xref: /openbmc/linux/fs/afs/write.c (revision 752beb5e)
1 /* handling of writes to regular files and writing back to the server
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/backing-dev.h>
13 #include <linux/slab.h>
14 #include <linux/fs.h>
15 #include <linux/pagemap.h>
16 #include <linux/writeback.h>
17 #include <linux/pagevec.h>
18 #include "internal.h"
19 
20 /*
21  * mark a page as having been made dirty and thus needing writeback
22  */
23 int afs_set_page_dirty(struct page *page)
24 {
25 	_enter("");
26 	return __set_page_dirty_nobuffers(page);
27 }
28 
29 /*
30  * partly or wholly fill a page that's under preparation for writing
31  */
32 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
33 			 loff_t pos, unsigned int len, struct page *page)
34 {
35 	struct afs_read *req;
36 	size_t p;
37 	void *data;
38 	int ret;
39 
40 	_enter(",,%llu", (unsigned long long)pos);
41 
42 	if (pos >= vnode->vfs_inode.i_size) {
43 		p = pos & ~PAGE_MASK;
44 		ASSERTCMP(p + len, <=, PAGE_SIZE);
45 		data = kmap(page);
46 		memset(data + p, 0, len);
47 		kunmap(page);
48 		return 0;
49 	}
50 
51 	req = kzalloc(sizeof(struct afs_read) + sizeof(struct page *),
52 		      GFP_KERNEL);
53 	if (!req)
54 		return -ENOMEM;
55 
56 	refcount_set(&req->usage, 1);
57 	req->pos = pos;
58 	req->len = len;
59 	req->nr_pages = 1;
60 	req->pages = req->array;
61 	req->pages[0] = page;
62 	get_page(page);
63 
64 	ret = afs_fetch_data(vnode, key, req);
65 	afs_put_read(req);
66 	if (ret < 0) {
67 		if (ret == -ENOENT) {
68 			_debug("got NOENT from server"
69 			       " - marking file deleted and stale");
70 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
71 			ret = -ESTALE;
72 		}
73 	}
74 
75 	_leave(" = %d", ret);
76 	return ret;
77 }
78 
79 /*
80  * prepare to perform part of a write to a page
81  */
82 int afs_write_begin(struct file *file, struct address_space *mapping,
83 		    loff_t pos, unsigned len, unsigned flags,
84 		    struct page **pagep, void **fsdata)
85 {
86 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
87 	struct page *page;
88 	struct key *key = afs_file_key(file);
89 	unsigned long priv;
90 	unsigned f, from = pos & (PAGE_SIZE - 1);
91 	unsigned t, to = from + len;
92 	pgoff_t index = pos >> PAGE_SHIFT;
93 	int ret;
94 
95 	_enter("{%llx:%llu},{%lx},%u,%u",
96 	       vnode->fid.vid, vnode->fid.vnode, index, from, to);
97 
98 	/* We want to store information about how much of a page is altered in
99 	 * page->private.
100 	 */
101 	BUILD_BUG_ON(PAGE_SIZE > 32768 && sizeof(page->private) < 8);
102 
103 	page = grab_cache_page_write_begin(mapping, index, flags);
104 	if (!page)
105 		return -ENOMEM;
106 
107 	if (!PageUptodate(page) && len != PAGE_SIZE) {
108 		ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
109 		if (ret < 0) {
110 			unlock_page(page);
111 			put_page(page);
112 			_leave(" = %d [prep]", ret);
113 			return ret;
114 		}
115 		SetPageUptodate(page);
116 	}
117 
118 	/* page won't leak in error case: it eventually gets cleaned off LRU */
119 	*pagep = page;
120 
121 try_again:
122 	/* See if this page is already partially written in a way that we can
123 	 * merge the new write with.
124 	 */
125 	t = f = 0;
126 	if (PagePrivate(page)) {
127 		priv = page_private(page);
128 		f = priv & AFS_PRIV_MAX;
129 		t = priv >> AFS_PRIV_SHIFT;
130 		ASSERTCMP(f, <=, t);
131 	}
132 
133 	if (f != t) {
134 		if (PageWriteback(page)) {
135 			trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
136 					     page->index, priv);
137 			goto flush_conflicting_write;
138 		}
139 		/* If the file is being filled locally, allow inter-write
140 		 * spaces to be merged into writes.  If it's not, only write
141 		 * back what the user gives us.
142 		 */
143 		if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
144 		    (to < f || from > t))
145 			goto flush_conflicting_write;
146 		if (from < f)
147 			f = from;
148 		if (to > t)
149 			t = to;
150 	} else {
151 		f = from;
152 		t = to;
153 	}
154 
155 	priv = (unsigned long)t << AFS_PRIV_SHIFT;
156 	priv |= f;
157 	trace_afs_page_dirty(vnode, tracepoint_string("begin"),
158 			     page->index, priv);
159 	SetPagePrivate(page);
160 	set_page_private(page, priv);
161 	_leave(" = 0");
162 	return 0;
163 
164 	/* The previous write and this write aren't adjacent or overlapping, so
165 	 * flush the page out.
166 	 */
167 flush_conflicting_write:
168 	_debug("flush conflict");
169 	ret = write_one_page(page);
170 	if (ret < 0) {
171 		_leave(" = %d", ret);
172 		return ret;
173 	}
174 
175 	ret = lock_page_killable(page);
176 	if (ret < 0) {
177 		_leave(" = %d", ret);
178 		return ret;
179 	}
180 	goto try_again;
181 }
182 
183 /*
184  * finalise part of a write to a page
185  */
186 int afs_write_end(struct file *file, struct address_space *mapping,
187 		  loff_t pos, unsigned len, unsigned copied,
188 		  struct page *page, void *fsdata)
189 {
190 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
191 	struct key *key = afs_file_key(file);
192 	loff_t i_size, maybe_i_size;
193 	int ret;
194 
195 	_enter("{%llx:%llu},{%lx}",
196 	       vnode->fid.vid, vnode->fid.vnode, page->index);
197 
198 	maybe_i_size = pos + copied;
199 
200 	i_size = i_size_read(&vnode->vfs_inode);
201 	if (maybe_i_size > i_size) {
202 		spin_lock(&vnode->wb_lock);
203 		i_size = i_size_read(&vnode->vfs_inode);
204 		if (maybe_i_size > i_size)
205 			i_size_write(&vnode->vfs_inode, maybe_i_size);
206 		spin_unlock(&vnode->wb_lock);
207 	}
208 
209 	if (!PageUptodate(page)) {
210 		if (copied < len) {
211 			/* Try and load any missing data from the server.  The
212 			 * unmarshalling routine will take care of clearing any
213 			 * bits that are beyond the EOF.
214 			 */
215 			ret = afs_fill_page(vnode, key, pos + copied,
216 					    len - copied, page);
217 			if (ret < 0)
218 				goto out;
219 		}
220 		SetPageUptodate(page);
221 	}
222 
223 	set_page_dirty(page);
224 	if (PageDirty(page))
225 		_debug("dirtied");
226 	ret = copied;
227 
228 out:
229 	unlock_page(page);
230 	put_page(page);
231 	return ret;
232 }
233 
234 /*
235  * kill all the pages in the given range
236  */
237 static void afs_kill_pages(struct address_space *mapping,
238 			   pgoff_t first, pgoff_t last)
239 {
240 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
241 	struct pagevec pv;
242 	unsigned count, loop;
243 
244 	_enter("{%llx:%llu},%lx-%lx",
245 	       vnode->fid.vid, vnode->fid.vnode, first, last);
246 
247 	pagevec_init(&pv);
248 
249 	do {
250 		_debug("kill %lx-%lx", first, last);
251 
252 		count = last - first + 1;
253 		if (count > PAGEVEC_SIZE)
254 			count = PAGEVEC_SIZE;
255 		pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
256 		ASSERTCMP(pv.nr, ==, count);
257 
258 		for (loop = 0; loop < count; loop++) {
259 			struct page *page = pv.pages[loop];
260 			ClearPageUptodate(page);
261 			SetPageError(page);
262 			end_page_writeback(page);
263 			if (page->index >= first)
264 				first = page->index + 1;
265 			lock_page(page);
266 			generic_error_remove_page(mapping, page);
267 			unlock_page(page);
268 		}
269 
270 		__pagevec_release(&pv);
271 	} while (first <= last);
272 
273 	_leave("");
274 }
275 
276 /*
277  * Redirty all the pages in a given range.
278  */
279 static void afs_redirty_pages(struct writeback_control *wbc,
280 			      struct address_space *mapping,
281 			      pgoff_t first, pgoff_t last)
282 {
283 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
284 	struct pagevec pv;
285 	unsigned count, loop;
286 
287 	_enter("{%llx:%llu},%lx-%lx",
288 	       vnode->fid.vid, vnode->fid.vnode, first, last);
289 
290 	pagevec_init(&pv);
291 
292 	do {
293 		_debug("redirty %lx-%lx", first, last);
294 
295 		count = last - first + 1;
296 		if (count > PAGEVEC_SIZE)
297 			count = PAGEVEC_SIZE;
298 		pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
299 		ASSERTCMP(pv.nr, ==, count);
300 
301 		for (loop = 0; loop < count; loop++) {
302 			struct page *page = pv.pages[loop];
303 
304 			redirty_page_for_writepage(wbc, page);
305 			end_page_writeback(page);
306 			if (page->index >= first)
307 				first = page->index + 1;
308 		}
309 
310 		__pagevec_release(&pv);
311 	} while (first <= last);
312 
313 	_leave("");
314 }
315 
316 /*
317  * write to a file
318  */
319 static int afs_store_data(struct address_space *mapping,
320 			  pgoff_t first, pgoff_t last,
321 			  unsigned offset, unsigned to)
322 {
323 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
324 	struct afs_fs_cursor fc;
325 	struct afs_wb_key *wbk = NULL;
326 	struct list_head *p;
327 	int ret = -ENOKEY, ret2;
328 
329 	_enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
330 	       vnode->volume->name,
331 	       vnode->fid.vid,
332 	       vnode->fid.vnode,
333 	       vnode->fid.unique,
334 	       first, last, offset, to);
335 
336 	spin_lock(&vnode->wb_lock);
337 	p = vnode->wb_keys.next;
338 
339 	/* Iterate through the list looking for a valid key to use. */
340 try_next_key:
341 	while (p != &vnode->wb_keys) {
342 		wbk = list_entry(p, struct afs_wb_key, vnode_link);
343 		_debug("wbk %u", key_serial(wbk->key));
344 		ret2 = key_validate(wbk->key);
345 		if (ret2 == 0)
346 			goto found_key;
347 		if (ret == -ENOKEY)
348 			ret = ret2;
349 		p = p->next;
350 	}
351 
352 	spin_unlock(&vnode->wb_lock);
353 	afs_put_wb_key(wbk);
354 	_leave(" = %d [no keys]", ret);
355 	return ret;
356 
357 found_key:
358 	refcount_inc(&wbk->usage);
359 	spin_unlock(&vnode->wb_lock);
360 
361 	_debug("USE WB KEY %u", key_serial(wbk->key));
362 
363 	ret = -ERESTARTSYS;
364 	if (afs_begin_vnode_operation(&fc, vnode, wbk->key)) {
365 		while (afs_select_fileserver(&fc)) {
366 			fc.cb_break = afs_calc_vnode_cb_break(vnode);
367 			afs_fs_store_data(&fc, mapping, first, last, offset, to);
368 		}
369 
370 		afs_check_for_remote_deletion(&fc, fc.vnode);
371 		afs_vnode_commit_status(&fc, vnode, fc.cb_break);
372 		ret = afs_end_vnode_operation(&fc);
373 	}
374 
375 	switch (ret) {
376 	case 0:
377 		afs_stat_v(vnode, n_stores);
378 		atomic_long_add((last * PAGE_SIZE + to) -
379 				(first * PAGE_SIZE + offset),
380 				&afs_v2net(vnode)->n_store_bytes);
381 		break;
382 	case -EACCES:
383 	case -EPERM:
384 	case -ENOKEY:
385 	case -EKEYEXPIRED:
386 	case -EKEYREJECTED:
387 	case -EKEYREVOKED:
388 		_debug("next");
389 		spin_lock(&vnode->wb_lock);
390 		p = wbk->vnode_link.next;
391 		afs_put_wb_key(wbk);
392 		goto try_next_key;
393 	}
394 
395 	afs_put_wb_key(wbk);
396 	_leave(" = %d", ret);
397 	return ret;
398 }
399 
400 /*
401  * Synchronously write back the locked page and any subsequent non-locked dirty
402  * pages.
403  */
404 static int afs_write_back_from_locked_page(struct address_space *mapping,
405 					   struct writeback_control *wbc,
406 					   struct page *primary_page,
407 					   pgoff_t final_page)
408 {
409 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
410 	struct page *pages[8], *page;
411 	unsigned long count, priv;
412 	unsigned n, offset, to, f, t;
413 	pgoff_t start, first, last;
414 	int loop, ret;
415 
416 	_enter(",%lx", primary_page->index);
417 
418 	count = 1;
419 	if (test_set_page_writeback(primary_page))
420 		BUG();
421 
422 	/* Find all consecutive lockable dirty pages that have contiguous
423 	 * written regions, stopping when we find a page that is not
424 	 * immediately lockable, is not dirty or is missing, or we reach the
425 	 * end of the range.
426 	 */
427 	start = primary_page->index;
428 	priv = page_private(primary_page);
429 	offset = priv & AFS_PRIV_MAX;
430 	to = priv >> AFS_PRIV_SHIFT;
431 	trace_afs_page_dirty(vnode, tracepoint_string("store"),
432 			     primary_page->index, priv);
433 
434 	WARN_ON(offset == to);
435 	if (offset == to)
436 		trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
437 				     primary_page->index, priv);
438 
439 	if (start >= final_page ||
440 	    (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
441 		goto no_more;
442 
443 	start++;
444 	do {
445 		_debug("more %lx [%lx]", start, count);
446 		n = final_page - start + 1;
447 		if (n > ARRAY_SIZE(pages))
448 			n = ARRAY_SIZE(pages);
449 		n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
450 		_debug("fgpc %u", n);
451 		if (n == 0)
452 			goto no_more;
453 		if (pages[0]->index != start) {
454 			do {
455 				put_page(pages[--n]);
456 			} while (n > 0);
457 			goto no_more;
458 		}
459 
460 		for (loop = 0; loop < n; loop++) {
461 			page = pages[loop];
462 			if (to != PAGE_SIZE &&
463 			    !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
464 				break;
465 			if (page->index > final_page)
466 				break;
467 			if (!trylock_page(page))
468 				break;
469 			if (!PageDirty(page) || PageWriteback(page)) {
470 				unlock_page(page);
471 				break;
472 			}
473 
474 			priv = page_private(page);
475 			f = priv & AFS_PRIV_MAX;
476 			t = priv >> AFS_PRIV_SHIFT;
477 			if (f != 0 &&
478 			    !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
479 				unlock_page(page);
480 				break;
481 			}
482 			to = t;
483 
484 			trace_afs_page_dirty(vnode, tracepoint_string("store+"),
485 					     page->index, priv);
486 
487 			if (!clear_page_dirty_for_io(page))
488 				BUG();
489 			if (test_set_page_writeback(page))
490 				BUG();
491 			unlock_page(page);
492 			put_page(page);
493 		}
494 		count += loop;
495 		if (loop < n) {
496 			for (; loop < n; loop++)
497 				put_page(pages[loop]);
498 			goto no_more;
499 		}
500 
501 		start += loop;
502 	} while (start <= final_page && count < 65536);
503 
504 no_more:
505 	/* We now have a contiguous set of dirty pages, each with writeback
506 	 * set; the first page is still locked at this point, but all the rest
507 	 * have been unlocked.
508 	 */
509 	unlock_page(primary_page);
510 
511 	first = primary_page->index;
512 	last = first + count - 1;
513 
514 	_debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
515 
516 	ret = afs_store_data(mapping, first, last, offset, to);
517 	switch (ret) {
518 	case 0:
519 		ret = count;
520 		break;
521 
522 	default:
523 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
524 		/* Fall through */
525 	case -EACCES:
526 	case -EPERM:
527 	case -ENOKEY:
528 	case -EKEYEXPIRED:
529 	case -EKEYREJECTED:
530 	case -EKEYREVOKED:
531 		afs_redirty_pages(wbc, mapping, first, last);
532 		mapping_set_error(mapping, ret);
533 		break;
534 
535 	case -EDQUOT:
536 	case -ENOSPC:
537 		afs_redirty_pages(wbc, mapping, first, last);
538 		mapping_set_error(mapping, -ENOSPC);
539 		break;
540 
541 	case -EROFS:
542 	case -EIO:
543 	case -EREMOTEIO:
544 	case -EFBIG:
545 	case -ENOENT:
546 	case -ENOMEDIUM:
547 	case -ENXIO:
548 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
549 		afs_kill_pages(mapping, first, last);
550 		mapping_set_error(mapping, ret);
551 		break;
552 	}
553 
554 	_leave(" = %d", ret);
555 	return ret;
556 }
557 
558 /*
559  * write a page back to the server
560  * - the caller locked the page for us
561  */
562 int afs_writepage(struct page *page, struct writeback_control *wbc)
563 {
564 	int ret;
565 
566 	_enter("{%lx},", page->index);
567 
568 	ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
569 					      wbc->range_end >> PAGE_SHIFT);
570 	if (ret < 0) {
571 		_leave(" = %d", ret);
572 		return 0;
573 	}
574 
575 	wbc->nr_to_write -= ret;
576 
577 	_leave(" = 0");
578 	return 0;
579 }
580 
581 /*
582  * write a region of pages back to the server
583  */
584 static int afs_writepages_region(struct address_space *mapping,
585 				 struct writeback_control *wbc,
586 				 pgoff_t index, pgoff_t end, pgoff_t *_next)
587 {
588 	struct page *page;
589 	int ret, n;
590 
591 	_enter(",,%lx,%lx,", index, end);
592 
593 	do {
594 		n = find_get_pages_range_tag(mapping, &index, end,
595 					PAGECACHE_TAG_DIRTY, 1, &page);
596 		if (!n)
597 			break;
598 
599 		_debug("wback %lx", page->index);
600 
601 		/*
602 		 * at this point we hold neither the i_pages lock nor the
603 		 * page lock: the page may be truncated or invalidated
604 		 * (changing page->mapping to NULL), or even swizzled
605 		 * back from swapper_space to tmpfs file mapping
606 		 */
607 		ret = lock_page_killable(page);
608 		if (ret < 0) {
609 			put_page(page);
610 			_leave(" = %d", ret);
611 			return ret;
612 		}
613 
614 		if (page->mapping != mapping || !PageDirty(page)) {
615 			unlock_page(page);
616 			put_page(page);
617 			continue;
618 		}
619 
620 		if (PageWriteback(page)) {
621 			unlock_page(page);
622 			if (wbc->sync_mode != WB_SYNC_NONE)
623 				wait_on_page_writeback(page);
624 			put_page(page);
625 			continue;
626 		}
627 
628 		if (!clear_page_dirty_for_io(page))
629 			BUG();
630 		ret = afs_write_back_from_locked_page(mapping, wbc, page, end);
631 		put_page(page);
632 		if (ret < 0) {
633 			_leave(" = %d", ret);
634 			return ret;
635 		}
636 
637 		wbc->nr_to_write -= ret;
638 
639 		cond_resched();
640 	} while (index < end && wbc->nr_to_write > 0);
641 
642 	*_next = index;
643 	_leave(" = 0 [%lx]", *_next);
644 	return 0;
645 }
646 
647 /*
648  * write some of the pending data back to the server
649  */
650 int afs_writepages(struct address_space *mapping,
651 		   struct writeback_control *wbc)
652 {
653 	pgoff_t start, end, next;
654 	int ret;
655 
656 	_enter("");
657 
658 	if (wbc->range_cyclic) {
659 		start = mapping->writeback_index;
660 		end = -1;
661 		ret = afs_writepages_region(mapping, wbc, start, end, &next);
662 		if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
663 			ret = afs_writepages_region(mapping, wbc, 0, start,
664 						    &next);
665 		mapping->writeback_index = next;
666 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
667 		end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
668 		ret = afs_writepages_region(mapping, wbc, 0, end, &next);
669 		if (wbc->nr_to_write > 0)
670 			mapping->writeback_index = next;
671 	} else {
672 		start = wbc->range_start >> PAGE_SHIFT;
673 		end = wbc->range_end >> PAGE_SHIFT;
674 		ret = afs_writepages_region(mapping, wbc, start, end, &next);
675 	}
676 
677 	_leave(" = %d", ret);
678 	return ret;
679 }
680 
681 /*
682  * completion of write to server
683  */
684 void afs_pages_written_back(struct afs_vnode *vnode, struct afs_call *call)
685 {
686 	struct pagevec pv;
687 	unsigned long priv;
688 	unsigned count, loop;
689 	pgoff_t first = call->first, last = call->last;
690 
691 	_enter("{%llx:%llu},{%lx-%lx}",
692 	       vnode->fid.vid, vnode->fid.vnode, first, last);
693 
694 	pagevec_init(&pv);
695 
696 	do {
697 		_debug("done %lx-%lx", first, last);
698 
699 		count = last - first + 1;
700 		if (count > PAGEVEC_SIZE)
701 			count = PAGEVEC_SIZE;
702 		pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
703 					      first, count, pv.pages);
704 		ASSERTCMP(pv.nr, ==, count);
705 
706 		for (loop = 0; loop < count; loop++) {
707 			priv = page_private(pv.pages[loop]);
708 			trace_afs_page_dirty(vnode, tracepoint_string("clear"),
709 					     pv.pages[loop]->index, priv);
710 			set_page_private(pv.pages[loop], 0);
711 			end_page_writeback(pv.pages[loop]);
712 		}
713 		first += count;
714 		__pagevec_release(&pv);
715 	} while (first <= last);
716 
717 	afs_prune_wb_keys(vnode);
718 	_leave("");
719 }
720 
721 /*
722  * write to an AFS file
723  */
724 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
725 {
726 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
727 	ssize_t result;
728 	size_t count = iov_iter_count(from);
729 
730 	_enter("{%llx:%llu},{%zu},",
731 	       vnode->fid.vid, vnode->fid.vnode, count);
732 
733 	if (IS_SWAPFILE(&vnode->vfs_inode)) {
734 		printk(KERN_INFO
735 		       "AFS: Attempt to write to active swap file!\n");
736 		return -EBUSY;
737 	}
738 
739 	if (!count)
740 		return 0;
741 
742 	result = generic_file_write_iter(iocb, from);
743 
744 	_leave(" = %zd", result);
745 	return result;
746 }
747 
748 /*
749  * flush any dirty pages for this process, and check for write errors.
750  * - the return status from this call provides a reliable indication of
751  *   whether any write errors occurred for this process.
752  */
753 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
754 {
755 	struct inode *inode = file_inode(file);
756 	struct afs_vnode *vnode = AFS_FS_I(inode);
757 
758 	_enter("{%llx:%llu},{n=%pD},%d",
759 	       vnode->fid.vid, vnode->fid.vnode, file,
760 	       datasync);
761 
762 	return file_write_and_wait_range(file, start, end);
763 }
764 
765 /*
766  * notification that a previously read-only page is about to become writable
767  * - if it returns an error, the caller will deliver a bus error signal
768  */
769 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
770 {
771 	struct file *file = vmf->vma->vm_file;
772 	struct inode *inode = file_inode(file);
773 	struct afs_vnode *vnode = AFS_FS_I(inode);
774 	unsigned long priv;
775 
776 	_enter("{{%llx:%llu}},{%lx}",
777 	       vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
778 
779 	sb_start_pagefault(inode->i_sb);
780 
781 	/* Wait for the page to be written to the cache before we allow it to
782 	 * be modified.  We then assume the entire page will need writing back.
783 	 */
784 #ifdef CONFIG_AFS_FSCACHE
785 	fscache_wait_on_page_write(vnode->cache, vmf->page);
786 #endif
787 
788 	if (PageWriteback(vmf->page) &&
789 	    wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
790 		return VM_FAULT_RETRY;
791 
792 	if (lock_page_killable(vmf->page) < 0)
793 		return VM_FAULT_RETRY;
794 
795 	/* We mustn't change page->private until writeback is complete as that
796 	 * details the portion of the page we need to write back and we might
797 	 * need to redirty the page if there's a problem.
798 	 */
799 	wait_on_page_writeback(vmf->page);
800 
801 	priv = (unsigned long)PAGE_SIZE << AFS_PRIV_SHIFT; /* To */
802 	priv |= 0; /* From */
803 	trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
804 			     vmf->page->index, priv);
805 	SetPagePrivate(vmf->page);
806 	set_page_private(vmf->page, priv);
807 
808 	sb_end_pagefault(inode->i_sb);
809 	return VM_FAULT_LOCKED;
810 }
811 
812 /*
813  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
814  */
815 void afs_prune_wb_keys(struct afs_vnode *vnode)
816 {
817 	LIST_HEAD(graveyard);
818 	struct afs_wb_key *wbk, *tmp;
819 
820 	/* Discard unused keys */
821 	spin_lock(&vnode->wb_lock);
822 
823 	if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
824 	    !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
825 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
826 			if (refcount_read(&wbk->usage) == 1)
827 				list_move(&wbk->vnode_link, &graveyard);
828 		}
829 	}
830 
831 	spin_unlock(&vnode->wb_lock);
832 
833 	while (!list_empty(&graveyard)) {
834 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
835 		list_del(&wbk->vnode_link);
836 		afs_put_wb_key(wbk);
837 	}
838 }
839 
840 /*
841  * Clean up a page during invalidation.
842  */
843 int afs_launder_page(struct page *page)
844 {
845 	struct address_space *mapping = page->mapping;
846 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
847 	unsigned long priv;
848 	unsigned int f, t;
849 	int ret = 0;
850 
851 	_enter("{%lx}", page->index);
852 
853 	priv = page_private(page);
854 	if (clear_page_dirty_for_io(page)) {
855 		f = 0;
856 		t = PAGE_SIZE;
857 		if (PagePrivate(page)) {
858 			f = priv & AFS_PRIV_MAX;
859 			t = priv >> AFS_PRIV_SHIFT;
860 		}
861 
862 		trace_afs_page_dirty(vnode, tracepoint_string("launder"),
863 				     page->index, priv);
864 		ret = afs_store_data(mapping, page->index, page->index, t, f);
865 	}
866 
867 	trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
868 			     page->index, priv);
869 	set_page_private(page, 0);
870 	ClearPagePrivate(page);
871 
872 #ifdef CONFIG_AFS_FSCACHE
873 	if (PageFsCache(page)) {
874 		fscache_wait_on_page_write(vnode->cache, page);
875 		fscache_uncache_page(vnode->cache, page);
876 	}
877 #endif
878 	return ret;
879 }
880