diff options
Diffstat (limited to 'WHATSNEW')
| -rw-r--r-- | WHATSNEW | 454 |
1 files changed, 454 insertions, 0 deletions
@@ -1984,3 +1984,457 @@ arch-specfic bugs fixed: - on mips, return from start function passed to clone crashed (runaway exec) - printf %a precision specifier malfunctioned except on ld80 archs - async thread cancellation crashed on powerpc64 and sh-fdpic + + +1.1.21 release notes + +new features: +- setting default thread stack size via PT_GNU_STACK program header +- arm vfork implementation +- arm tlsdesc/gnu2 tls dialect support +- name_to_handle_at and name_to_handle_at syscall wrappers +- header-level support for new linux features through 4.18 + +optimizations: +- glob rewrite with much better performance and stack usage properties +- single-threaded and already-locked fast paths for getc/putc variants +- single-instruction fma implementations for arm, s390x, powerpc, & x86_64 +- single-instruction fabs and sqrt implementations for powerpc +- size and performance from making all internal-only functions/data hidden +- made &errno and pthread_self results cachable again (attribute((const))) +- significant speedup in strtod with short inputs +- new tsearch AVL tree implementation, smaller and faster +- special-cased nop calls to wmemmove +- fixed erroneously suboptimal skip conditions in strstr and memmem + +hardening: +- default thread stack guard size increased from 4k to 8k + +compatibility: +- default thread stack size increased from 80k to 128k +- building for arm as thumb2 with clang internal assembler now works +- aio threads could overflow stack on kernels that break MINSIGSTKSZ ABI +- aio threads no longer call malloc (problematic with malloc replacement) +- pthread_sigmask/sigprocmask now ignore an invalid how when not changing mask + +bugs fixed: +- soft deadlock regression in stdio FILE locks with >2 threads contending +- deadlock and buffered data loss race in fclose +- race condition leading to possible crash in dcngettext plural forms +- glob failed to see past searchable-but-unreadable path components +- getdelim wrongly realloc'd buffer that was already exactly right size +- getdelim failed to set stream orientation on early error +- ttyname[_r] reported wrong error when given bad fd +- pthread_key_delete left old tsd values exposed if slot was reused +- freeaddrinfo failed to support freeing sublists +- access to optopt was broken by copy relocations +- memccpy returned wrong result if first byte past buffer end matched +- wordexp read past end of input string ending in backslash +- sem_wait and sem_timedwait were wrongly not interruptible by signals +- getspnam[_r] wrongly treated not-found as an error + +arch-specfic bugs fixed: +- soft deadlocks (missing futex wake) on powerpc locking +- dlsym returned wrong address for thread-local symbols on ppc/mips/m68k + + +1.1.22 release notes + +new features: +- priority-inheritance mutexes +- membarrier syscall, pre-registration to use it, fallback emulation +- header-level support for new linux features in 4.19, 4.20, 5.0 + +major internal changes: +- complete, async-safe view of all existent threads as global list +- robust __synccall based on new thread list +- new dynamic TLS is installed synchronously at dlopen +- TLSDESC resolver functions no longer make bad ABI assumptions to call C +- resolved shared library dependencies are now recorded + +compatibility & conformance: +- dependency-order shared library constructor execution +- sigaltstack no longer rejects SS_AUTODISARM, future flags +- FILE is now a complete (dummy) type in pre-C11 feature profiles +- setvbuf reports failure on invalid arguments +- TSVTX is exposed unconditionally in tar.h +- multithreaded set*id() no longer depends on /proc +- key slot reuse after pthread_key_delete no longer depends on /proc + +bugs fixed: +- failures in multithreaded set*id() with concurrent thread creation/exit +- interposed free was called from invalid/inconsistent contexts +- freeaddrinfo performed invalid free of some partial results lists +- dlsym dependency order search had false negatives and false positives +- dn_skipname gave wrong results for labels with 8-bit content +- dcngettext clobbered errno, often breaking printing of error messages +- sscanf read past end of buffer under certain conditions (1.1.21 regression) +- pthread_key_create spuriously failed under race condition (1.1.21 regression) +- fdopendir wrongly succeeded with O_PATH file descriptors +- gets behaved incorrectly in presence of null bytes +- namespace violations in c11 tsd and mutex function dependencies +- incorrect prototype for makecontext (unimplemented) + +arch-specfic bugs fixed: +- s390x had wrong values for POSIX_FADV_DONTNEED/_NOREUSE + + + +1.1.23 release notes + +new features: +- riscv64 port +- configure now allows customizing AR and RANLIB vars +- header-level support for new linux features in 5.1 + +major internal changes: +- removed extern __syscall; syscall header code is now fully self-contained + +performance: +- new math library implementation for log/exp/pow +- aarch64 dynamic tlsdesc function is streamlined + +compatibility & conformance: +- O_TTY_INIT is now defined +- sys/types.h no longer pollutes namespace with sys/sysmacros.h in any profile +- powerpc asm is now compatible with clang internal assembler + +changes for new POSIX interpretations: +- fgetwc now sets stream error indicator on encoding errors +- fmemopen no longer rejects 0 size + +bugs fixed: +- static TLS for shared libraries was allocated wrong on "Variant I" archs +- crash in dladdr reading through uninitialized pointer on non-match +- sigaltstack wrongly errored out on invalid ss_size when doing SS_DISABLE +- getdents function misbehaved with buffer length larger than INT_MAX +- set*id could deadlock after fork from multithreaded process + +arch-specfic bugs fixed: +- s390x SO_PEERSEC definition was wrong +- passing of 64-bit syscall arguments was broken on microblaze +- posix_fadvise was broken on mips due to missing 7-arg syscall support +- vrregset_t layout and member naming was wrong on powerpc64 + + + +1.1.24 release notes + +new features: +- GLOB_TILDE extension to glob +- non-stub catgets localization API, using netbsd binary catalog format +- posix_spawn file actions for [f]chdir (extension, pending future standard) +- secure_getenv function (extension) +- copy_file_range syscall wrapper (Linux extension) +- header-level support for new linux features in 5.2 + +performance: +- new fast path for lrint (generic C version) on 32-bit archs + +major internal changes: +- functions involving time are overhauled to be time64-ready in 32-bit archs +- x32 uses the new time64 code paths to replace nasty hacks in syscall glue + +compatibility & conformance: +- support for powerpc[64] unaligned relocation types +- powerpc[64] and sh sys/user.h no longer clash with kernel asm/ptrace.h +- select no longer modifies timeout on failure (or at all) +- mips64 stat results are no longer limited to 32-bit time range +- optreset (BSD extension) now has a public declaration +- support for clang inconsistencies in wchar_t type vs some 32-bit archs +- mips r6 syscall asm no longer has invalid lo/hi register clobbers +- vestigial asm declarations of __tls_get_new are removed (broke some tooling) +- riscv64 mcontext_t mismatch glibc's member naming is corrected + +bugs fixed: +- glob failed to match broken symlinks consistently +- invalid use of interposed calloc to allocate initial TLS +- various dlsym symbol resolution logic errors +- semctl with SEM_STAT_ANY didn't work +- pthread_create with explicit scheduling was subject to priority inversion +- pthread_create failure path had data race for thread count +- timer_create with SIGEV_THREAD notification had data race getting timer id +- wide printf family failed to support l modifier for float formats + +arch-specific bugs fixed: +- x87 floating point stack imbalance in math asm (i386-only CVE-2019-14697) +- x32 clock_adjtime, getrusage, wait3, wait4 produced junk (struct mismatches) +- lseek broken on x32 and mipsn32 with large file offsets +- riscv64 atomics weren't compiler barriers +- riscv64 atomics had broken asm constraints (missing earlyclobber flag) +- arm clone() was broken when compiled as thumb if start function returned +- mipsr6 setjmp/longjmp did not preserve fpu register state correctly + + + +1.2.0 release notes + +new features: +- time_t is now 64-bit on all archs (not just 64-bit archs) +- character type & case mapping data updated to Unicode 12.1.0 +- header-level support for new linux features in 5.3 and 5.4 + +performance: +- new O(1) wchar_t case mapping implementation +- i386 now uses C math code for exp, faster than old asm +- mips math asm + +compatibility & conformance: +- endian.h now aims to conform to future POSIX definition +- support older compilers that don't accept powerpc math asm constraints +- fdpic code in ldso was incompatible with valid optimizations in gcc 9+ +- RLIMIT_RTTIME was missing from sys/resource.h + +bugs fixed: +- wcwidth wrongly returned 0 for most of planes 4 and up +- missing case mapping between U+03F3 and U+037F +- wrong cacosh results for arguments with negative imaginary part +- wrong catanf/catanl results for various classes of arguments +- wrong return value for ungetc with argument outside [0,UCHAR_MAX] +- posix_openpt with no ptys available produced wrong errno + +arch-specific bugs fixed: +- sigcontext/regset definition mistakes & omissions on m68k, powerpc64 +- fesetenv(FE_DFL_ENV) crashed on riscv64 +- sh2 dynamic linker was broken since 1.1.21 (crash in stage 2b) +- arm dynamic linker chose wrong tls/atomic variants since 1.1.21 +- some math library functions returned excess precision on i386 +- unconfirmed regression in fchmodat AT_SYMLINK_NOFOLLOW on mips* + + + +1.2.1 release notes + +major changes: +- new malloc implementation (mallocng & overhauled bump allocator) + +new features: +- DNS queries via res_* now set AD flag, report zone signedness (DNSSEC) +- PTHREAD_NULL macro (POSIX-future) + +performance: +- optimized memcpy and memset for aarch64 +- optimized memcpy for arm now supports big endian +- optimized x86_64 remquol +- improved strerror without linear search + +bugs fixed: +- lock-skipping for processes that returned to single-threaded was wrong +- AF_UNSPEC dns lookups mishandled single failure in paired A+AAAA +- res_send and res_query returned wrong value on errors from nameserver +- corrupted sysvipc timestamps on 32-bit archs with old kernels +- incorrect parsing of timezone offsets after overly-long zone name +- clock_adjtime was broken on 32-bit archs (time64) +- pthread_kill as not async-signal-safe +- pthread_cancel was not async-cancel-safe +- large-ulp errors in various math functions in non-default rounding modes + +arch-specific bugs fixed: +- arm clock_gettime was broken on some hw due to bad time64 vdso +- m68k sqrtl lacked long double precision +- mips* syscall mechanism regressions on older kernels +- mips* had negated error codes for some syscalls (kernel bug) +- mips* SIGEMT was wrongly called SIGSTKFLT +- sh fesetround didn't work correctly on sh + + + +1.2.2 release notes + +major changes: +- child restrictions lifted after fork of multithreaded parent + +new features: +- _Fork function (POSIX-future) +- reallocarray function (extension from OpenBSD, now widespread) +- gettid function (kernel tid as supported concept) +- SIGEV_THREAD_ID sigevent API (Linux extension) +- tcgetwinsize and tcsetwinsize functions (POSIX-future) + +performance: +- faster software sqrt on archs without native sqrt instruction + +compatibility: +- realpath no longer depends on procfs availability & accuracy +- time zone parser now always prefers 64-bit tables if present +- crypt_blowfish now supports $2b$ prefix +- res_query now reports errors via h_errno +- set*id and setrlimit are now safe in vforked/cloned child +- setgroups now applies to all threads +- dlopen debugger notification is improved, should work with lldb +- setrlimit no longer needs __synccall broadcast on linux 2.6.36+ +- faccessat with AT_EACCESS no longer needs child process on linux 5.8+ + +bugs fixed: +- buffer overflow and infinite loop errors in wcsnrtombs (CVE-2020-28928) +- sem_close unmapped still-referenced semaphores +- fork of process with active aio could deadlock or crash paren +- pthread_cond_wait was broken with priority-inheritance mutex +- getgrouplist wrongly failed when nscd reported an empty list +- abort could leak modified SIGABRT disposition to fork or posix_spawn child +- regression with mallocng: malloc_usable_size(0) crashed +- readlink wrongly gave EINVAL on zero length dest buffer +- sqrtl was severely inaccurate (not correctly rounded) on ldquad archs +- assert failure wrongly flushed stdio (possible deadlock) +- MUSL_LOCPATH search was broken with multiple components +- missing newline in herror output +- possible deadlock in pthread_exit with pshared mutex or barrier usage +- pthread_mutexattr_getprotocol didn't read back protocol +- v4l2 ioctl translation for pre-time64 kernels didn't work + +arch-specific bugs fixed: +- x86_64 longjmp failed to handle 0 argument reliably +- i386 __set_thread_area fallback for pre-2.6 kernels didn't work +- missing O_LARGEFILE macro value on x86_64, x32, mips64 +- unpredictable s390x breakage from failure to preserve call-saved registers + + + +1.2.3 release notes + +new features: +- qsort_r function (POSIX-future) +- pthread_getname_np extension function +- hard float on SPE FPU for powerpc-sf +- SEEK_DATA and SEEK_HOLE exposed in unistd.h (Linux extensions) + +compatibility: +- free now preserves errno (POSIX-future requirement) +- setjmp is declared explicitly with returns_twice for non-GCC compilers +- macro version of isascii is no longer defined for C++ +- dynamic linker now tolerates zero-length LOAD segments +- epoll_[p]wait is now a cancellation point +- pwd/grp functions no longer fail on systems without AF_UNIX support +- POSIX TZ parsing is stricter to allow more names to fallback to files +- NULL is now defined as nullptr when used in C++11 or later +- gettext now accepts null pointer as argument + +bugs fixed: +- old regression in wcwidth of Hangul combining (vowel/final) letters +- duplocale used wrong malloc when malloc was replaced (1.2.2 regression) +- fmaf rounded wrong on archs without FE_TOWARDZERO (all softfloat archs) +- popen didn't honor requirement not to leak other popen pipe fds to child +- aligned_alloc and variants crashed on allocation failure +- dl_iterate_phdr reported incorrect module TLS pointers +- mishandling of some inputs in acoshf and expm1f and functions using them +- potentially wrong-sign zero in cproj functions at infinity +- multiple bugs in legacy function cuserid +- minor posix_spawn file actions API conformance issues +- pthread_setname_np fd leak +- out-of-bound read in zoneinfo handling with distant-past times +- out-of-tree builds lacked generated debug cfi for x86 asm + +arch-specific bugs fixed: +- powerpc (32-bit) struct shmid_ds layout was wrong for some fields +- time64 struct layout was wrong in sound ioctl fallback (32-bit archs) + + + +1.2.4 release notes + +new features: +- large dns record lookups via tcp fallback +- new getaddrinfo EAI_NODATA result to distinguish NODATA/NxDomain +- support for new RELR compressed format for relative relocations +- sysconf keys for querying signal stack size requirements +- real vfork on riscv64 + +performance: +- mallocng no longer uses MADV_FREE (high performance cost, little gain) +- vdso clock_gettime is supported once again on 32-bit arm + +compatibility: +- gethostbyname family now distinguishes NO_DATA from HOST_NOT_FOUND +- res_send now works with caller-provided edns0 queries +- arpa/nameser.h RR types list is now up-to-date +- previously-missing POSIX confstr keys have been added +- mntent interfaces now accept missing fields +- alt signal stack, if any, is now used for internal signals +- the LFS64 macros are no longer exposed without _LARGEFILE64_SOURCE +- memmem (POSIX-future) is now exposed in default feature profile +- pthread_atfork now admits calls from an application-provided malloc +- debugger tracking of shared libraries now works on MIPS PIE binaries +- sendmsg now supports up to SCM_MAX_FD fds in SCM_RIGHTS messages + +bugs fixed: +- gethostbyname[2]_r wrongly returned nonzero (error) on negative result +- parallel v4/v6 address queries could fail on query id collisions +- spurious getaddrinfo/AI_ADDRCONFIG failures due to errno clobbering +- dns search domains ending in dot (including lone dot) broke lookups +- ipv6 servers in resolv.conf broke lookups on systems with v6 disabled +- systems with bindv6only failed to query both v4 and v6 nameservers +- res_mkquery mishandled consecutive final dots in name +- res_send could malfunction for very small answer buffer sizes +- resolver dns backend accepted answers with wrong (A vs AAAA) RR type +- getservbyport_r returned junk or ENOENT (vs ERANGE) on buffer size errors +- dns result parsing of malformed responses could process uninitialized data +- freopen didn't reset stream orientation (byte/wide) & encoding rule +- fwprintf didn't print most fields on open_wmemstream FILEs +- wide printf %lc ignored field width +- wide printf erroneously processed %n after encoding errors +- use of wide printf %9$ argument slot overflowed undersized buffer +- swprintf malfunctioned on nul character in output +- strverscmp ordered digit sequences vs nondigits incorrectly +- timer_create/SIGEV_THREAD failure leaked the thread +- semaphores were subject to missed-wake under certain usage patterns +- several possible rare deadlocks with lock handling at thread exit +- several possible rare deadlocks with aio and multithreaded fork +- dynamic linker relro processing was broken on archs w/variable pagesize +- async cancellation could run cancellation handlers in invalid context +- pthread_detach was wrongly a cancellation point in rare race code path +- use-after-close/double-close errors in mq_notify error paths +- mq_notify event thread wrongly ran with signals unmasked +- wcs{,n}cmp, wmemcmp returned wrong results when difference overflowed +- accept4, pipe2, and dup3 handled unknown flags wrong in fallback cases +- CPU_SETSIZE macro had wrong unit +- select fallback for pre-time64 kernels truncated timeout (vs clamping) + +arch-specific bugs fixed: +- x32 new socketcalls took fallback path due to pointer sign extension +- x32 wait4 didn't fill rusage structure (time64 regression) +- x32 semtimedop mismatched timespec ABI with kernel (time64 regression) +- sigaction signal mask was bogus on or1k, microblaze, mips, and riscv +- powerpc-sf longjmp asm clobbered value argument +- or1k poll function passed timeout to syscall in wrong form + + + +1.2.5 release notes + +new features: +- statx function (linux extension; via syscall and fallback using fstatat) +- clone function is now usable and gives _Fork-like consistency in child +- statvfs now provides f_type result +- preadv2 and pwritev2 (linux extension) syscall wrappers +- riscv64 TLSDESC support + +new ports: +- loongarch64 +- riscv32 + +compatibility: +- DNS resolver can now handle answers with long CNAME chains +- string.h no longer provides (C23-incompat) non-prototype decl of basename +- fstatat statx backend now matches stat syscall non-automounting behavior +- mntent interfaces now handle escaped whitespace in paths/options + +standards updates: +- printf %lc of nul wchar now produces output +- snprintf and swprintf no longer fail on n > INT_MAX +- ppoll is now exposed in default feature profile + +bugs fixed: +- some long DNS answers were wrongly rejected despite new TCP support +- glob could wrongly return GLOB_NOMATCH if aborted before any matches +- multithreaded set*id could malfunction from thread sequencing logic bug +- certain use of threads after fork could deadlock thread-list lock +- posix_spawn child could deadlock in race with async parent death +- mbrtowc return value was wrong if argument n exceeded UINT_MAX +- 80-bit extended acoshl and powl got some corner cases wrong +- syslog incorrectly generated localized timestamps + +arch-specific bugs fixed: +- arm (32-bit) TLSDESC malfunctioned due to addends being processed wrong +- riscv64 icache flush operation was non-functional +- sh sigsetjmp failed to properly restore call-saved register r8 on return +- sh dlsym RTLD_NEXT did not identify calling module correctly |