x86: x86-64 ia32 ptrace debugreg cleanup
[deliverable/linux.git] / arch / x86 / kernel / ptrace_32.c
CommitLineData
1da177e4
LT
1/* By Ross Biro 1/23/92 */
2/*
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
5 */
6
7#include <linux/kernel.h>
8#include <linux/sched.h>
9#include <linux/mm.h>
10#include <linux/smp.h>
1da177e4
LT
11#include <linux/errno.h>
12#include <linux/ptrace.h>
13#include <linux/user.h>
14#include <linux/security.h>
15#include <linux/audit.h>
16#include <linux/seccomp.h>
7ed20e1a 17#include <linux/signal.h>
1da177e4
LT
18
19#include <asm/uaccess.h>
20#include <asm/pgtable.h>
21#include <asm/system.h>
22#include <asm/processor.h>
23#include <asm/i387.h>
24#include <asm/debugreg.h>
25#include <asm/ldt.h>
26#include <asm/desc.h>
27
28/*
29 * does not yet catch signals sent when the child dies.
30 * in exit.c or in signal.c.
31 */
32
9f155b98
CE
33/*
34 * Determines which flags the user has access to [1 = access, 0 = no access].
3c36c6aa 35 * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), NT(14), IOPL(12-13), IF(9).
9f155b98
CE
36 * Also masks reserved bits (31-22, 15, 5, 3, 1).
37 */
3c36c6aa 38#define FLAG_MASK 0x00050dd5
1da177e4 39
62a97d44 40static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
1da177e4 41{
62a97d44
RM
42 BUILD_BUG_ON(offsetof(struct pt_regs, ebx) != 0);
43 if (regno > FS)
44 --regno;
45 return &regs->ebx + regno;
1da177e4
LT
46}
47
48static int putreg(struct task_struct *child,
49 unsigned long regno, unsigned long value)
50{
62a97d44
RM
51 struct pt_regs *regs = task_pt_regs(child);
52 regno >>= 2;
53 switch (regno) {
464d1a78 54 case GS:
1da177e4
LT
55 if (value && (value & 3) != 3)
56 return -EIO;
464d1a78 57 child->thread.gs = value;
1da177e4 58 return 0;
1da177e4
LT
59 case DS:
60 case ES:
464d1a78 61 case FS:
1da177e4
LT
62 if (value && (value & 3) != 3)
63 return -EIO;
64 value &= 0xffff;
65 break;
66 case SS:
67 case CS:
68 if ((value & 3) != 3)
69 return -EIO;
70 value &= 0xffff;
71 break;
72 case EFL:
73 value &= FLAG_MASK;
e1f28773
RM
74 /*
75 * If the user value contains TF, mark that
76 * it was not "us" (the debugger) that set it.
77 * If not, make sure it stays set if we had.
78 */
79 if (value & X86_EFLAGS_TF)
80 clear_tsk_thread_flag(child, TIF_FORCED_TF);
81 else if (test_tsk_thread_flag(child, TIF_FORCED_TF))
82 value |= X86_EFLAGS_TF;
62a97d44 83 value |= regs->eflags & ~FLAG_MASK;
1da177e4
LT
84 break;
85 }
62a97d44 86 *pt_regs_access(regs, regno) = value;
1da177e4
LT
87 return 0;
88}
89
62a97d44 90static unsigned long getreg(struct task_struct *child, unsigned long regno)
1da177e4 91{
62a97d44 92 struct pt_regs *regs = task_pt_regs(child);
1da177e4
LT
93 unsigned long retval = ~0UL;
94
62a97d44
RM
95 regno >>= 2;
96 switch (regno) {
e1f28773
RM
97 case EFL:
98 /*
99 * If the debugger set TF, hide it from the readout.
100 */
62a97d44 101 retval = regs->eflags;
e1f28773
RM
102 if (test_tsk_thread_flag(child, TIF_FORCED_TF))
103 retval &= ~X86_EFLAGS_TF;
104 break;
464d1a78
JF
105 case GS:
106 retval = child->thread.gs;
1da177e4 107 break;
1da177e4
LT
108 case DS:
109 case ES:
464d1a78 110 case FS:
1da177e4
LT
111 case SS:
112 case CS:
113 retval = 0xffff;
114 /* fall through */
115 default:
62a97d44 116 retval &= *pt_regs_access(regs, regno);
1da177e4
LT
117 }
118 return retval;
119}
120
1da177e4
LT
121/*
122 * Called by kernel/ptrace.c when detaching..
123 *
124 * Make sure the single step bit is not set.
125 */
126void ptrace_disable(struct task_struct *child)
127{
7f232343 128 user_disable_single_step(child);
ab1c23c2 129 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
1da177e4
LT
130}
131
481bed45 132long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1da177e4 133{
1da177e4
LT
134 struct user * dummy = NULL;
135 int i, ret;
136 unsigned long __user *datap = (unsigned long __user *)data;
137
1da177e4
LT
138 switch (request) {
139 /* when I and D space are separate, these will need to be fixed. */
140 case PTRACE_PEEKTEXT: /* read word at location addr. */
76647323
AD
141 case PTRACE_PEEKDATA:
142 ret = generic_ptrace_peekdata(child, addr, data);
1da177e4 143 break;
1da177e4
LT
144
145 /* read the word at location addr in the USER area. */
146 case PTRACE_PEEKUSR: {
147 unsigned long tmp;
148
149 ret = -EIO;
150 if ((addr & 3) || addr < 0 ||
151 addr > sizeof(struct user) - 3)
152 break;
153
154 tmp = 0; /* Default return condition */
155 if(addr < FRAME_SIZE*sizeof(long))
156 tmp = getreg(child, addr);
157 if(addr >= (long) &dummy->u_debugreg[0] &&
158 addr <= (long) &dummy->u_debugreg[7]){
159 addr -= (long) &dummy->u_debugreg[0];
160 addr = addr >> 2;
161 tmp = child->thread.debugreg[addr];
162 }
163 ret = put_user(tmp, datap);
164 break;
165 }
166
167 /* when I and D space are separate, this will have to be fixed. */
168 case PTRACE_POKETEXT: /* write the word at location addr. */
169 case PTRACE_POKEDATA:
f284ce72 170 ret = generic_ptrace_pokedata(child, addr, data);
1da177e4
LT
171 break;
172
173 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
174 ret = -EIO;
175 if ((addr & 3) || addr < 0 ||
176 addr > sizeof(struct user) - 3)
177 break;
178
179 if (addr < FRAME_SIZE*sizeof(long)) {
180 ret = putreg(child, addr, data);
181 break;
182 }
183 /* We need to be very careful here. We implicitly
184 want to modify a portion of the task_struct, and we
185 have to be selective about what portions we allow someone
186 to modify. */
187
188 ret = -EIO;
189 if(addr >= (long) &dummy->u_debugreg[0] &&
190 addr <= (long) &dummy->u_debugreg[7]){
191
192 if(addr == (long) &dummy->u_debugreg[4]) break;
193 if(addr == (long) &dummy->u_debugreg[5]) break;
194 if(addr < (long) &dummy->u_debugreg[4] &&
195 ((unsigned long) data) >= TASK_SIZE-3) break;
196
197 /* Sanity-check data. Take one half-byte at once with
198 * check = (val >> (16 + 4*i)) & 0xf. It contains the
199 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
200 * 2 and 3 are LENi. Given a list of invalid values,
201 * we do mask |= 1 << invalid_value, so that
202 * (mask >> check) & 1 is a correct test for invalid
203 * values.
204 *
205 * R/Wi contains the type of the breakpoint /
206 * watchpoint, LENi contains the length of the watched
207 * data in the watchpoint case.
208 *
209 * The invalid values are:
210 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
211 * - R/Wi == 0x10 (break on I/O reads or writes), so
212 * mask |= 0x4444.
213 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
214 * 0x1110.
215 *
216 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
217 *
218 * See the Intel Manual "System Programming Guide",
219 * 15.2.4
220 *
221 * Note that LENi == 0x10 is defined on x86_64 in long
222 * mode (i.e. even for 32-bit userspace software, but
223 * 64-bit kernel), so the x86_64 mask value is 0x5454.
224 * See the AMD manual no. 24593 (AMD64 System
225 * Programming)*/
226
227 if(addr == (long) &dummy->u_debugreg[7]) {
228 data &= ~DR_CONTROL_RESERVED;
229 for(i=0; i<4; i++)
230 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
231 goto out_tsk;
b3cf2576
SE
232 if (data)
233 set_tsk_thread_flag(child, TIF_DEBUG);
234 else
235 clear_tsk_thread_flag(child, TIF_DEBUG);
1da177e4 236 }
1da177e4
LT
237 addr -= (long) &dummy->u_debugreg;
238 addr = addr >> 2;
239 child->thread.debugreg[addr] = data;
240 ret = 0;
241 }
242 break;
243
1da177e4
LT
244 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
245 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
246 ret = -EIO;
247 break;
248 }
249 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
250 __put_user(getreg(child, i), datap);
251 datap++;
252 }
253 ret = 0;
254 break;
255 }
256
257 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
258 unsigned long tmp;
259 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
260 ret = -EIO;
261 break;
262 }
263 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
264 __get_user(tmp, datap);
265 putreg(child, i, tmp);
266 datap++;
267 }
268 ret = 0;
269 break;
270 }
271
272 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
273 if (!access_ok(VERIFY_WRITE, datap,
274 sizeof(struct user_i387_struct))) {
275 ret = -EIO;
276 break;
277 }
278 ret = 0;
279 if (!tsk_used_math(child))
280 init_fpu(child);
281 get_fpregs((struct user_i387_struct __user *)data, child);
282 break;
283 }
284
285 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
286 if (!access_ok(VERIFY_READ, datap,
287 sizeof(struct user_i387_struct))) {
288 ret = -EIO;
289 break;
290 }
291 set_stopped_child_used_math(child);
292 set_fpregs(child, (struct user_i387_struct __user *)data);
293 ret = 0;
294 break;
295 }
296
297 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
298 if (!access_ok(VERIFY_WRITE, datap,
299 sizeof(struct user_fxsr_struct))) {
300 ret = -EIO;
301 break;
302 }
303 if (!tsk_used_math(child))
304 init_fpu(child);
305 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
306 break;
307 }
308
309 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
310 if (!access_ok(VERIFY_READ, datap,
311 sizeof(struct user_fxsr_struct))) {
312 ret = -EIO;
313 break;
314 }
315 set_stopped_child_used_math(child);
316 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
317 break;
318 }
319
320 case PTRACE_GET_THREAD_AREA:
efd1ca52
RM
321 if (addr < 0)
322 return -EIO;
323 ret = do_get_thread_area(child, addr,
324 (struct user_desc __user *) data);
1da177e4
LT
325 break;
326
327 case PTRACE_SET_THREAD_AREA:
efd1ca52
RM
328 if (addr < 0)
329 return -EIO;
330 ret = do_set_thread_area(child, addr,
331 (struct user_desc __user *) data, 0);
1da177e4
LT
332 break;
333
334 default:
335 ret = ptrace_request(child, request, addr, data);
336 break;
337 }
481bed45 338 out_tsk:
1da177e4
LT
339 return ret;
340}
341
342void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
343{
344 struct siginfo info;
345
346 tsk->thread.trap_no = 1;
347 tsk->thread.error_code = error_code;
348
349 memset(&info, 0, sizeof(info));
350 info.si_signo = SIGTRAP;
351 info.si_code = TRAP_BRKPT;
352
353 /* User-mode eip? */
fa1e1bdf 354 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
1da177e4 355
27b46d76 356 /* Send us the fake SIGTRAP */
1da177e4
LT
357 force_sig_info(SIGTRAP, &info, tsk);
358}
359
360/* notification of system call entry/exit
361 * - triggered by current->work.syscall_trace
362 */
363__attribute__((regparm(3)))
ed75e8d5 364int do_syscall_trace(struct pt_regs *regs, int entryexit)
1da177e4 365{
4c7fc722
AA
366 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
367 /*
368 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
369 * interception
370 */
1b38f006 371 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
4c7fc722 372 int ret = 0;
1b38f006 373
1da177e4 374 /* do the secure computing check first */
4c7fc722
AA
375 if (!entryexit)
376 secure_computing(regs->orig_eax);
1da177e4 377
ab1c23c2
BS
378 if (unlikely(current->audit_context)) {
379 if (entryexit)
5411be59 380 audit_syscall_exit(AUDITSC_RESULT(regs->eax),
4c7fc722 381 regs->eax);
ab1c23c2
BS
382 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
383 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
384 * not used, entry.S will call us only on syscall exit, not
385 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
386 * calling send_sigtrap() on syscall entry.
387 *
388 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
389 * is_singlestep is false, despite his name, so we will still do
390 * the correct thing.
391 */
392 else if (is_singlestep)
393 goto out;
394 }
1da177e4
LT
395
396 if (!(current->ptrace & PT_PTRACED))
2fd6f58b 397 goto out;
1da177e4 398
1b38f006
BS
399 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
400 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
401 * here. We have to check this and return */
402 if (is_sysemu && entryexit)
403 return 0;
ed75e8d5 404
1da177e4 405 /* Fake a debug trap */
c8c86cec 406 if (is_singlestep)
1da177e4
LT
407 send_sigtrap(current, regs, 0);
408
c8c86cec 409 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
2fd6f58b 410 goto out;
1da177e4
LT
411
412 /* the 0x80 provides a way for the tracing parent to distinguish
413 between a syscall stop and SIGTRAP delivery */
ed75e8d5 414 /* Note that the debugger could change the result of test_thread_flag!*/
4c7fc722 415 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1da177e4
LT
416
417 /*
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
421 */
422 if (current->exit_code) {
423 send_sig(current->exit_code, current, 1);
424 current->exit_code = 0;
425 }
ed75e8d5 426 ret = is_sysemu;
4c7fc722 427out:
2fd6f58b 428 if (unlikely(current->audit_context) && !entryexit)
5411be59 429 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_eax,
2fd6f58b 430 regs->ebx, regs->ecx, regs->edx, regs->esi);
c8c86cec
BS
431 if (ret == 0)
432 return 0;
433
1b38f006 434 regs->orig_eax = -1; /* force skip of syscall restarting */
c8c86cec 435 if (unlikely(current->audit_context))
5411be59 436 audit_syscall_exit(AUDITSC_RESULT(regs->eax), regs->eax);
c8c86cec 437 return 1;
1da177e4 438}
This page took 0.377215 seconds and 5 git commands to generate.