2 * Copyright (C) 2010 The Android Open Source Project
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
16 #include "ext4_utils.h"
17 #include "make_ext4fs.h"
18 #include "ext4_extents.h"
20 #include "ext4fixup.h"
22 #include <sparse/sparse.h>
24 #include <sys/types.h>
26 #include <sys/types.h>
34 #if defined(__APPLE__) && defined(__MACH__)
39 /* The inode block count for a file/directory is in units of 512 byte blocks,
40 * _NOT_ the filesystem block size!
42 #define INODE_BLOCK_SIZE 512
44 #define MAX_EXT4_BLOCK_SIZE 4096
46 /* The two modes the recurse_dir() can be in */
47 #define SANITY_CHECK_PASS 1
48 #define MARK_INODE_NUMS 2
49 #define UPDATE_INODE_NUMS 3
51 /* Magic numbers to indicate what state the update process is in */
52 #define MAGIC_STATE_MARKING_INUMS 0x7000151515565512ll
53 #define MAGIC_STATE_UPDATING_INUMS 0x6121131211735123ll
54 #define MAGIC_STATE_UPDATING_SB 0x15e1715151558477ll
56 /* Internal state variables corresponding to the magic numbers */
58 #define STATE_MARKING_INUMS 1
59 #define STATE_UPDATING_INUMS 2
60 #define STATE_UPDATING_SB 3
62 /* Used for automated testing of this programs ability to stop and be restarted wthout error */
63 static int bail_phase = 0;
64 static int bail_loc = 0;
65 static int bail_count = 0;
69 static int verbose = 0;
70 static int no_write = 0;
72 static int new_inodes_per_group = 0;
74 static int no_write_fixup_state = 0;
76 static int compute_new_inum(unsigned int old_inum)
78 unsigned int group, offset;
80 group = (old_inum - 1) / info.inodes_per_group;
81 offset = (old_inum -1) % info.inodes_per_group;
83 return (group * new_inodes_per_group) + offset + 1;
86 static int get_fs_fixup_state(int fd)
88 unsigned long long magic;
92 return no_write_fixup_state;
95 lseek64(fd, 0, SEEK_SET);
96 len = read(fd, &magic, sizeof(magic));
97 if (len != sizeof(magic)) {
98 critical_error("cannot read fixup_state\n");
102 case MAGIC_STATE_MARKING_INUMS:
103 ret = STATE_MARKING_INUMS;
105 case MAGIC_STATE_UPDATING_INUMS:
106 ret = STATE_UPDATING_INUMS;
108 case MAGIC_STATE_UPDATING_SB:
109 ret = STATE_UPDATING_SB;
117 static int set_fs_fixup_state(int fd, int state)
119 unsigned long long magic;
120 struct ext4_super_block sb;
124 no_write_fixup_state = state;
129 case STATE_MARKING_INUMS:
130 magic = MAGIC_STATE_MARKING_INUMS;
132 case STATE_UPDATING_INUMS:
133 magic = MAGIC_STATE_UPDATING_INUMS;
135 case STATE_UPDATING_SB:
136 magic = MAGIC_STATE_UPDATING_SB;
144 lseek64(fd, 0, SEEK_SET);
145 len = write(fd, &magic, sizeof(magic));
146 if (len != sizeof(magic)) {
147 critical_error("cannot write fixup_state\n");
152 /* If we are in the process of updating the filesystem, make it unmountable */
155 /* we are done, so make the filesystem mountable again */
156 sb.s_desc_size &= ~1;
160 write_sb(fd, 1024, &sb);
166 static int read_inode(int fd, unsigned int inum, struct ext4_inode *inode)
168 unsigned int bg_num, bg_offset;
169 off64_t inode_offset;
172 bg_num = (inum-1) / info.inodes_per_group;
173 bg_offset = (inum-1) % info.inodes_per_group;
175 inode_offset = ((unsigned long long)aux_info.bg_desc[bg_num].bg_inode_table * info.block_size) +
176 (bg_offset * info.inode_size);
178 if (lseek64(fd, inode_offset, SEEK_SET) < 0) {
179 critical_error_errno("failed to seek to inode %d\n", inum);
182 len=read(fd, inode, sizeof(*inode));
183 if (len != sizeof(*inode)) {
184 critical_error_errno("failed to read inode %d\n", inum);
190 static int read_block(int fd, unsigned long long block_num, void *block)
195 off = block_num * info.block_size;
197 if (lseek64(fd, off, SEEK_SET) , 0) {
198 critical_error_errno("failed to seek to block %lld\n", block_num);
201 len=read(fd, block, info.block_size);
202 if (len != info.block_size) {
203 critical_error_errno("failed to read block %lld\n", block_num);
209 static int write_block(int fd, unsigned long long block_num, void *block)
218 off = block_num * info.block_size;
220 if (lseek64(fd, off, SEEK_SET) < 0) {
221 critical_error_errno("failed to seek to block %lld\n", block_num);
224 len=write(fd, block, info.block_size);
225 if (len != info.block_size) {
226 critical_error_errno("failed to write block %lld\n", block_num);
232 static void check_inode_bitmap(int fd, unsigned int bg_num)
234 unsigned int inode_bitmap_block_num;
235 unsigned char block[MAX_EXT4_BLOCK_SIZE];
236 int i, bitmap_updated = 0;
238 /* Using the bg_num, aux_info.bg_desc[], info.inodes_per_group and
239 * new_inodes_per_group, retrieve the inode bitmap, and make sure
240 * the bits between the old and new size are clear
242 inode_bitmap_block_num = aux_info.bg_desc[bg_num].bg_inode_bitmap;
244 read_block(fd, inode_bitmap_block_num, block);
246 for (i = info.inodes_per_group; i < new_inodes_per_group; i++) {
247 if (bitmap_get_bit(block, i)) {
248 bitmap_clear_bit(block, i);
253 if (bitmap_updated) {
255 printf("Warning: updated inode bitmap for block group %d\n", bg_num);
257 write_block(fd, inode_bitmap_block_num, block);
263 /* Update the superblock and bgdesc of the specified block group */
264 static int update_superblocks_and_bg_desc(int fd, int state)
267 struct ext4_super_block sb;
268 unsigned int num_block_groups, total_new_inodes;
274 /* Compute how many more inodes are now available */
275 num_block_groups = DIV_ROUND_UP(aux_info.len_blocks, info.blocks_per_group);
276 total_new_inodes = num_block_groups * (new_inodes_per_group - sb.s_inodes_per_group);
279 printf("created %d additional inodes\n", total_new_inodes);
282 /* Update the free inodes count in each block group descriptor */
283 for (i = 0; i < num_block_groups; i++) {
284 if (state == STATE_UPDATING_SB) {
285 aux_info.bg_desc[i].bg_free_inodes_count += (new_inodes_per_group - sb.s_inodes_per_group);
287 check_inode_bitmap(fd, i);
290 /* First some sanity checks */
291 if ((sb.s_inodes_count + total_new_inodes) != (new_inodes_per_group * num_block_groups)) {
292 critical_error("Failed sanity check on new inode count\n");
294 if (new_inodes_per_group % (info.block_size/info.inode_size)) {
295 critical_error("Failed sanity check on new inode per group alignment\n");
298 /* Update the free inodes count in the superblock */
299 sb.s_inodes_count += total_new_inodes;
300 sb.s_free_inodes_count += total_new_inodes;
301 sb.s_inodes_per_group = new_inodes_per_group;
303 for (i = 0; i < aux_info.groups; i++) {
304 if (ext4_bg_has_super_block(i)) {
305 unsigned int sb_offset;
308 /* The first superblock is offset by 1K to leave room for boot sectors */
314 sb.s_block_group_nr = i;
315 /* Don't write out the backup superblocks with the bit set in the s_desc_size
316 * which prevents the filesystem from mounting. The bit for the primary
317 * superblock will be cleared on the final call to set_fs_fixup_state() */
319 sb.s_desc_size &= ~1;
324 (unsigned long long)i
325 * info.blocks_per_group * info.block_size
330 ret = lseek64(fd, ((unsigned long long)i * info.blocks_per_group * info.block_size) +
331 (info.block_size * (aux_info.first_data_block + 1)), SEEK_SET);
333 critical_error_errno("failed to seek to block group descriptors");
336 ret = write(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
338 critical_error_errno("failed to write block group descriptors");
339 if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
340 critical_error("failed to write all of block group descriptors");
343 if ((bail_phase == 4) && ((unsigned int)bail_count == i)) {
344 critical_error("bailing at phase 4\n");
352 static int get_direct_blocks(struct ext4_inode *inode, unsigned long long *block_list,
356 unsigned int ret = 0;
357 unsigned int sectors_per_block;
359 sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
360 while ((i < (inode->i_blocks_lo / sectors_per_block)) && (i < EXT4_NDIR_BLOCKS)) {
361 block_list[i] = inode->i_block[i];
367 if ((inode->i_blocks_lo / sectors_per_block) > EXT4_NDIR_BLOCKS) {
374 static int get_indirect_blocks(int fd, struct ext4_inode *inode,
375 unsigned long long *block_list, unsigned int *count)
378 unsigned int *indirect_block;
379 unsigned int sectors_per_block;
381 sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
383 indirect_block = (unsigned int *)malloc(info.block_size);
384 if (indirect_block == 0) {
385 critical_error("failed to allocate memory for indirect_block\n");
388 read_block(fd, inode->i_block[EXT4_NDIR_BLOCKS], indirect_block);
390 for(i = 0; i < (inode->i_blocks_lo / sectors_per_block - EXT4_NDIR_BLOCKS); i++) {
391 block_list[EXT4_NDIR_BLOCKS+i] = indirect_block[i];
396 free(indirect_block);
401 static int get_block_list_indirect(int fd, struct ext4_inode *inode, unsigned long long *block_list)
403 unsigned int count=0;
405 if (get_direct_blocks(inode, block_list, &count)) {
406 get_indirect_blocks(fd, inode, block_list, &count);
412 static int get_extent_ents(struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
415 struct ext4_extent *extent;
416 off64_t fs_block_num;
418 if (ext_hdr->eh_depth != 0) {
419 critical_error("get_extent_ents called with eh_depth != 0\n");
422 /* The extent entries immediately follow the header, so add 1 to the pointer
423 * and cast it to an extent pointer.
425 extent = (struct ext4_extent *)(ext_hdr + 1);
427 for (i = 0; i < ext_hdr->eh_entries; i++) {
428 fs_block_num = ((off64_t)extent->ee_start_hi << 32) | extent->ee_start_lo;
429 for (j = 0; j < extent->ee_len; j++) {
430 block_list[extent->ee_block+j] = fs_block_num+j;
438 static int get_extent_idx(int fd, struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
441 struct ext4_extent_idx *extent_idx;
442 struct ext4_extent_header *tmp_ext_hdr;
443 off64_t fs_block_num;
444 unsigned char block[MAX_EXT4_BLOCK_SIZE];
447 if (ext_hdr->eh_depth == 0) {
448 critical_error("get_extent_idx called with eh_depth == 0\n");
451 /* The extent entries immediately follow the header, so add 1 to the pointer
452 * and cast it to an extent pointer.
454 extent_idx = (struct ext4_extent_idx *)(ext_hdr + 1);
456 for (i = 0; i < ext_hdr->eh_entries; i++) {
457 fs_block_num = ((off64_t)extent_idx->ei_leaf_hi << 32) | extent_idx->ei_leaf_lo;
458 read_block(fd, fs_block_num, block);
459 tmp_ext_hdr = (struct ext4_extent_header *)block;
461 if (tmp_ext_hdr->eh_depth == 0) {
462 get_extent_ents(tmp_ext_hdr, block_list); /* leaf node, fill in block_list */
464 get_extent_idx(fd, tmp_ext_hdr, block_list); /* recurse down the tree */
471 static int get_block_list_extents(int fd, struct ext4_inode *inode, unsigned long long *block_list)
473 struct ext4_extent_header *extent_hdr;
475 extent_hdr = (struct ext4_extent_header *)inode->i_block;
477 if (extent_hdr->eh_magic != EXT4_EXT_MAGIC) {
478 critical_error("extent header has unexpected magic value 0x%4.4x\n",
479 extent_hdr->eh_magic);
482 if (extent_hdr->eh_depth == 0) {
483 get_extent_ents((struct ext4_extent_header *)inode->i_block, block_list);
487 get_extent_idx(fd, (struct ext4_extent_header *)inode->i_block, block_list);
492 static int is_entry_dir(int fd, struct ext4_dir_entry_2 *dirp, int pass)
494 struct ext4_inode inode;
497 if (dirp->file_type == EXT4_FT_DIR) {
499 } else if (dirp->file_type == EXT4_FT_UNKNOWN) {
500 /* Somebody was too lazy to fill in the dir entry,
501 * so we have to go fetch it from the inode. Grrr.
503 /* if UPDATE_INODE_NUMS pass and the inode high bit is not
504 * set return false so we don't recurse down the tree that is
505 * already updated. Otherwise, fetch inode, and return answer.
507 if ((pass == UPDATE_INODE_NUMS) && !(dirp->inode & 0x80000000)) {
510 read_inode(fd, (dirp->inode & 0x7fffffff), &inode);
511 if (S_ISDIR(inode.i_mode)) {
520 static int recurse_dir(int fd, struct ext4_inode *inode, char *dirbuf, int dirsize, int mode)
522 unsigned long long *block_list;
523 unsigned int num_blocks;
524 struct ext4_dir_entry_2 *dirp, *prev_dirp = 0;
526 unsigned int i, leftover_space, is_dir;
527 struct ext4_inode tmp_inode;
532 case SANITY_CHECK_PASS:
533 case MARK_INODE_NUMS:
534 case UPDATE_INODE_NUMS:
537 critical_error("recurse_dir() called witn unknown mode!\n");
540 if (dirsize % info.block_size) {
541 critical_error("dirsize %d not a multiple of block_size %d. This is unexpected!\n",
542 dirsize, info.block_size);
545 num_blocks = dirsize / info.block_size;
547 block_list = malloc((num_blocks + 1) * sizeof(*block_list));
548 if (block_list == 0) {
549 critical_error("failed to allocate memory for block_list\n");
552 if (inode->i_flags & EXT4_EXTENTS_FL) {
553 get_block_list_extents(fd, inode, block_list);
555 /* A directory that requires doubly or triply indirect blocks in huge indeed,
556 * and will almost certainly not exist, especially since make_ext4fs only creates
557 * directories with extents, and the kernel will too, but check to make sure the
558 * directory is not that big and give an error if so. Our limit is 12 direct blocks,
559 * plus block_size/4 singly indirect blocks, which for a filesystem with 4K blocks
560 * is a directory 1036 blocks long, or 4,243,456 bytes long! Assuming an average
561 * filename length of 20 (which I think is generous) thats 20 + 8 bytes overhead
562 * per entry, or 151,552 entries in the directory!
564 if (num_blocks > (info.block_size / 4 + EXT4_NDIR_BLOCKS)) {
565 critical_error("Non-extent based directory is too big!\n");
567 get_block_list_indirect(fd, inode, block_list);
570 /* Read in all the blocks for this directory */
571 for (i = 0; i < num_blocks; i++) {
572 read_block(fd, block_list[i], dirbuf + (i * info.block_size));
575 dirp = (struct ext4_dir_entry_2 *)dirbuf;
576 while (dirp < (struct ext4_dir_entry_2 *)(dirbuf + dirsize)) {
578 leftover_space = (char *)(dirbuf + dirsize) - (char *)dirp;
579 if (((mode == SANITY_CHECK_PASS) || (mode == UPDATE_INODE_NUMS)) &&
580 (leftover_space <= 8) && prev_dirp) {
581 /* This is a bug in an older version of make_ext4fs, where it
582 * didn't properly include the rest of the block in rec_len.
583 * Update rec_len on the previous entry to include the rest of
584 * the block and exit the loop.
587 printf("fixing up short rec_len for diretory entry for %s\n", name);
589 prev_dirp->rec_len += leftover_space;
593 if (dirp->inode == 0) {
594 /* This is the last entry in the directory */
598 strncpy(name, dirp->name, dirp->name_len);
599 name[dirp->name_len]='\0';
601 /* Only recurse on pass UPDATE_INODE_NUMS if the high bit is set.
602 * Otherwise, this inode entry has already been updated
603 * and we'll do the wrong thing. Also don't recurse on . or ..,
604 * and certainly not on non-directories!
606 /* Hrm, looks like filesystems made by fastboot on stingray set the file_type
607 * flag, but the lost+found directory has the type set to Unknown, which
608 * seems to imply I need to read the inode and get it.
610 is_dir = is_entry_dir(fd, dirp, mode);
611 if ( is_dir && (strcmp(name, ".") && strcmp(name, "..")) &&
612 ((mode == SANITY_CHECK_PASS) || (mode == MARK_INODE_NUMS) ||
613 ((mode == UPDATE_INODE_NUMS) && (dirp->inode & 0x80000000))) ) {
614 /* A directory! Recurse! */
615 read_inode(fd, dirp->inode & 0x7fffffff, &tmp_inode);
617 if (!S_ISDIR(tmp_inode.i_mode)) {
618 critical_error("inode %d for name %s does not point to a directory\n",
619 dirp->inode & 0x7fffffff, name);
622 printf("inode %d %s use extents\n", dirp->inode & 0x7fffffff,
623 (tmp_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
626 tmp_dirsize = tmp_inode.i_blocks_lo * INODE_BLOCK_SIZE;
628 printf("dir size = %d bytes\n", tmp_dirsize);
631 tmp_dirbuf = malloc(tmp_dirsize);
632 if (tmp_dirbuf == 0) {
633 critical_error("failed to allocate memory for tmp_dirbuf\n");
636 recurse_dir(fd, &tmp_inode, tmp_dirbuf, tmp_dirsize, mode);
643 printf("Directory %s\n", name);
645 printf("Non-directory %s\n", name);
649 /* Process entry based on current mode. Either set high bit or change inode number */
650 if (mode == MARK_INODE_NUMS) {
651 dirp->inode |= 0x80000000;
652 } else if (mode == UPDATE_INODE_NUMS) {
653 if (dirp->inode & 0x80000000) {
654 dirp->inode = compute_new_inum(dirp->inode & 0x7fffffff);
658 if ((bail_phase == mode) && (bail_loc == 1) && (bail_count == count)) {
659 critical_error("Bailing at phase %d, loc 1 and count %d\n", mode, count);
662 /* Point dirp at the next entry */
664 dirp = (struct ext4_dir_entry_2*)((char *)dirp + dirp->rec_len);
667 /* Write out all the blocks for this directory */
668 for (i = 0; i < num_blocks; i++) {
669 write_block(fd, block_list[i], dirbuf + (i * info.block_size));
670 if ((bail_phase == mode) && (bail_loc == 2) && (bail_count <= count)) {
671 critical_error("Bailing at phase %d, loc 2 and count %d\n", mode, count);
680 int ext4fixup(char *fsdev)
682 return ext4fixup_internal(fsdev, 0, 0, 0, 0, 0);
685 int ext4fixup_internal(char *fsdev, int v_flag, int n_flag,
686 int stop_phase, int stop_loc, int stop_count)
689 struct ext4_inode root_inode;
690 unsigned int dirsize;
693 if (setjmp(setjmp_env))
694 return EXIT_FAILURE; /* Handle a call to longjmp() */
699 bail_phase = stop_phase;
701 bail_count = stop_count;
703 fd = open(fsdev, O_RDWR);
706 critical_error_errno("failed to open filesystem image");
708 read_ext(fd, verbose);
710 if (info.feat_incompat & EXT4_FEATURE_INCOMPAT_RECOVER) {
711 critical_error("Filesystem needs recovery first, mount and unmount to do that\n");
714 /* Clear the low bit which is set while this tool is in progress.
715 * If the tool crashes, it will still be set when we restart.
716 * The low bit is set to make the filesystem unmountable while
717 * it is being fixed up. Also allow 0, which means the old ext2
720 if (((aux_info.sb->s_desc_size & ~1) != sizeof(struct ext2_group_desc)) &&
721 ((aux_info.sb->s_desc_size & ~1) != 0))
722 critical_error("error: bg_desc_size != sizeof(struct ext2_group_desc)\n");
724 if ((info.feat_incompat & EXT4_FEATURE_INCOMPAT_FILETYPE) == 0) {
725 critical_error("Expected filesystem to have filetype flag set\n");
728 #if 0 // If we have to fix the directory rec_len issue, we can't use this check
729 /* Check to see if the inodes/group is copacetic */
730 if (info.inodes_per_blockgroup % (info.block_size/info.inode_size) == 0) {
731 /* This filesystem has either already been updated, or was
735 printf("%s: filesystem correct, no work to do\n", me);
741 /* Compute what the new value of inodes_per_blockgroup will be when we're done */
742 new_inodes_per_group=EXT4_ALIGN(info.inodes_per_group,(info.block_size/info.inode_size));
744 read_inode(fd, EXT4_ROOT_INO, &root_inode);
746 if (!S_ISDIR(root_inode.i_mode)) {
747 critical_error("root inode %d does not point to a directory\n", EXT4_ROOT_INO);
750 printf("inode %d %s use extents\n", EXT4_ROOT_INO,
751 (root_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
754 dirsize = root_inode.i_blocks_lo * INODE_BLOCK_SIZE;
756 printf("root dir size = %d bytes\n", dirsize);
759 dirbuf = malloc(dirsize);
761 critical_error("failed to allocate memory for dirbuf\n");
764 /* Perform a sanity check pass first, try to catch any errors that will occur
765 * before we actually change anything, so we don't leave a filesystem in a
766 * corrupted, unrecoverable state. Set no_write, make it quiet, and do a recurse
767 * pass and a update_superblock pass. Set flags back to requested state when done.
768 * Only perform sanity check if the state is unset. If the state is _NOT_ unset,
769 * then the tool has already been run and interrupted, and it presumably ran and
770 * passed sanity checked before it got interrupted. It is _NOT_ safe to run sanity
771 * check if state is unset because it assumes inodes are to be computed using the
772 * old inodes/group, but some inode numbers may be updated to the new number.
774 if (get_fs_fixup_state(fd) == STATE_UNSET) {
777 recurse_dir(fd, &root_inode, dirbuf, dirsize, SANITY_CHECK_PASS);
778 update_superblocks_and_bg_desc(fd, STATE_UNSET);
782 set_fs_fixup_state(fd, STATE_MARKING_INUMS);
785 if (get_fs_fixup_state(fd) == STATE_MARKING_INUMS) {
786 count = 0; /* Reset debugging counter */
787 if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, MARK_INODE_NUMS)) {
788 set_fs_fixup_state(fd, STATE_UPDATING_INUMS);
792 if (get_fs_fixup_state(fd) == STATE_UPDATING_INUMS) {
793 count = 0; /* Reset debugging counter */
794 if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, UPDATE_INODE_NUMS)) {
795 set_fs_fixup_state(fd, STATE_UPDATING_SB);
799 if (get_fs_fixup_state(fd) == STATE_UPDATING_SB) {
800 /* set the new inodes/blockgroup number,
801 * and sets the state back to 0.
803 if (!update_superblocks_and_bg_desc(fd, STATE_UPDATING_SB)) {
804 set_fs_fixup_state(fd, STATE_UNSET);