xref: /openbmc/linux/fs/erofs/super.c (revision 8d8a09b093d7073465c824f74caf315c073d3875) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017-2018 HUAWEI, Inc.
4  *             http://www.huawei.com/
5  * Created by Gao Xiang <gaoxiang25@huawei.com>
6  */
7 #include <linux/module.h>
8 #include <linux/buffer_head.h>
9 #include <linux/statfs.h>
10 #include <linux/parser.h>
11 #include <linux/seq_file.h>
12 #include "xattr.h"
13 
14 #define CREATE_TRACE_POINTS
15 #include <trace/events/erofs.h>
16 
17 static struct kmem_cache *erofs_inode_cachep __read_mostly;
18 
19 static void init_once(void *ptr)
20 {
21 	struct erofs_vnode *vi = ptr;
22 
23 	inode_init_once(&vi->vfs_inode);
24 }
25 
26 static int __init erofs_init_inode_cache(void)
27 {
28 	erofs_inode_cachep = kmem_cache_create("erofs_inode",
29 					       sizeof(struct erofs_vnode), 0,
30 					       SLAB_RECLAIM_ACCOUNT,
31 					       init_once);
32 
33 	return erofs_inode_cachep ? 0 : -ENOMEM;
34 }
35 
36 static void erofs_exit_inode_cache(void)
37 {
38 	kmem_cache_destroy(erofs_inode_cachep);
39 }
40 
41 static struct inode *alloc_inode(struct super_block *sb)
42 {
43 	struct erofs_vnode *vi =
44 		kmem_cache_alloc(erofs_inode_cachep, GFP_KERNEL);
45 
46 	if (!vi)
47 		return NULL;
48 
49 	/* zero out everything except vfs_inode */
50 	memset(vi, 0, offsetof(struct erofs_vnode, vfs_inode));
51 	return &vi->vfs_inode;
52 }
53 
54 static void free_inode(struct inode *inode)
55 {
56 	struct erofs_vnode *vi = EROFS_V(inode);
57 
58 	/* be careful RCU symlink path (see ext4_inode_info->i_data)! */
59 	if (is_inode_fast_symlink(inode))
60 		kfree(inode->i_link);
61 
62 	kfree(vi->xattr_shared_xattrs);
63 
64 	kmem_cache_free(erofs_inode_cachep, vi);
65 }
66 
67 static bool check_layout_compatibility(struct super_block *sb,
68 				       struct erofs_super_block *layout)
69 {
70 	const unsigned int requirements = le32_to_cpu(layout->requirements);
71 
72 	EROFS_SB(sb)->requirements = requirements;
73 
74 	/* check if current kernel meets all mandatory requirements */
75 	if (requirements & (~EROFS_ALL_REQUIREMENTS)) {
76 		errln("unidentified requirements %x, please upgrade kernel version",
77 		      requirements & ~EROFS_ALL_REQUIREMENTS);
78 		return false;
79 	}
80 	return true;
81 }
82 
83 static int superblock_read(struct super_block *sb)
84 {
85 	struct erofs_sb_info *sbi;
86 	struct buffer_head *bh;
87 	struct erofs_super_block *layout;
88 	unsigned int blkszbits;
89 	int ret;
90 
91 	bh = sb_bread(sb, 0);
92 
93 	if (!bh) {
94 		errln("cannot read erofs superblock");
95 		return -EIO;
96 	}
97 
98 	sbi = EROFS_SB(sb);
99 	layout = (struct erofs_super_block *)((u8 *)bh->b_data
100 		 + EROFS_SUPER_OFFSET);
101 
102 	ret = -EINVAL;
103 	if (le32_to_cpu(layout->magic) != EROFS_SUPER_MAGIC_V1) {
104 		errln("cannot find valid erofs superblock");
105 		goto out;
106 	}
107 
108 	blkszbits = layout->blkszbits;
109 	/* 9(512 bytes) + LOG_SECTORS_PER_BLOCK == LOG_BLOCK_SIZE */
110 	if (blkszbits != LOG_BLOCK_SIZE) {
111 		errln("blksize %u isn't supported on this platform",
112 		      1 << blkszbits);
113 		goto out;
114 	}
115 
116 	if (!check_layout_compatibility(sb, layout))
117 		goto out;
118 
119 	sbi->blocks = le32_to_cpu(layout->blocks);
120 	sbi->meta_blkaddr = le32_to_cpu(layout->meta_blkaddr);
121 #ifdef CONFIG_EROFS_FS_XATTR
122 	sbi->xattr_blkaddr = le32_to_cpu(layout->xattr_blkaddr);
123 #endif
124 	sbi->islotbits = ffs(sizeof(struct erofs_inode_v1)) - 1;
125 	sbi->root_nid = le16_to_cpu(layout->root_nid);
126 	sbi->inos = le64_to_cpu(layout->inos);
127 
128 	sbi->build_time = le64_to_cpu(layout->build_time);
129 	sbi->build_time_nsec = le32_to_cpu(layout->build_time_nsec);
130 
131 	memcpy(&sb->s_uuid, layout->uuid, sizeof(layout->uuid));
132 
133 	ret = strscpy(sbi->volume_name, layout->volume_name,
134 		      sizeof(layout->volume_name));
135 	if (ret < 0) {	/* -E2BIG */
136 		errln("bad volume name without NIL terminator");
137 		ret = -EFSCORRUPTED;
138 		goto out;
139 	}
140 	ret = 0;
141 out:
142 	brelse(bh);
143 	return ret;
144 }
145 
146 #ifdef CONFIG_EROFS_FAULT_INJECTION
147 const char *erofs_fault_name[FAULT_MAX] = {
148 	[FAULT_KMALLOC]		= "kmalloc",
149 	[FAULT_READ_IO]		= "read IO error",
150 };
151 
152 static void __erofs_build_fault_attr(struct erofs_sb_info *sbi,
153 				     unsigned int rate)
154 {
155 	struct erofs_fault_info *ffi = &sbi->fault_info;
156 
157 	if (rate) {
158 		atomic_set(&ffi->inject_ops, 0);
159 		ffi->inject_rate = rate;
160 		ffi->inject_type = (1 << FAULT_MAX) - 1;
161 	} else {
162 		memset(ffi, 0, sizeof(struct erofs_fault_info));
163 	}
164 
165 	set_opt(sbi, FAULT_INJECTION);
166 }
167 
168 static int erofs_build_fault_attr(struct erofs_sb_info *sbi,
169 				  substring_t *args)
170 {
171 	int rate = 0;
172 
173 	if (args->from && match_int(args, &rate))
174 		return -EINVAL;
175 
176 	__erofs_build_fault_attr(sbi, rate);
177 	return 0;
178 }
179 
180 static unsigned int erofs_get_fault_rate(struct erofs_sb_info *sbi)
181 {
182 	return sbi->fault_info.inject_rate;
183 }
184 #else
185 static void __erofs_build_fault_attr(struct erofs_sb_info *sbi,
186 				     unsigned int rate)
187 {
188 }
189 
190 static int erofs_build_fault_attr(struct erofs_sb_info *sbi,
191 				  substring_t *args)
192 {
193 	infoln("fault_injection options not supported");
194 	return 0;
195 }
196 
197 static unsigned int erofs_get_fault_rate(struct erofs_sb_info *sbi)
198 {
199 	return 0;
200 }
201 #endif
202 
203 #ifdef CONFIG_EROFS_FS_ZIP
204 static int erofs_build_cache_strategy(struct erofs_sb_info *sbi,
205 				      substring_t *args)
206 {
207 	const char *cs = match_strdup(args);
208 	int err = 0;
209 
210 	if (!cs) {
211 		errln("Not enough memory to store cache strategy");
212 		return -ENOMEM;
213 	}
214 
215 	if (!strcmp(cs, "disabled")) {
216 		sbi->cache_strategy = EROFS_ZIP_CACHE_DISABLED;
217 	} else if (!strcmp(cs, "readahead")) {
218 		sbi->cache_strategy = EROFS_ZIP_CACHE_READAHEAD;
219 	} else if (!strcmp(cs, "readaround")) {
220 		sbi->cache_strategy = EROFS_ZIP_CACHE_READAROUND;
221 	} else {
222 		errln("Unrecognized cache strategy \"%s\"", cs);
223 		err = -EINVAL;
224 	}
225 	kfree(cs);
226 	return err;
227 }
228 #else
229 static int erofs_build_cache_strategy(struct erofs_sb_info *sbi,
230 				      substring_t *args)
231 {
232 	infoln("EROFS compression is disabled, so cache strategy is ignored");
233 	return 0;
234 }
235 #endif
236 
237 /* set up default EROFS parameters */
238 static void default_options(struct erofs_sb_info *sbi)
239 {
240 #ifdef CONFIG_EROFS_FS_ZIP
241 	sbi->cache_strategy = EROFS_ZIP_CACHE_READAROUND;
242 	sbi->max_sync_decompress_pages = 3;
243 #endif
244 #ifdef CONFIG_EROFS_FS_XATTR
245 	set_opt(sbi, XATTR_USER);
246 #endif
247 #ifdef CONFIG_EROFS_FS_POSIX_ACL
248 	set_opt(sbi, POSIX_ACL);
249 #endif
250 }
251 
252 enum {
253 	Opt_user_xattr,
254 	Opt_nouser_xattr,
255 	Opt_acl,
256 	Opt_noacl,
257 	Opt_fault_injection,
258 	Opt_cache_strategy,
259 	Opt_err
260 };
261 
262 static match_table_t erofs_tokens = {
263 	{Opt_user_xattr, "user_xattr"},
264 	{Opt_nouser_xattr, "nouser_xattr"},
265 	{Opt_acl, "acl"},
266 	{Opt_noacl, "noacl"},
267 	{Opt_fault_injection, "fault_injection=%u"},
268 	{Opt_cache_strategy, "cache_strategy=%s"},
269 	{Opt_err, NULL}
270 };
271 
272 static int parse_options(struct super_block *sb, char *options)
273 {
274 	substring_t args[MAX_OPT_ARGS];
275 	char *p;
276 	int err;
277 
278 	if (!options)
279 		return 0;
280 
281 	while ((p = strsep(&options, ","))) {
282 		int token;
283 
284 		if (!*p)
285 			continue;
286 
287 		args[0].to = args[0].from = NULL;
288 		token = match_token(p, erofs_tokens, args);
289 
290 		switch (token) {
291 #ifdef CONFIG_EROFS_FS_XATTR
292 		case Opt_user_xattr:
293 			set_opt(EROFS_SB(sb), XATTR_USER);
294 			break;
295 		case Opt_nouser_xattr:
296 			clear_opt(EROFS_SB(sb), XATTR_USER);
297 			break;
298 #else
299 		case Opt_user_xattr:
300 			infoln("user_xattr options not supported");
301 			break;
302 		case Opt_nouser_xattr:
303 			infoln("nouser_xattr options not supported");
304 			break;
305 #endif
306 #ifdef CONFIG_EROFS_FS_POSIX_ACL
307 		case Opt_acl:
308 			set_opt(EROFS_SB(sb), POSIX_ACL);
309 			break;
310 		case Opt_noacl:
311 			clear_opt(EROFS_SB(sb), POSIX_ACL);
312 			break;
313 #else
314 		case Opt_acl:
315 			infoln("acl options not supported");
316 			break;
317 		case Opt_noacl:
318 			infoln("noacl options not supported");
319 			break;
320 #endif
321 		case Opt_fault_injection:
322 			err = erofs_build_fault_attr(EROFS_SB(sb), args);
323 			if (err)
324 				return err;
325 			break;
326 		case Opt_cache_strategy:
327 			err = erofs_build_cache_strategy(EROFS_SB(sb), args);
328 			if (err)
329 				return err;
330 			break;
331 		default:
332 			errln("Unrecognized mount option \"%s\" or missing value", p);
333 			return -EINVAL;
334 		}
335 	}
336 	return 0;
337 }
338 
339 #ifdef CONFIG_EROFS_FS_ZIP
340 static const struct address_space_operations managed_cache_aops;
341 
342 static int managed_cache_releasepage(struct page *page, gfp_t gfp_mask)
343 {
344 	int ret = 1;	/* 0 - busy */
345 	struct address_space *const mapping = page->mapping;
346 
347 	DBG_BUGON(!PageLocked(page));
348 	DBG_BUGON(mapping->a_ops != &managed_cache_aops);
349 
350 	if (PagePrivate(page))
351 		ret = erofs_try_to_free_cached_page(mapping, page);
352 
353 	return ret;
354 }
355 
356 static void managed_cache_invalidatepage(struct page *page,
357 					 unsigned int offset,
358 					 unsigned int length)
359 {
360 	const unsigned int stop = length + offset;
361 
362 	DBG_BUGON(!PageLocked(page));
363 
364 	/* Check for potential overflow in debug mode */
365 	DBG_BUGON(stop > PAGE_SIZE || stop < length);
366 
367 	if (offset == 0 && stop == PAGE_SIZE)
368 		while (!managed_cache_releasepage(page, GFP_NOFS))
369 			cond_resched();
370 }
371 
372 static const struct address_space_operations managed_cache_aops = {
373 	.releasepage = managed_cache_releasepage,
374 	.invalidatepage = managed_cache_invalidatepage,
375 };
376 
377 static int erofs_init_managed_cache(struct super_block *sb)
378 {
379 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
380 	struct inode *const inode = new_inode(sb);
381 
382 	if (!inode)
383 		return -ENOMEM;
384 
385 	set_nlink(inode, 1);
386 	inode->i_size = OFFSET_MAX;
387 
388 	inode->i_mapping->a_ops = &managed_cache_aops;
389 	mapping_set_gfp_mask(inode->i_mapping,
390 			     GFP_NOFS | __GFP_HIGHMEM | __GFP_MOVABLE);
391 	sbi->managed_cache = inode;
392 	return 0;
393 }
394 #else
395 static int erofs_init_managed_cache(struct super_block *sb) { return 0; }
396 #endif
397 
398 static int erofs_fill_super(struct super_block *sb, void *data, int silent)
399 {
400 	struct inode *inode;
401 	struct erofs_sb_info *sbi;
402 	int err;
403 
404 	infoln("fill_super, device -> %s", sb->s_id);
405 	infoln("options -> %s", (char *)data);
406 
407 	sb->s_magic = EROFS_SUPER_MAGIC;
408 
409 	if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
410 		errln("failed to set erofs blksize");
411 		return -EINVAL;
412 	}
413 
414 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
415 	if (!sbi)
416 		return -ENOMEM;
417 
418 	sb->s_fs_info = sbi;
419 	err = superblock_read(sb);
420 	if (err)
421 		return err;
422 
423 	sb->s_flags |= SB_RDONLY | SB_NOATIME;
424 	sb->s_maxbytes = MAX_LFS_FILESIZE;
425 	sb->s_time_gran = 1;
426 
427 	sb->s_op = &erofs_sops;
428 
429 #ifdef CONFIG_EROFS_FS_XATTR
430 	sb->s_xattr = erofs_xattr_handlers;
431 #endif
432 	/* set erofs default mount options */
433 	default_options(sbi);
434 
435 	err = parse_options(sb, data);
436 	if (err)
437 		return err;
438 
439 	if (!silent)
440 		infoln("root inode @ nid %llu", ROOT_NID(sbi));
441 
442 	if (test_opt(sbi, POSIX_ACL))
443 		sb->s_flags |= SB_POSIXACL;
444 	else
445 		sb->s_flags &= ~SB_POSIXACL;
446 
447 #ifdef CONFIG_EROFS_FS_ZIP
448 	INIT_RADIX_TREE(&sbi->workstn_tree, GFP_ATOMIC);
449 #endif
450 
451 	/* get the root inode */
452 	inode = erofs_iget(sb, ROOT_NID(sbi), true);
453 	if (IS_ERR(inode))
454 		return PTR_ERR(inode);
455 
456 	if (!S_ISDIR(inode->i_mode)) {
457 		errln("rootino(nid %llu) is not a directory(i_mode %o)",
458 		      ROOT_NID(sbi), inode->i_mode);
459 		iput(inode);
460 		return -EINVAL;
461 	}
462 
463 	sb->s_root = d_make_root(inode);
464 	if (!sb->s_root)
465 		return -ENOMEM;
466 
467 	erofs_shrinker_register(sb);
468 	/* sb->s_umount is already locked, SB_ACTIVE and SB_BORN are not set */
469 	err = erofs_init_managed_cache(sb);
470 	if (err)
471 		return err;
472 
473 	if (!silent)
474 		infoln("mounted on %s with opts: %s.", sb->s_id, (char *)data);
475 	return 0;
476 }
477 
478 static struct dentry *erofs_mount(struct file_system_type *fs_type, int flags,
479 				  const char *dev_name, void *data)
480 {
481 	return mount_bdev(fs_type, flags, dev_name, data, erofs_fill_super);
482 }
483 
484 /*
485  * could be triggered after deactivate_locked_super()
486  * is called, thus including umount and failed to initialize.
487  */
488 static void erofs_kill_sb(struct super_block *sb)
489 {
490 	struct erofs_sb_info *sbi;
491 
492 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
493 	infoln("unmounting for %s", sb->s_id);
494 
495 	kill_block_super(sb);
496 
497 	sbi = EROFS_SB(sb);
498 	if (!sbi)
499 		return;
500 	kfree(sbi);
501 	sb->s_fs_info = NULL;
502 }
503 
504 /* called when ->s_root is non-NULL */
505 static void erofs_put_super(struct super_block *sb)
506 {
507 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
508 
509 	DBG_BUGON(!sbi);
510 
511 	erofs_shrinker_unregister(sb);
512 #ifdef CONFIG_EROFS_FS_ZIP
513 	iput(sbi->managed_cache);
514 	sbi->managed_cache = NULL;
515 #endif
516 }
517 
518 static struct file_system_type erofs_fs_type = {
519 	.owner          = THIS_MODULE,
520 	.name           = "erofs",
521 	.mount          = erofs_mount,
522 	.kill_sb        = erofs_kill_sb,
523 	.fs_flags       = FS_REQUIRES_DEV,
524 };
525 MODULE_ALIAS_FS("erofs");
526 
527 static int __init erofs_module_init(void)
528 {
529 	int err;
530 
531 	erofs_check_ondisk_layout_definitions();
532 	infoln("initializing erofs " EROFS_VERSION);
533 
534 	err = erofs_init_inode_cache();
535 	if (err)
536 		goto icache_err;
537 
538 	err = erofs_init_shrinker();
539 	if (err)
540 		goto shrinker_err;
541 
542 	err = z_erofs_init_zip_subsystem();
543 	if (err)
544 		goto zip_err;
545 
546 	err = register_filesystem(&erofs_fs_type);
547 	if (err)
548 		goto fs_err;
549 
550 	infoln("successfully to initialize erofs");
551 	return 0;
552 
553 fs_err:
554 	z_erofs_exit_zip_subsystem();
555 zip_err:
556 	erofs_exit_shrinker();
557 shrinker_err:
558 	erofs_exit_inode_cache();
559 icache_err:
560 	return err;
561 }
562 
563 static void __exit erofs_module_exit(void)
564 {
565 	unregister_filesystem(&erofs_fs_type);
566 	z_erofs_exit_zip_subsystem();
567 	erofs_exit_shrinker();
568 	erofs_exit_inode_cache();
569 	infoln("successfully finalize erofs");
570 }
571 
572 /* get filesystem statistics */
573 static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
574 {
575 	struct super_block *sb = dentry->d_sb;
576 	struct erofs_sb_info *sbi = EROFS_SB(sb);
577 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
578 
579 	buf->f_type = sb->s_magic;
580 	buf->f_bsize = EROFS_BLKSIZ;
581 	buf->f_blocks = sbi->blocks;
582 	buf->f_bfree = buf->f_bavail = 0;
583 
584 	buf->f_files = ULLONG_MAX;
585 	buf->f_ffree = ULLONG_MAX - sbi->inos;
586 
587 	buf->f_namelen = EROFS_NAME_LEN;
588 
589 	buf->f_fsid.val[0] = (u32)id;
590 	buf->f_fsid.val[1] = (u32)(id >> 32);
591 	return 0;
592 }
593 
594 static int erofs_show_options(struct seq_file *seq, struct dentry *root)
595 {
596 	struct erofs_sb_info *sbi __maybe_unused = EROFS_SB(root->d_sb);
597 
598 #ifdef CONFIG_EROFS_FS_XATTR
599 	if (test_opt(sbi, XATTR_USER))
600 		seq_puts(seq, ",user_xattr");
601 	else
602 		seq_puts(seq, ",nouser_xattr");
603 #endif
604 #ifdef CONFIG_EROFS_FS_POSIX_ACL
605 	if (test_opt(sbi, POSIX_ACL))
606 		seq_puts(seq, ",acl");
607 	else
608 		seq_puts(seq, ",noacl");
609 #endif
610 	if (test_opt(sbi, FAULT_INJECTION))
611 		seq_printf(seq, ",fault_injection=%u",
612 			   erofs_get_fault_rate(sbi));
613 #ifdef CONFIG_EROFS_FS_ZIP
614 	if (sbi->cache_strategy == EROFS_ZIP_CACHE_DISABLED) {
615 		seq_puts(seq, ",cache_strategy=disabled");
616 	} else if (sbi->cache_strategy == EROFS_ZIP_CACHE_READAHEAD) {
617 		seq_puts(seq, ",cache_strategy=readahead");
618 	} else if (sbi->cache_strategy == EROFS_ZIP_CACHE_READAROUND) {
619 		seq_puts(seq, ",cache_strategy=readaround");
620 	} else {
621 		seq_puts(seq, ",cache_strategy=(unknown)");
622 		DBG_BUGON(1);
623 	}
624 #endif
625 	return 0;
626 }
627 
628 static int erofs_remount(struct super_block *sb, int *flags, char *data)
629 {
630 	struct erofs_sb_info *sbi = EROFS_SB(sb);
631 	unsigned int org_mnt_opt = sbi->mount_opt;
632 	unsigned int org_inject_rate = erofs_get_fault_rate(sbi);
633 	int err;
634 
635 	DBG_BUGON(!sb_rdonly(sb));
636 	err = parse_options(sb, data);
637 	if (err)
638 		goto out;
639 
640 	if (test_opt(sbi, POSIX_ACL))
641 		sb->s_flags |= SB_POSIXACL;
642 	else
643 		sb->s_flags &= ~SB_POSIXACL;
644 
645 	*flags |= SB_RDONLY;
646 	return 0;
647 out:
648 	__erofs_build_fault_attr(sbi, org_inject_rate);
649 	sbi->mount_opt = org_mnt_opt;
650 
651 	return err;
652 }
653 
654 const struct super_operations erofs_sops = {
655 	.put_super = erofs_put_super,
656 	.alloc_inode = alloc_inode,
657 	.free_inode = free_inode,
658 	.statfs = erofs_statfs,
659 	.show_options = erofs_show_options,
660 	.remount_fs = erofs_remount,
661 };
662 
663 module_init(erofs_module_init);
664 module_exit(erofs_module_exit);
665 
666 MODULE_DESCRIPTION("Enhanced ROM File System");
667 MODULE_AUTHOR("Gao Xiang, Chao Yu, Miao Xie, CONSUMER BG, HUAWEI Inc.");
668 MODULE_LICENSE("GPL");
669 
670