Re: Hopefully, kmalloc() will always succeed, but if it doesn'tthen....

[Jan Engelhardt]

>> So, if memory allocation to 'a' fails, it is going to kfree 'b'. But since
>> 'b'
>> is not initialized, kfree may crash (unless DEBUG is defined).

... in which case we will be notified:

$ cat test.c
#include <linux/slab.h>

void func(void) {
    char *a, *b;
    if((a = kmalloc(10, GFP_KERNEL)) == NULL)
        goto err;
    if((b = kmalloc(10, GFP_KERNEL)) == NULL)
        goto err;

 err:
    kfree(a);
    kfree(b);
    return;
}

$ make -C /erk/kernel/linux-2.6.19-rc2 M=$PWD
  CC [M]  /dev/shm/test.o
/dev/shm/test.c: In function âfuncâ:
/dev/shm/test.c:4: warning: âbâ may be used uninitialized in this 
function


Compared to the whole source tree, the kernel has very few "may be 
uninitialized" spots. And stochastically, it is quite unlikely that all 
of them are caused by a construct like the above.


>> I have seen the same case at many places when allocating in a loop.
>
> So you found a bug. Why not send a patch to fix it?


	-`J'
--