Comment by halestock

3 hours ago

Pretty impressive to introduce 1400 CVEs in a project that's only ~7 months old.

These are CVEs in the base image and in standard lib dependencies. For example, just scanned an unhardened image I built today:

  Unhardened: docker.io/nanoco/nanoclaw:agent-alpha
  71 packages, 344 unique CVEs, linux/arm64

  PACKAGE         VERSION        TYP   C   H   M   L   N  TOT
  -----------------------------------------------------------
  expat           2.5.0          deb   0   4  18   1   2   25
  curl            7.88.1         deb   4   4   6   0   7   21
  hono            4.12.14        npm   0   1  18   2   0   21
  libtiff         4.5.0          deb   0   2   1   1  15   20
  perl            5.36.0         deb   5   6   3   0   3   17
  pnpm            10.33.0        npm   0   8   7   0   0   15
  glibc           2.36           deb   1   2   2   1   7   13
  openjpeg        2.5.0          deb   0   0   3   1   9   13
  cups            2.4.2          deb   0   2   8   0   1   11
  glib2           2.74.6         deb   1   7   1   0   1   10
  tar             1.34(+2)       deb   1   1   7   0   1   10
  llvm            15.0.6         deb   0   0   0   1   9   10
  sqlite3         3.40.1         deb   1   2   3   0   3    9
  nss             3.87.1         deb   1   0   3   0   4    8
  avahi           0.8            deb   0   0   8   0   0    8
  util-linux      2.38.1         deb   0   0   3   0   2    7
  elf             0.188          deb   0   0   0   0   7    7
  libssh2         1.10.0         deb   1   4   1   0   0    6
  openldap        2.5.13         deb   0   1   0   0   5    6
  chromium        151.0.7922.108 deb   0   5   0   0   0    5
  -----------------------------------------------------------
  UNIQUE CVEs                         16  68 121  17 119  344

  (+51 more packages, 102 findings)

  C/H/M/L/N = critical/high/medium/low/negligible.
  Counts are unique CVEs: binaries from one source package are
  grouped (libcurl4 + libcurl3-gnutls + curl = curl), so a CVE
  hitting three of them counts once, not three times.

  • Are these real findings, or a situation in which fixes have been backported? At one place I worked, the corpsec guys were wildly incompetent and would try to bury me in "CVEs" in my systems that were nothing but "vulnerable" software versions with all of the "identified" vulnerabilities fixed by Debian backported patches.

If the thing measuring whether there are CVEs is also the thing creating said CVEs, are we sure they are even CVEs? Deduped? Etc.

It’s like it’s made of CVEs. First 50-100 should be a good sign if it’s cleaner to start over.