I suppose an executable file with SetUID bit set should be running as its owner but I cannot really reproduce it. I tried the following.
$ cat prepare.sh cp /bin/bash . chown root.root bash chmod 4770 bash # Verified $ sudo sh prepare.sh $ ./bash $ id -u 1000 $ exit $
$ cat test.c #include<stdio.h> #include<unistd.h> int main(){ printf("%d,%d\n", getuid(), geteuid()); return 0; } $ gcc -o test test.c $ chmod 4770 test # Verified $ sudo chown root.root test $ ./test 1000,1000 $ # Why???
However
$ su # ./bash # id -u 0 # ./test 0,0 # exit # exit $
Note: The mount point has no nosuid
nor noexec
set.
Can anyone explain why it's failing to work on Ubuntu 16.04 LTS?
3 Answers 3
For the compiled executable, from man 2 chown
:
When the owner or group of an executable file are changed by an
unprivileged user the S_ISUID and S_ISGID mode bits are cleared. POSIX
does not specify whether this also should happen when root does the
chown(); the Linux behavior depends on the kernel version.
Reversing the chown
and chmod
order works for me:
$ sudo chmod 4770 foo
$ sudo chown root:root foo
$ stat foo
File: 'foo'
Size: 8712 Blocks: 24 IO Block: 4096 regular file
Device: 801h/2049d Inode: 967977 Links: 1
Access: (0770/-rwxrwx---) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2017年04月18日 15:15:15.074425000 +0900
Modify: 2017年04月18日 15:15:15.074425000 +0900
Change: 2017年04月18日 15:15:33.683725000 +0900
Birth: -
$ sudo chmod 4777 foo
$ ./foo
1000,0
In your first case, it's Bash that doesn't like being run as setuid.
If Bash is started with the effective user (group) id not equal to the real user (group) id,..., and the effective user id is set to the real user id.
See: Bash's manual on startup files, also Setuid bit seems to have no effect on bash .
In the second case, it's the order of chmod
and chown
that matters, as muru already answered. Changing the owner resets the setuid bit.
-
Oh, I didn't notice the OP was using the script setup a setuid bash.muru– muru2017年04月18日 08:18:12 +00:00Commented Apr 18, 2017 at 8:18
-
Is it possible to run shell scripts with the setuid bit via other shells? Or are shell scripts off the table entirely if I need to use setuid?maples– maples2022年04月09日 00:21:50 +00:00Commented Apr 9, 2022 at 0:21
-
@maples, having a script setuid is different from having the shell binary itself setuid. The latter is what they were doing here, and that's what Bash is extra picky about, other shells I tried didn't care. But setuid scripts also have issues, see Allow setuid on shell scripts and the links in the answers. The answers there also say that most unixes ignore the setuid bit on scripts to begin with, so it doesn't matter what the shell does at that point.ilkkachu– ilkkachu2022年04月09日 09:40:05 +00:00Commented Apr 9, 2022 at 9:40
-
1@maples, anyway, if you need to run a script with elevated privileges, the simple and ok-ish way would be to have your users run it through
sudo
. It's better than plain setuid, since sudo knows to do at least some cleanup when changing UIDs, which setuid doesn't really do. Of course, a better way would be to have a privileged background process listen to a unix socket and receive requests through it; that would get rid of all issues with stuff getting inherited from a parent process to a child.ilkkachu– ilkkachu2022年04月09日 09:43:28 +00:00Commented Apr 9, 2022 at 9:43
It could also be that the filesystem containing the test executable was mounted with the nosuid
option; I have heard that newer distributions will do this by default for /tmp
, and there are good arguments for applying it to /home
as well. nosuid
causes the kernel to ignore the setuid and setgid bits on all executables within the filesystem. (The unrelated thing that happens when you make a directory setgid is unaffected.)
sudo
? There's a bug or a typo in it, though, thechmod
is missing a file name.