initialize.c revision da3fc25bce5880c5d036d97727299ab09b415219
1/*
2 * initialize.c --- initialize a filesystem handle given superblock
3 * 	parameters.  Used by mke2fs when initializing a filesystem.
4 *
5 * Copyright (C) 1994, 1995, 1996 Theodore Ts'o.
6 *
7 * %Begin-Header%
8 * This file may be redistributed under the terms of the GNU Library
9 * General Public License, version 2.
10 * %End-Header%
11 */
12
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
27#include "ext2_fs.h"
28#include "ext2fs.h"
29
30#if defined(__linux__)    &&	defined(EXT2_OS_LINUX)
31#define CREATOR_OS EXT2_OS_LINUX
32#else
33#if defined(__GNU__)     &&	defined(EXT2_OS_HURD)
34#define CREATOR_OS EXT2_OS_HURD
35#else
36#if defined(__FreeBSD__) &&	defined(EXT2_OS_FREEBSD)
37#define CREATOR_OS EXT2_OS_FREEBSD
38#else
39#if defined(LITES) 	   &&	defined(EXT2_OS_LITES)
40#define CREATOR_OS EXT2_OS_LITES
41#else
42#define CREATOR_OS EXT2_OS_LINUX /* by default */
43#endif /* defined(LITES) && defined(EXT2_OS_LITES) */
44#endif /* defined(__FreeBSD__) && defined(EXT2_OS_FREEBSD) */
45#endif /* defined(__GNU__)     && defined(EXT2_OS_HURD) */
46#endif /* defined(__linux__)   && defined(EXT2_OS_LINUX) */
47
48/*
49 * Note we override the kernel include file's idea of what the default
50 * check interval (never) should be.  It's a good idea to check at
51 * least *occasionally*, specially since servers will never rarely get
52 * to reboot, since Linux is so robust these days.  :-)
53 *
54 * 180 days (six months) seems like a good value.
55 */
56#ifdef EXT2_DFL_CHECKINTERVAL
57#undef EXT2_DFL_CHECKINTERVAL
58#endif
59#define EXT2_DFL_CHECKINTERVAL (86400L * 180L)
60
61/*
62 * Calculate the number of GDT blocks to reserve for online filesystem growth.
63 * The absolute maximum number of GDT blocks we can reserve is determined by
64 * the number of block pointers that can fit into a single block.
65 */
66static unsigned int calc_reserved_gdt_blocks(ext2_filsys fs)
67{
68	struct ext2_super_block *sb = fs->super;
69	unsigned long bpg = sb->s_blocks_per_group;
70	unsigned int gdpb = EXT2_DESC_PER_BLOCK(sb);
71	unsigned long max_blocks = 0xffffffff;
72	unsigned long rsv_groups;
73	unsigned int rsv_gdb;
74
75	/* We set it at 1024x the current filesystem size, or
76	 * the upper block count limit (2^32), whichever is lower.
77	 */
78	if (ext2fs_blocks_count(sb) < max_blocks / 1024)
79		max_blocks = ext2fs_blocks_count(sb) * 1024;
80	/*
81	 * ext2fs_div64_ceil() is unnecessary because max_blocks is
82	 * max _GDT_ blocks, which is limited to 32 bits.
83	 */
84	rsv_groups = ext2fs_div_ceil(max_blocks - sb->s_first_data_block, bpg);
85	rsv_gdb = ext2fs_div_ceil(rsv_groups, gdpb) - fs->desc_blocks;
86	if (rsv_gdb > EXT2_ADDR_PER_BLOCK(sb))
87		rsv_gdb = EXT2_ADDR_PER_BLOCK(sb);
88#ifdef RES_GDT_DEBUG
89	printf("max_blocks %lu, rsv_groups = %lu, rsv_gdb = %u\n",
90	       max_blocks, rsv_groups, rsv_gdb);
91#endif
92
93	return rsv_gdb;
94}
95
96errcode_t ext2fs_initialize(const char *name, int flags,
97			    struct ext2_super_block *param,
98			    io_manager manager, ext2_filsys *ret_fs)
99{
100	ext2_filsys	fs;
101	errcode_t	retval;
102	struct ext2_super_block *super;
103	int		frags_per_block;
104	unsigned int	rem;
105	unsigned int	overhead = 0;
106	unsigned int	ipg;
107	dgrp_t		i;
108	blk_t		numblocks;
109	int		rsv_gdt;
110	int		csum_flag;
111	int		io_flags;
112	char		*buf = 0;
113	char		c;
114
115	if (!param || !ext2fs_blocks_count(param))
116		return EXT2_ET_INVALID_ARGUMENT;
117
118	retval = ext2fs_get_mem(sizeof(struct struct_ext2_filsys), &fs);
119	if (retval)
120		return retval;
121
122	memset(fs, 0, sizeof(struct struct_ext2_filsys));
123	fs->magic = EXT2_ET_MAGIC_EXT2FS_FILSYS;
124	fs->flags = flags | EXT2_FLAG_RW;
125	fs->umask = 022;
126#ifdef WORDS_BIGENDIAN
127	fs->flags |= EXT2_FLAG_SWAP_BYTES;
128#endif
129	io_flags = IO_FLAG_RW;
130	if (flags & EXT2_FLAG_EXCLUSIVE)
131		io_flags |= IO_FLAG_EXCLUSIVE;
132	retval = manager->open(name, io_flags, &fs->io);
133	if (retval)
134		goto cleanup;
135	fs->image_io = fs->io;
136	fs->io->app_data = fs;
137	retval = ext2fs_get_mem(strlen(name)+1, &fs->device_name);
138	if (retval)
139		goto cleanup;
140
141	strcpy(fs->device_name, name);
142	retval = ext2fs_get_mem(SUPERBLOCK_SIZE, &super);
143	if (retval)
144		goto cleanup;
145	fs->super = super;
146
147	memset(super, 0, SUPERBLOCK_SIZE);
148
149#define set_field(field, default) (super->field = param->field ? \
150				   param->field : (default))
151
152	super->s_magic = EXT2_SUPER_MAGIC;
153	super->s_state = EXT2_VALID_FS;
154
155	set_field(s_log_block_size, 0);	/* default blocksize: 1024 bytes */
156	set_field(s_log_frag_size, 0); /* default fragsize: 1024 bytes */
157	set_field(s_first_data_block, super->s_log_block_size ? 0 : 1);
158	set_field(s_max_mnt_count, EXT2_DFL_MAX_MNT_COUNT);
159	set_field(s_errors, EXT2_ERRORS_DEFAULT);
160	set_field(s_feature_compat, 0);
161	set_field(s_feature_incompat, 0);
162	set_field(s_feature_ro_compat, 0);
163	set_field(s_first_meta_bg, 0);
164	set_field(s_raid_stride, 0);		/* default stride size: 0 */
165	set_field(s_raid_stripe_width, 0);	/* default stripe width: 0 */
166	set_field(s_log_groups_per_flex, 0);
167	set_field(s_flags, 0);
168	if (super->s_feature_incompat & ~EXT2_LIB_FEATURE_INCOMPAT_SUPP) {
169		retval = EXT2_ET_UNSUPP_FEATURE;
170		goto cleanup;
171	}
172	if (super->s_feature_ro_compat & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
173		retval = EXT2_ET_RO_UNSUPP_FEATURE;
174		goto cleanup;
175	}
176
177	set_field(s_rev_level, EXT2_GOOD_OLD_REV);
178	if (super->s_rev_level >= EXT2_DYNAMIC_REV) {
179		set_field(s_first_ino, EXT2_GOOD_OLD_FIRST_INO);
180		set_field(s_inode_size, EXT2_GOOD_OLD_INODE_SIZE);
181		if (super->s_inode_size >= sizeof(struct ext2_inode_large)) {
182			int extra_isize = sizeof(struct ext2_inode_large) -
183				EXT2_GOOD_OLD_INODE_SIZE;
184			set_field(s_min_extra_isize, extra_isize);
185			set_field(s_want_extra_isize, extra_isize);
186		}
187	} else {
188		super->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
189		super->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
190	}
191
192	set_field(s_checkinterval, EXT2_DFL_CHECKINTERVAL);
193	super->s_mkfs_time = super->s_lastcheck = fs->now ? fs->now : time(NULL);
194
195	super->s_creator_os = CREATOR_OS;
196
197	fs->blocksize = EXT2_BLOCK_SIZE(super);
198	fs->fragsize = EXT2_FRAG_SIZE(super);
199	frags_per_block = fs->blocksize / fs->fragsize;
200
201	/* default: (fs->blocksize*8) blocks/group, up to 2^16 (GDT limit) */
202	set_field(s_blocks_per_group, fs->blocksize * 8);
203	if (super->s_blocks_per_group > EXT2_MAX_BLOCKS_PER_GROUP(super))
204		super->s_blocks_per_group = EXT2_MAX_BLOCKS_PER_GROUP(super);
205	super->s_frags_per_group = super->s_blocks_per_group * frags_per_block;
206
207	ext2fs_blocks_count_set(super, ext2fs_blocks_count(param));
208	ext2fs_r_blocks_count_set(super, ext2fs_r_blocks_count(param));
209	if (ext2fs_r_blocks_count(super) >= ext2fs_blocks_count(param)) {
210		retval = EXT2_ET_INVALID_ARGUMENT;
211		goto cleanup;
212	}
213
214	/*
215	 * If we're creating an external journal device, we don't need
216	 * to bother with the rest.
217	 */
218	if (super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
219		fs->group_desc_count = 0;
220		ext2fs_mark_super_dirty(fs);
221		*ret_fs = fs;
222		return 0;
223	}
224
225retry:
226	fs->group_desc_count = (blk_t) ext2fs_div64_ceil(
227		ext2fs_blocks_count(super) - super->s_first_data_block,
228		EXT2_BLOCKS_PER_GROUP(super));
229	if (fs->group_desc_count == 0) {
230		retval = EXT2_ET_TOOSMALL;
231		goto cleanup;
232	}
233
234	if (super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT)
235		super->s_desc_size = EXT2_MIN_DESC_SIZE_64BIT;
236
237	fs->desc_blocks = ext2fs_div_ceil(fs->group_desc_count,
238					  EXT2_DESC_PER_BLOCK(super));
239
240	i = fs->blocksize >= 4096 ? 1 : 4096 / fs->blocksize;
241
242	if (super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT &&
243	    (ext2fs_blocks_count(super) / i) > (1ULL << 32))
244		set_field(s_inodes_count, ~0U);
245	else
246		set_field(s_inodes_count, ext2fs_blocks_count(super) / i);
247
248	/*
249	 * Make sure we have at least EXT2_FIRST_INO + 1 inodes, so
250	 * that we have enough inodes for the filesystem(!)
251	 */
252	if (super->s_inodes_count < EXT2_FIRST_INODE(super)+1)
253		super->s_inodes_count = EXT2_FIRST_INODE(super)+1;
254
255	/*
256	 * There should be at least as many inodes as the user
257	 * requested.  Figure out how many inodes per group that
258	 * should be.  But make sure that we don't allocate more than
259	 * one bitmap's worth of inodes each group.
260	 */
261	ipg = ext2fs_div_ceil(super->s_inodes_count, fs->group_desc_count);
262	if (ipg > fs->blocksize * 8) {
263		if (super->s_blocks_per_group >= 256) {
264			/* Try again with slightly different parameters */
265			super->s_blocks_per_group -= 8;
266			ext2fs_blocks_count_set(super,
267						ext2fs_blocks_count(param));
268			super->s_frags_per_group = super->s_blocks_per_group *
269				frags_per_block;
270			goto retry;
271		} else {
272			retval = EXT2_ET_TOO_MANY_INODES;
273			goto cleanup;
274		}
275	}
276
277	if (ipg > (unsigned) EXT2_MAX_INODES_PER_GROUP(super))
278		ipg = EXT2_MAX_INODES_PER_GROUP(super);
279
280ipg_retry:
281	super->s_inodes_per_group = ipg;
282
283	/*
284	 * Make sure the number of inodes per group completely fills
285	 * the inode table blocks in the descriptor.  If not, add some
286	 * additional inodes/group.  Waste not, want not...
287	 */
288	fs->inode_blocks_per_group = (((super->s_inodes_per_group *
289					EXT2_INODE_SIZE(super)) +
290				       EXT2_BLOCK_SIZE(super) - 1) /
291				      EXT2_BLOCK_SIZE(super));
292	super->s_inodes_per_group = ((fs->inode_blocks_per_group *
293				      EXT2_BLOCK_SIZE(super)) /
294				     EXT2_INODE_SIZE(super));
295	/*
296	 * Finally, make sure the number of inodes per group is a
297	 * multiple of 8.  This is needed to simplify the bitmap
298	 * splicing code.
299	 */
300	super->s_inodes_per_group &= ~7;
301	fs->inode_blocks_per_group = (((super->s_inodes_per_group *
302					EXT2_INODE_SIZE(super)) +
303				       EXT2_BLOCK_SIZE(super) - 1) /
304				      EXT2_BLOCK_SIZE(super));
305
306	/*
307	 * adjust inode count to reflect the adjusted inodes_per_group
308	 */
309	if ((__u64)super->s_inodes_per_group * fs->group_desc_count > ~0U) {
310		ipg--;
311		goto ipg_retry;
312	}
313	super->s_inodes_count = super->s_inodes_per_group *
314		fs->group_desc_count;
315	super->s_free_inodes_count = super->s_inodes_count;
316
317	/*
318	 * check the number of reserved group descriptor table blocks
319	 */
320	if (super->s_feature_compat & EXT2_FEATURE_COMPAT_RESIZE_INODE)
321		rsv_gdt = calc_reserved_gdt_blocks(fs);
322	else
323		rsv_gdt = 0;
324	set_field(s_reserved_gdt_blocks, rsv_gdt);
325	if (super->s_reserved_gdt_blocks > EXT2_ADDR_PER_BLOCK(super)) {
326		retval = EXT2_ET_RES_GDT_BLOCKS;
327		goto cleanup;
328	}
329
330	/*
331	 * Calculate the maximum number of bookkeeping blocks per
332	 * group.  It includes the superblock, the block group
333	 * descriptors, the block bitmap, the inode bitmap, the inode
334	 * table, and the reserved gdt blocks.
335	 */
336	overhead = (int) (3 + fs->inode_blocks_per_group +
337			  fs->desc_blocks + super->s_reserved_gdt_blocks);
338
339	/* This can only happen if the user requested too many inodes */
340	if (overhead > super->s_blocks_per_group) {
341		retval = EXT2_ET_TOO_MANY_INODES;
342		goto cleanup;
343	}
344
345	/*
346	 * See if the last group is big enough to support the
347	 * necessary data structures.  If not, we need to get rid of
348	 * it.  We need to recalculate the overhead for the last block
349	 * group, since it might or might not have a superblock
350	 * backup.
351	 */
352	overhead = (int) (2 + fs->inode_blocks_per_group);
353	if (ext2fs_bg_has_super(fs, fs->group_desc_count - 1))
354		overhead += 1 + fs->desc_blocks + super->s_reserved_gdt_blocks;
355	rem = ((ext2fs_blocks_count(super) - super->s_first_data_block) %
356	       super->s_blocks_per_group);
357	if ((fs->group_desc_count == 1) && rem && (rem < overhead)) {
358		retval = EXT2_ET_TOOSMALL;
359		goto cleanup;
360	}
361	if (rem && (rem < overhead+50)) {
362		ext2fs_blocks_count_set(super, ext2fs_blocks_count(super) -
363					rem);
364
365		goto retry;
366	}
367
368	/*
369	 * At this point we know how big the filesystem will be.  So
370	 * we can do any and all allocations that depend on the block
371	 * count.
372	 */
373
374	retval = ext2fs_get_mem(strlen(fs->device_name) + 80, &buf);
375	if (retval)
376		goto cleanup;
377
378	strcpy(buf, "block bitmap for ");
379	strcat(buf, fs->device_name);
380	retval = ext2fs_allocate_block_bitmap(fs, buf, &fs->block_map);
381	if (retval)
382		goto cleanup;
383
384	strcpy(buf, "inode bitmap for ");
385	strcat(buf, fs->device_name);
386	retval = ext2fs_allocate_inode_bitmap(fs, buf, &fs->inode_map);
387	if (retval)
388		goto cleanup;
389
390	ext2fs_free_mem(&buf);
391
392	retval = ext2fs_get_array(fs->desc_blocks, fs->blocksize,
393				&fs->group_desc);
394	if (retval)
395		goto cleanup;
396
397	memset(fs->group_desc, 0, (size_t) fs->desc_blocks * fs->blocksize);
398
399	/*
400	 * Reserve the superblock and group descriptors for each
401	 * group, and fill in the correct group statistics for group.
402	 * Note that although the block bitmap, inode bitmap, and
403	 * inode table have not been allocated (and in fact won't be
404	 * by this routine), they are accounted for nevertheless.
405	 *
406	 * If FLEX_BG meta-data grouping is used, only account for the
407	 * superblock and group descriptors (the inode tables and
408	 * bitmaps will be accounted for when allocated).
409	 */
410	ext2fs_free_blocks_count_set(super, 0);
411	csum_flag = EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
412					       EXT4_FEATURE_RO_COMPAT_GDT_CSUM);
413	for (i = 0; i < fs->group_desc_count; i++) {
414		/*
415		 * Don't set the BLOCK_UNINIT group for the last group
416		 * because the block bitmap needs to be padded.
417		 */
418		if (csum_flag) {
419			if (i != fs->group_desc_count - 1)
420				ext2fs_bg_flags_set(fs, i,
421						    EXT2_BG_BLOCK_UNINIT);
422			ext2fs_bg_flags_set(fs, i, EXT2_BG_INODE_UNINIT);
423			numblocks = super->s_inodes_per_group;
424			if (i == 0)
425				numblocks -= super->s_first_ino;
426			ext2fs_bg_itable_unused_set(fs, i, numblocks);
427		}
428		numblocks = ext2fs_reserve_super_and_bgd(fs, i, fs->block_map);
429		if (fs->super->s_log_groups_per_flex)
430			numblocks += 2 + fs->inode_blocks_per_group;
431
432		ext2fs_free_blocks_count_set(super,
433					     ext2fs_free_blocks_count(super) +
434					     numblocks);
435		ext2fs_bg_free_blocks_count_set(fs, i, numblocks);
436		ext2fs_bg_free_inodes_count_set(fs, i, fs->super->s_inodes_per_group);
437		ext2fs_bg_used_dirs_count_set(fs, i, 0);
438		ext2fs_group_desc_csum_set(fs, i);
439	}
440
441	c = (char) 255;
442	if (((int) c) == -1) {
443		super->s_flags |= EXT2_FLAGS_SIGNED_HASH;
444	} else {
445		super->s_flags |= EXT2_FLAGS_UNSIGNED_HASH;
446	}
447
448	ext2fs_mark_super_dirty(fs);
449	ext2fs_mark_bb_dirty(fs);
450	ext2fs_mark_ib_dirty(fs);
451
452	io_channel_set_blksize(fs->io, fs->blocksize);
453
454	*ret_fs = fs;
455	return 0;
456cleanup:
457	free(buf);
458	ext2fs_free(fs);
459	return retval;
460}
461