Whamcloud - gitweb
Merge branch 'maint' into next
[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         unsigned        reserved_inos;
102         char            *buf = 0;
103         char            c;
104         double          reserved_ratio;
105
106         if (!param || !ext2fs_blocks_count(param))
107                 return EXT2_ET_INVALID_ARGUMENT;
108
109         retval = ext2fs_get_mem(sizeof(struct struct_ext2_filsys), &fs);
110         if (retval)
111                 return retval;
112
113         memset(fs, 0, sizeof(struct struct_ext2_filsys));
114         fs->magic = EXT2_ET_MAGIC_EXT2FS_FILSYS;
115         fs->flags = flags | EXT2_FLAG_RW;
116         fs->umask = 022;
117         fs->default_bitmap_type = EXT2FS_BMAP64_RBTREE;
118 #ifdef WORDS_BIGENDIAN
119         fs->flags |= EXT2_FLAG_SWAP_BYTES;
120 #endif
121         io_flags = IO_FLAG_RW;
122         if (flags & EXT2_FLAG_EXCLUSIVE)
123                 io_flags |= IO_FLAG_EXCLUSIVE;
124         if (flags & EXT2_FLAG_DIRECT_IO)
125                 io_flags |= IO_FLAG_DIRECT_IO;
126         retval = manager->open(name, io_flags, &fs->io);
127         if (retval)
128                 goto cleanup;
129         fs->image_io = fs->io;
130         fs->io->app_data = fs;
131         retval = ext2fs_get_mem(strlen(name)+1, &fs->device_name);
132         if (retval)
133                 goto cleanup;
134
135         strcpy(fs->device_name, name);
136         retval = ext2fs_get_mem(SUPERBLOCK_SIZE, &super);
137         if (retval)
138                 goto cleanup;
139         fs->super = super;
140
141         memset(super, 0, SUPERBLOCK_SIZE);
142
143 #define set_field(field, default) (super->field = param->field ? \
144                                    param->field : (default))
145 #define assign_field(field)     (super->field = param->field)
146
147         super->s_magic = EXT2_SUPER_MAGIC;
148         super->s_state = EXT2_VALID_FS;
149
150         bigalloc_flag = EXT2_HAS_RO_COMPAT_FEATURE(param,
151                                    EXT4_FEATURE_RO_COMPAT_BIGALLOC);
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         if (super->s_feature_incompat & ~EXT2_LIB_FEATURE_INCOMPAT_SUPP) {
177                 retval = EXT2_ET_UNSUPP_FEATURE;
178                 goto cleanup;
179         }
180         if (super->s_feature_ro_compat & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
181                 retval = EXT2_ET_RO_UNSUPP_FEATURE;
182                 goto cleanup;
183         }
184
185         set_field(s_rev_level, EXT2_GOOD_OLD_REV);
186         if (super->s_rev_level >= EXT2_DYNAMIC_REV) {
187                 set_field(s_first_ino, EXT2_GOOD_OLD_FIRST_INO);
188                 set_field(s_inode_size, EXT2_GOOD_OLD_INODE_SIZE);
189                 if (super->s_inode_size >= sizeof(struct ext2_inode_large)) {
190                         int extra_isize = sizeof(struct ext2_inode_large) -
191                                 EXT2_GOOD_OLD_INODE_SIZE;
192                         set_field(s_min_extra_isize, extra_isize);
193                         set_field(s_want_extra_isize, extra_isize);
194                 }
195         } else {
196                 super->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
197                 super->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
198         }
199
200         set_field(s_checkinterval, 0);
201         super->s_mkfs_time = super->s_lastcheck = fs->now ? fs->now : time(NULL);
202
203         super->s_creator_os = CREATOR_OS;
204
205         fs->fragsize = fs->blocksize = EXT2_BLOCK_SIZE(super);
206         fs->cluster_ratio_bits = super->s_log_cluster_size -
207                 super->s_log_block_size;
208
209         if (bigalloc_flag) {
210                 if (param->s_blocks_per_group &&
211                     param->s_clusters_per_group &&
212                     ((param->s_clusters_per_group * EXT2FS_CLUSTER_RATIO(fs)) !=
213                      param->s_blocks_per_group)) {
214                         retval = EXT2_ET_INVALID_ARGUMENT;
215                         goto cleanup;
216                 }
217                 if (param->s_clusters_per_group)
218                         assign_field(s_clusters_per_group);
219                 else if (param->s_blocks_per_group)
220                         super->s_clusters_per_group = 
221                                 param->s_blocks_per_group /
222                                 EXT2FS_CLUSTER_RATIO(fs);
223                 else
224                         super->s_clusters_per_group = fs->blocksize * 8;
225                 if (super->s_clusters_per_group > EXT2_MAX_CLUSTERS_PER_GROUP(super))
226                         super->s_clusters_per_group = EXT2_MAX_CLUSTERS_PER_GROUP(super);
227                 super->s_blocks_per_group = EXT2FS_C2B(fs,
228                                        super->s_clusters_per_group);
229         } else {
230                 set_field(s_blocks_per_group, fs->blocksize * 8);
231                 if (super->s_blocks_per_group > EXT2_MAX_BLOCKS_PER_GROUP(super))
232                         super->s_blocks_per_group = EXT2_MAX_BLOCKS_PER_GROUP(super);
233                 super->s_clusters_per_group = super->s_blocks_per_group;
234         }
235
236         ext2fs_blocks_count_set(super, ext2fs_blocks_count(param) &
237                                 ~((blk64_t) EXT2FS_CLUSTER_MASK(fs)));
238         ext2fs_r_blocks_count_set(super, ext2fs_r_blocks_count(param));
239         if (ext2fs_r_blocks_count(super) >= ext2fs_blocks_count(param)) {
240                 retval = EXT2_ET_INVALID_ARGUMENT;
241                 goto cleanup;
242         }
243
244         set_field(s_mmp_update_interval, 0);
245
246         /*
247          * If we're creating an external journal device, we don't need
248          * to bother with the rest.
249          */
250         if (super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV) {
251                 fs->group_desc_count = 0;
252                 ext2fs_mark_super_dirty(fs);
253                 *ret_fs = fs;
254                 return 0;
255         }
256
257 retry:
258         fs->group_desc_count = (dgrp_t) ext2fs_div64_ceil(
259                 ext2fs_blocks_count(super) - super->s_first_data_block,
260                 EXT2_BLOCKS_PER_GROUP(super));
261         if (fs->group_desc_count == 0) {
262                 retval = EXT2_ET_TOOSMALL;
263                 goto cleanup;
264         }
265
266         if (super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT)
267                 super->s_desc_size = EXT2_MIN_DESC_SIZE_64BIT;
268
269         fs->desc_blocks = ext2fs_div_ceil(fs->group_desc_count,
270                                           EXT2_DESC_PER_BLOCK(super));
271
272         i = fs->blocksize >= 4096 ? 1 : 4096 / fs->blocksize;
273
274         if (super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_64BIT &&
275             (ext2fs_blocks_count(super) / i) > (1ULL << 32))
276                 set_field(s_inodes_count, ~0U);
277         else
278                 set_field(s_inodes_count, ext2fs_blocks_count(super) / i);
279
280         /*
281          * Make sure we have at least EXT2_FIRST_INO + 1 inodes, so
282          * that we have enough inodes for the filesystem(!)
283          */
284         if (super->s_inodes_count < EXT2_FIRST_INODE(super)+1)
285                 super->s_inodes_count = EXT2_FIRST_INODE(super)+1;
286
287         /*
288          * There should be at least as many inodes as the user
289          * requested.  Figure out how many inodes per group that
290          * should be.  But make sure that we don't allocate more than
291          * one bitmap's worth of inodes each group.
292          */
293         ipg = ext2fs_div_ceil(super->s_inodes_count, fs->group_desc_count);
294         if (ipg > fs->blocksize * 8) {
295                 if (!bigalloc_flag && super->s_blocks_per_group >= 256) {
296                         /* Try again with slightly different parameters */
297                         super->s_blocks_per_group -= 8;
298                         ext2fs_blocks_count_set(super,
299                                                 ext2fs_blocks_count(param));
300                         super->s_clusters_per_group = super->s_blocks_per_group;
301                         goto retry;
302                 } else {
303                         retval = EXT2_ET_TOO_MANY_INODES;
304                         goto cleanup;
305                 }
306         }
307
308         if (ipg > (unsigned) EXT2_MAX_INODES_PER_GROUP(super))
309                 ipg = EXT2_MAX_INODES_PER_GROUP(super);
310
311 ipg_retry:
312         super->s_inodes_per_group = ipg;
313
314         /*
315          * Make sure the number of inodes per group completely fills
316          * the inode table blocks in the descriptor.  If not, add some
317          * additional inodes/group.  Waste not, want not...
318          */
319         fs->inode_blocks_per_group = (((super->s_inodes_per_group *
320                                         EXT2_INODE_SIZE(super)) +
321                                        EXT2_BLOCK_SIZE(super) - 1) /
322                                       EXT2_BLOCK_SIZE(super));
323         super->s_inodes_per_group = ((fs->inode_blocks_per_group *
324                                       EXT2_BLOCK_SIZE(super)) /
325                                      EXT2_INODE_SIZE(super));
326         /*
327          * Finally, make sure the number of inodes per group is a
328          * multiple of 8.  This is needed to simplify the bitmap
329          * splicing code.
330          */
331         if (super->s_inodes_per_group < 8)
332                 super->s_inodes_per_group = 8;
333         super->s_inodes_per_group &= ~7;
334         fs->inode_blocks_per_group = (((super->s_inodes_per_group *
335                                         EXT2_INODE_SIZE(super)) +
336                                        EXT2_BLOCK_SIZE(super) - 1) /
337                                       EXT2_BLOCK_SIZE(super));
338
339         /*
340          * adjust inode count to reflect the adjusted inodes_per_group
341          */
342         if ((__u64)super->s_inodes_per_group * fs->group_desc_count > ~0U) {
343                 ipg--;
344                 goto ipg_retry;
345         }
346         super->s_inodes_count = super->s_inodes_per_group *
347                 fs->group_desc_count;
348         super->s_free_inodes_count = super->s_inodes_count;
349
350         /*
351          * check the number of reserved group descriptor table blocks
352          */
353         if (super->s_feature_compat & EXT2_FEATURE_COMPAT_RESIZE_INODE)
354                 rsv_gdt = calc_reserved_gdt_blocks(fs);
355         else
356                 rsv_gdt = 0;
357         set_field(s_reserved_gdt_blocks, rsv_gdt);
358         if (super->s_reserved_gdt_blocks > EXT2_ADDR_PER_BLOCK(super)) {
359                 retval = EXT2_ET_RES_GDT_BLOCKS;
360                 goto cleanup;
361         }
362
363         /*
364          * Calculate the maximum number of bookkeeping blocks per
365          * group.  It includes the superblock, the block group
366          * descriptors, the block bitmap, the inode bitmap, the inode
367          * table, and the reserved gdt blocks.
368          */
369         overhead = (int) (3 + fs->inode_blocks_per_group +
370                           fs->desc_blocks + super->s_reserved_gdt_blocks);
371
372         /* This can only happen if the user requested too many inodes */
373         if (overhead > super->s_blocks_per_group) {
374                 retval = EXT2_ET_TOO_MANY_INODES;
375                 goto cleanup;
376         }
377
378         /*
379          * See if the last group is big enough to support the
380          * necessary data structures.  If not, we need to get rid of
381          * it.  We need to recalculate the overhead for the last block
382          * group, since it might or might not have a superblock
383          * backup.
384          */
385         overhead = (int) (2 + fs->inode_blocks_per_group);
386         if (ext2fs_bg_has_super(fs, fs->group_desc_count - 1))
387                 overhead += 1 + fs->desc_blocks + super->s_reserved_gdt_blocks;
388         rem = ((ext2fs_blocks_count(super) - super->s_first_data_block) %
389                super->s_blocks_per_group);
390         if ((fs->group_desc_count == 1) && rem && (rem < overhead)) {
391                 retval = EXT2_ET_TOOSMALL;
392                 goto cleanup;
393         }
394         if (rem && (rem < overhead+50)) {
395                 ext2fs_blocks_count_set(super, ext2fs_blocks_count(super) -
396                                         rem);
397                 /*
398                  * If blocks count is changed, we need to recalculate
399                  * reserved blocks count not to exceed 50%.
400                  */
401                 reserved_ratio = 100.0 * ext2fs_r_blocks_count(param) /
402                         ext2fs_blocks_count(param);
403                 ext2fs_r_blocks_count_set(super, reserved_ratio *
404                         ext2fs_blocks_count(super) / 100.0);
405
406                 goto retry;
407         }
408
409         /*
410          * At this point we know how big the filesystem will be.  So
411          * we can do any and all allocations that depend on the block
412          * count.
413          */
414
415         retval = ext2fs_get_mem(strlen(fs->device_name) + 80, &buf);
416         if (retval)
417                 goto cleanup;
418
419         strcpy(buf, "block bitmap for ");
420         strcat(buf, fs->device_name);
421         retval = ext2fs_allocate_subcluster_bitmap(fs, buf, &fs->block_map);
422         if (retval)
423                 goto cleanup;
424
425         strcpy(buf, "inode bitmap for ");
426         strcat(buf, fs->device_name);
427         retval = ext2fs_allocate_inode_bitmap(fs, buf, &fs->inode_map);
428         if (retval)
429                 goto cleanup;
430
431         ext2fs_free_mem(&buf);
432
433         retval = ext2fs_get_array(fs->desc_blocks, fs->blocksize,
434                                 &fs->group_desc);
435         if (retval)
436                 goto cleanup;
437
438         memset(fs->group_desc, 0, (size_t) fs->desc_blocks * fs->blocksize);
439
440         /*
441          * Reserve the superblock and group descriptors for each
442          * group, and fill in the correct group statistics for group.
443          * Note that although the block bitmap, inode bitmap, and
444          * inode table have not been allocated (and in fact won't be
445          * by this routine), they are accounted for nevertheless.
446          *
447          * If FLEX_BG meta-data grouping is used, only account for the
448          * superblock and group descriptors (the inode tables and
449          * bitmaps will be accounted for when allocated).
450          */
451         free_blocks = 0;
452         csum_flag = ext2fs_has_group_desc_csum(fs);
453         reserved_inos = super->s_first_ino;
454         for (i = 0; i < fs->group_desc_count; i++) {
455                 /*
456                  * Don't set the BLOCK_UNINIT group for the last group
457                  * because the block bitmap needs to be padded.
458                  */
459                 if (csum_flag) {
460                         if (i != fs->group_desc_count - 1)
461                                 ext2fs_bg_flags_set(fs, i,
462                                                     EXT2_BG_BLOCK_UNINIT);
463                         ext2fs_bg_flags_set(fs, i, EXT2_BG_INODE_UNINIT);
464                         numblocks = super->s_inodes_per_group;
465                         if (reserved_inos) {
466                                 if (numblocks > reserved_inos) {
467                                         numblocks -= reserved_inos;
468                                         reserved_inos = 0;
469                                 } else {
470                                         reserved_inos -= numblocks;
471                                         numblocks = 0;
472                                 }
473                         }
474                         ext2fs_bg_itable_unused_set(fs, i, numblocks);
475                 }
476                 numblocks = ext2fs_reserve_super_and_bgd(fs, i, fs->block_map);
477                 if (fs->super->s_log_groups_per_flex)
478                         numblocks += 2 + fs->inode_blocks_per_group;
479
480                 free_blocks += numblocks;
481                 ext2fs_bg_free_blocks_count_set(fs, i, numblocks);
482                 ext2fs_bg_free_inodes_count_set(fs, i, fs->super->s_inodes_per_group);
483                 ext2fs_bg_used_dirs_count_set(fs, i, 0);
484                 ext2fs_group_desc_csum_set(fs, i);
485         }
486         free_blocks &= ~EXT2FS_CLUSTER_MASK(fs);
487         ext2fs_free_blocks_count_set(super, free_blocks);
488
489         c = (char) 255;
490         if (((int) c) == -1) {
491                 super->s_flags |= EXT2_FLAGS_SIGNED_HASH;
492         } else {
493                 super->s_flags |= EXT2_FLAGS_UNSIGNED_HASH;
494         }
495
496         ext2fs_mark_super_dirty(fs);
497         ext2fs_mark_bb_dirty(fs);
498         ext2fs_mark_ib_dirty(fs);
499
500         io_channel_set_blksize(fs->io, fs->blocksize);
501
502         *ret_fs = fs;
503         return 0;
504 cleanup:
505         free(buf);
506         ext2fs_free(fs);
507         return retval;
508 }