xref: /linux/fs/ubifs/file.c (revision 2775df6e5e324be9dc375f7db2c8d3042df72bbf)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * This file is part of UBIFS.
4  *
5  * Copyright (C) 2006-2008 Nokia Corporation.
6  *
7  * Authors: Artem Bityutskiy (Битюцкий Артём)
8  *          Adrian Hunter
9  */
10 
11 /*
12  * This file implements VFS file and inode operations for regular files, device
13  * nodes and symlinks as well as address space operations.
14  *
15  * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
16  * the page is dirty and is used for optimization purposes - dirty pages are
17  * not budgeted so the flag shows that 'ubifs_write_end()' should not release
18  * the budget for this page. The @PG_checked flag is set if full budgeting is
19  * required for the page e.g., when it corresponds to a file hole or it is
20  * beyond the file size. The budgeting is done in 'ubifs_write_begin()', because
21  * it is OK to fail in this function, and the budget is released in
22  * 'ubifs_write_end()'. So the @PG_private and @PG_checked flags carry
23  * information about how the page was budgeted, to make it possible to release
24  * the budget properly.
25  *
26  * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we
27  * implement. However, this is not true for 'ubifs_writepage()', which may be
28  * called with @i_mutex unlocked. For example, when flusher thread is doing
29  * background write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex.
30  * At "normal" work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g.
31  * in the "sys_write -> alloc_pages -> direct reclaim path". So, in
32  * 'ubifs_writepage()' we are only guaranteed that the page is locked.
33  *
34  * Similarly, @i_mutex is not always locked in 'ubifs_read_folio()', e.g., the
35  * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
36  * ondemand_readahead -> read_folio"). In case of readahead, @I_SYNC flag is not
37  * set as well. However, UBIFS disables readahead.
38  */
39 
40 #include "ubifs.h"
41 #include <linux/mount.h>
42 #include <linux/slab.h>
43 #include <linux/migrate.h>
44 
read_block(struct inode * inode,void * addr,unsigned int block,struct ubifs_data_node * dn)45 static int read_block(struct inode *inode, void *addr, unsigned int block,
46 		      struct ubifs_data_node *dn)
47 {
48 	struct ubifs_info *c = inode->i_sb->s_fs_info;
49 	int err, len, out_len;
50 	union ubifs_key key;
51 	unsigned int dlen;
52 
53 	data_key_init(c, &key, inode->i_ino, block);
54 	err = ubifs_tnc_lookup(c, &key, dn);
55 	if (err) {
56 		if (err == -ENOENT)
57 			/* Not found, so it must be a hole */
58 			memset(addr, 0, UBIFS_BLOCK_SIZE);
59 		return err;
60 	}
61 
62 	ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
63 		     ubifs_inode(inode)->creat_sqnum);
64 	len = le32_to_cpu(dn->size);
65 	if (len <= 0 || len > UBIFS_BLOCK_SIZE)
66 		goto dump;
67 
68 	dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
69 
70 	if (IS_ENCRYPTED(inode)) {
71 		err = ubifs_decrypt(inode, dn, &dlen, block);
72 		if (err)
73 			goto dump;
74 	}
75 
76 	out_len = UBIFS_BLOCK_SIZE;
77 	err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
78 			       le16_to_cpu(dn->compr_type));
79 	if (err || len != out_len)
80 		goto dump;
81 
82 	/*
83 	 * Data length can be less than a full block, even for blocks that are
84 	 * not the last in the file (e.g., as a result of making a hole and
85 	 * appending data). Ensure that the remainder is zeroed out.
86 	 */
87 	if (len < UBIFS_BLOCK_SIZE)
88 		memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
89 
90 	return 0;
91 
92 dump:
93 	ubifs_err(c, "bad data node (block %u, inode %lu)",
94 		  block, inode->i_ino);
95 	ubifs_dump_node(c, dn, UBIFS_MAX_DATA_NODE_SZ);
96 	return -EINVAL;
97 }
98 
do_readpage(struct folio * folio)99 static int do_readpage(struct folio *folio)
100 {
101 	void *addr;
102 	int err = 0, i;
103 	unsigned int block, beyond;
104 	struct ubifs_data_node *dn = NULL;
105 	struct inode *inode = folio->mapping->host;
106 	struct ubifs_info *c = inode->i_sb->s_fs_info;
107 	loff_t i_size = i_size_read(inode);
108 
109 	dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
110 		inode->i_ino, folio->index, i_size, folio->flags);
111 	ubifs_assert(c, !folio_test_checked(folio));
112 	ubifs_assert(c, !folio->private);
113 
114 	addr = kmap_local_folio(folio, 0);
115 
116 	block = folio->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
117 	beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
118 	if (block >= beyond) {
119 		/* Reading beyond inode */
120 		folio_set_checked(folio);
121 		addr = folio_zero_tail(folio, 0, addr);
122 		goto out;
123 	}
124 
125 	dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
126 	if (!dn) {
127 		err = -ENOMEM;
128 		goto out;
129 	}
130 
131 	i = 0;
132 	while (1) {
133 		int ret;
134 
135 		if (block >= beyond) {
136 			/* Reading beyond inode */
137 			err = -ENOENT;
138 			memset(addr, 0, UBIFS_BLOCK_SIZE);
139 		} else {
140 			ret = read_block(inode, addr, block, dn);
141 			if (ret) {
142 				err = ret;
143 				if (err != -ENOENT)
144 					break;
145 			} else if (block + 1 == beyond) {
146 				int dlen = le32_to_cpu(dn->size);
147 				int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
148 
149 				if (ilen && ilen < dlen)
150 					memset(addr + ilen, 0, dlen - ilen);
151 			}
152 		}
153 		if (++i >= (UBIFS_BLOCKS_PER_PAGE << folio_order(folio)))
154 			break;
155 		block += 1;
156 		addr += UBIFS_BLOCK_SIZE;
157 		if (folio_test_highmem(folio) && (offset_in_page(addr) == 0)) {
158 			kunmap_local(addr - UBIFS_BLOCK_SIZE);
159 			addr = kmap_local_folio(folio, i * UBIFS_BLOCK_SIZE);
160 		}
161 	}
162 
163 	if (err) {
164 		struct ubifs_info *c = inode->i_sb->s_fs_info;
165 		if (err == -ENOENT) {
166 			/* Not found, so it must be a hole */
167 			folio_set_checked(folio);
168 			dbg_gen("hole");
169 			err = 0;
170 		} else {
171 			ubifs_err(c, "cannot read page %lu of inode %lu, error %d",
172 				  folio->index, inode->i_ino, err);
173 		}
174 	}
175 
176 out:
177 	kfree(dn);
178 	if (!err)
179 		folio_mark_uptodate(folio);
180 	flush_dcache_folio(folio);
181 	kunmap_local(addr);
182 	return err;
183 }
184 
185 /**
186  * release_new_page_budget - release budget of a new page.
187  * @c: UBIFS file-system description object
188  *
189  * This is a helper function which releases budget corresponding to the budget
190  * of one new page of data.
191  */
release_new_page_budget(struct ubifs_info * c)192 static void release_new_page_budget(struct ubifs_info *c)
193 {
194 	struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
195 
196 	ubifs_release_budget(c, &req);
197 }
198 
199 /**
200  * release_existing_page_budget - release budget of an existing page.
201  * @c: UBIFS file-system description object
202  *
203  * This is a helper function which releases budget corresponding to the budget
204  * of changing one page of data which already exists on the flash media.
205  */
release_existing_page_budget(struct ubifs_info * c)206 static void release_existing_page_budget(struct ubifs_info *c)
207 {
208 	struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
209 
210 	ubifs_release_budget(c, &req);
211 }
212 
write_begin_slow(struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop)213 static int write_begin_slow(struct address_space *mapping,
214 			    loff_t pos, unsigned len, struct folio **foliop)
215 {
216 	struct inode *inode = mapping->host;
217 	struct ubifs_info *c = inode->i_sb->s_fs_info;
218 	pgoff_t index = pos >> PAGE_SHIFT;
219 	struct ubifs_budget_req req = { .new_page = 1 };
220 	int err, appending = !!(pos + len > inode->i_size);
221 	struct folio *folio;
222 
223 	dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
224 		inode->i_ino, pos, len, inode->i_size);
225 
226 	/*
227 	 * At the slow path we have to budget before locking the folio, because
228 	 * budgeting may force write-back, which would wait on locked folios and
229 	 * deadlock if we had the folio locked. At this point we do not know
230 	 * anything about the folio, so assume that this is a new folio which is
231 	 * written to a hole. This corresponds to largest budget. Later the
232 	 * budget will be amended if this is not true.
233 	 */
234 	if (appending)
235 		/* We are appending data, budget for inode change */
236 		req.dirtied_ino = 1;
237 
238 	err = ubifs_budget_space(c, &req);
239 	if (unlikely(err))
240 		return err;
241 
242 	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
243 			mapping_gfp_mask(mapping));
244 	if (IS_ERR(folio)) {
245 		ubifs_release_budget(c, &req);
246 		return PTR_ERR(folio);
247 	}
248 
249 	if (!folio_test_uptodate(folio)) {
250 		if (pos == folio_pos(folio) && len >= folio_size(folio))
251 			folio_set_checked(folio);
252 		else {
253 			err = do_readpage(folio);
254 			if (err) {
255 				folio_unlock(folio);
256 				folio_put(folio);
257 				ubifs_release_budget(c, &req);
258 				return err;
259 			}
260 		}
261 	}
262 
263 	if (folio->private)
264 		/*
265 		 * The folio is dirty, which means it was budgeted twice:
266 		 *   o first time the budget was allocated by the task which
267 		 *     made the folio dirty and set the private field;
268 		 *   o and then we budgeted for it for the second time at the
269 		 *     very beginning of this function.
270 		 *
271 		 * So what we have to do is to release the folio budget we
272 		 * allocated.
273 		 */
274 		release_new_page_budget(c);
275 	else if (!folio_test_checked(folio))
276 		/*
277 		 * We are changing a folio which already exists on the media.
278 		 * This means that changing the folio does not make the amount
279 		 * of indexing information larger, and this part of the budget
280 		 * which we have already acquired may be released.
281 		 */
282 		ubifs_convert_page_budget(c);
283 
284 	if (appending) {
285 		struct ubifs_inode *ui = ubifs_inode(inode);
286 
287 		/*
288 		 * 'ubifs_write_end()' is optimized from the fast-path part of
289 		 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
290 		 * if data is appended.
291 		 */
292 		mutex_lock(&ui->ui_mutex);
293 		if (ui->dirty)
294 			/*
295 			 * The inode is dirty already, so we may free the
296 			 * budget we allocated.
297 			 */
298 			ubifs_release_dirty_inode_budget(c, ui);
299 	}
300 
301 	*foliop = folio;
302 	return 0;
303 }
304 
305 /**
306  * allocate_budget - allocate budget for 'ubifs_write_begin()'.
307  * @c: UBIFS file-system description object
308  * @folio: folio to allocate budget for
309  * @ui: UBIFS inode object the page belongs to
310  * @appending: non-zero if the page is appended
311  *
312  * This is a helper function for 'ubifs_write_begin()' which allocates budget
313  * for the operation. The budget is allocated differently depending on whether
314  * this is appending, whether the page is dirty or not, and so on. This
315  * function leaves the @ui->ui_mutex locked in case of appending.
316  *
317  * Returns: %0 in case of success and %-ENOSPC in case of failure.
318  */
allocate_budget(struct ubifs_info * c,struct folio * folio,struct ubifs_inode * ui,int appending)319 static int allocate_budget(struct ubifs_info *c, struct folio *folio,
320 			   struct ubifs_inode *ui, int appending)
321 {
322 	struct ubifs_budget_req req = { .fast = 1 };
323 
324 	if (folio->private) {
325 		if (!appending)
326 			/*
327 			 * The folio is dirty and we are not appending, which
328 			 * means no budget is needed at all.
329 			 */
330 			return 0;
331 
332 		mutex_lock(&ui->ui_mutex);
333 		if (ui->dirty)
334 			/*
335 			 * The page is dirty and we are appending, so the inode
336 			 * has to be marked as dirty. However, it is already
337 			 * dirty, so we do not need any budget. We may return,
338 			 * but @ui->ui_mutex hast to be left locked because we
339 			 * should prevent write-back from flushing the inode
340 			 * and freeing the budget. The lock will be released in
341 			 * 'ubifs_write_end()'.
342 			 */
343 			return 0;
344 
345 		/*
346 		 * The page is dirty, we are appending, the inode is clean, so
347 		 * we need to budget the inode change.
348 		 */
349 		req.dirtied_ino = 1;
350 	} else {
351 		if (folio_test_checked(folio))
352 			/*
353 			 * The page corresponds to a hole and does not
354 			 * exist on the media. So changing it makes
355 			 * the amount of indexing information
356 			 * larger, and we have to budget for a new
357 			 * page.
358 			 */
359 			req.new_page = 1;
360 		else
361 			/*
362 			 * Not a hole, the change will not add any new
363 			 * indexing information, budget for page
364 			 * change.
365 			 */
366 			req.dirtied_page = 1;
367 
368 		if (appending) {
369 			mutex_lock(&ui->ui_mutex);
370 			if (!ui->dirty)
371 				/*
372 				 * The inode is clean but we will have to mark
373 				 * it as dirty because we are appending. This
374 				 * needs a budget.
375 				 */
376 				req.dirtied_ino = 1;
377 		}
378 	}
379 
380 	return ubifs_budget_space(c, &req);
381 }
382 
383 /*
384  * This function is called when a page of data is going to be written. Since
385  * the page of data will not necessarily go to the flash straight away, UBIFS
386  * has to reserve space on the media for it, which is done by means of
387  * budgeting.
388  *
389  * This is the hot-path of the file-system and we are trying to optimize it as
390  * much as possible. For this reasons it is split on 2 parts - slow and fast.
391  *
392  * There many budgeting cases:
393  *     o a new page is appended - we have to budget for a new page and for
394  *       changing the inode; however, if the inode is already dirty, there is
395  *       no need to budget for it;
396  *     o an existing clean page is changed - we have budget for it; if the page
397  *       does not exist on the media (a hole), we have to budget for a new
398  *       page; otherwise, we may budget for changing an existing page; the
399  *       difference between these cases is that changing an existing page does
400  *       not introduce anything new to the FS indexing information, so it does
401  *       not grow, and smaller budget is acquired in this case;
402  *     o an existing dirty page is changed - no need to budget at all, because
403  *       the page budget has been acquired by earlier, when the page has been
404  *       marked dirty.
405  *
406  * UBIFS budgeting sub-system may force write-back if it thinks there is no
407  * space to reserve. This imposes some locking restrictions and makes it
408  * impossible to take into account the above cases, and makes it impossible to
409  * optimize budgeting.
410  *
411  * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
412  * there is a plenty of flash space and the budget will be acquired quickly,
413  * without forcing write-back. The slow path does not make this assumption.
414  */
ubifs_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct folio ** foliop,void ** fsdata)415 static int ubifs_write_begin(struct file *file, struct address_space *mapping,
416 			     loff_t pos, unsigned len,
417 			     struct folio **foliop, void **fsdata)
418 {
419 	struct inode *inode = mapping->host;
420 	struct ubifs_info *c = inode->i_sb->s_fs_info;
421 	struct ubifs_inode *ui = ubifs_inode(inode);
422 	pgoff_t index = pos >> PAGE_SHIFT;
423 	int err, appending = !!(pos + len > inode->i_size);
424 	int skipped_read = 0;
425 	struct folio *folio;
426 
427 	ubifs_assert(c, ubifs_inode(inode)->ui_size == inode->i_size);
428 	ubifs_assert(c, !c->ro_media && !c->ro_mount);
429 
430 	if (unlikely(c->ro_error))
431 		return -EROFS;
432 
433 	/* Try out the fast-path part first */
434 	folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
435 			mapping_gfp_mask(mapping));
436 	if (IS_ERR(folio))
437 		return PTR_ERR(folio);
438 
439 	if (!folio_test_uptodate(folio)) {
440 		/* The page is not loaded from the flash */
441 		if (pos == folio_pos(folio) && len >= folio_size(folio)) {
442 			/*
443 			 * We change whole page so no need to load it. But we
444 			 * do not know whether this page exists on the media or
445 			 * not, so we assume the latter because it requires
446 			 * larger budget. The assumption is that it is better
447 			 * to budget a bit more than to read the page from the
448 			 * media. Thus, we are setting the @PG_checked flag
449 			 * here.
450 			 */
451 			folio_set_checked(folio);
452 			skipped_read = 1;
453 		} else {
454 			err = do_readpage(folio);
455 			if (err) {
456 				folio_unlock(folio);
457 				folio_put(folio);
458 				return err;
459 			}
460 		}
461 	}
462 
463 	err = allocate_budget(c, folio, ui, appending);
464 	if (unlikely(err)) {
465 		ubifs_assert(c, err == -ENOSPC);
466 		/*
467 		 * If we skipped reading the page because we were going to
468 		 * write all of it, then it is not up to date.
469 		 */
470 		if (skipped_read)
471 			folio_clear_checked(folio);
472 		/*
473 		 * Budgeting failed which means it would have to force
474 		 * write-back but didn't, because we set the @fast flag in the
475 		 * request. Write-back cannot be done now, while we have the
476 		 * page locked, because it would deadlock. Unlock and free
477 		 * everything and fall-back to slow-path.
478 		 */
479 		if (appending) {
480 			ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
481 			mutex_unlock(&ui->ui_mutex);
482 		}
483 		folio_unlock(folio);
484 		folio_put(folio);
485 
486 		return write_begin_slow(mapping, pos, len, foliop);
487 	}
488 
489 	/*
490 	 * Whee, we acquired budgeting quickly - without involving
491 	 * garbage-collection, committing or forcing write-back. We return
492 	 * with @ui->ui_mutex locked if we are appending pages, and unlocked
493 	 * otherwise. This is an optimization (slightly hacky though).
494 	 */
495 	*foliop = folio;
496 	return 0;
497 }
498 
499 /**
500  * cancel_budget - cancel budget.
501  * @c: UBIFS file-system description object
502  * @folio: folio to cancel budget for
503  * @ui: UBIFS inode object the page belongs to
504  * @appending: non-zero if the page is appended
505  *
506  * This is a helper function for a page write operation. It unlocks the
507  * @ui->ui_mutex in case of appending.
508  */
cancel_budget(struct ubifs_info * c,struct folio * folio,struct ubifs_inode * ui,int appending)509 static void cancel_budget(struct ubifs_info *c, struct folio *folio,
510 			  struct ubifs_inode *ui, int appending)
511 {
512 	if (appending) {
513 		if (!ui->dirty)
514 			ubifs_release_dirty_inode_budget(c, ui);
515 		mutex_unlock(&ui->ui_mutex);
516 	}
517 	if (!folio->private) {
518 		if (folio_test_checked(folio))
519 			release_new_page_budget(c);
520 		else
521 			release_existing_page_budget(c);
522 	}
523 }
524 
ubifs_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio,void * fsdata)525 static int ubifs_write_end(struct file *file, struct address_space *mapping,
526 			   loff_t pos, unsigned len, unsigned copied,
527 			   struct folio *folio, void *fsdata)
528 {
529 	struct inode *inode = mapping->host;
530 	struct ubifs_inode *ui = ubifs_inode(inode);
531 	struct ubifs_info *c = inode->i_sb->s_fs_info;
532 	loff_t end_pos = pos + len;
533 	int appending = !!(end_pos > inode->i_size);
534 
535 	dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
536 		inode->i_ino, pos, folio->index, len, copied, inode->i_size);
537 
538 	if (unlikely(copied < len && !folio_test_uptodate(folio))) {
539 		/*
540 		 * VFS copied less data to the folio than it intended and
541 		 * declared in its '->write_begin()' call via the @len
542 		 * argument. If the folio was not up-to-date,
543 		 * the 'ubifs_write_begin()' function did
544 		 * not load it from the media (for optimization reasons). This
545 		 * means that part of the folio contains garbage. So read the
546 		 * folio now.
547 		 */
548 		dbg_gen("copied %d instead of %d, read page and repeat",
549 			copied, len);
550 		cancel_budget(c, folio, ui, appending);
551 		folio_clear_checked(folio);
552 
553 		/*
554 		 * Return 0 to force VFS to repeat the whole operation, or the
555 		 * error code if 'do_readpage()' fails.
556 		 */
557 		copied = do_readpage(folio);
558 		goto out;
559 	}
560 
561 	if (len == folio_size(folio))
562 		folio_mark_uptodate(folio);
563 
564 	if (!folio->private) {
565 		folio_attach_private(folio, (void *)1);
566 		atomic_long_inc(&c->dirty_pg_cnt);
567 		filemap_dirty_folio(mapping, folio);
568 	}
569 
570 	if (appending) {
571 		i_size_write(inode, end_pos);
572 		ui->ui_size = end_pos;
573 		/*
574 		 * We do not set @I_DIRTY_PAGES (which means that
575 		 * the inode has dirty pages), this was done in
576 		 * filemap_dirty_folio().
577 		 */
578 		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
579 		ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
580 		mutex_unlock(&ui->ui_mutex);
581 	}
582 
583 out:
584 	folio_unlock(folio);
585 	folio_put(folio);
586 	return copied;
587 }
588 
589 /**
590  * populate_page - copy data nodes into a page for bulk-read.
591  * @c: UBIFS file-system description object
592  * @folio: folio
593  * @bu: bulk-read information
594  * @n: next zbranch slot
595  *
596  * Returns: %0 on success and a negative error code on failure.
597  */
populate_page(struct ubifs_info * c,struct folio * folio,struct bu_info * bu,int * n)598 static int populate_page(struct ubifs_info *c, struct folio *folio,
599 			 struct bu_info *bu, int *n)
600 {
601 	int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
602 	struct inode *inode = folio->mapping->host;
603 	loff_t i_size = i_size_read(inode);
604 	unsigned int page_block;
605 	void *addr, *zaddr;
606 	pgoff_t end_index;
607 
608 	dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
609 		inode->i_ino, folio->index, i_size, folio->flags);
610 
611 	addr = zaddr = kmap_local_folio(folio, 0);
612 
613 	end_index = (i_size - 1) >> PAGE_SHIFT;
614 	if (!i_size || folio->index > end_index) {
615 		hole = 1;
616 		addr = folio_zero_tail(folio, 0, addr);
617 		goto out_hole;
618 	}
619 
620 	page_block = folio->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
621 	while (1) {
622 		int err, len, out_len, dlen;
623 
624 		if (nn >= bu->cnt) {
625 			hole = 1;
626 			memset(addr, 0, UBIFS_BLOCK_SIZE);
627 		} else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
628 			struct ubifs_data_node *dn;
629 
630 			dn = bu->buf + (bu->zbranch[nn].offs - offs);
631 
632 			ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
633 				     ubifs_inode(inode)->creat_sqnum);
634 
635 			len = le32_to_cpu(dn->size);
636 			if (len <= 0 || len > UBIFS_BLOCK_SIZE)
637 				goto out_err;
638 
639 			dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
640 			out_len = UBIFS_BLOCK_SIZE;
641 
642 			if (IS_ENCRYPTED(inode)) {
643 				err = ubifs_decrypt(inode, dn, &dlen, page_block);
644 				if (err)
645 					goto out_err;
646 			}
647 
648 			err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
649 					       le16_to_cpu(dn->compr_type));
650 			if (err || len != out_len)
651 				goto out_err;
652 
653 			if (len < UBIFS_BLOCK_SIZE)
654 				memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
655 
656 			nn += 1;
657 			read = (i << UBIFS_BLOCK_SHIFT) + len;
658 		} else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
659 			nn += 1;
660 			continue;
661 		} else {
662 			hole = 1;
663 			memset(addr, 0, UBIFS_BLOCK_SIZE);
664 		}
665 		if (++i >= UBIFS_BLOCKS_PER_PAGE)
666 			break;
667 		addr += UBIFS_BLOCK_SIZE;
668 		page_block += 1;
669 		if (folio_test_highmem(folio) && (offset_in_page(addr) == 0)) {
670 			kunmap_local(addr - UBIFS_BLOCK_SIZE);
671 			addr = kmap_local_folio(folio, i * UBIFS_BLOCK_SIZE);
672 		}
673 	}
674 
675 	if (end_index == folio->index) {
676 		int len = i_size & (PAGE_SIZE - 1);
677 
678 		if (len && len < read)
679 			memset(zaddr + len, 0, read - len);
680 	}
681 
682 out_hole:
683 	if (hole) {
684 		folio_set_checked(folio);
685 		dbg_gen("hole");
686 	}
687 
688 	folio_mark_uptodate(folio);
689 	flush_dcache_folio(folio);
690 	kunmap_local(addr);
691 	*n = nn;
692 	return 0;
693 
694 out_err:
695 	flush_dcache_folio(folio);
696 	kunmap_local(addr);
697 	ubifs_err(c, "bad data node (block %u, inode %lu)",
698 		  page_block, inode->i_ino);
699 	return -EINVAL;
700 }
701 
702 /**
703  * ubifs_do_bulk_read - do bulk-read.
704  * @c: UBIFS file-system description object
705  * @bu: bulk-read information
706  * @folio1: first folio to read
707  *
708  * Returns: %1 if the bulk-read is done, otherwise %0 is returned.
709  */
ubifs_do_bulk_read(struct ubifs_info * c,struct bu_info * bu,struct folio * folio1)710 static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
711 			      struct folio *folio1)
712 {
713 	pgoff_t offset = folio1->index, end_index;
714 	struct address_space *mapping = folio1->mapping;
715 	struct inode *inode = mapping->host;
716 	struct ubifs_inode *ui = ubifs_inode(inode);
717 	int err, page_idx, page_cnt, ret = 0, n = 0;
718 	int allocate = bu->buf ? 0 : 1;
719 	loff_t isize;
720 	gfp_t ra_gfp_mask = readahead_gfp_mask(mapping) & ~__GFP_FS;
721 
722 	err = ubifs_tnc_get_bu_keys(c, bu);
723 	if (err)
724 		goto out_warn;
725 
726 	if (bu->eof) {
727 		/* Turn off bulk-read at the end of the file */
728 		ui->read_in_a_row = 1;
729 		ui->bulk_read = 0;
730 	}
731 
732 	page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
733 	if (!page_cnt) {
734 		/*
735 		 * This happens when there are multiple blocks per page and the
736 		 * blocks for the first page we are looking for, are not
737 		 * together. If all the pages were like this, bulk-read would
738 		 * reduce performance, so we turn it off for a while.
739 		 */
740 		goto out_bu_off;
741 	}
742 
743 	if (bu->cnt) {
744 		if (allocate) {
745 			/*
746 			 * Allocate bulk-read buffer depending on how many data
747 			 * nodes we are going to read.
748 			 */
749 			bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
750 				      bu->zbranch[bu->cnt - 1].len -
751 				      bu->zbranch[0].offs;
752 			ubifs_assert(c, bu->buf_len > 0);
753 			ubifs_assert(c, bu->buf_len <= c->leb_size);
754 			bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
755 			if (!bu->buf)
756 				goto out_bu_off;
757 		}
758 
759 		err = ubifs_tnc_bulk_read(c, bu);
760 		if (err)
761 			goto out_warn;
762 	}
763 
764 	err = populate_page(c, folio1, bu, &n);
765 	if (err)
766 		goto out_warn;
767 
768 	folio_unlock(folio1);
769 	ret = 1;
770 
771 	isize = i_size_read(inode);
772 	if (isize == 0)
773 		goto out_free;
774 	end_index = ((isize - 1) >> PAGE_SHIFT);
775 
776 	for (page_idx = 1; page_idx < page_cnt; page_idx++) {
777 		pgoff_t page_offset = offset + page_idx;
778 		struct folio *folio;
779 
780 		if (page_offset > end_index)
781 			break;
782 		folio = __filemap_get_folio(mapping, page_offset,
783 				 FGP_LOCK|FGP_ACCESSED|FGP_CREAT|FGP_NOWAIT,
784 				 ra_gfp_mask);
785 		if (IS_ERR(folio))
786 			break;
787 		if (!folio_test_uptodate(folio))
788 			err = populate_page(c, folio, bu, &n);
789 		folio_unlock(folio);
790 		folio_put(folio);
791 		if (err)
792 			break;
793 	}
794 
795 	ui->last_page_read = offset + page_idx - 1;
796 
797 out_free:
798 	if (allocate)
799 		kfree(bu->buf);
800 	return ret;
801 
802 out_warn:
803 	ubifs_warn(c, "ignoring error %d and skipping bulk-read", err);
804 	goto out_free;
805 
806 out_bu_off:
807 	ui->read_in_a_row = ui->bulk_read = 0;
808 	goto out_free;
809 }
810 
811 /**
812  * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
813  * @folio: folio from which to start bulk-read.
814  *
815  * Some flash media are capable of reading sequentially at faster rates. UBIFS
816  * bulk-read facility is designed to take advantage of that, by reading in one
817  * go consecutive data nodes that are also located consecutively in the same
818  * LEB.
819  *
820  * Returns: %1 if a bulk-read is done and %0 otherwise.
821  */
ubifs_bulk_read(struct folio * folio)822 static int ubifs_bulk_read(struct folio *folio)
823 {
824 	struct inode *inode = folio->mapping->host;
825 	struct ubifs_info *c = inode->i_sb->s_fs_info;
826 	struct ubifs_inode *ui = ubifs_inode(inode);
827 	pgoff_t index = folio->index, last_page_read = ui->last_page_read;
828 	struct bu_info *bu;
829 	int err = 0, allocated = 0;
830 
831 	ui->last_page_read = index;
832 	if (!c->bulk_read)
833 		return 0;
834 
835 	/*
836 	 * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
837 	 * so don't bother if we cannot lock the mutex.
838 	 */
839 	if (!mutex_trylock(&ui->ui_mutex))
840 		return 0;
841 
842 	if (index != last_page_read + 1) {
843 		/* Turn off bulk-read if we stop reading sequentially */
844 		ui->read_in_a_row = 1;
845 		if (ui->bulk_read)
846 			ui->bulk_read = 0;
847 		goto out_unlock;
848 	}
849 
850 	if (!ui->bulk_read) {
851 		ui->read_in_a_row += 1;
852 		if (ui->read_in_a_row < 3)
853 			goto out_unlock;
854 		/* Three reads in a row, so switch on bulk-read */
855 		ui->bulk_read = 1;
856 	}
857 
858 	/*
859 	 * If possible, try to use pre-allocated bulk-read information, which
860 	 * is protected by @c->bu_mutex.
861 	 */
862 	if (mutex_trylock(&c->bu_mutex))
863 		bu = &c->bu;
864 	else {
865 		bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
866 		if (!bu)
867 			goto out_unlock;
868 
869 		bu->buf = NULL;
870 		allocated = 1;
871 	}
872 
873 	bu->buf_len = c->max_bu_buf_len;
874 	data_key_init(c, &bu->key, inode->i_ino,
875 		      folio->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
876 	err = ubifs_do_bulk_read(c, bu, folio);
877 
878 	if (!allocated)
879 		mutex_unlock(&c->bu_mutex);
880 	else
881 		kfree(bu);
882 
883 out_unlock:
884 	mutex_unlock(&ui->ui_mutex);
885 	return err;
886 }
887 
ubifs_read_folio(struct file * file,struct folio * folio)888 static int ubifs_read_folio(struct file *file, struct folio *folio)
889 {
890 	if (ubifs_bulk_read(folio))
891 		return 0;
892 	do_readpage(folio);
893 	folio_unlock(folio);
894 	return 0;
895 }
896 
do_writepage(struct folio * folio,size_t len)897 static int do_writepage(struct folio *folio, size_t len)
898 {
899 	int err = 0, blen;
900 	unsigned int block;
901 	void *addr;
902 	size_t offset = 0;
903 	union ubifs_key key;
904 	struct inode *inode = folio->mapping->host;
905 	struct ubifs_info *c = inode->i_sb->s_fs_info;
906 
907 #ifdef UBIFS_DEBUG
908 	struct ubifs_inode *ui = ubifs_inode(inode);
909 	spin_lock(&ui->ui_lock);
910 	ubifs_assert(c, folio->index <= ui->synced_i_size >> PAGE_SHIFT);
911 	spin_unlock(&ui->ui_lock);
912 #endif
913 
914 	folio_start_writeback(folio);
915 
916 	addr = kmap_local_folio(folio, offset);
917 	block = folio->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
918 	for (;;) {
919 		blen = min_t(size_t, len, UBIFS_BLOCK_SIZE);
920 		data_key_init(c, &key, inode->i_ino, block);
921 		err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
922 		if (err)
923 			break;
924 		len -= blen;
925 		if (!len)
926 			break;
927 		block += 1;
928 		addr += blen;
929 		if (folio_test_highmem(folio) && !offset_in_page(addr)) {
930 			kunmap_local(addr - blen);
931 			offset += PAGE_SIZE;
932 			addr = kmap_local_folio(folio, offset);
933 		}
934 	}
935 	kunmap_local(addr);
936 	if (err) {
937 		mapping_set_error(folio->mapping, err);
938 		ubifs_err(c, "cannot write folio %lu of inode %lu, error %d",
939 			  folio->index, inode->i_ino, err);
940 		ubifs_ro_mode(c, err);
941 	}
942 
943 	ubifs_assert(c, folio->private != NULL);
944 	if (folio_test_checked(folio))
945 		release_new_page_budget(c);
946 	else
947 		release_existing_page_budget(c);
948 
949 	atomic_long_dec(&c->dirty_pg_cnt);
950 	folio_detach_private(folio);
951 	folio_clear_checked(folio);
952 
953 	folio_unlock(folio);
954 	folio_end_writeback(folio);
955 	return err;
956 }
957 
958 /*
959  * When writing-back dirty inodes, VFS first writes-back pages belonging to the
960  * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
961  * situation when a we have an inode with size 0, then a megabyte of data is
962  * appended to the inode, then write-back starts and flushes some amount of the
963  * dirty pages, the journal becomes full, commit happens and finishes, and then
964  * an unclean reboot happens. When the file system is mounted next time, the
965  * inode size would still be 0, but there would be many pages which are beyond
966  * the inode size, they would be indexed and consume flash space. Because the
967  * journal has been committed, the replay would not be able to detect this
968  * situation and correct the inode size. This means UBIFS would have to scan
969  * whole index and correct all inode sizes, which is long an unacceptable.
970  *
971  * To prevent situations like this, UBIFS writes pages back only if they are
972  * within the last synchronized inode size, i.e. the size which has been
973  * written to the flash media last time. Otherwise, UBIFS forces inode
974  * write-back, thus making sure the on-flash inode contains current inode size,
975  * and then keeps writing pages back.
976  *
977  * Some locking issues explanation. 'ubifs_writepage()' first is called with
978  * the page locked, and it locks @ui_mutex. However, write-back does take inode
979  * @i_mutex, which means other VFS operations may be run on this inode at the
980  * same time. And the problematic one is truncation to smaller size, from where
981  * we have to call 'truncate_setsize()', which first changes @inode->i_size,
982  * then drops the truncated pages. And while dropping the pages, it takes the
983  * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
984  * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
985  * This means that @inode->i_size is changed while @ui_mutex is unlocked.
986  *
987  * XXX(truncate): with the new truncate sequence this is not true anymore,
988  * and the calls to truncate_setsize can be move around freely.  They should
989  * be moved to the very end of the truncate sequence.
990  *
991  * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
992  * inode size. How do we do this if @inode->i_size may became smaller while we
993  * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
994  * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
995  * internally and updates it under @ui_mutex.
996  *
997  * Q: why we do not worry that if we race with truncation, we may end up with a
998  * situation when the inode is truncated while we are in the middle of
999  * 'do_writepage()', so we do write beyond inode size?
1000  * A: If we are in the middle of 'do_writepage()', truncation would be locked
1001  * on the page lock and it would not write the truncated inode node to the
1002  * journal before we have finished.
1003  */
ubifs_writepage(struct folio * folio,struct writeback_control * wbc,void * data)1004 static int ubifs_writepage(struct folio *folio, struct writeback_control *wbc,
1005 		void *data)
1006 {
1007 	struct inode *inode = folio->mapping->host;
1008 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1009 	struct ubifs_inode *ui = ubifs_inode(inode);
1010 	loff_t i_size =  i_size_read(inode), synced_i_size;
1011 	int err, len = folio_size(folio);
1012 
1013 	dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1014 		inode->i_ino, folio->index, folio->flags);
1015 	ubifs_assert(c, folio->private != NULL);
1016 
1017 	/* Is the folio fully outside @i_size? (truncate in progress) */
1018 	if (folio_pos(folio) >= i_size) {
1019 		err = 0;
1020 		goto out_unlock;
1021 	}
1022 
1023 	spin_lock(&ui->ui_lock);
1024 	synced_i_size = ui->synced_i_size;
1025 	spin_unlock(&ui->ui_lock);
1026 
1027 	/* Is the folio fully inside i_size? */
1028 	if (folio_pos(folio) + len <= i_size) {
1029 		if (folio_pos(folio) + len > synced_i_size) {
1030 			err = inode->i_sb->s_op->write_inode(inode, NULL);
1031 			if (err)
1032 				goto out_redirty;
1033 			/*
1034 			 * The inode has been written, but the write-buffer has
1035 			 * not been synchronized, so in case of an unclean
1036 			 * reboot we may end up with some pages beyond inode
1037 			 * size, but they would be in the journal (because
1038 			 * commit flushes write buffers) and recovery would deal
1039 			 * with this.
1040 			 */
1041 		}
1042 		return do_writepage(folio, len);
1043 	}
1044 
1045 	/*
1046 	 * The folio straddles @i_size. It must be zeroed out on each and every
1047 	 * writepage invocation because it may be mmapped. "A file is mapped
1048 	 * in multiples of the page size. For a file that is not a multiple of
1049 	 * the page size, the remaining memory is zeroed when mapped, and
1050 	 * writes to that region are not written out to the file."
1051 	 */
1052 	len = i_size - folio_pos(folio);
1053 	folio_zero_segment(folio, len, folio_size(folio));
1054 
1055 	if (i_size > synced_i_size) {
1056 		err = inode->i_sb->s_op->write_inode(inode, NULL);
1057 		if (err)
1058 			goto out_redirty;
1059 	}
1060 
1061 	return do_writepage(folio, len);
1062 out_redirty:
1063 	/*
1064 	 * folio_redirty_for_writepage() won't call ubifs_dirty_inode() because
1065 	 * it passes I_DIRTY_PAGES flag while calling __mark_inode_dirty(), so
1066 	 * there is no need to do space budget for dirty inode.
1067 	 */
1068 	folio_redirty_for_writepage(wbc, folio);
1069 out_unlock:
1070 	folio_unlock(folio);
1071 	return err;
1072 }
1073 
ubifs_writepages(struct address_space * mapping,struct writeback_control * wbc)1074 static int ubifs_writepages(struct address_space *mapping,
1075 		struct writeback_control *wbc)
1076 {
1077 	return write_cache_pages(mapping, wbc, ubifs_writepage, NULL);
1078 }
1079 
1080 /**
1081  * do_attr_changes - change inode attributes.
1082  * @inode: inode to change attributes for
1083  * @attr: describes attributes to change
1084  */
do_attr_changes(struct inode * inode,const struct iattr * attr)1085 static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1086 {
1087 	if (attr->ia_valid & ATTR_UID)
1088 		inode->i_uid = attr->ia_uid;
1089 	if (attr->ia_valid & ATTR_GID)
1090 		inode->i_gid = attr->ia_gid;
1091 	if (attr->ia_valid & ATTR_ATIME)
1092 		inode_set_atime_to_ts(inode, attr->ia_atime);
1093 	if (attr->ia_valid & ATTR_MTIME)
1094 		inode_set_mtime_to_ts(inode, attr->ia_mtime);
1095 	if (attr->ia_valid & ATTR_CTIME)
1096 		inode_set_ctime_to_ts(inode, attr->ia_ctime);
1097 	if (attr->ia_valid & ATTR_MODE) {
1098 		umode_t mode = attr->ia_mode;
1099 
1100 		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1101 			mode &= ~S_ISGID;
1102 		inode->i_mode = mode;
1103 	}
1104 }
1105 
1106 /**
1107  * do_truncation - truncate an inode.
1108  * @c: UBIFS file-system description object
1109  * @inode: inode to truncate
1110  * @attr: inode attribute changes description
1111  *
1112  * This function implements VFS '->setattr()' call when the inode is truncated
1113  * to a smaller size.
1114  *
1115  * Returns: %0 in case of success and a negative error code
1116  * in case of failure.
1117  */
do_truncation(struct ubifs_info * c,struct inode * inode,const struct iattr * attr)1118 static int do_truncation(struct ubifs_info *c, struct inode *inode,
1119 			 const struct iattr *attr)
1120 {
1121 	int err;
1122 	struct ubifs_budget_req req;
1123 	loff_t old_size = inode->i_size, new_size = attr->ia_size;
1124 	int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1125 	struct ubifs_inode *ui = ubifs_inode(inode);
1126 
1127 	dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1128 	memset(&req, 0, sizeof(struct ubifs_budget_req));
1129 
1130 	/*
1131 	 * If this is truncation to a smaller size, and we do not truncate on a
1132 	 * block boundary, budget for changing one data block, because the last
1133 	 * block will be re-written.
1134 	 */
1135 	if (new_size & (UBIFS_BLOCK_SIZE - 1))
1136 		req.dirtied_page = 1;
1137 
1138 	req.dirtied_ino = 1;
1139 	/* A funny way to budget for truncation node */
1140 	req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1141 	err = ubifs_budget_space(c, &req);
1142 	if (err) {
1143 		/*
1144 		 * Treat truncations to zero as deletion and always allow them,
1145 		 * just like we do for '->unlink()'.
1146 		 */
1147 		if (new_size || err != -ENOSPC)
1148 			return err;
1149 		budgeted = 0;
1150 	}
1151 
1152 	truncate_setsize(inode, new_size);
1153 
1154 	if (offset) {
1155 		pgoff_t index = new_size >> PAGE_SHIFT;
1156 		struct folio *folio;
1157 
1158 		folio = filemap_lock_folio(inode->i_mapping, index);
1159 		if (!IS_ERR(folio)) {
1160 			if (folio_test_dirty(folio)) {
1161 				/*
1162 				 * 'ubifs_jnl_truncate()' will try to truncate
1163 				 * the last data node, but it contains
1164 				 * out-of-date data because the page is dirty.
1165 				 * Write the page now, so that
1166 				 * 'ubifs_jnl_truncate()' will see an already
1167 				 * truncated (and up to date) data node.
1168 				 */
1169 				ubifs_assert(c, folio->private != NULL);
1170 
1171 				folio_clear_dirty_for_io(folio);
1172 				if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1173 					offset = offset_in_folio(folio,
1174 							new_size);
1175 				err = do_writepage(folio, offset);
1176 				folio_put(folio);
1177 				if (err)
1178 					goto out_budg;
1179 				/*
1180 				 * We could now tell 'ubifs_jnl_truncate()' not
1181 				 * to read the last block.
1182 				 */
1183 			} else {
1184 				/*
1185 				 * We could 'kmap()' the page and pass the data
1186 				 * to 'ubifs_jnl_truncate()' to save it from
1187 				 * having to read it.
1188 				 */
1189 				folio_unlock(folio);
1190 				folio_put(folio);
1191 			}
1192 		}
1193 	}
1194 
1195 	mutex_lock(&ui->ui_mutex);
1196 	ui->ui_size = inode->i_size;
1197 	/* Truncation changes inode [mc]time */
1198 	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1199 	/* Other attributes may be changed at the same time as well */
1200 	do_attr_changes(inode, attr);
1201 	err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1202 	mutex_unlock(&ui->ui_mutex);
1203 
1204 out_budg:
1205 	if (budgeted)
1206 		ubifs_release_budget(c, &req);
1207 	else {
1208 		c->bi.nospace = c->bi.nospace_rp = 0;
1209 		smp_wmb();
1210 	}
1211 	return err;
1212 }
1213 
1214 /**
1215  * do_setattr - change inode attributes.
1216  * @c: UBIFS file-system description object
1217  * @inode: inode to change attributes for
1218  * @attr: inode attribute changes description
1219  *
1220  * This function implements VFS '->setattr()' call for all cases except
1221  * truncations to smaller size.
1222  *
1223  * Returns: %0 in case of success and a negative
1224  * error code in case of failure.
1225  */
do_setattr(struct ubifs_info * c,struct inode * inode,const struct iattr * attr)1226 static int do_setattr(struct ubifs_info *c, struct inode *inode,
1227 		      const struct iattr *attr)
1228 {
1229 	int err, release;
1230 	loff_t new_size = attr->ia_size;
1231 	struct ubifs_inode *ui = ubifs_inode(inode);
1232 	struct ubifs_budget_req req = { .dirtied_ino = 1,
1233 				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1234 
1235 	err = ubifs_budget_space(c, &req);
1236 	if (err)
1237 		return err;
1238 
1239 	if (attr->ia_valid & ATTR_SIZE) {
1240 		dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1241 		truncate_setsize(inode, new_size);
1242 	}
1243 
1244 	mutex_lock(&ui->ui_mutex);
1245 	if (attr->ia_valid & ATTR_SIZE) {
1246 		/* Truncation changes inode [mc]time */
1247 		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1248 		/* 'truncate_setsize()' changed @i_size, update @ui_size */
1249 		ui->ui_size = inode->i_size;
1250 	}
1251 
1252 	do_attr_changes(inode, attr);
1253 
1254 	release = ui->dirty;
1255 	if (attr->ia_valid & ATTR_SIZE)
1256 		/*
1257 		 * Inode length changed, so we have to make sure
1258 		 * @I_DIRTY_DATASYNC is set.
1259 		 */
1260 		 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1261 	else
1262 		mark_inode_dirty_sync(inode);
1263 	mutex_unlock(&ui->ui_mutex);
1264 
1265 	if (release)
1266 		ubifs_release_budget(c, &req);
1267 	if (IS_SYNC(inode))
1268 		err = inode->i_sb->s_op->write_inode(inode, NULL);
1269 	return err;
1270 }
1271 
ubifs_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)1272 int ubifs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1273 		  struct iattr *attr)
1274 {
1275 	int err;
1276 	struct inode *inode = d_inode(dentry);
1277 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1278 
1279 	dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1280 		inode->i_ino, inode->i_mode, attr->ia_valid);
1281 	err = setattr_prepare(&nop_mnt_idmap, dentry, attr);
1282 	if (err)
1283 		return err;
1284 
1285 	err = dbg_check_synced_i_size(c, inode);
1286 	if (err)
1287 		return err;
1288 
1289 	err = fscrypt_prepare_setattr(dentry, attr);
1290 	if (err)
1291 		return err;
1292 
1293 	if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1294 		/* Truncation to a smaller size */
1295 		err = do_truncation(c, inode, attr);
1296 	else
1297 		err = do_setattr(c, inode, attr);
1298 
1299 	return err;
1300 }
1301 
ubifs_invalidate_folio(struct folio * folio,size_t offset,size_t length)1302 static void ubifs_invalidate_folio(struct folio *folio, size_t offset,
1303 				 size_t length)
1304 {
1305 	struct inode *inode = folio->mapping->host;
1306 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1307 
1308 	ubifs_assert(c, folio_test_private(folio));
1309 	if (offset || length < folio_size(folio))
1310 		/* Partial folio remains dirty */
1311 		return;
1312 
1313 	if (folio_test_checked(folio))
1314 		release_new_page_budget(c);
1315 	else
1316 		release_existing_page_budget(c);
1317 
1318 	atomic_long_dec(&c->dirty_pg_cnt);
1319 	folio_detach_private(folio);
1320 	folio_clear_checked(folio);
1321 }
1322 
ubifs_fsync(struct file * file,loff_t start,loff_t end,int datasync)1323 int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1324 {
1325 	struct inode *inode = file->f_mapping->host;
1326 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1327 	int err;
1328 
1329 	dbg_gen("syncing inode %lu", inode->i_ino);
1330 
1331 	if (c->ro_mount)
1332 		/*
1333 		 * For some really strange reasons VFS does not filter out
1334 		 * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1335 		 */
1336 		return 0;
1337 
1338 	err = file_write_and_wait_range(file, start, end);
1339 	if (err)
1340 		return err;
1341 	inode_lock(inode);
1342 
1343 	/* Synchronize the inode unless this is a 'datasync()' call. */
1344 	if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1345 		err = inode->i_sb->s_op->write_inode(inode, NULL);
1346 		if (err)
1347 			goto out;
1348 	}
1349 
1350 	/*
1351 	 * Nodes related to this inode may still sit in a write-buffer. Flush
1352 	 * them.
1353 	 */
1354 	err = ubifs_sync_wbufs_by_inode(c, inode);
1355 out:
1356 	inode_unlock(inode);
1357 	return err;
1358 }
1359 
1360 /**
1361  * mctime_update_needed - check if mtime or ctime update is needed.
1362  * @inode: the inode to do the check for
1363  * @now: current time
1364  *
1365  * This helper function checks if the inode mtime/ctime should be updated or
1366  * not. If current values of the time-stamps are within the UBIFS inode time
1367  * granularity, they are not updated. This is an optimization.
1368  *
1369  * Returns: %1 if time update is needed, %0 if not
1370  */
mctime_update_needed(const struct inode * inode,const struct timespec64 * now)1371 static inline int mctime_update_needed(const struct inode *inode,
1372 				       const struct timespec64 *now)
1373 {
1374 	struct timespec64 ctime = inode_get_ctime(inode);
1375 	struct timespec64 mtime = inode_get_mtime(inode);
1376 
1377 	if (!timespec64_equal(&mtime, now) || !timespec64_equal(&ctime, now))
1378 		return 1;
1379 	return 0;
1380 }
1381 
1382 /**
1383  * ubifs_update_time - update time of inode.
1384  * @inode: inode to update
1385  * @flags: time updating control flag determines updating
1386  *	    which time fields of @inode
1387  *
1388  * This function updates time of the inode.
1389  *
1390  * Returns: %0 for success or a negative error code otherwise.
1391  */
ubifs_update_time(struct inode * inode,int flags)1392 int ubifs_update_time(struct inode *inode, int flags)
1393 {
1394 	struct ubifs_inode *ui = ubifs_inode(inode);
1395 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1396 	struct ubifs_budget_req req = { .dirtied_ino = 1,
1397 			.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1398 	int err, release;
1399 
1400 	if (!IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT)) {
1401 		generic_update_time(inode, flags);
1402 		return 0;
1403 	}
1404 
1405 	err = ubifs_budget_space(c, &req);
1406 	if (err)
1407 		return err;
1408 
1409 	mutex_lock(&ui->ui_mutex);
1410 	inode_update_timestamps(inode, flags);
1411 	release = ui->dirty;
1412 	__mark_inode_dirty(inode, I_DIRTY_SYNC);
1413 	mutex_unlock(&ui->ui_mutex);
1414 	if (release)
1415 		ubifs_release_budget(c, &req);
1416 	return 0;
1417 }
1418 
1419 /**
1420  * update_mctime - update mtime and ctime of an inode.
1421  * @inode: inode to update
1422  *
1423  * This function updates mtime and ctime of the inode if it is not equivalent to
1424  * current time.
1425  *
1426  * Returns: %0 in case of success and a negative error code in
1427  * case of failure.
1428  */
update_mctime(struct inode * inode)1429 static int update_mctime(struct inode *inode)
1430 {
1431 	struct timespec64 now = current_time(inode);
1432 	struct ubifs_inode *ui = ubifs_inode(inode);
1433 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1434 
1435 	if (mctime_update_needed(inode, &now)) {
1436 		int err, release;
1437 		struct ubifs_budget_req req = { .dirtied_ino = 1,
1438 				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1439 
1440 		err = ubifs_budget_space(c, &req);
1441 		if (err)
1442 			return err;
1443 
1444 		mutex_lock(&ui->ui_mutex);
1445 		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1446 		release = ui->dirty;
1447 		mark_inode_dirty_sync(inode);
1448 		mutex_unlock(&ui->ui_mutex);
1449 		if (release)
1450 			ubifs_release_budget(c, &req);
1451 	}
1452 
1453 	return 0;
1454 }
1455 
ubifs_write_iter(struct kiocb * iocb,struct iov_iter * from)1456 static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1457 {
1458 	int err = update_mctime(file_inode(iocb->ki_filp));
1459 	if (err)
1460 		return err;
1461 
1462 	return generic_file_write_iter(iocb, from);
1463 }
1464 
ubifs_dirty_folio(struct address_space * mapping,struct folio * folio)1465 static bool ubifs_dirty_folio(struct address_space *mapping,
1466 		struct folio *folio)
1467 {
1468 	bool ret;
1469 	struct ubifs_info *c = mapping->host->i_sb->s_fs_info;
1470 
1471 	ret = filemap_dirty_folio(mapping, folio);
1472 	/*
1473 	 * An attempt to dirty a page without budgeting for it - should not
1474 	 * happen.
1475 	 */
1476 	ubifs_assert(c, ret == false);
1477 	return ret;
1478 }
1479 
ubifs_release_folio(struct folio * folio,gfp_t unused_gfp_flags)1480 static bool ubifs_release_folio(struct folio *folio, gfp_t unused_gfp_flags)
1481 {
1482 	struct inode *inode = folio->mapping->host;
1483 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1484 
1485 	if (folio_test_writeback(folio))
1486 		return false;
1487 
1488 	/*
1489 	 * Page is private but not dirty, weird? There is one condition
1490 	 * making it happened. ubifs_writepage skipped the page because
1491 	 * page index beyonds isize (for example. truncated by other
1492 	 * process named A), then the page is invalidated by fadvise64
1493 	 * syscall before being truncated by process A.
1494 	 */
1495 	ubifs_assert(c, folio_test_private(folio));
1496 	if (folio_test_checked(folio))
1497 		release_new_page_budget(c);
1498 	else
1499 		release_existing_page_budget(c);
1500 
1501 	atomic_long_dec(&c->dirty_pg_cnt);
1502 	folio_detach_private(folio);
1503 	folio_clear_checked(folio);
1504 	return true;
1505 }
1506 
1507 /*
1508  * mmap()d file has taken write protection fault and is being made writable.
1509  * UBIFS must ensure page is budgeted for.
1510  */
ubifs_vm_page_mkwrite(struct vm_fault * vmf)1511 static vm_fault_t ubifs_vm_page_mkwrite(struct vm_fault *vmf)
1512 {
1513 	struct folio *folio = page_folio(vmf->page);
1514 	struct inode *inode = file_inode(vmf->vma->vm_file);
1515 	struct ubifs_info *c = inode->i_sb->s_fs_info;
1516 	struct timespec64 now = current_time(inode);
1517 	struct ubifs_budget_req req = { .new_page = 1 };
1518 	int err, update_time;
1519 
1520 	dbg_gen("ino %lu, pg %lu, i_size %lld",	inode->i_ino, folio->index,
1521 		i_size_read(inode));
1522 	ubifs_assert(c, !c->ro_media && !c->ro_mount);
1523 
1524 	if (unlikely(c->ro_error))
1525 		return VM_FAULT_SIGBUS; /* -EROFS */
1526 
1527 	/*
1528 	 * We have not locked @folio so far so we may budget for changing the
1529 	 * folio. Note, we cannot do this after we locked the folio, because
1530 	 * budgeting may cause write-back which would cause deadlock.
1531 	 *
1532 	 * At the moment we do not know whether the folio is dirty or not, so we
1533 	 * assume that it is not and budget for a new folio. We could look at
1534 	 * the @PG_private flag and figure this out, but we may race with write
1535 	 * back and the folio state may change by the time we lock it, so this
1536 	 * would need additional care. We do not bother with this at the
1537 	 * moment, although it might be good idea to do. Instead, we allocate
1538 	 * budget for a new folio and amend it later on if the folio was in fact
1539 	 * dirty.
1540 	 *
1541 	 * The budgeting-related logic of this function is similar to what we
1542 	 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1543 	 * for more comments.
1544 	 */
1545 	update_time = mctime_update_needed(inode, &now);
1546 	if (update_time)
1547 		/*
1548 		 * We have to change inode time stamp which requires extra
1549 		 * budgeting.
1550 		 */
1551 		req.dirtied_ino = 1;
1552 
1553 	err = ubifs_budget_space(c, &req);
1554 	if (unlikely(err)) {
1555 		if (err == -ENOSPC)
1556 			ubifs_warn(c, "out of space for mmapped file (inode number %lu)",
1557 				   inode->i_ino);
1558 		return VM_FAULT_SIGBUS;
1559 	}
1560 
1561 	folio_lock(folio);
1562 	if (unlikely(folio->mapping != inode->i_mapping ||
1563 		     folio_pos(folio) >= i_size_read(inode))) {
1564 		/* Folio got truncated out from underneath us */
1565 		goto sigbus;
1566 	}
1567 
1568 	if (folio->private)
1569 		release_new_page_budget(c);
1570 	else {
1571 		if (!folio_test_checked(folio))
1572 			ubifs_convert_page_budget(c);
1573 		folio_attach_private(folio, (void *)1);
1574 		atomic_long_inc(&c->dirty_pg_cnt);
1575 		filemap_dirty_folio(folio->mapping, folio);
1576 	}
1577 
1578 	if (update_time) {
1579 		int release;
1580 		struct ubifs_inode *ui = ubifs_inode(inode);
1581 
1582 		mutex_lock(&ui->ui_mutex);
1583 		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1584 		release = ui->dirty;
1585 		mark_inode_dirty_sync(inode);
1586 		mutex_unlock(&ui->ui_mutex);
1587 		if (release)
1588 			ubifs_release_dirty_inode_budget(c, ui);
1589 	}
1590 
1591 	folio_wait_stable(folio);
1592 	return VM_FAULT_LOCKED;
1593 
1594 sigbus:
1595 	folio_unlock(folio);
1596 	ubifs_release_budget(c, &req);
1597 	return VM_FAULT_SIGBUS;
1598 }
1599 
1600 static const struct vm_operations_struct ubifs_file_vm_ops = {
1601 	.fault        = filemap_fault,
1602 	.map_pages = filemap_map_pages,
1603 	.page_mkwrite = ubifs_vm_page_mkwrite,
1604 };
1605 
ubifs_file_mmap(struct file * file,struct vm_area_struct * vma)1606 static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1607 {
1608 	int err;
1609 
1610 	err = generic_file_mmap(file, vma);
1611 	if (err)
1612 		return err;
1613 	vma->vm_ops = &ubifs_file_vm_ops;
1614 
1615 	if (IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT))
1616 		file_accessed(file);
1617 
1618 	return 0;
1619 }
1620 
ubifs_get_link(struct dentry * dentry,struct inode * inode,struct delayed_call * done)1621 static const char *ubifs_get_link(struct dentry *dentry,
1622 					    struct inode *inode,
1623 					    struct delayed_call *done)
1624 {
1625 	struct ubifs_inode *ui = ubifs_inode(inode);
1626 
1627 	if (!IS_ENCRYPTED(inode))
1628 		return ui->data;
1629 
1630 	if (!dentry)
1631 		return ERR_PTR(-ECHILD);
1632 
1633 	return fscrypt_get_symlink(inode, ui->data, ui->data_len, done);
1634 }
1635 
ubifs_symlink_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)1636 static int ubifs_symlink_getattr(struct mnt_idmap *idmap,
1637 				 const struct path *path, struct kstat *stat,
1638 				 u32 request_mask, unsigned int query_flags)
1639 {
1640 	ubifs_getattr(idmap, path, stat, request_mask, query_flags);
1641 
1642 	if (IS_ENCRYPTED(d_inode(path->dentry)))
1643 		return fscrypt_symlink_getattr(path, stat);
1644 	return 0;
1645 }
1646 
1647 const struct address_space_operations ubifs_file_address_operations = {
1648 	.read_folio     = ubifs_read_folio,
1649 	.writepages     = ubifs_writepages,
1650 	.write_begin    = ubifs_write_begin,
1651 	.write_end      = ubifs_write_end,
1652 	.invalidate_folio = ubifs_invalidate_folio,
1653 	.dirty_folio	= ubifs_dirty_folio,
1654 	.migrate_folio	= filemap_migrate_folio,
1655 	.release_folio	= ubifs_release_folio,
1656 };
1657 
1658 const struct inode_operations ubifs_file_inode_operations = {
1659 	.setattr     = ubifs_setattr,
1660 	.getattr     = ubifs_getattr,
1661 	.listxattr   = ubifs_listxattr,
1662 	.update_time = ubifs_update_time,
1663 	.fileattr_get = ubifs_fileattr_get,
1664 	.fileattr_set = ubifs_fileattr_set,
1665 };
1666 
1667 const struct inode_operations ubifs_symlink_inode_operations = {
1668 	.get_link    = ubifs_get_link,
1669 	.setattr     = ubifs_setattr,
1670 	.getattr     = ubifs_symlink_getattr,
1671 	.listxattr   = ubifs_listxattr,
1672 	.update_time = ubifs_update_time,
1673 };
1674 
1675 const struct file_operations ubifs_file_operations = {
1676 	.llseek         = generic_file_llseek,
1677 	.read_iter      = generic_file_read_iter,
1678 	.write_iter     = ubifs_write_iter,
1679 	.mmap           = ubifs_file_mmap,
1680 	.fsync          = ubifs_fsync,
1681 	.unlocked_ioctl = ubifs_ioctl,
1682 	.splice_read	= filemap_splice_read,
1683 	.splice_write	= iter_file_splice_write,
1684 	.open		= fscrypt_file_open,
1685 #ifdef CONFIG_COMPAT
1686 	.compat_ioctl   = ubifs_compat_ioctl,
1687 #endif
1688 };
1689