[sr-dev] pthread mutex and openssl 1.1

Greg Troxel gdt at lexort.com
Tue Mar 17 00:49:39 CET 2020


My next issue in the NetBSD build is about mutexes.  I gather this is
very much about openssl 1.1, and all of this is only about non-Linux
since there futexes are used.  (pkgsrc has switched to openssl 1.1
basically the beginning of 2020.)

I am guessing that the basic issue is that the pthread mutexes are
placed in shared memory and used from multiple processes, and that
regardless of setting attributes, the point is that these need to work
in that environment, and if not, some other synchronization scheme needs
to be chosen.   And, I gather the problem is lack of a wakeup on release
to blocked threads in other processes.   Close?



There is -DKSR_PTHREAD_MUTEX_SHARED defined:

ifeq ($(LIBSSL_SET_MUTEX_SHARED), 1)
CC_PMUTEX_OPTS = -pthread -DKSR_PTHREAD_MUTEX_SHARED
LD_PMUTEX_OPTS = -pthread -rdynamic -Wl,-Bsymbolic-functions
else
CC_PMUTEX_OPTS =
LD_PMUTEX_OPTS =
endif

and this seems to get turned on regardless of which mutex option an OS
uses.

Later, in the netbsd if chunk

        ifneq ($(found_lock_method), yes)
                C_DEFS+= -DUSE_SYSV_SEM  # try pthread sems
                found_lock_method=yes

but then -DKSR_PTHREAD_MUTEX_SHARED is still on, and the code that usss
pthread_mutexattr_setpshared is compiled in.  I am thinking it might not
be used, because of SYSV_SEM.

(Btw, the check for NetBSD > 2.0.0 and kqueue can be removed; we support
only 8 and 9 now, and while I don't like to gratuitously desupport older
versions, I really doubt current kamailo would build on anything earlier
than 6 anyway.  Similar for FreeBSD probably, but I don't want to speak
for them.)


Thanks,
Greg






More information about the sr-dev mailing list