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