commits leading up to this one have moved the vast majority of
libc-internal interface declarations to appropriate internal headers,
allowing them to be type-checked and setting the stage to limit their
visibility. the ones that have not yet been moved are mostly
namespace-protected aliases for standard/public interfaces, which
exist to facilitate implementing plain C functions in terms of POSIX
functionality, or C or POSIX functionality in terms of extensions that
are not standardized. some don't quite fit this description, but are
"internally public" interfacs between subsystems of libc.
rather than create a number of newly-named headers to declare these
functions, and having to add explicit include directives for them to
every source file where they're needed, I have introduced a method of
wrapping the corresponding public headers.
parallel to the public headers in $(srcdir)/include, we now have
wrappers in $(srcdir)/src/include that come earlier in the include
path order. they include the public header they're wrapping, then add
declarations for namespace-protected versions of the same interfaces
and any "internally public" interfaces for the subsystem they
correspond to.
along these lines, the wrapper for features.h is now responsible for
the definition of the hidden, weak, and weak_alias macros. this means
source files will no longer need to include any special headers to
access these features.
over time, it is my expectation that the scope of what is "internally
public" will expand, reducing the number of source files which need to
include *_impl.h and related headers down to those which are actually
implementing the corresponding subsystems, not just using them.
eliminate gratuitous glue function for reporting the version, which
was probably leftover from the old dynamic linker design which lacked
a clear barrier for when/how it could access global data. put the
declaration for the data object that replaces it in libc.h where it
can be type checked.
this cleans up what had become widespread direct inline use of "GNU C"
style attributes directly in the source, and lowers the barrier to
increased use of hidden visibility, which will be useful to recovering
some of the efficiency lost when the protected visibility hack was
dropped in commit dc2f368e56, especially
on archs where the PLT ABI is costly.
this both allows removal of some of the main remaining uses of the
SHARED macro and clears one obstacle to static-linked dlopen support,
which may be added at some point in the future.
specialized single-TLS-module versions of __copy_tls and __reset_tls
are removed and replaced with code adapted from their dynamic-linked
versions, capable of operating on a whole chain of TLS modules, and
use of the dynamic linker's DSO chain (which contains large struct dso
objects) by these functions is replaced with a new chain of struct
tls_module objects containing only the information needed for
implementing TLS. this may also yield some performance benefit
initializing TLS for a new thread when a large number of modules
without TLS have been loaded, since since there is no need to walk
structures for modules without TLS.
functions which open in-memory FILE stream variants all shared a tail
with __fdopen, adding the FILE structure to stdio's open file list.
replacing this common tail with a function call reduces code size and
duplication of logic. the list is also partially encapsulated now.
function signatures were chosen to facilitate tail call optimization
and reduce the need for additional accessor functions.
with these changes, static linked programs that do not use stdio no
longer have an open file list at all.
previously, LC_MESSAGES was treated specially as the only category
which could be set to a locale name without a definition file, in
order to facilitate gettext message translations when no libc locale
was available. LC_NUMERIC was completely un-settable, and LC_CTYPE
stored a flag intended to be used for a possible future byte-based C
locale, instead of storing a __locale_map pointer like the other
categories use.
this patch changes all categories to be represented by pointers to
__locale_map structures, and allows locale names without definition
files to be treated as valid locales with trivial definition when used
in any category. outwardly visible functional changes should be minor,
limited mainly to the strings read back from setlocale and the way
gettext handles translations in categories other than LC_MESSAGES.
various internal refactoring has also been performed, and improvements
in const correctness have been made.
the code being removed used atomics to track whether any threads might
be using a locale other than the current global locale, and whether
any threads might have abstract 8-bit (non-UTF-8) LC_CTYPE active, a
feature which was never committed (still pending). the motivations
were to support early execution prior to setup of the thread pointer,
to partially support systems (ancient kernels) where thread pointer
setup is not possible, and to avoid high performance cost on archs
where accessing the thread pointer may be very slow.
since commit 19a1fe670a, the thread
pointer is always available, so these hacks are no longer needed.
removing them greatly simplifies the affected code.
these were hacks to work around toolchains that could not properly
optimize PIC accesses based on visibility and would generate GOT
lookups even for hidden data, which broke the old dynamic linker.
since commit f3ddd17380 it no longer
matters; the dynamic linker does not assume accessibility of this data
until stage 3.
since 1.1.0, musl has nominally required a thread pointer to be setup.
most of the remaining code that was checking for its availability was
doing so for the sake of being usable by the dynamic linker. as of
commit 71f099cb7d, this is no longer
necessary; the thread pointer is now valid before any libc code
(outside of dynamic linker bootstrap functions) runs.
this commit essentially concludes "phase 3" of the "transition path
for removing lazy init of thread pointer" project that began during
the 1.1.0 release cycle.
the memory model we use internally for atomics permits plain loads of
values which may be subject to concurrent modification without
requiring that a special load function be used. since a compiler is
free to make transformations that alter the number of loads or the way
in which loads are performed, the compiler is theoretically free to
break this usage. the most obvious concern is with atomic cas
constructs: something of the form tmp=*p;a_cas(p,tmp,f(tmp)); could be
transformed to a_cas(p,*p,f(*p)); where the latter is intended to show
multiple loads of *p whose resulting values might fail to be equal;
this would break the atomicity of the whole operation. but even more
fundamental breakage is possible.
with the changes being made now, objects that may be modified by
atomics are modeled as volatile, and the atomic operations performed
on them by other threads are modeled as asynchronous stores by
hardware which happens to be acting on the request of another thread.
such modeling of course does not itself address memory synchronization
between cores/cpus, but that aspect was already handled. this all
seems less than ideal, but it's the best we can do without mandating a
C11 compiler and using the C11 model for atomics.
in the case of pthread_once_t, the ABI type of the underlying object
is not volatile-qualified. so we are assuming that accessing the
object through a volatile-qualified lvalue via casts yields volatile
access semantics. the language of the C standard is somewhat unclear
on this matter, but this is an assumption the linux kernel also makes,
and seems to be the correct interpretation of the standard.
there is still no code which actually uses the loaded locale files, so
the main observable effect of this commit is that calls to setlocale
store and give back the names of the selected locales for the
remaining categories (LC_TIME, LC_COLLATE, LC_MONETARY) if a locale
file by the requested name could be loaded.
this commit adds non-stub implementations of setlocale, duplocale,
newlocale, and uselocale, along with the data structures and minimal
code needed for representing the active locale on a per-thread basis
and optimizing the common case where thread-local locale settings are
not in use.
at this point, the data structures only contain what is necessary to
represent LC_CTYPE (a single flag) and LC_MESSAGES (a name for use in
finding message translation files). representation for the other
categories will be added later; the expectation is that a single
pointer will suffice for each.
for LC_CTYPE, the strings "C" and "POSIX" are treated as special; any
other string is accepted and treated as "C.UTF-8". for other
categories, any string is accepted after being truncated to a maximum
supported length (currently 15 bytes). for LC_MESSAGES, the name is
kept regardless of whether libc itself can use such a message
translation locale, since applications using catgets or gettext should
be able to use message locales libc is not aware of. for other
categories, names which are not successfully loaded as locales (which,
at present, means all names) are treated as aliases for "C". setlocale
never fails.
locale settings are not yet used anywhere, so this commit should have
no visible effects except for the contents of the string returned by
setlocale.
this is the first step in an overhaul aimed at greatly simplifying and
optimizing everything dealing with thread-local state.
previously, the thread pointer was initialized lazily on first access,
or at program startup if stack protector was in use, or at certain
random places where inconsistent state could be reached if it were not
initialized early. while believed to be fully correct, the logic was
fragile and non-obvious.
in the first phase of the thread pointer overhaul, support is retained
(and in some cases improved) for systems/situation where loading the
thread pointer fails, e.g. old kernels.
some notes on specific changes:
- the confusing use of libc.main_thread as an indicator that the
thread pointer is initialized is eliminated in favor of an explicit
has_thread_pointer predicate.
- sigaction no longer needs to ensure that the thread pointer is
initialized before installing a signal handler (this was needed to
prevent a situation where the signal handler caused the thread
pointer to be initialized and the subsequent sigreturn cleared it
again) but it still needs to ensure that implementation-internal
thread-related signals are not blocked.
- pthread tsd initialization for the main thread is deferred in a new
manner to minimize bloat in the static-linked __init_tp code.
- pthread_setcancelstate no longer needs special handling for the
situation before the thread pointer is initialized. it simply fails
on systems that cannot support a thread pointer, which are
non-conforming anyway.
- pthread_cleanup_push/pop now check for missing thread pointer and
nop themselves out in this case, so stdio no longer needs to avoid
the cancellable path when the thread pointer is not available.
a number of cases remain where certain interfaces may crash if the
system does not support a thread pointer. at this point, these should
be limited to pthread interfaces, and the number of such cases should
be fewer than before.
if a multithreaded program became non-multithreaded (i.e. all other
threads exited) while one thread held an internal lock, the remaining
thread would fail to release the lock. the the program then became
multithreaded again at a later time, any further attempts to obtain
the lock would deadlock permanently.
the underlying cause is that the value of libc.threads_minus_1 at
unlock time might not match the value at lock time. one solution would
be returning a flag to the caller indicating whether the lock was
taken and needs to be unlocked, but there is a simpler solution: using
the lock itself as such a flag.
note that this flag is not needed anyway for correctness; if the lock
is not held, the unlock code is harmless. however, the memory
synchronization properties associated with a_store are costly on some
archs, so it's best to avoid executing the unlock code when it is
unnecessary.
PAGE_SIZE was hardcoded to 4096, which is historically what most
systems use, but on several archs it is a kernel config parameter,
user space can only know it at execution time from the aux vector.
PAGE_SIZE and PAGESIZE are not defined on archs where page size is
a runtime parameter, applications should use sysconf(_SC_PAGE_SIZE)
to query it. Internally libc code defines PAGE_SIZE to libc.page_size,
which is set to aux[AT_PAGESZ] in __init_libc and early in __dynlink
as well. (Note that libc.page_size can be accessed without GOT, ie.
before relocations are done)
Some fpathconf settings are hardcoded to 4096, these should be actually
queried from the filesystem using statfs.
modern (4.7.x and later) gcc uses init/fini arrays, rather than the
legacy _init/_fini function pasting and crtbegin/crtend ctors/dtors
system, on most or all archs. some archs had already switched a long
time ago. without following this change, global ctors/dtors will cease
to work under musl when building with new gcc versions.
the most surprising part of this patch is that it actually reduces the
size of the init code, for both static and shared libc. this is
achieved by (1) unifying the handling main program and shared
libraries in the dynamic linker, and (2) eliminating the
glibc-inspired rube goldberg machine for passing around init and fini
function pointers. to clarify, some background:
the function signature for __libc_start_main was based on glibc, as
part of the original goal of being able to run some glibc-linked
binaries. it worked by having the crt1 code, which is linked into
every application, static or dynamic, obtain and pass pointers to the
init and fini functions, which __libc_start_main is then responsible
for using and recording for later use, as necessary. however, in
neither the static-linked nor dynamic-linked case do we actually need
crt1.o's help. with dynamic linking, all the pointers are available in
the _DYNAMIC block. with static linking, it's safe to simply access
the _init/_fini and __init_array_start, etc. symbols directly.
obviously changing the __libc_start_main function signature in an
incompatible way would break both old musl-linked programs and
glibc-linked programs, so let's not do that. instead, the function can
just ignore the information it doesn't need. new archs need not even
provide the useless args in their versions of crt1.o. existing archs
should continue to provide it as long as there is an interest in
having newly-linked applications be able to run on old versions of
musl; at some point in the future, this support can be removed.
patch by Jens Gustedt.
previously, the intended policy was to use __environ in code that must
conform to the ISO C namespace requirements, and environ elsewhere.
this policy was not followed in practice anyway, making things
confusing. on top of that, Jens reported that certain combinations of
link-time optimization options were breaking with the inconsistent
references; this seems to be a compiler or linker bug, but having it
go away is a nice side effect of the changes made here.
on x86 and some other archs, functions which make function calls which
might go through a PLT incur a significant overhead cost loading the
GOT register prior to making the call. this load is utterly useless in
musl, since all calls are bound at library-creation time using
-Bsymbolic-functions, but the compiler has no way of knowing this, and
attempts to set the default visibility to protected have failed due to
bugs in GCC and binutils.
this commit simply manually assigns hidden/protected visibility, as
appropriate, to a few internal-use-only functions which have many
callers, or which have callers that are hot paths like getc/putc. it
shaves about 5k off the i386 libc.so with -Os. many of the
improvements are in syscall wrappers, where the benefit is just size
and performance improvement is unmeasurable noise amid the syscall
overhead. however, stdio may be measurably faster.
if in the future there are toolchains that can do the same thing
globally without introducing linking bugs, it might be worth
considering removing these workarounds.
with this change, pcc-built musl libc.so seems to work correctly. the
problem is that pcc generates GOT lookups for external-linkage symbols
even if they are hidden, rather than using GOT-relative addressing.
the entire reason we're using hidden visibility on the __libc object
is to make it accessible prior to relocations -- not to mention
inexpensive to access. unfortunately, the workaround makes it even
more expensive on pcc.
when the pcc issue is fixed, an appropriate version test should be
added so new pcc can use the much more efficient variant.
unlike other implementations, this one reserves memory for new TLS in
all pre-existing threads at dlopen-time, and dlopen will fail with no
resources consumed and no new libraries loaded if memory is not
available. memory is not immediately distributed to running threads;
that would be too complex and too costly. instead, assurances are made
that threads needing the new TLS can obtain it in an async-signal-safe
way from a buffer belonging to the dynamic linker/new module (via
atomic fetch-and-add based allocator).
I've re-appropriated the lock that was previously used for __synccall
(synchronizing set*id() syscalls between threads) as a general
pthread_create lock. it's a "backwards" rwlock where the "read"
operation is safe atomic modification of the live thread count, which
multiple threads can perform at the same time, and the "write"
operation is making sure the count does not increase during an
operation that depends on it remaining bounded (__synccall or dlopen).
in static-linked programs that don't use __synccall, this lock is a
no-op and has no cost.
the design for TLS in dynamic-linked programs is mostly complete too,
but I have not yet implemented it. cost is nonzero but still low for
programs which do not use TLS and/or do not use threads (a few hundred
bytes of new code, plus dependency on memcpy). i believe it can be
made smaller at some point by merging __init_tls and __init_security
into __libc_start_main and avoiding duplicate auxv-parsing code.
at the same time, I've also slightly changed the logic pthread_create
uses to allocate guard pages to ensure that guard pages are not
counted towards commit charge.
it's expected that this will be needed/useful only in asm, so I've
given it its own symbol that can be addressed in pc-relative ways from
asm rather than adding a field in the __libc structure which would
require hard-coding the offset wherever it's used.
these will NOT be used when compiling with -D_LARGEFILE64_SOURCE on
musl; instead, they exist in the hopes of eventually being able to run
some glibc-linked apps with musl sitting in place of glibc.
also remove the (apparently incorrect) fcntl alias.
i made a best attempt, but the intended semantics of this function are
fundamentally contradictory. there is no consistent way to handle
ownership of locks when forking a multi-threaded process. the code
could have worked by accident for programs that only used normal
mutexes and nothing else (since they don't actually store or care
about their owner), but that's about it. broken-by-design interfaces
that aren't even in glibc (only solaris) don't belong in musl.
i did some testing trying to switch malloc to use the new internal
lock with priority inheritance, and my malloc contention test got
20-100 times slower. if priority inheritance futexes are this slow,
it's simply too high a price to pay for avoiding priority inversion.
maybe we can consider them somewhere down the road once the kernel
folks get their act together on this (and perferably don't link it to
glibc's inefficient lock API)...
as such, i've switch __lock to use malloc's implementation of
lightweight locks, and updated all the users of the code to use an
array with a waiter count for their locks. this should give optimal
performance in the vast majority of cases, and it's simple.
malloc is still using its own internal copy of the lock code because
it seems to yield measurably better performance with -O3 when it's
inlined (20% or more difference in the contention stress test).
we use priority inheritance futexes if possible so that the library
cannot hit internal priority inversion deadlocks in the presence of
realtime priority scheduling (full support to be added later).
in gcc 3, the visibility attribute must be placed on both the
declaration and on the definition. if it's omitted from the
definition, the compiler fails to emit the ".hidden" directive in the
assembly, and the linker will either generate textrels (if supported,
such as on i386) or refuse to link (on targets where certain types of
textrels are forbidden or impossible without further assumptions about
memory layout, such as on x86_64).
this patch also unifies the decision about when to use visibility into
libc.h and makes the visibility in the utf-8 state machine tables
based on libc.h rather than a duplicate test.
this behavior (opening fds 0-2 for a suid program) is explicitly
allowed (but not required) by POSIX to protect badly-written suid
programs from clobbering files they later open.
this commit does add some cost in startup code, but the availability
of auxv and the security flag will be useful elsewhere in the future.
in particular auxv is needed for static-linked vdso support, which is
still waiting to be committed (sorry nik!)
fix up clone signature to match the actual behavior. the new
__syncall_wait function allows a __synccall callback to wait for other
threads to continue without returning, so that it can resume action
after the caller finishes. this interface could be made significantly
more general/powerful with minimal effort, but i'll wait to do that
until it's actually useful for something.
previously, stdio used spinlocks, which would be unacceptable if we
ever add support for thread priorities, and which yielded
pathologically bad performance if an application attempted to use
flockfile on a key file as a major/primary locking mechanism.
i had held off on making this change for fear that it would hurt
performance in the non-threaded case, but actually support for
recursive locking had already inflicted that cost. by having the
internal locking functions store a flag indicating whether they need
to perform unlocking, rather than using the actual recursive lock
counter, i was able to combine the conditionals at unlock time,
eliminating any additional cost, and also avoid a nasty corner case
where a huge number of calls to ftrylockfile could cause deadlock
later at the point of internal locking.
this commit also fixes some issues with usage of pthread_self
conflicting with __attribute__((const)) which resulted in crashes with
some compiler versions/optimizations, mainly in flockfile prior to
pthread_create.
changing credentials in a multi-threaded program is extremely
difficult on linux because it requires synchronizing the change
between all threads, which have their own thread-local credentials on
the kernel side. this is further complicated by the fact that changing
the real uid can fail due to exceeding RLIMIT_NPROC, making it
possible that the syscall will succeed in some threads but fail in
others.
the old __rsyscall approach being replaced was robust in that it would
report failure if any one thread failed, but in this case, the program
would be left in an inconsistent state where individual threads might
have different uid. (this was not as bad as glibc, which would
sometimes even fail to report the failure entirely!)
the new approach being committed refuses to change real user id when
it cannot temporarily set the rlimit to infinity. this is completely
POSIX conformant since POSIX does not require an implementation to
allow real-user-id changes for non-privileged processes whatsoever.
still, setting the real uid can fail due to memory allocation in the
kernel, but this can only happen if there is not already a cached
object for the target user. thus, we forcibly serialize the syscalls
attempts, and fail the entire operation on the first failure. this
*should* lead to an all-or-nothing success/failure result, but it's
still fragile and highly dependent on kernel developers not breaking
things worse than they're already broken.
ideally linux will eventually add a CLONE_USERCRED flag that would
give POSIX conformant credential changes without any hacks from
userspace, and all of this code would become redundant and could be
removed ~10 years down the line when everyone has abandoned the old
broken kernels. i'm not holding my breath...
some functions that should have been testing whether pthread_self()
had been called and initialized the thread pointer were instead
testing whether pthread_create() had been called and actually made the
program "threaded". while it's unlikely any mismatch would occur in
real-world problems, this could have introduced subtle bugs. now, we
store the address of the main thread's thread descriptor in the libc
structure and use its presence as a flag that the thread register is
initialized. note that after fork, the calling thread (not necessarily
the original main thread) is the new main thread.
the goal is to be able to use pthread_setcancelstate internally in
the implementation, whenever a function might want to use functions
which are cancellation points but avoid becoming a cancellation point
itself. i could have just used a separate internal function for
temporarily inhibiting cancellation, but the solution in this commit
is better because (1) it's one less implementation-specific detail in
functions that need to use it, and (2) application code can also get
the same benefit.
previously, pthread_setcancelstate dependend on pthread_self, which
would pull in unwanted thread setup overhead for non-threaded
programs. now, it temporarily stores the state in the global libc
struct if threads have not been initialized, and later moves it if
needed. this way we can instead use __pthread_self, which has no
dependencies and assumes that the thread register is already valid.
this patch improves the correctness, simplicity, and size of
cancellation-related code. modulo any small errors, it should now be
completely conformant, safe, and resource-leak free.
the notion of entering and exiting cancellation-point context has been
completely eliminated and replaced with alternative syscall assembly
code for cancellable syscalls. the assembly is responsible for setting
up execution context information (stack pointer and address of the
syscall instruction) which the cancellation signal handler can use to
determine whether the interrupted code was in a cancellable state.
these changes eliminate race conditions in the previous generation of
cancellation handling code (whereby a cancellation request received
just prior to the syscall would not be processed, leaving the syscall
to block, potentially indefinitely), and remedy an issue where
non-cancellable syscalls made from signal handlers became cancellable
if the signal handler interrupted a cancellation point.
x86_64 asm is untested and may need a second try to get it right.
this is something of a tradeoff, as now set*id() functions, rather
than pthread_create, are what pull in the code overhead for dealing
with linux's refusal to implement proper POSIX thread-vs-process
semantics. my motivations are:
1. it's cleaner this way, especially cleaner to optimize out the
rsyscall locking overhead from pthread_create when it's not needed.
2. it's expected that only a tiny number of core system programs will
ever use set*id() functions, whereas many programs may want to use
threads, and making thread overhead tiny is an incentive for "light"
programs to try threads.
with these small changes, libc functions which need to call functions
which are cancellation points, but which themselves must not be
cancellation points, can use the CANCELPT_INHIBIT and CANCELPT_RESUME
macros to temporarily inhibit all cancellation.
we want to keep atomically updated fields (locks and thread count) and
really anything writable far away from frequently-needed function
pointers. stuff some rarely-needed function pointers in between to
pad, hopefully up to a cache line boundary.
- there is no longer any risk of spoofing cancellation requests, since
the cancel flag is set in pthread_cancel rather than in the signal
handler.
- cancellation signal is no longer unblocked when running the
cancellation handlers. instead, pthread_create will cause any new
threads created from a cancellation handler to unblock their own
cancellation signal.
- various tweaks in preparation for POSIX timer support.
this commit addresses two issues:
1. a race condition, whereby a cancellation request occurring after a
syscall returned from kernelspace but before the subsequent
CANCELPT_END would cause cancellable resource-allocating syscalls
(like open) to leak resources.
2. signal handlers invoked while the thread was blocked at a
cancellation point behaved as if asynchronous cancellation mode wer in
effect, resulting in potentially dangerous state corruption if a
cancellation request occurs.
the glibc/nptl implementation of threads shares both of these issues.
with this commit, both are fixed. however, cancellation points
encountered in a signal handler will not be acted upon if the signal
was received while the thread was already at a cancellation point.
they will of course be acted upon after the signal handler returns, so
in real-world usage where signal handlers quickly return, it should
not be a problem. it's possible to solve this problem too by having
sigaction() wrap all signal handlers with a function that uses a
pthread_cleanup handler to catch cancellation, patch up the saved
context, and return into the cancellable function that will catch and
act upon the cancellation. however that would be a lot of complexity
for minimal if any benefit...
Rich Felker
Szabolcs Nagy