LoongArch architecture has been supported on linux kernels
since V5.19. This patch adds support for loongarch to ltrace,
More than 90 percent of testsuite passed. Now ltrace already
works well on loongarch.

Signed-off-by: Hui Li <lihui@loongson.cn>

Fix use-after-free error in ppc/plt.c

See merge request !11

Will Schmidt authored Sep 08, 2022 and Carlos O'Donell committed Sep 08, 2022

This resolves an error reported with a newer GCC.  The libsym
pointer is referenced after it is freed.

  plt.c: In function 'arch_elf_add_func_entry':
  plt.c:702:25: error: pointer 'libsym' used after 'free'
                                 [-Werror=use-after-free]
    702 |                         delete_symbol_chain(libsym);
        |                         ^~~~~~~~~~~~~~~~~~~~~~~~~~~
  plt.c:701:25: note: call to 'free' here
    701 |                         free(libsym);
        |                         ^~~~~~~~~~~~

Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>

Supplement PowerPC LWARX instruction checks (v2)

See merge request !13

Will Schmidt authored Sep 08, 2022 and Carlos O'Donell committed Sep 08, 2022

of the lwarx/stwcx Load and Reserve instructions.
This includes
        Byte in storage - lbarx/stbcx.
        Halfword in storage - lharx/sthcx.
        Quadword in storage - lqarx/stqcx.

These changes match what has been previously
added to GDB.

Signed-off-by: Will Schmidt <will_schmidt@vnet.ibm.com>
Reviewed-by: Carlos O'Donell <carlos@redhat.com>

After the build is complete we can immediately test installation.

Add a quick MR checker that builds and installs the branch.

If the kernel supports it AND glibc supports it, the ppc64le
syscalls may use the SCV variant instead of the SC variant.
The registers we care about have the same use so little change
is required.

RHBZ #1922305

This is downstream Fedora patch: ltrace-0.7.91-ppc64le-scv.patch
The downstream patch can be removed.

Signed-off-by: DJ Delorie <dj@redhat.com>

ltrace does not yet support AARCH64 ifuncs

This is downstream Fedora patch: ltrace-rh1225568.patch
The downstream patch can be removed.

Signed-off-by: DJ Delorie <dj@redhat.com>

Adds two features:

1. Parsing CET-enabled PLTs.
2. Using a second "insecure" PLT as a fallback.

This is downstream Fedora patch: ltrace-0.7.91-cet.patch
The downstream patch can be removed.

Signed-off-by: DJ Delorie <dj@redhat.com>

Fix aarch64 long parameters (via $r8) support.
Make system_call_params test compare more exactly

This is downstream Fedora patch: ltrace-0.7.91-aarch64-params.patch
The downstream patch can be removed.

Signed-off-by: DJ Delorie <dj@redhat.com>

Move initialization earlier to avoid gcc warnings

Fix FTBFS due to invalid code in ARM support (RHBZ #1423913).

This is downstream Fedora patch: ltrace-rh1423913.patch
The downstream patch can be removed.

Signed-off-by: DJ Delorie <dj@redhat.com>

Fix FTBFS due to new gcc 6 warnings.  (RHBZ #1347879)

This is downstream Fedora patch: ltrace-0.7.91-tautology.patch
The downstream patch can be removed.

Signed-off-by: DJ Delorie <dj@redhat.com>

Fix compilation with gcc 11 and glibc 2.33.

It is possible to have a switch statement value that causes an
invalid free. So set nrhs to a value that can be distinguished if
we are in the fall-through case.

expr.c: In function ‘expr_clone’:
expr.c:240:33: error: ‘nrhs’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
  240 |                                 free(nrhs);
      |                                 ^~~~~~~~~~
cc1: all warnings being treated as errors

This is downstream Fedora patch: ltrace-0.7.91-rh1799619.patch
The downstream patch can be removed.

Signed-off-by: Carlos O'Donell <carlos@redhat.com>

Fix compiling with gcc 11 and glibc 2.33 by using a default value for
the module and symbol name if we can't look them up. We use "[unknown]"
as the default.

output.c: In function ‘frame_callback’:
output.c:660:9: error: ‘%s’ directive argument is null [-Werror=format-overflow=]
  660 |         fprintf(options.output, " > %s(%s+0x%" PRIx64 ") [%" PRIx64 "]\n",
      |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                 modname, symname, off, pc);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors

This is downstream Fedora patch: ltrace-0.7.91-null.patch
This patch can be now be removed.

Signed-off-by: Carlos O'Donell <carlos@redhat.com>

Switch to using readdir to fix the use of a deprecated function.

proc.c: In function ‘process_tasks’:
proc.c:247:17: error: ‘readdir_r’ is deprecated [-Werror=deprecated-declarations]
  247 |                 if (readdir_r(d, &entry, &result) != 0) {
      |                 ^~
In file included from proc.c:31:
/usr/include/dirent.h:183:12: note: declared here
  183 | extern int readdir_r (DIR *__restrict __dirp,
      |            ^~~~~~~~~
cc1: all warnings being treated as errors

This is downstream Fedora patch: ltrace-0.7.91-tautology.patch
without the removal of the forced asserts. The downstream patches
to fix this can be removed.

Signed-off-by: Carlos O'Donell <carlos@redhat.com>

Files etc/libpthread.so-types.conf and etc/libc.so-types.conf was introduced in
commit 5ba9e10 ("Split type definitions from the bundled configs into their own
files"). However, they was not installed.

So latrce failed with messages like :
/usr/share/ltrace/libm.so.conf:333: error: unknown type around 'ldouble erfl(ldouble);

This patch declare these configuration files.

Signed-off-by: Jérôme Pouiller <jezz@sysmic.org>

Since "import" is not a keyword in C, it might be used as a type name.
However, a function prototype with "import" as the return type would
be interpreted as an import directive. So provide a new keyword,
"function", that can be used to force a line to be interpreted as a
prototype.

Naturally, the new keyword will also work if the return type is
"function" or "typedef".

Namely:

* PTHREAD_BARRIER_SERIAL_THREAD;
* Non-POSIX enumerators and functions from Glibc.

- Patch contributed by Faraz Shahbazker <faraz.shahbazker@imgtec.com>

* use signum as the type for pthread_kill's second parameter;
* import libpthread.so.conf into libc.so.conf, since libc.so also
  defines several functions from libpthread.so.

When a non-NULL protolib is returned and own_key is true, the key must
be freed.

- This bug was brought in by my on-the-fly edits to adapt Roman's
  patchset to the latest changes.

Handy for sharing declarations between libraries.

- The problem was reported by Roman Donchenko.

And, as an important side effect, make sure that any special characters in
the ltrace command lines are not interpreted by Tcl or the shell.

LTRACE and VALGRIND are now considered to be lists throughout the test
scripts.

- There was a memory error in `if (own_lens && lens != NULL)`
  (own_lens was only set when lens was != NULL), reported by Роман
  Донченко.  But we don't need to track the ownership of lenses at
  all, and the whole function name2lens is somewhat pointless as well.
  So simplify all that code.

symbols is only used (and freed) if only_exported_names is true, so
don't allocate it if it's false.

Attaching immediately often leads to attaching before the process reaches
the first sleep call. Use a delay.