pgcrypto: remove useless px_memset() and BF_ASM

Started by Dagfinn Ilmari Mannsåker8 months ago4 messageshackers
Jump to latest

Hi Hackers,

In the thread about centralised architecture detection, I noticed that
the BF_ASM macro in crypt-blowfish.c has never been defined to anything
but 0, and the _BF_body_r() function it would call has never existed, so
that can be got rid of.

While investigating at that, I also noticed that px_memset(), which has
the comment /* memset that must not be optimized away */, is only ever
called with zero for the value, which could be better written with
explicit_bzero() now that we have that.

Attached are patches for both.

- ilmari

Attachments:

t139267_1
0001-pgcrypto-remove-unused-BF_ASM-macro.patchtext/x-diffDownload+0-13
0002-pgcrypto-use-explicit_bzero-instead-of-px_memset-.-0.patchtext/x-diffDownload+49-59
#2Daniel Gustafsson
daniel@yesql.se
In reply to: Dagfinn Ilmari Mannsåker (#1)
Re: pgcrypto: remove useless px_memset() and BF_ASM

On 9 Apr 2026, at 13:51, Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> wrote:

In the thread about centralised architecture detection, I noticed that
the BF_ASM macro in crypt-blowfish.c has never been defined to anything
but 0, and the _BF_body_r() function it would call has never existed, so
that can be got rid of.

Agreed. I didn't do enough archaeology to figure out what upstream has/had or
why it was removed, but it's been dead for 25 odd years so it's about time to
remove.

While investigating at that, I also noticed that px_memset(), which has
the comment /* memset that must not be optimized away */, is only ever
called with zero for the value, which could be better written with
explicit_bzero() now that we have that.

One could imagine various tricks for rewriting px_memset to explicit_bzero in
order to reduce the churn, but since this code is very rarely backpatched into
it's not a big problem IMHO.

I'll have another look with fresh eyes once back in the office.

--
Daniel Gustafsson

#3Peter Eisentraut
peter@eisentraut.org
In reply to: Daniel Gustafsson (#2)
Re: pgcrypto: remove useless px_memset() and BF_ASM

On 09.04.26 21:07, Daniel Gustafsson wrote:

On 9 Apr 2026, at 13:51, Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> wrote:

In the thread about centralised architecture detection, I noticed that
the BF_ASM macro in crypt-blowfish.c has never been defined to anything
but 0, and the _BF_body_r() function it would call has never existed, so
that can be got rid of.

Agreed. I didn't do enough archaeology to figure out what upstream has/had or
why it was removed, but it's been dead for 25 odd years so it's about time to
remove.

While investigating at that, I also noticed that px_memset(), which has
the comment /* memset that must not be optimized away */, is only ever
called with zero for the value, which could be better written with
explicit_bzero() now that we have that.

One could imagine various tricks for rewriting px_memset to explicit_bzero in
order to reduce the churn, but since this code is very rarely backpatched into
it's not a big problem IMHO.

I have committed these two patches. One of them was also discussed in
another thread, but I have co-credited this thread.

In reply to: Peter Eisentraut (#3)
Re: pgcrypto: remove useless px_memset() and BF_ASM

Peter Eisentraut <peter@eisentraut.org> writes:

On 09.04.26 21:07, Daniel Gustafsson wrote:

On 9 Apr 2026, at 13:51, Dagfinn Ilmari Mannsåker <ilmari@ilmari.org> wrote:

In the thread about centralised architecture detection, I noticed
that the BF_ASM macro in crypt-blowfish.c has never been defined to
anything but 0, and the _BF_body_r() function it would call has
never existed, so that can be got rid of.

Agreed. I didn't do enough archaeology to figure out what upstream
has/had or why it was removed, but it's been dead for 25 odd years so
it's about time to remove.

While investigating at that, I also noticed that px_memset(), which
has the comment /* memset that must not be optimized away */, is
only ever called with zero for the value, which could be better
written with explicit_bzero() now that we have that.

One could imagine various tricks for rewriting px_memset to
explicit_bzero in order to reduce the churn, but since this code is
very rarely backpatched into it's not a big problem IMHO.

I have committed these two patches. One of them was also discussed in
another thread, but I have co-credited this thread.

Thanks!

- ilmari