there are several reasons for this. some of them are related to race
conditions that arise since fork is required to be async-signal-safe:
if fork or pthread_create is called from a signal handler after the
fork syscall has returned but before the subsequent userspace code has
finished, inconsistent state could result. also, there seem to be
kernel and/or strace bugs related to arrival of signals during fork,
at least on some versions, and simply blocking signals eliminates the
possibility of such bugs.

this commit does not add versioning support; it merely fixes incorrect
lookups of symbols in libraries that contain versioned symbols.
previously, the version information was completely ignored, and
empirically this seems to have resulted in the oldest version being
chosen, but I am uncertain if that behavior was even reliable.

the new behavior being introduced is to completely ignore symbols
which are marked "hidden" (this seems to be the confusing nomenclature
for non-current-version) when versioning is present. this should solve
all problems related to libraries with symbol versioning as long as
all binaries involved are up-to-date (compatible with the
latest-version symbols), and it's the needed behavior for dlsym under
all circumstances.

at this point, it is just the common base charset equivalent to
Windows CP 950, with no further extensions. HKSCS and possibly other
supersets will be added later. other aliases may need to be added too.

the (obsolete) standard allows either 0 or 1 for the decimal point
location in this case, but since the number of zero digits returned in
the output string (in this implementation) is one more than the number
of digits the caller requested, it makes sense for the decimal point
to be logically "after" the first digit. in a sense, this change goes
with the previous commit which fixed the value of the decimal point
location for non-zero inputs.

these functions are obsolete and have no modern standard. the text in
SUSv2 is highly ambiguous, specifying that "negative means to the left
of the returned digits", which suggested to me that 0 would mean to
the right of the first digit. however, this does not agree with
historic practice, and the Linux man pages are more clear, specifying
that a negative value means "that the decimal point is to the left of
the start of the string" (in which case, 0 would mean the start of the
string, in accordance with historic practice).

like for other character sets, stateful iso-2022 form is not supported
yet but everything else should work. all charset aliases are treated
the same, as Windows codepage 949, because reportedly the EUC-KR
charset name is in widespread (mis?)usage in email and on the web for
data which actually uses the extended characters outside the standard
93x94 grid. this could easily be changed if desired.

the principle of this converter for handling the giant bulk of rare
Hangul syllables outside of the standard KS X 1001 93x94 grid is the
same as the GB18030 converter's treatment of non-explicitly-coded
Unicode codepoints: sequences in the extension range are mapped to an
integer index N, and the converter explicitly computes the Nth Hangul
syllable not explicitly encoded in the character map. empirically,
this requires at most 7 passes over the grid. this approach reduces
the table size required for Korean legacy encodings from roughly 44k
to 17k and should have minimal performance impact on real-world text
conversions since the "slow" characters are rare. where it does have
impact, the cost is merely a large constant time factor.

unblocking it in the pthread_once init function is not sufficient,
since multiple threads, some of them with the signal blocked, could
already exist before this is called; timers started from such threads
would be non-functional.

this is needed for reused threads in the SIGEV_THREAD timer
notification system, and could be reused elsewhere in the future if
needed, though it should be refactored for such use.

for static linking, __init_tls.c is simply modified to export the TLS
info in a structure with external linkage, rather than using statics.
this perhaps makes the code more clear, since the statics were poorly
named for statics. the new __reset_tls.c is only linked if it is used.

for dynamic linking, the code is in dynlink.c. sharing code with
__copy_tls is not practical since __reset_tls must also re-zero
thread-local bss.

1. the thread result field was reused for storing a kernel timer id,
but would be overwritten if the application code exited or cancelled
the thread.

2. low pointer values were used as the indicator that the timer id is
a kernel timer id rather than a thread id. this is not portable, as
mmap may return low pointers on some conditions. instead, use the fact
that pointers must be aligned and kernel timer ids must be
non-negative to map pointers into the negative integer space.

3. signals were not blocked until after the timer thread started, so a
race condition could allow a signal handler to run in the timer thread
when it's not supposed to exist. this is mainly problematic if the
calling thread was the only thread where the signal was unblocked and
the signal handler assumes it runs in that thread.

this is another case of the kernel syscall failing to support flags
where it needs to, leading to horrible workarounds in userspace. this
time the workaround requires changing uid/gid, and that's not safe to
do in the current process. in the worst case, kernel resource limits
might prevent recovering the original values, and then there would be
no way to safely return. so, use the safe but horribly inefficient
alternative: forking. clone is used instead of fork to suppress
signals from the child.

fortunately this worst-case code is only needed when effective and
real ids mismatch, which mainly happens in suid programs.

it turns out Linux is buggy for faccessat, just like fchmodat: the
kernel does not actually take a flags argument. so we're going to have
to emulate it there.

this is mainly for ABI compat purposes.

this change is to align with a change in the glibc interface.

patch by nsz. the actual object the caller has storing the tree root
has type void *, so accessing it as struct node * is not valid.
instead, simply access the value, move it to a temporary of the
appropriate type and work from there, then move the result back.

check in configure to be polite (failing early if we're going to fail)
and in vfprintf.c since that is the point at which a mismatching type
would be extremely dangerous.

it was already declared in stdlib.h, but not defined anywhere.

this is a nonstandard extension.

on newer kernels, fchdir and fstat work anyway. this same fix should
be applied to any other syscalls that are similarly affected.

with this change, the current definitions of O_SEARCH and O_EXEC as
O_PATH are mostly conforming to POSIX requirements. the main remaining
issue is that O_NOFOLLOW has different semantics.

I intend to add more Linux workarounds that depend on using these
pathnames, and some of them will be in "syscall" functions that, from
an anti-bloat standpoint, should not depend on the whole snprintf
framework.

previously, the AT_SYMLINK_NOFOLLOW flag was ignored, giving
dangerously incorrect behavior -- the target of the symlink had its
modes changed to the modes (usually 0777) intended for the symlink).
this issue was amplified by the fact that musl provides lchmod, as a
wrapper for fchmodat, which some archival programs take as a sign that
symlink modes are supported and thus attempt to use.

emulating AT_SYMLINK_NOFOLLOW was a difficult problem, and I
originally believed it could not be solved, at least not without
depending on kernels newer than 3.5.x or so where O_PATH works halfway
well. however, it turns out that accessing O_PATH file descriptors via
their pseudo-symlink entries in /proc/self/fd works much better than
trying to use the fd directly, and works even on older kernels.
moreover, the kernel has permanently pegged these references to the
inode obtained by the O_PATH open, so there should not be race
conditions with the file being moved, deleted, replaced, etc.

this is the modern way, and the only way that makes any sense. glibc
has this complicated mechanism with RPATH and RUNPATH that controls
whether RPATH is processed before or after LD_LIBRARY_PATH, presumably
to support legacy binaries, but there is no compelling reason to
support this, and better behavior is obtained by just fixing the
search order.

this fixes an oversight in the previous commit.

previously, errno could be meaningless when the caller wrote it to the
dlerror string or stderr. try to make it meaningful. also, fix
incorrect check for over-long program headers and instead actually
support them by allocating memory if needed.

this can only happen for invalid library files, but they were not
detected reliably because the variable was uninitialized.

the access function cannot be used to check for existence, because it
operates using real uid/gid rather than effective to determine
accessibility; this matters for the non-final path components.
instead, use stat. failure of stat is success if only the final
component is missing (ENOENT) and otherwise is failure.

also refactor mkdtemp based on new shared temp code, removing
dependency on the deprecated mktemp, whose behavior made this logic
more difficult.

the concept of both versions is the same; they differ only in details.
for long runs, they use "rep movsl" or "rep movsq", and for small
runs, they use a trick, writing from both ends towards the middle,
that reduces the number of branches needed. in addition, if memset is
called multiple times with the same length, all branches will be
predicted; there are no loops.

for larger runs, there are likely faster approaches than "rep", at
least on some cpu models. for 32-bit, it's unlikely that there is any
faster approach that does not require non-baseline instructions; doing
anything fancier would require inspecting cpu capabilities. for
64-bit, there may very well be faster versions that work on all
models; further optimization could be explored in the future.

with these changes, memset is anywhere between 50% faster and 6 times
faster, depending on the cpu model and the length and alignment of the
destination buffer.

the original motivation for this patch was that qemu (and possibly
other syscall emulators) nop out madvise, resulting in an infinite
loop. however, there is another benefit to this change: madvise may
actually undo an explicit madvise the application intended for its
stack, whereas the mremap operation is a true nop. the logic here is
that mremap must fail if it cannot resize the mapping in-place, and
the caller knows that it cannot resize in-place because it knows the
next page of virtual memory is already occupied.

one of the arguments to memcmp may be shorter than the length l-3, and
memcmp is under no obligation not to access past the first byte that
differs. instead use strncmp which conveys the correct semantics. the
performance difference is negligible here and since the code is only
use for shared libc, both functions are already linked anyway.

the dev/inode for the main app and the dynamic linker ("interpreter")
are not available, so the subsequent checks don't work. in general we
don't want to make exact string matches to existing libraries prevent
loading new ones, since this breaks loading upgraded modules in
module-loading systems. so instead, special-case it.

the motivation for this fix is that calling dlopen on the names
returned by dl_iterate_phdr or walking the link map (obtained by
dlinfo) seem to be the only methods available to an application to
actually get a list of open dso handles.

reject elf files which are not ET_EXEC/ET_DYN type as bad exec format,
and reject ET_EXEC files when they cannot be loaded at the correct
address, since they are not relocatable at runtime. the main practical
benefit of this is to make dlopen of the main program fail rather than
producing an unsafe-to-use handle.

it's not clear to me why the linker even outputs these headers if they
are null, but apparently it does so. with the default startfiles, they
will never be null anyway, but this patch allows eliminating crti,
crtn, crtbegin, and crtend (leaving only crt1) if the toolchain is
using init_array/fini_array (or for a C-only, no-ctor environment).

in signal() it is needed since __sigaction uses restrict in parameters
and sharing the buffer is technically an aliasing error. do the same
for the syscall, as at least qemu-user does not handle it properly.

LC_GLOBAL_LOCALE refers to the global locale, controlled by setlocale,
not the thread-local locale in effect which these functions should be
using. neither LC_GLOBAL_LOCALE nor 0 has an argument to the *_l
functions has behavior defined by the standard, but 0 is a more
logical choice for requesting the callee to lookup the current locale.
in the future I may move the current locale lookup the the caller (the
non-_l-suffixed wrapper).

at this point, all of the locale logic is dummied out, so no harm was
done, but it should at least avoid misleading usage.

this change is in preparation for possibly adding support for the
field width and padding specifiers added in POSIX 2008.