xref: /openbmc/linux/fs/afs/write.c (revision 16380f52)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* handling of writes to regular files and writing back to the server
3  *
4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5  * Written by David Howells (dhowells@redhat.com)
6  */
7 
8 #include <linux/backing-dev.h>
9 #include <linux/slab.h>
10 #include <linux/fs.h>
11 #include <linux/pagemap.h>
12 #include <linux/writeback.h>
13 #include <linux/pagevec.h>
14 #include <linux/netfs.h>
15 #include "internal.h"
16 
17 static int afs_writepages_region(struct address_space *mapping,
18 				 struct writeback_control *wbc,
19 				 loff_t start, loff_t end, loff_t *_next,
20 				 bool max_one_loop);
21 
22 static void afs_write_to_cache(struct afs_vnode *vnode, loff_t start, size_t len,
23 			       loff_t i_size, bool caching);
24 
25 #ifdef CONFIG_AFS_FSCACHE
26 /*
27  * Mark a page as having been made dirty and thus needing writeback.  We also
28  * need to pin the cache object to write back to.
29  */
afs_dirty_folio(struct address_space * mapping,struct folio * folio)30 bool afs_dirty_folio(struct address_space *mapping, struct folio *folio)
31 {
32 	return fscache_dirty_folio(mapping, folio,
33 				afs_vnode_cache(AFS_FS_I(mapping->host)));
34 }
afs_folio_start_fscache(bool caching,struct folio * folio)35 static void afs_folio_start_fscache(bool caching, struct folio *folio)
36 {
37 	if (caching)
38 		folio_start_fscache(folio);
39 }
40 #else
afs_folio_start_fscache(bool caching,struct folio * folio)41 static void afs_folio_start_fscache(bool caching, struct folio *folio)
42 {
43 }
44 #endif
45 
46 /*
47  * Flush out a conflicting write.  This may extend the write to the surrounding
48  * pages if also dirty and contiguous to the conflicting region..
49  */
afs_flush_conflicting_write(struct address_space * mapping,struct folio * folio)50 static int afs_flush_conflicting_write(struct address_space *mapping,
51 				       struct folio *folio)
52 {
53 	struct writeback_control wbc = {
54 		.sync_mode	= WB_SYNC_ALL,
55 		.nr_to_write	= LONG_MAX,
56 		.range_start	= folio_pos(folio),
57 		.range_end	= LLONG_MAX,
58 	};
59 	loff_t next;
60 
61 	return afs_writepages_region(mapping, &wbc, folio_pos(folio), LLONG_MAX,
62 				     &next, true);
63 }
64 
65 /*
66  * prepare to perform part of a write to a page
67  */
afs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct page ** _page,void ** fsdata)68 int afs_write_begin(struct file *file, struct address_space *mapping,
69 		    loff_t pos, unsigned len,
70 		    struct page **_page, void **fsdata)
71 {
72 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
73 	struct folio *folio;
74 	unsigned long priv;
75 	unsigned f, from;
76 	unsigned t, to;
77 	pgoff_t index;
78 	int ret;
79 
80 	_enter("{%llx:%llu},%llx,%x",
81 	       vnode->fid.vid, vnode->fid.vnode, pos, len);
82 
83 	/* Prefetch area to be written into the cache if we're caching this
84 	 * file.  We need to do this before we get a lock on the page in case
85 	 * there's more than one writer competing for the same cache block.
86 	 */
87 	ret = netfs_write_begin(&vnode->netfs, file, mapping, pos, len, &folio, fsdata);
88 	if (ret < 0)
89 		return ret;
90 
91 	index = folio_index(folio);
92 	from = pos - index * PAGE_SIZE;
93 	to = from + len;
94 
95 try_again:
96 	/* See if this page is already partially written in a way that we can
97 	 * merge the new write with.
98 	 */
99 	if (folio_test_private(folio)) {
100 		priv = (unsigned long)folio_get_private(folio);
101 		f = afs_folio_dirty_from(folio, priv);
102 		t = afs_folio_dirty_to(folio, priv);
103 		ASSERTCMP(f, <=, t);
104 
105 		if (folio_test_writeback(folio)) {
106 			trace_afs_folio_dirty(vnode, tracepoint_string("alrdy"), folio);
107 			folio_unlock(folio);
108 			goto wait_for_writeback;
109 		}
110 		/* If the file is being filled locally, allow inter-write
111 		 * spaces to be merged into writes.  If it's not, only write
112 		 * back what the user gives us.
113 		 */
114 		if (!test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags) &&
115 		    (to < f || from > t))
116 			goto flush_conflicting_write;
117 	}
118 
119 	*_page = folio_file_page(folio, pos / PAGE_SIZE);
120 	_leave(" = 0");
121 	return 0;
122 
123 	/* The previous write and this write aren't adjacent or overlapping, so
124 	 * flush the page out.
125 	 */
126 flush_conflicting_write:
127 	trace_afs_folio_dirty(vnode, tracepoint_string("confl"), folio);
128 	folio_unlock(folio);
129 
130 	ret = afs_flush_conflicting_write(mapping, folio);
131 	if (ret < 0)
132 		goto error;
133 
134 wait_for_writeback:
135 	ret = folio_wait_writeback_killable(folio);
136 	if (ret < 0)
137 		goto error;
138 
139 	ret = folio_lock_killable(folio);
140 	if (ret < 0)
141 		goto error;
142 	goto try_again;
143 
144 error:
145 	folio_put(folio);
146 	_leave(" = %d", ret);
147 	return ret;
148 }
149 
150 /*
151  * finalise part of a write to a page
152  */
afs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * subpage,void * fsdata)153 int afs_write_end(struct file *file, struct address_space *mapping,
154 		  loff_t pos, unsigned len, unsigned copied,
155 		  struct page *subpage, void *fsdata)
156 {
157 	struct folio *folio = page_folio(subpage);
158 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
159 	unsigned long priv;
160 	unsigned int f, from = offset_in_folio(folio, pos);
161 	unsigned int t, to = from + copied;
162 	loff_t i_size, write_end_pos;
163 
164 	_enter("{%llx:%llu},{%lx}",
165 	       vnode->fid.vid, vnode->fid.vnode, folio_index(folio));
166 
167 	if (!folio_test_uptodate(folio)) {
168 		if (copied < len) {
169 			copied = 0;
170 			goto out;
171 		}
172 
173 		folio_mark_uptodate(folio);
174 	}
175 
176 	if (copied == 0)
177 		goto out;
178 
179 	write_end_pos = pos + copied;
180 
181 	i_size = i_size_read(&vnode->netfs.inode);
182 	if (write_end_pos > i_size) {
183 		write_seqlock(&vnode->cb_lock);
184 		i_size = i_size_read(&vnode->netfs.inode);
185 		if (write_end_pos > i_size)
186 			afs_set_i_size(vnode, write_end_pos);
187 		write_sequnlock(&vnode->cb_lock);
188 		fscache_update_cookie(afs_vnode_cache(vnode), NULL, &write_end_pos);
189 	}
190 
191 	if (folio_test_private(folio)) {
192 		priv = (unsigned long)folio_get_private(folio);
193 		f = afs_folio_dirty_from(folio, priv);
194 		t = afs_folio_dirty_to(folio, priv);
195 		if (from < f)
196 			f = from;
197 		if (to > t)
198 			t = to;
199 		priv = afs_folio_dirty(folio, f, t);
200 		folio_change_private(folio, (void *)priv);
201 		trace_afs_folio_dirty(vnode, tracepoint_string("dirty+"), folio);
202 	} else {
203 		priv = afs_folio_dirty(folio, from, to);
204 		folio_attach_private(folio, (void *)priv);
205 		trace_afs_folio_dirty(vnode, tracepoint_string("dirty"), folio);
206 	}
207 
208 	if (folio_mark_dirty(folio))
209 		_debug("dirtied %lx", folio_index(folio));
210 
211 out:
212 	folio_unlock(folio);
213 	folio_put(folio);
214 	return copied;
215 }
216 
217 /*
218  * kill all the pages in the given range
219  */
afs_kill_pages(struct address_space * mapping,loff_t start,loff_t len)220 static void afs_kill_pages(struct address_space *mapping,
221 			   loff_t start, loff_t len)
222 {
223 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
224 	struct folio *folio;
225 	pgoff_t index = start / PAGE_SIZE;
226 	pgoff_t last = (start + len - 1) / PAGE_SIZE, next;
227 
228 	_enter("{%llx:%llu},%llx @%llx",
229 	       vnode->fid.vid, vnode->fid.vnode, len, start);
230 
231 	do {
232 		_debug("kill %lx (to %lx)", index, last);
233 
234 		folio = filemap_get_folio(mapping, index);
235 		if (IS_ERR(folio)) {
236 			next = index + 1;
237 			continue;
238 		}
239 
240 		next = folio_next_index(folio);
241 
242 		folio_clear_uptodate(folio);
243 		folio_end_writeback(folio);
244 		folio_lock(folio);
245 		generic_error_remove_page(mapping, &folio->page);
246 		folio_unlock(folio);
247 		folio_put(folio);
248 
249 	} while (index = next, index <= last);
250 
251 	_leave("");
252 }
253 
254 /*
255  * Redirty all the pages in a given range.
256  */
afs_redirty_pages(struct writeback_control * wbc,struct address_space * mapping,loff_t start,loff_t len)257 static void afs_redirty_pages(struct writeback_control *wbc,
258 			      struct address_space *mapping,
259 			      loff_t start, loff_t len)
260 {
261 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
262 	struct folio *folio;
263 	pgoff_t index = start / PAGE_SIZE;
264 	pgoff_t last = (start + len - 1) / PAGE_SIZE, next;
265 
266 	_enter("{%llx:%llu},%llx @%llx",
267 	       vnode->fid.vid, vnode->fid.vnode, len, start);
268 
269 	do {
270 		_debug("redirty %llx @%llx", len, start);
271 
272 		folio = filemap_get_folio(mapping, index);
273 		if (IS_ERR(folio)) {
274 			next = index + 1;
275 			continue;
276 		}
277 
278 		next = index + folio_nr_pages(folio);
279 		folio_redirty_for_writepage(wbc, folio);
280 		folio_end_writeback(folio);
281 		folio_put(folio);
282 	} while (index = next, index <= last);
283 
284 	_leave("");
285 }
286 
287 /*
288  * completion of write to server
289  */
afs_pages_written_back(struct afs_vnode * vnode,loff_t start,unsigned int len)290 static void afs_pages_written_back(struct afs_vnode *vnode, loff_t start, unsigned int len)
291 {
292 	struct address_space *mapping = vnode->netfs.inode.i_mapping;
293 	struct folio *folio;
294 	pgoff_t end;
295 
296 	XA_STATE(xas, &mapping->i_pages, start / PAGE_SIZE);
297 
298 	_enter("{%llx:%llu},{%x @%llx}",
299 	       vnode->fid.vid, vnode->fid.vnode, len, start);
300 
301 	rcu_read_lock();
302 
303 	end = (start + len - 1) / PAGE_SIZE;
304 	xas_for_each(&xas, folio, end) {
305 		if (!folio_test_writeback(folio)) {
306 			kdebug("bad %x @%llx page %lx %lx",
307 			       len, start, folio_index(folio), end);
308 			ASSERT(folio_test_writeback(folio));
309 		}
310 
311 		trace_afs_folio_dirty(vnode, tracepoint_string("clear"), folio);
312 		folio_detach_private(folio);
313 		folio_end_writeback(folio);
314 	}
315 
316 	rcu_read_unlock();
317 
318 	afs_prune_wb_keys(vnode);
319 	_leave("");
320 }
321 
322 /*
323  * Find a key to use for the writeback.  We cached the keys used to author the
324  * writes on the vnode.  *_wbk will contain the last writeback key used or NULL
325  * and we need to start from there if it's set.
326  */
afs_get_writeback_key(struct afs_vnode * vnode,struct afs_wb_key ** _wbk)327 static int afs_get_writeback_key(struct afs_vnode *vnode,
328 				 struct afs_wb_key **_wbk)
329 {
330 	struct afs_wb_key *wbk = NULL;
331 	struct list_head *p;
332 	int ret = -ENOKEY, ret2;
333 
334 	spin_lock(&vnode->wb_lock);
335 	if (*_wbk)
336 		p = (*_wbk)->vnode_link.next;
337 	else
338 		p = vnode->wb_keys.next;
339 
340 	while (p != &vnode->wb_keys) {
341 		wbk = list_entry(p, struct afs_wb_key, vnode_link);
342 		_debug("wbk %u", key_serial(wbk->key));
343 		ret2 = key_validate(wbk->key);
344 		if (ret2 == 0) {
345 			refcount_inc(&wbk->usage);
346 			_debug("USE WB KEY %u", key_serial(wbk->key));
347 			break;
348 		}
349 
350 		wbk = NULL;
351 		if (ret == -ENOKEY)
352 			ret = ret2;
353 		p = p->next;
354 	}
355 
356 	spin_unlock(&vnode->wb_lock);
357 	if (*_wbk)
358 		afs_put_wb_key(*_wbk);
359 	*_wbk = wbk;
360 	return 0;
361 }
362 
afs_store_data_success(struct afs_operation * op)363 static void afs_store_data_success(struct afs_operation *op)
364 {
365 	struct afs_vnode *vnode = op->file[0].vnode;
366 
367 	op->ctime = op->file[0].scb.status.mtime_client;
368 	afs_vnode_commit_status(op, &op->file[0]);
369 	if (op->error == 0) {
370 		if (!op->store.laundering)
371 			afs_pages_written_back(vnode, op->store.pos, op->store.size);
372 		afs_stat_v(vnode, n_stores);
373 		atomic_long_add(op->store.size, &afs_v2net(vnode)->n_store_bytes);
374 	}
375 }
376 
377 static const struct afs_operation_ops afs_store_data_operation = {
378 	.issue_afs_rpc	= afs_fs_store_data,
379 	.issue_yfs_rpc	= yfs_fs_store_data,
380 	.success	= afs_store_data_success,
381 };
382 
383 /*
384  * write to a file
385  */
afs_store_data(struct afs_vnode * vnode,struct iov_iter * iter,loff_t pos,bool laundering)386 static int afs_store_data(struct afs_vnode *vnode, struct iov_iter *iter, loff_t pos,
387 			  bool laundering)
388 {
389 	struct afs_operation *op;
390 	struct afs_wb_key *wbk = NULL;
391 	loff_t size = iov_iter_count(iter);
392 	int ret = -ENOKEY;
393 
394 	_enter("%s{%llx:%llu.%u},%llx,%llx",
395 	       vnode->volume->name,
396 	       vnode->fid.vid,
397 	       vnode->fid.vnode,
398 	       vnode->fid.unique,
399 	       size, pos);
400 
401 	ret = afs_get_writeback_key(vnode, &wbk);
402 	if (ret) {
403 		_leave(" = %d [no keys]", ret);
404 		return ret;
405 	}
406 
407 	op = afs_alloc_operation(wbk->key, vnode->volume);
408 	if (IS_ERR(op)) {
409 		afs_put_wb_key(wbk);
410 		return -ENOMEM;
411 	}
412 
413 	afs_op_set_vnode(op, 0, vnode);
414 	op->file[0].dv_delta = 1;
415 	op->file[0].modification = true;
416 	op->store.pos = pos;
417 	op->store.size = size;
418 	op->store.laundering = laundering;
419 	op->flags |= AFS_OPERATION_UNINTR;
420 	op->ops = &afs_store_data_operation;
421 
422 try_next_key:
423 	afs_begin_vnode_operation(op);
424 
425 	op->store.write_iter = iter;
426 	op->store.i_size = max(pos + size, vnode->netfs.remote_i_size);
427 	op->mtime = vnode->netfs.inode.i_mtime;
428 
429 	afs_wait_for_operation(op);
430 
431 	switch (op->error) {
432 	case -EACCES:
433 	case -EPERM:
434 	case -ENOKEY:
435 	case -EKEYEXPIRED:
436 	case -EKEYREJECTED:
437 	case -EKEYREVOKED:
438 		_debug("next");
439 
440 		ret = afs_get_writeback_key(vnode, &wbk);
441 		if (ret == 0) {
442 			key_put(op->key);
443 			op->key = key_get(wbk->key);
444 			goto try_next_key;
445 		}
446 		break;
447 	}
448 
449 	afs_put_wb_key(wbk);
450 	_leave(" = %d", op->error);
451 	return afs_put_operation(op);
452 }
453 
454 /*
455  * Extend the region to be written back to include subsequent contiguously
456  * dirty pages if possible, but don't sleep while doing so.
457  *
458  * If this page holds new content, then we can include filler zeros in the
459  * writeback.
460  */
afs_extend_writeback(struct address_space * mapping,struct afs_vnode * vnode,long * _count,loff_t start,loff_t max_len,bool new_content,bool caching,unsigned int * _len)461 static void afs_extend_writeback(struct address_space *mapping,
462 				 struct afs_vnode *vnode,
463 				 long *_count,
464 				 loff_t start,
465 				 loff_t max_len,
466 				 bool new_content,
467 				 bool caching,
468 				 unsigned int *_len)
469 {
470 	struct folio_batch fbatch;
471 	struct folio *folio;
472 	unsigned long priv;
473 	unsigned int psize, filler = 0;
474 	unsigned int f, t;
475 	loff_t len = *_len;
476 	pgoff_t index = (start + len) / PAGE_SIZE;
477 	bool stop = true;
478 	unsigned int i;
479 
480 	XA_STATE(xas, &mapping->i_pages, index);
481 	folio_batch_init(&fbatch);
482 
483 	do {
484 		/* Firstly, we gather up a batch of contiguous dirty pages
485 		 * under the RCU read lock - but we can't clear the dirty flags
486 		 * there if any of those pages are mapped.
487 		 */
488 		rcu_read_lock();
489 
490 		xas_for_each(&xas, folio, ULONG_MAX) {
491 			stop = true;
492 			if (xas_retry(&xas, folio))
493 				continue;
494 			if (xa_is_value(folio))
495 				break;
496 			if (folio_index(folio) != index)
497 				break;
498 
499 			if (!folio_try_get(folio)) {
500 				xas_reset(&xas);
501 				continue;
502 			}
503 
504 			/* Has the page moved or been split? */
505 			if (unlikely(folio != xas_reload(&xas))) {
506 				folio_put(folio);
507 				break;
508 			}
509 
510 			if (!folio_trylock(folio)) {
511 				folio_put(folio);
512 				break;
513 			}
514 			if (!folio_test_dirty(folio) ||
515 			    folio_test_writeback(folio) ||
516 			    folio_test_fscache(folio)) {
517 				folio_unlock(folio);
518 				folio_put(folio);
519 				break;
520 			}
521 
522 			psize = folio_size(folio);
523 			priv = (unsigned long)folio_get_private(folio);
524 			f = afs_folio_dirty_from(folio, priv);
525 			t = afs_folio_dirty_to(folio, priv);
526 			if (f != 0 && !new_content) {
527 				folio_unlock(folio);
528 				folio_put(folio);
529 				break;
530 			}
531 
532 			len += filler + t;
533 			filler = psize - t;
534 			if (len >= max_len || *_count <= 0)
535 				stop = true;
536 			else if (t == psize || new_content)
537 				stop = false;
538 
539 			index += folio_nr_pages(folio);
540 			if (!folio_batch_add(&fbatch, folio))
541 				break;
542 			if (stop)
543 				break;
544 		}
545 
546 		if (!stop)
547 			xas_pause(&xas);
548 		rcu_read_unlock();
549 
550 		/* Now, if we obtained any folios, we can shift them to being
551 		 * writable and mark them for caching.
552 		 */
553 		if (!folio_batch_count(&fbatch))
554 			break;
555 
556 		for (i = 0; i < folio_batch_count(&fbatch); i++) {
557 			folio = fbatch.folios[i];
558 			trace_afs_folio_dirty(vnode, tracepoint_string("store+"), folio);
559 
560 			if (!folio_clear_dirty_for_io(folio))
561 				BUG();
562 			if (folio_start_writeback(folio))
563 				BUG();
564 			afs_folio_start_fscache(caching, folio);
565 
566 			*_count -= folio_nr_pages(folio);
567 			folio_unlock(folio);
568 		}
569 
570 		folio_batch_release(&fbatch);
571 		cond_resched();
572 	} while (!stop);
573 
574 	*_len = len;
575 }
576 
577 /*
578  * Synchronously write back the locked page and any subsequent non-locked dirty
579  * pages.
580  */
afs_write_back_from_locked_folio(struct address_space * mapping,struct writeback_control * wbc,struct folio * folio,loff_t start,loff_t end)581 static ssize_t afs_write_back_from_locked_folio(struct address_space *mapping,
582 						struct writeback_control *wbc,
583 						struct folio *folio,
584 						loff_t start, loff_t end)
585 {
586 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
587 	struct iov_iter iter;
588 	unsigned long priv;
589 	unsigned int offset, to, len, max_len;
590 	loff_t i_size = i_size_read(&vnode->netfs.inode);
591 	bool new_content = test_bit(AFS_VNODE_NEW_CONTENT, &vnode->flags);
592 	bool caching = fscache_cookie_enabled(afs_vnode_cache(vnode));
593 	long count = wbc->nr_to_write;
594 	int ret;
595 
596 	_enter(",%lx,%llx-%llx", folio_index(folio), start, end);
597 
598 	if (folio_start_writeback(folio))
599 		BUG();
600 	afs_folio_start_fscache(caching, folio);
601 
602 	count -= folio_nr_pages(folio);
603 
604 	/* Find all consecutive lockable dirty pages that have contiguous
605 	 * written regions, stopping when we find a page that is not
606 	 * immediately lockable, is not dirty or is missing, or we reach the
607 	 * end of the range.
608 	 */
609 	priv = (unsigned long)folio_get_private(folio);
610 	offset = afs_folio_dirty_from(folio, priv);
611 	to = afs_folio_dirty_to(folio, priv);
612 	trace_afs_folio_dirty(vnode, tracepoint_string("store"), folio);
613 
614 	len = to - offset;
615 	start += offset;
616 	if (start < i_size) {
617 		/* Trim the write to the EOF; the extra data is ignored.  Also
618 		 * put an upper limit on the size of a single storedata op.
619 		 */
620 		max_len = 65536 * 4096;
621 		max_len = min_t(unsigned long long, max_len, end - start + 1);
622 		max_len = min_t(unsigned long long, max_len, i_size - start);
623 
624 		if (len < max_len &&
625 		    (to == folio_size(folio) || new_content))
626 			afs_extend_writeback(mapping, vnode, &count,
627 					     start, max_len, new_content,
628 					     caching, &len);
629 		len = min_t(loff_t, len, max_len);
630 	}
631 
632 	/* We now have a contiguous set of dirty pages, each with writeback
633 	 * set; the first page is still locked at this point, but all the rest
634 	 * have been unlocked.
635 	 */
636 	folio_unlock(folio);
637 
638 	if (start < i_size) {
639 		_debug("write back %x @%llx [%llx]", len, start, i_size);
640 
641 		/* Speculatively write to the cache.  We have to fix this up
642 		 * later if the store fails.
643 		 */
644 		afs_write_to_cache(vnode, start, len, i_size, caching);
645 
646 		iov_iter_xarray(&iter, ITER_SOURCE, &mapping->i_pages, start, len);
647 		ret = afs_store_data(vnode, &iter, start, false);
648 	} else {
649 		_debug("write discard %x @%llx [%llx]", len, start, i_size);
650 
651 		/* The dirty region was entirely beyond the EOF. */
652 		fscache_clear_page_bits(mapping, start, len, caching);
653 		afs_pages_written_back(vnode, start, len);
654 		ret = 0;
655 	}
656 
657 	switch (ret) {
658 	case 0:
659 		wbc->nr_to_write = count;
660 		ret = len;
661 		break;
662 
663 	default:
664 		pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret);
665 		fallthrough;
666 	case -EACCES:
667 	case -EPERM:
668 	case -ENOKEY:
669 	case -EKEYEXPIRED:
670 	case -EKEYREJECTED:
671 	case -EKEYREVOKED:
672 	case -ENETRESET:
673 		afs_redirty_pages(wbc, mapping, start, len);
674 		mapping_set_error(mapping, ret);
675 		break;
676 
677 	case -EDQUOT:
678 	case -ENOSPC:
679 		afs_redirty_pages(wbc, mapping, start, len);
680 		mapping_set_error(mapping, -ENOSPC);
681 		break;
682 
683 	case -EROFS:
684 	case -EIO:
685 	case -EREMOTEIO:
686 	case -EFBIG:
687 	case -ENOENT:
688 	case -ENOMEDIUM:
689 	case -ENXIO:
690 		trace_afs_file_error(vnode, ret, afs_file_error_writeback_fail);
691 		afs_kill_pages(mapping, start, len);
692 		mapping_set_error(mapping, ret);
693 		break;
694 	}
695 
696 	_leave(" = %d", ret);
697 	return ret;
698 }
699 
700 /*
701  * write a region of pages back to the server
702  */
afs_writepages_region(struct address_space * mapping,struct writeback_control * wbc,loff_t start,loff_t end,loff_t * _next,bool max_one_loop)703 static int afs_writepages_region(struct address_space *mapping,
704 				 struct writeback_control *wbc,
705 				 loff_t start, loff_t end, loff_t *_next,
706 				 bool max_one_loop)
707 {
708 	struct folio *folio;
709 	struct folio_batch fbatch;
710 	ssize_t ret;
711 	unsigned int i;
712 	int n, skips = 0;
713 
714 	_enter("%llx,%llx,", start, end);
715 	folio_batch_init(&fbatch);
716 
717 	do {
718 		pgoff_t index = start / PAGE_SIZE;
719 
720 		n = filemap_get_folios_tag(mapping, &index, end / PAGE_SIZE,
721 					PAGECACHE_TAG_DIRTY, &fbatch);
722 
723 		if (!n)
724 			break;
725 		for (i = 0; i < n; i++) {
726 			folio = fbatch.folios[i];
727 			start = folio_pos(folio); /* May regress with THPs */
728 
729 			_debug("wback %lx", folio_index(folio));
730 
731 			/* At this point we hold neither the i_pages lock nor the
732 			 * page lock: the page may be truncated or invalidated
733 			 * (changing page->mapping to NULL), or even swizzled
734 			 * back from swapper_space to tmpfs file mapping
735 			 */
736 try_again:
737 			if (wbc->sync_mode != WB_SYNC_NONE) {
738 				ret = folio_lock_killable(folio);
739 				if (ret < 0) {
740 					folio_batch_release(&fbatch);
741 					return ret;
742 				}
743 			} else {
744 				if (!folio_trylock(folio))
745 					continue;
746 			}
747 
748 			if (folio->mapping != mapping ||
749 			    !folio_test_dirty(folio)) {
750 				start += folio_size(folio);
751 				folio_unlock(folio);
752 				continue;
753 			}
754 
755 			if (folio_test_writeback(folio) ||
756 			    folio_test_fscache(folio)) {
757 				folio_unlock(folio);
758 				if (wbc->sync_mode != WB_SYNC_NONE) {
759 					folio_wait_writeback(folio);
760 #ifdef CONFIG_AFS_FSCACHE
761 					folio_wait_fscache(folio);
762 #endif
763 					goto try_again;
764 				}
765 
766 				start += folio_size(folio);
767 				if (wbc->sync_mode == WB_SYNC_NONE) {
768 					if (skips >= 5 || need_resched()) {
769 						*_next = start;
770 						folio_batch_release(&fbatch);
771 						_leave(" = 0 [%llx]", *_next);
772 						return 0;
773 					}
774 					skips++;
775 				}
776 				continue;
777 			}
778 
779 			if (!folio_clear_dirty_for_io(folio))
780 				BUG();
781 			ret = afs_write_back_from_locked_folio(mapping, wbc,
782 					folio, start, end);
783 			if (ret < 0) {
784 				_leave(" = %zd", ret);
785 				folio_batch_release(&fbatch);
786 				return ret;
787 			}
788 
789 			start += ret;
790 		}
791 
792 		folio_batch_release(&fbatch);
793 		cond_resched();
794 	} while (wbc->nr_to_write > 0);
795 
796 	*_next = start;
797 	_leave(" = 0 [%llx]", *_next);
798 	return 0;
799 }
800 
801 /*
802  * write some of the pending data back to the server
803  */
afs_writepages(struct address_space * mapping,struct writeback_control * wbc)804 int afs_writepages(struct address_space *mapping,
805 		   struct writeback_control *wbc)
806 {
807 	struct afs_vnode *vnode = AFS_FS_I(mapping->host);
808 	loff_t start, next;
809 	int ret;
810 
811 	_enter("");
812 
813 	/* We have to be careful as we can end up racing with setattr()
814 	 * truncating the pagecache since the caller doesn't take a lock here
815 	 * to prevent it.
816 	 */
817 	if (wbc->sync_mode == WB_SYNC_ALL)
818 		down_read(&vnode->validate_lock);
819 	else if (!down_read_trylock(&vnode->validate_lock))
820 		return 0;
821 
822 	if (wbc->range_cyclic) {
823 		start = mapping->writeback_index * PAGE_SIZE;
824 		ret = afs_writepages_region(mapping, wbc, start, LLONG_MAX,
825 					    &next, false);
826 		if (ret == 0) {
827 			mapping->writeback_index = next / PAGE_SIZE;
828 			if (start > 0 && wbc->nr_to_write > 0) {
829 				ret = afs_writepages_region(mapping, wbc, 0,
830 							    start, &next, false);
831 				if (ret == 0)
832 					mapping->writeback_index =
833 						next / PAGE_SIZE;
834 			}
835 		}
836 	} else if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) {
837 		ret = afs_writepages_region(mapping, wbc, 0, LLONG_MAX,
838 					    &next, false);
839 		if (wbc->nr_to_write > 0 && ret == 0)
840 			mapping->writeback_index = next / PAGE_SIZE;
841 	} else {
842 		ret = afs_writepages_region(mapping, wbc,
843 					    wbc->range_start, wbc->range_end,
844 					    &next, false);
845 	}
846 
847 	up_read(&vnode->validate_lock);
848 	_leave(" = %d", ret);
849 	return ret;
850 }
851 
852 /*
853  * write to an AFS file
854  */
afs_file_write(struct kiocb * iocb,struct iov_iter * from)855 ssize_t afs_file_write(struct kiocb *iocb, struct iov_iter *from)
856 {
857 	struct afs_vnode *vnode = AFS_FS_I(file_inode(iocb->ki_filp));
858 	struct afs_file *af = iocb->ki_filp->private_data;
859 	ssize_t result;
860 	size_t count = iov_iter_count(from);
861 
862 	_enter("{%llx:%llu},{%zu},",
863 	       vnode->fid.vid, vnode->fid.vnode, count);
864 
865 	if (IS_SWAPFILE(&vnode->netfs.inode)) {
866 		printk(KERN_INFO
867 		       "AFS: Attempt to write to active swap file!\n");
868 		return -EBUSY;
869 	}
870 
871 	if (!count)
872 		return 0;
873 
874 	result = afs_validate(vnode, af->key);
875 	if (result < 0)
876 		return result;
877 
878 	result = generic_file_write_iter(iocb, from);
879 
880 	_leave(" = %zd", result);
881 	return result;
882 }
883 
884 /*
885  * flush any dirty pages for this process, and check for write errors.
886  * - the return status from this call provides a reliable indication of
887  *   whether any write errors occurred for this process.
888  */
afs_fsync(struct file * file,loff_t start,loff_t end,int datasync)889 int afs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
890 {
891 	struct afs_vnode *vnode = AFS_FS_I(file_inode(file));
892 	struct afs_file *af = file->private_data;
893 	int ret;
894 
895 	_enter("{%llx:%llu},{n=%pD},%d",
896 	       vnode->fid.vid, vnode->fid.vnode, file,
897 	       datasync);
898 
899 	ret = afs_validate(vnode, af->key);
900 	if (ret < 0)
901 		return ret;
902 
903 	return file_write_and_wait_range(file, start, end);
904 }
905 
906 /*
907  * notification that a previously read-only page is about to become writable
908  * - if it returns an error, the caller will deliver a bus error signal
909  */
afs_page_mkwrite(struct vm_fault * vmf)910 vm_fault_t afs_page_mkwrite(struct vm_fault *vmf)
911 {
912 	struct folio *folio = page_folio(vmf->page);
913 	struct file *file = vmf->vma->vm_file;
914 	struct inode *inode = file_inode(file);
915 	struct afs_vnode *vnode = AFS_FS_I(inode);
916 	struct afs_file *af = file->private_data;
917 	unsigned long priv;
918 	vm_fault_t ret = VM_FAULT_RETRY;
919 
920 	_enter("{{%llx:%llu}},{%lx}", vnode->fid.vid, vnode->fid.vnode, folio_index(folio));
921 
922 	afs_validate(vnode, af->key);
923 
924 	sb_start_pagefault(inode->i_sb);
925 
926 	/* Wait for the page to be written to the cache before we allow it to
927 	 * be modified.  We then assume the entire page will need writing back.
928 	 */
929 #ifdef CONFIG_AFS_FSCACHE
930 	if (folio_test_fscache(folio) &&
931 	    folio_wait_fscache_killable(folio) < 0)
932 		goto out;
933 #endif
934 
935 	if (folio_wait_writeback_killable(folio))
936 		goto out;
937 
938 	if (folio_lock_killable(folio) < 0)
939 		goto out;
940 
941 	/* We mustn't change folio->private until writeback is complete as that
942 	 * details the portion of the page we need to write back and we might
943 	 * need to redirty the page if there's a problem.
944 	 */
945 	if (folio_wait_writeback_killable(folio) < 0) {
946 		folio_unlock(folio);
947 		goto out;
948 	}
949 
950 	priv = afs_folio_dirty(folio, 0, folio_size(folio));
951 	priv = afs_folio_dirty_mmapped(priv);
952 	if (folio_test_private(folio)) {
953 		folio_change_private(folio, (void *)priv);
954 		trace_afs_folio_dirty(vnode, tracepoint_string("mkwrite+"), folio);
955 	} else {
956 		folio_attach_private(folio, (void *)priv);
957 		trace_afs_folio_dirty(vnode, tracepoint_string("mkwrite"), folio);
958 	}
959 	file_update_time(file);
960 
961 	ret = VM_FAULT_LOCKED;
962 out:
963 	sb_end_pagefault(inode->i_sb);
964 	return ret;
965 }
966 
967 /*
968  * Prune the keys cached for writeback.  The caller must hold vnode->wb_lock.
969  */
afs_prune_wb_keys(struct afs_vnode * vnode)970 void afs_prune_wb_keys(struct afs_vnode *vnode)
971 {
972 	LIST_HEAD(graveyard);
973 	struct afs_wb_key *wbk, *tmp;
974 
975 	/* Discard unused keys */
976 	spin_lock(&vnode->wb_lock);
977 
978 	if (!mapping_tagged(&vnode->netfs.inode.i_data, PAGECACHE_TAG_WRITEBACK) &&
979 	    !mapping_tagged(&vnode->netfs.inode.i_data, PAGECACHE_TAG_DIRTY)) {
980 		list_for_each_entry_safe(wbk, tmp, &vnode->wb_keys, vnode_link) {
981 			if (refcount_read(&wbk->usage) == 1)
982 				list_move(&wbk->vnode_link, &graveyard);
983 		}
984 	}
985 
986 	spin_unlock(&vnode->wb_lock);
987 
988 	while (!list_empty(&graveyard)) {
989 		wbk = list_entry(graveyard.next, struct afs_wb_key, vnode_link);
990 		list_del(&wbk->vnode_link);
991 		afs_put_wb_key(wbk);
992 	}
993 }
994 
995 /*
996  * Clean up a page during invalidation.
997  */
afs_launder_folio(struct folio * folio)998 int afs_launder_folio(struct folio *folio)
999 {
1000 	struct afs_vnode *vnode = AFS_FS_I(folio_inode(folio));
1001 	struct iov_iter iter;
1002 	struct bio_vec bv;
1003 	unsigned long priv;
1004 	unsigned int f, t;
1005 	int ret = 0;
1006 
1007 	_enter("{%lx}", folio->index);
1008 
1009 	priv = (unsigned long)folio_get_private(folio);
1010 	if (folio_clear_dirty_for_io(folio)) {
1011 		f = 0;
1012 		t = folio_size(folio);
1013 		if (folio_test_private(folio)) {
1014 			f = afs_folio_dirty_from(folio, priv);
1015 			t = afs_folio_dirty_to(folio, priv);
1016 		}
1017 
1018 		bvec_set_folio(&bv, folio, t - f, f);
1019 		iov_iter_bvec(&iter, ITER_SOURCE, &bv, 1, bv.bv_len);
1020 
1021 		trace_afs_folio_dirty(vnode, tracepoint_string("launder"), folio);
1022 		ret = afs_store_data(vnode, &iter, folio_pos(folio) + f, true);
1023 	}
1024 
1025 	trace_afs_folio_dirty(vnode, tracepoint_string("laundered"), folio);
1026 	folio_detach_private(folio);
1027 	folio_wait_fscache(folio);
1028 	return ret;
1029 }
1030 
1031 /*
1032  * Deal with the completion of writing the data to the cache.
1033  */
afs_write_to_cache_done(void * priv,ssize_t transferred_or_error,bool was_async)1034 static void afs_write_to_cache_done(void *priv, ssize_t transferred_or_error,
1035 				    bool was_async)
1036 {
1037 	struct afs_vnode *vnode = priv;
1038 
1039 	if (IS_ERR_VALUE(transferred_or_error) &&
1040 	    transferred_or_error != -ENOBUFS)
1041 		afs_invalidate_cache(vnode, 0);
1042 }
1043 
1044 /*
1045  * Save the write to the cache also.
1046  */
afs_write_to_cache(struct afs_vnode * vnode,loff_t start,size_t len,loff_t i_size,bool caching)1047 static void afs_write_to_cache(struct afs_vnode *vnode,
1048 			       loff_t start, size_t len, loff_t i_size,
1049 			       bool caching)
1050 {
1051 	fscache_write_to_cache(afs_vnode_cache(vnode),
1052 			       vnode->netfs.inode.i_mapping, start, len, i_size,
1053 			       afs_write_to_cache_done, vnode, caching);
1054 }
1055