]> bbs.cooldavid.org Git - net-next-2.6.git/blame - arch/x86/kernel/ptrace.c
x86: ptrace FLAG_MASK cleanup
[net-next-2.6.git] / arch / x86 / kernel / ptrace.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].
9f155b98 35 */
e39c2891
RM
36#define FLAG_MASK_32 ((unsigned long) \
37 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
38 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
39 X86_EFLAGS_SF | X86_EFLAGS_TF | \
40 X86_EFLAGS_DF | X86_EFLAGS_OF | \
41 X86_EFLAGS_RF | X86_EFLAGS_AC))
42
43#define FLAG_MASK FLAG_MASK_32
1da177e4 44
62a97d44 45static long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
1da177e4 46{
65ea5b03 47 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
62a97d44
RM
48 if (regno > FS)
49 --regno;
65ea5b03 50 return &regs->bx + regno;
1da177e4
LT
51}
52
53static int putreg(struct task_struct *child,
54 unsigned long regno, unsigned long value)
55{
62a97d44
RM
56 struct pt_regs *regs = task_pt_regs(child);
57 regno >>= 2;
58 switch (regno) {
9e714bed
RM
59 case GS:
60 if (value && (value & 3) != 3)
61 return -EIO;
62 child->thread.gs = value;
5fd4d16b
RM
63 if (child == current)
64 /*
65 * The user-mode %gs is not affected by
66 * kernel entry, so we must update the CPU.
67 */
68 loadsegment(gs, value);
9e714bed
RM
69 return 0;
70 case DS:
71 case ES:
72 case FS:
73 if (value && (value & 3) != 3)
74 return -EIO;
75 value &= 0xffff;
76 break;
77 case SS:
78 case CS:
79 if ((value & 3) != 3)
80 return -EIO;
81 value &= 0xffff;
82 break;
83 case EFL:
84 value &= FLAG_MASK;
85 /*
86 * If the user value contains TF, mark that
87 * it was not "us" (the debugger) that set it.
88 * If not, make sure it stays set if we had.
89 */
90 if (value & X86_EFLAGS_TF)
91 clear_tsk_thread_flag(child, TIF_FORCED_TF);
92 else if (test_tsk_thread_flag(child, TIF_FORCED_TF))
93 value |= X86_EFLAGS_TF;
94 value |= regs->flags & ~FLAG_MASK;
95 break;
1da177e4 96 }
62a97d44 97 *pt_regs_access(regs, regno) = value;
1da177e4
LT
98 return 0;
99}
100
62a97d44 101static unsigned long getreg(struct task_struct *child, unsigned long regno)
1da177e4 102{
62a97d44 103 struct pt_regs *regs = task_pt_regs(child);
1da177e4
LT
104 unsigned long retval = ~0UL;
105
62a97d44
RM
106 regno >>= 2;
107 switch (regno) {
9e714bed
RM
108 case EFL:
109 /*
110 * If the debugger set TF, hide it from the readout.
111 */
112 retval = regs->flags;
113 if (test_tsk_thread_flag(child, TIF_FORCED_TF))
114 retval &= ~X86_EFLAGS_TF;
115 break;
116 case GS:
117 retval = child->thread.gs;
5fd4d16b
RM
118 if (child == current)
119 savesegment(gs, retval);
9e714bed
RM
120 break;
121 case DS:
122 case ES:
123 case FS:
124 case SS:
125 case CS:
126 retval = 0xffff;
127 /* fall through */
128 default:
129 retval &= *pt_regs_access(regs, regno);
1da177e4
LT
130 }
131 return retval;
132}
133
d9771e8c
RM
134/*
135 * This function is trivial and will be inlined by the compiler.
136 * Having it separates the implementation details of debug
137 * registers from the interface details of ptrace.
138 */
139static unsigned long ptrace_get_debugreg(struct task_struct *child, int n)
140{
0f534093
RM
141 switch (n) {
142 case 0: return child->thread.debugreg0;
143 case 1: return child->thread.debugreg1;
144 case 2: return child->thread.debugreg2;
145 case 3: return child->thread.debugreg3;
146 case 6: return child->thread.debugreg6;
147 case 7: return child->thread.debugreg7;
148 }
149 return 0;
d9771e8c
RM
150}
151
152static int ptrace_set_debugreg(struct task_struct *child,
153 int n, unsigned long data)
154{
0f534093
RM
155 int i;
156
d9771e8c
RM
157 if (unlikely(n == 4 || n == 5))
158 return -EIO;
159
160 if (n < 4 && unlikely(data >= TASK_SIZE - 3))
161 return -EIO;
162
0f534093
RM
163 switch (n) {
164 case 0: child->thread.debugreg0 = data; break;
165 case 1: child->thread.debugreg1 = data; break;
166 case 2: child->thread.debugreg2 = data; break;
167 case 3: child->thread.debugreg3 = data; break;
168
169 case 6:
170 child->thread.debugreg6 = data;
171 break;
172
173 case 7:
d9771e8c
RM
174 /*
175 * Sanity-check data. Take one half-byte at once with
176 * check = (val >> (16 + 4*i)) & 0xf. It contains the
177 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
178 * 2 and 3 are LENi. Given a list of invalid values,
179 * we do mask |= 1 << invalid_value, so that
180 * (mask >> check) & 1 is a correct test for invalid
181 * values.
182 *
183 * R/Wi contains the type of the breakpoint /
184 * watchpoint, LENi contains the length of the watched
185 * data in the watchpoint case.
186 *
187 * The invalid values are:
188 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
189 * - R/Wi == 0x10 (break on I/O reads or writes), so
190 * mask |= 0x4444.
191 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
192 * 0x1110.
193 *
194 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
195 *
196 * See the Intel Manual "System Programming Guide",
197 * 15.2.4
198 *
199 * Note that LENi == 0x10 is defined on x86_64 in long
200 * mode (i.e. even for 32-bit userspace software, but
201 * 64-bit kernel), so the x86_64 mask value is 0x5454.
202 * See the AMD manual no. 24593 (AMD64 System Programming)
203 */
d9771e8c
RM
204 data &= ~DR_CONTROL_RESERVED;
205 for (i = 0; i < 4; i++)
206 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
207 return -EIO;
0f534093 208 child->thread.debugreg7 = data;
d9771e8c
RM
209 if (data)
210 set_tsk_thread_flag(child, TIF_DEBUG);
211 else
212 clear_tsk_thread_flag(child, TIF_DEBUG);
0f534093 213 break;
d9771e8c
RM
214 }
215
d9771e8c
RM
216 return 0;
217}
218
1da177e4
LT
219/*
220 * Called by kernel/ptrace.c when detaching..
221 *
222 * Make sure the single step bit is not set.
223 */
224void ptrace_disable(struct task_struct *child)
9e714bed 225{
7f232343 226 user_disable_single_step(child);
ab1c23c2 227 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
1da177e4
LT
228}
229
481bed45 230long arch_ptrace(struct task_struct *child, long request, long addr, long data)
1da177e4 231{
1da177e4
LT
232 struct user * dummy = NULL;
233 int i, ret;
234 unsigned long __user *datap = (unsigned long __user *)data;
235
1da177e4
LT
236 switch (request) {
237 /* when I and D space are separate, these will need to be fixed. */
9e714bed 238 case PTRACE_PEEKTEXT: /* read word at location addr. */
76647323
AD
239 case PTRACE_PEEKDATA:
240 ret = generic_ptrace_peekdata(child, addr, data);
1da177e4 241 break;
1da177e4
LT
242
243 /* read the word at location addr in the USER area. */
244 case PTRACE_PEEKUSR: {
245 unsigned long tmp;
246
247 ret = -EIO;
9e714bed 248 if ((addr & 3) || addr < 0 ||
1da177e4
LT
249 addr > sizeof(struct user) - 3)
250 break;
251
252 tmp = 0; /* Default return condition */
253 if(addr < FRAME_SIZE*sizeof(long))
254 tmp = getreg(child, addr);
255 if(addr >= (long) &dummy->u_debugreg[0] &&
256 addr <= (long) &dummy->u_debugreg[7]){
257 addr -= (long) &dummy->u_debugreg[0];
258 addr = addr >> 2;
d9771e8c 259 tmp = ptrace_get_debugreg(child, addr);
1da177e4
LT
260 }
261 ret = put_user(tmp, datap);
262 break;
263 }
264
265 /* when I and D space are separate, this will have to be fixed. */
266 case PTRACE_POKETEXT: /* write the word at location addr. */
267 case PTRACE_POKEDATA:
f284ce72 268 ret = generic_ptrace_pokedata(child, addr, data);
1da177e4
LT
269 break;
270
271 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
272 ret = -EIO;
9e714bed 273 if ((addr & 3) || addr < 0 ||
1da177e4
LT
274 addr > sizeof(struct user) - 3)
275 break;
276
277 if (addr < FRAME_SIZE*sizeof(long)) {
278 ret = putreg(child, addr, data);
279 break;
280 }
281 /* We need to be very careful here. We implicitly
282 want to modify a portion of the task_struct, and we
283 have to be selective about what portions we allow someone
284 to modify. */
285
286 ret = -EIO;
287 if(addr >= (long) &dummy->u_debugreg[0] &&
288 addr <= (long) &dummy->u_debugreg[7]){
1da177e4
LT
289 addr -= (long) &dummy->u_debugreg;
290 addr = addr >> 2;
d9771e8c 291 ret = ptrace_set_debugreg(child, addr, data);
1da177e4
LT
292 }
293 break;
294
1da177e4
LT
295 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
296 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
297 ret = -EIO;
298 break;
299 }
300 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
301 __put_user(getreg(child, i), datap);
302 datap++;
303 }
304 ret = 0;
305 break;
306 }
307
308 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
309 unsigned long tmp;
310 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
311 ret = -EIO;
312 break;
313 }
314 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
315 __get_user(tmp, datap);
316 putreg(child, i, tmp);
317 datap++;
318 }
319 ret = 0;
320 break;
321 }
322
323 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
324 if (!access_ok(VERIFY_WRITE, datap,
325 sizeof(struct user_i387_struct))) {
326 ret = -EIO;
327 break;
328 }
329 ret = 0;
330 if (!tsk_used_math(child))
331 init_fpu(child);
332 get_fpregs((struct user_i387_struct __user *)data, child);
333 break;
334 }
335
336 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
337 if (!access_ok(VERIFY_READ, datap,
338 sizeof(struct user_i387_struct))) {
339 ret = -EIO;
340 break;
341 }
342 set_stopped_child_used_math(child);
343 set_fpregs(child, (struct user_i387_struct __user *)data);
344 ret = 0;
345 break;
346 }
347
348 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
349 if (!access_ok(VERIFY_WRITE, datap,
350 sizeof(struct user_fxsr_struct))) {
351 ret = -EIO;
352 break;
353 }
354 if (!tsk_used_math(child))
355 init_fpu(child);
356 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
357 break;
358 }
359
360 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
361 if (!access_ok(VERIFY_READ, datap,
362 sizeof(struct user_fxsr_struct))) {
363 ret = -EIO;
364 break;
365 }
366 set_stopped_child_used_math(child);
367 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
368 break;
369 }
370
371 case PTRACE_GET_THREAD_AREA:
efd1ca52
RM
372 if (addr < 0)
373 return -EIO;
374 ret = do_get_thread_area(child, addr,
375 (struct user_desc __user *) data);
1da177e4
LT
376 break;
377
378 case PTRACE_SET_THREAD_AREA:
efd1ca52
RM
379 if (addr < 0)
380 return -EIO;
381 ret = do_set_thread_area(child, addr,
382 (struct user_desc __user *) data, 0);
1da177e4
LT
383 break;
384
385 default:
386 ret = ptrace_request(child, request, addr, data);
387 break;
388 }
d9771e8c 389
1da177e4
LT
390 return ret;
391}
392
393void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
394{
395 struct siginfo info;
396
397 tsk->thread.trap_no = 1;
398 tsk->thread.error_code = error_code;
399
400 memset(&info, 0, sizeof(info));
401 info.si_signo = SIGTRAP;
402 info.si_code = TRAP_BRKPT;
403
65ea5b03
PA
404 /* User-mode ip? */
405 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL;
1da177e4 406
27b46d76 407 /* Send us the fake SIGTRAP */
1da177e4
LT
408 force_sig_info(SIGTRAP, &info, tsk);
409}
410
411/* notification of system call entry/exit
412 * - triggered by current->work.syscall_trace
413 */
414__attribute__((regparm(3)))
ed75e8d5 415int do_syscall_trace(struct pt_regs *regs, int entryexit)
1da177e4 416{
4c7fc722
AA
417 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
418 /*
419 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
420 * interception
421 */
1b38f006 422 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
4c7fc722 423 int ret = 0;
1b38f006 424
1da177e4 425 /* do the secure computing check first */
4c7fc722 426 if (!entryexit)
65ea5b03 427 secure_computing(regs->orig_ax);
1da177e4 428
ab1c23c2
BS
429 if (unlikely(current->audit_context)) {
430 if (entryexit)
65ea5b03
PA
431 audit_syscall_exit(AUDITSC_RESULT(regs->ax),
432 regs->ax);
ab1c23c2
BS
433 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
434 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
435 * not used, entry.S will call us only on syscall exit, not
436 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
437 * calling send_sigtrap() on syscall entry.
438 *
439 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
440 * is_singlestep is false, despite his name, so we will still do
441 * the correct thing.
442 */
443 else if (is_singlestep)
444 goto out;
445 }
1da177e4
LT
446
447 if (!(current->ptrace & PT_PTRACED))
2fd6f58b 448 goto out;
1da177e4 449
1b38f006
BS
450 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
451 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
452 * here. We have to check this and return */
453 if (is_sysemu && entryexit)
454 return 0;
ed75e8d5 455
1da177e4 456 /* Fake a debug trap */
c8c86cec 457 if (is_singlestep)
1da177e4
LT
458 send_sigtrap(current, regs, 0);
459
c8c86cec 460 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
2fd6f58b 461 goto out;
1da177e4
LT
462
463 /* the 0x80 provides a way for the tracing parent to distinguish
464 between a syscall stop and SIGTRAP delivery */
ed75e8d5 465 /* Note that the debugger could change the result of test_thread_flag!*/
4c7fc722 466 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
1da177e4
LT
467
468 /*
469 * this isn't the same as continuing with a signal, but it will do
470 * for normal use. strace only continues with a signal if the
471 * stopping signal is not SIGTRAP. -brl
472 */
473 if (current->exit_code) {
474 send_sig(current->exit_code, current, 1);
475 current->exit_code = 0;
476 }
ed75e8d5 477 ret = is_sysemu;
4c7fc722 478out:
2fd6f58b 479 if (unlikely(current->audit_context) && !entryexit)
65ea5b03
PA
480 audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax,
481 regs->bx, regs->cx, regs->dx, regs->si);
c8c86cec
BS
482 if (ret == 0)
483 return 0;
484
65ea5b03 485 regs->orig_ax = -1; /* force skip of syscall restarting */
c8c86cec 486 if (unlikely(current->audit_context))
65ea5b03 487 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
c8c86cec 488 return 1;
1da177e4 489}