Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2000 Jeff Dike (jdike@karaya.com) |
| 3 | * Licensed under the GPL |
| 4 | */ |
| 5 | |
| 6 | #include "linux/sched.h" |
| 7 | #include "linux/mm.h" |
| 8 | #include "linux/errno.h" |
| 9 | #include "linux/smp_lock.h" |
| 10 | #include "linux/security.h" |
| 11 | #include "linux/ptrace.h" |
| 12 | #include "linux/audit.h" |
| 13 | #ifdef CONFIG_PROC_MM |
| 14 | #include "linux/proc_mm.h" |
| 15 | #endif |
| 16 | #include "asm/ptrace.h" |
| 17 | #include "asm/uaccess.h" |
| 18 | #include "kern_util.h" |
| 19 | #include "skas_ptrace.h" |
| 20 | #include "sysdep/ptrace.h" |
| 21 | |
| 22 | /* |
| 23 | * Called by kernel/ptrace.c when detaching.. |
| 24 | */ |
| 25 | void ptrace_disable(struct task_struct *child) |
| 26 | { |
| 27 | child->ptrace &= ~PT_DTRACE; |
| 28 | child->thread.singlestep_syscall = 0; |
| 29 | } |
| 30 | |
| 31 | long sys_ptrace(long request, long pid, long addr, long data) |
| 32 | { |
| 33 | struct task_struct *child; |
| 34 | int i, ret; |
| 35 | |
| 36 | lock_kernel(); |
| 37 | ret = -EPERM; |
| 38 | if (request == PTRACE_TRACEME) { |
| 39 | /* are we already being traced? */ |
| 40 | if (current->ptrace & PT_PTRACED) |
| 41 | goto out; |
| 42 | |
| 43 | ret = security_ptrace(current->parent, current); |
| 44 | if (ret) |
| 45 | goto out; |
| 46 | |
| 47 | /* set the ptrace bit in the process flags. */ |
| 48 | current->ptrace |= PT_PTRACED; |
| 49 | ret = 0; |
| 50 | goto out; |
| 51 | } |
| 52 | ret = -ESRCH; |
| 53 | read_lock(&tasklist_lock); |
| 54 | child = find_task_by_pid(pid); |
| 55 | if (child) |
| 56 | get_task_struct(child); |
| 57 | read_unlock(&tasklist_lock); |
| 58 | if (!child) |
| 59 | goto out; |
| 60 | |
| 61 | ret = -EPERM; |
| 62 | if (pid == 1) /* you may not mess with init */ |
| 63 | goto out_tsk; |
| 64 | |
| 65 | if (request == PTRACE_ATTACH) { |
| 66 | ret = ptrace_attach(child); |
| 67 | goto out_tsk; |
| 68 | } |
| 69 | |
| 70 | ret = ptrace_check_attach(child, request == PTRACE_KILL); |
| 71 | if (ret < 0) |
| 72 | goto out_tsk; |
| 73 | |
| 74 | switch (request) { |
| 75 | /* when I and D space are separate, these will need to be fixed. */ |
| 76 | case PTRACE_PEEKTEXT: /* read word at location addr. */ |
| 77 | case PTRACE_PEEKDATA: { |
| 78 | unsigned long tmp; |
| 79 | int copied; |
| 80 | |
| 81 | ret = -EIO; |
| 82 | copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0); |
| 83 | if (copied != sizeof(tmp)) |
| 84 | break; |
| 85 | ret = put_user(tmp, (unsigned long __user *) data); |
| 86 | break; |
| 87 | } |
| 88 | |
| 89 | /* read the word at location addr in the USER area. */ |
| 90 | case PTRACE_PEEKUSR: { |
| 91 | unsigned long tmp; |
| 92 | |
| 93 | ret = -EIO; |
| 94 | if ((addr & 3) || addr < 0) |
| 95 | break; |
| 96 | |
| 97 | tmp = 0; /* Default return condition */ |
| 98 | if(addr < MAX_REG_OFFSET){ |
| 99 | tmp = getreg(child, addr); |
| 100 | } |
| 101 | else if((addr >= offsetof(struct user, u_debugreg[0])) && |
| 102 | (addr <= offsetof(struct user, u_debugreg[7]))){ |
| 103 | addr -= offsetof(struct user, u_debugreg[0]); |
| 104 | addr = addr >> 2; |
| 105 | tmp = child->thread.arch.debugregs[addr]; |
| 106 | } |
| 107 | ret = put_user(tmp, (unsigned long __user *) data); |
| 108 | break; |
| 109 | } |
| 110 | |
| 111 | /* when I and D space are separate, this will have to be fixed. */ |
| 112 | case PTRACE_POKETEXT: /* write the word at location addr. */ |
| 113 | case PTRACE_POKEDATA: |
| 114 | ret = -EIO; |
| 115 | if (access_process_vm(child, addr, &data, sizeof(data), |
| 116 | 1) != sizeof(data)) |
| 117 | break; |
| 118 | ret = 0; |
| 119 | break; |
| 120 | |
| 121 | case PTRACE_POKEUSR: /* write the word at location addr in the USER area */ |
| 122 | ret = -EIO; |
| 123 | if ((addr & 3) || addr < 0) |
| 124 | break; |
| 125 | |
| 126 | if (addr < MAX_REG_OFFSET) { |
| 127 | ret = putreg(child, addr, data); |
| 128 | break; |
| 129 | } |
| 130 | #if 0 /* XXX x86_64 */ |
| 131 | else if((addr >= offsetof(struct user, u_debugreg[0])) && |
| 132 | (addr <= offsetof(struct user, u_debugreg[7]))){ |
| 133 | addr -= offsetof(struct user, u_debugreg[0]); |
| 134 | addr = addr >> 2; |
| 135 | if((addr == 4) || (addr == 5)) break; |
| 136 | child->thread.arch.debugregs[addr] = data; |
| 137 | ret = 0; |
| 138 | } |
| 139 | #endif |
| 140 | |
| 141 | break; |
| 142 | |
| 143 | case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */ |
| 144 | case PTRACE_CONT: { /* restart after signal. */ |
| 145 | ret = -EIO; |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 146 | if (!valid_signal(data)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 147 | break; |
| 148 | |
| 149 | child->ptrace &= ~PT_DTRACE; |
| 150 | child->thread.singlestep_syscall = 0; |
| 151 | if (request == PTRACE_SYSCALL) { |
| 152 | set_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 153 | } |
| 154 | else { |
| 155 | clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 156 | } |
| 157 | child->exit_code = data; |
| 158 | wake_up_process(child); |
| 159 | ret = 0; |
| 160 | break; |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * make the child exit. Best I can do is send it a sigkill. |
| 165 | * perhaps it should be put in the status that it wants to |
| 166 | * exit. |
| 167 | */ |
| 168 | case PTRACE_KILL: { |
| 169 | ret = 0; |
| 170 | if (child->exit_state == EXIT_ZOMBIE) /* already dead */ |
| 171 | break; |
| 172 | |
| 173 | child->ptrace &= ~PT_DTRACE; |
| 174 | child->thread.singlestep_syscall = 0; |
| 175 | child->exit_code = SIGKILL; |
| 176 | wake_up_process(child); |
| 177 | break; |
| 178 | } |
| 179 | |
| 180 | case PTRACE_SINGLESTEP: { /* set the trap flag. */ |
| 181 | ret = -EIO; |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 182 | if (!valid_signal(data)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | break; |
| 184 | clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 185 | child->ptrace |= PT_DTRACE; |
| 186 | child->thread.singlestep_syscall = 0; |
| 187 | child->exit_code = data; |
| 188 | /* give it a chance to run. */ |
| 189 | wake_up_process(child); |
| 190 | ret = 0; |
| 191 | break; |
| 192 | } |
| 193 | |
| 194 | case PTRACE_DETACH: |
| 195 | /* detach a process that was attached. */ |
| 196 | ret = ptrace_detach(child, data); |
| 197 | break; |
| 198 | |
| 199 | #ifdef PTRACE_GETREGS |
| 200 | case PTRACE_GETREGS: { /* Get all gp regs from the child. */ |
| 201 | if (!access_ok(VERIFY_WRITE, (unsigned long *)data, |
| 202 | MAX_REG_OFFSET)) { |
| 203 | ret = -EIO; |
| 204 | break; |
| 205 | } |
| 206 | for ( i = 0; i < MAX_REG_OFFSET; i += sizeof(long) ) { |
| 207 | __put_user(getreg(child, i), |
| 208 | (unsigned long __user *) data); |
| 209 | data += sizeof(long); |
| 210 | } |
| 211 | ret = 0; |
| 212 | break; |
| 213 | } |
| 214 | #endif |
| 215 | #ifdef PTRACE_SETREGS |
| 216 | case PTRACE_SETREGS: { /* Set all gp regs in the child. */ |
| 217 | unsigned long tmp = 0; |
| 218 | if (!access_ok(VERIFY_READ, (unsigned *)data, |
| 219 | MAX_REG_OFFSET)) { |
| 220 | ret = -EIO; |
| 221 | break; |
| 222 | } |
| 223 | for ( i = 0; i < MAX_REG_OFFSET; i += sizeof(long) ) { |
| 224 | __get_user(tmp, (unsigned long __user *) data); |
| 225 | putreg(child, i, tmp); |
| 226 | data += sizeof(long); |
| 227 | } |
| 228 | ret = 0; |
| 229 | break; |
| 230 | } |
| 231 | #endif |
| 232 | #ifdef PTRACE_GETFPREGS |
| 233 | case PTRACE_GETFPREGS: /* Get the child FPU state. */ |
| 234 | ret = get_fpregs(data, child); |
| 235 | break; |
| 236 | #endif |
| 237 | #ifdef PTRACE_SETFPREGS |
| 238 | case PTRACE_SETFPREGS: /* Set the child FPU state. */ |
| 239 | ret = set_fpregs(data, child); |
| 240 | break; |
| 241 | #endif |
| 242 | #ifdef PTRACE_GETFPXREGS |
| 243 | case PTRACE_GETFPXREGS: /* Get the child FPU state. */ |
| 244 | ret = get_fpxregs(data, child); |
| 245 | break; |
| 246 | #endif |
| 247 | #ifdef PTRACE_SETFPXREGS |
| 248 | case PTRACE_SETFPXREGS: /* Set the child FPU state. */ |
| 249 | ret = set_fpxregs(data, child); |
| 250 | break; |
| 251 | #endif |
| 252 | case PTRACE_FAULTINFO: { |
| 253 | struct ptrace_faultinfo fault; |
| 254 | |
| 255 | fault = ((struct ptrace_faultinfo) |
| 256 | { .is_write = child->thread.err, |
| 257 | .addr = child->thread.cr2 }); |
| 258 | ret = copy_to_user((unsigned long __user *) data, &fault, |
| 259 | sizeof(fault)); |
| 260 | if(ret) |
| 261 | break; |
| 262 | break; |
| 263 | } |
| 264 | case PTRACE_SIGPENDING: |
| 265 | ret = copy_to_user((unsigned long __user *) data, |
| 266 | &child->pending.signal, |
| 267 | sizeof(child->pending.signal)); |
| 268 | break; |
| 269 | |
| 270 | case PTRACE_LDT: { |
| 271 | struct ptrace_ldt ldt; |
| 272 | |
| 273 | if(copy_from_user(&ldt, (unsigned long __user *) data, |
| 274 | sizeof(ldt))){ |
| 275 | ret = -EIO; |
| 276 | break; |
| 277 | } |
| 278 | |
| 279 | /* This one is confusing, so just punt and return -EIO for |
| 280 | * now |
| 281 | */ |
| 282 | ret = -EIO; |
| 283 | break; |
| 284 | } |
| 285 | #ifdef CONFIG_PROC_MM |
| 286 | case PTRACE_SWITCH_MM: { |
| 287 | struct mm_struct *old = child->mm; |
| 288 | struct mm_struct *new = proc_mm_get_mm(data); |
| 289 | |
| 290 | if(IS_ERR(new)){ |
| 291 | ret = PTR_ERR(new); |
| 292 | break; |
| 293 | } |
| 294 | |
| 295 | atomic_inc(&new->mm_users); |
| 296 | child->mm = new; |
| 297 | child->active_mm = new; |
| 298 | mmput(old); |
| 299 | ret = 0; |
| 300 | break; |
| 301 | } |
| 302 | #endif |
| 303 | default: |
| 304 | ret = ptrace_request(child, request, addr, data); |
| 305 | break; |
| 306 | } |
| 307 | out_tsk: |
| 308 | put_task_struct(child); |
| 309 | out: |
| 310 | unlock_kernel(); |
| 311 | return ret; |
| 312 | } |
| 313 | |
| 314 | void send_sigtrap(struct task_struct *tsk, union uml_pt_regs *regs, |
| 315 | int error_code) |
| 316 | { |
| 317 | struct siginfo info; |
| 318 | |
| 319 | memset(&info, 0, sizeof(info)); |
| 320 | info.si_signo = SIGTRAP; |
| 321 | info.si_code = TRAP_BRKPT; |
| 322 | |
| 323 | /* User-mode eip? */ |
| 324 | info.si_addr = UPT_IS_USER(regs) ? (void __user *) UPT_IP(regs) : NULL; |
| 325 | |
| 326 | /* Send us the fakey SIGTRAP */ |
| 327 | force_sig_info(SIGTRAP, &info, tsk); |
| 328 | } |
| 329 | |
| 330 | /* XXX Check PT_DTRACE vs TIF_SINGLESTEP for singlestepping check and |
| 331 | * PT_PTRACED vs TIF_SYSCALL_TRACE for syscall tracing check |
| 332 | */ |
| 333 | void syscall_trace(union uml_pt_regs *regs, int entryexit) |
| 334 | { |
| 335 | int is_singlestep = (current->ptrace & PT_DTRACE) && entryexit; |
| 336 | int tracesysgood; |
| 337 | |
| 338 | if (unlikely(current->audit_context)) { |
| 339 | if (!entryexit) |
| 340 | audit_syscall_entry(current, |
| 341 | UPT_SYSCALL_NR(®s->regs), |
| 342 | UPT_SYSCALL_ARG1(®s->regs), |
| 343 | UPT_SYSCALL_ARG2(®s->regs), |
| 344 | UPT_SYSCALL_ARG3(®s->regs), |
| 345 | UPT_SYSCALL_ARG4(®s->regs)); |
| 346 | else |
| 347 | audit_syscall_exit(current, |
| 348 | UPT_SYSCALL_RET(®s->regs)); |
| 349 | } |
| 350 | |
| 351 | /* Fake a debug trap */ |
| 352 | if (is_singlestep) |
| 353 | send_sigtrap(current, regs, 0); |
| 354 | |
| 355 | if (!test_thread_flag(TIF_SYSCALL_TRACE)) |
| 356 | return; |
| 357 | |
| 358 | if (!(current->ptrace & PT_PTRACED)) |
| 359 | return; |
| 360 | |
| 361 | /* the 0x80 provides a way for the tracing parent to distinguish |
| 362 | between a syscall stop and SIGTRAP delivery */ |
| 363 | tracesysgood = (current->ptrace & PT_TRACESYSGOOD); |
| 364 | ptrace_notify(SIGTRAP | (tracesysgood ? 0x80 : 0)); |
| 365 | |
| 366 | if (entryexit) /* force do_signal() --> is_syscall() */ |
| 367 | set_thread_flag(TIF_SIGPENDING); |
| 368 | |
| 369 | /* this isn't the same as continuing with a signal, but it will do |
| 370 | * for normal use. strace only continues with a signal if the |
| 371 | * stopping signal is not SIGTRAP. -brl |
| 372 | */ |
| 373 | if (current->exit_code) { |
| 374 | send_sig(current->exit_code, current, 1); |
| 375 | current->exit_code = 0; |
| 376 | } |
| 377 | } |
| 378 | |
| 379 | /* |
| 380 | * Overrides for Emacs so that we follow Linus's tabbing style. |
| 381 | * Emacs will notice this stuff at the end of the file and automatically |
| 382 | * adjust the settings for this buffer only. This must remain at the end |
| 383 | * of the file. |
| 384 | * --------------------------------------------------------------------------- |
| 385 | * Local variables: |
| 386 | * c-file-style: "linux" |
| 387 | * End: |
| 388 | */ |