xref: /openbmc/linux/fs/afs/write.c (revision 00ae4ebc)
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 (copied == 0)
122 		goto out;
123 
124 	maybe_i_size = pos + copied;
125 
126 	i_size = i_size_read(&vnode->vfs_inode);
127 	if (maybe_i_size > i_size) {
128 		write_seqlock(&vnode->cb_lock);
129 		i_size = i_size_read(&vnode->vfs_inode);
130 		if (maybe_i_size > i_size)
131 			i_size_write(&vnode->vfs_inode, maybe_i_size);
132 		write_sequnlock(&vnode->cb_lock);
133 	}
134 
135 	ASSERT(PageUptodate(page));
136 
137 	if (PagePrivate(page)) {
138 		priv = page_private(page);
139 		f = afs_page_dirty_from(page, priv);
140 		t = afs_page_dirty_to(page, priv);
141 		if (from < f)
142 			f = from;
143 		if (to > t)
144 			t = to;
145 		priv = afs_page_dirty(page, f, t);
146 		set_page_private(page, priv);
147 		trace_afs_page_dirty(vnode, tracepoint_string("dirty+"), page);
148 	} else {
149 		priv = afs_page_dirty(page, from, to);
150 		attach_page_private(page, (void *)priv);
151 		trace_afs_page_dirty(vnode, tracepoint_string("dirty"), page);
152 	}
153 
154 	if (set_page_dirty(page))
155 		_debug("dirtied %lx", page->index);
156 
157 out:
158 	unlock_page(page);
159 	put_page(page);
160 	return copied;
161 }
162 
163 /*
164  * kill all the pages in the given range
165  */
166 static void afs_kill_pages(struct address_space *mapping,
167 			   loff_t start, loff_t len)
168 {
169 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
170 	struct pagevec pv;
171 	unsigned int loop, psize;
172 
173 	_enter("{%llx:%llu},%llx @%llx",
174 	       vnode->fid.vid, vnode->fid.vnode, len, start);
175 
176 	pagevec_init(&pv);
177 
178 	do {
179 		_debug("kill %llx @%llx", len, start);
180 
181 		pv.nr = find_get_pages_contig(mapping, start / PAGE_SIZE,
182 					      PAGEVEC_SIZE, pv.pages);
183 		if (pv.nr == 0)
184 			break;
185 
186 		for (loop = 0; loop < pv.nr; loop++) {
187 			struct page *page = pv.pages[loop];
188 
189 			if (page->index * PAGE_SIZE >= start + len)
190 				break;
191 
192 			psize = thp_size(page);
193 			start += psize;
194 			len -= psize;
195 			ClearPageUptodate(page);
196 			end_page_writeback(page);
197 			lock_page(page);
198 			generic_error_remove_page(mapping, page);
199 			unlock_page(page);
200 		}
201 
202 		__pagevec_release(&pv);
203 	} while (len > 0);
204 
205 	_leave("");
206 }
207 
208 /*
209  * Redirty all the pages in a given range.
210  */
211 static void afs_redirty_pages(struct writeback_control *wbc,
212 			      struct address_space *mapping,
213 			      loff_t start, loff_t len)
214 {
215 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
216 	struct pagevec pv;
217 	unsigned int loop, psize;
218 
219 	_enter("{%llx:%llu},%llx @%llx",
220 	       vnode->fid.vid, vnode->fid.vnode, len, start);
221 
222 	pagevec_init(&pv);
223 
224 	do {
225 		_debug("redirty %llx @%llx", len, start);
226 
227 		pv.nr = find_get_pages_contig(mapping, start / PAGE_SIZE,
228 					      PAGEVEC_SIZE, pv.pages);
229 		if (pv.nr == 0)
230 			break;
231 
232 		for (loop = 0; loop < pv.nr; loop++) {
233 			struct page *page = pv.pages[loop];
234 
235 			if (page->index * PAGE_SIZE >= start + len)
236 				break;
237 
238 			psize = thp_size(page);
239 			start += psize;
240 			len -= psize;
241 			redirty_page_for_writepage(wbc, page);
242 			end_page_writeback(page);
243 		}
244 
245 		__pagevec_release(&pv);
246 	} while (len > 0);
247 
248 	_leave("");
249 }
250 
251 /*
252  * completion of write to server
253  */
254 static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
255 {
256 	struct address_space *mapping = vnode->vfs_inode.i_mapping;
257 	struct page *page;
258 	pgoff_t end;
259 
260 	XA_STATE(xas, &mapping->i_pages, start / PAGE_SIZE);
261 
262 	_enter("{%llx:%llu},{%x @%llx}",
263 	       vnode->fid.vid, vnode->fid.vnode, len, start);
264 
265 	rcu_read_lock();
266 
267 	end = (start + len - 1) / PAGE_SIZE;
268 	xas_for_each(&xas, page, end) {
269 		if (!PageWriteback(page)) {
270 			kdebug("bad %x @%llx page %lx %lx", len, start, page->index, end);
271 			ASSERT(PageWriteback(page));
272 		}
273 
274 		trace_afs_page_dirty(vnode, tracepoint_string("clear"), page);
275 		detach_page_private(page);
276 		page_endio(page, true, 0);
277 	}
278 
279 	rcu_read_unlock();
280 
281 	afs_prune_wb_keys(vnode);
282 	_leave("");
283 }
284 
285 /*
286  * Find a key to use for the writeback.  We cached the keys used to author the
287  * writes on the vnode.  *_wbk will contain the last writeback key used or NULL
288  * and we need to start from there if it's set.
289  */
290 static int afs_get_writeback_key(struct afs_vnode *vnode,
291 				 struct afs_wb_key **_wbk)
292 {
293 	struct afs_wb_key *wbk = NULL;
294 	struct list_head *p;
295 	int ret = -ENOKEY, ret2;
296 
297 	spin_lock(&vnode->wb_lock);
298 	if (*_wbk)
299 		p = (*_wbk)->vnode_link.next;
300 	else
301 		p = vnode->wb_keys.next;
302 
303 	while (p != &vnode->wb_keys) {
304 		wbk = list_entry(p, struct afs_wb_key, vnode_link);
305 		_debug("wbk %u", key_serial(wbk->key));
306 		ret2 = key_validate(wbk->key);
307 		if (ret2 == 0) {
308 			refcount_inc(&wbk->usage);
309 			_debug("USE WB KEY %u", key_serial(wbk->key));
310 			break;
311 		}
312 
313 		wbk = NULL;
314 		if (ret == -ENOKEY)
315 			ret = ret2;
316 		p = p->next;
317 	}
318 
319 	spin_unlock(&vnode->wb_lock);
320 	if (*_wbk)
321 		afs_put_wb_key(*_wbk);
322 	*_wbk = wbk;
323 	return 0;
324 }
325 
326 static void afs_store_data_success(struct afs_operation *op)
327 {
328 	struct afs_vnode *vnode = op->file[0].vnode;
329 
330 	op->ctime = op->file[0].scb.status.mtime_client;
331 	afs_vnode_commit_status(op, &op->file[0]);
332 	if (op->error == 0) {
333 		if (!op->store.laundering)
334 			afs_pages_written_back(vnode, op->store.pos, op->store.size);
335 		afs_stat_v(vnode, n_stores);
336 		atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
337 	}
338 }
339 
340 static const struct afs_operation_ops afs_store_data_operation = {
341 	.issue_afs_rpc	= afs_fs_store_data,
342 	.issue_yfs_rpc	= yfs_fs_store_data,
343 	.success	= afs_store_data_success,
344 };
345 
346 /*
347  * write to a file
348  */
349 static int afs_store_data(struct afs_vnode *vnode, struct iov_iter *iter, loff_t pos,
350 			  bool laundering)
351 {
352 	struct afs_operation *op;
353 	struct afs_wb_key *wbk = NULL;
354 	loff_t size = iov_iter_count(iter), i_size;
355 	int ret = -ENOKEY;
356 
357 	_enter("%s{%llx:%llu.%u},%llx,%llx",
358 	       vnode->volume->name,
359 	       vnode->fid.vid,
360 	       vnode->fid.vnode,
361 	       vnode->fid.unique,
362 	       size, pos);
363 
364 	ret = afs_get_writeback_key(vnode, &wbk);
365 	if (ret) {
366 		_leave(" = %d [no keys]", ret);
367 		return ret;
368 	}
369 
370 	op = afs_alloc_operation(wbk->key, vnode->volume);
371 	if (IS_ERR(op)) {
372 		afs_put_wb_key(wbk);
373 		return -ENOMEM;
374 	}
375 
376 	i_size = i_size_read(&vnode->vfs_inode);
377 
378 	afs_op_set_vnode(op, 0, vnode);
379 	op->file[0].dv_delta = 1;
380 	op->file[0].modification = true;
381 	op->store.write_iter = iter;
382 	op->store.pos = pos;
383 	op->store.size = size;
384 	op->store.i_size = max(pos + size, i_size);
385 	op->store.laundering = laundering;
386 	op->mtime = vnode->vfs_inode.i_mtime;
387 	op->flags |= AFS_OPERATION_UNINTR;
388 	op->ops = &afs_store_data_operation;
389 
390 try_next_key:
391 	afs_begin_vnode_operation(op);
392 	afs_wait_for_operation(op);
393 
394 	switch (op->error) {
395 	case -EACCES:
396 	case -EPERM:
397 	case -ENOKEY:
398 	case -EKEYEXPIRED:
399 	case -EKEYREJECTED:
400 	case -EKEYREVOKED:
401 		_debug("next");
402 
403 		ret = afs_get_writeback_key(vnode, &wbk);
404 		if (ret == 0) {
405 			key_put(op->key);
406 			op->key = key_get(wbk->key);
407 			goto try_next_key;
408 		}
409 		break;
410 	}
411 
412 	afs_put_wb_key(wbk);
413 	_leave(" = %d", op->error);
414 	return afs_put_operation(op);
415 }
416 
417 /*
418  * Extend the region to be written back to include subsequent contiguously
419  * dirty pages if possible, but don't sleep while doing so.
420  *
421  * If this page holds new content, then we can include filler zeros in the
422  * writeback.
423  */
424 static void afs_extend_writeback(struct address_space *mapping,
425 				 struct afs_vnode *vnode,
426 				 long *_count,
427 				 loff_t start,
428 				 loff_t max_len,
429 				 bool new_content,
430 				 unsigned int *_len)
431 {
432 	struct pagevec pvec;
433 	struct page *page;
434 	unsigned long priv;
435 	unsigned int psize, filler = 0;
436 	unsigned int f, t;
437 	loff_t len = *_len;
438 	pgoff_t index = (start + len) / PAGE_SIZE;
439 	bool stop = true;
440 	unsigned int i;
441 
442 	XA_STATE(xas, &mapping->i_pages, index);
443 	pagevec_init(&pvec);
444 
445 	do {
446 		/* Firstly, we gather up a batch of contiguous dirty pages
447 		 * under the RCU read lock - but we can't clear the dirty flags
448 		 * there if any of those pages are mapped.
449 		 */
450 		rcu_read_lock();
451 
452 		xas_for_each(&xas, page, ULONG_MAX) {
453 			stop = true;
454 			if (xas_retry(&xas, page))
455 				continue;
456 			if (xa_is_value(page))
457 				break;
458 			if (page->index != index)
459 				break;
460 
461 			if (!page_cache_get_speculative(page)) {
462 				xas_reset(&xas);
463 				continue;
464 			}
465 
466 			/* Has the page moved or been split? */
467 			if (unlikely(page != xas_reload(&xas)))
468 				break;
469 
470 			if (!trylock_page(page))
471 				break;
472 			if (!PageDirty(page) || PageWriteback(page)) {
473 				unlock_page(page);
474 				break;
475 			}
476 
477 			psize = thp_size(page);
478 			priv = page_private(page);
479 			f = afs_page_dirty_from(page, priv);
480 			t = afs_page_dirty_to(page, priv);
481 			if (f != 0 && !new_content) {
482 				unlock_page(page);
483 				break;
484 			}
485 
486 			len += filler + t;
487 			filler = psize - t;
488 			if (len >= max_len || *_count <= 0)
489 				stop = true;
490 			else if (t == psize || new_content)
491 				stop = false;
492 
493 			index += thp_nr_pages(page);
494 			if (!pagevec_add(&pvec, page))
495 				break;
496 			if (stop)
497 				break;
498 		}
499 
500 		if (!stop)
501 			xas_pause(&xas);
502 		rcu_read_unlock();
503 
504 		/* Now, if we obtained any pages, we can shift them to being
505 		 * writable and mark them for caching.
506 		 */
507 		if (!pagevec_count(&pvec))
508 			break;
509 
510 		for (i = 0; i < pagevec_count(&pvec); i++) {
511 			page = pvec.pages[i];
512 			trace_afs_page_dirty(vnode, tracepoint_string("store+"), page);
513 
514 			if (!clear_page_dirty_for_io(page))
515 				BUG();
516 			if (test_set_page_writeback(page))
517 				BUG();
518 
519 			*_count -= thp_nr_pages(page);
520 			unlock_page(page);
521 		}
522 
523 		pagevec_release(&pvec);
524 		cond_resched();
525 	} while (!stop);
526 
527 	*_len = len;
528 }
529 
530 /*
531  * Synchronously write back the locked page and any subsequent non-locked dirty
532  * pages.
533  */
534 static ssize_t afs_write_back_from_locked_page(struct address_space *mapping,
535 					       struct writeback_control *wbc,
536 					       struct page *page,
537 					       loff_t start, loff_t end)
538 {
539 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
540 	struct iov_iter iter;
541 	unsigned long priv;
542 	unsigned int offset, to, len, max_len;
543 	loff_t i_size = i_size_read(&vnode->vfs_inode);
544 	bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
545 	long count = wbc->nr_to_write;
546 	int ret;
547 
548 	_enter(",%lx,%llx-%llx", page->index, start, end);
549 
550 	if (test_set_page_writeback(page))
551 		BUG();
552 
553 	count -= thp_nr_pages(page);
554 
555 	/* Find all consecutive lockable dirty pages that have contiguous
556 	 * written regions, stopping when we find a page that is not
557 	 * immediately lockable, is not dirty or is missing, or we reach the
558 	 * end of the range.
559 	 */
560 	priv = page_private(page);
561 	offset = afs_page_dirty_from(page, priv);
562 	to = afs_page_dirty_to(page, priv);
563 	trace_afs_page_dirty(vnode, tracepoint_string("store"), page);
564 
565 	len = to - offset;
566 	start += offset;
567 	if (start < i_size) {
568 		/* Trim the write to the EOF; the extra data is ignored.  Also
569 		 * put an upper limit on the size of a single storedata op.
570 		 */
571 		max_len = 65536 * 4096;
572 		max_len = min_t(unsigned long long, max_len, end - start + 1);
573 		max_len = min_t(unsigned long long, max_len, i_size - start);
574 
575 		if (len < max_len &&
576 		    (to == thp_size(page) || new_content))
577 			afs_extend_writeback(mapping, vnode, &count,
578 					     start, max_len, new_content, &len);
579 		len = min_t(loff_t, len, max_len);
580 	}
581 
582 	/* We now have a contiguous set of dirty pages, each with writeback
583 	 * set; the first page is still locked at this point, but all the rest
584 	 * have been unlocked.
585 	 */
586 	unlock_page(page);
587 
588 	if (start < i_size) {
589 		_debug("write back %x @%llx [%llx]", len, start, i_size);
590 
591 		iov_iter_xarray(&iter, WRITE, &mapping->i_pages, start, len);
592 		ret = afs_store_data(vnode, &iter, start, false);
593 	} else {
594 		_debug("write discard %x @%llx [%llx]", len, start, i_size);
595 
596 		/* The dirty region was entirely beyond the EOF. */
597 		afs_pages_written_back(vnode, start, len);
598 		ret = 0;
599 	}
600 
601 	switch (ret) {
602 	case 0:
603 		wbc->nr_to_write = count;
604 		ret = len;
605 		break;
606 
607 	default:
608 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
609 		fallthrough;
610 	case -EACCES:
611 	case -EPERM:
612 	case -ENOKEY:
613 	case -EKEYEXPIRED:
614 	case -EKEYREJECTED:
615 	case -EKEYREVOKED:
616 		afs_redirty_pages(wbc, mapping, start, len);
617 		mapping_set_error(mapping, ret);
618 		break;
619 
620 	case -EDQUOT:
621 	case -ENOSPC:
622 		afs_redirty_pages(wbc, mapping, start, len);
623 		mapping_set_error(mapping, -ENOSPC);
624 		break;
625 
626 	case -EROFS:
627 	case -EIO:
628 	case -EREMOTEIO:
629 	case -EFBIG:
630 	case -ENOENT:
631 	case -ENOMEDIUM:
632 	case -ENXIO:
633 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
634 		afs_kill_pages(mapping, start, len);
635 		mapping_set_error(mapping, ret);
636 		break;
637 	}
638 
639 	_leave(" = %d", ret);
640 	return ret;
641 }
642 
643 /*
644  * write a page back to the server
645  * - the caller locked the page for us
646  */
647 int afs_writepage(struct page *page, struct writeback_control *wbc)
648 {
649 	ssize_t ret;
650 	loff_t start;
651 
652 	_enter("{%lx},", page->index);
653 
654 	start = page->index * PAGE_SIZE;
655 	ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
656 					      start, LLONG_MAX - start);
657 	if (ret < 0) {
658 		_leave(" = %zd", ret);
659 		return ret;
660 	}
661 
662 	_leave(" = 0");
663 	return 0;
664 }
665 
666 /*
667  * write a region of pages back to the server
668  */
669 static int afs_writepages_region(struct address_space *mapping,
670 				 struct writeback_control *wbc,
671 				 loff_t start, loff_t end, loff_t *_next)
672 {
673 	struct page *page;
674 	ssize_t ret;
675 	int n;
676 
677 	_enter("%llx,%llx,", start, end);
678 
679 	do {
680 		pgoff_t index = start / PAGE_SIZE;
681 
682 		n = find_get_pages_range_tag(mapping, &index, end / PAGE_SIZE,
683 					     PAGECACHE_TAG_DIRTY, 1, &page);
684 		if (!n)
685 			break;
686 
687 		start = (loff_t)page->index * PAGE_SIZE; /* May regress with THPs */
688 
689 		_debug("wback %lx", page->index);
690 
691 		/* At this point we hold neither the i_pages lock nor the
692 		 * page lock: the page may be truncated or invalidated
693 		 * (changing page->mapping to NULL), or even swizzled
694 		 * back from swapper_space to tmpfs file mapping
695 		 */
696 		if (wbc->sync_mode != WB_SYNC_NONE) {
697 			ret = lock_page_killable(page);
698 			if (ret < 0) {
699 				put_page(page);
700 				return ret;
701 			}
702 		} else {
703 			if (!trylock_page(page)) {
704 				put_page(page);
705 				return 0;
706 			}
707 		}
708 
709 		if (page->mapping != mapping || !PageDirty(page)) {
710 			start += thp_size(page);
711 			unlock_page(page);
712 			put_page(page);
713 			continue;
714 		}
715 
716 		if (PageWriteback(page)) {
717 			unlock_page(page);
718 			if (wbc->sync_mode != WB_SYNC_NONE)
719 				wait_on_page_writeback(page);
720 			put_page(page);
721 			continue;
722 		}
723 
724 		if (!clear_page_dirty_for_io(page))
725 			BUG();
726 		ret = afs_write_back_from_locked_page(mapping, wbc, page, start, end);
727 		put_page(page);
728 		if (ret < 0) {
729 			_leave(" = %zd", ret);
730 			return ret;
731 		}
732 
733 		start += ret * PAGE_SIZE;
734 
735 		cond_resched();
736 	} while (wbc->nr_to_write > 0);
737 
738 	*_next = start;
739 	_leave(" = 0 [%llx]", *_next);
740 	return 0;
741 }
742 
743 /*
744  * write some of the pending data back to the server
745  */
746 int afs_writepages(struct address_space *mapping,
747 		   struct writeback_control *wbc)
748 {
749 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
750 	loff_t start, next;
751 	int ret;
752 
753 	_enter("");
754 
755 	/* We have to be careful as we can end up racing with setattr()
756 	 * truncating the pagecache since the caller doesn't take a lock here
757 	 * to prevent it.
758 	 */
759 	if (wbc->sync_mode == WB_SYNC_ALL)
760 		down_read(&vnode->validate_lock);
761 	else if (!down_read_trylock(&vnode->validate_lock))
762 		return 0;
763 
764 	if (wbc->range_cyclic) {
765 		start = mapping->writeback_index * PAGE_SIZE;
766 		ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX, &next);
767 		if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
768 			ret = afs_writepages_region(mapping, wbc, 0, start,
769 						    &next);
770 		mapping->writeback_index = next / PAGE_SIZE;
771 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
772 		ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX, &next);
773 		if (wbc->nr_to_write > 0)
774 			mapping->writeback_index = next;
775 	} else {
776 		ret = afs_writepages_region(mapping, wbc,
777 					    wbc->range_start, wbc->range_end, &next);
778 	}
779 
780 	up_read(&vnode->validate_lock);
781 	_leave(" = %d", ret);
782 	return ret;
783 }
784 
785 /*
786  * write to an AFS file
787  */
788 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
789 {
790 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
791 	ssize_t result;
792 	size_t count = iov_iter_count(from);
793 
794 	_enter("{%llx:%llu},{%zu},",
795 	       vnode->fid.vid, vnode->fid.vnode, count);
796 
797 	if (IS_SWAPFILE(&vnode->vfs_inode)) {
798 		printk(KERN_INFO
799 		       "AFS: Attempt to write to active swap file!\n");
800 		return -EBUSY;
801 	}
802 
803 	if (!count)
804 		return 0;
805 
806 	result = generic_file_write_iter(iocb, from);
807 
808 	_leave(" = %zd", result);
809 	return result;
810 }
811 
812 /*
813  * flush any dirty pages for this process, and check for write errors.
814  * - the return status from this call provides a reliable indication of
815  *   whether any write errors occurred for this process.
816  */
817 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
818 {
819 	struct inode *inode = file_inode(file);
820 	struct afs_vnode *vnode = AFS_FS_I(inode);
821 
822 	_enter("{%llx:%llu},{n=%pD},%d",
823 	       vnode->fid.vid, vnode->fid.vnode, file,
824 	       datasync);
825 
826 	return file_write_and_wait_range(file, start, end);
827 }
828 
829 /*
830  * notification that a previously read-only page is about to become writable
831  * - if it returns an error, the caller will deliver a bus error signal
832  */
833 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
834 {
835 	struct page *page = thp_head(vmf->page);
836 	struct file *file = vmf->vma->vm_file;
837 	struct inode *inode = file_inode(file);
838 	struct afs_vnode *vnode = AFS_FS_I(inode);
839 	unsigned long priv;
840 
841 	_enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, page->index);
842 
843 	sb_start_pagefault(inode->i_sb);
844 
845 	/* Wait for the page to be written to the cache before we allow it to
846 	 * be modified.  We then assume the entire page will need writing back.
847 	 */
848 #ifdef CONFIG_AFS_FSCACHE
849 	if (PageFsCache(page) &&
850 	    wait_on_page_fscache_killable(page) < 0)
851 		return VM_FAULT_RETRY;
852 #endif
853 
854 	if (wait_on_page_writeback_killable(page))
855 		return VM_FAULT_RETRY;
856 
857 	if (lock_page_killable(page) < 0)
858 		return VM_FAULT_RETRY;
859 
860 	/* We mustn't change page->private until writeback is complete as that
861 	 * details the portion of the page we need to write back and we might
862 	 * need to redirty the page if there's a problem.
863 	 */
864 	if (wait_on_page_writeback_killable(page) < 0) {
865 		unlock_page(page);
866 		return VM_FAULT_RETRY;
867 	}
868 
869 	priv = afs_page_dirty(page, 0, thp_size(page));
870 	priv = afs_page_dirty_mmapped(priv);
871 	if (PagePrivate(page)) {
872 		set_page_private(page, priv);
873 		trace_afs_page_dirty(vnode, tracepoint_string("mkwrite+"), page);
874 	} else {
875 		attach_page_private(page, (void *)priv);
876 		trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"), page);
877 	}
878 	file_update_time(file);
879 
880 	sb_end_pagefault(inode->i_sb);
881 	return VM_FAULT_LOCKED;
882 }
883 
884 /*
885  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
886  */
887 void afs_prune_wb_keys(struct afs_vnode *vnode)
888 {
889 	LIST_HEAD(graveyard);
890 	struct afs_wb_key *wbk, *tmp;
891 
892 	/* Discard unused keys */
893 	spin_lock(&vnode->wb_lock);
894 
895 	if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
896 	    !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
897 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
898 			if (refcount_read(&wbk->usage) == 1)
899 				list_move(&wbk->vnode_link, &graveyard);
900 		}
901 	}
902 
903 	spin_unlock(&vnode->wb_lock);
904 
905 	while (!list_empty(&graveyard)) {
906 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
907 		list_del(&wbk->vnode_link);
908 		afs_put_wb_key(wbk);
909 	}
910 }
911 
912 /*
913  * Clean up a page during invalidation.
914  */
915 int afs_launder_page(struct page *page)
916 {
917 	struct address_space *mapping = page->mapping;
918 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
919 	struct iov_iter iter;
920 	struct bio_vec bv[1];
921 	unsigned long priv;
922 	unsigned int f, t;
923 	int ret = 0;
924 
925 	_enter("{%lx}", page->index);
926 
927 	priv = page_private(page);
928 	if (clear_page_dirty_for_io(page)) {
929 		f = 0;
930 		t = thp_size(page);
931 		if (PagePrivate(page)) {
932 			f = afs_page_dirty_from(page, priv);
933 			t = afs_page_dirty_to(page, priv);
934 		}
935 
936 		bv[0].bv_page = page;
937 		bv[0].bv_offset = f;
938 		bv[0].bv_len = t - f;
939 		iov_iter_bvec(&iter, WRITE, bv, 1, bv[0].bv_len);
940 
941 		trace_afs_page_dirty(vnode, tracepoint_string("launder"), page);
942 		ret = afs_store_data(vnode, &iter, (loff_t)page->index * PAGE_SIZE,
943 				     true);
944 	}
945 
946 	trace_afs_page_dirty(vnode, tracepoint_string("laundered"), page);
947 	detach_page_private(page);
948 	wait_on_page_fscache(page);
949 	return ret;
950 }
951