glibc/hurd/STATUS

65 lines
1.8 KiB
Plaintext

Status of Hurd support in libc. Last updated 21 Feb 95.
Roland McGrath <roland@gnu.ai.mit.edu>
Everything not noted below is implemented, most of it tested. There are
various very small things unfinished or thought to be perhaps wrong
throughout the code, marked by comments containing `XXX'.
* We are not sure about possible races between setpgrp (A, pgrp) from
process B vs process A receiving proc_newids.
* The rest of libc (stdio et al) is not safe for multithreaded programs.
mutex locks should be added to various things.
* Recovery from faults in the signal thread is not implemented yet.
* longjmp needs to clean up reply port, intr_port; needs thought about.
* Cooperation with cthreads is not finished. If you link with cthreads,
libc internal code still does not use real condition variables.
sigsuspend currently does a busy wait where it should use a condition.
Signal state is per kernel thread; for unwired cthreads it should be per
cthread instead.
* sigaltstack/sigstack do not really work: the signal stack needs thread
variables and cthread data set up, which is not done.
* malloc is a kludge.
* Resource limits do not really work; current implementation is patchy and
inconsistent.
* libc implicitly uses some environment variables. This is a security
problem for setuid exec. Probably crt0 should remove the variables from
the environment if setuid.
* The miscellaneous msg.defs calls are only partially implemented.
* The default SIGINFO handler needs to be written.
* File locking is not implemented; the RPC interface is not there yet.
* The current getitimer/setitimer implementation is a kludge.
* mmap cannot do MAP_NOEXTEND.
* Unimplemented calls (from the 4.4 system call list):
acct
fstatfs
getfh
getfsstat
getrusage
madvise
mincore
mount
msync
profil
recvmsg
revoke
sendmsg
sstk
statfs
swapon
unmount