Hello,

This commit adds two new environment variables (AFL_FORKSRV_UID and AFL_FORKSRV_GID) that allow, when running the target using the fork server, to use specific user and group IDs.

Example of Linux session:

# Add required permissions to AFL binary
setcap "cap_setuid,cap_setgid,cap_chown+ep" afl-fuzz

# Fuzz the target using custom UID (1001) and GIDs (1001 main group, 23 and 42 supplementary groups)
AFL_FORKSRV_UID=1001 AFL_FORKSRV_GID=1001,23,42 afl-fuzz [AFL options and target]

Under the hood, the fork server calls setregid, setgroups and setreuid right before launching the target using execv. Unfortunately when the UID and GIDs of afl-fuzz and the forkserver are different, the communication between them is faulty: files and shared memory have default permission rights set to 600 and UID/GID set to the ones of afl-fuzz. When only the GID is modified this isn't problematic (access granted using the common UID) but when the UID is modified we need to set permission rights to 660 so the fork server can access entities using the common group ID. In case both are modified, we even need to change the group ownership of the entities so afl-buzz binary can access them using its UID and the fork server using its GID.

This new ownership scheme constitutes the most of the commit since the default permission and the potential new group has to be passed to all function creating communication entities (files and shared memory). I have modified the creation of the entities related to the forkserver and left the other one intact but I might have missed cases and thus I think a review by an experimented AFL developer is required.

Finally, I have used the constructions shown bellow several times. If there's macro more handy that automatically deals with errno please tell me so I can improve the code.

WARNF("Syscall failed: %s.", strerror(errno));
FATAL("Syscall failed: %s.", strerror(errno));

Thank you very much,

Pierre Graux