[Valgrind-announce] Valgrind-3.26.0 is available

[Mark W. <ma...@kl...>]

We are pleased to announce a new release of Valgrind, version 3.26.0,
available from https://valgrind.org/downloads/current.html

This release adds an upgrade to GPL version 3, build control for html
and/or pdf docs, added LibVEX_set_VexControl, removed Iop_Clz32/64 and
Iop_Ctz32/64, integrated LTP v20250930, 13 new Linux syscall wrappers,
new --modify-fds=yes, use log output protocol 6 with --xml=yes, new
--track-fds=bad, gdb qExecAndArgs packet support, rewrite of DWARF
inlined subroutine handling, new vgstack utility, handling of aligned
allocation with size of zero changed, checks for C23 free_sized and
free_aligned_sized.

See the release notes below for details of the changes.

Our thanks to all those who contribute to Valgrind's development.
This release represents a great deal of time, energy and effort on the
part of many people. It was a busy release, with more than 400 commits
by 12 people, fixing 90 bugs.

Happy and productive debugging and profiling,

-- The Valgrind Developers

Release 3.26.0 (24 Oct 2025)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This release supports X86/Linux, AMD64/Linux, ARM32/Linux, ARM64/Linux,
PPC32/Linux, PPC64BE/Linux, PPC64LE/Linux, S390X/Linux, MIPS32/Linux,
MIPS64/Linux, RISCV64/Linux, ARM/Android, ARM64/Android, MIPS32/Android,
X86/Android, X86/Solaris, AMD64/Solaris, AMD64/MacOSX 10.12, X86/FreeBSD,
AMD64/FreeBSD and ARM64/FreeBSD There is also preliminary support for
X86/macOS 10.13, AMD64/macOS 10.13 and nanoMIPS/Linux.

* ==================== CORE CHANGES ===================

* Upgrade to the GNU General Public License version 3.

* Control building documentation. When using make dist set the
  Makefile BUILD_DOCS to none, all or html. none, does not build any
  documentation.  all, builds all documentation. html, builds HTML
  docs but skips building PDFs. See also README_DEVELOPERS.

* New VEX API function LibVEX_set_VexControl

* The deprecated IROps: Iop_Clz32/64 and Iop_Ctz32/64 have been removed

* The Linux Test Project (LTP) integration has been updated to
  v20250930. The test output has been made compatible with bunsen.
  Various issues with the linux syscall wrappers have been fixed.

  New Linux syscall wrappers for: cachestat, futex_waitv, listmount,
  mount_setattr, mseal, quotactl_fd, remap_file_pages, setdomainname,
  statmount, swapoff, swapon, sysfs and ustat.

* --modify-fds=yes has been added. It acts like --modify-fds=high (the
  highest available file descriptor is returned first) except when
  when the lowers stdin/stdout/stderr (file descriptors 0, 1, 2) are
  available. With --modify-fds=yes 0, 1 or 2 are always returned first
  when still available before higher file descriptor numbers are.

* With --xml=yes log output protocol 6 is now always used (unlike
  protocol 5 which was only used with--track-fds). The main difference
  is that the xml output now contains error summaries. See also
  xml-output-protocol6.txt.

* Add "bad" option for --track-fds. When --track-fds=bad is specified,
  do not produce errors about unclosed file descriptors at program
  exit. Only produce errors for bad file descriptor usage, either
  double close or use of file descriptor that is (no longer) valid.

* vgdb will now handle the qExecAndArgs packet.

* DWARF inlined subroutine handling has been rewritten to work cross
  compile units. This should get rid of backtraces with
  "UnknownInlinedFun".

* ================== PLATFORM CHANGES =================

FreeBSD 15 (which is expected to ship in December 2025, after
Valgrind 3.26 is released) contains a change to ptrace that affects
use of Valgrind with vgdb. This impacts the mechanism that vgdb
uses to interrupt Valgrind if all threads are blocked and you want
to get back to the gdb prompt by hitting ctrl-c. This mechanism
is no longer reliable. On arm64 Valgrind will crash with an assert.
On amd64 syscalls may give spurious and incorrect return codes.

There is a workaround. Run the following command (as root).

 sysctl debug.ptrace_attach_transparent=0

 See also
 https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=290008

* ==================== TOOL CHANGES ===================

* There is a new utility script, "vgstack". It has two
  option, -h for minimal help, and -v for the version information.
  In normal use pass it the PID of a running Valgrind process
  and it will perform a vgdb attach and print the backtrace(s)
  of the guest executable.

* Memcheck handling of aligned allocation functions with a
  size of zero has changed.

  Firstly, 'free_aligned_sized' with a size of
  zero is no longer considered an error. This was intended so
  that deallocation had the same behaviour as allocation. In
  practice, platforms that allow aligned allocation with a
  size of zero will already generate an error at allocation.
  Other platforms will get an 'Invalid free' error. The case
  where the allocation and deallocation sizes are different
  with the deallocation size being zero is already covered by
  "Mismatched [alloc/dealloc] size" errors.

  Secondly, the three C aligned allocation functions memalign,
  aligned_alloc and posix_memalign have a different error message if
  used with a size of zero. Previously the error was "[function]
  invalid size value: [number]". This was an overstatement of the
  issue. The problem is that such usage is not portable across
  platforms. memalign and aligned_alloc are poorly documented, saying
  things like "Behavior is undefined if size is not an integral
  multiple of alignment.". Clearly this does not include negative
  integers though it does not say so explicitly. Does that include
  zero?  posix_memalign is well documented but says that using a size
  of 0 is implementation-defined. These functions now produce an error
  "Unsafe allocation with size of zero is implementation-defined".

  The associated suppression name has also changed from "BadSize" to
  "UnsafeZeroSize".

  Checks for C23 free_sized and free_aligned_sized have been added to
  Linux. Almost no libraries support these functions yet, with
  the exception being Google tcmalloc.

* ==================== FIXED BUGS ====================

The following bugs have been fixed or resolved.  Note that "n-i-bz"
stands for "not in bugzilla" -- that is, a bug that was reported to us
but never got a bugzilla entry.  We encourage you to file bugs in
bugzilla (https://bugs.kde.org/enter_bug.cgi?product=valgrind) rather
than mailing the developers (or mailing lists) directly -- bugs that
are not entered into bugzilla tend to get forgotten about or ignored.

286849  [PATCH] Interceptors for new/delete on Darwin were erroneously
        commented out in r12043
306098  s390x: Alternate opcode form for convert to/from fixed and friends
309100  s390x: Testcases for extended BFP
309554  Wrap syscall remap_file_pages (216)
331311  Valgrind shows open files in /proc/self/fd that don't work for the process
338803  Handling of dwz debug alt files or cross-CU is broken
368791  Handle swapon and swapoff syscalls as linux generic
369030  Wrap linux syscall: 171 (setdomainname)
388526  Inconsistent severity in message text: "WARNING: Serious error"
418756  MAP_FIXED_NOREPLACE mmap flag unsupported
454276  Some IPC syscalls is missing for x86 linux
476465  AArch64 ARMv8.3 LDAPR/LDAPRH/LDAPRB instructions not supported
493430  Review all syscalls that use or return (new) file descriptors
493434  Add --track-fds=bad mode (no "leak" tracking)
501741  syscall cachestat not wrapped
502359  Add --modify-fds=yes option
502968  Wrap linux specific syscalls 457 (listmount) and 458 (statmount)
503098  Incorrect NAN-boxing for float registers in RISC-V
503241  s390x: Support z17 changes to the NNPA instruction
503641  close_range syscalls started failing with 3.25.0
503677  duplicated-cond compiler warning in dis_RV64M
503817  s390x: fix 'ordered comparison of pointer with integer zero' compiler warnings
503914  mount syscall param filesystemtype may be NULL
503969  Make test results of make ltpchecks compatible with bunsen
504101  Add a "vgstack" script
504177  FILE DESCRIPTORS banner shows when closing some inherited fds
504265  FreeBSD: missing syscall wrappers for fchroot and setcred
504341  Valgrind killed by LTP syscall testcase setrlimit05
504466  Double close causes SEGV
504904  Hide "bad act handler address" warnings when -q (quiet) flag is set
504909  Hide "Bad oldset address" warnings when -q (quiet) flag is set
504919  Hide "client tried to modify addresses" warnings when -q (quiet) set
504936  Add FreeBSD amd64 sysarch subcommands AMD64_SET_TLSBASE and
        AMD64_GET_TLSBASE
505228  Wrap linux specific mseal syscall
505673  Valgrind crashes with an internal error and SIGBUS when
        the guest tries to open its own file with O_WRONLY|O_CREAT|O_TRUNC
506076  unimplemented fcntl command: 1028 (F_CREATED_QUERY)
506499  Unhandled syscall 592 (exterrctl - FreeBSD
506795  Better report which clone flags are problematic
506806  Fix execveat() with AT_FDCWD and relative path
506813  The execveat wrapper needs to do more checking
506816  futex2, futex_waitv WARNING: unhandled amd64-linux syscall: 449
506910  openat2 with RESOLVE_NO_MAGICLINKS succeeds on /proc/self/exe
506928  Wrap (deprecated) linux specific ustat syscall
506929  Wrap (deprecated) linux sysfs syscall
506930  valgrind allows SIGKILL being reset to SIG_DFL
506967  Implement and override mallinfo2
506970  mmap needs an EBADF fd_allowed check
507033  Remove deprecated Iop_Clz32/64 and Iop_Ctz32/64
507173  s390x: Crash when constant folding is disabled
507188  memcheck with track-fds=yes on x86 with popen: Assertion
507720  Review syscalls returning file descriptors (other platforms)
507721  Wire up illumos and Solaris mallinfo
507853  faccessat and faccessat2 should handle AT_FDCWD and absolute paths
507866  fanotify_mark dirfd isn't checked
507867  perf_event_open group_fd isn't checked
507868  futimesat doesn't handle AT_FDCWD
507869  Various at syscalls don't check dirfd argument
507873  Make fchmodat and fchmodat2 syscall wrappers accept AT_FDCWD
507897  Allow for patching LTP sources
507970  -Wcalloc-transposed-args warnings in valgrind-di-server.c
508027  Fix mips32 FTBFS
508029  Review the vmsplice syscall wrapper
508030  Add several missing syscall hooks to ppc64-linux
508093  VALGRIND_CLO_CHANGE does not update vex_control
508145  ppc64le needs ld.so hardwire for strcmp
508154  PRE(sys_fchownat) not handling VKI_AT_FDCWD
508638  Self-hosting not working on FreeBSD
508777  amd64-linux: add minimal scalar test
508778  syscall-wrapper waitid warns about infop=null
508779  PRE(sys_prlimit64): reorder check for memory validity
508869  x86-linux: simplify scalar test output
508958  FreeBSD: add getgroups and setgroups wrappers
509103  Fix tests/arm64/bug484935.c build with "-O2 -flto -ffat-lto-objects"
509107  memcheck/tests/duplicate_align_size_errors.cpp fails
509139  Update BadSize error messages
509258  FreeBSD: add jail_attach_jd and jail_remove_jd syscall wrappers
509406  FreeBSD 15 issues
509517  s390x: Even/odd lane confusion in various vector insns
509566  Wrap amd64-linux syscall: 442 (mount_setattr)
509572  s390x: Overhaul BFP testsuite
509590  Run the LTP tests with LTP_QUIET
509567  unhandled amd64-linux syscall: 443 (quotactl_fd)
509642  Add missing ppc64-linux syswraps
509643  Add missing s390x-linux syswraps
510169  Update the LTP version in valgrind testsuite to 20250930
510292  Silence false positive failure of LTP munmap01
510436  Don't warn about fcntl F_GETFD with --track-fds
510694  Handle qExecAndArgs remote protocol packet

To see details of a given bug, visit
  https://bugs.kde.org/show_bug.cgi?id=XXXXXX
where XXXXXX is the bug number as listed above.

(3.26.0.RC1: 17 Oct 2025)