[PATCH] erofs-utils: fsck: support extracting subtrees

[Gao Xiang]

On 2026/2/26 10:18, Inseob Kim wrote:
> On Thu, Feb 26, 2026 at 10:50 AM Gao Xiang <hsiangkao at linux.alibaba.com> wrote:
>>
>> Hi Inseob,
>>
>> On 2026/2/26 08:59, Inseob Kim wrote:
>>> Add --nid and --path options to fsck.erofs to allow users to check
>>> or extract specific sub-directories or files instead of the entire
>>> filesystem.
>>>
>>> This is useful for targeted data recovery or verifying specific
>>> image components without the overhead of a full traversal.
>>
>> Thanks for the patch!
> 
> Thank *you* for quick response!
> 
>>
>>>
>>> Signed-off-by: Inseob Kim <inseob at google.com>
>>> ---
>>>    fsck/main.c | 50 ++++++++++++++++++++++++++++++++++++++++----------
>>>    1 file changed, 40 insertions(+), 10 deletions(-)
>>>
>>> diff --git a/fsck/main.c b/fsck/main.c
>>> index ab697be..a7d9f46 100644
>>> --- a/fsck/main.c
>>> +++ b/fsck/main.c
>>> @@ -39,6 +39,8 @@ struct erofsfsck_cfg {
>>>        bool preserve_owner;
>>>        bool preserve_perms;
>>>        bool dump_xattrs;
>>> +     erofs_nid_t nid;
>>> +     const char *inode_path;
>>>        bool nosbcrc;
>>>    };
>>>    static struct erofsfsck_cfg fsckcfg;
>>> @@ -59,6 +61,8 @@ static struct option long_options[] = {
>>>        {"offset", required_argument, 0, 12},
>>>        {"xattrs", no_argument, 0, 13},
>>>        {"no-xattrs", no_argument, 0, 14},
>>> +     {"nid", required_argument, 0, 15},
>>> +     {"path", required_argument, 0, 16},
>>>        {"no-sbcrc", no_argument, 0, 512},
>>>        {0, 0, 0, 0},
>>>    };
>>> @@ -110,6 +114,8 @@ static void usage(int argc, char **argv)
>>>                " --extract[=X]          check if all files are well encoded, optionally\n"
>>>                "                        extract to X\n"
>>>                " --offset=#             skip # bytes at the beginning of IMAGE\n"
>>> +             " --nid=#                check or extract from the target inode of nid #\n"
>>> +             " --path=X               check or extract from the target inode of path X\n"
>>>                " --no-sbcrc             bypass the superblock checksum verification\n"
>>>                " --[no-]xattrs          whether to dump extended attributes (default off)\n"
>>>                "\n"
>>> @@ -245,6 +251,12 @@ static int erofsfsck_parse_options_cfg(int argc, char **argv)
>>>                case 14:
>>>                        fsckcfg.dump_xattrs = false;
>>>                        break;
>>> +             case 15:
>>> +                     fsckcfg.nid = (erofs_nid_t)atoll(optarg);
>>> +                     break;
>>> +             case 16:
>>> +                     fsckcfg.inode_path = optarg;
>>> +                     break;
>>>                case 512:
>>>                        fsckcfg.nosbcrc = true;
>>>                        break;
>>> @@ -981,7 +993,8 @@ static int erofsfsck_check_inode(erofs_nid_t pnid, erofs_nid_t nid)
>>>
>>>        if (S_ISDIR(inode.i_mode)) {
>>>                struct erofs_dir_context ctx = {
>>> -                     .flags = EROFS_READDIR_VALID_PNID,
>>> +                     .flags = (pnid == nid && nid != g_sbi.root_nid) ?
>>
>> Does it relax the validatation check?
>>
>> and does (nid == pnid && nid == fsckcfg.nid) work?
> 
> It shouldn't relax the existing validation check.
> `erofsfsck_check_inode` is called with `err =
> erofsfsck_check_inode(fsckcfg.nid, fsckcfg.nid);`.
> 
> * If a given path is not the root, `nid`'s parent `..` will differ
> from `pnid`, causing failure. This condition only relaxes the starting
> directory.

I just have a wild thought, if there is a directory which have

  foo/
  | - .. -> pointing to `foo` itself
    - a/ -> pointing to `foo` directory too

will (pnid == nid && nid != g_sbi.root_nid) relaxs
the check for a/ ?

I'm not sure but you could double check.

> * In the case of the root, `nid`'s parent `..` should indeed be
> itself. So we can still validate.
> 
> If you have any better suggestions, I'll follow them.
> 
>>
>>> +                             0 : EROFS_READDIR_VALID_PNID,
>>>                        .pnid = pnid,
>>>                        .dir = &inode,
>>>                        .cb = erofsfsck_dirent_iter,
>>> @@ -1033,6 +1046,8 @@ int main(int argc, char *argv[])
>>>        fsckcfg.preserve_owner = fsckcfg.superuser;
>>>        fsckcfg.preserve_perms = fsckcfg.superuser;
>>>        fsckcfg.dump_xattrs = false;
>>> +     fsckcfg.nid = 0;
>>> +     fsckcfg.inode_path = NULL;
>>>
>>>        err = erofsfsck_parse_options_cfg(argc, argv);
>>>        if (err) {
>>> @@ -1068,22 +1083,37 @@ int main(int argc, char *argv[])
>>>        if (fsckcfg.extract_path)
>>>                erofsfsck_hardlink_init();
>>>
>>> -     if (erofs_sb_has_fragments(&g_sbi) && g_sbi.packed_nid > 0) {
>>> -             err = erofs_packedfile_init(&g_sbi, false);
>>> +     if (fsckcfg.inode_path) {
>>> +             struct erofs_inode inode = { .sbi = &g_sbi };
>>> +
>>> +             err = erofs_ilookup(fsckcfg.inode_path, &inode);
>>>                if (err) {
>>> -                     erofs_err("failed to initialize packedfile: %s",
>>> -                               erofs_strerror(err));
>>> +                     erofs_err("failed to lookup %s", fsckcfg.inode_path);
>>>                        goto exit_hardlink;
>>>                }
>>
>> It would be better to check if it's a directory.
> 
> My intention was that we support both directories and files. Or should
> I create a separate flag like `--cat` in dump.erofs?

Ok, make sense.

Thanks,
Gao Xiang