]> CyberLeo.Net >> Repos - FreeBSD/FreeBSD.git/blob - sys/kern/sys_process.c
Merge from head
[FreeBSD/FreeBSD.git] / sys / kern / sys_process.c
1 /*-
2  * Copyright (c) 1994, Sean Eric Fagan
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *      This product includes software developed by Sean Eric Fagan.
16  * 4. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_compat.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/syscallsubr.h>
42 #include <sys/sysent.h>
43 #include <sys/sysproto.h>
44 #include <sys/pioctl.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/vnode.h>
48 #include <sys/ptrace.h>
49 #include <sys/rwlock.h>
50 #include <sys/sx.h>
51 #include <sys/malloc.h>
52 #include <sys/signalvar.h>
53
54 #include <machine/reg.h>
55
56 #include <security/audit/audit.h>
57
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_extern.h>
61 #include <vm/vm_map.h>
62 #include <vm/vm_kern.h>
63 #include <vm/vm_object.h>
64 #include <vm/vm_page.h>
65 #include <vm/vm_param.h>
66
67 #ifdef COMPAT_FREEBSD32
68 #include <sys/procfs.h>
69 #include <compat/freebsd32/freebsd32_signal.h>
70
71 struct ptrace_io_desc32 {
72         int             piod_op;
73         uint32_t        piod_offs;
74         uint32_t        piod_addr;
75         uint32_t        piod_len;
76 };
77
78 struct ptrace_vm_entry32 {
79         int             pve_entry;
80         int             pve_timestamp;
81         uint32_t        pve_start;
82         uint32_t        pve_end;
83         uint32_t        pve_offset;
84         u_int           pve_prot;
85         u_int           pve_pathlen;
86         int32_t         pve_fileid;
87         u_int           pve_fsid;
88         uint32_t        pve_path;
89 };
90
91 struct ptrace_lwpinfo32 {
92         lwpid_t pl_lwpid;       /* LWP described. */
93         int     pl_event;       /* Event that stopped the LWP. */
94         int     pl_flags;       /* LWP flags. */
95         sigset_t        pl_sigmask;     /* LWP signal mask */
96         sigset_t        pl_siglist;     /* LWP pending signal */
97         struct siginfo32 pl_siginfo;    /* siginfo for signal */
98         char    pl_tdname[MAXCOMLEN + 1];       /* LWP name. */
99         int     pl_child_pid;           /* New child pid */
100 };
101
102 #endif
103
104 /*
105  * Functions implemented using PROC_ACTION():
106  *
107  * proc_read_regs(proc, regs)
108  *      Get the current user-visible register set from the process
109  *      and copy it into the regs structure (<machine/reg.h>).
110  *      The process is stopped at the time read_regs is called.
111  *
112  * proc_write_regs(proc, regs)
113  *      Update the current register set from the passed in regs
114  *      structure.  Take care to avoid clobbering special CPU
115  *      registers or privileged bits in the PSL.
116  *      Depending on the architecture this may have fix-up work to do,
117  *      especially if the IAR or PCW are modified.
118  *      The process is stopped at the time write_regs is called.
119  *
120  * proc_read_fpregs, proc_write_fpregs
121  *      deal with the floating point register set, otherwise as above.
122  *
123  * proc_read_dbregs, proc_write_dbregs
124  *      deal with the processor debug register set, otherwise as above.
125  *
126  * proc_sstep(proc)
127  *      Arrange for the process to trap after executing a single instruction.
128  */
129
130 #define PROC_ACTION(action) do {                                        \
131         int error;                                                      \
132                                                                         \
133         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
134         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
135                 error = EIO;                                            \
136         else                                                            \
137                 error = (action);                                       \
138         return (error);                                                 \
139 } while(0)
140
141 int
142 proc_read_regs(struct thread *td, struct reg *regs)
143 {
144
145         PROC_ACTION(fill_regs(td, regs));
146 }
147
148 int
149 proc_write_regs(struct thread *td, struct reg *regs)
150 {
151
152         PROC_ACTION(set_regs(td, regs));
153 }
154
155 int
156 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
157 {
158
159         PROC_ACTION(fill_dbregs(td, dbregs));
160 }
161
162 int
163 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
164 {
165
166         PROC_ACTION(set_dbregs(td, dbregs));
167 }
168
169 /*
170  * Ptrace doesn't support fpregs at all, and there are no security holes
171  * or translations for fpregs, so we can just copy them.
172  */
173 int
174 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
175 {
176
177         PROC_ACTION(fill_fpregs(td, fpregs));
178 }
179
180 int
181 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
182 {
183
184         PROC_ACTION(set_fpregs(td, fpregs));
185 }
186
187 #ifdef COMPAT_FREEBSD32
188 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
189 int
190 proc_read_regs32(struct thread *td, struct reg32 *regs32)
191 {
192
193         PROC_ACTION(fill_regs32(td, regs32));
194 }
195
196 int
197 proc_write_regs32(struct thread *td, struct reg32 *regs32)
198 {
199
200         PROC_ACTION(set_regs32(td, regs32));
201 }
202
203 int
204 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
205 {
206
207         PROC_ACTION(fill_dbregs32(td, dbregs32));
208 }
209
210 int
211 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
212 {
213
214         PROC_ACTION(set_dbregs32(td, dbregs32));
215 }
216
217 int
218 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
219 {
220
221         PROC_ACTION(fill_fpregs32(td, fpregs32));
222 }
223
224 int
225 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
226 {
227
228         PROC_ACTION(set_fpregs32(td, fpregs32));
229 }
230 #endif
231
232 int
233 proc_sstep(struct thread *td)
234 {
235
236         PROC_ACTION(ptrace_single_step(td));
237 }
238
239 int
240 proc_rwmem(struct proc *p, struct uio *uio)
241 {
242         vm_map_t map;
243         vm_offset_t pageno;             /* page number */
244         vm_prot_t reqprot;
245         int error, fault_flags, page_offset, writing;
246
247         /*
248          * Assert that someone has locked this vmspace.  (Should be
249          * curthread but we can't assert that.)  This keeps the process
250          * from exiting out from under us until this operation completes.
251          */
252         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
253             p, p->p_pid));
254
255         /*
256          * The map we want...
257          */
258         map = &p->p_vmspace->vm_map;
259
260         /*
261          * If we are writing, then we request vm_fault() to create a private
262          * copy of each page.  Since these copies will not be writeable by the
263          * process, we must explicity request that they be dirtied.
264          */
265         writing = uio->uio_rw == UIO_WRITE;
266         reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
267         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
268
269         /*
270          * Only map in one page at a time.  We don't have to, but it
271          * makes things easier.  This way is trivial - right?
272          */
273         do {
274                 vm_offset_t uva;
275                 u_int len;
276                 vm_page_t m;
277
278                 uva = (vm_offset_t)uio->uio_offset;
279
280                 /*
281                  * Get the page number of this segment.
282                  */
283                 pageno = trunc_page(uva);
284                 page_offset = uva - pageno;
285
286                 /*
287                  * How many bytes to copy
288                  */
289                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
290
291                 /*
292                  * Fault and hold the page on behalf of the process.
293                  */
294                 error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
295                 if (error != KERN_SUCCESS) {
296                         if (error == KERN_RESOURCE_SHORTAGE)
297                                 error = ENOMEM;
298                         else
299                                 error = EFAULT;
300                         break;
301                 }
302
303                 /*
304                  * Now do the i/o move.
305                  */
306                 error = uiomove_fromphys(&m, page_offset, len, uio);
307
308                 /* Make the I-cache coherent for breakpoints. */
309                 if (writing && error == 0) {
310                         vm_map_lock_read(map);
311                         if (vm_map_check_protection(map, pageno, pageno +
312                             PAGE_SIZE, VM_PROT_EXECUTE))
313                                 vm_sync_icache(map, uva, len);
314                         vm_map_unlock_read(map);
315                 }
316
317                 /*
318                  * Release the page.
319                  */
320                 vm_page_lock(m);
321                 vm_page_unhold(m);
322                 vm_page_unlock(m);
323
324         } while (error == 0 && uio->uio_resid > 0);
325
326         return (error);
327 }
328
329 static int
330 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
331 {
332         struct vattr vattr;
333         vm_map_t map;
334         vm_map_entry_t entry;
335         vm_object_t obj, tobj, lobj;
336         struct vmspace *vm;
337         struct vnode *vp;
338         char *freepath, *fullpath;
339         u_int pathlen;
340         int error, index;
341
342         error = 0;
343         obj = NULL;
344
345         vm = vmspace_acquire_ref(p);
346         map = &vm->vm_map;
347         vm_map_lock_read(map);
348
349         do {
350                 entry = map->header.next;
351                 index = 0;
352                 while (index < pve->pve_entry && entry != &map->header) {
353                         entry = entry->next;
354                         index++;
355                 }
356                 if (index != pve->pve_entry) {
357                         error = EINVAL;
358                         break;
359                 }
360                 while (entry != &map->header &&
361                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
362                         entry = entry->next;
363                         index++;
364                 }
365                 if (entry == &map->header) {
366                         error = ENOENT;
367                         break;
368                 }
369
370                 /* We got an entry. */
371                 pve->pve_entry = index + 1;
372                 pve->pve_timestamp = map->timestamp;
373                 pve->pve_start = entry->start;
374                 pve->pve_end = entry->end - 1;
375                 pve->pve_offset = entry->offset;
376                 pve->pve_prot = entry->protection;
377
378                 /* Backing object's path needed? */
379                 if (pve->pve_pathlen == 0)
380                         break;
381
382                 pathlen = pve->pve_pathlen;
383                 pve->pve_pathlen = 0;
384
385                 obj = entry->object.vm_object;
386                 if (obj != NULL)
387                         VM_OBJECT_RLOCK(obj);
388         } while (0);
389
390         vm_map_unlock_read(map);
391         vmspace_free(vm);
392
393         pve->pve_fsid = VNOVAL;
394         pve->pve_fileid = VNOVAL;
395
396         if (error == 0 && obj != NULL) {
397                 lobj = obj;
398                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
399                         if (tobj != obj)
400                                 VM_OBJECT_RLOCK(tobj);
401                         if (lobj != obj)
402                                 VM_OBJECT_RUNLOCK(lobj);
403                         lobj = tobj;
404                         pve->pve_offset += tobj->backing_object_offset;
405                 }
406                 vp = vm_object_vnode(lobj);
407                 if (vp != NULL)
408                         vref(vp);
409                 if (lobj != obj)
410                         VM_OBJECT_RUNLOCK(lobj);
411                 VM_OBJECT_RUNLOCK(obj);
412
413                 if (vp != NULL) {
414                         freepath = NULL;
415                         fullpath = NULL;
416                         vn_fullpath(td, vp, &fullpath, &freepath);
417                         vn_lock(vp, LK_SHARED | LK_RETRY);
418                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
419                                 pve->pve_fileid = vattr.va_fileid;
420                                 pve->pve_fsid = vattr.va_fsid;
421                         }
422                         vput(vp);
423
424                         if (fullpath != NULL) {
425                                 pve->pve_pathlen = strlen(fullpath) + 1;
426                                 if (pve->pve_pathlen <= pathlen) {
427                                         error = copyout(fullpath, pve->pve_path,
428                                             pve->pve_pathlen);
429                                 } else
430                                         error = ENAMETOOLONG;
431                         }
432                         if (freepath != NULL)
433                                 free(freepath, M_TEMP);
434                 }
435         }
436         if (error == 0)
437                 CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
438                     p->p_pid, pve->pve_entry, pve->pve_start);
439
440         return (error);
441 }
442
443 #ifdef COMPAT_FREEBSD32
444 static int      
445 ptrace_vm_entry32(struct thread *td, struct proc *p,
446     struct ptrace_vm_entry32 *pve32)
447 {
448         struct ptrace_vm_entry pve;
449         int error;
450
451         pve.pve_entry = pve32->pve_entry;
452         pve.pve_pathlen = pve32->pve_pathlen;
453         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
454
455         error = ptrace_vm_entry(td, p, &pve);
456         if (error == 0) {
457                 pve32->pve_entry = pve.pve_entry;
458                 pve32->pve_timestamp = pve.pve_timestamp;
459                 pve32->pve_start = pve.pve_start;
460                 pve32->pve_end = pve.pve_end;
461                 pve32->pve_offset = pve.pve_offset;
462                 pve32->pve_prot = pve.pve_prot;
463                 pve32->pve_fileid = pve.pve_fileid;
464                 pve32->pve_fsid = pve.pve_fsid;
465         }
466
467         pve32->pve_pathlen = pve.pve_pathlen;
468         return (error);
469 }
470
471 static void
472 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
473     struct ptrace_lwpinfo32 *pl32)
474 {
475
476         pl32->pl_lwpid = pl->pl_lwpid;
477         pl32->pl_event = pl->pl_event;
478         pl32->pl_flags = pl->pl_flags;
479         pl32->pl_sigmask = pl->pl_sigmask;
480         pl32->pl_siglist = pl->pl_siglist;
481         siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
482         strcpy(pl32->pl_tdname, pl->pl_tdname);
483         pl32->pl_child_pid = pl->pl_child_pid;
484 }
485 #endif /* COMPAT_FREEBSD32 */
486
487 /*
488  * Process debugging system call.
489  */
490 #ifndef _SYS_SYSPROTO_H_
491 struct ptrace_args {
492         int     req;
493         pid_t   pid;
494         caddr_t addr;
495         int     data;
496 };
497 #endif
498
499 #ifdef COMPAT_FREEBSD32
500 /*
501  * This CPP subterfuge is to try and reduce the number of ifdefs in
502  * the body of the code.
503  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
504  * becomes either:
505  *   copyin(uap->addr, &r.reg, sizeof r.reg);
506  * or
507  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
508  * .. except this is done at runtime.
509  */
510 #define COPYIN(u, k, s)         wrap32 ? \
511         copyin(u, k ## 32, s ## 32) : \
512         copyin(u, k, s)
513 #define COPYOUT(k, u, s)        wrap32 ? \
514         copyout(k ## 32, u, s ## 32) : \
515         copyout(k, u, s)
516 #else
517 #define COPYIN(u, k, s)         copyin(u, k, s)
518 #define COPYOUT(k, u, s)        copyout(k, u, s)
519 #endif
520 int
521 sys_ptrace(struct thread *td, struct ptrace_args *uap)
522 {
523         /*
524          * XXX this obfuscation is to reduce stack usage, but the register
525          * structs may be too large to put on the stack anyway.
526          */
527         union {
528                 struct ptrace_io_desc piod;
529                 struct ptrace_lwpinfo pl;
530                 struct ptrace_vm_entry pve;
531                 struct dbreg dbreg;
532                 struct fpreg fpreg;
533                 struct reg reg;
534 #ifdef COMPAT_FREEBSD32
535                 struct dbreg32 dbreg32;
536                 struct fpreg32 fpreg32;
537                 struct reg32 reg32;
538                 struct ptrace_io_desc32 piod32;
539                 struct ptrace_lwpinfo32 pl32;
540                 struct ptrace_vm_entry32 pve32;
541 #endif
542         } r;
543         void *addr;
544         int error = 0;
545 #ifdef COMPAT_FREEBSD32
546         int wrap32 = 0;
547
548         if (SV_CURPROC_FLAG(SV_ILP32))
549                 wrap32 = 1;
550 #endif
551         AUDIT_ARG_PID(uap->pid);
552         AUDIT_ARG_CMD(uap->req);
553         AUDIT_ARG_VALUE(uap->data);
554         addr = &r;
555         switch (uap->req) {
556         case PT_GETREGS:
557         case PT_GETFPREGS:
558         case PT_GETDBREGS:
559         case PT_LWPINFO:
560                 break;
561         case PT_SETREGS:
562                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
563                 break;
564         case PT_SETFPREGS:
565                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
566                 break;
567         case PT_SETDBREGS:
568                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
569                 break;
570         case PT_IO:
571                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
572                 break;
573         case PT_VM_ENTRY:
574                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
575                 break;
576         default:
577                 addr = uap->addr;
578                 break;
579         }
580         if (error)
581                 return (error);
582
583         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
584         if (error)
585                 return (error);
586
587         switch (uap->req) {
588         case PT_VM_ENTRY:
589                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
590                 break;
591         case PT_IO:
592                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
593                 break;
594         case PT_GETREGS:
595                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
596                 break;
597         case PT_GETFPREGS:
598                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
599                 break;
600         case PT_GETDBREGS:
601                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
602                 break;
603         case PT_LWPINFO:
604                 error = copyout(&r.pl, uap->addr, uap->data);
605                 break;
606         }
607
608         return (error);
609 }
610 #undef COPYIN
611 #undef COPYOUT
612
613 #ifdef COMPAT_FREEBSD32
614 /*
615  *   PROC_READ(regs, td2, addr);
616  * becomes either:
617  *   proc_read_regs(td2, addr);
618  * or
619  *   proc_read_regs32(td2, addr);
620  * .. except this is done at runtime.  There is an additional
621  * complication in that PROC_WRITE disallows 32 bit consumers
622  * from writing to 64 bit address space targets.
623  */
624 #define PROC_READ(w, t, a)      wrap32 ? \
625         proc_read_ ## w ## 32(t, a) : \
626         proc_read_ ## w (t, a)
627 #define PROC_WRITE(w, t, a)     wrap32 ? \
628         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
629         proc_write_ ## w (t, a)
630 #else
631 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
632 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
633 #endif
634
635 int
636 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
637 {
638         struct iovec iov;
639         struct uio uio;
640         struct proc *curp, *p, *pp;
641         struct thread *td2 = NULL, *td3;
642         struct ptrace_io_desc *piod = NULL;
643         struct ptrace_lwpinfo *pl;
644         int error, write, tmp, num;
645         int proctree_locked = 0;
646         lwpid_t tid = 0, *buf;
647 #ifdef COMPAT_FREEBSD32
648         int wrap32 = 0, safe = 0;
649         struct ptrace_io_desc32 *piod32 = NULL;
650         struct ptrace_lwpinfo32 *pl32 = NULL;
651         struct ptrace_lwpinfo plr;
652 #endif
653
654         curp = td->td_proc;
655
656         /* Lock proctree before locking the process. */
657         switch (req) {
658         case PT_TRACE_ME:
659         case PT_ATTACH:
660         case PT_STEP:
661         case PT_CONTINUE:
662         case PT_TO_SCE:
663         case PT_TO_SCX:
664         case PT_SYSCALL:
665         case PT_FOLLOW_FORK:
666         case PT_DETACH:
667                 sx_xlock(&proctree_lock);
668                 proctree_locked = 1;
669                 break;
670         default:
671                 break;
672         }
673
674         write = 0;
675         if (req == PT_TRACE_ME) {
676                 p = td->td_proc;
677                 PROC_LOCK(p);
678         } else {
679                 if (pid <= PID_MAX) {
680                         if ((p = pfind(pid)) == NULL) {
681                                 if (proctree_locked)
682                                         sx_xunlock(&proctree_lock);
683                                 return (ESRCH);
684                         }
685                 } else {
686                         td2 = tdfind(pid, -1);
687                         if (td2 == NULL) {
688                                 if (proctree_locked)
689                                         sx_xunlock(&proctree_lock);
690                                 return (ESRCH);
691                         }
692                         p = td2->td_proc;
693                         tid = pid;
694                         pid = p->p_pid;
695                 }
696         }
697         AUDIT_ARG_PROCESS(p);
698
699         if ((p->p_flag & P_WEXIT) != 0) {
700                 error = ESRCH;
701                 goto fail;
702         }
703         if ((error = p_cansee(td, p)) != 0)
704                 goto fail;
705
706         if ((error = p_candebug(td, p)) != 0)
707                 goto fail;
708
709         /*
710          * System processes can't be debugged.
711          */
712         if ((p->p_flag & P_SYSTEM) != 0) {
713                 error = EINVAL;
714                 goto fail;
715         }
716
717         if (tid == 0) {
718                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
719                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
720                         td2 = p->p_xthread;
721                 } else {
722                         td2 = FIRST_THREAD_IN_PROC(p);
723                 }
724                 tid = td2->td_tid;
725         }
726
727 #ifdef COMPAT_FREEBSD32
728         /*
729          * Test if we're a 32 bit client and what the target is.
730          * Set the wrap controls accordingly.
731          */
732         if (SV_CURPROC_FLAG(SV_ILP32)) {
733                 if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
734                         safe = 1;
735                 wrap32 = 1;
736         }
737 #endif
738         /*
739          * Permissions check
740          */
741         switch (req) {
742         case PT_TRACE_ME:
743                 /* Always legal. */
744                 break;
745
746         case PT_ATTACH:
747                 /* Self */
748                 if (p->p_pid == td->td_proc->p_pid) {
749                         error = EINVAL;
750                         goto fail;
751                 }
752
753                 /* Already traced */
754                 if (p->p_flag & P_TRACED) {
755                         error = EBUSY;
756                         goto fail;
757                 }
758
759                 /* Can't trace an ancestor if you're being traced. */
760                 if (curp->p_flag & P_TRACED) {
761                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
762                                 if (pp == p) {
763                                         error = EINVAL;
764                                         goto fail;
765                                 }
766                         }
767                 }
768
769
770                 /* OK */
771                 break;
772
773         case PT_CLEARSTEP:
774                 /* Allow thread to clear single step for itself */
775                 if (td->td_tid == tid)
776                         break;
777
778                 /* FALLTHROUGH */
779         default:
780                 /* not being traced... */
781                 if ((p->p_flag & P_TRACED) == 0) {
782                         error = EPERM;
783                         goto fail;
784                 }
785
786                 /* not being traced by YOU */
787                 if (p->p_pptr != td->td_proc) {
788                         error = EBUSY;
789                         goto fail;
790                 }
791
792                 /* not currently stopped */
793                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
794                     p->p_suspcount != p->p_numthreads  ||
795                     (p->p_flag & P_WAITED) == 0) {
796                         error = EBUSY;
797                         goto fail;
798                 }
799
800                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
801                         static int count = 0;
802                         if (count++ == 0)
803                                 printf("P_STOPPED_TRACE not set.\n");
804                 }
805
806                 /* OK */
807                 break;
808         }
809
810         /* Keep this process around until we finish this request. */
811         _PHOLD(p);
812
813 #ifdef FIX_SSTEP
814         /*
815          * Single step fixup ala procfs
816          */
817         FIX_SSTEP(td2);
818 #endif
819
820         /*
821          * Actually do the requests
822          */
823
824         td->td_retval[0] = 0;
825
826         switch (req) {
827         case PT_TRACE_ME:
828                 /* set my trace flag and "owner" so it can read/write me */
829                 p->p_flag |= P_TRACED;
830                 if (p->p_flag & P_PPWAIT)
831                         p->p_flag |= P_PPTRACE;
832                 p->p_oppid = p->p_pptr->p_pid;
833                 CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
834                 break;
835
836         case PT_ATTACH:
837                 /* security check done above */
838                 /*
839                  * It would be nice if the tracing relationship was separate
840                  * from the parent relationship but that would require
841                  * another set of links in the proc struct or for "wait"
842                  * to scan the entire proc table.  To make life easier,
843                  * we just re-parent the process we're trying to trace.
844                  * The old parent is remembered so we can put things back
845                  * on a "detach".
846                  */
847                 p->p_flag |= P_TRACED;
848                 p->p_oppid = p->p_pptr->p_pid;
849                 if (p->p_pptr != td->td_proc) {
850                         proc_reparent(p, td->td_proc);
851                 }
852                 data = SIGSTOP;
853                 CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
854                     p->p_oppid);
855                 goto sendsig;   /* in PT_CONTINUE below */
856
857         case PT_CLEARSTEP:
858                 CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
859                     p->p_pid);
860                 error = ptrace_clear_single_step(td2);
861                 break;
862
863         case PT_SETSTEP:
864                 CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
865                     p->p_pid);
866                 error = ptrace_single_step(td2);
867                 break;
868
869         case PT_SUSPEND:
870                 CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
871                     p->p_pid);
872                 td2->td_dbgflags |= TDB_SUSPEND;
873                 thread_lock(td2);
874                 td2->td_flags |= TDF_NEEDSUSPCHK;
875                 thread_unlock(td2);
876                 break;
877
878         case PT_RESUME:
879                 CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
880                     p->p_pid);
881                 td2->td_dbgflags &= ~TDB_SUSPEND;
882                 break;
883
884         case PT_FOLLOW_FORK:
885                 CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
886                     p->p_flag & P_FOLLOWFORK ? "enabled" : "disabled",
887                     data ? "enabled" : "disabled");
888                 if (data)
889                         p->p_flag |= P_FOLLOWFORK;
890                 else
891                         p->p_flag &= ~P_FOLLOWFORK;
892                 break;
893
894         case PT_STEP:
895         case PT_CONTINUE:
896         case PT_TO_SCE:
897         case PT_TO_SCX:
898         case PT_SYSCALL:
899         case PT_DETACH:
900                 /* Zero means do not send any signal */
901                 if (data < 0 || data > _SIG_MAXSIG) {
902                         error = EINVAL;
903                         break;
904                 }
905
906                 switch (req) {
907                 case PT_STEP:
908                         CTR2(KTR_PTRACE, "PT_STEP: tid %d (pid %d)",
909                             td2->td_tid, p->p_pid);
910                         error = ptrace_single_step(td2);
911                         if (error)
912                                 goto out;
913                         break;
914                 case PT_CONTINUE:
915                 case PT_TO_SCE:
916                 case PT_TO_SCX:
917                 case PT_SYSCALL:
918                         if (addr != (void *)1) {
919                                 error = ptrace_set_pc(td2,
920                                     (u_long)(uintfptr_t)addr);
921                                 if (error)
922                                         goto out;
923                         }
924                         switch (req) {
925                         case PT_TO_SCE:
926                                 p->p_stops |= S_PT_SCE;
927                                 CTR2(KTR_PTRACE,
928                                     "PT_TO_SCE: pid %d, stops = %#x", p->p_pid,
929                                     p->p_stops);
930                                 break;
931                         case PT_TO_SCX:
932                                 p->p_stops |= S_PT_SCX;
933                                 CTR2(KTR_PTRACE,
934                                     "PT_TO_SCX: pid %d, stops = %#x", p->p_pid,
935                                     p->p_stops);
936                                 break;
937                         case PT_SYSCALL:
938                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
939                                 CTR2(KTR_PTRACE,
940                                     "PT_SYSCALL: pid %d, stops = %#x", p->p_pid,
941                                     p->p_stops);
942                                 break;
943                         case PT_CONTINUE:
944                                 CTR1(KTR_PTRACE,
945                                     "PT_CONTINUE: pid %d", p->p_pid);
946                                 break;
947                         }
948                         break;
949                 case PT_DETACH:
950                         /* reset process parent */
951                         if (p->p_oppid != p->p_pptr->p_pid) {
952                                 PROC_LOCK(p->p_pptr);
953                                 sigqueue_take(p->p_ksi);
954                                 PROC_UNLOCK(p->p_pptr);
955
956                                 pp = proc_realparent(p);
957                                 proc_reparent(p, pp);
958                                 if (pp == initproc)
959                                         p->p_sigparent = SIGCHLD;
960                                 CTR2(KTR_PTRACE,
961                                     "PT_DETACH: pid %d reparented to pid %d",
962                                     p->p_pid, pp->p_pid);
963                         } else
964                                 CTR1(KTR_PTRACE, "PT_DETACH: pid %d", p->p_pid);
965                         p->p_oppid = 0;
966                         p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
967                         p->p_stops = 0;
968
969                         /* should we send SIGCHLD? */
970                         /* childproc_continued(p); */
971                         break;
972                 }
973
974         sendsig:
975                 if (proctree_locked) {
976                         sx_xunlock(&proctree_lock);
977                         proctree_locked = 0;
978                 }
979                 p->p_xsig = data;
980                 p->p_xthread = NULL;
981                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
982                         /* deliver or queue signal */
983                         td2->td_dbgflags &= ~TDB_XSIG;
984                         td2->td_xsig = data;
985
986                         if (req == PT_DETACH) {
987                                 FOREACH_THREAD_IN_PROC(p, td3)
988                                         td3->td_dbgflags &= ~TDB_SUSPEND; 
989                         }
990                         /*
991                          * unsuspend all threads, to not let a thread run,
992                          * you should use PT_SUSPEND to suspend it before
993                          * continuing process.
994                          */
995                         PROC_SLOCK(p);
996                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
997                         thread_unsuspend(p);
998                         PROC_SUNLOCK(p);
999                         if (req == PT_ATTACH)
1000                                 kern_psignal(p, data);
1001                 } else {
1002                         if (data)
1003                                 kern_psignal(p, data);
1004                 }
1005                 break;
1006
1007         case PT_WRITE_I:
1008         case PT_WRITE_D:
1009                 td2->td_dbgflags |= TDB_USERWR;
1010                 write = 1;
1011                 /* FALLTHROUGH */
1012         case PT_READ_I:
1013         case PT_READ_D:
1014                 PROC_UNLOCK(p);
1015                 tmp = 0;
1016                 /* write = 0 set above */
1017                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1018                 iov.iov_len = sizeof(int);
1019                 uio.uio_iov = &iov;
1020                 uio.uio_iovcnt = 1;
1021                 uio.uio_offset = (off_t)(uintptr_t)addr;
1022                 uio.uio_resid = sizeof(int);
1023                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
1024                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1025                 uio.uio_td = td;
1026                 error = proc_rwmem(p, &uio);
1027                 if (uio.uio_resid != 0) {
1028                         /*
1029                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1030                          * so I think write() can bogusly return 0.
1031                          * XXX what happens for short writes?  We don't want
1032                          * to write partial data.
1033                          * XXX proc_rwmem() returns EPERM for other invalid
1034                          * addresses.  Convert this to EINVAL.  Does this
1035                          * clobber returns of EPERM for other reasons?
1036                          */
1037                         if (error == 0 || error == ENOSPC || error == EPERM)
1038                                 error = EINVAL; /* EOF */
1039                 }
1040                 if (!write)
1041                         td->td_retval[0] = tmp;
1042                 if (error == 0) {
1043                         if (write)
1044                                 CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1045                                     p->p_pid, addr, data);
1046                         else
1047                                 CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1048                                     p->p_pid, addr, tmp);
1049                 }
1050                 PROC_LOCK(p);
1051                 break;
1052
1053         case PT_IO:
1054 #ifdef COMPAT_FREEBSD32
1055                 if (wrap32) {
1056                         piod32 = addr;
1057                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1058                         iov.iov_len = piod32->piod_len;
1059                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1060                         uio.uio_resid = piod32->piod_len;
1061                 } else
1062 #endif
1063                 {
1064                         piod = addr;
1065                         iov.iov_base = piod->piod_addr;
1066                         iov.iov_len = piod->piod_len;
1067                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1068                         uio.uio_resid = piod->piod_len;
1069                 }
1070                 uio.uio_iov = &iov;
1071                 uio.uio_iovcnt = 1;
1072                 uio.uio_segflg = UIO_USERSPACE;
1073                 uio.uio_td = td;
1074 #ifdef COMPAT_FREEBSD32
1075                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1076 #else
1077                 tmp = piod->piod_op;
1078 #endif
1079                 switch (tmp) {
1080                 case PIOD_READ_D:
1081                 case PIOD_READ_I:
1082                         CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1083                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1084                         uio.uio_rw = UIO_READ;
1085                         break;
1086                 case PIOD_WRITE_D:
1087                 case PIOD_WRITE_I:
1088                         CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1089                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1090                         td2->td_dbgflags |= TDB_USERWR;
1091                         uio.uio_rw = UIO_WRITE;
1092                         break;
1093                 default:
1094                         error = EINVAL;
1095                         goto out;
1096                 }
1097                 PROC_UNLOCK(p);
1098                 error = proc_rwmem(p, &uio);
1099 #ifdef COMPAT_FREEBSD32
1100                 if (wrap32)
1101                         piod32->piod_len -= uio.uio_resid;
1102                 else
1103 #endif
1104                         piod->piod_len -= uio.uio_resid;
1105                 PROC_LOCK(p);
1106                 break;
1107
1108         case PT_KILL:
1109                 CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1110                 data = SIGKILL;
1111                 goto sendsig;   /* in PT_CONTINUE above */
1112
1113         case PT_SETREGS:
1114                 CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1115                     p->p_pid);
1116                 td2->td_dbgflags |= TDB_USERWR;
1117                 error = PROC_WRITE(regs, td2, addr);
1118                 break;
1119
1120         case PT_GETREGS:
1121                 CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1122                     p->p_pid);
1123                 error = PROC_READ(regs, td2, addr);
1124                 break;
1125
1126         case PT_SETFPREGS:
1127                 CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1128                     p->p_pid);
1129                 td2->td_dbgflags |= TDB_USERWR;
1130                 error = PROC_WRITE(fpregs, td2, addr);
1131                 break;
1132
1133         case PT_GETFPREGS:
1134                 CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1135                     p->p_pid);
1136                 error = PROC_READ(fpregs, td2, addr);
1137                 break;
1138
1139         case PT_SETDBREGS:
1140                 CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1141                     p->p_pid);
1142                 td2->td_dbgflags |= TDB_USERWR;
1143                 error = PROC_WRITE(dbregs, td2, addr);
1144                 break;
1145
1146         case PT_GETDBREGS:
1147                 CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1148                     p->p_pid);
1149                 error = PROC_READ(dbregs, td2, addr);
1150                 break;
1151
1152         case PT_LWPINFO:
1153                 if (data <= 0 ||
1154 #ifdef COMPAT_FREEBSD32
1155                     (!wrap32 && data > sizeof(*pl)) ||
1156                     (wrap32 && data > sizeof(*pl32))) {
1157 #else
1158                     data > sizeof(*pl)) {
1159 #endif
1160                         error = EINVAL;
1161                         break;
1162                 }
1163 #ifdef COMPAT_FREEBSD32
1164                 if (wrap32) {
1165                         pl = &plr;
1166                         pl32 = addr;
1167                 } else
1168 #endif
1169                 pl = addr;
1170                 pl->pl_lwpid = td2->td_tid;
1171                 pl->pl_event = PL_EVENT_NONE;
1172                 pl->pl_flags = 0;
1173                 if (td2->td_dbgflags & TDB_XSIG) {
1174                         pl->pl_event = PL_EVENT_SIGNAL;
1175                         if (td2->td_dbgksi.ksi_signo != 0 &&
1176 #ifdef COMPAT_FREEBSD32
1177                             ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1178                             pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1179                             (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1180                             pl_siginfo) + sizeof(struct siginfo32)))
1181 #else
1182                             data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1183                             + sizeof(pl->pl_siginfo)
1184 #endif
1185                         ){
1186                                 pl->pl_flags |= PL_FLAG_SI;
1187                                 pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1188                         }
1189                 }
1190                 if ((pl->pl_flags & PL_FLAG_SI) == 0)
1191                         bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1192                 if (td2->td_dbgflags & TDB_SCE)
1193                         pl->pl_flags |= PL_FLAG_SCE;
1194                 else if (td2->td_dbgflags & TDB_SCX)
1195                         pl->pl_flags |= PL_FLAG_SCX;
1196                 if (td2->td_dbgflags & TDB_EXEC)
1197                         pl->pl_flags |= PL_FLAG_EXEC;
1198                 if (td2->td_dbgflags & TDB_FORK) {
1199                         pl->pl_flags |= PL_FLAG_FORKED;
1200                         pl->pl_child_pid = td2->td_dbg_forked;
1201                 }
1202                 if (td2->td_dbgflags & TDB_CHILD)
1203                         pl->pl_flags |= PL_FLAG_CHILD;
1204                 pl->pl_sigmask = td2->td_sigmask;
1205                 pl->pl_siglist = td2->td_siglist;
1206                 strcpy(pl->pl_tdname, td2->td_name);
1207 #ifdef COMPAT_FREEBSD32
1208                 if (wrap32)
1209                         ptrace_lwpinfo_to32(pl, pl32);
1210 #endif
1211                 CTR5(KTR_PTRACE,
1212             "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d",
1213                     td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1214                     pl->pl_child_pid);
1215                 break;
1216
1217         case PT_GETNUMLWPS:
1218                 CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1219                     p->p_numthreads);
1220                 td->td_retval[0] = p->p_numthreads;
1221                 break;
1222
1223         case PT_GETLWPLIST:
1224                 CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1225                     p->p_pid, data, p->p_numthreads);
1226                 if (data <= 0) {
1227                         error = EINVAL;
1228                         break;
1229                 }
1230                 num = imin(p->p_numthreads, data);
1231                 PROC_UNLOCK(p);
1232                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1233                 tmp = 0;
1234                 PROC_LOCK(p);
1235                 FOREACH_THREAD_IN_PROC(p, td2) {
1236                         if (tmp >= num)
1237                                 break;
1238                         buf[tmp++] = td2->td_tid;
1239                 }
1240                 PROC_UNLOCK(p);
1241                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1242                 free(buf, M_TEMP);
1243                 if (!error)
1244                         td->td_retval[0] = tmp;
1245                 PROC_LOCK(p);
1246                 break;
1247
1248         case PT_VM_TIMESTAMP:
1249                 CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1250                     p->p_pid, p->p_vmspace->vm_map.timestamp);
1251                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1252                 break;
1253
1254         case PT_VM_ENTRY:
1255                 PROC_UNLOCK(p);
1256 #ifdef COMPAT_FREEBSD32
1257                 if (wrap32)
1258                         error = ptrace_vm_entry32(td, p, addr);
1259                 else
1260 #endif
1261                 error = ptrace_vm_entry(td, p, addr);
1262                 PROC_LOCK(p);
1263                 break;
1264
1265         default:
1266 #ifdef __HAVE_PTRACE_MACHDEP
1267                 if (req >= PT_FIRSTMACH) {
1268                         PROC_UNLOCK(p);
1269                         error = cpu_ptrace(td2, req, addr, data);
1270                         PROC_LOCK(p);
1271                 } else
1272 #endif
1273                         /* Unknown request. */
1274                         error = EINVAL;
1275                 break;
1276         }
1277
1278 out:
1279         /* Drop our hold on this process now that the request has completed. */
1280         _PRELE(p);
1281 fail:
1282         PROC_UNLOCK(p);
1283         if (proctree_locked)
1284                 sx_xunlock(&proctree_lock);
1285         return (error);
1286 }
1287 #undef PROC_READ
1288 #undef PROC_WRITE
1289
1290 /*
1291  * Stop a process because of a debugging event;
1292  * stay stopped until p->p_step is cleared
1293  * (cleared by PIOCCONT in procfs).
1294  */
1295 void
1296 stopevent(struct proc *p, unsigned int event, unsigned int val)
1297 {
1298
1299         PROC_LOCK_ASSERT(p, MA_OWNED);
1300         p->p_step = 1;
1301         CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1302             val);
1303         do {
1304                 if (event != S_EXIT)
1305                         p->p_xsig = val;
1306                 p->p_xthread = NULL;
1307                 p->p_stype = event;     /* Which event caused the stop? */
1308                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1309                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1310         } while (p->p_step);
1311 }