Re: [libvirt] [PATCH 2/2] Revert "Prevent more compiler optimization of mockable functions"

On Wed, Jul 12, 2017 at 01:49:04PM +0200, Martin Kletzander wrote: > On Wed, Jul 12, 2017 at 01:10:08PM +0200, Martin Kletzander wrote: > > On Wed, Jul 12, 2017 at 11:14:16AM +0100, Daniel P. Berrange wrote: > > > This reverts commit e4b980c853d2114b25fa805a84ea288384416221. > > > > > > When a binary links against a .a archive (as opposed to a shared library), > > > any symbols which are marked as 'weak' get silently dropped. As a result > > > when the binary later runs, those 'weak' functions have an address of > > > 0x0 and thus crash when run. > > > > > > This happened with virtlogd and virtlockd because they don't link to > > > libvirt.so, but instead just libvirt_util.a and libvirt_rpc.a. The > > > virRandomBits symbols was weak and so left out of the virtlogd & > > > virtlockd binaries, despite being required by virHashTable functions. > > > > > > Various other binaries like libvirt_lxc, libvirt_iohelper, etc also > > > link directly to .a files instead of libvirt.so, so are potentially > > > at risk of dropping symbols leading to a later runtime crash. > > > > > > This is normal linker behaviour because a weak symbol is not treated > > > as undefined, so nothing forces it to be pulled in from the .a You > > > have to force the linker to pull in weak symbols using -u$SYMNAME > > > which is not a practical approach. > > > > > > > How is this done by gnulib (or libc) when most their functions are weak > > aliases anyway? Can't we use the same approach they have? > > virtlo{g,ck}d link with libgnu.la as well and there is no problem with > > that, right? So I guess this _must_ be solvable somehow, IMHO. > > > > I'm just curious how that works. > > > > Martin > > I guess we would have to do something like the following, but for every > function. > > diff --git i/src/util/virrandom.c w/src/util/virrandom.c > index 41daa404b273..3d9fe7f85d97 100644 > --- i/src/util/virrandom.c > +++ w/src/util/virrandom.c > @@ -99,7 +99,7 @@ VIR_ONCE_GLOBAL_INIT(virRandom) > * > * Return: a random number with @nbits entropy > */ > -uint64_t virRandomBits(int nbits) > +static uint64_t __virRandomBits(int nbits) > { > uint64_t ret = 0; > int32_t bits; > @@ -125,6 +125,7 @@ uint64_t virRandomBits(int nbits) > virMutexUnlock(&randomLock); > return ret; > } > +uint64_t virRandomBits(int nbits) ATTRIBUTE_MOCKABLE __attribute__((alias("__virRandomBits"))); > > > /** > diff --git i/src/util/virrandom.h w/src/util/virrandom.h > index 990a456addf7..abda95aef506 100644 > --- i/src/util/virrandom.h > +++ w/src/util/virrandom.h > @@ -24,7 +24,7 @@ > > # include "internal.h" > > -uint64_t virRandomBits(int nbits) ATTRIBUTE_MOCKABLE; > +uint64_t virRandomBits(int nbits); > double virRandom(void); > uint32_t virRandomInt(uint32_t max); > int virRandomBytes(unsigned char *buf, size_t buflen) > -- > > And of course that could be macrofied so that ATTRIBUTE_MOCKABLE takes > function or something, etc. > > I like this more than reverting the patches. > Also, having the weak alias we can drop all the mocks and the problems with them and just redefine the functions we would like to mock in the tests (see tests/virfilewrapper.c), it would work on win32, it would not compile if we would forgot to make the function as an alias (so no need to check that in the syntax-check) and maybe provide other benefits that I can't think of right now.