You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Yesterday, CVE-2024-53899 was published, with the following description:
virtualenv before 20.26.6 allows command injection through the activation scripts for a virtual environment. Magic template strings are not quoted correctly when replacing.
Because uv vendors virtualenv activation scripts in crates/uv-virtualenv/src/activator/, I wondered if uv was also affected. Imitating the original reproducer in pypa/virtualenv#2768, and working with a uv checkout on the 0.5.4 tag:
This is good; the uname -a command was not executed, so this test did not reproduce the problem.
Checking older versions, it looks like this was fixed between uv 0.5.2 and 0.5.3, probably in #8984 (although I did not bother bisecting). To reiterate this very clearly, I was not able to trivially reproduce CVE-2024-53899 in uv 0.5.3 and later, but I was able to reproduce it (output from uname -a appeared in my console) in uv 0.5.2 and earlier.
Could you please comment on whether you believe the underlying issue is entirely mitigated in uv, or whether further work needs to be done? Thank you!
The text was updated successfully, but these errors were encountered:
While #8947 and the batch behavior are clearly bugs, I would like to remind users that you should never build virtual environments from untrusted input. There are various sources of arbitrary code execution, such as .pth files, wheel data writing to arbitrary paths in the venv and source dist builds running arbitrary code.
Yesterday, CVE-2024-53899 was published, with the following description:
This corresponds to pypa/virtualenv#2768 and is supposed to be fixed by pypa/virtualenv#2771.
Because
uv
vendorsvirtualenv
activation scripts incrates/uv-virtualenv/src/activator/
, I wondered ifuv
was also affected. Imitating the original reproducer in pypa/virtualenv#2768, and working with auv
checkout on the0.5.4
tag:This is good; the
uname -a
command was not executed, so this test did not reproduce the problem.Checking older versions, it looks like this was fixed between
uv
0.5.2 and 0.5.3, probably in #8984 (although I did not bother bisecting). To reiterate this very clearly, I was not able to trivially reproduce CVE-2024-53899 inuv
0.5.3 and later, but I was able to reproduce it (output fromuname -a
appeared in my console) inuv
0.5.2 and earlier.Could you please comment on whether you believe the underlying issue is entirely mitigated in
uv
, or whether further work needs to be done? Thank you!The text was updated successfully, but these errors were encountered: