illumos gate

I tested this using the following small program, once using realloc() and once using reallocf() and checking for memory leaks. This uses realloc[f] to allocate memory, resize memory and free memory. With realloc there was a resulting memory leak from the failed realloc case, with reallocf there was none, see below.

#include <stdio.h>
#include <stdlib.h>
#include <assert.h>

int
main()
{
        void *p;

        p = reallocf(NULL, 73);
        assert(p != NULL);

        p = reallocf(p, 1024);
        assert(p != NULL);

        // realloc will fail
        p = reallocf(p, SIZE_MAX - 4096);
        assert(p == NULL);

        p = reallocf(NULL, 81);
        assert(p != NULL);

        // realloc will free memory
        p = reallocf(p, 0);
        assert(p == NULL);
}

With realloc()

bloody% UMEM_DEBUG=default mdb ./realloc
> ::load libumem.so.1
> ::bp exit
> ::run
mdb: stop at exit
mdb: target stopped at:
libc_hwcap1.so.1`exit:  pushl  %ebx
> ::findleaks -d
CACHE     LEAKED   BUFCTL CALLER
08192010       1 081bc970 main+0x50
------------------------------------------------------------------------
   Total       1 buffer, 1152 bytes

umem_alloc_1152 leak: 1 buffer, 1152 bytes
            ADDR          BUFADDR        TIMESTAMP           THREAD
                            CACHE          LASTLOG         CONTENTS
         81bc970          81bfb00     31c37e8bc084                1
                          8192010                0                0
                 libumem.so.1`umem_cache_alloc_debug+0x216
                 libumem.so.1`umem_cache_alloc+0x1f5
                 libumem.so.1`umem_alloc+0x5c
                 libumem.so.1`umem_malloc+0x25
                 libumem.so.1`realloc+0x4a
                 main+0x50
                 _start_crt+0x96
                 _start+0x1a

>

With reallocf()

bloody% UMEM_DEBUG=default mdb ./reallocf
> ::load libumem.so.1
> ::bp exit
> ::run
mdb: stop at exit
mdb: target stopped at:
libc_hwcap1.so.1`exit:  pushl  %ebx
> ::findleaks -d
findleaks: no memory leaks detected

I also did a full build of OmniOS userland following this change and checked that, for example, the dma software was now using the new system function, and that it functions.

bloody:omnios.bloody:master% nm /usr/lib/smtp/dma/dma | grep reallocf
[289]   |             4251552|                  63|FUNC |GLOB |0    |14     |reallocf
bloody:omnios.bloody:master% nm build/dma/tmp/src/dma | grep reallocf
[289] |             4218520|                   0|FUNC |GLOB |0    |UNDEF  |reallocf