problem 1: mutex type from the attribute was being ignored by
pthread_mutex_init, so recursive/errorchecking mutexes were never
being used at all.
problem 2: ownership of recursive mutexes was not being enforced at
unlock time.
this code was written independently of musl, with support for a the
backwards, nonstandard "31-bit unicode" some libraries/apps might
want. unfortunately the extra code (inside #ifdef) makes the source
harder to read and makes code that should be simple look complex, so
i'm removing it. anyone who wants to use the old code can find it in
the history or from elsewhere.
also, change the visibility of the __fsmu8 state machine table to
hidden, if supported. this should improve performance slightly in
shared-library builds.
prefer using visibility=hidden for __libc internal data, rather than
an accessor function, if the compiler has visibility.
optimize with -O3 for PIC targets (shared library). without heavy
inlining, reloading the GOT register in small functions kills
performance. 20-30% size increase for a single libc.so is not a big
deal, compared to comparaible size increase in every static binaries.
use -Bsymbolic-functions, not -Bsymbolic. global variables are subject
to COPY relocations, and thus binding their addresses in the library
at link time will cause library functions to read the wrong (original)
copies instead of the copies made in the main program's bss section.
add entry point, _start, for dynamic linker.
this only made the function unnecessarily slow on systems with
unaligned access, but would of course crash on systems that can't do
unaligned accesses (none of which have ports yet).
prior to this change, a large portion of libc was unusable prior to
relocation by the dynamic linker, due to dependence on the global data
in the __libc structure and the need to obtain its address through the
GOT. with this patch, the accessor function __libc_loc is now able to
obtain the address of __libc via PC-relative addressing without using
the GOT. this means the majority of libc functionality is now
accessible right away.
naturally, the above statements all depend on having an architecture
where PC-relative addressing and jumps/calls are feasible, and a
compiler that generates the appropriate code.
this change is in preparation for upcoming PIC/shared library support.
the intent is to avoid going through the GOT, mainly so that dprintf
is operable immediately, prior to processing of relocations. having
dprintf accessible from the dynamic linker will make writing and
debugging the dynamic linker much easier.
this change is made with some reluctance, but i think it's for the
best. correct programs must handle either behavior, so there is little
advantage to having malloc(0) return NULL. and i managed to actually
make the malloc code slightly smaller with this change.
do not allow allocations that overflow ptrdiff_t; fix some overflow
checks that were not quite right but didn't matter due to address
layout implementation.
this is needed in the long term for ABI compatibility anyway, and in
the immediate, it helps with building broken programs like GNU screen
that try to prototype the functions themselves rather than using the
header.
the existence of a (kernelspace) thread must never have observable
effects after the thread count is decremented. if signals are not
blocked, it could end up handling the signal for rsyscall and
contributing towards the count of threads which have changed ids,
causing a thread to be missed. this could lead to one thread retaining
unwanted privilege level.
this change may also address other subtle race conditions in
application code that uses signals.
the historic mktemp is supposed to blank the template string on
failure, rather than returning 0. just zero the first character so
that mkstemp and mkdtemp can still retry with O(1) space requirement.
note that this presently does not handle consistency of the libc's own
global state during forking. as per POSIX 2008, if the parent process
was threaded, the child process may only call async-signal-safe
functions until one of the exec-family functions is called, so the
current behavior is believed to be conformant even if non-ideal. it
may be improved at some later time.
use current time in nanoseconds and some potentially-random (if aslr
is enabled) pointer values for the initial tempfile name generation,
and step via a cheap linear prng on collisions. limit the number of
retry attempts to prevent denial of service attacks even if an
attacker can guess the filenames.
Rich Felker