wireguard-openbsd/regress/libexec/ld.so/randomdata/ld.so-cookie, branch jd/queueboosts WireGuard implementation for the OpenBSD kernel https://git.zx2c4.com/wireguard-openbsd/atom/regress/libexec/ld.so/randomdata/ld.so-cookie?h=jd%2Fqueueboosts 2014-02-16T00:33:44Z Update ld.so-cookie test which is currently broken and gets in the 2014-02-16T00:33:44Z martynas martynas@openbsd.org 2014-02-16T00:33:44Z urn:sha1:e026d09d28b43a8ee26685c712c31793a2521911 way of SSP testing. For some reason this diff keeps escaping from cvs commit... Basic idea: * Use dl_iterate_phdr() to iterate all loaded ELF program headers. * Check that we actually find a header for ld.so. * For ld.so, check that we find a PT_OPENBSD_RANDOMIZE segment. * For each PT_OPENBSD_RANDOMIZE segment, check that at least one byte in the memory range is non-zero. Written by matthew@. "Please go ahaed" kettenis@. "Move fast" deraadt@, a week ago. Double check that we're not testing libc's __guard. Changing ld.so's 2012-08-17T15:11:12Z matthew matthew@openbsd.org 2012-08-17T15:11:12Z urn:sha1:b113c0bf3ac8e642fa6e14ffc88183c98d5edd35 __guard symbol to HIDDEN causes dlsym() to return a pointer to libc's __guard instead, which can cause the test to wrongly pass even if ld.so's random data segments are not being initialized properly. Use dlopen("ld.so") instead of dlopen("/usr/libexec/ld.so") so we get 2012-08-16T20:38:13Z matthew matthew@openbsd.org 2012-08-16T20:38:13Z urn:sha1:ccd5238a63fc9ae9b12c98bf727fda4b3a07e81d a reference to the already loaded ld.so instead of loading another instance. Add a test case to verify ld.so's randomdata section is initialized 2012-08-16T18:19:36Z matthew matthew@openbsd.org 2012-08-16T18:19:36Z urn:sha1:9c999f0bd28dada1a6060dc5a10b177aed447e4e too, since interpreters will be initialized by a slightly different code path than executables.