Linux 2.6.31-rc6
[linux-2.6] / kernel / ptrace.c
1 /*
2  * linux/kernel/ptrace.c
3  *
4  * (C) Copyright 1999 Linus Torvalds
5  *
6  * Common interfaces for "ptrace()" which we do not want
7  * to continually duplicate across every architecture.
8  */
9
10 #include <linux/capability.h>
11 #include <linux/module.h>
12 #include <linux/sched.h>
13 #include <linux/errno.h>
14 #include <linux/mm.h>
15 #include <linux/highmem.h>
16 #include <linux/pagemap.h>
17 #include <linux/smp_lock.h>
18 #include <linux/ptrace.h>
19 #include <linux/security.h>
20 #include <linux/signal.h>
21 #include <linux/audit.h>
22 #include <linux/pid_namespace.h>
23 #include <linux/syscalls.h>
24 #include <linux/uaccess.h>
25
26
27 /*
28  * ptrace a task: make the debugger its new parent and
29  * move it to the ptrace list.
30  *
31  * Must be called with the tasklist lock write-held.
32  */
33 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
34 {
35         BUG_ON(!list_empty(&child->ptrace_entry));
36         list_add(&child->ptrace_entry, &new_parent->ptraced);
37         child->parent = new_parent;
38 }
39
40 /*
41  * Turn a tracing stop into a normal stop now, since with no tracer there
42  * would be no way to wake it up with SIGCONT or SIGKILL.  If there was a
43  * signal sent that would resume the child, but didn't because it was in
44  * TASK_TRACED, resume it now.
45  * Requires that irqs be disabled.
46  */
47 static void ptrace_untrace(struct task_struct *child)
48 {
49         spin_lock(&child->sighand->siglock);
50         if (task_is_traced(child)) {
51                 /*
52                  * If the group stop is completed or in progress,
53                  * this thread was already counted as stopped.
54                  */
55                 if (child->signal->flags & SIGNAL_STOP_STOPPED ||
56                     child->signal->group_stop_count)
57                         __set_task_state(child, TASK_STOPPED);
58                 else
59                         signal_wake_up(child, 1);
60         }
61         spin_unlock(&child->sighand->siglock);
62 }
63
64 /*
65  * unptrace a task: move it back to its original parent and
66  * remove it from the ptrace list.
67  *
68  * Must be called with the tasklist lock write-held.
69  */
70 void __ptrace_unlink(struct task_struct *child)
71 {
72         BUG_ON(!child->ptrace);
73
74         child->ptrace = 0;
75         child->parent = child->real_parent;
76         list_del_init(&child->ptrace_entry);
77
78         arch_ptrace_untrace(child);
79         if (task_is_traced(child))
80                 ptrace_untrace(child);
81 }
82
83 /*
84  * Check that we have indeed attached to the thing..
85  */
86 int ptrace_check_attach(struct task_struct *child, int kill)
87 {
88         int ret = -ESRCH;
89
90         /*
91          * We take the read lock around doing both checks to close a
92          * possible race where someone else was tracing our child and
93          * detached between these two checks.  After this locked check,
94          * we are sure that this is our traced child and that can only
95          * be changed by us so it's not changing right after this.
96          */
97         read_lock(&tasklist_lock);
98         if ((child->ptrace & PT_PTRACED) && child->parent == current) {
99                 ret = 0;
100                 /*
101                  * child->sighand can't be NULL, release_task()
102                  * does ptrace_unlink() before __exit_signal().
103                  */
104                 spin_lock_irq(&child->sighand->siglock);
105                 if (task_is_stopped(child))
106                         child->state = TASK_TRACED;
107                 else if (!task_is_traced(child) && !kill)
108                         ret = -ESRCH;
109                 spin_unlock_irq(&child->sighand->siglock);
110         }
111         read_unlock(&tasklist_lock);
112
113         if (!ret && !kill)
114                 ret = wait_task_inactive(child, TASK_TRACED) ? 0 : -ESRCH;
115
116         /* All systems go.. */
117         return ret;
118 }
119
120 int __ptrace_may_access(struct task_struct *task, unsigned int mode)
121 {
122         const struct cred *cred = current_cred(), *tcred;
123
124         /* May we inspect the given task?
125          * This check is used both for attaching with ptrace
126          * and for allowing access to sensitive information in /proc.
127          *
128          * ptrace_attach denies several cases that /proc allows
129          * because setting up the necessary parent/child relationship
130          * or halting the specified task is impossible.
131          */
132         int dumpable = 0;
133         /* Don't let security modules deny introspection */
134         if (task == current)
135                 return 0;
136         rcu_read_lock();
137         tcred = __task_cred(task);
138         if ((cred->uid != tcred->euid ||
139              cred->uid != tcred->suid ||
140              cred->uid != tcred->uid  ||
141              cred->gid != tcred->egid ||
142              cred->gid != tcred->sgid ||
143              cred->gid != tcred->gid) &&
144             !capable(CAP_SYS_PTRACE)) {
145                 rcu_read_unlock();
146                 return -EPERM;
147         }
148         rcu_read_unlock();
149         smp_rmb();
150         if (task->mm)
151                 dumpable = get_dumpable(task->mm);
152         if (!dumpable && !capable(CAP_SYS_PTRACE))
153                 return -EPERM;
154
155         return security_ptrace_may_access(task, mode);
156 }
157
158 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
159 {
160         int err;
161         task_lock(task);
162         err = __ptrace_may_access(task, mode);
163         task_unlock(task);
164         return !err;
165 }
166
167 int ptrace_attach(struct task_struct *task)
168 {
169         int retval;
170
171         audit_ptrace(task);
172
173         retval = -EPERM;
174         if (unlikely(task->flags & PF_KTHREAD))
175                 goto out;
176         if (same_thread_group(task, current))
177                 goto out;
178
179         /*
180          * Protect exec's credential calculations against our interference;
181          * interference; SUID, SGID and LSM creds get determined differently
182          * under ptrace.
183          */
184         retval = -ERESTARTNOINTR;
185         if (mutex_lock_interruptible(&task->cred_guard_mutex))
186                 goto out;
187
188         task_lock(task);
189         retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
190         task_unlock(task);
191         if (retval)
192                 goto unlock_creds;
193
194         write_lock_irq(&tasklist_lock);
195         retval = -EPERM;
196         if (unlikely(task->exit_state))
197                 goto unlock_tasklist;
198         if (task->ptrace)
199                 goto unlock_tasklist;
200
201         task->ptrace = PT_PTRACED;
202         if (capable(CAP_SYS_PTRACE))
203                 task->ptrace |= PT_PTRACE_CAP;
204
205         __ptrace_link(task, current);
206         send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
207
208         retval = 0;
209 unlock_tasklist:
210         write_unlock_irq(&tasklist_lock);
211 unlock_creds:
212         mutex_unlock(&task->cred_guard_mutex);
213 out:
214         return retval;
215 }
216
217 /**
218  * ptrace_traceme  --  helper for PTRACE_TRACEME
219  *
220  * Performs checks and sets PT_PTRACED.
221  * Should be used by all ptrace implementations for PTRACE_TRACEME.
222  */
223 int ptrace_traceme(void)
224 {
225         int ret = -EPERM;
226
227         write_lock_irq(&tasklist_lock);
228         /* Are we already being traced? */
229         if (!current->ptrace) {
230                 ret = security_ptrace_traceme(current->parent);
231                 /*
232                  * Check PF_EXITING to ensure ->real_parent has not passed
233                  * exit_ptrace(). Otherwise we don't report the error but
234                  * pretend ->real_parent untraces us right after return.
235                  */
236                 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
237                         current->ptrace = PT_PTRACED;
238                         __ptrace_link(current, current->real_parent);
239                 }
240         }
241         write_unlock_irq(&tasklist_lock);
242
243         return ret;
244 }
245
246 /*
247  * Called with irqs disabled, returns true if childs should reap themselves.
248  */
249 static int ignoring_children(struct sighand_struct *sigh)
250 {
251         int ret;
252         spin_lock(&sigh->siglock);
253         ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
254               (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
255         spin_unlock(&sigh->siglock);
256         return ret;
257 }
258
259 /*
260  * Called with tasklist_lock held for writing.
261  * Unlink a traced task, and clean it up if it was a traced zombie.
262  * Return true if it needs to be reaped with release_task().
263  * (We can't call release_task() here because we already hold tasklist_lock.)
264  *
265  * If it's a zombie, our attachedness prevented normal parent notification
266  * or self-reaping.  Do notification now if it would have happened earlier.
267  * If it should reap itself, return true.
268  *
269  * If it's our own child, there is no notification to do.
270  * But if our normal children self-reap, then this child
271  * was prevented by ptrace and we must reap it now.
272  */
273 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
274 {
275         __ptrace_unlink(p);
276
277         if (p->exit_state == EXIT_ZOMBIE) {
278                 if (!task_detached(p) && thread_group_empty(p)) {
279                         if (!same_thread_group(p->real_parent, tracer))
280                                 do_notify_parent(p, p->exit_signal);
281                         else if (ignoring_children(tracer->sighand))
282                                 p->exit_signal = -1;
283                 }
284                 if (task_detached(p)) {
285                         /* Mark it as in the process of being reaped. */
286                         p->exit_state = EXIT_DEAD;
287                         return true;
288                 }
289         }
290
291         return false;
292 }
293
294 int ptrace_detach(struct task_struct *child, unsigned int data)
295 {
296         bool dead = false;
297
298         if (!valid_signal(data))
299                 return -EIO;
300
301         /* Architecture-specific hardware disable .. */
302         ptrace_disable(child);
303         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
304
305         write_lock_irq(&tasklist_lock);
306         /*
307          * This child can be already killed. Make sure de_thread() or
308          * our sub-thread doing do_wait() didn't do release_task() yet.
309          */
310         if (child->ptrace) {
311                 child->exit_code = data;
312                 dead = __ptrace_detach(current, child);
313                 if (!child->exit_state)
314                         wake_up_process(child);
315         }
316         write_unlock_irq(&tasklist_lock);
317
318         if (unlikely(dead))
319                 release_task(child);
320
321         return 0;
322 }
323
324 /*
325  * Detach all tasks we were using ptrace on.
326  */
327 void exit_ptrace(struct task_struct *tracer)
328 {
329         struct task_struct *p, *n;
330         LIST_HEAD(ptrace_dead);
331
332         write_lock_irq(&tasklist_lock);
333         list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
334                 if (__ptrace_detach(tracer, p))
335                         list_add(&p->ptrace_entry, &ptrace_dead);
336         }
337         write_unlock_irq(&tasklist_lock);
338
339         BUG_ON(!list_empty(&tracer->ptraced));
340
341         list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
342                 list_del_init(&p->ptrace_entry);
343                 release_task(p);
344         }
345 }
346
347 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
348 {
349         int copied = 0;
350
351         while (len > 0) {
352                 char buf[128];
353                 int this_len, retval;
354
355                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
356                 retval = access_process_vm(tsk, src, buf, this_len, 0);
357                 if (!retval) {
358                         if (copied)
359                                 break;
360                         return -EIO;
361                 }
362                 if (copy_to_user(dst, buf, retval))
363                         return -EFAULT;
364                 copied += retval;
365                 src += retval;
366                 dst += retval;
367                 len -= retval;
368         }
369         return copied;
370 }
371
372 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
373 {
374         int copied = 0;
375
376         while (len > 0) {
377                 char buf[128];
378                 int this_len, retval;
379
380                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
381                 if (copy_from_user(buf, src, this_len))
382                         return -EFAULT;
383                 retval = access_process_vm(tsk, dst, buf, this_len, 1);
384                 if (!retval) {
385                         if (copied)
386                                 break;
387                         return -EIO;
388                 }
389                 copied += retval;
390                 src += retval;
391                 dst += retval;
392                 len -= retval;
393         }
394         return copied;
395 }
396
397 static int ptrace_setoptions(struct task_struct *child, long data)
398 {
399         child->ptrace &= ~PT_TRACE_MASK;
400
401         if (data & PTRACE_O_TRACESYSGOOD)
402                 child->ptrace |= PT_TRACESYSGOOD;
403
404         if (data & PTRACE_O_TRACEFORK)
405                 child->ptrace |= PT_TRACE_FORK;
406
407         if (data & PTRACE_O_TRACEVFORK)
408                 child->ptrace |= PT_TRACE_VFORK;
409
410         if (data & PTRACE_O_TRACECLONE)
411                 child->ptrace |= PT_TRACE_CLONE;
412
413         if (data & PTRACE_O_TRACEEXEC)
414                 child->ptrace |= PT_TRACE_EXEC;
415
416         if (data & PTRACE_O_TRACEVFORKDONE)
417                 child->ptrace |= PT_TRACE_VFORK_DONE;
418
419         if (data & PTRACE_O_TRACEEXIT)
420                 child->ptrace |= PT_TRACE_EXIT;
421
422         return (data & ~PTRACE_O_MASK) ? -EINVAL : 0;
423 }
424
425 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
426 {
427         unsigned long flags;
428         int error = -ESRCH;
429
430         if (lock_task_sighand(child, &flags)) {
431                 error = -EINVAL;
432                 if (likely(child->last_siginfo != NULL)) {
433                         *info = *child->last_siginfo;
434                         error = 0;
435                 }
436                 unlock_task_sighand(child, &flags);
437         }
438         return error;
439 }
440
441 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
442 {
443         unsigned long flags;
444         int error = -ESRCH;
445
446         if (lock_task_sighand(child, &flags)) {
447                 error = -EINVAL;
448                 if (likely(child->last_siginfo != NULL)) {
449                         *child->last_siginfo = *info;
450                         error = 0;
451                 }
452                 unlock_task_sighand(child, &flags);
453         }
454         return error;
455 }
456
457
458 #ifdef PTRACE_SINGLESTEP
459 #define is_singlestep(request)          ((request) == PTRACE_SINGLESTEP)
460 #else
461 #define is_singlestep(request)          0
462 #endif
463
464 #ifdef PTRACE_SINGLEBLOCK
465 #define is_singleblock(request)         ((request) == PTRACE_SINGLEBLOCK)
466 #else
467 #define is_singleblock(request)         0
468 #endif
469
470 #ifdef PTRACE_SYSEMU
471 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
472 #else
473 #define is_sysemu_singlestep(request)   0
474 #endif
475
476 static int ptrace_resume(struct task_struct *child, long request, long data)
477 {
478         if (!valid_signal(data))
479                 return -EIO;
480
481         if (request == PTRACE_SYSCALL)
482                 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
483         else
484                 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
485
486 #ifdef TIF_SYSCALL_EMU
487         if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
488                 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
489         else
490                 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
491 #endif
492
493         if (is_singleblock(request)) {
494                 if (unlikely(!arch_has_block_step()))
495                         return -EIO;
496                 user_enable_block_step(child);
497         } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
498                 if (unlikely(!arch_has_single_step()))
499                         return -EIO;
500                 user_enable_single_step(child);
501         } else {
502                 user_disable_single_step(child);
503         }
504
505         child->exit_code = data;
506         wake_up_process(child);
507
508         return 0;
509 }
510
511 int ptrace_request(struct task_struct *child, long request,
512                    long addr, long data)
513 {
514         int ret = -EIO;
515         siginfo_t siginfo;
516
517         switch (request) {
518         case PTRACE_PEEKTEXT:
519         case PTRACE_PEEKDATA:
520                 return generic_ptrace_peekdata(child, addr, data);
521         case PTRACE_POKETEXT:
522         case PTRACE_POKEDATA:
523                 return generic_ptrace_pokedata(child, addr, data);
524
525 #ifdef PTRACE_OLDSETOPTIONS
526         case PTRACE_OLDSETOPTIONS:
527 #endif
528         case PTRACE_SETOPTIONS:
529                 ret = ptrace_setoptions(child, data);
530                 break;
531         case PTRACE_GETEVENTMSG:
532                 ret = put_user(child->ptrace_message, (unsigned long __user *) data);
533                 break;
534
535         case PTRACE_GETSIGINFO:
536                 ret = ptrace_getsiginfo(child, &siginfo);
537                 if (!ret)
538                         ret = copy_siginfo_to_user((siginfo_t __user *) data,
539                                                    &siginfo);
540                 break;
541
542         case PTRACE_SETSIGINFO:
543                 if (copy_from_user(&siginfo, (siginfo_t __user *) data,
544                                    sizeof siginfo))
545                         ret = -EFAULT;
546                 else
547                         ret = ptrace_setsiginfo(child, &siginfo);
548                 break;
549
550         case PTRACE_DETACH:      /* detach a process that was attached. */
551                 ret = ptrace_detach(child, data);
552                 break;
553
554 #ifdef PTRACE_SINGLESTEP
555         case PTRACE_SINGLESTEP:
556 #endif
557 #ifdef PTRACE_SINGLEBLOCK
558         case PTRACE_SINGLEBLOCK:
559 #endif
560 #ifdef PTRACE_SYSEMU
561         case PTRACE_SYSEMU:
562         case PTRACE_SYSEMU_SINGLESTEP:
563 #endif
564         case PTRACE_SYSCALL:
565         case PTRACE_CONT:
566                 return ptrace_resume(child, request, data);
567
568         case PTRACE_KILL:
569                 if (child->exit_state)  /* already dead */
570                         return 0;
571                 return ptrace_resume(child, request, SIGKILL);
572
573         default:
574                 break;
575         }
576
577         return ret;
578 }
579
580 static struct task_struct *ptrace_get_task_struct(pid_t pid)
581 {
582         struct task_struct *child;
583
584         rcu_read_lock();
585         child = find_task_by_vpid(pid);
586         if (child)
587                 get_task_struct(child);
588         rcu_read_unlock();
589
590         if (!child)
591                 return ERR_PTR(-ESRCH);
592         return child;
593 }
594
595 #ifndef arch_ptrace_attach
596 #define arch_ptrace_attach(child)       do { } while (0)
597 #endif
598
599 SYSCALL_DEFINE4(ptrace, long, request, long, pid, long, addr, long, data)
600 {
601         struct task_struct *child;
602         long ret;
603
604         /*
605          * This lock_kernel fixes a subtle race with suid exec
606          */
607         lock_kernel();
608         if (request == PTRACE_TRACEME) {
609                 ret = ptrace_traceme();
610                 if (!ret)
611                         arch_ptrace_attach(current);
612                 goto out;
613         }
614
615         child = ptrace_get_task_struct(pid);
616         if (IS_ERR(child)) {
617                 ret = PTR_ERR(child);
618                 goto out;
619         }
620
621         if (request == PTRACE_ATTACH) {
622                 ret = ptrace_attach(child);
623                 /*
624                  * Some architectures need to do book-keeping after
625                  * a ptrace attach.
626                  */
627                 if (!ret)
628                         arch_ptrace_attach(child);
629                 goto out_put_task_struct;
630         }
631
632         ret = ptrace_check_attach(child, request == PTRACE_KILL);
633         if (ret < 0)
634                 goto out_put_task_struct;
635
636         ret = arch_ptrace(child, request, addr, data);
637
638  out_put_task_struct:
639         put_task_struct(child);
640  out:
641         unlock_kernel();
642         return ret;
643 }
644
645 int generic_ptrace_peekdata(struct task_struct *tsk, long addr, long data)
646 {
647         unsigned long tmp;
648         int copied;
649
650         copied = access_process_vm(tsk, addr, &tmp, sizeof(tmp), 0);
651         if (copied != sizeof(tmp))
652                 return -EIO;
653         return put_user(tmp, (unsigned long __user *)data);
654 }
655
656 int generic_ptrace_pokedata(struct task_struct *tsk, long addr, long data)
657 {
658         int copied;
659
660         copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
661         return (copied == sizeof(data)) ? 0 : -EIO;
662 }
663
664 #if defined CONFIG_COMPAT
665 #include <linux/compat.h>
666
667 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
668                           compat_ulong_t addr, compat_ulong_t data)
669 {
670         compat_ulong_t __user *datap = compat_ptr(data);
671         compat_ulong_t word;
672         siginfo_t siginfo;
673         int ret;
674
675         switch (request) {
676         case PTRACE_PEEKTEXT:
677         case PTRACE_PEEKDATA:
678                 ret = access_process_vm(child, addr, &word, sizeof(word), 0);
679                 if (ret != sizeof(word))
680                         ret = -EIO;
681                 else
682                         ret = put_user(word, datap);
683                 break;
684
685         case PTRACE_POKETEXT:
686         case PTRACE_POKEDATA:
687                 ret = access_process_vm(child, addr, &data, sizeof(data), 1);
688                 ret = (ret != sizeof(data) ? -EIO : 0);
689                 break;
690
691         case PTRACE_GETEVENTMSG:
692                 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
693                 break;
694
695         case PTRACE_GETSIGINFO:
696                 ret = ptrace_getsiginfo(child, &siginfo);
697                 if (!ret)
698                         ret = copy_siginfo_to_user32(
699                                 (struct compat_siginfo __user *) datap,
700                                 &siginfo);
701                 break;
702
703         case PTRACE_SETSIGINFO:
704                 memset(&siginfo, 0, sizeof siginfo);
705                 if (copy_siginfo_from_user32(
706                             &siginfo, (struct compat_siginfo __user *) datap))
707                         ret = -EFAULT;
708                 else
709                         ret = ptrace_setsiginfo(child, &siginfo);
710                 break;
711
712         default:
713                 ret = ptrace_request(child, request, addr, data);
714         }
715
716         return ret;
717 }
718
719 asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
720                                   compat_long_t addr, compat_long_t data)
721 {
722         struct task_struct *child;
723         long ret;
724
725         /*
726          * This lock_kernel fixes a subtle race with suid exec
727          */
728         lock_kernel();
729         if (request == PTRACE_TRACEME) {
730                 ret = ptrace_traceme();
731                 goto out;
732         }
733
734         child = ptrace_get_task_struct(pid);
735         if (IS_ERR(child)) {
736                 ret = PTR_ERR(child);
737                 goto out;
738         }
739
740         if (request == PTRACE_ATTACH) {
741                 ret = ptrace_attach(child);
742                 /*
743                  * Some architectures need to do book-keeping after
744                  * a ptrace attach.
745                  */
746                 if (!ret)
747                         arch_ptrace_attach(child);
748                 goto out_put_task_struct;
749         }
750
751         ret = ptrace_check_attach(child, request == PTRACE_KILL);
752         if (!ret)
753                 ret = compat_arch_ptrace(child, request, addr, data);
754
755  out_put_task_struct:
756         put_task_struct(child);
757  out:
758         unlock_kernel();
759         return ret;
760 }
761 #endif  /* CONFIG_COMPAT */