1/*
2 * Kernel support for the ptrace() and syscall tracing interfaces.
3 *
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>
7 */
8
9#include <linux/kernel.h>
10#include <linux/sched.h>
11#include <linux/mm.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
19#include <asm/uaccess.h>
20#include <asm/pgtable.h>
21#include <asm/system.h>
22#include <asm/processor.h>
23#include <asm/offset.h>
24
25/* These are used in entry.S, syscall_restore_rfi.  We need to record the
26 * current stepping mode somewhere other than in PSW, because there is no
27 * concept of saving and restoring the users PSW over a syscall.  We choose
28 * to use these two bits in task->ptrace.  These bits must not clash with
29 * any PT_* defined in include/linux/sched.h, and must match with the bit
30 * tests in entry.S
31 */
32#define PT_SINGLESTEP	0x10000
33#define PT_BLOCKSTEP	0x20000
34
35/* PSW bits we allow the debugger to modify */
36#define USER_PSW_BITS	(PSW_N | PSW_V | PSW_CB)
37
38#undef DEBUG_PTRACE
39
40#ifdef DEBUG_PTRACE
41#define DBG(x)	printk x
42#else
43#define DBG(x)
44#endif
45
46#ifdef __LP64__
47
48#define CHILD_IS_32BIT	(child->personality == PER_LINUX_32BIT)
49
50/* This function is needed to translate 32 bit pt_regs offsets in to
51 * 64 bit pt_regs offsets.  For example, a 32 bit gdb under a 64 bit kernel
52 * will request offset 12 if it wants gr3, but the lower 32 bits of
53 * the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4).
54 * This code relies on a 32 bit pt_regs being comprised of 32 bit values
55 * except for the fp registers which (a) are 64 bits, and (b) follow
56 * the gr registers at the start of pt_regs.  The 32 bit pt_regs should
57 * be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[]
58 * being 64 bit in both cases.
59 */
60
61static long translate_usr_offset(long offset)
62{
63	if (offset < 0)
64		return -1;
65	else if (offset <= 32*4)	/* gr[0..31] */
66		return offset * 2 + 4;
67	else if (offset <= 32*4+32*8)	/* gr[0..31] + fr[0..31] */
68		return offset + 32*4;
69	else if (offset < sizeof(struct pt_regs)/2 + 32*4)
70		return offset * 2 + 4 - 32*8;
71	else
72		return -1;
73}
74#endif
75
76/*
77 * Called by kernel/ptrace.c when detaching..
78 *
79 * Make sure single step bits etc are not set.
80 */
81void ptrace_disable(struct task_struct *child)
82{
83	/* make sure the trap bits are not set */
84	pa_psw(child)->r = 0;
85	pa_psw(child)->t = 0;
86	pa_psw(child)->h = 0;
87	pa_psw(child)->l = 0;
88}
89
90long sys_ptrace(long request, pid_t pid, long addr, long data)
91{
92	struct task_struct *child;
93	long ret;
94#ifdef DEBUG_PTRACE
95	long oaddr=addr, odata=data;
96#endif
97
98	lock_kernel();
99	ret = -EPERM;
100	if (request == PTRACE_TRACEME) {
101		/* are we already being traced? */
102		if (current->ptrace & PT_PTRACED)
103			goto out;
104		/* set the ptrace bit in the process flags. */
105		current->ptrace |= PT_PTRACED;
106		ret = 0;
107		goto out;
108	}
109
110	ret = -ESRCH;
111	read_lock(&tasklist_lock);
112	child = find_task_by_pid(pid);
113	if (child)
114		get_task_struct(child);
115	read_unlock(&tasklist_lock);
116	if (!child)
117		goto out;
118	ret = -EPERM;
119	if (pid == 1)		/* no messing around with init! */
120		goto out_tsk;
121
122	if (request == PTRACE_ATTACH) {
123		ret = ptrace_attach(child);
124		goto out_tsk;
125	}
126	ret = -ESRCH;
127	if (!(child->ptrace & PT_PTRACED))
128		goto out_tsk;
129	if (child->state != TASK_STOPPED) {
130		if (request != PTRACE_KILL)
131			goto out_tsk;
132	}
133	if (child->p_pptr != current)
134		goto out_tsk;
135
136	switch (request) {
137	case PTRACE_PEEKTEXT: /* read word at location addr. */
138	case PTRACE_PEEKDATA: {
139		int copied;
140
141#ifdef __LP64__
142		if (CHILD_IS_32BIT) {
143			unsigned int tmp;
144
145			addr &= 0xffffffffL;
146			copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
147			ret = -EIO;
148			if (copied != sizeof(tmp))
149				goto out_tsk;
150			ret = put_user(tmp,(unsigned int *) data);
151			DBG(("sys_ptrace(PEEK%s, %d, %lx, %lx) returning %ld, data %x\n",
152				request == PTRACE_PEEKTEXT ? "TEXT" : "DATA",
153				pid, oaddr, odata, ret, tmp));
154		}
155		else
156#endif
157		{
158			unsigned long tmp;
159
160			copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
161			ret = -EIO;
162			if (copied != sizeof(tmp))
163				goto out_tsk;
164			ret = put_user(tmp,(unsigned long *) data);
165		}
166		goto out_tsk;
167	}
168
169	/* when I and D space are separate, this will have to be fixed. */
170	case PTRACE_POKETEXT: /* write the word at location addr. */
171	case PTRACE_POKEDATA:
172		ret = 0;
173#ifdef __LP64__
174		if (CHILD_IS_32BIT) {
175			unsigned int tmp = (unsigned int)data;
176			DBG(("sys_ptrace(POKE%s, %d, %lx, %lx)\n",
177				request == PTRACE_POKETEXT ? "TEXT" : "DATA",
178				pid, oaddr, odata));
179			addr &= 0xffffffffL;
180			if (access_process_vm(child, addr, &tmp, sizeof(tmp), 1) == sizeof(tmp))
181				goto out_tsk;
182		}
183		else
184#endif
185		{
186			if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
187				goto out_tsk;
188		}
189		ret = -EIO;
190		goto out_tsk;
191
192	/* Read the word at location addr in the USER area.  For ptraced
193	   processes, the kernel saves all regs on a syscall. */
194	case PTRACE_PEEKUSR: {
195		ret = -EIO;
196#ifdef __LP64__
197		if (CHILD_IS_32BIT) {
198			unsigned int tmp;
199
200			if (addr & (sizeof(int)-1))
201				goto out_tsk;
202			if ((addr = translate_usr_offset(addr)) < 0)
203				goto out_tsk;
204
205			tmp = *(unsigned int *) ((char *) task_regs(child) + addr);
206			ret = put_user(tmp, (unsigned int *) data);
207			DBG(("sys_ptrace(PEEKUSR, %d, %lx, %lx) returning %ld, addr %lx, data %x\n",
208				pid, oaddr, odata, ret, addr, tmp));
209		}
210		else
211#endif
212		{
213			unsigned long tmp;
214
215			if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs))
216				goto out_tsk;
217			tmp = *(unsigned long *) ((char *) task_regs(child) + addr);
218			ret = put_user(tmp, (unsigned long *) data);
219		}
220		goto out_tsk;
221	}
222
223	case PTRACE_POKEUSR:
224		ret = -EIO;
225		/* Some register values written here may be ignored in
226		 * entry.S:syscall_restore_rfi; e.g. iaoq is written with
227		 * r31/r31+4, and not with the values in pt_regs.
228		 */
229		 /* PT_PSW=0, so this is valid for 32 bit processes under 64
230		 * bit kernels.
231		 */
232		if (addr == PT_PSW) {
233			/* PT_PSW=0, so this is valid for 32 bit processes
234			 * under 64 bit kernels.
235			 *
236			 * Allow writing to Nullify, Divide-step-correction,
237			 * and carry/borrow bits.
238			 * BEWARE, if you set N, and then single step, it wont
239			 * stop on the nullified instruction.
240			 */
241			DBG(("sys_ptrace(POKEUSR, %d, %lx, %lx)\n",
242				pid, oaddr, odata));
243			data &= USER_PSW_BITS;
244			task_regs(child)->gr[0] &= ~USER_PSW_BITS;
245			task_regs(child)->gr[0] |= data;
246			ret = 0;
247			goto out_tsk;
248		}
249#ifdef __LP64__
250		if (CHILD_IS_32BIT) {
251			if (addr & (sizeof(int)-1))
252				goto out_tsk;
253			if ((addr = translate_usr_offset(addr)) < 0)
254				goto out_tsk;
255			DBG(("sys_ptrace(POKEUSR, %d, %lx, %lx) addr %lx\n",
256				pid, oaddr, odata, addr));
257			if (addr >= PT_FR0 && addr <= PT_FR31) {
258				/* Special case, fp regs are 64 bits anyway */
259				*(unsigned int *) ((char *) task_regs(child) + addr) = data;
260				ret = 0;
261			}
262			else if ((addr >= PT_GR1+4 && addr <= PT_GR31+4) ||
263					addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4 ||
264					addr == PT_SAR+4) {
265				/* Zero the top 32 bits */
266				*(unsigned int *) ((char *) task_regs(child) + addr - 4) = 0;
267				*(unsigned int *) ((char *) task_regs(child) + addr) = data;
268				ret = 0;
269			}
270			goto out_tsk;
271		}
272		else
273#endif
274		{
275			if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs))
276				goto out_tsk;
277			if ((addr >= PT_GR1 && addr <= PT_GR31) ||
278					addr == PT_IAOQ0 || addr == PT_IAOQ1 ||
279					(addr >= PT_FR0 && addr <= PT_FR31) ||
280					addr == PT_SAR) {
281				*(unsigned long *) ((char *) task_regs(child) + addr) = data;
282				ret = 0;
283			}
284			goto out_tsk;
285		}
286
287	case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
288	case PTRACE_CONT:
289		ret = -EIO;
290		DBG(("sys_ptrace(%s)\n",
291			request == PTRACE_SYSCALL ? "SYSCALL" : "CONT"));
292		if ((unsigned long) data > _NSIG)
293			goto out_tsk;
294		child->ptrace &= ~(PT_SINGLESTEP|PT_BLOCKSTEP);
295		if (request == PTRACE_SYSCALL)
296			child->ptrace |= PT_TRACESYS;
297		else
298			child->ptrace &= ~PT_TRACESYS;
299		child->exit_code = data;
300		goto out_wake_notrap;
301
302	case PTRACE_KILL:
303		/*
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.
307		 */
308		DBG(("sys_ptrace(KILL)\n"));
309		if (child->state == TASK_ZOMBIE)	/* already dead */
310			goto out_tsk;
311		child->exit_code = SIGKILL;
312		goto out_wake_notrap;
313
314	case PTRACE_SINGLEBLOCK:
315		DBG(("sys_ptrace(SINGLEBLOCK)\n"));
316		ret = -EIO;
317		if ((unsigned long) data > _NSIG)
318			goto out_tsk;
319		child->ptrace &= ~(PT_TRACESYS|PT_SINGLESTEP);
320		child->ptrace |= PT_BLOCKSTEP;
321		child->exit_code = data;
322
323		/* Enable taken branch trap. */
324		pa_psw(child)->r = 0;
325		pa_psw(child)->t = 1;
326		pa_psw(child)->h = 0;
327		pa_psw(child)->l = 0;
328		goto out_wake;
329
330	case PTRACE_SINGLESTEP:
331		DBG(("sys_ptrace(SINGLESTEP)\n"));
332		ret = -EIO;
333		if ((unsigned long) data > _NSIG)
334			goto out_tsk;
335		child->ptrace &= ~(PT_TRACESYS|PT_BLOCKSTEP);
336		child->ptrace |= PT_SINGLESTEP;
337		child->exit_code = data;
338
339		if (pa_psw(child)->n) {
340			struct siginfo si;
341
342			/* Nullified, just crank over the queue. */
343			task_regs(child)->iaoq[0] = task_regs(child)->iaoq[1];
344			task_regs(child)->iasq[0] = task_regs(child)->iasq[1];
345			task_regs(child)->iaoq[1] = task_regs(child)->iaoq[0] + 4;
346			pa_psw(child)->n = 0;
347			pa_psw(child)->x = 0;
348			pa_psw(child)->y = 0;
349			pa_psw(child)->z = 0;
350			pa_psw(child)->b = 0;
351			ptrace_disable(child);
352			/* Don't wake up the child, but let the
353			   parent know something happened. */
354			si.si_code = TRAP_TRACE;
355			si.si_addr = (void *) (task_regs(child)->iaoq[0] & ~3);
356			si.si_signo = SIGTRAP;
357			si.si_errno = 0;
358			force_sig_info(SIGTRAP, &si, child);
359			//notify_parent(child, SIGCHLD);
360			//ret = 0;
361			goto out_wake;
362		}
363
364		/* Enable recovery counter traps.  The recovery counter
365		 * itself will be set to zero on a task switch.  If the
366		 * task is suspended on a syscall then the syscall return
367		 * path will overwrite the recovery counter with a suitable
368		 * value such that it traps once back in user space.  We
369		 * disable interrupts in the childs PSW here also, to avoid
370		 * interrupts while the recovery counter is decrementing.
371		 */
372		pa_psw(child)->r = 1;
373		pa_psw(child)->t = 0;
374		pa_psw(child)->h = 0;
375		pa_psw(child)->l = 0;
376		/* give it a chance to run. */
377		goto out_wake;
378
379	case PTRACE_DETACH:
380		ret = ptrace_detach(child, data);
381		goto out_tsk;
382
383	default:
384		ret = -EIO;
385		goto out_tsk;
386	}
387
388out_wake_notrap:
389	ptrace_disable(child);
390out_wake:
391	wake_up_process(child);
392	ret = 0;
393out_tsk:
394	free_task_struct(child);
395out:
396	unlock_kernel();
397	DBG(("sys_ptrace(%ld, %d, %lx, %lx) returning %ld\n",
398		request, pid, oaddr, odata, ret));
399	return ret;
400}
401
402void syscall_trace(void)
403{
404	if ((current->ptrace & (PT_PTRACED|PT_TRACESYS)) !=
405			(PT_PTRACED|PT_TRACESYS))
406		return;
407	current->exit_code = SIGTRAP;
408	current->state = TASK_STOPPED;
409	notify_parent(current, SIGCHLD);
410	schedule();
411	/*
412	 * this isn't the same as continuing with a signal, but it will do
413	 * for normal use.  strace only continues with a signal if the
414	 * stopping signal is not SIGTRAP.  -brl
415	 */
416	if (current->exit_code) {
417		send_sig(current->exit_code, current, 1);
418		current->exit_code = 0;
419	}
420}
421