If a binary gets invoked by passing it as argument to ld.so the stack
still holds the auxiliary vector of ld.so when entering the _start
routine of the executable. So the invocation via ld.so is not fully
transparent to the executable. This causes problems if the executable
wants to scan the auxv itself.
grantpt was performing two consecutive calls to stat with the same
file name. Avoid this by creating a special version of the ptsname
function which allows to pass the stat result back to the caller.
The pt_chown program is completely transparently called. It might
not be able to live with the various file descriptors the program
has open at the time of the call (e.g., under SELinux). Close all
but the needed descriptor and connect stdin, stdout, and stderr
with /dev/null. pt_chown shouldn't print anything when called to
do real work.
This is a minor optimization. The tty group mustn't change so a
successful call to getgrnam will always return the same information.
Cache it and reuse it.
The ntp_gettime implementation of NTP exports the tai field the kernel
now produces. This requires an ABI change since the ntptimeval structure
changed. Upstream kept the same name, there is nothing to do. This
patch changes the ntptimeval structure but keeps the old ntp_gettime
definition. A new ntp_gettimex function which is transparently invoked
through the old name is introduced. This has the advantage that even
object files can remain compatible. This wouldn't be the case if
symbol versioning would be used to overload the name ntp_gettime.
I've noticed that sync_file_range is a stub on ppc/ppc64.
The kernel on these arches provides sync_file_range2 syscall with swapped
parameters.
The following completely untested patch ought to fix this.
Due to alignment of 64bit parameters there is a dummy second argument.
But other than that the syscall arguments are directly mapped to the
function arguments.
I've just committed STT_GNU_IFUNC ppc/ppc64 support into prelink,
and this patch is needed on the glibc side. Without it ld.so segfaults,
as in dl-conflict.c sym_map is always NULL. While dl-machine.h could use
RESOLVE_CONFLICT_FIND_MAP macro to compute it, it doesn't make sense,
because with prelink we know it is already properly relocated (all relative
relocations are applied by prelink).
As reported in http://bugzilla.redhat.com/533063 , preadv/pwritev prototypes
are wrong on 32-bit arches with -D_FILE_OFFSET_BITS=64 and as I've just
found, fallocate is wrong too.
The problem is that only off_t is remapped to the 64-bit type transparently,
__off_t is not.
The implementation of posix_openpt on Linux can fail in a few extra
ways if the appropriate pseudo filesystems are not mounted etc. In
some of these cases we have to explicitly set errno.
If a second call to ttyname is not for the same type of device (e.g.,
serial vs ptty) the prefix of the buffer was wrong. Don't rely on
the previous content, always reinitialize it.
The syscall conventions on some Linux archs prevented F_GETOWN from working
correctly in some situations. This can be rectified when using the new
F_GETOWN_EX command.
tst-longjmp_chk passes, tst-longjmp_chk2 fails but that is because
of some limitations of kernel signal delivery on sparc that I need
to fix, it has nothing to do with the longjmp_chk implementation.
(The problem with tst-longjmp_chk2 is that it tries to do a stack
fault SIGSEGV within a stack fault SIGSEGV , and the Linux kernel
will refuse to setup the signal stack and deliver the signal if the
register windows can't be written out to the stack first)
