this bug was introduced in the recent resolver overhaul commits. it
likely had visible symptoms. these were probably limited to wrongly
accepting truncated versions of over-long names (vs rejecting them),
as opposed to stack-based overflows or anything more severe, but no
extensive checks were made. there have been no releases where this bug
was present.

now that host and service lookup have been separated in the backend,
there's no need for service lookup functions to pull in the host
lookup code. moreover, dynamic allocation is no longer needed, so this
function should now be async-signal-safe. it's also significantly
smaller.

one change in getservbyname is also made: knowing that getservbyname_r
needs only two character pointers in the caller-provided buffer, some
wasted bss can be avoided.

these changes reduce the size of the function somewhat and remove many
of its dependencies, including free. in principle it should now be
async-signal-safe, but this has not been verified in detail.

minor changes to error handling are also made.

this is the first phase of the "resolver overhaul" project.

conceptually, the results of getaddrinfo are a direct product of a
list of address results and a list of service results. the new code
makes this explicit by computing these lists separately and combining
the results. this adds support for services that have both tcp and udp
versions, where the caller has not specified which it wants, and
eliminates a number of duplicate code paths which were all producing
the final output addrinfo structures, but in subtly different ways,
making it difficult to implement any of the features which were
missing.

in addition to the above benefits, the refactoring allows for legacy
functions like gethostbyname to be implemented without using the
getaddrinfo function itself. such changes to the legacy functions have
not yet been made, however.

further improvements include matching of service alias names from
/etc/services (previously only the primary name was supported),
returning multiple results from /etc/hosts (previously only the first
matching line was honored), and support for the AI_V4MAPPED and AI_ALL
flags.

features which remain unimplemented are IDN translations (encoding
non-ASCII hostnames for DNS lookup) and the AI_ADDRCONFIG flag.

at this point, the DNS-based name resolving code is still based on the
old interfaces in __dns.c, albeit somewhat simpler in its use of them.
there may be some dead code which could already be removed, but
changes to this layer will be a later phase of the resolver overhaul.

CONCAT(0x1p,LDBL_MANT_DIG) is not safe outside of libc,
use 2/LDBL_EPSILON instead.

fix was proposed by Morten Welinder.

On 32 bit mips the kernel uses -1UL/2 to mark RLIM_INFINITY (and
this is the definition in the userspace api), but since it is in
the middle of the valid range of limits and limits are often
compared with relational operators, various kernel side logic is
broken if larger than -1UL/2 limits are used. So we truncate the
limits to -1UL/2 in get/setrlimit and prlimit.

Even if the kernel side logic consistently treated -1UL/2 as greater
than any other limit value, there wouldn't be any clean workaround
that allowed using large limits:
* using -1UL/2 as RLIM_INFINITY in userspace would mean different
infinity value for get/setrlimt and prlimit (where infinity is always
-1ULL) and userspace logic could break easily (just like the kernel
is broken now) and more special case code would be needed for mips.
* translating -1UL/2 kernel side value to -1ULL in userspace would
mean that -1UL/2 limit cannot be set (eg. -1UL/2+1 had to be passed
to the kernel instead).

using the existence of SYS_stat64 as the condition for remapping other
related syscalls is no longer valid, since new archs that omit the old
syscalls will not have SYS_stat or SYS_stat64, but still potentially
need SYS_fstat and others remapped. it would probably be possible to
get by with just one or two extra conditionals, but just breaking them
all down into separate conditions is robust and not significantly
heavier for the preprocessor.

somehow the remapping of this syscall to the 64-bit version was
overlooked. the issue was found, and patch provided, by Stefan
Kristiansson. presumably the reason this bug was not caught earlier is
that the syscall takes a pointer to off_t rather than a value, so on
little-endian systems, everything appears to work as long as the
offset value fits in the low 31 bits. on big-endian systems, though,
sendfile was presumably completely non-functional.

such archs are expected to omit definitions of the SYS_* macros for
syscalls their kernels lack from arch/$ARCH/bits/syscall.h. the
preprocessor is then able to select the an appropriate implementation
for affected functions. two basic strategies are used on a
case-by-case basis:

where the old syscalls correspond to deprecated library-level
functions, the deprecated functions have been converted to wrappers
for the modern function, and the modern function has fallback code
(omitted at the preprocessor level on new archs) to make use of the
old syscalls if the new syscall fails with ENOSYS. this also improves
functionality on older kernels and eliminates the incentive to program
with deprecated library-level functions for the sake of compatibility
with older kernels.

in other situations where the old syscalls correspond to library-level
functions which are not deprecated but merely lack some new features,
such as the *at functions, the old syscalls are still used on archs
which support them. this may change at some point in the future if or
when fallback code is added to the new functions to make them usable
(possibly with reduced functionality) on old kernels.

calling exit more than once invokes undefined behavior. in some cases
it's desirable to detect undefined behavior and diagnose it via a
predictable crash, but the code here was silently covering up an
uncommon case (exit from more than one thread) and turning a much more
common case (recursive calls to exit) into a permanent hang.

these all now use the shared __randname function internally, rather
than duplicating logic for producing a random name. incorrect usage of
the access syscall (which works with real uid/gid, not effective) has
been removed, along with unnecessary heavy dependencies like snprintf.

this was messed up during a recent commit when the socketcall macros
were moved to the common internal/syscall.h, and the following commit
expanded the problem by adding more new content outside the guard.

this only matters on x32 (and perhaps future 32-on-64 abis for other
archs); otherwise the type is long anyway. the cast through uintptr_t
prevents nonsensical "sign extension" of pointers, and follows the
principle that uintptr_t is the canonical integer type to which
pointer conversion is safe.

open is handled specially because it is used from so many places, in
so many variants (2 or 3 arguments, setting errno or not, and
cancellable or not). trying to do it as a function would not only
increase bloat, but would also risk subtle breakage.

this is the first step towards supporting "new" archs where linux
lacks "old" syscalls.

%C, %U, %W, and %y handling were completely missing; %C wrongly
fell-through to unrelated cases, and the rest returned failure. for
now, they all parse numbers in the proper forms and range-check the
values, but they do not store the value anywhere.

it's not clear to me whether, as "derived" fields, %U and %W should
produce any result. they certainly cannot produce a result unless the
year and weekday are also converted, but in this case it might be
desirable for them to do so. clarification is needed on the intended
behavior of strptime in cases like this.

%C and %y have well-defined behavior as long as they are used together
(and %y is defined by itself but may change in the future).
implementing them (including their correct interaction) is left as a
later change to be made.

finally, strptime now rejects unknown/invalid format characters
instead of ignoring them.

some of these may have been from ancient (pre-SUSv2) POSIX versions;
more likely, they were from POSIX drafts or glibc interpretations of
what ancient versions of POSIX should have added (instead they made
they described functionality mandatory and/or dropped it completely).
others are purely glibc-isms, many of them ill-thought-out, like
providing ways to lookup the min/max values of types at runtime
(despite the impossibility of them changing at runtime and the
impossibility of representing ULONG_MAX in a return value of type
long).

since our sysconf implementation does not support or return meaningful
values for any of these, it's harmful to have the macros around;
applications' build scripts may detect and attempt to use them, only
to get -1/EINVAL as a result.

if removing them does break some applications, and it's determined
that the usage was reasonable, some of these could be added back on an
as-needed basis, but they should return actual meaningful values, not
junk like they were returning before.

based on patch by Timo Teräs. previously, the value zero was used as a
literal zero, meaning that all invalid sysconf "names", which should
result in sysconf returning -1, had to be explicitly listed. (in
addition, it was not possible for sysconf to set errno to EINVAL, as
there was no distinction between -1 as an error and -1 as a valid
result.)

now, the value 0 is used for invalid/undefined slots in the table and
a new switch table entry is used for returning literal zeros.

in addition, an off-by-one error in checking against the table size is
fixed.

perhaps some additional legacy DOS-era codepages would also be useful
to have, but these are the ones for which there has been demand. the
size of the diff is due to the fact that legacychars.h is updated in
such a way that new characters are inserted into the table in unicode
codepoint order; thus other mappings in codepages.h have changed to
reflect the new table indices of their characters.

this is no longer used for anything, and reportedly clashed with a
builtin on certain compilers.

previously, setting TZ to the pathname of a file which was not a valid
zoneinfo file would usually cause programs using local time zone based
operations to crash. the new code checks the file size and magic at
the beginning of the file, which seems sufficient to prevent
accidental misconfiguration from causing crashes. attempting to make
fully-robust validation would be futile unless we wanted to drop use
of mmap (shared zoneinfo) and instead read it into a local buffer,
since such validation would be subject to race conditions with
modification of the file.

being static allows it to be inlined in __libc_start_main; inlining
should take place at all levels since the function is called exactly
once. this further reduces mandatory startup code size for static
binaries.

there is no reason (and seemingly there never was any) for
__init_security to be its own function. it's linked unconditionally
so it can just be placed inline in __init_libc.

moving the call to __init_ssp from __init_security to __init_libc
makes __init_security a leaf function, which allows the compiler to
make it smaller. __init_libc is already non-leaf, and the additional
call makes no difference to the amount of register spillage.

in addition, it really made no sense for the call to __init_ssp to be
buried inside __init_security rather than parallel with other init
functions.

since the form TZ=name is reserved for POSIX-form time zone strings,
TZ=:name needs to be used when the zoneinfo filename is in the
top-level zoneinfo directory and therefore does not contain a slash.
previously the leading colon was merely dropped, making it impossible
to access such zones without a full absolute pathname.

changes based on patch by Timo Teräs.

in cases where the memorized match range from the right factor
exceeded the length of the left factor, it was wrongly treated as a
mismatch rather than a match.

issue reported by Yves Bastide.

so far the options are --library-path and --preload which override the
corresponding environment variables, and --list which forces the
behavior of ldd even if the invocation name is not ldd. both the
two-arg form and the one-arg form using an equals sign are supported.

based loosely on a patch proposed by Rune.

the vdso symbol lookup code is based on the original 2011 patch by
Nicholas J. Kain, with some streamlining, pointer arithmetic fixes,
and one symbol version matching fix.

on the consumer side (clock_gettime), per-arch macros for the
particular symbol name and version to lookup are added in
syscall_arch.h, and no vdso code is pulled in on archs which do not
define these macros. at this time, vdso is enabled only on x86_64.

the vdso support at the dynamic linker level is no longer useful to
libc, but is left in place for the sake of debuggers (which may need
the vdso in the link map to find its functions) and possibly use with
dlsym.

at the end of successful pthread_once, there was a race window during
which another thread calling pthread_once would momentarily change the
state back from 2 (finished) to 1 (in-progress). in this case, the
status was immediately changed back, but with no wake call, meaning
that waiters which arrived during this short window could block
forever. there are two possible fixes. one would be adding the wake to
the code path where it was missing. but it's better just to avoid
reverting the status at all, by using compare-and-swap instead of
swap.

it will be needed to implement some things in sysconf, and the syscall
can't easily be used directly because the x32 syscall uses the wrong
structure layout. the l (uncreative, for "linux") prefix is used since
the symbol name __sysinfo is already taken for AT_SYSINFO from the aux
vector.

the way the x32 override of this function works is also changed to be
simpler and avoid the useless jump instruction.

the syscall is deprecated (replaced by prlimit64) and does not work
correctly on x32. this change mildly increases size, but is likely
needed anyway for newer archs that might omit deprecated syscalls.

the previous handling of cases that could not fit in the 16-bit table
or which required non-constant results was extremely ugly and could
not scale. the new code remaps these keys into a contiguous range
that's efficient for a switch statement.