xref: /linux/fs/erofs/super.c (revision c7c707cbaa5ed277836364da4033e141ff985678)
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/statfs.h>
8 #include <linux/seq_file.h>
9 #include <linux/crc32c.h>
10 #include <linux/fs_context.h>
11 #include <linux/fs_parser.h>
12 #include <linux/exportfs.h>
13 #include <linux/backing-dev.h>
14 #include <linux/pseudo_fs.h>
15 #include "xattr.h"
16 
17 #define CREATE_TRACE_POINTS
18 #include <trace/events/erofs.h>
19 
20 static struct kmem_cache *erofs_inode_cachep __read_mostly;
21 
22 void _erofs_printk(struct super_block *sb, const char *fmt, ...)
23 {
24 	struct va_format vaf;
25 	va_list args;
26 	int level;
27 
28 	va_start(args, fmt);
29 
30 	level = printk_get_level(fmt);
31 	vaf.fmt = printk_skip_level(fmt);
32 	vaf.va = &args;
33 	if (sb)
34 		printk("%c%cerofs (device %s): %pV",
35 				KERN_SOH_ASCII, level, sb->s_id, &vaf);
36 	else
37 		printk("%c%cerofs: %pV", KERN_SOH_ASCII, level, &vaf);
38 	va_end(args);
39 }
40 
41 static int erofs_superblock_csum_verify(struct super_block *sb, void *sbdata)
42 {
43 	struct erofs_super_block *dsb = sbdata + EROFS_SUPER_OFFSET;
44 	u32 len = 1 << EROFS_SB(sb)->blkszbits, crc;
45 
46 	if (len > EROFS_SUPER_OFFSET)
47 		len -= EROFS_SUPER_OFFSET;
48 	len -= offsetof(struct erofs_super_block, checksum) +
49 			sizeof(dsb->checksum);
50 
51 	/* skip .magic(pre-verified) and .checksum(0) fields */
52 	crc = crc32c(0x5045B54A, (&dsb->checksum) + 1, len);
53 	if (crc == le32_to_cpu(dsb->checksum))
54 		return 0;
55 	erofs_err(sb, "invalid checksum 0x%08x, 0x%08x expected",
56 		  crc, le32_to_cpu(dsb->checksum));
57 	return -EBADMSG;
58 }
59 
60 static void erofs_inode_init_once(void *ptr)
61 {
62 	struct erofs_inode *vi = ptr;
63 
64 	inode_init_once(&vi->vfs_inode);
65 }
66 
67 static struct inode *erofs_alloc_inode(struct super_block *sb)
68 {
69 	struct erofs_inode *vi =
70 		alloc_inode_sb(sb, erofs_inode_cachep, GFP_KERNEL);
71 
72 	if (!vi)
73 		return NULL;
74 
75 	/* zero out everything except vfs_inode */
76 	memset(vi, 0, offsetof(struct erofs_inode, vfs_inode));
77 	return &vi->vfs_inode;
78 }
79 
80 static void erofs_free_inode(struct inode *inode)
81 {
82 	struct erofs_inode *vi = EROFS_I(inode);
83 
84 	if (inode->i_op == &erofs_fast_symlink_iops)
85 		kfree(inode->i_link);
86 	kfree(vi->xattr_shared_xattrs);
87 	kmem_cache_free(erofs_inode_cachep, vi);
88 }
89 
90 /* read variable-sized metadata, offset will be aligned by 4-byte */
91 void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
92 			  erofs_off_t *offset, int *lengthp)
93 {
94 	u8 *buffer, *ptr;
95 	int len, i, cnt;
96 
97 	*offset = round_up(*offset, 4);
98 	ptr = erofs_bread(buf, *offset, true);
99 	if (IS_ERR(ptr))
100 		return ptr;
101 
102 	len = le16_to_cpu(*(__le16 *)ptr);
103 	if (!len)
104 		len = U16_MAX + 1;
105 	buffer = kmalloc(len, GFP_KERNEL);
106 	if (!buffer)
107 		return ERR_PTR(-ENOMEM);
108 	*offset += sizeof(__le16);
109 	*lengthp = len;
110 
111 	for (i = 0; i < len; i += cnt) {
112 		cnt = min_t(int, sb->s_blocksize - erofs_blkoff(sb, *offset),
113 			    len - i);
114 		ptr = erofs_bread(buf, *offset, true);
115 		if (IS_ERR(ptr)) {
116 			kfree(buffer);
117 			return ptr;
118 		}
119 		memcpy(buffer + i, ptr, cnt);
120 		*offset += cnt;
121 	}
122 	return buffer;
123 }
124 
125 static int erofs_init_device(struct erofs_buf *buf, struct super_block *sb,
126 			     struct erofs_device_info *dif, erofs_off_t *pos)
127 {
128 	struct erofs_sb_info *sbi = EROFS_SB(sb);
129 	struct erofs_fscache *fscache;
130 	struct erofs_deviceslot *dis;
131 	struct file *file;
132 
133 	dis = erofs_read_metabuf(buf, sb, *pos, false);
134 	if (IS_ERR(dis))
135 		return PTR_ERR(dis);
136 
137 	if (!sbi->devs->flatdev && !dif->path) {
138 		if (!dis->tag[0]) {
139 			erofs_err(sb, "empty device tag @ pos %llu", *pos);
140 			return -EINVAL;
141 		}
142 		dif->path = kmemdup_nul(dis->tag, sizeof(dis->tag), GFP_KERNEL);
143 		if (!dif->path)
144 			return -ENOMEM;
145 	}
146 
147 	if (erofs_is_fscache_mode(sb)) {
148 		fscache = erofs_fscache_register_cookie(sb, dif->path, 0);
149 		if (IS_ERR(fscache))
150 			return PTR_ERR(fscache);
151 		dif->fscache = fscache;
152 	} else if (!sbi->devs->flatdev) {
153 		file = erofs_is_fileio_mode(sbi) ?
154 				filp_open(dif->path, O_RDONLY | O_LARGEFILE, 0) :
155 				bdev_file_open_by_path(dif->path,
156 						BLK_OPEN_READ, sb->s_type, NULL);
157 		if (IS_ERR(file)) {
158 			if (file == ERR_PTR(-ENOTBLK))
159 				return -EINVAL;
160 			return PTR_ERR(file);
161 		}
162 
163 		if (!erofs_is_fileio_mode(sbi)) {
164 			dif->dax_dev = fs_dax_get_by_bdev(file_bdev(file),
165 					&dif->dax_part_off, NULL, NULL);
166 		} else if (!S_ISREG(file_inode(file)->i_mode)) {
167 			fput(file);
168 			return -EINVAL;
169 		}
170 		if (!dif->dax_dev && test_opt(&sbi->opt, DAX_ALWAYS)) {
171 			erofs_info(sb, "DAX unsupported by %s. Turning off DAX.",
172 				   dif->path);
173 			clear_opt(&sbi->opt, DAX_ALWAYS);
174 		}
175 		dif->file = file;
176 	}
177 
178 	dif->blocks = le32_to_cpu(dis->blocks_lo);
179 	dif->uniaddr = le32_to_cpu(dis->uniaddr_lo);
180 	sbi->total_blocks += dif->blocks;
181 	*pos += EROFS_DEVT_SLOT_SIZE;
182 	return 0;
183 }
184 
185 static int erofs_scan_devices(struct super_block *sb,
186 			      struct erofs_super_block *dsb)
187 {
188 	struct erofs_sb_info *sbi = EROFS_SB(sb);
189 	unsigned int ondisk_extradevs;
190 	erofs_off_t pos;
191 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
192 	struct erofs_device_info *dif;
193 	int id, err = 0;
194 
195 	sbi->total_blocks = sbi->dif0.blocks;
196 	if (!erofs_sb_has_device_table(sbi))
197 		ondisk_extradevs = 0;
198 	else
199 		ondisk_extradevs = le16_to_cpu(dsb->extra_devices);
200 
201 	if (sbi->devs->extra_devices &&
202 	    ondisk_extradevs != sbi->devs->extra_devices) {
203 		erofs_err(sb, "extra devices don't match (ondisk %u, given %u)",
204 			  ondisk_extradevs, sbi->devs->extra_devices);
205 		return -EINVAL;
206 	}
207 
208 	if (test_opt(&sbi->opt, DAX_ALWAYS) && !sbi->dif0.dax_dev) {
209 		erofs_info(sb, "DAX unsupported by block device. Turning off DAX.");
210 		clear_opt(&sbi->opt, DAX_ALWAYS);
211 	}
212 	if (!ondisk_extradevs)
213 		return 0;
214 
215 	if (!sbi->devs->extra_devices && !erofs_is_fscache_mode(sb))
216 		sbi->devs->flatdev = true;
217 
218 	sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
219 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
220 	down_read(&sbi->devs->rwsem);
221 	if (sbi->devs->extra_devices) {
222 		idr_for_each_entry(&sbi->devs->tree, dif, id) {
223 			err = erofs_init_device(&buf, sb, dif, &pos);
224 			if (err)
225 				break;
226 		}
227 	} else {
228 		for (id = 0; id < ondisk_extradevs; id++) {
229 			dif = kzalloc(sizeof(*dif), GFP_KERNEL);
230 			if (!dif) {
231 				err = -ENOMEM;
232 				break;
233 			}
234 
235 			err = idr_alloc(&sbi->devs->tree, dif, 0, 0, GFP_KERNEL);
236 			if (err < 0) {
237 				kfree(dif);
238 				break;
239 			}
240 			++sbi->devs->extra_devices;
241 
242 			err = erofs_init_device(&buf, sb, dif, &pos);
243 			if (err)
244 				break;
245 		}
246 	}
247 	up_read(&sbi->devs->rwsem);
248 	erofs_put_metabuf(&buf);
249 	return err;
250 }
251 
252 static int erofs_read_superblock(struct super_block *sb)
253 {
254 	struct erofs_sb_info *sbi = EROFS_SB(sb);
255 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
256 	struct erofs_super_block *dsb;
257 	void *data;
258 	int ret;
259 
260 	data = erofs_read_metabuf(&buf, sb, 0, false);
261 	if (IS_ERR(data)) {
262 		erofs_err(sb, "cannot read erofs superblock");
263 		return PTR_ERR(data);
264 	}
265 
266 	dsb = (struct erofs_super_block *)(data + EROFS_SUPER_OFFSET);
267 	ret = -EINVAL;
268 	if (le32_to_cpu(dsb->magic) != EROFS_SUPER_MAGIC_V1) {
269 		erofs_err(sb, "cannot find valid erofs superblock");
270 		goto out;
271 	}
272 
273 	sbi->blkszbits = dsb->blkszbits;
274 	if (sbi->blkszbits < 9 || sbi->blkszbits > PAGE_SHIFT) {
275 		erofs_err(sb, "blkszbits %u isn't supported", sbi->blkszbits);
276 		goto out;
277 	}
278 	if (dsb->dirblkbits) {
279 		erofs_err(sb, "dirblkbits %u isn't supported", dsb->dirblkbits);
280 		goto out;
281 	}
282 
283 	sbi->feature_compat = le32_to_cpu(dsb->feature_compat);
284 	if (erofs_sb_has_sb_chksum(sbi)) {
285 		ret = erofs_superblock_csum_verify(sb, data);
286 		if (ret)
287 			goto out;
288 	}
289 
290 	ret = -EINVAL;
291 	sbi->feature_incompat = le32_to_cpu(dsb->feature_incompat);
292 	if (sbi->feature_incompat & ~EROFS_ALL_FEATURE_INCOMPAT) {
293 		erofs_err(sb, "unidentified incompatible feature %x, please upgrade kernel",
294 			  sbi->feature_incompat & ~EROFS_ALL_FEATURE_INCOMPAT);
295 		goto out;
296 	}
297 
298 	sbi->sb_size = 128 + dsb->sb_extslots * EROFS_SB_EXTSLOT_SIZE;
299 	if (sbi->sb_size > PAGE_SIZE - EROFS_SUPER_OFFSET) {
300 		erofs_err(sb, "invalid sb_extslots %u (more than a fs block)",
301 			  sbi->sb_size);
302 		goto out;
303 	}
304 	sbi->dif0.blocks = le32_to_cpu(dsb->blocks_lo);
305 	sbi->meta_blkaddr = le32_to_cpu(dsb->meta_blkaddr);
306 #ifdef CONFIG_EROFS_FS_XATTR
307 	sbi->xattr_blkaddr = le32_to_cpu(dsb->xattr_blkaddr);
308 	sbi->xattr_prefix_start = le32_to_cpu(dsb->xattr_prefix_start);
309 	sbi->xattr_prefix_count = dsb->xattr_prefix_count;
310 	sbi->xattr_filter_reserved = dsb->xattr_filter_reserved;
311 	if (erofs_sb_has_ishare_xattrs(sbi)) {
312 		if (dsb->ishare_xattr_prefix_id >= sbi->xattr_prefix_count) {
313 			erofs_err(sb, "invalid ishare xattr prefix id %u",
314 				  dsb->ishare_xattr_prefix_id);
315 			ret = -EFSCORRUPTED;
316 			goto out;
317 		}
318 		sbi->ishare_xattr_prefix_id = dsb->ishare_xattr_prefix_id;
319 	}
320 #endif
321 	sbi->islotbits = ilog2(sizeof(struct erofs_inode_compact));
322 	if (erofs_sb_has_48bit(sbi) && dsb->rootnid_8b) {
323 		sbi->root_nid = le64_to_cpu(dsb->rootnid_8b);
324 		sbi->dif0.blocks = sbi->dif0.blocks |
325 				((u64)le16_to_cpu(dsb->rb.blocks_hi) << 32);
326 	} else {
327 		sbi->root_nid = le16_to_cpu(dsb->rb.rootnid_2b);
328 	}
329 	sbi->packed_nid = le64_to_cpu(dsb->packed_nid);
330 	if (erofs_sb_has_metabox(sbi)) {
331 		ret = -EFSCORRUPTED;
332 		if (sbi->sb_size <= offsetof(struct erofs_super_block,
333 					     metabox_nid))
334 			goto out;
335 		sbi->metabox_nid = le64_to_cpu(dsb->metabox_nid);
336 		if (sbi->metabox_nid & BIT_ULL(EROFS_DIRENT_NID_METABOX_BIT))
337 			goto out;		/* self-loop detection */
338 	}
339 	sbi->inos = le64_to_cpu(dsb->inos);
340 
341 	sbi->epoch = (s64)le64_to_cpu(dsb->epoch);
342 	sbi->fixed_nsec = le32_to_cpu(dsb->fixed_nsec);
343 	super_set_uuid(sb, (void *)dsb->uuid, sizeof(dsb->uuid));
344 
345 	if (dsb->volume_name[0]) {
346 		sbi->volume_name = kstrndup(dsb->volume_name,
347 					    sizeof(dsb->volume_name), GFP_KERNEL);
348 		if (!sbi->volume_name) {
349 			ret = -ENOMEM;
350 			goto out;
351 		}
352 	}
353 
354 	if (IS_ENABLED(CONFIG_EROFS_FS_ZIP)) {
355 		ret = z_erofs_parse_cfgs(sb, dsb);
356 		if (ret < 0)
357 			goto out;
358 	} else if (dsb->u1.available_compr_algs ||
359 		   erofs_sb_has_lz4_0padding(sbi)) {
360 		erofs_err(sb, "compression disabled, unable to mount compressed EROFS");
361 		ret = -EOPNOTSUPP;
362 		goto out;
363 	}
364 
365 	ret = erofs_scan_devices(sb, dsb);
366 
367 	if (erofs_sb_has_48bit(sbi))
368 		erofs_info(sb, "EXPERIMENTAL 48-bit layout support in use. Use at your own risk!");
369 	if (erofs_sb_has_metabox(sbi))
370 		erofs_info(sb, "EXPERIMENTAL metadata compression support in use. Use at your own risk!");
371 	if (erofs_is_fscache_mode(sb))
372 		erofs_info(sb, "[deprecated] fscache-based on-demand read feature in use. Use at your own risk!");
373 out:
374 	erofs_put_metabuf(&buf);
375 	return ret;
376 }
377 
378 static void erofs_default_options(struct erofs_sb_info *sbi)
379 {
380 #ifdef CONFIG_EROFS_FS_ZIP
381 	sbi->opt.cache_strategy = EROFS_ZIP_CACHE_READAROUND;
382 	sbi->sync_decompress = EROFS_SYNC_DECOMPRESS_AUTO;
383 #endif
384 	if (IS_ENABLED(CONFIG_EROFS_FS_XATTR))
385 		set_opt(&sbi->opt, XATTR_USER);
386 	if (IS_ENABLED(CONFIG_EROFS_FS_POSIX_ACL))
387 		set_opt(&sbi->opt, POSIX_ACL);
388 }
389 
390 enum {
391 	Opt_user_xattr, Opt_acl, Opt_cache_strategy, Opt_dax, Opt_dax_enum,
392 	Opt_device, Opt_fsid, Opt_domain_id, Opt_directio, Opt_fsoffset,
393 	Opt_inode_share,
394 };
395 
396 static const struct constant_table erofs_param_cache_strategy[] = {
397 	{"disabled",	EROFS_ZIP_CACHE_DISABLED},
398 	{"readahead",	EROFS_ZIP_CACHE_READAHEAD},
399 	{"readaround",	EROFS_ZIP_CACHE_READAROUND},
400 	{}
401 };
402 
403 static const struct constant_table erofs_dax_param_enums[] = {
404 	{"always",	EROFS_MOUNT_DAX_ALWAYS},
405 	{"never",	EROFS_MOUNT_DAX_NEVER},
406 	{}
407 };
408 
409 static const struct fs_parameter_spec erofs_fs_parameters[] = {
410 	fsparam_flag_no("user_xattr",	Opt_user_xattr),
411 	fsparam_flag_no("acl",		Opt_acl),
412 	fsparam_enum("cache_strategy",	Opt_cache_strategy,
413 		     erofs_param_cache_strategy),
414 	fsparam_flag("dax",             Opt_dax),
415 	fsparam_enum("dax",		Opt_dax_enum, erofs_dax_param_enums),
416 	fsparam_string("device",	Opt_device),
417 	fsparam_string("fsid",		Opt_fsid),
418 	fsparam_string("domain_id",	Opt_domain_id),
419 	fsparam_flag_no("directio",	Opt_directio),
420 	fsparam_u64("fsoffset",		Opt_fsoffset),
421 	fsparam_flag("inode_share",	Opt_inode_share),
422 	{}
423 };
424 
425 static bool erofs_fc_set_dax_mode(struct fs_context *fc, unsigned int mode)
426 {
427 #ifdef CONFIG_FS_DAX
428 	struct erofs_sb_info *sbi = fc->s_fs_info;
429 
430 	switch (mode) {
431 	case EROFS_MOUNT_DAX_ALWAYS:
432 		set_opt(&sbi->opt, DAX_ALWAYS);
433 		clear_opt(&sbi->opt, DAX_NEVER);
434 		return true;
435 	case EROFS_MOUNT_DAX_NEVER:
436 		set_opt(&sbi->opt, DAX_NEVER);
437 		clear_opt(&sbi->opt, DAX_ALWAYS);
438 		return true;
439 	default:
440 		DBG_BUGON(1);
441 		return false;
442 	}
443 #else
444 	errorfc(fc, "dax options not supported");
445 	return false;
446 #endif
447 }
448 
449 static int erofs_fc_parse_param(struct fs_context *fc,
450 				struct fs_parameter *param)
451 {
452 	struct erofs_sb_info *sbi = fc->s_fs_info;
453 	struct fs_parse_result result;
454 	struct erofs_device_info *dif;
455 	int opt, ret;
456 
457 	opt = fs_parse(fc, erofs_fs_parameters, param, &result);
458 	if (opt < 0)
459 		return opt;
460 
461 	switch (opt) {
462 	case Opt_user_xattr:
463 #ifdef CONFIG_EROFS_FS_XATTR
464 		if (result.boolean)
465 			set_opt(&sbi->opt, XATTR_USER);
466 		else
467 			clear_opt(&sbi->opt, XATTR_USER);
468 #else
469 		errorfc(fc, "{,no}user_xattr options not supported");
470 #endif
471 		break;
472 	case Opt_acl:
473 #ifdef CONFIG_EROFS_FS_POSIX_ACL
474 		if (result.boolean)
475 			set_opt(&sbi->opt, POSIX_ACL);
476 		else
477 			clear_opt(&sbi->opt, POSIX_ACL);
478 #else
479 		errorfc(fc, "{,no}acl options not supported");
480 #endif
481 		break;
482 	case Opt_cache_strategy:
483 #ifdef CONFIG_EROFS_FS_ZIP
484 		sbi->opt.cache_strategy = result.uint_32;
485 #else
486 		errorfc(fc, "compression not supported, cache_strategy ignored");
487 #endif
488 		break;
489 	case Opt_dax:
490 		if (!erofs_fc_set_dax_mode(fc, EROFS_MOUNT_DAX_ALWAYS))
491 			return -EINVAL;
492 		break;
493 	case Opt_dax_enum:
494 		if (!erofs_fc_set_dax_mode(fc, result.uint_32))
495 			return -EINVAL;
496 		break;
497 	case Opt_device:
498 		dif = kzalloc(sizeof(*dif), GFP_KERNEL);
499 		if (!dif)
500 			return -ENOMEM;
501 		dif->path = kstrdup(param->string, GFP_KERNEL);
502 		if (!dif->path) {
503 			kfree(dif);
504 			return -ENOMEM;
505 		}
506 		down_write(&sbi->devs->rwsem);
507 		ret = idr_alloc(&sbi->devs->tree, dif, 0, 0, GFP_KERNEL);
508 		up_write(&sbi->devs->rwsem);
509 		if (ret < 0) {
510 			kfree(dif->path);
511 			kfree(dif);
512 			return ret;
513 		}
514 		++sbi->devs->extra_devices;
515 		break;
516 #ifdef CONFIG_EROFS_FS_ONDEMAND
517 	case Opt_fsid:
518 		kfree(sbi->fsid);
519 		sbi->fsid = kstrdup(param->string, GFP_KERNEL);
520 		if (!sbi->fsid)
521 			return -ENOMEM;
522 		break;
523 #endif
524 #if defined(CONFIG_EROFS_FS_ONDEMAND) || defined(CONFIG_EROFS_FS_PAGE_CACHE_SHARE)
525 	case Opt_domain_id:
526 		kfree_sensitive(sbi->domain_id);
527 		sbi->domain_id = no_free_ptr(param->string);
528 		break;
529 #else
530 	case Opt_fsid:
531 	case Opt_domain_id:
532 		errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
533 		break;
534 #endif
535 	case Opt_directio:
536 #ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
537 		if (result.boolean)
538 			set_opt(&sbi->opt, DIRECT_IO);
539 		else
540 			clear_opt(&sbi->opt, DIRECT_IO);
541 #else
542 		errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
543 #endif
544 		break;
545 	case Opt_fsoffset:
546 		sbi->dif0.fsoff = result.uint_64;
547 		break;
548 	case Opt_inode_share:
549 #ifdef CONFIG_EROFS_FS_PAGE_CACHE_SHARE
550 		set_opt(&sbi->opt, INODE_SHARE);
551 #else
552 		errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
553 #endif
554 		break;
555 	}
556 	return 0;
557 }
558 
559 static int erofs_encode_fh(struct inode *inode, u32 *fh, int *max_len,
560 			   struct inode *parent)
561 {
562 	erofs_nid_t nid = EROFS_I(inode)->nid;
563 	int len = parent ? 6 : 3;
564 
565 	if (*max_len < len) {
566 		*max_len = len;
567 		return FILEID_INVALID;
568 	}
569 
570 	fh[0] = (u32)(nid >> 32);
571 	fh[1] = (u32)(nid & 0xffffffff);
572 	fh[2] = inode->i_generation;
573 
574 	if (parent) {
575 		nid = EROFS_I(parent)->nid;
576 
577 		fh[3] = (u32)(nid >> 32);
578 		fh[4] = (u32)(nid & 0xffffffff);
579 		fh[5] = parent->i_generation;
580 	}
581 
582 	*max_len = len;
583 	return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
584 }
585 
586 static struct dentry *erofs_fh_to_dentry(struct super_block *sb,
587 		struct fid *fid, int fh_len, int fh_type)
588 {
589 	if ((fh_type != FILEID_INO64_GEN &&
590 	     fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
591 		return NULL;
592 
593 	return d_obtain_alias(erofs_iget(sb,
594 		((u64)fid->raw[0] << 32) | fid->raw[1]));
595 }
596 
597 static struct dentry *erofs_fh_to_parent(struct super_block *sb,
598 		struct fid *fid, int fh_len, int fh_type)
599 {
600 	if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
601 		return NULL;
602 
603 	return d_obtain_alias(erofs_iget(sb,
604 		((u64)fid->raw[3] << 32) | fid->raw[4]));
605 }
606 
607 static struct dentry *erofs_get_parent(struct dentry *child)
608 {
609 	erofs_nid_t nid;
610 	unsigned int d_type;
611 	int err;
612 
613 	err = erofs_namei(d_inode(child), &dotdot_name, &nid, &d_type);
614 	if (err)
615 		return ERR_PTR(err);
616 	return d_obtain_alias(erofs_iget(child->d_sb, nid));
617 }
618 
619 static const struct export_operations erofs_export_ops = {
620 	.encode_fh = erofs_encode_fh,
621 	.fh_to_dentry = erofs_fh_to_dentry,
622 	.fh_to_parent = erofs_fh_to_parent,
623 	.get_parent = erofs_get_parent,
624 };
625 
626 static void erofs_set_sysfs_name(struct super_block *sb)
627 {
628 	struct erofs_sb_info *sbi = EROFS_SB(sb);
629 
630 	if (sbi->domain_id && sbi->fsid)
631 		super_set_sysfs_name_generic(sb, "%s,%s", sbi->domain_id,
632 					     sbi->fsid);
633 	else if (sbi->fsid)
634 		super_set_sysfs_name_generic(sb, "%s", sbi->fsid);
635 	else if (erofs_is_fileio_mode(sbi))
636 		super_set_sysfs_name_generic(sb, "%s",
637 					     bdi_dev_name(sb->s_bdi));
638 	else
639 		super_set_sysfs_name_id(sb);
640 }
641 
642 static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
643 {
644 	struct inode *inode;
645 	struct erofs_sb_info *sbi = EROFS_SB(sb);
646 	int err;
647 
648 	sb->s_magic = EROFS_SUPER_MAGIC;
649 	sb->s_flags |= SB_RDONLY | SB_NOATIME;
650 	sb->s_maxbytes = MAX_LFS_FILESIZE;
651 	sb->s_op = &erofs_sops;
652 
653 	if (!sbi->domain_id && test_opt(&sbi->opt, INODE_SHARE)) {
654 		errorfc(fc, "domain_id is needed when inode_ishare is on");
655 		return -EINVAL;
656 	}
657 	if (test_opt(&sbi->opt, DAX_ALWAYS) && test_opt(&sbi->opt, INODE_SHARE)) {
658 		errorfc(fc, "FSDAX is not allowed when inode_ishare is on");
659 		return -EINVAL;
660 	}
661 
662 	sbi->blkszbits = PAGE_SHIFT;
663 	if (!sb->s_bdev) {
664 		/*
665 		 * (File-backed mounts) EROFS claims it's safe to nest other
666 		 * fs contexts (including its own) due to self-controlled RO
667 		 * accesses/contexts and no side-effect changes that need to
668 		 * context save & restore so it can reuse the current thread
669 		 * context.
670 		 * However, we still need to prevent kernel stack overflow due
671 		 * to filesystem nesting: just ensure that s_stack_depth is 0
672 		 * to disallow mounting EROFS on stacked filesystems.
673 		 * Note: s_stack_depth is not incremented here for now, since
674 		 * EROFS is the only fs supporting file-backed mounts for now.
675 		 * It MUST change if another fs plans to support them, which
676 		 * may also require adjusting FILESYSTEM_MAX_STACK_DEPTH.
677 		 */
678 		if (erofs_is_fileio_mode(sbi)) {
679 			inode = file_inode(sbi->dif0.file);
680 			if ((inode->i_sb->s_op == &erofs_sops &&
681 			     !inode->i_sb->s_bdev) ||
682 			    inode->i_sb->s_stack_depth) {
683 				erofs_err(sb, "file-backed mounts cannot be applied to stacked fses");
684 				return -ENOTBLK;
685 			}
686 		}
687 		sb->s_blocksize = PAGE_SIZE;
688 		sb->s_blocksize_bits = PAGE_SHIFT;
689 
690 		if (erofs_is_fscache_mode(sb)) {
691 			err = erofs_fscache_register_fs(sb);
692 			if (err)
693 				return err;
694 		}
695 		err = super_setup_bdi(sb);
696 		if (err)
697 			return err;
698 	} else {
699 		if (!sb_set_blocksize(sb, PAGE_SIZE)) {
700 			errorfc(fc, "failed to set initial blksize");
701 			return -EINVAL;
702 		}
703 
704 		sbi->dif0.dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
705 				&sbi->dif0.dax_part_off, NULL, NULL);
706 	}
707 
708 	err = erofs_read_superblock(sb);
709 	if (err)
710 		return err;
711 
712 	if (sb->s_blocksize_bits != sbi->blkszbits) {
713 		if (erofs_is_fscache_mode(sb)) {
714 			errorfc(fc, "unsupported blksize for fscache mode");
715 			return -EINVAL;
716 		}
717 
718 		if (erofs_is_fileio_mode(sbi)) {
719 			sb->s_blocksize = 1 << sbi->blkszbits;
720 			sb->s_blocksize_bits = sbi->blkszbits;
721 		} else if (!sb_set_blocksize(sb, 1 << sbi->blkszbits)) {
722 			errorfc(fc, "failed to set erofs blksize");
723 			return -EINVAL;
724 		}
725 	}
726 
727 	if (sbi->dif0.fsoff) {
728 		if (sbi->dif0.fsoff & (sb->s_blocksize - 1))
729 			return invalfc(fc, "fsoffset %llu is not aligned to block size %lu",
730 				       sbi->dif0.fsoff, sb->s_blocksize);
731 		if (erofs_is_fscache_mode(sb))
732 			return invalfc(fc, "cannot use fsoffset in fscache mode");
733 	}
734 
735 	if (test_opt(&sbi->opt, DAX_ALWAYS) && sbi->blkszbits != PAGE_SHIFT) {
736 		erofs_info(sb, "unsupported blocksize for DAX");
737 		clear_opt(&sbi->opt, DAX_ALWAYS);
738 	}
739 	if (test_opt(&sbi->opt, INODE_SHARE) && !erofs_sb_has_ishare_xattrs(sbi)) {
740 		erofs_info(sb, "on-disk ishare xattrs not found. Turning off inode_share.");
741 		clear_opt(&sbi->opt, INODE_SHARE);
742 	}
743 	if (test_opt(&sbi->opt, INODE_SHARE))
744 		erofs_info(sb, "EXPERIMENTAL EROFS page cache share support in use. Use at your own risk!");
745 
746 	sb->s_time_gran = 1;
747 	sb->s_xattr = erofs_xattr_handlers;
748 	sb->s_export_op = &erofs_export_ops;
749 
750 	if (test_opt(&sbi->opt, POSIX_ACL))
751 		sb->s_flags |= SB_POSIXACL;
752 	else
753 		sb->s_flags &= ~SB_POSIXACL;
754 
755 	err = z_erofs_init_super(sb);
756 	if (err)
757 		return err;
758 
759 	if (erofs_sb_has_fragments(sbi) && sbi->packed_nid) {
760 		inode = erofs_iget(sb, sbi->packed_nid);
761 		if (IS_ERR(inode))
762 			return PTR_ERR(inode);
763 		sbi->packed_inode = inode;
764 	}
765 	if (erofs_sb_has_metabox(sbi)) {
766 		inode = erofs_iget(sb, sbi->metabox_nid);
767 		if (IS_ERR(inode))
768 			return PTR_ERR(inode);
769 		sbi->metabox_inode = inode;
770 	}
771 
772 	inode = erofs_iget(sb, sbi->root_nid);
773 	if (IS_ERR(inode))
774 		return PTR_ERR(inode);
775 
776 	if (!S_ISDIR(inode->i_mode)) {
777 		erofs_err(sb, "rootino(nid %llu) is not a directory(i_mode %o)",
778 			  sbi->root_nid, inode->i_mode);
779 		iput(inode);
780 		return -EINVAL;
781 	}
782 	sb->s_root = d_make_root(inode);
783 	if (!sb->s_root)
784 		return -ENOMEM;
785 
786 	erofs_shrinker_register(sb);
787 	err = erofs_xattr_prefixes_init(sb);
788 	if (err)
789 		return err;
790 
791 	erofs_set_sysfs_name(sb);
792 	err = erofs_register_sysfs(sb);
793 	if (err)
794 		return err;
795 
796 	sbi->dir_ra_bytes = EROFS_DIR_RA_BYTES;
797 	erofs_info(sb, "mounted with root inode @ nid %llu.", sbi->root_nid);
798 	return 0;
799 }
800 
801 static int erofs_fc_get_tree(struct fs_context *fc)
802 {
803 	struct erofs_sb_info *sbi = fc->s_fs_info;
804 	int ret;
805 
806 	if (IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && sbi->fsid)
807 		return get_tree_nodev(fc, erofs_fc_fill_super);
808 
809 	ret = get_tree_bdev_flags(fc, erofs_fc_fill_super,
810 		IS_ENABLED(CONFIG_EROFS_FS_BACKED_BY_FILE) ?
811 			GET_TREE_BDEV_QUIET_LOOKUP : 0);
812 #ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
813 	if (ret == -ENOTBLK) {
814 		struct file *file;
815 
816 		if (!fc->source)
817 			return invalf(fc, "No source specified");
818 		file = filp_open(fc->source, O_RDONLY | O_LARGEFILE, 0);
819 		if (IS_ERR(file))
820 			return PTR_ERR(file);
821 		sbi->dif0.file = file;
822 
823 		if (S_ISREG(file_inode(sbi->dif0.file)->i_mode) &&
824 		    sbi->dif0.file->f_mapping->a_ops->read_folio)
825 			return get_tree_nodev(fc, erofs_fc_fill_super);
826 	}
827 #endif
828 	return ret;
829 }
830 
831 static int erofs_fc_reconfigure(struct fs_context *fc)
832 {
833 	struct super_block *sb = fc->root->d_sb;
834 	struct erofs_sb_info *sbi = EROFS_SB(sb);
835 	struct erofs_sb_info *new_sbi = fc->s_fs_info;
836 
837 	DBG_BUGON(!sb_rdonly(sb));
838 
839 	if (new_sbi->fsid || new_sbi->domain_id)
840 		erofs_info(sb, "ignoring reconfiguration for fsid|domain_id.");
841 
842 	if (test_opt(&new_sbi->opt, POSIX_ACL))
843 		fc->sb_flags |= SB_POSIXACL;
844 	else
845 		fc->sb_flags &= ~SB_POSIXACL;
846 
847 	sbi->opt = new_sbi->opt;
848 
849 	fc->sb_flags |= SB_RDONLY;
850 	return 0;
851 }
852 
853 static int erofs_release_device_info(int id, void *ptr, void *data)
854 {
855 	struct erofs_device_info *dif = ptr;
856 
857 	fs_put_dax(dif->dax_dev, NULL);
858 	if (dif->file)
859 		fput(dif->file);
860 	erofs_fscache_unregister_cookie(dif->fscache);
861 	dif->fscache = NULL;
862 	kfree(dif->path);
863 	kfree(dif);
864 	return 0;
865 }
866 
867 static void erofs_free_dev_context(struct erofs_dev_context *devs)
868 {
869 	if (!devs)
870 		return;
871 	idr_for_each(&devs->tree, &erofs_release_device_info, NULL);
872 	idr_destroy(&devs->tree);
873 	kfree(devs);
874 }
875 
876 static void erofs_sb_free(struct erofs_sb_info *sbi)
877 {
878 	erofs_free_dev_context(sbi->devs);
879 	kfree(sbi->fsid);
880 	kfree_sensitive(sbi->domain_id);
881 	if (sbi->dif0.file)
882 		fput(sbi->dif0.file);
883 	kfree(sbi->volume_name);
884 	kfree(sbi);
885 }
886 
887 static void erofs_fc_free(struct fs_context *fc)
888 {
889 	struct erofs_sb_info *sbi = fc->s_fs_info;
890 
891 	if (sbi) /* free here if an error occurs before transferring to sb */
892 		erofs_sb_free(sbi);
893 }
894 
895 static const struct fs_context_operations erofs_context_ops = {
896 	.parse_param	= erofs_fc_parse_param,
897 	.get_tree       = erofs_fc_get_tree,
898 	.reconfigure    = erofs_fc_reconfigure,
899 	.free		= erofs_fc_free,
900 };
901 
902 static int erofs_init_fs_context(struct fs_context *fc)
903 {
904 	struct erofs_sb_info *sbi;
905 
906 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
907 	if (!sbi)
908 		return -ENOMEM;
909 
910 	sbi->devs = kzalloc(sizeof(struct erofs_dev_context), GFP_KERNEL);
911 	if (!sbi->devs) {
912 		kfree(sbi);
913 		return -ENOMEM;
914 	}
915 	fc->s_fs_info = sbi;
916 
917 	idr_init(&sbi->devs->tree);
918 	init_rwsem(&sbi->devs->rwsem);
919 	erofs_default_options(sbi);
920 	fc->ops = &erofs_context_ops;
921 	return 0;
922 }
923 
924 static void erofs_drop_internal_inodes(struct erofs_sb_info *sbi)
925 {
926 	iput(sbi->packed_inode);
927 	sbi->packed_inode = NULL;
928 	iput(sbi->metabox_inode);
929 	sbi->metabox_inode = NULL;
930 #ifdef CONFIG_EROFS_FS_ZIP
931 	iput(sbi->managed_cache);
932 	sbi->managed_cache = NULL;
933 #endif
934 }
935 
936 static void erofs_kill_sb(struct super_block *sb)
937 {
938 	struct erofs_sb_info *sbi = EROFS_SB(sb);
939 
940 	if ((IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && sbi->fsid) ||
941 	    sbi->dif0.file)
942 		kill_anon_super(sb);
943 	else
944 		kill_block_super(sb);
945 	erofs_drop_internal_inodes(sbi);
946 	fs_put_dax(sbi->dif0.dax_dev, NULL);
947 	erofs_fscache_unregister_fs(sb);
948 	erofs_sb_free(sbi);
949 	sb->s_fs_info = NULL;
950 }
951 
952 static void erofs_put_super(struct super_block *sb)
953 {
954 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
955 
956 	erofs_unregister_sysfs(sb);
957 	erofs_shrinker_unregister(sb);
958 	erofs_xattr_prefixes_cleanup(sb);
959 	erofs_drop_internal_inodes(sbi);
960 	erofs_free_dev_context(sbi->devs);
961 	sbi->devs = NULL;
962 	erofs_fscache_unregister_fs(sb);
963 }
964 
965 static struct file_system_type erofs_fs_type = {
966 	.owner          = THIS_MODULE,
967 	.name           = "erofs",
968 	.init_fs_context = erofs_init_fs_context,
969 	.kill_sb        = erofs_kill_sb,
970 	.fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
971 };
972 MODULE_ALIAS_FS("erofs");
973 
974 #if defined(CONFIG_EROFS_FS_ONDEMAND) || defined(CONFIG_EROFS_FS_PAGE_CACHE_SHARE)
975 static void erofs_free_anon_inode(struct inode *inode)
976 {
977 	struct erofs_inode *vi = EROFS_I(inode);
978 
979 #ifdef CONFIG_EROFS_FS_PAGE_CACHE_SHARE
980 	kfree(vi->fingerprint.opaque);
981 #endif
982 	kmem_cache_free(erofs_inode_cachep, vi);
983 }
984 
985 static const struct super_operations erofs_anon_sops = {
986 	.alloc_inode = erofs_alloc_inode,
987 	.drop_inode = inode_just_drop,
988 	.free_inode = erofs_free_anon_inode,
989 };
990 
991 static int erofs_anon_init_fs_context(struct fs_context *fc)
992 {
993 	struct pseudo_fs_context *ctx;
994 
995 	ctx = init_pseudo(fc, EROFS_SUPER_MAGIC);
996 	if (!ctx)
997 		return -ENOMEM;
998 	ctx->ops = &erofs_anon_sops;
999 	return 0;
1000 }
1001 
1002 struct file_system_type erofs_anon_fs_type = {
1003 	.name           = "pseudo_erofs",
1004 	.init_fs_context = erofs_anon_init_fs_context,
1005 	.kill_sb        = kill_anon_super,
1006 };
1007 #endif
1008 
1009 static int __init erofs_module_init(void)
1010 {
1011 	int err;
1012 
1013 	erofs_check_ondisk_layout_definitions();
1014 
1015 	erofs_inode_cachep = kmem_cache_create("erofs_inode",
1016 			sizeof(struct erofs_inode), 0,
1017 			SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
1018 			erofs_inode_init_once);
1019 	if (!erofs_inode_cachep)
1020 		return -ENOMEM;
1021 
1022 	err = erofs_init_shrinker();
1023 	if (err)
1024 		goto shrinker_err;
1025 
1026 	err = z_erofs_init_subsystem();
1027 	if (err)
1028 		goto zip_err;
1029 
1030 	err = erofs_init_sysfs();
1031 	if (err)
1032 		goto sysfs_err;
1033 
1034 	err = erofs_init_ishare();
1035 	if (err)
1036 		goto ishare_err;
1037 
1038 	err = register_filesystem(&erofs_fs_type);
1039 	if (err)
1040 		goto fs_err;
1041 
1042 	return 0;
1043 
1044 fs_err:
1045 	erofs_exit_ishare();
1046 ishare_err:
1047 	erofs_exit_sysfs();
1048 sysfs_err:
1049 	z_erofs_exit_subsystem();
1050 zip_err:
1051 	erofs_exit_shrinker();
1052 shrinker_err:
1053 	kmem_cache_destroy(erofs_inode_cachep);
1054 	return err;
1055 }
1056 
1057 static void __exit erofs_module_exit(void)
1058 {
1059 	unregister_filesystem(&erofs_fs_type);
1060 
1061 	/* Ensure all RCU free inodes / pclusters are safe to be destroyed. */
1062 	rcu_barrier();
1063 
1064 	erofs_exit_ishare();
1065 	erofs_exit_sysfs();
1066 	z_erofs_exit_subsystem();
1067 	erofs_exit_shrinker();
1068 	kmem_cache_destroy(erofs_inode_cachep);
1069 }
1070 
1071 static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
1072 {
1073 	struct super_block *sb = dentry->d_sb;
1074 	struct erofs_sb_info *sbi = EROFS_SB(sb);
1075 
1076 	buf->f_type = sb->s_magic;
1077 	buf->f_bsize = sb->s_blocksize;
1078 	buf->f_blocks = sbi->total_blocks;
1079 	buf->f_bfree = buf->f_bavail = 0;
1080 	buf->f_files = ULLONG_MAX;
1081 	buf->f_ffree = ULLONG_MAX - sbi->inos;
1082 	buf->f_namelen = EROFS_NAME_LEN;
1083 
1084 	if (uuid_is_null(&sb->s_uuid))
1085 		buf->f_fsid = u64_to_fsid(!sb->s_bdev ? 0 :
1086 				huge_encode_dev(sb->s_bdev->bd_dev));
1087 	else
1088 		buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
1089 	return 0;
1090 }
1091 
1092 static int erofs_show_options(struct seq_file *seq, struct dentry *root)
1093 {
1094 	struct erofs_sb_info *sbi = EROFS_SB(root->d_sb);
1095 	struct erofs_mount_opts *opt = &sbi->opt;
1096 
1097 	if (IS_ENABLED(CONFIG_EROFS_FS_XATTR))
1098 		seq_puts(seq, test_opt(opt, XATTR_USER) ?
1099 				",user_xattr" : ",nouser_xattr");
1100 	if (IS_ENABLED(CONFIG_EROFS_FS_POSIX_ACL))
1101 		seq_puts(seq, test_opt(opt, POSIX_ACL) ? ",acl" : ",noacl");
1102 	if (IS_ENABLED(CONFIG_EROFS_FS_ZIP))
1103 		seq_printf(seq, ",cache_strategy=%s",
1104 			  erofs_param_cache_strategy[opt->cache_strategy].name);
1105 	if (test_opt(opt, DAX_ALWAYS))
1106 		seq_puts(seq, ",dax=always");
1107 	if (test_opt(opt, DAX_NEVER))
1108 		seq_puts(seq, ",dax=never");
1109 	if (erofs_is_fileio_mode(sbi) && test_opt(opt, DIRECT_IO))
1110 		seq_puts(seq, ",directio");
1111 #ifdef CONFIG_EROFS_FS_ONDEMAND
1112 	if (sbi->fsid)
1113 		seq_printf(seq, ",fsid=%s", sbi->fsid);
1114 	if (sbi->domain_id)
1115 		seq_printf(seq, ",domain_id=%s", sbi->domain_id);
1116 #endif
1117 	if (sbi->dif0.fsoff)
1118 		seq_printf(seq, ",fsoffset=%llu", sbi->dif0.fsoff);
1119 	if (test_opt(opt, INODE_SHARE))
1120 		seq_puts(seq, ",inode_share");
1121 	return 0;
1122 }
1123 
1124 static void erofs_evict_inode(struct inode *inode)
1125 {
1126 	if (IS_DAX(inode))
1127 		dax_break_layout_final(inode);
1128 	erofs_ishare_free_inode(inode);
1129 	truncate_inode_pages_final(&inode->i_data);
1130 	clear_inode(inode);
1131 }
1132 
1133 const struct super_operations erofs_sops = {
1134 	.put_super = erofs_put_super,
1135 	.alloc_inode = erofs_alloc_inode,
1136 	.free_inode = erofs_free_inode,
1137 	.evict_inode = erofs_evict_inode,
1138 	.statfs = erofs_statfs,
1139 	.show_options = erofs_show_options,
1140 };
1141 
1142 module_init(erofs_module_init);
1143 module_exit(erofs_module_exit);
1144 
1145 MODULE_DESCRIPTION("Enhanced ROM File System");
1146 MODULE_AUTHOR("Gao Xiang, Chao Yu, Miao Xie, CONSUMER BG, HUAWEI Inc.");
1147 MODULE_LICENSE("GPL");
1148