xref: /linux/fs/afs/write.c (revision a4eb44a6435d6d8f9e642407a4a06f65eb90ca04)
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 void afs_write_to_cache(struct afs_vnode *vnode, loff_t start, size_t len,
18 			       loff_t i_size, bool caching);
19 
20 #ifdef CONFIG_AFS_FSCACHE
21 /*
22  * Mark a page as having been made dirty and thus needing writeback.  We also
23  * need to pin the cache object to write back to.
24  */
25 int afs_set_page_dirty(struct page *page)
26 {
27 	return fscache_set_page_dirty(page, afs_vnode_cache(AFS_FS_I(page->mapping->host)));
28 }
29 static void afs_folio_start_fscache(bool caching, struct folio *folio)
30 {
31 	if (caching)
32 		folio_start_fscache(folio);
33 }
34 #else
35 static void afs_folio_start_fscache(bool caching, struct folio *folio)
36 {
37 }
38 #endif
39 
40 /*
41  * prepare to perform part of a write to a page
42  */
43 int afs_write_begin(struct file *file, struct address_space *mapping,
44 		    loff_t pos, unsigned len, unsigned flags,
45 		    struct page **_page, void **fsdata)
46 {
47 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
48 	struct folio *folio;
49 	unsigned long priv;
50 	unsigned f, from;
51 	unsigned t, to;
52 	pgoff_t index;
53 	int ret;
54 
55 	_enter("{%llx:%llu},%llx,%x",
56 	       vnode->fid.vid, vnode->fid.vnode, pos, len);
57 
58 	/* Prefetch area to be written into the cache if we're caching this
59 	 * file.  We need to do this before we get a lock on the page in case
60 	 * there's more than one writer competing for the same cache block.
61 	 */
62 	ret = netfs_write_begin(file, mapping, pos, len, flags, &folio, fsdata,
63 				&afs_req_ops, NULL);
64 	if (ret < 0)
65 		return ret;
66 
67 	index = folio_index(folio);
68 	from = pos - index * PAGE_SIZE;
69 	to = from + len;
70 
71 try_again:
72 	/* See if this page is already partially written in a way that we can
73 	 * merge the new write with.
74 	 */
75 	if (folio_test_private(folio)) {
76 		priv = (unsigned long)folio_get_private(folio);
77 		f = afs_folio_dirty_from(folio, priv);
78 		t = afs_folio_dirty_to(folio, priv);
79 		ASSERTCMP(f, <=, t);
80 
81 		if (folio_test_writeback(folio)) {
82 			trace_afs_folio_dirty(vnode, tracepoint_string("alrdy"), folio);
83 			goto flush_conflicting_write;
84 		}
85 		/* If the file is being filled locally, allow inter-write
86 		 * spaces to be merged into writes.  If it's not, only write
87 		 * back what the user gives us.
88 		 */
89 		if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
90 		    (to < f || from > t))
91 			goto flush_conflicting_write;
92 	}
93 
94 	*_page = &folio->page;
95 	_leave(" = 0");
96 	return 0;
97 
98 	/* The previous write and this write aren't adjacent or overlapping, so
99 	 * flush the page out.
100 	 */
101 flush_conflicting_write:
102 	_debug("flush conflict");
103 	ret = folio_write_one(folio);
104 	if (ret < 0)
105 		goto error;
106 
107 	ret = folio_lock_killable(folio);
108 	if (ret < 0)
109 		goto error;
110 	goto try_again;
111 
112 error:
113 	folio_put(folio);
114 	_leave(" = %d", ret);
115 	return ret;
116 }
117 
118 /*
119  * finalise part of a write to a page
120  */
121 int afs_write_end(struct file *file, struct address_space *mapping,
122 		  loff_t pos, unsigned len, unsigned copied,
123 		  struct page *subpage, void *fsdata)
124 {
125 	struct folio *folio = page_folio(subpage);
126 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
127 	unsigned long priv;
128 	unsigned int f, from = offset_in_folio(folio, pos);
129 	unsigned int t, to = from + copied;
130 	loff_t i_size, write_end_pos;
131 
132 	_enter("{%llx:%llu},{%lx}",
133 	       vnode->fid.vid, vnode->fid.vnode, folio_index(folio));
134 
135 	if (!folio_test_uptodate(folio)) {
136 		if (copied < len) {
137 			copied = 0;
138 			goto out;
139 		}
140 
141 		folio_mark_uptodate(folio);
142 	}
143 
144 	if (copied == 0)
145 		goto out;
146 
147 	write_end_pos = pos + copied;
148 
149 	i_size = i_size_read(&vnode->vfs_inode);
150 	if (write_end_pos > i_size) {
151 		write_seqlock(&vnode->cb_lock);
152 		i_size = i_size_read(&vnode->vfs_inode);
153 		if (write_end_pos > i_size)
154 			afs_set_i_size(vnode, write_end_pos);
155 		write_sequnlock(&vnode->cb_lock);
156 		fscache_update_cookie(afs_vnode_cache(vnode), NULL, &write_end_pos);
157 	}
158 
159 	if (folio_test_private(folio)) {
160 		priv = (unsigned long)folio_get_private(folio);
161 		f = afs_folio_dirty_from(folio, priv);
162 		t = afs_folio_dirty_to(folio, priv);
163 		if (from < f)
164 			f = from;
165 		if (to > t)
166 			t = to;
167 		priv = afs_folio_dirty(folio, f, t);
168 		folio_change_private(folio, (void *)priv);
169 		trace_afs_folio_dirty(vnode, tracepoint_string("dirty+"), folio);
170 	} else {
171 		priv = afs_folio_dirty(folio, from, to);
172 		folio_attach_private(folio, (void *)priv);
173 		trace_afs_folio_dirty(vnode, tracepoint_string("dirty"), folio);
174 	}
175 
176 	if (folio_mark_dirty(folio))
177 		_debug("dirtied %lx", folio_index(folio));
178 
179 out:
180 	folio_unlock(folio);
181 	folio_put(folio);
182 	return copied;
183 }
184 
185 /*
186  * kill all the pages in the given range
187  */
188 static void afs_kill_pages(struct address_space *mapping,
189 			   loff_t start, loff_t len)
190 {
191 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
192 	struct folio *folio;
193 	pgoff_t index = start / PAGE_SIZE;
194 	pgoff_t last = (start + len - 1) / PAGE_SIZE, next;
195 
196 	_enter("{%llx:%llu},%llx @%llx",
197 	       vnode->fid.vid, vnode->fid.vnode, len, start);
198 
199 	do {
200 		_debug("kill %lx (to %lx)", index, last);
201 
202 		folio = filemap_get_folio(mapping, index);
203 		if (!folio) {
204 			next = index + 1;
205 			continue;
206 		}
207 
208 		next = folio_next_index(folio);
209 
210 		folio_clear_uptodate(folio);
211 		folio_end_writeback(folio);
212 		folio_lock(folio);
213 		generic_error_remove_page(mapping, &folio->page);
214 		folio_unlock(folio);
215 		folio_put(folio);
216 
217 	} while (index = next, index <= last);
218 
219 	_leave("");
220 }
221 
222 /*
223  * Redirty all the pages in a given range.
224  */
225 static void afs_redirty_pages(struct writeback_control *wbc,
226 			      struct address_space *mapping,
227 			      loff_t start, loff_t len)
228 {
229 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
230 	struct folio *folio;
231 	pgoff_t index = start / PAGE_SIZE;
232 	pgoff_t last = (start + len - 1) / PAGE_SIZE, next;
233 
234 	_enter("{%llx:%llu},%llx @%llx",
235 	       vnode->fid.vid, vnode->fid.vnode, len, start);
236 
237 	do {
238 		_debug("redirty %llx @%llx", len, start);
239 
240 		folio = filemap_get_folio(mapping, index);
241 		if (!folio) {
242 			next = index + 1;
243 			continue;
244 		}
245 
246 		next = index + folio_nr_pages(folio);
247 		folio_redirty_for_writepage(wbc, folio);
248 		folio_end_writeback(folio);
249 		folio_put(folio);
250 	} while (index = next, index <= last);
251 
252 	_leave("");
253 }
254 
255 /*
256  * completion of write to server
257  */
258 static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
259 {
260 	struct address_space *mapping = vnode->vfs_inode.i_mapping;
261 	struct folio *folio;
262 	pgoff_t end;
263 
264 	XA_STATE(xas, &mapping->i_pages, start / PAGE_SIZE);
265 
266 	_enter("{%llx:%llu},{%x @%llx}",
267 	       vnode->fid.vid, vnode->fid.vnode, len, start);
268 
269 	rcu_read_lock();
270 
271 	end = (start + len - 1) / PAGE_SIZE;
272 	xas_for_each(&xas, folio, end) {
273 		if (!folio_test_writeback(folio)) {
274 			kdebug("bad %x @%llx page %lx %lx",
275 			       len, start, folio_index(folio), end);
276 			ASSERT(folio_test_writeback(folio));
277 		}
278 
279 		trace_afs_folio_dirty(vnode, tracepoint_string("clear"), folio);
280 		folio_detach_private(folio);
281 		folio_end_writeback(folio);
282 	}
283 
284 	rcu_read_unlock();
285 
286 	afs_prune_wb_keys(vnode);
287 	_leave("");
288 }
289 
290 /*
291  * Find a key to use for the writeback.  We cached the keys used to author the
292  * writes on the vnode.  *_wbk will contain the last writeback key used or NULL
293  * and we need to start from there if it's set.
294  */
295 static int afs_get_writeback_key(struct afs_vnode *vnode,
296 				 struct afs_wb_key **_wbk)
297 {
298 	struct afs_wb_key *wbk = NULL;
299 	struct list_head *p;
300 	int ret = -ENOKEY, ret2;
301 
302 	spin_lock(&vnode->wb_lock);
303 	if (*_wbk)
304 		p = (*_wbk)->vnode_link.next;
305 	else
306 		p = vnode->wb_keys.next;
307 
308 	while (p != &vnode->wb_keys) {
309 		wbk = list_entry(p, struct afs_wb_key, vnode_link);
310 		_debug("wbk %u", key_serial(wbk->key));
311 		ret2 = key_validate(wbk->key);
312 		if (ret2 == 0) {
313 			refcount_inc(&wbk->usage);
314 			_debug("USE WB KEY %u", key_serial(wbk->key));
315 			break;
316 		}
317 
318 		wbk = NULL;
319 		if (ret == -ENOKEY)
320 			ret = ret2;
321 		p = p->next;
322 	}
323 
324 	spin_unlock(&vnode->wb_lock);
325 	if (*_wbk)
326 		afs_put_wb_key(*_wbk);
327 	*_wbk = wbk;
328 	return 0;
329 }
330 
331 static void afs_store_data_success(struct afs_operation *op)
332 {
333 	struct afs_vnode *vnode = op->file[0].vnode;
334 
335 	op->ctime = op->file[0].scb.status.mtime_client;
336 	afs_vnode_commit_status(op, &op->file[0]);
337 	if (op->error == 0) {
338 		if (!op->store.laundering)
339 			afs_pages_written_back(vnode, op->store.pos, op->store.size);
340 		afs_stat_v(vnode, n_stores);
341 		atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
342 	}
343 }
344 
345 static const struct afs_operation_ops afs_store_data_operation = {
346 	.issue_afs_rpc	= afs_fs_store_data,
347 	.issue_yfs_rpc	= yfs_fs_store_data,
348 	.success	= afs_store_data_success,
349 };
350 
351 /*
352  * write to a file
353  */
354 static int afs_store_data(struct afs_vnode *vnode, struct iov_iter *iter, loff_t pos,
355 			  bool laundering)
356 {
357 	struct afs_operation *op;
358 	struct afs_wb_key *wbk = NULL;
359 	loff_t size = iov_iter_count(iter), i_size;
360 	int ret = -ENOKEY;
361 
362 	_enter("%s{%llx:%llu.%u},%llx,%llx",
363 	       vnode->volume->name,
364 	       vnode->fid.vid,
365 	       vnode->fid.vnode,
366 	       vnode->fid.unique,
367 	       size, pos);
368 
369 	ret = afs_get_writeback_key(vnode, &wbk);
370 	if (ret) {
371 		_leave(" = %d [no keys]", ret);
372 		return ret;
373 	}
374 
375 	op = afs_alloc_operation(wbk->key, vnode->volume);
376 	if (IS_ERR(op)) {
377 		afs_put_wb_key(wbk);
378 		return -ENOMEM;
379 	}
380 
381 	i_size = i_size_read(&vnode->vfs_inode);
382 
383 	afs_op_set_vnode(op, 0, vnode);
384 	op->file[0].dv_delta = 1;
385 	op->file[0].modification = true;
386 	op->store.write_iter = iter;
387 	op->store.pos = pos;
388 	op->store.size = size;
389 	op->store.i_size = max(pos + size, i_size);
390 	op->store.laundering = laundering;
391 	op->mtime = vnode->vfs_inode.i_mtime;
392 	op->flags |= AFS_OPERATION_UNINTR;
393 	op->ops = &afs_store_data_operation;
394 
395 try_next_key:
396 	afs_begin_vnode_operation(op);
397 	afs_wait_for_operation(op);
398 
399 	switch (op->error) {
400 	case -EACCES:
401 	case -EPERM:
402 	case -ENOKEY:
403 	case -EKEYEXPIRED:
404 	case -EKEYREJECTED:
405 	case -EKEYREVOKED:
406 		_debug("next");
407 
408 		ret = afs_get_writeback_key(vnode, &wbk);
409 		if (ret == 0) {
410 			key_put(op->key);
411 			op->key = key_get(wbk->key);
412 			goto try_next_key;
413 		}
414 		break;
415 	}
416 
417 	afs_put_wb_key(wbk);
418 	_leave(" = %d", op->error);
419 	return afs_put_operation(op);
420 }
421 
422 /*
423  * Extend the region to be written back to include subsequent contiguously
424  * dirty pages if possible, but don't sleep while doing so.
425  *
426  * If this page holds new content, then we can include filler zeros in the
427  * writeback.
428  */
429 static void afs_extend_writeback(struct address_space *mapping,
430 				 struct afs_vnode *vnode,
431 				 long *_count,
432 				 loff_t start,
433 				 loff_t max_len,
434 				 bool new_content,
435 				 bool caching,
436 				 unsigned int *_len)
437 {
438 	struct pagevec pvec;
439 	struct folio *folio;
440 	unsigned long priv;
441 	unsigned int psize, filler = 0;
442 	unsigned int f, t;
443 	loff_t len = *_len;
444 	pgoff_t index = (start + len) / PAGE_SIZE;
445 	bool stop = true;
446 	unsigned int i;
447 
448 	XA_STATE(xas, &mapping->i_pages, index);
449 	pagevec_init(&pvec);
450 
451 	do {
452 		/* Firstly, we gather up a batch of contiguous dirty pages
453 		 * under the RCU read lock - but we can't clear the dirty flags
454 		 * there if any of those pages are mapped.
455 		 */
456 		rcu_read_lock();
457 
458 		xas_for_each(&xas, folio, ULONG_MAX) {
459 			stop = true;
460 			if (xas_retry(&xas, folio))
461 				continue;
462 			if (xa_is_value(folio))
463 				break;
464 			if (folio_index(folio) != index)
465 				break;
466 
467 			if (!folio_try_get_rcu(folio)) {
468 				xas_reset(&xas);
469 				continue;
470 			}
471 
472 			/* Has the page moved or been split? */
473 			if (unlikely(folio != xas_reload(&xas))) {
474 				folio_put(folio);
475 				break;
476 			}
477 
478 			if (!folio_trylock(folio)) {
479 				folio_put(folio);
480 				break;
481 			}
482 			if (!folio_test_dirty(folio) ||
483 			    folio_test_writeback(folio) ||
484 			    folio_test_fscache(folio)) {
485 				folio_unlock(folio);
486 				folio_put(folio);
487 				break;
488 			}
489 
490 			psize = folio_size(folio);
491 			priv = (unsigned long)folio_get_private(folio);
492 			f = afs_folio_dirty_from(folio, priv);
493 			t = afs_folio_dirty_to(folio, priv);
494 			if (f != 0 && !new_content) {
495 				folio_unlock(folio);
496 				folio_put(folio);
497 				break;
498 			}
499 
500 			len += filler + t;
501 			filler = psize - t;
502 			if (len >= max_len || *_count <= 0)
503 				stop = true;
504 			else if (t == psize || new_content)
505 				stop = false;
506 
507 			index += folio_nr_pages(folio);
508 			if (!pagevec_add(&pvec, &folio->page))
509 				break;
510 			if (stop)
511 				break;
512 		}
513 
514 		if (!stop)
515 			xas_pause(&xas);
516 		rcu_read_unlock();
517 
518 		/* Now, if we obtained any pages, we can shift them to being
519 		 * writable and mark them for caching.
520 		 */
521 		if (!pagevec_count(&pvec))
522 			break;
523 
524 		for (i = 0; i < pagevec_count(&pvec); i++) {
525 			folio = page_folio(pvec.pages[i]);
526 			trace_afs_folio_dirty(vnode, tracepoint_string("store+"), folio);
527 
528 			if (!folio_clear_dirty_for_io(folio))
529 				BUG();
530 			if (folio_start_writeback(folio))
531 				BUG();
532 			afs_folio_start_fscache(caching, folio);
533 
534 			*_count -= folio_nr_pages(folio);
535 			folio_unlock(folio);
536 		}
537 
538 		pagevec_release(&pvec);
539 		cond_resched();
540 	} while (!stop);
541 
542 	*_len = len;
543 }
544 
545 /*
546  * Synchronously write back the locked page and any subsequent non-locked dirty
547  * pages.
548  */
549 static ssize_t afs_write_back_from_locked_folio(struct address_space *mapping,
550 						struct writeback_control *wbc,
551 						struct folio *folio,
552 						loff_t start, loff_t end)
553 {
554 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
555 	struct iov_iter iter;
556 	unsigned long priv;
557 	unsigned int offset, to, len, max_len;
558 	loff_t i_size = i_size_read(&vnode->vfs_inode);
559 	bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
560 	bool caching = fscache_cookie_enabled(afs_vnode_cache(vnode));
561 	long count = wbc->nr_to_write;
562 	int ret;
563 
564 	_enter(",%lx,%llx-%llx", folio_index(folio), start, end);
565 
566 	if (folio_start_writeback(folio))
567 		BUG();
568 	afs_folio_start_fscache(caching, folio);
569 
570 	count -= folio_nr_pages(folio);
571 
572 	/* Find all consecutive lockable dirty pages that have contiguous
573 	 * written regions, stopping when we find a page that is not
574 	 * immediately lockable, is not dirty or is missing, or we reach the
575 	 * end of the range.
576 	 */
577 	priv = (unsigned long)folio_get_private(folio);
578 	offset = afs_folio_dirty_from(folio, priv);
579 	to = afs_folio_dirty_to(folio, priv);
580 	trace_afs_folio_dirty(vnode, tracepoint_string("store"), folio);
581 
582 	len = to - offset;
583 	start += offset;
584 	if (start < i_size) {
585 		/* Trim the write to the EOF; the extra data is ignored.  Also
586 		 * put an upper limit on the size of a single storedata op.
587 		 */
588 		max_len = 65536 * 4096;
589 		max_len = min_t(unsigned long long, max_len, end - start + 1);
590 		max_len = min_t(unsigned long long, max_len, i_size - start);
591 
592 		if (len < max_len &&
593 		    (to == folio_size(folio) || new_content))
594 			afs_extend_writeback(mapping, vnode, &count,
595 					     start, max_len, new_content,
596 					     caching, &len);
597 		len = min_t(loff_t, len, max_len);
598 	}
599 
600 	/* We now have a contiguous set of dirty pages, each with writeback
601 	 * set; the first page is still locked at this point, but all the rest
602 	 * have been unlocked.
603 	 */
604 	folio_unlock(folio);
605 
606 	if (start < i_size) {
607 		_debug("write back %x @%llx [%llx]", len, start, i_size);
608 
609 		/* Speculatively write to the cache.  We have to fix this up
610 		 * later if the store fails.
611 		 */
612 		afs_write_to_cache(vnode, start, len, i_size, caching);
613 
614 		iov_iter_xarray(&iter, WRITE, &mapping->i_pages, start, len);
615 		ret = afs_store_data(vnode, &iter, start, false);
616 	} else {
617 		_debug("write discard %x @%llx [%llx]", len, start, i_size);
618 
619 		/* The dirty region was entirely beyond the EOF. */
620 		fscache_clear_page_bits(afs_vnode_cache(vnode),
621 					mapping, start, len, caching);
622 		afs_pages_written_back(vnode, start, len);
623 		ret = 0;
624 	}
625 
626 	switch (ret) {
627 	case 0:
628 		wbc->nr_to_write = count;
629 		ret = len;
630 		break;
631 
632 	default:
633 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
634 		fallthrough;
635 	case -EACCES:
636 	case -EPERM:
637 	case -ENOKEY:
638 	case -EKEYEXPIRED:
639 	case -EKEYREJECTED:
640 	case -EKEYREVOKED:
641 		afs_redirty_pages(wbc, mapping, start, len);
642 		mapping_set_error(mapping, ret);
643 		break;
644 
645 	case -EDQUOT:
646 	case -ENOSPC:
647 		afs_redirty_pages(wbc, mapping, start, len);
648 		mapping_set_error(mapping, -ENOSPC);
649 		break;
650 
651 	case -EROFS:
652 	case -EIO:
653 	case -EREMOTEIO:
654 	case -EFBIG:
655 	case -ENOENT:
656 	case -ENOMEDIUM:
657 	case -ENXIO:
658 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
659 		afs_kill_pages(mapping, start, len);
660 		mapping_set_error(mapping, ret);
661 		break;
662 	}
663 
664 	_leave(" = %d", ret);
665 	return ret;
666 }
667 
668 /*
669  * write a page back to the server
670  * - the caller locked the page for us
671  */
672 int afs_writepage(struct page *subpage, struct writeback_control *wbc)
673 {
674 	struct folio *folio = page_folio(subpage);
675 	ssize_t ret;
676 	loff_t start;
677 
678 	_enter("{%lx},", folio_index(folio));
679 
680 #ifdef CONFIG_AFS_FSCACHE
681 	folio_wait_fscache(folio);
682 #endif
683 
684 	start = folio_index(folio) * PAGE_SIZE;
685 	ret = afs_write_back_from_locked_folio(folio_mapping(folio), wbc,
686 					       folio, start, LLONG_MAX - start);
687 	if (ret < 0) {
688 		_leave(" = %zd", ret);
689 		return ret;
690 	}
691 
692 	_leave(" = 0");
693 	return 0;
694 }
695 
696 /*
697  * write a region of pages back to the server
698  */
699 static int afs_writepages_region(struct address_space *mapping,
700 				 struct writeback_control *wbc,
701 				 loff_t start, loff_t end, loff_t *_next)
702 {
703 	struct folio *folio;
704 	struct page *head_page;
705 	ssize_t ret;
706 	int n, skips = 0;
707 
708 	_enter("%llx,%llx,", start, end);
709 
710 	do {
711 		pgoff_t index = start / PAGE_SIZE;
712 
713 		n = find_get_pages_range_tag(mapping, &index, end / PAGE_SIZE,
714 					     PAGECACHE_TAG_DIRTY, 1, &head_page);
715 		if (!n)
716 			break;
717 
718 		folio = page_folio(head_page);
719 		start = folio_pos(folio); /* May regress with THPs */
720 
721 		_debug("wback %lx", folio_index(folio));
722 
723 		/* At this point we hold neither the i_pages lock nor the
724 		 * page lock: the page may be truncated or invalidated
725 		 * (changing page->mapping to NULL), or even swizzled
726 		 * back from swapper_space to tmpfs file mapping
727 		 */
728 		if (wbc->sync_mode != WB_SYNC_NONE) {
729 			ret = folio_lock_killable(folio);
730 			if (ret < 0) {
731 				folio_put(folio);
732 				return ret;
733 			}
734 		} else {
735 			if (!folio_trylock(folio)) {
736 				folio_put(folio);
737 				return 0;
738 			}
739 		}
740 
741 		if (folio_mapping(folio) != mapping ||
742 		    !folio_test_dirty(folio)) {
743 			start += folio_size(folio);
744 			folio_unlock(folio);
745 			folio_put(folio);
746 			continue;
747 		}
748 
749 		if (folio_test_writeback(folio) ||
750 		    folio_test_fscache(folio)) {
751 			folio_unlock(folio);
752 			if (wbc->sync_mode != WB_SYNC_NONE) {
753 				folio_wait_writeback(folio);
754 #ifdef CONFIG_AFS_FSCACHE
755 				folio_wait_fscache(folio);
756 #endif
757 			} else {
758 				start += folio_size(folio);
759 			}
760 			folio_put(folio);
761 			if (wbc->sync_mode == WB_SYNC_NONE) {
762 				if (skips >= 5 || need_resched())
763 					break;
764 				skips++;
765 			}
766 			continue;
767 		}
768 
769 		if (!folio_clear_dirty_for_io(folio))
770 			BUG();
771 		ret = afs_write_back_from_locked_folio(mapping, wbc, folio, start, end);
772 		folio_put(folio);
773 		if (ret < 0) {
774 			_leave(" = %zd", ret);
775 			return ret;
776 		}
777 
778 		start += ret;
779 
780 		cond_resched();
781 	} while (wbc->nr_to_write > 0);
782 
783 	*_next = start;
784 	_leave(" = 0 [%llx]", *_next);
785 	return 0;
786 }
787 
788 /*
789  * write some of the pending data back to the server
790  */
791 int afs_writepages(struct address_space *mapping,
792 		   struct writeback_control *wbc)
793 {
794 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
795 	loff_t start, next;
796 	int ret;
797 
798 	_enter("");
799 
800 	/* We have to be careful as we can end up racing with setattr()
801 	 * truncating the pagecache since the caller doesn't take a lock here
802 	 * to prevent it.
803 	 */
804 	if (wbc->sync_mode == WB_SYNC_ALL)
805 		down_read(&vnode->validate_lock);
806 	else if (!down_read_trylock(&vnode->validate_lock))
807 		return 0;
808 
809 	if (wbc->range_cyclic) {
810 		start = mapping->writeback_index * PAGE_SIZE;
811 		ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX, &next);
812 		if (ret == 0) {
813 			mapping->writeback_index = next / PAGE_SIZE;
814 			if (start > 0 && wbc->nr_to_write > 0) {
815 				ret = afs_writepages_region(mapping, wbc, 0,
816 							    start, &next);
817 				if (ret == 0)
818 					mapping->writeback_index =
819 						next / PAGE_SIZE;
820 			}
821 		}
822 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
823 		ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX, &next);
824 		if (wbc->nr_to_write > 0 && ret == 0)
825 			mapping->writeback_index = next / PAGE_SIZE;
826 	} else {
827 		ret = afs_writepages_region(mapping, wbc,
828 					    wbc->range_start, wbc->range_end, &next);
829 	}
830 
831 	up_read(&vnode->validate_lock);
832 	_leave(" = %d", ret);
833 	return ret;
834 }
835 
836 /*
837  * write to an AFS file
838  */
839 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
840 {
841 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
842 	struct afs_file *af = iocb->ki_filp->private_data;
843 	ssize_t result;
844 	size_t count = iov_iter_count(from);
845 
846 	_enter("{%llx:%llu},{%zu},",
847 	       vnode->fid.vid, vnode->fid.vnode, count);
848 
849 	if (IS_SWAPFILE(&vnode->vfs_inode)) {
850 		printk(KERN_INFO
851 		       "AFS: Attempt to write to active swap file!\n");
852 		return -EBUSY;
853 	}
854 
855 	if (!count)
856 		return 0;
857 
858 	result = afs_validate(vnode, af->key);
859 	if (result < 0)
860 		return result;
861 
862 	result = generic_file_write_iter(iocb, from);
863 
864 	_leave(" = %zd", result);
865 	return result;
866 }
867 
868 /*
869  * flush any dirty pages for this process, and check for write errors.
870  * - the return status from this call provides a reliable indication of
871  *   whether any write errors occurred for this process.
872  */
873 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
874 {
875 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
876 	struct afs_file *af = file->private_data;
877 	int ret;
878 
879 	_enter("{%llx:%llu},{n=%pD},%d",
880 	       vnode->fid.vid, vnode->fid.vnode, file,
881 	       datasync);
882 
883 	ret = afs_validate(vnode, af->key);
884 	if (ret < 0)
885 		return ret;
886 
887 	return file_write_and_wait_range(file, start, end);
888 }
889 
890 /*
891  * notification that a previously read-only page is about to become writable
892  * - if it returns an error, the caller will deliver a bus error signal
893  */
894 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
895 {
896 	struct folio *folio = page_folio(vmf->page);
897 	struct file *file = vmf->vma->vm_file;
898 	struct inode *inode = file_inode(file);
899 	struct afs_vnode *vnode = AFS_FS_I(inode);
900 	struct afs_file *af = file->private_data;
901 	unsigned long priv;
902 	vm_fault_t ret = VM_FAULT_RETRY;
903 
904 	_enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, folio_index(folio));
905 
906 	afs_validate(vnode, af->key);
907 
908 	sb_start_pagefault(inode->i_sb);
909 
910 	/* Wait for the page to be written to the cache before we allow it to
911 	 * be modified.  We then assume the entire page will need writing back.
912 	 */
913 #ifdef CONFIG_AFS_FSCACHE
914 	if (folio_test_fscache(folio) &&
915 	    folio_wait_fscache_killable(folio) < 0)
916 		goto out;
917 #endif
918 
919 	if (folio_wait_writeback_killable(folio))
920 		goto out;
921 
922 	if (folio_lock_killable(folio) < 0)
923 		goto out;
924 
925 	/* We mustn't change folio->private until writeback is complete as that
926 	 * details the portion of the page we need to write back and we might
927 	 * need to redirty the page if there's a problem.
928 	 */
929 	if (folio_wait_writeback_killable(folio) < 0) {
930 		folio_unlock(folio);
931 		goto out;
932 	}
933 
934 	priv = afs_folio_dirty(folio, 0, folio_size(folio));
935 	priv = afs_folio_dirty_mmapped(priv);
936 	if (folio_test_private(folio)) {
937 		folio_change_private(folio, (void *)priv);
938 		trace_afs_folio_dirty(vnode, tracepoint_string("mkwrite+"), folio);
939 	} else {
940 		folio_attach_private(folio, (void *)priv);
941 		trace_afs_folio_dirty(vnode, tracepoint_string("mkwrite"), folio);
942 	}
943 	file_update_time(file);
944 
945 	ret = VM_FAULT_LOCKED;
946 out:
947 	sb_end_pagefault(inode->i_sb);
948 	return ret;
949 }
950 
951 /*
952  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
953  */
954 void afs_prune_wb_keys(struct afs_vnode *vnode)
955 {
956 	LIST_HEAD(graveyard);
957 	struct afs_wb_key *wbk, *tmp;
958 
959 	/* Discard unused keys */
960 	spin_lock(&vnode->wb_lock);
961 
962 	if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
963 	    !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
964 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
965 			if (refcount_read(&wbk->usage) == 1)
966 				list_move(&wbk->vnode_link, &graveyard);
967 		}
968 	}
969 
970 	spin_unlock(&vnode->wb_lock);
971 
972 	while (!list_empty(&graveyard)) {
973 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
974 		list_del(&wbk->vnode_link);
975 		afs_put_wb_key(wbk);
976 	}
977 }
978 
979 /*
980  * Clean up a page during invalidation.
981  */
982 int afs_launder_page(struct page *subpage)
983 {
984 	struct folio *folio = page_folio(subpage);
985 	struct afs_vnode *vnode = AFS_FS_I(folio_inode(folio));
986 	struct iov_iter iter;
987 	struct bio_vec bv[1];
988 	unsigned long priv;
989 	unsigned int f, t;
990 	int ret = 0;
991 
992 	_enter("{%lx}", folio_index(folio));
993 
994 	priv = (unsigned long)folio_get_private(folio);
995 	if (folio_clear_dirty_for_io(folio)) {
996 		f = 0;
997 		t = folio_size(folio);
998 		if (folio_test_private(folio)) {
999 			f = afs_folio_dirty_from(folio, priv);
1000 			t = afs_folio_dirty_to(folio, priv);
1001 		}
1002 
1003 		bv[0].bv_page = &folio->page;
1004 		bv[0].bv_offset = f;
1005 		bv[0].bv_len = t - f;
1006 		iov_iter_bvec(&iter, WRITE, bv, 1, bv[0].bv_len);
1007 
1008 		trace_afs_folio_dirty(vnode, tracepoint_string("launder"), folio);
1009 		ret = afs_store_data(vnode, &iter, folio_pos(folio) + f, true);
1010 	}
1011 
1012 	trace_afs_folio_dirty(vnode, tracepoint_string("laundered"), folio);
1013 	folio_detach_private(folio);
1014 	folio_wait_fscache(folio);
1015 	return ret;
1016 }
1017 
1018 /*
1019  * Deal with the completion of writing the data to the cache.
1020  */
1021 static void afs_write_to_cache_done(void *priv, ssize_t transferred_or_error,
1022 				    bool was_async)
1023 {
1024 	struct afs_vnode *vnode = priv;
1025 
1026 	if (IS_ERR_VALUE(transferred_or_error) &&
1027 	    transferred_or_error != -ENOBUFS)
1028 		afs_invalidate_cache(vnode, 0);
1029 }
1030 
1031 /*
1032  * Save the write to the cache also.
1033  */
1034 static void afs_write_to_cache(struct afs_vnode *vnode,
1035 			       loff_t start, size_t len, loff_t i_size,
1036 			       bool caching)
1037 {
1038 	fscache_write_to_cache(afs_vnode_cache(vnode),
1039 			       vnode->vfs_inode.i_mapping, start, len, i_size,
1040 			       afs_write_to_cache_done, vnode, caching);
1041 }
1042