xref: /linux/fs/cramfs/inode.c (revision 987b741c52c7c6c68d46fbaeb95b8d1087f10b7f)
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 page *pages[BLKS_PER_BUF];
187 	unsigned i, blocknr, buffer;
188 	unsigned long devsize;
189 	char *data;
190 
191 	if (!len)
192 		return NULL;
193 	blocknr = offset >> PAGE_SHIFT;
194 	offset &= PAGE_SIZE - 1;
195 
196 	/* Check if an existing buffer already has the data.. */
197 	for (i = 0; i < READ_BUFFERS; i++) {
198 		unsigned int blk_offset;
199 
200 		if (buffer_dev[i] != sb)
201 			continue;
202 		if (blocknr < buffer_blocknr[i])
203 			continue;
204 		blk_offset = (blocknr - buffer_blocknr[i]) << PAGE_SHIFT;
205 		blk_offset += offset;
206 		if (blk_offset > BUFFER_SIZE ||
207 		    blk_offset + len > BUFFER_SIZE)
208 			continue;
209 		return read_buffers[i] + blk_offset;
210 	}
211 
212 	devsize = mapping->host->i_size >> PAGE_SHIFT;
213 
214 	/* Ok, read in BLKS_PER_BUF pages completely first. */
215 	for (i = 0; i < BLKS_PER_BUF; i++) {
216 		struct page *page = NULL;
217 
218 		if (blocknr + i < devsize) {
219 			page = read_mapping_page(mapping, blocknr + i, NULL);
220 			/* synchronous error? */
221 			if (IS_ERR(page))
222 				page = NULL;
223 		}
224 		pages[i] = page;
225 	}
226 
227 	for (i = 0; i < BLKS_PER_BUF; i++) {
228 		struct page *page = pages[i];
229 
230 		if (page) {
231 			wait_on_page_locked(page);
232 			if (!PageUptodate(page)) {
233 				/* asynchronous error */
234 				put_page(page);
235 				pages[i] = NULL;
236 			}
237 		}
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(data, kmap(page), PAGE_SIZE);
251 			kunmap(page);
252 			put_page(page);
253 		} else
254 			memset(data, 0, PAGE_SIZE);
255 		data += PAGE_SIZE;
256 	}
257 	return read_buffers[buffer] + offset;
258 }
259 
260 /*
261  * Return a pointer to the linearly addressed cramfs image in memory.
262  */
263 static void *cramfs_direct_read(struct super_block *sb, unsigned int offset,
264 				unsigned int len)
265 {
266 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
267 
268 	if (!len)
269 		return NULL;
270 	if (len > sbi->size || offset > sbi->size - len)
271 		return page_address(ZERO_PAGE(0));
272 	return sbi->linear_virt_addr + offset;
273 }
274 
275 /*
276  * Returns a pointer to a buffer containing at least LEN bytes of
277  * filesystem starting at byte offset OFFSET into the filesystem.
278  */
279 static void *cramfs_read(struct super_block *sb, unsigned int offset,
280 			 unsigned int len)
281 {
282 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
283 
284 	if (IS_ENABLED(CONFIG_CRAMFS_MTD) && sbi->linear_virt_addr)
285 		return cramfs_direct_read(sb, offset, len);
286 	else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV))
287 		return cramfs_blkdev_read(sb, offset, len);
288 	else
289 		return NULL;
290 }
291 
292 /*
293  * For a mapping to be possible, we need a range of uncompressed and
294  * contiguous blocks. Return the offset for the first block and number of
295  * valid blocks for which that is true, or zero otherwise.
296  */
297 static u32 cramfs_get_block_range(struct inode *inode, u32 pgoff, u32 *pages)
298 {
299 	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
300 	int i;
301 	u32 *blockptrs, first_block_addr;
302 
303 	/*
304 	 * We can dereference memory directly here as this code may be
305 	 * reached only when there is a direct filesystem image mapping
306 	 * available in memory.
307 	 */
308 	blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode) + pgoff * 4);
309 	first_block_addr = blockptrs[0] & ~CRAMFS_BLK_FLAGS;
310 	i = 0;
311 	do {
312 		u32 block_off = i * (PAGE_SIZE >> CRAMFS_BLK_DIRECT_PTR_SHIFT);
313 		u32 expect = (first_block_addr + block_off) |
314 			     CRAMFS_BLK_FLAG_DIRECT_PTR |
315 			     CRAMFS_BLK_FLAG_UNCOMPRESSED;
316 		if (blockptrs[i] != expect) {
317 			pr_debug("range: block %d/%d got %#x expects %#x\n",
318 				 pgoff+i, pgoff + *pages - 1,
319 				 blockptrs[i], expect);
320 			if (i == 0)
321 				return 0;
322 			break;
323 		}
324 	} while (++i < *pages);
325 
326 	*pages = i;
327 	return first_block_addr << CRAMFS_BLK_DIRECT_PTR_SHIFT;
328 }
329 
330 #ifdef CONFIG_MMU
331 
332 /*
333  * Return true if the last page of a file in the filesystem image contains
334  * some other data that doesn't belong to that file. It is assumed that the
335  * last block is CRAMFS_BLK_FLAG_DIRECT_PTR | CRAMFS_BLK_FLAG_UNCOMPRESSED
336  * (verified by cramfs_get_block_range() and directly accessible in memory.
337  */
338 static bool cramfs_last_page_is_shared(struct inode *inode)
339 {
340 	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
341 	u32 partial, last_page, blockaddr, *blockptrs;
342 	char *tail_data;
343 
344 	partial = offset_in_page(inode->i_size);
345 	if (!partial)
346 		return false;
347 	last_page = inode->i_size >> PAGE_SHIFT;
348 	blockptrs = (u32 *)(sbi->linear_virt_addr + OFFSET(inode));
349 	blockaddr = blockptrs[last_page] & ~CRAMFS_BLK_FLAGS;
350 	blockaddr <<= CRAMFS_BLK_DIRECT_PTR_SHIFT;
351 	tail_data = sbi->linear_virt_addr + blockaddr + partial;
352 	return memchr_inv(tail_data, 0, PAGE_SIZE - partial) ? true : false;
353 }
354 
355 static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct *vma)
356 {
357 	struct inode *inode = file_inode(file);
358 	struct cramfs_sb_info *sbi = CRAMFS_SB(inode->i_sb);
359 	unsigned int pages, max_pages, offset;
360 	unsigned long address, pgoff = vma->vm_pgoff;
361 	char *bailout_reason;
362 	int ret;
363 
364 	ret = generic_file_readonly_mmap(file, vma);
365 	if (ret)
366 		return ret;
367 
368 	/*
369 	 * Now try to pre-populate ptes for this vma with a direct
370 	 * mapping avoiding memory allocation when possible.
371 	 */
372 
373 	/* Could COW work here? */
374 	bailout_reason = "vma is writable";
375 	if (vma->vm_flags & VM_WRITE)
376 		goto bailout;
377 
378 	max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
379 	bailout_reason = "beyond file limit";
380 	if (pgoff >= max_pages)
381 		goto bailout;
382 	pages = min(vma_pages(vma), max_pages - pgoff);
383 
384 	offset = cramfs_get_block_range(inode, pgoff, &pages);
385 	bailout_reason = "unsuitable block layout";
386 	if (!offset)
387 		goto bailout;
388 	address = sbi->linear_phys_addr + offset;
389 	bailout_reason = "data is not page aligned";
390 	if (!PAGE_ALIGNED(address))
391 		goto bailout;
392 
393 	/* Don't map the last page if it contains some other data */
394 	if (pgoff + pages == max_pages && cramfs_last_page_is_shared(inode)) {
395 		pr_debug("mmap: %pD: last page is shared\n", file);
396 		pages--;
397 	}
398 
399 	if (!pages) {
400 		bailout_reason = "no suitable block remaining";
401 		goto bailout;
402 	}
403 
404 	if (pages == vma_pages(vma)) {
405 		/*
406 		 * The entire vma is mappable. remap_pfn_range() will
407 		 * make it distinguishable from a non-direct mapping
408 		 * in /proc/<pid>/maps by substituting the file offset
409 		 * with the actual physical address.
410 		 */
411 		ret = remap_pfn_range(vma, vma->vm_start, address >> PAGE_SHIFT,
412 				      pages * PAGE_SIZE, vma->vm_page_prot);
413 	} else {
414 		/*
415 		 * Let's create a mixed map if we can't map it all.
416 		 * The normal paging machinery will take care of the
417 		 * unpopulated ptes via cramfs_readpage().
418 		 */
419 		int i;
420 		vma->vm_flags |= VM_MIXEDMAP;
421 		for (i = 0; i < pages && !ret; i++) {
422 			vm_fault_t vmf;
423 			unsigned long off = i * PAGE_SIZE;
424 			pfn_t pfn = phys_to_pfn_t(address + off, PFN_DEV);
425 			vmf = vmf_insert_mixed(vma, vma->vm_start + off, pfn);
426 			if (vmf & VM_FAULT_ERROR)
427 				ret = vm_fault_to_errno(vmf, 0);
428 		}
429 	}
430 
431 	if (!ret)
432 		pr_debug("mapped %pD[%lu] at 0x%08lx (%u/%lu pages) "
433 			 "to vma 0x%08lx, page_prot 0x%llx\n", file,
434 			 pgoff, address, pages, vma_pages(vma), vma->vm_start,
435 			 (unsigned long long)pgprot_val(vma->vm_page_prot));
436 	return ret;
437 
438 bailout:
439 	pr_debug("%pD[%lu]: direct mmap impossible: %s\n",
440 		 file, pgoff, bailout_reason);
441 	/* Didn't manage any direct map, but normal paging is still possible */
442 	return 0;
443 }
444 
445 #else /* CONFIG_MMU */
446 
447 static int cramfs_physmem_mmap(struct file *file, struct vm_area_struct *vma)
448 {
449 	return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -ENOSYS;
450 }
451 
452 static unsigned long cramfs_physmem_get_unmapped_area(struct file *file,
453 			unsigned long addr, unsigned long len,
454 			unsigned long pgoff, unsigned long flags)
455 {
456 	struct inode *inode = file_inode(file);
457 	struct super_block *sb = inode->i_sb;
458 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
459 	unsigned int pages, block_pages, max_pages, offset;
460 
461 	pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT;
462 	max_pages = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
463 	if (pgoff >= max_pages || pages > max_pages - pgoff)
464 		return -EINVAL;
465 	block_pages = pages;
466 	offset = cramfs_get_block_range(inode, pgoff, &block_pages);
467 	if (!offset || block_pages != pages)
468 		return -ENOSYS;
469 	addr = sbi->linear_phys_addr + offset;
470 	pr_debug("get_unmapped for %pD ofs %#lx siz %lu at 0x%08lx\n",
471 		 file, pgoff*PAGE_SIZE, len, addr);
472 	return addr;
473 }
474 
475 static unsigned int cramfs_physmem_mmap_capabilities(struct file *file)
476 {
477 	return NOMMU_MAP_COPY | NOMMU_MAP_DIRECT |
478 	       NOMMU_MAP_READ | NOMMU_MAP_EXEC;
479 }
480 
481 #endif /* CONFIG_MMU */
482 
483 static const struct file_operations cramfs_physmem_fops = {
484 	.llseek			= generic_file_llseek,
485 	.read_iter		= generic_file_read_iter,
486 	.splice_read		= generic_file_splice_read,
487 	.mmap			= cramfs_physmem_mmap,
488 #ifndef CONFIG_MMU
489 	.get_unmapped_area	= cramfs_physmem_get_unmapped_area,
490 	.mmap_capabilities	= cramfs_physmem_mmap_capabilities,
491 #endif
492 };
493 
494 static void cramfs_kill_sb(struct super_block *sb)
495 {
496 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
497 
498 	if (IS_ENABLED(CONFIG_CRAMFS_MTD) && sb->s_mtd) {
499 		if (sbi && sbi->mtd_point_size)
500 			mtd_unpoint(sb->s_mtd, 0, sbi->mtd_point_size);
501 		kill_mtd_super(sb);
502 	} else if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV) && sb->s_bdev) {
503 		kill_block_super(sb);
504 	}
505 	kfree(sbi);
506 }
507 
508 static int cramfs_reconfigure(struct fs_context *fc)
509 {
510 	sync_filesystem(fc->root->d_sb);
511 	fc->sb_flags |= SB_RDONLY;
512 	return 0;
513 }
514 
515 static int cramfs_read_super(struct super_block *sb, struct fs_context *fc,
516 			     struct cramfs_super *super)
517 {
518 	struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
519 	unsigned long root_offset;
520 	bool silent = fc->sb_flags & SB_SILENT;
521 
522 	/* We don't know the real size yet */
523 	sbi->size = PAGE_SIZE;
524 
525 	/* Read the first block and get the superblock from it */
526 	mutex_lock(&read_mutex);
527 	memcpy(super, cramfs_read(sb, 0, sizeof(*super)), sizeof(*super));
528 	mutex_unlock(&read_mutex);
529 
530 	/* Do sanity checks on the superblock */
531 	if (super->magic != CRAMFS_MAGIC) {
532 		/* check for wrong endianness */
533 		if (super->magic == CRAMFS_MAGIC_WEND) {
534 			if (!silent)
535 				errorfc(fc, "wrong endianness");
536 			return -EINVAL;
537 		}
538 
539 		/* check at 512 byte offset */
540 		mutex_lock(&read_mutex);
541 		memcpy(super,
542 		       cramfs_read(sb, 512, sizeof(*super)),
543 		       sizeof(*super));
544 		mutex_unlock(&read_mutex);
545 		if (super->magic != CRAMFS_MAGIC) {
546 			if (super->magic == CRAMFS_MAGIC_WEND && !silent)
547 				errorfc(fc, "wrong endianness");
548 			else if (!silent)
549 				errorfc(fc, "wrong magic");
550 			return -EINVAL;
551 		}
552 	}
553 
554 	/* get feature flags first */
555 	if (super->flags & ~CRAMFS_SUPPORTED_FLAGS) {
556 		errorfc(fc, "unsupported filesystem features");
557 		return -EINVAL;
558 	}
559 
560 	/* Check that the root inode is in a sane state */
561 	if (!S_ISDIR(super->root.mode)) {
562 		errorfc(fc, "root is not a directory");
563 		return -EINVAL;
564 	}
565 	/* correct strange, hard-coded permissions of mkcramfs */
566 	super->root.mode |= 0555;
567 
568 	root_offset = super->root.offset << 2;
569 	if (super->flags & CRAMFS_FLAG_FSID_VERSION_2) {
570 		sbi->size = super->size;
571 		sbi->blocks = super->fsid.blocks;
572 		sbi->files = super->fsid.files;
573 	} else {
574 		sbi->size = 1<<28;
575 		sbi->blocks = 0;
576 		sbi->files = 0;
577 	}
578 	sbi->magic = super->magic;
579 	sbi->flags = super->flags;
580 	if (root_offset == 0)
581 		infofc(fc, "empty filesystem");
582 	else if (!(super->flags & CRAMFS_FLAG_SHIFTED_ROOT_OFFSET) &&
583 		 ((root_offset != sizeof(struct cramfs_super)) &&
584 		  (root_offset != 512 + sizeof(struct cramfs_super))))
585 	{
586 		errorfc(fc, "bad root offset %lu", root_offset);
587 		return -EINVAL;
588 	}
589 
590 	return 0;
591 }
592 
593 static int cramfs_finalize_super(struct super_block *sb,
594 				 struct cramfs_inode *cramfs_root)
595 {
596 	struct inode *root;
597 
598 	/* Set it all up.. */
599 	sb->s_flags |= SB_RDONLY;
600 	sb->s_time_min = 0;
601 	sb->s_time_max = 0;
602 	sb->s_op = &cramfs_ops;
603 	root = get_cramfs_inode(sb, cramfs_root, 0);
604 	if (IS_ERR(root))
605 		return PTR_ERR(root);
606 	sb->s_root = d_make_root(root);
607 	if (!sb->s_root)
608 		return -ENOMEM;
609 	return 0;
610 }
611 
612 static int cramfs_blkdev_fill_super(struct super_block *sb, struct fs_context *fc)
613 {
614 	struct cramfs_sb_info *sbi;
615 	struct cramfs_super super;
616 	int i, err;
617 
618 	sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
619 	if (!sbi)
620 		return -ENOMEM;
621 	sb->s_fs_info = sbi;
622 
623 	/* Invalidate the read buffers on mount: think disk change.. */
624 	for (i = 0; i < READ_BUFFERS; i++)
625 		buffer_blocknr[i] = -1;
626 
627 	err = cramfs_read_super(sb, fc, &super);
628 	if (err)
629 		return err;
630 	return cramfs_finalize_super(sb, &super.root);
631 }
632 
633 static int cramfs_mtd_fill_super(struct super_block *sb, struct fs_context *fc)
634 {
635 	struct cramfs_sb_info *sbi;
636 	struct cramfs_super super;
637 	int err;
638 
639 	sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
640 	if (!sbi)
641 		return -ENOMEM;
642 	sb->s_fs_info = sbi;
643 
644 	/* Map only one page for now.  Will remap it when fs size is known. */
645 	err = mtd_point(sb->s_mtd, 0, PAGE_SIZE, &sbi->mtd_point_size,
646 			&sbi->linear_virt_addr, &sbi->linear_phys_addr);
647 	if (err || sbi->mtd_point_size != PAGE_SIZE) {
648 		pr_err("unable to get direct memory access to mtd:%s\n",
649 		       sb->s_mtd->name);
650 		return err ? : -ENODATA;
651 	}
652 
653 	pr_info("checking physical address %pap for linear cramfs image\n",
654 		&sbi->linear_phys_addr);
655 	err = cramfs_read_super(sb, fc, &super);
656 	if (err)
657 		return err;
658 
659 	/* Remap the whole filesystem now */
660 	pr_info("linear cramfs image on mtd:%s appears to be %lu KB in size\n",
661 		sb->s_mtd->name, sbi->size/1024);
662 	mtd_unpoint(sb->s_mtd, 0, PAGE_SIZE);
663 	err = mtd_point(sb->s_mtd, 0, sbi->size, &sbi->mtd_point_size,
664 			&sbi->linear_virt_addr, &sbi->linear_phys_addr);
665 	if (err || sbi->mtd_point_size != sbi->size) {
666 		pr_err("unable to get direct memory access to mtd:%s\n",
667 		       sb->s_mtd->name);
668 		return err ? : -ENODATA;
669 	}
670 
671 	return cramfs_finalize_super(sb, &super.root);
672 }
673 
674 static int cramfs_statfs(struct dentry *dentry, struct kstatfs *buf)
675 {
676 	struct super_block *sb = dentry->d_sb;
677 	u64 id = 0;
678 
679 	if (sb->s_bdev)
680 		id = huge_encode_dev(sb->s_bdev->bd_dev);
681 	else if (sb->s_dev)
682 		id = huge_encode_dev(sb->s_dev);
683 
684 	buf->f_type = CRAMFS_MAGIC;
685 	buf->f_bsize = PAGE_SIZE;
686 	buf->f_blocks = CRAMFS_SB(sb)->blocks;
687 	buf->f_bfree = 0;
688 	buf->f_bavail = 0;
689 	buf->f_files = CRAMFS_SB(sb)->files;
690 	buf->f_ffree = 0;
691 	buf->f_fsid = u64_to_fsid(id);
692 	buf->f_namelen = CRAMFS_MAXPATHLEN;
693 	return 0;
694 }
695 
696 /*
697  * Read a cramfs directory entry.
698  */
699 static int cramfs_readdir(struct file *file, struct dir_context *ctx)
700 {
701 	struct inode *inode = file_inode(file);
702 	struct super_block *sb = inode->i_sb;
703 	char *buf;
704 	unsigned int offset;
705 
706 	/* Offset within the thing. */
707 	if (ctx->pos >= inode->i_size)
708 		return 0;
709 	offset = ctx->pos;
710 	/* Directory entries are always 4-byte aligned */
711 	if (offset & 3)
712 		return -EINVAL;
713 
714 	buf = kmalloc(CRAMFS_MAXPATHLEN, GFP_KERNEL);
715 	if (!buf)
716 		return -ENOMEM;
717 
718 	while (offset < inode->i_size) {
719 		struct cramfs_inode *de;
720 		unsigned long nextoffset;
721 		char *name;
722 		ino_t ino;
723 		umode_t mode;
724 		int namelen;
725 
726 		mutex_lock(&read_mutex);
727 		de = cramfs_read(sb, OFFSET(inode) + offset, sizeof(*de)+CRAMFS_MAXPATHLEN);
728 		name = (char *)(de+1);
729 
730 		/*
731 		 * Namelengths on disk are shifted by two
732 		 * and the name padded out to 4-byte boundaries
733 		 * with zeroes.
734 		 */
735 		namelen = de->namelen << 2;
736 		memcpy(buf, name, namelen);
737 		ino = cramino(de, OFFSET(inode) + offset);
738 		mode = de->mode;
739 		mutex_unlock(&read_mutex);
740 		nextoffset = offset + sizeof(*de) + namelen;
741 		for (;;) {
742 			if (!namelen) {
743 				kfree(buf);
744 				return -EIO;
745 			}
746 			if (buf[namelen-1])
747 				break;
748 			namelen--;
749 		}
750 		if (!dir_emit(ctx, buf, namelen, ino, mode >> 12))
751 			break;
752 
753 		ctx->pos = offset = nextoffset;
754 	}
755 	kfree(buf);
756 	return 0;
757 }
758 
759 /*
760  * Lookup and fill in the inode data..
761  */
762 static struct dentry *cramfs_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
763 {
764 	unsigned int offset = 0;
765 	struct inode *inode = NULL;
766 	int sorted;
767 
768 	mutex_lock(&read_mutex);
769 	sorted = CRAMFS_SB(dir->i_sb)->flags & CRAMFS_FLAG_SORTED_DIRS;
770 	while (offset < dir->i_size) {
771 		struct cramfs_inode *de;
772 		char *name;
773 		int namelen, retval;
774 		int dir_off = OFFSET(dir) + offset;
775 
776 		de = cramfs_read(dir->i_sb, dir_off, sizeof(*de)+CRAMFS_MAXPATHLEN);
777 		name = (char *)(de+1);
778 
779 		/* Try to take advantage of sorted directories */
780 		if (sorted && (dentry->d_name.name[0] < name[0]))
781 			break;
782 
783 		namelen = de->namelen << 2;
784 		offset += sizeof(*de) + namelen;
785 
786 		/* Quick check that the name is roughly the right length */
787 		if (((dentry->d_name.len + 3) & ~3) != namelen)
788 			continue;
789 
790 		for (;;) {
791 			if (!namelen) {
792 				inode = ERR_PTR(-EIO);
793 				goto out;
794 			}
795 			if (name[namelen-1])
796 				break;
797 			namelen--;
798 		}
799 		if (namelen != dentry->d_name.len)
800 			continue;
801 		retval = memcmp(dentry->d_name.name, name, namelen);
802 		if (retval > 0)
803 			continue;
804 		if (!retval) {
805 			inode = get_cramfs_inode(dir->i_sb, de, dir_off);
806 			break;
807 		}
808 		/* else (retval < 0) */
809 		if (sorted)
810 			break;
811 	}
812 out:
813 	mutex_unlock(&read_mutex);
814 	return d_splice_alias(inode, dentry);
815 }
816 
817 static int cramfs_readpage(struct file *file, struct page *page)
818 {
819 	struct inode *inode = page->mapping->host;
820 	u32 maxblock;
821 	int bytes_filled;
822 	void *pgdata;
823 
824 	maxblock = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
825 	bytes_filled = 0;
826 	pgdata = kmap(page);
827 
828 	if (page->index < maxblock) {
829 		struct super_block *sb = inode->i_sb;
830 		u32 blkptr_offset = OFFSET(inode) + page->index * 4;
831 		u32 block_ptr, block_start, block_len;
832 		bool uncompressed, direct;
833 
834 		mutex_lock(&read_mutex);
835 		block_ptr = *(u32 *) cramfs_read(sb, blkptr_offset, 4);
836 		uncompressed = (block_ptr & CRAMFS_BLK_FLAG_UNCOMPRESSED);
837 		direct = (block_ptr & CRAMFS_BLK_FLAG_DIRECT_PTR);
838 		block_ptr &= ~CRAMFS_BLK_FLAGS;
839 
840 		if (direct) {
841 			/*
842 			 * The block pointer is an absolute start pointer,
843 			 * shifted by 2 bits. The size is included in the
844 			 * first 2 bytes of the data block when compressed,
845 			 * or PAGE_SIZE otherwise.
846 			 */
847 			block_start = block_ptr << CRAMFS_BLK_DIRECT_PTR_SHIFT;
848 			if (uncompressed) {
849 				block_len = PAGE_SIZE;
850 				/* if last block: cap to file length */
851 				if (page->index == maxblock - 1)
852 					block_len =
853 						offset_in_page(inode->i_size);
854 			} else {
855 				block_len = *(u16 *)
856 					cramfs_read(sb, block_start, 2);
857 				block_start += 2;
858 			}
859 		} else {
860 			/*
861 			 * The block pointer indicates one past the end of
862 			 * the current block (start of next block). If this
863 			 * is the first block then it starts where the block
864 			 * pointer table ends, otherwise its start comes
865 			 * from the previous block's pointer.
866 			 */
867 			block_start = OFFSET(inode) + maxblock * 4;
868 			if (page->index)
869 				block_start = *(u32 *)
870 					cramfs_read(sb, blkptr_offset - 4, 4);
871 			/* Beware... previous ptr might be a direct ptr */
872 			if (unlikely(block_start & CRAMFS_BLK_FLAG_DIRECT_PTR)) {
873 				/* See comments on earlier code. */
874 				u32 prev_start = block_start;
875 				block_start = prev_start & ~CRAMFS_BLK_FLAGS;
876 				block_start <<= CRAMFS_BLK_DIRECT_PTR_SHIFT;
877 				if (prev_start & CRAMFS_BLK_FLAG_UNCOMPRESSED) {
878 					block_start += PAGE_SIZE;
879 				} else {
880 					block_len = *(u16 *)
881 						cramfs_read(sb, block_start, 2);
882 					block_start += 2 + block_len;
883 				}
884 			}
885 			block_start &= ~CRAMFS_BLK_FLAGS;
886 			block_len = block_ptr - block_start;
887 		}
888 
889 		if (block_len == 0)
890 			; /* hole */
891 		else if (unlikely(block_len > 2*PAGE_SIZE ||
892 				  (uncompressed && block_len > PAGE_SIZE))) {
893 			mutex_unlock(&read_mutex);
894 			pr_err("bad data blocksize %u\n", block_len);
895 			goto err;
896 		} else if (uncompressed) {
897 			memcpy(pgdata,
898 			       cramfs_read(sb, block_start, block_len),
899 			       block_len);
900 			bytes_filled = block_len;
901 		} else {
902 			bytes_filled = cramfs_uncompress_block(pgdata,
903 				 PAGE_SIZE,
904 				 cramfs_read(sb, block_start, block_len),
905 				 block_len);
906 		}
907 		mutex_unlock(&read_mutex);
908 		if (unlikely(bytes_filled < 0))
909 			goto err;
910 	}
911 
912 	memset(pgdata + bytes_filled, 0, PAGE_SIZE - bytes_filled);
913 	flush_dcache_page(page);
914 	kunmap(page);
915 	SetPageUptodate(page);
916 	unlock_page(page);
917 	return 0;
918 
919 err:
920 	kunmap(page);
921 	ClearPageUptodate(page);
922 	SetPageError(page);
923 	unlock_page(page);
924 	return 0;
925 }
926 
927 static const struct address_space_operations cramfs_aops = {
928 	.readpage = cramfs_readpage
929 };
930 
931 /*
932  * Our operations:
933  */
934 
935 /*
936  * A directory can only readdir
937  */
938 static const struct file_operations cramfs_directory_operations = {
939 	.llseek		= generic_file_llseek,
940 	.read		= generic_read_dir,
941 	.iterate_shared	= cramfs_readdir,
942 };
943 
944 static const struct inode_operations cramfs_dir_inode_operations = {
945 	.lookup		= cramfs_lookup,
946 };
947 
948 static const struct super_operations cramfs_ops = {
949 	.statfs		= cramfs_statfs,
950 };
951 
952 static int cramfs_get_tree(struct fs_context *fc)
953 {
954 	int ret = -ENOPROTOOPT;
955 
956 	if (IS_ENABLED(CONFIG_CRAMFS_MTD)) {
957 		ret = get_tree_mtd(fc, cramfs_mtd_fill_super);
958 		if (!ret)
959 			return 0;
960 	}
961 	if (IS_ENABLED(CONFIG_CRAMFS_BLOCKDEV))
962 		ret = get_tree_bdev(fc, cramfs_blkdev_fill_super);
963 	return ret;
964 }
965 
966 static const struct fs_context_operations cramfs_context_ops = {
967 	.get_tree	= cramfs_get_tree,
968 	.reconfigure	= cramfs_reconfigure,
969 };
970 
971 /*
972  * Set up the filesystem mount context.
973  */
974 static int cramfs_init_fs_context(struct fs_context *fc)
975 {
976 	fc->ops = &cramfs_context_ops;
977 	return 0;
978 }
979 
980 static struct file_system_type cramfs_fs_type = {
981 	.owner		= THIS_MODULE,
982 	.name		= "cramfs",
983 	.init_fs_context = cramfs_init_fs_context,
984 	.kill_sb	= cramfs_kill_sb,
985 	.fs_flags	= FS_REQUIRES_DEV,
986 };
987 MODULE_ALIAS_FS("cramfs");
988 
989 static int __init init_cramfs_fs(void)
990 {
991 	int rv;
992 
993 	rv = cramfs_uncompress_init();
994 	if (rv < 0)
995 		return rv;
996 	rv = register_filesystem(&cramfs_fs_type);
997 	if (rv < 0)
998 		cramfs_uncompress_exit();
999 	return rv;
1000 }
1001 
1002 static void __exit exit_cramfs_fs(void)
1003 {
1004 	cramfs_uncompress_exit();
1005 	unregister_filesystem(&cramfs_fs_type);
1006 }
1007 
1008 module_init(init_cramfs_fs)
1009 module_exit(exit_cramfs_fs)
1010 MODULE_LICENSE("GPL");
1011