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/priv.h>
  45 #include <sys/proc.h>
  46 #include <sys/procctl.h>
  47 #include <sys/vnode.h>
  48 #include <sys/ptrace.h>
  49 #include <sys/sx.h>
  50 #include <sys/malloc.h>
  51 #include <sys/signalvar.h>
  52
  53 #include <machine/reg.h>
  54
  55 #include <security/audit/audit.h>
  56
  57 #include <vm/vm.h>
  58 #include <vm/pmap.h>
  59 #include <vm/vm_extern.h>
  60 #include <vm/vm_map.h>
  61 #include <vm/vm_kern.h>
  62 #include <vm/vm_object.h>
  63 #include <vm/vm_page.h>
  64 #include <vm/vm_pager.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         pid_t   pl_child_pid;           /* New child pid */
 100         u_int           pl_syscall_code;
 101         u_int           pl_syscall_narg;
 102 };
 103
 104 #endif
 105
 106 /*
 107  * Functions implemented using PROC_ACTION():
 108  *
 109  * proc_read_regs(proc, regs)
 110  *      Get the current user-visible register set from the process
 111  *      and copy it into the regs structure (<machine/reg.h>).
 112  *      The process is stopped at the time read_regs is called.
 113  *
 114  * proc_write_regs(proc, regs)
 115  *      Update the current register set from the passed in regs
 116  *      structure.  Take care to avoid clobbering special CPU
 117  *      registers or privileged bits in the PSL.
 118  *      Depending on the architecture this may have fix-up work to do,
 119  *      especially if the IAR or PCW are modified.
 120  *      The process is stopped at the time write_regs is called.
 121  *
 122  * proc_read_fpregs, proc_write_fpregs
 123  *      deal with the floating point register set, otherwise as above.
 124  *
 125  * proc_read_dbregs, proc_write_dbregs
 126  *      deal with the processor debug register set, otherwise as above.
 127  *
 128  * proc_sstep(proc)
 129  *      Arrange for the process to trap after executing a single instruction.
 130  */
 131
 132 #define PROC_ACTION(action) do {                                        \
 133         int error;                                                      \
 134                                                                         \
 135         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 136         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
 137                 error = EIO;                                            \
 138         else                                                            \
 139                 error = (action);                                       \
 140         return (error);                                                 \
 141 } while(0)
 142
 143 int
 144 proc_read_regs(struct thread *td, struct reg *regs)
 145 {
 146
 147         PROC_ACTION(fill_regs(td, regs));
 148 }
 149
 150 int
 151 proc_write_regs(struct thread *td, struct reg *regs)
 152 {
 153
 154         PROC_ACTION(set_regs(td, regs));
 155 }
 156
 157 int
 158 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 159 {
 160
 161         PROC_ACTION(fill_dbregs(td, dbregs));
 162 }
 163
 164 int
 165 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 166 {
 167
 168         PROC_ACTION(set_dbregs(td, dbregs));
 169 }
 170
 171 /*
 172  * Ptrace doesn't support fpregs at all, and there are no security holes
 173  * or translations for fpregs, so we can just copy them.
 174  */
 175 int
 176 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 177 {
 178
 179         PROC_ACTION(fill_fpregs(td, fpregs));
 180 }
 181
 182 int
 183 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 184 {
 185
 186         PROC_ACTION(set_fpregs(td, fpregs));
 187 }
 188
 189 #ifdef COMPAT_FREEBSD32
 190 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 191 int
 192 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 193 {
 194
 195         PROC_ACTION(fill_regs32(td, regs32));
 196 }
 197
 198 int
 199 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 200 {
 201
 202         PROC_ACTION(set_regs32(td, regs32));
 203 }
 204
 205 int
 206 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 207 {
 208
 209         PROC_ACTION(fill_dbregs32(td, dbregs32));
 210 }
 211
 212 int
 213 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 214 {
 215
 216         PROC_ACTION(set_dbregs32(td, dbregs32));
 217 }
 218
 219 int
 220 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 221 {
 222
 223         PROC_ACTION(fill_fpregs32(td, fpregs32));
 224 }
 225
 226 int
 227 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 228 {
 229
 230         PROC_ACTION(set_fpregs32(td, fpregs32));
 231 }
 232 #endif
 233
 234 int
 235 proc_sstep(struct thread *td)
 236 {
 237
 238         PROC_ACTION(ptrace_single_step(td));
 239 }
 240
 241 int
 242 proc_rwmem(struct proc *p, struct uio *uio)
 243 {
 244         vm_map_t map;
 245         vm_offset_t pageno;             /* page number */
 246         vm_prot_t reqprot;
 247         int error, fault_flags, page_offset, writing;
 248
 249         /*
 250          * Assert that someone has locked this vmspace.  (Should be
 251          * curthread but we can't assert that.)  This keeps the process
 252          * from exiting out from under us until this operation completes.
 253          */
 254         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 255             p, p->p_pid));
 256
 257         /*
 258          * The map we want...
 259          */
 260         map = &p->p_vmspace->vm_map;
 261
 262         /*
 263          * If we are writing, then we request vm_fault() to create a private
 264          * copy of each page.  Since these copies will not be writeable by the
 265          * process, we must explicity request that they be dirtied.
 266          */
 267         writing = uio->uio_rw == UIO_WRITE;
 268         reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
 269         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
 270
 271         /*
 272          * Only map in one page at a time.  We don't have to, but it
 273          * makes things easier.  This way is trivial - right?
 274          */
 275         do {
 276                 vm_offset_t uva;
 277                 u_int len;
 278                 vm_page_t m;
 279
 280                 uva = (vm_offset_t)uio->uio_offset;
 281
 282                 /*
 283                  * Get the page number of this segment.
 284                  */
 285                 pageno = trunc_page(uva);
 286                 page_offset = uva - pageno;
 287
 288                 /*
 289                  * How many bytes to copy
 290                  */
 291                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 292
 293                 /*
 294                  * Fault and hold the page on behalf of the process.
 295                  */
 296                 error = vm_fault_hold(map, pageno, reqprot, fault_flags, &m);
 297                 if (error != KERN_SUCCESS) {
 298                         if (error == KERN_RESOURCE_SHORTAGE)
 299                                 error = ENOMEM;
 300                         else
 301                                 error = EFAULT;
 302                         break;
 303                 }
 304
 305                 /*
 306                  * Now do the i/o move.
 307                  */
 308                 error = uiomove_fromphys(&m, page_offset, len, uio);
 309
 310                 /* Make the I-cache coherent for breakpoints. */
 311                 if (writing && error == 0) {
 312                         vm_map_lock_read(map);
 313                         if (vm_map_check_protection(map, pageno, pageno +
 314                             PAGE_SIZE, VM_PROT_EXECUTE))
 315                                 vm_sync_icache(map, uva, len);
 316                         vm_map_unlock_read(map);
 317                 }
 318
 319                 /*
 320                  * Release the page.
 321                  */
 322                 vm_page_lock(m);
 323                 vm_page_unhold(m);
 324                 vm_page_unlock(m);
 325
 326         } while (error == 0 && uio->uio_resid > 0);
 327
 328         return (error);
 329 }
 330
 331 static int
 332 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
 333 {
 334         struct vattr vattr;
 335         vm_map_t map;
 336         vm_map_entry_t entry;
 337         vm_object_t obj, tobj, lobj;
 338         struct vmspace *vm;
 339         struct vnode *vp;
 340         char *freepath, *fullpath;
 341         u_int pathlen;
 342         int error, index, vfslocked;
 343
 344         error = 0;
 345         obj = NULL;
 346
 347         vm = vmspace_acquire_ref(p);
 348         map = &vm->vm_map;
 349         vm_map_lock_read(map);
 350
 351         do {
 352                 entry = map->header.next;
 353                 index = 0;
 354                 while (index < pve->pve_entry && entry != &map->header) {
 355                         entry = entry->next;
 356                         index++;
 357                 }
 358                 if (index != pve->pve_entry) {
 359                         error = EINVAL;
 360                         break;
 361                 }
 362                 while (entry != &map->header &&
 363                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
 364                         entry = entry->next;
 365                         index++;
 366                 }
 367                 if (entry == &map->header) {
 368                         error = ENOENT;
 369                         break;
 370                 }
 371
 372                 /* We got an entry. */
 373                 pve->pve_entry = index + 1;
 374                 pve->pve_timestamp = map->timestamp;
 375                 pve->pve_start = entry->start;
 376                 pve->pve_end = entry->end - 1;
 377                 pve->pve_offset = entry->offset;
 378                 pve->pve_prot = entry->protection;
 379
 380                 /* Backing object's path needed? */
 381                 if (pve->pve_pathlen == 0)
 382                         break;
 383
 384                 pathlen = pve->pve_pathlen;
 385                 pve->pve_pathlen = 0;
 386
 387                 obj = entry->object.vm_object;
 388                 if (obj != NULL)
 389                         VM_OBJECT_LOCK(obj);
 390         } while (0);
 391
 392         vm_map_unlock_read(map);
 393         vmspace_free(vm);
 394
 395         pve->pve_fsid = VNOVAL;
 396         pve->pve_fileid = VNOVAL;
 397
 398         if (error == 0 && obj != NULL) {
 399                 lobj = obj;
 400                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
 401                         if (tobj != obj)
 402                                 VM_OBJECT_LOCK(tobj);
 403                         if (lobj != obj)
 404                                 VM_OBJECT_UNLOCK(lobj);
 405                         lobj = tobj;
 406                         pve->pve_offset += tobj->backing_object_offset;
 407                 }
 408                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 409                 if (vp != NULL)
 410                         vref(vp);
 411                 if (lobj != obj)
 412                         VM_OBJECT_UNLOCK(lobj);
 413                 VM_OBJECT_UNLOCK(obj);
 414
 415                 if (vp != NULL) {
 416                         freepath = NULL;
 417                         fullpath = NULL;
 418                         vn_fullpath(td, vp, &fullpath, &freepath);
 419                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 420                         vn_lock(vp, LK_SHARED | LK_RETRY);
 421                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 422                                 pve->pve_fileid = vattr.va_fileid;
 423                                 pve->pve_fsid = vattr.va_fsid;
 424                         }
 425                         vput(vp);
 426                         VFS_UNLOCK_GIANT(vfslocked);
 427
 428                         if (fullpath != NULL) {
 429                                 pve->pve_pathlen = strlen(fullpath) + 1;
 430                                 if (pve->pve_pathlen <= pathlen) {
 431                                         error = copyout(fullpath, pve->pve_path,
 432                                             pve->pve_pathlen);
 433                                 } else
 434                                         error = ENAMETOOLONG;
 435                         }
 436                         if (freepath != NULL)
 437                                 free(freepath, M_TEMP);
 438                 }
 439         }
 440         if (error == 0)
 441                 CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
 442                     p->p_pid, pve->pve_entry, pve->pve_start);
 443
 444         return (error);
 445 }
 446
 447 #ifdef COMPAT_FREEBSD32
 448 static int
 449 ptrace_vm_entry32(struct thread *td, struct proc *p,
 450     struct ptrace_vm_entry32 *pve32)
 451 {
 452         struct ptrace_vm_entry pve;
 453         int error;
 454
 455         pve.pve_entry = pve32->pve_entry;
 456         pve.pve_pathlen = pve32->pve_pathlen;
 457         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 458
 459         error = ptrace_vm_entry(td, p, &pve);
 460         if (error == 0) {
 461                 pve32->pve_entry = pve.pve_entry;
 462                 pve32->pve_timestamp = pve.pve_timestamp;
 463                 pve32->pve_start = pve.pve_start;
 464                 pve32->pve_end = pve.pve_end;
 465                 pve32->pve_offset = pve.pve_offset;
 466                 pve32->pve_prot = pve.pve_prot;
 467                 pve32->pve_fileid = pve.pve_fileid;
 468                 pve32->pve_fsid = pve.pve_fsid;
 469         }
 470
 471         pve32->pve_pathlen = pve.pve_pathlen;
 472         return (error);
 473 }
 474
 475 static void
 476 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
 477     struct ptrace_lwpinfo32 *pl32)
 478 {
 479
 480         pl32->pl_lwpid = pl->pl_lwpid;
 481         pl32->pl_event = pl->pl_event;
 482         pl32->pl_flags = pl->pl_flags;
 483         pl32->pl_sigmask = pl->pl_sigmask;
 484         pl32->pl_siglist = pl->pl_siglist;
 485         siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
 486         strcpy(pl32->pl_tdname, pl->pl_tdname);
 487         pl32->pl_child_pid = pl->pl_child_pid;
 488         pl32->pl_syscall_code = pl->pl_syscall_code;
 489         pl32->pl_syscall_narg = pl->pl_syscall_narg;
 490 }
 491 #endif /* COMPAT_FREEBSD32 */
 492
 493 /*
 494  * Process debugging system call.
 495  */
 496 #ifndef _SYS_SYSPROTO_H_
 497 struct ptrace_args {
 498         int     req;
 499         pid_t   pid;
 500         caddr_t addr;
 501         int     data;
 502 };
 503 #endif
 504
 505 #ifdef COMPAT_FREEBSD32
 506 /*
 507  * This CPP subterfuge is to try and reduce the number of ifdefs in
 508  * the body of the code.
 509  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 510  * becomes either:
 511  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 512  * or
 513  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 514  * .. except this is done at runtime.
 515  */
 516 #define COPYIN(u, k, s)         wrap32 ? \
 517         copyin(u, k ## 32, s ## 32) : \
 518         copyin(u, k, s)
 519 #define COPYOUT(k, u, s)        wrap32 ? \
 520         copyout(k ## 32, u, s ## 32) : \
 521         copyout(k, u, s)
 522 #else
 523 #define COPYIN(u, k, s)         copyin(u, k, s)
 524 #define COPYOUT(k, u, s)        copyout(k, u, s)
 525 #endif
 526 int
 527 sys_ptrace(struct thread *td, struct ptrace_args *uap)
 528 {
 529         /*
 530          * XXX this obfuscation is to reduce stack usage, but the register
 531          * structs may be too large to put on the stack anyway.
 532          */
 533         union {
 534                 struct ptrace_io_desc piod;
 535                 struct ptrace_lwpinfo pl;
 536                 struct ptrace_vm_entry pve;
 537                 struct dbreg dbreg;
 538                 struct fpreg fpreg;
 539                 struct reg reg;
 540 #ifdef COMPAT_FREEBSD32
 541                 struct dbreg32 dbreg32;
 542                 struct fpreg32 fpreg32;
 543                 struct reg32 reg32;
 544                 struct ptrace_io_desc32 piod32;
 545                 struct ptrace_lwpinfo32 pl32;
 546                 struct ptrace_vm_entry32 pve32;
 547 #endif
 548         } r;
 549         void *addr;
 550         int error = 0;
 551 #ifdef COMPAT_FREEBSD32
 552         int wrap32 = 0;
 553
 554         if (SV_CURPROC_FLAG(SV_ILP32))
 555                 wrap32 = 1;
 556 #endif
 557         AUDIT_ARG_PID(uap->pid);
 558         AUDIT_ARG_CMD(uap->req);
 559         AUDIT_ARG_VALUE(uap->data);
 560         addr = &r;
 561         switch (uap->req) {
 562         case PT_GETREGS:
 563         case PT_GETFPREGS:
 564         case PT_GETDBREGS:
 565         case PT_LWPINFO:
 566                 break;
 567         case PT_SETREGS:
 568                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 569                 break;
 570         case PT_SETFPREGS:
 571                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 572                 break;
 573         case PT_SETDBREGS:
 574                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 575                 break;
 576         case PT_IO:
 577                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 578                 break;
 579         case PT_VM_ENTRY:
 580                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 581                 break;
 582         default:
 583                 addr = uap->addr;
 584                 break;
 585         }
 586         if (error)
 587                 return (error);
 588
 589         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 590         if (error)
 591                 return (error);
 592
 593         switch (uap->req) {
 594         case PT_VM_ENTRY:
 595                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 596                 break;
 597         case PT_IO:
 598                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 599                 break;
 600         case PT_GETREGS:
 601                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 602                 break;
 603         case PT_GETFPREGS:
 604                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 605                 break;
 606         case PT_GETDBREGS:
 607                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 608                 break;
 609         case PT_LWPINFO:
 610                 error = copyout(&r.pl, uap->addr, uap->data);
 611                 break;
 612         }
 613
 614         return (error);
 615 }
 616 #undef COPYIN
 617 #undef COPYOUT
 618
 619 #ifdef COMPAT_FREEBSD32
 620 /*
 621  *   PROC_READ(regs, td2, addr);
 622  * becomes either:
 623  *   proc_read_regs(td2, addr);
 624  * or
 625  *   proc_read_regs32(td2, addr);
 626  * .. except this is done at runtime.  There is an additional
 627  * complication in that PROC_WRITE disallows 32 bit consumers
 628  * from writing to 64 bit address space targets.
 629  */
 630 #define PROC_READ(w, t, a)      wrap32 ? \
 631         proc_read_ ## w ## 32(t, a) : \
 632         proc_read_ ## w (t, a)
 633 #define PROC_WRITE(w, t, a)     wrap32 ? \
 634         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 635         proc_write_ ## w (t, a)
 636 #else
 637 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 638 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 639 #endif
 640
 641 int
 642 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 643 {
 644         struct iovec iov;
 645         struct uio uio;
 646         struct proc *curp, *p, *pp;
 647         struct thread *td2 = NULL;
 648         struct ptrace_io_desc *piod = NULL;
 649         struct ptrace_lwpinfo *pl;
 650         int error, write, tmp, num;
 651         int proctree_locked = 0;
 652         lwpid_t tid = 0, *buf;
 653 #ifdef COMPAT_FREEBSD32
 654         int wrap32 = 0, safe = 0;
 655         struct ptrace_io_desc32 *piod32 = NULL;
 656         struct ptrace_lwpinfo32 *pl32 = NULL;
 657         struct ptrace_lwpinfo plr;
 658 #endif
 659
 660         curp = td->td_proc;
 661
 662         /* Lock proctree before locking the process. */
 663         switch (req) {
 664         case PT_TRACE_ME:
 665         case PT_ATTACH:
 666         case PT_STEP:
 667         case PT_CONTINUE:
 668         case PT_TO_SCE:
 669         case PT_TO_SCX:
 670         case PT_SYSCALL:
 671         case PT_FOLLOW_FORK:
 672         case PT_DETACH:
 673                 sx_xlock(&proctree_lock);
 674                 proctree_locked = 1;
 675                 break;
 676         default:
 677                 break;
 678         }
 679
 680         write = 0;
 681         if (req == PT_TRACE_ME) {
 682                 p = td->td_proc;
 683                 PROC_LOCK(p);
 684         } else {
 685                 if (pid <= PID_MAX) {
 686                         if ((p = pfind(pid)) == NULL) {
 687                                 if (proctree_locked)
 688                                         sx_xunlock(&proctree_lock);
 689                                 return (ESRCH);
 690                         }
 691                 } else {
 692                         td2 = tdfind(pid, -1);
 693                         if (td2 == NULL) {
 694                                 if (proctree_locked)
 695                                         sx_xunlock(&proctree_lock);
 696                                 return (ESRCH);
 697                         }
 698                         p = td2->td_proc;
 699                         tid = pid;
 700                         pid = p->p_pid;
 701                 }
 702         }
 703         AUDIT_ARG_PROCESS(p);
 704
 705         if ((p->p_flag & P_WEXIT) != 0) {
 706                 error = ESRCH;
 707                 goto fail;
 708         }
 709         if ((error = p_cansee(td, p)) != 0)
 710                 goto fail;
 711
 712         if ((error = p_candebug(td, p)) != 0)
 713                 goto fail;
 714
 715         /*
 716          * System processes can't be debugged.
 717          */
 718         if ((p->p_flag & P_SYSTEM) != 0) {
 719                 error = EINVAL;
 720                 goto fail;
 721         }
 722
 723         if (tid == 0) {
 724                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 725                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 726                         td2 = p->p_xthread;
 727                 } else {
 728                         td2 = FIRST_THREAD_IN_PROC(p);
 729                 }
 730                 tid = td2->td_tid;
 731         }
 732
 733 #ifdef COMPAT_FREEBSD32
 734         /*
 735          * Test if we're a 32 bit client and what the target is.
 736          * Set the wrap controls accordingly.
 737          */
 738         if (SV_CURPROC_FLAG(SV_ILP32)) {
 739                 if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
 740                         safe = 1;
 741                 wrap32 = 1;
 742         }
 743 #endif
 744         /*
 745          * Permissions check
 746          */
 747         switch (req) {
 748         case PT_TRACE_ME:
 749                 /* Always legal. */
 750                 break;
 751
 752         case PT_ATTACH:
 753                 /* Self */
 754                 if (p->p_pid == td->td_proc->p_pid) {
 755                         error = EINVAL;
 756                         goto fail;
 757                 }
 758
 759                 /* Already traced */
 760                 if (p->p_flag & P_TRACED) {
 761                         error = EBUSY;
 762                         goto fail;
 763                 }
 764
 765                 /* Can't trace an ancestor if you're being traced. */
 766                 if (curp->p_flag & P_TRACED) {
 767                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 768                                 if (pp == p) {
 769                                         error = EINVAL;
 770                                         goto fail;
 771                                 }
 772                         }
 773                 }
 774
 775
 776                 /* OK */
 777                 break;
 778
 779         case PT_CLEARSTEP:
 780                 /* Allow thread to clear single step for itself */
 781                 if (td->td_tid == tid)
 782                         break;
 783
 784                 /* FALLTHROUGH */
 785         default:
 786                 /* not being traced... */
 787                 if ((p->p_flag & P_TRACED) == 0) {
 788                         error = EPERM;
 789                         goto fail;
 790                 }
 791
 792                 /* not being traced by YOU */
 793                 if (p->p_pptr != td->td_proc) {
 794                         error = EBUSY;
 795                         goto fail;
 796                 }
 797
 798                 /* not currently stopped */
 799                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 800                     p->p_suspcount != p->p_numthreads  ||
 801                     (p->p_flag & P_WAITED) == 0) {
 802                         error = EBUSY;
 803                         goto fail;
 804                 }
 805
 806                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 807                         static int count = 0;
 808                         if (count++ == 0)
 809                                 printf("P_STOPPED_TRACE not set.\n");
 810                 }
 811
 812                 /* OK */
 813                 break;
 814         }
 815
 816         /* Keep this process around until we finish this request. */
 817         _PHOLD(p);
 818
 819 #ifdef FIX_SSTEP
 820         /*
 821          * Single step fixup ala procfs
 822          */
 823         FIX_SSTEP(td2);
 824 #endif
 825
 826         /*
 827          * Actually do the requests
 828          */
 829
 830         td->td_retval[0] = 0;
 831
 832         switch (req) {
 833         case PT_TRACE_ME:
 834                 /* set my trace flag and "owner" so it can read/write me */
 835                 p->p_flag |= P_TRACED;
 836                 if (p->p_flag & P_PPWAIT)
 837                         p->p_flag |= P_PPTRACE;
 838                 p->p_oppid = p->p_pptr->p_pid;
 839                 CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
 840                 break;
 841
 842         case PT_ATTACH:
 843                 /* security check done above */
 844                 /*
 845                  * It would be nice if the tracing relationship was separate
 846                  * from the parent relationship but that would require
 847                  * another set of links in the proc struct or for "wait"
 848                  * to scan the entire proc table.  To make life easier,
 849                  * we just re-parent the process we're trying to trace.
 850                  * The old parent is remembered so we can put things back
 851                  * on a "detach".
 852                  */
 853                 p->p_flag |= P_TRACED;
 854                 p->p_oppid = p->p_pptr->p_pid;
 855                 if (p->p_pptr != td->td_proc) {
 856                         proc_reparent(p, td->td_proc);
 857                 }
 858                 data = SIGSTOP;
 859                 CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
 860                     p->p_oppid);
 861                 goto sendsig;   /* in PT_CONTINUE below */
 862
 863         case PT_CLEARSTEP:
 864                 CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
 865                     p->p_pid);
 866                 error = ptrace_clear_single_step(td2);
 867                 break;
 868
 869         case PT_SETSTEP:
 870                 CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
 871                     p->p_pid);
 872                 error = ptrace_single_step(td2);
 873                 break;
 874
 875         case PT_SUSPEND:
 876                 CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
 877                     p->p_pid);
 878                 td2->td_dbgflags |= TDB_SUSPEND;
 879                 thread_lock(td2);
 880                 td2->td_flags |= TDF_NEEDSUSPCHK;
 881                 thread_unlock(td2);
 882                 break;
 883
 884         case PT_RESUME:
 885                 CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
 886                     p->p_pid);
 887                 td2->td_dbgflags &= ~TDB_SUSPEND;
 888                 break;
 889
 890         case PT_FOLLOW_FORK:
 891                 CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
 892                     p->p_flag & P_FOLLOWFORK ? "enabled" : "disabled",
 893                     data ? "enabled" : "disabled");
 894                 if (data)
 895                         p->p_flag |= P_FOLLOWFORK;
 896                 else
 897                         p->p_flag &= ~P_FOLLOWFORK;
 898                 break;
 899
 900         case PT_STEP:
 901         case PT_CONTINUE:
 902         case PT_TO_SCE:
 903         case PT_TO_SCX:
 904         case PT_SYSCALL:
 905         case PT_DETACH:
 906                 /* Zero means do not send any signal */
 907                 if (data < 0 || data > _SIG_MAXSIG) {
 908                         error = EINVAL;
 909                         break;
 910                 }
 911
 912                 switch (req) {
 913                 case PT_STEP:
 914                         CTR2(KTR_PTRACE, "PT_STEP: tid %d (pid %d)",
 915                             td2->td_tid, p->p_pid);
 916                         error = ptrace_single_step(td2);
 917                         if (error)
 918                                 goto out;
 919                         break;
 920                 case PT_CONTINUE:
 921                 case PT_TO_SCE:
 922                 case PT_TO_SCX:
 923                 case PT_SYSCALL:
 924                         if (addr != (void *)1) {
 925                                 error = ptrace_set_pc(td2,
 926                                     (u_long)(uintfptr_t)addr);
 927                                 if (error)
 928                                         goto out;
 929                         }
 930                         switch (req) {
 931                         case PT_TO_SCE:
 932                                 p->p_stops |= S_PT_SCE;
 933                                 CTR4(KTR_PTRACE,
 934                     "PT_TO_SCE: pid %d, stops = %#x, PC = %#lx, sig = %d",
 935                                     p->p_pid, p->p_stops,
 936                                     (u_long)(uintfptr_t)addr, data);
 937                                 break;
 938                         case PT_TO_SCX:
 939                                 p->p_stops |= S_PT_SCX;
 940                                 CTR4(KTR_PTRACE,
 941                     "PT_TO_SCX: pid %d, stops = %#x, PC = %#lx, sig = %d",
 942                                     p->p_pid, p->p_stops,
 943                                     (u_long)(uintfptr_t)addr, data);
 944                                 break;
 945                         case PT_SYSCALL:
 946                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
 947                                 CTR4(KTR_PTRACE,
 948                     "PT_SYSCALL: pid %d, stops = %#x, PC = %#lx, sig = %d",
 949                                     p->p_pid, p->p_stops,
 950                                     (u_long)(uintfptr_t)addr, data);
 951                                 break;
 952                         case PT_CONTINUE:
 953                                 CTR3(KTR_PTRACE,
 954                                     "PT_CONTINUE: pid %d, PC = %#lx, sig = %d",
 955                                     p->p_pid, (u_long)(uintfptr_t)addr, data);
 956                                 break;
 957                         }
 958                         break;
 959                 case PT_DETACH:
 960                         /*
 961                          * Reset the process parent.
 962                          *
 963                          * NB: This clears P_TRACED before reparenting
 964                          * a detached process back to its original
 965                          * parent.  Otherwise the debugee will be set
 966                          * as an orphan of the debugger.
 967                          */
 968                         p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
 969                         if (p->p_oppid != p->p_pptr->p_pid) {
 970                                 PROC_LOCK(p->p_pptr);
 971                                 sigqueue_take(p->p_ksi);
 972                                 PROC_UNLOCK(p->p_pptr);
 973
 974                                 pp = proc_realparent(p);
 975                                 proc_reparent(p, pp);
 976                                 if (pp == initproc)
 977                                         p->p_sigparent = SIGCHLD;
 978                                 CTR3(KTR_PTRACE,
 979                             "PT_DETACH: pid %d reparented to pid %d, sig %d",
 980                                     p->p_pid, pp->p_pid, data);
 981                         } else
 982                                 CTR2(KTR_PTRACE, "PT_DETACH: pid %d, sig %d",
 983                                     p->p_pid, data);
 984                         p->p_oppid = 0;
 985
 986                         /* should we send SIGCHLD? */
 987                         /* childproc_continued(p); */
 988                         break;
 989                 }
 990
 991         sendsig:
 992                 if (proctree_locked) {
 993                         sx_xunlock(&proctree_lock);
 994                         proctree_locked = 0;
 995                 }
 996                 p->p_xstat = data;
 997                 p->p_xthread = NULL;
 998                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 999                         /* deliver or queue signal */
1000                         td2->td_dbgflags &= ~TDB_XSIG;
1001                         td2->td_xsig = data;
1002
1003                         if (req == PT_DETACH) {
1004                                 struct thread *td3;
1005                                 FOREACH_THREAD_IN_PROC(p, td3) {
1006                                         td3->td_dbgflags &= ~TDB_SUSPEND;
1007                                 }
1008                         }
1009                         /*
1010                          * unsuspend all threads, to not let a thread run,
1011                          * you should use PT_SUSPEND to suspend it before
1012                          * continuing process.
1013                          */
1014                         PROC_SLOCK(p);
1015                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
1016                         thread_unsuspend(p);
1017                         PROC_SUNLOCK(p);
1018                 } else {
1019                         if (data)
1020                                 kern_psignal(p, data);
1021                 }
1022                 break;
1023
1024         case PT_WRITE_I:
1025         case PT_WRITE_D:
1026                 td2->td_dbgflags |= TDB_USERWR;
1027                 write = 1;
1028                 /* FALLTHROUGH */
1029         case PT_READ_I:
1030         case PT_READ_D:
1031                 PROC_UNLOCK(p);
1032                 tmp = 0;
1033                 /* write = 0 set above */
1034                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1035                 iov.iov_len = sizeof(int);
1036                 uio.uio_iov = &iov;
1037                 uio.uio_iovcnt = 1;
1038                 uio.uio_offset = (off_t)(uintptr_t)addr;
1039                 uio.uio_resid = sizeof(int);
1040                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
1041                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1042                 uio.uio_td = td;
1043                 error = proc_rwmem(p, &uio);
1044                 if (uio.uio_resid != 0) {
1045                         /*
1046                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1047                          * so I think write() can bogusly return 0.
1048                          * XXX what happens for short writes?  We don't want
1049                          * to write partial data.
1050                          * XXX proc_rwmem() returns EPERM for other invalid
1051                          * addresses.  Convert this to EINVAL.  Does this
1052                          * clobber returns of EPERM for other reasons?
1053                          */
1054                         if (error == 0 || error == ENOSPC || error == EPERM)
1055                                 error = EINVAL; /* EOF */
1056                 }
1057                 if (!write)
1058                         td->td_retval[0] = tmp;
1059                 if (error == 0) {
1060                         if (write)
1061                                 CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1062                                     p->p_pid, addr, data);
1063                         else
1064                                 CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1065                                     p->p_pid, addr, tmp);
1066                 }
1067                 PROC_LOCK(p);
1068                 break;
1069
1070         case PT_IO:
1071 #ifdef COMPAT_FREEBSD32
1072                 if (wrap32) {
1073                         piod32 = addr;
1074                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1075                         iov.iov_len = piod32->piod_len;
1076                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1077                         uio.uio_resid = piod32->piod_len;
1078                 } else
1079 #endif
1080                 {
1081                         piod = addr;
1082                         iov.iov_base = piod->piod_addr;
1083                         iov.iov_len = piod->piod_len;
1084                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1085                         uio.uio_resid = piod->piod_len;
1086                 }
1087                 uio.uio_iov = &iov;
1088                 uio.uio_iovcnt = 1;
1089                 uio.uio_segflg = UIO_USERSPACE;
1090                 uio.uio_td = td;
1091 #ifdef COMPAT_FREEBSD32
1092                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1093 #else
1094                 tmp = piod->piod_op;
1095 #endif
1096                 switch (tmp) {
1097                 case PIOD_READ_D:
1098                 case PIOD_READ_I:
1099                         CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1100                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1101                         uio.uio_rw = UIO_READ;
1102                         break;
1103                 case PIOD_WRITE_D:
1104                 case PIOD_WRITE_I:
1105                         CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1106                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1107                         td2->td_dbgflags |= TDB_USERWR;
1108                         uio.uio_rw = UIO_WRITE;
1109                         break;
1110                 default:
1111                         error = EINVAL;
1112                         goto out;
1113                 }
1114                 PROC_UNLOCK(p);
1115                 error = proc_rwmem(p, &uio);
1116 #ifdef COMPAT_FREEBSD32
1117                 if (wrap32)
1118                         piod32->piod_len -= uio.uio_resid;
1119                 else
1120 #endif
1121                         piod->piod_len -= uio.uio_resid;
1122                 PROC_LOCK(p);
1123                 break;
1124
1125         case PT_KILL:
1126                 CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1127                 data = SIGKILL;
1128                 goto sendsig;   /* in PT_CONTINUE above */
1129
1130         case PT_SETREGS:
1131                 CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1132                     p->p_pid);
1133                 td2->td_dbgflags |= TDB_USERWR;
1134                 error = PROC_WRITE(regs, td2, addr);
1135                 break;
1136
1137         case PT_GETREGS:
1138                 CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1139                     p->p_pid);
1140                 error = PROC_READ(regs, td2, addr);
1141                 break;
1142
1143         case PT_SETFPREGS:
1144                 CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1145                     p->p_pid);
1146                 td2->td_dbgflags |= TDB_USERWR;
1147                 error = PROC_WRITE(fpregs, td2, addr);
1148                 break;
1149
1150         case PT_GETFPREGS:
1151                 CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1152                     p->p_pid);
1153                 error = PROC_READ(fpregs, td2, addr);
1154                 break;
1155
1156         case PT_SETDBREGS:
1157                 CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1158                     p->p_pid);
1159                 td2->td_dbgflags |= TDB_USERWR;
1160                 error = PROC_WRITE(dbregs, td2, addr);
1161                 break;
1162
1163         case PT_GETDBREGS:
1164                 CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1165                     p->p_pid);
1166                 error = PROC_READ(dbregs, td2, addr);
1167                 break;
1168
1169         case PT_LWPINFO:
1170                 if (data <= 0 ||
1171 #ifdef COMPAT_FREEBSD32
1172                     (!wrap32 && data > sizeof(*pl)) ||
1173                     (wrap32 && data > sizeof(*pl32))) {
1174 #else
1175                     data > sizeof(*pl)) {
1176 #endif
1177                         error = EINVAL;
1178                         break;
1179                 }
1180 #ifdef COMPAT_FREEBSD32
1181                 if (wrap32) {
1182                         pl = &plr;
1183                         pl32 = addr;
1184                 } else
1185 #endif
1186                 pl = addr;
1187                 pl->pl_lwpid = td2->td_tid;
1188                 pl->pl_event = PL_EVENT_NONE;
1189                 pl->pl_flags = 0;
1190                 if (td2->td_dbgflags & TDB_XSIG) {
1191                         pl->pl_event = PL_EVENT_SIGNAL;
1192                         if (td2->td_dbgksi.ksi_signo != 0 &&
1193 #ifdef COMPAT_FREEBSD32
1194                             ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1195                             pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1196                             (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1197                             pl_siginfo) + sizeof(struct siginfo32)))
1198 #else
1199                             data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1200                             + sizeof(pl->pl_siginfo)
1201 #endif
1202                         ){
1203                                 pl->pl_flags |= PL_FLAG_SI;
1204                                 pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1205                         }
1206                 }
1207                 if ((pl->pl_flags & PL_FLAG_SI) == 0)
1208                         bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1209                 if (td2->td_dbgflags & TDB_SCE)
1210                         pl->pl_flags |= PL_FLAG_SCE;
1211                 else if (td2->td_dbgflags & TDB_SCX)
1212                         pl->pl_flags |= PL_FLAG_SCX;
1213                 if (td2->td_dbgflags & TDB_EXEC)
1214                         pl->pl_flags |= PL_FLAG_EXEC;
1215                 if (td2->td_dbgflags & TDB_FORK) {
1216                         pl->pl_flags |= PL_FLAG_FORKED;
1217                         pl->pl_child_pid = td2->td_dbg_forked;
1218                 }
1219                 if (td2->td_dbgflags & TDB_CHILD)
1220                         pl->pl_flags |= PL_FLAG_CHILD;
1221                 pl->pl_sigmask = td2->td_sigmask;
1222                 pl->pl_siglist = td2->td_siglist;
1223                 strcpy(pl->pl_tdname, td2->td_name);
1224                 if ((td2->td_dbgflags & (TDB_SCE | TDB_SCX)) != 0) {
1225                         pl->pl_syscall_code = td2->td_dbg_sc_code;
1226                         pl->pl_syscall_narg = td2->td_dbg_sc_narg;
1227                 } else {
1228                         pl->pl_syscall_code = 0;
1229                         pl->pl_syscall_narg = 0;
1230                 }
1231 #ifdef COMPAT_FREEBSD32
1232                 if (wrap32)
1233                         ptrace_lwpinfo_to32(pl, pl32);
1234 #endif
1235                 CTR6(KTR_PTRACE,
1236     "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d syscall %d",
1237                     td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1238                     pl->pl_child_pid, pl->pl_syscall_code);
1239                 break;
1240
1241         case PT_GETNUMLWPS:
1242                 CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1243                     p->p_numthreads);
1244                 td->td_retval[0] = p->p_numthreads;
1245                 break;
1246
1247         case PT_GETLWPLIST:
1248                 CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1249                     p->p_pid, data, p->p_numthreads);
1250                 if (data <= 0) {
1251                         error = EINVAL;
1252                         break;
1253                 }
1254                 num = imin(p->p_numthreads, data);
1255                 PROC_UNLOCK(p);
1256                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1257                 tmp = 0;
1258                 PROC_LOCK(p);
1259                 FOREACH_THREAD_IN_PROC(p, td2) {
1260                         if (tmp >= num)
1261                                 break;
1262                         buf[tmp++] = td2->td_tid;
1263                 }
1264                 PROC_UNLOCK(p);
1265                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1266                 free(buf, M_TEMP);
1267                 if (!error)
1268                         td->td_retval[0] = tmp;
1269                 PROC_LOCK(p);
1270                 break;
1271
1272         case PT_VM_TIMESTAMP:
1273                 CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1274                     p->p_pid, p->p_vmspace->vm_map.timestamp);
1275                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1276                 break;
1277
1278         case PT_VM_ENTRY:
1279                 PROC_UNLOCK(p);
1280 #ifdef COMPAT_FREEBSD32
1281                 if (wrap32)
1282                         error = ptrace_vm_entry32(td, p, addr);
1283                 else
1284 #endif
1285                 error = ptrace_vm_entry(td, p, addr);
1286                 PROC_LOCK(p);
1287                 break;
1288
1289         default:
1290 #ifdef __HAVE_PTRACE_MACHDEP
1291                 if (req >= PT_FIRSTMACH) {
1292                         PROC_UNLOCK(p);
1293                         error = cpu_ptrace(td2, req, addr, data);
1294                         PROC_LOCK(p);
1295                 } else
1296 #endif
1297                         /* Unknown request. */
1298                         error = EINVAL;
1299                 break;
1300         }
1301
1302 out:
1303         /* Drop our hold on this process now that the request has completed. */
1304         _PRELE(p);
1305 fail:
1306         PROC_UNLOCK(p);
1307         if (proctree_locked)
1308                 sx_xunlock(&proctree_lock);
1309         return (error);
1310 }
1311 #undef PROC_READ
1312 #undef PROC_WRITE
1313
1314 /*
1315  * Stop a process because of a debugging event;
1316  * stay stopped until p->p_step is cleared
1317  * (cleared by PIOCCONT in procfs).
1318  */
1319 void
1320 stopevent(struct proc *p, unsigned int event, unsigned int val)
1321 {
1322
1323         PROC_LOCK_ASSERT(p, MA_OWNED);
1324         p->p_step = 1;
1325         CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1326             val);
1327         do {
1328                 p->p_xstat = val;
1329                 p->p_xthread = NULL;
1330                 p->p_stype = event;     /* Which event caused the stop? */
1331                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1332                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1333         } while (p->p_step);
1334 }
1335
1336 static int
1337 protect_setchild(struct thread *td, struct proc *p, int flags)
1338 {
1339
1340         PROC_LOCK_ASSERT(p, MA_OWNED);
1341         if (p->p_flag & P_SYSTEM || p_cansched(td, p) != 0)
1342                 return (0);
1343         if (flags & PPROT_SET) {
1344                 p->p_flag |= P_PROTECTED;
1345                 if (flags & PPROT_INHERIT)
1346                         p->p_flag2 |= P2_INHERIT_PROTECTED;
1347         } else {
1348                 p->p_flag &= ~P_PROTECTED;
1349                 p->p_flag2 &= ~P2_INHERIT_PROTECTED;
1350         }
1351         return (1);
1352 }
1353
1354 static int
1355 protect_setchildren(struct thread *td, struct proc *top, int flags)
1356 {
1357         struct proc *p;
1358         int ret;
1359
1360         p = top;
1361         ret = 0;
1362         sx_assert(&proctree_lock, SX_LOCKED);
1363         for (;;) {
1364                 ret |= protect_setchild(td, p, flags);
1365                 PROC_UNLOCK(p);
1366                 /*
1367                  * If this process has children, descend to them next,
1368                  * otherwise do any siblings, and if done with this level,
1369                  * follow back up the tree (but not past top).
1370                  */
1371                 if (!LIST_EMPTY(&p->p_children))
1372                         p = LIST_FIRST(&p->p_children);
1373                 else for (;;) {
1374                         if (p == top) {
1375                                 PROC_LOCK(p);
1376                                 return (ret);
1377                         }
1378                         if (LIST_NEXT(p, p_sibling)) {
1379                                 p = LIST_NEXT(p, p_sibling);
1380                                 break;
1381                         }
1382                         p = p->p_pptr;
1383                 }
1384                 PROC_LOCK(p);
1385         }
1386 }
1387
1388 static int
1389 protect_set(struct thread *td, struct proc *p, int flags)
1390 {
1391         int error, ret;
1392
1393         switch (PPROT_OP(flags)) {
1394         case PPROT_SET:
1395         case PPROT_CLEAR:
1396                 break;
1397         default:
1398                 return (EINVAL);
1399         }
1400
1401         if ((PPROT_FLAGS(flags) & ~(PPROT_DESCEND | PPROT_INHERIT)) != 0)
1402                 return (EINVAL);
1403
1404         error = priv_check(td, PRIV_VM_MADV_PROTECT);
1405         if (error)
1406                 return (error);
1407
1408         if (flags & PPROT_DESCEND)
1409                 ret = protect_setchildren(td, p, flags);
1410         else
1411                 ret = protect_setchild(td, p, flags);
1412         if (ret == 0)
1413                 return (EPERM);
1414         return (0);
1415 }
1416
1417 #ifndef _SYS_SYSPROTO_H_
1418 struct procctl_args {
1419         idtype_t idtype;
1420         id_t    id;
1421         int     com;
1422         void    *data;
1423 };
1424 #endif
1425 /* ARGSUSED */
1426 int
1427 sys_procctl(struct thread *td, struct procctl_args *uap)
1428 {
1429         int error, flags;
1430         void *data;
1431
1432         switch (uap->com) {
1433         case PROC_SPROTECT:
1434                 error = copyin(uap->data, &flags, sizeof(flags));
1435                 if (error)
1436                         return (error);
1437                 data = &flags;
1438                 break;
1439         default:
1440                 return (EINVAL);
1441         }
1442
1443         return (kern_procctl(td, uap->idtype, uap->id, uap->com, data));
1444 }
1445
1446 static int
1447 kern_procctl_single(struct thread *td, struct proc *p, int com, void *data)
1448 {
1449
1450         PROC_LOCK_ASSERT(p, MA_OWNED);
1451         switch (com) {
1452         case PROC_SPROTECT:
1453                 return (protect_set(td, p, *(int *)data));
1454         default:
1455                 return (EINVAL);
1456         }
1457 }
1458
1459 int
1460 kern_procctl(struct thread *td, idtype_t idtype, id_t id, int com, void *data)
1461 {
1462         struct pgrp *pg;
1463         struct proc *p;
1464         int error, first_error, ok;
1465
1466         sx_slock(&proctree_lock);
1467         switch (idtype) {
1468         case P_PID:
1469                 p = pfind(id);
1470                 if (p == NULL) {
1471                         error = ESRCH;
1472                         break;
1473                 }
1474                 if (p->p_state == PRS_NEW)
1475                         error = ESRCH;
1476                 else
1477                         error = p_cansee(td, p);
1478                 if (error == 0)
1479                         error = kern_procctl_single(td, p, com, data);
1480                 PROC_UNLOCK(p);
1481                 break;
1482         case P_PGID:
1483                 /*
1484                  * Attempt to apply the operation to all members of the
1485                  * group.  Ignore processes in the group that can't be
1486                  * seen.  Ignore errors so long as at least one process is
1487                  * able to complete the request successfully.
1488                  */
1489                 pg = pgfind(id);
1490                 if (pg == NULL) {
1491                         error = ESRCH;
1492                         break;
1493                 }
1494                 PGRP_UNLOCK(pg);
1495                 ok = 0;
1496                 first_error = 0;
1497                 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1498                         PROC_LOCK(p);
1499                         if (p->p_state == PRS_NEW || p_cansee(td, p) != 0) {
1500                                 PROC_UNLOCK(p);
1501                                 continue;
1502                         }
1503                         error = kern_procctl_single(td, p, com, data);
1504                         PROC_UNLOCK(p);
1505                         if (error == 0)
1506                                 ok = 1;
1507                         else if (first_error == 0)
1508                                 first_error = error;
1509                 }
1510                 if (ok)
1511                         error = 0;
1512                 else if (first_error != 0)
1513                         error = first_error;
1514                 else
1515                         /*
1516                          * Was not able to see any processes in the
1517                          * process group.
1518                          */
1519                         error = ESRCH;
1520                 break;
1521         default:
1522                 error = EINVAL;
1523                 break;
1524         }
1525         sx_sunlock(&proctree_lock);
1526         return (error);
1527 }