Adventure With Stack Smashing Protector

Adam 'pi3' Zabrocki
11 November 2013
Adventure with Stack Smashing Protector (SSP)

Introduction.
I was heavily playing with Stack Smashing Protector a few years ago. Some of my research (observation) I
decided to publish on phrack magazine but not everything. Two years ago my professional life moved to
the Windows environment and unfortunately I didnt have time to play with UNIX world as much as
before. One weekend I decided to reanalyze SSP code again and this write-up is describing a few of my
observations Ive made during the work
which can be shortly summarized as (details can be found at Random ideas section):
Not security related
1. We can change programs name (from SSP perspective) via overwriting memory region where
pointer to "argv[0]" points to.
2. We can crash Stack Smashing Protector code in many ways:
a. Via corrupting memory region pointed by "__environ" variable.
b. Via setting "LIBC_FATAL_STDERR_" to the edge of valid addresses.
c. Via forcing "alloca()" to fail e.g. stack exhaustion.
d. There is one more bug which Im analyzing more comprehensively at point 4. It may
indirectly force SSP to crash. It exists in DWARF stack (state) machine which is responsible
for gathering information about the stack trace ("__backtrace()") and prints it.
3. We can slightly control SSPs execution flow. (Un)Fortunately it doesnt have any influence for the
main execution (what about security?). Following scenarios are possible:
a. Force SSP to open "/dev/tty"
b. Force SSP not to open "/dev/tty" and assign to the "fd" descriptor
"STDERR_FILENO" value:
#define
STDERR_FILENO
/* Standard error output.
*/
c. Crash SSP via 2b. scenario

4. We can crash indirectly SSP via unwinding algorithm (read-AV or we can be killed by
"gcc_unreachable" or "gcc_assert" function) DWARF stack (state) machine:
a. Simulate FDE object was not found
b. Simulate FDE object was found.
Somehow security related (look at Random ideas section for details):
1. We can force SSP to allocate a lot of memory and cause Denial of Service via Resource Exhaustion
attack.
2. Theoretical Information leak:
a. Stack cookie information leak.
b. Any kind of information leak
c. File corruption.
1
Adam 'pi3' Zabrocki

11 November 2013
Stack Smashing Protector (SSP) a.k.a ProPolice under the microscope.

GNU Compiler Collection (GCC) includes SSP implementation and is the most commonly used one. Let's
analyze the full code chain called during SSP execution. If corruption on the stack is detected (stack canary
is not valid), the following function is called:
"debug/stack_chk_fail.c"
void
__attribute__ ((noreturn))
__stack_chk_fail (void)
{
__fortify_fail ("stack smashing detected");
}
Nothing fancy here so lets move forward:
"debug/fortify_fail.c"
void
__attribute__ ((noreturn))
__fortify_fail (msg)
const char *msg;
{
/* The loop is added only to keep gcc happy. */
while (1)
__libc_message (2, "*** %s ***: %s terminated\n",
msg, __libc_argv[0] ?: "<unknown>");
}
libc_hidden_def (__fortify_fail)
First discovery in the very early stage is __libc_argv[0] cannot be trusted, because the memory area
where it's pointing to can be modified at the time of crash (each pointer on the stack may be corrupted).
Moving forward:
"sysdeps/unix/sysv/linux/libc_fatal.c"
/* Abort with an error message. */
void
__libc_message (int do_abort, const char *fmt, ...)
{
va_list ap;
va_list ap_copy;
int fd = -1;
va_start (ap, fmt);
va_copy (ap_copy, ap);
#ifdef FATAL_PREPARE
FATAL_PREPARE;
#endif
Adam 'pi3' Zabrocki

11 November 2013
/* Open a descriptor for /dev/tty unless the user explicitly
requests errors on standard error. */
const char *on_2 = __libc_secure_getenv ("LIBC_FATAL_STDERR_");
if (on_2 == NULL || *on_2 == '\0')
fd = open_not_cancel_2 (_PATH_TTY, O_RDWR | O_NOCTTY |
O_NDELAY);
if (fd == -1)
fd = STDERR_FILENO;
struct str_list *list = NULL;
int nlist = 0;
const char *cp = fmt;
while (*cp != '\0')
{
/* Find the next "%s" or the end of the string.
const char *next = cp;
while (next[0] != '%' || next[1] != 's')
{
next = __strchrnul (next + 1, '%');
*/
if (next[0] == '\0')
break;
}
/* Determine what to print. */
const char *str;
size_t len;
if (cp[0] == '%' && cp[1] == 's')
{
str = va_arg (ap, const char *);
len = strlen (str);
cp += 2;
}
else
{
str = cp;
len = next - cp;
cp = next;
}
struct str_list *newp = alloca (sizeof (struct str_list));
newp->str = str;
newp->len = len;
newp->next = list;
list = newp;
++nlist;
}
bool written = false;
if (nlist > 0)
3
Adam 'pi3' Zabrocki

11 November 2013
{
struct iovec *iov = alloca (nlist * sizeof (struct iovec));
ssize_t total = 0;
for (int cnt = nlist - 1; cnt >= 0; --cnt)
{
iov[cnt].iov_base = (void *) list->str;
iov[cnt].iov_len = list->len;
total += list->len;
list = list->next;
}
INTERNAL_SYSCALL_DECL (err);
ssize_t cnt;
do
cnt = INTERNAL_SYSCALL (writev, err, 3, fd, iov, nlist);
while (INTERNAL_SYSCALL_ERROR_P (cnt, err)
&& INTERNAL_SYSCALL_ERRNO (cnt, err) == EINTR);
if (cnt == total)
written = true;
if (do_abort)
{
total = ((total + 1 + GLRO(dl_pagesize) - 1)
& ~(GLRO(dl_pagesize) - 1));
struct abort_msg_s *buf = __mmap (NULL, total,
PROT_READ | PROT_WRITE,
MAP_ANON | MAP_PRIVATE, -1, 0);
if (__builtin_expect (buf != MAP_FAILED, 1))
{
buf->size = total;
char *wp = buf->msg;
for (int cnt = 0; cnt < nlist; ++cnt)
wp = mempcpy (wp, iov[cnt].iov_base, iov[cnt].iov_len);
*wp = '\0';
/* We have to free the old buffer since the application
might
catch the SIGABRT signal. */
struct abort_msg_s *old = atomic_exchange_acq
(&__abort_msg,
buf);
if (old != NULL)
__munmap (old, old->size);
}
}
}
va_end (ap);
/* If we
had no success writing the message, use syslog.

4
*/
Adam 'pi3' Zabrocki

11 November 2013
if (! written)
vsyslog (LOG_ERR, fmt, ap_copy);
va_end (ap_copy);
if (do_abort)
{
if (do_abort > 1 && written)
{
void *addrs[64];
#define naddrs (sizeof (addrs) / sizeof (addrs[0]))
int n = __backtrace (addrs, naddrs);
if (n > 2)
{
#define strnsize(str) str, strlen (str)
#define writestr(str) write_not_cancel (fd, str)
writestr (strnsize ("======= Backtrace: =========\n"));
__backtrace_symbols_fd (addrs + 1, n - 1, fd);
writestr (strnsize ("======= Memory map: ========\n"));
int fd2 = open_not_cancel_2 ("/proc/self/maps",
O_RDONLY);
char buf[1024];
ssize_t n2;
while ((n2 = read_not_cancel (fd2, buf, sizeof (buf)))
> 0)
if (write_not_cancel (fd, buf, n2) != n2)
break;
close_not_cancel_no_status (fd2);
}
}
/* Terminate the process.
abort ();
*/
}
}
What is interesting in this function? At first, before function abort() is called a lot of code (too much
;)) is executed. This is not amazing idea, because corrupted process cannot be trusted, and execution of
any code (especially the code which relies on any pointer[s]) is unexpected.
Let's look closer for the following line:
Which essentially executes:
"stdlib/secure-getenv.c"
char *
__libc_secure_getenv (name)
const char *name;
{
5
Adam 'pi3' Zabrocki

11 November 2013
return __libc_enable_secure ? NULL : getenv (name);
}
Moving forward:
"stdlib/getenv.c"
char *
getenv (name)
const char *name;
{
size_t len = strlen (name);
char **ep;
uint16_t name_start;
if (__environ == NULL || name[0] == '\0')
return NULL;
if (name[1] == '\0')
{
/* The name of the variable consists of only one character.
Therefore
the first two characters of the environment entry are this
character
and a '=' character. */
#if __BYTE_ORDER == __LITTLE_ENDIAN || !_STRING_ARCH_unaligned
name_start = ('=' << 8) | *(const unsigned char *) name;
#else
# if __BYTE_ORDER == __BIG_ENDIAN
name_start = '=' | ((*(const unsigned char *) name) << 8);
# else
#error "Funny byte order."
# endif
#endif
for (ep = __environ; *ep != NULL; ++ep)
{
#if _STRING_ARCH_unaligned
uint16_t ep_start = *(uint16_t *) *ep;
#else
uint16_t ep_start = (((unsigned char *) *ep)[0]
| (((unsigned char *) *ep)[1] << 8));
#endif
if (name_start == ep_start)
return &(*ep)[2];
}
}
else
{
name_start = *(const uint16_t *) name;
#else
name_start = (((const unsigned char *) name)[0]
6
Adam 'pi3' Zabrocki

11 November 2013
| (((const unsigned char *) name)[1] << 8));
#endif
len -= 2;
name += 2;
for (ep = __environ; *ep != NULL; ++ep)
{
uint16_t ep_start = *(uint16_t *) *ep;
#else
| (((unsigned char *) *ep)[1] << 8));
#endif
if (name_start == ep_start && !strncmp (*ep + 2, name, len)
&& (*ep)[len + 2] == '=')
return &(*ep)[len + 3];
}
}
return NULL;
}
libc_hidden_def (getenv)
Essentially this function goes through the environment block/array (pointer to the pointers) and checks if
first two characters are equivalent to the arguments'. If yes it checks the rest of the string. The last step
of this function is to verify if character '=' exists and in this case address to the next byte is returned.
Again, this code relies on the pointers which can be corrupted. We are able to easily crash this code here
when it references blocks from the environment block/array (which I will prove later).
Additionally, bytes after '=' character is not verified and just simple address is returned.
After executing getenv() function, following code is executed:
/* Open a descriptor for /dev/tty unless the user explicitly
if (on_2 == NULL || *on_2 == '\0')
O_NDELAY);
if (fd == -1)
fd = STDERR_FILENO;
If not NULL is return by __libc_secure_getenv(), the code references the address and verifies if
it is pointing to the NULL. When both checks are passed function open_not_cancel_2() is called. In
the end just default open() function is executed. What is _PATH_TTY?
#define _PATH_TTY
"/dev/tty"
We are able to execute (control) 3 possible scenarios:
Adam 'pi3' Zabrocki

11 November 2013
1. Force SSP to open /dev/tty.
2. Force SSP to NOT open /dev/tty and just assign to the 'fd' descriptor value STDERR_FILENO
which means:
#define
STDERR_FILENO
*/
3. Crash SSP via setting LIBC_FATAL_STDERR_ to the edge of valid addresses

SSP can be crashed in much easier (than 3rd scenario) way just by completely messing up environmental
block (variable __environ).
Another interesting piece of code dynamically allocates memory via alloca() function. For example:
...
struct str_list *newp = alloca (sizeof (struct str_list));
newp->str = str;
newp->len = len;
newp->next = list;
list = newp;
++nlist;
...
ssize_t total = 0;
{
total += list->len;
list = list->next;
}
...
Again, this code is dangerous because SIGSEGV can be received in some specific scenarios. Let me quote
following information from the GNU:
"Normally, gcc(1) translates calls to alloca() with inlined code.
This is not done when either the -ansi, -std=c89, -std=c99, or the
-std=c11 option is given and the header <alloca.h> is not included.
Otherwise (without an -ansi or -std=c* option) the glibc version
of <stdlib.h> includes <alloca.h> and that contains the lines:
#ifdef __GNUC__
#define alloca(size)
#endif
__builtin_alloca (size)
with messy consequences if one has a private version of this

function.
Adam 'pi3' Zabrocki

11 November 2013
The fact that the code is inlined means that it is impossible to
take the address of this function, or to change its behavior by
linking with a different library.
The inlined code often consists of a single instruction adjusting
the stack pointer, and does not check for stack overflow. Thus,
there is no NULL error return."
and continuing:
"There is no error indication if the stack frame cannot be extended.
(However, after a failed allocation, the program is likely to
receive a SIGSEGV signal if it attempts to access the unallocated
space.)"
OK, so now let's analyze something more interesting. Let's look for the following code:
--- CUT --struct str_list *list = NULL;
int nlist = 0;
const char *cp = fmt;
while (*cp != '\0')
{
...
...
/* Determine what to print. */
const char *str;
size_t len;
if (cp[0] == '%' && cp[1] == 's')
{
str = va_arg (ap, const char *);
[1] len = strlen (str);
cp += 2;
}
...
...
}
bool written = false;
if (nlist > 0)
{
ssize_t total = 0;
{
[2] total += list->len;
list = list->next;
9
Adam 'pi3' Zabrocki

11 November 2013
}
...
...
if (do_abort)
{
[3] total = ((total + 1 + GLRO(dl_pagesize) - 1)
& ~(GLRO(dl_pagesize) - 1));
[4] struct abort_msg_s *buf = __mmap (NULL, total,
PROT_READ | PROT_WRITE,
MAP_ANON | MAP_PRIVATE, -1, 0);
if (__builtin_expect (buf != MAP_FAILED, 1))
{
[5]
buf->size = total;
[6]
char *wp = buf->msg;
for (int cnt = 0; cnt < nlist; ++cnt)
[7]
wp = mempcpy (wp, iov[cnt].iov_base, iov[cnt].iov_len);
*wp = '\0';
/* We have to free the old buffer since the application
might
catch the SIGABRT signal. */
struct abort_msg_s *old = atomic_exchange_acq
(&__abort_msg,
buf);
if (old != NULL)
__munmap (old, old->size);
}
}
}
...
...
...
}
--- CUT --I've added a few tags with numbers. Let's start from the [1] tag. If function format includes any string
argument ("%s") sequence it will be extracted (va_arg()), assigned to the "str" variable, length will
be calculated (via strlen()) and assigned to the "len" variable. This scenario will be executed for every
"%s" formatter.
This is very important, because we can control one of the argument to the string formatter. If you look
closer at the beginning of this write-up you will realize why. Short reminder:
__libc_message (2, "*** %s ***: %s terminated\n",
msg, __libc_argv[0] ?: "<unknown>");
We can overflow memory where pointer "__libc_argv[0]" points to and change the displayed name
of crashed application. What is even more important, we can change behavior of the "len" and "str"
variables. In fact in some way we can control "len" variable.
10
Adam 'pi3' Zabrocki

11 November 2013
Next, if we move to the [2] tag, you may discover that we can indirectly control "total" variable as well.
This variable is updated each passing of the loop. The code inside just go through the list (built before)
and for every substring calculates their length and updates "iov" "database". The "total" variable
keeps the full length calculated from each substring.
At line [3], variable "total" is recalculated (aligned to the page) and at line [4] used as a size argument
to the "__mmap" function (dynamic memory allocation). Most of you probably realize that we can control
how much memory will be dynamically allocated. Next (line [5]) "total" is kept in the newly allocated
buffers metadata. At line [6] dynamic buffer is assigned to the temporary pointer. Line [7] is inside of the
loop which extracts previously created "iov" "database" and copy all data to the newly allocated
memory.
What can we get via this scenario? We can force SSP to allocate big chunk of memory which later will be
referenced and some data copied. This may results with small resources exhaustion attack.
The last stage of "__libc_message" function is to execute following code in case "do_abort" is
declared (which is in our situation):
if (do_abort)
{
if (do_abort > 1 && written)
{
void *addrs[64];
#define naddrs (sizeof (addrs) / sizeof (addrs[0]))
int n = __backtrace (addrs, naddrs);
if (n > 2)
{
#define strnsize(str) str, strlen (str)
#define writestr(str) write_not_cancel (fd, str)
writestr (strnsize ("======= Backtrace: =========\n"));
__backtrace_symbols_fd (addrs + 1, n - 1, fd);
writestr (strnsize ("======= Memory map: ========\n"));
int fd2 = open_not_cancel_2 ("/proc/self/maps",
O_RDONLY);
char buf[1024];
ssize_t n2;
while ((n2 = read_not_cancel (fd2, buf, sizeof (buf)))
> 0)
if (write_not_cancel (fd, buf, n2) != n2)
break;
}
}
/* Terminate the process.
abort ();
}
11
*/
Adam 'pi3' Zabrocki

11 November 2013
The most important and interesting is "__backtrace" function. Its very easy to crash the SSP code on
read access violation (AV) in the depth of function calls through the "__backtrace". Backtracing in gcc
heavily use DWARF. Before we analyze source code in details, its good time to describe a bit more what
and how is DWARF used for
"Debugging With Attributed Record Formats" DWARF.
DWARF is a debugging format used to describe programs in C and other similar programming languages.
It is most widely associated with the ELF object format but it has been used with other object file formats.
Additionally gcc uses DWARF mechanism for stack unwinding in general, and also for C++ exception
handling.
To handle an exception, the stack must be unwound. Unfortunately this problem cant be shortened just
to the walk the call stack following return address pointers to find all call frames. Mainly because this
information are not enough to restore execution to an exception handler as well as this process does not
respect register state. To solve this problems Call-Frame Information section (unwinding information) of
the DWARF standard has been adopted (with some changes) for exception handling.
Quoting the excellent research paper "Exploiting the hard-working dwarf" by James Oakley and Sergey
Bratus:
"Conceptually, what this unwinding information describes is a large
table. The rows of the table correspond to machine instructions in
the program text, and the columns correspond to registers and
Canonical Frame Address (CFA). Each row describes how to restore
the machine state (the values of the registers and CFA) for every
instruction at the previous call frame as if control were to return
up the stack from that instruction. DWARF allows for an arbitrary
number of registers, identified merely by number. It is up to
individual ABIs to define a mapping between DWARF register numbers
and the hardware registers. The DWARF registers are not required
to map to actual hardware registers, but may be used internally,
as is often done with a DWARF register for the return address.
Each cell of this table holds a rule detailing how the contents of
the register will be restored for the previous call frame. DWARF
allows for several types of rules, and the curious reader is invited
to find them in the DWARF standard. Most registers are restored
either from another register or from a memory location accessed at
some offset from the CFA.
We note that this table, if constructed in its entirety, would be
absurdly large, larger than the text of the program itself. There
are many empty cells and many duplicated entries in columns. Much
of the DWARF call frame information standard is essentially a
compression technique, allowing to provide sufficient information
at runtime to build parts of the table as needed without the full,
prohibitively large, table ever being built or stored. This
compression is performed by introducing the concept of Frame
Description Entities (FDEs) and DWARF instructions. An FDE
corresponds to a logical block of program text and describes how
unwinding may be done from within that block. Each FDE contains a
series of DWARF instructions. Each instruction either specifies
12
Adam 'pi3' Zabrocki

11 November 2013
one of the column rules (registers) as from our table above or
specifies which text locations the register rules apply to."
More details may be found in DWARF Debugging Standard Website.
It is also worthiest to understand how exception handler is encoded and handled because conception of
backtracing in gcc is very similar (to be honest, almost the same excluding call to the EH personality
routine) to the backtracing which we want to analyze ("__backtrace" function). I want to quote again
James Oakleys and Sergey Bratuss paper:
"DWARF is designed as a debugging format, where the debugger is in
control of how far to unwind the stack. DWARF therefore does not
provide any mechanism to govern halting the unwinding process. What
it does provide is the means for augmentation to the standard.
Certain DWARF data structures include an augmentation string, the
contents of which are implementation defined, allowing a DWARF
producer to communicate to a compatible DWARF consumer information
not controlled by the standard. The augmentations to be used on
Linux and x86 64 are well-defined. These augmentations allow a
language-specific data area (LSDA) and personality routine to be
associated with every FDE.
When unwinding an FDE, the exception handling process is required
to call the personality routine associated with the FDE. The
personality routine interprets the LSDA and determines if a handler
for the exception has been found. The actual contents of the LSDA
are not defined by any standard, and two separate compilation units
originally written in different languages and using different LSDA
formats may coexist in the same program, as they will be served by
separate personality routines.
The result of these design decisions is that the encoding of where
exception handlers are located and what type of exceptions they
handle is mostly nonstandardized. The best known source of
information on the format used by gcc is the verbose assembly code
generated by gcc. (...) In an ELF binary, the section .gcc except
table contains the LSDAs. In the environment we are concerned with,
an LSDA breaks the text region described by the corresponding FDE
into call sites. Each call site corresponds to code within a try
block (to use C++ terminology) and has a pointer to a chain of C++
typeinfo descriptors. These objects are used by the personality
routine to determine whether the thrown exception can be handled
in the current frame.
(...)
During Exception Process, libgcc computes the machine state as a
result of the unwinding, directly restores the necessary registers,
and then returns into the handler code, which is known as the
landing pad. We note that, at least in current (4.5.2) gcc
implementations, this means that at the time execution is first
returned to the handler code, the data from which the registers
were restored will still be present below the stack pointer until
it is overwritten"
13
Adam 'pi3' Zabrocki

11 November 2013
Now we have solid knowledge about DWARF itself and some expectation how gcc should use it in
backtracing algorithm. Lets analyze following code:
"sysdeps/x86_64/backtrace.c"
int
__backtrace (array, size)
void **array;
int size;
{
struct trace_arg arg = { .array = array, .cfa = 0, .size = size,
.cnt = -1 };
#ifdef SHARED
__libc_once_define (static, once);
__libc_once (once, init);
if (unwind_backtrace == NULL)
return 0;
#endif
if (size >= 1)
unwind_backtrace (backtrace_helper, &arg);
/* _Unwind_Backtrace seems to put NULL address above
_start. Fix it up here. */
if (arg.cnt > 1 && arg.array[arg.cnt - 1] == NULL)
--arg.cnt;
return arg.cnt != -1 ? arg.cnt : 0;
}
weak_alias (__backtrace, backtrace)
libc_hidden_def (__backtrace)
Where:
static _Unwind_Reason_Code (*unwind_backtrace) (_Unwind_Trace_Fn,
void *);
...
static void *libgcc_handle;
...
...
libgcc_handle = __libc_dlopen ("libgcc_s.so.1");
...
unwind_backtrace
=
__libc_dlsym
(libgcc_handle,
"_Unwind_Backtrace");
...
Before we move to the "_Unwind_Backtrace" function, lets see helper function passed as an
argument to it:
static _Unwind_Reason_Code
backtrace_helper (struct _Unwind_Context *ctx, void *a)
14
Adam 'pi3' Zabrocki

11 November 2013
{
struct trace_arg *arg = a;
/* We are first called with address in the __backtrace function.
Skip it. */
if (arg->cnt != -1)
{
arg->array[arg->cnt] = (void *) unwind_getip (ctx);
/* Check whether we make any progress. */
_Unwind_Word cfa = unwind_getcfa (ctx);
if (arg->cnt > 0 && arg->array[arg->cnt
>array[arg->cnt]
&& cfa == arg->cfa)
return _URC_END_OF_STACK;
arg->cfa = cfa;
}
if (++arg->cnt == arg->size)
return _URC_NO_REASON;
}
1]
==
arg-
Where:
unwind_getip = __libc_dlsym (libgcc_handle, "_Unwind_GetIP");
unwind_getcfa = (__libc_dlsym (libgcc_handle, "_Unwind_GetCFA")
?: dummy_getcfa);
inline _Unwind_Ptr
_Unwind_GetIP (struct _Unwind_Context *context)
{
return (_Unwind_Ptr) context->ra;
}
_Unwind_Word
_Unwind_GetCFA (struct _Unwind_Context *context)
{
return (_Unwind_Ptr) context->cfa;
}
In short, helper function is responsible for checking if there is any progress in stack unwinding by
analyzing CFA. It also prevents from the looping around the same frames.
Returning to the main unwinding function:
"libgcc/unwind.inc"
/* Perform stack backtrace through unwind data.
*/
_Unwind_Reason_Code LIBGCC2_UNWIND_ATTRIBUTE
_Unwind_Backtrace(_Unwind_Trace_Fn trace, void * trace_argument)
{
15
Adam 'pi3' Zabrocki

11 November 2013
struct _Unwind_Context context;
_Unwind_Reason_Code code;
uw_init_context (&context);
while (1)
{
_Unwind_FrameState fs;
/* Set up fs to describe the FDE for the caller of context.
*/
code = uw_frame_state_for (&context, &fs);
if (code != _URC_NO_REASON && code != _URC_END_OF_STACK)
return _URC_FATAL_PHASE1_ERROR;
/* Call trace function. */
if ((*trace) (&context, trace_argument) != _URC_NO_REASON)
/* We're done at end of stack.
if (code == _URC_END_OF_STACK)
break;
*/
/* Update context to describe the same frame as fs.

uw_update_context (&context, &fs);
*/
}
return code;
}
In short this function is responsible for setting up current frame state ("fs" variable) based on current
context. After that "context" is updated for the next frame and parsing starts again. This infinitive loop
will break if algorithm detects that current frame is the last one ("_URC_END_OF_STACK").
Lets move to the most important function in this algorithm:
"libgcc/unwind-dw2.c"
/* Given the _Unwind_Context CONTEXT for a stack frame, look up
the FDE for
its caller and decode it into FS. This function also sets the
args_size and lsda members of CONTEXT, as they are really
information
about the caller's frame. */
uw_frame_state_for
(struct
_Unwind_Context
_Unwind_FrameState *fs)
{
const struct dwarf_fde *fde;
const struct dwarf_cie *cie;
const unsigned char *aug, *insn, *end;
16
*context,
Adam 'pi3' Zabrocki

11 November 2013
memset (fs, 0, sizeof (*fs));
context->args_size = 0;
context->lsda = 0;
if (context->ra == 0)
fde = _Unwind_Find_FDE
(context) - 1,
(context->ra
_Unwind_IsSignalFrame
&context->bases);
if (fde == NULL)
{
#ifdef MD_FALLBACK_FRAME_STATE_FOR
/* Couldn't find frame unwind info for this function.
target-specific
fallback
mechanism.
This
necessarily
not provide a personality routine or LSDA. */
return MD_FALLBACK_FRAME_STATE_FOR (context, fs);
#else
#endif
}
Try a
will
fs->pc = context->bases.func;
cie = get_cie (fde);
insn = extract_cie_info (cie, context, fs);
if (insn == NULL)
/* CIE contained unknown augmentation. */
/* First decode all the insns in the CIE. */
end = (const unsigned char *) next_fde ((const struct dwarf_fde
*) cie);
execute_cfa_program (insn, end, context, fs);
/* Locate augmentation for the fde. */
aug = (const unsigned char *) fde + sizeof (*fde);
aug += 2 * size_of_encoded_value (fs->fde_encoding);
insn = NULL;
if (fs->saw_z)
{
_uleb128_t i;
aug = read_uleb128 (aug, &i);
insn = aug + i;
}
if (fs->lsda_encoding != DW_EH_PE_omit)
{
_Unwind_Ptr lsda;
aug = read_encoded_value (context, fs->lsda_encoding, aug,
&lsda);
17
Adam 'pi3' Zabrocki

11 November 2013
context->lsda = (void *) lsda;
}
/* Then the insns in the FDE up to our target PC.
if (insn == NULL)
insn = aug;
end = (const unsigned char *) next_fde (fde);
*/
}
If return address of current context is 0 (which is indicator for the end of stack) function immediately
returns. Otherwise complicated "_Unwind_Find_FDE" function is called. The main goal of it is to find
FDE object based on current context and return address:
const fde *
_Unwind_Find_FDE (void *pc, struct dwarf_eh_bases *bases)
{
struct object *ob;
const fde *f = NULL;
init_object_mutex_once ();
__gthread_mutex_lock (&object_mutex);
/* Linear search through the classified objects, to find the one
containing the pc. Note that pc_begin is sorted descending,
and
we expect objects to be non-overlapping. */
for (ob = seen_objects; ob; ob = ob->next)
if (pc >= ob->pc_begin)
{
f = search_object (ob, pc);
if (f)
goto fini;
break;
}
/* Classify and search the objects we've not yet processed.
while ((ob = unseen_objects))
{
struct object **p;
unseen_objects = ob->next;
f = search_object (ob, pc);
/* Insert the object into the classified list.
for (p = &seen_objects; *p ; p = &(*p)->next)
if ((*p)->pc_begin < ob->pc_begin)
break;
ob->next = *p;
18
*/
*/
Adam 'pi3' Zabrocki

11 November 2013
*p = ob;
if (f)
goto fini;
}
fini:
__gthread_mutex_unlock (&object_mutex);
if (f)
{
int encoding;
_Unwind_Ptr func;
bases->tbase = ob->tbase;
bases->dbase = ob->dbase;
encoding = ob->s.b.encoding;
if (ob->s.b.mixed_encoding)
encoding = get_fde_encoding (f);
read_encoded_value_with_base
(encoding,
base_from_object
(encoding, ob),
f->pc_begin, &func);
bases->func = (void *) func;
}
return f;
}
This function is responsible to find out FDE object based on current return address read from the frame
(which can be fully controllable by us). The key function is "search_object":
static const fde *
search_object (struct object* ob, void *pc)
{
/* If the data hasn't been sorted, try to do this now.
have
more memory available than last time we tried. */
if (! ob->s.b.sorted)
{
init_object (ob);
We may
/* Despite the above comment, the normal reason to get here

is
that we've not processed this object before.
range
check is in order. */
if (pc < ob->pc_begin)
return NULL;
}
if (ob->s.b.sorted)
19
A quick
Adam 'pi3' Zabrocki

11 November 2013
{
if (ob->s.b.mixed_encoding)
return binary_search_mixed_encoding_fdes (ob, pc);
else if (ob->s.b.encoding == DW_EH_PE_absptr)
return binary_search_unencoded_fdes (ob, pc);
else
return binary_search_single_encoding_fdes (ob, pc);
}
else
{
/* Long slow laborious linear search, cos we've no memory.
*/
if (ob->s.b.from_array)
{
fde **p;
for (p = ob->u.array; *p ; p++)
{
const fde *f = linear_search_fdes (ob, *p, pc);
if (f)
return f;
}
return NULL;
}
else
return linear_search_fdes (ob, ob->u.single, pc);
}
}
In
general,
different
type
of
searching
algorithm
is
executed
("binary_search_mixed_encoding_fdes",
"binary_search_unencoded_fdes",
"binary_search_single_encoding_fdes", "linear_search_fdes"). Each of the function
depends on return address as a range of search. Because at stack overflow bugs we fully control return
address we can point it to the memory where special prepared bytes can be recognized as correct (or not)
and specific existing in the process FDE object can be chosen.
Next based on what "_Unwind_Find_FDE" found (or not) CIE object may be calculated. Quoting gcc
internal source code comments:
/*
CIE - Common Information Element
FDE - Frame Descriptor Element
There is one per function, and it describes where the function code
is located, and what the register lifetimes and stack layout are
within the function.
The data structures are defined in the DWARF specification,
although not in a very readable way (see LITERATURE).
Every time an exception is thrown, the code needs to locate the
FDE for the current function, and starts to look for exception
regions from that FDE. This works in a two-level search:
a) in a linear search, find the shared image (i.e. DLL)
20
Adam 'pi3' Zabrocki

11 November 2013
containing the PC
b) using the FDE table for that shared object, locate the FDE
using binary search (which requires the sorting). */
This is quite interesting satiation because we can choose which code path to execute. Lets at first simulate
(analyze) easier one none of the FDE objects was found. In this case following lines are executed:
if (fde == NULL)
{
#ifdef MD_FALLBACK_FRAME_STATE_FOR
target-specific
fallback
mechanism.
This
necessarily
#else
#endif
}
Try a
will
"MD_FALLBACK_FRAME_STATE_FOR" is defined by default so:

obj-x86_64-redhat-linux/x86_64-redhat-linux/libgcc/md-unwindsupport.h:
#define MD_FALLBACK_FRAME_STATE_FOR x86_64_fallback_frame_state
obj-x86_64-redhat-linux/x86_64-redhat-linux/libgcc/md-unwindsupport.h:
#define MD_FALLBACK_FRAME_STATE_FOR x86_fallback_frame_state
Im using 64 bits VM for this research so this case will be analyzed. Fortunately there is not much
differences between them:
x86_64_fallback_frame_state (struct _Unwind_Context *context,
{
unsigned char *pc = context->ra;
struct sigcontext *sc;
long new_cfa;
/* movq $__NR_rt_sigreturn, %rax ; syscall. */
#ifdef __LP64__
#define RT_SIGRETURN_SYSCALL
0x050f0000000fc0c7ULL
#else
0x050f40000201c0c7ULL
#endif
if (*(unsigned char *)(pc+0) == 0x48
&& *(unsigned long long *)(pc+1) == RT_SIGRETURN_SYSCALL)
{
struct ucontext *uc_ = context->cfa;
21
Adam 'pi3' Zabrocki

11 November 2013
/* The void * cast is necessary to avoid an aliasing warning.
The aliasing warning is correct, but should not be a
problem
because it does not alias anything. */
sc = (struct sigcontext *) (void *) &uc_->uc_mcontext;
}
else
new_cfa = sc->rsp;
fs->regs.cfa_how = CFA_REG_OFFSET;
/* Register 7 is rsp */
fs->regs.cfa_reg = 7;
fs->regs.cfa_offset = new_cfa - (long) context->cfa;
/* The SVR4 register numbering macros aren't usable in libgcc.
*/
fs->regs.reg[0].how = REG_SAVED_OFFSET;
fs->regs.reg[0].loc.offset = (long)&sc->rax - new_cfa;
fs->regs.reg[1].loc.offset = (long)&sc->rdx - new_cfa;
fs->regs.reg[2].loc.offset = (long)&sc->rcx - new_cfa;
fs->regs.reg[3].loc.offset = (long)&sc->rbx - new_cfa;
fs->regs.reg[4].loc.offset = (long)&sc->rsi - new_cfa;
fs->regs.reg[5].loc.offset = (long)&sc->rdi - new_cfa;
fs->regs.reg[6].loc.offset = (long)&sc->rbp - new_cfa;
fs->regs.reg[8].loc.offset = (long)&sc->r8 - new_cfa;
fs->regs.reg[16].loc.offset = (long)&sc->rip - new_cfa;
fs->retaddr_column = 16;
fs->signal_frame = 1;
22
Adam 'pi3' Zabrocki

11 November 2013
}
Following line:
Assigns our controllable return address to the "pc" pointer (program counter). Without any validation
following references are done:
&& *(unsigned long long *)(pc+1) == RT_SIGRETURN_SYSCALL)
Thats why SSP by default crashes at this lines of code whenever return address is overwritten by random
address. What will happen if we point it to the controllable and valid memory (which of course may be
safely referenced)?
sc = (struct sigcontext *) (void *) &uc_->uc_mcontext;
At this point we control signal context:
The rest of the code fills in frame state ("_Unwind_FrameState *fs") using our controllable values.
After this operation code will return to the main unwinding loop ("_Unwind_Backtrace" function).
Next helper function overtake the control via following call:
if ((*trace) (&context, trace_argument) != _URC_NO_REASON)
As I described this function before, helper function is responsible for checking if there is any progress in
stack unwinding by analyzing CFA. It also prevents from the looping around the same frames. What is
important it uses following data:
return (_Unwind_Ptr) context->ra;
return (_Unwind_Ptr) context->cfa;
<- _Unwind_GetIP function

<- _Unwind_GetCFA function
Both values are fully controllable. In the end of the unwinding loop, context is updated
("uw_update_context"). At this point new frame is found and parsed (using our fully controllable
data):
/* CONTEXT describes the unwind state for a frame, and FS describes
the FDE
of its caller. Update CONTEXT to refer to the caller as well.
Note
that the args_size and lsda members are not updated here, but
later in
uw_frame_state_for. */
static void
23
Adam 'pi3' Zabrocki

11 November 2013
uw_update_context
(struct
_Unwind_Context
{
uw_update_context_1 (context, fs);
*context,
/* In general this unwinder doesn't make any distinction between

undefined and same_value rule.
Call-saved registers are
assumed
to have same_value rule by default and explicit undefined
rule is handled like same_value. The only exception is
DW_CFA_undefined on retaddr_column which is supposed to
mark outermost frame in DWARF 3. */
if
(fs->regs.reg[DWARF_REG_TO_UNWIND_COLUMN
(fs>retaddr_column)].how
== REG_UNDEFINED)
/* uw_frame_state_for uses context->ra == 0 check to find
outermost
stack frame. */
context->ra = 0;
else
/* Compute the return address now, since the return address
column
can change from frame to frame. */
context->ra = __builtin_extract_return_addr
(_Unwind_GetPtr (context, fs->retaddr_column));
}
This is just a wrapper to "uw_update_context_1". Before we analyze it, let's quickly look for further
if-else block. We are interested in "else" case which updates return address in the context:
Quoting the gcc documentation:
"Built-in Function:
void * __builtin_extract_return_addr (void *addr)
The address as returned by __builtin_return_address may have to be
fed through this function to get the actual encoded address. For
example, on the 31-bit S/390 platform the highest bit has to be
masked out, or on SPARC platforms an offset has to be added for
the true next instruction to be executed.
If no fixup is needed, this function simply passes through addr. "
What does "_Unwind_GetPtr" do?
static inline void *
_Unwind_GetPtr (struct _Unwind_Context *context, int index)
24
Adam 'pi3' Zabrocki

11 November 2013
{
return (void *)(_Unwind_Ptr) _Unwind_GetGR (context, index);
}
where:
_Unwind_GetGR (struct _Unwind_Context *context, int index)
{
int size;
_Unwind_Context_Reg_Val val;
#ifdef DWARF_ZERO_REG
if (index == DWARF_ZERO_REG)
return 0;
#endif
index = DWARF_REG_TO_UNWIND_COLUMN (index);
gcc_assert (index < (int) sizeof(dwarf_reg_size_table));
size = dwarf_reg_size_table[index];
val = context->reg[index];
if
(_Unwind_IsExtendedContext
(context)
>by_value[index])
return _Unwind_Get_Unwind_Word (val);
&&
context-
/* This will segfault if the register hasn't been saved. */

if (size == sizeof(_Unwind_Ptr))
return * (_Unwind_Ptr *) (_Unwind_Internal_Ptr) val;
else
{
gcc_assert (size == sizeof(_Unwind_Word));
return * (_Unwind_Word *) (_Unwind_Internal_Ptr) val;
}
}
In short, this function takes from the context register, value corresponded to the "index" value. It will
be value from the return address register in our case.
What does "uw_update_context_1" function do? It's complicated function which plays with CFA. In
short function trying to calculate CFA through the saved frame pointer. If frame pointer is not saved (might
happen in many architectures or in case of "-fomit-frame-pointer" flag) tracking new CFA is done
via analyzing previous one.
After recalculating new CFA, context is updated by the current registers value in that specific frame. In
some cases "execute_stack_op" function is executed. It's again complicated function which operates
on gcc internal structures. In this case function:
"Decode a DW_OP stack program"
Which is DWARF expression. If every function is finished and new return address is calculated, whole main
loop is executed again to analyze newly calculated context for current frame (newly found one). The whole
25
Adam 'pi3' Zabrocki

11 November 2013
story starts again. If newly calculated frame have return address pointing somewhere in the unreachable
memory, program will crash at read AV (immediate dereference of return address pointer which
shouldnt be trusted).
(Un)Fortunately I was not able to change read AV to any kind of write AV or anything controllable which
can give me code execution. Maybe I'm too stupid to play with DWARF algorithm and someone finds a
way how to do that. Be aware that we control almost whole context and internal structures, but I was not
able to find a way of controlling any metadata in this algorithm (we can look at it as state/stack machine),
excluding CFA itself and context which is used for dumping necessary informations (debugging, so memory
sections are parsed etc.) and calculating next/new frame...
It's also common to be killed by "gcc_unreachable" function. It's called whenever some internal
function detects that values in the context which points to the critical data are not as it supposed to be.
Similar situation can happen by "gcc_assert" function. Everything need to be perfect aligned and has
perfect values if we dont want to be killed...
In further section of this write-up Im going to simulate this scenario under debugger (gdb).
OK, this was the case if algorithm didnt find any FDE. Would be nice to see what might happen in case
any FDE was found.
The last scenario (hard one) is in case of calculating FDE object. In this case is even worse and more
complicated ;) In theory we have bigger chance of creating write AV / code exec, (un)fortunately I was not
able to do that neither. Lets start
Function "uw_frame_state_for" instead of calling "x86_64_fallback_frame_state" goes
further...
if (insn == NULL)
end = (const unsigned char *) next_fde ((const struct dwarf_fde
*) cie);
If FDE was found, CIE object is calculated relative to FDE:
static inline const struct dwarf_cie *
get_cie (const struct dwarf_fde *f)
{
return (const void *)&f->CIE_delta - f->CIE_delta;
}
26
Adam 'pi3' Zabrocki

11 November 2013
"extract_cie_info" function parses current CIE object and extract necessary information which are
assigned to the frame state ("fs") structure. Additional this function return pointer to the byte after the
augmentation or NULL if undecipherable augmentation was encountered. From this pointer next FDE is
calculated to get all possible instructions for the current FDE. Now it's time to execute BIG and complicated
function - "execute_cfa_program".
At first, how next FDE is calculated? In very simple way:
static inline const fde *
next_fde (const fde *f)
{
return (const fde *) ((const char *) f + f->length + sizeof (f>length));
}
Let's back to the main problem. What does "execute_cfa_program" do? Quoting internal
comments:
/* Decode DWARF 2 call frame information. Takes pointers the
instruction sequence to decode, current register information
and
CIE info, and the PC range to evaluate. */
Further:
/* The comparison with the return address uses < rather than <=
because
we are only interested in the effects of code before the call;
for a
noreturn function, the return address may point to unrelated
code with
a different stack configuration that we are not interested
in. We
assume that the call itself is unwind info-neutral; if not,
or if
there are delay instructions that adjust the stack, these must
be
reflected at the point immediately before the call insn.
In signal frames, return address is after last completed
instruction,
so we add 1 to return address to make the comparison <=. */
Apparently this function emulates DWARF instruction and/or expression. It may be seen as a core of
DWARF state (stack) machine. In the mean time frame status ("fs") is updated using currently parsed
data. If unexpected bytes are parsed, process is killed via "gcc_unreachable" function.
This function is executed twice in the "uw_frame_state_for" function - for current FDE and upper
one. Next function is returned to the main loop and this process might happen again (or previously
analyzed one). I was not able to force this algorithm (state machine) to execute my code or to do writeAV. Again only Read AV or killing process was achieved.
27
Adam 'pi3' Zabrocki

11 November 2013
The main problem is we cant create own FDE but we can still use existing one by confusing DWARF
machine (via controlling return address). Every program has hundreds of existing FDEs. Even if developer
didnt write any Exception Handler (EH), dynamic libraries may have one. Additionally gcc may create
some. Following listening shows how many potential FDEs exists in example program and in glibc:
[pi3@localhost ~]$ readelf -w ./test|grep FDE|sort -u|uniq
00000018 00000014 0000001c FDE cie=00000000 pc=00400540..0040056a
00000048 00000024 0000001c FDE cie=00000030 pc=004004d0..00400540
00000070 0000001c 00000044 FDE cie=00000030 pc=00400630..004006b3
00000090 00000044 00000064 FDE cie=00000030 pc=004006c0..00400725
000000d8 00000014 000000ac FDE cie=00000030 pc=00400730..00400732
[pi3@localhost ~]$ readelf -w /lib/libc-2.17.so|grep FDE|sort u|uniq|wc -l
3665
FDE with DWARF expressions (not only DWARF instructions):
[pi3@localhost ~]$ readelf -w ./test|grep DW_OP
DW_CFA_def_cfa_expression (DW_OP_breg7 (rsp): 8; DW_OP_breg16
(rip): 0; DW_OP_lit15; DW_OP_and; DW_OP_lit11;
DW_OP_ge; DW_OP_lit3; DW_OP_shl; DW_OP_plus)
[pi3@localhost ~]$ readelf -w /lib/libc-2.17.so|grep DW_OP|wc -l
2140
The Lord of the rings and DWARF stories ;)
At this point I would like to dream a bit What may happen if we were able to create own FDE? We would
be able to create any DWARF instruction and/or expression! In that case we can try to exploit DWARF
state (stack) machine itself. Is there any potential code for it? Apparently yes gcc fixed important bug in
DWARF on May 17, 2013. Lets look for DWARF DW_CFA_register instruction before fix:
case DW_CFA_register:
{
_uleb128_t reg2;
insn_ptr = read_uleb128 (insn_ptr, &reg);
insn_ptr = read_uleb128 (insn_ptr, &reg2);
fs->regs.reg[DWARF_REG_TO_UNWIND_COLUMN (reg)].how =
REG_SAVED_REG;
fs->regs.reg[DWARF_REG_TO_UNWIND_COLUMN (reg)].loc.reg
=
(_Unwind_Word)reg2;
}
break;
and after:
case DW_CFA_register:
{
_uleb128_t reg2;
insn_ptr = read_uleb128 (insn_ptr, &reg);
28
Adam 'pi3' Zabrocki

11 November 2013
insn_ptr = read_uleb128 (insn_ptr, &reg2);
reg = DWARF_REG_TO_UNWIND_COLUMN (reg);
if (UNWIND_COLUMN_IN_RANGE (reg))
{
fs->regs.reg[reg].how = REG_SAVED_REG;
fs->regs.reg[reg].loc.reg = (_Unwind_Word)reg2;
}
}
break;
where:
#define UNWIND_COLUMN_IN_RANGE(x) \
__builtin_expect((x) <= DWARF_FRAME_REGISTERS, 1)
And:
#define DWARF_FRAME_REGISTERS 17
Any code compiled by gcc without following patch, is trivial to exploit but this is just random ideas.
Btw. Very interesting challenge was introduce in codegate 2014 CTF which required DWARF exploiting as
well. Only one team solved this problem PPP. In their case EH was executed by throwing an exception
in SIGSEGV handler. They had primitive to overwrite "frame_hdr_cache_head" pointer which points
to the resolved FDEs (EH). Because of that they was able to create own FDE and CIE object, which are
parsed in the DWARF exception handling algorithm. If personality routine exist (knowledge based on the
controllable CIE object), pointer is extracted, "fs" updated and in the end EH called. I do recommend to
read Brian Paks write-up on his blog:
http://www.bpak.org/blog/2014/02/codegate-2014-membership-800pt-pwnable-write-up/
In further section of this write-up Im going to simulate similar scenario under debugger (gdb).
Random ideas
Not security related
OK, let's summarize what can be done (with SPP) from the non-security perspective:
1. We can change programs name (from SSP perspective) via overwriting memory region where
pointer to "argv[0]" points to.
2. We can crash Stack Smashing Protector code in many ways:
a. Via corrupting memory region pointed by "__environ" variable.
b. Via setting "LIBC_FATAL_STDERR_" to the edge of valid addresses.
c. Via forcing "alloca()" to fail e.g. stack exhaustion.
d. There is one more bug which Im analyzing more comprehensively at point 4. It may
indirectly force SSP to crash. It exists in DWARF stack (state) machine which is responsible
for gathering information about the stack trace ("__backtrace()") and prints it.
29
Adam 'pi3' Zabrocki

11 November 2013
3. We can slightly control SSPs execution flow. (Un)Fortunately it doesnt have any influence for the
main execution (what about security?). Following scenarios are possible:
a. Force SSP to open "/dev/tty"
b. Force SSP not to open "/dev/tty" and assign to the "fd" descriptor
"STDERR_FILENO" value:
#define
STDERR_FILENO
*/
c. Crash SSP via 2b. scenario

4. We can crash indirectly SSP via unwinding algorithm (read-AV or we can be killed by
"gcc_unreachable" or "gcc_assert" function) DWARF stack (state) machine:
a. Simulate FDE object was not found
b. Simulate FDE object was found.
Somehow security related

1. We can force SSP to allocate a lot of memory and cause Denial of Service via Resource
Exhaustion attack.
This need to be explained a little bit more... We are controlling a following variable:
ssize_t total = 0;
Which at one point is used as second argument in the "mmap()" function:
void *mmap(void *addr, size_t len, int prot, int flags, int
fildes, off_t off);
Someone may realize that "ssize_t" is cast to the "size_t" type. Apparently it doesnt matter
here. What is important, how is "(s)size_t" defined? C99 standard from 2007 says:
"7.17 Common denitions <stddef.h>
size_t
which is the unsigned integer type of the result of the sizeof
operator; "
and reading further:
"The types used for size_t and ptrdiff_t should not have an integer
conversion rank greater than that of signed long int unless the
implementation supports objects large enough to make this
necessary."
30
Adam 'pi3' Zabrocki

11 November 2013
Which effectively means for 32 bits "(s)size_t" == "int" but for 64 bits "(s)size_t" ==
"long". This type always covers whole process memory address space. Because of that we cant
overflow "total" variable. We may only control one component used for calculation. This
component is calculated via "strlen()" function and will never returns need number (around
0xFFFFF000 on 32 bits and around 0xFFFFFFFFFFFFF000 for 64 bits).
(Un)Fortunately we are still able to force SSP to dynamically allocate relatively large piece of
memory and force it to recopy existing data from the process memory space to the newly
allocated buffer. If you do it in very careful way you can try to point to the data which was paged
out (swapped out) which force system to execute relatively heavy operation of paging in this
data again to the users working set. Next because each page in the newly allocated memory
will be referenced (copy operation via "memcpy()" function) system will generate page fault
for them and will be forced to make real allocation (make it available in current working set)
and recopy physical data from one physical page to another. Additional if the longest consistent
chunk of memory you can find, it will be more effective attack.
Of course its still controversial if we may assign that scenario to the Resource Exhaustion bucket
or not From my perspective it should be but I can also understand if someone disagree with
that.
2. Theoretical Information leak.
If you look closer to the Not security related ideas at point 3b in some very rare situations
(impossible in real world?) remote information leak vulnerability may exists. This scenario
statically assign to the "fd" descriptor value 2 which by default corresponds to the output stream
for errors. Unfortunately you have no guarantee that process didnt change that. It is possible that
application map this descriptor to some opened clients socket (e.g. via common "dup2()"
function). This scenario may happened for any application (library?) which emulates
pseudoterminals etc. Even more rare situation may happened, if application for some reason
closes descriptor number 2 and later tries to open anything (file, device, etc) or tries to create
socket, by default this number (2) will be reused (may be done e.g. by some kind of vulnerability
which allows you to close what you want). SSP sends the output like application name (which is
read from the pointer which may be corrupted), stack trace, etc. exactly to the "fd" descriptor. If
you are able to corrupt application name pointer to something you want to leak, it will be send to
the descriptor 2 (which could be a socket corresponded to the clients connection).
Possible theoretical attack scenarios:
a) Stack cookie Information leak. If application call any protected function
after the stack overflow happened, cookie wont be overflowed and it will
be still save to leak it from the stack. This scenario allows you to defeat
SSP protection in two shots. First, you force SSP to leak stack cookie.
Second shot, you prepare fully working overflow stream (which will
include correct cookie value). Of course this scenario will be possible only
in "fork()"-like applications (exclude applications which do "fork()"
+ "exec*()" like OpenSSH or Postfix).
31
Adam 'pi3' Zabrocki

11 November 2013
You also need to know stack segment address. If ASLR is enabled, first
you can leak stack segment address and next continue original stack
cookie information leak attack.
b) Any kind of information leak. You can leak whatever you like. It may be
useful for ASLR defeating as normal image process leak (shared library
base address, if PIE binary, programs image base address). Leak any kind
of the secret from application if wont be destroyed via this theoretical
attack itself.
c) File corruption. If descriptor will be assigned (correspond) to any file,
SSPs output will corrupt this file. Another theoretical scenario is when
you corrupt programs name pointer to the data which you fully control,
you can corrupt the file with the data which you exactly want. Especially
dangerous if critical files are opened (like "passwd / shadow /
services", etc.).
Lazy practice
Im too lazy to test all possible scenarios thats why I did only few of them
At first let's create very simple vulnerable program:
#include <stdio.h>
int main(int argc, char *argv[]) {
char buf[100];
memset(buf,0x0,sizeof(buf));
if (argv[1])
strcpy(buf,argv[1]);
printf("DONE!\n");
return 0;
}
and compile with "-fstack-protector-all" flag:
[pi3@localhost ~]$ gcc test.c -o test -g -ggdb -fstack-protectorall
test.c: In function main:
test.c:7:4: warning: incompatible implicit declaration of builtin function memset [enabled by default]
memset(buf,0x0,sizeof(buf));
^
32
Adam 'pi3' Zabrocki

11 November 2013
test.c:9:7: warning: incompatible implicit declaration of builtin function strcpy [enabled by default]
strcpy(buf,argv[1]);
^
[pi3@localhost ~]$ ./test `perl -e 'print "A"x110'`
DONE!
*** stack smashing detected ***: ./test terminated
======= Backtrace: =========
/lib64/libc.so.6(__fortify_fail+0x37)[0x35c190d6b7]
/lib64/libc.so.6(__fortify_fail+0x0)[0x35c190d680]
./test[0x40075a]
/lib64/libc.so.6(__libc_start_main+0xf5)[0x35c1821b75]
./test[0x4005f9]
======= Memory map: ========
00400000-00401000 r-xp 00000000 fd: 02 262194
/home/pi3/test
00600000-00601000 r--p 00000000 fd: 02 262194
/home/pi3/test
00601000-00602000 rw-p 00001000 fd: 02 262194
/home/pi3/test
018e7000-01908000 rw-p 00000000 00:00 0
[heap]
35c1000000-35c1021000 r-xp 00000000 fd:01 1061612
/usr/lib64/ld-2.17.so
35c1220000-35c1221000 r--p 00020000 fd:01 1061612
35c1221000-35c1222000 rw-p 00021000 fd:01 1061612
35c1222000-35c1223000 rw-p 00000000 00:00 0
35c1800000-35c19b6000 r-xp 00000000 fd:01 1061613
/usr/lib64/libc-2.17.so
35c19b6000-35c1bb6000 ---p 001b6000 fd:01 1061613
35c1bb6000-35c1bba000 r--p 001b6000 fd:01 1061613
35c1bba000-35c1bbc000 rw-p 001ba000 fd:01 1061613
35c1bbc000-35c1bc1000 rw-p 00000000 00:00 0
35c4000000-35c4015000 r-xp 00000000 fd:01 1061670
/usr/lib64/libgcc_s-4.8.1-20130603.so.1
35c4015000-35c4214000 ---p 00015000 fd:01 1061670
35c4214000-35c4215000 r--p 00014000 fd:01 1061670
35c4215000-35c4216000 rw-p 00015000 fd:01 1061670
7f3ab2d94000-7f3ab2d97000 rw-p 00000000 00:00 0
7f3ab2da7000-7f3ab2dab000 rw-p 00000000 00:00 0
7fff57436000-7fff57457000 rw-p 00000000 00:00 0
[stack]
7fff575d4000-7fff575d6000 r-xp 00000000 00:00 0
[vdso]
33
Adam 'pi3' Zabrocki

11 November 2013
ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0
[vsyscall]
Aborted (core dumped)
[pi3@localhost ~]$
As you can see SSP works correctly, detects overflow, appropriate information was gained, and process
was killed. Everything was printed to the terminal. As we saw in the SSP/Glibc code, SSP also printed
following line:
*** stack smashing detected ***: ./test terminated
Now let's try to cause some bugs in the SSP.
At first, lets try to check if it is possible to change program's name - which is a key point for causing
theoretical security related bugs.
Normal SSP run:
(gdb) r `perl -e 'print "A"x110'`
Starting program: /home/pi3/test `perl -e 'print "A"x110'`
DONE!
*** stack smashing detected ***: /home/pi3/test terminated
======= Backtrace: =========
/home/pi3/test[0x4006b1]
/lib64/libc.so.6(__libc_start_main+0xf5)[0x35c1821b75]
/home/pi3/test[0x400569]
======= Memory map: ========
00400000-00401000 r-xp 00000000 fd: 02 262194
/home/pi3/test
00600000-00601000 r--p 00000000 fd: 02 262194
/home/pi3/test
00601000-00602000 rw-p 00001000 fd: 02 262194
/home/pi3/test
00602000-00623000 rw-p 00000000 00:00 0
[heap]
35c1000000-35c1021000 r-xp 00000000 fd:01 1061612
35c1220000-35c1221000 r--p 00020000 fd:01 1061612
35c1221000-35c1222000 rw-p 00021000 fd:01 1061612
35c1222000-35c1223000 rw-p 00000000 00:00 0
35c1800000-35c19b6000 r-xp 00000000 fd:01 1061613
35c19b6000-35c1bb6000 ---p 001b6000 fd:01 1061613
35c1bb6000-35c1bba000 r--p 001b6000 fd:01 1061613
34
Adam 'pi3' Zabrocki

11 November 2013
35c1bbc000-35c1bc1000 rw-p 00000000 00:00 0
35c4000000-35c4015000 r-xp 00000000 fd:01 1061670
35c4015000-35c4214000 ---p 00015000 fd:01 1061670
35c4214000-35c4215000 r--p 00014000 fd:01 1061670
35c4215000-35c4216000 rw-p 00015000 fd:01 1061670
7ffff7fe6000-7ffff7fe9000 rw-p 00000000 00:00 0
7ffff7ffa000-7ffff7ffd000 rw-p 00000000 00:00 0
7ffff7ffd000-7ffff7fff000 r-xp 00000000 00:00 0
[vdso]
7ffffffde000-7ffffffff000 rw-p 00000000 00:00 0
[stack]
[vsyscall]
Program received signal SIGABRT, Aborted.
0x00000035c1835a19 in __GI_raise (sig=sig@entry=6) at
../nptl/sysdeps/unix/sysv/linux/raise.c:56
56
return INLINE_SYSCALL (tgkill, 3, pid, selftid, sig);
(gdb) print __libc_argv[0]
$10 = 0x7fffffffe445 "/home/pi3/test"
(gdb)
Re-run and overflow arguments (argv):

The program being debugged has been started already.
Start it from the beginning? (y or n) y
DONE!
Program received signal SIGSEGV, Segmentation fault.
__GI_getenv (name=0x35c197bc64 "BC_FATAL_STDERR_",
name@entry=0x35c197bc62 "LIBC_FATAL_STDERR_") at getenv.c:89
89
if (name_start == ep_start && !strncmp (*ep + 2, name,
len)
(gdb) print __libc_argv[0]
$11 = 0x4141414141414141 <Address 0x4141414141414141 out of
bounds>
(gdb)
Done. As we expected its possible. Also, stack trace is not printed and program crashed somewhere
we hit one of the described bugs.
35
Adam 'pi3' Zabrocki

11 November 2013
Not security related:
Crash 2a:
DONE!
__GI_getenv (name=0x35c197bc64 "BC_FATAL_STDERR_",
89
len)
(gdb) bt
#0 __GI_getenv (name=0x35c197bc64 "BC_FATAL_STDERR_",
#1 0x00000035c18391c2 in __GI___libc_secure_getenv
(name=name@entry=0x35c197bc62 "LIBC_FATAL_STDERR_") at securegetenv.c:30
#2 0x00000035c1875a9a in __libc_message
(do_abort=do_abort@entry=2,
fmt=fmt@entry=0x35c197d302 "*** %s ***: %s terminated\n") at
../sysdeps/unix/sysv/linux/libc_fatal.c:66
#3 0x00000035c190d6b7 in __GI___fortify_fail
(msg=msg@entry=0x35c197d2ea "stack smashing detected") at
fortify_fail.c:31
#4 0x00000035c190d680 in __stack_chk_fail () at
stack_chk_fail.c:28
#5 0x00000000004006b1 in main (argc=2, argv=0x7fffffffce58) at
test.c:15
(gdb) list
84 #else
85
86
| (((unsigned char *) *ep)[1] << 8));
87 #endif
88
89
len)
90
&& (*ep)[len + 2] == '=')
91
return &(*ep)[len + 3];
92
}
93
}
(gdb) x/i $rip
=> 0x35c183892d <__GI_getenv+173>: cmp
(%rbx),%r12w
(gdb) i r rbx
rbx
0x4141414141414141
4702111234474983745
(gdb) print ep
$12 = (char **) 0x7fffffffce70
(gdb) print *ep
36
Adam 'pi3' Zabrocki

11 November 2013
$13 = 0x4141414141414141 <Address 0x4141414141414141 out of
bounds>
(gdb)
Crash 2d:
DONE!
x86_64_fallback_frame_state (context=0x7fffffffd3a0,
context=0x7fffffffd3a0, fs=0x7fffffffd490) at ./md-unwindsupport.h:58
58
(gdb) bt
#0 x86_64_fallback_frame_state (context=0x7fffffffd3a0,
context=0x7fffffffd3a0, fs=0x7fffffffd490)
at ./md-unwind-support.h:58
#1 uw_frame_state_for (context=context@entry=0x7fffffffd3a0,
fs=fs@entry=0x7fffffffd490)
at ../../../libgcc/unwind-dw2.c:1253
#2 0x00000035c400ff19 in _Unwind_Backtrace (trace=0x35c1909bc0
<backtrace_helper>, trace_argument=0x7fffffffd650)
at ../../../libgcc/unwind.inc:290
#3 0x00000035c1909d36 in __GI___backtrace
(array=array@entry=0x7fffffffd830, size=size@entry=64)
at ../sysdeps/x86_64/backtrace.c:109
#4 0x00000035c1875d64 in __libc_message
fortify_fail.c:31
stack_chk_fail.c:28
#7 0x00000000004006b1 in main (argc=2, argv=0x7fffffffe0b8) at
test.c:15
(gdb) print context->ra
$14 = (void *) 0x4141414141414141
(gdb) x/i $rip
=> 0x35c400f018 <uw_frame_state_for+1080>: cmpb
$0x48,(%rcx)
(gdb) i r rcx
rcx
0x4141414141414141
4702111234474983745
(gdb) list 50
45 x86_64_fallback_frame_state (struct _Unwind_Context *context,
37
Adam 'pi3' Zabrocki

11 November 2013
46
47 {
48
49
50
long new_cfa;
51
52
/* movq $__NR_rt_sigreturn, %rax ; syscall. */
53 #ifdef __LP64__
54 #define RT_SIGRETURN_SYSCALL
0x050f0000000fc0c7ULL
55 #else
56 #define RT_SIGRETURN_SYSCALL
0x050f40000201c0c7ULL
57 #endif
58
59
&& *(unsigned long long *)(pc+1) ==
RT_SIGRETURN_SYSCALL)
60
{
61
62
/* The void * cast is necessary to avoid an aliasing
warning.
Scenario 3a:
[pi3@localhost ~]$ gdb -q -p 16473
Attaching to process 16473
Reading symbols from /home/pi3/test...done.
Reading symbols from /lib64/libc.so.6...Reading symbols from
/usr/lib/debug/lib64/libc-2.17.so.debug...done.
done.
Loaded symbols for /lib64/libc.so.6
Reading symbols from /lib64/ld-linux-x86-64.so.2...Reading
symbols from /usr/lib/debug/lib64/ld-2.17.so.debug...done.
done.
Loaded symbols for /lib64/ld-linux-x86-64.so.2
0x00000035c18e7650 in __read_nocancel () at
../sysdeps/unix/syscall-template.S:81
81 T_PSEUDO (SYSCALL_SYMBOL, SYSCALL_NAME, SYSCALL_NARGS)
(gdb) break libc_fatal.c:66
Breakpoint 1 at 0x35c1875a37: file
../sysdeps/unix/sysv/linux/libc_fatal.c, line 66.
(gdb) c
Continuing.
Breakpoint 1, __libc_message (do_abort=do_abort@entry=2,
fmt=fmt@entry=0x35c197d302 "*** %s ***: %s terminated\n")
at ../sysdeps/unix/sysv/linux/libc_fatal.c:66
66
const char *on_2 = __libc_secure_getenv
("LIBC_FATAL_STDERR_");
(gdb) list
61
FATAL_PREPARE;
62 #endif
38
Adam 'pi3' Zabrocki

11 November 2013
63
64
/* Open a descriptor for /dev/tty unless the user
explicitly
65
66
67
if (on_2 == NULL || *on_2 == '\0')
68
O_NDELAY);
69
70
if (fd == -1)
(gdb) print on_2
$1 = <optimized out>
Breakpoint 2 at 0x35c1875a9a: file
(gdb) c
Continuing.
67
if (on_2 == NULL || *on_2 == '\0')
(gdb) print on_2
$2 = 0x0
Breakpoint 3 at 0x35c1875abb: file
(gdb) c
Continuing.
70
if (fd == -1)
<some inline debug jump>
(gdb) ni
68
O_NDELAY);
(gdb) print fd
$4 = -1
(gdb) ni
70
if (fd == -1)
(gdb) print fd
$5 = 3
(gdb)
and double check in the list of opened file descriptors for this process:
39
Adam 'pi3' Zabrocki

11 November 2013
[pi3@localhost ~]$ ls
total 0
dr-x------. 2 pi3 pi3
dr-xr-xr-x. 9 pi3 pi3
lrwx------. 1 pi3 pi3
lrwx------. 1 pi3 pi3
lrwx------. 1 pi3 pi3
lrwx------. 1 pi3 pi3
[pi3@localhost ~]$
-al /proc/16473/fd
0
0
64
64
64
64
Sep
Sep
Sep
Sep
Sep
Sep
29
29
29
29
29
29
11:02
11:02
11:06
11:06
11:02
11:06
.
..
0 ->
1 ->
2 ->
3 ->
/dev/pts/3
/dev/pts/3
/dev/pts/3
/dev/tty
Scenario 3b:
done.
done.
Breakpoint 2 at 0x35c1875a9a: file
(gdb) c
Continuing.
66
Breakpoint 3 at 0x35c1875abb: file
(gdb) print on_2
(gdb) c
Continuing.
40
Adam 'pi3' Zabrocki

11 November 2013
67
if (on_2 == NULL || *on_2 == '\0')
(gdb) print on_2
$2 = 0x7fff2b8aafe0 "/tmp/pi3"
(gdb) c
Continuing.
and double check:
[pi3@localhost ~]$ ls
total 0
dr-x------. 2 pi3 pi3
dr-xr-xr-x. 9 pi3 pi3
lrwx------. 1 pi3 pi3
lrwx------. 1 pi3 pi3
lrwx------. 1 pi3 pi3
[pi3@localhost ~]$
-al /proc/16531/fd
0
0
64
64
64
Sep
Sep
Sep
Sep
Sep
29
29
29
29
29
11:11
11:11
11:14
11:14
11:11
.
..
0 -> /dev/pts/3
1 -> /dev/pts/3
2 -> /dev/pts/3
Scenario 4a:
At the beginning lets just prove that simple return address overflow might lead to read-AV crash. First
overflow cookie without touching return address:
[pi3@localhost ~]$ gdb -q ./test
Reading symbols from /home/pi3/test...(no debugging symbols
found)...done.
DONE!
======= Backtrace: =========
/home/pi3/test[0x4006b1]
/lib64/libc.so.6(__libc_start_main+0x80)[0x35c1821b00]
/home/pi3/test[0x400569]
======= Memory map: ========
00400000-00401000 r-xp 00000000 fd:02 262194
/home/pi3/test
00600000-00601000 r--p 00000000 fd:02 262194
/home/pi3/test
00601000-00602000 rw-p 00001000 fd:02 262194
/home/pi3/test
00602000-00623000 rw-p 00000000 00:00 0
[heap]
35c1000000-35c1021000 r-xp 00000000 fd:01 1061612
35c1220000-35c1221000 r--p 00020000 fd:01 1061612
35c1221000-35c1222000 rw-p 00021000 fd:01 1061612
41
Adam 'pi3' Zabrocki

11 November 2013
35c1222000-35c1223000 rw-p 00000000 00:00 0
35c1800000-35c19b6000 r-xp 00000000 fd:01 1061613
35c19b6000-35c1bb6000 ---p 001b6000 fd:01 1061613
35c1bb6000-35c1bba000 r--p 001b6000 fd:01 1061613
35c1bbc000-35c1bc1000 rw-p 00000000 00:00 0
35c4000000-35c4015000 r-xp 00000000 fd:01 1061670
35c4015000-35c4214000 ---p 00015000 fd:01 1061670
35c4214000-35c4215000 r--p 00014000 fd:01 1061670
35c4215000-35c4216000 rw-p 00015000 fd:01 1061670
7ffff7fe6000-7ffff7fe9000 rw-p 00000000 00:00 0
7ffff7ffa000-7ffff7ffd000 rw-p 00000000 00:00 0
7ffff7ffd000-7ffff7fff000 r-xp 00000000 00:00 0
[vdso]
7ffffffde000-7ffffffff000 rw-p 00000000 00:00 0
[stack]
[vsyscall]
56
(gdb) bt
#0 0x00000035c1835a19 in __GI_raise (sig=sig@entry=6) at
#1 0x00000035c1837128 in __GI_abort () at abort.c:90
fortify_fail.c:31
stack_chk_fail.c:28
#5 0x00000000004006b1 in main ()
(gdb) list libc_fatal.c:196
191
192
}
193
}
194
195
/* Terminate the process. */
196
abort ();
42
Adam 'pi3' Zabrocki

11 November 2013
197
198
199
200
}
}
Overwrite return address:

DONE!
x86_64_fallback_frame_state (context=0x7fffffffd420,
context=0x7fffffffd420, fs=0x7fffffffd510) at ./md-unwindsupport.h:58
58
(gdb) bt
#0 x86_64_fallback_frame_state (context=0x7fffffffd420,
context=0x7fffffffd420, fs=0x7fffffffd510)
#1 uw_frame_state_for (context=context@entry=0x7fffffffd420,
#2 0x00000035c400ff19 in _Unwind_Backtrace (trace=0x35c1909bc0
<backtrace_helper>, trace_argument=0x7fffffffd6d0)
(array=array@entry=0x7fffffffd8b0, size=size@entry=64)
fortify_fail.c:31
stack_chk_fail.c:28
#7 0x00000000004006b1 in main ()
(gdb) print pc
$2 = (unsigned char *) 0x41414141 <Address 0x41414141 out of
bounds>
(gdb)
As you can see, instead of killing the process, SIGSEGV was received. The crash happened exactly where
we predict in our previous analyze. Lets now simulate that we fully control memory where return address
point to (signal frame):
43
Adam 'pi3' Zabrocki

11 November 2013
DONE!
Breakpoint 9, uw_frame_state_for
(context=context@entry=0x7fffffffd370,
fs=fs@entry=0x7fffffffd1c0)
1233 {
(gdb) c
...
<many times>
...
$161 = (void *) 0x4141414141414141
(gdb) set context->ra = 0x7ffffffde101
...
<lets pad the memory>
...
(gdb) set $pos = 0
(gdb) while ($pos < 4000)
>set *(0x7ffffffde101+$pos++) = 0x4141414141414141
>end
(gdb) while ($pos < 4000)
>set *(0x7ffffffde101-$pos++) = 0x4141414141414141
>end
(gdb) list
1228
args_size and lsda members of CONTEXT, as they are
really information
1229
about the caller's frame. */
1230
1231 static _Unwind_Reason_Code
1232 uw_frame_state_for (struct _Unwind_Context *context,
1233 {
1234
const struct dwarf_fde *fde;
1235
const struct dwarf_cie *cie;
1236
const unsigned char *aug, *insn, *end;
1237
(gdb)
1238
memset (fs, 0, sizeof (*fs));
1239
context->args_size = 0;
1240
context->lsda = 0;
1241
1242
1243
1244
1245
fde = _Unwind_Find_FDE (context->ra +
_Unwind_IsSignalFrame (context) - 1,
44
Adam 'pi3' Zabrocki

11 November 2013
1246
&context->bases);
1247
if (fde == NULL)
(gdb)
1248
{
1249 #ifdef MD_FALLBACK_FRAME_STATE_FOR
1250
Try a
1251
target-specific fallback mechanism. This will
necessarily
1252
1253
1254 #else
1255
1256 #endif
1257
}
(gdb) b 1247
Breakpoint 20 at 0x35c400ec8c: file ../../../libgcc/unwind-dw2.c,
line 1247.
(gdb) c
Continuing.
1247
if (fde == NULL)
(gdb) del 20
(gdb) list
1242
1243
1244
1245
1246
&context->bases);
1247
if (fde == NULL)
1248
{
1249 #ifdef MD_FALLBACK_FRAME_STATE_FOR
1250
Try a
1251
target-specific fallback mechanism. This will
necessarily
(gdb) si
1245
(gdb)
1247
if (fde == NULL)
(gdb)
1232 uw_frame_state_for (struct _Unwind_Context *context,
(gdb)
0x00000035c400f00c
1232 uw_frame_state_for (struct
_Unwind_Context *context, _Unwind_FrameState *fs)
45
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
x86_64_fallback_frame_state (context=0x7fffffffd370,
context=0x7fffffffd370, fs=0x7fffffffd460) at ./md-unwindsupport.h:68
68
(gdb)
58
(gdb) list
53
#ifdef __LP64__
54
0x050f0000000fc0c7ULL
55
#else
56
0x050f40000201c0c7ULL
57
#endif
58
59
60
{
61
62
warning.
(gdb)
63
The aliasing warning is correct, but should not be
a problem
64
65
sc = (struct sigcontext *) (void *) &uc_>uc_mcontext;
66
}
67
else
68
69
70
new_cfa = sc->rsp;
71
72
(gdb) x/20i $rip
=> 0x35c400f018 <uw_frame_state_for+1080>: cmpb
$0x48,(%rcx)
0x35c400f01b <uw_frame_state_for+1083>: jne
0x35c400eded
<uw_frame_state_for+525>
0x35c400f021 <uw_frame_state_for+1089>: movabs
$0x50f0000000fc0c7,%rsi
0x35c400f02b <uw_frame_state_for+1099>: cmp
%rsi,0x1(%rcx)
0x35c400f02f <uw_frame_state_for+1103>: jne
0x35c400eded
0x35c400f035 <uw_frame_state_for+1109>: mov
0xa0(%rdx),%rax
0x35c400f03c <uw_frame_state_for+1116>: lea
0x90(%rdx),%rsi
0x35c400f043 <uw_frame_state_for+1123>: movl
$0x1,0x140(%r12)
0x35c400f04f <uw_frame_state_for+1135>: movq
$0x7,0x130(%r12)
0x35c400f05b <uw_frame_state_for+1147>: movl
$0x1,0x8(%r12)
46
Adam 'pi3' Zabrocki

11 November 2013
0x35c400f064 <uw_frame_state_for+1156>: movl
$0x1,0x18(%r12)
0x35c400f06d <uw_frame_state_for+1165>: movl
$0x1,0x28(%r12)
%rax,%rcx
0x35c400f079 <uw_frame_state_for+1177>: sub
%rax,%rsi
0x35c400f07c <uw_frame_state_for+1180>: movl
$0x1,0x38(%r12)
0x35c400f085 <uw_frame_state_for+1189>: sub
%rdx,%rcx
%rsi,(%r12)
0x35c400f08c <uw_frame_state_for+1196>: lea
0x88(%rdx),%rsi
%rcx,0x128(%r12)
0x35c400f09b <uw_frame_state_for+1211>: lea
0x28(%rdx),%rcx
(gdb) x/x $rcx
0x7ffffffde101: 0x41
...
<we need to change the memory layout to pass the checks>
...
(gdb) set *$rcx=0x48
(gdb) x/x $rcx
(gdb) echo 0x5 0f 00 00 00 0f c0 c7 \n
0x5 0f 00 00 00 0f c0 c7
(gdb) set *($rcx+5)=00
(gdb) set *($rcx+6)=00
(gdb) set *($rcx+7)=0x0f
(gdb) set *($rcx+8)=0x05
(gdb) x/8x $rcx+1
0x7ffffffde102: 0xc7 0xc0 0x0f 0x00 0x00 0x00 0x0f 0x05
(gdb) si
0x00000035c400f01b
58
(gdb)
59
(gdb)
0x00000035c400f02b
59
&& *(unsigned long long *)(pc+1)
== RT_SIGRETURN_SYSCALL)
(gdb)
0x00000035c400f02f
59
&& *(unsigned long long *)(pc+1)
== RT_SIGRETURN_SYSCALL)
(gdb) list
54
0x050f0000000fc0c7ULL
55
#else
56
0x050f40000201c0c7ULL
57
#endif
58
59
60
{
47
Adam 'pi3' Zabrocki

11 November 2013
61
62
warning.
63
The aliasing warning is correct, but should not be
a problem
(gdb)
64
65
sc = (struct sigcontext *) (void *) &uc_>uc_mcontext;
66
}
67
else
68
69
70
new_cfa = sc->rsp;
71
72
73
(gdb) si
70
new_cfa = sc->rsp;
(gdb) print sc
$163 = (struct sigcontext *) 0x7fffffffdfd8
(gdb) print/x *sc
$165 = {r8 = 0x4141414141414141, r9 = 0x4141414141414141, r10 =
0x4141414141414141, r11 = 0x4141414141414141,
r12 = 0x4141414141414141, r13 = 0x4141414141414141, r14 =
0x4141414141414141, r15 = 0x4141414141414141,
rdi = 0x4141414141414141, rsi = 0x4141414141414141, rbp =
0x4141414141414141, rbx = 0x4141414141414141,
rdx = 0x4141414141414141, rax = 0x4141414141414141, rcx =
0x4141414141414141, rsp = 0x4141414141414141,
rip = 0x7f0041414141, eflags = 0x0, cs = 0x569, gs = 0x40, fs =
0x0, __pad0 = 0x0, err = 0x7fffffffe078, trapno = 0x1c,
oldmask = 0x2, cr2 = 0x7fffffffe367, {fpstate = 0x7fffffffe376,
__fpstate_word = 0x7fffffffe376}, __reserved1 = {0x0,
0x7fffffffe4a3, 0x7fffffffe4ae, 0x7fffffffe4c0,
0x7fffffffe4df, 0x7fffffffe514, 0x7fffffffe52b, 0x7fffffffe53b}}
(gdb) si
78
(gdb)
71
(gdb)
73
(gdb)
77
(gdb)
79
(gdb)
81
(gdb)
74
(gdb)
78
48
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
83
(gdb)
74
(gdb)
78
(gdb)
80
(gdb)
74
(gdb)
78
(gdb)
85
(gdb)
80
(gdb)
87
(gdb)
89
(gdb)
92
(gdb)
80
(gdb)
82
(gdb)
92
(gdb)
94
(gdb)
91
(gdb)
82
(gdb)
93
(gdb)
95
(gdb)
94
(gdb)
82
(gdb)
84
(gdb)
94
(gdb)
96
(gdb)
97
(gdb)
84
49
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
99
(gdb)
101
(gdb)
96
(gdb)
84
(gdb)
86
(gdb)
96
(gdb)
98
(gdb)
103
(gdb)
86
(gdb)
105
(gdb)
98
(gdb)
86
(gdb)
88
(gdb)
98
(gdb)
100
(gdb)
88
(gdb)
100
(gdb)
88
(gdb)
90
(gdb)
100
(gdb)
102
(gdb)
90
(gdb)
102
(gdb)
90
(gdb)
102
(gdb)
104
50
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400f1be
104
fs->regs.reg[14].loc.offset =
(long)&sc->r14 - new_cfa;
(gdb)
0x00000035c400f1c1
104
(long)&sc->r14 - new_cfa;
(gdb)
106
(gdb)
108
(gdb)
0x00000035c400f1d4
108
(long)&sc->rip - new_cfa;
(gdb)
106
(gdb)
111
(gdb)
106
(gdb)
107
(gdb)
108
(gdb)
109
fs->retaddr_column = 16;
(gdb)
110
(gdb)
0x00000035c400f20d
110
(gdb)
uw_frame_state_for (context=context@entry=0x7fffffffd370,
fs=fs@entry=0x7fffffffd460) at ../../../libgcc/unwind-dw2.c:1296
1296 }
(gdb)
0x00000035c400edf1
1296 }
(gdb) list
1291
insn = aug;
1292
1293
1294
1295
1296 }
1297 ^L
1298 typedef struct frame_state
1299 {
1300
void *cfa;
(gdb) si
0x00000035c400edf2
1296 }
(gdb) si
0x00000035c400edf3
1296 }
(gdb)
0x00000035c400edf5
1296 }
51
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400edf7
1296 }
(gdb)
0x00000035c400edf9
1296 }
(gdb)
0x00000035c400edfb
1296 }
(gdb) si
_Unwind_Backtrace (trace=0x35c1909bc0 <backtrace_helper>,
trace_argument=0x7fffffffd620) at ../../../libgcc/unwind.inc:291
291
if (code != _URC_NO_REASON && code !=
_URC_END_OF_STACK)
(gdb)
290
(gdb)
291
_URC_END_OF_STACK)
(gdb)
0x00000035c400ff21
291
if (code != _URC_NO_REASON &&
code != _URC_END_OF_STACK)
(gdb)
0x00000035c400ff23
291
if (code != _URC_NO_REASON &&
code != _URC_END_OF_STACK)
(gdb)
295
if ((*trace) (&context, trace_argument) !=
_URC_NO_REASON)
(gdb)
0x00000035c400fef3
295
if ((*trace) (&context,
trace_argument) != _URC_NO_REASON)
(gdb) n
299
(gdb) print code
$167 = _URC_NO_REASON
(gdb) n
303
(gdb) print fs->regs.cfa_how
$168 = CFA_REG_OFFSET
(gdb) p/x context->cfa
$169 = 0x7fffffffdfb0
(gdb) si
0x00000035c400ff06
303
uw_update_context (&context,
&fs);
(gdb)
0x00000035c400ff09
303
&fs);
(gdb)
uw_update_context (context=context@entry=0x7fffffffd370,
1505 {
(gdb)
0x00000035c400eb51
1505 {
(gdb)
0x00000035c400eb54
1505 {
52
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400eb55
1505 {
(gdb)
0x00000035c400eb58
1505 {
(gdb)
1506
uw_update_context_1 (context, fs);
(gdb)
uw_update_context_1 (context=context@entry=0x7fffffffd370,
1382 {
(gdb)
1383
struct _Unwind_Context orig_context = *context;
(gdb) n
1382 {
(gdb)
1383
(gdb)
1382 {
(gdb)
1383
(gdb)
1405
if (!_Unwind_GetGRPtr (&orig_context,
__builtin_dwarf_sp_column ()))
(gdb)
1406
_Unwind_SetSpColumn (&orig_context, context->cfa,
&tmp_sp);
(gdb)
1407
_Unwind_SetGRPtr (context, __builtin_dwarf_sp_column (),
NULL);
(gdb)
1411
switch (fs->regs.cfa_how)
(gdb)
1407
_Unwind_SetGRPtr (context, __builtin_dwarf_sp_column (),
NULL);
(gdb)
1411
switch (fs->regs.cfa_how)
(gdb)
1414
cfa = _Unwind_GetPtr (&orig_context, fs>regs.cfa_reg);
(gdb)
1415
cfa += fs->regs.cfa_offset;
(gdb)
1416
break;
(gdb)
1436
switch (fs->regs.reg[i].how)
(gdb)
1432
context->cfa = cfa;
(gdb)
1467
_Unwind_SetGRPtr (context, i, (void *) val);
(gdb)
1436
(gdb)
53
Adam 'pi3' Zabrocki

11 November 2013
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
_Unwind_SetGRPtr (context, i,
for (i = 0; i < DWARF_FRAME_REGISTERS + 1; ++i)
54
Adam 'pi3' Zabrocki

11 November 2013
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1443
(gdb)
1435
(gdb)
1436
(gdb)
1435
(gdb)
1491
(gdb)
_Unwind_SetSignalFrame (context, fs->signal_frame);
55
Adam 'pi3' Zabrocki

11 November 2013
1496 }
(gdb)
uw_update_context (context=context@entry=0x7fffffffd370,
1514
if (fs->regs.reg[DWARF_REG_TO_UNWIND_COLUMN (fs>retaddr_column)].how
(gdb)
1523
(gdb)
1522
(gdb) n
1524 }
$170 = (void *) 0x7f0041414141 <- Newly calculated return address
(gdb) list
1519
else
1520
/* Compute the return address now, since the return
address column
1521
can change from frame to frame. */
1522
1523
1524 }
1525
1526 static void
1527 uw_advance_context (struct _Unwind_Context *context,
1528 {
(gdb) x/20i $rip
=> 0x35c400ebba <uw_update_context+106>:
add
$0x8,%rsp
0x35c400ebbe <uw_update_context+110>:
pop
%rbx
0x35c400ebbf <uw_update_context+111>:
pop
%rbp
0x35c400ebc0 <uw_update_context+112>:
retq
nopl
0x0(%rax)
movq
$0x0,0x98(%rbx)
0x35c400ebd3 <uw_update_context+131>:
add
$0x8,%rsp
pop
%rbx
pop
%rbp
retq
0x35c400ebda: nopw
0x0(%rax,%rax,1)
0x35c400ebe0 <uw_frame_state_for>: push
%r15
0x35c400ebe2 <uw_frame_state_for+2>:
mov
$0x180,%edx
push
%r14
push
%r13
0x35c400ebeb <uw_frame_state_for+11>:
mov
%rdi,%r13
0x35c400ebee <uw_frame_state_for+14>:
mov
%rsi,%rdi
0x35c400ebf1 <uw_frame_state_for+17>:
push
%r12
mov
%rsi,%r12
push
%rbp
(gdb) i r rcx
rcx
0x7f0041414141
139639071523137
(gdb) p/x *fs
56
Adam 'pi3' Zabrocki

11 November 2013
$171 = {regs = {reg = {{loc = {reg = 0xbebf3ebebebe9eff, offset =
0xbebf3ebebebe9eff, exp = 0xbebf3ebebebe9eff}, how = 0x1},
{loc = {reg = 0xbebf3ebebebe9ef7, offset =
0xbebf3ebebebe9ef7, exp = 0xbebf3ebebebe9ef7}, how = 0x1}, {loc =
{
reg = 0xbebf3ebebebe9f07, offset = 0xbebf3ebebebe9f07,
exp = 0xbebf3ebebebe9f07}, how = 0x1}, {loc = {
reg = 0xbebf3ebebebe9eef, offset = 0xbebf3ebebebe9eef,
exp = 0xbebf3ebebebe9eef}, how = 0x1}, {loc = {
reg = 0xbebf3ebebebe9edf, offset = 0xbebf3ebebebe9edf,
exp = 0xbebf3ebebebe9edf}, how = 0x1}, {loc = {
reg = 0xbebf3ebebebe9ed7, offset = 0xbebf3ebebebe9ed7,
exp = 0xbebf3ebebebe9ed7}, how = 0x1}, {loc = {
reg = 0xbebf3ebebebe9ee7, offset = 0xbebf3ebebebe9ee7,
exp = 0xbebf3ebebebe9ee7}, how = 0x1}, {loc = {reg = 0x0,
offset = 0x0, exp = 0x0}, how = 0x0}, {loc = {reg =
0xbebf3ebebebe9e97, offset = 0xbebf3ebebebe9e97,
exp = 0xbebf3ebebebe9e97}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9e9f, offset = 0xbebf3ebebebe9e9f,
exp = 0xbebf3ebebebe9e9f}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9ea7, offset = 0xbebf3ebebebe9ea7,
exp = 0xbebf3ebebebe9ea7}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9eaf, offset = 0xbebf3ebebebe9eaf,
exp = 0xbebf3ebebebe9eaf}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9eb7, offset = 0xbebf3ebebebe9eb7,
exp = 0xbebf3ebebebe9eb7}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9ebf, offset = 0xbebf3ebebebe9ebf,
exp = 0xbebf3ebebebe9ebf}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9ec7, offset = 0xbebf3ebebebe9ec7,
exp = 0xbebf3ebebebe9ec7}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9ecf, offset = 0xbebf3ebebebe9ecf,
exp = 0xbebf3ebebebe9ecf}, how = 0x1}, {loc = {reg =
0xbebf3ebebebe9f17, offset = 0xbebf3ebebebe9f17,
exp = 0xbebf3ebebebe9f17}, how = 0x1}, {loc = {reg =
0x0, offset = 0x0, exp = 0x0}, how = 0x0}}, prev = 0x0,
cfa_offset = 0x4140c14141416191, cfa_reg = 0x7, cfa_exp =
0x0, cfa_how = 0x1}, pc = 0x0, personality = 0x0,
data_align = 0x0, code_align = 0x0, retaddr_column = 0x10,
fde_encoding = 0x0, lsda_encoding = 0x0, saw_z = 0x0,
signal_frame = 0x1, eh_ptr = 0x0}
(gdb)
The latest values in the context and frame state variables corresponds to the relative offsets from the
0x4141414141414141 values. Next interation of the main loop eventually cause the read-AV because
currently calculated return address is pointing to the unreachable memory. We returned to the starting
point.
57
Adam 'pi3' Zabrocki

11 November 2013
Scenario 4b lets try to check if DWARF instructions are emulated if we fully control FDE:
DONE!
fs=fs@entry=0x7fffffffd1c0)
1233 {
(gdb) c
Continuing.
...
<keep going until we hit what we want...>
...
(gdb) source pi3-test
<- prepare the memory and do padding
Breakpoint 18 at 0x35c400ec8c: file ../../../libgcc/unwind-dw2.c,
line 1247.
1247
if (fde == NULL)
1245
1247
if (fde == NULL)
1259
get_cie (f=<optimized out>) at ../../../libgcc/unwind-dw2fde.h:157
157
1259
get_cie (f=0x7ffffffde101) at ../../../libgcc/unwind-dw2fde.h:157
157
0x00000035c400ecaf
157
return (const void *)&f->CIE_delta f->CIE_delta;
(gdb) si
extract_cie_info (fs=0x7fffffffd460, context=0x7fffffffd370,
cie=0x7fffbebc9fc4) at ../../../libgcc/unwind-dw2.c:415
415
const unsigned char *aug = cie->augmentation;
(gdb) set cie = context->ra
(gdb) print *cie
$138 = {length = 1094795585, CIE_id = 1094795585, version = 65
'A', augmentation = 0x7ffffffde10a 'A' <repeats 200 times>...}
58
Adam 'pi3' Zabrocki

11 November 2013
(gdb) set cie->version = 0x1
...
<fix the memory to avoid killing by gcc_unreachable() or
gcc_assert()>
...
(gdb) print *cie
$139 = {length = 1094795585, CIE_id = 1094795585, version = 1
'\001',
augmentation = 0x7ffffffde10a 'A' <repeats 200 times>...}
(gdb) si
416
const unsigned char *p = aug + strlen ((const char *)aug)
+ 1;
(gdb)
0x00000035c400ecb9
416
const unsigned char *p = aug +
strlen ((const char *)aug) + 1;
(gdb) n
423
if (aug[0] == 'e' && aug[1] == 'h')
(gdb) si
416
const unsigned char *p = aug + strlen ((const char *)aug)
+ 1;
(gdb)
423
if (aug[0] == 'e' && aug[1] == 'h')
(gdb)
433
if (__builtin_expect (cie->version >= 4, 0))
(gdb) x/20i $rip
=> 0x35c400ecce <uw_frame_state_for+238>:
cmpb
$0x3,0x8(%r14)
0x35c400ecd3 <uw_frame_state_for+243>:
ja
0x35c400f592
0x35c400ecd9 <uw_frame_state_for+249>:
xor
%esi,%esi
0x35c400ecdb <uw_frame_state_for+251>:
xor
%ecx,%ecx
0x35c400ecdd <uw_frame_state_for+253>:
nopl
(%rax)
0x35c400ece0 <uw_frame_state_for+256>:
add
$0x1,%rdi
movzbl 0x1(%rdi),%edx
mov
%rdx,%rax
0x35c400eceb <uw_frame_state_for+267>:
and
$0x7f,%eax
0x35c400ecee <uw_frame_state_for+270>:
shl
%cl,%rax
0x35c400ecf1 <uw_frame_state_for+273>:
add
$0x7,%ecx
or
%rax,%rsi
test
%dl,%dl
js
0x35c400ece0
0x35c400ecfb <uw_frame_state_for+283>:
mov
%rsi,0x160(%r12)
0x35c400ed03 <uw_frame_state_for+291>:
lea
0x40(%rsp),%rsi
callq 0x35c400d090
<read_sleb128>
0x35c400ed0d <uw_frame_state_for+301>:
mov
%rax,%rbx
mov
0x40(%rsp),%rax
xor
%esi,%esi
59
Adam 'pi3' Zabrocki

11 November 2013
(gdb) x/x $r14+0x8
(gdb) si
0x00000035c400ecd3
433
if (__builtin_expect (cie->version
>= 4, 0))
(gdb)
x86_64_fallback_frame_state (context=<optimized out>,
context=<optimized out>, fs=<optimized out>)
68
(gdb) x/20i $rip
=> 0x35c400ecd9 <uw_frame_state_for+249>:
xor
%esi,%esi
0x35c400ecdb <uw_frame_state_for+251>:
xor
%ecx,%ecx
0x35c400ecdd <uw_frame_state_for+253>:
nopl
(%rax)
add
$0x1,%rdi
mov
%rdx,%rax
and
$0x7f,%eax
shl
%cl,%rax
add
$0x7,%ecx
or
%rax,%rsi
test
%dl,%dl
js
0x35c400ece0
mov
%rsi,0x160(%r12)
lea
0x40(%rsp),%rsi
callq 0x35c400d090
<read_sleb128>
mov
%rax,%rbx
mov
0x40(%rsp),%rax
xor
%esi,%esi
xor
%ecx,%ecx
mov
%rax,0x158(%r12)
(gdb) si
0x00000035c400ecdb
68
(gdb)
0x00000035c400ecdd
68
(gdb) x/20i $rip
=> 0x35c400ecdd <uw_frame_state_for+253>:
nopl
(%rax)
add
$0x1,%rdi
mov
%rdx,%rax
and
$0x7f,%eax
shl
%cl,%rax
add
$0x7,%ecx
or
%rax,%rsi
60
Adam 'pi3' Zabrocki

11 November 2013
test
%dl,%dl
js
0x35c400ece0
mov
%rsi,0x160(%r12)
lea
0x40(%rsp),%rsi
callq 0x35c400d090
<read_sleb128>
mov
%rax,%rbx
mov
0x40(%rsp),%rax
xor
%esi,%esi
xor
%ecx,%ecx
mov
%rax,0x158(%r12)
cmpb
$0x1,0x8(%r14)
je
0x35c400f270
(gdb) si
read_uleb128 (val=<optimized out>, p=0x7ffffffdf0a5 "") at
../../../libgcc/unwind-pe.h:140
140
byte = *p++;
(gdb)
0x00000035c400ece4
140
byte = *p++;
(gdb)
141
result |= ((_uleb128_t)byte & 0x7f) << shift;
(gdb)
0x00000035c400eceb
141
result |= ((_uleb128_t)byte &
0x7f) << shift;
(gdb)
0x00000035c400ecee
141
0x7f) << shift;
(gdb)
142
shift += 7;
(gdb)
141
(gdb)
144
while (byte & 0x80);
(gdb)
0x00000035c400ecf9
144
(gdb)
cie=0x7ffffffde101) at ../../../libgcc/unwind-dw2.c:442
442
fs->code_align = (_Unwind_Word)utmp;
(gdb) print utmp
$140 = 0
(gdb) si
443
p = read_sleb128 (p, &stmp);
(gdb) x/20i $rip
=> 0x35c400ed03 <uw_frame_state_for+291>:
lea
0x40(%rsp),%rsi
61
Adam 'pi3' Zabrocki

11 November 2013
callq 0x35c400d090
<read_sleb128>
mov
%rax,%rbx
mov
0x40(%rsp),%rax
xor
%esi,%esi
xor
%ecx,%ecx
mov
%rax,0x158(%r12)
cmpb
$0x1,0x8(%r14)
je
0x35c400f270
0x35c400ed2c <uw_frame_state_for+332>:
nopl
0x0(%rax)
add
$0x1,%rbx
movzbl 0x1(%rbx),%edx
mov
%rdx,%rax
0x35c400ed3b <uw_frame_state_for+347>:
and
$0x7f,%eax
0x35c400ed3e <uw_frame_state_for+350>:
shl
%cl,%rax
add
$0x7,%ecx
or
%rax,%rsi
test
%dl,%dl
js
0x35c400ed30
0x35c400ed4b <uw_frame_state_for+363>:
mov
%rsi,0x168(%r12)
(gdb) si
0x00000035c400ed08
443
(gdb)
read_sleb128 (p=0x7ffffffdf0a6 "", val=val@entry=0x7fffffffd320)
at ../../../libgcc/unwind-pe.h:154
154 {
(gdb)
159
result = 0;
(gdb)
155
unsigned int shift = 0;
(gdb)
0x00000035c400d098
155
(gdb)
162
byte = *p++;
(gdb)
0x00000035c400d0a4
162
byte = *p++;
(gdb)
163
(gdb)
0x00000035c400d0ab
163
0x7f) << shift;
(gdb)
0x00000035c400d0ae
163
0x7f) << shift;
(gdb)
164
shift += 7;
62
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
163
(gdb)
166
(gdb)
0x00000035c400d0ba
166
(gdb)
169
if (shift < 8 * sizeof(result) && (byte & 0x40) != 0)
(gdb)
0x00000035c400d0bf
169
if (shift < 8 * sizeof(result) &&
(byte & 0x40) != 0)
(gdb)
0x00000035c400d0c1
169
(byte & 0x40) != 0)
(gdb)
0x00000035c400d0c4
169
(byte & 0x40) != 0)
(gdb)
172
*val = (_sleb128_t) result;
(gdb)
174 }
(gdb)
0x00000035c400ed0d in extract_cie_info (fs=0x7fffffffd460,
context=0x7fffffffd370, cie=0x7ffffffde101)
443
(gdb)
444
fs->data_align = (_Unwind_Sword)stmp;
(gdb)
445
if (cie->version == 1)
(gdb)
0x00000035c400ed17
445
(gdb)
444
fs->data_align = (_Unwind_Sword)stmp;
(gdb)
445
(gdb)
0x00000035c400ed26
445
(gdb)
446
fs->retaddr_column = *p++;
(gdb)
0x00000035c400f273
446
(gdb)
0x00000035c400f277
446
(gdb)
0x00000035c400f27f
446
(gdb)
452
fs->lsda_encoding = DW_EH_PE_omit;
(gdb)
457
if (*aug == 'z')
(gdb)
417
const unsigned char *ret = NULL;
63
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
457
if (*aug == 'z')
(gdb)
0x00000035c400ed65
457
if (*aug == 'z')
(gdb)
read_encoded_value_with_base (val=<optimized out>, p=<optimized
out>, base=<optimized out>, encoding=<optimized out>)
at ../../../libgcc/unwind-pe.h:225
225
p = read_sleb128 (p, &tmp);
(gdb)
0x00000035c400ed70
225
(gdb)
207
switch (encoding & 0x0f)
(gdb)
225
(gdb)
0x00000035c400ed80
225
(gdb)
cie=0x7ffffffde101) at ../../../libgcc/unwind-dw2.c:467
467
while (*aug != '\0')
(gdb)
0x00000035c400edab
467
while (*aug != '\0')
(gdb)
470
if (aug[0] == 'L')
(gdb)
0x00000035c400edb3
470
if (aug[0] == 'L')
(gdb)
477
else if (aug[0] == 'R')
(gdb)
0x00000035c400ed8a
477
else if (aug[0] == 'R')
(gdb)
484
else if (aug[0] == 'P')
(gdb)
0x00000035c400ed8e
484
else if (aug[0] == 'P')
(gdb)
494
else if (aug[0] == 'S')
(gdb)
0x00000035c400ed92
494
(gdb)
494
(gdb)
1263
if (insn == NULL)
(gdb)
0x00000035c400f2b6
1263
if (insn == NULL)
(gdb) list
1258
1259
1260
1261
64
Adam 'pi3' Zabrocki

11 November 2013
1262
1263
if (insn == NULL)
1264
1265
1266
1267
(gdb) x/20i $rip
=> 0x35c400f2b6 <uw_frame_state_for+1750>: jne
0x35c400eeac
0x35c400f2bc <uw_frame_state_for+1756>: add
$0x58,%rsp
0x35c400f2c0 <uw_frame_state_for+1760>: mov
$0x3,%eax
0x35c400f2c5 <uw_frame_state_for+1765>: pop
%rbx
%rbp
%r12
%r13
0x35c400f2cb <uw_frame_state_for+1771>: pop
%r14
0x35c400f2cd <uw_frame_state_for+1773>: pop
%r15
0x35c400f2cf <uw_frame_state_for+1775>: retq
0x35c400f2d0 <uw_frame_state_for+1776>: lea
0x1(%rsi),%rdi
0x35c400f2d4 <uw_frame_state_for+1780>: movb
$0x0,(%rsi)
0x35c400f2d7 <uw_frame_state_for+1783>: mov
$0x7f,%dl
0x35c400f2d9 <uw_frame_state_for+1785>: test
$0x2,%dil
0x35c400f2dd <uw_frame_state_for+1789>: je
0x35c400ec10
0x35c400f2e3 <uw_frame_state_for+1795>: nopl
0x0(%rax,%rax,1)
0x35c400f2e8 <uw_frame_state_for+1800>: xor
%ecx,%ecx
0x35c400f2ea <uw_frame_state_for+1802>: add
$0x2,%rdi
0x35c400f2ee <uw_frame_state_for+1806>: sub
$0x2,%edx
0x35c400f2f1 <uw_frame_state_for+1809>: mov
%cx,-0x2(%rdi)
$142 = (void *) 0x7ffffffde101
(gdb) set $rdi = context->ra
(gdb) si
1263
if (insn == NULL)
(gdb) x/20i $rip
=> 0x35c400f2b3 <uw_frame_state_for+1747>: test
%rdi,%rdi
0x35c400f2b6 <uw_frame_state_for+1750>: jne
0x35c400eeac
0x35c400f2bc <uw_frame_state_for+1756>: add
$0x58,%rsp
0x35c400f2c0 <uw_frame_state_for+1760>: mov
$0x3,%eax
%rbx
%rbp
%r12
%r13
0x35c400f2cb <uw_frame_state_for+1771>: pop
%r14
0x35c400f2cd <uw_frame_state_for+1773>: pop
%r15
0x35c400f2cf <uw_frame_state_for+1775>: retq
0x35c400f2d0 <uw_frame_state_for+1776>: lea
0x1(%rsi),%rdi
0x35c400f2d4 <uw_frame_state_for+1780>: movb
$0x0,(%rsi)
65
Adam 'pi3' Zabrocki

11 November 2013
0x35c400f2d7 <uw_frame_state_for+1783>: mov
$0x7f,%dl
0x35c400f2d9 <uw_frame_state_for+1785>: test
$0x2,%dil
0x35c400f2dd <uw_frame_state_for+1789>: je
0x35c400ec10
0x35c400f2e3 <uw_frame_state_for+1795>: nopl
0x0(%rax,%rax,1)
0x35c400f2e8 <uw_frame_state_for+1800>: xor
%ecx,%ecx
0x35c400f2ea <uw_frame_state_for+1802>: add
$0x2,%rdi
0x35c400f2ee <uw_frame_state_for+1806>: sub
$0x2,%edx
(gdb) i r rdi r9
rdi
0x7ffffffde101
140737488216321
r9
0x7ffffffde101
140737488216321
(gdb) si
0x00000035c400f2b6
1263
if (insn == NULL)
(gdb)
next_fde (f=0x7ffffffde101) at ../../../libgcc/unwind-dw2fde.h:163
163
return (const fde *) ((const char *) f + f->length +
sizeof (f->length));
(gdb) si
1269
(gdb) print context
$144 = (struct _Unwind_Context *) 0x7fffffffd370
(gdb) print *context
$145 = {reg = {0x0, 0x0, 0x0, 0x7fffffffdef8, 0x0, 0x0,
0x7fffffffdfa0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x7fffffffdec0,
0x7fffffffdec8, 0x7fffffffded0, 0x7fffffffded8,
0x7fffffffdfa8, 0x0}, cfa = 0x7fffffffdfb0, ra = 0x7ffffffde101,
lsda = 0x0, bases = {tbase = 0x0, dbase = 0x0, func = 0x400630
<main>}, flags = 4611686018427387904, version = 0,
args_size = 0, by_value = '\000' <repeats 17 times>}
(gdb) si
0x00000035c400eeb2
1269
execute_cfa_program (insn, end,
context, fs);
(gdb)
163
(gdb) list
158 }
159
160 static inline const fde *
161 next_fde (const fde *f)
162 {
163
164 }
165
66
Adam 'pi3' Zabrocki

11 November 2013
166 extern const fde * _Unwind_Find_FDE (void *, struct
dwarf_eh_bases *);
167
(gdb) si
1269
(gdb)
execute_cfa_program (insn_ptr=0x7ffffffde101 "AAAAAAAA\001", 'A'
<repeats 191 times>...,
insn_end=0x8000413f2246 <Address 0x8000413f2246 out of
bounds>, context=context@entry=0x7fffffffd370,
fs=fs@entry=0x7fffffffd460) at ../../../libgcc/unwinddw2.c:942
942 {
(gdb) si
0x00000035c400d431
942 {
(gdb) si
0x00000035c400d434
942 {
(gdb)
0x00000035c400d436
942 {
(gdb)
0x00000035c400d439
942 {
(gdb)
0x00000035c400d43b
942 {
(gdb)
0x00000035c400d43d
942 {
(gdb)
0x00000035c400d440
942 {
(gdb)
0x00000035c400d442
942 {
(gdb)
0x00000035c400d443
942 {
(gdb)
0x00000035c400d446
942 {
(gdb)
957
while (insn_ptr < insn_end
(gdb) print insn_ptr
$146 = (const unsigned char *) 0x7ffffffde101 "AAAAAAAA\001", 'A'
<repeats 191 times>...
(gdb) print insn_end
$147 = (const unsigned char *) 0x8000413f2246 <Address
0x8000413f2246 out of bounds>
(gdb) print insn_end - insn_ptr
$148 = 1094795589
(gdb) si
946
fs->regs.prev = NULL;
(gdb)
957
(gdb)
_Unwind_IsSignalFrame (context=<optimized out>) at
../../../libgcc/unwind-dw2.c:202
67
Adam 'pi3' Zabrocki

11 November 2013
202
return (context->flags & SIGNAL_FRAME_BIT) ? 1 : 0;
(gdb) list
197 read_8s (const void *p) { const union unaligned *up = p;
return up->s8; }
198 ^L
199 static inline _Unwind_Word
200 _Unwind_IsSignalFrame (struct _Unwind_Context *context)
201 {
202
203 }
204
205 static inline void
206 _Unwind_SetSignalFrame (struct _Unwind_Context *context,
int val)
(gdb) si
958
&& fs->pc < context->ra + _Unwind_IsSignalFrame
(context))
(gdb) list
953
there are delay instructions that adjust the stack,
these must be
954
reflected at the point immediately before the call
insn.
955
In signal frames, return address is after last
completed instruction,
956
so we add 1 to return address to make the comparison
<=. */
957
958
(context))
959
{
960
unsigned char insn = *insn_ptr++;
961
_uleb128_t reg, utmp;
962
_sleb128_t offset, stmp;
(gdb) print fs->pc
$149 = (void *) 0x400630 <main>
$150 = (void *) 0x7ffffffde101
(gdb) si
0x00000035c400d46c
958
&& fs->pc < context->ra +
_Unwind_IsSignalFrame (context))
(gdb) x/20i $rip
=> 0x35c400d46c <execute_cfa_program+60>:
mov
%rdx,%r14
0x35c400d46f <execute_cfa_program+63>:
shr
$0x3f,%rax
0x35c400d473 <execute_cfa_program+67>:
add
0x98(%rdx),%rax
0x35c400d47a <execute_cfa_program+74>:
cmp
%rax,%rcx
68
Adam 'pi3' Zabrocki

11 November 2013
0x35c400d47d <execute_cfa_program+77>:
jae
0x35c400d50b
<execute_cfa_program+219>
lea
0x38(%rbp),%rax
lea
0x5162(%rip),%rdx
# 0x35c40125f0
0x35c400d48e <execute_cfa_program+94>:
xor
%r8d,%r8d
mov
%rax,0x48(%rbp)
0x35c400d495 <execute_cfa_program+101>: nopl
(%rax)
0x35c400d498 <execute_cfa_program+104>: movzbl (%rbx),%eax
0x35c400d49b <execute_cfa_program+107>: lea
0x1(%rbx),%r12
0x35c400d49f <execute_cfa_program+111>: mov
%eax,%esi
0x35c400d4a1 <execute_cfa_program+113>: and
$0xffffffc0,%esi
0x35c400d4a4 <execute_cfa_program+116>: cmp
$0x40,%sil
0x35c400d4a8 <execute_cfa_program+120>: je
0x35c400d4d0
0x35c400d4aa <execute_cfa_program+122>: cmp
$0x80,%sil
0x35c400d4ae <execute_cfa_program+126>: je
0x35c400d520
0x35c400d4b0 <execute_cfa_program+128>: cmp
$0xc0,%sil
0x35c400d4b4 <execute_cfa_program+132>: je
0x35c400d578
(gdb) si
_Unwind_IsSignalFrame (context=0x7fffffffd370) at
202
(gdb)
958
(context))
(gdb) si
0x00000035c400d47a
958
(gdb)
0x00000035c400d47d
958
(gdb)
1169
insn_ptr = read_sleb128 (insn_ptr, &stmp);
(gdb)
985
else switch (insn)
(gdb) print insn
(gdb) b *0x35c400d495
Breakpoint 19 at 0x35c400d495: file ../../../libgcc/unwind-dw2.c,
line 1169.
69
Adam 'pi3' Zabrocki

11 November 2013
(gdb) c
Continuing.
Breakpoint 19, 0x00000035c400d495 in execute_cfa_program (
insn_ptr=0x7ffffffde101 "AAAAAAAA\001", 'A' <repeats 191
times>...,
1169
(gdb) x/20i $rip-10
0x35c400d48b <execute_cfa_program+91>:
push
%rcx
0x35c400d48c <execute_cfa_program+92>:
add
%al,(%rax)
xor
%r8d,%r8d
mov
%rax,0x48(%rbp)
=> 0x35c400d495 <execute_cfa_program+101>: nopl
(%rax)
0x1(%rbx),%r12
%eax,%esi
$0xffffffc0,%esi
$0x40,%sil
0x35c400d4d0
$0x80,%sil
0x35c400d520
$0xc0,%sil
0x35c400d578
0x35c400d4ba <execute_cfa_program+138>: cmp
$0x2f,%al
0x35c400d4bc <execute_cfa_program+140>: ja
0x35c400d5a9
0x35c400d4c2 <execute_cfa_program+146>: movslq
(%rdx,%rax,4),%rax
0x35c400d4c6 <execute_cfa_program+150>: add
%rdx,%rax
0x35c400d4c9 <execute_cfa_program+153>: jmpq
*%rax
(gdb) x/x $rbx
(gdb) set *(0x7ffffffde101) = 0x1
(gdb) x/8x 0x7ffffffde101
0x7ffffffde101: 0x01 0x00 0x00 0x00 0x41 0x41 0x41 0x41
(gdb) set *(0x7ffffffde101) = 0x80 <- emulate DWARF instruction
...
gcc_assert()>
...
(gdb) si
960
(gdb)
70
Adam 'pi3' Zabrocki

11 November 2013
0x00000035c400d49b
960
unsigned char insn =
(gdb)
964
if ((insn & 0xc0) == DW_CFA_advance_loc)
(gdb)
0x00000035c400d4a1
964
if ((insn & 0xc0) ==
DW_CFA_advance_loc)
(gdb)
0x00000035c400d4a4
964
DW_CFA_advance_loc)
(gdb)
0x00000035c400d4a8
964
DW_CFA_advance_loc)
(gdb)
966
else if ((insn & 0xc0) == DW_CFA_offset)
(gdb)
0x00000035c400d4ae
966
else if ((insn & 0xc0) ==
DW_CFA_offset)
(gdb)
968
reg = insn & 0x3f;
(gdb)
0x00000035c400d523
968
reg = insn & 0x3f;
(gdb)
read_uleb128 (val=<optimized out>, p=<optimized out>) at
137
result = 0;
(gdb)
execute_cfa_program (insn_ptr=0x7ffffffde102 "",
insn_end=0x8000413f2246 <Address 0x8000413f2246 out of bounds>,
context=context@entry=0x7fffffffd370,
968
reg = insn & 0x3f;
(gdb) x/20i $rip
=> 0x35c400d528 <execute_cfa_program+248>: and
$0x3f,%edi
0x35c400d52b <execute_cfa_program+251>: xor
%ecx,%ecx
0x35c400d52d <execute_cfa_program+253>: nopl
(%rax)
0x35c400d530 <execute_cfa_program+256>: add
$0x1,%rbx
0x35c400d534 <execute_cfa_program+260>: movzbl 0x1(%rbx),%r9d
0x35c400d539 <execute_cfa_program+265>: mov
%r9,%rax
0x35c400d53c <execute_cfa_program+268>: and
$0x7f,%eax
0x35c400d53f <execute_cfa_program+271>: shl
%cl,%rax
0x35c400d542 <execute_cfa_program+274>: add
$0x7,%ecx
0x35c400d545 <execute_cfa_program+277>: or
%rax,%rsi
0x35c400d548 <execute_cfa_program+280>: test
%r9b,%r9b
0x35c400d54b <execute_cfa_program+283>: js
0x35c400d530
0x35c400d54d <execute_cfa_program+285>: imul
0x158(%r13),%rsi
0x35c400d555 <execute_cfa_program+293>: cmp
$0x11,%rdi
0x35c400d559 <execute_cfa_program+297>: ja
0x35c400d4e8
0x35c400d55b <execute_cfa_program+299>: shl
$0x4,%rdi
71
Adam 'pi3' Zabrocki

11 November 2013
0x35c400d55f <execute_cfa_program+303>: add
%r13,%rdi
0x35c400d562 <execute_cfa_program+306>: movl
$0x1,0x8(%rdi)
%rsi,(%rdi)
0x35c400d56c <execute_cfa_program+316>: jmpq
0x35c400d4e8
(gdb) si
read_uleb128 (val=<synthetic pointer>, p=0x7ffffffde102 "") at
133
(gdb)
0x00000035c400d52d
133
(gdb)
140
byte = *p++;
(gdb)
0x00000035c400d534
140
byte = *p++;
(gdb)
141
(gdb)
0x00000035c400d53c
141
0x7f) << shift;
(gdb)
0x00000035c400d53f
141
0x7f) << shift;
(gdb)
142
shift += 7;
(gdb)
141
(gdb)
144
(gdb)
0x00000035c400d54b
144
(gdb)
970
offset = (_Unwind_Sword) utmp * fs->data_align;
(gdb)
972
(gdb)
0x00000035c400d559
972
if (UNWIND_COLUMN_IN_RANGE
(reg))
(gdb)
0x00000035c400d55b
972
(reg))
(gdb)
0x00000035c400d55f
972
(reg))
(gdb)
974
fs->regs.reg[reg].how = REG_SAVED_OFFSET;
(gdb)
975
fs->regs.reg[reg].loc.offset = offset;
72
Adam 'pi3' Zabrocki

11 November 2013
(gdb) print fs
$153 = (_Unwind_FrameState *) 0x7fffffffd460
(gdb) print *fs
$154 = {regs = {reg = {{loc = {reg = 0, offset = 0, exp = 0x0},
how = REG_SAVED_OFFSET}, {loc = {reg = 0, offset = 0,
exp = 0x0}, how = REG_UNSAVED} <repeats 17 times>},
prev = 0x0, cfa_offset = 0, cfa_reg = 0, cfa_exp = 0x0,
cfa_how = CFA_UNSET}, pc = 0x400630 <main>, personality =
0x0, data_align = 0, code_align = 0, retaddr_column = 0,
fde_encoding = 0 '\000', lsda_encoding = 255 '\377', saw_z = 0
'\000', signal_frame = 0 '\000', eh_ptr = 0x0}
(gdb) si
0x00000035c400d56c
975
fs>regs.reg[reg].loc.offset = offset;
(gdb) print offset
$155 = 0
(gdb) si
957
(gdb)
0x00000035c400d4eb
957
(gdb)
202
(gdb)
958
(context))
(gdb)
202
(gdb)
958
(context))
(gdb)
0x00000035c400d506
958
(gdb)
0x00000035c400d509
958
(gdb)
960
(gdb)
73
Adam 'pi3' Zabrocki

11 November 2013
0x00000035c400d49b
960
*insn_ptr++;
(gdb)
964
(gdb)
0x00000035c400d4a1
964
DW_CFA_advance_loc)
(gdb) x/20i $rip
=> 0x35c400d4a1 <execute_cfa_program+113>: and
$0xffffffc0,%esi
$0x40,%sil
0x35c400d4d0
$0x80,%sil
0x35c400d520
$0xc0,%sil
0x35c400d578
$0x2f,%al
0x35c400d5a9
(%rdx,%rax,4),%rax
%rdx,%rax
*%rax
0x35c400d4cb <execute_cfa_program+155>: nopl
0x0(%rax,%rax,1)
0x35c400d4d0 <execute_cfa_program+160>: and
$0x3f,%eax
0x35c400d4d3 <execute_cfa_program+163>: mov
%r12,%rbx
0x35c400d4d6 <execute_cfa_program+166>: imul
0x160(%r13),%rax
0x35c400d4de <execute_cfa_program+174>: add
%rcx,%rax
0x35c400d4e1 <execute_cfa_program+177>: mov
%rax,0x148(%r13)
0x35c400d4e8 <execute_cfa_program+184>: cmp
%r15,%rbx
0x35c400d4eb <execute_cfa_program+187>: jae
0x35c400d50b
(gdb) set $esi = 0xc8
(gdb) set $esi = 0x2f
...
gcc_assert()>
...
(gdb) si
0x00000035c400d4a4
964
DW_CFA_advance_loc)
(gdb)
0x00000035c400d4a8
964
DW_CFA_advance_loc)
(gdb)
966
74
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400d4ae
966
DW_CFA_offset)
(gdb)
978
else if ((insn & 0xc0) == DW_CFA_restore)
(gdb)
0x00000035c400d4b4
978
DW_CFA_restore)
(gdb)
985
else switch (insn)
(gdb)
0x00000035c400d4bc
985
else switch (insn)
(gdb)
0x00000035c400d4c2
985
else switch (insn)
(gdb)
0x00000035c400d4c6
985
else switch (insn)
(gdb)
0x00000035c400d4c9
985
else switch (insn)
(gdb)
960
(gdb)
0x00000035c400d5b1
960
*insn_ptr++;
(gdb)
957
(gdb) x/20i $rip
=> 0x35c400d4e8 <execute_cfa_program+184>: cmp
%r15,%rbx
0x35c400d4eb <execute_cfa_program+187>: jae
0x35c400d50b
0x35c400d4ed <execute_cfa_program+189>: mov
0xc0(%r14),%rax
0x35c400d4f4 <execute_cfa_program+196>: mov
0x148(%r13),%rcx
0x35c400d4fb <execute_cfa_program+203>: shr
$0x3f,%rax
0x35c400d4ff <execute_cfa_program+207>: add
0x98(%r14),%rax
%rax,%rcx
0x35c400d509 <execute_cfa_program+217>: jb
0x35c400d498
0x28(%rbp),%rsp
0x35c400d50f <execute_cfa_program+223>: pop
%rbx
0x35c400d510 <execute_cfa_program+224>: pop
%r12
%r13
%r14
%r15
%rbp
0x35c400d519 <execute_cfa_program+233>: retq
0x35c400d51a <execute_cfa_program+234>: nopw
0x0(%rax,%rax,1)
%rax,%rdi
%r12,%rbx
75
Adam 'pi3' Zabrocki

11 November 2013
0x35c400d526 <execute_cfa_program+246>: xor
%esi,%esi
(gdb) i r r15 rbx
r15
0x8000413f2246
140738583011910
rbx
0x7ffffffde104
140737488216324
(gdb) set $r15 = $rbx
...
<Try to jump out from the while loop because it will takie
forever ant requires fixing memory object for each interation>
...
(gdb) i r r15 rbx
r15
0x7ffffffde104
140737488216324
rbx
0x7ffffffde104
140737488216324
(gdb) si
0x00000035c400d4eb
957
(gdb)
1224 }
(gdb)
0x00000035c400d50f
1224 }
(gdb)
0x00000035c400d510
1224 }
(gdb)
0x00000035c400d512
1224 }
(gdb)
0x00000035c400d514
1224 }
(gdb)
0x00000035c400d516
1224 }
(gdb)
0x00000035c400d518
1224 }
(gdb)
0x00000035c400d519
1224 }
(gdb)
1273
(gdb)
1272
aug = (const unsigned char *) fde + sizeof (*fde);
(gdb)
size_of_encoded_value (encoding=0 '\000') at
74
if (encoding == DW_EH_PE_omit)
(gdb)
0x00000035c400eece
74
if (encoding == DW_EH_PE_omit)
(gdb)
77
switch (encoding & 0x07)
(gdb)
0x00000035c400eed7
77
(gdb)
0x00000035c400eed9
77
(gdb)
0x00000035c400eedf
77
(gdb)
77
76
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400f51b
77
(gdb)
0x00000035c400eef6
77
(gdb)
1273
(gdb)
1274
insn = NULL;
(gdb)
1275
if (fs->saw_z)
(gdb)
0x00000035c400ef09
1275
if (fs->saw_z)
(gdb)
1281
if (fs->lsda_encoding != DW_EH_PE_omit)
(gdb)
0x00000035c400ef38
1281
if (fs->lsda_encoding !=
DW_EH_PE_omit)
(gdb)
0x00000035c400ef3c
1281
if (fs->lsda_encoding !=
DW_EH_PE_omit)
(gdb)
next_fde (f=<optimized out>) at ../../../libgcc/unwind-dw2fde.h:163
163
(gdb)
1291
insn = aug;
(gdb)
1293
(gdb)
1291
insn = aug;
(gdb)
1293
(gdb)
0x00000035c400f472
1293
execute_cfa_program (insn, end,
context, fs);
(gdb)
163
(gdb)
1293
(gdb)
execute_cfa_program (insn_ptr=insn_ptr@entry=0x7ffffffde119 'A'
77
Adam 'pi3' Zabrocki

11 November 2013
insn_end=0x7ffffffde185 'A' <repeats 200 times>...,
context=context@entry=0x7fffffffd370, fs=fs@entry=0x7fffffffd460)
942 {
(gdb)
0x00000035c400d431
942 {
(gdb)
0x00000035c400d434
942 {
(gdb)
0x00000035c400d436
942 {
(gdb)
0x00000035c400d439
942 {
(gdb)
0x00000035c400d43b
942 {
(gdb)
0x00000035c400d43d
942 {
(gdb)
0x00000035c400d440
942 {
(gdb)
0x00000035c400d442
942 {
(gdb)
0x00000035c400d443
942 {
(gdb)
0x00000035c400d446
942 {
(gdb)
957
(gdb) n
946
fs->regs.prev = NULL;
(gdb)
957
(gdb)
958
(context))
(gdb)
1169
(gdb)
985
else switch (insn)
(gdb) x/20i $rip
=> 0x35c400d487 <execute_cfa_program+87>:
lea
0x5162(%rip),%rdx
# 0x35c40125f0
xor
%r8d,%r8d
mov
%rax,0x48(%rbp)
0x35c400d495 <execute_cfa_program+101>: nopl
(%rax)
0x1(%rbx),%r12
%eax,%esi
$0xffffffc0,%esi
$0x40,%sil
0x35c400d4d0
78
Adam 'pi3' Zabrocki

11 November 2013
$0x80,%sil
0x35c400d520
$0xc0,%sil
0x35c400d578
$0x2f,%al
0x35c400d5a9
(%rdx,%rax,4),%rax
%rdx,%rax
*%rax
0x35c400d4cb <execute_cfa_program+155>: nopl
0x0(%rax,%rax,1)
(gdb) x/x $rbx
(gdb) set *$rbx = 0x80 <- Lets try the same again ;)
...
gcc_assert()>
...
(gdb) si
943
struct frame_state_reg_info *unused_rs = NULL;
(gdb)
1169
(gdb)
0x00000035c400d495
1169
insn_ptr = read_sleb128
(insn_ptr, &stmp);
(gdb)
960
(gdb)
0x00000035c400d49b
960
*insn_ptr++;
(gdb)
964
(gdb)
0x00000035c400d4a1
964
DW_CFA_advance_loc)
(gdb)
0x00000035c400d4a4
964
DW_CFA_advance_loc)
(gdb)
0x00000035c400d4a8
964
DW_CFA_advance_loc)
(gdb)
966
(gdb)
0x00000035c400d4ae
966
DW_CFA_offset)
(gdb)
968
reg = insn & 0x3f;
79
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400d523
968
reg = insn & 0x3f;
(gdb)
read_uleb128 (val=<optimized out>, p=<optimized out>) at
137
result = 0;
(gdb)
execute_cfa_program (insn_ptr=0x7ffffffde11a "",
insn_ptr@entry=0x7ffffffde119 "\200",
968
reg = insn & 0x3f;
(gdb)
read_uleb128 (val=<synthetic pointer>, p=0x7ffffffde11a "") at
133
(gdb)
0x00000035c400d52d
133
(gdb)
140
byte = *p++;
(gdb)
0x00000035c400d534
140
byte = *p++;
(gdb)
141
(gdb)
0x00000035c400d53c
141
0x7f) << shift;
(gdb)
0x00000035c400d53f
141
0x7f) << shift;
(gdb)
142
shift += 7;
(gdb)
141
(gdb)
144
(gdb)
0x00000035c400d54b
144
(gdb)
execute_cfa_program (insn_ptr=0x7ffffffde11b "",
insn_ptr@entry=0x7ffffffde119 "\200",
970
offset = (_Unwind_Sword) utmp * fs->data_align;
(gdb)
972
(gdb)
0x00000035c400d559
972
(reg))
(gdb)
80
Adam 'pi3' Zabrocki

11 November 2013
0x00000035c400d55b
972
(reg))
(gdb)
0x00000035c400d55f
972
(reg))
(gdb)
974
fs->regs.reg[reg].how = REG_SAVED_OFFSET;
(gdb)
975
fs->regs.reg[reg].loc.offset = offset;
(gdb)
0x00000035c400d56c
975
fs>regs.reg[reg].loc.offset = offset;
(gdb)
957
(gdb)
0x00000035c400d4eb
957
(gdb) x/20i $rip-10
0x35c400d4e1 <execute_cfa_program+177>: mov
%rax,0x148(%r13)
0x35c400d4e8 <execute_cfa_program+184>: cmp
%r15,%rbx
=> 0x35c400d4eb <execute_cfa_program+187>: jae
0x35c400d50b
0x35c400d4ed <execute_cfa_program+189>: mov
0xc0(%r14),%rax
0x35c400d4f4 <execute_cfa_program+196>: mov
0x148(%r13),%rcx
0x35c400d4fb <execute_cfa_program+203>: shr
$0x3f,%rax
0x35c400d4ff <execute_cfa_program+207>: add
0x98(%r14),%rax
%rax,%rcx
0x35c400d509 <execute_cfa_program+217>: jb
0x35c400d498
0x28(%rbp),%rsp
0x35c400d50f <execute_cfa_program+223>: pop
%rbx
%r12
%r13
%r14
%r15
%rbp
0x35c400d519 <execute_cfa_program+233>: retq
0x35c400d51a <execute_cfa_program+234>: nopw
0x0(%rax,%rax,1)
%rax,%rdi
%r12,%rbx
(gdb) i r r15 rbx
r15
0x7ffffffde185
140737488216453
rbx
0x7ffffffde11b
140737488216347
(gdb) set $r15 = $rbx
(gdb) set $rip = 0x35c400d4e8
...
81
Adam 'pi3' Zabrocki

11 November 2013
<fix while loop again <- its 2nd entry to this function (for
upper FDE)>
...
(gdb) si
0x00000035c400d4eb
957
(gdb)
1224 }
(gdb)
0x00000035c400d50f
1224 }
(gdb)
0x00000035c400d510
1224 }
(gdb)
0x00000035c400d512
1224 }
(gdb)
0x00000035c400d514
1224 }
(gdb)
0x00000035c400d516
1224 }
(gdb)
0x00000035c400d518
1224 }
(gdb)
0x00000035c400d519
1224 }
(gdb)
1296 }
(gdb) list
1291
insn = aug;
1292
1293
1294
1295
1296 }
1297 ^L
1298 typedef struct frame_state
1299 {
1300
void *cfa;
(gdb)
1301
void *eh_ptr;
1302
long cfa_offset;
1303
long args_size;
1304
long reg_or_offset[PRE_GCC3_DWARF_FRAME_REGISTERS+1];
1305
unsigned short cfa_reg;
1306
unsigned short retaddr_column;
1307
char saved[PRE_GCC3_DWARF_FRAME_REGISTERS+1];
1308 } frame_state;
1309
1310 struct frame_state * __frame_state_for (void *, struct
frame_state *);
(gdb) si
1295
(gdb)
1296 }
82
Adam 'pi3' Zabrocki

11 November 2013
(gdb)
0x00000035c400f486
1296 }
(gdb)
0x00000035c400f487
1296 }
(gdb)
0x00000035c400f489
1296 }
(gdb)
0x00000035c400f48b
1296 }
(gdb)
0x00000035c400f48d
1296 }
(gdb)
0x00000035c400f48f
1296 }
(gdb)
_Unwind_Backtrace (trace=0x35c1909bc0 <backtrace_helper>,
trace_argument=0x7fffffffd620) at ../../../libgcc/unwind.inc:291
291
_URC_END_OF_STACK)
(gdb)
290
(gdb) n
291
_URC_END_OF_STACK)
(gdb) print code
(gdb) n
295
if ((*trace) (&context, trace_argument) !=
_URC_NO_REASON)
(gdb)
299
(gdb) print code
(gdb) n
303
(gdb) si
0x00000035c400ff06
303
&fs);
(gdb) n
56
(gdb) bt
#0 0x00000035c1835a19 in __GI_raise (sig=sig@entry=6) at
#2 0x00000035c400e8d5 in uw_update_context_1
83
Adam 'pi3' Zabrocki

11 November 2013
#3 0x00000035c400eb61 in uw_update_context
#4 0x00000035c400ff0e in _Unwind_Backtrace (trace=0x35c1909bc0
<backtrace_helper>, trace_argument=0x7fffffffd620)
(array=array@entry=0x7fffffffd800, size=size@entry=64)
fortify_fail.c:31
stack_chk_fail.c:28
#9 0x00000000004006b1 in main ()
(gdb) up
90
raise (SIGABRT);
(gdb)
#2 0x00000035c400e8d5 in uw_update_context_1
1430
gcc_unreachable (); <- one of the object wasnt
patched ;)
(gdb) print fs->regs.cfa_how
$159 = CFA_UNSET
(gdb) print *fs
$160 = {regs = {reg = {{loc = {reg = 0, offset = 0, exp = 0x0},
how = REG_SAVED_OFFSET}, {loc = {reg = 0, offset = 0,
exp = 0x0}, how = REG_UNSAVED} <repeats 17 times>},
prev = 0x0, cfa_offset = 0, cfa_reg = 0, cfa_exp = 0x0,
cfa_how = CFA_UNSET}, pc = 0x400630 <main>, personality =
0x0, data_align = 0, code_align = 0, retaddr_column = 0,
fde_encoding = 0 '\000', lsda_encoding = 255 '\377', saw_z = 0
'\000', signal_frame = 0 '\000', eh_ptr = 0x0}
84
Adam 'pi3' Zabrocki

11 November 2013
Somehow security related:
Scenario 1 force SSP to recopy existing processs memory region to the newly allocated one. It is very
difficult because to hit vulnerable code, you need to fix all possible crashes which I've described before
(and reproduced). This mean fixing environment block, pointers to argument etc. I've switched off ASLR
+ did some memory patching just to monitor control flow and to see possible memory exhaustion. In
my scenario I've allocated 2GB of memory where I've pointed as pointer to the program's name.

done.
done.
(gdb) b libc_fatal.c:66
(gdb) c
Continuing.
66
(gdb) ni
57
va_start (ap, fmt);
(gdb)
0x00000035c1875a45 57
va_start (ap, fmt);
(gdb)
0x00000035c1875a4c 57
va_start (ap, fmt);
(gdb)
0x00000035c1875a56 57
va_start (ap, fmt);
(gdb)
52 {
(gdb)
57
va_start (ap, fmt);
(gdb)
58
(gdb)
0x00000035c1875a75 58
(gdb)
85
Adam 'pi3' Zabrocki

11 November 2013
0x00000035c1875a7c 58
(gdb)
0x00000035c1875a80 58
(gdb)
0x00000035c1875a87 58
(gdb)
0x00000035c1875a8e 58
(gdb)
66
(gdb)
67
if (on_2 == NULL || *on_2 == '\0')
(gdb) si
0x00000035c1875a9d 67
if (on_2 == NULL || *on_2 == '\0')
(gdb) x/20i 0x00000035c1875a9d
=> 0x35c1875a9d <__libc_message+205>:
je
0x35c1875aa8
<__libc_message+216>
0x35c1875a9f <__libc_message+207>:
cmpb
$0x0,(%rax)
0x35c1875aa2 <__libc_message+210>:
jne
0x35c1875c68
0x35c1875aa8 <__libc_message+216>:
lea
0x1061c6(%rip),%rdi
# 0x35c197bc75
0x35c1875aaf <__libc_message+223>:
xor
%eax,%eax
0x35c1875ab1 <__libc_message+225>:
mov
$0x902,%esi
0x35c1875ab6 <__libc_message+230>:
callq 0x35c18e7459
<__open_nocancel>
0x35c1875abb <__libc_message+235>:
cmp
$0xffffffff,%eax
0x35c1875abe <__libc_message+238>:
mov
%eax,-0x718(%rbp)
0x35c1875ac4 <__libc_message+244>:
je
0x35c1875c68
0x35c1875aca <__libc_message+250>:
mov
-0x720(%rbp),%rax
0x35c1875ad1 <__libc_message+257>:
xor
%r14d,%r14d
xor
%r13d,%r13d
movzbl (%rax),%r12d
0x35c1875adb <__libc_message+267>:
mov
%rax,%rbx
0x35c1875ade <__libc_message+270>:
test
%r12b,%r12b
0x35c1875ae1 <__libc_message+273>:
je
0x35c1875c8e
0x35c1875ae7 <__libc_message+279>:
nopw
0x0(%rax,%rax,1)
0x35c1875af0 <__libc_message+288>:
mov
%r12d,%edx
0x35c1875af3 <__libc_message+291>:
mov
%rbx,%rax
(gdb) i r rax
rax
0x7fff8b4a4fe0
140735530291168
(gdb) x/x $rax
0x7fff8b4a4fe0: 0x706d742f
(gdb) si
0x00000035c1875a9f 67
if (on_2 == NULL || *on_2 == '\0')
(gdb) si
0x00000035c1875aa2 67
if (on_2 == NULL || *on_2 == '\0')
(gdb) x/10i 0x35c1875c68
0x35c1875c68 <__libc_message+664>:
movl
$0x2,-0x718(%rbp)
86
Adam 'pi3' Zabrocki

11 November 2013
jmpq
0x35c1875aca
mov
-0x708(%rbp),%rax
0x35c1875c7e <__libc_message+686>:
lea
0x8(%rax),%rdx
mov
%rdx,-0x708(%rbp)
jmpq
0x35c1875b8f
0x35c1875c8e <__libc_message+702>:
mov
-0x720(%rbp),%rsi
lea
-0x6f8(%rbp),%rdx
0x35c1875c9c <__libc_message+716>:
mov
$0x3,%edi
0x35c1875ca1 <__libc_message+721>:
callq 0x35c18f0430
<__vsyslog>
(gdb) i r rip
rip
0x35c1875c68 0x35c1875c68 <__libc_message+664>
(gdb) bt
#0 __libc_message (do_abort=do_abort@entry=2,
fortify_fail.c:31
stack_chk_fail.c:28
#3 0x000000000040075a in main (argc=2, argv=0x7fff8b4a32e8) at
test.c:15
...
...
<bla bla bla>
...
...
(gdb) c
Continuing.
95
len = strlen (str);
(gdb) print str
$5 = 0x35c197d2ea "stack smashing detected"
(gdb) c
Continuing.
106
newp->str = str;
...
...
<bla bla bla>
...
...
(gdb) ni
87
Adam 'pi3' Zabrocki

11 November 2013
0x00000035c1875b9a
(gdb)
96
cp += 2;
(gdb) print len
$11 = -2147483647
95
len = strlen (str);
For some reason my gdb has small bug here:

(gdb) print len
$11 = -2147483647
Program was correctly compiled with debug symbols (flags: "-ggdb" and "g") but gdb could find correct
definition of "len" variable and dumped it as integer (which is the default type in case when gdb dont
know the type).
Greetings:
Mateusz 'j00ru' Jurczyk, Gynvael Coldwind for reviewing and great confrontation (discussion) of ideas.
Rafa 'nergal' Wojtczuk for great discussion and helping with DWARF.
Brian Pak for great research and at least for reading this paper ;)
sergio and #trololol for motivation and at least for reading this paper ;)
References:
1.
2.
3.
4.
5.
http://www.phrack.org/issues.html?issue=67&id=13&mode=txt
http://xorl.wordpress.com/2010/10/14/linux-glibc-stack-canary-values/
http://www.bpak.org/blog/2014/02/codegate-2014-membership-800pt-pwnable-write-up/
https://www.usenix.org/legacy/event/woot11/tech/final_files/Oakley.pdf
http://dwarfstd.org/
Best regards,
Adam 'pi3' Zabrocki
http://pi3.com.pl
88

Adventure With Stack Smashing Protector

Uploaded by

Copyright:

Available Formats

Adventure With Stack Smashing Protector

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Adventure With Stack Smashing Protector

Uploaded by

Copyright:

Available Formats

Adam 'pi3' Zabrocki