xref: /openbmc/linux/fs/erofs/super.c (revision f9834f18) 
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 <linux/crc32c.h>
13 #include "xattr.h"
14 
15 #define CREATE_TRACE_POINTS
16 #include <trace/events/erofs.h>
17 
18 static struct kmem_cache *erofs_inode_cachep __read_mostly;
19 
20 void _erofs_err(struct super_block *sb, const char *function,
21 		const char *fmt, ...)
22 {
23 	struct va_format vaf;
24 	va_list args;
25 
26 	va_start(args, fmt);
27 
28 	vaf.fmt = fmt;
29 	vaf.va = &args;
30 
31 	pr_err("(device %s): %s: %pV", sb->s_id, function, &vaf);
32 	va_end(args);
33 }
34 
35 void _erofs_info(struct super_block *sb, const char *function,
36 		 const char *fmt, ...)
37 {
38 	struct va_format vaf;
39 	va_list args;
40 
41 	va_start(args, fmt);
42 
43 	vaf.fmt = fmt;
44 	vaf.va = &args;
45 
46 	pr_info("(device %s): %pV", sb->s_id, &vaf);
47 	va_end(args);
48 }
49 
50 static int erofs_superblock_csum_verify(struct super_block *sb, void *sbdata)
51 {
52 	struct erofs_super_block *dsb;
53 	u32 expected_crc, crc;
54 
55 	dsb = kmemdup(sbdata + EROFS_SUPER_OFFSET,
56 		      EROFS_BLKSIZ - EROFS_SUPER_OFFSET, GFP_KERNEL);
57 	if (!dsb)
58 		return -ENOMEM;
59 
60 	expected_crc = le32_to_cpu(dsb->checksum);
61 	dsb->checksum = 0;
62 	/* to allow for x86 boot sectors and other oddities. */
63 	crc = crc32c(~0, dsb, EROFS_BLKSIZ - EROFS_SUPER_OFFSET);
64 	kfree(dsb);
65 
66 	if (crc != expected_crc) {
67 		erofs_err(sb, "invalid checksum 0x%08x, 0x%08x expected",
68 			  crc, expected_crc);
69 		return -EBADMSG;
70 	}
71 	return 0;
72 }
73 
74 static void erofs_inode_init_once(void *ptr)
75 {
76 	struct erofs_inode *vi = ptr;
77 
78 	inode_init_once(&vi->vfs_inode);
79 }
80 
81 static struct inode *erofs_alloc_inode(struct super_block *sb)
82 {
83 	struct erofs_inode *vi =
84 		kmem_cache_alloc(erofs_inode_cachep, GFP_KERNEL);
85 
86 	if (!vi)
87 		return NULL;
88 
89 	/* zero out everything except vfs_inode */
90 	memset(vi, 0, offsetof(struct erofs_inode, vfs_inode));
91 	return &vi->vfs_inode;
92 }
93 
94 static void erofs_free_inode(struct inode *inode)
95 {
96 	struct erofs_inode *vi = EROFS_I(inode);
97 
98 	/* be careful of RCU symlink path */
99 	if (inode->i_op == &erofs_fast_symlink_iops)
100 		kfree(inode->i_link);
101 	kfree(vi->xattr_shared_xattrs);
102 
103 	kmem_cache_free(erofs_inode_cachep, vi);
104 }
105 
106 static bool check_layout_compatibility(struct super_block *sb,
107 				       struct erofs_super_block *dsb)
108 {
109 	const unsigned int feature = le32_to_cpu(dsb->feature_incompat);
110 
111 	EROFS_SB(sb)->feature_incompat = feature;
112 
113 	/* check if current kernel meets all mandatory requirements */
114 	if (feature & (~EROFS_ALL_FEATURE_INCOMPAT)) {
115 		erofs_err(sb,
116 			  "unidentified incompatible feature %x, please upgrade kernel version",
117 			   feature & ~EROFS_ALL_FEATURE_INCOMPAT);
118 		return false;
119 	}
120 	return true;
121 }
122 
123 static int erofs_read_superblock(struct super_block *sb)
124 {
125 	struct erofs_sb_info *sbi;
126 	struct page *page;
127 	struct erofs_super_block *dsb;
128 	unsigned int blkszbits;
129 	void *data;
130 	int ret;
131 
132 	page = read_mapping_page(sb->s_bdev->bd_inode->i_mapping, 0, NULL);
133 	if (IS_ERR(page)) {
134 		erofs_err(sb, "cannot read erofs superblock");
135 		return PTR_ERR(page);
136 	}
137 
138 	sbi = EROFS_SB(sb);
139 
140 	data = kmap(page);
141 	dsb = (struct erofs_super_block *)(data + EROFS_SUPER_OFFSET);
142 
143 	ret = -EINVAL;
144 	if (le32_to_cpu(dsb->magic) != EROFS_SUPER_MAGIC_V1) {
145 		erofs_err(sb, "cannot find valid erofs superblock");
146 		goto out;
147 	}
148 
149 	sbi->feature_compat = le32_to_cpu(dsb->feature_compat);
150 	if (sbi->feature_compat & EROFS_FEATURE_COMPAT_SB_CHKSUM) {
151 		ret = erofs_superblock_csum_verify(sb, data);
152 		if (ret)
153 			goto out;
154 	}
155 
156 	blkszbits = dsb->blkszbits;
157 	/* 9(512 bytes) + LOG_SECTORS_PER_BLOCK == LOG_BLOCK_SIZE */
158 	if (blkszbits != LOG_BLOCK_SIZE) {
159 		erofs_err(sb, "blksize %u isn't supported on this platform",
160 			  1 << blkszbits);
161 		goto out;
162 	}
163 
164 	if (!check_layout_compatibility(sb, dsb))
165 		goto out;
166 
167 	sbi->blocks = le32_to_cpu(dsb->blocks);
168 	sbi->meta_blkaddr = le32_to_cpu(dsb->meta_blkaddr);
169 #ifdef CONFIG_EROFS_FS_XATTR
170 	sbi->xattr_blkaddr = le32_to_cpu(dsb->xattr_blkaddr);
171 #endif
172 	sbi->islotbits = ilog2(sizeof(struct erofs_inode_compact));
173 	sbi->root_nid = le16_to_cpu(dsb->root_nid);
174 	sbi->inos = le64_to_cpu(dsb->inos);
175 
176 	sbi->build_time = le64_to_cpu(dsb->build_time);
177 	sbi->build_time_nsec = le32_to_cpu(dsb->build_time_nsec);
178 
179 	memcpy(&sb->s_uuid, dsb->uuid, sizeof(dsb->uuid));
180 
181 	ret = strscpy(sbi->volume_name, dsb->volume_name,
182 		      sizeof(dsb->volume_name));
183 	if (ret < 0) {	/* -E2BIG */
184 		erofs_err(sb, "bad volume name without NIL terminator");
185 		ret = -EFSCORRUPTED;
186 		goto out;
187 	}
188 	ret = 0;
189 out:
190 	kunmap(page);
191 	put_page(page);
192 	return ret;
193 }
194 
195 #ifdef CONFIG_EROFS_FS_ZIP
196 static int erofs_build_cache_strategy(struct super_block *sb,
197 				      substring_t *args)
198 {
199 	struct erofs_sb_info *sbi = EROFS_SB(sb);
200 	const char *cs = match_strdup(args);
201 	int err = 0;
202 
203 	if (!cs) {
204 		erofs_err(sb, "Not enough memory to store cache strategy");
205 		return -ENOMEM;
206 	}
207 
208 	if (!strcmp(cs, "disabled")) {
209 		sbi->cache_strategy = EROFS_ZIP_CACHE_DISABLED;
210 	} else if (!strcmp(cs, "readahead")) {
211 		sbi->cache_strategy = EROFS_ZIP_CACHE_READAHEAD;
212 	} else if (!strcmp(cs, "readaround")) {
213 		sbi->cache_strategy = EROFS_ZIP_CACHE_READAROUND;
214 	} else {
215 		erofs_err(sb, "Unrecognized cache strategy \"%s\"", cs);
216 		err = -EINVAL;
217 	}
218 	kfree(cs);
219 	return err;
220 }
221 #else
222 static int erofs_build_cache_strategy(struct super_block *sb,
223 				      substring_t *args)
224 {
225 	erofs_info(sb, "EROFS compression is disabled, so cache strategy is ignored");
226 	return 0;
227 }
228 #endif
229 
230 /* set up default EROFS parameters */
231 static void erofs_default_options(struct erofs_sb_info *sbi)
232 {
233 #ifdef CONFIG_EROFS_FS_ZIP
234 	sbi->cache_strategy = EROFS_ZIP_CACHE_READAROUND;
235 	sbi->max_sync_decompress_pages = 3;
236 #endif
237 #ifdef CONFIG_EROFS_FS_XATTR
238 	set_opt(sbi, XATTR_USER);
239 #endif
240 #ifdef CONFIG_EROFS_FS_POSIX_ACL
241 	set_opt(sbi, POSIX_ACL);
242 #endif
243 }
244 
245 enum {
246 	Opt_user_xattr,
247 	Opt_nouser_xattr,
248 	Opt_acl,
249 	Opt_noacl,
250 	Opt_cache_strategy,
251 	Opt_err
252 };
253 
254 static match_table_t erofs_tokens = {
255 	{Opt_user_xattr, "user_xattr"},
256 	{Opt_nouser_xattr, "nouser_xattr"},
257 	{Opt_acl, "acl"},
258 	{Opt_noacl, "noacl"},
259 	{Opt_cache_strategy, "cache_strategy=%s"},
260 	{Opt_err, NULL}
261 };
262 
263 static int erofs_parse_options(struct super_block *sb, char *options)
264 {
265 	substring_t args[MAX_OPT_ARGS];
266 	char *p;
267 	int err;
268 
269 	if (!options)
270 		return 0;
271 
272 	while ((p = strsep(&options, ","))) {
273 		int token;
274 
275 		if (!*p)
276 			continue;
277 
278 		args[0].to = args[0].from = NULL;
279 		token = match_token(p, erofs_tokens, args);
280 
281 		switch (token) {
282 #ifdef CONFIG_EROFS_FS_XATTR
283 		case Opt_user_xattr:
284 			set_opt(EROFS_SB(sb), XATTR_USER);
285 			break;
286 		case Opt_nouser_xattr:
287 			clear_opt(EROFS_SB(sb), XATTR_USER);
288 			break;
289 #else
290 		case Opt_user_xattr:
291 			erofs_info(sb, "user_xattr options not supported");
292 			break;
293 		case Opt_nouser_xattr:
294 			erofs_info(sb, "nouser_xattr options not supported");
295 			break;
296 #endif
297 #ifdef CONFIG_EROFS_FS_POSIX_ACL
298 		case Opt_acl:
299 			set_opt(EROFS_SB(sb), POSIX_ACL);
300 			break;
301 		case Opt_noacl:
302 			clear_opt(EROFS_SB(sb), POSIX_ACL);
303 			break;
304 #else
305 		case Opt_acl:
306 			erofs_info(sb, "acl options not supported");
307 			break;
308 		case Opt_noacl:
309 			erofs_info(sb, "noacl options not supported");
310 			break;
311 #endif
312 		case Opt_cache_strategy:
313 			err = erofs_build_cache_strategy(sb, args);
314 			if (err)
315 				return err;
316 			break;
317 		default:
318 			erofs_err(sb, "Unrecognized mount option \"%s\" or missing value", p);
319 			return -EINVAL;
320 		}
321 	}
322 	return 0;
323 }
324 
325 #ifdef CONFIG_EROFS_FS_ZIP
326 static const struct address_space_operations managed_cache_aops;
327 
328 static int erofs_managed_cache_releasepage(struct page *page, gfp_t gfp_mask)
329 {
330 	int ret = 1;	/* 0 - busy */
331 	struct address_space *const mapping = page->mapping;
332 
333 	DBG_BUGON(!PageLocked(page));
334 	DBG_BUGON(mapping->a_ops != &managed_cache_aops);
335 
336 	if (PagePrivate(page))
337 		ret = erofs_try_to_free_cached_page(mapping, page);
338 
339 	return ret;
340 }
341 
342 static void erofs_managed_cache_invalidatepage(struct page *page,
343 					       unsigned int offset,
344 					       unsigned int length)
345 {
346 	const unsigned int stop = length + offset;
347 
348 	DBG_BUGON(!PageLocked(page));
349 
350 	/* Check for potential overflow in debug mode */
351 	DBG_BUGON(stop > PAGE_SIZE || stop < length);
352 
353 	if (offset == 0 && stop == PAGE_SIZE)
354 		while (!erofs_managed_cache_releasepage(page, GFP_NOFS))
355 			cond_resched();
356 }
357 
358 static const struct address_space_operations managed_cache_aops = {
359 	.releasepage = erofs_managed_cache_releasepage,
360 	.invalidatepage = erofs_managed_cache_invalidatepage,
361 };
362 
363 static int erofs_init_managed_cache(struct super_block *sb)
364 {
365 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
366 	struct inode *const inode = new_inode(sb);
367 
368 	if (!inode)
369 		return -ENOMEM;
370 
371 	set_nlink(inode, 1);
372 	inode->i_size = OFFSET_MAX;
373 
374 	inode->i_mapping->a_ops = &managed_cache_aops;
375 	mapping_set_gfp_mask(inode->i_mapping,
376 			     GFP_NOFS | __GFP_HIGHMEM | __GFP_MOVABLE);
377 	sbi->managed_cache = inode;
378 	return 0;
379 }
380 #else
381 static int erofs_init_managed_cache(struct super_block *sb) { return 0; }
382 #endif
383 
384 static int erofs_fill_super(struct super_block *sb, void *data, int silent)
385 {
386 	struct inode *inode;
387 	struct erofs_sb_info *sbi;
388 	int err;
389 
390 	sb->s_magic = EROFS_SUPER_MAGIC;
391 
392 	if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
393 		erofs_err(sb, "failed to set erofs blksize");
394 		return -EINVAL;
395 	}
396 
397 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
398 	if (!sbi)
399 		return -ENOMEM;
400 
401 	sb->s_fs_info = sbi;
402 	err = erofs_read_superblock(sb);
403 	if (err)
404 		return err;
405 
406 	sb->s_flags |= SB_RDONLY | SB_NOATIME;
407 	sb->s_maxbytes = MAX_LFS_FILESIZE;
408 	sb->s_time_gran = 1;
409 
410 	sb->s_op = &erofs_sops;
411 
412 #ifdef CONFIG_EROFS_FS_XATTR
413 	sb->s_xattr = erofs_xattr_handlers;
414 #endif
415 	/* set erofs default mount options */
416 	erofs_default_options(sbi);
417 
418 	err = erofs_parse_options(sb, data);
419 	if (err)
420 		return err;
421 
422 	if (test_opt(sbi, POSIX_ACL))
423 		sb->s_flags |= SB_POSIXACL;
424 	else
425 		sb->s_flags &= ~SB_POSIXACL;
426 
427 #ifdef CONFIG_EROFS_FS_ZIP
428 	INIT_RADIX_TREE(&sbi->workstn_tree, GFP_ATOMIC);
429 #endif
430 
431 	/* get the root inode */
432 	inode = erofs_iget(sb, ROOT_NID(sbi), true);
433 	if (IS_ERR(inode))
434 		return PTR_ERR(inode);
435 
436 	if (!S_ISDIR(inode->i_mode)) {
437 		erofs_err(sb, "rootino(nid %llu) is not a directory(i_mode %o)",
438 			  ROOT_NID(sbi), inode->i_mode);
439 		iput(inode);
440 		return -EINVAL;
441 	}
442 
443 	sb->s_root = d_make_root(inode);
444 	if (!sb->s_root)
445 		return -ENOMEM;
446 
447 	erofs_shrinker_register(sb);
448 	/* sb->s_umount is already locked, SB_ACTIVE and SB_BORN are not set */
449 	err = erofs_init_managed_cache(sb);
450 	if (err)
451 		return err;
452 
453 	erofs_info(sb, "mounted with opts: %s, root inode @ nid %llu.",
454 		   (char *)data, ROOT_NID(sbi));
455 	return 0;
456 }
457 
458 static struct dentry *erofs_mount(struct file_system_type *fs_type, int flags,
459 				  const char *dev_name, void *data)
460 {
461 	return mount_bdev(fs_type, flags, dev_name, data, erofs_fill_super);
462 }
463 
464 /*
465  * could be triggered after deactivate_locked_super()
466  * is called, thus including umount and failed to initialize.
467  */
468 static void erofs_kill_sb(struct super_block *sb)
469 {
470 	struct erofs_sb_info *sbi;
471 
472 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
473 
474 	kill_block_super(sb);
475 
476 	sbi = EROFS_SB(sb);
477 	if (!sbi)
478 		return;
479 	kfree(sbi);
480 	sb->s_fs_info = NULL;
481 }
482 
483 /* called when ->s_root is non-NULL */
484 static void erofs_put_super(struct super_block *sb)
485 {
486 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
487 
488 	DBG_BUGON(!sbi);
489 
490 	erofs_shrinker_unregister(sb);
491 #ifdef CONFIG_EROFS_FS_ZIP
492 	iput(sbi->managed_cache);
493 	sbi->managed_cache = NULL;
494 #endif
495 }
496 
497 static struct file_system_type erofs_fs_type = {
498 	.owner          = THIS_MODULE,
499 	.name           = "erofs",
500 	.mount          = erofs_mount,
501 	.kill_sb        = erofs_kill_sb,
502 	.fs_flags       = FS_REQUIRES_DEV,
503 };
504 MODULE_ALIAS_FS("erofs");
505 
506 static int __init erofs_module_init(void)
507 {
508 	int err;
509 
510 	erofs_check_ondisk_layout_definitions();
511 
512 	erofs_inode_cachep = kmem_cache_create("erofs_inode",
513 					       sizeof(struct erofs_inode), 0,
514 					       SLAB_RECLAIM_ACCOUNT,
515 					       erofs_inode_init_once);
516 	if (!erofs_inode_cachep) {
517 		err = -ENOMEM;
518 		goto icache_err;
519 	}
520 
521 	err = erofs_init_shrinker();
522 	if (err)
523 		goto shrinker_err;
524 
525 	err = z_erofs_init_zip_subsystem();
526 	if (err)
527 		goto zip_err;
528 
529 	err = register_filesystem(&erofs_fs_type);
530 	if (err)
531 		goto fs_err;
532 
533 	return 0;
534 
535 fs_err:
536 	z_erofs_exit_zip_subsystem();
537 zip_err:
538 	erofs_exit_shrinker();
539 shrinker_err:
540 	kmem_cache_destroy(erofs_inode_cachep);
541 icache_err:
542 	return err;
543 }
544 
545 static void __exit erofs_module_exit(void)
546 {
547 	unregister_filesystem(&erofs_fs_type);
548 	z_erofs_exit_zip_subsystem();
549 	erofs_exit_shrinker();
550 
551 	/* Ensure all RCU free inodes are safe before cache is destroyed. */
552 	rcu_barrier();
553 	kmem_cache_destroy(erofs_inode_cachep);
554 }
555 
556 /* get filesystem statistics */
557 static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
558 {
559 	struct super_block *sb = dentry->d_sb;
560 	struct erofs_sb_info *sbi = EROFS_SB(sb);
561 	u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
562 
563 	buf->f_type = sb->s_magic;
564 	buf->f_bsize = EROFS_BLKSIZ;
565 	buf->f_blocks = sbi->blocks;
566 	buf->f_bfree = buf->f_bavail = 0;
567 
568 	buf->f_files = ULLONG_MAX;
569 	buf->f_ffree = ULLONG_MAX - sbi->inos;
570 
571 	buf->f_namelen = EROFS_NAME_LEN;
572 
573 	buf->f_fsid.val[0] = (u32)id;
574 	buf->f_fsid.val[1] = (u32)(id >> 32);
575 	return 0;
576 }
577 
578 static int erofs_show_options(struct seq_file *seq, struct dentry *root)
579 {
580 	struct erofs_sb_info *sbi __maybe_unused = EROFS_SB(root->d_sb);
581 
582 #ifdef CONFIG_EROFS_FS_XATTR
583 	if (test_opt(sbi, XATTR_USER))
584 		seq_puts(seq, ",user_xattr");
585 	else
586 		seq_puts(seq, ",nouser_xattr");
587 #endif
588 #ifdef CONFIG_EROFS_FS_POSIX_ACL
589 	if (test_opt(sbi, POSIX_ACL))
590 		seq_puts(seq, ",acl");
591 	else
592 		seq_puts(seq, ",noacl");
593 #endif
594 #ifdef CONFIG_EROFS_FS_ZIP
595 	if (sbi->cache_strategy == EROFS_ZIP_CACHE_DISABLED) {
596 		seq_puts(seq, ",cache_strategy=disabled");
597 	} else if (sbi->cache_strategy == EROFS_ZIP_CACHE_READAHEAD) {
598 		seq_puts(seq, ",cache_strategy=readahead");
599 	} else if (sbi->cache_strategy == EROFS_ZIP_CACHE_READAROUND) {
600 		seq_puts(seq, ",cache_strategy=readaround");
601 	}
602 #endif
603 	return 0;
604 }
605 
606 static int erofs_remount(struct super_block *sb, int *flags, char *data)
607 {
608 	struct erofs_sb_info *sbi = EROFS_SB(sb);
609 	unsigned int org_mnt_opt = sbi->mount_opt;
610 	int err;
611 
612 	DBG_BUGON(!sb_rdonly(sb));
613 	err = erofs_parse_options(sb, data);
614 	if (err)
615 		goto out;
616 
617 	if (test_opt(sbi, POSIX_ACL))
618 		sb->s_flags |= SB_POSIXACL;
619 	else
620 		sb->s_flags &= ~SB_POSIXACL;
621 
622 	*flags |= SB_RDONLY;
623 	return 0;
624 out:
625 	sbi->mount_opt = org_mnt_opt;
626 	return err;
627 }
628 
629 const struct super_operations erofs_sops = {
630 	.put_super = erofs_put_super,
631 	.alloc_inode = erofs_alloc_inode,
632 	.free_inode = erofs_free_inode,
633 	.statfs = erofs_statfs,
634 	.show_options = erofs_show_options,
635 	.remount_fs = erofs_remount,
636 };
637 
638 module_init(erofs_module_init);
639 module_exit(erofs_module_exit);
640 
641 MODULE_DESCRIPTION("Enhanced ROM File System");
642 MODULE_AUTHOR("Gao Xiang, Chao Yu, Miao Xie, CONSUMER BG, HUAWEI Inc.");
643 MODULE_LICENSE("GPL");
644 
645