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>
21 #include <asm/uaccess.h>
22 #include <asm/pgtable.h>
23 #include <asm/system.h>
24 #include <asm/processor.h>
25 #include <asm/offsets.h>
27 /* PSW bits we allow the debugger to modify */
28 #define USER_PSW_BITS (PSW_N | PSW_V | PSW_CB)
33 #define DBG(x...) printk(x)
40 /* This function is needed to translate 32 bit pt_regs offsets in to
41 * 64 bit pt_regs offsets. For example, a 32 bit gdb under a 64 bit kernel
42 * will request offset 12 if it wants gr3, but the lower 32 bits of
43 * the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4).
44 * This code relies on a 32 bit pt_regs being comprised of 32 bit values
45 * except for the fp registers which (a) are 64 bits, and (b) follow
46 * the gr registers at the start of pt_regs. The 32 bit pt_regs should
47 * be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[]
48 * being 64 bit in both cases.
51 static long translate_usr_offset(long offset)
55 else if (offset <= 32*4) /* gr[0..31] */
56 return offset * 2 + 4;
57 else if (offset <= 32*4+32*8) /* gr[0..31] + fr[0..31] */
59 else if (offset < sizeof(struct pt_regs)/2 + 32*4)
60 return offset * 2 + 4 - 32*8;
67 * Called by kernel/ptrace.c when detaching..
69 * Make sure single step bits etc are not set.
71 void ptrace_disable(struct task_struct *child)
73 /* make sure the trap bits are not set */
80 long sys_ptrace(long request, pid_t pid, long addr, long data)
82 struct task_struct *child;
85 long oaddr=addr, odata=data;
90 if (request == PTRACE_TRACEME) {
91 /* are we already being traced? */
92 if (current->ptrace & PT_PTRACED)
95 ret = security_ptrace(current->parent, current);
99 /* set the ptrace bit in the process flags. */
100 current->ptrace |= PT_PTRACED;
106 read_lock(&tasklist_lock);
107 child = find_task_by_pid(pid);
109 get_task_struct(child);
110 read_unlock(&tasklist_lock);
113 if (!vx_check(vx_task_xid(child), VX_WATCH|VX_IDENT))
117 if (pid == 1) /* no messing around with init! */
120 if (request == PTRACE_ATTACH) {
121 ret = ptrace_attach(child);
125 ret = ptrace_check_attach(child, request == PTRACE_KILL);
130 case PTRACE_PEEKTEXT: /* read word at location addr. */
131 case PTRACE_PEEKDATA: {
135 if (is_compat_task(child)) {
139 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
141 if (copied != sizeof(tmp))
143 ret = put_user(tmp,(unsigned int *) data);
144 DBG("sys_ptrace(PEEK%s, %d, %lx, %lx) returning %ld, data %x\n",
145 request == PTRACE_PEEKTEXT ? "TEXT" : "DATA",
146 pid, oaddr, odata, ret, tmp);
153 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
155 if (copied != sizeof(tmp))
157 ret = put_user(tmp,(unsigned long *) data);
162 /* when I and D space are separate, this will have to be fixed. */
163 case PTRACE_POKETEXT: /* write the word at location addr. */
164 case PTRACE_POKEDATA:
167 if (is_compat_task(child)) {
168 unsigned int tmp = (unsigned int)data;
169 DBG("sys_ptrace(POKE%s, %d, %lx, %lx)\n",
170 request == PTRACE_POKETEXT ? "TEXT" : "DATA",
173 if (access_process_vm(child, addr, &tmp, sizeof(tmp), 1) == sizeof(tmp))
179 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
185 /* Read the word at location addr in the USER area. For ptraced
186 processes, the kernel saves all regs on a syscall. */
187 case PTRACE_PEEKUSR: {
190 if (is_compat_task(child)) {
193 if (addr & (sizeof(int)-1))
195 if ((addr = translate_usr_offset(addr)) < 0)
198 tmp = *(unsigned int *) ((char *) task_regs(child) + addr);
199 ret = put_user(tmp, (unsigned int *) data);
200 DBG("sys_ptrace(PEEKUSR, %d, %lx, %lx) returning %ld, addr %lx, data %x\n",
201 pid, oaddr, odata, ret, addr, tmp);
208 if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs))
210 tmp = *(unsigned long *) ((char *) task_regs(child) + addr);
211 ret = put_user(tmp, (unsigned long *) data);
216 /* Write the word at location addr in the USER area. This will need
217 to change when the kernel no longer saves all regs on a syscall.
218 FIXME. There is a problem at the moment in that r3-r18 are only
219 saved if the process is ptraced on syscall entry, and even then
220 those values are overwritten by actual register values on syscall
224 /* Some register values written here may be ignored in
225 * entry.S:syscall_restore_rfi; e.g. iaoq is written with
226 * r31/r31+4, and not with the values in pt_regs.
228 /* PT_PSW=0, so this is valid for 32 bit processes under 64
231 if (addr == PT_PSW) {
232 /* PT_PSW=0, so this is valid for 32 bit processes
233 * under 64 bit kernels.
235 * Allow writing to Nullify, Divide-step-correction,
236 * and carry/borrow bits.
237 * BEWARE, if you set N, and then single step, it won't
238 * stop on the nullified instruction.
240 DBG("sys_ptrace(POKEUSR, %d, %lx, %lx)\n",
242 data &= USER_PSW_BITS;
243 task_regs(child)->gr[0] &= ~USER_PSW_BITS;
244 task_regs(child)->gr[0] |= data;
249 if (is_compat_task(child)) {
250 if (addr & (sizeof(int)-1))
252 if ((addr = translate_usr_offset(addr)) < 0)
254 DBG("sys_ptrace(POKEUSR, %d, %lx, %lx) addr %lx\n",
255 pid, oaddr, odata, addr);
256 if (addr >= PT_FR0 && addr <= PT_FR31 + 4) {
257 /* Special case, fp regs are 64 bits anyway */
258 *(unsigned int *) ((char *) task_regs(child) + addr) = data;
261 else if ((addr >= PT_GR1+4 && addr <= PT_GR31+4) ||
262 addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4 ||
264 /* Zero the top 32 bits */
265 *(unsigned int *) ((char *) task_regs(child) + addr - 4) = 0;
266 *(unsigned int *) ((char *) task_regs(child) + addr) = data;
274 if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs))
276 if ((addr >= PT_GR1 && addr <= PT_GR31) ||
277 addr == PT_IAOQ0 || addr == PT_IAOQ1 ||
278 (addr >= PT_FR0 && addr <= PT_FR31 + 4) ||
280 *(unsigned long *) ((char *) task_regs(child) + addr) = data;
286 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
289 DBG("sys_ptrace(%s)\n",
290 request == PTRACE_SYSCALL ? "SYSCALL" : "CONT");
291 if ((unsigned long) data > _NSIG)
293 child->ptrace &= ~(PT_SINGLESTEP|PT_BLOCKSTEP);
294 if (request == PTRACE_SYSCALL) {
295 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
297 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
299 child->exit_code = data;
300 goto out_wake_notrap;
304 * make the child exit. Best I can do is send it a
305 * sigkill. perhaps it should be put in the status
306 * that it wants to exit.
308 DBG("sys_ptrace(KILL)\n");
309 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
311 child->exit_code = SIGKILL;
312 goto out_wake_notrap;
314 case PTRACE_SINGLEBLOCK:
315 DBG("sys_ptrace(SINGLEBLOCK)\n");
317 if ((unsigned long) data > _NSIG)
319 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
320 child->ptrace &= ~PT_SINGLESTEP;
321 child->ptrace |= PT_BLOCKSTEP;
322 child->exit_code = data;
324 /* Enable taken branch trap. */
325 pa_psw(child)->r = 0;
326 pa_psw(child)->t = 1;
327 pa_psw(child)->h = 0;
328 pa_psw(child)->l = 0;
331 case PTRACE_SINGLESTEP:
332 DBG("sys_ptrace(SINGLESTEP)\n");
334 if ((unsigned long) data > _NSIG)
337 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
338 child->ptrace &= ~PT_BLOCKSTEP;
339 child->ptrace |= PT_SINGLESTEP;
340 child->exit_code = data;
342 if (pa_psw(child)->n) {
345 /* Nullified, just crank over the queue. */
346 task_regs(child)->iaoq[0] = task_regs(child)->iaoq[1];
347 task_regs(child)->iasq[0] = task_regs(child)->iasq[1];
348 task_regs(child)->iaoq[1] = task_regs(child)->iaoq[0] + 4;
349 pa_psw(child)->n = 0;
350 pa_psw(child)->x = 0;
351 pa_psw(child)->y = 0;
352 pa_psw(child)->z = 0;
353 pa_psw(child)->b = 0;
354 ptrace_disable(child);
355 /* Don't wake up the child, but let the
356 parent know something happened. */
357 si.si_code = TRAP_TRACE;
358 si.si_addr = (void __user *) (task_regs(child)->iaoq[0] & ~3);
359 si.si_signo = SIGTRAP;
361 force_sig_info(SIGTRAP, &si, child);
362 //notify_parent(child, SIGCHLD);
367 /* Enable recovery counter traps. The recovery counter
368 * itself will be set to zero on a task switch. If the
369 * task is suspended on a syscall then the syscall return
370 * path will overwrite the recovery counter with a suitable
371 * value such that it traps once back in user space. We
372 * disable interrupts in the childs PSW here also, to avoid
373 * interrupts while the recovery counter is decrementing.
375 pa_psw(child)->r = 1;
376 pa_psw(child)->t = 0;
377 pa_psw(child)->h = 0;
378 pa_psw(child)->l = 0;
379 /* give it a chance to run. */
383 ret = ptrace_detach(child, data);
386 case PTRACE_GETEVENTMSG:
387 ret = put_user(child->ptrace_message, (unsigned int __user *) data);
391 ret = ptrace_request(child, request, addr, data);
396 ptrace_disable(child);
398 wake_up_process(child);
401 put_task_struct(child);
404 DBG("sys_ptrace(%ld, %d, %lx, %lx) returning %ld\n",
405 request, pid, oaddr, odata, ret);
409 void syscall_trace(void)
411 if (!test_thread_flag(TIF_SYSCALL_TRACE))
413 if (!(current->ptrace & PT_PTRACED))
415 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
418 * this isn't the same as continuing with a signal, but it will do
419 * for normal use. strace only continues with a signal if the
420 * stopping signal is not SIGTRAP. -brl
422 if (current->exit_code) {
423 send_sig(current->exit_code, current, 1);
424 current->exit_code = 0;