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 "ext4_extents.h"
19 #include "ext4fixup.h"
21 #include <sparse/sparse.h>
23 #include <sys/types.h>
25 #include <sys/types.h>
30 #if defined(__APPLE__) && defined(__MACH__)
35 /* The inode block count for a file/directory is in units of 512 byte blocks,
36 * _NOT_ the filesystem block size!
38 #define INODE_BLOCK_SIZE 512
40 #define MAX_EXT4_BLOCK_SIZE 4096
42 /* The two modes the recurse_dir() can be in */
43 #define SANITY_CHECK_PASS 1
44 #define MARK_INODE_NUMS 2
45 #define UPDATE_INODE_NUMS 3
47 /* Magic numbers to indicate what state the update process is in */
48 #define MAGIC_STATE_MARKING_INUMS 0x7000151515565512ll
49 #define MAGIC_STATE_UPDATING_INUMS 0x6121131211735123ll
50 #define MAGIC_STATE_UPDATING_SB 0x15e1715151558477ll
52 /* Internal state variables corresponding to the magic numbers */
54 #define STATE_MARKING_INUMS 1
55 #define STATE_UPDATING_INUMS 2
56 #define STATE_UPDATING_SB 3
58 /* Used for automated testing of this programs ability to stop and be restarted wthout error */
59 static int bail_phase = 0;
60 static int bail_loc = 0;
61 static int bail_count = 0;
65 static int verbose = 0;
66 static int no_write = 0;
68 static int new_inodes_per_group = 0;
70 static int no_write_fixup_state = 0;
72 static int compute_new_inum(unsigned int old_inum)
74 unsigned int group, offset;
76 group = (old_inum - 1) / info.inodes_per_group;
77 offset = (old_inum -1) % info.inodes_per_group;
79 return (group * new_inodes_per_group) + offset + 1;
82 static int get_fs_fixup_state(int fd)
84 unsigned long long magic;
88 return no_write_fixup_state;
91 lseek64(fd, 0, SEEK_SET);
92 len = read(fd, &magic, sizeof(magic));
93 if (len != sizeof(magic)) {
94 critical_error("cannot read fixup_state\n");
98 case MAGIC_STATE_MARKING_INUMS:
99 ret = STATE_MARKING_INUMS;
101 case MAGIC_STATE_UPDATING_INUMS:
102 ret = STATE_UPDATING_INUMS;
104 case MAGIC_STATE_UPDATING_SB:
105 ret = STATE_UPDATING_SB;
113 static int set_fs_fixup_state(int fd, int state)
115 unsigned long long magic;
116 struct ext4_super_block sb;
120 no_write_fixup_state = state;
125 case STATE_MARKING_INUMS:
126 magic = MAGIC_STATE_MARKING_INUMS;
128 case STATE_UPDATING_INUMS:
129 magic = MAGIC_STATE_UPDATING_INUMS;
131 case STATE_UPDATING_SB:
132 magic = MAGIC_STATE_UPDATING_SB;
140 lseek64(fd, 0, SEEK_SET);
141 len = write(fd, &magic, sizeof(magic));
142 if (len != sizeof(magic)) {
143 critical_error("cannot write fixup_state\n");
148 /* If we are in the process of updating the filesystem, make it unmountable */
151 /* we are done, so make the filesystem mountable again */
152 sb.s_desc_size &= ~1;
156 write_sb(fd, 1024, &sb);
162 static int read_inode(int fd, unsigned int inum, struct ext4_inode *inode)
164 unsigned int bg_num, bg_offset;
165 off64_t inode_offset;
168 bg_num = (inum-1) / info.inodes_per_group;
169 bg_offset = (inum-1) % info.inodes_per_group;
171 inode_offset = ((unsigned long long)aux_info.bg_desc[bg_num].bg_inode_table * info.block_size) +
172 (bg_offset * info.inode_size);
174 if (lseek64(fd, inode_offset, SEEK_SET) < 0) {
175 critical_error_errno("failed to seek to inode %d\n", inum);
178 len=read(fd, inode, sizeof(*inode));
179 if (len != sizeof(*inode)) {
180 critical_error_errno("failed to read inode %d\n", inum);
186 static int read_block(int fd, unsigned long long block_num, void *block)
191 off = block_num * info.block_size;
193 if (lseek64(fd, off, SEEK_SET) , 0) {
194 critical_error_errno("failed to seek to block %lld\n", block_num);
197 len=read(fd, block, info.block_size);
198 if (len != info.block_size) {
199 critical_error_errno("failed to read block %lld\n", block_num);
205 static int write_block(int fd, unsigned long long block_num, void *block)
214 off = block_num * info.block_size;
216 if (lseek64(fd, off, SEEK_SET) < 0) {
217 critical_error_errno("failed to seek to block %lld\n", block_num);
220 len=write(fd, block, info.block_size);
221 if (len != info.block_size) {
222 critical_error_errno("failed to write block %lld\n", block_num);
228 static void check_inode_bitmap(int fd, unsigned int bg_num)
230 unsigned int inode_bitmap_block_num;
231 unsigned char block[MAX_EXT4_BLOCK_SIZE];
232 int i, bitmap_updated = 0;
234 /* Using the bg_num, aux_info.bg_desc[], info.inodes_per_group and
235 * new_inodes_per_group, retrieve the inode bitmap, and make sure
236 * the bits between the old and new size are clear
238 inode_bitmap_block_num = aux_info.bg_desc[bg_num].bg_inode_bitmap;
240 read_block(fd, inode_bitmap_block_num, block);
242 for (i = info.inodes_per_group; i < new_inodes_per_group; i++) {
243 if (bitmap_get_bit(block, i)) {
244 bitmap_clear_bit(block, i);
249 if (bitmap_updated) {
251 printf("Warning: updated inode bitmap for block group %d\n", bg_num);
253 write_block(fd, inode_bitmap_block_num, block);
259 /* Update the superblock and bgdesc of the specified block group */
260 static int update_superblocks_and_bg_desc(int fd, int state)
263 struct ext4_super_block sb;
264 unsigned int num_block_groups, total_new_inodes;
270 /* Compute how many more inodes are now available */
271 num_block_groups = DIV_ROUND_UP(aux_info.len_blocks, info.blocks_per_group);
272 total_new_inodes = num_block_groups * (new_inodes_per_group - sb.s_inodes_per_group);
275 printf("created %d additional inodes\n", total_new_inodes);
278 /* Update the free inodes count in each block group descriptor */
279 for (i = 0; i < num_block_groups; i++) {
280 if (state == STATE_UPDATING_SB) {
281 aux_info.bg_desc[i].bg_free_inodes_count += (new_inodes_per_group - sb.s_inodes_per_group);
283 check_inode_bitmap(fd, i);
286 /* First some sanity checks */
287 if ((sb.s_inodes_count + total_new_inodes) != (new_inodes_per_group * num_block_groups)) {
288 critical_error("Failed sanity check on new inode count\n");
290 if (new_inodes_per_group % (info.block_size/info.inode_size)) {
291 critical_error("Failed sanity check on new inode per group alignment\n");
294 /* Update the free inodes count in the superblock */
295 sb.s_inodes_count += total_new_inodes;
296 sb.s_free_inodes_count += total_new_inodes;
297 sb.s_inodes_per_group = new_inodes_per_group;
299 for (i = 0; i < aux_info.groups; i++) {
300 if (ext4_bg_has_super_block(i)) {
301 unsigned int sb_offset;
304 /* The first superblock is offset by 1K to leave room for boot sectors */
310 sb.s_block_group_nr = i;
311 /* Don't write out the backup superblocks with the bit set in the s_desc_size
312 * which prevents the filesystem from mounting. The bit for the primary
313 * superblock will be cleared on the final call to set_fs_fixup_state() */
315 sb.s_desc_size &= ~1;
320 (unsigned long long)i
321 * info.blocks_per_group * info.block_size
326 ret = lseek64(fd, ((unsigned long long)i * info.blocks_per_group * info.block_size) +
327 (info.block_size * (aux_info.first_data_block + 1)), SEEK_SET);
329 critical_error_errno("failed to seek to block group descriptors");
332 ret = write(fd, aux_info.bg_desc, info.block_size * aux_info.bg_desc_blocks);
334 critical_error_errno("failed to write block group descriptors");
335 if (ret != (int)info.block_size * (int)aux_info.bg_desc_blocks)
336 critical_error("failed to write all of block group descriptors");
339 if ((bail_phase == 4) && ((unsigned int)bail_count == i)) {
340 critical_error("bailing at phase 4\n");
348 static int get_direct_blocks(struct ext4_inode *inode, unsigned long long *block_list,
352 unsigned int ret = 0;
353 unsigned int sectors_per_block;
355 sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
356 while ((i < (inode->i_blocks_lo / sectors_per_block)) && (i < EXT4_NDIR_BLOCKS)) {
357 block_list[i] = inode->i_block[i];
363 if ((inode->i_blocks_lo / sectors_per_block) > EXT4_NDIR_BLOCKS) {
370 static int get_indirect_blocks(int fd, struct ext4_inode *inode,
371 unsigned long long *block_list, unsigned int *count)
374 unsigned int *indirect_block;
375 unsigned int sectors_per_block;
377 sectors_per_block = info.block_size / INODE_BLOCK_SIZE;
379 indirect_block = (unsigned int *)malloc(info.block_size);
380 if (indirect_block == 0) {
381 critical_error("failed to allocate memory for indirect_block\n");
384 read_block(fd, inode->i_block[EXT4_NDIR_BLOCKS], indirect_block);
386 for(i = 0; i < (inode->i_blocks_lo / sectors_per_block - EXT4_NDIR_BLOCKS); i++) {
387 block_list[EXT4_NDIR_BLOCKS+i] = indirect_block[i];
392 free(indirect_block);
397 static int get_block_list_indirect(int fd, struct ext4_inode *inode, unsigned long long *block_list)
399 unsigned int count=0;
401 if (get_direct_blocks(inode, block_list, &count)) {
402 get_indirect_blocks(fd, inode, block_list, &count);
408 static int get_extent_ents(struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
411 struct ext4_extent *extent;
412 off64_t fs_block_num;
414 if (ext_hdr->eh_depth != 0) {
415 critical_error("get_extent_ents called with eh_depth != 0\n");
418 /* The extent entries immediately follow the header, so add 1 to the pointer
419 * and cast it to an extent pointer.
421 extent = (struct ext4_extent *)(ext_hdr + 1);
423 for (i = 0; i < ext_hdr->eh_entries; i++) {
424 fs_block_num = ((off64_t)extent->ee_start_hi << 32) | extent->ee_start_lo;
425 for (j = 0; j < extent->ee_len; j++) {
426 block_list[extent->ee_block+j] = fs_block_num+j;
434 static int get_extent_idx(int fd, struct ext4_extent_header *ext_hdr, unsigned long long *block_list)
437 struct ext4_extent_idx *extent_idx;
438 struct ext4_extent_header *tmp_ext_hdr;
439 off64_t fs_block_num;
440 unsigned char block[MAX_EXT4_BLOCK_SIZE];
443 if (ext_hdr->eh_depth == 0) {
444 critical_error("get_extent_idx called with eh_depth == 0\n");
447 /* The extent entries immediately follow the header, so add 1 to the pointer
448 * and cast it to an extent pointer.
450 extent_idx = (struct ext4_extent_idx *)(ext_hdr + 1);
452 for (i = 0; i < ext_hdr->eh_entries; i++) {
453 fs_block_num = ((off64_t)extent_idx->ei_leaf_hi << 32) | extent_idx->ei_leaf_lo;
454 read_block(fd, fs_block_num, block);
455 tmp_ext_hdr = (struct ext4_extent_header *)block;
457 if (tmp_ext_hdr->eh_depth == 0) {
458 get_extent_ents(tmp_ext_hdr, block_list); /* leaf node, fill in block_list */
460 get_extent_idx(fd, tmp_ext_hdr, block_list); /* recurse down the tree */
467 static int get_block_list_extents(int fd, struct ext4_inode *inode, unsigned long long *block_list)
469 struct ext4_extent_header *extent_hdr;
471 extent_hdr = (struct ext4_extent_header *)inode->i_block;
473 if (extent_hdr->eh_magic != EXT4_EXT_MAGIC) {
474 critical_error("extent header has unexpected magic value 0x%4.4x\n",
475 extent_hdr->eh_magic);
478 if (extent_hdr->eh_depth == 0) {
479 get_extent_ents((struct ext4_extent_header *)inode->i_block, block_list);
483 get_extent_idx(fd, (struct ext4_extent_header *)inode->i_block, block_list);
488 static int is_entry_dir(int fd, struct ext4_dir_entry_2 *dirp, int pass)
490 struct ext4_inode inode;
493 if (dirp->file_type == EXT4_FT_DIR) {
495 } else if (dirp->file_type == EXT4_FT_UNKNOWN) {
496 /* Somebody was too lazy to fill in the dir entry,
497 * so we have to go fetch it from the inode. Grrr.
499 /* if UPDATE_INODE_NUMS pass and the inode high bit is not
500 * set return false so we don't recurse down the tree that is
501 * already updated. Otherwise, fetch inode, and return answer.
503 if ((pass == UPDATE_INODE_NUMS) && !(dirp->inode & 0x80000000)) {
506 read_inode(fd, (dirp->inode & 0x7fffffff), &inode);
507 if (S_ISDIR(inode.i_mode)) {
516 static int recurse_dir(int fd, struct ext4_inode *inode, char *dirbuf, int dirsize, int mode)
518 unsigned long long *block_list;
519 unsigned int num_blocks;
520 struct ext4_dir_entry_2 *dirp, *prev_dirp = 0;
522 unsigned int i, leftover_space, is_dir;
523 struct ext4_inode tmp_inode;
528 case SANITY_CHECK_PASS:
529 case MARK_INODE_NUMS:
530 case UPDATE_INODE_NUMS:
533 critical_error("recurse_dir() called witn unknown mode!\n");
536 if (dirsize % info.block_size) {
537 critical_error("dirsize %d not a multiple of block_size %d. This is unexpected!\n",
538 dirsize, info.block_size);
541 num_blocks = dirsize / info.block_size;
543 block_list = malloc((num_blocks + 1) * sizeof(*block_list));
544 if (block_list == 0) {
545 critical_error("failed to allocate memory for block_list\n");
548 if (inode->i_flags & EXT4_EXTENTS_FL) {
549 get_block_list_extents(fd, inode, block_list);
551 /* A directory that requires doubly or triply indirect blocks in huge indeed,
552 * and will almost certainly not exist, especially since make_ext4fs only creates
553 * directories with extents, and the kernel will too, but check to make sure the
554 * directory is not that big and give an error if so. Our limit is 12 direct blocks,
555 * plus block_size/4 singly indirect blocks, which for a filesystem with 4K blocks
556 * is a directory 1036 blocks long, or 4,243,456 bytes long! Assuming an average
557 * filename length of 20 (which I think is generous) thats 20 + 8 bytes overhead
558 * per entry, or 151,552 entries in the directory!
560 if (num_blocks > (info.block_size / 4 + EXT4_NDIR_BLOCKS)) {
561 critical_error("Non-extent based directory is too big!\n");
563 get_block_list_indirect(fd, inode, block_list);
566 /* Read in all the blocks for this directory */
567 for (i = 0; i < num_blocks; i++) {
568 read_block(fd, block_list[i], dirbuf + (i * info.block_size));
571 dirp = (struct ext4_dir_entry_2 *)dirbuf;
572 while (dirp < (struct ext4_dir_entry_2 *)(dirbuf + dirsize)) {
574 leftover_space = (char *)(dirbuf + dirsize) - (char *)dirp;
575 if (((mode == SANITY_CHECK_PASS) || (mode == UPDATE_INODE_NUMS)) &&
576 (leftover_space <= 8) && prev_dirp) {
577 /* This is a bug in an older version of make_ext4fs, where it
578 * didn't properly include the rest of the block in rec_len.
579 * Update rec_len on the previous entry to include the rest of
580 * the block and exit the loop.
583 printf("fixing up short rec_len for diretory entry for %s\n", name);
585 prev_dirp->rec_len += leftover_space;
589 if (dirp->inode == 0) {
590 /* This is the last entry in the directory */
594 strncpy(name, dirp->name, dirp->name_len);
595 name[dirp->name_len]='\0';
597 /* Only recurse on pass UPDATE_INODE_NUMS if the high bit is set.
598 * Otherwise, this inode entry has already been updated
599 * and we'll do the wrong thing. Also don't recurse on . or ..,
600 * and certainly not on non-directories!
602 /* Hrm, looks like filesystems made by fastboot on stingray set the file_type
603 * flag, but the lost+found directory has the type set to Unknown, which
604 * seems to imply I need to read the inode and get it.
606 is_dir = is_entry_dir(fd, dirp, mode);
607 if ( is_dir && (strcmp(name, ".") && strcmp(name, "..")) &&
608 ((mode == SANITY_CHECK_PASS) || (mode == MARK_INODE_NUMS) ||
609 ((mode == UPDATE_INODE_NUMS) && (dirp->inode & 0x80000000))) ) {
610 /* A directory! Recurse! */
611 read_inode(fd, dirp->inode & 0x7fffffff, &tmp_inode);
613 if (!S_ISDIR(tmp_inode.i_mode)) {
614 critical_error("inode %d for name %s does not point to a directory\n",
615 dirp->inode & 0x7fffffff, name);
618 printf("inode %d %s use extents\n", dirp->inode & 0x7fffffff,
619 (tmp_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
622 tmp_dirsize = tmp_inode.i_blocks_lo * INODE_BLOCK_SIZE;
624 printf("dir size = %d bytes\n", tmp_dirsize);
627 tmp_dirbuf = malloc(tmp_dirsize);
628 if (tmp_dirbuf == 0) {
629 critical_error("failed to allocate memory for tmp_dirbuf\n");
632 recurse_dir(fd, &tmp_inode, tmp_dirbuf, tmp_dirsize, mode);
639 printf("Directory %s\n", name);
641 printf("Non-directory %s\n", name);
645 /* Process entry based on current mode. Either set high bit or change inode number */
646 if (mode == MARK_INODE_NUMS) {
647 dirp->inode |= 0x80000000;
648 } else if (mode == UPDATE_INODE_NUMS) {
649 if (dirp->inode & 0x80000000) {
650 dirp->inode = compute_new_inum(dirp->inode & 0x7fffffff);
654 if ((bail_phase == mode) && (bail_loc == 1) && (bail_count == count)) {
655 critical_error("Bailing at phase %d, loc 1 and count %d\n", mode, count);
658 /* Point dirp at the next entry */
660 dirp = (struct ext4_dir_entry_2*)((char *)dirp + dirp->rec_len);
663 /* Write out all the blocks for this directory */
664 for (i = 0; i < num_blocks; i++) {
665 write_block(fd, block_list[i], dirbuf + (i * info.block_size));
666 if ((bail_phase == mode) && (bail_loc == 2) && (bail_count <= count)) {
667 critical_error("Bailing at phase %d, loc 2 and count %d\n", mode, count);
676 int ext4fixup(char *fsdev)
678 return ext4fixup_internal(fsdev, 0, 0, 0, 0, 0);
681 int ext4fixup_internal(char *fsdev, int v_flag, int n_flag,
682 int stop_phase, int stop_loc, int stop_count)
685 struct ext4_inode root_inode;
686 unsigned int dirsize;
689 if (setjmp(setjmp_env))
690 return EXIT_FAILURE; /* Handle a call to longjmp() */
695 bail_phase = stop_phase;
697 bail_count = stop_count;
699 fd = open(fsdev, O_RDWR);
702 critical_error_errno("failed to open filesystem image");
704 read_ext(fd, verbose);
706 if (info.feat_incompat & EXT4_FEATURE_INCOMPAT_RECOVER) {
707 critical_error("Filesystem needs recovery first, mount and unmount to do that\n");
710 /* Clear the low bit which is set while this tool is in progress.
711 * If the tool crashes, it will still be set when we restart.
712 * The low bit is set to make the filesystem unmountable while
713 * it is being fixed up. Also allow 0, which means the old ext2
716 if (((aux_info.sb->s_desc_size & ~1) != sizeof(struct ext2_group_desc)) &&
717 ((aux_info.sb->s_desc_size & ~1) != 0))
718 critical_error("error: bg_desc_size != sizeof(struct ext2_group_desc)\n");
720 if ((info.feat_incompat & EXT4_FEATURE_INCOMPAT_FILETYPE) == 0) {
721 critical_error("Expected filesystem to have filetype flag set\n");
724 #if 0 // If we have to fix the directory rec_len issue, we can't use this check
725 /* Check to see if the inodes/group is copacetic */
726 if (info.inodes_per_blockgroup % (info.block_size/info.inode_size) == 0) {
727 /* This filesystem has either already been updated, or was
731 printf("%s: filesystem correct, no work to do\n", me);
737 /* Compute what the new value of inodes_per_blockgroup will be when we're done */
738 new_inodes_per_group=EXT4_ALIGN(info.inodes_per_group,(info.block_size/info.inode_size));
740 read_inode(fd, EXT4_ROOT_INO, &root_inode);
742 if (!S_ISDIR(root_inode.i_mode)) {
743 critical_error("root inode %d does not point to a directory\n", EXT4_ROOT_INO);
746 printf("inode %d %s use extents\n", EXT4_ROOT_INO,
747 (root_inode.i_flags & EXT4_EXTENTS_FL) ? "does" : "does not");
750 dirsize = root_inode.i_blocks_lo * INODE_BLOCK_SIZE;
752 printf("root dir size = %d bytes\n", dirsize);
755 dirbuf = malloc(dirsize);
757 critical_error("failed to allocate memory for dirbuf\n");
760 /* Perform a sanity check pass first, try to catch any errors that will occur
761 * before we actually change anything, so we don't leave a filesystem in a
762 * corrupted, unrecoverable state. Set no_write, make it quiet, and do a recurse
763 * pass and a update_superblock pass. Set flags back to requested state when done.
764 * Only perform sanity check if the state is unset. If the state is _NOT_ unset,
765 * then the tool has already been run and interrupted, and it presumably ran and
766 * passed sanity checked before it got interrupted. It is _NOT_ safe to run sanity
767 * check if state is unset because it assumes inodes are to be computed using the
768 * old inodes/group, but some inode numbers may be updated to the new number.
770 if (get_fs_fixup_state(fd) == STATE_UNSET) {
773 recurse_dir(fd, &root_inode, dirbuf, dirsize, SANITY_CHECK_PASS);
774 update_superblocks_and_bg_desc(fd, STATE_UNSET);
778 set_fs_fixup_state(fd, STATE_MARKING_INUMS);
781 if (get_fs_fixup_state(fd) == STATE_MARKING_INUMS) {
782 count = 0; /* Reset debugging counter */
783 if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, MARK_INODE_NUMS)) {
784 set_fs_fixup_state(fd, STATE_UPDATING_INUMS);
788 if (get_fs_fixup_state(fd) == STATE_UPDATING_INUMS) {
789 count = 0; /* Reset debugging counter */
790 if (!recurse_dir(fd, &root_inode, dirbuf, dirsize, UPDATE_INODE_NUMS)) {
791 set_fs_fixup_state(fd, STATE_UPDATING_SB);
795 if (get_fs_fixup_state(fd) == STATE_UPDATING_SB) {
796 /* set the new inodes/blockgroup number,
797 * and sets the state back to 0.
799 if (!update_superblocks_and_bg_desc(fd, STATE_UPDATING_SB)) {
800 set_fs_fixup_state(fd, STATE_UNSET);