GitLab

intrigeri authored Nov 16, 2019 and segfault committed Jan 05, 2020

These options are "aimed at preventing possible information leaks and making the
control-flow bugs that depend on uninitialized values more deterministic"¹.
All kmalloc()s effectively become kzalloc()s and all kfree()s effectively become
kzfree()s².

In passing, apart of the defense-in-depth security benefits intended by the
authors of this Linux feature, init_on_free=1 may ensure we clean more kernel
memory at shutdown time.

Benchmarks show:

* a negligible performance hit with init_on_alloc=1
* a 7-25% performance hit with init_on_free=1

Let's see if/how this affects our use cases.

[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6471384af2a6530696fc0203bafe4de41a23c9ef
[2] https://outflux.net/blog/archives/2019/11/14/security-things-in-linux-v5-3/

I removed them in 7727622d
but segfault rightfully observed that they can be helpful.

… by commit cced26b6.

segfault authored Jan 05, 2020 and intrigeri committed Jan 05, 2020

- Use argparse's "required" parameter instead of checking if the
  argument is None.
- Drop stream parameter from logging.basicConfig because sys.stderr is used by
  default.
- Don't catch exceptions only for printing them to stderr and exiting
  with exit code 1, because that is exactly what happens if an exception
  is not caught.
- Don't use sys.exit(main()) with a main function that doesn't return
  anything.

Clarification wrt. "Don't catch exceptions only for printing them […]": the
behavior _is_ different in that before this commit, we would hide the resulting
stacktrace. But there's no situation in which we expect main() to raise an
exception here, so IMO hiding exceptions was a bug. Thank you for fixing
it :) -- intrigeri

Merge branch 'bugfix/17279-workaround-lack-of-usable-tor-transifex-branches' into stable (Closes: #17279)

Merge branch 'feature/15281-single-squashfs-diff' into stable (Closes: #15281, #15279, #15283, #15286)

Fixes a regression introduced by the fix for #17359: when using the recommended
pre-commit hook, that fix prevents creating commits that do not modify any
PO file.

As per https://metacpan.org/pod/LWP::UserAgent#request,

  Note that errors writing to the content file (for example due to permission
  denied or the filesystem being full) will be reported via the Client-Aborted
  or X-Died response headers, and not the is_success method.

We were already handling X-Died specially, but not Client-Aborted,
so there could be cases when we later treated the download as failed
(due to size or hash mismatch) but did not report about the origin
of the problem, which would make debugging of user bug reports harder.

Currently translated at 90.9% (10 of 11 strings)

Currently translated at 100.0% (5 of 5 strings)

Currently translated at 77.8% (7 of 9 strings)

Currently translated at 100.0% (11 of 11 strings)

Currently translated at 90.9% (10 of 11 strings)

Currently translated at 100.0% (5 of 5 strings)

Currently translated at 100.0% (5 of 5 strings)

Currently translated at 80.0% (4 of 5 strings)

Currently translated at 80.0% (4 of 5 strings)

Currently translated at 60.0% (3 of 5 strings)

Currently translated at 60.0% (3 of 5 strings)

Currently translated at 40.0% (2 of 5 strings)

Currently translated at 40.0% (2 of 5 strings)

Currently translated at 20.0% (1 of 5 strings)