2 * resize2fs.c --- ext2 main routine
4 * Copyright (C) 1997, 1998 by Theodore Ts'o and
7 * Copyright (C) 1999, 2000 by Theosore Ts'o
10 * This file may be redistributed under the terms of the GNU Public
16 * Resizing a filesystem consists of the following phases:
18 * 1. Adjust superblock and write out new parts of the inode
20 * 2. Determine blocks which need to be relocated, and copy the
21 * contents of blocks from their old locations to the new ones.
22 * 3. Scan the inode table, doing the following:
23 * a. If blocks have been moved, update the block
24 * pointers in the inodes and indirect blocks to
25 * point at the new block locations.
26 * b. If parts of the inode table need to be evacuated,
27 * copy inodes from their old locations to their
29 * c. If (b) needs to be done, note which blocks contain
30 * directory information, since we will need to
31 * update the directory information.
32 * 4. Update the directory blocks with the new inode locations.
33 * 5. Move the inode tables, if necessary.
37 #include "resize2fs.h"
40 #ifdef __linux__ /* Kludge for debugging */
41 #define RESIZE2FS_DEBUG
44 static void fix_uninit_block_bitmaps(ext2_filsys fs);
45 static errcode_t adjust_superblock(ext2_resize_t rfs, blk64_t new_size);
46 static errcode_t blocks_to_move(ext2_resize_t rfs);
47 static errcode_t block_mover(ext2_resize_t rfs);
48 static errcode_t inode_scan_and_fix(ext2_resize_t rfs);
49 static errcode_t inode_ref_fix(ext2_resize_t rfs);
50 static errcode_t move_itables(ext2_resize_t rfs);
51 static errcode_t fix_resize_inode(ext2_filsys fs);
52 static errcode_t ext2fs_calculate_summary_stats(ext2_filsys fs);
53 static errcode_t fix_sb_journal_backup(ext2_filsys fs);
54 static errcode_t mark_table_blocks(ext2_filsys fs,
55 ext2fs_block_bitmap bmap);
58 * Some helper CPP macros
60 #define IS_BLOCK_BM(fs, i, blk) ((blk) == ext2fs_block_bitmap_loc((fs),(i)))
61 #define IS_INODE_BM(fs, i, blk) ((blk) == ext2fs_inode_bitmap_loc((fs),(i)))
63 #define IS_INODE_TB(fs, i, blk) (((blk) >= ext2fs_inode_table_loc((fs), (i))) && \
64 ((blk) < (ext2fs_inode_table_loc((fs), (i)) + \
65 (fs)->inode_blocks_per_group)))
67 /* Some bigalloc helper macros which are more succint... */
68 #define B2C(x) EXT2FS_B2C(fs, (x))
69 #define C2B(x) EXT2FS_C2B(fs, (x))
70 #define EQ_CLSTR(x, y) (B2C(x) == B2C(y))
71 #define LE_CLSTR(x, y) (B2C(x) <= B2C(y))
72 #define LT_CLSTR(x, y) (B2C(x) < B2C(y))
73 #define GE_CLSTR(x, y) (B2C(x) >= B2C(y))
74 #define GT_CLSTR(x, y) (B2C(x) > B2C(y))
76 static int lazy_itable_init;
79 * This is the top-level routine which does the dirty deed....
81 errcode_t resize_fs(ext2_filsys fs, blk64_t *new_size, int flags,
82 errcode_t (*progress)(ext2_resize_t rfs, int pass,
84 unsigned long max_val))
88 struct resource_track rtrack, overall_track;
91 * Create the data structure
93 retval = ext2fs_get_mem(sizeof(struct ext2_resize_struct), &rfs);
97 memset(rfs, 0, sizeof(struct ext2_resize_struct));
102 rfs->progress = progress;
104 init_resource_track(&overall_track, "overall resize2fs", fs->io);
105 init_resource_track(&rtrack, "read_bitmaps", fs->io);
106 retval = ext2fs_read_bitmaps(fs);
109 print_resource_track(rfs, &rtrack, fs->io);
111 fs->super->s_state |= EXT2_ERROR_FS;
112 ext2fs_mark_super_dirty(fs);
115 init_resource_track(&rtrack, "fix_uninit_block_bitmaps 1", fs->io);
116 fix_uninit_block_bitmaps(fs);
117 print_resource_track(rfs, &rtrack, fs->io);
118 retval = ext2fs_dup_handle(fs, &rfs->new_fs);
122 init_resource_track(&rtrack, "adjust_superblock", fs->io);
123 retval = adjust_superblock(rfs, *new_size);
126 print_resource_track(rfs, &rtrack, fs->io);
129 init_resource_track(&rtrack, "fix_uninit_block_bitmaps 2", fs->io);
130 fix_uninit_block_bitmaps(rfs->new_fs);
131 print_resource_track(rfs, &rtrack, fs->io);
132 /* Clear the block bitmap uninit flag for the last block group */
133 ext2fs_bg_flags_clear(rfs->new_fs, rfs->new_fs->group_desc_count - 1,
134 EXT2_BG_BLOCK_UNINIT);
136 *new_size = ext2fs_blocks_count(rfs->new_fs->super);
138 init_resource_track(&rtrack, "blocks_to_move", fs->io);
139 retval = blocks_to_move(rfs);
142 print_resource_track(rfs, &rtrack, fs->io);
144 #ifdef RESIZE2FS_DEBUG
145 if (rfs->flags & RESIZE_DEBUG_BMOVE)
146 printf("Number of free blocks: %llu/%llu, Needed: %llu\n",
147 ext2fs_free_blocks_count(rfs->old_fs->super),
148 ext2fs_free_blocks_count(rfs->new_fs->super),
152 init_resource_track(&rtrack, "block_mover", fs->io);
153 retval = block_mover(rfs);
156 print_resource_track(rfs, &rtrack, fs->io);
158 init_resource_track(&rtrack, "inode_scan_and_fix", fs->io);
159 retval = inode_scan_and_fix(rfs);
162 print_resource_track(rfs, &rtrack, fs->io);
164 init_resource_track(&rtrack, "inode_ref_fix", fs->io);
165 retval = inode_ref_fix(rfs);
168 print_resource_track(rfs, &rtrack, fs->io);
170 init_resource_track(&rtrack, "move_itables", fs->io);
171 retval = move_itables(rfs);
174 print_resource_track(rfs, &rtrack, fs->io);
176 init_resource_track(&rtrack, "calculate_summary_stats", fs->io);
177 retval = ext2fs_calculate_summary_stats(rfs->new_fs);
180 print_resource_track(rfs, &rtrack, fs->io);
182 init_resource_track(&rtrack, "fix_resize_inode", fs->io);
183 retval = fix_resize_inode(rfs->new_fs);
186 print_resource_track(rfs, &rtrack, fs->io);
188 init_resource_track(&rtrack, "fix_sb_journal_backup", fs->io);
189 retval = fix_sb_journal_backup(rfs->new_fs);
192 print_resource_track(rfs, &rtrack, fs->io);
194 rfs->new_fs->super->s_state &= ~EXT2_ERROR_FS;
195 rfs->new_fs->flags &= ~EXT2_FLAG_MASTER_SB_ONLY;
197 print_resource_track(rfs, &overall_track, fs->io);
198 retval = ext2fs_close(rfs->new_fs);
204 ext2fs_free(rfs->old_fs);
206 ext2fs_free_mem(&rfs->itable_buf);
207 if (rfs->reserve_blocks)
208 ext2fs_free_block_bitmap(rfs->reserve_blocks);
209 if (rfs->move_blocks)
210 ext2fs_free_block_bitmap(rfs->move_blocks);
211 ext2fs_free_mem(&rfs);
217 ext2fs_free(rfs->new_fs);
219 ext2fs_free_mem(&rfs->itable_buf);
220 ext2fs_free_mem(&rfs);
225 * Clean up the bitmaps for unitialized bitmaps
227 static void fix_uninit_block_bitmaps(ext2_filsys fs)
233 if (!ext2fs_has_group_desc_csum(fs))
236 for (g=0; g < fs->group_desc_count; g++) {
237 if (!(ext2fs_bg_flags_test(fs, g, EXT2_BG_BLOCK_UNINIT)))
240 blk = ext2fs_group_first_block2(fs, g);
241 lblk = ext2fs_group_last_block2(fs, g);
242 ext2fs_unmark_block_bitmap_range2(fs->block_map, blk,
245 ext2fs_reserve_super_and_bgd(fs, g, fs->block_map);
246 ext2fs_mark_block_bitmap2(fs->block_map,
247 ext2fs_block_bitmap_loc(fs, g));
248 ext2fs_mark_block_bitmap2(fs->block_map,
249 ext2fs_inode_bitmap_loc(fs, g));
250 for (i = 0, blk = ext2fs_inode_table_loc(fs, g);
251 i < (unsigned int) fs->inode_blocks_per_group;
253 ext2fs_mark_block_bitmap2(fs->block_map, blk);
257 /* --------------------------------------------------------------------
259 * Resize processing, phase 1.
261 * In this phase we adjust the in-memory superblock information, and
262 * initialize any new parts of the inode table. The new parts of the
263 * inode table are created in virgin disk space, so we can abort here
264 * without any side effects.
265 * --------------------------------------------------------------------
269 * If the group descriptor's bitmap and inode table blocks are valid,
270 * release them in the new filesystem data structure, and mark them as
271 * reserved so the old inode table blocks don't get overwritten.
273 static void free_gdp_blocks(ext2_filsys fs,
274 ext2fs_block_bitmap reserve_blocks,
281 blk = ext2fs_block_bitmap_loc(old_fs, group);
283 (blk < ext2fs_blocks_count(fs->super))) {
284 ext2fs_block_alloc_stats2(fs, blk, -1);
285 ext2fs_mark_block_bitmap2(reserve_blocks, blk);
288 blk = ext2fs_inode_bitmap_loc(old_fs, group);
290 (blk < ext2fs_blocks_count(fs->super))) {
291 ext2fs_block_alloc_stats2(fs, blk, -1);
292 ext2fs_mark_block_bitmap2(reserve_blocks, blk);
295 blk = ext2fs_inode_table_loc(old_fs, group);
297 (blk >= ext2fs_blocks_count(fs->super)))
301 j < fs->inode_blocks_per_group; j++, blk++) {
302 if (blk >= ext2fs_blocks_count(fs->super))
304 ext2fs_block_alloc_stats2(fs, blk, -1);
305 ext2fs_mark_block_bitmap2(reserve_blocks, blk);
310 * This routine is shared by the online and offline resize routines.
311 * All of the information which is adjusted in memory is done here.
313 errcode_t adjust_fs_info(ext2_filsys fs, ext2_filsys old_fs,
314 ext2fs_block_bitmap reserve_blocks, blk64_t new_size)
317 blk64_t overhead = 0;
319 blk64_t blk, group_block;
321 blk64_t old_numblocks, numblocks, adjblocks;
322 unsigned long i, j, old_desc_blocks;
323 unsigned int meta_bg, meta_bg_size;
324 int has_super, csum_flag;
325 unsigned long long new_inodes; /* u64 to check for overflow */
328 ext2fs_blocks_count_set(fs->super, new_size);
331 fs->group_desc_count = ext2fs_div64_ceil(ext2fs_blocks_count(fs->super) -
332 fs->super->s_first_data_block,
333 EXT2_BLOCKS_PER_GROUP(fs->super));
334 if (fs->group_desc_count == 0)
335 return EXT2_ET_TOOSMALL;
336 fs->desc_blocks = ext2fs_div_ceil(fs->group_desc_count,
337 EXT2_DESC_PER_BLOCK(fs->super));
340 * Overhead is the number of bookkeeping blocks per group. It
341 * includes the superblock backup, the group descriptor
342 * backups, the inode bitmap, the block bitmap, and the inode
345 overhead = (int) (2 + fs->inode_blocks_per_group);
347 if (ext2fs_bg_has_super(fs, fs->group_desc_count - 1))
348 overhead += 1 + fs->desc_blocks +
349 fs->super->s_reserved_gdt_blocks;
352 * See if the last group is big enough to support the
353 * necessary data structures. If not, we need to get rid of
356 rem = (ext2fs_blocks_count(fs->super) - fs->super->s_first_data_block) %
357 fs->super->s_blocks_per_group;
358 if ((fs->group_desc_count == 1) && rem && (rem < overhead))
359 return EXT2_ET_TOOSMALL;
360 if ((fs->group_desc_count > 1) && rem && (rem < overhead+50)) {
361 ext2fs_blocks_count_set(fs->super,
362 ext2fs_blocks_count(fs->super) - rem);
366 * Adjust the number of inodes
368 new_inodes =(unsigned long long) fs->super->s_inodes_per_group * fs->group_desc_count;
369 if (new_inodes > ~0U) {
370 fprintf(stderr, _("inodes (%llu) must be less than %u"),
372 return EXT2_ET_TOO_MANY_INODES;
374 fs->super->s_inodes_count = fs->super->s_inodes_per_group *
375 fs->group_desc_count;
378 * Adjust the number of free blocks
380 blk = ext2fs_blocks_count(old_fs->super);
381 if (blk > ext2fs_blocks_count(fs->super))
382 ext2fs_free_blocks_count_set(fs->super,
383 ext2fs_free_blocks_count(fs->super) -
384 (blk - ext2fs_blocks_count(fs->super)));
386 ext2fs_free_blocks_count_set(fs->super,
387 ext2fs_free_blocks_count(fs->super) +
388 (ext2fs_blocks_count(fs->super) - blk));
391 * Adjust the number of reserved blocks
393 percent = (ext2fs_r_blocks_count(old_fs->super) * 100.0) /
394 ext2fs_blocks_count(old_fs->super);
395 ext2fs_r_blocks_count_set(fs->super,
396 (percent * ext2fs_blocks_count(fs->super) /
400 * Adjust the bitmaps for size
402 retval = ext2fs_resize_inode_bitmap2(fs->super->s_inodes_count,
403 fs->super->s_inodes_count,
405 if (retval) goto errout;
407 real_end = (((blk64_t) EXT2_BLOCKS_PER_GROUP(fs->super) *
408 fs->group_desc_count)) - 1 +
409 fs->super->s_first_data_block;
410 retval = ext2fs_resize_block_bitmap2(new_size - 1,
411 real_end, fs->block_map);
412 if (retval) goto errout;
415 * If we are growing the file system, also grow the size of
416 * the reserve_blocks bitmap
418 if (reserve_blocks && new_size > ext2fs_blocks_count(old_fs->super)) {
419 retval = ext2fs_resize_block_bitmap2(new_size - 1,
420 real_end, reserve_blocks);
421 if (retval) goto errout;
425 * Reallocate the group descriptors as necessary.
427 if (old_fs->desc_blocks != fs->desc_blocks) {
428 retval = ext2fs_resize_mem(old_fs->desc_blocks *
430 fs->desc_blocks * fs->blocksize,
434 if (fs->desc_blocks > old_fs->desc_blocks)
435 memset((char *) fs->group_desc +
436 (old_fs->desc_blocks * fs->blocksize), 0,
437 (fs->desc_blocks - old_fs->desc_blocks) *
442 * If the resize_inode feature is set, and we are changing the
443 * number of descriptor blocks, then adjust
444 * s_reserved_gdt_blocks if possible to avoid needing to move
445 * the inode table either now or in the future.
447 if ((fs->super->s_feature_compat &
448 EXT2_FEATURE_COMPAT_RESIZE_INODE) &&
449 (old_fs->desc_blocks != fs->desc_blocks)) {
452 new = ((int) fs->super->s_reserved_gdt_blocks) +
453 (old_fs->desc_blocks - fs->desc_blocks);
456 if (new > (int) fs->blocksize/4)
457 new = fs->blocksize/4;
458 fs->super->s_reserved_gdt_blocks = new;
462 * If we are shrinking the number of block groups, we're done
465 if (old_fs->group_desc_count > fs->group_desc_count) {
467 * Check the block groups that we are chopping off
468 * and free any blocks associated with their metadata
470 for (i = fs->group_desc_count;
471 i < old_fs->group_desc_count; i++)
472 free_gdp_blocks(fs, reserve_blocks, old_fs, i);
478 * Fix the count of the last (old) block group
480 old_numblocks = (ext2fs_blocks_count(old_fs->super) -
481 old_fs->super->s_first_data_block) %
482 old_fs->super->s_blocks_per_group;
484 old_numblocks = old_fs->super->s_blocks_per_group;
485 if (old_fs->group_desc_count == fs->group_desc_count) {
486 numblocks = (ext2fs_blocks_count(fs->super) -
487 fs->super->s_first_data_block) %
488 fs->super->s_blocks_per_group;
490 numblocks = fs->super->s_blocks_per_group;
492 numblocks = fs->super->s_blocks_per_group;
493 i = old_fs->group_desc_count - 1;
494 ext2fs_bg_free_blocks_count_set(fs, i, ext2fs_bg_free_blocks_count(fs, i) + (numblocks - old_numblocks));
495 ext2fs_group_desc_csum_set(fs, i);
498 * If the number of block groups is staying the same, we're
499 * done and can exit now. (If the number block groups is
500 * shrinking, we had exited earlier.)
502 if (old_fs->group_desc_count >= fs->group_desc_count) {
508 * Initialize the new block group descriptors
510 group_block = ext2fs_group_first_block2(fs,
511 old_fs->group_desc_count);
512 csum_flag = ext2fs_has_group_desc_csum(fs);
513 if (access("/sys/fs/ext4/features/lazy_itable_init", F_OK) == 0)
514 lazy_itable_init = 1;
515 if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_META_BG)
516 old_desc_blocks = fs->super->s_first_meta_bg;
518 old_desc_blocks = fs->desc_blocks +
519 fs->super->s_reserved_gdt_blocks;
522 * If we changed the number of block_group descriptor blocks,
523 * we need to make sure they are all marked as reserved in the
524 * file systems's block allocation map.
526 for (i = 0; i < old_fs->group_desc_count; i++)
527 ext2fs_reserve_super_and_bgd(fs, i, fs->block_map);
529 for (i = old_fs->group_desc_count;
530 i < fs->group_desc_count; i++) {
531 memset(ext2fs_group_desc(fs, fs->group_desc, i), 0,
532 sizeof(struct ext2_group_desc));
535 ext2fs_bg_flags_zap(fs, i);
537 ext2fs_bg_flags_set(fs, i, EXT2_BG_INODE_UNINIT);
538 if (!lazy_itable_init)
539 ext2fs_bg_flags_set(fs, i,
540 EXT2_BG_INODE_ZEROED);
541 ext2fs_bg_itable_unused_set(fs, i,
542 fs->super->s_inodes_per_group);
545 numblocks = ext2fs_group_blocks_count(fs, i);
546 if ((i < fs->group_desc_count - 1) && csum_flag)
547 ext2fs_bg_flags_set(fs, i, EXT2_BG_BLOCK_UNINIT);
549 has_super = ext2fs_bg_has_super(fs, i);
551 ext2fs_block_alloc_stats2(fs, group_block, +1);
554 meta_bg_size = EXT2_DESC_PER_BLOCK(fs->super);
555 meta_bg = i / meta_bg_size;
556 if (!(fs->super->s_feature_incompat &
557 EXT2_FEATURE_INCOMPAT_META_BG) ||
558 (meta_bg < fs->super->s_first_meta_bg)) {
560 for (j=0; j < old_desc_blocks; j++)
561 ext2fs_block_alloc_stats2(fs,
562 group_block + 1 + j, +1);
563 adjblocks += old_desc_blocks;
568 if (((i % meta_bg_size) == 0) ||
569 ((i % meta_bg_size) == 1) ||
570 ((i % meta_bg_size) == (meta_bg_size-1)))
571 ext2fs_block_alloc_stats2(fs,
572 group_block + has_super, +1);
575 adjblocks += 2 + fs->inode_blocks_per_group;
577 numblocks -= adjblocks;
578 ext2fs_free_blocks_count_set(fs->super,
579 ext2fs_free_blocks_count(fs->super) - adjblocks);
580 fs->super->s_free_inodes_count +=
581 fs->super->s_inodes_per_group;
582 ext2fs_bg_free_blocks_count_set(fs, i, numblocks);
583 ext2fs_bg_free_inodes_count_set(fs, i,
584 fs->super->s_inodes_per_group);
585 ext2fs_bg_used_dirs_count_set(fs, i, 0);
586 ext2fs_group_desc_csum_set(fs, i);
588 retval = ext2fs_allocate_group_table(fs, i, 0);
589 if (retval) goto errout;
591 group_block += fs->super->s_blocks_per_group;
596 * Mark all of the metadata blocks as reserved so they won't
597 * get allocated by the call to ext2fs_allocate_group_table()
598 * in blocks_to_move(), where we allocate new blocks to
599 * replace those allocation bitmap and inode table blocks
600 * which have to get relocated to make space for an increased
601 * number of the block group descriptors.
604 mark_table_blocks(fs, reserve_blocks);
611 * This routine adjusts the superblock and other data structures, both
612 * in disk as well as in memory...
614 static errcode_t adjust_superblock(ext2_resize_t rfs, blk64_t new_size)
621 unsigned long max_group;
624 ext2fs_mark_super_dirty(fs);
625 ext2fs_mark_bb_dirty(fs);
626 ext2fs_mark_ib_dirty(fs);
628 retval = ext2fs_allocate_block_bitmap(fs, _("reserved blocks"),
629 &rfs->reserve_blocks);
633 retval = adjust_fs_info(fs, rfs->old_fs, rfs->reserve_blocks, new_size);
638 * Check to make sure there are enough inodes
640 if ((rfs->old_fs->super->s_inodes_count -
641 rfs->old_fs->super->s_free_inodes_count) >
642 rfs->new_fs->super->s_inodes_count) {
648 * If we are shrinking the number block groups, we're done and
651 if (rfs->old_fs->group_desc_count > fs->group_desc_count) {
657 * If the number of block groups is staying the same, we're
658 * done and can exit now. (If the number block groups is
659 * shrinking, we had exited earlier.)
661 if (rfs->old_fs->group_desc_count >= fs->group_desc_count) {
667 * If we are using uninit_bg (aka GDT_CSUM) and the kernel
668 * supports lazy inode initialization, we can skip
669 * initializing the inode table.
671 if (lazy_itable_init && ext2fs_has_group_desc_csum(fs)) {
677 * Initialize the inode table
679 retval = ext2fs_get_array(fs->blocksize, fs->inode_blocks_per_group,
684 memset(rfs->itable_buf, 0, fs->blocksize * fs->inode_blocks_per_group);
685 group_block = ext2fs_group_first_block2(fs,
686 rfs->old_fs->group_desc_count);
687 adj = rfs->old_fs->group_desc_count;
688 max_group = fs->group_desc_count - adj;
690 retval = rfs->progress(rfs, E2_RSZ_EXTEND_ITABLE_PASS,
695 for (i = rfs->old_fs->group_desc_count;
696 i < fs->group_desc_count; i++) {
698 * Write out the new inode table
700 retval = io_channel_write_blk64(fs->io,
701 ext2fs_inode_table_loc(fs, i),
702 fs->inode_blocks_per_group,
704 if (retval) goto errout;
706 io_channel_flush(fs->io);
708 retval = rfs->progress(rfs, E2_RSZ_EXTEND_ITABLE_PASS,
709 i - adj + 1, max_group);
713 group_block += fs->super->s_blocks_per_group;
715 io_channel_flush(fs->io);
722 /* --------------------------------------------------------------------
724 * Resize processing, phase 2.
726 * In this phase we adjust determine which blocks need to be moved, in
727 * blocks_to_move(). We then copy the blocks to their ultimate new
728 * destinations using block_mover(). Since we are copying blocks to
729 * their new locations, again during this pass we can abort without
731 * --------------------------------------------------------------------
735 * This helper function creates a block bitmap with all of the
736 * filesystem meta-data blocks.
738 static errcode_t mark_table_blocks(ext2_filsys fs,
739 ext2fs_block_bitmap bmap)
744 for (i = 0; i < fs->group_desc_count; i++) {
745 ext2fs_reserve_super_and_bgd(fs, i, bmap);
748 * Mark the blocks used for the inode table
750 blk = ext2fs_inode_table_loc(fs, i);
752 ext2fs_mark_block_bitmap_range2(bmap, blk,
753 fs->inode_blocks_per_group);
756 * Mark block used for the block bitmap
758 blk = ext2fs_block_bitmap_loc(fs, i);
760 ext2fs_mark_block_bitmap2(bmap, blk);
763 * Mark block used for the inode bitmap
765 blk = ext2fs_inode_bitmap_loc(fs, i);
767 ext2fs_mark_block_bitmap2(bmap, blk);
773 * This function checks to see if a particular block (either a
774 * superblock or a block group descriptor) overlaps with an inode or
775 * block bitmap block, or with the inode table.
777 static void mark_fs_metablock(ext2_resize_t rfs,
778 ext2fs_block_bitmap meta_bmap,
779 int group, blk64_t blk)
781 ext2_filsys fs = rfs->new_fs;
783 ext2fs_mark_block_bitmap2(rfs->reserve_blocks, blk);
784 ext2fs_block_alloc_stats2(fs, blk, +1);
787 * Check to see if we overlap with the inode or block bitmap,
788 * or the inode tables. If not, and the block is in use, then
789 * mark it as a block to be moved.
791 if (IS_BLOCK_BM(fs, group, blk)) {
792 ext2fs_block_bitmap_loc_set(fs, group, 0);
793 rfs->needed_blocks++;
796 if (IS_INODE_BM(fs, group, blk)) {
797 ext2fs_inode_bitmap_loc_set(fs, group, 0);
798 rfs->needed_blocks++;
801 if (IS_INODE_TB(fs, group, blk)) {
802 ext2fs_inode_table_loc_set(fs, group, 0);
803 rfs->needed_blocks++;
806 if (fs->super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_FLEX_BG) {
809 for (i=0; i < rfs->old_fs->group_desc_count; i++) {
810 if (IS_BLOCK_BM(fs, i, blk)) {
811 ext2fs_block_bitmap_loc_set(fs, i, 0);
812 rfs->needed_blocks++;
815 if (IS_INODE_BM(fs, i, blk)) {
816 ext2fs_inode_bitmap_loc_set(fs, i, 0);
817 rfs->needed_blocks++;
820 if (IS_INODE_TB(fs, i, blk)) {
821 ext2fs_inode_table_loc_set(fs, i, 0);
822 rfs->needed_blocks++;
828 if (ext2fs_has_group_desc_csum(fs) &&
829 (ext2fs_bg_flags_test(fs, group, EXT2_BG_BLOCK_UNINIT))) {
831 * If the block bitmap is uninitialized, which means
832 * nothing other than standard metadata in use.
835 } else if (ext2fs_test_block_bitmap2(rfs->old_fs->block_map, blk) &&
836 !ext2fs_test_block_bitmap2(meta_bmap, blk)) {
837 ext2fs_mark_block_bitmap2(rfs->move_blocks, blk);
838 rfs->needed_blocks++;
844 * This routine marks and unmarks reserved blocks in the new block
845 * bitmap. It also determines which blocks need to be moved and
846 * places this information into the move_blocks bitmap.
848 static errcode_t blocks_to_move(ext2_resize_t rfs)
851 dgrp_t i, max_groups, g;
852 blk64_t blk, group_blk;
853 blk64_t old_blocks, new_blocks;
855 unsigned int meta_bg, meta_bg_size;
857 ext2_filsys fs, old_fs;
858 ext2fs_block_bitmap meta_bmap;
862 old_fs = rfs->old_fs;
863 if (ext2fs_blocks_count(old_fs->super) > ext2fs_blocks_count(fs->super))
866 retval = ext2fs_allocate_block_bitmap(fs, _("blocks to be moved"),
871 retval = ext2fs_allocate_block_bitmap(fs, _("meta-data blocks"),
876 retval = mark_table_blocks(old_fs, meta_bmap);
883 * If we're shrinking the filesystem, we need to move any
884 * group's metadata blocks (either allocation bitmaps or the
885 * inode table) which are beyond the end of the new
888 new_size = ext2fs_blocks_count(fs->super);
889 if (new_size < ext2fs_blocks_count(old_fs->super)) {
890 for (g = 0; g < fs->group_desc_count; g++) {
893 * ext2fs_allocate_group_table will re-allocate any
894 * metadata blocks whose location is set to zero.
896 if (ext2fs_block_bitmap_loc(fs, g) >= new_size) {
897 ext2fs_block_bitmap_loc_set(fs, g, 0);
900 if (ext2fs_inode_bitmap_loc(fs, g) >= new_size) {
901 ext2fs_inode_bitmap_loc_set(fs, g, 0);
904 if ((ext2fs_inode_table_loc(fs, g) +
905 fs->inode_blocks_per_group) > new_size) {
906 ext2fs_inode_table_loc_set(fs, g, 0);
911 retval = ext2fs_allocate_group_table(fs, g, 0);
919 * If we're shrinking the filesystem, we need to move all of
920 * the blocks that don't fit any more
922 for (blk = ext2fs_blocks_count(fs->super);
923 blk < ext2fs_blocks_count(old_fs->super); blk++) {
924 g = ext2fs_group_of_blk2(fs, blk);
925 if (ext2fs_has_group_desc_csum(fs) &&
926 ext2fs_bg_flags_test(old_fs, g, EXT2_BG_BLOCK_UNINIT)) {
928 * The block bitmap is uninitialized, so skip
929 * to the next block group.
931 blk = ext2fs_group_first_block2(fs, g+1) - 1;
934 if (ext2fs_test_block_bitmap2(old_fs->block_map, blk) &&
935 !ext2fs_test_block_bitmap2(meta_bmap, blk)) {
936 ext2fs_mark_block_bitmap2(rfs->move_blocks, blk);
937 rfs->needed_blocks++;
939 ext2fs_mark_block_bitmap2(rfs->reserve_blocks, blk);
942 if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_META_BG) {
943 old_blocks = old_fs->super->s_first_meta_bg;
944 new_blocks = fs->super->s_first_meta_bg;
946 old_blocks = old_fs->desc_blocks + old_fs->super->s_reserved_gdt_blocks;
947 new_blocks = fs->desc_blocks + fs->super->s_reserved_gdt_blocks;
950 if (old_blocks == new_blocks) {
955 max_groups = fs->group_desc_count;
956 if (max_groups > old_fs->group_desc_count)
957 max_groups = old_fs->group_desc_count;
958 group_blk = old_fs->super->s_first_data_block;
960 * If we're reducing the number of descriptor blocks, this
961 * makes life easy. :-) We just have to mark some extra
964 if (old_blocks > new_blocks) {
965 for (i = 0; i < max_groups; i++) {
966 if (!ext2fs_bg_has_super(fs, i)) {
967 group_blk += fs->super->s_blocks_per_group;
970 for (blk = group_blk+1+new_blocks;
971 blk < group_blk+1+old_blocks; blk++) {
972 ext2fs_block_alloc_stats2(fs, blk, -1);
973 rfs->needed_blocks--;
975 group_blk += fs->super->s_blocks_per_group;
981 * If we're increasing the number of descriptor blocks, life
982 * gets interesting....
984 meta_bg_size = EXT2_DESC_PER_BLOCK(fs->super);
985 flex_bg = fs->super->s_feature_incompat &
986 EXT4_FEATURE_INCOMPAT_FLEX_BG;
987 /* first reserve all of the existing fs meta blocks */
988 for (i = 0; i < max_groups; i++) {
989 has_super = ext2fs_bg_has_super(fs, i);
991 mark_fs_metablock(rfs, meta_bmap, i, group_blk);
993 meta_bg = i / meta_bg_size;
994 if (!(fs->super->s_feature_incompat &
995 EXT2_FEATURE_INCOMPAT_META_BG) ||
996 (meta_bg < fs->super->s_first_meta_bg)) {
998 for (blk = group_blk+1;
999 blk < group_blk + 1 + new_blocks; blk++)
1000 mark_fs_metablock(rfs, meta_bmap,
1006 if (((i % meta_bg_size) == 0) ||
1007 ((i % meta_bg_size) == 1) ||
1008 ((i % meta_bg_size) == (meta_bg_size-1)))
1009 mark_fs_metablock(rfs, meta_bmap, i,
1010 group_blk + has_super);
1014 * Reserve the existing meta blocks that we know
1015 * aren't to be moved.
1017 * For flex_bg file systems, in order to avoid
1018 * overwriting fs metadata (especially inode table
1019 * blocks) belonging to a different block group when
1020 * we are relocating the inode tables, we need to
1021 * reserve all existing fs metadata blocks.
1023 if (ext2fs_block_bitmap_loc(fs, i))
1024 ext2fs_mark_block_bitmap2(rfs->reserve_blocks,
1025 ext2fs_block_bitmap_loc(fs, i));
1026 else if (flex_bg && i < old_fs->group_desc_count)
1027 ext2fs_mark_block_bitmap2(rfs->reserve_blocks,
1028 ext2fs_block_bitmap_loc(old_fs, i));
1030 if (ext2fs_inode_bitmap_loc(fs, i))
1031 ext2fs_mark_block_bitmap2(rfs->reserve_blocks,
1032 ext2fs_inode_bitmap_loc(fs, i));
1033 else if (flex_bg && i < old_fs->group_desc_count)
1034 ext2fs_mark_block_bitmap2(rfs->reserve_blocks,
1035 ext2fs_inode_bitmap_loc(old_fs, i));
1037 if (ext2fs_inode_table_loc(fs, i))
1038 ext2fs_mark_block_bitmap_range2(rfs->reserve_blocks,
1039 ext2fs_inode_table_loc(fs, i),
1040 fs->inode_blocks_per_group);
1041 else if (flex_bg && i < old_fs->group_desc_count)
1042 ext2fs_mark_block_bitmap_range2(rfs->reserve_blocks,
1043 ext2fs_inode_table_loc(old_fs, i),
1044 old_fs->inode_blocks_per_group);
1046 group_blk += rfs->new_fs->super->s_blocks_per_group;
1049 /* Allocate the missing data structures */
1050 for (i = 0; i < max_groups; i++) {
1051 if (ext2fs_inode_table_loc(fs, i) &&
1052 ext2fs_inode_bitmap_loc(fs, i) &&
1053 ext2fs_block_bitmap_loc(fs, i))
1056 retval = ext2fs_allocate_group_table(fs, i,
1057 rfs->reserve_blocks);
1062 * For those structures that have changed, we need to
1065 if (ext2fs_block_bitmap_loc(old_fs, i) !=
1066 (blk = ext2fs_block_bitmap_loc(fs, i))) {
1067 ext2fs_block_alloc_stats2(fs, blk, +1);
1068 if (ext2fs_test_block_bitmap2(old_fs->block_map, blk) &&
1069 !ext2fs_test_block_bitmap2(meta_bmap, blk))
1070 ext2fs_mark_block_bitmap2(rfs->move_blocks,
1073 if (ext2fs_inode_bitmap_loc(old_fs, i) !=
1074 (blk = ext2fs_inode_bitmap_loc(fs, i))) {
1075 ext2fs_block_alloc_stats2(fs, blk, +1);
1076 if (ext2fs_test_block_bitmap2(old_fs->block_map, blk) &&
1077 !ext2fs_test_block_bitmap2(meta_bmap, blk))
1078 ext2fs_mark_block_bitmap2(rfs->move_blocks,
1083 * The inode table, if we need to relocate it, is
1084 * handled specially. We have to reserve the blocks
1085 * for both the old and the new inode table, since we
1086 * can't have the inode table be destroyed during the
1087 * block relocation phase.
1089 if (ext2fs_inode_table_loc(fs, i) == ext2fs_inode_table_loc(old_fs, i))
1090 continue; /* inode table not moved */
1092 rfs->needed_blocks += fs->inode_blocks_per_group;
1095 * Mark the new inode table as in use in the new block
1096 * allocation bitmap, and move any blocks that might
1099 for (blk = ext2fs_inode_table_loc(fs, i), j=0;
1100 j < fs->inode_blocks_per_group ; j++, blk++) {
1101 ext2fs_block_alloc_stats2(fs, blk, +1);
1102 if (ext2fs_test_block_bitmap2(old_fs->block_map, blk) &&
1103 !ext2fs_test_block_bitmap2(meta_bmap, blk))
1104 ext2fs_mark_block_bitmap2(rfs->move_blocks,
1109 * Make sure the old inode table is reserved in the
1110 * block reservation bitmap.
1112 for (blk = ext2fs_inode_table_loc(rfs->old_fs, i), j=0;
1113 j < fs->inode_blocks_per_group ; j++, blk++)
1114 ext2fs_mark_block_bitmap2(rfs->reserve_blocks, blk);
1120 ext2fs_free_block_bitmap(meta_bmap);
1126 * This helper function tries to allocate a new block. We try to
1127 * avoid hitting the original group descriptor blocks at least at
1128 * first, since we want to make it possible to recover from a badly
1129 * aborted resize operation as much as possible.
1131 * In the future, I may further modify this routine to balance out
1132 * where we get the new blocks across the various block groups.
1133 * Ideally we would allocate blocks that corresponded with the block
1134 * group of the containing inode, and keep contiguous blocks
1135 * together. However, this very difficult to do efficiently, since we
1136 * don't have the necessary information up front.
1140 #define DESPERATION 2
1142 static void init_block_alloc(ext2_resize_t rfs)
1144 rfs->alloc_state = AVOID_OLD;
1145 rfs->new_blk = rfs->new_fs->super->s_first_data_block;
1147 /* HACK for testing */
1148 if (ext2fs_blocks_count(rfs->new_fs->super) >
1149 ext2fs_blocks_count(rfs->old_fs->super))
1150 rfs->new_blk = ext2fs_blocks_count(rfs->old_fs->super);
1154 static blk64_t get_new_block(ext2_resize_t rfs)
1156 ext2_filsys fs = rfs->new_fs;
1159 if (rfs->new_blk >= ext2fs_blocks_count(fs->super)) {
1160 if (rfs->alloc_state == DESPERATION)
1163 #ifdef RESIZE2FS_DEBUG
1164 if (rfs->flags & RESIZE_DEBUG_BMOVE)
1165 printf("Going into desperation mode "
1166 "for block allocations\n");
1168 rfs->alloc_state = DESPERATION;
1169 rfs->new_blk = fs->super->s_first_data_block;
1172 if (ext2fs_test_block_bitmap2(fs->block_map, rfs->new_blk) ||
1173 ext2fs_test_block_bitmap2(rfs->reserve_blocks,
1175 ((rfs->alloc_state == AVOID_OLD) &&
1176 (rfs->new_blk < ext2fs_blocks_count(rfs->old_fs->super)) &&
1177 ext2fs_test_block_bitmap2(rfs->old_fs->block_map,
1182 return rfs->new_blk;
1186 static errcode_t resize2fs_get_alloc_block(ext2_filsys fs, blk64_t goal,
1189 ext2_resize_t rfs = (ext2_resize_t) fs->priv_data;
1192 blk = get_new_block(rfs);
1196 #ifdef RESIZE2FS_DEBUG
1197 if (rfs->flags & 0xF)
1198 printf("get_alloc_block allocating %llu\n", blk);
1201 ext2fs_mark_block_bitmap2(rfs->old_fs->block_map, blk);
1202 ext2fs_mark_block_bitmap2(rfs->new_fs->block_map, blk);
1203 *ret = (blk64_t) blk;
1207 static errcode_t block_mover(ext2_resize_t rfs)
1209 blk64_t blk, old_blk, new_blk;
1210 ext2_filsys fs = rfs->new_fs;
1211 ext2_filsys old_fs = rfs->old_fs;
1216 ext2_badblocks_list badblock_list = 0;
1217 int bb_modified = 0;
1219 fs->get_alloc_block = resize2fs_get_alloc_block;
1220 old_fs->get_alloc_block = resize2fs_get_alloc_block;
1222 retval = ext2fs_read_bb_inode(old_fs, &badblock_list);
1226 new_blk = fs->super->s_first_data_block;
1227 if (!rfs->itable_buf) {
1228 retval = ext2fs_get_array(fs->blocksize,
1229 fs->inode_blocks_per_group,
1234 retval = ext2fs_create_extent_table(&rfs->bmap, 0);
1239 * The first step is to figure out where all of the blocks
1242 to_move = moved = 0;
1243 init_block_alloc(rfs);
1244 for (blk = B2C(old_fs->super->s_first_data_block);
1245 blk < ext2fs_blocks_count(old_fs->super);
1246 blk += EXT2FS_CLUSTER_RATIO(fs)) {
1247 if (!ext2fs_test_block_bitmap2(old_fs->block_map, blk))
1249 if (!ext2fs_test_block_bitmap2(rfs->move_blocks, blk))
1251 if (ext2fs_badblocks_list_test(badblock_list, blk)) {
1252 ext2fs_badblocks_list_del(badblock_list, blk);
1257 new_blk = get_new_block(rfs);
1262 ext2fs_block_alloc_stats2(fs, new_blk, +1);
1263 ext2fs_add_extent_entry(rfs->bmap, B2C(blk), B2C(new_blk));
1269 ext2fs_free_extent_table(rfs->bmap);
1277 * Step two is to actually move the blocks
1279 retval = ext2fs_iterate_extent(rfs->bmap, 0, 0, 0);
1280 if (retval) goto errout;
1282 if (rfs->progress) {
1283 retval = (rfs->progress)(rfs, E2_RSZ_BLOCK_RELOC_PASS,
1289 retval = ext2fs_iterate_extent(rfs->bmap, &old_blk, &new_blk, &size);
1290 if (retval) goto errout;
1293 old_blk = C2B(old_blk);
1294 new_blk = C2B(new_blk);
1296 #ifdef RESIZE2FS_DEBUG
1297 if (rfs->flags & RESIZE_DEBUG_BMOVE)
1298 printf("Moving %llu blocks %llu->%llu\n",
1299 size, old_blk, new_blk);
1303 if (c > fs->inode_blocks_per_group)
1304 c = fs->inode_blocks_per_group;
1305 retval = io_channel_read_blk64(fs->io, old_blk, c,
1307 if (retval) goto errout;
1308 retval = io_channel_write_blk64(fs->io, new_blk, c,
1310 if (retval) goto errout;
1315 if (rfs->progress) {
1316 io_channel_flush(fs->io);
1317 retval = (rfs->progress)(rfs,
1318 E2_RSZ_BLOCK_RELOC_PASS,
1324 io_channel_flush(fs->io);
1328 if (badblock_list) {
1329 if (!retval && bb_modified)
1330 retval = ext2fs_update_bb_inode(old_fs,
1332 ext2fs_badblocks_list_free(badblock_list);
1338 /* --------------------------------------------------------------------
1340 * Resize processing, phase 3
1342 * --------------------------------------------------------------------
1347 * The extent translation table is stored in clusters so we need to
1348 * take special care when mapping a source block number to its
1349 * destination block number.
1351 static __u64 extent_translate(ext2_filsys fs, ext2_extent extent, __u64 old_loc)
1353 __u64 new_block = C2B(ext2fs_extent_translate(extent, B2C(old_loc)));
1356 new_block += old_loc & (EXT2FS_CLUSTER_RATIO(fs) - 1);
1360 struct process_block_struct {
1364 struct ext2_inode * inode;
1371 static int process_block(ext2_filsys fs, blk64_t *block_nr,
1372 e2_blkcnt_t blockcnt,
1373 blk64_t ref_block EXT2FS_ATTR((unused)),
1374 int ref_offset EXT2FS_ATTR((unused)), void *priv_data)
1376 struct process_block_struct *pb;
1378 blk64_t block, new_block;
1381 pb = (struct process_block_struct *) priv_data;
1383 if (pb->rfs->bmap) {
1384 new_block = extent_translate(fs, pb->rfs->bmap, block);
1386 *block_nr = new_block;
1387 ret |= BLOCK_CHANGED;
1389 #ifdef RESIZE2FS_DEBUG
1390 if (pb->rfs->flags & RESIZE_DEBUG_BMOVE)
1391 printf("ino=%u, blockcnt=%lld, %llu->%llu\n",
1392 pb->old_ino, blockcnt, block,
1400 * If we moved inodes and metadata_csum is enabled, we must force the
1401 * extent block to be rewritten with new checksum.
1403 if (EXT2_HAS_RO_COMPAT_FEATURE(fs->super,
1404 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
1406 pb->old_ino != pb->ino)
1407 ret |= BLOCK_CHANGED;
1410 retval = ext2fs_add_dir_block2(fs->dblist, pb->ino,
1411 block, (int) blockcnt);
1423 static errcode_t progress_callback(ext2_filsys fs,
1424 ext2_inode_scan scan EXT2FS_ATTR((unused)),
1425 dgrp_t group, void * priv_data)
1427 ext2_resize_t rfs = (ext2_resize_t) priv_data;
1431 * This check is to protect against old ext2 libraries. It
1432 * shouldn't be needed against new libraries.
1437 if (rfs->progress) {
1438 io_channel_flush(fs->io);
1439 retval = (rfs->progress)(rfs, E2_RSZ_INODE_SCAN_PASS,
1440 group+1, fs->group_desc_count);
1448 static errcode_t migrate_ea_block(ext2_resize_t rfs, ext2_ino_t ino,
1449 struct ext2_inode *inode, int *changed)
1455 /* No EA block or no remapping? Quit early. */
1456 if (ext2fs_file_acl_block(rfs->old_fs, inode) == 0 && !rfs->bmap)
1458 new_block = extent_translate(rfs->old_fs, rfs->bmap,
1459 ext2fs_file_acl_block(rfs->old_fs, inode));
1463 /* Set the new ACL block */
1464 ext2fs_file_acl_block_set(rfs->old_fs, inode, new_block);
1466 /* Update checksum */
1467 if (EXT2_HAS_RO_COMPAT_FEATURE(rfs->new_fs->super,
1468 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
1469 err = ext2fs_get_mem(rfs->old_fs->blocksize, &buf);
1472 rfs->old_fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
1473 err = ext2fs_read_ext_attr3(rfs->old_fs, new_block, buf, ino);
1474 rfs->old_fs->flags &= ~EXT2_FLAG_IGNORE_CSUM_ERRORS;
1477 err = ext2fs_write_ext_attr3(rfs->old_fs, new_block, buf, ino);
1484 ext2fs_free_mem(&buf);
1488 static errcode_t inode_scan_and_fix(ext2_resize_t rfs)
1490 struct process_block_struct pb;
1491 ext2_ino_t ino, new_inode;
1492 struct ext2_inode *inode = NULL;
1493 ext2_inode_scan scan = NULL;
1495 char *block_buf = 0;
1496 ext2_ino_t start_to_move;
1500 if ((rfs->old_fs->group_desc_count <=
1501 rfs->new_fs->group_desc_count) &&
1506 * Save the original size of the old filesystem, and
1507 * temporarily set the size to be the new size if the new size
1508 * is larger. We need to do this to avoid catching an error
1509 * by the block iterator routines
1511 orig_size = ext2fs_blocks_count(rfs->old_fs->super);
1512 if (orig_size < ext2fs_blocks_count(rfs->new_fs->super))
1513 ext2fs_blocks_count_set(rfs->old_fs->super,
1514 ext2fs_blocks_count(rfs->new_fs->super));
1516 retval = ext2fs_open_inode_scan(rfs->old_fs, 0, &scan);
1517 if (retval) goto errout;
1519 retval = ext2fs_init_dblist(rfs->old_fs, 0);
1520 if (retval) goto errout;
1521 retval = ext2fs_get_array(rfs->old_fs->blocksize, 3, &block_buf);
1522 if (retval) goto errout;
1524 start_to_move = (rfs->new_fs->group_desc_count *
1525 rfs->new_fs->super->s_inodes_per_group);
1527 if (rfs->progress) {
1528 retval = (rfs->progress)(rfs, E2_RSZ_INODE_SCAN_PASS,
1529 0, rfs->old_fs->group_desc_count);
1533 ext2fs_set_inode_callback(scan, progress_callback, (void *) rfs);
1537 new_inode = EXT2_FIRST_INODE(rfs->new_fs->super);
1538 inode_size = EXT2_INODE_SIZE(rfs->new_fs->super);
1539 inode = malloc(inode_size);
1545 * First, copy all of the inodes that need to be moved
1546 * elsewhere in the inode table
1549 retval = ext2fs_get_next_inode_full(scan, &ino, inode, inode_size);
1550 if (retval) goto errout;
1554 if (inode->i_links_count == 0 && ino != EXT2_RESIZE_INO)
1555 continue; /* inode not in use */
1557 pb.is_dir = LINUX_S_ISDIR(inode->i_mode);
1560 /* Remap EA block */
1561 retval = migrate_ea_block(rfs, ino, inode, &pb.changed);
1566 if (ino <= start_to_move)
1567 goto remap_blocks; /* Don't need to move inode. */
1570 * Find a new inode. Now that extents and directory blocks
1571 * are tied to the inode number through the checksum, we must
1572 * set up the new inode before we start rewriting blocks.
1574 retval = ext2fs_new_inode(rfs->new_fs, 0, 0, 0, &new_inode);
1578 ext2fs_inode_alloc_stats2(rfs->new_fs, new_inode, +1,
1580 inode->i_ctime = time(0);
1581 retval = ext2fs_write_inode_full(rfs->old_fs, new_inode,
1587 #ifdef RESIZE2FS_DEBUG
1588 if (rfs->flags & RESIZE_DEBUG_INODEMAP)
1589 printf("Inode moved %u->%u\n", ino, new_inode);
1592 retval = ext2fs_create_extent_table(&rfs->imap, 0);
1596 ext2fs_add_extent_entry(rfs->imap, ino, new_inode);
1600 retval = ext2fs_write_inode_full(rfs->old_fs,
1607 * Update inodes to point to new blocks; schedule directory
1608 * blocks for inode remapping. Need to write out dir blocks
1609 * with new inode numbers if we have metadata_csum enabled.
1611 if (ext2fs_inode_has_valid_blocks2(rfs->old_fs, inode) &&
1612 (rfs->bmap || pb.is_dir)) {
1615 pb.has_extents = inode->i_flags & EXT4_EXTENTS_FL;
1616 rfs->old_fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
1617 retval = ext2fs_block_iterate3(rfs->old_fs,
1618 new_inode, 0, block_buf,
1619 process_block, &pb);
1620 rfs->old_fs->flags &= ~EXT2_FLAG_IGNORE_CSUM_ERRORS;
1629 io_channel_flush(rfs->old_fs->io);
1632 ext2fs_blocks_count_set(rfs->old_fs->super, orig_size);
1634 ext2fs_free_extent_table(rfs->bmap);
1638 ext2fs_close_inode_scan(scan);
1640 ext2fs_free_mem(&block_buf);
1645 /* --------------------------------------------------------------------
1647 * Resize processing, phase 4.
1649 * --------------------------------------------------------------------
1655 unsigned int max_dirs;
1659 static int check_and_change_inodes(ext2_ino_t dir,
1660 int entry EXT2FS_ATTR((unused)),
1661 struct ext2_dir_entry *dirent, int offset,
1662 int blocksize EXT2FS_ATTR((unused)),
1663 char *buf EXT2FS_ATTR((unused)),
1666 struct istruct *is = (struct istruct *) priv_data;
1667 struct ext2_inode inode;
1668 ext2_ino_t new_inode;
1672 if (is->rfs->progress && offset == 0) {
1673 io_channel_flush(is->rfs->old_fs->io);
1674 is->err = (is->rfs->progress)(is->rfs,
1675 E2_RSZ_INODE_REF_UPD_PASS,
1676 ++is->num, is->max_dirs);
1678 return DIRENT_ABORT;
1682 * If we have checksums enabled and the inode wasn't present in the
1683 * old fs, then we must rewrite all dir blocks with new checksums.
1685 if (EXT2_HAS_RO_COMPAT_FEATURE(is->rfs->old_fs->super,
1686 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
1687 !ext2fs_test_inode_bitmap2(is->rfs->old_fs->inode_map, dir))
1688 ret |= DIRENT_CHANGED;
1693 new_inode = ext2fs_extent_translate(is->rfs->imap, dirent->inode);
1697 #ifdef RESIZE2FS_DEBUG
1698 if (is->rfs->flags & RESIZE_DEBUG_INODEMAP)
1699 printf("Inode translate (dir=%u, name=%.*s, %u->%u)\n",
1700 dir, ext2fs_dirent_name_len(dirent), dirent->name,
1701 dirent->inode, new_inode);
1704 dirent->inode = new_inode;
1706 /* Update the directory mtime and ctime */
1707 retval = ext2fs_read_inode(is->rfs->old_fs, dir, &inode);
1709 inode.i_mtime = inode.i_ctime = time(0);
1710 is->err = ext2fs_write_inode(is->rfs->old_fs, dir, &inode);
1712 return ret | DIRENT_ABORT;
1715 return ret | DIRENT_CHANGED;
1718 static errcode_t inode_ref_fix(ext2_resize_t rfs)
1727 * Now, we iterate over all of the directories to update the
1731 is.max_dirs = ext2fs_dblist_count2(rfs->old_fs->dblist);
1735 if (rfs->progress) {
1736 retval = (rfs->progress)(rfs, E2_RSZ_INODE_REF_UPD_PASS,
1742 rfs->old_fs->flags |= EXT2_FLAG_IGNORE_CSUM_ERRORS;
1743 retval = ext2fs_dblist_dir_iterate(rfs->old_fs->dblist,
1744 DIRENT_FLAG_INCLUDE_EMPTY, 0,
1745 check_and_change_inodes, &is);
1746 rfs->old_fs->flags &= ~EXT2_FLAG_IGNORE_CSUM_ERRORS;
1754 if (rfs->progress && (is.num < is.max_dirs))
1755 (rfs->progress)(rfs, E2_RSZ_INODE_REF_UPD_PASS,
1756 is.max_dirs, is.max_dirs);
1759 ext2fs_free_extent_table(rfs->imap);
1765 /* --------------------------------------------------------------------
1767 * Resize processing, phase 5.
1769 * In this phase we actually move the inode table around, and then
1770 * update the summary statistics. This is scary, since aborting here
1771 * will potentially scramble the filesystem. (We are moving the
1772 * inode tables around in place, and so the potential for lost data,
1773 * or at the very least scrambling the mapping between filenames and
1774 * inode numbers is very high in case of a power failure here.)
1775 * --------------------------------------------------------------------
1780 * A very scary routine --- this one moves the inode table around!!!
1782 * After this you have to use the rfs->new_fs file handle to read and
1785 static errcode_t move_itables(ext2_resize_t rfs)
1789 dgrp_t i, max_groups;
1790 ext2_filsys fs = rfs->new_fs;
1792 blk64_t old_blk, new_blk, blk;
1794 int j, to_move, moved;
1796 max_groups = fs->group_desc_count;
1797 if (max_groups > rfs->old_fs->group_desc_count)
1798 max_groups = rfs->old_fs->group_desc_count;
1800 size = fs->blocksize * fs->inode_blocks_per_group;
1801 if (!rfs->itable_buf) {
1802 retval = ext2fs_get_mem(size, &rfs->itable_buf);
1808 * Figure out how many inode tables we need to move
1810 to_move = moved = 0;
1811 for (i=0; i < max_groups; i++)
1812 if (ext2fs_inode_table_loc(rfs->old_fs, i) !=
1813 ext2fs_inode_table_loc(fs, i))
1819 if (rfs->progress) {
1820 retval = rfs->progress(rfs, E2_RSZ_MOVE_ITABLE_PASS,
1826 rfs->old_fs->flags |= EXT2_FLAG_MASTER_SB_ONLY;
1828 for (i=0; i < max_groups; i++) {
1829 old_blk = ext2fs_inode_table_loc(rfs->old_fs, i);
1830 new_blk = ext2fs_inode_table_loc(fs, i);
1831 diff = new_blk - old_blk;
1833 #ifdef RESIZE2FS_DEBUG
1834 if (rfs->flags & RESIZE_DEBUG_ITABLEMOVE)
1835 printf("Itable move group %d block %llu->%llu (diff %lld)\n",
1836 i, old_blk, new_blk, diff);
1842 retval = io_channel_read_blk64(fs->io, old_blk,
1843 fs->inode_blocks_per_group,
1848 * The end of the inode table segment often contains
1849 * all zeros, and we're often only moving the inode
1850 * table down a block or two. If so, we can optimize
1851 * things by not rewriting blocks that we know to be zero
1854 for (cp = rfs->itable_buf+size-1, n=0; n < size; n++, cp--)
1857 n = n >> EXT2_BLOCK_SIZE_BITS(fs->super);
1858 #ifdef RESIZE2FS_DEBUG
1859 if (rfs->flags & RESIZE_DEBUG_ITABLEMOVE)
1860 printf("%d blocks of zeros...\n", n);
1862 num = fs->inode_blocks_per_group;
1866 retval = io_channel_write_blk64(fs->io, new_blk,
1867 num, rfs->itable_buf);
1869 io_channel_write_blk64(fs->io, old_blk,
1870 num, rfs->itable_buf);
1874 retval = io_channel_write_blk64(fs->io,
1875 old_blk + fs->inode_blocks_per_group,
1876 diff, (rfs->itable_buf +
1877 (fs->inode_blocks_per_group - diff) *
1883 for (blk = ext2fs_inode_table_loc(rfs->old_fs, i), j=0;
1884 j < fs->inode_blocks_per_group ; j++, blk++)
1885 ext2fs_block_alloc_stats2(fs, blk, -1);
1887 ext2fs_inode_table_loc_set(rfs->old_fs, i, new_blk);
1888 ext2fs_group_desc_csum_set(rfs->old_fs, i);
1889 ext2fs_mark_super_dirty(rfs->old_fs);
1890 ext2fs_flush(rfs->old_fs);
1892 if (rfs->progress) {
1893 retval = rfs->progress(rfs, E2_RSZ_MOVE_ITABLE_PASS,
1899 mark_table_blocks(fs, fs->block_map);
1901 #ifdef RESIZE2FS_DEBUG
1902 if (rfs->flags & RESIZE_DEBUG_ITABLEMOVE)
1903 printf("Inode table move finished.\n");
1912 * Fix the resize inode
1914 static errcode_t fix_resize_inode(ext2_filsys fs)
1916 struct ext2_inode inode;
1918 char *block_buf = NULL;
1920 if (!(fs->super->s_feature_compat &
1921 EXT2_FEATURE_COMPAT_RESIZE_INODE))
1924 retval = ext2fs_get_mem(fs->blocksize, &block_buf);
1925 if (retval) goto errout;
1927 retval = ext2fs_read_inode(fs, EXT2_RESIZE_INO, &inode);
1928 if (retval) goto errout;
1930 ext2fs_iblk_set(fs, &inode, 1);
1932 retval = ext2fs_write_inode(fs, EXT2_RESIZE_INO, &inode);
1933 if (retval) goto errout;
1935 if (!inode.i_block[EXT2_DIND_BLOCK]) {
1937 * Avoid zeroing out block #0; that's rude. This
1938 * should never happen anyway since the filesystem
1939 * should be fsck'ed and we assume it is consistent.
1941 fprintf(stderr, "%s",
1942 _("Should never happen: resize inode corrupt!\n"));
1946 memset(block_buf, 0, fs->blocksize);
1948 retval = io_channel_write_blk64(fs->io, inode.i_block[EXT2_DIND_BLOCK],
1950 if (retval) goto errout;
1952 retval = ext2fs_create_resize_inode(fs);
1958 ext2fs_free_mem(&block_buf);
1963 * Finally, recalculate the summary information
1965 static errcode_t ext2fs_calculate_summary_stats(ext2_filsys fs)
1969 unsigned int group = 0;
1970 unsigned int count = 0;
1971 blk64_t total_blocks_free = 0;
1972 int total_inodes_free = 0;
1975 blk64_t super_blk, old_desc_blk, new_desc_blk;
1976 int old_desc_blocks;
1979 * First calculate the block statistics
1981 uninit = ext2fs_bg_flags_test(fs, group, EXT2_BG_BLOCK_UNINIT);
1982 ext2fs_super_and_bgd_loc2(fs, group, &super_blk, &old_desc_blk,
1984 if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_META_BG)
1985 old_desc_blocks = fs->super->s_first_meta_bg;
1987 old_desc_blocks = fs->desc_blocks +
1988 fs->super->s_reserved_gdt_blocks;
1989 for (blk = B2C(fs->super->s_first_data_block);
1990 blk < ext2fs_blocks_count(fs->super);
1991 blk += EXT2FS_CLUSTER_RATIO(fs)) {
1993 !(EQ_CLSTR(blk, super_blk) ||
1994 ((old_desc_blk && old_desc_blocks &&
1995 GE_CLSTR(blk, old_desc_blk) &&
1996 LT_CLSTR(blk, old_desc_blk + old_desc_blocks))) ||
1997 ((new_desc_blk && EQ_CLSTR(blk, new_desc_blk))) ||
1998 EQ_CLSTR(blk, ext2fs_block_bitmap_loc(fs, group)) ||
1999 EQ_CLSTR(blk, ext2fs_inode_bitmap_loc(fs, group)) ||
2000 ((GE_CLSTR(blk, ext2fs_inode_table_loc(fs, group)) &&
2001 LT_CLSTR(blk, ext2fs_inode_table_loc(fs, group)
2002 + fs->inode_blocks_per_group))))) ||
2003 (!ext2fs_fast_test_block_bitmap2(fs->block_map, blk))) {
2005 total_blocks_free++;
2008 if ((count == fs->super->s_clusters_per_group) ||
2009 EQ_CLSTR(blk, ext2fs_blocks_count(fs->super)-1)) {
2010 ext2fs_bg_free_blocks_count_set(fs, group, group_free);
2011 ext2fs_group_desc_csum_set(fs, group);
2013 if (group >= fs->group_desc_count)
2017 uninit = ext2fs_bg_flags_test(fs, group, EXT2_BG_BLOCK_UNINIT);
2018 ext2fs_super_and_bgd_loc2(fs, group, &super_blk,
2021 if (fs->super->s_feature_incompat &
2022 EXT2_FEATURE_INCOMPAT_META_BG)
2023 old_desc_blocks = fs->super->s_first_meta_bg;
2025 old_desc_blocks = fs->desc_blocks +
2026 fs->super->s_reserved_gdt_blocks;
2029 total_blocks_free = C2B(total_blocks_free);
2030 ext2fs_free_blocks_count_set(fs->super, total_blocks_free);
2033 * Next, calculate the inode statistics
2039 /* Protect loop from wrap-around if s_inodes_count maxed */
2040 uninit = ext2fs_bg_flags_test(fs, group, EXT2_BG_INODE_UNINIT);
2041 for (ino = 1; ino <= fs->super->s_inodes_count && ino > 0; ino++) {
2043 !ext2fs_fast_test_inode_bitmap2(fs->inode_map, ino)) {
2045 total_inodes_free++;
2048 if ((count == fs->super->s_inodes_per_group) ||
2049 (ino == fs->super->s_inodes_count)) {
2050 ext2fs_bg_free_inodes_count_set(fs, group, group_free);
2051 ext2fs_group_desc_csum_set(fs, group);
2053 if (group >= fs->group_desc_count)
2057 uninit = ext2fs_bg_flags_test(fs, group, EXT2_BG_INODE_UNINIT);
2060 fs->super->s_free_inodes_count = total_inodes_free;
2061 ext2fs_mark_super_dirty(fs);
2066 * Journal may have been relocated; update the backup journal blocks
2067 * in the superblock.
2069 static errcode_t fix_sb_journal_backup(ext2_filsys fs)
2072 struct ext2_inode inode;
2074 if (!(fs->super->s_feature_compat & EXT3_FEATURE_COMPAT_HAS_JOURNAL))
2077 /* External journal? Nothing to do. */
2078 if (fs->super->s_journal_dev && !fs->super->s_journal_inum)
2081 retval = ext2fs_read_inode(fs, fs->super->s_journal_inum, &inode);
2084 memcpy(fs->super->s_jnl_blocks, inode.i_block, EXT2_N_BLOCKS*4);
2085 fs->super->s_jnl_blocks[15] = inode.i_size_high;
2086 fs->super->s_jnl_blocks[16] = inode.i_size;
2087 fs->super->s_jnl_backup_type = EXT3_JNL_BACKUP_BLOCKS;
2088 ext2fs_mark_super_dirty(fs);
2092 static int calc_group_overhead(ext2_filsys fs, blk64_t grp,
2093 int old_desc_blocks)
2095 blk64_t super_blk, old_desc_blk, new_desc_blk;
2098 /* inode table blocks plus allocation bitmaps */
2099 overhead = fs->inode_blocks_per_group + 2;
2101 ext2fs_super_and_bgd_loc2(fs, grp, &super_blk,
2102 &old_desc_blk, &new_desc_blk, 0);
2103 if ((grp == 0) || super_blk)
2106 overhead += old_desc_blocks;
2107 else if (new_desc_blk)
2114 * calcluate the minimum number of blocks the given fs can be resized to
2116 blk64_t calculate_minimum_resize_size(ext2_filsys fs, int flags)
2118 ext2_ino_t inode_count;
2120 blk64_t blks_needed, data_blocks;
2121 blk64_t grp, data_needed, last_start;
2122 blk64_t overhead = 0;
2123 int old_desc_blocks;
2124 int extra_groups = 0;
2125 int flexbg_size = 1 << fs->super->s_log_groups_per_flex;
2128 * first figure out how many group descriptors we need to
2129 * handle the number of inodes we have
2131 inode_count = fs->super->s_inodes_count -
2132 fs->super->s_free_inodes_count;
2133 blks_needed = ext2fs_div_ceil(inode_count,
2134 fs->super->s_inodes_per_group) *
2135 EXT2_BLOCKS_PER_GROUP(fs->super);
2136 groups = ext2fs_div64_ceil(blks_needed,
2137 EXT2_BLOCKS_PER_GROUP(fs->super));
2138 #ifdef RESIZE2FS_DEBUG
2139 if (flags & RESIZE_DEBUG_MIN_CALC)
2140 printf("fs has %d inodes, %d groups required.\n",
2141 inode_count, groups);
2145 * number of old-style block group descriptor blocks
2147 if (fs->super->s_feature_incompat & EXT2_FEATURE_INCOMPAT_META_BG)
2148 old_desc_blocks = fs->super->s_first_meta_bg;
2150 old_desc_blocks = fs->desc_blocks +
2151 fs->super->s_reserved_gdt_blocks;
2153 /* calculate how many blocks are needed for data */
2154 data_needed = ext2fs_blocks_count(fs->super) -
2155 ext2fs_free_blocks_count(fs->super);
2157 for (grp = 0; grp < fs->group_desc_count; grp++)
2158 data_needed -= calc_group_overhead(fs, grp, old_desc_blocks);
2159 #ifdef RESIZE2FS_DEBUG
2160 if (flags & RESIZE_DEBUG_MIN_CALC)
2161 printf("fs requires %llu data blocks.\n", data_needed);
2165 * For ext4 we need to allow for up to a flex_bg worth of
2166 * inode tables of slack space so the resize operation can be
2167 * guaranteed to finish.
2169 if (fs->super->s_feature_incompat & EXT4_FEATURE_INCOMPAT_FLEX_BG) {
2170 extra_groups = flexbg_size - (groups & (flexbg_size - 1));
2171 data_needed += fs->inode_blocks_per_group * extra_groups;
2172 extra_groups = groups % flexbg_size;
2176 * figure out how many data blocks we have given the number of groups
2177 * we need for our inodes
2179 data_blocks = groups * EXT2_BLOCKS_PER_GROUP(fs->super);
2181 for (grp = 0; grp < groups; grp++) {
2182 overhead = calc_group_overhead(fs, grp, old_desc_blocks);
2185 * we want to keep track of how much data we can store in
2186 * the groups leading up to the last group so we can determine
2187 * how big the last group needs to be
2189 if (grp != (groups - 1))
2190 last_start += EXT2_BLOCKS_PER_GROUP(fs->super) -
2193 data_blocks -= overhead;
2195 #ifdef RESIZE2FS_DEBUG
2196 if (flags & RESIZE_DEBUG_MIN_CALC)
2197 printf("With %d group(s), we have %llu blocks available.\n",
2198 groups, data_blocks);
2202 * if we need more group descriptors in order to accomodate our data
2203 * then we need to add them here
2205 while (data_needed > data_blocks) {
2206 blk64_t remainder = data_needed - data_blocks;
2209 /* figure out how many more groups we need for the data */
2210 extra_grps = ext2fs_div64_ceil(remainder,
2211 EXT2_BLOCKS_PER_GROUP(fs->super));
2213 data_blocks += extra_grps * EXT2_BLOCKS_PER_GROUP(fs->super);
2215 /* ok we have to account for the last group */
2216 overhead = calc_group_overhead(fs, groups-1, old_desc_blocks);
2217 last_start += EXT2_BLOCKS_PER_GROUP(fs->super) - overhead;
2219 for (grp = groups; grp < groups+extra_grps; grp++) {
2220 overhead = calc_group_overhead(fs, grp,
2224 * again, we need to see how much data we cram into
2225 * all of the groups leading up to the last group
2227 if (grp != (groups + extra_grps - 1))
2228 last_start += EXT2_BLOCKS_PER_GROUP(fs->super)
2231 data_blocks -= overhead;
2234 groups += extra_grps;
2235 extra_groups += extra_grps;
2236 if (fs->super->s_feature_incompat
2237 & EXT4_FEATURE_INCOMPAT_FLEX_BG
2238 && extra_groups > flexbg_size) {
2240 * For ext4 we need to allow for up to a flex_bg worth
2241 * of inode tables of slack space so the resize
2242 * operation can be guaranteed to finish.
2244 extra_groups = flexbg_size -
2245 (groups & (flexbg_size - 1));
2246 data_needed += (fs->inode_blocks_per_group *
2248 extra_groups = groups % flexbg_size;
2250 #ifdef RESIZE2FS_DEBUG
2251 if (flags & RESIZE_DEBUG_MIN_CALC)
2252 printf("Added %d extra group(s), "
2253 "data_needed %llu, data_blocks %llu, "
2254 "last_start %llu\n",
2255 extra_grps, data_needed, data_blocks,
2260 /* now for the fun voodoo */
2261 overhead = calc_group_overhead(fs, groups-1, old_desc_blocks);
2262 #ifdef RESIZE2FS_DEBUG
2263 if (flags & RESIZE_DEBUG_MIN_CALC)
2264 printf("Last group's overhead is %llu\n", overhead);
2268 * if this is the case then the last group is going to have data in it
2269 * so we need to adjust the size of the last group accordingly
2271 if (last_start < data_needed) {
2272 blk64_t remainder = data_needed - last_start;
2274 #ifdef RESIZE2FS_DEBUG
2275 if (flags & RESIZE_DEBUG_MIN_CALC)
2276 printf("Need %llu data blocks in last group\n",
2280 * 50 is a magic number that mkfs/resize uses to see if its
2281 * even worth making/resizing the fs. basically you need to
2282 * have at least 50 blocks in addition to the blocks needed
2283 * for the metadata in the last group
2286 overhead += remainder;
2292 overhead += fs->super->s_first_data_block;
2293 #ifdef RESIZE2FS_DEBUG
2294 if (flags & RESIZE_DEBUG_MIN_CALC)
2295 printf("Final size of last group is %lld\n", overhead);
2299 * since our last group doesn't have to be BLOCKS_PER_GROUP large, we
2300 * only do groups-1, and then add the number of blocks needed to
2301 * handle the group descriptor metadata+data that we need
2303 blks_needed = (groups-1) * EXT2_BLOCKS_PER_GROUP(fs->super);
2304 blks_needed += overhead;
2307 * Make sure blks_needed covers the end of the inode table in
2308 * the last block group.
2310 overhead = ext2fs_inode_table_loc(fs, groups-1) +
2311 fs->inode_blocks_per_group;
2312 if (blks_needed < overhead)
2313 blks_needed = overhead;
2315 #ifdef RESIZE2FS_DEBUG
2316 if (flags & RESIZE_DEBUG_MIN_CALC)
2317 printf("Estimated blocks needed: %llu\n", blks_needed);
2321 * If at this point we've already added up more "needed" than
2322 * the current size, just return current size as minimum.
2324 if (blks_needed >= ext2fs_blocks_count(fs->super))
2325 return ext2fs_blocks_count(fs->super);
2327 * We need to reserve a few extra blocks if extents are
2328 * enabled, in case we need to grow the extent tree. The more
2329 * we shrink the file system, the more space we need.
2331 if (fs->super->s_feature_incompat & EXT3_FEATURE_INCOMPAT_EXTENTS) {
2332 blk64_t safe_margin = (ext2fs_blocks_count(fs->super) -
2334 #ifdef RESIZE2FS_DEBUG
2335 if (flags & RESIZE_DEBUG_MIN_CALC)
2336 printf("Extents safety margin: %llu\n", safe_margin);
2338 blks_needed += safe_margin;