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