kernel_optimize_test/arch/parisc/kernel/ptrace.c
Helge Deller ed79b86d8a parisc: fix bug in compat_arch_ptrace
Commit 81e192d6ce ("parisc: convert to
generic compat_sys_ptrace") introduced a bug which segfaults the parisc
64bit kernel when stracing 32bit applications:

  Kernel Fault: Code=15 regs=00000000bafa42b0 (Addr=00000001baf5ab57)
       YZrvWESTHLNXBCVMcbcbcbcbOGFRQPDI
  PSW: 00001000000001101111111100001011 Tainted: G        W
  r00-03  000000ff0806ff0b 000000004068edc0 00000000401203f8 00000000fb3e2508
  r04-07  0000000040686dc0 00000000baf5a800 fffffffffffffffc fffffffffb3e2508
  r08-11  00000000baf5a800 000000000004b068 00000000000402b0 0000000000040d68
  r12-15  0000000000042a9c 0000000000040a9c 0000000000040d60 0000000000042e9c
  r16-19  000000000004b060 000000000004b058 0000000000042d9c ffffffffffffffff
  r20-23  000000000800000b 0000000000000000 000000000800000b fffffffffb3e2508
  r24-27  00000000fffffffc 0000000000000003 00000000fffffffc 0000000040686dc0
  r28-31  00000001baf5a7ff 00000000bafa4280 00000000bafa42b0 00000000000001d7
  sr00-03  0000000000fca000 0000000000000000 0000000000000000 0000000000fca000
  sr04-07  0000000000000000 0000000000000000 0000000000000000 0000000000000000

  IASQ: 0000000000000000 0000000000000000 IAOQ: 0000000040120400 0000000040120404
   IIR: 4b9a06b0    ISR: 0000000000000000  IOR: 00000001baf5ab57
   CPU:        0   CR30: 00000000bafa4000 CR31: 00000000d22344e0
   ORIG_R28: 00000000fb3e2248
   IAOQ[0]: compat_arch_ptrace+0xb8/0x160
   IAOQ[1]: compat_arch_ptrace+0xbc/0x160
   RP(r2): compat_arch_ptrace+0xb0/0x160
  Backtrace:
   [<00000000401612ac>] compat_sys_ptrace+0x15c/0x180
   [<0000000040104ef8>] syscall_exit+0x0/0x14

The problem is that compat_arch_ptrace() enters with an addr value of
type compat_ulong_t and calls translate_usr_offset() to translate the
address offset into a struct pt_regs offset like this:

	addr = translate_usr_offset(addr)

this means that any return value of translate_usr_offset() is stored
back as compat_ulong_t type into the addr variable.

But since translate_usr_offset() returns -1 for invalid offsets, addr
can now get the value 0xffffffff which then fails the next return-value
sanity check and thus the kernel tries to access invalid memory:

	if (addr < 0)
		break;

Fix this bug by modifying translate_usr_offset() to take and return
values of type compat_ulong_t, and by returning the value
"sizeof(struct pt_regs)" as an error indicator.

Additionally change the sanity check to check for return values
for >= sizeof(struct pt_regs).

This patch survived my compile and run-tests.

Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-11-20 13:09:17 -08:00

285 lines
8.0 KiB
C

/*
* Kernel support for the ptrace() and syscall tracing interfaces.
*
* Copyright (C) 2000 Hewlett-Packard Co, Linuxcare Inc.
* Copyright (C) 2000 Matthew Wilcox <matthew@wil.cx>
* Copyright (C) 2000 David Huggins-Daines <dhd@debian.org>
* Copyright (C) 2008 Helge Deller <deller@gmx.de>
*/
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/mm.h>
#include <linux/smp.h>
#include <linux/errno.h>
#include <linux/ptrace.h>
#include <linux/user.h>
#include <linux/personality.h>
#include <linux/security.h>
#include <linux/compat.h>
#include <linux/signal.h>
#include <asm/uaccess.h>
#include <asm/pgtable.h>
#include <asm/system.h>
#include <asm/processor.h>
#include <asm/asm-offsets.h>
/* PSW bits we allow the debugger to modify */
#define USER_PSW_BITS (PSW_N | PSW_V | PSW_CB)
/*
* Called by kernel/ptrace.c when detaching..
*
* Make sure single step bits etc are not set.
*/
void ptrace_disable(struct task_struct *task)
{
task->ptrace &= ~(PT_SINGLESTEP|PT_BLOCKSTEP);
/* make sure the trap bits are not set */
pa_psw(task)->r = 0;
pa_psw(task)->t = 0;
pa_psw(task)->h = 0;
pa_psw(task)->l = 0;
}
/*
* The following functions are called by ptrace_resume() when
* enabling or disabling single/block tracing.
*/
void user_disable_single_step(struct task_struct *task)
{
ptrace_disable(task);
}
void user_enable_single_step(struct task_struct *task)
{
task->ptrace &= ~PT_BLOCKSTEP;
task->ptrace |= PT_SINGLESTEP;
if (pa_psw(task)->n) {
struct siginfo si;
/* Nullified, just crank over the queue. */
task_regs(task)->iaoq[0] = task_regs(task)->iaoq[1];
task_regs(task)->iasq[0] = task_regs(task)->iasq[1];
task_regs(task)->iaoq[1] = task_regs(task)->iaoq[0] + 4;
pa_psw(task)->n = 0;
pa_psw(task)->x = 0;
pa_psw(task)->y = 0;
pa_psw(task)->z = 0;
pa_psw(task)->b = 0;
ptrace_disable(task);
/* Don't wake up the task, but let the
parent know something happened. */
si.si_code = TRAP_TRACE;
si.si_addr = (void __user *) (task_regs(task)->iaoq[0] & ~3);
si.si_signo = SIGTRAP;
si.si_errno = 0;
force_sig_info(SIGTRAP, &si, task);
/* notify_parent(task, SIGCHLD); */
return;
}
/* Enable recovery counter traps. The recovery counter
* itself will be set to zero on a task switch. If the
* task is suspended on a syscall then the syscall return
* path will overwrite the recovery counter with a suitable
* value such that it traps once back in user space. We
* disable interrupts in the tasks PSW here also, to avoid
* interrupts while the recovery counter is decrementing.
*/
pa_psw(task)->r = 1;
pa_psw(task)->t = 0;
pa_psw(task)->h = 0;
pa_psw(task)->l = 0;
}
void user_enable_block_step(struct task_struct *task)
{
task->ptrace &= ~PT_SINGLESTEP;
task->ptrace |= PT_BLOCKSTEP;
/* Enable taken branch trap. */
pa_psw(task)->r = 0;
pa_psw(task)->t = 1;
pa_psw(task)->h = 0;
pa_psw(task)->l = 0;
}
long arch_ptrace(struct task_struct *child, long request, long addr, long data)
{
unsigned long tmp;
long ret = -EIO;
switch (request) {
/* Read the word at location addr in the USER area. For ptraced
processes, the kernel saves all regs on a syscall. */
case PTRACE_PEEKUSR:
if ((addr & (sizeof(long)-1)) ||
(unsigned long) addr >= sizeof(struct pt_regs))
break;
tmp = *(unsigned long *) ((char *) task_regs(child) + addr);
ret = put_user(tmp, (unsigned long *) data);
break;
/* Write the word at location addr in the USER area. This will need
to change when the kernel no longer saves all regs on a syscall.
FIXME. There is a problem at the moment in that r3-r18 are only
saved if the process is ptraced on syscall entry, and even then
those values are overwritten by actual register values on syscall
exit. */
case PTRACE_POKEUSR:
/* Some register values written here may be ignored in
* entry.S:syscall_restore_rfi; e.g. iaoq is written with
* r31/r31+4, and not with the values in pt_regs.
*/
if (addr == PT_PSW) {
/* Allow writing to Nullify, Divide-step-correction,
* and carry/borrow bits.
* BEWARE, if you set N, and then single step, it won't
* stop on the nullified instruction.
*/
data &= USER_PSW_BITS;
task_regs(child)->gr[0] &= ~USER_PSW_BITS;
task_regs(child)->gr[0] |= data;
ret = 0;
break;
}
if ((addr & (sizeof(long)-1)) ||
(unsigned long) addr >= sizeof(struct pt_regs))
break;
if ((addr >= PT_GR1 && addr <= PT_GR31) ||
addr == PT_IAOQ0 || addr == PT_IAOQ1 ||
(addr >= PT_FR0 && addr <= PT_FR31 + 4) ||
addr == PT_SAR) {
*(unsigned long *) ((char *) task_regs(child) + addr) = data;
ret = 0;
}
break;
default:
ret = ptrace_request(child, request, addr, data);
break;
}
return ret;
}
#ifdef CONFIG_COMPAT
/* This function is needed to translate 32 bit pt_regs offsets in to
* 64 bit pt_regs offsets. For example, a 32 bit gdb under a 64 bit kernel
* will request offset 12 if it wants gr3, but the lower 32 bits of
* the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4).
* This code relies on a 32 bit pt_regs being comprised of 32 bit values
* except for the fp registers which (a) are 64 bits, and (b) follow
* the gr registers at the start of pt_regs. The 32 bit pt_regs should
* be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[]
* being 64 bit in both cases.
*/
static compat_ulong_t translate_usr_offset(compat_ulong_t offset)
{
if (offset < 0)
return sizeof(struct pt_regs);
else if (offset <= 32*4) /* gr[0..31] */
return offset * 2 + 4;
else if (offset <= 32*4+32*8) /* gr[0..31] + fr[0..31] */
return offset + 32*4;
else if (offset < sizeof(struct pt_regs)/2 + 32*4)
return offset * 2 + 4 - 32*8;
else
return sizeof(struct pt_regs);
}
long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
compat_ulong_t addr, compat_ulong_t data)
{
compat_uint_t tmp;
long ret = -EIO;
switch (request) {
case PTRACE_PEEKUSR:
if (addr & (sizeof(compat_uint_t)-1))
break;
addr = translate_usr_offset(addr);
if (addr >= sizeof(struct pt_regs))
break;
tmp = *(compat_uint_t *) ((char *) task_regs(child) + addr);
ret = put_user(tmp, (compat_uint_t *) (unsigned long) data);
break;
/* Write the word at location addr in the USER area. This will need
to change when the kernel no longer saves all regs on a syscall.
FIXME. There is a problem at the moment in that r3-r18 are only
saved if the process is ptraced on syscall entry, and even then
those values are overwritten by actual register values on syscall
exit. */
case PTRACE_POKEUSR:
/* Some register values written here may be ignored in
* entry.S:syscall_restore_rfi; e.g. iaoq is written with
* r31/r31+4, and not with the values in pt_regs.
*/
if (addr == PT_PSW) {
/* Since PT_PSW==0, it is valid for 32 bit processes
* under 64 bit kernels as well.
*/
ret = arch_ptrace(child, request, addr, data);
} else {
if (addr & (sizeof(compat_uint_t)-1))
break;
addr = translate_usr_offset(addr);
if (addr >= sizeof(struct pt_regs))
break;
if (addr >= PT_FR0 && addr <= PT_FR31 + 4) {
/* Special case, fp regs are 64 bits anyway */
*(__u64 *) ((char *) task_regs(child) + addr) = data;
ret = 0;
}
else if ((addr >= PT_GR1+4 && addr <= PT_GR31+4) ||
addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4 ||
addr == PT_SAR+4) {
/* Zero the top 32 bits */
*(__u32 *) ((char *) task_regs(child) + addr - 4) = 0;
*(__u32 *) ((char *) task_regs(child) + addr) = data;
ret = 0;
}
}
break;
default:
ret = compat_ptrace_request(child, request, addr, data);
break;
}
return ret;
}
#endif
void syscall_trace(void)
{
if (!test_thread_flag(TIF_SYSCALL_TRACE))
return;
if (!(current->ptrace & PT_PTRACED))
return;
ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
? 0x80 : 0));
/*
* this isn't the same as continuing with a signal, but it will do
* for normal use. strace only continues with a signal if the
* stopping signal is not SIGTRAP. -brl
*/
if (current->exit_code) {
send_sig(current->exit_code, current, 1);
current->exit_code = 0;
}
}