Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Kernel support for the ptrace() and syscall tracing interfaces. |
| 3 | * |
| 4 | * Copyright (C) 2000 Hewlett-Packard Co, Linuxcare Inc. |
| 5 | * Copyright (C) 2000 Matthew Wilcox <matthew@wil.cx> |
| 6 | * Copyright (C) 2000 David Huggins-Daines <dhd@debian.org> |
| 7 | */ |
| 8 | |
| 9 | #include <linux/kernel.h> |
| 10 | #include <linux/sched.h> |
| 11 | #include <linux/mm.h> |
| 12 | #include <linux/smp.h> |
| 13 | #include <linux/smp_lock.h> |
| 14 | #include <linux/errno.h> |
| 15 | #include <linux/ptrace.h> |
| 16 | #include <linux/user.h> |
| 17 | #include <linux/personality.h> |
| 18 | #include <linux/security.h> |
| 19 | #include <linux/compat.h> |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 20 | #include <linux/signal.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | |
| 22 | #include <asm/uaccess.h> |
| 23 | #include <asm/pgtable.h> |
| 24 | #include <asm/system.h> |
| 25 | #include <asm/processor.h> |
Sam Ravnborg | 0013a85 | 2005-09-09 20:57:26 +0200 | [diff] [blame^] | 26 | #include <asm/asm-offsets.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | |
| 28 | /* PSW bits we allow the debugger to modify */ |
| 29 | #define USER_PSW_BITS (PSW_N | PSW_V | PSW_CB) |
| 30 | |
| 31 | #undef DEBUG_PTRACE |
| 32 | |
| 33 | #ifdef DEBUG_PTRACE |
| 34 | #define DBG(x...) printk(x) |
| 35 | #else |
| 36 | #define DBG(x...) |
| 37 | #endif |
| 38 | |
| 39 | #ifdef __LP64__ |
| 40 | |
| 41 | /* This function is needed to translate 32 bit pt_regs offsets in to |
| 42 | * 64 bit pt_regs offsets. For example, a 32 bit gdb under a 64 bit kernel |
| 43 | * will request offset 12 if it wants gr3, but the lower 32 bits of |
| 44 | * the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4). |
| 45 | * This code relies on a 32 bit pt_regs being comprised of 32 bit values |
| 46 | * except for the fp registers which (a) are 64 bits, and (b) follow |
| 47 | * the gr registers at the start of pt_regs. The 32 bit pt_regs should |
| 48 | * be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[] |
| 49 | * being 64 bit in both cases. |
| 50 | */ |
| 51 | |
| 52 | static long translate_usr_offset(long offset) |
| 53 | { |
| 54 | if (offset < 0) |
| 55 | return -1; |
| 56 | else if (offset <= 32*4) /* gr[0..31] */ |
| 57 | return offset * 2 + 4; |
| 58 | else if (offset <= 32*4+32*8) /* gr[0..31] + fr[0..31] */ |
| 59 | return offset + 32*4; |
| 60 | else if (offset < sizeof(struct pt_regs)/2 + 32*4) |
| 61 | return offset * 2 + 4 - 32*8; |
| 62 | else |
| 63 | return -1; |
| 64 | } |
| 65 | #endif |
| 66 | |
| 67 | /* |
| 68 | * Called by kernel/ptrace.c when detaching.. |
| 69 | * |
| 70 | * Make sure single step bits etc are not set. |
| 71 | */ |
| 72 | void ptrace_disable(struct task_struct *child) |
| 73 | { |
| 74 | /* make sure the trap bits are not set */ |
| 75 | pa_psw(child)->r = 0; |
| 76 | pa_psw(child)->t = 0; |
| 77 | pa_psw(child)->h = 0; |
| 78 | pa_psw(child)->l = 0; |
| 79 | } |
| 80 | |
| 81 | long sys_ptrace(long request, pid_t pid, long addr, long data) |
| 82 | { |
| 83 | struct task_struct *child; |
| 84 | long ret; |
| 85 | #ifdef DEBUG_PTRACE |
| 86 | long oaddr=addr, odata=data; |
| 87 | #endif |
| 88 | |
| 89 | lock_kernel(); |
| 90 | ret = -EPERM; |
| 91 | if (request == PTRACE_TRACEME) { |
| 92 | /* are we already being traced? */ |
| 93 | if (current->ptrace & PT_PTRACED) |
| 94 | goto out; |
| 95 | |
| 96 | ret = security_ptrace(current->parent, current); |
| 97 | if (ret) |
| 98 | goto out; |
| 99 | |
| 100 | /* set the ptrace bit in the process flags. */ |
| 101 | current->ptrace |= PT_PTRACED; |
| 102 | ret = 0; |
| 103 | goto out; |
| 104 | } |
| 105 | |
| 106 | ret = -ESRCH; |
| 107 | read_lock(&tasklist_lock); |
| 108 | child = find_task_by_pid(pid); |
| 109 | if (child) |
| 110 | get_task_struct(child); |
| 111 | read_unlock(&tasklist_lock); |
| 112 | if (!child) |
| 113 | goto out; |
| 114 | ret = -EPERM; |
| 115 | if (pid == 1) /* no messing around with init! */ |
| 116 | goto out_tsk; |
| 117 | |
| 118 | if (request == PTRACE_ATTACH) { |
| 119 | ret = ptrace_attach(child); |
| 120 | goto out_tsk; |
| 121 | } |
| 122 | |
| 123 | ret = ptrace_check_attach(child, request == PTRACE_KILL); |
| 124 | if (ret < 0) |
| 125 | goto out_tsk; |
| 126 | |
| 127 | switch (request) { |
| 128 | case PTRACE_PEEKTEXT: /* read word at location addr. */ |
| 129 | case PTRACE_PEEKDATA: { |
| 130 | int copied; |
| 131 | |
| 132 | #ifdef __LP64__ |
| 133 | if (is_compat_task(child)) { |
| 134 | unsigned int tmp; |
| 135 | |
| 136 | addr &= 0xffffffffL; |
| 137 | copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0); |
| 138 | ret = -EIO; |
| 139 | if (copied != sizeof(tmp)) |
| 140 | goto out_tsk; |
| 141 | ret = put_user(tmp,(unsigned int *) data); |
| 142 | DBG("sys_ptrace(PEEK%s, %d, %lx, %lx) returning %ld, data %x\n", |
| 143 | request == PTRACE_PEEKTEXT ? "TEXT" : "DATA", |
| 144 | pid, oaddr, odata, ret, tmp); |
| 145 | } |
| 146 | else |
| 147 | #endif |
| 148 | { |
| 149 | unsigned long tmp; |
| 150 | |
| 151 | copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0); |
| 152 | ret = -EIO; |
| 153 | if (copied != sizeof(tmp)) |
| 154 | goto out_tsk; |
| 155 | ret = put_user(tmp,(unsigned long *) data); |
| 156 | } |
| 157 | goto out_tsk; |
| 158 | } |
| 159 | |
| 160 | /* when I and D space are separate, this will have to be fixed. */ |
| 161 | case PTRACE_POKETEXT: /* write the word at location addr. */ |
| 162 | case PTRACE_POKEDATA: |
| 163 | ret = 0; |
| 164 | #ifdef __LP64__ |
| 165 | if (is_compat_task(child)) { |
| 166 | unsigned int tmp = (unsigned int)data; |
| 167 | DBG("sys_ptrace(POKE%s, %d, %lx, %lx)\n", |
| 168 | request == PTRACE_POKETEXT ? "TEXT" : "DATA", |
| 169 | pid, oaddr, odata); |
| 170 | addr &= 0xffffffffL; |
| 171 | if (access_process_vm(child, addr, &tmp, sizeof(tmp), 1) == sizeof(tmp)) |
| 172 | goto out_tsk; |
| 173 | } |
| 174 | else |
| 175 | #endif |
| 176 | { |
| 177 | if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data)) |
| 178 | goto out_tsk; |
| 179 | } |
| 180 | ret = -EIO; |
| 181 | goto out_tsk; |
| 182 | |
| 183 | /* Read the word at location addr in the USER area. For ptraced |
| 184 | processes, the kernel saves all regs on a syscall. */ |
| 185 | case PTRACE_PEEKUSR: { |
| 186 | ret = -EIO; |
| 187 | #ifdef __LP64__ |
| 188 | if (is_compat_task(child)) { |
| 189 | unsigned int tmp; |
| 190 | |
| 191 | if (addr & (sizeof(int)-1)) |
| 192 | goto out_tsk; |
| 193 | if ((addr = translate_usr_offset(addr)) < 0) |
| 194 | goto out_tsk; |
| 195 | |
| 196 | tmp = *(unsigned int *) ((char *) task_regs(child) + addr); |
| 197 | ret = put_user(tmp, (unsigned int *) data); |
| 198 | DBG("sys_ptrace(PEEKUSR, %d, %lx, %lx) returning %ld, addr %lx, data %x\n", |
| 199 | pid, oaddr, odata, ret, addr, tmp); |
| 200 | } |
| 201 | else |
| 202 | #endif |
| 203 | { |
| 204 | unsigned long tmp; |
| 205 | |
| 206 | if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs)) |
| 207 | goto out_tsk; |
| 208 | tmp = *(unsigned long *) ((char *) task_regs(child) + addr); |
| 209 | ret = put_user(tmp, (unsigned long *) data); |
| 210 | } |
| 211 | goto out_tsk; |
| 212 | } |
| 213 | |
| 214 | /* Write the word at location addr in the USER area. This will need |
| 215 | to change when the kernel no longer saves all regs on a syscall. |
| 216 | FIXME. There is a problem at the moment in that r3-r18 are only |
| 217 | saved if the process is ptraced on syscall entry, and even then |
| 218 | those values are overwritten by actual register values on syscall |
| 219 | exit. */ |
| 220 | case PTRACE_POKEUSR: |
| 221 | ret = -EIO; |
| 222 | /* Some register values written here may be ignored in |
| 223 | * entry.S:syscall_restore_rfi; e.g. iaoq is written with |
| 224 | * r31/r31+4, and not with the values in pt_regs. |
| 225 | */ |
| 226 | /* PT_PSW=0, so this is valid for 32 bit processes under 64 |
| 227 | * bit kernels. |
| 228 | */ |
| 229 | if (addr == PT_PSW) { |
| 230 | /* PT_PSW=0, so this is valid for 32 bit processes |
| 231 | * under 64 bit kernels. |
| 232 | * |
| 233 | * Allow writing to Nullify, Divide-step-correction, |
| 234 | * and carry/borrow bits. |
| 235 | * BEWARE, if you set N, and then single step, it won't |
| 236 | * stop on the nullified instruction. |
| 237 | */ |
| 238 | DBG("sys_ptrace(POKEUSR, %d, %lx, %lx)\n", |
| 239 | pid, oaddr, odata); |
| 240 | data &= USER_PSW_BITS; |
| 241 | task_regs(child)->gr[0] &= ~USER_PSW_BITS; |
| 242 | task_regs(child)->gr[0] |= data; |
| 243 | ret = 0; |
| 244 | goto out_tsk; |
| 245 | } |
| 246 | #ifdef __LP64__ |
| 247 | if (is_compat_task(child)) { |
| 248 | if (addr & (sizeof(int)-1)) |
| 249 | goto out_tsk; |
| 250 | if ((addr = translate_usr_offset(addr)) < 0) |
| 251 | goto out_tsk; |
| 252 | DBG("sys_ptrace(POKEUSR, %d, %lx, %lx) addr %lx\n", |
| 253 | pid, oaddr, odata, addr); |
| 254 | if (addr >= PT_FR0 && addr <= PT_FR31 + 4) { |
| 255 | /* Special case, fp regs are 64 bits anyway */ |
| 256 | *(unsigned int *) ((char *) task_regs(child) + addr) = data; |
| 257 | ret = 0; |
| 258 | } |
| 259 | else if ((addr >= PT_GR1+4 && addr <= PT_GR31+4) || |
| 260 | addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4 || |
| 261 | addr == PT_SAR+4) { |
| 262 | /* Zero the top 32 bits */ |
| 263 | *(unsigned int *) ((char *) task_regs(child) + addr - 4) = 0; |
| 264 | *(unsigned int *) ((char *) task_regs(child) + addr) = data; |
| 265 | ret = 0; |
| 266 | } |
| 267 | goto out_tsk; |
| 268 | } |
| 269 | else |
| 270 | #endif |
| 271 | { |
| 272 | if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs)) |
| 273 | goto out_tsk; |
| 274 | if ((addr >= PT_GR1 && addr <= PT_GR31) || |
| 275 | addr == PT_IAOQ0 || addr == PT_IAOQ1 || |
| 276 | (addr >= PT_FR0 && addr <= PT_FR31 + 4) || |
| 277 | addr == PT_SAR) { |
| 278 | *(unsigned long *) ((char *) task_regs(child) + addr) = data; |
| 279 | ret = 0; |
| 280 | } |
| 281 | goto out_tsk; |
| 282 | } |
| 283 | |
| 284 | case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */ |
| 285 | case PTRACE_CONT: |
| 286 | ret = -EIO; |
| 287 | DBG("sys_ptrace(%s)\n", |
| 288 | request == PTRACE_SYSCALL ? "SYSCALL" : "CONT"); |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 289 | if (!valid_signal(data)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | goto out_tsk; |
| 291 | child->ptrace &= ~(PT_SINGLESTEP|PT_BLOCKSTEP); |
| 292 | if (request == PTRACE_SYSCALL) { |
| 293 | set_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 294 | } else { |
| 295 | clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 296 | } |
| 297 | child->exit_code = data; |
| 298 | goto out_wake_notrap; |
| 299 | |
| 300 | case PTRACE_KILL: |
| 301 | /* |
| 302 | * make the child exit. Best I can do is send it a |
| 303 | * sigkill. perhaps it should be put in the status |
| 304 | * that it wants to exit. |
| 305 | */ |
| 306 | DBG("sys_ptrace(KILL)\n"); |
| 307 | if (child->exit_state == EXIT_ZOMBIE) /* already dead */ |
| 308 | goto out_tsk; |
| 309 | child->exit_code = SIGKILL; |
| 310 | goto out_wake_notrap; |
| 311 | |
| 312 | case PTRACE_SINGLEBLOCK: |
| 313 | DBG("sys_ptrace(SINGLEBLOCK)\n"); |
| 314 | ret = -EIO; |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 315 | if (!valid_signal(data)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | goto out_tsk; |
| 317 | clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 318 | child->ptrace &= ~PT_SINGLESTEP; |
| 319 | child->ptrace |= PT_BLOCKSTEP; |
| 320 | child->exit_code = data; |
| 321 | |
| 322 | /* Enable taken branch trap. */ |
| 323 | pa_psw(child)->r = 0; |
| 324 | pa_psw(child)->t = 1; |
| 325 | pa_psw(child)->h = 0; |
| 326 | pa_psw(child)->l = 0; |
| 327 | goto out_wake; |
| 328 | |
| 329 | case PTRACE_SINGLESTEP: |
| 330 | DBG("sys_ptrace(SINGLESTEP)\n"); |
| 331 | ret = -EIO; |
Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 332 | if (!valid_signal(data)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | goto out_tsk; |
| 334 | |
| 335 | clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); |
| 336 | child->ptrace &= ~PT_BLOCKSTEP; |
| 337 | child->ptrace |= PT_SINGLESTEP; |
| 338 | child->exit_code = data; |
| 339 | |
| 340 | if (pa_psw(child)->n) { |
| 341 | struct siginfo si; |
| 342 | |
| 343 | /* Nullified, just crank over the queue. */ |
| 344 | task_regs(child)->iaoq[0] = task_regs(child)->iaoq[1]; |
| 345 | task_regs(child)->iasq[0] = task_regs(child)->iasq[1]; |
| 346 | task_regs(child)->iaoq[1] = task_regs(child)->iaoq[0] + 4; |
| 347 | pa_psw(child)->n = 0; |
| 348 | pa_psw(child)->x = 0; |
| 349 | pa_psw(child)->y = 0; |
| 350 | pa_psw(child)->z = 0; |
| 351 | pa_psw(child)->b = 0; |
| 352 | ptrace_disable(child); |
| 353 | /* Don't wake up the child, but let the |
| 354 | parent know something happened. */ |
| 355 | si.si_code = TRAP_TRACE; |
| 356 | si.si_addr = (void __user *) (task_regs(child)->iaoq[0] & ~3); |
| 357 | si.si_signo = SIGTRAP; |
| 358 | si.si_errno = 0; |
| 359 | force_sig_info(SIGTRAP, &si, child); |
| 360 | //notify_parent(child, SIGCHLD); |
| 361 | //ret = 0; |
| 362 | goto out_wake; |
| 363 | } |
| 364 | |
| 365 | /* Enable recovery counter traps. The recovery counter |
| 366 | * itself will be set to zero on a task switch. If the |
| 367 | * task is suspended on a syscall then the syscall return |
| 368 | * path will overwrite the recovery counter with a suitable |
| 369 | * value such that it traps once back in user space. We |
| 370 | * disable interrupts in the childs PSW here also, to avoid |
| 371 | * interrupts while the recovery counter is decrementing. |
| 372 | */ |
| 373 | pa_psw(child)->r = 1; |
| 374 | pa_psw(child)->t = 0; |
| 375 | pa_psw(child)->h = 0; |
| 376 | pa_psw(child)->l = 0; |
| 377 | /* give it a chance to run. */ |
| 378 | goto out_wake; |
| 379 | |
| 380 | case PTRACE_DETACH: |
| 381 | ret = ptrace_detach(child, data); |
| 382 | goto out_tsk; |
| 383 | |
| 384 | case PTRACE_GETEVENTMSG: |
| 385 | ret = put_user(child->ptrace_message, (unsigned int __user *) data); |
| 386 | goto out_tsk; |
| 387 | |
| 388 | default: |
| 389 | ret = ptrace_request(child, request, addr, data); |
| 390 | goto out_tsk; |
| 391 | } |
| 392 | |
| 393 | out_wake_notrap: |
| 394 | ptrace_disable(child); |
| 395 | out_wake: |
| 396 | wake_up_process(child); |
| 397 | ret = 0; |
| 398 | out_tsk: |
| 399 | put_task_struct(child); |
| 400 | out: |
| 401 | unlock_kernel(); |
| 402 | DBG("sys_ptrace(%ld, %d, %lx, %lx) returning %ld\n", |
| 403 | request, pid, oaddr, odata, ret); |
| 404 | return ret; |
| 405 | } |
| 406 | |
| 407 | void syscall_trace(void) |
| 408 | { |
| 409 | if (!test_thread_flag(TIF_SYSCALL_TRACE)) |
| 410 | return; |
| 411 | if (!(current->ptrace & PT_PTRACED)) |
| 412 | return; |
| 413 | ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) |
| 414 | ? 0x80 : 0)); |
| 415 | /* |
| 416 | * this isn't the same as continuing with a signal, but it will do |
| 417 | * for normal use. strace only continues with a signal if the |
| 418 | * stopping signal is not SIGTRAP. -brl |
| 419 | */ |
| 420 | if (current->exit_code) { |
| 421 | send_sig(current->exit_code, current, 1); |
| 422 | current->exit_code = 0; |
| 423 | } |
| 424 | } |