Re: [PATCH] mcheck.3, mtrace.3: update examples for glibc 2.34

Re: [PATCH] mcheck.3, mtrace.3: update examples for glibc 2.34

[Alejandro Colomar]

[-- Attachment #1: Type: text/plain, Size: 1974 bytes --]

Hi Vincent,

On Tue, Oct 24, 2023 at 04:31:17PM +0200, Vincent Lefevre wrote:
> The glibc 2.34 NEWS file says that mtrace() and mcheck() have now been
> disabled in the main C library, and preloading libc_malloc_debug.so is

Can you please clarify "disabled"?  What is the behavior if that library
is not preloaded?  My attempt to run this program is that mcheck(3)
returns -1.  So, the preload interposes the dummy function that only
returns -1 with an actual mcheck(3) implementation?

Are these functions deprecated by glibc?  If so, we should probably use
[[deprecated]] in the SYNOPSIS.  Or did they just move the functionality
for other reasons but without deprecating?

Should we document the need for LD_PRELOAD in the LIBRARY section?

Thanks,
Alex

> needed to get this functionality back.
> 
> So the examples now need a "LD_PRELOAD=libc_malloc_debug.so".
> 
> Signed-off-by: Vincent Lefevre <vincent@vinc17.net>
> ---
>  man3/mcheck.3 | 2 +-
>  man3/mtrace.3 | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/man3/mcheck.3 b/man3/mcheck.3
> index 285ea946c..e17650a4c 100644
> --- a/man3/mcheck.3
> +++ b/man3/mcheck.3
> @@ -169,7 +169,7 @@ when running the program:
>  .PP
>  .in +4n
>  .EX
> -.RB "$" " ./a.out"
> +.RB "$ " "LD_PRELOAD=libc_malloc_debug.so ./a.out"
>  About to free
>  \&
>  About to free a second time
> diff --git a/man3/mtrace.3 b/man3/mtrace.3
> index f30511cd0..d0cde42c9 100644
> --- a/man3/mtrace.3
> +++ b/man3/mtrace.3
> @@ -153,7 +153,7 @@ diagnosed memory leaks at two different locations in the program:
>  .EX
>  .RB "$ " "cc \-g t_mtrace.c \-o t_mtrace"
>  .RB "$ " "export MALLOC_TRACE=/tmp/t"
> -.RB "$ " "./t_mtrace"
> +.RB "$ " "LD_PRELOAD=libc_malloc_debug.so ./t_mtrace"
>  .RB "$ " "mtrace ./t_mtrace $MALLOC_TRACE"
>  Memory not freed:
>  -\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-
> -- 
> 2.42.0
> 

-- 
<https://www.alejandro-colomar.es/>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH] mcheck.3, mtrace.3: update examples for glibc 2.34

[Adhemerval Zanella Netto]

On 24/10/23 12:32, Alejandro Colomar wrote:
> Hi Vincent,
> 
> On Tue, Oct 24, 2023 at 04:31:17PM +0200, Vincent Lefevre wrote:
>> The glibc 2.34 NEWS file says that mtrace() and mcheck() have now been
>> disabled in the main C library, and preloading libc_malloc_debug.so is
> 
> Can you please clarify "disabled"?  What is the behavior if that library
> is not preloaded?  My attempt to run this program is that mcheck(3)
> returns -1.  So, the preload interposes the dummy function that only
> returns -1 with an actual mcheck(3) implementation?

The libc mcheck and mcheck_pedantic now return -1, while mprobe returns
MCHECK_DISABLED.  All the implementation that used to live in libc.so
were moved to libc_malloc_debug.so.

> 
> Are these functions deprecated by glibc?  If so, we should probably use
> [[deprecated]] in the SYNOPSIS.  Or did they just move the functionality
> for other reasons but without deprecating?

There is no deprecation because the functionality is still fully supported,
albeit in a different library. Setting to deprecated would require to move
the prototypes to a different header.

> 
> Should we document the need for LD_PRELOAD in the LIBRARY section?

Yes, I think it would be better.

> 
> Thanks,
> Alex
> 
>> needed to get this functionality back.
>>
>> So the examples now need a "LD_PRELOAD=libc_malloc_debug.so".
>>
>> Signed-off-by: Vincent Lefevre <vincent@vinc17.net>
>> ---
>>  man3/mcheck.3 | 2 +-
>>  man3/mtrace.3 | 2 +-
>>  2 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/man3/mcheck.3 b/man3/mcheck.3
>> index 285ea946c..e17650a4c 100644
>> --- a/man3/mcheck.3
>> +++ b/man3/mcheck.3
>> @@ -169,7 +169,7 @@ when running the program:
>>  .PP
>>  .in +4n
>>  .EX
>> -.RB "$" " ./a.out"
>> +.RB "$ " "LD_PRELOAD=libc_malloc_debug.so ./a.out"
>>  About to free
>>  \&
>>  About to free a second time
>> diff --git a/man3/mtrace.3 b/man3/mtrace.3
>> index f30511cd0..d0cde42c9 100644
>> --- a/man3/mtrace.3
>> +++ b/man3/mtrace.3
>> @@ -153,7 +153,7 @@ diagnosed memory leaks at two different locations in the program:
>>  .EX
>>  .RB "$ " "cc \-g t_mtrace.c \-o t_mtrace"
>>  .RB "$ " "export MALLOC_TRACE=/tmp/t"
>> -.RB "$ " "./t_mtrace"
>> +.RB "$ " "LD_PRELOAD=libc_malloc_debug.so ./t_mtrace"
>>  .RB "$ " "mtrace ./t_mtrace $MALLOC_TRACE"
>>  Memory not freed:
>>  -\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-\-
>> -- 
>> 2.42.0
>>
> 

Re: [PATCH] mcheck.3, mtrace.3: update examples for glibc 2.34

[Alejandro Colomar]

[-- Attachment #1: Type: text/plain, Size: 2578 bytes --]

Hi Adhemerval,

On Tue, Oct 24, 2023 at 01:28:06PM -0300, Adhemerval Zanella Netto wrote:
> On 24/10/23 12:32, Alejandro Colomar wrote:
> > On Tue, Oct 24, 2023 at 04:31:17PM +0200, Vincent Lefevre wrote:
> >> The glibc 2.34 NEWS file says that mtrace() and mcheck() have now been
> >> disabled in the main C library, and preloading libc_malloc_debug.so is
> > 
> > Can you please clarify "disabled"?  What is the behavior if that library
> > is not preloaded?  My attempt to run this program is that mcheck(3)
> > returns -1.  So, the preload interposes the dummy function that only
> > returns -1 with an actual mcheck(3) implementation?
> 
> The libc mcheck and mcheck_pedantic now return -1, while mprobe returns
> MCHECK_DISABLED.  All the implementation that used to live in libc.so
> were moved to libc_malloc_debug.so.

Ok.  This should probably be documented.

> 
> > 
> > Are these functions deprecated by glibc?  If so, we should probably use
> > [[deprecated]] in the SYNOPSIS.  Or did they just move the functionality
> > for other reasons but without deprecating?
> 
> There is no deprecation because the functionality is still fully supported,
> albeit in a different library. Setting to deprecated would require to move
> the prototypes to a different header.

Why was it moved to a different library?  I guess it wasn't often used,
maybe?  This should probably go to a HISTORY section.

Regarding the move, maybe I'm doing something wrong, but I'd expect the
following to work (but it doesn't):


	$ cc -Wall -Wextra mcheck.c -lc_malloc_debug -lc
	mcheck.c: In function ‘main’:
	mcheck.c:21:12: warning: pointer ‘p’ used after ‘free’ [-Wuse-after-free]
	   21 |            free(p);
	      |            ^~~~~~~
	mcheck.c:19:12: note: call to ‘free’ here
	   19 |            free(p);
	      |            ^~~~~~~
	$ ./a.out 
	mcheck() failed
	$ ldd ./a.out 
		linux-vdso.so.1 (0x00007fff048de000)
		libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f82fe810000)
		/lib64/ld-linux-x86-64.so.2 (0x00007f82fea0c000)

Do you know why this program isn't linking correctly to
libc_malloc_debug?

Thanks,
Alex

> > 
> > Should we document the need for LD_PRELOAD in the LIBRARY section?
> 
> Yes, I think it would be better.
> 
> > 
> > Thanks,
> > Alex
> > 
> >> needed to get this functionality back.
> >>
> >> So the examples now need a "LD_PRELOAD=libc_malloc_debug.so".
> >>
> >> Signed-off-by: Vincent Lefevre <vincent@vinc17.net>

-- 
<https://www.alejandro-colomar.es/>

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH] mcheck.3, mtrace.3: update examples for glibc 2.34

[DJ Delorie]

Alejandro Colomar <alx@kernel.org> writes:
> Why was it moved to a different library?

Security.  Moving unneeded code out of libc.so makes it unavailable to
hackers.  The malloc-related code in glibc is a common target for such
things.