xref: /linux/fs/afs/write.c (revision 724ba6751532055db75992fc6ae21c3e322e94a7)
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 "internal.h"
16 
17 static int afs_writepages_region(struct address_space *mapping,
18 				 struct writeback_control *wbc,
19 				 loff_t start, loff_t end, loff_t *_next,
20 				 bool max_one_loop);
21 
22 static void afs_write_to_cache(struct afs_vnode *vnode, loff_t start, size_t len,
23 			       loff_t i_size, bool caching);
24 
25 #ifdef CONFIG_AFS_FSCACHE
26 /*
27  * Mark a page as having been made dirty and thus needing writeback.  We also
28  * need to pin the cache object to write back to.
29  */
30 bool afs_dirty_folio(struct address_space *mapping, struct folio *folio)
31 {
32 	return fscache_dirty_folio(mapping, folio,
33 				afs_vnode_cache(AFS_FS_I(mapping->host)));
34 }
35 static void afs_folio_start_fscache(bool caching, struct folio *folio)
36 {
37 	if (caching)
38 		folio_start_fscache(folio);
39 }
40 #else
41 static void afs_folio_start_fscache(bool caching, struct folio *folio)
42 {
43 }
44 #endif
45 
46 /*
47  * Flush out a conflicting write.  This may extend the write to the surrounding
48  * pages if also dirty and contiguous to the conflicting region..
49  */
50 static int afs_flush_conflicting_write(struct address_space *mapping,
51 				       struct folio *folio)
52 {
53 	struct writeback_control wbc = {
54 		.sync_mode	= WB_SYNC_ALL,
55 		.nr_to_write	= LONG_MAX,
56 		.range_start	= folio_pos(folio),
57 		.range_end	= LLONG_MAX,
58 	};
59 	loff_t next;
60 
61 	return afs_writepages_region(mapping, &wbc, folio_pos(folio), LLONG_MAX,
62 				     &next, true);
63 }
64 
65 /*
66  * prepare to perform part of a write to a page
67  */
68 int afs_write_begin(struct file *file, struct address_space *mapping,
69 		    loff_t pos, unsigned len,
70 		    struct page **_page, void **fsdata)
71 {
72 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
73 	struct folio *folio;
74 	unsigned long priv;
75 	unsigned f, from;
76 	unsigned t, to;
77 	pgoff_t index;
78 	int ret;
79 
80 	_enter("{%llx:%llu},%llx,%x",
81 	       vnode->fid.vid, vnode->fid.vnode, pos, len);
82 
83 	/* Prefetch area to be written into the cache if we're caching this
84 	 * file.  We need to do this before we get a lock on the page in case
85 	 * there's more than one writer competing for the same cache block.
86 	 */
87 	ret = netfs_write_begin(&vnode->netfs, file, mapping, pos, len, &folio, fsdata);
88 	if (ret < 0)
89 		return ret;
90 
91 	index = folio_index(folio);
92 	from = pos - index * PAGE_SIZE;
93 	to = from + len;
94 
95 try_again:
96 	/* See if this page is already partially written in a way that we can
97 	 * merge the new write with.
98 	 */
99 	if (folio_test_private(folio)) {
100 		priv = (unsigned long)folio_get_private(folio);
101 		f = afs_folio_dirty_from(folio, priv);
102 		t = afs_folio_dirty_to(folio, priv);
103 		ASSERTCMP(f, <=, t);
104 
105 		if (folio_test_writeback(folio)) {
106 			trace_afs_folio_dirty(vnode, tracepoint_string("alrdy"), folio);
107 			folio_unlock(folio);
108 			goto wait_for_writeback;
109 		}
110 		/* If the file is being filled locally, allow inter-write
111 		 * spaces to be merged into writes.  If it's not, only write
112 		 * back what the user gives us.
113 		 */
114 		if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
115 		    (to < f || from > t))
116 			goto flush_conflicting_write;
117 	}
118 
119 	*_page = folio_file_page(folio, pos / PAGE_SIZE);
120 	_leave(" = 0");
121 	return 0;
122 
123 	/* The previous write and this write aren't adjacent or overlapping, so
124 	 * flush the page out.
125 	 */
126 flush_conflicting_write:
127 	trace_afs_folio_dirty(vnode, tracepoint_string("confl"), folio);
128 	folio_unlock(folio);
129 
130 	ret = afs_flush_conflicting_write(mapping, folio);
131 	if (ret < 0)
132 		goto error;
133 
134 wait_for_writeback:
135 	ret = folio_wait_writeback_killable(folio);
136 	if (ret < 0)
137 		goto error;
138 
139 	ret = folio_lock_killable(folio);
140 	if (ret < 0)
141 		goto error;
142 	goto try_again;
143 
144 error:
145 	folio_put(folio);
146 	_leave(" = %d", ret);
147 	return ret;
148 }
149 
150 /*
151  * finalise part of a write to a page
152  */
153 int afs_write_end(struct file *file, struct address_space *mapping,
154 		  loff_t pos, unsigned len, unsigned copied,
155 		  struct page *subpage, void *fsdata)
156 {
157 	struct folio *folio = page_folio(subpage);
158 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
159 	unsigned long priv;
160 	unsigned int f, from = offset_in_folio(folio, pos);
161 	unsigned int t, to = from + copied;
162 	loff_t i_size, write_end_pos;
163 
164 	_enter("{%llx:%llu},{%lx}",
165 	       vnode->fid.vid, vnode->fid.vnode, folio_index(folio));
166 
167 	if (!folio_test_uptodate(folio)) {
168 		if (copied < len) {
169 			copied = 0;
170 			goto out;
171 		}
172 
173 		folio_mark_uptodate(folio);
174 	}
175 
176 	if (copied == 0)
177 		goto out;
178 
179 	write_end_pos = pos + copied;
180 
181 	i_size = i_size_read(&vnode->netfs.inode);
182 	if (write_end_pos > i_size) {
183 		write_seqlock(&vnode->cb_lock);
184 		i_size = i_size_read(&vnode->netfs.inode);
185 		if (write_end_pos > i_size)
186 			afs_set_i_size(vnode, write_end_pos);
187 		write_sequnlock(&vnode->cb_lock);
188 		fscache_update_cookie(afs_vnode_cache(vnode), NULL, &write_end_pos);
189 	}
190 
191 	if (folio_test_private(folio)) {
192 		priv = (unsigned long)folio_get_private(folio);
193 		f = afs_folio_dirty_from(folio, priv);
194 		t = afs_folio_dirty_to(folio, priv);
195 		if (from < f)
196 			f = from;
197 		if (to > t)
198 			t = to;
199 		priv = afs_folio_dirty(folio, f, t);
200 		folio_change_private(folio, (void *)priv);
201 		trace_afs_folio_dirty(vnode, tracepoint_string("dirty+"), folio);
202 	} else {
203 		priv = afs_folio_dirty(folio, from, to);
204 		folio_attach_private(folio, (void *)priv);
205 		trace_afs_folio_dirty(vnode, tracepoint_string("dirty"), folio);
206 	}
207 
208 	if (folio_mark_dirty(folio))
209 		_debug("dirtied %lx", folio_index(folio));
210 
211 out:
212 	folio_unlock(folio);
213 	folio_put(folio);
214 	return copied;
215 }
216 
217 /*
218  * kill all the pages in the given range
219  */
220 static void afs_kill_pages(struct address_space *mapping,
221 			   loff_t start, loff_t len)
222 {
223 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
224 	struct folio *folio;
225 	pgoff_t index = start / PAGE_SIZE;
226 	pgoff_t last = (start + len - 1) / PAGE_SIZE, next;
227 
228 	_enter("{%llx:%llu},%llx @%llx",
229 	       vnode->fid.vid, vnode->fid.vnode, len, start);
230 
231 	do {
232 		_debug("kill %lx (to %lx)", index, last);
233 
234 		folio = filemap_get_folio(mapping, index);
235 		if (IS_ERR(folio)) {
236 			next = index + 1;
237 			continue;
238 		}
239 
240 		next = folio_next_index(folio);
241 
242 		folio_clear_uptodate(folio);
243 		folio_end_writeback(folio);
244 		folio_lock(folio);
245 		generic_error_remove_page(mapping, &folio->page);
246 		folio_unlock(folio);
247 		folio_put(folio);
248 
249 	} while (index = next, index <= last);
250 
251 	_leave("");
252 }
253 
254 /*
255  * Redirty all the pages in a given range.
256  */
257 static void afs_redirty_pages(struct writeback_control *wbc,
258 			      struct address_space *mapping,
259 			      loff_t start, loff_t len)
260 {
261 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
262 	struct folio *folio;
263 	pgoff_t index = start / PAGE_SIZE;
264 	pgoff_t last = (start + len - 1) / PAGE_SIZE, next;
265 
266 	_enter("{%llx:%llu},%llx @%llx",
267 	       vnode->fid.vid, vnode->fid.vnode, len, start);
268 
269 	do {
270 		_debug("redirty %llx @%llx", len, start);
271 
272 		folio = filemap_get_folio(mapping, index);
273 		if (IS_ERR(folio)) {
274 			next = index + 1;
275 			continue;
276 		}
277 
278 		next = index + folio_nr_pages(folio);
279 		folio_redirty_for_writepage(wbc, folio);
280 		folio_end_writeback(folio);
281 		folio_put(folio);
282 	} while (index = next, index <= last);
283 
284 	_leave("");
285 }
286 
287 /*
288  * completion of write to server
289  */
290 static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
291 {
292 	struct address_space *mapping = vnode->netfs.inode.i_mapping;
293 	struct folio *folio;
294 	pgoff_t end;
295 
296 	XA_STATE(xas, &mapping->i_pages, start / PAGE_SIZE);
297 
298 	_enter("{%llx:%llu},{%x @%llx}",
299 	       vnode->fid.vid, vnode->fid.vnode, len, start);
300 
301 	rcu_read_lock();
302 
303 	end = (start + len - 1) / PAGE_SIZE;
304 	xas_for_each(&xas, folio, end) {
305 		if (!folio_test_writeback(folio)) {
306 			kdebug("bad %x @%llx page %lx %lx",
307 			       len, start, folio_index(folio), end);
308 			ASSERT(folio_test_writeback(folio));
309 		}
310 
311 		trace_afs_folio_dirty(vnode, tracepoint_string("clear"), folio);
312 		folio_detach_private(folio);
313 		folio_end_writeback(folio);
314 	}
315 
316 	rcu_read_unlock();
317 
318 	afs_prune_wb_keys(vnode);
319 	_leave("");
320 }
321 
322 /*
323  * Find a key to use for the writeback.  We cached the keys used to author the
324  * writes on the vnode.  *_wbk will contain the last writeback key used or NULL
325  * and we need to start from there if it's set.
326  */
327 static int afs_get_writeback_key(struct afs_vnode *vnode,
328 				 struct afs_wb_key **_wbk)
329 {
330 	struct afs_wb_key *wbk = NULL;
331 	struct list_head *p;
332 	int ret = -ENOKEY, ret2;
333 
334 	spin_lock(&vnode->wb_lock);
335 	if (*_wbk)
336 		p = (*_wbk)->vnode_link.next;
337 	else
338 		p = vnode->wb_keys.next;
339 
340 	while (p != &vnode->wb_keys) {
341 		wbk = list_entry(p, struct afs_wb_key, vnode_link);
342 		_debug("wbk %u", key_serial(wbk->key));
343 		ret2 = key_validate(wbk->key);
344 		if (ret2 == 0) {
345 			refcount_inc(&wbk->usage);
346 			_debug("USE WB KEY %u", key_serial(wbk->key));
347 			break;
348 		}
349 
350 		wbk = NULL;
351 		if (ret == -ENOKEY)
352 			ret = ret2;
353 		p = p->next;
354 	}
355 
356 	spin_unlock(&vnode->wb_lock);
357 	if (*_wbk)
358 		afs_put_wb_key(*_wbk);
359 	*_wbk = wbk;
360 	return 0;
361 }
362 
363 static void afs_store_data_success(struct afs_operation *op)
364 {
365 	struct afs_vnode *vnode = op->file[0].vnode;
366 
367 	op->ctime = op->file[0].scb.status.mtime_client;
368 	afs_vnode_commit_status(op, &op->file[0]);
369 	if (op->error == 0) {
370 		if (!op->store.laundering)
371 			afs_pages_written_back(vnode, op->store.pos, op->store.size);
372 		afs_stat_v(vnode, n_stores);
373 		atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
374 	}
375 }
376 
377 static const struct afs_operation_ops afs_store_data_operation = {
378 	.issue_afs_rpc	= afs_fs_store_data,
379 	.issue_yfs_rpc	= yfs_fs_store_data,
380 	.success	= afs_store_data_success,
381 };
382 
383 /*
384  * write to a file
385  */
386 static int afs_store_data(struct afs_vnode *vnode, struct iov_iter *iter, loff_t pos,
387 			  bool laundering)
388 {
389 	struct afs_operation *op;
390 	struct afs_wb_key *wbk = NULL;
391 	loff_t size = iov_iter_count(iter);
392 	int ret = -ENOKEY;
393 
394 	_enter("%s{%llx:%llu.%u},%llx,%llx",
395 	       vnode->volume->name,
396 	       vnode->fid.vid,
397 	       vnode->fid.vnode,
398 	       vnode->fid.unique,
399 	       size, pos);
400 
401 	ret = afs_get_writeback_key(vnode, &wbk);
402 	if (ret) {
403 		_leave(" = %d [no keys]", ret);
404 		return ret;
405 	}
406 
407 	op = afs_alloc_operation(wbk->key, vnode->volume);
408 	if (IS_ERR(op)) {
409 		afs_put_wb_key(wbk);
410 		return -ENOMEM;
411 	}
412 
413 	afs_op_set_vnode(op, 0, vnode);
414 	op->file[0].dv_delta = 1;
415 	op->file[0].modification = true;
416 	op->store.write_iter = iter;
417 	op->store.pos = pos;
418 	op->store.size = size;
419 	op->store.i_size = max(pos + size, vnode->netfs.remote_i_size);
420 	op->store.laundering = laundering;
421 	op->mtime = vnode->netfs.inode.i_mtime;
422 	op->flags |= AFS_OPERATION_UNINTR;
423 	op->ops = &afs_store_data_operation;
424 
425 try_next_key:
426 	afs_begin_vnode_operation(op);
427 	afs_wait_for_operation(op);
428 
429 	switch (op->error) {
430 	case -EACCES:
431 	case -EPERM:
432 	case -ENOKEY:
433 	case -EKEYEXPIRED:
434 	case -EKEYREJECTED:
435 	case -EKEYREVOKED:
436 		_debug("next");
437 
438 		ret = afs_get_writeback_key(vnode, &wbk);
439 		if (ret == 0) {
440 			key_put(op->key);
441 			op->key = key_get(wbk->key);
442 			goto try_next_key;
443 		}
444 		break;
445 	}
446 
447 	afs_put_wb_key(wbk);
448 	_leave(" = %d", op->error);
449 	return afs_put_operation(op);
450 }
451 
452 /*
453  * Extend the region to be written back to include subsequent contiguously
454  * dirty pages if possible, but don't sleep while doing so.
455  *
456  * If this page holds new content, then we can include filler zeros in the
457  * writeback.
458  */
459 static void afs_extend_writeback(struct address_space *mapping,
460 				 struct afs_vnode *vnode,
461 				 long *_count,
462 				 loff_t start,
463 				 loff_t max_len,
464 				 bool new_content,
465 				 bool caching,
466 				 unsigned int *_len)
467 {
468 	struct pagevec pvec;
469 	struct folio *folio;
470 	unsigned long priv;
471 	unsigned int psize, filler = 0;
472 	unsigned int f, t;
473 	loff_t len = *_len;
474 	pgoff_t index = (start + len) / PAGE_SIZE;
475 	bool stop = true;
476 	unsigned int i;
477 
478 	XA_STATE(xas, &mapping->i_pages, index);
479 	pagevec_init(&pvec);
480 
481 	do {
482 		/* Firstly, we gather up a batch of contiguous dirty pages
483 		 * under the RCU read lock - but we can't clear the dirty flags
484 		 * there if any of those pages are mapped.
485 		 */
486 		rcu_read_lock();
487 
488 		xas_for_each(&xas, folio, ULONG_MAX) {
489 			stop = true;
490 			if (xas_retry(&xas, folio))
491 				continue;
492 			if (xa_is_value(folio))
493 				break;
494 			if (folio_index(folio) != index)
495 				break;
496 
497 			if (!folio_try_get_rcu(folio)) {
498 				xas_reset(&xas);
499 				continue;
500 			}
501 
502 			/* Has the page moved or been split? */
503 			if (unlikely(folio != xas_reload(&xas))) {
504 				folio_put(folio);
505 				break;
506 			}
507 
508 			if (!folio_trylock(folio)) {
509 				folio_put(folio);
510 				break;
511 			}
512 			if (!folio_test_dirty(folio) ||
513 			    folio_test_writeback(folio) ||
514 			    folio_test_fscache(folio)) {
515 				folio_unlock(folio);
516 				folio_put(folio);
517 				break;
518 			}
519 
520 			psize = folio_size(folio);
521 			priv = (unsigned long)folio_get_private(folio);
522 			f = afs_folio_dirty_from(folio, priv);
523 			t = afs_folio_dirty_to(folio, priv);
524 			if (f != 0 && !new_content) {
525 				folio_unlock(folio);
526 				folio_put(folio);
527 				break;
528 			}
529 
530 			len += filler + t;
531 			filler = psize - t;
532 			if (len >= max_len || *_count <= 0)
533 				stop = true;
534 			else if (t == psize || new_content)
535 				stop = false;
536 
537 			index += folio_nr_pages(folio);
538 			if (!pagevec_add(&pvec, &folio->page))
539 				break;
540 			if (stop)
541 				break;
542 		}
543 
544 		if (!stop)
545 			xas_pause(&xas);
546 		rcu_read_unlock();
547 
548 		/* Now, if we obtained any pages, we can shift them to being
549 		 * writable and mark them for caching.
550 		 */
551 		if (!pagevec_count(&pvec))
552 			break;
553 
554 		for (i = 0; i < pagevec_count(&pvec); i++) {
555 			folio = page_folio(pvec.pages[i]);
556 			trace_afs_folio_dirty(vnode, tracepoint_string("store+"), folio);
557 
558 			if (!folio_clear_dirty_for_io(folio))
559 				BUG();
560 			if (folio_start_writeback(folio))
561 				BUG();
562 			afs_folio_start_fscache(caching, folio);
563 
564 			*_count -= folio_nr_pages(folio);
565 			folio_unlock(folio);
566 		}
567 
568 		pagevec_release(&pvec);
569 		cond_resched();
570 	} while (!stop);
571 
572 	*_len = len;
573 }
574 
575 /*
576  * Synchronously write back the locked page and any subsequent non-locked dirty
577  * pages.
578  */
579 static ssize_t afs_write_back_from_locked_folio(struct address_space *mapping,
580 						struct writeback_control *wbc,
581 						struct folio *folio,
582 						loff_t start, loff_t end)
583 {
584 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
585 	struct iov_iter iter;
586 	unsigned long priv;
587 	unsigned int offset, to, len, max_len;
588 	loff_t i_size = i_size_read(&vnode->netfs.inode);
589 	bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
590 	bool caching = fscache_cookie_enabled(afs_vnode_cache(vnode));
591 	long count = wbc->nr_to_write;
592 	int ret;
593 
594 	_enter(",%lx,%llx-%llx", folio_index(folio), start, end);
595 
596 	if (folio_start_writeback(folio))
597 		BUG();
598 	afs_folio_start_fscache(caching, folio);
599 
600 	count -= folio_nr_pages(folio);
601 
602 	/* Find all consecutive lockable dirty pages that have contiguous
603 	 * written regions, stopping when we find a page that is not
604 	 * immediately lockable, is not dirty or is missing, or we reach the
605 	 * end of the range.
606 	 */
607 	priv = (unsigned long)folio_get_private(folio);
608 	offset = afs_folio_dirty_from(folio, priv);
609 	to = afs_folio_dirty_to(folio, priv);
610 	trace_afs_folio_dirty(vnode, tracepoint_string("store"), folio);
611 
612 	len = to - offset;
613 	start += offset;
614 	if (start < i_size) {
615 		/* Trim the write to the EOF; the extra data is ignored.  Also
616 		 * put an upper limit on the size of a single storedata op.
617 		 */
618 		max_len = 65536 * 4096;
619 		max_len = min_t(unsigned long long, max_len, end - start + 1);
620 		max_len = min_t(unsigned long long, max_len, i_size - start);
621 
622 		if (len < max_len &&
623 		    (to == folio_size(folio) || new_content))
624 			afs_extend_writeback(mapping, vnode, &count,
625 					     start, max_len, new_content,
626 					     caching, &len);
627 		len = min_t(loff_t, len, max_len);
628 	}
629 
630 	/* We now have a contiguous set of dirty pages, each with writeback
631 	 * set; the first page is still locked at this point, but all the rest
632 	 * have been unlocked.
633 	 */
634 	folio_unlock(folio);
635 
636 	if (start < i_size) {
637 		_debug("write back %x @%llx [%llx]", len, start, i_size);
638 
639 		/* Speculatively write to the cache.  We have to fix this up
640 		 * later if the store fails.
641 		 */
642 		afs_write_to_cache(vnode, start, len, i_size, caching);
643 
644 		iov_iter_xarray(&iter, ITER_SOURCE, &mapping->i_pages, start, len);
645 		ret = afs_store_data(vnode, &iter, start, false);
646 	} else {
647 		_debug("write discard %x @%llx [%llx]", len, start, i_size);
648 
649 		/* The dirty region was entirely beyond the EOF. */
650 		fscache_clear_page_bits(mapping, start, len, caching);
651 		afs_pages_written_back(vnode, start, len);
652 		ret = 0;
653 	}
654 
655 	switch (ret) {
656 	case 0:
657 		wbc->nr_to_write = count;
658 		ret = len;
659 		break;
660 
661 	default:
662 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
663 		fallthrough;
664 	case -EACCES:
665 	case -EPERM:
666 	case -ENOKEY:
667 	case -EKEYEXPIRED:
668 	case -EKEYREJECTED:
669 	case -EKEYREVOKED:
670 	case -ENETRESET:
671 		afs_redirty_pages(wbc, mapping, start, len);
672 		mapping_set_error(mapping, ret);
673 		break;
674 
675 	case -EDQUOT:
676 	case -ENOSPC:
677 		afs_redirty_pages(wbc, mapping, start, len);
678 		mapping_set_error(mapping, -ENOSPC);
679 		break;
680 
681 	case -EROFS:
682 	case -EIO:
683 	case -EREMOTEIO:
684 	case -EFBIG:
685 	case -ENOENT:
686 	case -ENOMEDIUM:
687 	case -ENXIO:
688 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
689 		afs_kill_pages(mapping, start, len);
690 		mapping_set_error(mapping, ret);
691 		break;
692 	}
693 
694 	_leave(" = %d", ret);
695 	return ret;
696 }
697 
698 /*
699  * write a region of pages back to the server
700  */
701 static int afs_writepages_region(struct address_space *mapping,
702 				 struct writeback_control *wbc,
703 				 loff_t start, loff_t end, loff_t *_next,
704 				 bool max_one_loop)
705 {
706 	struct folio *folio;
707 	struct folio_batch fbatch;
708 	ssize_t ret;
709 	unsigned int i;
710 	int n, skips = 0;
711 
712 	_enter("%llx,%llx,", start, end);
713 	folio_batch_init(&fbatch);
714 
715 	do {
716 		pgoff_t index = start / PAGE_SIZE;
717 
718 		n = filemap_get_folios_tag(mapping, &index, end / PAGE_SIZE,
719 					PAGECACHE_TAG_DIRTY, &fbatch);
720 
721 		if (!n)
722 			break;
723 		for (i = 0; i < n; i++) {
724 			folio = fbatch.folios[i];
725 			start = folio_pos(folio); /* May regress with THPs */
726 
727 			_debug("wback %lx", folio_index(folio));
728 
729 			/* At this point we hold neither the i_pages lock nor the
730 			 * page lock: the page may be truncated or invalidated
731 			 * (changing page->mapping to NULL), or even swizzled
732 			 * back from swapper_space to tmpfs file mapping
733 			 */
734 			if (wbc->sync_mode != WB_SYNC_NONE) {
735 				ret = folio_lock_killable(folio);
736 				if (ret < 0) {
737 					folio_batch_release(&fbatch);
738 					return ret;
739 				}
740 			} else {
741 				if (!folio_trylock(folio))
742 					continue;
743 			}
744 
745 			if (folio->mapping != mapping ||
746 			    !folio_test_dirty(folio)) {
747 				start += folio_size(folio);
748 				folio_unlock(folio);
749 				continue;
750 			}
751 
752 			if (folio_test_writeback(folio) ||
753 			    folio_test_fscache(folio)) {
754 				folio_unlock(folio);
755 				if (wbc->sync_mode != WB_SYNC_NONE) {
756 					folio_wait_writeback(folio);
757 #ifdef CONFIG_AFS_FSCACHE
758 					folio_wait_fscache(folio);
759 #endif
760 				} else {
761 					start += folio_size(folio);
762 				}
763 				if (wbc->sync_mode == WB_SYNC_NONE) {
764 					if (skips >= 5 || need_resched()) {
765 						*_next = start;
766 						_leave(" = 0 [%llx]", *_next);
767 						return 0;
768 					}
769 					skips++;
770 				}
771 				continue;
772 			}
773 
774 			if (!folio_clear_dirty_for_io(folio))
775 				BUG();
776 			ret = afs_write_back_from_locked_folio(mapping, wbc,
777 					folio, start, end);
778 			if (ret < 0) {
779 				_leave(" = %zd", ret);
780 				folio_batch_release(&fbatch);
781 				return ret;
782 			}
783 
784 			start += ret;
785 		}
786 
787 		folio_batch_release(&fbatch);
788 		cond_resched();
789 	} while (wbc->nr_to_write > 0);
790 
791 	*_next = start;
792 	_leave(" = 0 [%llx]", *_next);
793 	return 0;
794 }
795 
796 /*
797  * write some of the pending data back to the server
798  */
799 int afs_writepages(struct address_space *mapping,
800 		   struct writeback_control *wbc)
801 {
802 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
803 	loff_t start, next;
804 	int ret;
805 
806 	_enter("");
807 
808 	/* We have to be careful as we can end up racing with setattr()
809 	 * truncating the pagecache since the caller doesn't take a lock here
810 	 * to prevent it.
811 	 */
812 	if (wbc->sync_mode == WB_SYNC_ALL)
813 		down_read(&vnode->validate_lock);
814 	else if (!down_read_trylock(&vnode->validate_lock))
815 		return 0;
816 
817 	if (wbc->range_cyclic) {
818 		start = mapping->writeback_index * PAGE_SIZE;
819 		ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX,
820 					    &next, false);
821 		if (ret == 0) {
822 			mapping->writeback_index = next / PAGE_SIZE;
823 			if (start > 0 && wbc->nr_to_write > 0) {
824 				ret = afs_writepages_region(mapping, wbc, 0,
825 							    start, &next, false);
826 				if (ret == 0)
827 					mapping->writeback_index =
828 						next / PAGE_SIZE;
829 			}
830 		}
831 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
832 		ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX,
833 					    &next, false);
834 		if (wbc->nr_to_write > 0 && ret == 0)
835 			mapping->writeback_index = next / PAGE_SIZE;
836 	} else {
837 		ret = afs_writepages_region(mapping, wbc,
838 					    wbc->range_start, wbc->range_end,
839 					    &next, false);
840 	}
841 
842 	up_read(&vnode->validate_lock);
843 	_leave(" = %d", ret);
844 	return ret;
845 }
846 
847 /*
848  * write to an AFS file
849  */
850 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
851 {
852 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
853 	struct afs_file *af = iocb->ki_filp->private_data;
854 	ssize_t result;
855 	size_t count = iov_iter_count(from);
856 
857 	_enter("{%llx:%llu},{%zu},",
858 	       vnode->fid.vid, vnode->fid.vnode, count);
859 
860 	if (IS_SWAPFILE(&vnode->netfs.inode)) {
861 		printk(KERN_INFO
862 		       "AFS: Attempt to write to active swap file!\n");
863 		return -EBUSY;
864 	}
865 
866 	if (!count)
867 		return 0;
868 
869 	result = afs_validate(vnode, af->key);
870 	if (result < 0)
871 		return result;
872 
873 	result = generic_file_write_iter(iocb, from);
874 
875 	_leave(" = %zd", result);
876 	return result;
877 }
878 
879 /*
880  * flush any dirty pages for this process, and check for write errors.
881  * - the return status from this call provides a reliable indication of
882  *   whether any write errors occurred for this process.
883  */
884 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
885 {
886 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
887 	struct afs_file *af = file->private_data;
888 	int ret;
889 
890 	_enter("{%llx:%llu},{n=%pD},%d",
891 	       vnode->fid.vid, vnode->fid.vnode, file,
892 	       datasync);
893 
894 	ret = afs_validate(vnode, af->key);
895 	if (ret < 0)
896 		return ret;
897 
898 	return file_write_and_wait_range(file, start, end);
899 }
900 
901 /*
902  * notification that a previously read-only page is about to become writable
903  * - if it returns an error, the caller will deliver a bus error signal
904  */
905 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
906 {
907 	struct folio *folio = page_folio(vmf->page);
908 	struct file *file = vmf->vma->vm_file;
909 	struct inode *inode = file_inode(file);
910 	struct afs_vnode *vnode = AFS_FS_I(inode);
911 	struct afs_file *af = file->private_data;
912 	unsigned long priv;
913 	vm_fault_t ret = VM_FAULT_RETRY;
914 
915 	_enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, folio_index(folio));
916 
917 	afs_validate(vnode, af->key);
918 
919 	sb_start_pagefault(inode->i_sb);
920 
921 	/* Wait for the page to be written to the cache before we allow it to
922 	 * be modified.  We then assume the entire page will need writing back.
923 	 */
924 #ifdef CONFIG_AFS_FSCACHE
925 	if (folio_test_fscache(folio) &&
926 	    folio_wait_fscache_killable(folio) < 0)
927 		goto out;
928 #endif
929 
930 	if (folio_wait_writeback_killable(folio))
931 		goto out;
932 
933 	if (folio_lock_killable(folio) < 0)
934 		goto out;
935 
936 	/* We mustn't change folio->private until writeback is complete as that
937 	 * details the portion of the page we need to write back and we might
938 	 * need to redirty the page if there's a problem.
939 	 */
940 	if (folio_wait_writeback_killable(folio) < 0) {
941 		folio_unlock(folio);
942 		goto out;
943 	}
944 
945 	priv = afs_folio_dirty(folio, 0, folio_size(folio));
946 	priv = afs_folio_dirty_mmapped(priv);
947 	if (folio_test_private(folio)) {
948 		folio_change_private(folio, (void *)priv);
949 		trace_afs_folio_dirty(vnode, tracepoint_string("mkwrite+"), folio);
950 	} else {
951 		folio_attach_private(folio, (void *)priv);
952 		trace_afs_folio_dirty(vnode, tracepoint_string("mkwrite"), folio);
953 	}
954 	file_update_time(file);
955 
956 	ret = VM_FAULT_LOCKED;
957 out:
958 	sb_end_pagefault(inode->i_sb);
959 	return ret;
960 }
961 
962 /*
963  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
964  */
965 void afs_prune_wb_keys(struct afs_vnode *vnode)
966 {
967 	LIST_HEAD(graveyard);
968 	struct afs_wb_key *wbk, *tmp;
969 
970 	/* Discard unused keys */
971 	spin_lock(&vnode->wb_lock);
972 
973 	if (!mapping_tagged(&vnode->netfs.inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
974 	    !mapping_tagged(&vnode->netfs.inode.i_data, PAGECACHE_TAG_DIRTY)) {
975 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
976 			if (refcount_read(&wbk->usage) == 1)
977 				list_move(&wbk->vnode_link, &graveyard);
978 		}
979 	}
980 
981 	spin_unlock(&vnode->wb_lock);
982 
983 	while (!list_empty(&graveyard)) {
984 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
985 		list_del(&wbk->vnode_link);
986 		afs_put_wb_key(wbk);
987 	}
988 }
989 
990 /*
991  * Clean up a page during invalidation.
992  */
993 int afs_launder_folio(struct folio *folio)
994 {
995 	struct afs_vnode *vnode = AFS_FS_I(folio_inode(folio));
996 	struct iov_iter iter;
997 	struct bio_vec bv;
998 	unsigned long priv;
999 	unsigned int f, t;
1000 	int ret = 0;
1001 
1002 	_enter("{%lx}", folio->index);
1003 
1004 	priv = (unsigned long)folio_get_private(folio);
1005 	if (folio_clear_dirty_for_io(folio)) {
1006 		f = 0;
1007 		t = folio_size(folio);
1008 		if (folio_test_private(folio)) {
1009 			f = afs_folio_dirty_from(folio, priv);
1010 			t = afs_folio_dirty_to(folio, priv);
1011 		}
1012 
1013 		bvec_set_folio(&bv, folio, t - f, f);
1014 		iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, bv.bv_len);
1015 
1016 		trace_afs_folio_dirty(vnode, tracepoint_string("launder"), folio);
1017 		ret = afs_store_data(vnode, &iter, folio_pos(folio) + f, true);
1018 	}
1019 
1020 	trace_afs_folio_dirty(vnode, tracepoint_string("laundered"), folio);
1021 	folio_detach_private(folio);
1022 	folio_wait_fscache(folio);
1023 	return ret;
1024 }
1025 
1026 /*
1027  * Deal with the completion of writing the data to the cache.
1028  */
1029 static void afs_write_to_cache_done(void *priv, ssize_t transferred_or_error,
1030 				    bool was_async)
1031 {
1032 	struct afs_vnode *vnode = priv;
1033 
1034 	if (IS_ERR_VALUE(transferred_or_error) &&
1035 	    transferred_or_error != -ENOBUFS)
1036 		afs_invalidate_cache(vnode, 0);
1037 }
1038 
1039 /*
1040  * Save the write to the cache also.
1041  */
1042 static void afs_write_to_cache(struct afs_vnode *vnode,
1043 			       loff_t start, size_t len, loff_t i_size,
1044 			       bool caching)
1045 {
1046 	fscache_write_to_cache(afs_vnode_cache(vnode),
1047 			       vnode->netfs.inode.i_mapping, start, len, i_size,
1048 			       afs_write_to_cache_done, vnode, caching);
1049 }
1050