xref: /openbmc/linux/fs/cramfs/inode.c (revision 85250a24)
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/pfn_t.h>
21 #include <linux/ramfs.h>
22 #include <linux/init.h>
23 #include <linux/string.h>
24 #include <linux/blkdev.h>
25 #include <linux/mtd/mtd.h>
26 #include <linux/mtd/super.h>
27 #include <linux/fs_context.h>
28 #include <linux/slab.h>
29 #include <linux/vfs.h>
30 #include <linux/mutex.h>
31 #include <uapi/linux/cramfs_fs.h>
32 #include <linux/uaccess.h>
33 
34 #include "internal.h"
35 
36 /*
37  * cramfs super-block data in memory
38  */
39 struct cramfs_sb_info {
40 	unsigned long magic;
41 	unsigned long size;
42 	unsigned long blocks;
43 	unsigned long files;
44 	unsigned long flags;
45 	void *linear_virt_addr;
46 	resource_size_t linear_phys_addr;
47 	size_t mtd_point_size;
48 };
49 
50 static inline struct cramfs_sb_info *CRAMFS_SB(struct super_block *sb)
51 {
52 	return sb->s_fs_info;
53 }
54 
55 static const struct super_operations cramfs_ops;
56 static const struct inode_operations cramfs_dir_inode_operations;
57 static const struct file_operations cramfs_directory_operations;
58 static const struct file_operations cramfs_physmem_fops;
59 static const struct address_space_operations cramfs_aops;
60 
61 static DEFINE_MUTEX(read_mutex);
62 
63 
64 /* These macros may change in future, to provide better st_ino semantics. */
65 #define OFFSET(x)	((x)->i_ino)
66 
67 static unsigned long cramino(const struct cramfs_inode *cino, unsigned int offset)
68 {
69 	if (!cino->offset)
70 		return offset + 1;
71 	if (!cino->size)
72 		return offset + 1;
73 
74 	/*
75 	 * The file mode test fixes buggy mkcramfs implementations where
76 	 * cramfs_inode->offset is set to a non zero value for entries
77 	 * which did not contain data, like devices node and fifos.
78 	 */
79 	switch (cino->mode & S_IFMT) {
80 	case S_IFREG:
81 	case S_IFDIR:
82 	case S_IFLNK:
83 		return cino->offset << 2;
84 	default:
85 		break;
86 	}
87 	return offset + 1;
88 }
89 
90 static struct inode *get_cramfs_inode(struct super_block *sb,
91 	const struct cramfs_inode *cramfs_inode, unsigned int offset)
92 {
93 	struct inode *inode;
94 	static struct timespec64 zerotime;
95 
96 	inode = iget_locked(sb, cramino(cramfs_inode, offset));
97 	if (!inode)
98 		return ERR_PTR(-ENOMEM);
99 	if (!(inode->i_state & I_NEW))
100 		return inode;
101 
102 	switch (cramfs_inode->mode & S_IFMT) {
103 	case S_IFREG:
104 		inode->i_fop = &generic_ro_fops;
105 		inode->i_data.a_ops = &cramfs_aops;
106 		if (IS_ENABLED(CONFIG_CRAMFS_MTD) &&
107 		    CRAMFS_SB(sb)->flags & CRAMFS_FLAG_EXT_BLOCK_POINTERS &&
108 		    CRAMFS_SB(sb)->linear_phys_addr)
109 			inode->i_fop = &cramfs_physmem_fops;
110 		break;
111 	case S_IFDIR:
112 		inode->i_op = &cramfs_dir_inode_operations;
113 		inode->i_fop = &cramfs_directory_operations;
114 		break;
115 	case S_IFLNK:
116 		inode->i_op = &page_symlink_inode_operations;
117 		inode_nohighmem(inode);
118 		inode->i_data.a_ops = &cramfs_aops;
119 		break;
120 	default:
121 		init_special_inode(inode, cramfs_inode->mode,
122 				old_decode_dev(cramfs_inode->size));
123 	}
124 
125 	inode->i_mode = cramfs_inode->mode;
126 	i_uid_write(inode, cramfs_inode->uid);
127 	i_gid_write(inode, cramfs_inode->gid);
128 
129 	/* if the lower 2 bits are zero, the inode contains data */
130 	if (!(inode->i_ino & 3)) {
131 		inode->i_size = cramfs_inode->size;
132 		inode->i_blocks = (cramfs_inode->size - 1) / 512 + 1;
133 	}
134 
135 	/* Struct copy intentional */
136 	inode->i_mtime = inode->i_atime = inode->i_ctime = zerotime;
137 	/* inode->i_nlink is left 1 - arguably wrong for directories,
138 	   but it's the best we can do without reading the directory
139 	   contents.  1 yields the right result in GNU find, even
140 	   without -noleaf option. */
141 
142 	unlock_new_inode(inode);
143 
144 	return inode;
145 }
146 
147 /*
148  * We have our own block cache: don't fill up the buffer cache
149  * with the rom-image, because the way the filesystem is set
150  * up the accesses should be fairly regular and cached in the
151  * page cache and dentry tree anyway..
152  *
153  * This also acts as a way to guarantee contiguous areas of up to
154  * BLKS_PER_BUF*PAGE_SIZE, so that the caller doesn't need to
155  * worry about end-of-buffer issues even when decompressing a full
156  * page cache.
157  *
158  * Note: This is all optimized away at compile time when
159  *       CONFIG_CRAMFS_BLOCKDEV=n.
160  */
161 #define READ_BUFFERS (2)
162 /* NEXT_BUFFER(): Loop over [0..(READ_BUFFERS-1)]. */
163 #define NEXT_BUFFER(_ix) ((_ix) ^ 1)
164 
165 /*
166  * BLKS_PER_BUF_SHIFT should be at least 2 to allow for "compressed"
167  * data that takes up more space than the original and with unlucky
168  * alignment.
169  */
170 #define BLKS_PER_BUF_SHIFT	(2)
171 #define BLKS_PER_BUF		(1 << BLKS_PER_BUF_SHIFT)
172 #define BUFFER_SIZE		(BLKS_PER_BUF*PAGE_SIZE)
173 
174 static unsigned char read_buffers[READ_BUFFERS][BUFFER_SIZE];
175 static unsigned buffer_blocknr[READ_BUFFERS];
176 static struct super_block *buffer_dev[READ_BUFFERS];
177 static int next_buffer;
178 
179 /*
180  * Populate our block cache and return a pointer to it.
181  */
182 static void *cramfs_blkdev_read(struct super_block *sb, unsigned int offset,
183 				unsigned int len)
184 {
185 	struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
186 	struct file_ra_state ra;
187 	struct page *pages[BLKS_PER_BUF];
188 	unsigned i, blocknr, buffer;
189 	unsigned long devsize;
190 	char *data;
191 
192 	if (!len)
193 		return NULL;
194 	blocknr = offset >> PAGE_SHIFT;
195 	offset &= PAGE_SIZE - 1;
196 
197 	/* Check if an existing buffer already has the data.. */
198 	for (i = 0; i < READ_BUFFERS; i++) {
199 		unsigned int blk_offset;
200 
201 		if (buffer_dev[i] != sb)
202 			continue;
203 		if (blocknr < buffer_blocknr[i])
204 			continue;
205 		blk_offset = (blocknr - buffer_blocknr[i]) << PAGE_SHIFT;
206 		blk_offset += offset;
207 		if (blk_offset > BUFFER_SIZE ||
208 		    blk_offset + len > BUFFER_SIZE)
209 			continue;
210 		return read_buffers[i] + blk_offset;
211 	}
212 
213 	devsize = bdev_nr_bytes(sb->s_bdev) >> PAGE_SHIFT;
214 
215 	/* Ok, read in BLKS_PER_BUF pages completely first. */
216 	file_ra_state_init(&ra, mapping);
217 	page_cache_sync_readahead(mapping, &ra, NULL, blocknr, BLKS_PER_BUF);
218 
219 	for (i = 0; i < BLKS_PER_BUF; i++) {
220 		struct page *page = NULL;
221 
222 		if (blocknr + i < devsize) {
223 			page = read_mapping_page(mapping, blocknr + i, NULL);
224 			/* synchronous error? */
225 			if (IS_ERR(page))
226 				page = NULL;
227 		}
228 		pages[i] = page;
229 	}
230 
231 	buffer = next_buffer;
232 	next_buffer = NEXT_BUFFER(buffer);
233 	buffer_blocknr[buffer] = blocknr;
234 	buffer_dev[buffer] = sb;
235 
236 	data = read_buffers[buffer];
237 	for (i = 0; i < BLKS_PER_BUF; i++) {
238 		struct page *page = pages[i];
239 
240 		if (page) {
241 			memcpy(data, kmap(page), PAGE_SIZE);
242 			kunmap(page);
243 			put_page(page);
244 		} else
245 			memset(data, 0, PAGE_SIZE);
246 		data += PAGE_SIZE;
247 	}
248 	return read_buffers[buffer] + offset;
249 }
250 
251 /*
252  * Return a pointer to the linearly addressed cramfs image in memory.
253  */
254 static void *cramfs_direct_read(struct super_block *sb, unsigned int offset,
255 				unsigned int len)
256 {
257 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
258 
259 	if (!len)
260 		return NULL;
261 	if (len > sbi->size || offset > sbi->size - len)
262 		return page_address(ZERO_PAGE(0));
263 	return sbi->linear_virt_addr + offset;
264 }
265 
266 /*
267  * Returns a pointer to a buffer containing at least LEN bytes of
268  * filesystem starting at byte offset OFFSET into the filesystem.
269  */
270 static void *cramfs_read(struct super_block *sb, unsigned int offset,
271 			 unsigned int len)
272 {
273 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
274 
275 	if (IS_ENABLED(CONFIG_CRAMFS_MTD) && sbi->linear_virt_addr)
276 		return cramfs_direct_read(sb, offset, len);
277 	else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV))
278 		return cramfs_blkdev_read(sb, offset, len);
279 	else
280 		return NULL;
281 }
282 
283 /*
284  * For a mapping to be possible, we need a range of uncompressed and
285  * contiguous blocks. Return the offset for the first block and number of
286  * valid blocks for which that is true, or zero otherwise.
287  */
288 static u32 cramfs_get_block_range(struct inode *inode, u32 pgoff, u32 *pages)
289 {
290 	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
291 	int i;
292 	u32 *blockptrs, first_block_addr;
293 
294 	/*
295 	 * We can dereference memory directly here as this code may be
296 	 * reached only when there is a direct filesystem image mapping
297 	 * available in memory.
298 	 */
299 	blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode) + pgoff * 4);
300 	first_block_addr = blockptrs[0] & ~CRAMFS_BLK_FLAGS;
301 	i = 0;
302 	do {
303 		u32 block_off = i * (PAGE_SIZE >> CRAMFS_BLK_DIRECT_PTR_SHIFT);
304 		u32 expect = (first_block_addr + block_off) |
305 			     CRAMFS_BLK_FLAG_DIRECT_PTR |
306 			     CRAMFS_BLK_FLAG_UNCOMPRESSED;
307 		if (blockptrs[i] != expect) {
308 			pr_debug("range: block %d/%d got %#x expects %#x\n",
309 				 pgoff+i, pgoff + *pages - 1,
310 				 blockptrs[i], expect);
311 			if (i == 0)
312 				return 0;
313 			break;
314 		}
315 	} while (++i < *pages);
316 
317 	*pages = i;
318 	return first_block_addr << CRAMFS_BLK_DIRECT_PTR_SHIFT;
319 }
320 
321 #ifdef CONFIG_MMU
322 
323 /*
324  * Return true if the last page of a file in the filesystem image contains
325  * some other data that doesn't belong to that file. It is assumed that the
326  * last block is CRAMFS_BLK_FLAG_DIRECT_PTR | CRAMFS_BLK_FLAG_UNCOMPRESSED
327  * (verified by cramfs_get_block_range() and directly accessible in memory.
328  */
329 static bool cramfs_last_page_is_shared(struct inode *inode)
330 {
331 	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
332 	u32 partial, last_page, blockaddr, *blockptrs;
333 	char *tail_data;
334 
335 	partial = offset_in_page(inode->i_size);
336 	if (!partial)
337 		return false;
338 	last_page = inode->i_size >> PAGE_SHIFT;
339 	blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode));
340 	blockaddr = blockptrs[last_page] & ~CRAMFS_BLK_FLAGS;
341 	blockaddr <<= CRAMFS_BLK_DIRECT_PTR_SHIFT;
342 	tail_data = sbi->linear_virt_addr + blockaddr + partial;
343 	return memchr_inv(tail_data, 0, PAGE_SIZE - partial) ? true : false;
344 }
345 
346 static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct *vma)
347 {
348 	struct inode *inode = file_inode(file);
349 	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
350 	unsigned int pages, max_pages, offset;
351 	unsigned long address, pgoff = vma->vm_pgoff;
352 	char *bailout_reason;
353 	int ret;
354 
355 	ret = generic_file_readonly_mmap(file, vma);
356 	if (ret)
357 		return ret;
358 
359 	/*
360 	 * Now try to pre-populate ptes for this vma with a direct
361 	 * mapping avoiding memory allocation when possible.
362 	 */
363 
364 	/* Could COW work here? */
365 	bailout_reason = "vma is writable";
366 	if (vma->vm_flags & VM_WRITE)
367 		goto bailout;
368 
369 	max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
370 	bailout_reason = "beyond file limit";
371 	if (pgoff >= max_pages)
372 		goto bailout;
373 	pages = min(vma_pages(vma), max_pages - pgoff);
374 
375 	offset = cramfs_get_block_range(inode, pgoff, &pages);
376 	bailout_reason = "unsuitable block layout";
377 	if (!offset)
378 		goto bailout;
379 	address = sbi->linear_phys_addr + offset;
380 	bailout_reason = "data is not page aligned";
381 	if (!PAGE_ALIGNED(address))
382 		goto bailout;
383 
384 	/* Don't map the last page if it contains some other data */
385 	if (pgoff + pages == max_pages && cramfs_last_page_is_shared(inode)) {
386 		pr_debug("mmap: %pD: last page is shared\n", file);
387 		pages--;
388 	}
389 
390 	if (!pages) {
391 		bailout_reason = "no suitable block remaining";
392 		goto bailout;
393 	}
394 
395 	if (pages == vma_pages(vma)) {
396 		/*
397 		 * The entire vma is mappable. remap_pfn_range() will
398 		 * make it distinguishable from a non-direct mapping
399 		 * in /proc/<pid>/maps by substituting the file offset
400 		 * with the actual physical address.
401 		 */
402 		ret = remap_pfn_range(vma, vma->vm_start, address >> PAGE_SHIFT,
403 				      pages * PAGE_SIZE, vma->vm_page_prot);
404 	} else {
405 		/*
406 		 * Let's create a mixed map if we can't map it all.
407 		 * The normal paging machinery will take care of the
408 		 * unpopulated ptes via cramfs_read_folio().
409 		 */
410 		int i;
411 		vma->vm_flags |= VM_MIXEDMAP;
412 		for (i = 0; i < pages && !ret; i++) {
413 			vm_fault_t vmf;
414 			unsigned long off = i * PAGE_SIZE;
415 			pfn_t pfn = phys_to_pfn_t(address + off, PFN_DEV);
416 			vmf = vmf_insert_mixed(vma, vma->vm_start + off, pfn);
417 			if (vmf & VM_FAULT_ERROR)
418 				ret = vm_fault_to_errno(vmf, 0);
419 		}
420 	}
421 
422 	if (!ret)
423 		pr_debug("mapped %pD[%lu] at 0x%08lx (%u/%lu pages) "
424 			 "to vma 0x%08lx, page_prot 0x%llx\n", file,
425 			 pgoff, address, pages, vma_pages(vma), vma->vm_start,
426 			 (unsigned long long)pgprot_val(vma->vm_page_prot));
427 	return ret;
428 
429 bailout:
430 	pr_debug("%pD[%lu]: direct mmap impossible: %s\n",
431 		 file, pgoff, bailout_reason);
432 	/* Didn't manage any direct map, but normal paging is still possible */
433 	return 0;
434 }
435 
436 #else /* CONFIG_MMU */
437 
438 static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct *vma)
439 {
440 	return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -ENOSYS;
441 }
442 
443 static unsigned long cramfs_physmem_get_unmapped_area(struct file *file,
444 			unsigned long addr, unsigned long len,
445 			unsigned long pgoff, unsigned long flags)
446 {
447 	struct inode *inode = file_inode(file);
448 	struct super_block *sb = inode->i_sb;
449 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
450 	unsigned int pages, block_pages, max_pages, offset;
451 
452 	pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
453 	max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
454 	if (pgoff >= max_pages || pages > max_pages - pgoff)
455 		return -EINVAL;
456 	block_pages = pages;
457 	offset = cramfs_get_block_range(inode, pgoff, &block_pages);
458 	if (!offset || block_pages != pages)
459 		return -ENOSYS;
460 	addr = sbi->linear_phys_addr + offset;
461 	pr_debug("get_unmapped for %pD ofs %#lx siz %lu at 0x%08lx\n",
462 		 file, pgoff*PAGE_SIZE, len, addr);
463 	return addr;
464 }
465 
466 static unsigned int cramfs_physmem_mmap_capabilities(struct file *file)
467 {
468 	return NOMMU_MAP_COPY | NOMMU_MAP_DIRECT |
469 	       NOMMU_MAP_READ | NOMMU_MAP_EXEC;
470 }
471 
472 #endif /* CONFIG_MMU */
473 
474 static const struct file_operations cramfs_physmem_fops = {
475 	.llseek			= generic_file_llseek,
476 	.read_iter		= generic_file_read_iter,
477 	.splice_read		= generic_file_splice_read,
478 	.mmap			= cramfs_physmem_mmap,
479 #ifndef CONFIG_MMU
480 	.get_unmapped_area	= cramfs_physmem_get_unmapped_area,
481 	.mmap_capabilities	= cramfs_physmem_mmap_capabilities,
482 #endif
483 };
484 
485 static void cramfs_kill_sb(struct super_block *sb)
486 {
487 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
488 
489 	if (IS_ENABLED(CONFIG_CRAMFS_MTD) && sb->s_mtd) {
490 		if (sbi && sbi->mtd_point_size)
491 			mtd_unpoint(sb->s_mtd, 0, sbi->mtd_point_size);
492 		kill_mtd_super(sb);
493 	} else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV) && sb->s_bdev) {
494 		kill_block_super(sb);
495 	}
496 	kfree(sbi);
497 }
498 
499 static int cramfs_reconfigure(struct fs_context *fc)
500 {
501 	sync_filesystem(fc->root->d_sb);
502 	fc->sb_flags |= SB_RDONLY;
503 	return 0;
504 }
505 
506 static int cramfs_read_super(struct super_block *sb, struct fs_context *fc,
507 			     struct cramfs_super *super)
508 {
509 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
510 	unsigned long root_offset;
511 	bool silent = fc->sb_flags & SB_SILENT;
512 
513 	/* We don't know the real size yet */
514 	sbi->size = PAGE_SIZE;
515 
516 	/* Read the first block and get the superblock from it */
517 	mutex_lock(&read_mutex);
518 	memcpy(super, cramfs_read(sb, 0, sizeof(*super)), sizeof(*super));
519 	mutex_unlock(&read_mutex);
520 
521 	/* Do sanity checks on the superblock */
522 	if (super->magic != CRAMFS_MAGIC) {
523 		/* check for wrong endianness */
524 		if (super->magic == CRAMFS_MAGIC_WEND) {
525 			if (!silent)
526 				errorfc(fc, "wrong endianness");
527 			return -EINVAL;
528 		}
529 
530 		/* check at 512 byte offset */
531 		mutex_lock(&read_mutex);
532 		memcpy(super,
533 		       cramfs_read(sb, 512, sizeof(*super)),
534 		       sizeof(*super));
535 		mutex_unlock(&read_mutex);
536 		if (super->magic != CRAMFS_MAGIC) {
537 			if (super->magic == CRAMFS_MAGIC_WEND && !silent)
538 				errorfc(fc, "wrong endianness");
539 			else if (!silent)
540 				errorfc(fc, "wrong magic");
541 			return -EINVAL;
542 		}
543 	}
544 
545 	/* get feature flags first */
546 	if (super->flags & ~CRAMFS_SUPPORTED_FLAGS) {
547 		errorfc(fc, "unsupported filesystem features");
548 		return -EINVAL;
549 	}
550 
551 	/* Check that the root inode is in a sane state */
552 	if (!S_ISDIR(super->root.mode)) {
553 		errorfc(fc, "root is not a directory");
554 		return -EINVAL;
555 	}
556 	/* correct strange, hard-coded permissions of mkcramfs */
557 	super->root.mode |= 0555;
558 
559 	root_offset = super->root.offset << 2;
560 	if (super->flags & CRAMFS_FLAG_FSID_VERSION_2) {
561 		sbi->size = super->size;
562 		sbi->blocks = super->fsid.blocks;
563 		sbi->files = super->fsid.files;
564 	} else {
565 		sbi->size = 1<<28;
566 		sbi->blocks = 0;
567 		sbi->files = 0;
568 	}
569 	sbi->magic = super->magic;
570 	sbi->flags = super->flags;
571 	if (root_offset == 0)
572 		infofc(fc, "empty filesystem");
573 	else if (!(super->flags & CRAMFS_FLAG_SHIFTED_ROOT_OFFSET) &&
574 		 ((root_offset != sizeof(struct cramfs_super)) &&
575 		  (root_offset != 512 + sizeof(struct cramfs_super))))
576 	{
577 		errorfc(fc, "bad root offset %lu", root_offset);
578 		return -EINVAL;
579 	}
580 
581 	return 0;
582 }
583 
584 static int cramfs_finalize_super(struct super_block *sb,
585 				 struct cramfs_inode *cramfs_root)
586 {
587 	struct inode *root;
588 
589 	/* Set it all up.. */
590 	sb->s_flags |= SB_RDONLY;
591 	sb->s_time_min = 0;
592 	sb->s_time_max = 0;
593 	sb->s_op = &cramfs_ops;
594 	root = get_cramfs_inode(sb, cramfs_root, 0);
595 	if (IS_ERR(root))
596 		return PTR_ERR(root);
597 	sb->s_root = d_make_root(root);
598 	if (!sb->s_root)
599 		return -ENOMEM;
600 	return 0;
601 }
602 
603 static int cramfs_blkdev_fill_super(struct super_block *sb, struct fs_context *fc)
604 {
605 	struct cramfs_sb_info *sbi;
606 	struct cramfs_super super;
607 	int i, err;
608 
609 	sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
610 	if (!sbi)
611 		return -ENOMEM;
612 	sb->s_fs_info = sbi;
613 
614 	/* Invalidate the read buffers on mount: think disk change.. */
615 	for (i = 0; i < READ_BUFFERS; i++)
616 		buffer_blocknr[i] = -1;
617 
618 	err = cramfs_read_super(sb, fc, &super);
619 	if (err)
620 		return err;
621 	return cramfs_finalize_super(sb, &super.root);
622 }
623 
624 static int cramfs_mtd_fill_super(struct super_block *sb, struct fs_context *fc)
625 {
626 	struct cramfs_sb_info *sbi;
627 	struct cramfs_super super;
628 	int err;
629 
630 	sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
631 	if (!sbi)
632 		return -ENOMEM;
633 	sb->s_fs_info = sbi;
634 
635 	/* Map only one page for now.  Will remap it when fs size is known. */
636 	err = mtd_point(sb->s_mtd, 0, PAGE_SIZE, &sbi->mtd_point_size,
637 			&sbi->linear_virt_addr, &sbi->linear_phys_addr);
638 	if (err || sbi->mtd_point_size != PAGE_SIZE) {
639 		pr_err("unable to get direct memory access to mtd:%s\n",
640 		       sb->s_mtd->name);
641 		return err ? : -ENODATA;
642 	}
643 
644 	pr_info("checking physical address %pap for linear cramfs image\n",
645 		&sbi->linear_phys_addr);
646 	err = cramfs_read_super(sb, fc, &super);
647 	if (err)
648 		return err;
649 
650 	/* Remap the whole filesystem now */
651 	pr_info("linear cramfs image on mtd:%s appears to be %lu KB in size\n",
652 		sb->s_mtd->name, sbi->size/1024);
653 	mtd_unpoint(sb->s_mtd, 0, PAGE_SIZE);
654 	err = mtd_point(sb->s_mtd, 0, sbi->size, &sbi->mtd_point_size,
655 			&sbi->linear_virt_addr, &sbi->linear_phys_addr);
656 	if (err || sbi->mtd_point_size != sbi->size) {
657 		pr_err("unable to get direct memory access to mtd:%s\n",
658 		       sb->s_mtd->name);
659 		return err ? : -ENODATA;
660 	}
661 
662 	return cramfs_finalize_super(sb, &super.root);
663 }
664 
665 static int cramfs_statfs(struct dentry *dentry, struct kstatfs *buf)
666 {
667 	struct super_block *sb = dentry->d_sb;
668 	u64 id = 0;
669 
670 	if (sb->s_bdev)
671 		id = huge_encode_dev(sb->s_bdev->bd_dev);
672 	else if (sb->s_dev)
673 		id = huge_encode_dev(sb->s_dev);
674 
675 	buf->f_type = CRAMFS_MAGIC;
676 	buf->f_bsize = PAGE_SIZE;
677 	buf->f_blocks = CRAMFS_SB(sb)->blocks;
678 	buf->f_bfree = 0;
679 	buf->f_bavail = 0;
680 	buf->f_files = CRAMFS_SB(sb)->files;
681 	buf->f_ffree = 0;
682 	buf->f_fsid = u64_to_fsid(id);
683 	buf->f_namelen = CRAMFS_MAXPATHLEN;
684 	return 0;
685 }
686 
687 /*
688  * Read a cramfs directory entry.
689  */
690 static int cramfs_readdir(struct file *file, struct dir_context *ctx)
691 {
692 	struct inode *inode = file_inode(file);
693 	struct super_block *sb = inode->i_sb;
694 	char *buf;
695 	unsigned int offset;
696 
697 	/* Offset within the thing. */
698 	if (ctx->pos >= inode->i_size)
699 		return 0;
700 	offset = ctx->pos;
701 	/* Directory entries are always 4-byte aligned */
702 	if (offset & 3)
703 		return -EINVAL;
704 
705 	buf = kmalloc(CRAMFS_MAXPATHLEN, GFP_KERNEL);
706 	if (!buf)
707 		return -ENOMEM;
708 
709 	while (offset < inode->i_size) {
710 		struct cramfs_inode *de;
711 		unsigned long nextoffset;
712 		char *name;
713 		ino_t ino;
714 		umode_t mode;
715 		int namelen;
716 
717 		mutex_lock(&read_mutex);
718 		de = cramfs_read(sb, OFFSET(inode) + offset, sizeof(*de)+CRAMFS_MAXPATHLEN);
719 		name = (char *)(de+1);
720 
721 		/*
722 		 * Namelengths on disk are shifted by two
723 		 * and the name padded out to 4-byte boundaries
724 		 * with zeroes.
725 		 */
726 		namelen = de->namelen << 2;
727 		memcpy(buf, name, namelen);
728 		ino = cramino(de, OFFSET(inode) + offset);
729 		mode = de->mode;
730 		mutex_unlock(&read_mutex);
731 		nextoffset = offset + sizeof(*de) + namelen;
732 		for (;;) {
733 			if (!namelen) {
734 				kfree(buf);
735 				return -EIO;
736 			}
737 			if (buf[namelen-1])
738 				break;
739 			namelen--;
740 		}
741 		if (!dir_emit(ctx, buf, namelen, ino, mode >> 12))
742 			break;
743 
744 		ctx->pos = offset = nextoffset;
745 	}
746 	kfree(buf);
747 	return 0;
748 }
749 
750 /*
751  * Lookup and fill in the inode data..
752  */
753 static struct dentry *cramfs_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
754 {
755 	unsigned int offset = 0;
756 	struct inode *inode = NULL;
757 	int sorted;
758 
759 	mutex_lock(&read_mutex);
760 	sorted = CRAMFS_SB(dir->i_sb)->flags & CRAMFS_FLAG_SORTED_DIRS;
761 	while (offset < dir->i_size) {
762 		struct cramfs_inode *de;
763 		char *name;
764 		int namelen, retval;
765 		int dir_off = OFFSET(dir) + offset;
766 
767 		de = cramfs_read(dir->i_sb, dir_off, sizeof(*de)+CRAMFS_MAXPATHLEN);
768 		name = (char *)(de+1);
769 
770 		/* Try to take advantage of sorted directories */
771 		if (sorted && (dentry->d_name.name[0] < name[0]))
772 			break;
773 
774 		namelen = de->namelen << 2;
775 		offset += sizeof(*de) + namelen;
776 
777 		/* Quick check that the name is roughly the right length */
778 		if (((dentry->d_name.len + 3) & ~3) != namelen)
779 			continue;
780 
781 		for (;;) {
782 			if (!namelen) {
783 				inode = ERR_PTR(-EIO);
784 				goto out;
785 			}
786 			if (name[namelen-1])
787 				break;
788 			namelen--;
789 		}
790 		if (namelen != dentry->d_name.len)
791 			continue;
792 		retval = memcmp(dentry->d_name.name, name, namelen);
793 		if (retval > 0)
794 			continue;
795 		if (!retval) {
796 			inode = get_cramfs_inode(dir->i_sb, de, dir_off);
797 			break;
798 		}
799 		/* else (retval < 0) */
800 		if (sorted)
801 			break;
802 	}
803 out:
804 	mutex_unlock(&read_mutex);
805 	return d_splice_alias(inode, dentry);
806 }
807 
808 static int cramfs_read_folio(struct file *file, struct folio *folio)
809 {
810 	struct page *page = &folio->page;
811 	struct inode *inode = page->mapping->host;
812 	u32 maxblock;
813 	int bytes_filled;
814 	void *pgdata;
815 
816 	maxblock = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
817 	bytes_filled = 0;
818 	pgdata = kmap(page);
819 
820 	if (page->index < maxblock) {
821 		struct super_block *sb = inode->i_sb;
822 		u32 blkptr_offset = OFFSET(inode) + page->index * 4;
823 		u32 block_ptr, block_start, block_len;
824 		bool uncompressed, direct;
825 
826 		mutex_lock(&read_mutex);
827 		block_ptr = *(u32 *) cramfs_read(sb, blkptr_offset, 4);
828 		uncompressed = (block_ptr & CRAMFS_BLK_FLAG_UNCOMPRESSED);
829 		direct = (block_ptr & CRAMFS_BLK_FLAG_DIRECT_PTR);
830 		block_ptr &= ~CRAMFS_BLK_FLAGS;
831 
832 		if (direct) {
833 			/*
834 			 * The block pointer is an absolute start pointer,
835 			 * shifted by 2 bits. The size is included in the
836 			 * first 2 bytes of the data block when compressed,
837 			 * or PAGE_SIZE otherwise.
838 			 */
839 			block_start = block_ptr << CRAMFS_BLK_DIRECT_PTR_SHIFT;
840 			if (uncompressed) {
841 				block_len = PAGE_SIZE;
842 				/* if last block: cap to file length */
843 				if (page->index == maxblock - 1)
844 					block_len =
845 						offset_in_page(inode->i_size);
846 			} else {
847 				block_len = *(u16 *)
848 					cramfs_read(sb, block_start, 2);
849 				block_start += 2;
850 			}
851 		} else {
852 			/*
853 			 * The block pointer indicates one past the end of
854 			 * the current block (start of next block). If this
855 			 * is the first block then it starts where the block
856 			 * pointer table ends, otherwise its start comes
857 			 * from the previous block's pointer.
858 			 */
859 			block_start = OFFSET(inode) + maxblock * 4;
860 			if (page->index)
861 				block_start = *(u32 *)
862 					cramfs_read(sb, blkptr_offset - 4, 4);
863 			/* Beware... previous ptr might be a direct ptr */
864 			if (unlikely(block_start & CRAMFS_BLK_FLAG_DIRECT_PTR)) {
865 				/* See comments on earlier code. */
866 				u32 prev_start = block_start;
867 				block_start = prev_start & ~CRAMFS_BLK_FLAGS;
868 				block_start <<= CRAMFS_BLK_DIRECT_PTR_SHIFT;
869 				if (prev_start & CRAMFS_BLK_FLAG_UNCOMPRESSED) {
870 					block_start += PAGE_SIZE;
871 				} else {
872 					block_len = *(u16 *)
873 						cramfs_read(sb, block_start, 2);
874 					block_start += 2 + block_len;
875 				}
876 			}
877 			block_start &= ~CRAMFS_BLK_FLAGS;
878 			block_len = block_ptr - block_start;
879 		}
880 
881 		if (block_len == 0)
882 			; /* hole */
883 		else if (unlikely(block_len > 2*PAGE_SIZE ||
884 				  (uncompressed && block_len > PAGE_SIZE))) {
885 			mutex_unlock(&read_mutex);
886 			pr_err("bad data blocksize %u\n", block_len);
887 			goto err;
888 		} else if (uncompressed) {
889 			memcpy(pgdata,
890 			       cramfs_read(sb, block_start, block_len),
891 			       block_len);
892 			bytes_filled = block_len;
893 		} else {
894 			bytes_filled = cramfs_uncompress_block(pgdata,
895 				 PAGE_SIZE,
896 				 cramfs_read(sb, block_start, block_len),
897 				 block_len);
898 		}
899 		mutex_unlock(&read_mutex);
900 		if (unlikely(bytes_filled < 0))
901 			goto err;
902 	}
903 
904 	memset(pgdata + bytes_filled, 0, PAGE_SIZE - bytes_filled);
905 	flush_dcache_page(page);
906 	kunmap(page);
907 	SetPageUptodate(page);
908 	unlock_page(page);
909 	return 0;
910 
911 err:
912 	kunmap(page);
913 	ClearPageUptodate(page);
914 	SetPageError(page);
915 	unlock_page(page);
916 	return 0;
917 }
918 
919 static const struct address_space_operations cramfs_aops = {
920 	.read_folio = cramfs_read_folio
921 };
922 
923 /*
924  * Our operations:
925  */
926 
927 /*
928  * A directory can only readdir
929  */
930 static const struct file_operations cramfs_directory_operations = {
931 	.llseek		= generic_file_llseek,
932 	.read		= generic_read_dir,
933 	.iterate_shared	= cramfs_readdir,
934 };
935 
936 static const struct inode_operations cramfs_dir_inode_operations = {
937 	.lookup		= cramfs_lookup,
938 };
939 
940 static const struct super_operations cramfs_ops = {
941 	.statfs		= cramfs_statfs,
942 };
943 
944 static int cramfs_get_tree(struct fs_context *fc)
945 {
946 	int ret = -ENOPROTOOPT;
947 
948 	if (IS_ENABLED(CONFIG_CRAMFS_MTD)) {
949 		ret = get_tree_mtd(fc, cramfs_mtd_fill_super);
950 		if (!ret)
951 			return 0;
952 	}
953 	if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV))
954 		ret = get_tree_bdev(fc, cramfs_blkdev_fill_super);
955 	return ret;
956 }
957 
958 static const struct fs_context_operations cramfs_context_ops = {
959 	.get_tree	= cramfs_get_tree,
960 	.reconfigure	= cramfs_reconfigure,
961 };
962 
963 /*
964  * Set up the filesystem mount context.
965  */
966 static int cramfs_init_fs_context(struct fs_context *fc)
967 {
968 	fc->ops = &cramfs_context_ops;
969 	return 0;
970 }
971 
972 static struct file_system_type cramfs_fs_type = {
973 	.owner		= THIS_MODULE,
974 	.name		= "cramfs",
975 	.init_fs_context = cramfs_init_fs_context,
976 	.kill_sb	= cramfs_kill_sb,
977 	.fs_flags	= FS_REQUIRES_DEV,
978 };
979 MODULE_ALIAS_FS("cramfs");
980 
981 static int __init init_cramfs_fs(void)
982 {
983 	int rv;
984 
985 	rv = cramfs_uncompress_init();
986 	if (rv < 0)
987 		return rv;
988 	rv = register_filesystem(&cramfs_fs_type);
989 	if (rv < 0)
990 		cramfs_uncompress_exit();
991 	return rv;
992 }
993 
994 static void __exit exit_cramfs_fs(void)
995 {
996 	cramfs_uncompress_exit();
997 	unregister_filesystem(&cramfs_fs_type);
998 }
999 
1000 module_init(init_cramfs_fs)
1001 module_exit(exit_cramfs_fs)
1002 MODULE_LICENSE("GPL");
1003