xref: /openbmc/linux/fs/erofs/super.c (revision 7b4e372c) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017-2018 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  * Copyright (C) 2021, Alibaba Cloud
6  */
7 #include <linux/module.h>
8 #include <linux/statfs.h>
9 #include <linux/parser.h>
10 #include <linux/seq_file.h>
11 #include <linux/crc32c.h>
12 #include <linux/fs_context.h>
13 #include <linux/fs_parser.h>
14 #include <linux/dax.h>
15 #include <linux/exportfs.h>
16 #include "xattr.h"
17 
18 #define CREATE_TRACE_POINTS
19 #include <trace/events/erofs.h>
20 
21 static struct kmem_cache *erofs_inode_cachep __read_mostly;
22 
23 void _erofs_err(struct super_block *sb, const char *function,
24 		const char *fmt, ...)
25 {
26 	struct va_format vaf;
27 	va_list args;
28 
29 	va_start(args, fmt);
30 
31 	vaf.fmt = fmt;
32 	vaf.va = &args;
33 
34 	pr_err("(device %s): %s: %pV", sb->s_id, function, &vaf);
35 	va_end(args);
36 }
37 
38 void _erofs_info(struct super_block *sb, const char *function,
39 		 const char *fmt, ...)
40 {
41 	struct va_format vaf;
42 	va_list args;
43 
44 	va_start(args, fmt);
45 
46 	vaf.fmt = fmt;
47 	vaf.va = &args;
48 
49 	pr_info("(device %s): %pV", sb->s_id, &vaf);
50 	va_end(args);
51 }
52 
53 static int erofs_superblock_csum_verify(struct super_block *sb, void *sbdata)
54 {
55 	size_t len = 1 << EROFS_SB(sb)->blkszbits;
56 	struct erofs_super_block *dsb;
57 	u32 expected_crc, crc;
58 
59 	if (len > EROFS_SUPER_OFFSET)
60 		len -= EROFS_SUPER_OFFSET;
61 
62 	dsb = kmemdup(sbdata + EROFS_SUPER_OFFSET, len, GFP_KERNEL);
63 	if (!dsb)
64 		return -ENOMEM;
65 
66 	expected_crc = le32_to_cpu(dsb->checksum);
67 	dsb->checksum = 0;
68 	/* to allow for x86 boot sectors and other oddities. */
69 	crc = crc32c(~0, dsb, len);
70 	kfree(dsb);
71 
72 	if (crc != expected_crc) {
73 		erofs_err(sb, "invalid checksum 0x%08x, 0x%08x expected",
74 			  crc, expected_crc);
75 		return -EBADMSG;
76 	}
77 	return 0;
78 }
79 
80 static void erofs_inode_init_once(void *ptr)
81 {
82 	struct erofs_inode *vi = ptr;
83 
84 	inode_init_once(&vi->vfs_inode);
85 }
86 
87 static struct inode *erofs_alloc_inode(struct super_block *sb)
88 {
89 	struct erofs_inode *vi =
90 		alloc_inode_sb(sb, erofs_inode_cachep, GFP_KERNEL);
91 
92 	if (!vi)
93 		return NULL;
94 
95 	/* zero out everything except vfs_inode */
96 	memset(vi, 0, offsetof(struct erofs_inode, vfs_inode));
97 	return &vi->vfs_inode;
98 }
99 
100 static void erofs_free_inode(struct inode *inode)
101 {
102 	struct erofs_inode *vi = EROFS_I(inode);
103 
104 	/* be careful of RCU symlink path */
105 	if (inode->i_op == &erofs_fast_symlink_iops)
106 		kfree(inode->i_link);
107 	kfree(vi->xattr_shared_xattrs);
108 
109 	kmem_cache_free(erofs_inode_cachep, vi);
110 }
111 
112 static bool check_layout_compatibility(struct super_block *sb,
113 				       struct erofs_super_block *dsb)
114 {
115 	const unsigned int feature = le32_to_cpu(dsb->feature_incompat);
116 
117 	EROFS_SB(sb)->feature_incompat = feature;
118 
119 	/* check if current kernel meets all mandatory requirements */
120 	if (feature & (~EROFS_ALL_FEATURE_INCOMPAT)) {
121 		erofs_err(sb,
122 			  "unidentified incompatible feature %x, please upgrade kernel version",
123 			   feature & ~EROFS_ALL_FEATURE_INCOMPAT);
124 		return false;
125 	}
126 	return true;
127 }
128 
129 /* read variable-sized metadata, offset will be aligned by 4-byte */
130 void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
131 			  erofs_off_t *offset, int *lengthp)
132 {
133 	u8 *buffer, *ptr;
134 	int len, i, cnt;
135 
136 	*offset = round_up(*offset, 4);
137 	ptr = erofs_bread(buf, erofs_blknr(sb, *offset), EROFS_KMAP);
138 	if (IS_ERR(ptr))
139 		return ptr;
140 
141 	len = le16_to_cpu(*(__le16 *)&ptr[erofs_blkoff(sb, *offset)]);
142 	if (!len)
143 		len = U16_MAX + 1;
144 	buffer = kmalloc(len, GFP_KERNEL);
145 	if (!buffer)
146 		return ERR_PTR(-ENOMEM);
147 	*offset += sizeof(__le16);
148 	*lengthp = len;
149 
150 	for (i = 0; i < len; i += cnt) {
151 		cnt = min_t(int, sb->s_blocksize - erofs_blkoff(sb, *offset),
152 			    len - i);
153 		ptr = erofs_bread(buf, erofs_blknr(sb, *offset), EROFS_KMAP);
154 		if (IS_ERR(ptr)) {
155 			kfree(buffer);
156 			return ptr;
157 		}
158 		memcpy(buffer + i, ptr + erofs_blkoff(sb, *offset), cnt);
159 		*offset += cnt;
160 	}
161 	return buffer;
162 }
163 
164 #ifdef CONFIG_EROFS_FS_ZIP
165 static int erofs_load_compr_cfgs(struct super_block *sb,
166 				 struct erofs_super_block *dsb)
167 {
168 	struct erofs_sb_info *sbi = EROFS_SB(sb);
169 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
170 	unsigned int algs, alg;
171 	erofs_off_t offset;
172 	int size, ret = 0;
173 
174 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
175 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
176 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
177 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
178 		return -EINVAL;
179 	}
180 
181 	erofs_init_metabuf(&buf, sb);
182 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
183 	alg = 0;
184 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
185 		void *data;
186 
187 		if (!(algs & 1))
188 			continue;
189 
190 		data = erofs_read_metadata(sb, &buf, &offset, &size);
191 		if (IS_ERR(data)) {
192 			ret = PTR_ERR(data);
193 			break;
194 		}
195 
196 		switch (alg) {
197 		case Z_EROFS_COMPRESSION_LZ4:
198 			ret = z_erofs_load_lz4_config(sb, dsb, data, size);
199 			break;
200 		case Z_EROFS_COMPRESSION_LZMA:
201 			ret = z_erofs_load_lzma_config(sb, dsb, data, size);
202 			break;
203 		default:
204 			DBG_BUGON(1);
205 			ret = -EFAULT;
206 		}
207 		kfree(data);
208 		if (ret)
209 			break;
210 	}
211 	erofs_put_metabuf(&buf);
212 	return ret;
213 }
214 #else
215 static int erofs_load_compr_cfgs(struct super_block *sb,
216 				 struct erofs_super_block *dsb)
217 {
218 	if (dsb->u1.available_compr_algs) {
219 		erofs_err(sb, "try to load compressed fs when compression is disabled");
220 		return -EINVAL;
221 	}
222 	return 0;
223 }
224 #endif
225 
226 static int erofs_init_device(struct erofs_buf *buf, struct super_block *sb,
227 			     struct erofs_device_info *dif, erofs_off_t *pos)
228 {
229 	struct erofs_sb_info *sbi = EROFS_SB(sb);
230 	struct erofs_fscache *fscache;
231 	struct erofs_deviceslot *dis;
232 	struct block_device *bdev;
233 	void *ptr;
234 
235 	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(sb, *pos), EROFS_KMAP);
236 	if (IS_ERR(ptr))
237 		return PTR_ERR(ptr);
238 	dis = ptr + erofs_blkoff(sb, *pos);
239 
240 	if (!dif->path) {
241 		if (!dis->tag[0]) {
242 			erofs_err(sb, "empty device tag @ pos %llu", *pos);
243 			return -EINVAL;
244 		}
245 		dif->path = kmemdup_nul(dis->tag, sizeof(dis->tag), GFP_KERNEL);
246 		if (!dif->path)
247 			return -ENOMEM;
248 	}
249 
250 	if (erofs_is_fscache_mode(sb)) {
251 		fscache = erofs_fscache_register_cookie(sb, dif->path, 0);
252 		if (IS_ERR(fscache))
253 			return PTR_ERR(fscache);
254 		dif->fscache = fscache;
255 	} else if (!sbi->devs->flatdev) {
256 		bdev = blkdev_get_by_path(dif->path, FMODE_READ | FMODE_EXCL,
257 					  sb->s_type);
258 		if (IS_ERR(bdev))
259 			return PTR_ERR(bdev);
260 		dif->bdev = bdev;
261 		dif->dax_dev = fs_dax_get_by_bdev(bdev, &dif->dax_part_off,
262 						  NULL, NULL);
263 	}
264 
265 	dif->blocks = le32_to_cpu(dis->blocks);
266 	dif->mapped_blkaddr = le32_to_cpu(dis->mapped_blkaddr);
267 	sbi->total_blocks += dif->blocks;
268 	*pos += EROFS_DEVT_SLOT_SIZE;
269 	return 0;
270 }
271 
272 static int erofs_scan_devices(struct super_block *sb,
273 			      struct erofs_super_block *dsb)
274 {
275 	struct erofs_sb_info *sbi = EROFS_SB(sb);
276 	unsigned int ondisk_extradevs;
277 	erofs_off_t pos;
278 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
279 	struct erofs_device_info *dif;
280 	int id, err = 0;
281 
282 	sbi->total_blocks = sbi->primarydevice_blocks;
283 	if (!erofs_sb_has_device_table(sbi))
284 		ondisk_extradevs = 0;
285 	else
286 		ondisk_extradevs = le16_to_cpu(dsb->extra_devices);
287 
288 	if (sbi->devs->extra_devices &&
289 	    ondisk_extradevs != sbi->devs->extra_devices) {
290 		erofs_err(sb, "extra devices don't match (ondisk %u, given %u)",
291 			  ondisk_extradevs, sbi->devs->extra_devices);
292 		return -EINVAL;
293 	}
294 	if (!ondisk_extradevs)
295 		return 0;
296 
297 	if (!sbi->devs->extra_devices && !erofs_is_fscache_mode(sb))
298 		sbi->devs->flatdev = true;
299 
300 	sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
301 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
302 	down_read(&sbi->devs->rwsem);
303 	if (sbi->devs->extra_devices) {
304 		idr_for_each_entry(&sbi->devs->tree, dif, id) {
305 			err = erofs_init_device(&buf, sb, dif, &pos);
306 			if (err)
307 				break;
308 		}
309 	} else {
310 		for (id = 0; id < ondisk_extradevs; id++) {
311 			dif = kzalloc(sizeof(*dif), GFP_KERNEL);
312 			if (!dif) {
313 				err = -ENOMEM;
314 				break;
315 			}
316 
317 			err = idr_alloc(&sbi->devs->tree, dif, 0, 0, GFP_KERNEL);
318 			if (err < 0) {
319 				kfree(dif);
320 				break;
321 			}
322 			++sbi->devs->extra_devices;
323 
324 			err = erofs_init_device(&buf, sb, dif, &pos);
325 			if (err)
326 				break;
327 		}
328 	}
329 	up_read(&sbi->devs->rwsem);
330 	erofs_put_metabuf(&buf);
331 	return err;
332 }
333 
334 static int erofs_read_superblock(struct super_block *sb)
335 {
336 	struct erofs_sb_info *sbi;
337 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
338 	struct erofs_super_block *dsb;
339 	void *data;
340 	int ret;
341 
342 	data = erofs_read_metabuf(&buf, sb, 0, EROFS_KMAP);
343 	if (IS_ERR(data)) {
344 		erofs_err(sb, "cannot read erofs superblock");
345 		return PTR_ERR(data);
346 	}
347 
348 	sbi = EROFS_SB(sb);
349 	dsb = (struct erofs_super_block *)(data + EROFS_SUPER_OFFSET);
350 
351 	ret = -EINVAL;
352 	if (le32_to_cpu(dsb->magic) != EROFS_SUPER_MAGIC_V1) {
353 		erofs_err(sb, "cannot find valid erofs superblock");
354 		goto out;
355 	}
356 
357 	sbi->blkszbits  = dsb->blkszbits;
358 	if (sbi->blkszbits < 9 || sbi->blkszbits > PAGE_SHIFT) {
359 		erofs_err(sb, "blkszbits %u isn't supported", sbi->blkszbits);
360 		goto out;
361 	}
362 	if (dsb->dirblkbits) {
363 		erofs_err(sb, "dirblkbits %u isn't supported", dsb->dirblkbits);
364 		goto out;
365 	}
366 
367 	sbi->feature_compat = le32_to_cpu(dsb->feature_compat);
368 	if (erofs_sb_has_sb_chksum(sbi)) {
369 		ret = erofs_superblock_csum_verify(sb, data);
370 		if (ret)
371 			goto out;
372 	}
373 
374 	ret = -EINVAL;
375 	if (!check_layout_compatibility(sb, dsb))
376 		goto out;
377 
378 	sbi->sb_size = 128 + dsb->sb_extslots * EROFS_SB_EXTSLOT_SIZE;
379 	if (sbi->sb_size > PAGE_SIZE - EROFS_SUPER_OFFSET) {
380 		erofs_err(sb, "invalid sb_extslots %u (more than a fs block)",
381 			  sbi->sb_size);
382 		goto out;
383 	}
384 	sbi->primarydevice_blocks = le32_to_cpu(dsb->blocks);
385 	sbi->meta_blkaddr = le32_to_cpu(dsb->meta_blkaddr);
386 #ifdef CONFIG_EROFS_FS_XATTR
387 	sbi->xattr_blkaddr = le32_to_cpu(dsb->xattr_blkaddr);
388 	sbi->xattr_prefix_start = le32_to_cpu(dsb->xattr_prefix_start);
389 	sbi->xattr_prefix_count = dsb->xattr_prefix_count;
390 #endif
391 	sbi->islotbits = ilog2(sizeof(struct erofs_inode_compact));
392 	sbi->root_nid = le16_to_cpu(dsb->root_nid);
393 	sbi->packed_nid = le64_to_cpu(dsb->packed_nid);
394 	sbi->inos = le64_to_cpu(dsb->inos);
395 
396 	sbi->build_time = le64_to_cpu(dsb->build_time);
397 	sbi->build_time_nsec = le32_to_cpu(dsb->build_time_nsec);
398 
399 	memcpy(&sb->s_uuid, dsb->uuid, sizeof(dsb->uuid));
400 
401 	ret = strscpy(sbi->volume_name, dsb->volume_name,
402 		      sizeof(dsb->volume_name));
403 	if (ret < 0) {	/* -E2BIG */
404 		erofs_err(sb, "bad volume name without NIL terminator");
405 		ret = -EFSCORRUPTED;
406 		goto out;
407 	}
408 
409 	/* parse on-disk compression configurations */
410 	if (erofs_sb_has_compr_cfgs(sbi))
411 		ret = erofs_load_compr_cfgs(sb, dsb);
412 	else
413 		ret = z_erofs_load_lz4_config(sb, dsb, NULL, 0);
414 	if (ret < 0)
415 		goto out;
416 
417 	/* handle multiple devices */
418 	ret = erofs_scan_devices(sb, dsb);
419 
420 	if (erofs_is_fscache_mode(sb))
421 		erofs_info(sb, "EXPERIMENTAL fscache-based on-demand read feature in use. Use at your own risk!");
422 	if (erofs_sb_has_fragments(sbi))
423 		erofs_info(sb, "EXPERIMENTAL compressed fragments feature in use. Use at your own risk!");
424 	if (erofs_sb_has_dedupe(sbi))
425 		erofs_info(sb, "EXPERIMENTAL global deduplication feature in use. Use at your own risk!");
426 out:
427 	erofs_put_metabuf(&buf);
428 	return ret;
429 }
430 
431 /* set up default EROFS parameters */
432 static void erofs_default_options(struct erofs_fs_context *ctx)
433 {
434 #ifdef CONFIG_EROFS_FS_ZIP
435 	ctx->opt.cache_strategy = EROFS_ZIP_CACHE_READAROUND;
436 	ctx->opt.max_sync_decompress_pages = 3;
437 	ctx->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_AUTO;
438 #endif
439 #ifdef CONFIG_EROFS_FS_XATTR
440 	set_opt(&ctx->opt, XATTR_USER);
441 #endif
442 #ifdef CONFIG_EROFS_FS_POSIX_ACL
443 	set_opt(&ctx->opt, POSIX_ACL);
444 #endif
445 }
446 
447 enum {
448 	Opt_user_xattr,
449 	Opt_acl,
450 	Opt_cache_strategy,
451 	Opt_dax,
452 	Opt_dax_enum,
453 	Opt_device,
454 	Opt_fsid,
455 	Opt_domain_id,
456 	Opt_err
457 };
458 
459 static const struct constant_table erofs_param_cache_strategy[] = {
460 	{"disabled",	EROFS_ZIP_CACHE_DISABLED},
461 	{"readahead",	EROFS_ZIP_CACHE_READAHEAD},
462 	{"readaround",	EROFS_ZIP_CACHE_READAROUND},
463 	{}
464 };
465 
466 static const struct constant_table erofs_dax_param_enums[] = {
467 	{"always",	EROFS_MOUNT_DAX_ALWAYS},
468 	{"never",	EROFS_MOUNT_DAX_NEVER},
469 	{}
470 };
471 
472 static const struct fs_parameter_spec erofs_fs_parameters[] = {
473 	fsparam_flag_no("user_xattr",	Opt_user_xattr),
474 	fsparam_flag_no("acl",		Opt_acl),
475 	fsparam_enum("cache_strategy",	Opt_cache_strategy,
476 		     erofs_param_cache_strategy),
477 	fsparam_flag("dax",             Opt_dax),
478 	fsparam_enum("dax",		Opt_dax_enum, erofs_dax_param_enums),
479 	fsparam_string("device",	Opt_device),
480 	fsparam_string("fsid",		Opt_fsid),
481 	fsparam_string("domain_id",	Opt_domain_id),
482 	{}
483 };
484 
485 static bool erofs_fc_set_dax_mode(struct fs_context *fc, unsigned int mode)
486 {
487 #ifdef CONFIG_FS_DAX
488 	struct erofs_fs_context *ctx = fc->fs_private;
489 
490 	switch (mode) {
491 	case EROFS_MOUNT_DAX_ALWAYS:
492 		warnfc(fc, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
493 		set_opt(&ctx->opt, DAX_ALWAYS);
494 		clear_opt(&ctx->opt, DAX_NEVER);
495 		return true;
496 	case EROFS_MOUNT_DAX_NEVER:
497 		set_opt(&ctx->opt, DAX_NEVER);
498 		clear_opt(&ctx->opt, DAX_ALWAYS);
499 		return true;
500 	default:
501 		DBG_BUGON(1);
502 		return false;
503 	}
504 #else
505 	errorfc(fc, "dax options not supported");
506 	return false;
507 #endif
508 }
509 
510 static int erofs_fc_parse_param(struct fs_context *fc,
511 				struct fs_parameter *param)
512 {
513 	struct erofs_fs_context *ctx = fc->fs_private;
514 	struct fs_parse_result result;
515 	struct erofs_device_info *dif;
516 	int opt, ret;
517 
518 	opt = fs_parse(fc, erofs_fs_parameters, param, &result);
519 	if (opt < 0)
520 		return opt;
521 
522 	switch (opt) {
523 	case Opt_user_xattr:
524 #ifdef CONFIG_EROFS_FS_XATTR
525 		if (result.boolean)
526 			set_opt(&ctx->opt, XATTR_USER);
527 		else
528 			clear_opt(&ctx->opt, XATTR_USER);
529 #else
530 		errorfc(fc, "{,no}user_xattr options not supported");
531 #endif
532 		break;
533 	case Opt_acl:
534 #ifdef CONFIG_EROFS_FS_POSIX_ACL
535 		if (result.boolean)
536 			set_opt(&ctx->opt, POSIX_ACL);
537 		else
538 			clear_opt(&ctx->opt, POSIX_ACL);
539 #else
540 		errorfc(fc, "{,no}acl options not supported");
541 #endif
542 		break;
543 	case Opt_cache_strategy:
544 #ifdef CONFIG_EROFS_FS_ZIP
545 		ctx->opt.cache_strategy = result.uint_32;
546 #else
547 		errorfc(fc, "compression not supported, cache_strategy ignored");
548 #endif
549 		break;
550 	case Opt_dax:
551 		if (!erofs_fc_set_dax_mode(fc, EROFS_MOUNT_DAX_ALWAYS))
552 			return -EINVAL;
553 		break;
554 	case Opt_dax_enum:
555 		if (!erofs_fc_set_dax_mode(fc, result.uint_32))
556 			return -EINVAL;
557 		break;
558 	case Opt_device:
559 		dif = kzalloc(sizeof(*dif), GFP_KERNEL);
560 		if (!dif)
561 			return -ENOMEM;
562 		dif->path = kstrdup(param->string, GFP_KERNEL);
563 		if (!dif->path) {
564 			kfree(dif);
565 			return -ENOMEM;
566 		}
567 		down_write(&ctx->devs->rwsem);
568 		ret = idr_alloc(&ctx->devs->tree, dif, 0, 0, GFP_KERNEL);
569 		up_write(&ctx->devs->rwsem);
570 		if (ret < 0) {
571 			kfree(dif->path);
572 			kfree(dif);
573 			return ret;
574 		}
575 		++ctx->devs->extra_devices;
576 		break;
577 #ifdef CONFIG_EROFS_FS_ONDEMAND
578 	case Opt_fsid:
579 		kfree(ctx->fsid);
580 		ctx->fsid = kstrdup(param->string, GFP_KERNEL);
581 		if (!ctx->fsid)
582 			return -ENOMEM;
583 		break;
584 	case Opt_domain_id:
585 		kfree(ctx->domain_id);
586 		ctx->domain_id = kstrdup(param->string, GFP_KERNEL);
587 		if (!ctx->domain_id)
588 			return -ENOMEM;
589 		break;
590 #else
591 	case Opt_fsid:
592 	case Opt_domain_id:
593 		errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
594 		break;
595 #endif
596 	default:
597 		return -ENOPARAM;
598 	}
599 	return 0;
600 }
601 
602 static struct inode *erofs_nfs_get_inode(struct super_block *sb,
603 					 u64 ino, u32 generation)
604 {
605 	return erofs_iget(sb, ino);
606 }
607 
608 static struct dentry *erofs_fh_to_dentry(struct super_block *sb,
609 		struct fid *fid, int fh_len, int fh_type)
610 {
611 	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
612 				    erofs_nfs_get_inode);
613 }
614 
615 static struct dentry *erofs_fh_to_parent(struct super_block *sb,
616 		struct fid *fid, int fh_len, int fh_type)
617 {
618 	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
619 				    erofs_nfs_get_inode);
620 }
621 
622 static struct dentry *erofs_get_parent(struct dentry *child)
623 {
624 	erofs_nid_t nid;
625 	unsigned int d_type;
626 	int err;
627 
628 	err = erofs_namei(d_inode(child), &dotdot_name, &nid, &d_type);
629 	if (err)
630 		return ERR_PTR(err);
631 	return d_obtain_alias(erofs_iget(child->d_sb, nid));
632 }
633 
634 static const struct export_operations erofs_export_ops = {
635 	.fh_to_dentry = erofs_fh_to_dentry,
636 	.fh_to_parent = erofs_fh_to_parent,
637 	.get_parent = erofs_get_parent,
638 };
639 
640 static int erofs_fc_fill_pseudo_super(struct super_block *sb, struct fs_context *fc)
641 {
642 	static const struct tree_descr empty_descr = {""};
643 
644 	return simple_fill_super(sb, EROFS_SUPER_MAGIC, &empty_descr);
645 }
646 
647 static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
648 {
649 	struct inode *inode;
650 	struct erofs_sb_info *sbi;
651 	struct erofs_fs_context *ctx = fc->fs_private;
652 	int err;
653 
654 	sb->s_magic = EROFS_SUPER_MAGIC;
655 	sb->s_flags |= SB_RDONLY | SB_NOATIME;
656 	sb->s_maxbytes = MAX_LFS_FILESIZE;
657 	sb->s_op = &erofs_sops;
658 
659 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
660 	if (!sbi)
661 		return -ENOMEM;
662 
663 	sb->s_fs_info = sbi;
664 	sbi->opt = ctx->opt;
665 	sbi->devs = ctx->devs;
666 	ctx->devs = NULL;
667 	sbi->fsid = ctx->fsid;
668 	ctx->fsid = NULL;
669 	sbi->domain_id = ctx->domain_id;
670 	ctx->domain_id = NULL;
671 
672 	sbi->blkszbits = PAGE_SHIFT;
673 	if (erofs_is_fscache_mode(sb)) {
674 		sb->s_blocksize = PAGE_SIZE;
675 		sb->s_blocksize_bits = PAGE_SHIFT;
676 
677 		err = erofs_fscache_register_fs(sb);
678 		if (err)
679 			return err;
680 
681 		err = super_setup_bdi(sb);
682 		if (err)
683 			return err;
684 	} else {
685 		if (!sb_set_blocksize(sb, PAGE_SIZE)) {
686 			errorfc(fc, "failed to set initial blksize");
687 			return -EINVAL;
688 		}
689 
690 		sbi->dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
691 						  &sbi->dax_part_off,
692 						  NULL, NULL);
693 	}
694 
695 	err = erofs_read_superblock(sb);
696 	if (err)
697 		return err;
698 
699 	if (sb->s_blocksize_bits != sbi->blkszbits) {
700 		if (erofs_is_fscache_mode(sb)) {
701 			errorfc(fc, "unsupported blksize for fscache mode");
702 			return -EINVAL;
703 		}
704 		if (!sb_set_blocksize(sb, 1 << sbi->blkszbits)) {
705 			errorfc(fc, "failed to set erofs blksize");
706 			return -EINVAL;
707 		}
708 	}
709 
710 	if (test_opt(&sbi->opt, DAX_ALWAYS)) {
711 		if (!sbi->dax_dev) {
712 			errorfc(fc, "DAX unsupported by block device. Turning off DAX.");
713 			clear_opt(&sbi->opt, DAX_ALWAYS);
714 		} else if (sbi->blkszbits != PAGE_SHIFT) {
715 			errorfc(fc, "unsupported blocksize for DAX");
716 			clear_opt(&sbi->opt, DAX_ALWAYS);
717 		}
718 	}
719 
720 	sb->s_time_gran = 1;
721 	sb->s_xattr = erofs_xattr_handlers;
722 	sb->s_export_op = &erofs_export_ops;
723 
724 	if (test_opt(&sbi->opt, POSIX_ACL))
725 		sb->s_flags |= SB_POSIXACL;
726 	else
727 		sb->s_flags &= ~SB_POSIXACL;
728 
729 #ifdef CONFIG_EROFS_FS_ZIP
730 	xa_init(&sbi->managed_pslots);
731 #endif
732 
733 	/* get the root inode */
734 	inode = erofs_iget(sb, ROOT_NID(sbi));
735 	if (IS_ERR(inode))
736 		return PTR_ERR(inode);
737 
738 	if (!S_ISDIR(inode->i_mode)) {
739 		erofs_err(sb, "rootino(nid %llu) is not a directory(i_mode %o)",
740 			  ROOT_NID(sbi), inode->i_mode);
741 		iput(inode);
742 		return -EINVAL;
743 	}
744 
745 	sb->s_root = d_make_root(inode);
746 	if (!sb->s_root)
747 		return -ENOMEM;
748 
749 	erofs_shrinker_register(sb);
750 	/* sb->s_umount is already locked, SB_ACTIVE and SB_BORN are not set */
751 	if (erofs_sb_has_fragments(sbi) && sbi->packed_nid) {
752 		sbi->packed_inode = erofs_iget(sb, sbi->packed_nid);
753 		if (IS_ERR(sbi->packed_inode)) {
754 			err = PTR_ERR(sbi->packed_inode);
755 			sbi->packed_inode = NULL;
756 			return err;
757 		}
758 	}
759 	err = erofs_init_managed_cache(sb);
760 	if (err)
761 		return err;
762 
763 	err = erofs_xattr_prefixes_init(sb);
764 	if (err)
765 		return err;
766 
767 	err = erofs_register_sysfs(sb);
768 	if (err)
769 		return err;
770 
771 	erofs_info(sb, "mounted with root inode @ nid %llu.", ROOT_NID(sbi));
772 	return 0;
773 }
774 
775 static int erofs_fc_anon_get_tree(struct fs_context *fc)
776 {
777 	return get_tree_nodev(fc, erofs_fc_fill_pseudo_super);
778 }
779 
780 static int erofs_fc_get_tree(struct fs_context *fc)
781 {
782 	struct erofs_fs_context *ctx = fc->fs_private;
783 
784 	if (IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && ctx->fsid)
785 		return get_tree_nodev(fc, erofs_fc_fill_super);
786 
787 	return get_tree_bdev(fc, erofs_fc_fill_super);
788 }
789 
790 static int erofs_fc_reconfigure(struct fs_context *fc)
791 {
792 	struct super_block *sb = fc->root->d_sb;
793 	struct erofs_sb_info *sbi = EROFS_SB(sb);
794 	struct erofs_fs_context *ctx = fc->fs_private;
795 
796 	DBG_BUGON(!sb_rdonly(sb));
797 
798 	if (ctx->fsid || ctx->domain_id)
799 		erofs_info(sb, "ignoring reconfiguration for fsid|domain_id.");
800 
801 	if (test_opt(&ctx->opt, POSIX_ACL))
802 		fc->sb_flags |= SB_POSIXACL;
803 	else
804 		fc->sb_flags &= ~SB_POSIXACL;
805 
806 	sbi->opt = ctx->opt;
807 
808 	fc->sb_flags |= SB_RDONLY;
809 	return 0;
810 }
811 
812 static int erofs_release_device_info(int id, void *ptr, void *data)
813 {
814 	struct erofs_device_info *dif = ptr;
815 
816 	fs_put_dax(dif->dax_dev, NULL);
817 	if (dif->bdev)
818 		blkdev_put(dif->bdev, FMODE_READ | FMODE_EXCL);
819 	erofs_fscache_unregister_cookie(dif->fscache);
820 	dif->fscache = NULL;
821 	kfree(dif->path);
822 	kfree(dif);
823 	return 0;
824 }
825 
826 static void erofs_free_dev_context(struct erofs_dev_context *devs)
827 {
828 	if (!devs)
829 		return;
830 	idr_for_each(&devs->tree, &erofs_release_device_info, NULL);
831 	idr_destroy(&devs->tree);
832 	kfree(devs);
833 }
834 
835 static void erofs_fc_free(struct fs_context *fc)
836 {
837 	struct erofs_fs_context *ctx = fc->fs_private;
838 
839 	erofs_free_dev_context(ctx->devs);
840 	kfree(ctx->fsid);
841 	kfree(ctx->domain_id);
842 	kfree(ctx);
843 }
844 
845 static const struct fs_context_operations erofs_context_ops = {
846 	.parse_param	= erofs_fc_parse_param,
847 	.get_tree       = erofs_fc_get_tree,
848 	.reconfigure    = erofs_fc_reconfigure,
849 	.free		= erofs_fc_free,
850 };
851 
852 static const struct fs_context_operations erofs_anon_context_ops = {
853 	.get_tree       = erofs_fc_anon_get_tree,
854 };
855 
856 static int erofs_init_fs_context(struct fs_context *fc)
857 {
858 	struct erofs_fs_context *ctx;
859 
860 	/* pseudo mount for anon inodes */
861 	if (fc->sb_flags & SB_KERNMOUNT) {
862 		fc->ops = &erofs_anon_context_ops;
863 		return 0;
864 	}
865 
866 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
867 	if (!ctx)
868 		return -ENOMEM;
869 	ctx->devs = kzalloc(sizeof(struct erofs_dev_context), GFP_KERNEL);
870 	if (!ctx->devs) {
871 		kfree(ctx);
872 		return -ENOMEM;
873 	}
874 	fc->fs_private = ctx;
875 
876 	idr_init(&ctx->devs->tree);
877 	init_rwsem(&ctx->devs->rwsem);
878 	erofs_default_options(ctx);
879 	fc->ops = &erofs_context_ops;
880 	return 0;
881 }
882 
883 /*
884  * could be triggered after deactivate_locked_super()
885  * is called, thus including umount and failed to initialize.
886  */
887 static void erofs_kill_sb(struct super_block *sb)
888 {
889 	struct erofs_sb_info *sbi;
890 
891 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
892 
893 	/* pseudo mount for anon inodes */
894 	if (sb->s_flags & SB_KERNMOUNT) {
895 		kill_anon_super(sb);
896 		return;
897 	}
898 
899 	if (erofs_is_fscache_mode(sb))
900 		kill_anon_super(sb);
901 	else
902 		kill_block_super(sb);
903 
904 	sbi = EROFS_SB(sb);
905 	if (!sbi)
906 		return;
907 
908 	erofs_free_dev_context(sbi->devs);
909 	fs_put_dax(sbi->dax_dev, NULL);
910 	erofs_fscache_unregister_fs(sb);
911 	kfree(sbi->fsid);
912 	kfree(sbi->domain_id);
913 	kfree(sbi);
914 	sb->s_fs_info = NULL;
915 }
916 
917 /* called when ->s_root is non-NULL */
918 static void erofs_put_super(struct super_block *sb)
919 {
920 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
921 
922 	DBG_BUGON(!sbi);
923 
924 	erofs_unregister_sysfs(sb);
925 	erofs_shrinker_unregister(sb);
926 	erofs_xattr_prefixes_cleanup(sb);
927 #ifdef CONFIG_EROFS_FS_ZIP
928 	iput(sbi->managed_cache);
929 	sbi->managed_cache = NULL;
930 #endif
931 	iput(sbi->packed_inode);
932 	sbi->packed_inode = NULL;
933 	erofs_free_dev_context(sbi->devs);
934 	sbi->devs = NULL;
935 	erofs_fscache_unregister_fs(sb);
936 }
937 
938 struct file_system_type erofs_fs_type = {
939 	.owner          = THIS_MODULE,
940 	.name           = "erofs",
941 	.init_fs_context = erofs_init_fs_context,
942 	.kill_sb        = erofs_kill_sb,
943 	.fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
944 };
945 MODULE_ALIAS_FS("erofs");
946 
947 static int __init erofs_module_init(void)
948 {
949 	int err;
950 
951 	erofs_check_ondisk_layout_definitions();
952 
953 	erofs_inode_cachep = kmem_cache_create("erofs_inode",
954 					       sizeof(struct erofs_inode), 0,
955 					       SLAB_RECLAIM_ACCOUNT,
956 					       erofs_inode_init_once);
957 	if (!erofs_inode_cachep) {
958 		err = -ENOMEM;
959 		goto icache_err;
960 	}
961 
962 	err = erofs_init_shrinker();
963 	if (err)
964 		goto shrinker_err;
965 
966 	err = z_erofs_lzma_init();
967 	if (err)
968 		goto lzma_err;
969 
970 	erofs_pcpubuf_init();
971 	err = z_erofs_init_zip_subsystem();
972 	if (err)
973 		goto zip_err;
974 
975 	err = erofs_init_sysfs();
976 	if (err)
977 		goto sysfs_err;
978 
979 	err = register_filesystem(&erofs_fs_type);
980 	if (err)
981 		goto fs_err;
982 
983 	return 0;
984 
985 fs_err:
986 	erofs_exit_sysfs();
987 sysfs_err:
988 	z_erofs_exit_zip_subsystem();
989 zip_err:
990 	z_erofs_lzma_exit();
991 lzma_err:
992 	erofs_exit_shrinker();
993 shrinker_err:
994 	kmem_cache_destroy(erofs_inode_cachep);
995 icache_err:
996 	return err;
997 }
998 
999 static void __exit erofs_module_exit(void)
1000 {
1001 	unregister_filesystem(&erofs_fs_type);
1002 
1003 	/* Ensure all RCU free inodes / pclusters are safe to be destroyed. */
1004 	rcu_barrier();
1005 
1006 	erofs_exit_sysfs();
1007 	z_erofs_exit_zip_subsystem();
1008 	z_erofs_lzma_exit();
1009 	erofs_exit_shrinker();
1010 	kmem_cache_destroy(erofs_inode_cachep);
1011 	erofs_pcpubuf_exit();
1012 }
1013 
1014 /* get filesystem statistics */
1015 static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
1016 {
1017 	struct super_block *sb = dentry->d_sb;
1018 	struct erofs_sb_info *sbi = EROFS_SB(sb);
1019 	u64 id = 0;
1020 
1021 	if (!erofs_is_fscache_mode(sb))
1022 		id = huge_encode_dev(sb->s_bdev->bd_dev);
1023 
1024 	buf->f_type = sb->s_magic;
1025 	buf->f_bsize = sb->s_blocksize;
1026 	buf->f_blocks = sbi->total_blocks;
1027 	buf->f_bfree = buf->f_bavail = 0;
1028 
1029 	buf->f_files = ULLONG_MAX;
1030 	buf->f_ffree = ULLONG_MAX - sbi->inos;
1031 
1032 	buf->f_namelen = EROFS_NAME_LEN;
1033 
1034 	buf->f_fsid    = u64_to_fsid(id);
1035 	return 0;
1036 }
1037 
1038 static int erofs_show_options(struct seq_file *seq, struct dentry *root)
1039 {
1040 	struct erofs_sb_info *sbi = EROFS_SB(root->d_sb);
1041 	struct erofs_mount_opts *opt = &sbi->opt;
1042 
1043 #ifdef CONFIG_EROFS_FS_XATTR
1044 	if (test_opt(opt, XATTR_USER))
1045 		seq_puts(seq, ",user_xattr");
1046 	else
1047 		seq_puts(seq, ",nouser_xattr");
1048 #endif
1049 #ifdef CONFIG_EROFS_FS_POSIX_ACL
1050 	if (test_opt(opt, POSIX_ACL))
1051 		seq_puts(seq, ",acl");
1052 	else
1053 		seq_puts(seq, ",noacl");
1054 #endif
1055 #ifdef CONFIG_EROFS_FS_ZIP
1056 	if (opt->cache_strategy == EROFS_ZIP_CACHE_DISABLED)
1057 		seq_puts(seq, ",cache_strategy=disabled");
1058 	else if (opt->cache_strategy == EROFS_ZIP_CACHE_READAHEAD)
1059 		seq_puts(seq, ",cache_strategy=readahead");
1060 	else if (opt->cache_strategy == EROFS_ZIP_CACHE_READAROUND)
1061 		seq_puts(seq, ",cache_strategy=readaround");
1062 #endif
1063 	if (test_opt(opt, DAX_ALWAYS))
1064 		seq_puts(seq, ",dax=always");
1065 	if (test_opt(opt, DAX_NEVER))
1066 		seq_puts(seq, ",dax=never");
1067 #ifdef CONFIG_EROFS_FS_ONDEMAND
1068 	if (sbi->fsid)
1069 		seq_printf(seq, ",fsid=%s", sbi->fsid);
1070 	if (sbi->domain_id)
1071 		seq_printf(seq, ",domain_id=%s", sbi->domain_id);
1072 #endif
1073 	return 0;
1074 }
1075 
1076 const struct super_operations erofs_sops = {
1077 	.put_super = erofs_put_super,
1078 	.alloc_inode = erofs_alloc_inode,
1079 	.free_inode = erofs_free_inode,
1080 	.statfs = erofs_statfs,
1081 	.show_options = erofs_show_options,
1082 };
1083 
1084 module_init(erofs_module_init);
1085 module_exit(erofs_module_exit);
1086 
1087 MODULE_DESCRIPTION("Enhanced ROM File System");
1088 MODULE_AUTHOR("Gao Xiang, Chao Yu, Miao Xie, CONSUMER BG, HUAWEI Inc.");
1089 MODULE_LICENSE("GPL");
1090