Whamcloud - gitweb
e2fsprogs: Fix thinko in ext2fs_initialize bigalloc case
[tools/e2fsprogs.git] / lib / ext2fs / initialize.c
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 "config.h"
14 #include <stdio.h>
15 #include <string.h>
16 #if HAVE_UNISTD_H
17 #include <unistd.h>
18 #endif
19 #include <fcntl.h>
20 #include <time.h>
21 #if HAVE_SYS_STAT_H
22 #include <sys/stat.h>
23 #endif
24 #if HAVE_SYS_TYPES_H
25 #include <sys/types.h>
26 #endif
27
28 #include "ext2_fs.h"
29 #include "ext2fs.h"
30
31 #if defined(__linux__)    &&    defined(EXT2_OS_LINUX)
32 #define CREATOR_OS EXT2_OS_LINUX
33 #else
34 #if defined(__GNU__)     &&     defined(EXT2_OS_HURD)
35 #define CREATOR_OS EXT2_OS_HURD
36 #else
37 #if defined(__FreeBSD__) &&     defined(EXT2_OS_FREEBSD)
38 #define CREATOR_OS EXT2_OS_FREEBSD
39 #else
40 #if defined(LITES)         &&   defined(EXT2_OS_LITES)
41 #define CREATOR_OS EXT2_OS_LITES
42 #else
43 #define CREATOR_OS EXT2_OS_LINUX /* by default */
44 #endif /* defined(LITES) && defined(EXT2_OS_LITES) */
45 #endif /* defined(__FreeBSD__) && defined(EXT2_OS_FREEBSD) */
46 #endif /* defined(__GNU__)     && defined(EXT2_OS_HURD) */
47 #endif /* defined(__linux__)   && defined(EXT2_OS_LINUX) */
48
49 /*
50  * Calculate the number of GDT blocks to reserve for online filesystem growth.
51  * The absolute maximum number of GDT blocks we can reserve is determined by
52  * the number of block pointers that can fit into a single block.
53  */
54 static unsigned int calc_reserved_gdt_blocks(ext2_filsys fs)
55 {
56         struct ext2_super_block *sb = fs->super;
57         unsigned long bpg = sb->s_blocks_per_group;
58         unsigned int gdpb = EXT2_DESC_PER_BLOCK(sb);
59         unsigned long max_blocks = 0xffffffff;
60         unsigned long rsv_groups;
61         unsigned int rsv_gdb;
62
63         /* We set it at 1024x the current filesystem size, or
64          * the upper block count limit (2^32), whichever is lower.
65          */
66         if (ext2fs_blocks_count(sb) < max_blocks / 1024)
67                 max_blocks = ext2fs_blocks_count(sb) * 1024;
68         /*
69          * ext2fs_div64_ceil() is unnecessary because max_blocks is
70          * max _GDT_ blocks, which is limited to 32 bits.
71          */
72         rsv_groups = ext2fs_div_ceil(max_blocks - sb->s_first_data_block, bpg);
73         rsv_gdb = ext2fs_div_ceil(rsv_groups, gdpb) - fs->desc_blocks;
74         if (rsv_gdb > EXT2_ADDR_PER_BLOCK(sb))
75                 rsv_gdb = EXT2_ADDR_PER_BLOCK(sb);
76 #ifdef RES_GDT_DEBUG
77         printf("max_blocks %lu, rsv_groups = %lu, rsv_gdb = %u\n",
78                max_blocks, rsv_groups, rsv_gdb);
79 #endif
80
81         return rsv_gdb;
82 }
83
84 errcode_t ext2fs_initialize(const char *name, int flags,
85                             struct ext2_super_block *param,
86                             io_manager manager, ext2_filsys *ret_fs)
87 {
88         ext2_filsys     fs;
89         errcode_t       retval;
90         struct ext2_super_block *super;
91         unsigned int    rem;
92         unsigned int    overhead = 0;
93         unsigned int    ipg;
94         dgrp_t          i;
95         blk64_t         free_blocks;
96         blk_t           numblocks;
97         int             rsv_gdt;
98         int             csum_flag;
99         int             bigalloc_flag;
100         int             io_flags;
101         char            *buf = 0;
102         char            c;
103
104         if (!param || !ext2fs_blocks_count(param))
105                 return EXT2_ET_INVALID_ARGUMENT;
106
107         retval = ext2fs_get_mem(sizeof(struct struct_ext2_filsys), &fs);
108         if (retval)
109                 return retval;
110
111         memset(fs, 0, sizeof(struct struct_ext2_filsys));
112         fs->magic = EXT2_ET_MAGIC_EXT2FS_FILSYS;
113         fs->flags = flags | EXT2_FLAG_RW;
114         fs->umask = 022;
115 #ifdef WORDS_BIGENDIAN
116         fs->flags |= EXT2_FLAG_SWAP_BYTES;
117 #endif
118         io_flags = IO_FLAG_RW;
119         if (flags & EXT2_FLAG_EXCLUSIVE)
120                 io_flags |= IO_FLAG_EXCLUSIVE;
121         retval = manager->open(name, io_flags, &fs->io);
122         if (retval)
123                 goto cleanup;
124         fs->image_io = fs->io;
125         fs->io->app_data = fs;
126         retval = ext2fs_get_mem(strlen(name)+1, &fs->device_name);
127         if (retval)
128                 goto cleanup;
129
130         strcpy(fs->device_name, name);
131         retval = ext2fs_get_mem(SUPERBLOCK_SIZE, &super);
132         if (retval)
133                 goto cleanup;
134         fs->super = super;
135
136         memset(super, 0, SUPERBLOCK_SIZE);
137
138 #define set_field(field, default) (super->field = param->field ? \
139                                    param->field : (default))
140 #define assign_field(field)     (super->field = param->field)
141
142         super->s_magic = EXT2_SUPER_MAGIC;
143         super->s_state = EXT2_VALID_FS;
144
145         bigalloc_flag = EXT2_HAS_RO_COMPAT_FEATURE(param,
146                                    EXT4_FEATURE_RO_COMPAT_BIGALLOC);
147
148         assign_field(s_log_block_size);
149
150         if (bigalloc_flag) {
151                 set_field(s_log_cluster_size, super->s_log_block_size+4);
152                 if (super->s_log_block_size > super->s_log_cluster_size) {
153                         retval = EXT2_ET_INVALID_ARGUMENT;
154                         goto cleanup;
155                 }
156         } else
157                 super->s_log_cluster_size = super->s_log_block_size;
158
159         set_field(s_first_data_block, super->s_log_cluster_size ? 0 : 1);
160         set_field(s_max_mnt_count, 0);
161         set_field(s_errors, EXT2_ERRORS_DEFAULT);
162         set_field(s_feature_compat, 0);
163         set_field(s_feature_incompat, 0);
164         set_field(s_feature_ro_compat, 0);
165         set_field(s_default_mount_opts, 0);
166         set_field(s_first_meta_bg, 0);
167         set_field(s_raid_stride, 0);            /* default stride size: 0 */
168         set_field(s_raid_stripe_width, 0);      /* default stripe width: 0 */
169         set_field(s_log_groups_per_flex, 0);
170         set_field(s_flags, 0);
171         if (super->s_feature_incompat & ~EXT2_LIB_FEATURE_INCOMPAT_SUPP) {
172                 retval = EXT2_ET_UNSUPP_FEATURE;
173                 goto cleanup;
174         }
175         if (super->s_feature_ro_compat & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
176                 retval = EXT2_ET_RO_UNSUPP_FEATURE;
177                 goto cleanup;
178         }
179
180         set_field(s_rev_level, EXT2_GOOD_OLD_REV);
181         if (super->s_rev_level >= EXT2_DYNAMIC_REV) {
182                 set_field(s_first_ino, EXT2_GOOD_OLD_FIRST_INO);
183                 set_field(s_inode_size, EXT2_GOOD_OLD_INODE_SIZE);
184                 if (super->s_inode_size >= sizeof(struct ext2_inode_large)) {
185                         int extra_isize = sizeof(struct ext2_inode_large) -
186                                 EXT2_GOOD_OLD_INODE_SIZE;
187                         set_field(s_min_extra_isize, extra_isize);
188                         set_field(s_want_extra_isize, extra_isize);
189                 }
190         } else {
191                 super->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
192                 super->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
193         }
194
195         set_field(s_checkinterval, 0);
196         super->s_mkfs_time = super->s_lastcheck = fs->now ? fs->now : time(NULL);
197
198         super->s_creator_os = CREATOR_OS;
199
200         fs->fragsize = fs->blocksize = EXT2_BLOCK_SIZE(super);
201         fs->cluster_ratio_bits = super->s_log_cluster_size -
202                 super->s_log_block_size;
203
204         if (bigalloc_flag) {
205                 if (param->s_blocks_per_group &&
206                     param->s_clusters_per_group &&
207                     ((param->s_clusters_per_group * EXT2FS_CLUSTER_RATIO(fs)) !=
208                      param->s_blocks_per_group)) {
209                         retval = EXT2_ET_INVALID_ARGUMENT;
210                         goto cleanup;
211                 }
212                 if (param->s_clusters_per_group)
213                         assign_field(s_clusters_per_group);
214                 else if (param->s_blocks_per_group)
215                         super->s_clusters_per_group = 
216                                 param->s_blocks_per_group /
217                                 EXT2FS_CLUSTER_RATIO(fs);
218                 else
219                         super->s_clusters_per_group = fs->blocksize * 8;
220                 if (super->s_clusters_per_group > EXT2_MAX_CLUSTERS_PER_GROUP(super))
221                         super->s_clusters_per_group = EXT2_MAX_CLUSTERS_PER_GROUP(super);
222                 super->s_blocks_per_group = EXT2FS_C2B(fs,
223                                        super->s_clusters_per_group);
224         } else {
225                 set_field(s_blocks_per_group, fs->blocksize * 8);
226                 if (super->s_blocks_per_group > EXT2_MAX_BLOCKS_PER_GROUP(super))
227                         super->s_blocks_per_group = EXT2_MAX_BLOCKS_PER_GROUP(super);
228                 super->s_clusters_per_group = super->s_blocks_per_group;
229         }
230
231         ext2fs_blocks_count_set(super, ext2fs_blocks_count(param) &
232                                 ~((blk64_t) EXT2FS_CLUSTER_MASK(fs)));
233         ext2fs_r_blocks_count_set(super, ext2fs_r_blocks_count(param));
234         if (ext2fs_r_blocks_count(super) >= ext2fs_blocks_count(param)) {
235                 retval = EXT2_ET_INVALID_ARGUMENT;
236                 goto cleanup;
237         }
238
239         /*
240          * If we're creating an external journal device, we don't need
241          * to bother with the rest.
242          */
243         if (super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
244                 fs->group_desc_count = 0;
245                 ext2fs_mark_super_dirty(fs);
246                 *ret_fs = fs;
247                 return 0;
248         }
249
250 retry:
251         fs->group_desc_count = (blk_t) ext2fs_div64_ceil(
252                 ext2fs_blocks_count(super) - super->s_first_data_block,
253                 EXT2_BLOCKS_PER_GROUP(super));
254         if (fs->group_desc_count == 0) {
255                 retval = EXT2_ET_TOOSMALL;
256                 goto cleanup;
257         }
258
259         if (super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT)
260                 super->s_desc_size = EXT2_MIN_DESC_SIZE_64BIT;
261
262         fs->desc_blocks = ext2fs_div_ceil(fs->group_desc_count,
263                                           EXT2_DESC_PER_BLOCK(super));
264
265         i = fs->blocksize >= 4096 ? 1 : 4096 / fs->blocksize;
266
267         if (super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT &&
268             (ext2fs_blocks_count(super) / i) > (1ULL << 32))
269                 set_field(s_inodes_count, ~0U);
270         else
271                 set_field(s_inodes_count, ext2fs_blocks_count(super) / i);
272
273         /*
274          * Make sure we have at least EXT2_FIRST_INO + 1 inodes, so
275          * that we have enough inodes for the filesystem(!)
276          */
277         if (super->s_inodes_count < EXT2_FIRST_INODE(super)+1)
278                 super->s_inodes_count = EXT2_FIRST_INODE(super)+1;
279
280         /*
281          * There should be at least as many inodes as the user
282          * requested.  Figure out how many inodes per group that
283          * should be.  But make sure that we don't allocate more than
284          * one bitmap's worth of inodes each group.
285          */
286         ipg = ext2fs_div_ceil(super->s_inodes_count, fs->group_desc_count);
287         if (ipg > fs->blocksize * 8) {
288                 if (!bigalloc_flag && super->s_blocks_per_group >= 256) {
289                         /* Try again with slightly different parameters */
290                         super->s_blocks_per_group -= 8;
291                         ext2fs_blocks_count_set(super,
292                                                 ext2fs_blocks_count(param));
293                         super->s_clusters_per_group = super->s_blocks_per_group;
294                         goto retry;
295                 } else {
296                         retval = EXT2_ET_TOO_MANY_INODES;
297                         goto cleanup;
298                 }
299         }
300
301         if (ipg > (unsigned) EXT2_MAX_INODES_PER_GROUP(super))
302                 ipg = EXT2_MAX_INODES_PER_GROUP(super);
303
304 ipg_retry:
305         super->s_inodes_per_group = ipg;
306
307         /*
308          * Make sure the number of inodes per group completely fills
309          * the inode table blocks in the descriptor.  If not, add some
310          * additional inodes/group.  Waste not, want not...
311          */
312         fs->inode_blocks_per_group = (((super->s_inodes_per_group *
313                                         EXT2_INODE_SIZE(super)) +
314                                        EXT2_BLOCK_SIZE(super) - 1) /
315                                       EXT2_BLOCK_SIZE(super));
316         super->s_inodes_per_group = ((fs->inode_blocks_per_group *
317                                       EXT2_BLOCK_SIZE(super)) /
318                                      EXT2_INODE_SIZE(super));
319         /*
320          * Finally, make sure the number of inodes per group is a
321          * multiple of 8.  This is needed to simplify the bitmap
322          * splicing code.
323          */
324         if (super->s_inodes_per_group < 8)
325                 super->s_inodes_per_group = 8;
326         super->s_inodes_per_group &= ~7;
327         fs->inode_blocks_per_group = (((super->s_inodes_per_group *
328                                         EXT2_INODE_SIZE(super)) +
329                                        EXT2_BLOCK_SIZE(super) - 1) /
330                                       EXT2_BLOCK_SIZE(super));
331
332         /*
333          * adjust inode count to reflect the adjusted inodes_per_group
334          */
335         if ((__u64)super->s_inodes_per_group * fs->group_desc_count > ~0U) {
336                 ipg--;
337                 goto ipg_retry;
338         }
339         super->s_inodes_count = super->s_inodes_per_group *
340                 fs->group_desc_count;
341         super->s_free_inodes_count = super->s_inodes_count;
342
343         /*
344          * check the number of reserved group descriptor table blocks
345          */
346         if (super->s_feature_compat & EXT2_FEATURE_COMPAT_RESIZE_INODE)
347                 rsv_gdt = calc_reserved_gdt_blocks(fs);
348         else
349                 rsv_gdt = 0;
350         set_field(s_reserved_gdt_blocks, rsv_gdt);
351         if (super->s_reserved_gdt_blocks > EXT2_ADDR_PER_BLOCK(super)) {
352                 retval = EXT2_ET_RES_GDT_BLOCKS;
353                 goto cleanup;
354         }
355
356         /*
357          * Calculate the maximum number of bookkeeping blocks per
358          * group.  It includes the superblock, the block group
359          * descriptors, the block bitmap, the inode bitmap, the inode
360          * table, and the reserved gdt blocks.
361          */
362         overhead = (int) (3 + fs->inode_blocks_per_group +
363                           fs->desc_blocks + super->s_reserved_gdt_blocks);
364
365         /* This can only happen if the user requested too many inodes */
366         if (overhead > super->s_blocks_per_group) {
367                 retval = EXT2_ET_TOO_MANY_INODES;
368                 goto cleanup;
369         }
370
371         /*
372          * See if the last group is big enough to support the
373          * necessary data structures.  If not, we need to get rid of
374          * it.  We need to recalculate the overhead for the last block
375          * group, since it might or might not have a superblock
376          * backup.
377          */
378         overhead = (int) (2 + fs->inode_blocks_per_group);
379         if (ext2fs_bg_has_super(fs, fs->group_desc_count - 1))
380                 overhead += 1 + fs->desc_blocks + super->s_reserved_gdt_blocks;
381         rem = ((ext2fs_blocks_count(super) - super->s_first_data_block) %
382                super->s_blocks_per_group);
383         if ((fs->group_desc_count == 1) && rem && (rem < overhead)) {
384                 retval = EXT2_ET_TOOSMALL;
385                 goto cleanup;
386         }
387         if (rem && (rem < overhead+50)) {
388                 ext2fs_blocks_count_set(super, ext2fs_blocks_count(super) -
389                                         rem);
390
391                 goto retry;
392         }
393
394         /*
395          * At this point we know how big the filesystem will be.  So
396          * we can do any and all allocations that depend on the block
397          * count.
398          */
399
400         retval = ext2fs_get_mem(strlen(fs->device_name) + 80, &buf);
401         if (retval)
402                 goto cleanup;
403
404         strcpy(buf, "block bitmap for ");
405         strcat(buf, fs->device_name);
406         retval = ext2fs_allocate_subcluster_bitmap(fs, buf, &fs->block_map);
407         if (retval)
408                 goto cleanup;
409
410         strcpy(buf, "inode bitmap for ");
411         strcat(buf, fs->device_name);
412         retval = ext2fs_allocate_inode_bitmap(fs, buf, &fs->inode_map);
413         if (retval)
414                 goto cleanup;
415
416         ext2fs_free_mem(&buf);
417
418         retval = ext2fs_get_array(fs->desc_blocks, fs->blocksize,
419                                 &fs->group_desc);
420         if (retval)
421                 goto cleanup;
422
423         memset(fs->group_desc, 0, (size_t) fs->desc_blocks * fs->blocksize);
424
425         /*
426          * Reserve the superblock and group descriptors for each
427          * group, and fill in the correct group statistics for group.
428          * Note that although the block bitmap, inode bitmap, and
429          * inode table have not been allocated (and in fact won't be
430          * by this routine), they are accounted for nevertheless.
431          *
432          * If FLEX_BG meta-data grouping is used, only account for the
433          * superblock and group descriptors (the inode tables and
434          * bitmaps will be accounted for when allocated).
435          */
436         free_blocks = 0;
437         csum_flag = EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
438                                                EXT4_FEATURE_RO_COMPAT_GDT_CSUM);
439         for (i = 0; i < fs->group_desc_count; i++) {
440                 /*
441                  * Don't set the BLOCK_UNINIT group for the last group
442                  * because the block bitmap needs to be padded.
443                  */
444                 if (csum_flag) {
445                         if (i != fs->group_desc_count - 1)
446                                 ext2fs_bg_flags_set(fs, i,
447                                                     EXT2_BG_BLOCK_UNINIT);
448                         ext2fs_bg_flags_set(fs, i, EXT2_BG_INODE_UNINIT);
449                         numblocks = super->s_inodes_per_group;
450                         if (i == 0)
451                                 numblocks -= super->s_first_ino;
452                         ext2fs_bg_itable_unused_set(fs, i, numblocks);
453                 }
454                 numblocks = ext2fs_reserve_super_and_bgd(fs, i, fs->block_map);
455                 if (fs->super->s_log_groups_per_flex)
456                         numblocks += 2 + fs->inode_blocks_per_group;
457
458                 free_blocks += numblocks;
459                 ext2fs_bg_free_blocks_count_set(fs, i, numblocks);
460                 ext2fs_bg_free_inodes_count_set(fs, i, fs->super->s_inodes_per_group);
461                 ext2fs_bg_used_dirs_count_set(fs, i, 0);
462                 ext2fs_group_desc_csum_set(fs, i);
463         }
464         free_blocks &= ~EXT2FS_CLUSTER_MASK(fs);
465         ext2fs_free_blocks_count_set(super, free_blocks);
466
467         c = (char) 255;
468         if (((int) c) == -1) {
469                 super->s_flags |= EXT2_FLAGS_SIGNED_HASH;
470         } else {
471                 super->s_flags |= EXT2_FLAGS_UNSIGNED_HASH;
472         }
473
474         ext2fs_mark_super_dirty(fs);
475         ext2fs_mark_bb_dirty(fs);
476         ext2fs_mark_ib_dirty(fs);
477
478         io_channel_set_blksize(fs->io, fs->blocksize);
479
480         *ret_fs = fs;
481         return 0;
482 cleanup:
483         free(buf);
484         ext2fs_free(fs);
485         return retval;
486 }