv[0].bv_page = page;
953 		bv[0].bv_offset = f;
954 		bv[0].bv_len = t - f;
955 		iov_iter_bvec(&iter, WRITE, bv, 1, bv[0].bv_len);
956 
957 		trace_afs_page_dirty(vnode, tracepoint_string("launder"), page);
958 		ret = afs_store_data(vnode, &iter, (loff_t)page->index * PAGE_SIZE,
959 				     true);
960 	}
961 
962 	trace_afs_page_dirty(vnode, tracepoint_string("laundered"), page);
963 	detach_page_private(page);
964 	wait_on_page_fscache(page);
965 	return ret;
966 }
967