illumos gate

We saw a panic with the following stack trace:

fasttrap_pid_enable+0x143(ffffff5565883ab0, 151c4, ffffff57fbaeff18)
dtrace_ecb_enable+0x68(ffffff87e5522230)
dtrace_ecb_create_enable+0x4d(ffffff5676229980, ffffff436f4dc7c0)
dtrace_match+0x111(ffffff01efd91950, e, 0, 1, fffffffff80bf680, ffffff436f4dc7c0)
dtrace_probe_enable+0x81(ffffff4b2821a498, ffffff436f4dc7c0)
dtrace_enabling_match+0x76(ffffff436f4dc7c0, 0)
dtrace_enabling_matchall+0x4a()
dtrace_register+0x315(ffffff01efd91b30, ffffff01efd91cc8, 1c, ffffff43eaa4a9c8, 
ffffffffc01192a0, ffffff5466033be8)
fasttrap_provider_lookup+0x2f3(2dcc, ffffff43eb5a8bae, ffffff01efd91cc8)
fasttrap_meta_provide+0x93(0, ffffff01efd91cc0, 2dcc)
dtrace_helper_provide_one+0xf2(ffffff43ee186140, ffffff43eb5a80e0, 2dcc)
dtrace_helper_provide+0x6f(ffffff43ee186140, 2dcc)
dtrace_helper_provider_register+0x163(ffffff58cc298008, ffffff454aadf640, 0)
dtrace_helpers_duplicate+0x1df(ffffff43757700b0, ffffff58cc298008)
cfork+0x9e2(0, 1, 0)
forksys+0x3c(0, 0)
sys_syscall+0x17a()

The problem is that fasttrap_pid_enable() assumes that any failure to sprlock() a process from within a forking process must be because the target process is unborn – not that it is dead. As a result, we dereference the return value of prfind() – which is potentially NULL if the process is actually gone and we just happen to be within the context of a forking process. (Under this condition, we need to simply bail out.)