[heitbaum]

• Fixes Compile error with gcc-14.1 and 32-bit / 64-bit targets #87

./vclog/vclog.c:211:34: error: argument 2 value '4294967288' exceeds maximum object size 2147483647 [-Werror=alloc-size-larger-than=]
211 | payload_buffer = realloc(payload_buffer, payload_buffer_size);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from ./vclog/vclog.c:17:
../../toolchain/armv7ve-libreelec-linux-gnueabihf/sysroot/usr/include/stdlib.h:683:14: note: in a call to allocation function 'realloc' declared here
683 | extern void *realloc (void *__ptr, size_t __size)
| ^~~~~~~

[pelwell]

Does this follow-up patch (which should be in heitbaum:patch-2 now) also work? It looks more correct to me.

[heitbaum]

Does this follow-up patch (which should be in heitbaum:patch-2 now) also work? It looks more correct to me.

I didn’t like my version of 88 either (which is why I separated it from 87) but the size_t doesn’t work either. It matches 4GB too.

FAILED: vclog/CMakeFiles/vclog.dir/vclog.c.o 
/var/media/DATA/home-rudi/LibreELEC.kernel11/build.LibreELEC-RPi2.arm-12.0-devel/toolchain/bin/armv7ve-libreelec-linux-gnueabihf-gcc   -march=armv7ve -mtune=cortex-a7 -mabi=aapcs-linux -Wno-psabi -Wa,-mno-warn-deprecated -mfloat-abi=hard -mfpu=neon-vfpv4 -Wall -pipe  -O2 -fomit-frame-pointer -DNDEBUG -Wall -Wextra -Werror -pedantic -DDUMMYOPT -MD -MT vclog/CMakeFiles/vclog.dir/vclog.c.o -MF vclog/CMakeFiles/vclog.dir/vclog.c.o.d -o vclog/CMakeFiles/vclog.dir/vclog.c.o -c /var/media/DATA/home-rudi/LibreELEC.kernel11/build.LibreELEC-RPi2.arm-12.0-devel/build/bcm2835-utils-6dc6f5f3d129a6c9423316ac1a53efb19a5c40d1/vclog/vclog.c
/var/media/DATA/home-rudi/LibreELEC.kernel11/build.LibreELEC-RPi2.arm-12.0-devel/build/bcm2835-utils-6dc6f5f3d129a6c9423316ac1a53efb19a5c40d1/vclog/vclog.c: In function 'main':
/var/media/DATA/home-rudi/LibreELEC.kernel11/build.LibreELEC-RPi2.arm-12.0-devel/build/bcm2835-utils-6dc6f5f3d129a6c9423316ac1a53efb19a5c40d1/vclog/vclog.c:211:34: error: argument 2 value '4294967288' exceeds maximum object size 2147483647 [-Werror=alloc-size-larger-than=]
  211 |                 payload_buffer = realloc(payload_buffer, payload_buffer_size);
      |                                  ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /var/media/DATA/home-rudi/LibreELEC.kernel11/build.LibreELEC-RPi2.arm-12.0-devel/build/bcm2835-utils-6dc6f5f3d129a6c9423316ac1a53efb19a5c40d1/vclog/vclog.c:17:
/var/media/DATA/home-rudi/LibreELEC.kernel11/build.LibreELEC-RPi2.arm-12.0-devel/toolchain/armv7ve-libreelec-linux-gnueabihf/sysroot/usr/include/stdlib.h:683:14: note: in a call to allocation function 'realloc' declared here
  683 | extern void *realloc (void *__ptr, size_t __size)
      |              ^~~~~~~
cc1: all warnings being treated as errors

[stintel]

Hitting this problem in OpenWrt master while building for Raspberry Pi Zero W. @heitbaum's fix (dc0fd47) works. Adding @pelwell's follow-up patch (94977ce) breaks things again.

[popcornmix]

I've pushed 5c5e2fd, which replaces both earlier attempts.

Cause: if log_copy_wrap() returns early (offset >= log_size), nothing is copied, msg.size stays 0 (from the msg.size = 0 added in b9c6321) and payload_len = 0 - sizeof(msg_hdr_t) = 0xfffffff8. The existing payload_len > log_size check can't rule that out at compile time, since log_size is only known at run time, so gcc sees a path where realloc() is passed 4294967288.

First version (MIN(INT32_MAX, payload_buffer_size)): silences the warning, but payload_buffer_size can then be larger than the buffer actually allocated, so a later copy of payload_len bytes could overflow it. It hides the problem rather than fixing it.

Second version (size_t payload_buffer_size): on 32-bit targets size_t is still 32 bits, so the value is unchanged and the error remains (as reported above).

This version: treats msg.size < sizeof(msg_hdr_t) as a corrupt entry. That removes the wrapped value and, as msg.size is a uint16_t, limits payload_len to under 64 KiB. The net change of the PR is now:

-            if (payload_len > log_size)
+            if (msg.size < sizeof(msg_hdr_t) || payload_len > log_size)

Testing: the error only shows with 32-bit ARM gcc 14/15 at -O3. It doesn't appear at -O2/-Os, on arm64 (gcc 14.2) or on x86 (-m32/-m64, gcc 14.2). Using ARM gcc 14.1 and 15.2 on godbolt with -march=armv7ve -mfloat-abi=hard -mfpu=neon-vfpv4 -Wall -Wextra -Werror -pedantic:

Version	-O3
current master	error
size_t version	error
MIN(INT32_MAX, ...) version	builds
this PR (5c5e2fd)	builds
this PR merged with current master (gcc 14.1)	builds

This PR, alone and merged with current master, also builds clean at -O2 and -Os (ARM gcc 14.1). It merges into master without conflicts.

[popcornmix]

@pelwell okay with this version? It will want a "squash and merge".