2 * Kernel support for the ptrace() and syscall tracing interfaces.
4 * Copyright (C) 2000 Hewlett-Packard Co, Linuxcare Inc.
5 * Copyright (C) 2000 Matthew Wilcox <matthew@wil.cx>
6 * Copyright (C) 2000 David Huggins-Daines <dhd@debian.org>
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
12 #include <linux/smp.h>
13 #include <linux/smp_lock.h>
14 #include <linux/errno.h>
15 #include <linux/ptrace.h>
16 #include <linux/user.h>
17 #include <linux/personality.h>
18 #include <linux/security.h>
19 #include <linux/compat.h>
20 #include <linux/signal.h>
22 #include <asm/uaccess.h>
23 #include <asm/pgtable.h>
24 #include <asm/system.h>
25 #include <asm/processor.h>
26 #include <asm/offsets.h>
28 /* PSW bits we allow the debugger to modify */
29 #define USER_PSW_BITS (PSW_N | PSW_V | PSW_CB)
34 #define DBG(x...) printk(x)
41 /* This function is needed to translate 32 bit pt_regs offsets in to
42 * 64 bit pt_regs offsets. For example, a 32 bit gdb under a 64 bit kernel
43 * will request offset 12 if it wants gr3, but the lower 32 bits of
44 * the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4).
45 * This code relies on a 32 bit pt_regs being comprised of 32 bit values
46 * except for the fp registers which (a) are 64 bits, and (b) follow
47 * the gr registers at the start of pt_regs. The 32 bit pt_regs should
48 * be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[]
49 * being 64 bit in both cases.
52 static long translate_usr_offset(long offset
)
56 else if (offset
<= 32*4) /* gr[0..31] */
57 return offset
* 2 + 4;
58 else if (offset
<= 32*4+32*8) /* gr[0..31] + fr[0..31] */
60 else if (offset
< sizeof(struct pt_regs
)/2 + 32*4)
61 return offset
* 2 + 4 - 32*8;
68 * Called by kernel/ptrace.c when detaching..
70 * Make sure single step bits etc are not set.
72 void ptrace_disable(struct task_struct
*child
)
74 /* make sure the trap bits are not set */
81 long sys_ptrace(long request
, pid_t pid
, long addr
, long data
)
83 struct task_struct
*child
;
86 long oaddr
=addr
, odata
=data
;
91 if (request
== PTRACE_TRACEME
) {
92 /* are we already being traced? */
93 if (current
->ptrace
& PT_PTRACED
)
96 ret
= security_ptrace(current
->parent
, current
);
100 /* set the ptrace bit in the process flags. */
101 current
->ptrace
|= PT_PTRACED
;
107 read_lock(&tasklist_lock
);
108 child
= find_task_by_pid(pid
);
110 get_task_struct(child
);
111 read_unlock(&tasklist_lock
);
115 if (pid
== 1) /* no messing around with init! */
118 if (request
== PTRACE_ATTACH
) {
119 ret
= ptrace_attach(child
);
123 ret
= ptrace_check_attach(child
, request
== PTRACE_KILL
);
128 case PTRACE_PEEKTEXT
: /* read word at location addr. */
129 case PTRACE_PEEKDATA
: {
133 if (is_compat_task(child
)) {
137 copied
= access_process_vm(child
, addr
, &tmp
, sizeof(tmp
), 0);
139 if (copied
!= sizeof(tmp
))
141 ret
= put_user(tmp
,(unsigned int *) data
);
142 DBG("sys_ptrace(PEEK%s, %d, %lx, %lx) returning %ld, data %x\n",
143 request
== PTRACE_PEEKTEXT
? "TEXT" : "DATA",
144 pid
, oaddr
, odata
, ret
, tmp
);
151 copied
= access_process_vm(child
, addr
, &tmp
, sizeof(tmp
), 0);
153 if (copied
!= sizeof(tmp
))
155 ret
= put_user(tmp
,(unsigned long *) data
);
160 /* when I and D space are separate, this will have to be fixed. */
161 case PTRACE_POKETEXT
: /* write the word at location addr. */
162 case PTRACE_POKEDATA
:
165 if (is_compat_task(child
)) {
166 unsigned int tmp
= (unsigned int)data
;
167 DBG("sys_ptrace(POKE%s, %d, %lx, %lx)\n",
168 request
== PTRACE_POKETEXT
? "TEXT" : "DATA",
171 if (access_process_vm(child
, addr
, &tmp
, sizeof(tmp
), 1) == sizeof(tmp
))
177 if (access_process_vm(child
, addr
, &data
, sizeof(data
), 1) == sizeof(data
))
183 /* Read the word at location addr in the USER area. For ptraced
184 processes, the kernel saves all regs on a syscall. */
185 case PTRACE_PEEKUSR
: {
188 if (is_compat_task(child
)) {
191 if (addr
& (sizeof(int)-1))
193 if ((addr
= translate_usr_offset(addr
)) < 0)
196 tmp
= *(unsigned int *) ((char *) task_regs(child
) + addr
);
197 ret
= put_user(tmp
, (unsigned int *) data
);
198 DBG("sys_ptrace(PEEKUSR, %d, %lx, %lx) returning %ld, addr %lx, data %x\n",
199 pid
, oaddr
, odata
, ret
, addr
, tmp
);
206 if ((addr
& (sizeof(long)-1)) || (unsigned long) addr
>= sizeof(struct pt_regs
))
208 tmp
= *(unsigned long *) ((char *) task_regs(child
) + addr
);
209 ret
= put_user(tmp
, (unsigned long *) data
);
214 /* Write the word at location addr in the USER area. This will need
215 to change when the kernel no longer saves all regs on a syscall.
216 FIXME. There is a problem at the moment in that r3-r18 are only
217 saved if the process is ptraced on syscall entry, and even then
218 those values are overwritten by actual register values on syscall
222 /* Some register values written here may be ignored in
223 * entry.S:syscall_restore_rfi; e.g. iaoq is written with
224 * r31/r31+4, and not with the values in pt_regs.
226 /* PT_PSW=0, so this is valid for 32 bit processes under 64
229 if (addr
== PT_PSW
) {
230 /* PT_PSW=0, so this is valid for 32 bit processes
231 * under 64 bit kernels.
233 * Allow writing to Nullify, Divide-step-correction,
234 * and carry/borrow bits.
235 * BEWARE, if you set N, and then single step, it won't
236 * stop on the nullified instruction.
238 DBG("sys_ptrace(POKEUSR, %d, %lx, %lx)\n",
240 data
&= USER_PSW_BITS
;
241 task_regs(child
)->gr
[0] &= ~USER_PSW_BITS
;
242 task_regs(child
)->gr
[0] |= data
;
247 if (is_compat_task(child
)) {
248 if (addr
& (sizeof(int)-1))
250 if ((addr
= translate_usr_offset(addr
)) < 0)
252 DBG("sys_ptrace(POKEUSR, %d, %lx, %lx) addr %lx\n",
253 pid
, oaddr
, odata
, addr
);
254 if (addr
>= PT_FR0
&& addr
<= PT_FR31
+ 4) {
255 /* Special case, fp regs are 64 bits anyway */
256 *(unsigned int *) ((char *) task_regs(child
) + addr
) = data
;
259 else if ((addr
>= PT_GR1
+4 && addr
<= PT_GR31
+4) ||
260 addr
== PT_IAOQ0
+4 || addr
== PT_IAOQ1
+4 ||
262 /* Zero the top 32 bits */
263 *(unsigned int *) ((char *) task_regs(child
) + addr
- 4) = 0;
264 *(unsigned int *) ((char *) task_regs(child
) + addr
) = data
;
272 if ((addr
& (sizeof(long)-1)) || (unsigned long) addr
>= sizeof(struct pt_regs
))
274 if ((addr
>= PT_GR1
&& addr
<= PT_GR31
) ||
275 addr
== PT_IAOQ0
|| addr
== PT_IAOQ1
||
276 (addr
>= PT_FR0
&& addr
<= PT_FR31
+ 4) ||
278 *(unsigned long *) ((char *) task_regs(child
) + addr
) = data
;
284 case PTRACE_SYSCALL
: /* continue and stop at next (return from) syscall */
287 DBG("sys_ptrace(%s)\n",
288 request
== PTRACE_SYSCALL
? "SYSCALL" : "CONT");
289 if (!valid_signal(data
))
291 child
->ptrace
&= ~(PT_SINGLESTEP
|PT_BLOCKSTEP
);
292 if (request
== PTRACE_SYSCALL
) {
293 set_tsk_thread_flag(child
, TIF_SYSCALL_TRACE
);
295 clear_tsk_thread_flag(child
, TIF_SYSCALL_TRACE
);
297 child
->exit_code
= data
;
298 goto out_wake_notrap
;
302 * make the child exit. Best I can do is send it a
303 * sigkill. perhaps it should be put in the status
304 * that it wants to exit.
306 DBG("sys_ptrace(KILL)\n");
307 if (child
->exit_state
== EXIT_ZOMBIE
) /* already dead */
309 child
->exit_code
= SIGKILL
;
310 goto out_wake_notrap
;
312 case PTRACE_SINGLEBLOCK
:
313 DBG("sys_ptrace(SINGLEBLOCK)\n");
315 if (!valid_signal(data
))
317 clear_tsk_thread_flag(child
, TIF_SYSCALL_TRACE
);
318 child
->ptrace
&= ~PT_SINGLESTEP
;
319 child
->ptrace
|= PT_BLOCKSTEP
;
320 child
->exit_code
= data
;
322 /* Enable taken branch trap. */
323 pa_psw(child
)->r
= 0;
324 pa_psw(child
)->t
= 1;
325 pa_psw(child
)->h
= 0;
326 pa_psw(child
)->l
= 0;
329 case PTRACE_SINGLESTEP
:
330 DBG("sys_ptrace(SINGLESTEP)\n");
332 if (!valid_signal(data
))
335 clear_tsk_thread_flag(child
, TIF_SYSCALL_TRACE
);
336 child
->ptrace
&= ~PT_BLOCKSTEP
;
337 child
->ptrace
|= PT_SINGLESTEP
;
338 child
->exit_code
= data
;
340 if (pa_psw(child
)->n
) {
343 /* Nullified, just crank over the queue. */
344 task_regs(child
)->iaoq
[0] = task_regs(child
)->iaoq
[1];
345 task_regs(child
)->iasq
[0] = task_regs(child
)->iasq
[1];
346 task_regs(child
)->iaoq
[1] = task_regs(child
)->iaoq
[0] + 4;
347 pa_psw(child
)->n
= 0;
348 pa_psw(child
)->x
= 0;
349 pa_psw(child
)->y
= 0;
350 pa_psw(child
)->z
= 0;
351 pa_psw(child
)->b
= 0;
352 ptrace_disable(child
);
353 /* Don't wake up the child, but let the
354 parent know something happened. */
355 si
.si_code
= TRAP_TRACE
;
356 si
.si_addr
= (void __user
*) (task_regs(child
)->iaoq
[0] & ~3);
357 si
.si_signo
= SIGTRAP
;
359 force_sig_info(SIGTRAP
, &si
, child
);
360 //notify_parent(child, SIGCHLD);
365 /* Enable recovery counter traps. The recovery counter
366 * itself will be set to zero on a task switch. If the
367 * task is suspended on a syscall then the syscall return
368 * path will overwrite the recovery counter with a suitable
369 * value such that it traps once back in user space. We
370 * disable interrupts in the childs PSW here also, to avoid
371 * interrupts while the recovery counter is decrementing.
373 pa_psw(child
)->r
= 1;
374 pa_psw(child
)->t
= 0;
375 pa_psw(child
)->h
= 0;
376 pa_psw(child
)->l
= 0;
377 /* give it a chance to run. */
381 ret
= ptrace_detach(child
, data
);
384 case PTRACE_GETEVENTMSG
:
385 ret
= put_user(child
->ptrace_message
, (unsigned int __user
*) data
);
389 ret
= ptrace_request(child
, request
, addr
, data
);
394 ptrace_disable(child
);
396 wake_up_process(child
);
399 put_task_struct(child
);
402 DBG("sys_ptrace(%ld, %d, %lx, %lx) returning %ld\n",
403 request
, pid
, oaddr
, odata
, ret
);
407 void syscall_trace(void)
409 if (!test_thread_flag(TIF_SYSCALL_TRACE
))
411 if (!(current
->ptrace
& PT_PTRACED
))
413 ptrace_notify(SIGTRAP
| ((current
->ptrace
& PT_TRACESYSGOOD
)
416 * this isn't the same as continuing with a signal, but it will do
417 * for normal use. strace only continues with a signal if the
418 * stopping signal is not SIGTRAP. -brl
420 if (current
->exit_code
) {
421 send_sig(current
->exit_code
, current
, 1);
422 current
->exit_code
= 0;