Whamcloud - gitweb
resize2fs, libext2fs: fix bugs in sparse_super2 support
[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         int             has_bg;
102         unsigned        reserved_inos;
103         char            *buf = 0;
104         char            c;
105         double          reserved_ratio;
106
107         if (!param || !ext2fs_blocks_count(param))
108                 return EXT2_ET_INVALID_ARGUMENT;
109
110         retval = ext2fs_get_mem(sizeof(struct struct_ext2_filsys), &fs);
111         if (retval)
112                 return retval;
113
114         memset(fs, 0, sizeof(struct struct_ext2_filsys));
115         fs->magic = EXT2_ET_MAGIC_EXT2FS_FILSYS;
116         fs->flags = flags | EXT2_FLAG_RW;
117         fs->umask = 022;
118         fs->default_bitmap_type = EXT2FS_BMAP64_RBTREE;
119 #ifdef WORDS_BIGENDIAN
120         fs->flags |= EXT2_FLAG_SWAP_BYTES;
121 #endif
122         io_flags = IO_FLAG_RW;
123         if (flags & EXT2_FLAG_EXCLUSIVE)
124                 io_flags |= IO_FLAG_EXCLUSIVE;
125         if (flags & EXT2_FLAG_DIRECT_IO)
126                 io_flags |= IO_FLAG_DIRECT_IO;
127         retval = manager->open(name, io_flags, &fs->io);
128         if (retval)
129                 goto cleanup;
130         fs->image_io = fs->io;
131         fs->io->app_data = fs;
132         retval = ext2fs_get_mem(strlen(name)+1, &fs->device_name);
133         if (retval)
134                 goto cleanup;
135
136         strcpy(fs->device_name, name);
137         retval = ext2fs_get_mem(SUPERBLOCK_SIZE, &super);
138         if (retval)
139                 goto cleanup;
140         fs->super = super;
141
142         memset(super, 0, SUPERBLOCK_SIZE);
143
144 #define set_field(field, default) (super->field = param->field ? \
145                                    param->field : (default))
146 #define assign_field(field)     (super->field = param->field)
147
148         super->s_magic = EXT2_SUPER_MAGIC;
149         super->s_state = EXT2_VALID_FS;
150
151         bigalloc_flag = ext2fs_has_feature_bigalloc(param);
152
153         assign_field(s_log_block_size);
154
155         if (bigalloc_flag) {
156                 set_field(s_log_cluster_size, super->s_log_block_size+4);
157                 if (super->s_log_block_size > super->s_log_cluster_size) {
158                         retval = EXT2_ET_INVALID_ARGUMENT;
159                         goto cleanup;
160                 }
161         } else
162                 super->s_log_cluster_size = super->s_log_block_size;
163
164         set_field(s_first_data_block, super->s_log_cluster_size ? 0 : 1);
165         set_field(s_max_mnt_count, 0);
166         set_field(s_errors, EXT2_ERRORS_DEFAULT);
167         set_field(s_feature_compat, 0);
168         set_field(s_feature_incompat, 0);
169         set_field(s_feature_ro_compat, 0);
170         set_field(s_default_mount_opts, 0);
171         set_field(s_first_meta_bg, 0);
172         set_field(s_raid_stride, 0);            /* default stride size: 0 */
173         set_field(s_raid_stripe_width, 0);      /* default stripe width: 0 */
174         set_field(s_log_groups_per_flex, 0);
175         set_field(s_flags, 0);
176         assign_field(s_backup_bgs[0]);
177         assign_field(s_backup_bgs[1]);
178         if (super->s_feature_incompat & ~EXT2_LIB_FEATURE_INCOMPAT_SUPP) {
179                 retval = EXT2_ET_UNSUPP_FEATURE;
180                 goto cleanup;
181         }
182         if (super->s_feature_ro_compat & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
183                 retval = EXT2_ET_RO_UNSUPP_FEATURE;
184                 goto cleanup;
185         }
186
187         set_field(s_rev_level, EXT2_GOOD_OLD_REV);
188         if (super->s_rev_level >= EXT2_DYNAMIC_REV) {
189                 set_field(s_first_ino, EXT2_GOOD_OLD_FIRST_INO);
190                 set_field(s_inode_size, EXT2_GOOD_OLD_INODE_SIZE);
191                 if (super->s_inode_size >= sizeof(struct ext2_inode_large)) {
192                         int extra_isize = sizeof(struct ext2_inode_large) -
193                                 EXT2_GOOD_OLD_INODE_SIZE;
194                         set_field(s_min_extra_isize, extra_isize);
195                         set_field(s_want_extra_isize, extra_isize);
196                 }
197         } else {
198                 super->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
199                 super->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
200         }
201
202         set_field(s_checkinterval, 0);
203         super->s_mkfs_time = super->s_lastcheck = fs->now ? fs->now : time(NULL);
204
205         super->s_creator_os = CREATOR_OS;
206
207         fs->fragsize = fs->blocksize = EXT2_BLOCK_SIZE(super);
208         fs->cluster_ratio_bits = super->s_log_cluster_size -
209                 super->s_log_block_size;
210
211         if (bigalloc_flag) {
212                 unsigned long long bpg;
213
214                 if (param->s_blocks_per_group &&
215                     param->s_clusters_per_group &&
216                     ((param->s_clusters_per_group * EXT2FS_CLUSTER_RATIO(fs)) !=
217                      param->s_blocks_per_group)) {
218                         retval = EXT2_ET_INVALID_ARGUMENT;
219                         goto cleanup;
220                 }
221                 if (param->s_clusters_per_group)
222                         assign_field(s_clusters_per_group);
223                 else if (param->s_blocks_per_group)
224                         super->s_clusters_per_group = 
225                                 param->s_blocks_per_group /
226                                 EXT2FS_CLUSTER_RATIO(fs);
227                 else if (super->s_log_cluster_size + 15 < 32)
228                         super->s_clusters_per_group = fs->blocksize * 8;
229                 else
230                         super->s_clusters_per_group = (fs->blocksize - 1) * 8;
231                 if (super->s_clusters_per_group > EXT2_MAX_CLUSTERS_PER_GROUP(super))
232                         super->s_clusters_per_group = EXT2_MAX_CLUSTERS_PER_GROUP(super);
233                 bpg = EXT2FS_C2B(fs,
234                         (unsigned long long) super->s_clusters_per_group);
235                 if (bpg >= (((unsigned long long) 1) << 32)) {
236                         retval = EXT2_ET_INVALID_ARGUMENT;
237                         goto cleanup;
238                 }
239                 super->s_blocks_per_group = bpg;
240         } else {
241                 set_field(s_blocks_per_group, fs->blocksize * 8);
242                 if (super->s_blocks_per_group > EXT2_MAX_BLOCKS_PER_GROUP(super))
243                         super->s_blocks_per_group = EXT2_MAX_BLOCKS_PER_GROUP(super);
244                 super->s_clusters_per_group = super->s_blocks_per_group;
245         }
246
247         ext2fs_blocks_count_set(super, ext2fs_blocks_count(param) &
248                                 ~((blk64_t) EXT2FS_CLUSTER_MASK(fs)));
249         ext2fs_r_blocks_count_set(super, ext2fs_r_blocks_count(param));
250         if (ext2fs_r_blocks_count(super) >= ext2fs_blocks_count(param)) {
251                 retval = EXT2_ET_INVALID_ARGUMENT;
252                 goto cleanup;
253         }
254
255         set_field(s_mmp_update_interval, 0);
256
257         /*
258          * If we're creating an external journal device, we don't need
259          * to bother with the rest.
260          */
261         if (ext2fs_has_feature_journal_dev(super)) {
262                 fs->group_desc_count = 0;
263                 ext2fs_mark_super_dirty(fs);
264                 *ret_fs = fs;
265                 return 0;
266         }
267
268 retry:
269         fs->group_desc_count = (dgrp_t) ext2fs_div64_ceil(
270                 ext2fs_blocks_count(super) - super->s_first_data_block,
271                 EXT2_BLOCKS_PER_GROUP(super));
272         if (fs->group_desc_count == 0) {
273                 retval = EXT2_ET_TOOSMALL;
274                 goto cleanup;
275         }
276
277         set_field(s_desc_size,
278                   ext2fs_has_feature_64bit(super) ?
279                   EXT2_MIN_DESC_SIZE_64BIT : 0);
280
281         fs->desc_blocks = ext2fs_div_ceil(fs->group_desc_count,
282                                           EXT2_DESC_PER_BLOCK(super));
283
284         i = fs->blocksize >= 4096 ? 1 : 4096 / fs->blocksize;
285
286         if (ext2fs_has_feature_64bit(super) &&
287             (ext2fs_blocks_count(super) / i) > (1ULL << 32))
288                 set_field(s_inodes_count, ~0U);
289         else
290                 set_field(s_inodes_count, ext2fs_blocks_count(super) / i);
291
292         /*
293          * Make sure we have at least EXT2_FIRST_INO + 1 inodes, so
294          * that we have enough inodes for the filesystem(!)
295          */
296         if (super->s_inodes_count < EXT2_FIRST_INODE(super)+1)
297                 super->s_inodes_count = EXT2_FIRST_INODE(super)+1;
298
299         /*
300          * There should be at least as many inodes as the user
301          * requested.  Figure out how many inodes per group that
302          * should be.  But make sure that we don't allocate more than
303          * one bitmap's worth of inodes each group.
304          */
305         ipg = ext2fs_div_ceil(super->s_inodes_count, fs->group_desc_count);
306         if (ipg > fs->blocksize * 8) {
307                 if (!bigalloc_flag && super->s_blocks_per_group >= 256) {
308                         /* Try again with slightly different parameters */
309                         super->s_blocks_per_group -= 8;
310                         ext2fs_blocks_count_set(super,
311                                                 ext2fs_blocks_count(param));
312                         super->s_clusters_per_group = super->s_blocks_per_group;
313                         goto retry;
314                 } else {
315                         retval = EXT2_ET_TOO_MANY_INODES;
316                         goto cleanup;
317                 }
318         }
319
320         if (ipg > (unsigned) EXT2_MAX_INODES_PER_GROUP(super))
321                 ipg = EXT2_MAX_INODES_PER_GROUP(super);
322
323 ipg_retry:
324         super->s_inodes_per_group = ipg;
325
326         /*
327          * Make sure the number of inodes per group completely fills
328          * the inode table blocks in the descriptor.  If not, add some
329          * additional inodes/group.  Waste not, want not...
330          */
331         fs->inode_blocks_per_group = (((super->s_inodes_per_group *
332                                         EXT2_INODE_SIZE(super)) +
333                                        EXT2_BLOCK_SIZE(super) - 1) /
334                                       EXT2_BLOCK_SIZE(super));
335         super->s_inodes_per_group = ((fs->inode_blocks_per_group *
336                                       EXT2_BLOCK_SIZE(super)) /
337                                      EXT2_INODE_SIZE(super));
338         /*
339          * Finally, make sure the number of inodes per group is a
340          * multiple of 8.  This is needed to simplify the bitmap
341          * splicing code.
342          */
343         if (super->s_inodes_per_group < 8)
344                 super->s_inodes_per_group = 8;
345         super->s_inodes_per_group &= ~7;
346         fs->inode_blocks_per_group = (((super->s_inodes_per_group *
347                                         EXT2_INODE_SIZE(super)) +
348                                        EXT2_BLOCK_SIZE(super) - 1) /
349                                       EXT2_BLOCK_SIZE(super));
350
351         /*
352          * adjust inode count to reflect the adjusted inodes_per_group
353          */
354         if ((__u64)super->s_inodes_per_group * fs->group_desc_count > ~0U) {
355                 ipg--;
356                 goto ipg_retry;
357         }
358         super->s_inodes_count = super->s_inodes_per_group *
359                 fs->group_desc_count;
360         super->s_free_inodes_count = super->s_inodes_count;
361
362         /*
363          * check the number of reserved group descriptor table blocks
364          */
365         if (ext2fs_has_feature_resize_inode(super))
366                 rsv_gdt = calc_reserved_gdt_blocks(fs);
367         else
368                 rsv_gdt = 0;
369         set_field(s_reserved_gdt_blocks, rsv_gdt);
370         if (super->s_reserved_gdt_blocks > EXT2_ADDR_PER_BLOCK(super)) {
371                 retval = EXT2_ET_RES_GDT_BLOCKS;
372                 goto cleanup;
373         }
374
375         /*
376          * Calculate the maximum number of bookkeeping blocks per
377          * group.  It includes the superblock, the block group
378          * descriptors, the block bitmap, the inode bitmap, the inode
379          * table, and the reserved gdt blocks.
380          */
381         overhead = (int) (3 + fs->inode_blocks_per_group +
382                           super->s_reserved_gdt_blocks);
383
384         /* Enable meta_bg if we'd lose more than 3/4 of a BG to GDT blocks. */
385         if (super->s_reserved_gdt_blocks + fs->desc_blocks >
386             super->s_blocks_per_group * 3 / 4)
387                 ext2fs_set_feature_meta_bg(fs->super);
388
389         if (ext2fs_has_feature_meta_bg(fs->super))
390                 overhead++;
391         else
392                 overhead += fs->desc_blocks;
393
394         /* This can only happen if the user requested too many inodes */
395         if (overhead > super->s_blocks_per_group) {
396                 retval = EXT2_ET_TOO_MANY_INODES;
397                 goto cleanup;
398         }
399
400         /*
401          * See if the last group is big enough to support the
402          * necessary data structures.  If not, we need to get rid of
403          * it.  We need to recalculate the overhead for the last block
404          * group, since it might or might not have a superblock
405          * backup.
406          */
407         overhead = (int) (2 + fs->inode_blocks_per_group);
408         has_bg = 0;
409         if (ext2fs_has_feature_sparse_super2(super)) {
410                 /*
411                  * We have to do this manually since
412                  * super->s_backup_bgs hasn't been set up yet.
413                  */
414                 if (fs->group_desc_count == 2)
415                         has_bg = param->s_backup_bgs[0] != 0;
416                 else
417                         has_bg = param->s_backup_bgs[1] != 0;
418         } else
419                 has_bg = ext2fs_bg_has_super(fs, fs->group_desc_count - 1);
420         if (has_bg)
421                 overhead += 1 + fs->desc_blocks + super->s_reserved_gdt_blocks;
422         rem = ((ext2fs_blocks_count(super) - super->s_first_data_block) %
423                super->s_blocks_per_group);
424         if ((fs->group_desc_count == 1) && rem && (rem < overhead)) {
425                 retval = EXT2_ET_TOOSMALL;
426                 goto cleanup;
427         }
428         if (rem && (rem < overhead+50)) {
429                 ext2fs_blocks_count_set(super, ext2fs_blocks_count(super) -
430                                         rem);
431                 /*
432                  * If blocks count is changed, we need to recalculate
433                  * reserved blocks count not to exceed 50%.
434                  */
435                 reserved_ratio = 100.0 * ext2fs_r_blocks_count(param) /
436                         ext2fs_blocks_count(param);
437                 ext2fs_r_blocks_count_set(super, reserved_ratio *
438                         ext2fs_blocks_count(super) / 100.0);
439
440                 goto retry;
441         }
442
443         /*
444          * At this point we know how big the filesystem will be.  So
445          * we can do any and all allocations that depend on the block
446          * count.
447          */
448
449         /* Set up the locations of the backup superblocks */
450         if (ext2fs_has_feature_sparse_super2(super)) {
451                 if (super->s_backup_bgs[0] >= fs->group_desc_count)
452                         super->s_backup_bgs[0] = fs->group_desc_count - 1;
453                 if (super->s_backup_bgs[1] >= fs->group_desc_count)
454                         super->s_backup_bgs[1] = fs->group_desc_count - 1;
455                 if (super->s_backup_bgs[0] == super->s_backup_bgs[1])
456                         super->s_backup_bgs[1] = 0;
457                 if (super->s_backup_bgs[0] > super->s_backup_bgs[1]) {
458                         __u32 t = super->s_backup_bgs[0];
459                         super->s_backup_bgs[0] = super->s_backup_bgs[1];
460                         super->s_backup_bgs[1] = t;
461                 }
462         }
463
464         retval = ext2fs_get_mem(strlen(fs->device_name) + 80, &buf);
465         if (retval)
466                 goto cleanup;
467
468         strcpy(buf, "block bitmap for ");
469         strcat(buf, fs->device_name);
470         retval = ext2fs_allocate_subcluster_bitmap(fs, buf, &fs->block_map);
471         if (retval)
472                 goto cleanup;
473
474         strcpy(buf, "inode bitmap for ");
475         strcat(buf, fs->device_name);
476         retval = ext2fs_allocate_inode_bitmap(fs, buf, &fs->inode_map);
477         if (retval)
478                 goto cleanup;
479
480         ext2fs_free_mem(&buf);
481
482         retval = ext2fs_get_array(fs->desc_blocks, fs->blocksize,
483                                 &fs->group_desc);
484         if (retval)
485                 goto cleanup;
486
487         memset(fs->group_desc, 0, (size_t) fs->desc_blocks * fs->blocksize);
488
489         /*
490          * Reserve the superblock and group descriptors for each
491          * group, and fill in the correct group statistics for group.
492          * Note that although the block bitmap, inode bitmap, and
493          * inode table have not been allocated (and in fact won't be
494          * by this routine), they are accounted for nevertheless.
495          *
496          * If FLEX_BG meta-data grouping is used, only account for the
497          * superblock and group descriptors (the inode tables and
498          * bitmaps will be accounted for when allocated).
499          */
500         free_blocks = 0;
501         csum_flag = ext2fs_has_group_desc_csum(fs);
502         reserved_inos = super->s_first_ino;
503         for (i = 0; i < fs->group_desc_count; i++) {
504                 /*
505                  * Don't set the BLOCK_UNINIT group for the last group
506                  * because the block bitmap needs to be padded.
507                  */
508                 if (csum_flag) {
509                         if (i != fs->group_desc_count - 1)
510                                 ext2fs_bg_flags_set(fs, i,
511                                                     EXT2_BG_BLOCK_UNINIT);
512                         ext2fs_bg_flags_set(fs, i, EXT2_BG_INODE_UNINIT);
513                         numblocks = super->s_inodes_per_group;
514                         if (reserved_inos) {
515                                 if (numblocks > reserved_inos) {
516                                         numblocks -= reserved_inos;
517                                         reserved_inos = 0;
518                                 } else {
519                                         reserved_inos -= numblocks;
520                                         numblocks = 0;
521                                 }
522                         }
523                         ext2fs_bg_itable_unused_set(fs, i, numblocks);
524                 }
525                 numblocks = ext2fs_reserve_super_and_bgd(fs, i, fs->block_map);
526                 if (fs->super->s_log_groups_per_flex)
527                         numblocks += 2 + fs->inode_blocks_per_group;
528
529                 free_blocks += numblocks;
530                 ext2fs_bg_free_blocks_count_set(fs, i, numblocks);
531                 ext2fs_bg_free_inodes_count_set(fs, i, fs->super->s_inodes_per_group);
532                 ext2fs_bg_used_dirs_count_set(fs, i, 0);
533                 ext2fs_group_desc_csum_set(fs, i);
534         }
535         free_blocks &= ~EXT2FS_CLUSTER_MASK(fs);
536         ext2fs_free_blocks_count_set(super, free_blocks);
537
538         c = (char) 255;
539         if (((int) c) == -1) {
540                 super->s_flags |= EXT2_FLAGS_SIGNED_HASH;
541         } else {
542                 super->s_flags |= EXT2_FLAGS_UNSIGNED_HASH;
543         }
544
545         ext2fs_mark_super_dirty(fs);
546         ext2fs_mark_bb_dirty(fs);
547         ext2fs_mark_ib_dirty(fs);
548
549         io_channel_set_blksize(fs->io, fs->blocksize);
550
551         *ret_fs = fs;
552         return 0;
553 cleanup:
554         free(buf);
555         ext2fs_free(fs);
556         return retval;
557 }