Stack smashing detected
Stack Smashing here is actually caused due to a protection mechanism used by gcc to detect buffer overflow errors. For example in the following snippet:
#include <stdio.h>
void func()
{
char array[10];
gets(array);
}
int main(int argc, char **argv)
{
func();
}
The compiler, (in this case gcc) adds protection variables (called canaries) which have known values. An input string of size greater than 10 causes corruption of this variable resulting in SIGABRT to terminate the program.
To get some insight, you can try disabling this protection of gcc using option -fno-stack-protector while compiling. In that case you will get a different error, most likely a segmentation fault as you are trying to access an illegal memory location. Note that -fstack-protector should always be turned on for release builds as it is a security feature.
You can get some information about the point of overflow by running the program with a debugger. Valgrind doesn't work well with stack-related errors, but like a debugger, it may help you pin-point the location and reason for the crash.
Minimal reproduction example with disassembly analysis
main.c
void myfunc(char *const src, int len) {
int i;
for (i = 0; i < len; ++i) {
src[i] = 42;
}
}
int main(void) {
char arr[] = {'a', 'b', 'c', 'd'};
int len = sizeof(arr);
myfunc(arr, len + 1);
return 0;
}
GitHub upstream.
Compile and run:
gcc -fstack-protector-all -g -O0 -std=c99 main.c
ulimit -c unlimited && rm -f core
./a.out
fails as desired:
*** stack smashing detected ***: terminated
Aborted (core dumped)
Tested on Ubuntu 20.04, GCC 10.2.0.
On Ubuntu 16.04, GCC 6.4.0, I could reproduce with -fstack-protector instead of -fstack-protector-all, but it stopped blowing up when I tested on GCC 10.2.0 as per Geng Jiawen's comment. man gcc clarifies that as suggested by the option name, the -all version adds checks more aggressively, and therefore presumably incurs a larger performance loss:
-fstack-protector
Emit extra code to check for buffer overflows, such as stack smashing attacks. This is done by adding a guard variable to functions with vulnerable objects. This includes functions that call "alloca", and functions with buffers larger than or equal to 8 bytes. The guards are initialized when a function is entered and then checked when the function exits. If a guard check fails, an error message is printed and the program exits. Only variables that are actually allocated on the stack are considered, optimized away variables or variables allocated in registers don't count.
-fstack-protector-all
Like -fstack-protector except that all functions are protected.
Disassembly
Now we look at the disassembly:
objdump -D a.out
which contains:
int main (void){
400579: 55 push %rbp
40057a: 48 89 e5 mov %rsp,%rbp
# Allocate 0x10 of stack space.
40057d: 48 83 ec 10 sub $0x10,%rsp
# Put the 8 byte canary from %fs:0x28 to -0x8(%rbp),
# which is right at the bottom of the stack.
400581: 64 48 8b 04 25 28 00 mov %fs:0x28,%rax
400588: 00 00
40058a: 48 89 45 f8 mov %rax,-0x8(%rbp)
40058e: 31 c0 xor %eax,%eax
char arr[] = {'a', 'b', 'c', 'd'};
400590: c6 45 f4 61 movb $0x61,-0xc(%rbp)
400594: c6 45 f5 62 movb $0x62,-0xb(%rbp)
400598: c6 45 f6 63 movb $0x63,-0xa(%rbp)
40059c: c6 45 f7 64 movb $0x64,-0x9(%rbp)
int len = sizeof(arr);
4005a0: c7 45 f0 04 00 00 00 movl $0x4,-0x10(%rbp)
myfunc(arr, len + 1);
4005a7: 8b 45 f0 mov -0x10(%rbp),%eax
4005aa: 8d 50 01 lea 0x1(%rax),%edx
4005ad: 48 8d 45 f4 lea -0xc(%rbp),%rax
4005b1: 89 d6 mov %edx,%esi
4005b3: 48 89 c7 mov %rax,%rdi
4005b6: e8 8b ff ff ff callq 400546 <myfunc>
return 0;
4005bb: b8 00 00 00 00 mov $0x0,%eax
}
# Check that the canary at -0x8(%rbp) hasn't changed after calling myfunc.
# If it has, jump to the failure point __stack_chk_fail.
4005c0: 48 8b 4d f8 mov -0x8(%rbp),%rcx
4005c4: 64 48 33 0c 25 28 00 xor %fs:0x28,%rcx
4005cb: 00 00
4005cd: 74 05 je 4005d4 <main+0x5b>
4005cf: e8 4c fe ff ff callq 400420 <__stack_chk_fail@plt>
# Otherwise, exit normally.
4005d4: c9 leaveq
4005d5: c3 retq
4005d6: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
4005dd: 00 00 00
Notice the handy comments automatically added by objdump's artificial intelligence module.
If you run this program multiple times through GDB, you will see that:
- the canary gets a different random value every time
- the last loop of
myfuncis exactly what modifies the address of the canary
The canary randomized by setting it with %fs:0x28, which contains a random value as explained at:
- https://unix.stackexchange.com/questions/453749/what-sets-fs0x28-stack-canary
- Why does this memory address %fs:0x28 ( fs[0x28] ) have a random value?
Debug attempts
From now on, we modify the code:
myfunc(arr, len + 1);
to be instead:
myfunc(arr, len);
myfunc(arr, len + 1); /* line 12 */
myfunc(arr, len);
to be more interesting.
We will then try to see if we can pinpoint the culprit + 1 call with a method more automated than just reading and understanding the entire source code.
gcc -fsanitize=address to enable Google's Address Sanitizer (ASan)
If you recompile with this flag and run the program, it outputs:
#0 0x4008bf in myfunc /home/ciro/test/main.c:4
#1 0x40099b in main /home/ciro/test/main.c:12
#2 0x7fcd2e13d82f in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x2082f)
#3 0x400798 in _start (/home/ciro/test/a.out+0x40079
followed by some more colored output.
This clearly pinpoints the problematic line 12.
The source code for this is at: https://github.com/google/sanitizers but as we saw from the example it is already upstreamed into GCC.
ASan can also detect other memory problems such as memory leaks: How to find memory leak in a C++ code/project?
Valgrind SGCheck
As mentioned by others, Valgrind is not good at solving this kind of problem.
It does have an experimental tool called SGCheck:
SGCheck is a tool for finding overruns of stack and global arrays. It works by using a heuristic approach derived from an observation about the likely forms of stack and global array accesses.
So I was not very surprised when it did not find the error:
valgrind --tool=exp-sgcheck ./a.out
The error message should look like this apparently: Valgrind missing error
GDB
An important observation is that if you run the program through GDB, or examine the core file after the fact:
gdb -nh -q a.out core
then, as we saw on the assembly, GDB should point you to the end of the function that did the canary check:
(gdb) bt
#0 0x00007f0f66e20428 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:54
#1 0x00007f0f66e2202a in __GI_abort () at abort.c:89
#2 0x00007f0f66e627ea in __libc_message (do_abort=do_abort@entry=1, fmt=fmt@entry=0x7f0f66f7a49f "*** %s ***: %s terminated\n") at ../sysdeps/posix/libc_fatal.c:175
#3 0x00007f0f66f0415c in __GI___fortify_fail (msg=<optimized out>, msg@entry=0x7f0f66f7a481 "stack smashing detected") at fortify_fail.c:37
#4 0x00007f0f66f04100 in __stack_chk_fail () at stack_chk_fail.c:28
#5 0x00000000004005f6 in main () at main.c:15
(gdb) f 5
#5 0x00000000004005f6 in main () at main.c:15
15 }
(gdb)
And therefore the problem is likely in one of the calls that this function made.
Next we try to pinpoint the exact failing call by first single stepping up just after the canary is set:
400581: 64 48 8b 04 25 28 00 mov %fs:0x28,%rax
400588: 00 00
40058a: 48 89 45 f8 mov %rax,-0x8(%rbp)
and watching the address:
(gdb) p $rbp - 0x8
$1 = (void *) 0x7fffffffcf18
(gdb) watch 0x7fffffffcf18
Hardware watchpoint 2: *0x7fffffffcf18
(gdb) c
Continuing.
Hardware watchpoint 2: *0x7fffffffcf18
Old value = 1800814336
New value = 1800814378
myfunc (src=0x7fffffffcf14 "*****?Vk\266", <incomplete sequence \355\216>, len=5) at main.c:3
3 for (i = 0; i < len; ++i) {
(gdb) p len
$2 = 5
(gdb) p i
$3 = 4
(gdb) bt
#0 myfunc (src=0x7fffffffcf14 "*****?Vk\266", <incomplete sequence \355\216>, len=5) at main.c:3
#1 0x00000000004005cc in main () at main.c:12
Now, this does leaves us at the right offending instruction: len = 5 and i = 4, and in this particular case, did point us to the culprit line 12.
However, the backtrace is corrupted, and contains some trash. A correct backtrace would look like:
#0 myfunc (src=0x7fffffffcf14 "abcd", len=4) at main.c:3
#1 0x00000000004005b8 in main () at main.c:11
so maybe this could corrupt the stack and prevent you from seeing the trace.
Also, this method requires knowing what is the last call of the canary checking function otherwise you will have false positives, which will not always be feasible, unless you use reverse debugging.
Please look at the following situation:
ab@cd-x:$ cat test_overflow.c
#include <stdio.h>
#include <string.h>
int check_password(char *password){
int flag = 0;
char buffer[20];
strcpy(buffer, password);
if(strcmp(buffer, "mypass") == 0){
flag = 1;
}
if(strcmp(buffer, "yourpass") == 0){
flag = 1;
}
return flag;
}
int main(int argc, char *argv[]){
if(argc >= 2){
if(check_password(argv[1])){
printf("%s", "Access granted\n");
}else{
printf("%s", "Access denied\n");
}
}else{
printf("%s", "Please enter password!\n");
}
}
ab@cd-x:$ gcc -g -fno-stack-protector test_overflow.c
ab@cd-x:$ ./a.out mypass
Access granted
ab@cd-x:$ ./a.out yourpass
Access granted
ab@cd-x:$ ./a.out wepass
Access denied
ab@cd-x:$ ./a.out wepassssssssssssssssss
Access granted
ab@cd-x:$ gcc -g -fstack-protector test_overflow.c
ab@cd-x:$ ./a.out wepass
Access denied
ab@cd-x:$ ./a.out mypass
Access granted
ab@cd-x:$ ./a.out yourpass
Access granted
ab@cd-x:$ ./a.out wepassssssssssssssssss
*** stack smashing detected ***: ./a.out terminated
======= Backtrace: =========
/lib/tls/i686/cmov/libc.so.6(__fortify_fail+0x48)[0xce0ed8]
/lib/tls/i686/cmov/libc.so.6(__fortify_fail+0x0)[0xce0e90]
./a.out[0x8048524]
./a.out[0x8048545]
/lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xe6)[0xc16b56]
./a.out[0x8048411]
======= Memory map: ========
007d9000-007f5000 r-xp 00000000 08:06 5776 /lib/libgcc_s.so.1
007f5000-007f6000 r--p 0001b000 08:06 5776 /lib/libgcc_s.so.1
007f6000-007f7000 rw-p 0001c000 08:06 5776 /lib/libgcc_s.so.1
0090a000-0090b000 r-xp 00000000 00:00 0 [vdso]
00c00000-00d3e000 r-xp 00000000 08:06 1183 /lib/tls/i686/cmov/libc-2.10.1.so
00d3e000-00d3f000 ---p 0013e000 08:06 1183 /lib/tls/i686/cmov/libc-2.10.1.so
00d3f000-00d41000 r--p 0013e000 08:06 1183 /lib/tls/i686/cmov/libc-2.10.1.so
00d41000-00d42000 rw-p 00140000 08:06 1183 /lib/tls/i686/cmov/libc-2.10.1.so
00d42000-00d45000 rw-p 00000000 00:00 0
00e0c000-00e27000 r-xp 00000000 08:06 4213 /lib/ld-2.10.1.so
00e27000-00e28000 r--p 0001a000 08:06 4213 /lib/ld-2.10.1.so
00e28000-00e29000 rw-p 0001b000 08:06 4213 /lib/ld-2.10.1.so
08048000-08049000 r-xp 00000000 08:05 1056811 /dos/hacking/test/a.out
08049000-0804a000 r--p 00000000 08:05 1056811 /dos/hacking/test/a.out
0804a000-0804b000 rw-p 00001000 08:05 1056811 /dos/hacking/test/a.out
08675000-08696000 rw-p 00000000 00:00 0 [heap]
b76fe000-b76ff000 rw-p 00000000 00:00 0
b7717000-b7719000 rw-p 00000000 00:00 0
bfc1c000-bfc31000 rw-p 00000000 00:00 0 [stack]
Aborted
ab@cd-x:$
When I disabled the stack smashing protector no errors were detected, which should have happened when I used "./a.out wepassssssssssssssssss"
So to answer your question above, the message "** stack smashing detected : xxx" was displayed because your stack smashing protector was active and found that there is stack overflow in your program.
Just find out where that occurs, and fix it.
You could try to debug the problem using valgrind:
The Valgrind distribution currently includes six production-quality tools: a memory error detector, two thread error detectors, a cache and branch-prediction profiler, a call-graph generating cache profiler, and a heap profiler. It also includes two experimental tools: a heap/stack/global array overrun detector, and a SimPoint basic block vector generator. It runs on the following platforms: X86/Linux, AMD64/Linux, PPC32/Linux, PPC64/Linux, and X86/Darwin (Mac OS X).