• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "internal.h"
15 
16 /*
17  * mark a page as having been made dirty and thus needing writeback
18  */
afs_set_page_dirty(struct page * page)19 int afs_set_page_dirty(struct page *page)
20 {
21 	_enter("");
22 	return __set_page_dirty_nobuffers(page);
23 }
24 
25 /*
26  * partly or wholly fill a page that's under preparation for writing
27  */
afs_fill_page(struct afs_vnode * vnode,struct key * key,loff_t pos,unsigned int len,struct page * page)28 static int afs_fill_page(struct afs_vnode *vnode, struct key *key,
29 			 loff_t pos, unsigned int len, struct page *page)
30 {
31 	struct afs_read *req;
32 	size_t p;
33 	void *data;
34 	int ret;
35 
36 	_enter(",,%llu", (unsigned long long)pos);
37 
38 	if (pos >= vnode->vfs_inode.i_size) {
39 		p = pos & ~PAGE_MASK;
40 		ASSERTCMP(p + len, <=, PAGE_SIZE);
41 		data = kmap(page);
42 		memset(data + p, 0, len);
43 		kunmap(page);
44 		return 0;
45 	}
46 
47 	req = kzalloc(struct_size(req, array, 1), GFP_KERNEL);
48 	if (!req)
49 		return -ENOMEM;
50 
51 	refcount_set(&req->usage, 1);
52 	req->pos = pos;
53 	req->len = len;
54 	req->nr_pages = 1;
55 	req->pages = req->array;
56 	req->pages[0] = page;
57 	get_page(page);
58 
59 	ret = afs_fetch_data(vnode, key, req);
60 	afs_put_read(req);
61 	if (ret < 0) {
62 		if (ret == -ENOENT) {
63 			_debug("got NOENT from server"
64 			       " - marking file deleted and stale");
65 			set_bit(AFS_VNODE_DELETED, &vnode->flags);
66 			ret = -ESTALE;
67 		}
68 	}
69 
70 	_leave(" = %d", ret);
71 	return ret;
72 }
73 
74 /*
75  * prepare to perform part of a write to a page
76  */
afs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned flags,struct page ** _page,void ** fsdata)77 int afs_write_begin(struct file *file, struct address_space *mapping,
78 		    loff_t pos, unsigned len, unsigned flags,
79 		    struct page **_page, void **fsdata)
80 {
81 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
82 	struct page *page;
83 	struct key *key = afs_file_key(file);
84 	unsigned long priv;
85 	unsigned f, from = pos & (PAGE_SIZE - 1);
86 	unsigned t, to = from + len;
87 	pgoff_t index = pos >> PAGE_SHIFT;
88 	int ret;
89 
90 	_enter("{%llx:%llu},{%lx},%u,%u",
91 	       vnode->fid.vid, vnode->fid.vnode, index, from, to);
92 
93 	page = grab_cache_page_write_begin(mapping, index, flags);
94 	if (!page)
95 		return -ENOMEM;
96 
97 	if (!PageUptodate(page) && len != PAGE_SIZE) {
98 		ret = afs_fill_page(vnode, key, pos & PAGE_MASK, PAGE_SIZE, page);
99 		if (ret < 0) {
100 			unlock_page(page);
101 			put_page(page);
102 			_leave(" = %d [prep]", ret);
103 			return ret;
104 		}
105 		SetPageUptodate(page);
106 	}
107 
108 try_again:
109 	/* See if this page is already partially written in a way that we can
110 	 * merge the new write with.
111 	 */
112 	t = f = 0;
113 	if (PagePrivate(page)) {
114 		priv = page_private(page);
115 		f = afs_page_dirty_from(priv);
116 		t = afs_page_dirty_to(priv);
117 		ASSERTCMP(f, <=, t);
118 	}
119 
120 	if (f != t) {
121 		if (PageWriteback(page)) {
122 			trace_afs_page_dirty(vnode, tracepoint_string("alrdy"),
123 					     page->index, priv);
124 			goto flush_conflicting_write;
125 		}
126 		/* If the file is being filled locally, allow inter-write
127 		 * spaces to be merged into writes.  If it's not, only write
128 		 * back what the user gives us.
129 		 */
130 		if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
131 		    (to < f || from > t))
132 			goto flush_conflicting_write;
133 	}
134 
135 	*_page = page;
136 	_leave(" = 0");
137 	return 0;
138 
139 	/* The previous write and this write aren't adjacent or overlapping, so
140 	 * flush the page out.
141 	 */
142 flush_conflicting_write:
143 	_debug("flush conflict");
144 	ret = write_one_page(page);
145 	if (ret < 0)
146 		goto error;
147 
148 	ret = lock_page_killable(page);
149 	if (ret < 0)
150 		goto error;
151 	goto try_again;
152 
153 error:
154 	put_page(page);
155 	_leave(" = %d", ret);
156 	return ret;
157 }
158 
159 /*
160  * finalise part of a write to a page
161  */
afs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)162 int afs_write_end(struct file *file, struct address_space *mapping,
163 		  loff_t pos, unsigned len, unsigned copied,
164 		  struct page *page, void *fsdata)
165 {
166 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
167 	struct key *key = afs_file_key(file);
168 	unsigned long priv;
169 	unsigned int f, from = pos & (PAGE_SIZE - 1);
170 	unsigned int t, to = from + copied;
171 	loff_t i_size, maybe_i_size;
172 	int ret = 0;
173 
174 	_enter("{%llx:%llu},{%lx}",
175 	       vnode->fid.vid, vnode->fid.vnode, page->index);
176 
177 	if (copied == 0)
178 		goto out;
179 
180 	maybe_i_size = pos + copied;
181 
182 	i_size = i_size_read(&vnode->vfs_inode);
183 	if (maybe_i_size > i_size) {
184 		write_seqlock(&vnode->cb_lock);
185 		i_size = i_size_read(&vnode->vfs_inode);
186 		if (maybe_i_size > i_size)
187 			afs_set_i_size(vnode, maybe_i_size);
188 		write_sequnlock(&vnode->cb_lock);
189 	}
190 
191 	if (!PageUptodate(page)) {
192 		if (copied < len) {
193 			/* Try and load any missing data from the server.  The
194 			 * unmarshalling routine will take care of clearing any
195 			 * bits that are beyond the EOF.
196 			 */
197 			ret = afs_fill_page(vnode, key, pos + copied,
198 					    len - copied, page);
199 			if (ret < 0)
200 				goto out;
201 		}
202 		SetPageUptodate(page);
203 	}
204 
205 	if (PagePrivate(page)) {
206 		priv = page_private(page);
207 		f = afs_page_dirty_from(priv);
208 		t = afs_page_dirty_to(priv);
209 		if (from < f)
210 			f = from;
211 		if (to > t)
212 			t = to;
213 		priv = afs_page_dirty(f, t);
214 		set_page_private(page, priv);
215 		trace_afs_page_dirty(vnode, tracepoint_string("dirty+"),
216 				     page->index, priv);
217 	} else {
218 		priv = afs_page_dirty(from, to);
219 		attach_page_private(page, (void *)priv);
220 		trace_afs_page_dirty(vnode, tracepoint_string("dirty"),
221 				     page->index, priv);
222 	}
223 
224 	set_page_dirty(page);
225 	if (PageDirty(page))
226 		_debug("dirtied");
227 	ret = copied;
228 
229 out:
230 	unlock_page(page);
231 	put_page(page);
232 	return ret;
233 }
234 
235 /*
236  * kill all the pages in the given range
237  */
afs_kill_pages(struct address_space * mapping,pgoff_t first,pgoff_t last)238 static void afs_kill_pages(struct address_space *mapping,
239 			   pgoff_t first, pgoff_t last)
240 {
241 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
242 	struct pagevec pv;
243 	unsigned count, loop;
244 
245 	_enter("{%llx:%llu},%lx-%lx",
246 	       vnode->fid.vid, vnode->fid.vnode, first, last);
247 
248 	pagevec_init(&pv);
249 
250 	do {
251 		_debug("kill %lx-%lx", first, last);
252 
253 		count = last - first + 1;
254 		if (count > PAGEVEC_SIZE)
255 			count = PAGEVEC_SIZE;
256 		pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
257 		ASSERTCMP(pv.nr, ==, count);
258 
259 		for (loop = 0; loop < count; loop++) {
260 			struct page *page = pv.pages[loop];
261 			ClearPageUptodate(page);
262 			SetPageError(page);
263 			end_page_writeback(page);
264 			if (page->index >= first)
265 				first = page->index + 1;
266 			lock_page(page);
267 			generic_error_remove_page(mapping, page);
268 			unlock_page(page);
269 		}
270 
271 		__pagevec_release(&pv);
272 	} while (first <= last);
273 
274 	_leave("");
275 }
276 
277 /*
278  * Redirty all the pages in a given range.
279  */
afs_redirty_pages(struct writeback_control * wbc,struct address_space * mapping,pgoff_t first,pgoff_t last)280 static void afs_redirty_pages(struct writeback_control *wbc,
281 			      struct address_space *mapping,
282 			      pgoff_t first, pgoff_t last)
283 {
284 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
285 	struct pagevec pv;
286 	unsigned count, loop;
287 
288 	_enter("{%llx:%llu},%lx-%lx",
289 	       vnode->fid.vid, vnode->fid.vnode, first, last);
290 
291 	pagevec_init(&pv);
292 
293 	do {
294 		_debug("redirty %lx-%lx", first, last);
295 
296 		count = last - first + 1;
297 		if (count > PAGEVEC_SIZE)
298 			count = PAGEVEC_SIZE;
299 		pv.nr = find_get_pages_contig(mapping, first, count, pv.pages);
300 		ASSERTCMP(pv.nr, ==, count);
301 
302 		for (loop = 0; loop < count; loop++) {
303 			struct page *page = pv.pages[loop];
304 
305 			redirty_page_for_writepage(wbc, page);
306 			end_page_writeback(page);
307 			if (page->index >= first)
308 				first = page->index + 1;
309 		}
310 
311 		__pagevec_release(&pv);
312 	} while (first <= last);
313 
314 	_leave("");
315 }
316 
317 /*
318  * completion of write to server
319  */
afs_pages_written_back(struct afs_vnode * vnode,pgoff_t first,pgoff_t last)320 static void afs_pages_written_back(struct afs_vnode *vnode,
321 				   pgoff_t first, pgoff_t last)
322 {
323 	struct pagevec pv;
324 	unsigned long priv;
325 	unsigned count, loop;
326 
327 	_enter("{%llx:%llu},{%lx-%lx}",
328 	       vnode->fid.vid, vnode->fid.vnode, first, last);
329 
330 	pagevec_init(&pv);
331 
332 	do {
333 		_debug("done %lx-%lx", first, last);
334 
335 		count = last - first + 1;
336 		if (count > PAGEVEC_SIZE)
337 			count = PAGEVEC_SIZE;
338 		pv.nr = find_get_pages_contig(vnode->vfs_inode.i_mapping,
339 					      first, count, pv.pages);
340 		ASSERTCMP(pv.nr, ==, count);
341 
342 		for (loop = 0; loop < count; loop++) {
343 			priv = (unsigned long)detach_page_private(pv.pages[loop]);
344 			trace_afs_page_dirty(vnode, tracepoint_string("clear"),
345 					     pv.pages[loop]->index, priv);
346 			end_page_writeback(pv.pages[loop]);
347 		}
348 		first += count;
349 		__pagevec_release(&pv);
350 	} while (first <= last);
351 
352 	afs_prune_wb_keys(vnode);
353 	_leave("");
354 }
355 
356 /*
357  * Find a key to use for the writeback.  We cached the keys used to author the
358  * writes on the vnode.  *_wbk will contain the last writeback key used or NULL
359  * and we need to start from there if it's set.
360  */
afs_get_writeback_key(struct afs_vnode * vnode,struct afs_wb_key ** _wbk)361 static int afs_get_writeback_key(struct afs_vnode *vnode,
362 				 struct afs_wb_key **_wbk)
363 {
364 	struct afs_wb_key *wbk = NULL;
365 	struct list_head *p;
366 	int ret = -ENOKEY, ret2;
367 
368 	spin_lock(&vnode->wb_lock);
369 	if (*_wbk)
370 		p = (*_wbk)->vnode_link.next;
371 	else
372 		p = vnode->wb_keys.next;
373 
374 	while (p != &vnode->wb_keys) {
375 		wbk = list_entry(p, struct afs_wb_key, vnode_link);
376 		_debug("wbk %u", key_serial(wbk->key));
377 		ret2 = key_validate(wbk->key);
378 		if (ret2 == 0) {
379 			refcount_inc(&wbk->usage);
380 			_debug("USE WB KEY %u", key_serial(wbk->key));
381 			break;
382 		}
383 
384 		wbk = NULL;
385 		if (ret == -ENOKEY)
386 			ret = ret2;
387 		p = p->next;
388 	}
389 
390 	spin_unlock(&vnode->wb_lock);
391 	if (*_wbk)
392 		afs_put_wb_key(*_wbk);
393 	*_wbk = wbk;
394 	return 0;
395 }
396 
afs_store_data_success(struct afs_operation * op)397 static void afs_store_data_success(struct afs_operation *op)
398 {
399 	struct afs_vnode *vnode = op->file[0].vnode;
400 
401 	op->ctime = op->file[0].scb.status.mtime_client;
402 	afs_vnode_commit_status(op, &op->file[0]);
403 	if (op->error == 0) {
404 		if (!op->store.laundering)
405 			afs_pages_written_back(vnode, op->store.first, op->store.last);
406 		afs_stat_v(vnode, n_stores);
407 		atomic_long_add((op->store.last * PAGE_SIZE + op->store.last_to) -
408 				(op->store.first * PAGE_SIZE + op->store.first_offset),
409 				&afs_v2net(vnode)->n_store_bytes);
410 	}
411 }
412 
413 static const struct afs_operation_ops afs_store_data_operation = {
414 	.issue_afs_rpc	= afs_fs_store_data,
415 	.issue_yfs_rpc	= yfs_fs_store_data,
416 	.success	= afs_store_data_success,
417 };
418 
419 /*
420  * write to a file
421  */
afs_store_data(struct address_space * mapping,pgoff_t first,pgoff_t last,unsigned offset,unsigned to,bool laundering)422 static int afs_store_data(struct address_space *mapping,
423 			  pgoff_t first, pgoff_t last,
424 			  unsigned offset, unsigned to, bool laundering)
425 {
426 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
427 	struct afs_operation *op;
428 	struct afs_wb_key *wbk = NULL;
429 	int ret;
430 
431 	_enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
432 	       vnode->volume->name,
433 	       vnode->fid.vid,
434 	       vnode->fid.vnode,
435 	       vnode->fid.unique,
436 	       first, last, offset, to);
437 
438 	ret = afs_get_writeback_key(vnode, &wbk);
439 	if (ret) {
440 		_leave(" = %d [no keys]", ret);
441 		return ret;
442 	}
443 
444 	op = afs_alloc_operation(wbk->key, vnode->volume);
445 	if (IS_ERR(op)) {
446 		afs_put_wb_key(wbk);
447 		return -ENOMEM;
448 	}
449 
450 	afs_op_set_vnode(op, 0, vnode);
451 	op->file[0].dv_delta = 1;
452 	op->store.mapping = mapping;
453 	op->file[0].modification = true;
454 	op->store.first = first;
455 	op->store.last = last;
456 	op->store.first_offset = offset;
457 	op->store.last_to = to;
458 	op->store.laundering = laundering;
459 	op->mtime = vnode->vfs_inode.i_mtime;
460 	op->flags |= AFS_OPERATION_UNINTR;
461 	op->ops = &afs_store_data_operation;
462 
463 try_next_key:
464 	afs_begin_vnode_operation(op);
465 	afs_wait_for_operation(op);
466 
467 	switch (op->error) {
468 	case -EACCES:
469 	case -EPERM:
470 	case -ENOKEY:
471 	case -EKEYEXPIRED:
472 	case -EKEYREJECTED:
473 	case -EKEYREVOKED:
474 		_debug("next");
475 
476 		ret = afs_get_writeback_key(vnode, &wbk);
477 		if (ret == 0) {
478 			key_put(op->key);
479 			op->key = key_get(wbk->key);
480 			goto try_next_key;
481 		}
482 		break;
483 	}
484 
485 	afs_put_wb_key(wbk);
486 	_leave(" = %d", op->error);
487 	return afs_put_operation(op);
488 }
489 
490 /*
491  * Synchronously write back the locked page and any subsequent non-locked dirty
492  * pages.
493  */
afs_write_back_from_locked_page(struct address_space * mapping,struct writeback_control * wbc,struct page * primary_page,pgoff_t final_page)494 static int afs_write_back_from_locked_page(struct address_space *mapping,
495 					   struct writeback_control *wbc,
496 					   struct page *primary_page,
497 					   pgoff_t final_page)
498 {
499 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
500 	struct page *pages[8], *page;
501 	unsigned long count, priv;
502 	unsigned n, offset, to, f, t;
503 	pgoff_t start, first, last;
504 	loff_t i_size, end;
505 	int loop, ret;
506 
507 	_enter(",%lx", primary_page->index);
508 
509 	count = 1;
510 	if (test_set_page_writeback(primary_page))
511 		BUG();
512 
513 	/* Find all consecutive lockable dirty pages that have contiguous
514 	 * written regions, stopping when we find a page that is not
515 	 * immediately lockable, is not dirty or is missing, or we reach the
516 	 * end of the range.
517 	 */
518 	start = primary_page->index;
519 	priv = page_private(primary_page);
520 	offset = afs_page_dirty_from(priv);
521 	to = afs_page_dirty_to(priv);
522 	trace_afs_page_dirty(vnode, tracepoint_string("store"),
523 			     primary_page->index, priv);
524 
525 	WARN_ON(offset == to);
526 	if (offset == to)
527 		trace_afs_page_dirty(vnode, tracepoint_string("WARN"),
528 				     primary_page->index, priv);
529 
530 	if (start >= final_page ||
531 	    (to < PAGE_SIZE && !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)))
532 		goto no_more;
533 
534 	start++;
535 	do {
536 		_debug("more %lx [%lx]", start, count);
537 		n = final_page - start + 1;
538 		if (n > ARRAY_SIZE(pages))
539 			n = ARRAY_SIZE(pages);
540 		n = find_get_pages_contig(mapping, start, ARRAY_SIZE(pages), pages);
541 		_debug("fgpc %u", n);
542 		if (n == 0)
543 			goto no_more;
544 		if (pages[0]->index != start) {
545 			do {
546 				put_page(pages[--n]);
547 			} while (n > 0);
548 			goto no_more;
549 		}
550 
551 		for (loop = 0; loop < n; loop++) {
552 			page = pages[loop];
553 			if (to != PAGE_SIZE &&
554 			    !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags))
555 				break;
556 			if (page->index > final_page)
557 				break;
558 			if (!trylock_page(page))
559 				break;
560 			if (!PageDirty(page) || PageWriteback(page)) {
561 				unlock_page(page);
562 				break;
563 			}
564 
565 			priv = page_private(page);
566 			f = afs_page_dirty_from(priv);
567 			t = afs_page_dirty_to(priv);
568 			if (f != 0 &&
569 			    !test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags)) {
570 				unlock_page(page);
571 				break;
572 			}
573 			to = t;
574 
575 			trace_afs_page_dirty(vnode, tracepoint_string("store+"),
576 					     page->index, priv);
577 
578 			if (!clear_page_dirty_for_io(page))
579 				BUG();
580 			if (test_set_page_writeback(page))
581 				BUG();
582 			unlock_page(page);
583 			put_page(page);
584 		}
585 		count += loop;
586 		if (loop < n) {
587 			for (; loop < n; loop++)
588 				put_page(pages[loop]);
589 			goto no_more;
590 		}
591 
592 		start += loop;
593 	} while (start <= final_page && count < 65536);
594 
595 no_more:
596 	/* We now have a contiguous set of dirty pages, each with writeback
597 	 * set; the first page is still locked at this point, but all the rest
598 	 * have been unlocked.
599 	 */
600 	unlock_page(primary_page);
601 
602 	first = primary_page->index;
603 	last = first + count - 1;
604 
605 	end = (loff_t)last * PAGE_SIZE + to;
606 	i_size = i_size_read(&vnode->vfs_inode);
607 
608 	_debug("write back %lx[%u..] to %lx[..%u]", first, offset, last, to);
609 	if (end > i_size)
610 		to = i_size & ~PAGE_MASK;
611 
612 	ret = afs_store_data(mapping, first, last, offset, to, false);
613 	switch (ret) {
614 	case 0:
615 		ret = count;
616 		break;
617 
618 	default:
619 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
620 		fallthrough;
621 	case -EACCES:
622 	case -EPERM:
623 	case -ENOKEY:
624 	case -EKEYEXPIRED:
625 	case -EKEYREJECTED:
626 	case -EKEYREVOKED:
627 		afs_redirty_pages(wbc, mapping, first, last);
628 		mapping_set_error(mapping, ret);
629 		break;
630 
631 	case -EDQUOT:
632 	case -ENOSPC:
633 		afs_redirty_pages(wbc, mapping, first, last);
634 		mapping_set_error(mapping, -ENOSPC);
635 		break;
636 
637 	case -EROFS:
638 	case -EIO:
639 	case -EREMOTEIO:
640 	case -EFBIG:
641 	case -ENOENT:
642 	case -ENOMEDIUM:
643 	case -ENXIO:
644 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
645 		afs_kill_pages(mapping, first, last);
646 		mapping_set_error(mapping, ret);
647 		break;
648 	}
649 
650 	_leave(" = %d", ret);
651 	return ret;
652 }
653 
654 /*
655  * write a page back to the server
656  * - the caller locked the page for us
657  */
afs_writepage(struct page * page,struct writeback_control * wbc)658 int afs_writepage(struct page *page, struct writeback_control *wbc)
659 {
660 	int ret;
661 
662 	_enter("{%lx},", page->index);
663 
664 	ret = afs_write_back_from_locked_page(page->mapping, wbc, page,
665 					      wbc->range_end >> PAGE_SHIFT);
666 	if (ret < 0) {
667 		_leave(" = %d", ret);
668 		return 0;
669 	}
670 
671 	wbc->nr_to_write -= ret;
672 
673 	_leave(" = 0");
674 	return 0;
675 }
676 
677 /*
678  * write a region of pages back to the server
679  */
afs_writepages_region(struct address_space * mapping,struct writeback_control * wbc,pgoff_t index,pgoff_t end,pgoff_t * _next)680 static int afs_writepages_region(struct address_space *mapping,
681 				 struct writeback_control *wbc,
682 				 pgoff_t index, pgoff_t end, pgoff_t *_next)
683 {
684 	struct page *page;
685 	int ret, n;
686 
687 	_enter(",,%lx,%lx,", index, end);
688 
689 	do {
690 		n = find_get_pages_range_tag(mapping, &index, end,
691 					PAGECACHE_TAG_DIRTY, 1, &page);
692 		if (!n)
693 			break;
694 
695 		_debug("wback %lx", page->index);
696 
697 		/*
698 		 * at this point we hold neither the i_pages lock nor the
699 		 * page lock: the page may be truncated or invalidated
700 		 * (changing page->mapping to NULL), or even swizzled
701 		 * back from swapper_space to tmpfs file mapping
702 		 */
703 		ret = lock_page_killable(page);
704 		if (ret < 0) {
705 			put_page(page);
706 			_leave(" = %d", ret);
707 			return ret;
708 		}
709 
710 		if (page->mapping != mapping || !PageDirty(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, end);
727 		put_page(page);
728 		if (ret < 0) {
729 			_leave(" = %d", ret);
730 			return ret;
731 		}
732 
733 		wbc->nr_to_write -= ret;
734 
735 		cond_resched();
736 	} while (index < end && wbc->nr_to_write > 0);
737 
738 	*_next = index;
739 	_leave(" = 0 [%lx]", *_next);
740 	return 0;
741 }
742 
743 /*
744  * write some of the pending data back to the server
745  */
afs_writepages(struct address_space * mapping,struct writeback_control * wbc)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 	pgoff_t start, end, 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;
766 		end = -1;
767 		ret = afs_writepages_region(mapping, wbc, start, end, &next);
768 		if (start > 0 && wbc->nr_to_write > 0 && ret == 0)
769 			ret = afs_writepages_region(mapping, wbc, 0, start,
770 						    &next);
771 		mapping->writeback_index = next;
772 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
773 		end = (pgoff_t)(LLONG_MAX >> PAGE_SHIFT);
774 		ret = afs_writepages_region(mapping, wbc, 0, end, &next);
775 		if (wbc->nr_to_write > 0)
776 			mapping->writeback_index = next;
777 	} else {
778 		start = wbc->range_start >> PAGE_SHIFT;
779 		end = wbc->range_end >> PAGE_SHIFT;
780 		ret = afs_writepages_region(mapping, wbc, start, end, &next);
781 	}
782 
783 	up_read(&vnode->validate_lock);
784 	_leave(" = %d", ret);
785 	return ret;
786 }
787 
788 /*
789  * write to an AFS file
790  */
afs_file_write(struct kiocb * iocb,struct iov_iter * from)791 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
792 {
793 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
794 	ssize_t result;
795 	size_t count = iov_iter_count(from);
796 
797 	_enter("{%llx:%llu},{%zu},",
798 	       vnode->fid.vid, vnode->fid.vnode, count);
799 
800 	if (IS_SWAPFILE(&vnode->vfs_inode)) {
801 		printk(KERN_INFO
802 		       "AFS: Attempt to write to active swap file!\n");
803 		return -EBUSY;
804 	}
805 
806 	if (!count)
807 		return 0;
808 
809 	result = generic_file_write_iter(iocb, from);
810 
811 	_leave(" = %zd", result);
812 	return result;
813 }
814 
815 /*
816  * flush any dirty pages for this process, and check for write errors.
817  * - the return status from this call provides a reliable indication of
818  *   whether any write errors occurred for this process.
819  */
afs_fsync(struct file * file,loff_t start,loff_t end,int datasync)820 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
821 {
822 	struct inode *inode = file_inode(file);
823 	struct afs_vnode *vnode = AFS_FS_I(inode);
824 
825 	_enter("{%llx:%llu},{n=%pD},%d",
826 	       vnode->fid.vid, vnode->fid.vnode, file,
827 	       datasync);
828 
829 	return file_write_and_wait_range(file, start, end);
830 }
831 
832 /*
833  * notification that a previously read-only page is about to become writable
834  * - if it returns an error, the caller will deliver a bus error signal
835  */
afs_page_mkwrite(struct vm_fault * vmf)836 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
837 {
838 	struct file *file = vmf->vma->vm_file;
839 	struct inode *inode = file_inode(file);
840 	struct afs_vnode *vnode = AFS_FS_I(inode);
841 	unsigned long priv;
842 
843 	_enter("{{%llx:%llu}},{%lx}",
844 	       vnode->fid.vid, vnode->fid.vnode, vmf->page->index);
845 
846 	sb_start_pagefault(inode->i_sb);
847 
848 	/* Wait for the page to be written to the cache before we allow it to
849 	 * be modified.  We then assume the entire page will need writing back.
850 	 */
851 #ifdef CONFIG_AFS_FSCACHE
852 	fscache_wait_on_page_write(vnode->cache, vmf->page);
853 #endif
854 
855 	if (PageWriteback(vmf->page) &&
856 	    wait_on_page_bit_killable(vmf->page, PG_writeback) < 0)
857 		return VM_FAULT_RETRY;
858 
859 	if (lock_page_killable(vmf->page) < 0)
860 		return VM_FAULT_RETRY;
861 
862 	/* We mustn't change page->private until writeback is complete as that
863 	 * details the portion of the page we need to write back and we might
864 	 * need to redirty the page if there's a problem.
865 	 */
866 	wait_on_page_writeback(vmf->page);
867 
868 	priv = afs_page_dirty(0, PAGE_SIZE);
869 	priv = afs_page_dirty_mmapped(priv);
870 	trace_afs_page_dirty(vnode, tracepoint_string("mkwrite"),
871 			     vmf->page->index, priv);
872 	if (PagePrivate(vmf->page))
873 		set_page_private(vmf->page, priv);
874 	else
875 		attach_page_private(vmf->page, (void *)priv);
876 	file_update_time(file);
877 
878 	sb_end_pagefault(inode->i_sb);
879 	return VM_FAULT_LOCKED;
880 }
881 
882 /*
883  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
884  */
afs_prune_wb_keys(struct afs_vnode * vnode)885 void afs_prune_wb_keys(struct afs_vnode *vnode)
886 {
887 	LIST_HEAD(graveyard);
888 	struct afs_wb_key *wbk, *tmp;
889 
890 	/* Discard unused keys */
891 	spin_lock(&vnode->wb_lock);
892 
893 	if (!mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
894 	    !mapping_tagged(&vnode->vfs_inode.i_data, PAGECACHE_TAG_DIRTY)) {
895 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
896 			if (refcount_read(&wbk->usage) == 1)
897 				list_move(&wbk->vnode_link, &graveyard);
898 		}
899 	}
900 
901 	spin_unlock(&vnode->wb_lock);
902 
903 	while (!list_empty(&graveyard)) {
904 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
905 		list_del(&wbk->vnode_link);
906 		afs_put_wb_key(wbk);
907 	}
908 }
909 
910 /*
911  * Clean up a page during invalidation.
912  */
afs_launder_page(struct page * page)913 int afs_launder_page(struct page *page)
914 {
915 	struct address_space *mapping = page->mapping;
916 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
917 	unsigned long priv;
918 	unsigned int f, t;
919 	int ret = 0;
920 
921 	_enter("{%lx}", page->index);
922 
923 	priv = page_private(page);
924 	if (clear_page_dirty_for_io(page)) {
925 		f = 0;
926 		t = PAGE_SIZE;
927 		if (PagePrivate(page)) {
928 			f = afs_page_dirty_from(priv);
929 			t = afs_page_dirty_to(priv);
930 		}
931 
932 		trace_afs_page_dirty(vnode, tracepoint_string("launder"),
933 				     page->index, priv);
934 		ret = afs_store_data(mapping, page->index, page->index, t, f, true);
935 	}
936 
937 	priv = (unsigned long)detach_page_private(page);
938 	trace_afs_page_dirty(vnode, tracepoint_string("laundered"),
939 			     page->index, priv);
940 
941 #ifdef CONFIG_AFS_FSCACHE
942 	if (PageFsCache(page)) {
943 		fscache_wait_on_page_write(vnode->cache, page);
944 		fscache_uncache_page(vnode->cache, page);
945 	}
946 #endif
947 	return ret;
948 }
949