We cap our number of CPU worker threads to at least 2 even if we have a
single core. But also, before we used to always add one extra thread
regardless of the number of core.
This meant that we were off when re-using the get_num_cpus() function
when calculating our onionskin work overhead because we were always off
by one.
This commit makes it that we always use the number of thread our actual
thread pool was configured with.
Fixes#40719
Signed-off-by: David Goulet <dgoulet@torproject.org>
Until now, there was this magic number (64) used as the maximum number
of tasks a CPU worker can take at once.
This commit makes it a consensus parameter so our future selves can
think of a better value depending on network conditions.
Part of #40704
Signed-off-by: David Goulet <dgoulet@torproject.org>
* it also uses sys/param.h to track its version;
* present that to tor_libc_get_version_str() as libc version;
while here, we also fix the return of FreeBSD version
* __FreeBSD_version is the correct var tracking the OSVERSION
* we use OSVERSION here (defined by __FreeBSD__);
* it's part of the <sys/param.h> include;
* that tracks all noteworthy changes made to the base system.
* __BSD_VISIBLE is defined by systems like FreeBSD and OpenBSD;
* that also extends to DragonFlyBSD;
* it's used on stdlib.h and ctypes.h on those systems.
This BUG() was added when the code was written to see if this callback
was ever executed after we marked the handle as EOF. It turns out, it
does, but we handle it gracefully. We can therefore remove the BUG().
Fixes tpo/core/tor#40596.
Bug 1: We were purporting to calculate milliseconds per tick, when we
*should* have been computing ticks per millisecond.
Bug 2: Instead of computing either one of those, we were _actually_
computing femtoseconds per tick.
These two bugs covered for one another on x86 hardware, where 1 tick
== 1 nanosecond. But on M1 OSX, 1 tick is about 41 nanoseconds,
causing surprising results.
Fixes bug 40684; bugfix on 0.3.3.1-alpha.
LibreSSL is now closer to OpenSSL 1.1 than OpenSSL 1.0. According to
https://undeadly.org/cgi?action=article;sid=20220116121253, this is the
intention of OpenBSD developers.
According to #40630, many special cases are needed to compile Tor against
LibreSSL 3.5 when using Tor's OpenSSL 1.0 compatibility mode, whereas only a
small number of #defines are required when using OpenSSL 1.1 compatibility
mode. One additional workaround is required for LibreSSL 3.4 compatibility.
Compiles and passes unit tests with LibreSSL 3.4.3 and 3.5.1.
For some syscalls the kernel ABI uses 32 bit signed integers. Whether
these 32 bit integer values are sign extended or zero extended to the
native 64 bit register sizes is undefined and dependent on the {arch,
compiler, libc} being used. Instead of trying to detect which cases
zero-extend and which cases sign-extend, this commit uses a masked
equality check on the lower 32 bits of the value.
The chown/chmod/rename syscalls have never existed on AArch64, and libc
implements the POSIX functions via the fchownat/fchmodat/renameat
syscalls instead.
Add new filter functions for fchownat/fchmodat/renameat, not made
architecture specific since the syscalls exists everywhere else too.
However, in order to limit seccomp filter space usage, we only insert
rules for one of {chown, chown32, fchownat} depending on the
architecture (resp. {chmod, fchmodat}, {rename, renameat}).
New glibc versions not sign-extending 32 bit negative constants seems to
not be a thing on AArch64. I suspect that this might not be the only
architecture where the sign-extensions is happening, and the correct fix
might be instead to use a proper 32 bit comparison for the first openat
parameter. For now, band-aid fix this so the sandbox can work again on
AArch64.
Using tor_free is wrong; event_free must be called for objects obtained from
event_new. Additionally, this slightly simplifies the code.
Also, add a static_assert to prevent further instances.