xref: /linux/fs/erofs/super.c (revision 0da908c291070d89482f6211dbe81d4d43c3f7cb)
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_fscache *fscache;
228 	struct erofs_deviceslot *dis;
229 	struct block_device *bdev;
230 	void *ptr;
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 		fscache = erofs_fscache_register_cookie(sb, dif->path, 0);
249 		if (IS_ERR(fscache))
250 			return PTR_ERR(fscache);
251 		dif->fscache = fscache;
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 						  NULL, NULL);
260 	}
261 
262 	dif->blocks = le32_to_cpu(dis->blocks);
263 	dif->mapped_blkaddr = le32_to_cpu(dis->mapped_blkaddr);
264 	sbi->total_blocks += dif->blocks;
265 	*pos += EROFS_DEVT_SLOT_SIZE;
266 	return 0;
267 }
268 
269 static int erofs_scan_devices(struct super_block *sb,
270 			      struct erofs_super_block *dsb)
271 {
272 	struct erofs_sb_info *sbi = EROFS_SB(sb);
273 	unsigned int ondisk_extradevs;
274 	erofs_off_t pos;
275 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
276 	struct erofs_device_info *dif;
277 	int id, err = 0;
278 
279 	sbi->total_blocks = sbi->primarydevice_blocks;
280 	if (!erofs_sb_has_device_table(sbi))
281 		ondisk_extradevs = 0;
282 	else
283 		ondisk_extradevs = le16_to_cpu(dsb->extra_devices);
284 
285 	if (sbi->devs->extra_devices &&
286 	    ondisk_extradevs != sbi->devs->extra_devices) {
287 		erofs_err(sb, "extra devices don't match (ondisk %u, given %u)",
288 			  ondisk_extradevs, sbi->devs->extra_devices);
289 		return -EINVAL;
290 	}
291 	if (!ondisk_extradevs)
292 		return 0;
293 
294 	sbi->device_id_mask = roundup_pow_of_two(ondisk_extradevs + 1) - 1;
295 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
296 	down_read(&sbi->devs->rwsem);
297 	if (sbi->devs->extra_devices) {
298 		idr_for_each_entry(&sbi->devs->tree, dif, id) {
299 			err = erofs_init_device(&buf, sb, dif, &pos);
300 			if (err)
301 				break;
302 		}
303 	} else {
304 		for (id = 0; id < ondisk_extradevs; id++) {
305 			dif = kzalloc(sizeof(*dif), GFP_KERNEL);
306 			if (!dif) {
307 				err = -ENOMEM;
308 				break;
309 			}
310 
311 			err = idr_alloc(&sbi->devs->tree, dif, 0, 0, GFP_KERNEL);
312 			if (err < 0) {
313 				kfree(dif);
314 				break;
315 			}
316 			++sbi->devs->extra_devices;
317 
318 			err = erofs_init_device(&buf, sb, dif, &pos);
319 			if (err)
320 				break;
321 		}
322 	}
323 	up_read(&sbi->devs->rwsem);
324 	erofs_put_metabuf(&buf);
325 	return err;
326 }
327 
328 static int erofs_read_superblock(struct super_block *sb)
329 {
330 	struct erofs_sb_info *sbi;
331 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
332 	struct erofs_super_block *dsb;
333 	unsigned int blkszbits;
334 	void *data;
335 	int ret;
336 
337 	data = erofs_read_metabuf(&buf, sb, 0, EROFS_KMAP);
338 	if (IS_ERR(data)) {
339 		erofs_err(sb, "cannot read erofs superblock");
340 		return PTR_ERR(data);
341 	}
342 
343 	sbi = EROFS_SB(sb);
344 	dsb = (struct erofs_super_block *)(data + EROFS_SUPER_OFFSET);
345 
346 	ret = -EINVAL;
347 	if (le32_to_cpu(dsb->magic) != EROFS_SUPER_MAGIC_V1) {
348 		erofs_err(sb, "cannot find valid erofs superblock");
349 		goto out;
350 	}
351 
352 	sbi->feature_compat = le32_to_cpu(dsb->feature_compat);
353 	if (erofs_sb_has_sb_chksum(sbi)) {
354 		ret = erofs_superblock_csum_verify(sb, data);
355 		if (ret)
356 			goto out;
357 	}
358 
359 	ret = -EINVAL;
360 	blkszbits = dsb->blkszbits;
361 	/* 9(512 bytes) + LOG_SECTORS_PER_BLOCK == LOG_BLOCK_SIZE */
362 	if (blkszbits != LOG_BLOCK_SIZE) {
363 		erofs_err(sb, "blkszbits %u isn't supported on this platform",
364 			  blkszbits);
365 		goto out;
366 	}
367 
368 	if (!check_layout_compatibility(sb, dsb))
369 		goto out;
370 
371 	sbi->sb_size = 128 + dsb->sb_extslots * EROFS_SB_EXTSLOT_SIZE;
372 	if (sbi->sb_size > EROFS_BLKSIZ) {
373 		erofs_err(sb, "invalid sb_extslots %u (more than a fs block)",
374 			  sbi->sb_size);
375 		goto out;
376 	}
377 	sbi->primarydevice_blocks = le32_to_cpu(dsb->blocks);
378 	sbi->meta_blkaddr = le32_to_cpu(dsb->meta_blkaddr);
379 #ifdef CONFIG_EROFS_FS_XATTR
380 	sbi->xattr_blkaddr = le32_to_cpu(dsb->xattr_blkaddr);
381 #endif
382 	sbi->islotbits = ilog2(sizeof(struct erofs_inode_compact));
383 	sbi->root_nid = le16_to_cpu(dsb->root_nid);
384 #ifdef CONFIG_EROFS_FS_ZIP
385 	sbi->packed_inode = NULL;
386 	if (erofs_sb_has_fragments(sbi) && dsb->packed_nid) {
387 		sbi->packed_inode =
388 			erofs_iget(sb, le64_to_cpu(dsb->packed_nid));
389 		if (IS_ERR(sbi->packed_inode)) {
390 			ret = PTR_ERR(sbi->packed_inode);
391 			goto out;
392 		}
393 	}
394 #endif
395 	sbi->inos = le64_to_cpu(dsb->inos);
396 
397 	sbi->build_time = le64_to_cpu(dsb->build_time);
398 	sbi->build_time_nsec = le32_to_cpu(dsb->build_time_nsec);
399 
400 	memcpy(&sb->s_uuid, dsb->uuid, sizeof(dsb->uuid));
401 
402 	ret = strscpy(sbi->volume_name, dsb->volume_name,
403 		      sizeof(dsb->volume_name));
404 	if (ret < 0) {	/* -E2BIG */
405 		erofs_err(sb, "bad volume name without NIL terminator");
406 		ret = -EFSCORRUPTED;
407 		goto out;
408 	}
409 
410 	/* parse on-disk compression configurations */
411 	if (erofs_sb_has_compr_cfgs(sbi))
412 		ret = erofs_load_compr_cfgs(sb, dsb);
413 	else
414 		ret = z_erofs_load_lz4_config(sb, dsb, NULL, 0);
415 	if (ret < 0)
416 		goto out;
417 
418 	/* handle multiple devices */
419 	ret = erofs_scan_devices(sb, dsb);
420 
421 	if (erofs_sb_has_ztailpacking(sbi))
422 		erofs_info(sb, "EXPERIMENTAL compressed inline data feature in use. Use at your own risk!");
423 	if (erofs_is_fscache_mode(sb))
424 		erofs_info(sb, "EXPERIMENTAL fscache-based on-demand read feature in use. Use at your own risk!");
425 	if (erofs_sb_has_fragments(sbi))
426 		erofs_info(sb, "EXPERIMENTAL compressed fragments feature in use. Use at your own risk!");
427 	if (erofs_sb_has_dedupe(sbi))
428 		erofs_info(sb, "EXPERIMENTAL global deduplication feature in use. Use at your own risk!");
429 out:
430 	erofs_put_metabuf(&buf);
431 	return ret;
432 }
433 
434 /* set up default EROFS parameters */
435 static void erofs_default_options(struct erofs_fs_context *ctx)
436 {
437 #ifdef CONFIG_EROFS_FS_ZIP
438 	ctx->opt.cache_strategy = EROFS_ZIP_CACHE_READAROUND;
439 	ctx->opt.max_sync_decompress_pages = 3;
440 	ctx->opt.sync_decompress = EROFS_SYNC_DECOMPRESS_AUTO;
441 #endif
442 #ifdef CONFIG_EROFS_FS_XATTR
443 	set_opt(&ctx->opt, XATTR_USER);
444 #endif
445 #ifdef CONFIG_EROFS_FS_POSIX_ACL
446 	set_opt(&ctx->opt, POSIX_ACL);
447 #endif
448 }
449 
450 enum {
451 	Opt_user_xattr,
452 	Opt_acl,
453 	Opt_cache_strategy,
454 	Opt_dax,
455 	Opt_dax_enum,
456 	Opt_device,
457 	Opt_fsid,
458 	Opt_domain_id,
459 	Opt_err
460 };
461 
462 static const struct constant_table erofs_param_cache_strategy[] = {
463 	{"disabled",	EROFS_ZIP_CACHE_DISABLED},
464 	{"readahead",	EROFS_ZIP_CACHE_READAHEAD},
465 	{"readaround",	EROFS_ZIP_CACHE_READAROUND},
466 	{}
467 };
468 
469 static const struct constant_table erofs_dax_param_enums[] = {
470 	{"always",	EROFS_MOUNT_DAX_ALWAYS},
471 	{"never",	EROFS_MOUNT_DAX_NEVER},
472 	{}
473 };
474 
475 static const struct fs_parameter_spec erofs_fs_parameters[] = {
476 	fsparam_flag_no("user_xattr",	Opt_user_xattr),
477 	fsparam_flag_no("acl",		Opt_acl),
478 	fsparam_enum("cache_strategy",	Opt_cache_strategy,
479 		     erofs_param_cache_strategy),
480 	fsparam_flag("dax",             Opt_dax),
481 	fsparam_enum("dax",		Opt_dax_enum, erofs_dax_param_enums),
482 	fsparam_string("device",	Opt_device),
483 	fsparam_string("fsid",		Opt_fsid),
484 	fsparam_string("domain_id",	Opt_domain_id),
485 	{}
486 };
487 
488 static bool erofs_fc_set_dax_mode(struct fs_context *fc, unsigned int mode)
489 {
490 #ifdef CONFIG_FS_DAX
491 	struct erofs_fs_context *ctx = fc->fs_private;
492 
493 	switch (mode) {
494 	case EROFS_MOUNT_DAX_ALWAYS:
495 		warnfc(fc, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
496 		set_opt(&ctx->opt, DAX_ALWAYS);
497 		clear_opt(&ctx->opt, DAX_NEVER);
498 		return true;
499 	case EROFS_MOUNT_DAX_NEVER:
500 		set_opt(&ctx->opt, DAX_NEVER);
501 		clear_opt(&ctx->opt, DAX_ALWAYS);
502 		return true;
503 	default:
504 		DBG_BUGON(1);
505 		return false;
506 	}
507 #else
508 	errorfc(fc, "dax options not supported");
509 	return false;
510 #endif
511 }
512 
513 static int erofs_fc_parse_param(struct fs_context *fc,
514 				struct fs_parameter *param)
515 {
516 	struct erofs_fs_context *ctx = fc->fs_private;
517 	struct fs_parse_result result;
518 	struct erofs_device_info *dif;
519 	int opt, ret;
520 
521 	opt = fs_parse(fc, erofs_fs_parameters, param, &result);
522 	if (opt < 0)
523 		return opt;
524 
525 	switch (opt) {
526 	case Opt_user_xattr:
527 #ifdef CONFIG_EROFS_FS_XATTR
528 		if (result.boolean)
529 			set_opt(&ctx->opt, XATTR_USER);
530 		else
531 			clear_opt(&ctx->opt, XATTR_USER);
532 #else
533 		errorfc(fc, "{,no}user_xattr options not supported");
534 #endif
535 		break;
536 	case Opt_acl:
537 #ifdef CONFIG_EROFS_FS_POSIX_ACL
538 		if (result.boolean)
539 			set_opt(&ctx->opt, POSIX_ACL);
540 		else
541 			clear_opt(&ctx->opt, POSIX_ACL);
542 #else
543 		errorfc(fc, "{,no}acl options not supported");
544 #endif
545 		break;
546 	case Opt_cache_strategy:
547 #ifdef CONFIG_EROFS_FS_ZIP
548 		ctx->opt.cache_strategy = result.uint_32;
549 #else
550 		errorfc(fc, "compression not supported, cache_strategy ignored");
551 #endif
552 		break;
553 	case Opt_dax:
554 		if (!erofs_fc_set_dax_mode(fc, EROFS_MOUNT_DAX_ALWAYS))
555 			return -EINVAL;
556 		break;
557 	case Opt_dax_enum:
558 		if (!erofs_fc_set_dax_mode(fc, result.uint_32))
559 			return -EINVAL;
560 		break;
561 	case Opt_device:
562 		dif = kzalloc(sizeof(*dif), GFP_KERNEL);
563 		if (!dif)
564 			return -ENOMEM;
565 		dif->path = kstrdup(param->string, GFP_KERNEL);
566 		if (!dif->path) {
567 			kfree(dif);
568 			return -ENOMEM;
569 		}
570 		down_write(&ctx->devs->rwsem);
571 		ret = idr_alloc(&ctx->devs->tree, dif, 0, 0, GFP_KERNEL);
572 		up_write(&ctx->devs->rwsem);
573 		if (ret < 0) {
574 			kfree(dif->path);
575 			kfree(dif);
576 			return ret;
577 		}
578 		++ctx->devs->extra_devices;
579 		break;
580 	case Opt_fsid:
581 #ifdef CONFIG_EROFS_FS_ONDEMAND
582 		kfree(ctx->fsid);
583 		ctx->fsid = kstrdup(param->string, GFP_KERNEL);
584 		if (!ctx->fsid)
585 			return -ENOMEM;
586 #else
587 		errorfc(fc, "fsid option not supported");
588 #endif
589 		break;
590 	case Opt_domain_id:
591 #ifdef CONFIG_EROFS_FS_ONDEMAND
592 		kfree(ctx->domain_id);
593 		ctx->domain_id = kstrdup(param->string, GFP_KERNEL);
594 		if (!ctx->domain_id)
595 			return -ENOMEM;
596 #else
597 		errorfc(fc, "domain_id option not supported");
598 #endif
599 		break;
600 	default:
601 		return -ENOPARAM;
602 	}
603 	return 0;
604 }
605 
606 #ifdef CONFIG_EROFS_FS_ZIP
607 static const struct address_space_operations managed_cache_aops;
608 
609 static bool erofs_managed_cache_release_folio(struct folio *folio, gfp_t gfp)
610 {
611 	bool ret = true;
612 	struct address_space *const mapping = folio->mapping;
613 
614 	DBG_BUGON(!folio_test_locked(folio));
615 	DBG_BUGON(mapping->a_ops != &managed_cache_aops);
616 
617 	if (folio_test_private(folio))
618 		ret = erofs_try_to_free_cached_page(&folio->page);
619 
620 	return ret;
621 }
622 
623 /*
624  * It will be called only on inode eviction. In case that there are still some
625  * decompression requests in progress, wait with rescheduling for a bit here.
626  * We could introduce an extra locking instead but it seems unnecessary.
627  */
628 static void erofs_managed_cache_invalidate_folio(struct folio *folio,
629 					       size_t offset, size_t length)
630 {
631 	const size_t stop = length + offset;
632 
633 	DBG_BUGON(!folio_test_locked(folio));
634 
635 	/* Check for potential overflow in debug mode */
636 	DBG_BUGON(stop > folio_size(folio) || stop < length);
637 
638 	if (offset == 0 && stop == folio_size(folio))
639 		while (!erofs_managed_cache_release_folio(folio, GFP_NOFS))
640 			cond_resched();
641 }
642 
643 static const struct address_space_operations managed_cache_aops = {
644 	.release_folio = erofs_managed_cache_release_folio,
645 	.invalidate_folio = erofs_managed_cache_invalidate_folio,
646 };
647 
648 static int erofs_init_managed_cache(struct super_block *sb)
649 {
650 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
651 	struct inode *const inode = new_inode(sb);
652 
653 	if (!inode)
654 		return -ENOMEM;
655 
656 	set_nlink(inode, 1);
657 	inode->i_size = OFFSET_MAX;
658 
659 	inode->i_mapping->a_ops = &managed_cache_aops;
660 	mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
661 	sbi->managed_cache = inode;
662 	return 0;
663 }
664 #else
665 static int erofs_init_managed_cache(struct super_block *sb) { return 0; }
666 #endif
667 
668 static struct inode *erofs_nfs_get_inode(struct super_block *sb,
669 					 u64 ino, u32 generation)
670 {
671 	return erofs_iget(sb, ino);
672 }
673 
674 static struct dentry *erofs_fh_to_dentry(struct super_block *sb,
675 		struct fid *fid, int fh_len, int fh_type)
676 {
677 	return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
678 				    erofs_nfs_get_inode);
679 }
680 
681 static struct dentry *erofs_fh_to_parent(struct super_block *sb,
682 		struct fid *fid, int fh_len, int fh_type)
683 {
684 	return generic_fh_to_parent(sb, fid, fh_len, fh_type,
685 				    erofs_nfs_get_inode);
686 }
687 
688 static struct dentry *erofs_get_parent(struct dentry *child)
689 {
690 	erofs_nid_t nid;
691 	unsigned int d_type;
692 	int err;
693 
694 	err = erofs_namei(d_inode(child), &dotdot_name, &nid, &d_type);
695 	if (err)
696 		return ERR_PTR(err);
697 	return d_obtain_alias(erofs_iget(child->d_sb, nid));
698 }
699 
700 static const struct export_operations erofs_export_ops = {
701 	.fh_to_dentry = erofs_fh_to_dentry,
702 	.fh_to_parent = erofs_fh_to_parent,
703 	.get_parent = erofs_get_parent,
704 };
705 
706 static int erofs_fc_fill_pseudo_super(struct super_block *sb, struct fs_context *fc)
707 {
708 	static const struct tree_descr empty_descr = {""};
709 
710 	return simple_fill_super(sb, EROFS_SUPER_MAGIC, &empty_descr);
711 }
712 
713 static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
714 {
715 	struct inode *inode;
716 	struct erofs_sb_info *sbi;
717 	struct erofs_fs_context *ctx = fc->fs_private;
718 	int err;
719 
720 	sb->s_magic = EROFS_SUPER_MAGIC;
721 	sb->s_flags |= SB_RDONLY | SB_NOATIME;
722 	sb->s_maxbytes = MAX_LFS_FILESIZE;
723 	sb->s_op = &erofs_sops;
724 
725 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
726 	if (!sbi)
727 		return -ENOMEM;
728 
729 	sb->s_fs_info = sbi;
730 	sbi->opt = ctx->opt;
731 	sbi->devs = ctx->devs;
732 	ctx->devs = NULL;
733 	sbi->fsid = ctx->fsid;
734 	ctx->fsid = NULL;
735 	sbi->domain_id = ctx->domain_id;
736 	ctx->domain_id = NULL;
737 
738 	if (erofs_is_fscache_mode(sb)) {
739 		sb->s_blocksize = EROFS_BLKSIZ;
740 		sb->s_blocksize_bits = LOG_BLOCK_SIZE;
741 
742 		err = erofs_fscache_register_fs(sb);
743 		if (err)
744 			return err;
745 
746 		err = super_setup_bdi(sb);
747 		if (err)
748 			return err;
749 	} else {
750 		if (!sb_set_blocksize(sb, EROFS_BLKSIZ)) {
751 			erofs_err(sb, "failed to set erofs blksize");
752 			return -EINVAL;
753 		}
754 
755 		sbi->dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
756 						  &sbi->dax_part_off,
757 						  NULL, NULL);
758 	}
759 
760 	err = erofs_read_superblock(sb);
761 	if (err)
762 		return err;
763 
764 	if (test_opt(&sbi->opt, DAX_ALWAYS)) {
765 		BUILD_BUG_ON(EROFS_BLKSIZ != PAGE_SIZE);
766 
767 		if (!sbi->dax_dev) {
768 			errorfc(fc, "DAX unsupported by block device. Turning off DAX.");
769 			clear_opt(&sbi->opt, DAX_ALWAYS);
770 		}
771 	}
772 
773 	sb->s_time_gran = 1;
774 	sb->s_xattr = erofs_xattr_handlers;
775 	sb->s_export_op = &erofs_export_ops;
776 
777 	if (test_opt(&sbi->opt, POSIX_ACL))
778 		sb->s_flags |= SB_POSIXACL;
779 	else
780 		sb->s_flags &= ~SB_POSIXACL;
781 
782 #ifdef CONFIG_EROFS_FS_ZIP
783 	xa_init(&sbi->managed_pslots);
784 #endif
785 
786 	/* get the root inode */
787 	inode = erofs_iget(sb, ROOT_NID(sbi));
788 	if (IS_ERR(inode))
789 		return PTR_ERR(inode);
790 
791 	if (!S_ISDIR(inode->i_mode)) {
792 		erofs_err(sb, "rootino(nid %llu) is not a directory(i_mode %o)",
793 			  ROOT_NID(sbi), inode->i_mode);
794 		iput(inode);
795 		return -EINVAL;
796 	}
797 
798 	sb->s_root = d_make_root(inode);
799 	if (!sb->s_root)
800 		return -ENOMEM;
801 
802 	erofs_shrinker_register(sb);
803 	/* sb->s_umount is already locked, SB_ACTIVE and SB_BORN are not set */
804 	err = erofs_init_managed_cache(sb);
805 	if (err)
806 		return err;
807 
808 	err = erofs_register_sysfs(sb);
809 	if (err)
810 		return err;
811 
812 	erofs_info(sb, "mounted with root inode @ nid %llu.", ROOT_NID(sbi));
813 	return 0;
814 }
815 
816 static int erofs_fc_anon_get_tree(struct fs_context *fc)
817 {
818 	return get_tree_nodev(fc, erofs_fc_fill_pseudo_super);
819 }
820 
821 static int erofs_fc_get_tree(struct fs_context *fc)
822 {
823 	struct erofs_fs_context *ctx = fc->fs_private;
824 
825 	if (IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && ctx->fsid)
826 		return get_tree_nodev(fc, erofs_fc_fill_super);
827 
828 	return get_tree_bdev(fc, erofs_fc_fill_super);
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_fs_context *ctx = fc->fs_private;
836 
837 	DBG_BUGON(!sb_rdonly(sb));
838 
839 	if (ctx->fsid || ctx->domain_id)
840 		erofs_info(sb, "ignoring reconfiguration for fsid|domain_id.");
841 
842 	if (test_opt(&ctx->opt, POSIX_ACL))
843 		fc->sb_flags |= SB_POSIXACL;
844 	else
845 		fc->sb_flags &= ~SB_POSIXACL;
846 
847 	sbi->opt = ctx->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->bdev)
859 		blkdev_put(dif->bdev, FMODE_READ | FMODE_EXCL);
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_fc_free(struct fs_context *fc)
877 {
878 	struct erofs_fs_context *ctx = fc->fs_private;
879 
880 	erofs_free_dev_context(ctx->devs);
881 	kfree(ctx->fsid);
882 	kfree(ctx->domain_id);
883 	kfree(ctx);
884 }
885 
886 static const struct fs_context_operations erofs_context_ops = {
887 	.parse_param	= erofs_fc_parse_param,
888 	.get_tree       = erofs_fc_get_tree,
889 	.reconfigure    = erofs_fc_reconfigure,
890 	.free		= erofs_fc_free,
891 };
892 
893 static const struct fs_context_operations erofs_anon_context_ops = {
894 	.get_tree       = erofs_fc_anon_get_tree,
895 };
896 
897 static int erofs_init_fs_context(struct fs_context *fc)
898 {
899 	struct erofs_fs_context *ctx;
900 
901 	/* pseudo mount for anon inodes */
902 	if (fc->sb_flags & SB_KERNMOUNT) {
903 		fc->ops = &erofs_anon_context_ops;
904 		return 0;
905 	}
906 
907 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
908 	if (!ctx)
909 		return -ENOMEM;
910 	ctx->devs = kzalloc(sizeof(struct erofs_dev_context), GFP_KERNEL);
911 	if (!ctx->devs) {
912 		kfree(ctx);
913 		return -ENOMEM;
914 	}
915 	fc->fs_private = ctx;
916 
917 	idr_init(&ctx->devs->tree);
918 	init_rwsem(&ctx->devs->rwsem);
919 	erofs_default_options(ctx);
920 	fc->ops = &erofs_context_ops;
921 	return 0;
922 }
923 
924 /*
925  * could be triggered after deactivate_locked_super()
926  * is called, thus including umount and failed to initialize.
927  */
928 static void erofs_kill_sb(struct super_block *sb)
929 {
930 	struct erofs_sb_info *sbi;
931 
932 	WARN_ON(sb->s_magic != EROFS_SUPER_MAGIC);
933 
934 	/* pseudo mount for anon inodes */
935 	if (sb->s_flags & SB_KERNMOUNT) {
936 		kill_anon_super(sb);
937 		return;
938 	}
939 
940 	if (erofs_is_fscache_mode(sb))
941 		kill_anon_super(sb);
942 	else
943 		kill_block_super(sb);
944 
945 	sbi = EROFS_SB(sb);
946 	if (!sbi)
947 		return;
948 
949 	erofs_free_dev_context(sbi->devs);
950 	fs_put_dax(sbi->dax_dev, NULL);
951 	erofs_fscache_unregister_fs(sb);
952 	kfree(sbi->fsid);
953 	kfree(sbi->domain_id);
954 	kfree(sbi);
955 	sb->s_fs_info = NULL;
956 }
957 
958 /* called when ->s_root is non-NULL */
959 static void erofs_put_super(struct super_block *sb)
960 {
961 	struct erofs_sb_info *const sbi = EROFS_SB(sb);
962 
963 	DBG_BUGON(!sbi);
964 
965 	erofs_unregister_sysfs(sb);
966 	erofs_shrinker_unregister(sb);
967 #ifdef CONFIG_EROFS_FS_ZIP
968 	iput(sbi->managed_cache);
969 	sbi->managed_cache = NULL;
970 	iput(sbi->packed_inode);
971 	sbi->packed_inode = NULL;
972 #endif
973 	erofs_fscache_unregister_fs(sb);
974 }
975 
976 struct file_system_type erofs_fs_type = {
977 	.owner          = THIS_MODULE,
978 	.name           = "erofs",
979 	.init_fs_context = erofs_init_fs_context,
980 	.kill_sb        = erofs_kill_sb,
981 	.fs_flags       = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
982 };
983 MODULE_ALIAS_FS("erofs");
984 
985 static int __init erofs_module_init(void)
986 {
987 	int err;
988 
989 	erofs_check_ondisk_layout_definitions();
990 
991 	erofs_inode_cachep = kmem_cache_create("erofs_inode",
992 					       sizeof(struct erofs_inode), 0,
993 					       SLAB_RECLAIM_ACCOUNT,
994 					       erofs_inode_init_once);
995 	if (!erofs_inode_cachep) {
996 		err = -ENOMEM;
997 		goto icache_err;
998 	}
999 
1000 	err = erofs_init_shrinker();
1001 	if (err)
1002 		goto shrinker_err;
1003 
1004 	err = z_erofs_lzma_init();
1005 	if (err)
1006 		goto lzma_err;
1007 
1008 	erofs_pcpubuf_init();
1009 	err = z_erofs_init_zip_subsystem();
1010 	if (err)
1011 		goto zip_err;
1012 
1013 	err = erofs_init_sysfs();
1014 	if (err)
1015 		goto sysfs_err;
1016 
1017 	err = register_filesystem(&erofs_fs_type);
1018 	if (err)
1019 		goto fs_err;
1020 
1021 	return 0;
1022 
1023 fs_err:
1024 	erofs_exit_sysfs();
1025 sysfs_err:
1026 	z_erofs_exit_zip_subsystem();
1027 zip_err:
1028 	z_erofs_lzma_exit();
1029 lzma_err:
1030 	erofs_exit_shrinker();
1031 shrinker_err:
1032 	kmem_cache_destroy(erofs_inode_cachep);
1033 icache_err:
1034 	return err;
1035 }
1036 
1037 static void __exit erofs_module_exit(void)
1038 {
1039 	unregister_filesystem(&erofs_fs_type);
1040 
1041 	/* Ensure all RCU free inodes / pclusters are safe to be destroyed. */
1042 	rcu_barrier();
1043 
1044 	erofs_exit_sysfs();
1045 	z_erofs_exit_zip_subsystem();
1046 	z_erofs_lzma_exit();
1047 	erofs_exit_shrinker();
1048 	kmem_cache_destroy(erofs_inode_cachep);
1049 	erofs_pcpubuf_exit();
1050 }
1051 
1052 /* get filesystem statistics */
1053 static int erofs_statfs(struct dentry *dentry, struct kstatfs *buf)
1054 {
1055 	struct super_block *sb = dentry->d_sb;
1056 	struct erofs_sb_info *sbi = EROFS_SB(sb);
1057 	u64 id = 0;
1058 
1059 	if (!erofs_is_fscache_mode(sb))
1060 		id = huge_encode_dev(sb->s_bdev->bd_dev);
1061 
1062 	buf->f_type = sb->s_magic;
1063 	buf->f_bsize = EROFS_BLKSIZ;
1064 	buf->f_blocks = sbi->total_blocks;
1065 	buf->f_bfree = buf->f_bavail = 0;
1066 
1067 	buf->f_files = ULLONG_MAX;
1068 	buf->f_ffree = ULLONG_MAX - sbi->inos;
1069 
1070 	buf->f_namelen = EROFS_NAME_LEN;
1071 
1072 	buf->f_fsid    = u64_to_fsid(id);
1073 	return 0;
1074 }
1075 
1076 static int erofs_show_options(struct seq_file *seq, struct dentry *root)
1077 {
1078 	struct erofs_sb_info *sbi = EROFS_SB(root->d_sb);
1079 	struct erofs_mount_opts *opt = &sbi->opt;
1080 
1081 #ifdef CONFIG_EROFS_FS_XATTR
1082 	if (test_opt(opt, XATTR_USER))
1083 		seq_puts(seq, ",user_xattr");
1084 	else
1085 		seq_puts(seq, ",nouser_xattr");
1086 #endif
1087 #ifdef CONFIG_EROFS_FS_POSIX_ACL
1088 	if (test_opt(opt, POSIX_ACL))
1089 		seq_puts(seq, ",acl");
1090 	else
1091 		seq_puts(seq, ",noacl");
1092 #endif
1093 #ifdef CONFIG_EROFS_FS_ZIP
1094 	if (opt->cache_strategy == EROFS_ZIP_CACHE_DISABLED)
1095 		seq_puts(seq, ",cache_strategy=disabled");
1096 	else if (opt->cache_strategy == EROFS_ZIP_CACHE_READAHEAD)
1097 		seq_puts(seq, ",cache_strategy=readahead");
1098 	else if (opt->cache_strategy == EROFS_ZIP_CACHE_READAROUND)
1099 		seq_puts(seq, ",cache_strategy=readaround");
1100 #endif
1101 	if (test_opt(opt, DAX_ALWAYS))
1102 		seq_puts(seq, ",dax=always");
1103 	if (test_opt(opt, DAX_NEVER))
1104 		seq_puts(seq, ",dax=never");
1105 #ifdef CONFIG_EROFS_FS_ONDEMAND
1106 	if (sbi->fsid)
1107 		seq_printf(seq, ",fsid=%s", sbi->fsid);
1108 	if (sbi->domain_id)
1109 		seq_printf(seq, ",domain_id=%s", sbi->domain_id);
1110 #endif
1111 	return 0;
1112 }
1113 
1114 const struct super_operations erofs_sops = {
1115 	.put_super = erofs_put_super,
1116 	.alloc_inode = erofs_alloc_inode,
1117 	.free_inode = erofs_free_inode,
1118 	.statfs = erofs_statfs,
1119 	.show_options = erofs_show_options,
1120 };
1121 
1122 module_init(erofs_module_init);
1123 module_exit(erofs_module_exit);
1124 
1125 MODULE_DESCRIPTION("Enhanced ROM File System");
1126 MODULE_AUTHOR("Gao Xiang, Chao Yu, Miao Xie, CONSUMER BG, HUAWEI Inc.");
1127 MODULE_LICENSE("GPL");
1128