xref: /openbmc/linux/fs/erofs/super.c (revision d9e9d230)
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/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 <linux/fs_context.h>
14 #include <linux/fs_parser.h>
15 #include <linux/dax.h>
16 #include <linux/exportfs.h>
17 #include "xattr.h"
18 
19 #define CREATE_TRACE_POINTS
20 #include <trace/events/erofs.h>
21 
22 static struct kmem_cache *erofs_inode_cachep __read_mostly;
23 
24 void _erofs_err(struct super_block *sb, const char *function,
25 		const char *fmt, ...)
26 {
27 	struct va_format vaf;
28 	va_list args;
29 
30 	va_start(args, fmt);
31 
32 	vaf.fmt = fmt;
33 	vaf.va = &args;
34 
35 	pr_err("(device %s): %s: %pV", sb->s_id, function, &vaf);
36 	va_end(args);
37 }
38 
39 void _erofs_info(struct super_block *sb, const char *function,
40 		 const char *fmt, ...)
41 {
42 	struct va_format vaf;
43 	va_list args;
44 
45 	va_start(args, fmt);
46 
47 	vaf.fmt = fmt;
48 	vaf.va = &args;
49 
50 	pr_info("(device %s): %pV", sb->s_id, &vaf);
51 	va_end(args);
52 }
53 
54 static int erofs_superblock_csum_verify(struct super_block *sb, void *sbdata)
55 {
56 	struct erofs_super_block *dsb;
57 	u32 expected_crc, crc;
58 
59 	dsb = kmemdup(sbdata + EROFS_SUPER_OFFSET,
60 		      EROFS_BLKSIZ - EROFS_SUPER_OFFSET, GFP_KERNEL);
61 	if (!dsb)
62 		return -ENOMEM;
63 
64 	expected_crc = le32_to_cpu(dsb->checksum);
65 	dsb->checksum = 0;
66 	/* to allow for x86 boot sectors and other oddities. */
67 	crc = crc32c(~0, dsb, EROFS_BLKSIZ - EROFS_SUPER_OFFSET);
68 	kfree(dsb);
69 
70 	if (crc != expected_crc) {
71 		erofs_err(sb, "invalid checksum 0x%08x, 0x%08x expected",
72 			  crc, expected_crc);
73 		return -EBADMSG;
74 	}
75 	return 0;
76 }
77 
78 static void erofs_inode_init_once(void *ptr)
79 {
80 	struct erofs_inode *vi = ptr;
81 
82 	inode_init_once(&vi->vfs_inode);
83 }
84 
85 static struct inode *erofs_alloc_inode(struct super_block *sb)
86 {
87 	struct erofs_inode *vi =
88 		alloc_inode_sb(sb, erofs_inode_cachep, GFP_KERNEL);
89 
90 	if (!vi)
91 		return NULL;
92 
93 	/* zero out everything except vfs_inode */
94 	memset(vi, 0, offsetof(struct erofs_inode, vfs_inode));
95 	return &vi->vfs_inode;
96 }
97 
98 static void erofs_free_inode(struct inode *inode)
99 {
100 	struct erofs_inode *vi = EROFS_I(inode);
101 
102 	/* be careful of RCU symlink path */
103 	if (inode->i_op == &erofs_fast_symlink_iops)
104 		kfree(inode->i_link);
105 	kfree(vi->xattr_shared_xattrs);
106 
107 	kmem_cache_free(erofs_inode_cachep, vi);
108 }
109 
110 static bool check_layout_compatibility(struct super_block *sb,
111 				       struct erofs_super_block *dsb)
112 {
113 	const unsigned int feature = le32_to_cpu(dsb->feature_incompat);
114 
115 	EROFS_SB(sb)->feature_incompat = feature;
116 
117 	/* check if current kernel meets all mandatory requirements */
118 	if (feature & (~EROFS_ALL_FEATURE_INCOMPAT)) {
119 		erofs_err(sb,
120 			  "unidentified incompatible feature %x, please upgrade kernel version",
121 			   feature & ~EROFS_ALL_FEATURE_INCOMPAT);
122 		return false;
123 	}
124 	return true;
125 }
126 
127 #ifdef CONFIG_EROFS_FS_ZIP
128 /* read variable-sized metadata, offset will be aligned by 4-byte */
129 static void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
130 				 erofs_off_t *offset, int *lengthp)
131 {
132 	u8 *buffer, *ptr;
133 	int len, i, cnt;
134 
135 	*offset = round_up(*offset, 4);
136 	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset), EROFS_KMAP);
137 	if (IS_ERR(ptr))
138 		return ptr;
139 
140 	len = le16_to_cpu(*(__le16 *)&ptr[erofs_blkoff(*offset)]);
141 	if (!len)
142 		len = U16_MAX + 1;
143 	buffer = kmalloc(len, GFP_KERNEL);
144 	if (!buffer)
145 		return ERR_PTR(-ENOMEM);
146 	*offset += sizeof(__le16);
147 	*lengthp = len;
148 
149 	for (i = 0; i < len; i += cnt) {
150 		cnt = min(EROFS_BLKSIZ - (int)erofs_blkoff(*offset), len - i);
151 		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
152 					 EROFS_KMAP);
153 		if (IS_ERR(ptr)) {
154 			kfree(buffer);
155 			return ptr;
156 		}
157 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
158 		*offset += cnt;
159 	}
160 	return buffer;
161 }
162 
163 static int erofs_load_compr_cfgs(struct super_block *sb,
164 				 struct erofs_super_block *dsb)
165 {
166 	struct erofs_sb_info *sbi = EROFS_SB(sb);
167 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
168 	unsigned int algs, alg;
169 	erofs_off_t offset;
170 	int size, ret = 0;
171 
172 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
173 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
174 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
175 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
176 		return -EINVAL;
177 	}
178 
179 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
180 	alg = 0;
181 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
182 		void *data;
183 
184 		if (!(algs & 1))
185 			continue;
186 
187 		data = erofs_read_metadata(sb, &buf, &offset, &size);
188 		if (IS_ERR(data)) {
189 			ret = PTR_ERR(data);
190 			break;
191 		}
192 
193 		switch (alg) {
194 		case Z_EROFS_COMPRESSION_LZ4:
195 			ret = z_erofs_load_lz4_config(sb, dsb, data, size);
196 			break;
197 		case Z_EROFS_COMPRESSION_LZMA:
198 			ret = z_erofs_load_lzma_config(sb, dsb, data, size);
199 			break;
200 		default:
201 			DBG_BUGON(1);
202 			ret = -EFAULT;
203 		}
204 		kfree(data);
205 		if (ret)
206 			break;
207 	}
208 	erofs_put_metabuf(&buf);
209 	return ret;
210 }
211 #else
212 static int erofs_load_compr_cfgs(struct super_block *sb,
213 				 struct erofs_super_block *dsb)
214 {
215 	if (dsb->u1.available_compr_algs) {
216 		erofs_err(sb, "try to load compressed fs when compression is disabled");
217 		return -EINVAL;
218 	}
219 	return 0;
220 }
221 #endif
222 
223 static int erofs_init_device(struct erofs_buf *buf, struct super_block *sb,
224 			     struct erofs_device_info *dif, erofs_off_t *pos)
225 {
226 	struct erofs_sb_info *sbi = EROFS_SB(sb);
227 	struct erofs_deviceslot *dis;
228 	struct block_device *bdev;
229 	void *ptr;
230 	int ret;
231 
232 	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*pos), EROFS_KMAP);
233 	if (IS_ERR(ptr))
234 		return PTR_ERR(ptr);
235 	dis = ptr + erofs_blkoff(*pos);
236 
237 	if (!dif->path) {
238 		if (!dis->tag[0]) {
239 			erofs_err(sb, "empty device tag @ pos %llu", *pos);
240 			return -EINVAL;
241 		}
242 		dif->path = kmemdup_nul(dis->tag, sizeof(dis->tag), GFP_KERNEL);
243 		if (!dif->path)
244 			return -ENOMEM;
245 	}
246 
247 	if (erofs_is_fscache_mode(sb)) {
248 		ret = erofs_fscache_register_cookie(sb, &dif->fscache,
249 				dif->path, false);
250 		if (ret)
251 			return ret;
252 	} else {
253 		bdev = blkdev_get_by_path(dif->path, FMODE_READ | FMODE_EXCL,
254 					  sb->s_type);
255 		if (IS_ERR(bdev))
256 			return PTR_ERR(bdev);
257 		dif->bdev = bdev;
258 		dif->dax_dev = fs_dax_get_by_bdev(bdev, &dif->dax_part_off);
259 	}
260 
261 	dif->blocks = le32_to_cpu(dis->blocks);
262 	dif->mapped_blkaddr = le32_to_cpu(dis->mapped_blkaddr);
263 	sbi->total_blocks += dif->blocks;
264 	*pos += EROFS_DEVT_SLOT_SIZE;
265 	return 0;
266 }
267 
268 static int erofs_scan_devices(struct super_block *sb,
269 			      struct erofs_super_block *dsb)
270 {
271 	struct erofs_sb_info *sbi = EROFS_SB(sb);
272 	unsigned int ondisk_extradevs;
273 	erofs_off_t pos;
274 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
275 	struct erofs_device_info *dif;
276 	int id, err = 0;
277 
278 	sbi->total_blocks = sbi->primarydevice_blocks;
279 	if (!erofs_sb_has_device_table(sbi))
280 		ondisk_extradevs = 0;
281 	else
282 		ondisk_extradevs = le16_to_cpu(dsb->extra_devices);
283 
284 	if (sbi->devs->extra_devices &&
285 	    ondisk_extradevs != sbi->devs->extra_devices) {
286 		erofs_err(sb, "extra devices don't match (ondisk %u, given %u)",
287 			  ondisk_extradevs, sbi->devs->extra_devices);
288 		return -EINVAL;
289 	}
290 	if (!ondisk_extradevs)
291 		return 0;
292 
293 	sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
294 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
295 	down_read(&sbi->devs->rwsem);
296 	if (sbi->devs->extra_devices) {
297 		idr_for_each_entry(&sbi->devs->tree, dif, id) {
298 			err = erofs_init_device(&buf, sb, dif, &pos);
299 			if (err)
300 				break;
301 		}
302 	} else {
303 		for (id = 0; id < ondisk_extradevs; id++) {
304 			dif = kzalloc(sizeof(*dif), GFP_KERNEL);
305 			if (!dif) {
306 				err = -ENOMEM;
307 				break;
308 			}
309 
310 			err = idr_alloc(&sbi->devs->tree, dif, 0, 0, GFP_KERNEL);
311 			if (err < 0) {
312 				kfree(dif);
313 				break;
314 			}
315 			++sbi->devs->extra_devices;
316 
317 			err = erofs_init_device(&buf, sb, dif, &pos);
318 			if (err)
319 				break;
320 		}
321 	}
322 	up_read(&sbi->devs->rwsem);
323 	erofs_put_metabuf(&buf);
324 	return err;
325 }
326 
327 static int erofs_read_superblock(struct super_block *sb)
328 {
329 	struct erofs_sb_info *sbi;
330 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
331 	struct erofs_super_block *dsb;
332 	unsigned int blkszbits;
333 	void *data;
334 	int ret;
335 
336 	data = erofs_read_metabuf(&buf, sb, 0, EROFS_KMAP);
337 	if (IS_ERR(data)) {
338 		erofs_err(sb, "cannot read erofs superblock");
339 		return PTR_ERR(data);
340 	}
341 
342 	sbi = EROFS_SB(sb);
343 	dsb = (struct erofs_super_block *)(data + EROFS_SUPER_OFFSET);
344 
345 	ret = -EINVAL;
346 	if (le32_to_cpu(dsb->magic) != EROFS_SUPER_MAGIC_V1) {
347 		erofs_err(sb, "cannot find valid erofs superblock");
348 		goto out;
349 	}
350 
351 	sbi->feature_compat = le32_to_cpu(dsb->feature_compat);
352 	if (erofs_sb_has_sb_chksum(sbi)) {
353 		ret = erofs_superblock_csum_verify(sb, data);
354 		if (ret)
355 			goto out;
356 	}
357 
358 	ret = -EINVAL;
359 	blkszbits = dsb->blkszbits;
360 	/* 9(512 bytes) + LOG_SECTORS_PER_BLOCK == LOG_BLOCK_SIZE */
361 	if (blkszbits != LOG_BLOCK_SIZE) {
362 		erofs_err(sb, "blkszbits %u isn't supported on this platform",
363 			  blkszbits);
364 		goto out;
365 	}
366 
367 	if (!check_layout_compatibility(sb, dsb))
368 		goto out;
369 
370 	sbi->sb_size = 128 + dsb->sb_extslots * EROFS_SB_EXTSLOT_SIZE;
371 	if (sbi->sb_size > EROFS_BLKSIZ) {
372 		erofs_err(sb, "invalid sb_extslots %u (more than a fs block)",
373 			  sbi->sb_size);
374 		goto out;
375 	}
376 	sbi->primarydevice_blocks = le32_to_cpu(dsb->blocks);
377 	sbi->meta_blkaddr = le32_to_cpu(dsb->meta_blkaddr);
378 #ifdef CONFIG_EROFS_FS_XATTR
379 	sbi->xattr_blkaddr = le32_to_cpu(dsb->xattr_blkaddr);
380 #endif
381 	sbi->islotbits = ilog2(sizeof(struct erofs_inode_compact));
382 	sbi->root_nid = le16_to_cpu(dsb->root_nid);
383 	sbi->inos = le64_to_cpu(dsb->inos);
384 
385 	sbi->build_time = le64_to_cpu(dsb->build_time);
386 	sbi->build_time_nsec = le32_to_cpu(dsb->build_time_nsec);
387 
388 	memcpy(&sb->s_uuid, dsb->uuid, sizeof(dsb->uuid));
389 
390 	ret = strscpy(sbi->volume_name, dsb->volume_name,
391 		      sizeof(dsb->volume_name));
392 	if (ret < 0) {	/* -E2BIG */
393 		erofs_err(sb, "bad volume name without NIL terminator");
394 		ret = -EFSCORRUPTED;
395 		goto out;
396 	}
397 
398 	/* parse on-disk compression configurations */
399 	if (erofs_sb_has_compr_cfgs(sbi))
400 		ret = erofs_load_compr_cfgs(sb, dsb);
401 	else
402 		ret = z_erofs_load_lz4_config(sb, dsb, NULL, 0);
403 	if (ret < 0)
404 		goto out;
405 
406 	/* handle multiple devices */
407 	ret = erofs_scan_devices(sb, dsb);
408 
409 	if (erofs_sb_has_ztailpacking(sbi))
410 		erofs_info(sb, "EXPERIMENTAL compressed inline data feature in use. Use at your own risk!");
411 	if (erofs_is_fscache_mode(sb))
412 		erofs_info(sb, "EXPERIMENTAL fscache-based on-demand read feature in use. Use at your own risk!");
413 out:
414 	erofs_put_metabuf(&buf);
415 	return ret;
416 }
417 
418 /* set up default EROFS parameters */
419 static void erofs_default_options(struct erofs_fs_context *ctx)
420 {
421 #ifdef CONFIG_EROFS_FS_ZIP
422 	ctx->opt.cache_strategy = EROFS_ZIP_CACHE_READAROUND;
423 	ctx->opt.max_sync_decompress_pages = 3;
424 	ctx->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_AUTO;
425 #endif
426 #ifdef CONFIG_EROFS_FS_XATTR
427 	set_opt(&ctx->opt, XATTR_USER);
428 #endif
429 #ifdef CONFIG_EROFS_FS_POSIX_ACL
430 	set_opt(&ctx->opt, POSIX_ACL);
431 #endif
432 }
433 
434 enum {
435 	Opt_user_xattr,
436 	Opt_acl,
437 	Opt_cache_strategy,
438 	Opt_dax,
439 	Opt_dax_enum,
440 	Opt_device,
441 	Opt_fsid,
442 	Opt_err
443 };
444 
445 static const struct constant_table erofs_param_cache_strategy[] = {
446 	{"disabled",	EROFS_ZIP_CACHE_DISABLED},
447 	{"readahead",	EROFS_ZIP_CACHE_READAHEAD},
448 	{"readaround",	EROFS_ZIP_CACHE_READAROUND},
449 	{}
450 };
451 
452 static const struct constant_table erofs_dax_param_enums[] = {
453 	{"always",	EROFS_MOUNT_DAX_ALWAYS},
454 	{"never",	EROFS_MOUNT_DAX_NEVER},
455 	{}
456 };
457 
458 static const struct fs_parameter_spec erofs_fs_parameters[] = {
459 	fsparam_flag_no("user_xattr",	Opt_user_xattr),
460 	fsparam_flag_no("acl",		Opt_acl),
461 	fsparam_enum("cache_strategy",	Opt_cache_strategy,
462 		     erofs_param_cache_strategy),
463 	fsparam_flag("dax",             Opt_dax),
464 	fsparam_enum("dax",		Opt_dax_enum, erofs_dax_param_enums),
465 	fsparam_string("device",	Opt_device),
466 	fsparam_string("fsid",		Opt_fsid),
467 	{}
468 };
469 
470 static bool erofs_fc_set_dax_mode(struct fs_context *fc, unsigned int mode)
471 {
472 #ifdef CONFIG_FS_DAX
473 	struct erofs_fs_context *ctx = fc->fs_private;
474 
475 	switch (mode) {
476 	case EROFS_MOUNT_DAX_ALWAYS:
477 		warnfc(fc, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
478 		set_opt(&ctx->opt, DAX_ALWAYS);
479 		clear_opt(&ctx->opt, DAX_NEVER);
480 		return true;
481 	case EROFS_MOUNT_DAX_NEVER:
482 		set_opt(&ctx->opt, DAX_NEVER);
483 		clear_opt(&ctx->opt, DAX_ALWAYS);
484 		return true;
485 	default:
486 		DBG_BUGON(1);
487 		return false;
488 	}
489 #else
490 	errorfc(fc, "dax options not supported");
491 	return false;
492 #endif
493 }
494 
495 static int erofs_fc_parse_param(struct fs_context *fc,
496 				struct fs_parameter *param)
497 {
498 	struct erofs_fs_context *ctx = fc->fs_private;
499 	struct fs_parse_result result;
500 	struct erofs_device_info *dif;
501 	int opt, ret;
502 
503 	opt = fs_parse(fc, erofs_fs_parameters, param, &result);
504 	if (opt < 0)
505 		return opt;
506 
507 	switch (opt) {
508 	case Opt_user_xattr:
509 #ifdef CONFIG_EROFS_FS_XATTR
510 		if (result.boolean)
511 			set_opt(&ctx->opt, XATTR_USER);
512 		else
513 			clear_opt(&ctx->opt, XATTR_USER);
514 #else
515 		errorfc(fc, "{,no}user_xattr options not supported");
516 #endif
517 		break;
518 	case Opt_acl:
519 #ifdef CONFIG_EROFS_FS_POSIX_ACL
520 		if (result.boolean)
521 			set_opt(&ctx->opt, POSIX_ACL);
522 		else
523 			clear_opt(&ctx->opt, POSIX_ACL);
524 #else
525 		errorfc(fc, "{,no}acl options not supported");
526 #endif
527 		break;
528 	case Opt_cache_strategy:
529 #ifdef CONFIG_EROFS_FS_ZIP
530 		ctx->opt.cache_strategy = result.uint_32;
531 #else
532 		errorfc(fc, "compression not supported, cache_strategy ignored");
533 #endif
534 		break;
535 	case Opt_dax:
536 		if (!erofs_fc_set_dax_mode(fc, EROFS_MOUNT_DAX_ALWAYS))
537 			return -EINVAL;
538 		break;
539 	case Opt_dax_enum:
540 		if (!erofs_fc_set_dax_mode(fc, result.uint_32))
541 			return -EINVAL;
542 		break;
543 	case Opt_device:
544 		dif = kzalloc(sizeof(*dif), GFP_KERNEL);
545 		if (!dif)
546 			return -ENOMEM;
547 		dif->path = kstrdup(param->string, GFP_KERNEL);
548 		if (!dif->path) {
549 			kfree(dif);
550 			return -ENOMEM;
551 		}
552 		down_write(&ctx->devs->rwsem);
553 		ret = idr_alloc(&ctx->devs->tree, dif, 0, 0, GFP_KERNEL);
554 		up_write(&ctx->devs->rwsem);
555 		if (ret < 0) {
556 			kfree(dif->path);
557 			kfree(dif);
558 			return ret;
559 		}
560 		++ctx->devs->extra_devices;
561 		break;
562 	case Opt_fsid:
563 #ifdef CONFIG_EROFS_FS_ONDEMAND
564 		kfree(ctx->opt.fsid);
565 		ctx->opt.fsid = kstrdup(param->string, GFP_KERNEL);
566 		if (!ctx->opt.fsid)
567 			return -ENOMEM;
568 #else
569 		errorfc(fc, "fsid option not supported");
570 #endif
571 		break;
572 	default:
573 		return -ENOPARAM;
574 	}
575 	return 0;
576 }
577 
578 #ifdef CONFIG_EROFS_FS_ZIP
579 static const struct address_space_operations managed_cache_aops;
580 
581 static bool erofs_managed_cache_release_folio(struct folio *folio, gfp_t gfp)
582 {
583 	bool ret = true;
584 	struct address_space *const mapping = folio->mapping;
585 
586 	DBG_BUGON(!folio_test_locked(folio));
587 	DBG_BUGON(mapping->a_ops != &managed_cache_aops);
588 
589 	if (folio_test_private(folio))
590 		ret = erofs_try_to_free_cached_page(&folio->page);
591 
592 	return ret;
593 }
594 
595 /*
596  * It will be called only on inode eviction. In case that there are still some
597  * decompression requests in progress, wait with rescheduling for a bit here.
598  * We could introduce an extra locking instead but it seems unnecessary.
599  */
600 static void erofs_managed_cache_invalidate_folio(struct folio *folio,
601 					       size_t offset, size_t length)
602 {
603 	const size_t stop = length + offset;
604 
605 	DBG_BUGON(!folio_test_locked(folio));
606 
607 	/* Check for potential overflow in debug mode */
608 	DBG_BUGON(stop > folio_size(folio) || stop < length);
609 
610 	if (offset == 0 && stop == folio_size(folio))
611 		while (!erofs_managed_cache_release_folio(folio, GFP_NOFS))
612 			cond_resched();
613 }
614 
615 static const struct address_space_operations managed_cache_aops = {
616 	.release_folio = erofs_managed_cache_release_folio,
617 	.invalidate_folio = erofs_managed_cache_invalidate_folio,
618 };
619 
620 static int erofs_init_managed_cache(struct super_block *sb)
621 {
622 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
623 	struct inode *const inode = new_inode(sb);
624 
625 	if (!inode)
626 		return -ENOMEM;
627 
628 	set_nlink(inode, 1);
629 	inode->i_size = OFFSET_MAX;
630 
631 	inode->i_mapping->a_ops = &managed_cache_aops;
632 	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
633 	sbi->managed_cache = inode;
634 	return 0;
635 }
636 #else
637 static int erofs_init_managed_cache(struct super_block *sb) { return 0; }
638 #endif
639 
640 static struct inode *erofs_nfs_get_inode(struct super_block *sb,
641 					 u64 ino, u32 generation)
642 {
643 	return erofs_iget(sb, ino, false);
644 }
645 
646 static struct dentry *erofs_fh_to_dentry(struct super_block *sb,
647 		struct fid *fid, int fh_len, int fh_type)
648 {
649 	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
650 				    erofs_nfs_get_inode);
651 }
652 
653 static struct dentry *erofs_fh_to_parent(struct super_block *sb,
654 		struct fid *fid, int fh_len, int fh_type)
655 {
656 	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
657 				    erofs_nfs_get_inode);
658 }
659 
660 static struct dentry *erofs_get_parent(struct dentry *child)
661 {
662 	erofs_nid_t nid;
663 	unsigned int d_type;
664 	int err;
665 
666 	err = erofs_namei(d_inode(child), &dotdot_name, &nid, &d_type);
667 	if (err)
668 		return ERR_PTR(err);
669 	return d_obtain_alias(erofs_iget(child->d_sb, nid, d_type == FT_DIR));
670 }
671 
672 static const struct export_operations erofs_export_ops = {
673 	.fh_to_dentry = erofs_fh_to_dentry,
674 	.fh_to_parent = erofs_fh_to_parent,
675 	.get_parent = erofs_get_parent,
676 };
677 
678 static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
679 {
680 	struct inode *inode;
681 	struct erofs_sb_info *sbi;
682 	struct erofs_fs_context *ctx = fc->fs_private;
683 	int err;
684 
685 	sb->s_magic = EROFS_SUPER_MAGIC;
686 	sb->s_flags |= SB_RDONLY | SB_NOATIME;
687 	sb->s_maxbytes = MAX_LFS_FILESIZE;
688 	sb->s_op = &erofs_sops;
689 
690 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
691 	if (!sbi)
692 		return -ENOMEM;
693 
694 	sb->s_fs_info = sbi;
695 	sbi->opt = ctx->opt;
696 	ctx->opt.fsid = NULL;
697 	sbi->devs = ctx->devs;
698 	ctx->devs = NULL;
699 
700 	if (erofs_is_fscache_mode(sb)) {
701 		sb->s_blocksize = EROFS_BLKSIZ;
702 		sb->s_blocksize_bits = LOG_BLOCK_SIZE;
703 
704 		err = erofs_fscache_register_fs(sb);
705 		if (err)
706 			return err;
707 
708 		err = erofs_fscache_register_cookie(sb, &sbi->s_fscache,
709 						    sbi->opt.fsid, true);
710 		if (err)
711 			return err;
712 
713 		err = super_setup_bdi(sb);
714 		if (err)
715 			return err;
716 	} else {
717 		if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
718 			erofs_err(sb, "failed to set erofs blksize");
719 			return -EINVAL;
720 		}
721 
722 		sbi->dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
723 						  &sbi->dax_part_off);
724 	}
725 
726 	err = erofs_read_superblock(sb);
727 	if (err)
728 		return err;
729 
730 	if (test_opt(&sbi->opt, DAX_ALWAYS)) {
731 		BUILD_BUG_ON(EROFS_BLKSIZ != PAGE_SIZE);
732 
733 		if (!sbi->dax_dev) {
734 			errorfc(fc, "DAX unsupported by block device. Turning off DAX.");
735 			clear_opt(&sbi->opt, DAX_ALWAYS);
736 		}
737 	}
738 
739 	sb->s_time_gran = 1;
740 	sb->s_xattr = erofs_xattr_handlers;
741 	sb->s_export_op = &erofs_export_ops;
742 
743 	if (test_opt(&sbi->opt, POSIX_ACL))
744 		sb->s_flags |= SB_POSIXACL;
745 	else
746 		sb->s_flags &= ~SB_POSIXACL;
747 
748 #ifdef CONFIG_EROFS_FS_ZIP
749 	xa_init(&sbi->managed_pslots);
750 #endif
751 
752 	/* get the root inode */
753 	inode = erofs_iget(sb, ROOT_NID(sbi), true);
754 	if (IS_ERR(inode))
755 		return PTR_ERR(inode);
756 
757 	if (!S_ISDIR(inode->i_mode)) {
758 		erofs_err(sb, "rootino(nid %llu) is not a directory(i_mode %o)",
759 			  ROOT_NID(sbi), inode->i_mode);
760 		iput(inode);
761 		return -EINVAL;
762 	}
763 
764 	sb->s_root = d_make_root(inode);
765 	if (!sb->s_root)
766 		return -ENOMEM;
767 
768 	erofs_shrinker_register(sb);
769 	/* sb->s_umount is already locked, SB_ACTIVE and SB_BORN are not set */
770 	err = erofs_init_managed_cache(sb);
771 	if (err)
772 		return err;
773 
774 	err = erofs_register_sysfs(sb);
775 	if (err)
776 		return err;
777 
778 	erofs_info(sb, "mounted with root inode @ nid %llu.", ROOT_NID(sbi));
779 	return 0;
780 }
781 
782 static int erofs_fc_get_tree(struct fs_context *fc)
783 {
784 	struct erofs_fs_context *ctx = fc->fs_private;
785 
786 	if (IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && ctx->opt.fsid)
787 		return get_tree_nodev(fc, erofs_fc_fill_super);
788 
789 	return get_tree_bdev(fc, erofs_fc_fill_super);
790 }
791 
792 static int erofs_fc_reconfigure(struct fs_context *fc)
793 {
794 	struct super_block *sb = fc->root->d_sb;
795 	struct erofs_sb_info *sbi = EROFS_SB(sb);
796 	struct erofs_fs_context *ctx = fc->fs_private;
797 
798 	DBG_BUGON(!sb_rdonly(sb));
799 
800 	if (test_opt(&ctx->opt, POSIX_ACL))
801 		fc->sb_flags |= SB_POSIXACL;
802 	else
803 		fc->sb_flags &= ~SB_POSIXACL;
804 
805 	sbi->opt = ctx->opt;
806 
807 	fc->sb_flags |= SB_RDONLY;
808 	return 0;
809 }
810 
811 static int erofs_release_device_info(int id, void *ptr, void *data)
812 {
813 	struct erofs_device_info *dif = ptr;
814 
815 	fs_put_dax(dif->dax_dev);
816 	if (dif->bdev)
817 		blkdev_put(dif->bdev, FMODE_READ | FMODE_EXCL);
818 	erofs_fscache_unregister_cookie(&dif->fscache);
819 	kfree(dif->path);
820 	kfree(dif);
821 	return 0;
822 }
823 
824 static void erofs_free_dev_context(struct erofs_dev_context *devs)
825 {
826 	if (!devs)
827 		return;
828 	idr_for_each(&devs->tree, &erofs_release_device_info, NULL);
829 	idr_destroy(&devs->tree);
830 	kfree(devs);
831 }
832 
833 static void erofs_fc_free(struct fs_context *fc)
834 {
835 	struct erofs_fs_context *ctx = fc->fs_private;
836 
837 	erofs_free_dev_context(ctx->devs);
838 	kfree(ctx->opt.fsid);
839 	kfree(ctx);
840 }
841 
842 static const struct fs_context_operations erofs_context_ops = {
843 	.parse_param	= erofs_fc_parse_param,
844 	.get_tree       = erofs_fc_get_tree,
845 	.reconfigure    = erofs_fc_reconfigure,
846 	.free		= erofs_fc_free,
847 };
848 
849 static int erofs_init_fs_context(struct fs_context *fc)
850 {
851 	struct erofs_fs_context *ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
852 
853 	if (!ctx)
854 		return -ENOMEM;
855 	ctx->devs = kzalloc(sizeof(struct erofs_dev_context), GFP_KERNEL);
856 	if (!ctx->devs) {
857 		kfree(ctx);
858 		return -ENOMEM;
859 	}
860 	fc->fs_private = ctx;
861 
862 	idr_init(&ctx->devs->tree);
863 	init_rwsem(&ctx->devs->rwsem);
864 	erofs_default_options(ctx);
865 	fc->ops = &erofs_context_ops;
866 	return 0;
867 }
868 
869 /*
870  * could be triggered after deactivate_locked_super()
871  * is called, thus including umount and failed to initialize.
872  */
873 static void erofs_kill_sb(struct super_block *sb)
874 {
875 	struct erofs_sb_info *sbi;
876 
877 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
878 
879 	if (erofs_is_fscache_mode(sb))
880 		generic_shutdown_super(sb);
881 	else
882 		kill_block_super(sb);
883 
884 	sbi = EROFS_SB(sb);
885 	if (!sbi)
886 		return;
887 
888 	erofs_free_dev_context(sbi->devs);
889 	fs_put_dax(sbi->dax_dev);
890 	erofs_fscache_unregister_cookie(&sbi->s_fscache);
891 	erofs_fscache_unregister_fs(sb);
892 	kfree(sbi->opt.fsid);
893 	kfree(sbi);
894 	sb->s_fs_info = NULL;
895 }
896 
897 /* called when ->s_root is non-NULL */
898 static void erofs_put_super(struct super_block *sb)
899 {
900 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
901 
902 	DBG_BUGON(!sbi);
903 
904 	erofs_unregister_sysfs(sb);
905 	erofs_shrinker_unregister(sb);
906 #ifdef CONFIG_EROFS_FS_ZIP
907 	iput(sbi->managed_cache);
908 	sbi->managed_cache = NULL;
909 #endif
910 	erofs_fscache_unregister_cookie(&sbi->s_fscache);
911 }
912 
913 static struct file_system_type erofs_fs_type = {
914 	.owner          = THIS_MODULE,
915 	.name           = "erofs",
916 	.init_fs_context = erofs_init_fs_context,
917 	.kill_sb        = erofs_kill_sb,
918 	.fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
919 };
920 MODULE_ALIAS_FS("erofs");
921 
922 static int __init erofs_module_init(void)
923 {
924 	int err;
925 
926 	erofs_check_ondisk_layout_definitions();
927 
928 	erofs_inode_cachep = kmem_cache_create("erofs_inode",
929 					       sizeof(struct erofs_inode), 0,
930 					       SLAB_RECLAIM_ACCOUNT,
931 					       erofs_inode_init_once);
932 	if (!erofs_inode_cachep) {
933 		err = -ENOMEM;
934 		goto icache_err;
935 	}
936 
937 	err = erofs_init_shrinker();
938 	if (err)
939 		goto shrinker_err;
940 
941 	err = z_erofs_lzma_init();
942 	if (err)
943 		goto lzma_err;
944 
945 	erofs_pcpubuf_init();
946 	err = z_erofs_init_zip_subsystem();
947 	if (err)
948 		goto zip_err;
949 
950 	err = erofs_init_sysfs();
951 	if (err)
952 		goto sysfs_err;
953 
954 	err = register_filesystem(&erofs_fs_type);
955 	if (err)
956 		goto fs_err;
957 
958 	return 0;
959 
960 fs_err:
961 	erofs_exit_sysfs();
962 sysfs_err:
963 	z_erofs_exit_zip_subsystem();
964 zip_err:
965 	z_erofs_lzma_exit();
966 lzma_err:
967 	erofs_exit_shrinker();
968 shrinker_err:
969 	kmem_cache_destroy(erofs_inode_cachep);
970 icache_err:
971 	return err;
972 }
973 
974 static void __exit erofs_module_exit(void)
975 {
976 	unregister_filesystem(&erofs_fs_type);
977 
978 	/* Ensure all RCU free inodes / pclusters are safe to be destroyed. */
979 	rcu_barrier();
980 
981 	erofs_exit_sysfs();
982 	z_erofs_exit_zip_subsystem();
983 	z_erofs_lzma_exit();
984 	erofs_exit_shrinker();
985 	kmem_cache_destroy(erofs_inode_cachep);
986 	erofs_pcpubuf_exit();
987 }
988 
989 /* get filesystem statistics */
990 static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
991 {
992 	struct super_block *sb = dentry->d_sb;
993 	struct erofs_sb_info *sbi = EROFS_SB(sb);
994 	u64 id = 0;
995 
996 	if (!erofs_is_fscache_mode(sb))
997 		id = huge_encode_dev(sb->s_bdev->bd_dev);
998 
999 	buf->f_type = sb->s_magic;
1000 	buf->f_bsize = EROFS_BLKSIZ;
1001 	buf->f_blocks = sbi->total_blocks;
1002 	buf->f_bfree = buf->f_bavail = 0;
1003 
1004 	buf->f_files = ULLONG_MAX;
1005 	buf->f_ffree = ULLONG_MAX - sbi->inos;
1006 
1007 	buf->f_namelen = EROFS_NAME_LEN;
1008 
1009 	buf->f_fsid    = u64_to_fsid(id);
1010 	return 0;
1011 }
1012 
1013 static int erofs_show_options(struct seq_file *seq, struct dentry *root)
1014 {
1015 	struct erofs_sb_info *sbi = EROFS_SB(root->d_sb);
1016 	struct erofs_mount_opts *opt = &sbi->opt;
1017 
1018 #ifdef CONFIG_EROFS_FS_XATTR
1019 	if (test_opt(opt, XATTR_USER))
1020 		seq_puts(seq, ",user_xattr");
1021 	else
1022 		seq_puts(seq, ",nouser_xattr");
1023 #endif
1024 #ifdef CONFIG_EROFS_FS_POSIX_ACL
1025 	if (test_opt(opt, POSIX_ACL))
1026 		seq_puts(seq, ",acl");
1027 	else
1028 		seq_puts(seq, ",noacl");
1029 #endif
1030 #ifdef CONFIG_EROFS_FS_ZIP
1031 	if (opt->cache_strategy == EROFS_ZIP_CACHE_DISABLED)
1032 		seq_puts(seq, ",cache_strategy=disabled");
1033 	else if (opt->cache_strategy == EROFS_ZIP_CACHE_READAHEAD)
1034 		seq_puts(seq, ",cache_strategy=readahead");
1035 	else if (opt->cache_strategy == EROFS_ZIP_CACHE_READAROUND)
1036 		seq_puts(seq, ",cache_strategy=readaround");
1037 #endif
1038 	if (test_opt(opt, DAX_ALWAYS))
1039 		seq_puts(seq, ",dax=always");
1040 	if (test_opt(opt, DAX_NEVER))
1041 		seq_puts(seq, ",dax=never");
1042 #ifdef CONFIG_EROFS_FS_ONDEMAND
1043 	if (opt->fsid)
1044 		seq_printf(seq, ",fsid=%s", opt->fsid);
1045 #endif
1046 	return 0;
1047 }
1048 
1049 const struct super_operations erofs_sops = {
1050 	.put_super = erofs_put_super,
1051 	.alloc_inode = erofs_alloc_inode,
1052 	.free_inode = erofs_free_inode,
1053 	.statfs = erofs_statfs,
1054 	.show_options = erofs_show_options,
1055 };
1056 
1057 module_init(erofs_module_init);
1058 module_exit(erofs_module_exit);
1059 
1060 MODULE_DESCRIPTION("Enhanced ROM File System");
1061 MODULE_AUTHOR("Gao Xiang, Chao Yu, Miao Xie, CONSUMER BG, HUAWEI Inc.");
1062 MODULE_LICENSE("GPL");
1063