as a result of commit e8e4e56a8c,
the later code path for setting optarg to a null pointer is no longer
necessary, and removing it eliminates an indention level and arguably
makes the code more readable.
the standard getopt does not touch optarg unless processing an option
with an argument. however, programs using the GNU getopt API, which we
attempt to provide in getopt_long, expect optarg to be a null pointer
after processing an option without an argument.
before argument permutation support was added, such programs typically
detected its absence and used their own replacement getopt_long,
masking the discrepency in behavior.
multi-threaded set*id and setrlimit use the internal __synccall
function to work around the kernel's wrongful treatment of these
process properties as thread-local. the old implementation of
__synccall failed to be AS-safe, despite POSIX requiring setuid and
setgid to be AS-safe, and was not rigorous in assuring that all
threads were caught. in a worst case, threads late in the process of
exiting could retain permissions after setuid reported success, in
which case attacks to regain dropped permissions may have been
possible under the right conditions.
the new implementation of __synccall depends on the presence of
/proc/self/task and will fail if it can't be opened, but is able to
determine that it has caught all threads, and does not use any locks
except its own. it thereby achieves AS-safety simply by blocking
signals to preclude re-entry in the same thread.
with this commit, all known conformance and safety issues in set*id
functions should be fixed.
per POSIX, the EINTR condition is an optional error for these
functions, not a mandatory one. since old kernels (pre-2.6.22) failed
to honor SA_RESTART for the futex syscall, it's dangerous to trust
EINTR from the kernel. thankfully POSIX offers an easy way out.
in the current version of __synccall, the callback is always run, so
failure to handle this case did not matter. however, the upcoming
overhaul of __synccall will have failure cases, in which case the
callback does not run and errno is already set. the changes being
committed now are in preparation for that.
the code being removed was introduced to work around "partial failure"
of multi-threaded set*id() operations, where some threads would
succeed in changing their ids but an RLIMIT_NPROC setting would
prevent the rest from succeeding, leaving the process in an
inconsistent and dangerous state. however, the workaround code did not
handle important usage cases like swapping real and effective uids
then restoring their original values, and the wrongful kernel
enforcement of RLIMIT_NPROC at setuid time was removed in Linux 3.1,
making the workaround obsolete.
since the partial failure still is dangerous on old kernels, and could
in principle happen on post-fix kernels as well if set*id() syscalls
fail for another spurious reason such as resource-related failures,
new code is added to detect and forcibly kill the process if/when such
a situation arises. future documentation releases should be updated to
reflect that setting RLIMIT_NPROC to RLIM_INFINITY is necessary to
avoid this forced-kill on old kernels. ideally, at some point the
kernel will get proper multi-threaded set*id() syscalls capable of
performing their actions atomically, and all of the userspace code to
emulate them can be treated as a fallback for outdated kernels.
opening /dev/tty then using ttyname_r on it does not produce a
canonical terminal name; it simply yields "/dev/tty".
it would be possible to make ctermid determine the actual controlling
terminal device via field 7 of /proc/self/stat, but doing so would
introduce a buffer overflow into applications built with L_ctermid==9,
which glibc defines, adversely affecting the quality of ABI compat.
commit b72cd07f17 added support for a
this feature in getopt, but it was later broken in the case where
getopt_long is used as a side effect of the changes made in commit
91184c4f16, which prevented the
underlying getopt call from seeing the leading '-' or '+' character in
optstring.
this commit changes the logic in the getopt_long core to check for a
leading colon, possibly after the leading '-' or '+', without
depending on the latter having been skipped by the caller. a minor
incorrectness in the return value for one error condition in
getopt_long is also fixed when opterr has been set to zero but
optstring has no leading ':'.
based on patch by Dima Krasner, with minor improvements for code size.
connect can fail if there is no listening syslogd, in which case a
useless socket was kept open, preventing subsequent syslog call from
attempting to connect again.
per the rules for hexadecimal integer constants, the previous
definitions were correctly treated as having unsigned type except
possibly when used in preprocessor conditionals, where all artithmetic
takes place as intmax_t or uintmax_t. the explicit 'u' suffix ensures
that they are treated as unsigned in all contexts.
based on discussion with and patches by Felix Janda. these changes
started as an effort to factor forkpty in terms of login_tty, which
returns an error and skips fd reassignment and closing if setting the
controlling terminal failed. the previous forkpty code was unable to
handle errors in the child, and did not attempt to; it just silently
ignored them. but this would have been unacceptable when switching to
using login_tty, since the child would start with the wrong stdin,
stdout, and stderr and thereby clobber the parent's files.
the new code uses the same technique as the posix_spawn implementation
to convey any possible error in the child to the parent so that the
parent can report failure to the caller. it is also safe against
thread cancellation and against signal delivery in the child prior to
the determination of success.
being a nonstandard function, this isn't strictly necessary, but it's
inexpensive and avoids unpleasant surprises. eventually I would like
all functions in libc to be safe against cancellation, either ignoring
it or acting on it cleanly.
not only is this semantically more correct; it also reduces code size
slightly by eliminating the need for the compiler to assume the
possibility of aliasing.
if writing the error message fails, POSIX requires that ferror(stderr)
be set. and as a function that operates on a stdio stream, getopt is
required to lock the stream it uses, stderr.
fwrite calls are used instead of fprintf since there is a demand from
some users not to pull in heavy stdio machinery via getopt. this
mimics the original code using write.
this shaves off a useless syscall for getting the caller's pid and
brings raise into alignment with other functions which were adapted to
use tkill rather than tgkill.
commit 83dc6eb087 documents the
rationale for this change, and in particular why the tgkill syscall is
useless for its designed purpose of avoiding races.
formally, it seems a sign is only required when the '+' modifier
appears in the format specifier, in which case either '+' or '-' must
be present in the output. but the specification is written such that
an optional negative sign is part of the output format anyway, and the
simplest approach to fixing the problem is removing the code that was
suppressing the sign.
it's unclear whether compilers which provide pure imaginary types
might produce a pure imaginary expression for 1.0fi. using 0.0f+1.0fi
ensures that the result is explicitly complex and makes this obvious
to human readers too.
based on patches by Jens Gustedt. these macros need to be usable in
static initializers, and the old definitions were not.
there is no portable way to provide correct definitions for these
macros unless the compiler supports pure imaginary types. a portable
definition is provided for this case even though there are presently
no compilers that can use it. gcc and compatible compilers provide a
builtin function that can be used, but clang fails to support this and
instead requires a construct which is a constraint violation and which
is only a constant expression as a clang-specific extension.
since these macros are a namespace violation in pre-C11 profiles, and
since no known pre-C11 compilers provide any way to define them
correctly anyway, the definitions have been made conditional on C11.
this avoids assuming the presence of C11 macro definitions in the
public complex.h, which need changes potentially incompatible with the
way these macros are being used internally.
based on patch by Timo Teräs, with some corrections to bounds checking
code and other minor changes.
while they are borderline scope creep, the functions added are fairly
small and are roughly the minimum code needed to use the results of
the res_query API without re-implementing error-prone DNS packet
parsing, and they are used in practice by some kerberos related
software and possibly other things. at this time there is no intent to
implement further nameser.h API functions.
previously, write errors neither stopped further output attempts nor
caused the function to return an error to the caller. this could
result in silent loss of output, possibly in the middle of output in
the event of a non-permanent error.
the simplest solution is temporarily clearing the error flag for the
target stream, then suppressing further output when the error flag is
set and checking/restoring it at the end of the operation to determine
the correct return value.
since the wide version of the code internally calls the narrow fprintf
to perform some of its underlying operations, initial clearing of the
error flag is suppressed when performing a narrow vfprintf on a
wide-oriented stream. this is not a problem since the behavior of
narrow operations on wide-oriented streams is undefined.
if argv permutation is used, the option terminator "--" should be
moved before any skipped non-option arguments rather than being left
in the argv tail where the caller will see and interpret it.
in the case where an initial '+' was passed in optstring (a
getopt_long feature to suppress argv permutation), getopt would fail
to see a possible subsequent ':', resulting in incorrect handling of
missing arguments.
the write function is a cancellation point and accesses thread-local
state belonging to the calling thread in the parent process. since
cancellation is blocked for the duration of posix_spawn, this is
probably safe, but it's fragile and unnecessary. making the syscall
directly is just as easy and clearly safe.
the resolution of austin group issue #370 removes the requirement that
posix_spawn fail when the close file action is performed on an
already-closed fd. since there are no other meaningful errors for
close, just ignoring the return value completely is the simplest fix.
the previous hard-coded offsets of +1 and +2 contained a hidden
assumption that the option character matched was single-byte, despite
this implementation of getopt attempting to support multibyte option
characters. this patch reworks the matching logic to leave the final
index pointing just past the matched character so that fixed offsets
can be used to check for ':'.
the sched_getaffinity syscall only fills a cpu set up to the set size
used/supported by the kernel. the rest is left untouched and userspace
is responsible for zero-filling it based on the return value of the
syscall.
the new DT_RUNPATH semantics for search order are always used, and
since binutils had always set both DT_RPATH and DT_RUNPATH when the
latter was used, processing only DT_RPATH worked fine. however, recent
binutils has stopped generating DT_RPATH when DT_RUNPATH is used,
which broke support for this feature completely.
this file had been a mess that went unnoticed ever since it was
imported. some lines used spaces for indention while others used tabs,
and tabs were used for alignment.
commit 27828f7e9a fixed compatibility
with clang's internal assembler, but broke compatibility with gas and
the traditional arm asm syntax by switching to the arm "unified
assembler language" (UAL). recent versions of gas also support UAL,
but require the .syntax directive to be used to switch to it. clang on
the other hand defaults to UAL. and old versions of gas (still
relevant) don't support UAL at all.
for the conditional ldm/stm instructions, "ia" is default and can just
be omitted, resulting in a mnemonic that's compatible with both
traditional and UAL syntax. but for byte/halfword loads and stores,
there seems to be no mnemonic compatible with both, and thus .word is
used to produce the desired opcode explicitly. the .inst directive is
not used because it is not compatible with older assemblers.
except powerpc, which still lacks inline syscalls simply because
nobody has written the code, these are all fallbacks used to work
around a clang bug that probably does not exist in versions of clang
that can compile musl. however, it's useful to have the generic
non-inline code anyway, as it eases the task of porting to new archs:
writing inline syscall code is now optional. this approach could also
help support compilers which don't understand inline asm or lack
support for the needed register constraints.
mips could not be unified because it has special fixup code for broken
layout of the kernel's struct stat.
the register constraints in the non-clang case were tested to work on
clang back to 3.2, and earlier versions of clang have known bugs that
preclude building musl.
there may be other reasons to prefer not to use inline syscalls, but
if so the function-call-based implementations should be added back in
a unified way for all archs.
calls to __aeabi_read_tp may be generated by the compiler to access
TLS on pre-v6 targets. previously, this function was hard-coded to
call the kuser helper, which would crash on kernels with kuser helper
removed.
to fix the problem most efficiently, the definition of __aeabi_read_tp
is moved so that it's an alias for the new __a_gettp. however, on v7+
targets, code to initialize the runtime choice of thread-pointer
loading code is not even compiled, meaning that defining
__aeabi_read_tp would have caused an immediate crash due to using the
default implementation of __a_gettp with a HCF instruction.
fortunately there is an elegant solution which reduces overall code
size: putting the native thread-pointer loading instruction in the
default code path for __a_gettp, so that separate default/native code
paths are not needed. this function should never be called before
__set_thread_area anyway, and if it is called early on pre-v6
hardware, the old behavior (crashing) is maintained.
ideally __aeabi_read_tp would not be called at all on v7+ targets
anyway -- in fact, prior to the overhaul, the same problem existed,
but it was never caught by users building for v7+ with kuser disabled.
however, it's possible for calls to __aeabi_read_tp to end up in a v7+
binary if some of the object files were built for pre-v7 targets, e.g.
in the case of static libraries that were built separately, so this
case needs to be handled.
previously, builds for pre-armv6 targets hard-coded use of the "kuser
helper" system for atomics and thread-pointer access, resulting in
binaries that fail to run (crash) on systems where this functionality
has been disabled (as a security/hardening measure) in the kernel.
additionally, builds for armv6 hard-coded an outdated/deprecated
memory barrier instruction which may require emulation (extremely
slow) on future models.
this overhaul replaces the behavior for all pre-armv7 builds (both of
the above cases) to perform runtime detection of the appropriate
mechanisms for barrier, atomic compare-and-swap, and thread pointer
access. detection is based on information provided by the kernel in
auxv: presence of the HWCAP_TLS bit for AT_HWCAP and the architecture
version encoded in AT_PLATFORM. direct use of the instructions is
preferred when possible, since probing for the existence of the kuser
helper page would be difficult and would incur runtime cost.
for builds targeting armv7 or later, the runtime detection code is not
compiled at all, and much more efficient versions of the non-cas
atomic operations are provided by using ldrex/strex directly rather
than wrapping cas.
in this case there are two conflicting rules in play: that an explicit
precision of zero with the value zero produces no output, and that the
'#' modifier for octal increases the precision sufficiently to yield a
leading zero. ISO C (7.19.6.1 paragraph 6 in C99+TC3) includes a
parenthetical remark to clarify that the precision-increasing behavior
takes precedence, but the corresponding text in POSIX off of which I
based the implementation is missing this remark.
this issue was covered in WG14 DR#151.
the kernel syscall interface for or1k does not expect 64-bit arguments
to be aligned to "even" register boundaries. this incorrect alignment
broke truncate/ftruncate and as well as a few less-common syscalls.
this change is a workaround for the inability of current compilers to
perform "shrink wrapping" optimizations. in casual testing, it roughly
doubled the performance of pthread_once when called on an
already-finished once control object.
these functions need to be fast when the init routine has already run,
since they may be called very often from code which depends on global
initialization having taken place. as such, a fast path bypassing
atomic cas on the once control object was used to avoid heavy memory
contention. however, on archs with weakly ordered memory, the fast
path failed to ensure that the caller actually observes the side
effects of the init routine.
preliminary performance testing showed that simply removing the fast
path was not practical; a performance drop of roughly 85x was observed
with 20 threads hammering the same once control on a 24-core machine.
so the new explicit barrier operation from atomic.h is used to retain
the fast path while ensuring memory visibility.
performance may be reduced on some archs where the barrier actually
makes a difference, but the previous behavior was unsafe and incorrect
on these archs. future improvements to the implementation of a_barrier
should reduce the impact.
previously, the hours were considered as a signed quantity while
minutes and seconds were always treated as positive offsets. however,
semantically the '-' sign should negate the whole hh:mm:ss offset.
this bug only affected timezones east of GMT with non-whole-hours
offsets, such as those used in India and Nepal.
previously the external definitions of these functions were omitted on
archs where long double is the same as double, since the code paths in
the math.h macros which would call them are unreachable. however, even
if they are unreachable, the definitions are still mandatory. omitting
them is invalid C, and in the case of a non-optimizing compiler, will
result in a link error.
per the text accepted for inclusion in POSIX, behavior is unspecified
when any of the access mode bits are set. since it's impossible to
consistently report this usage error (O_RDONLY could not be detected
since its value happens to be zero), the most consistent way to handle
them is just to ignore them.
previously, if a caller erroneously passed O_WRONLY, the resulting
access mode would be O_WRONLY|O_RDWR, which has the value 3, and this
resulted in a file descriptor which rejects both read and write
attempts when it is subsequently used.
this function is specified to leave the last byte with "unspecified
disposition" when the length is odd, so for the most part correct
programs should not be calling swab with odd lengths. however, doing
so is permitted, and should not write past the end of the destination
buffer.
patch by Jens Gustedt. this fixes a bug reported by Nadav Har'El. the
underlying issue was that a left-shift by 16 bits after promotion of
unsigned short to int caused integer overflow. while some compilers
define this overflow case as "shifting into the sign bit", doing so
doesn't help; the sign bit then gets extended through the upper bits
in subsequent arithmetic as unsigned long long. this patch imposes a
promotion to unsigned prior to the shift, so that the result is
well-defined and matches the specified behavior.
commit 5345c9b884 added a linked list to
track the FILE streams currently locked (via flockfile) by a thread.
due to a failure to fully link newly added members, removal from the
list could leave behind references which could later result in writes
to already-freed memory and possibly other memory corruption.
implicit stdio locking was unaffected; the list is only used in
conjunction with explicit flockfile locking.
this bug was not present in any releases; it was introduced and fixed
during the same release cycle.
patch by Timo Teräs, who discovered and tracked down the bug.
incorrect behavior occurred only in cases where the input overflows
unsigned long long, not just the (possibly lower) range limit for the
result type. in this case, processing of the '-' sign character was
not suppressed, and the function returned a value of 1 despite setting
errno to ERANGE.
the C11 _Alignas keyword is not present in C++, and despite it being
in the reserved namespace and thus reasonable to support even in
non-C11 modes, compilers seem to fail to support it.
as a result of commit ab8f6a6e42, this
definition is now equivalent to the actual "default profile" which
appears immediately below in features.h, and which defines both
_BSD_SOURCE and _XOPEN_SOURCE.
the intent of providing a _DEFAULT_SOURCE, which glibc also now
provides, is to give applications a way to "get back" the default
feature profile when it was lost either by compiler flags that inhibit
it (such as -std=c99) or by library-provided predefined macros (such
as -D_POSIX_C_SOURCE=200809L) which may inhibit exposure of features
that were otherwise visible by default and which the application may
need. without _DEFAULT_SOURCE, the application had encode knowledge of
a particular libc's defaults, and such knowledge was fragile and
subject to bitrot.
eventually the names _GNU_SOURCE and _BSD_SOURCE should be phased out
in favor of the more-descriptive and more-accurate _ALL_SOURCE and
_DEFAULT_SOURCE, leaving the old names as aliases but using the new
ones internally. however this is a more invasive change that would
require extensive regression testing, so it is deferred.
the vast majority of these failures seem to have been oversights at
the time _BSD_SOURCE was added, or perhaps shortly afterward. the one
which may have had some reason behind it is omission of setpgrp from
the _BSD_SOURCE feature profile, since the standard setpgrp interface
conflicts with a legacy (pre-POSIX) BSD interface by the same name.
however, such omission is not aligned with our general policy in this
area (for example, handling of similar _GNU_SOURCE cases) and should
not be preserved.
based on patch by Jens Gustedt.
the main difficulty here is handling the difference between start
function signatures and thread return types for C11 threads versus
POSIX threads. pointers to void are assumed to be able to represent
faithfully all values of int. the function pointer for the thread
start function is cast to an incorrect type for passing through
pthread_create, but is cast back to its correct type before calling so
that the behavior of the call is well-defined.
changes to the existing threads implementation were kept minimal to
reduce the risk of regressions, and duplication of code that carries
implementation-specific assumptions was avoided for ease and safety of
future maintenance.
based on patch by Jens Gustedt.
mtx_t and cnd_t are defined in such a way that they are formally
"compatible types" with pthread_mutex_t and pthread_cond_t,
respectively, when accessed from a different translation unit. this
makes it possible to implement the C11 functions using the pthread
functions (which will dereference them with the pthread types) without
having to use the same types, which would necessitate either namespace
violations (exposing pthread type names in threads.h) or incompatible
changes to the C++ name mangling ABI for the pthread types.
for the rest of the types, things are much simpler; using identical
types is possible without any namespace considerations.
this error resulted in an out-of-bounds read, as opposed to a reported
error, when calling the function with an argument one greater than the
max valid index.
if the loop stopped due to reaching the end of the string, the
subsequent increment could possibly move the position one past the end
of the buffer. no further writes happen, the reads cannot fault anyway
unless the stack completely lacks any zero bytes, and reading junk
should not yield an incorrect result from the function either.
nonetheless the code was wrong and needs to be fixed.
the condition was probably intended to be !*p rather than !p, but
neither is needed here. the subsequent code naturally handles the case
where it's already at end of string.
U+00DF ('ß') has had an uppercase form (U+1E9E) available since
Unicode 5.1, but Unicode lacks the case mappings for it due to
stability policy. when I added support for the new character in commit
1a63a9fc30, I omitted the mapping in the
lowercase-to-uppercase direction. this choice was not based on any
actual information, only assumptions.
this commit adds bidirectional case mappings between U+00DF and
U+1E9E, and removes the special-case hack that allowed U+00DF to be
identified as lowecase despite lacking a mapping. aside from strong
evidence that this is the "right" behavior for real-world usage of
these characters, several factors informed this decision:
- the other "potentially correct" mapping, to "SS", is not
representable in the C case-mapping system anyway.
- leaving one letter in lowercase form when transforming a string to
uppercase is obviously wrong.
- having a character which is nominally lowercase but which is fixed
under case mapping violates reasonable invariants.
per POSIX these functions are both cancellation points, so they must
act on any cancellation request which is pending prior to the call.
previously, only the code path where actual waiting took place could
act on cancellation.
the outer getnameinfo function already has a properly-sized temporary
buffer for storing the reverse dns (ptr) result. there is no reason
for the callback to use a secondary buffer and copy it on success, and
doing so potentially expanded the impact of the dn_expand bug that was
fixed in commit 49d2c8c6bc.
this change reduces the code size by a small amount, and also reduces
the run-time stack space requirements by about 256 bytes.
previously, fgets, fputs, fread, and fwrite completely omitted locking
and access to the FILE object when their arguments yielded a zero
length read or write operation independent of the FILE state. this
optimization was invalid; it wrongly skipped marking the stream as
byte-oriented (a C conformance bug) and exposed observably missing
synchronization (a POSIX conformance bug) where one of these functions
could wrongly complete despite another thread provably holding the
lock.
the C standard requires that "the contents of the array remain
unchanged" in this case.
this patch also changes the behavior on read errors, but in that case
"the array contents are indeterminate", so the application cannot
inspect them anyway.
this function is needed for some important practical applications of
ABI compatibility, and may be useful for supporting some non-portable
software at the source level too.
I was hesitant to add a function which imposes any constraints on
malloc internals; however, it turns out that any malloc implementation
which has realloc must already have an efficient way to determine the
size of existing allocations, so no additional constraint is imposed.
for now, some internal malloc definitions are duplicated in the new
source file. if/when malloc is refactored to put them in a shared
internal header file, these could be removed.
since malloc_usable_size is conventionally declared in malloc.h, the
empty stub version of this file was no longer suitable. it's updated
to provide the standard allocator functions, nonstandard ones (even if
stdlib.h would not expose them based on the feature test macros in
effect), and any malloc-extension functions provided (currently, only
malloc_usable_size).
if there is already a waiter for a lock, spinning on the lock is
essentially an attempt to steal it from whichever waiter would obtain
it via any priority rules in place, and is therefore undesirable. in
the current implementation, there is always an inherent race window at
unlock during which a newly-arriving thread may steal the lock from
the existing waiters, but we should aim to keep this window minimal
rather than enlarging it.
empirically, this increases the maximum rate of wait/post operations
between two threads by 20-150 times on machines I tested, including
x86 and arm. conceptually, it makes sense to do some spinning because
semaphores are intended to be usable as a notification mechanism
between threads, not just as locks, and low-latency notification is a
valuable property to have.
the previous spin limit of 10000 was utterly unreasonable.
empirically, it could consume up to 200000 cycles, whereas a failed
futex wait (EAGAIN) typically takes 1000 cycles or less, and even a
true wait/wake round seems much less expensive.
the new counts (100 for general wait, 200 in barrier) were simply
chosen to be in the range of what's reasonable without having adverse
effects on casual micro-benchmark tests I have been running. they may
still be too high, from a standpoint of not wasting cpu cycles, but at
least they're a lot better than before. rigorous testing across
different archs and cpu models should be performed at some point to
determine whether further adjustments should be made.
conceptually, a_spin needs to be at least a compiler barrier, so the
compiler will not optimize out loops (and the load on each iteration)
while spinning. it should also be a memory barrier, or the spinning
thread might keep spinning without noticing stores from other threads,
thus delaying for longer than it should.
ideally, an optimal a_spin implementation that avoids unnecessary
cache/memory contention should be chosen for each arch, but for now,
the easiest thing is to perform a useless a_cas on the calling
thread's stack.
this is analogous commit fffc5cda10
which fixed the corresponding issue for mutexes.
the robust list can't be used here because the locks do not share a
common layout with mutexes. at some point it may make sense to simply
incorporate a mutex object into the FILE structure and use it, but
that would be a much more invasive change, and it doesn't mesh well
with the current design that uses a simpler code path for internal
locking and pulls in the recursive-mutex-like code when the flockfile
API is used explicitly.
the subsequent code in pthread_create and the code which copies TLS
initialization images to the new thread's TLS space assume that the
memory provided to them is zero-initialized, which is true when it's
obtained by pthread_create using mmap. however, when the caller
provides a stack using pthread_attr_setstack, pthread_create cannot
make any assumptions about the contents. simply zero-filling the
relevant memory in this case is the simplest and safest fix.
unfortunately this needs to be able to vary by arch, because of a huge
mess GCC made: the GCC definition, which became the ABI, depends on
quirks in GCC's definition of __alignof__, which does not match the
formal alignment of the type.
GCC's __alignof__ unexpectedly exposes the an implementation detail,
its "preferred alignment" for the type, rather than the formal/ABI
alignment of the type, which it only actually uses in structures. on
most archs the two values are the same, but on some (at least i386)
the preferred alignment is greater than the ABI alignment.
I considered using _Alignas(8) unconditionally, but on at least one
arch (or1k), the alignment of max_align_t with GCC's definition is
only 4 (even the "preferred alignment" for these types is only 4).
the main idea of the changes made is to have waiters wait directly on
the "barrier" lock that was used to prevent them from making forward
progress too early rather than first waiting on the atomic state value
and then attempting to lock the barrier.
in addition, adjustments to the mutex waiter count are optimized.
previously, each waking waiter decremented the count (unless it was
the first) then immediately incremented it again for the next waiter
(unless it was the last). this was a roundabout was of achieving the
equivalent of incrementing it once for the first waiter and
decrementing it once for the last.
previously, wake order could be unpredictable: if a waiter happened to
leave its futex wait on the state early, e.g. due to EAGAIN while
restarting after a signal handler, it could acquire the mutex out of
turn. handling this required ugly O(n) list walking in the unwait
function and accounting to remove waiters that already woke from the
list.
with the new changes, the "barrier" locks in each waiter node are only
unlocked in turn. in addition to simplifying the code, this seems to
improve performance slightly, probably by reducing the number of
accesses threads make to each other's stacks.
as an additional benefit, unrecoverable mutex re-locking errors
(mainly ENOTRECOVERABLE for robust mutexes) no longer need to be
handled with deadlock; they can be reported to the caller, since the
unlocking sequence makes it unnecessary to rely on the mutex to
synchronize access to the waiter list.
Rich Felker