7

This problem is related to Samba and inodes are not necessary.

I have a problem handling a certain file that has some special characters in it. If I search it by its inode it will list the file:

 $ find . -inum 90505400 -exec ls {} \;
./12 String Quartet No. 16 in F Major Op. 135: Der schwer gefa?te Entschlu?: Grave, ma non troppo tratto (Mu? es sein ?) - Allegro (Es mu? sein !).flac

However, if I then proceed to use cp or rm on the file it will throw a file not found error (in German 'Datei oder Verzeichnis nicht gefunden'):

 $ find . -inum 90505400 -exec cp {} ne.flac \;
cp: './12 String Quartet No. 16 in F Major Op. 135: Der schwer gefa?te Entschlu?: Grave, ma non troppo tratto (Mu? es sein ?) - Allegro (Es mu? sein !).flac' kann nicht zum Lesen geöffnet werden: Datei oder Verzeichnis nicht gefunden

I wonder, if I can copy the file with another command that uses the inode directly. I also had this problem for some time now. I can remove all files with rm *, but I would like to fix the broken filename.

It is an ext4 filesystem which I mount on a Raspi from an external USB HDD with this line (changed obfuscated paths and IPs):

UUID=e3f9d42a-9703-4e47-9185-33be24b81c46   /mnt/test   ext4    rw,auto,defaults,nofail,x-systemd.device-timeout=15 0 2

I then share it with samba:

[mybook]
path=/mnt/test
public = yes
browseable = yes
writeable = yes
comment = test
printable = no
guest ok = no

And I mount this on a Lubuntu 16 with this:

//192.168.1.190/test         /home/ben/test               cifs auto,nofail,username=XXX,password=XXX,uid=1000,gid=1000

I connect to the Lubuntu 16 through VNC from a Macbook. Or I SSH directly into it. I am just telling this for full information.

I also mount the share on that Macbook (and others) in Finder. Finder does not display the filename correctly.

enter image description here

After a useful comment from a user, I realized I should try to manipulate the file on the host with the original filesystem instead of trying to do it over samba.

SSHing into the host reveals this filename (look at the sign with 0xF022 after '135'):

'12 String Quartet No. 16 in F Major Op. 135 Der schwer gefa?te Entschlu? Grave, ma non troppo tratto (Mu? es sein ) - Allegro (Es mu? sein !).flac'

I then was able to copy the file with cp on the host itself.

(In case anybody wonders how I came to the filename: I split a summed up flac file with it's cue sheet into the separate files and they got named automatically.)

Improve this question
9
  • 1
    Is the file on an NTFS filesystem? Commented Jan 23, 2022 at 18:10
  • @bomben, heh, ok, done. Glad that it helped. Commented Jan 23, 2022 at 21:04
  • 1
    The quotes around the {} does not do a thing. That's cargo cult coding. Commented Jan 23, 2022 at 21:46
  • 1
    @they I edited my question because I think it was stupid to not check it on the host system but instead doing something via inodes over samba. I wanted to prevent people reading a lot of text just to recognise it does not at all relate to their problem. Commented Jan 23, 2022 at 21:58
  • 1
    The unicode character following Op. 135 shows as 0xF022 which falls into en.wikipedia.org/wiki/Private_Use_Areas - i.e. not globally unique unicode character. Commented Jan 24, 2022 at 10:16

2 Answers 2

Reset to default
11

All of open() (for copying), rename() and unlink() (removal) work by filenames. There's really nothing that would work on an inode directly, apart from low-level tools like debugfs.

If you can remove the file with rm *, you should be able to rename it with mv ./12* someothername.flac, or copy it with cp ./12* newfile.flac (assuming ./12* matches just that file). find in itself shouldn't be that different.

But you mentioned Mac, and I think Mac requires filenames to be valid UTF-8 and that might cause issues if the filenames are broken. Linux doesn't names that are invalid UTF-8, but of course there, too, some tools might react oddly. (I haven't tested.) Having Samba in there might not help either.

Assuming that has something to do with the issue, you could try to SSH in to the host with the filesystem, skipping the intermediary parts, and rename the files there.

Improve this answer
1
  • Thanks again! Obviously, doing it on the host filesystem instead of over samba should have been my first approach. Commented Jan 23, 2022 at 21:47
7

It is not possible to open a file via its inode. This is a deliberate aspect of the design of the operating system, because opening a file via its inode would bypass permissions.

Calling find . -inum $inode_number -exec … {} \; is as close as it gets to act on a file based on its inode. But this does use the file name, and it is guaranteed to work on a non-buggy system.

The ? in the output represent bytes that do not make up a valid character. Presumably one piece of software is presenting a legacy 8-bit encoding to a tool that expects UTF-8.

I'm not sure if this could be a symptom of macOS tools working with a file name with an invalid encoding. Just in case, try running the commands in a C locale. This means that all file names will be treated as sequences of bytes, so there are no invalid characters (as far as the user land tools are concerned — the kernel might still have trouble if the remote server is giving it bad data).

LC_ALL=C find . -inum 90505400 -exec cp {} ne.flac \;

Another approach you can try is to run export LC_ALL=C then type mv 12 and press Tab for completion.

If that doesn't work, then the problem is a buggy filesystem which reacts differently when told to retrieve the metadata of a file and when told to open a file. This could be a bug or a misconfiguration on the Samba client or on the Samba server. Or perhaps Samba is fully configured for UTF-8 and is unable to cope with a file name on the server that is not encoded in UTF-8. I recommend checking on the server whether the file name is encoded in UTF-8. If it isn't, rename it on the server. If it is, you have a problem with your Samba configuration.

Improve this answer
1
  • @bomben Since you saw ? in the file name on the server as well, it's likely that both machines are configured for UTF-8 and the file name is in an 8-bit encoding (presumably latin1). And so on the server, `mv 12<TAB> '12 String Quartet No. 16 in F Major Op. 135 Der schwer gefaßte Entschluß Grave, ma non troppo tratto (Muß es sein) - Allegro (Es muß sein!).flac' should work. Commented Jan 23, 2022 at 22:06

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.