inode.c

   1/*
   2 * Compressed rom filesystem for Linux.
   3 *
   4 * Copyright (C) 1999 Linus Torvalds.
   5 *
   6 * This file is released under the GPL.
   7 */
   8
   9/*
  10 * These are the VFS interfaces to the compressed rom filesystem.
  11 * The actual compression is based on zlib, see the other files.
  12 */
  13
  14#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15
  16#include <linux/module.h>
  17#include <linux/fs.h>
  18#include <linux/file.h>
  19#include <linux/pagemap.h>
  20#include <linux/ramfs.h>
  21#include <linux/init.h>
  22#include <linux/string.h>
  23#include <linux/blkdev.h>
  24#include <linux/mtd/mtd.h>
  25#include <linux/mtd/super.h>
  26#include <linux/fs_context.h>
  27#include <linux/slab.h>
  28#include <linux/vfs.h>
  29#include <linux/mutex.h>
  30#include <uapi/linux/cramfs_fs.h>
  31#include <linux/uaccess.h>
  32
  33#include "internal.h"
  34
  35/*
  36 * cramfs super-block data in memory
  37 */
  38struct cramfs_sb_info {
  39	unsigned long magic;
  40	unsigned long size;
  41	unsigned long blocks;
  42	unsigned long files;
  43	unsigned long flags;
  44	void *linear_virt_addr;
  45	resource_size_t linear_phys_addr;
  46	size_t mtd_point_size;
  47};
  48
  49static inline struct cramfs_sb_info *CRAMFS_SB(struct super_block *sb)
  50{
  51	return sb->s_fs_info;
  52}
  53
  54static const struct super_operations cramfs_ops;
  55static const struct inode_operations cramfs_dir_inode_operations;
  56static const struct file_operations cramfs_directory_operations;
  57static const struct file_operations cramfs_physmem_fops;
  58static const struct address_space_operations cramfs_aops;
  59
  60static DEFINE_MUTEX(read_mutex);
  61
  62
  63/* These macros may change in future, to provide better st_ino semantics. */
  64#define OFFSET(x)	((x)->i_ino)
  65
  66static unsigned long cramino(const struct cramfs_inode *cino, unsigned int offset)
  67{
  68	if (!cino->offset)
  69		return offset + 1;
  70	if (!cino->size)
  71		return offset + 1;
  72
  73	/*
  74	 * The file mode test fixes buggy mkcramfs implementations where
  75	 * cramfs_inode->offset is set to a non zero value for entries
  76	 * which did not contain data, like devices node and fifos.
  77	 */
  78	switch (cino->mode & S_IFMT) {
  79	case S_IFREG:
  80	case S_IFDIR:
  81	case S_IFLNK:
  82		return cino->offset << 2;
  83	default:
  84		break;
  85	}
  86	return offset + 1;
  87}
  88
  89static struct inode *get_cramfs_inode(struct super_block *sb,
  90	const struct cramfs_inode *cramfs_inode, unsigned int offset)
  91{
  92	struct inode *inode;
  93	static struct timespec64 zerotime;
  94
  95	inode = iget_locked(sb, cramino(cramfs_inode, offset));
  96	if (!inode)
  97		return ERR_PTR(-ENOMEM);
  98	if (!(inode_state_read_once(inode) & I_NEW))
  99		return inode;
 100
 101	switch (cramfs_inode->mode & S_IFMT) {
 102	case S_IFREG:
 103		inode->i_fop = &generic_ro_fops;
 104		inode->i_data.a_ops = &cramfs_aops;
 105		if (IS_ENABLED(CONFIG_CRAMFS_MTD) &&
 106		    CRAMFS_SB(sb)->flags & CRAMFS_FLAG_EXT_BLOCK_POINTERS &&
 107		    CRAMFS_SB(sb)->linear_phys_addr)
 108			inode->i_fop = &cramfs_physmem_fops;
 109		break;
 110	case S_IFDIR:
 111		inode->i_op = &cramfs_dir_inode_operations;
 112		inode->i_fop = &cramfs_directory_operations;
 113		break;
 114	case S_IFLNK:
 115		inode->i_op = &page_symlink_inode_operations;
 116		inode_nohighmem(inode);
 117		inode->i_data.a_ops = &cramfs_aops;
 118		break;
 119	case S_IFCHR:
 120	case S_IFBLK:
 121	case S_IFIFO:
 122	case S_IFSOCK:
 123		init_special_inode(inode, cramfs_inode->mode,
 124				old_decode_dev(cramfs_inode->size));
 125		break;
 126	default:
 127		printk(KERN_DEBUG "CRAMFS: Invalid file type 0%04o for inode %lu.\n",
 128		       inode->i_mode, inode->i_ino);
 129		iget_failed(inode);
 130		return ERR_PTR(-EIO);
 131	}
 132
 133	inode->i_mode = cramfs_inode->mode;
 134	i_uid_write(inode, cramfs_inode->uid);
 135	i_gid_write(inode, cramfs_inode->gid);
 136
 137	/* if the lower 2 bits are zero, the inode contains data */
 138	if (!(inode->i_ino & 3)) {
 139		inode->i_size = cramfs_inode->size;
 140		inode->i_blocks = (cramfs_inode->size - 1) / 512 + 1;
 141	}
 142
 143	/* Struct copy intentional */
 144	inode_set_mtime_to_ts(inode,
 145			      inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, zerotime)));
 146	/* inode->i_nlink is left 1 - arguably wrong for directories,
 147	   but it's the best we can do without reading the directory
 148	   contents.  1 yields the right result in GNU find, even
 149	   without -noleaf option. */
 150
 151	unlock_new_inode(inode);
 152
 153	return inode;
 154}
 155
 156/*
 157 * We have our own block cache: don't fill up the buffer cache
 158 * with the rom-image, because the way the filesystem is set
 159 * up the accesses should be fairly regular and cached in the
 160 * page cache and dentry tree anyway..
 161 *
 162 * This also acts as a way to guarantee contiguous areas of up to
 163 * BLKS_PER_BUF*PAGE_SIZE, so that the caller doesn't need to
 164 * worry about end-of-buffer issues even when decompressing a full
 165 * page cache.
 166 *
 167 * Note: This is all optimized away at compile time when
 168 *       CONFIG_CRAMFS_BLOCKDEV=n.
 169 */
 170#define READ_BUFFERS (2)
 171/* NEXT_BUFFER(): Loop over [0..(READ_BUFFERS-1)]. */
 172#define NEXT_BUFFER(_ix) ((_ix) ^ 1)
 173
 174/*
 175 * BLKS_PER_BUF_SHIFT should be at least 2 to allow for "compressed"
 176 * data that takes up more space than the original and with unlucky
 177 * alignment.
 178 */
 179#define BLKS_PER_BUF_SHIFT	(2)
 180#define BLKS_PER_BUF		(1 << BLKS_PER_BUF_SHIFT)
 181#define BUFFER_SIZE		(BLKS_PER_BUF*PAGE_SIZE)
 182
 183static unsigned char read_buffers[READ_BUFFERS][BUFFER_SIZE];
 184static unsigned buffer_blocknr[READ_BUFFERS];
 185static struct super_block *buffer_dev[READ_BUFFERS];
 186static int next_buffer;
 187
 188/*
 189 * Populate our block cache and return a pointer to it.
 190 */
 191static void *cramfs_blkdev_read(struct super_block *sb, unsigned int offset,
 192				unsigned int len)
 193{
 194	struct address_space *mapping = sb->s_bdev->bd_mapping;
 195	struct file_ra_state ra = {};
 196	struct page *pages[BLKS_PER_BUF];
 197	unsigned i, blocknr, buffer;
 198	unsigned long devsize;
 199	char *data;
 200
 201	if (!len)
 202		return NULL;
 203	blocknr = offset >> PAGE_SHIFT;
 204	offset &= PAGE_SIZE - 1;
 205
 206	/* Check if an existing buffer already has the data.. */
 207	for (i = 0; i < READ_BUFFERS; i++) {
 208		unsigned int blk_offset;
 209
 210		if (buffer_dev[i] != sb)
 211			continue;
 212		if (blocknr < buffer_blocknr[i])
 213			continue;
 214		blk_offset = (blocknr - buffer_blocknr[i]) << PAGE_SHIFT;
 215		blk_offset += offset;
 216		if (blk_offset > BUFFER_SIZE ||
 217		    blk_offset + len > BUFFER_SIZE)
 218			continue;
 219		return read_buffers[i] + blk_offset;
 220	}
 221
 222	devsize = bdev_nr_bytes(sb->s_bdev) >> PAGE_SHIFT;
 223
 224	/* Ok, read in BLKS_PER_BUF pages completely first. */
 225	file_ra_state_init(&ra, mapping);
 226	page_cache_sync_readahead(mapping, &ra, NULL, blocknr, BLKS_PER_BUF);
 227
 228	for (i = 0; i < BLKS_PER_BUF; i++) {
 229		struct page *page = NULL;
 230
 231		if (blocknr + i < devsize) {
 232			page = read_mapping_page(mapping, blocknr + i, NULL);
 233			/* synchronous error? */
 234			if (IS_ERR(page))
 235				page = NULL;
 236		}
 237		pages[i] = page;
 238	}
 239
 240	buffer = next_buffer;
 241	next_buffer = NEXT_BUFFER(buffer);
 242	buffer_blocknr[buffer] = blocknr;
 243	buffer_dev[buffer] = sb;
 244
 245	data = read_buffers[buffer];
 246	for (i = 0; i < BLKS_PER_BUF; i++) {
 247		struct page *page = pages[i];
 248
 249		if (page) {
 250			memcpy_from_page(data, page, 0, PAGE_SIZE);
 251			put_page(page);
 252		} else
 253			memset(data, 0, PAGE_SIZE);
 254		data += PAGE_SIZE;
 255	}
 256	return read_buffers[buffer] + offset;
 257}
 258
 259/*
 260 * Return a pointer to the linearly addressed cramfs image in memory.
 261 */
 262static void *cramfs_direct_read(struct super_block *sb, unsigned int offset,
 263				unsigned int len)
 264{
 265	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
 266
 267	if (!len)
 268		return NULL;
 269	if (len > sbi->size || offset > sbi->size - len)
 270		return page_address(ZERO_PAGE(0));
 271	return sbi->linear_virt_addr + offset;
 272}
 273
 274/*
 275 * Returns a pointer to a buffer containing at least LEN bytes of
 276 * filesystem starting at byte offset OFFSET into the filesystem.
 277 */
 278static void *cramfs_read(struct super_block *sb, unsigned int offset,
 279			 unsigned int len)
 280{
 281	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
 282
 283	if (IS_ENABLED(CONFIG_CRAMFS_MTD) && sbi->linear_virt_addr)
 284		return cramfs_direct_read(sb, offset, len);
 285	else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV))
 286		return cramfs_blkdev_read(sb, offset, len);
 287	else
 288		return NULL;
 289}
 290
 291/*
 292 * For a mapping to be possible, we need a range of uncompressed and
 293 * contiguous blocks. Return the offset for the first block and number of
 294 * valid blocks for which that is true, or zero otherwise.
 295 */
 296static u32 cramfs_get_block_range(struct inode *inode, u32 pgoff, u32 *pages)
 297{
 298	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
 299	int i;
 300	u32 *blockptrs, first_block_addr;
 301
 302	/*
 303	 * We can dereference memory directly here as this code may be
 304	 * reached only when there is a direct filesystem image mapping
 305	 * available in memory.
 306	 */
 307	blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode) + pgoff * 4);
 308	first_block_addr = blockptrs[0] & ~CRAMFS_BLK_FLAGS;
 309	i = 0;
 310	do {
 311		u32 block_off = i * (PAGE_SIZE >> CRAMFS_BLK_DIRECT_PTR_SHIFT);
 312		u32 expect = (first_block_addr + block_off) |
 313			     CRAMFS_BLK_FLAG_DIRECT_PTR |
 314			     CRAMFS_BLK_FLAG_UNCOMPRESSED;
 315		if (blockptrs[i] != expect) {
 316			pr_debug("range: block %d/%d got %#x expects %#x\n",
 317				 pgoff+i, pgoff + *pages - 1,
 318				 blockptrs[i], expect);
 319			if (i == 0)
 320				return 0;
 321			break;
 322		}
 323	} while (++i < *pages);
 324
 325	*pages = i;
 326	return first_block_addr << CRAMFS_BLK_DIRECT_PTR_SHIFT;
 327}
 328
 329#ifdef CONFIG_MMU
 330
 331/*
 332 * Return true if the last page of a file in the filesystem image contains
 333 * some other data that doesn't belong to that file. It is assumed that the
 334 * last block is CRAMFS_BLK_FLAG_DIRECT_PTR | CRAMFS_BLK_FLAG_UNCOMPRESSED
 335 * (verified by cramfs_get_block_range() and directly accessible in memory.
 336 */
 337static bool cramfs_last_page_is_shared(struct inode *inode)
 338{
 339	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
 340	u32 partial, last_page, blockaddr, *blockptrs;
 341	char *tail_data;
 342
 343	partial = offset_in_page(inode->i_size);
 344	if (!partial)
 345		return false;
 346	last_page = inode->i_size >> PAGE_SHIFT;
 347	blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode));
 348	blockaddr = blockptrs[last_page] & ~CRAMFS_BLK_FLAGS;
 349	blockaddr <<= CRAMFS_BLK_DIRECT_PTR_SHIFT;
 350	tail_data = sbi->linear_virt_addr + blockaddr + partial;
 351	return memchr_inv(tail_data, 0, PAGE_SIZE - partial) ? true : false;
 352}
 353
 354static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct *vma)
 355{
 356	struct inode *inode = file_inode(file);
 357	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
 358	unsigned int pages, max_pages, offset;
 359	unsigned long address, pgoff = vma->vm_pgoff;
 360	char *bailout_reason;
 361	int ret;
 362
 363	ret = generic_file_readonly_mmap(file, vma);
 364	if (ret)
 365		return ret;
 366
 367	/*
 368	 * Now try to pre-populate ptes for this vma with a direct
 369	 * mapping avoiding memory allocation when possible.
 370	 */
 371
 372	/* Could COW work here? */
 373	bailout_reason = "vma is writable";
 374	if (vma->vm_flags & VM_WRITE)
 375		goto bailout;
 376
 377	max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 378	bailout_reason = "beyond file limit";
 379	if (pgoff >= max_pages)
 380		goto bailout;
 381	pages = min(vma_pages(vma), max_pages - pgoff);
 382
 383	offset = cramfs_get_block_range(inode, pgoff, &pages);
 384	bailout_reason = "unsuitable block layout";
 385	if (!offset)
 386		goto bailout;
 387	address = sbi->linear_phys_addr + offset;
 388	bailout_reason = "data is not page aligned";
 389	if (!PAGE_ALIGNED(address))
 390		goto bailout;
 391
 392	/* Don't map the last page if it contains some other data */
 393	if (pgoff + pages == max_pages && cramfs_last_page_is_shared(inode)) {
 394		pr_debug("mmap: %pD: last page is shared\n", file);
 395		pages--;
 396	}
 397
 398	if (!pages) {
 399		bailout_reason = "no suitable block remaining";
 400		goto bailout;
 401	}
 402
 403	if (pages == vma_pages(vma)) {
 404		/*
 405		 * The entire vma is mappable. remap_pfn_range() will
 406		 * make it distinguishable from a non-direct mapping
 407		 * in /proc/<pid>/maps by substituting the file offset
 408		 * with the actual physical address.
 409		 */
 410		ret = remap_pfn_range(vma, vma->vm_start, address >> PAGE_SHIFT,
 411				      pages * PAGE_SIZE, vma->vm_page_prot);
 412	} else {
 413		/*
 414		 * Let's create a mixed map if we can't map it all.
 415		 * The normal paging machinery will take care of the
 416		 * unpopulated ptes via cramfs_read_folio().
 417		 */
 418		int i;
 419		vm_flags_set(vma, VM_MIXEDMAP);
 420		for (i = 0; i < pages && !ret; i++) {
 421			vm_fault_t vmf;
 422			unsigned long off = i * PAGE_SIZE;
 423			vmf = vmf_insert_mixed(vma, vma->vm_start + off,
 424					PHYS_PFN(address + off));
 425			if (vmf & VM_FAULT_ERROR)
 426				ret = vm_fault_to_errno(vmf, 0);
 427		}
 428	}
 429
 430	if (!ret)
 431		pr_debug("mapped %pD[%lu] at 0x%08lx (%u/%lu pages) "
 432			 "to vma 0x%08lx, page_prot 0x%llx\n", file,
 433			 pgoff, address, pages, vma_pages(vma), vma->vm_start,
 434			 (unsigned long long)pgprot_val(vma->vm_page_prot));
 435	return ret;
 436
 437bailout:
 438	pr_debug("%pD[%lu]: direct mmap impossible: %s\n",
 439		 file, pgoff, bailout_reason);
 440	/* Didn't manage any direct map, but normal paging is still possible */
 441	return 0;
 442}
 443
 444#else /* CONFIG_MMU */
 445
 446static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct *vma)
 447{
 448	return is_nommu_shared_mapping(vma->vm_flags) ? 0 : -ENOSYS;
 449}
 450
 451static unsigned long cramfs_physmem_get_unmapped_area(struct file *file,
 452			unsigned long addr, unsigned long len,
 453			unsigned long pgoff, unsigned long flags)
 454{
 455	struct inode *inode = file_inode(file);
 456	struct super_block *sb = inode->i_sb;
 457	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
 458	unsigned int pages, block_pages, max_pages, offset;
 459
 460	pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
 461	max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 462	if (pgoff >= max_pages || pages > max_pages - pgoff)
 463		return -EINVAL;
 464	block_pages = pages;
 465	offset = cramfs_get_block_range(inode, pgoff, &block_pages);
 466	if (!offset || block_pages != pages)
 467		return -ENOSYS;
 468	addr = sbi->linear_phys_addr + offset;
 469	pr_debug("get_unmapped for %pD ofs %#lx siz %lu at 0x%08lx\n",
 470		 file, pgoff*PAGE_SIZE, len, addr);
 471	return addr;
 472}
 473
 474static unsigned int cramfs_physmem_mmap_capabilities(struct file *file)
 475{
 476	return NOMMU_MAP_COPY | NOMMU_MAP_DIRECT |
 477	       NOMMU_MAP_READ | NOMMU_MAP_EXEC;
 478}
 479
 480#endif /* CONFIG_MMU */
 481
 482static const struct file_operations cramfs_physmem_fops = {
 483	.llseek			= generic_file_llseek,
 484	.read_iter		= generic_file_read_iter,
 485	.splice_read		= filemap_splice_read,
 486	.mmap			= cramfs_physmem_mmap,
 487#ifndef CONFIG_MMU
 488	.get_unmapped_area	= cramfs_physmem_get_unmapped_area,
 489	.mmap_capabilities	= cramfs_physmem_mmap_capabilities,
 490#endif
 491};
 492
 493static void cramfs_kill_sb(struct super_block *sb)
 494{
 495	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
 496
 497	generic_shutdown_super(sb);
 498
 499	if (IS_ENABLED(CONFIG_CRAMFS_MTD) && sb->s_mtd) {
 500		if (sbi && sbi->mtd_point_size)
 501			mtd_unpoint(sb->s_mtd, 0, sbi->mtd_point_size);
 502		put_mtd_device(sb->s_mtd);
 503		sb->s_mtd = NULL;
 504	} else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV) && sb->s_bdev) {
 505		sync_blockdev(sb->s_bdev);
 506		bdev_fput(sb->s_bdev_file);
 507	}
 508	kfree(sbi);
 509}
 510
 511static int cramfs_reconfigure(struct fs_context *fc)
 512{
 513	sync_filesystem(fc->root->d_sb);
 514	fc->sb_flags |= SB_RDONLY;
 515	return 0;
 516}
 517
 518static int cramfs_read_super(struct super_block *sb, struct fs_context *fc,
 519			     struct cramfs_super *super)
 520{
 521	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
 522	unsigned long root_offset;
 523	bool silent = fc->sb_flags & SB_SILENT;
 524
 525	/* We don't know the real size yet */
 526	sbi->size = PAGE_SIZE;
 527
 528	/* Read the first block and get the superblock from it */
 529	mutex_lock(&read_mutex);
 530	memcpy(super, cramfs_read(sb, 0, sizeof(*super)), sizeof(*super));
 531	mutex_unlock(&read_mutex);
 532
 533	/* Do sanity checks on the superblock */
 534	if (super->magic != CRAMFS_MAGIC) {
 535		/* check for wrong endianness */
 536		if (super->magic == CRAMFS_MAGIC_WEND) {
 537			if (!silent)
 538				errorfc(fc, "wrong endianness");
 539			return -EINVAL;
 540		}
 541
 542		/* check at 512 byte offset */
 543		mutex_lock(&read_mutex);
 544		memcpy(super,
 545		       cramfs_read(sb, 512, sizeof(*super)),
 546		       sizeof(*super));
 547		mutex_unlock(&read_mutex);
 548		if (super->magic != CRAMFS_MAGIC) {
 549			if (super->magic == CRAMFS_MAGIC_WEND && !silent)
 550				errorfc(fc, "wrong endianness");
 551			else if (!silent)
 552				errorfc(fc, "wrong magic");
 553			return -EINVAL;
 554		}
 555	}
 556
 557	/* get feature flags first */
 558	if (super->flags & ~CRAMFS_SUPPORTED_FLAGS) {
 559		errorfc(fc, "unsupported filesystem features");
 560		return -EINVAL;
 561	}
 562
 563	/* Check that the root inode is in a sane state */
 564	if (!S_ISDIR(super->root.mode)) {
 565		errorfc(fc, "root is not a directory");
 566		return -EINVAL;
 567	}
 568	/* correct strange, hard-coded permissions of mkcramfs */
 569	super->root.mode |= 0555;
 570
 571	root_offset = super->root.offset << 2;
 572	if (super->flags & CRAMFS_FLAG_FSID_VERSION_2) {
 573		sbi->size = super->size;
 574		sbi->blocks = super->fsid.blocks;
 575		sbi->files = super->fsid.files;
 576	} else {
 577		sbi->size = 1<<28;
 578		sbi->blocks = 0;
 579		sbi->files = 0;
 580	}
 581	sbi->magic = super->magic;
 582	sbi->flags = super->flags;
 583	if (root_offset == 0)
 584		infofc(fc, "empty filesystem");
 585	else if (!(super->flags & CRAMFS_FLAG_SHIFTED_ROOT_OFFSET) &&
 586		 ((root_offset != sizeof(struct cramfs_super)) &&
 587		  (root_offset != 512 + sizeof(struct cramfs_super))))
 588	{
 589		errorfc(fc, "bad root offset %lu", root_offset);
 590		return -EINVAL;
 591	}
 592
 593	return 0;
 594}
 595
 596static int cramfs_finalize_super(struct super_block *sb,
 597				 struct cramfs_inode *cramfs_root)
 598{
 599	struct inode *root;
 600
 601	/* Set it all up.. */
 602	sb->s_flags |= SB_RDONLY;
 603	sb->s_time_min = 0;
 604	sb->s_time_max = 0;
 605	sb->s_op = &cramfs_ops;
 606	root = get_cramfs_inode(sb, cramfs_root, 0);
 607	if (IS_ERR(root))
 608		return PTR_ERR(root);
 609	sb->s_root = d_make_root(root);
 610	if (!sb->s_root)
 611		return -ENOMEM;
 612	return 0;
 613}
 614
 615static int cramfs_blkdev_fill_super(struct super_block *sb, struct fs_context *fc)
 616{
 617	struct cramfs_sb_info *sbi;
 618	struct cramfs_super super;
 619	int i, err;
 620
 621	sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
 622	if (!sbi)
 623		return -ENOMEM;
 624	sb->s_fs_info = sbi;
 625
 626	/* Invalidate the read buffers on mount: think disk change.. */
 627	for (i = 0; i < READ_BUFFERS; i++)
 628		buffer_blocknr[i] = -1;
 629
 630	err = cramfs_read_super(sb, fc, &super);
 631	if (err)
 632		return err;
 633	return cramfs_finalize_super(sb, &super.root);
 634}
 635
 636static int cramfs_mtd_fill_super(struct super_block *sb, struct fs_context *fc)
 637{
 638	struct cramfs_sb_info *sbi;
 639	struct cramfs_super super;
 640	int err;
 641
 642	sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
 643	if (!sbi)
 644		return -ENOMEM;
 645	sb->s_fs_info = sbi;
 646
 647	/* Map only one page for now.  Will remap it when fs size is known. */
 648	err = mtd_point(sb->s_mtd, 0, PAGE_SIZE, &sbi->mtd_point_size,
 649			&sbi->linear_virt_addr, &sbi->linear_phys_addr);
 650	if (err || sbi->mtd_point_size != PAGE_SIZE) {
 651		pr_err("unable to get direct memory access to mtd:%s\n",
 652		       sb->s_mtd->name);
 653		return err ? : -ENODATA;
 654	}
 655
 656	pr_info("checking physical address %pap for linear cramfs image\n",
 657		&sbi->linear_phys_addr);
 658	err = cramfs_read_super(sb, fc, &super);
 659	if (err)
 660		return err;
 661
 662	/* Remap the whole filesystem now */
 663	pr_info("linear cramfs image on mtd:%s appears to be %lu KB in size\n",
 664		sb->s_mtd->name, sbi->size/1024);
 665	mtd_unpoint(sb->s_mtd, 0, PAGE_SIZE);
 666	err = mtd_point(sb->s_mtd, 0, sbi->size, &sbi->mtd_point_size,
 667			&sbi->linear_virt_addr, &sbi->linear_phys_addr);
 668	if (err || sbi->mtd_point_size != sbi->size) {
 669		pr_err("unable to get direct memory access to mtd:%s\n",
 670		       sb->s_mtd->name);
 671		return err ? : -ENODATA;
 672	}
 673
 674	return cramfs_finalize_super(sb, &super.root);
 675}
 676
 677static int cramfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 678{
 679	struct super_block *sb = dentry->d_sb;
 680	u64 id = 0;
 681
 682	if (sb->s_bdev)
 683		id = huge_encode_dev(sb->s_bdev->bd_dev);
 684	else if (sb->s_dev)
 685		id = huge_encode_dev(sb->s_dev);
 686
 687	buf->f_type = CRAMFS_MAGIC;
 688	buf->f_bsize = PAGE_SIZE;
 689	buf->f_blocks = CRAMFS_SB(sb)->blocks;
 690	buf->f_bfree = 0;
 691	buf->f_bavail = 0;
 692	buf->f_files = CRAMFS_SB(sb)->files;
 693	buf->f_ffree = 0;
 694	buf->f_fsid = u64_to_fsid(id);
 695	buf->f_namelen = CRAMFS_MAXPATHLEN;
 696	return 0;
 697}
 698
 699/*
 700 * Read a cramfs directory entry.
 701 */
 702static int cramfs_readdir(struct file *file, struct dir_context *ctx)
 703{
 704	struct inode *inode = file_inode(file);
 705	struct super_block *sb = inode->i_sb;
 706	char *buf;
 707	unsigned int offset;
 708
 709	/* Offset within the thing. */
 710	if (ctx->pos >= inode->i_size)
 711		return 0;
 712	offset = ctx->pos;
 713	/* Directory entries are always 4-byte aligned */
 714	if (offset & 3)
 715		return -EINVAL;
 716
 717	buf = kmalloc(CRAMFS_MAXPATHLEN, GFP_KERNEL);
 718	if (!buf)
 719		return -ENOMEM;
 720
 721	while (offset < inode->i_size) {
 722		struct cramfs_inode *de;
 723		unsigned long nextoffset;
 724		char *name;
 725		ino_t ino;
 726		umode_t mode;
 727		int namelen;
 728
 729		mutex_lock(&read_mutex);
 730		de = cramfs_read(sb, OFFSET(inode) + offset, sizeof(*de)+CRAMFS_MAXPATHLEN);
 731		name = (char *)(de+1);
 732
 733		/*
 734		 * Namelengths on disk are shifted by two
 735		 * and the name padded out to 4-byte boundaries
 736		 * with zeroes.
 737		 */
 738		namelen = de->namelen << 2;
 739		memcpy(buf, name, namelen);
 740		ino = cramino(de, OFFSET(inode) + offset);
 741		mode = de->mode;
 742		mutex_unlock(&read_mutex);
 743		nextoffset = offset + sizeof(*de) + namelen;
 744		for (;;) {
 745			if (!namelen) {
 746				kfree(buf);
 747				return -EIO;
 748			}
 749			if (buf[namelen-1])
 750				break;
 751			namelen--;
 752		}
 753		if (!dir_emit(ctx, buf, namelen, ino, mode >> 12))
 754			break;
 755
 756		ctx->pos = offset = nextoffset;
 757	}
 758	kfree(buf);
 759	return 0;
 760}
 761
 762/*
 763 * Lookup and fill in the inode data..
 764 */
 765static struct dentry *cramfs_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
 766{
 767	unsigned int offset = 0;
 768	struct inode *inode = NULL;
 769	int sorted;
 770
 771	mutex_lock(&read_mutex);
 772	sorted = CRAMFS_SB(dir->i_sb)->flags & CRAMFS_FLAG_SORTED_DIRS;
 773	while (offset < dir->i_size) {
 774		struct cramfs_inode *de;
 775		char *name;
 776		int namelen, retval;
 777		int dir_off = OFFSET(dir) + offset;
 778
 779		de = cramfs_read(dir->i_sb, dir_off, sizeof(*de)+CRAMFS_MAXPATHLEN);
 780		name = (char *)(de+1);
 781
 782		/* Try to take advantage of sorted directories */
 783		if (sorted && (dentry->d_name.name[0] < name[0]))
 784			break;
 785
 786		namelen = de->namelen << 2;
 787		offset += sizeof(*de) + namelen;
 788
 789		/* Quick check that the name is roughly the right length */
 790		if (((dentry->d_name.len + 3) & ~3) != namelen)
 791			continue;
 792
 793		for (;;) {
 794			if (!namelen) {
 795				inode = ERR_PTR(-EIO);
 796				goto out;
 797			}
 798			if (name[namelen-1])
 799				break;
 800			namelen--;
 801		}
 802		if (namelen != dentry->d_name.len)
 803			continue;
 804		retval = memcmp(dentry->d_name.name, name, namelen);
 805		if (retval > 0)
 806			continue;
 807		if (!retval) {
 808			inode = get_cramfs_inode(dir->i_sb, de, dir_off);
 809			break;
 810		}
 811		/* else (retval < 0) */
 812		if (sorted)
 813			break;
 814	}
 815out:
 816	mutex_unlock(&read_mutex);
 817	return d_splice_alias(inode, dentry);
 818}
 819
 820static int cramfs_read_folio(struct file *file, struct folio *folio)
 821{
 822	struct inode *inode = folio->mapping->host;
 823	u32 maxblock;
 824	int bytes_filled;
 825	void *pgdata;
 826	bool success = false;
 827
 828	maxblock = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 829	bytes_filled = 0;
 830	pgdata = kmap_local_folio(folio, 0);
 831
 832	if (folio->index < maxblock) {
 833		struct super_block *sb = inode->i_sb;
 834		u32 blkptr_offset = OFFSET(inode) + folio->index * 4;
 835		u32 block_ptr, block_start, block_len;
 836		bool uncompressed, direct;
 837
 838		mutex_lock(&read_mutex);
 839		block_ptr = *(u32 *) cramfs_read(sb, blkptr_offset, 4);
 840		uncompressed = (block_ptr & CRAMFS_BLK_FLAG_UNCOMPRESSED);
 841		direct = (block_ptr & CRAMFS_BLK_FLAG_DIRECT_PTR);
 842		block_ptr &= ~CRAMFS_BLK_FLAGS;
 843
 844		if (direct) {
 845			/*
 846			 * The block pointer is an absolute start pointer,
 847			 * shifted by 2 bits. The size is included in the
 848			 * first 2 bytes of the data block when compressed,
 849			 * or PAGE_SIZE otherwise.
 850			 */
 851			block_start = block_ptr << CRAMFS_BLK_DIRECT_PTR_SHIFT;
 852			if (uncompressed) {
 853				block_len = PAGE_SIZE;
 854				/* if last block: cap to file length */
 855				if (folio->index == maxblock - 1)
 856					block_len =
 857						offset_in_page(inode->i_size);
 858			} else {
 859				block_len = *(u16 *)
 860					cramfs_read(sb, block_start, 2);
 861				block_start += 2;
 862			}
 863		} else {
 864			/*
 865			 * The block pointer indicates one past the end of
 866			 * the current block (start of next block). If this
 867			 * is the first block then it starts where the block
 868			 * pointer table ends, otherwise its start comes
 869			 * from the previous block's pointer.
 870			 */
 871			block_start = OFFSET(inode) + maxblock * 4;
 872			if (folio->index)
 873				block_start = *(u32 *)
 874					cramfs_read(sb, blkptr_offset - 4, 4);
 875			/* Beware... previous ptr might be a direct ptr */
 876			if (unlikely(block_start & CRAMFS_BLK_FLAG_DIRECT_PTR)) {
 877				/* See comments on earlier code. */
 878				u32 prev_start = block_start;
 879				block_start = prev_start & ~CRAMFS_BLK_FLAGS;
 880				block_start <<= CRAMFS_BLK_DIRECT_PTR_SHIFT;
 881				if (prev_start & CRAMFS_BLK_FLAG_UNCOMPRESSED) {
 882					block_start += PAGE_SIZE;
 883				} else {
 884					block_len = *(u16 *)
 885						cramfs_read(sb, block_start, 2);
 886					block_start += 2 + block_len;
 887				}
 888			}
 889			block_start &= ~CRAMFS_BLK_FLAGS;
 890			block_len = block_ptr - block_start;
 891		}
 892
 893		if (block_len == 0)
 894			; /* hole */
 895		else if (unlikely(block_len > 2*PAGE_SIZE ||
 896				  (uncompressed && block_len > PAGE_SIZE))) {
 897			mutex_unlock(&read_mutex);
 898			pr_err("bad data blocksize %u\n", block_len);
 899			goto err;
 900		} else if (uncompressed) {
 901			memcpy(pgdata,
 902			       cramfs_read(sb, block_start, block_len),
 903			       block_len);
 904			bytes_filled = block_len;
 905		} else {
 906			bytes_filled = cramfs_uncompress_block(pgdata,
 907				 PAGE_SIZE,
 908				 cramfs_read(sb, block_start, block_len),
 909				 block_len);
 910		}
 911		mutex_unlock(&read_mutex);
 912		if (unlikely(bytes_filled < 0))
 913			goto err;
 914	}
 915
 916	memset(pgdata + bytes_filled, 0, PAGE_SIZE - bytes_filled);
 917	flush_dcache_folio(folio);
 918
 919	success = true;
 920err:
 921	kunmap_local(pgdata);
 922	folio_end_read(folio, success);
 923	return 0;
 924}
 925
 926static const struct address_space_operations cramfs_aops = {
 927	.read_folio = cramfs_read_folio
 928};
 929
 930/*
 931 * Our operations:
 932 */
 933
 934/*
 935 * A directory can only readdir
 936 */
 937static const struct file_operations cramfs_directory_operations = {
 938	.llseek		= generic_file_llseek,
 939	.read		= generic_read_dir,
 940	.iterate_shared	= cramfs_readdir,
 941};
 942
 943static const struct inode_operations cramfs_dir_inode_operations = {
 944	.lookup		= cramfs_lookup,
 945};
 946
 947static const struct super_operations cramfs_ops = {
 948	.statfs		= cramfs_statfs,
 949};
 950
 951static int cramfs_get_tree(struct fs_context *fc)
 952{
 953	int ret = -ENOPROTOOPT;
 954
 955	if (IS_ENABLED(CONFIG_CRAMFS_MTD)) {
 956		ret = get_tree_mtd(fc, cramfs_mtd_fill_super);
 957		if (!ret)
 958			return 0;
 959	}
 960	if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV))
 961		ret = get_tree_bdev(fc, cramfs_blkdev_fill_super);
 962	return ret;
 963}
 964
 965static const struct fs_context_operations cramfs_context_ops = {
 966	.get_tree	= cramfs_get_tree,
 967	.reconfigure	= cramfs_reconfigure,
 968};
 969
 970/*
 971 * Set up the filesystem mount context.
 972 */
 973static int cramfs_init_fs_context(struct fs_context *fc)
 974{
 975	fc->ops = &cramfs_context_ops;
 976	return 0;
 977}
 978
 979static struct file_system_type cramfs_fs_type = {
 980	.owner		= THIS_MODULE,
 981	.name		= "cramfs",
 982	.init_fs_context = cramfs_init_fs_context,
 983	.kill_sb	= cramfs_kill_sb,
 984	.fs_flags	= FS_REQUIRES_DEV,
 985};
 986MODULE_ALIAS_FS("cramfs");
 987
 988static int __init init_cramfs_fs(void)
 989{
 990	int rv;
 991
 992	rv = cramfs_uncompress_init();
 993	if (rv < 0)
 994		return rv;
 995	rv = register_filesystem(&cramfs_fs_type);
 996	if (rv < 0)
 997		cramfs_uncompress_exit();
 998	return rv;
 999}
1000
1001static void __exit exit_cramfs_fs(void)
1002{
1003	cramfs_uncompress_exit();
1004	unregister_filesystem(&cramfs_fs_type);
1005}
1006
1007module_init(init_cramfs_fs)
1008module_exit(exit_cramfs_fs)
1009MODULE_DESCRIPTION("Compressed ROM file system support");
1010MODULE_LICENSE("GPL");