| Summary: | net-mail/cmd5checkpw is installed setuid, but does not drop euid | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Gentoo Security | Reporter: | Florian Westphal <westphal> | ||||
| Component: | Vulnerabilities | Assignee: | Gentoo Security <security> | ||||
| Status: | RESOLVED FIXED | ||||||
| Severity: | minor | CC: | net-mail+disabled, robbat2 | ||||
| Priority: | High | ||||||
| Version: | unspecified | ||||||
| Hardware: | All | ||||||
| OS: | Linux | ||||||
| Whiteboard: | B4 [glsa] | ||||||
| Package list: | Runtime testing required: | --- | |||||
| Attachments: |
|
||||||
Created attachment 48674 [details, diff]
cmd5checkpw: set euid to uid of calling user
net-mail herd, please comment (on the need to be SUID and on the patch) Florian: did you try to contact upstream yet ? I emailed brush@elysium.pl about this a few minutes ago. (same Bugreport + patch) (I thought this was a Gentoo specific bug at first before seeing that upstream docs suggest making cmd5checkpw setuid) Reassigning this as a vulnerability, since it's a clear local information leak. Florian: any answer from upstream ? No reply from upstream until now. The last 'news' item on the project homepage is dated 09.10.2000... net-mail: please comment. Can cmd5checkwp not be setuid ? If not, what do you think of the patch ? langthang asked me to comment on this bug. There we go.
Quoting the manpage of cmd5checkpw:
FILES
/etc/poppasswd - this file contains pairs of logins and clear text
passwords separated by ":". It looks like this:
login1:password1
login2:password2
Best way to protect it is to make it readable only for one specific
user different than you normal system users and make cmd5checkpw suid
that user.
Therefore, I would say that cmd5checkpw has to be setuid if /etc/poppasswd is only readable by a specific user. But I also think that dropping the effective uid wouldn't hurt. If nobody else (robbat2?) sees a problem in here, we should apply the patch.
Robin, what do you think about this patch? Can we apply it? Upstream looks dead... net-mail: please apply the patch or drop the package. The patch is now applied to cmd5checkpw-0.22-r2. The ebuild is currently in ~ARCH for testing. Please test it and comment on this bug again. Then we'll make a stabilization request to all affected architectures. Thx Micheal, reopening for stable marking. Arches please test and mark cmd5checkpw-0.22-r2 stable. x86 stable Stable on ppc and hppa. Stable on alpha. sparc stable. how is this being tested?(noone on amd64 apparently uses it) Mike: I am not a cmd5checkpw user but it looks like a password changing system that will access a /etc/poppasswd (owned by the cmd5checkpw user and -rw-------).
Try creating a /etc/poppasswd file with pairs of logins and clear text passwords like this :
login1:password1
And validate you can change the password as a regular user. You can also vamidate that the exploit in bug description is no longer working.
Stable on mips. stable on amd64 security, pls vote on GLSA need I vote no glsa, please feel free to disagree ;) Local users can get plaintext POP passwords for their coworkers... I vote yes. not absolutely necessary, but a GLSA on this might be a good idea voting for one GLSA 200502-30 arm should park stable to benefit from GLSA |
net-mail/cmd5checkpw is installed setuid cmd5checkpw, but it does not drop privileges before calling execvp(), i.e. the invoked program retains the cmd5checkpw euid. Local users that know at least one valid /etc/poppasswd user/password combination can read the /etc/poppasswd file. Reproducible: Always Steps to Reproduce: 1. emerge net-mail/cmd5checkwp 2. create /etc/poppasswd to contain: user:pass secret:secret 3: $ id uid=1001(fw) gid=100(users) groups=5(tty),10(wheel),16(cron),100(users) $ perl -e 'print("user\0pass\0\pass\0");' > test $ 3<test $ /bin/cmd5checkpw id uid=1001(fw) gid=100(users) euid=1000(cmd5checkpw) [..] Actual Results: user obtains euid=1000(cmd5checkpw). Expected Results: Drop euid before execvp(). If cmd5checkpw really needs to be setuid, it should set its effective uid to that the real uid of the calling process. I'll add a patch to do this, but i'd prefer cmd5checkpw to not be setuid (this might break things though)