openfs.c revision 585bca68f89aa2a99e39f9234a10f825eb243589
1/*
2 * openfs.c --- open an ext2 filesystem
3 *
4 * Copyright (C) 1993, 1994, 1995, 1996 Theodore Ts'o.
5 *
6 * %Begin-Header%
7 * This file may be redistributed under the terms of the GNU Library
8 * General Public License, version 2.
9 * %End-Header%
10 */
11
12#include "config.h"
13#include <stdio.h>
14#include <string.h>
15#if HAVE_UNISTD_H
16#include <unistd.h>
17#endif
18#include <fcntl.h>
19#include <time.h>
20#if HAVE_SYS_STAT_H
21#include <sys/stat.h>
22#endif
23#if HAVE_SYS_TYPES_H
24#include <sys/types.h>
25#endif
26#ifdef HAVE_ERRNO_H
27#include <errno.h>
28#endif
29
30#include "ext2_fs.h"
31
32
33#include "ext2fs.h"
34#include "e2image.h"
35
36blk64_t ext2fs_descriptor_block_loc2(ext2_filsys fs, blk64_t group_block,
37				     dgrp_t i)
38{
39	int	bg;
40	int	has_super = 0, group_zero_adjust = 0;
41	blk64_t	ret_blk;
42
43	/*
44	 * On a bigalloc FS with 1K blocks, block 0 is reserved for non-ext4
45	 * stuff, so adjust for that if we're being asked for group 0.
46	 */
47	if (i == 0 && fs->blocksize == 1024 && EXT2FS_CLUSTER_RATIO(fs) > 1)
48		group_zero_adjust = 1;
49
50	if (!(fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_META_BG) ||
51	    (i < fs->super->s_first_meta_bg))
52		return group_block + i + 1 + group_zero_adjust;
53
54	bg = EXT2_DESC_PER_BLOCK(fs->super) * i;
55	if (ext2fs_bg_has_super(fs, bg))
56		has_super = 1;
57	ret_blk = ext2fs_group_first_block2(fs, bg);
58	/*
59	 * If group_block is not the normal value, we're trying to use
60	 * the backup group descriptors and superblock --- so use the
61	 * alternate location of the second block group in the
62	 * metablock group.  Ideally we should be testing each bg
63	 * descriptor block individually for correctness, but we don't
64	 * have the infrastructure in place to do that.
65	 */
66	if (group_block != fs->super->s_first_data_block &&
67	    ((ret_blk + has_super + fs->super->s_blocks_per_group) <
68	     ext2fs_blocks_count(fs->super))) {
69		ret_blk += fs->super->s_blocks_per_group;
70
71		/*
72		 * If we're going to jump forward a block group, make sure
73		 * that we adjust has_super to account for the next group's
74		 * backup superblock (or lack thereof).
75		 */
76		if (ext2fs_bg_has_super(fs, bg + 1))
77			has_super = 1;
78		else
79			has_super = 0;
80	}
81	return ret_blk + has_super + group_zero_adjust;
82}
83
84blk_t ext2fs_descriptor_block_loc(ext2_filsys fs, blk_t group_block, dgrp_t i)
85{
86	return ext2fs_descriptor_block_loc2(fs, group_block, i);
87}
88
89errcode_t ext2fs_open(const char *name, int flags, int superblock,
90		      unsigned int block_size, io_manager manager,
91		      ext2_filsys *ret_fs)
92{
93	return ext2fs_open2(name, 0, flags, superblock, block_size,
94			    manager, ret_fs);
95}
96
97/*
98 *  Note: if superblock is non-zero, block-size must also be non-zero.
99 * 	Superblock and block_size can be zero to use the default size.
100 *
101 * Valid flags for ext2fs_open()
102 *
103 * 	EXT2_FLAG_RW	- Open the filesystem for read/write.
104 * 	EXT2_FLAG_FORCE - Open the filesystem even if some of the
105 *				features aren't supported.
106 *	EXT2_FLAG_JOURNAL_DEV_OK - Open an ext3 journal device
107 *	EXT2_FLAG_SKIP_MMP - Open without multi-mount protection check.
108 *	EXT2_FLAG_64BITS - Allow 64-bit bitfields (needed for large
109 *				filesystems)
110 */
111errcode_t ext2fs_open2(const char *name, const char *io_options,
112		       int flags, int superblock,
113		       unsigned int block_size, io_manager manager,
114		       ext2_filsys *ret_fs)
115{
116	ext2_filsys	fs;
117	errcode_t	retval;
118	unsigned long	i, first_meta_bg;
119	__u32		features;
120	unsigned int	blocks_per_group, io_flags;
121	blk64_t		group_block, blk;
122	char		*dest, *cp;
123	int		group_zero_adjust = 0;
124#ifdef WORDS_BIGENDIAN
125	unsigned int	groups_per_block;
126	struct ext2_group_desc *gdp;
127	int		j;
128#endif
129
130	EXT2_CHECK_MAGIC(manager, EXT2_ET_MAGIC_IO_MANAGER);
131
132	retval = ext2fs_get_mem(sizeof(struct struct_ext2_filsys), &fs);
133	if (retval)
134		return retval;
135
136	memset(fs, 0, sizeof(struct struct_ext2_filsys));
137	fs->magic = EXT2_ET_MAGIC_EXT2FS_FILSYS;
138	fs->flags = flags;
139	/* don't overwrite sb backups unless flag is explicitly cleared */
140	fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
141	fs->umask = 022;
142	retval = ext2fs_get_mem(strlen(name)+1, &fs->device_name);
143	if (retval)
144		goto cleanup;
145	strcpy(fs->device_name, name);
146	cp = strchr(fs->device_name, '?');
147	if (!io_options && cp) {
148		*cp++ = 0;
149		io_options = cp;
150	}
151
152	io_flags = 0;
153	if (flags & EXT2_FLAG_RW)
154		io_flags |= IO_FLAG_RW;
155	if (flags & EXT2_FLAG_EXCLUSIVE)
156		io_flags |= IO_FLAG_EXCLUSIVE;
157	if (flags & EXT2_FLAG_DIRECT_IO)
158		io_flags |= IO_FLAG_DIRECT_IO;
159	retval = manager->open(fs->device_name, io_flags, &fs->io);
160	if (retval)
161		goto cleanup;
162	if (io_options &&
163	    (retval = io_channel_set_options(fs->io, io_options)))
164		goto cleanup;
165	fs->image_io = fs->io;
166	fs->io->app_data = fs;
167	retval = io_channel_alloc_buf(fs->io, -SUPERBLOCK_SIZE, &fs->super);
168	if (retval)
169		goto cleanup;
170	if (flags & EXT2_FLAG_IMAGE_FILE) {
171		retval = ext2fs_get_mem(sizeof(struct ext2_image_hdr),
172					&fs->image_header);
173		if (retval)
174			goto cleanup;
175		retval = io_channel_read_blk(fs->io, 0,
176					     -(int)sizeof(struct ext2_image_hdr),
177					     fs->image_header);
178		if (retval)
179			goto cleanup;
180		if (fs->image_header->magic_number != EXT2_ET_MAGIC_E2IMAGE)
181			return EXT2_ET_MAGIC_E2IMAGE;
182		superblock = 1;
183		block_size = fs->image_header->fs_blocksize;
184	}
185
186	/*
187	 * If the user specifies a specific block # for the
188	 * superblock, then he/she must also specify the block size!
189	 * Otherwise, read the master superblock located at offset
190	 * SUPERBLOCK_OFFSET from the start of the partition.
191	 *
192	 * Note: we only save a backup copy of the superblock if we
193	 * are reading the superblock from the primary superblock location.
194	 */
195	if (superblock) {
196		if (!block_size) {
197			retval = EXT2_ET_INVALID_ARGUMENT;
198			goto cleanup;
199		}
200		io_channel_set_blksize(fs->io, block_size);
201		group_block = superblock;
202		fs->orig_super = 0;
203	} else {
204		io_channel_set_blksize(fs->io, SUPERBLOCK_OFFSET);
205		superblock = 1;
206		group_block = 0;
207		retval = ext2fs_get_mem(SUPERBLOCK_SIZE, &fs->orig_super);
208		if (retval)
209			goto cleanup;
210	}
211	retval = io_channel_read_blk(fs->io, superblock, -SUPERBLOCK_SIZE,
212				     fs->super);
213	if (retval)
214		goto cleanup;
215	if (fs->orig_super)
216		memcpy(fs->orig_super, fs->super, SUPERBLOCK_SIZE);
217
218#ifdef WORDS_BIGENDIAN
219	fs->flags |= EXT2_FLAG_SWAP_BYTES;
220	ext2fs_swap_super(fs->super);
221#else
222	if (fs->flags & EXT2_FLAG_SWAP_BYTES) {
223		retval = EXT2_ET_UNIMPLEMENTED;
224		goto cleanup;
225	}
226#endif
227
228	if (fs->super->s_magic != EXT2_SUPER_MAGIC) {
229		retval = EXT2_ET_BAD_MAGIC;
230		goto cleanup;
231	}
232	if (fs->super->s_rev_level > EXT2_LIB_CURRENT_REV) {
233		retval = EXT2_ET_REV_TOO_HIGH;
234		goto cleanup;
235	}
236
237	/*
238	 * Check for feature set incompatibility
239	 */
240	if (!(flags & EXT2_FLAG_FORCE)) {
241		features = fs->super->s_feature_incompat;
242#ifdef EXT2_LIB_SOFTSUPP_INCOMPAT
243		if (flags & EXT2_FLAG_SOFTSUPP_FEATURES)
244			features &= ~EXT2_LIB_SOFTSUPP_INCOMPAT;
245#endif
246		if (features & ~EXT2_LIB_FEATURE_INCOMPAT_SUPP) {
247			retval = EXT2_ET_UNSUPP_FEATURE;
248			goto cleanup;
249		}
250
251		features = fs->super->s_feature_ro_compat;
252#ifdef EXT2_LIB_SOFTSUPP_RO_COMPAT
253		if (flags & EXT2_FLAG_SOFTSUPP_FEATURES)
254			features &= ~EXT2_LIB_SOFTSUPP_RO_COMPAT;
255#endif
256		if ((flags & EXT2_FLAG_RW) &&
257		    (features & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP)) {
258			retval = EXT2_ET_RO_UNSUPP_FEATURE;
259			goto cleanup;
260		}
261
262		if (!(flags & EXT2_FLAG_JOURNAL_DEV_OK) &&
263		    (fs->super->s_feature_incompat &
264		     EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
265			retval = EXT2_ET_UNSUPP_FEATURE;
266			goto cleanup;
267		}
268	}
269
270	if ((fs->super->s_log_block_size + EXT2_MIN_BLOCK_LOG_SIZE) >
271	    EXT2_MAX_BLOCK_LOG_SIZE) {
272		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
273		goto cleanup;
274	}
275
276	/*
277	 * bigalloc requires cluster-aware bitfield operations, which at the
278	 * moment means we need EXT2_FLAG_64BITS.
279	 */
280	if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
281				       EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
282	    !(flags & EXT2_FLAG_64BITS)) {
283		retval = EXT2_ET_CANT_USE_LEGACY_BITMAPS;
284		goto cleanup;
285	}
286
287	if (!EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
288					EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
289	    (fs->super->s_log_block_size != fs->super->s_log_cluster_size)) {
290		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
291		goto cleanup;
292	}
293	fs->fragsize = fs->blocksize = EXT2_BLOCK_SIZE(fs->super);
294	if (EXT2_INODE_SIZE(fs->super) < EXT2_GOOD_OLD_INODE_SIZE) {
295		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
296		goto cleanup;
297	}
298	fs->cluster_ratio_bits = fs->super->s_log_cluster_size -
299		fs->super->s_log_block_size;
300	if (EXT2_BLOCKS_PER_GROUP(fs->super) !=
301	    EXT2_CLUSTERS_PER_GROUP(fs->super) << fs->cluster_ratio_bits) {
302		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
303		goto cleanup;
304	}
305	fs->inode_blocks_per_group = ((EXT2_INODES_PER_GROUP(fs->super) *
306				       EXT2_INODE_SIZE(fs->super) +
307				       EXT2_BLOCK_SIZE(fs->super) - 1) /
308				      EXT2_BLOCK_SIZE(fs->super));
309	if (block_size) {
310		if (block_size != fs->blocksize) {
311			retval = EXT2_ET_UNEXPECTED_BLOCK_SIZE;
312			goto cleanup;
313		}
314	}
315	/*
316	 * Set the blocksize to the filesystem's blocksize.
317	 */
318	io_channel_set_blksize(fs->io, fs->blocksize);
319
320	/*
321	 * If this is an external journal device, don't try to read
322	 * the group descriptors, because they're not there.
323	 */
324	if (fs->super->s_feature_incompat &
325	    EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
326		fs->group_desc_count = 0;
327		*ret_fs = fs;
328		return 0;
329	}
330
331	if (EXT2_INODES_PER_GROUP(fs->super) == 0) {
332		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
333		goto cleanup;
334	}
335
336	/*
337	 * Read group descriptors
338	 */
339	blocks_per_group = EXT2_BLOCKS_PER_GROUP(fs->super);
340	if (blocks_per_group == 0 ||
341	    blocks_per_group > EXT2_MAX_BLOCKS_PER_GROUP(fs->super) ||
342	    fs->inode_blocks_per_group > EXT2_MAX_INODES_PER_GROUP(fs->super) ||
343           EXT2_DESC_PER_BLOCK(fs->super) == 0 ||
344           fs->super->s_first_data_block >= ext2fs_blocks_count(fs->super)) {
345		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
346		goto cleanup;
347	}
348	fs->group_desc_count = ext2fs_div64_ceil(ext2fs_blocks_count(fs->super) -
349						 fs->super->s_first_data_block,
350						 blocks_per_group);
351	if (fs->group_desc_count * EXT2_INODES_PER_GROUP(fs->super) !=
352	    fs->super->s_inodes_count) {
353		retval = EXT2_ET_CORRUPT_SUPERBLOCK;
354		goto cleanup;
355	}
356	fs->desc_blocks = ext2fs_div_ceil(fs->group_desc_count,
357					  EXT2_DESC_PER_BLOCK(fs->super));
358	retval = ext2fs_get_array(fs->desc_blocks, fs->blocksize,
359				&fs->group_desc);
360	if (retval)
361		goto cleanup;
362	if (!group_block)
363		group_block = fs->super->s_first_data_block;
364	/*
365	 * On a FS with a 1K blocksize, block 0 is reserved for bootloaders
366	 * so we must increment block numbers to any group 0 items.
367	 *
368	 * However, we cannot touch group_block directly because in the meta_bg
369	 * case, the ext2fs_descriptor_block_loc2() function will interpret
370	 * group_block != s_first_data_block to mean that we want to access the
371	 * backup group descriptors.  This is not what we want if the caller
372	 * set superblock == 0 (i.e. auto-detect the superblock), which is
373	 * what's going on here.
374	 */
375	if (group_block == 0 && fs->blocksize == 1024)
376		group_zero_adjust = 1;
377	dest = (char *) fs->group_desc;
378#ifdef WORDS_BIGENDIAN
379	groups_per_block = EXT2_DESC_PER_BLOCK(fs->super);
380#endif
381	if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_META_BG)
382		first_meta_bg = fs->super->s_first_meta_bg;
383	else
384		first_meta_bg = fs->desc_blocks;
385	if (first_meta_bg) {
386		retval = io_channel_read_blk(fs->io, group_block +
387					     group_zero_adjust + 1,
388					     first_meta_bg, dest);
389		if (retval)
390			goto cleanup;
391#ifdef WORDS_BIGENDIAN
392		gdp = (struct ext2_group_desc *) dest;
393		for (j=0; j < groups_per_block*first_meta_bg; j++) {
394			gdp = ext2fs_group_desc(fs, fs->group_desc, j);
395			ext2fs_swap_group_desc2(fs, gdp);
396		}
397#endif
398		dest += fs->blocksize*first_meta_bg;
399	}
400	for (i=first_meta_bg ; i < fs->desc_blocks; i++) {
401		blk = ext2fs_descriptor_block_loc2(fs, group_block, i);
402		retval = io_channel_read_blk64(fs->io, blk, 1, dest);
403		if (retval)
404			goto cleanup;
405#ifdef WORDS_BIGENDIAN
406		for (j=0; j < groups_per_block; j++) {
407			gdp = ext2fs_group_desc(fs, fs->group_desc,
408						i * groups_per_block + j);
409			ext2fs_swap_group_desc2(fs, gdp);
410		}
411#endif
412		dest += fs->blocksize;
413	}
414
415	fs->stride = fs->super->s_raid_stride;
416
417	/*
418	 * If recovery is from backup superblock, Clear _UNININT flags &
419	 * reset bg_itable_unused to zero
420	 */
421	if (superblock > 1 && EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
422					EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
423		dgrp_t group;
424
425		for (group = 0; group < fs->group_desc_count; group++) {
426			ext2fs_bg_flags_clear(fs, group, EXT2_BG_BLOCK_UNINIT);
427			ext2fs_bg_flags_clear(fs, group, EXT2_BG_INODE_UNINIT);
428			ext2fs_bg_itable_unused_set(fs, group, 0);
429			/* The checksum will be reset later, but fix it here
430			 * anyway to avoid printing a lot of spurious errors. */
431			ext2fs_group_desc_csum_set(fs, group);
432		}
433		if (fs->flags & EXT2_FLAG_RW)
434			ext2fs_mark_super_dirty(fs);
435	}
436
437	if ((fs->super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_MMP) &&
438	    !(flags & EXT2_FLAG_SKIP_MMP) &&
439	    (flags & (EXT2_FLAG_RW | EXT2_FLAG_EXCLUSIVE))) {
440		retval = ext2fs_mmp_start(fs);
441		if (retval) {
442			fs->flags |= EXT2_FLAG_SKIP_MMP; /* just do cleanup */
443			ext2fs_mmp_stop(fs);
444			goto cleanup;
445		}
446	}
447
448	fs->flags &= ~EXT2_FLAG_NOFREE_ON_ERROR;
449	*ret_fs = fs;
450
451	return 0;
452cleanup:
453	if (!(flags & EXT2_FLAG_NOFREE_ON_ERROR)) {
454		ext2fs_free(fs);
455		fs = NULL;
456	}
457	*ret_fs = fs;
458	return retval;
459}
460
461/*
462 * Set/get the filesystem data I/O channel.
463 *
464 * These functions are only valid if EXT2_FLAG_IMAGE_FILE is true.
465 */
466errcode_t ext2fs_get_data_io(ext2_filsys fs, io_channel *old_io)
467{
468	if ((fs->flags & EXT2_FLAG_IMAGE_FILE) == 0)
469		return EXT2_ET_NOT_IMAGE_FILE;
470	if (old_io) {
471		*old_io = (fs->image_io == fs->io) ? 0 : fs->io;
472	}
473	return 0;
474}
475
476errcode_t ext2fs_set_data_io(ext2_filsys fs, io_channel new_io)
477{
478	if ((fs->flags & EXT2_FLAG_IMAGE_FILE) == 0)
479		return EXT2_ET_NOT_IMAGE_FILE;
480	fs->io = new_io ? new_io : fs->image_io;
481	return 0;
482}
483
484errcode_t ext2fs_rewrite_to_io(ext2_filsys fs, io_channel new_io)
485{
486	errcode_t err;
487
488	if ((fs->flags & EXT2_FLAG_IMAGE_FILE) == 0)
489		return EXT2_ET_NOT_IMAGE_FILE;
490	err = io_channel_set_blksize(new_io, fs->blocksize);
491	if (err)
492		return err;
493	if ((new_io == fs->image_io) || (new_io == fs->io))
494		return 0;
495	if ((fs->image_io != fs->io) &&
496	    fs->image_io)
497		io_channel_close(fs->image_io);
498	if (fs->io)
499		io_channel_close(fs->io);
500	fs->io = fs->image_io = new_io;
501	fs->flags |= EXT2_FLAG_DIRTY | EXT2_FLAG_RW |
502		EXT2_FLAG_BB_DIRTY | EXT2_FLAG_IB_DIRTY;
503	fs->flags &= ~EXT2_FLAG_IMAGE_FILE;
504	return 0;
505}
506