super.c

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