this is a POSIX requirement.
also remove the gratuitous locking shenanigans and simply access f->fd
under control of the lock. there is no advantage to not doing so, and
it made the correctness non-obvious at best.
__aeabi_read_tp used to call c code, but that was incorrect as the
arm runtime abi specifies special pcs for this function: it is only
allowed to clobber r0, ip, lr and cpsr.
maintainer's note: the old code explicitly saved and restored all
general-purpose registers which are call-clobbered in the normal
calling convention, so it's unlikely that any real-world compilers
produced code that could break. however theoretically they could have
chosen to use floating point registers, in which case the caller's
values of those registers would be clobbered.
with async cancellation enabled, pthread_cancel(pthread_self())
deadlocked due to pthread_kill holding killlock which is needed by
pthread_exit.
this could be solved by making pthread_kill block signals around the
critical section, at least when the target thread is itself, but the
issue only arises for cancellation, and otherwise would just be
imposing unnecessary cost.
instead just have pthread_cancel explicitly check for async
self-cancellation and call pthread_exit(PTHREAD_CANCELED) directly
rather than going through the signal machinery.
This manifests itself in mktime if tm_isdst = 1 and the current TZ= is
a POSIX timezone specification. mktime would see that tm_isdst was set
to 0 by __secs_to_zone, and subtract 'oppoff' (dst_off) - gmtoff from
the resultant time. This meant that mktime returned a time that was
exactly double the GMT offset of the desired timezone when tm_isdst
was = 1.
commit 610c5a8524 changed the thread
pointer setup so tp points at the end of the pthread struct on arm,
but failed to update __aeabi_read_tp so it was off by 8.
this broke tls access in code that is compiled with -mtp=soft, which
is the default when target arch is pre armv6k or thumb1.
maintainer's note: no release versions are affected.
this is an obsolete error code from RFS, an obsolete predecessor of
NFS. POSIX documents it only as "Reserved", but maintains the
requirement that it be defined. as long as it is defined, it needs a
string for strerror to produce; the one chosen matches glibc and
documentation from other language runtimes I could find.
the code to perform rounding to the desired precision wrongly assumed
the long double mantissa was an integral number of nibbles (hex
digits) in length. this is true for 80-bit extended precision (64-bit
mantissa) but not for double (53) or quad (113).
scale the rounding value by 1<<(LDBL_MANT_DIG%4) to compensate.
the text of the specification for getopt's handling of options that
require an argument, which requires updating optarg and optind, does
not exclude the error case where the end of the argument list has been
reached. in that case, it is expected that optarg be assigned
argv[argc] (normally null) and optind be incremented by 2, resulting
in a value of argc+1.
POSIX requires the symlink function to fail with ENAMETOOLONG if the
link contents to be written exceed SYMLINK_MAX in length, but neither
Linux nor our syscall wrapper code enforce this. the value 255 for
SYMLINK_MAX is not meaningful and does not seem to have been motivated
by anything except perhaps a wrong assumption that a definition was
mandatory. it has been present (though moving through bits to
top-level limits.h) since the beginning of the project history.
[f]pathconf is entitled to return -1 as the limit for conf names for
which there is no hard limit, with the usual POSIX note that an
indefinite limit does not imply an infinite limit. in principle we
might should report a limit for filesystems that impose one, but such
functionality is not currently present for any of the pathconf limits,
and adding it is beyond the scope of fixing the incorrect limit.
Call SYS_exit on return from fn in __clone. This is the expected
behavior of this function. Without this the child task will crash on
return from fn, since it will return to nowhere.
due to moved code, commit b8742f3260
inadvertently used the return value of __clone, rather than the return
value of SYS_sched_setscheduler in the new thread, to check whether it
needed to report failure. since a successful __clone returns the tid
of the new thread, which is never zero, this caused pthread_create
always to return with an invalid error number in the code path for
PTHREAD_EXPLICIT_SCHED.
this regression was not present in any releases.
the sign character produced came from the sign of tm_gmtoff/3600 as an
integer division, which is zero for negative offsets smaller in
magnitude than 3600. instead of printing the hours and minutes as
separate fields, print them as a single value of the form
hours*100+minutes, which naturally has the correct sign.
this fixes a major gap in the intended functionality of
pthread_setattr_default_np. if application/library code creating a
thread does not pass a null attribute pointer to pthread_create, but
sets up an attribute object to change other properties while leaving
the stack alone, the created thread will get a stack with size
DEFAULT_STACK_SIZE. this makes pthread_setattr_default_np useless for
working around stack overflow issues in such applications, and leaves
a major risk of regression if previously-working code switches from
using a null attribute pointer to an attribute object.
this change aligns the behavior more closely with the glibc
pthread_setattr_default_np functionality too, albeit via a different
mechanism. glibc encodes "default" specially in the attribute object
and reads the actual default at thread creation time. with this
commit, we now copy the current default into the attribute object at
pthread_attr_init time, so that applications that query the properties
of the attribute object will see the right values.
maintainer's note: the key observation here is that the compared
element is the first slot of the second ceil(half) of the array, and
thus can be removed for further comparison when it does not match, so
that we descend into the second ceil(half)-1 rather than ceil(half)
elements. this change ensures that nel strictly decreases with each
iteration, so that the case of != but nel==1 does not need to be
special-cased anymore.
commit 32482f61da reduced the number of
int members before the dirent buf from 4 to 3, thereby misaligning it
mod sizeof(off_t), producing invalid accesses on any arch where
alignof(off_t)==sizeof(off_t).
rather than re-adding wasted padding, reorder the struct to meet the
requirement and add a comment and static assertion to prevent this
from getting broken again.
this flag is notoriously under-/mis-specified, and in the past it was
implemented as a nop, essentially considering the absence of a
loopback interface with 127.0.0.1 and ::1 addresses an unsupported
configuration. however, common real-world container environments omit
IPv6 support (even for the network-namespaced loopback interface), and
some kernels omit IPv6 support entirely. future systems on the other
hand might omit IPv4 entirely.
treat these as supported configurations and suppress results of the
unconfigured/unsupported address families when AI_ADDRCONFIG is
requested. use routability of the loopback address to make the
determination; unlike other implementations, we do not exclude
loopback from the "an address is configured" condition, since there is
no basis in the specification for such exclusion. obtaining a result
with AI_ADDRCONFIG does not imply routability of the result, and
applications must still be able to cope with unroutable results even
if they pass AI_ADDRCONFIG.
commit 0b80a7b040, which added non-stub
setvbuf, applied the UNGET pushback adjustment to the size of the
buffer passed in, but inadvertently omitted offsetting the start by
the same amount, thereby allowing unget to clobber up to 8 bytes
before the start of the buffer. this bug was introduced in the present
release cycle; no releases are affected.
to produce sorted results roughly corresponding to RFC 3484/6724,
__lookup_name computes routability and choice of source address via
dummy UDP connect operations (which do not produce any packets). since
at the logical level, the properties fed into the sort key are
computed on ipv6 addresses, the code was written to use the v4mapped
ipv6 form of ipv4 addresses and share a common code path for them all.
however, on kernels where ipv6 support has been completely omitted,
this causes ipv4 to appear equally unroutable as ipv6, thereby putting
unreachable ipv6 addresses before ipv4 addresses in the results.
instead, use only ipv4 sockets to compute routability for ipv4
addresses. some gratuitous conversion back and forth is left so that
the logic is not affected by these changes. it may be possible to
simplify the ipv4 case considerably, thereby reducing code size and
complexity.
maintainer's note: past sentiment was that, despite being imperfect
and unable to force clearing of all possible copies of sensitive data
(e.g. in registers, register spills, signal contexts left on the
stack, etc.) this function would be added if major implementations
agreed on it, which has happened -- several BSDs and glibc all include
it.
maintainer's note: this change is for conformance with RFC 5952,
4.2.2, which explicitly forbids use of :: to shorten a single 16-bit 0
field when producing the canonical text representation for an IPv6
address. fixes a test failure reported by Philip Homburg, who also
submitted a patch, but this fix is simpler and should produce smaller
code.
if a final dot was included in the queried host name to anchor it to
the dns root/suppress search domains, and the result was not a CNAME,
the returned canonical name included the final dot. this was not
consistent with other implementations, confused some applications, and
does not seem desirable.
POSIX specifies returning a pointer to, or to a copy of, the input
nodename, when the canonical name is not available, but does not
attempt to specify what constitutes "not available". in the case of
search, we already have an implementation-defined "availability" of a
canonical name as the fully-qualified name resulting from search, so
defining it similarly in the no-search case seems reasonable in
addition to being consistent with other implementations.
as a bonus, fix the case where more than one trailing dot is included,
since otherwise the changes made here would wrongly cause lookups with
two trailing dots to succeed. previously this case resulted in
malformed dns queries and produced EAI_AGAIN after a timeout. now it
fails immediately with EAI_NONAME.
mlock2 syscall was added in linux v4.4 and glibc has api for it.
It falls back to mlock in case of flags==0, so that case works
even on older kernels.
MLOCK_ONFAULT is moved under _GNU_SOURCE following glibc.
the mode member of struct ipc_perm is specified by POSIX to have type
mode_t, which is uniformly defined as unsigned int. however, Linux
defines it with type __kernel_mode_t, and defines __kernel_mode_t as
unsigned short on some archs. since there is a subsequent padding
field, treating it as a 32-bit unsigned int works on little endian
archs, but the order is backwards on big endian archs with the
erroneous definition.
since multiple archs are affected, remedy the situation with fixup
code in the affected functions (shmctl, semctl, and msgctl) rather
than repeating the same shims in syscall_arch.h for every affected
arch.
three ABIs are supported: the default with 68881 80-bit fpu format and
results returned in floating point registers, softfloat-only with the
same format, and coldfire fpu with IEEE single/double only. only the
first is tested at all, and only under qemu which has fpu emulation
bugs.
basic functionality smoke tests have been performed for the most
common arch-specific breakage via libc-test and qemu user-level
emulation. some sysvipc failures remain, but are shared with other big
endian archs and will be fixed separately.
since x86 and m68k are the only archs with 80-bit long double and each
has mandatory endianness, select the variant via endianness.
differences are minor: apparently just byte order and representation
of infinities. the m68k format is not well-documented anywhere I could
find, so if other differences are found they may require additional
changes later.
In TLS variant I the TLS is above TP (or above a fixed offset from TP)
but on some targets there is a reserved gap above TP before TLS starts.
This matters for the local-exec tls access model when the offsets of
TLS variables from the TP are hard coded by the linker into the
executable, so the libc must compute these offsets the same way as the
linker. The tls offset of the main module has to be
alignup(GAP_ABOVE_TP, main_tls_align).
If there is no TLS in the main module then the gap can be ignored
since musl does not use it and the tls access models of shared
libraries are not affected.
The previous setup only worked if (tls_align & -GAP_ABOVE_TP) == 0
(i.e. TLS did not require large alignment) because the gap was
treated as a fixed offset from TP. Now the TP points at the end
of the pthread struct (which is aligned) and there is a gap above
it (which may also need alignment).
The fix required changing TP_ADJ and __pthread_self on affected
targets (aarch64, arm and sh) and in the tlsdesc asm the offset to
access the dtv changed too.
since this iconv implementation's output is stateless, it's necessary
to know before writing anything to the output buffer whether the
conversion of the current input character will fit.
previously we used a hard-coded table of the output size needed for
each supported output encoding, but failed to update the table when
adding support for conversion to jis-based encodings and again when
adding separate encoding identifiers for implicit-endianness utf-16/32
and ucs-2/4 variants, resulting in out-of-bound table reads and
incorrect size checks. no buffer overflow was possible, but the
affected characters could be converted incorrectly, and iconv could
potentially produce an incorrect return value as a result.
remove the hard-coded table, and instead perform the recursive iconv
conversion to a temporary buffer, measuring the output size and
transferring it to the actual output buffer only if the whole
converted result fits.
this case is handled with a recursive call to iconv using a
specially-constructed conversion descriptor. the constant 0 was used
as the offset for utf-8, since utf-8 appears first in the charmaps
table, but the offset used needs to point into the charmap entry, past
the name/aliases at the beginning, to the byte identifying the
encoding. as a result of this error, junk was produced.
instead, call find_charmap so we don't have to hard-code a nontrivial
offset. with this change, the code has been tested and found to work
in the case of converting the affected hkscs characters to utf-8.
maintainer's notes:
commit 95c6044e2a split UTF-32 and
UTF-32BE but neglected to add a case for the former as a destination
encoding, resulting in it wrongly being handled by the default case.
the intent was that the value of the macro be chosen to encode "big
endian" in the low bits, so that no code would be needed, but this was
botched; instead, handle it the way UCS2 is handled.
maintainer's notes:
commit a223dbd27a added the reverse
conversions to JIS-based encodings, but omitted the check for remining
buffer space in the case where the next character to be written was
single-byte, allowing conversion to continue past the end of the
destination buffer.
the wrapper start function that performs scheduling operations is
unreachable if pthread_attr_setinheritsched is never called, so move
it there rather than the pthread_create source file, saving some code
size for static-linked programs.
eliminate the awkward startlock mechanism and corresponding fields of
the pthread structure that were only used at startup.
instead of having pthread_create perform the scheduling operations and
having the new thread wait for them to be completed, start the new
thread with a wrapper start function that performs its own scheduling,
sending the result code back via a futex. this way the new thread can
use storage from the calling thread's stack rather than permanent
fields in the pthread structure.
over time the pthread structure has accumulated a lot of cruft taking
up size. this commit removes unused fields and packs booleans and
other small data more efficiently. changes which would also require
changing code are not included at this time.
non-volatile booleans are packed as unsigned char bitfield members.
the canceldisable and cancelasync fields need volatile qualification
due to how they're accessed from the cancellation signal handler and
cancellable syscalls called from signal handlers. since volatile
bitfield semantics are not clearly defined, discrete char objects are
used instead.
the pid field is completely removed; it has been unused since commit
83dc6eb087.
the tid field's type is changed to int because its use is as a value
in futexes, which are defined as plain int. it has no conceptual
relationship to pid_t. also, its position is not ABI.
startlock is reduced to a length-1 array. the second element was
presumably intended as a waiter count, but it was never used and made
no sense, since there is at most one waiter.
previously, some accesses to the detached state (from pthread_join and
pthread_getattr_np) were unsynchronized; they were harmless in
programs with well-defined behavior, but ugly. other accesses (in
pthread_exit and pthread_detach) were synchronized by a poorly named
"exitlock", with an ad-hoc trylock operation on it open-coded in
pthread_detach, whose only purpose was establishing protocol for which
thread is responsible for deallocation of detached-thread resources.
instead, use an atomic detach_state and unify it with the futex used
to wait for thread exit. this eliminates 2 members from the pthread
structure, gets rid of the hackish lock usage, and makes rigorous the
trap added in commit 80bf595255 for
catching attempts to join detached threads. it should also make
attempt to detach an already-detached thread reliably trap.
if the last thread exited via pthread_exit, the logic that marked it
dead did not account for the possibility of it targeting itself via
atexit handlers. for example, an atexit handler calling
pthread_kill(pthread_self(), SIGKILL) would return success
(previously, ESRCH) rather than causing termination via the signal.
move the release of killlock after the determination is made whether
the exiting thread is the last thread. in the case where it's not,
move the release all the way to the end of the function. this way we
can clear the tid rather than spending storage on a dedicated
dead-flag. clearing the tid is also preferable in that it hardens
against inadvertent use of the value after the thread has terminated
but before it is joined.
posix documents in the rationale and future directions for
pthread_kill that, since the lifetime of the thread id for a joinable
thread lasts until it is joined, ESRCH is not a correct error for
pthread_kill to produce when the target thread has exited but not yet
been joined, and that conforming applications cannot attempt to detect
this state. future versions of the standard may explicitly require
that ESRCH not be returned for this case.
the tid field in the pthread structure is not volatile, and really
shouldn't be, so as not to limit the compiler's ability to reorder,
merge, or split loads in code paths that may be relevant to
performance (like controlling lock ownership).
however, use of objects which are not volatile or atomic with futex
wait is inherently broken, since the compiler is free to transform a
single load into multiple loads, thereby using a different value for
the controlling expression of the loop and the value passed to the
futex syscall, leading the syscall to block instead of returning.
reportedly glibc's pthread_join was actually affected by an equivalent
issue in glibc on s390.
add a separate, dedicated join_futex object for pthread_join to use.
the static const zero set ended up getting put in bss instead of
rodata, wasting writable memory, and the call to memcmp was
size-inefficient. generally for nonstandard extension functions we try
to avoid poking at any internals directly, but the way the zero set
was setup was arguably already doing so.
to support the GNU extension of allocating a buffer for getcwd's
result when a null pointer is passed without incurring a link
dependency on free, we use a PATH_MAX-sized buffer on the stack and
only duplicate it to allocated storage after the operation succeeds.
unfortunately this imposed excessive stack usage on all callers,
including those not making use of the GNU extension.
instead, use a VLA to make stack allocation conditional.
for getopt_long, partial (prefix) matches of long options always begin
with "--" and thus can never be ambiguous with a short option. for
getopt_long_only, though, a single-character option can match both a
short option and as a prefix for a long option. in this case, we
wrongly interpreted it as a prefix for the long option.
introduce a new pass, only in long-only mode, to check the prefix
match against short options before accepting it. the only reason
there's a slightly nontrivial loop being introduced rather than strchr
is that our getopt already supports multibyte short options, and
getopt_long_long should handle them consistently. a temp buffer and
strstr could have been used, but the code to set it up would be just
as large as what's introduced here and it would unnecessarily pull in
relatively large code for strstr.
commit 618b18c78e removed the previous
detection and hardening since it was incorrect. commit
72141795d4 already handled all that
remained for hardening the static-linked case. in the dynamic-linked
case, have the dynamic linker check whether malloc was replaced and
make that information available.
with these changes, the properties documented in commit
c9f415d7ea are restored: if calloc is
not provided, it will behave as malloc+memset, and any of the
memalign-family functions not provided will fail with ENOMEM.
this change serves multiple purposes:
1. it ensures that static linking of memalign-family functions will
pull in the system malloc implementation, thereby causing link errors
if an attempt is made to link the system memalign functions with a
replacement malloc (incomplete allocator replacement).
2. it eliminates calls to free that are unpaired with allocations,
which are confusing when setting breakpoints or tracing execution.
as a bonus, making __bin_chunk external may discourage aggressive and
unnecessary inlining of it.
the generated code should be mostly unchanged, except for explicit use
of C_INUSE in place of copying the low bits from existing chunk
headers/footers.
these changes also remove mild UB due to dubious arithmetic on
pointers into imaginary size_t[] arrays.
Rich Felker
Szabolcs Nagy
A. Wilcox
Segev Finer
Fangrui Song
Alexander Monakov
Patrick Oppenlander
David Carlier
Arthur Jones
Daniel Sabogal
Will Dietz