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> | |
2047b08b RM |
27 | #include <asm/prctl.h> |
28 | #include <asm/proto.h> | |
1da177e4 LT |
29 | |
30 | /* | |
31 | * does not yet catch signals sent when the child dies. | |
32 | * in exit.c or in signal.c. | |
33 | */ | |
34 | ||
9f155b98 CE |
35 | /* |
36 | * Determines which flags the user has access to [1 = access, 0 = no access]. | |
9f155b98 | 37 | */ |
e39c2891 RM |
38 | #define FLAG_MASK_32 ((unsigned long) \ |
39 | (X86_EFLAGS_CF | X86_EFLAGS_PF | \ | |
40 | X86_EFLAGS_AF | X86_EFLAGS_ZF | \ | |
41 | X86_EFLAGS_SF | X86_EFLAGS_TF | \ | |
42 | X86_EFLAGS_DF | X86_EFLAGS_OF | \ | |
43 | X86_EFLAGS_RF | X86_EFLAGS_AC)) | |
44 | ||
2047b08b RM |
45 | /* |
46 | * Determines whether a value may be installed in a segment register. | |
47 | */ | |
48 | static inline bool invalid_selector(u16 value) | |
49 | { | |
50 | return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL); | |
51 | } | |
52 | ||
53 | #ifdef CONFIG_X86_32 | |
54 | ||
e39c2891 | 55 | #define FLAG_MASK FLAG_MASK_32 |
1da177e4 | 56 | |
62a97d44 | 57 | static long *pt_regs_access(struct pt_regs *regs, unsigned long regno) |
1da177e4 | 58 | { |
65ea5b03 | 59 | BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0); |
06ee1b68 | 60 | regno >>= 2; |
62a97d44 RM |
61 | if (regno > FS) |
62 | --regno; | |
65ea5b03 | 63 | return ®s->bx + regno; |
1da177e4 LT |
64 | } |
65 | ||
06ee1b68 | 66 | static u16 get_segment_reg(struct task_struct *task, unsigned long offset) |
1da177e4 | 67 | { |
06ee1b68 RM |
68 | /* |
69 | * Returning the value truncates it to 16 bits. | |
70 | */ | |
71 | unsigned int retval; | |
72 | if (offset != offsetof(struct user_regs_struct, gs)) | |
73 | retval = *pt_regs_access(task_pt_regs(task), offset); | |
74 | else { | |
75 | retval = task->thread.gs; | |
76 | if (task == current) | |
77 | savesegment(gs, retval); | |
78 | } | |
79 | return retval; | |
80 | } | |
81 | ||
82 | static int set_segment_reg(struct task_struct *task, | |
83 | unsigned long offset, u16 value) | |
84 | { | |
85 | /* | |
86 | * The value argument was already truncated to 16 bits. | |
87 | */ | |
2047b08b | 88 | if (invalid_selector(value)) |
06ee1b68 RM |
89 | return -EIO; |
90 | ||
91 | if (offset != offsetof(struct user_regs_struct, gs)) | |
92 | *pt_regs_access(task_pt_regs(task), offset) = value; | |
93 | else { | |
94 | task->thread.gs = value; | |
95 | if (task == current) | |
5fd4d16b RM |
96 | /* |
97 | * The user-mode %gs is not affected by | |
98 | * kernel entry, so we must update the CPU. | |
99 | */ | |
100 | loadsegment(gs, value); | |
1da177e4 | 101 | } |
06ee1b68 | 102 | |
1da177e4 LT |
103 | return 0; |
104 | } | |
105 | ||
2047b08b RM |
106 | static unsigned long debugreg_addr_limit(struct task_struct *task) |
107 | { | |
108 | return TASK_SIZE - 3; | |
109 | } | |
110 | ||
111 | #else /* CONFIG_X86_64 */ | |
112 | ||
113 | #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT) | |
114 | ||
115 | static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset) | |
116 | { | |
117 | BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0); | |
118 | return ®s->r15 + (offset / sizeof(regs->r15)); | |
119 | } | |
120 | ||
121 | static u16 get_segment_reg(struct task_struct *task, unsigned long offset) | |
122 | { | |
123 | /* | |
124 | * Returning the value truncates it to 16 bits. | |
125 | */ | |
126 | unsigned int seg; | |
127 | ||
128 | switch (offset) { | |
129 | case offsetof(struct user_regs_struct, fs): | |
130 | if (task == current) { | |
131 | /* Older gas can't assemble movq %?s,%r?? */ | |
132 | asm("movl %%fs,%0" : "=r" (seg)); | |
133 | return seg; | |
134 | } | |
135 | return task->thread.fsindex; | |
136 | case offsetof(struct user_regs_struct, gs): | |
137 | if (task == current) { | |
138 | asm("movl %%gs,%0" : "=r" (seg)); | |
139 | return seg; | |
140 | } | |
141 | return task->thread.gsindex; | |
142 | case offsetof(struct user_regs_struct, ds): | |
143 | if (task == current) { | |
144 | asm("movl %%ds,%0" : "=r" (seg)); | |
145 | return seg; | |
146 | } | |
147 | return task->thread.ds; | |
148 | case offsetof(struct user_regs_struct, es): | |
149 | if (task == current) { | |
150 | asm("movl %%es,%0" : "=r" (seg)); | |
151 | return seg; | |
152 | } | |
153 | return task->thread.es; | |
154 | ||
155 | case offsetof(struct user_regs_struct, cs): | |
156 | case offsetof(struct user_regs_struct, ss): | |
157 | break; | |
158 | } | |
159 | return *pt_regs_access(task_pt_regs(task), offset); | |
160 | } | |
161 | ||
162 | static int set_segment_reg(struct task_struct *task, | |
163 | unsigned long offset, u16 value) | |
164 | { | |
165 | /* | |
166 | * The value argument was already truncated to 16 bits. | |
167 | */ | |
168 | if (invalid_selector(value)) | |
169 | return -EIO; | |
170 | ||
171 | switch (offset) { | |
172 | case offsetof(struct user_regs_struct,fs): | |
173 | /* | |
174 | * If this is setting fs as for normal 64-bit use but | |
175 | * setting fs_base has implicitly changed it, leave it. | |
176 | */ | |
177 | if ((value == FS_TLS_SEL && task->thread.fsindex == 0 && | |
178 | task->thread.fs != 0) || | |
179 | (value == 0 && task->thread.fsindex == FS_TLS_SEL && | |
180 | task->thread.fs == 0)) | |
181 | break; | |
182 | task->thread.fsindex = value; | |
183 | if (task == current) | |
184 | loadsegment(fs, task->thread.fsindex); | |
185 | break; | |
186 | case offsetof(struct user_regs_struct,gs): | |
187 | /* | |
188 | * If this is setting gs as for normal 64-bit use but | |
189 | * setting gs_base has implicitly changed it, leave it. | |
190 | */ | |
191 | if ((value == GS_TLS_SEL && task->thread.gsindex == 0 && | |
192 | task->thread.gs != 0) || | |
193 | (value == 0 && task->thread.gsindex == GS_TLS_SEL && | |
194 | task->thread.gs == 0)) | |
195 | break; | |
196 | task->thread.gsindex = value; | |
197 | if (task == current) | |
198 | load_gs_index(task->thread.gsindex); | |
199 | break; | |
200 | case offsetof(struct user_regs_struct,ds): | |
201 | task->thread.ds = value; | |
202 | if (task == current) | |
203 | loadsegment(ds, task->thread.ds); | |
204 | break; | |
205 | case offsetof(struct user_regs_struct,es): | |
206 | task->thread.es = value; | |
207 | if (task == current) | |
208 | loadsegment(es, task->thread.es); | |
209 | break; | |
210 | ||
211 | /* | |
212 | * Can't actually change these in 64-bit mode. | |
213 | */ | |
214 | case offsetof(struct user_regs_struct,cs): | |
215 | #ifdef CONFIG_IA32_EMULATION | |
216 | if (test_tsk_thread_flag(task, TIF_IA32)) | |
217 | task_pt_regs(task)->cs = value; | |
2047b08b | 218 | #endif |
cb757c41 | 219 | break; |
2047b08b RM |
220 | case offsetof(struct user_regs_struct,ss): |
221 | #ifdef CONFIG_IA32_EMULATION | |
222 | if (test_tsk_thread_flag(task, TIF_IA32)) | |
223 | task_pt_regs(task)->ss = value; | |
2047b08b | 224 | #endif |
cb757c41 | 225 | break; |
2047b08b RM |
226 | } |
227 | ||
228 | return 0; | |
229 | } | |
230 | ||
231 | static unsigned long debugreg_addr_limit(struct task_struct *task) | |
232 | { | |
233 | #ifdef CONFIG_IA32_EMULATION | |
234 | if (test_tsk_thread_flag(task, TIF_IA32)) | |
235 | return IA32_PAGE_OFFSET - 3; | |
236 | #endif | |
237 | return TASK_SIZE64 - 7; | |
238 | } | |
239 | ||
240 | #endif /* CONFIG_X86_32 */ | |
241 | ||
06ee1b68 | 242 | static unsigned long get_flags(struct task_struct *task) |
1da177e4 | 243 | { |
06ee1b68 RM |
244 | unsigned long retval = task_pt_regs(task)->flags; |
245 | ||
246 | /* | |
247 | * If the debugger set TF, hide it from the readout. | |
248 | */ | |
249 | if (test_tsk_thread_flag(task, TIF_FORCED_TF)) | |
250 | retval &= ~X86_EFLAGS_TF; | |
1da177e4 | 251 | |
1da177e4 LT |
252 | return retval; |
253 | } | |
254 | ||
06ee1b68 RM |
255 | static int set_flags(struct task_struct *task, unsigned long value) |
256 | { | |
257 | struct pt_regs *regs = task_pt_regs(task); | |
258 | ||
259 | /* | |
260 | * If the user value contains TF, mark that | |
261 | * it was not "us" (the debugger) that set it. | |
262 | * If not, make sure it stays set if we had. | |
263 | */ | |
264 | if (value & X86_EFLAGS_TF) | |
265 | clear_tsk_thread_flag(task, TIF_FORCED_TF); | |
266 | else if (test_tsk_thread_flag(task, TIF_FORCED_TF)) | |
267 | value |= X86_EFLAGS_TF; | |
268 | ||
269 | regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK); | |
270 | ||
271 | return 0; | |
272 | } | |
273 | ||
274 | static int putreg(struct task_struct *child, | |
275 | unsigned long offset, unsigned long value) | |
276 | { | |
277 | switch (offset) { | |
278 | case offsetof(struct user_regs_struct, cs): | |
279 | case offsetof(struct user_regs_struct, ds): | |
280 | case offsetof(struct user_regs_struct, es): | |
281 | case offsetof(struct user_regs_struct, fs): | |
282 | case offsetof(struct user_regs_struct, gs): | |
283 | case offsetof(struct user_regs_struct, ss): | |
284 | return set_segment_reg(child, offset, value); | |
285 | ||
286 | case offsetof(struct user_regs_struct, flags): | |
287 | return set_flags(child, value); | |
2047b08b RM |
288 | |
289 | #ifdef CONFIG_X86_64 | |
290 | case offsetof(struct user_regs_struct,fs_base): | |
291 | if (value >= TASK_SIZE_OF(child)) | |
292 | return -EIO; | |
293 | /* | |
294 | * When changing the segment base, use do_arch_prctl | |
295 | * to set either thread.fs or thread.fsindex and the | |
296 | * corresponding GDT slot. | |
297 | */ | |
298 | if (child->thread.fs != value) | |
299 | return do_arch_prctl(child, ARCH_SET_FS, value); | |
300 | return 0; | |
301 | case offsetof(struct user_regs_struct,gs_base): | |
302 | /* | |
303 | * Exactly the same here as the %fs handling above. | |
304 | */ | |
305 | if (value >= TASK_SIZE_OF(child)) | |
306 | return -EIO; | |
307 | if (child->thread.gs != value) | |
308 | return do_arch_prctl(child, ARCH_SET_GS, value); | |
309 | return 0; | |
310 | #endif | |
06ee1b68 RM |
311 | } |
312 | ||
313 | *pt_regs_access(task_pt_regs(child), offset) = value; | |
314 | return 0; | |
315 | } | |
316 | ||
317 | static unsigned long getreg(struct task_struct *task, unsigned long offset) | |
318 | { | |
319 | switch (offset) { | |
320 | case offsetof(struct user_regs_struct, cs): | |
321 | case offsetof(struct user_regs_struct, ds): | |
322 | case offsetof(struct user_regs_struct, es): | |
323 | case offsetof(struct user_regs_struct, fs): | |
324 | case offsetof(struct user_regs_struct, gs): | |
325 | case offsetof(struct user_regs_struct, ss): | |
326 | return get_segment_reg(task, offset); | |
327 | ||
328 | case offsetof(struct user_regs_struct, flags): | |
329 | return get_flags(task); | |
2047b08b RM |
330 | |
331 | #ifdef CONFIG_X86_64 | |
332 | case offsetof(struct user_regs_struct, fs_base): { | |
333 | /* | |
334 | * do_arch_prctl may have used a GDT slot instead of | |
335 | * the MSR. To userland, it appears the same either | |
336 | * way, except the %fs segment selector might not be 0. | |
337 | */ | |
338 | unsigned int seg = task->thread.fsindex; | |
339 | if (task->thread.fs != 0) | |
340 | return task->thread.fs; | |
341 | if (task == current) | |
342 | asm("movl %%fs,%0" : "=r" (seg)); | |
343 | if (seg != FS_TLS_SEL) | |
344 | return 0; | |
345 | return get_desc_base(&task->thread.tls_array[FS_TLS]); | |
346 | } | |
347 | case offsetof(struct user_regs_struct, gs_base): { | |
348 | /* | |
349 | * Exactly the same here as the %fs handling above. | |
350 | */ | |
351 | unsigned int seg = task->thread.gsindex; | |
352 | if (task->thread.gs != 0) | |
353 | return task->thread.gs; | |
354 | if (task == current) | |
355 | asm("movl %%gs,%0" : "=r" (seg)); | |
356 | if (seg != GS_TLS_SEL) | |
357 | return 0; | |
358 | return get_desc_base(&task->thread.tls_array[GS_TLS]); | |
359 | } | |
360 | #endif | |
06ee1b68 RM |
361 | } |
362 | ||
363 | return *pt_regs_access(task_pt_regs(task), offset); | |
364 | } | |
365 | ||
d9771e8c RM |
366 | /* |
367 | * This function is trivial and will be inlined by the compiler. | |
368 | * Having it separates the implementation details of debug | |
369 | * registers from the interface details of ptrace. | |
370 | */ | |
371 | static unsigned long ptrace_get_debugreg(struct task_struct *child, int n) | |
372 | { | |
0f534093 RM |
373 | switch (n) { |
374 | case 0: return child->thread.debugreg0; | |
375 | case 1: return child->thread.debugreg1; | |
376 | case 2: return child->thread.debugreg2; | |
377 | case 3: return child->thread.debugreg3; | |
378 | case 6: return child->thread.debugreg6; | |
379 | case 7: return child->thread.debugreg7; | |
380 | } | |
381 | return 0; | |
d9771e8c RM |
382 | } |
383 | ||
384 | static int ptrace_set_debugreg(struct task_struct *child, | |
385 | int n, unsigned long data) | |
386 | { | |
0f534093 RM |
387 | int i; |
388 | ||
d9771e8c RM |
389 | if (unlikely(n == 4 || n == 5)) |
390 | return -EIO; | |
391 | ||
2047b08b | 392 | if (n < 4 && unlikely(data >= debugreg_addr_limit(child))) |
d9771e8c RM |
393 | return -EIO; |
394 | ||
0f534093 RM |
395 | switch (n) { |
396 | case 0: child->thread.debugreg0 = data; break; | |
397 | case 1: child->thread.debugreg1 = data; break; | |
398 | case 2: child->thread.debugreg2 = data; break; | |
399 | case 3: child->thread.debugreg3 = data; break; | |
400 | ||
401 | case 6: | |
2047b08b RM |
402 | if ((data & ~0xffffffffUL) != 0) |
403 | return -EIO; | |
0f534093 RM |
404 | child->thread.debugreg6 = data; |
405 | break; | |
406 | ||
407 | case 7: | |
d9771e8c RM |
408 | /* |
409 | * Sanity-check data. Take one half-byte at once with | |
410 | * check = (val >> (16 + 4*i)) & 0xf. It contains the | |
411 | * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits | |
412 | * 2 and 3 are LENi. Given a list of invalid values, | |
413 | * we do mask |= 1 << invalid_value, so that | |
414 | * (mask >> check) & 1 is a correct test for invalid | |
415 | * values. | |
416 | * | |
417 | * R/Wi contains the type of the breakpoint / | |
418 | * watchpoint, LENi contains the length of the watched | |
419 | * data in the watchpoint case. | |
420 | * | |
421 | * The invalid values are: | |
2047b08b | 422 | * - LENi == 0x10 (undefined), so mask |= 0x0f00. [32-bit] |
d9771e8c RM |
423 | * - R/Wi == 0x10 (break on I/O reads or writes), so |
424 | * mask |= 0x4444. | |
425 | * - R/Wi == 0x00 && LENi != 0x00, so we have mask |= | |
426 | * 0x1110. | |
427 | * | |
428 | * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54. | |
429 | * | |
430 | * See the Intel Manual "System Programming Guide", | |
431 | * 15.2.4 | |
432 | * | |
433 | * Note that LENi == 0x10 is defined on x86_64 in long | |
434 | * mode (i.e. even for 32-bit userspace software, but | |
435 | * 64-bit kernel), so the x86_64 mask value is 0x5454. | |
436 | * See the AMD manual no. 24593 (AMD64 System Programming) | |
437 | */ | |
2047b08b RM |
438 | #ifdef CONFIG_X86_32 |
439 | #define DR7_MASK 0x5f54 | |
440 | #else | |
441 | #define DR7_MASK 0x5554 | |
442 | #endif | |
d9771e8c RM |
443 | data &= ~DR_CONTROL_RESERVED; |
444 | for (i = 0; i < 4; i++) | |
2047b08b | 445 | if ((DR7_MASK >> ((data >> (16 + 4*i)) & 0xf)) & 1) |
d9771e8c | 446 | return -EIO; |
0f534093 | 447 | child->thread.debugreg7 = data; |
d9771e8c RM |
448 | if (data) |
449 | set_tsk_thread_flag(child, TIF_DEBUG); | |
450 | else | |
451 | clear_tsk_thread_flag(child, TIF_DEBUG); | |
0f534093 | 452 | break; |
d9771e8c RM |
453 | } |
454 | ||
d9771e8c RM |
455 | return 0; |
456 | } | |
457 | ||
1da177e4 LT |
458 | /* |
459 | * Called by kernel/ptrace.c when detaching.. | |
460 | * | |
461 | * Make sure the single step bit is not set. | |
462 | */ | |
463 | void ptrace_disable(struct task_struct *child) | |
9e714bed | 464 | { |
7f232343 | 465 | user_disable_single_step(child); |
e9c86c78 | 466 | #ifdef TIF_SYSCALL_EMU |
ab1c23c2 | 467 | clear_tsk_thread_flag(child, TIF_SYSCALL_EMU); |
e9c86c78 | 468 | #endif |
1da177e4 LT |
469 | } |
470 | ||
481bed45 | 471 | long arch_ptrace(struct task_struct *child, long request, long addr, long data) |
1da177e4 | 472 | { |
1da177e4 LT |
473 | int i, ret; |
474 | unsigned long __user *datap = (unsigned long __user *)data; | |
475 | ||
1da177e4 LT |
476 | switch (request) { |
477 | /* when I and D space are separate, these will need to be fixed. */ | |
9e714bed | 478 | case PTRACE_PEEKTEXT: /* read word at location addr. */ |
76647323 AD |
479 | case PTRACE_PEEKDATA: |
480 | ret = generic_ptrace_peekdata(child, addr, data); | |
1da177e4 | 481 | break; |
1da177e4 LT |
482 | |
483 | /* read the word at location addr in the USER area. */ | |
484 | case PTRACE_PEEKUSR: { | |
485 | unsigned long tmp; | |
486 | ||
487 | ret = -EIO; | |
e9c86c78 RM |
488 | if ((addr & (sizeof(data) - 1)) || addr < 0 || |
489 | addr >= sizeof(struct user)) | |
1da177e4 LT |
490 | break; |
491 | ||
492 | tmp = 0; /* Default return condition */ | |
e9c86c78 | 493 | if (addr < sizeof(struct user_regs_struct)) |
1da177e4 | 494 | tmp = getreg(child, addr); |
e9c86c78 RM |
495 | else if (addr >= offsetof(struct user, u_debugreg[0]) && |
496 | addr <= offsetof(struct user, u_debugreg[7])) { | |
497 | addr -= offsetof(struct user, u_debugreg[0]); | |
498 | tmp = ptrace_get_debugreg(child, addr / sizeof(data)); | |
1da177e4 LT |
499 | } |
500 | ret = put_user(tmp, datap); | |
501 | break; | |
502 | } | |
503 | ||
504 | /* when I and D space are separate, this will have to be fixed. */ | |
505 | case PTRACE_POKETEXT: /* write the word at location addr. */ | |
506 | case PTRACE_POKEDATA: | |
f284ce72 | 507 | ret = generic_ptrace_pokedata(child, addr, data); |
1da177e4 LT |
508 | break; |
509 | ||
510 | case PTRACE_POKEUSR: /* write the word at location addr in the USER area */ | |
511 | ret = -EIO; | |
e9c86c78 RM |
512 | if ((addr & (sizeof(data) - 1)) || addr < 0 || |
513 | addr >= sizeof(struct user)) | |
1da177e4 LT |
514 | break; |
515 | ||
e9c86c78 | 516 | if (addr < sizeof(struct user_regs_struct)) |
1da177e4 | 517 | ret = putreg(child, addr, data); |
e9c86c78 RM |
518 | else if (addr >= offsetof(struct user, u_debugreg[0]) && |
519 | addr <= offsetof(struct user, u_debugreg[7])) { | |
520 | addr -= offsetof(struct user, u_debugreg[0]); | |
521 | ret = ptrace_set_debugreg(child, | |
522 | addr / sizeof(data), data); | |
1da177e4 | 523 | } |
e9c86c78 | 524 | break; |
1da177e4 | 525 | |
1da177e4 | 526 | case PTRACE_GETREGS: { /* Get all gp regs from the child. */ |
e9c86c78 | 527 | if (!access_ok(VERIFY_WRITE, datap, sizeof(struct user_regs_struct))) { |
1da177e4 LT |
528 | ret = -EIO; |
529 | break; | |
530 | } | |
e9c86c78 | 531 | for (i = 0; i < sizeof(struct user_regs_struct); i += sizeof(long)) { |
1da177e4 LT |
532 | __put_user(getreg(child, i), datap); |
533 | datap++; | |
534 | } | |
535 | ret = 0; | |
536 | break; | |
537 | } | |
538 | ||
539 | case PTRACE_SETREGS: { /* Set all gp regs in the child. */ | |
540 | unsigned long tmp; | |
e9c86c78 | 541 | if (!access_ok(VERIFY_READ, datap, sizeof(struct user_regs_struct))) { |
1da177e4 LT |
542 | ret = -EIO; |
543 | break; | |
544 | } | |
e9c86c78 | 545 | for (i = 0; i < sizeof(struct user_regs_struct); i += sizeof(long)) { |
1da177e4 LT |
546 | __get_user(tmp, datap); |
547 | putreg(child, i, tmp); | |
548 | datap++; | |
549 | } | |
550 | ret = 0; | |
551 | break; | |
552 | } | |
553 | ||
554 | case PTRACE_GETFPREGS: { /* Get the child FPU state. */ | |
555 | if (!access_ok(VERIFY_WRITE, datap, | |
556 | sizeof(struct user_i387_struct))) { | |
557 | ret = -EIO; | |
558 | break; | |
559 | } | |
560 | ret = 0; | |
561 | if (!tsk_used_math(child)) | |
562 | init_fpu(child); | |
563 | get_fpregs((struct user_i387_struct __user *)data, child); | |
564 | break; | |
565 | } | |
566 | ||
567 | case PTRACE_SETFPREGS: { /* Set the child FPU state. */ | |
568 | if (!access_ok(VERIFY_READ, datap, | |
569 | sizeof(struct user_i387_struct))) { | |
570 | ret = -EIO; | |
571 | break; | |
572 | } | |
573 | set_stopped_child_used_math(child); | |
574 | set_fpregs(child, (struct user_i387_struct __user *)data); | |
575 | ret = 0; | |
576 | break; | |
577 | } | |
578 | ||
e9c86c78 | 579 | #ifdef CONFIG_X86_32 |
1da177e4 LT |
580 | case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */ |
581 | if (!access_ok(VERIFY_WRITE, datap, | |
582 | sizeof(struct user_fxsr_struct))) { | |
583 | ret = -EIO; | |
584 | break; | |
585 | } | |
586 | if (!tsk_used_math(child)) | |
587 | init_fpu(child); | |
588 | ret = get_fpxregs((struct user_fxsr_struct __user *)data, child); | |
589 | break; | |
590 | } | |
591 | ||
592 | case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */ | |
593 | if (!access_ok(VERIFY_READ, datap, | |
594 | sizeof(struct user_fxsr_struct))) { | |
595 | ret = -EIO; | |
596 | break; | |
597 | } | |
598 | set_stopped_child_used_math(child); | |
599 | ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data); | |
600 | break; | |
601 | } | |
e9c86c78 | 602 | #endif |
1da177e4 | 603 | |
e9c86c78 | 604 | #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION |
1da177e4 | 605 | case PTRACE_GET_THREAD_AREA: |
efd1ca52 RM |
606 | if (addr < 0) |
607 | return -EIO; | |
608 | ret = do_get_thread_area(child, addr, | |
609 | (struct user_desc __user *) data); | |
1da177e4 LT |
610 | break; |
611 | ||
612 | case PTRACE_SET_THREAD_AREA: | |
efd1ca52 RM |
613 | if (addr < 0) |
614 | return -EIO; | |
615 | ret = do_set_thread_area(child, addr, | |
616 | (struct user_desc __user *) data, 0); | |
1da177e4 | 617 | break; |
e9c86c78 RM |
618 | #endif |
619 | ||
620 | #ifdef CONFIG_X86_64 | |
621 | /* normal 64bit interface to access TLS data. | |
622 | Works just like arch_prctl, except that the arguments | |
623 | are reversed. */ | |
624 | case PTRACE_ARCH_PRCTL: | |
625 | ret = do_arch_prctl(child, data, addr); | |
626 | break; | |
627 | #endif | |
1da177e4 LT |
628 | |
629 | default: | |
630 | ret = ptrace_request(child, request, addr, data); | |
631 | break; | |
632 | } | |
d9771e8c | 633 | |
1da177e4 LT |
634 | return ret; |
635 | } | |
636 | ||
cb757c41 RM |
637 | #ifdef CONFIG_IA32_EMULATION |
638 | ||
639 | #include <asm/user32.h> | |
640 | ||
641 | #define R32(l,q) \ | |
642 | case offsetof(struct user32, regs.l): \ | |
643 | regs->q = value; break | |
644 | ||
645 | #define SEG32(rs) \ | |
646 | case offsetof(struct user32, regs.rs): \ | |
647 | return set_segment_reg(child, \ | |
648 | offsetof(struct user_regs_struct, rs), \ | |
649 | value); \ | |
650 | break | |
651 | ||
652 | static int putreg32(struct task_struct *child, unsigned regno, u32 value) | |
653 | { | |
654 | struct pt_regs *regs = task_pt_regs(child); | |
655 | ||
656 | switch (regno) { | |
657 | ||
658 | SEG32(cs); | |
659 | SEG32(ds); | |
660 | SEG32(es); | |
661 | SEG32(fs); | |
662 | SEG32(gs); | |
663 | SEG32(ss); | |
664 | ||
665 | R32(ebx, bx); | |
666 | R32(ecx, cx); | |
667 | R32(edx, dx); | |
668 | R32(edi, di); | |
669 | R32(esi, si); | |
670 | R32(ebp, bp); | |
671 | R32(eax, ax); | |
672 | R32(orig_eax, orig_ax); | |
673 | R32(eip, ip); | |
674 | R32(esp, sp); | |
675 | ||
676 | case offsetof(struct user32, regs.eflags): | |
677 | return set_flags(child, value); | |
678 | ||
679 | case offsetof(struct user32, u_debugreg[0]) ... | |
680 | offsetof(struct user32, u_debugreg[7]): | |
681 | regno -= offsetof(struct user32, u_debugreg[0]); | |
682 | return ptrace_set_debugreg(child, regno / 4, value); | |
683 | ||
684 | default: | |
685 | if (regno > sizeof(struct user32) || (regno & 3)) | |
686 | return -EIO; | |
687 | ||
688 | /* | |
689 | * Other dummy fields in the virtual user structure | |
690 | * are ignored | |
691 | */ | |
692 | break; | |
693 | } | |
694 | return 0; | |
695 | } | |
696 | ||
697 | #undef R32 | |
698 | #undef SEG32 | |
699 | ||
700 | #define R32(l,q) \ | |
701 | case offsetof(struct user32, regs.l): \ | |
702 | *val = regs->q; break | |
703 | ||
704 | #define SEG32(rs) \ | |
705 | case offsetof(struct user32, regs.rs): \ | |
706 | *val = get_segment_reg(child, \ | |
707 | offsetof(struct user_regs_struct, rs)); \ | |
708 | break | |
709 | ||
710 | static int getreg32(struct task_struct *child, unsigned regno, u32 *val) | |
711 | { | |
712 | struct pt_regs *regs = task_pt_regs(child); | |
713 | ||
714 | switch (regno) { | |
715 | ||
716 | SEG32(ds); | |
717 | SEG32(es); | |
718 | SEG32(fs); | |
719 | SEG32(gs); | |
720 | ||
721 | R32(cs, cs); | |
722 | R32(ss, ss); | |
723 | R32(ebx, bx); | |
724 | R32(ecx, cx); | |
725 | R32(edx, dx); | |
726 | R32(edi, di); | |
727 | R32(esi, si); | |
728 | R32(ebp, bp); | |
729 | R32(eax, ax); | |
730 | R32(orig_eax, orig_ax); | |
731 | R32(eip, ip); | |
732 | R32(esp, sp); | |
733 | ||
734 | case offsetof(struct user32, regs.eflags): | |
735 | *val = get_flags(child); | |
736 | break; | |
737 | ||
738 | case offsetof(struct user32, u_debugreg[0]) ... | |
739 | offsetof(struct user32, u_debugreg[7]): | |
740 | regno -= offsetof(struct user32, u_debugreg[0]); | |
741 | *val = ptrace_get_debugreg(child, regno / 4); | |
742 | break; | |
743 | ||
744 | default: | |
745 | if (regno > sizeof(struct user32) || (regno & 3)) | |
746 | return -EIO; | |
747 | ||
748 | /* | |
749 | * Other dummy fields in the virtual user structure | |
750 | * are ignored | |
751 | */ | |
752 | *val = 0; | |
753 | break; | |
754 | } | |
755 | return 0; | |
756 | } | |
757 | ||
758 | #undef R32 | |
759 | #undef SEG32 | |
760 | ||
761 | #endif /* CONFIG_IA32_EMULATION */ | |
762 | ||
86976cd8 RM |
763 | #ifdef CONFIG_X86_32 |
764 | ||
1da177e4 LT |
765 | void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code) |
766 | { | |
767 | struct siginfo info; | |
768 | ||
769 | tsk->thread.trap_no = 1; | |
770 | tsk->thread.error_code = error_code; | |
771 | ||
772 | memset(&info, 0, sizeof(info)); | |
773 | info.si_signo = SIGTRAP; | |
774 | info.si_code = TRAP_BRKPT; | |
775 | ||
65ea5b03 PA |
776 | /* User-mode ip? */ |
777 | info.si_addr = user_mode_vm(regs) ? (void __user *) regs->ip : NULL; | |
1da177e4 | 778 | |
27b46d76 | 779 | /* Send us the fake SIGTRAP */ |
1da177e4 LT |
780 | force_sig_info(SIGTRAP, &info, tsk); |
781 | } | |
782 | ||
783 | /* notification of system call entry/exit | |
784 | * - triggered by current->work.syscall_trace | |
785 | */ | |
786 | __attribute__((regparm(3))) | |
ed75e8d5 | 787 | int do_syscall_trace(struct pt_regs *regs, int entryexit) |
1da177e4 | 788 | { |
4c7fc722 AA |
789 | int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU); |
790 | /* | |
791 | * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall | |
792 | * interception | |
793 | */ | |
1b38f006 | 794 | int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP); |
4c7fc722 | 795 | int ret = 0; |
1b38f006 | 796 | |
1da177e4 | 797 | /* do the secure computing check first */ |
4c7fc722 | 798 | if (!entryexit) |
65ea5b03 | 799 | secure_computing(regs->orig_ax); |
1da177e4 | 800 | |
ab1c23c2 BS |
801 | if (unlikely(current->audit_context)) { |
802 | if (entryexit) | |
65ea5b03 PA |
803 | audit_syscall_exit(AUDITSC_RESULT(regs->ax), |
804 | regs->ax); | |
ab1c23c2 BS |
805 | /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only |
806 | * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is | |
807 | * not used, entry.S will call us only on syscall exit, not | |
808 | * entry; so when TIF_SYSCALL_AUDIT is used we must avoid | |
809 | * calling send_sigtrap() on syscall entry. | |
810 | * | |
811 | * Note that when PTRACE_SYSEMU_SINGLESTEP is used, | |
812 | * is_singlestep is false, despite his name, so we will still do | |
813 | * the correct thing. | |
814 | */ | |
815 | else if (is_singlestep) | |
816 | goto out; | |
817 | } | |
1da177e4 LT |
818 | |
819 | if (!(current->ptrace & PT_PTRACED)) | |
2fd6f58b | 820 | goto out; |
1da177e4 | 821 | |
1b38f006 BS |
822 | /* If a process stops on the 1st tracepoint with SYSCALL_TRACE |
823 | * and then is resumed with SYSEMU_SINGLESTEP, it will come in | |
824 | * here. We have to check this and return */ | |
825 | if (is_sysemu && entryexit) | |
826 | return 0; | |
ed75e8d5 | 827 | |
1da177e4 | 828 | /* Fake a debug trap */ |
c8c86cec | 829 | if (is_singlestep) |
1da177e4 LT |
830 | send_sigtrap(current, regs, 0); |
831 | ||
c8c86cec | 832 | if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu) |
2fd6f58b | 833 | goto out; |
1da177e4 LT |
834 | |
835 | /* the 0x80 provides a way for the tracing parent to distinguish | |
836 | between a syscall stop and SIGTRAP delivery */ | |
ed75e8d5 | 837 | /* Note that the debugger could change the result of test_thread_flag!*/ |
4c7fc722 | 838 | ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0)); |
1da177e4 LT |
839 | |
840 | /* | |
841 | * this isn't the same as continuing with a signal, but it will do | |
842 | * for normal use. strace only continues with a signal if the | |
843 | * stopping signal is not SIGTRAP. -brl | |
844 | */ | |
845 | if (current->exit_code) { | |
846 | send_sig(current->exit_code, current, 1); | |
847 | current->exit_code = 0; | |
848 | } | |
ed75e8d5 | 849 | ret = is_sysemu; |
4c7fc722 | 850 | out: |
2fd6f58b | 851 | if (unlikely(current->audit_context) && !entryexit) |
65ea5b03 PA |
852 | audit_syscall_entry(AUDIT_ARCH_I386, regs->orig_ax, |
853 | regs->bx, regs->cx, regs->dx, regs->si); | |
c8c86cec BS |
854 | if (ret == 0) |
855 | return 0; | |
856 | ||
65ea5b03 | 857 | regs->orig_ax = -1; /* force skip of syscall restarting */ |
c8c86cec | 858 | if (unlikely(current->audit_context)) |
65ea5b03 | 859 | audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax); |
c8c86cec | 860 | return 1; |
1da177e4 | 861 | } |
86976cd8 RM |
862 | |
863 | #else /* CONFIG_X86_64 */ | |
864 | ||
865 | static void syscall_trace(struct pt_regs *regs) | |
866 | { | |
867 | ||
868 | #if 0 | |
869 | printk("trace %s ip %lx sp %lx ax %d origrax %d caller %lx tiflags %x ptrace %x\n", | |
870 | current->comm, | |
871 | regs->ip, regs->sp, regs->ax, regs->orig_ax, __builtin_return_address(0), | |
872 | current_thread_info()->flags, current->ptrace); | |
873 | #endif | |
874 | ||
875 | ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) | |
876 | ? 0x80 : 0)); | |
877 | /* | |
878 | * this isn't the same as continuing with a signal, but it will do | |
879 | * for normal use. strace only continues with a signal if the | |
880 | * stopping signal is not SIGTRAP. -brl | |
881 | */ | |
882 | if (current->exit_code) { | |
883 | send_sig(current->exit_code, current, 1); | |
884 | current->exit_code = 0; | |
885 | } | |
886 | } | |
887 | ||
888 | asmlinkage void syscall_trace_enter(struct pt_regs *regs) | |
889 | { | |
890 | /* do the secure computing check first */ | |
891 | secure_computing(regs->orig_ax); | |
892 | ||
893 | if (test_thread_flag(TIF_SYSCALL_TRACE) | |
894 | && (current->ptrace & PT_PTRACED)) | |
895 | syscall_trace(regs); | |
896 | ||
897 | if (unlikely(current->audit_context)) { | |
898 | if (test_thread_flag(TIF_IA32)) { | |
899 | audit_syscall_entry(AUDIT_ARCH_I386, | |
900 | regs->orig_ax, | |
901 | regs->bx, regs->cx, | |
902 | regs->dx, regs->si); | |
903 | } else { | |
904 | audit_syscall_entry(AUDIT_ARCH_X86_64, | |
905 | regs->orig_ax, | |
906 | regs->di, regs->si, | |
907 | regs->dx, regs->r10); | |
908 | } | |
909 | } | |
910 | } | |
911 | ||
912 | asmlinkage void syscall_trace_leave(struct pt_regs *regs) | |
913 | { | |
914 | if (unlikely(current->audit_context)) | |
915 | audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax); | |
916 | ||
917 | if ((test_thread_flag(TIF_SYSCALL_TRACE) | |
918 | || test_thread_flag(TIF_SINGLESTEP)) | |
919 | && (current->ptrace & PT_PTRACED)) | |
920 | syscall_trace(regs); | |
921 | } | |
922 | ||
923 | #endif /* CONFIG_X86_32 */ |