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         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, vfslocked;
 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_LOCK(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_LOCK(tobj);
 401                         if (lobj != obj)
 402                                 VM_OBJECT_UNLOCK(lobj);
 403                         lobj = tobj;
 404                         pve->pve_offset += tobj->backing_object_offset;
 405                 }
 406                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 407                 if (vp != NULL)
 408                         vref(vp);
 409                 if (lobj != obj)
 410                         VM_OBJECT_UNLOCK(lobj);
 411                 VM_OBJECT_UNLOCK(obj);
 412
 413                 if (vp != NULL) {
 414                         freepath = NULL;
 415                         fullpath = NULL;
 416                         vn_fullpath(td, vp, &fullpath, &freepath);
 417                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 418                         vn_lock(vp, LK_SHARED | LK_RETRY);
 419                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 420                                 pve->pve_fileid = vattr.va_fileid;
 421                                 pve->pve_fsid = vattr.va_fsid;
 422                         }
 423                         vput(vp);
 424                         VFS_UNLOCK_GIANT(vfslocked);
 425
 426                         if (fullpath != NULL) {
 427                                 pve->pve_pathlen = strlen(fullpath) + 1;
 428                                 if (pve->pve_pathlen <= pathlen) {
 429                                         error = copyout(fullpath, pve->pve_path,
 430                                             pve->pve_pathlen);
 431                                 } else
 432                                         error = ENAMETOOLONG;
 433                         }
 434                         if (freepath != NULL)
 435                                 free(freepath, M_TEMP);
 436                 }
 437         }
 438         if (error == 0)
 439                 CTR3(KTR_PTRACE, "PT_VM_ENTRY: pid %d, entry %d, start %p",
 440                     p->p_pid, pve->pve_entry, pve->pve_start);
 441
 442         return (error);
 443 }
 444
 445 #ifdef COMPAT_FREEBSD32
 446 static int
 447 ptrace_vm_entry32(struct thread *td, struct proc *p,
 448     struct ptrace_vm_entry32 *pve32)
 449 {
 450         struct ptrace_vm_entry pve;
 451         int error;
 452
 453         pve.pve_entry = pve32->pve_entry;
 454         pve.pve_pathlen = pve32->pve_pathlen;
 455         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 456
 457         error = ptrace_vm_entry(td, p, &pve);
 458         if (error == 0) {
 459                 pve32->pve_entry = pve.pve_entry;
 460                 pve32->pve_timestamp = pve.pve_timestamp;
 461                 pve32->pve_start = pve.pve_start;
 462                 pve32->pve_end = pve.pve_end;
 463                 pve32->pve_offset = pve.pve_offset;
 464                 pve32->pve_prot = pve.pve_prot;
 465                 pve32->pve_fileid = pve.pve_fileid;
 466                 pve32->pve_fsid = pve.pve_fsid;
 467         }
 468
 469         pve32->pve_pathlen = pve.pve_pathlen;
 470         return (error);
 471 }
 472
 473 static void
 474 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
 475     struct ptrace_lwpinfo32 *pl32)
 476 {
 477
 478         pl32->pl_lwpid = pl->pl_lwpid;
 479         pl32->pl_event = pl->pl_event;
 480         pl32->pl_flags = pl->pl_flags;
 481         pl32->pl_sigmask = pl->pl_sigmask;
 482         pl32->pl_siglist = pl->pl_siglist;
 483         siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
 484         strcpy(pl32->pl_tdname, pl->pl_tdname);
 485         pl32->pl_child_pid = pl->pl_child_pid;
 486 }
 487 #endif /* COMPAT_FREEBSD32 */
 488
 489 /*
 490  * Process debugging system call.
 491  */
 492 #ifndef _SYS_SYSPROTO_H_
 493 struct ptrace_args {
 494         int     req;
 495         pid_t   pid;
 496         caddr_t addr;
 497         int     data;
 498 };
 499 #endif
 500
 501 #ifdef COMPAT_FREEBSD32
 502 /*
 503  * This CPP subterfuge is to try and reduce the number of ifdefs in
 504  * the body of the code.
 505  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 506  * becomes either:
 507  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 508  * or
 509  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 510  * .. except this is done at runtime.
 511  */
 512 #define COPYIN(u, k, s)         wrap32 ? \
 513         copyin(u, k ## 32, s ## 32) : \
 514         copyin(u, k, s)
 515 #define COPYOUT(k, u, s)        wrap32 ? \
 516         copyout(k ## 32, u, s ## 32) : \
 517         copyout(k, u, s)
 518 #else
 519 #define COPYIN(u, k, s)         copyin(u, k, s)
 520 #define COPYOUT(k, u, s)        copyout(k, u, s)
 521 #endif
 522 int
 523 sys_ptrace(struct thread *td, struct ptrace_args *uap)
 524 {
 525         /*
 526          * XXX this obfuscation is to reduce stack usage, but the register
 527          * structs may be too large to put on the stack anyway.
 528          */
 529         union {
 530                 struct ptrace_io_desc piod;
 531                 struct ptrace_lwpinfo pl;
 532                 struct ptrace_vm_entry pve;
 533                 struct dbreg dbreg;
 534                 struct fpreg fpreg;
 535                 struct reg reg;
 536 #ifdef COMPAT_FREEBSD32
 537                 struct dbreg32 dbreg32;
 538                 struct fpreg32 fpreg32;
 539                 struct reg32 reg32;
 540                 struct ptrace_io_desc32 piod32;
 541                 struct ptrace_lwpinfo32 pl32;
 542                 struct ptrace_vm_entry32 pve32;
 543 #endif
 544         } r;
 545         void *addr;
 546         int error = 0;
 547 #ifdef COMPAT_FREEBSD32
 548         int wrap32 = 0;
 549
 550         if (SV_CURPROC_FLAG(SV_ILP32))
 551                 wrap32 = 1;
 552 #endif
 553         AUDIT_ARG_PID(uap->pid);
 554         AUDIT_ARG_CMD(uap->req);
 555         AUDIT_ARG_VALUE(uap->data);
 556         addr = &r;
 557         switch (uap->req) {
 558         case PT_GETREGS:
 559         case PT_GETFPREGS:
 560         case PT_GETDBREGS:
 561         case PT_LWPINFO:
 562                 break;
 563         case PT_SETREGS:
 564                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 565                 break;
 566         case PT_SETFPREGS:
 567                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 568                 break;
 569         case PT_SETDBREGS:
 570                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 571                 break;
 572         case PT_IO:
 573                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 574                 break;
 575         case PT_VM_ENTRY:
 576                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 577                 break;
 578         default:
 579                 addr = uap->addr;
 580                 break;
 581         }
 582         if (error)
 583                 return (error);
 584
 585         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 586         if (error)
 587                 return (error);
 588
 589         switch (uap->req) {
 590         case PT_VM_ENTRY:
 591                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 592                 break;
 593         case PT_IO:
 594                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 595                 break;
 596         case PT_GETREGS:
 597                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 598                 break;
 599         case PT_GETFPREGS:
 600                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 601                 break;
 602         case PT_GETDBREGS:
 603                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 604                 break;
 605         case PT_LWPINFO:
 606                 error = copyout(&r.pl, uap->addr, uap->data);
 607                 break;
 608         }
 609
 610         return (error);
 611 }
 612 #undef COPYIN
 613 #undef COPYOUT
 614
 615 #ifdef COMPAT_FREEBSD32
 616 /*
 617  *   PROC_READ(regs, td2, addr);
 618  * becomes either:
 619  *   proc_read_regs(td2, addr);
 620  * or
 621  *   proc_read_regs32(td2, addr);
 622  * .. except this is done at runtime.  There is an additional
 623  * complication in that PROC_WRITE disallows 32 bit consumers
 624  * from writing to 64 bit address space targets.
 625  */
 626 #define PROC_READ(w, t, a)      wrap32 ? \
 627         proc_read_ ## w ## 32(t, a) : \
 628         proc_read_ ## w (t, a)
 629 #define PROC_WRITE(w, t, a)     wrap32 ? \
 630         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 631         proc_write_ ## w (t, a)
 632 #else
 633 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 634 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 635 #endif
 636
 637 int
 638 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 639 {
 640         struct iovec iov;
 641         struct uio uio;
 642         struct proc *curp, *p, *pp;
 643         struct thread *td2 = NULL;
 644         struct ptrace_io_desc *piod = NULL;
 645         struct ptrace_lwpinfo *pl;
 646         int error, write, tmp, num;
 647         int proctree_locked = 0;
 648         lwpid_t tid = 0, *buf;
 649 #ifdef COMPAT_FREEBSD32
 650         int wrap32 = 0, safe = 0;
 651         struct ptrace_io_desc32 *piod32 = NULL;
 652         struct ptrace_lwpinfo32 *pl32 = NULL;
 653         struct ptrace_lwpinfo plr;
 654 #endif
 655
 656         curp = td->td_proc;
 657
 658         /* Lock proctree before locking the process. */
 659         switch (req) {
 660         case PT_TRACE_ME:
 661         case PT_ATTACH:
 662         case PT_STEP:
 663         case PT_CONTINUE:
 664         case PT_TO_SCE:
 665         case PT_TO_SCX:
 666         case PT_SYSCALL:
 667         case PT_FOLLOW_FORK:
 668         case PT_DETACH:
 669                 sx_xlock(&proctree_lock);
 670                 proctree_locked = 1;
 671                 break;
 672         default:
 673                 break;
 674         }
 675
 676         write = 0;
 677         if (req == PT_TRACE_ME) {
 678                 p = td->td_proc;
 679                 PROC_LOCK(p);
 680         } else {
 681                 if (pid <= PID_MAX) {
 682                         if ((p = pfind(pid)) == NULL) {
 683                                 if (proctree_locked)
 684                                         sx_xunlock(&proctree_lock);
 685                                 return (ESRCH);
 686                         }
 687                 } else {
 688                         td2 = tdfind(pid, -1);
 689                         if (td2 == NULL) {
 690                                 if (proctree_locked)
 691                                         sx_xunlock(&proctree_lock);
 692                                 return (ESRCH);
 693                         }
 694                         p = td2->td_proc;
 695                         tid = pid;
 696                         pid = p->p_pid;
 697                 }
 698         }
 699         AUDIT_ARG_PROCESS(p);
 700
 701         if ((p->p_flag & P_WEXIT) != 0) {
 702                 error = ESRCH;
 703                 goto fail;
 704         }
 705         if ((error = p_cansee(td, p)) != 0)
 706                 goto fail;
 707
 708         if ((error = p_candebug(td, p)) != 0)
 709                 goto fail;
 710
 711         /*
 712          * System processes can't be debugged.
 713          */
 714         if ((p->p_flag & P_SYSTEM) != 0) {
 715                 error = EINVAL;
 716                 goto fail;
 717         }
 718
 719         if (tid == 0) {
 720                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 721                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 722                         td2 = p->p_xthread;
 723                 } else {
 724                         td2 = FIRST_THREAD_IN_PROC(p);
 725                 }
 726                 tid = td2->td_tid;
 727         }
 728
 729 #ifdef COMPAT_FREEBSD32
 730         /*
 731          * Test if we're a 32 bit client and what the target is.
 732          * Set the wrap controls accordingly.
 733          */
 734         if (SV_CURPROC_FLAG(SV_ILP32)) {
 735                 if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
 736                         safe = 1;
 737                 wrap32 = 1;
 738         }
 739 #endif
 740         /*
 741          * Permissions check
 742          */
 743         switch (req) {
 744         case PT_TRACE_ME:
 745                 /* Always legal. */
 746                 break;
 747
 748         case PT_ATTACH:
 749                 /* Self */
 750                 if (p->p_pid == td->td_proc->p_pid) {
 751                         error = EINVAL;
 752                         goto fail;
 753                 }
 754
 755                 /* Already traced */
 756                 if (p->p_flag & P_TRACED) {
 757                         error = EBUSY;
 758                         goto fail;
 759                 }
 760
 761                 /* Can't trace an ancestor if you're being traced. */
 762                 if (curp->p_flag & P_TRACED) {
 763                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 764                                 if (pp == p) {
 765                                         error = EINVAL;
 766                                         goto fail;
 767                                 }
 768                         }
 769                 }
 770
 771
 772                 /* OK */
 773                 break;
 774
 775         case PT_CLEARSTEP:
 776                 /* Allow thread to clear single step for itself */
 777                 if (td->td_tid == tid)
 778                         break;
 779
 780                 /* FALLTHROUGH */
 781         default:
 782                 /* not being traced... */
 783                 if ((p->p_flag & P_TRACED) == 0) {
 784                         error = EPERM;
 785                         goto fail;
 786                 }
 787
 788                 /* not being traced by YOU */
 789                 if (p->p_pptr != td->td_proc) {
 790                         error = EBUSY;
 791                         goto fail;
 792                 }
 793
 794                 /* not currently stopped */
 795                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 796                     p->p_suspcount != p->p_numthreads  ||
 797                     (p->p_flag & P_WAITED) == 0) {
 798                         error = EBUSY;
 799                         goto fail;
 800                 }
 801
 802                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 803                         static int count = 0;
 804                         if (count++ == 0)
 805                                 printf("P_STOPPED_TRACE not set.\n");
 806                 }
 807
 808                 /* OK */
 809                 break;
 810         }
 811
 812         /* Keep this process around until we finish this request. */
 813         _PHOLD(p);
 814
 815 #ifdef FIX_SSTEP
 816         /*
 817          * Single step fixup ala procfs
 818          */
 819         FIX_SSTEP(td2);
 820 #endif
 821
 822         /*
 823          * Actually do the requests
 824          */
 825
 826         td->td_retval[0] = 0;
 827
 828         switch (req) {
 829         case PT_TRACE_ME:
 830                 /* set my trace flag and "owner" so it can read/write me */
 831                 p->p_flag |= P_TRACED;
 832                 if (p->p_flag & P_PPWAIT)
 833                         p->p_flag |= P_PPTRACE;
 834                 p->p_oppid = p->p_pptr->p_pid;
 835                 CTR1(KTR_PTRACE, "PT_TRACE_ME: pid %d", p->p_pid);
 836                 break;
 837
 838         case PT_ATTACH:
 839                 /* security check done above */
 840                 /*
 841                  * It would be nice if the tracing relationship was separate
 842                  * from the parent relationship but that would require
 843                  * another set of links in the proc struct or for "wait"
 844                  * to scan the entire proc table.  To make life easier,
 845                  * we just re-parent the process we're trying to trace.
 846                  * The old parent is remembered so we can put things back
 847                  * on a "detach".
 848                  */
 849                 p->p_flag |= P_TRACED;
 850                 p->p_oppid = p->p_pptr->p_pid;
 851                 if (p->p_pptr != td->td_proc) {
 852                         proc_reparent(p, td->td_proc);
 853                 }
 854                 data = SIGSTOP;
 855                 CTR2(KTR_PTRACE, "PT_ATTACH: pid %d, oppid %d", p->p_pid,
 856                     p->p_oppid);
 857                 goto sendsig;   /* in PT_CONTINUE below */
 858
 859         case PT_CLEARSTEP:
 860                 CTR2(KTR_PTRACE, "PT_CLEARSTEP: tid %d (pid %d)", td2->td_tid,
 861                     p->p_pid);
 862                 error = ptrace_clear_single_step(td2);
 863                 break;
 864
 865         case PT_SETSTEP:
 866                 CTR2(KTR_PTRACE, "PT_SETSTEP: tid %d (pid %d)", td2->td_tid,
 867                     p->p_pid);
 868                 error = ptrace_single_step(td2);
 869                 break;
 870
 871         case PT_SUSPEND:
 872                 CTR2(KTR_PTRACE, "PT_SUSPEND: tid %d (pid %d)", td2->td_tid,
 873                     p->p_pid);
 874                 td2->td_dbgflags |= TDB_SUSPEND;
 875                 thread_lock(td2);
 876                 td2->td_flags |= TDF_NEEDSUSPCHK;
 877                 thread_unlock(td2);
 878                 break;
 879
 880         case PT_RESUME:
 881                 CTR2(KTR_PTRACE, "PT_RESUME: tid %d (pid %d)", td2->td_tid,
 882                     p->p_pid);
 883                 td2->td_dbgflags &= ~TDB_SUSPEND;
 884                 break;
 885
 886         case PT_FOLLOW_FORK:
 887                 CTR3(KTR_PTRACE, "PT_FOLLOW_FORK: pid %d %s -> %s", p->p_pid,
 888                     p->p_flag & P_FOLLOWFORK ? "enabled" : "disabled",
 889                     data ? "enabled" : "disabled");
 890                 if (data)
 891                         p->p_flag |= P_FOLLOWFORK;
 892                 else
 893                         p->p_flag &= ~P_FOLLOWFORK;
 894                 break;
 895
 896         case PT_STEP:
 897         case PT_CONTINUE:
 898         case PT_TO_SCE:
 899         case PT_TO_SCX:
 900         case PT_SYSCALL:
 901         case PT_DETACH:
 902                 /* Zero means do not send any signal */
 903                 if (data < 0 || data > _SIG_MAXSIG) {
 904                         error = EINVAL;
 905                         break;
 906                 }
 907
 908                 switch (req) {
 909                 case PT_STEP:
 910                         CTR2(KTR_PTRACE, "PT_STEP: tid %d (pid %d)",
 911                             td2->td_tid, p->p_pid);
 912                         error = ptrace_single_step(td2);
 913                         if (error)
 914                                 goto out;
 915                         break;
 916                 case PT_CONTINUE:
 917                 case PT_TO_SCE:
 918                 case PT_TO_SCX:
 919                 case PT_SYSCALL:
 920                         if (addr != (void *)1) {
 921                                 error = ptrace_set_pc(td2,
 922                                     (u_long)(uintfptr_t)addr);
 923                                 if (error)
 924                                         goto out;
 925                         }
 926                         switch (req) {
 927                         case PT_TO_SCE:
 928                                 p->p_stops |= S_PT_SCE;
 929                                 CTR2(KTR_PTRACE,
 930                                     "PT_TO_SCE: pid %d, stops = %#x", p->p_pid,
 931                                     p->p_stops);
 932                                 break;
 933                         case PT_TO_SCX:
 934                                 p->p_stops |= S_PT_SCX;
 935                                 CTR2(KTR_PTRACE,
 936                                     "PT_TO_SCX: pid %d, stops = %#x", p->p_pid,
 937                                     p->p_stops);
 938                                 break;
 939                         case PT_SYSCALL:
 940                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
 941                                 CTR2(KTR_PTRACE,
 942                                     "PT_SYSCALL: pid %d, stops = %#x", p->p_pid,
 943                                     p->p_stops);
 944                                 break;
 945                         case PT_CONTINUE:
 946                                 CTR1(KTR_PTRACE,
 947                                     "PT_CONTINUE: pid %d", p->p_pid);
 948                                 break;
 949                         }
 950                         break;
 951                 case PT_DETACH:
 952                         /* reset process parent */
 953                         if (p->p_oppid != p->p_pptr->p_pid) {
 954                                 struct proc *pp;
 955
 956                                 PROC_LOCK(p->p_pptr);
 957                                 sigqueue_take(p->p_ksi);
 958                                 PROC_UNLOCK(p->p_pptr);
 959
 960                                 PROC_UNLOCK(p);
 961                                 pp = pfind(p->p_oppid);
 962                                 if (pp == NULL)
 963                                         pp = initproc;
 964                                 else
 965                                         PROC_UNLOCK(pp);
 966                                 PROC_LOCK(p);
 967                                 proc_reparent(p, pp);
 968                                 if (pp == initproc)
 969                                         p->p_sigparent = SIGCHLD;
 970                                 CTR2(KTR_PTRACE,
 971                                     "PT_DETACH: pid %d reparented to pid %d",
 972                                     p->p_pid, pp->p_pid);
 973                         } else
 974                                 CTR1(KTR_PTRACE, "PT_DETACH: pid %d", p->p_pid);
 975                         p->p_oppid = 0;
 976                         p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
 977
 978                         /* should we send SIGCHLD? */
 979                         /* childproc_continued(p); */
 980                         break;
 981                 }
 982
 983         sendsig:
 984                 if (proctree_locked) {
 985                         sx_xunlock(&proctree_lock);
 986                         proctree_locked = 0;
 987                 }
 988                 p->p_xstat = data;
 989                 p->p_xthread = NULL;
 990                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 991                         /* deliver or queue signal */
 992                         td2->td_dbgflags &= ~TDB_XSIG;
 993                         td2->td_xsig = data;
 994
 995                         if (req == PT_DETACH) {
 996                                 struct thread *td3;
 997                                 FOREACH_THREAD_IN_PROC(p, td3) {
 998                                         td3->td_dbgflags &= ~TDB_SUSPEND;
 999                                 }
1000                         }
1001                         /*
1002                          * unsuspend all threads, to not let a thread run,
1003                          * you should use PT_SUSPEND to suspend it before
1004                          * continuing process.
1005                          */
1006                         PROC_SLOCK(p);
1007                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
1008                         thread_unsuspend(p);
1009                         PROC_SUNLOCK(p);
1010                 } else {
1011                         if (data)
1012                                 kern_psignal(p, data);
1013                 }
1014                 break;
1015
1016         case PT_WRITE_I:
1017         case PT_WRITE_D:
1018                 td2->td_dbgflags |= TDB_USERWR;
1019                 write = 1;
1020                 /* FALLTHROUGH */
1021         case PT_READ_I:
1022         case PT_READ_D:
1023                 PROC_UNLOCK(p);
1024                 tmp = 0;
1025                 /* write = 0 set above */
1026                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1027                 iov.iov_len = sizeof(int);
1028                 uio.uio_iov = &iov;
1029                 uio.uio_iovcnt = 1;
1030                 uio.uio_offset = (off_t)(uintptr_t)addr;
1031                 uio.uio_resid = sizeof(int);
1032                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
1033                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1034                 uio.uio_td = td;
1035                 error = proc_rwmem(p, &uio);
1036                 if (uio.uio_resid != 0) {
1037                         /*
1038                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1039                          * so I think write() can bogusly return 0.
1040                          * XXX what happens for short writes?  We don't want
1041                          * to write partial data.
1042                          * XXX proc_rwmem() returns EPERM for other invalid
1043                          * addresses.  Convert this to EINVAL.  Does this
1044                          * clobber returns of EPERM for other reasons?
1045                          */
1046                         if (error == 0 || error == ENOSPC || error == EPERM)
1047                                 error = EINVAL; /* EOF */
1048                 }
1049                 if (!write)
1050                         td->td_retval[0] = tmp;
1051                 if (error == 0) {
1052                         if (write)
1053                                 CTR3(KTR_PTRACE, "PT_WRITE: pid %d: %p <= %#x",
1054                                     p->p_pid, addr, data);
1055                         else
1056                                 CTR3(KTR_PTRACE, "PT_READ: pid %d: %p >= %#x",
1057                                     p->p_pid, addr, tmp);
1058                 }
1059                 PROC_LOCK(p);
1060                 break;
1061
1062         case PT_IO:
1063 #ifdef COMPAT_FREEBSD32
1064                 if (wrap32) {
1065                         piod32 = addr;
1066                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1067                         iov.iov_len = piod32->piod_len;
1068                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1069                         uio.uio_resid = piod32->piod_len;
1070                 } else
1071 #endif
1072                 {
1073                         piod = addr;
1074                         iov.iov_base = piod->piod_addr;
1075                         iov.iov_len = piod->piod_len;
1076                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1077                         uio.uio_resid = piod->piod_len;
1078                 }
1079                 uio.uio_iov = &iov;
1080                 uio.uio_iovcnt = 1;
1081                 uio.uio_segflg = UIO_USERSPACE;
1082                 uio.uio_td = td;
1083 #ifdef COMPAT_FREEBSD32
1084                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1085 #else
1086                 tmp = piod->piod_op;
1087 #endif
1088                 switch (tmp) {
1089                 case PIOD_READ_D:
1090                 case PIOD_READ_I:
1091                         CTR3(KTR_PTRACE, "PT_IO: pid %d: READ (%p, %#x)",
1092                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1093                         uio.uio_rw = UIO_READ;
1094                         break;
1095                 case PIOD_WRITE_D:
1096                 case PIOD_WRITE_I:
1097                         CTR3(KTR_PTRACE, "PT_IO: pid %d: WRITE (%p, %#x)",
1098                             p->p_pid, (uintptr_t)uio.uio_offset, uio.uio_resid);
1099                         td2->td_dbgflags |= TDB_USERWR;
1100                         uio.uio_rw = UIO_WRITE;
1101                         break;
1102                 default:
1103                         error = EINVAL;
1104                         goto out;
1105                 }
1106                 PROC_UNLOCK(p);
1107                 error = proc_rwmem(p, &uio);
1108 #ifdef COMPAT_FREEBSD32
1109                 if (wrap32)
1110                         piod32->piod_len -= uio.uio_resid;
1111                 else
1112 #endif
1113                         piod->piod_len -= uio.uio_resid;
1114                 PROC_LOCK(p);
1115                 break;
1116
1117         case PT_KILL:
1118                 CTR1(KTR_PTRACE, "PT_KILL: pid %d", p->p_pid);
1119                 data = SIGKILL;
1120                 goto sendsig;   /* in PT_CONTINUE above */
1121
1122         case PT_SETREGS:
1123                 CTR2(KTR_PTRACE, "PT_SETREGS: tid %d (pid %d)", td2->td_tid,
1124                     p->p_pid);
1125                 td2->td_dbgflags |= TDB_USERWR;
1126                 error = PROC_WRITE(regs, td2, addr);
1127                 break;
1128
1129         case PT_GETREGS:
1130                 CTR2(KTR_PTRACE, "PT_GETREGS: tid %d (pid %d)", td2->td_tid,
1131                     p->p_pid);
1132                 error = PROC_READ(regs, td2, addr);
1133                 break;
1134
1135         case PT_SETFPREGS:
1136                 CTR2(KTR_PTRACE, "PT_SETFPREGS: tid %d (pid %d)", td2->td_tid,
1137                     p->p_pid);
1138                 td2->td_dbgflags |= TDB_USERWR;
1139                 error = PROC_WRITE(fpregs, td2, addr);
1140                 break;
1141
1142         case PT_GETFPREGS:
1143                 CTR2(KTR_PTRACE, "PT_GETFPREGS: tid %d (pid %d)", td2->td_tid,
1144                     p->p_pid);
1145                 error = PROC_READ(fpregs, td2, addr);
1146                 break;
1147
1148         case PT_SETDBREGS:
1149                 CTR2(KTR_PTRACE, "PT_SETDBREGS: tid %d (pid %d)", td2->td_tid,
1150                     p->p_pid);
1151                 td2->td_dbgflags |= TDB_USERWR;
1152                 error = PROC_WRITE(dbregs, td2, addr);
1153                 break;
1154
1155         case PT_GETDBREGS:
1156                 CTR2(KTR_PTRACE, "PT_GETDBREGS: tid %d (pid %d)", td2->td_tid,
1157                     p->p_pid);
1158                 error = PROC_READ(dbregs, td2, addr);
1159                 break;
1160
1161         case PT_LWPINFO:
1162                 if (data <= 0 ||
1163 #ifdef COMPAT_FREEBSD32
1164                     (!wrap32 && data > sizeof(*pl)) ||
1165                     (wrap32 && data > sizeof(*pl32))) {
1166 #else
1167                     data > sizeof(*pl)) {
1168 #endif
1169                         error = EINVAL;
1170                         break;
1171                 }
1172 #ifdef COMPAT_FREEBSD32
1173                 if (wrap32) {
1174                         pl = &plr;
1175                         pl32 = addr;
1176                 } else
1177 #endif
1178                 pl = addr;
1179                 pl->pl_lwpid = td2->td_tid;
1180                 pl->pl_event = PL_EVENT_NONE;
1181                 pl->pl_flags = 0;
1182                 if (td2->td_dbgflags & TDB_XSIG) {
1183                         pl->pl_event = PL_EVENT_SIGNAL;
1184                         if (td2->td_dbgksi.ksi_signo != 0 &&
1185 #ifdef COMPAT_FREEBSD32
1186                             ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1187                             pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1188                             (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1189                             pl_siginfo) + sizeof(struct siginfo32)))
1190 #else
1191                             data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1192                             + sizeof(pl->pl_siginfo)
1193 #endif
1194                         ){
1195                                 pl->pl_flags |= PL_FLAG_SI;
1196                                 pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1197                         }
1198                 }
1199                 if ((pl->pl_flags & PL_FLAG_SI) == 0)
1200                         bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1201                 if (td2->td_dbgflags & TDB_SCE)
1202                         pl->pl_flags |= PL_FLAG_SCE;
1203                 else if (td2->td_dbgflags & TDB_SCX)
1204                         pl->pl_flags |= PL_FLAG_SCX;
1205                 if (td2->td_dbgflags & TDB_EXEC)
1206                         pl->pl_flags |= PL_FLAG_EXEC;
1207                 if (td2->td_dbgflags & TDB_FORK) {
1208                         pl->pl_flags |= PL_FLAG_FORKED;
1209                         pl->pl_child_pid = td2->td_dbg_forked;
1210                 }
1211                 if (td2->td_dbgflags & TDB_CHILD)
1212                         pl->pl_flags |= PL_FLAG_CHILD;
1213                 pl->pl_sigmask = td2->td_sigmask;
1214                 pl->pl_siglist = td2->td_siglist;
1215                 strcpy(pl->pl_tdname, td2->td_name);
1216 #ifdef COMPAT_FREEBSD32
1217                 if (wrap32)
1218                         ptrace_lwpinfo_to32(pl, pl32);
1219 #endif
1220                 CTR5(KTR_PTRACE,
1221             "PT_LWPINFO: tid %d (pid %d) event %d flags %#x child pid %d",
1222                     td2->td_tid, p->p_pid, pl->pl_event, pl->pl_flags,
1223                     pl->pl_child_pid);
1224                 break;
1225
1226         case PT_GETNUMLWPS:
1227                 CTR2(KTR_PTRACE, "PT_GETNUMLWPS: pid %d: %d threads", p->p_pid,
1228                     p->p_numthreads);
1229                 td->td_retval[0] = p->p_numthreads;
1230                 break;
1231
1232         case PT_GETLWPLIST:
1233                 CTR3(KTR_PTRACE, "PT_GETLWPLIST: pid %d: data %d, actual %d",
1234                     p->p_pid, data, p->p_numthreads);
1235                 if (data <= 0) {
1236                         error = EINVAL;
1237                         break;
1238                 }
1239                 num = imin(p->p_numthreads, data);
1240                 PROC_UNLOCK(p);
1241                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1242                 tmp = 0;
1243                 PROC_LOCK(p);
1244                 FOREACH_THREAD_IN_PROC(p, td2) {
1245                         if (tmp >= num)
1246                                 break;
1247                         buf[tmp++] = td2->td_tid;
1248                 }
1249                 PROC_UNLOCK(p);
1250                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1251                 free(buf, M_TEMP);
1252                 if (!error)
1253                         td->td_retval[0] = tmp;
1254                 PROC_LOCK(p);
1255                 break;
1256
1257         case PT_VM_TIMESTAMP:
1258                 CTR2(KTR_PTRACE, "PT_VM_TIMESTAMP: pid %d: timestamp %d",
1259                     p->p_pid, p->p_vmspace->vm_map.timestamp);
1260                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1261                 break;
1262
1263         case PT_VM_ENTRY:
1264                 PROC_UNLOCK(p);
1265 #ifdef COMPAT_FREEBSD32
1266                 if (wrap32)
1267                         error = ptrace_vm_entry32(td, p, addr);
1268                 else
1269 #endif
1270                 error = ptrace_vm_entry(td, p, addr);
1271                 PROC_LOCK(p);
1272                 break;
1273
1274         default:
1275 #ifdef __HAVE_PTRACE_MACHDEP
1276                 if (req >= PT_FIRSTMACH) {
1277                         PROC_UNLOCK(p);
1278                         error = cpu_ptrace(td2, req, addr, data);
1279                         PROC_LOCK(p);
1280                 } else
1281 #endif
1282                         /* Unknown request. */
1283                         error = EINVAL;
1284                 break;
1285         }
1286
1287 out:
1288         /* Drop our hold on this process now that the request has completed. */
1289         _PRELE(p);
1290 fail:
1291         PROC_UNLOCK(p);
1292         if (proctree_locked)
1293                 sx_xunlock(&proctree_lock);
1294         return (error);
1295 }
1296 #undef PROC_READ
1297 #undef PROC_WRITE
1298
1299 /*
1300  * Stop a process because of a debugging event;
1301  * stay stopped until p->p_step is cleared
1302  * (cleared by PIOCCONT in procfs).
1303  */
1304 void
1305 stopevent(struct proc *p, unsigned int event, unsigned int val)
1306 {
1307
1308         PROC_LOCK_ASSERT(p, MA_OWNED);
1309         p->p_step = 1;
1310         CTR3(KTR_PTRACE, "stopevent: pid %d event %u val %u", p->p_pid, event,
1311             val);
1312         do {
1313                 p->p_xstat = val;
1314                 p->p_xthread = NULL;
1315                 p->p_stype = event;     /* Which event caused the stop? */
1316                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1317                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1318         } while (p->p_step);
1319 }
1320
1321 static int
1322 protect_setchild(struct thread *td, struct proc *p, int flags)
1323 {
1324
1325         PROC_LOCK_ASSERT(p, MA_OWNED);
1326         if (p->p_flag & P_SYSTEM || p_cansched(td, p) != 0)
1327                 return (0);
1328         if (flags & PPROT_SET) {
1329                 p->p_flag |= P_PROTECTED;
1330                 if (flags & PPROT_INHERIT)
1331                         p->p_flag2 |= P2_INHERIT_PROTECTED;
1332         } else {
1333                 p->p_flag &= ~P_PROTECTED;
1334                 p->p_flag2 &= ~P2_INHERIT_PROTECTED;
1335         }
1336         return (1);
1337 }
1338
1339 static int
1340 protect_setchildren(struct thread *td, struct proc *top, int flags)
1341 {
1342         struct proc *p;
1343         int ret;
1344
1345         p = top;
1346         ret = 0;
1347         sx_assert(&proctree_lock, SX_LOCKED);
1348         for (;;) {
1349                 ret |= protect_setchild(td, p, flags);
1350                 PROC_UNLOCK(p);
1351                 /*
1352                  * If this process has children, descend to them next,
1353                  * otherwise do any siblings, and if done with this level,
1354                  * follow back up the tree (but not past top).
1355                  */
1356                 if (!LIST_EMPTY(&p->p_children))
1357                         p = LIST_FIRST(&p->p_children);
1358                 else for (;;) {
1359                         if (p == top) {
1360                                 PROC_LOCK(p);
1361                                 return (ret);
1362                         }
1363                         if (LIST_NEXT(p, p_sibling)) {
1364                                 p = LIST_NEXT(p, p_sibling);
1365                                 break;
1366                         }
1367                         p = p->p_pptr;
1368                 }
1369                 PROC_LOCK(p);
1370         }
1371 }
1372
1373 static int
1374 protect_set(struct thread *td, struct proc *p, int flags)
1375 {
1376         int error, ret;
1377
1378         switch (PPROT_OP(flags)) {
1379         case PPROT_SET:
1380         case PPROT_CLEAR:
1381                 break;
1382         default:
1383                 return (EINVAL);
1384         }
1385
1386         if ((PPROT_FLAGS(flags) & ~(PPROT_DESCEND | PPROT_INHERIT)) != 0)
1387                 return (EINVAL);
1388
1389         error = priv_check(td, PRIV_VM_MADV_PROTECT);
1390         if (error)
1391                 return (error);
1392
1393         if (flags & PPROT_DESCEND)
1394                 ret = protect_setchildren(td, p, flags);
1395         else
1396                 ret = protect_setchild(td, p, flags);
1397         if (ret == 0)
1398                 return (EPERM);
1399         return (0);
1400 }
1401
1402 #ifndef _SYS_SYSPROTO_H_
1403 struct procctl_args {
1404         idtype_t idtype;
1405         id_t    id;
1406         int     com;
1407         void    *data;
1408 };
1409 #endif
1410 /* ARGSUSED */
1411 int
1412 sys_procctl(struct thread *td, struct procctl_args *uap)
1413 {
1414         int error, flags;
1415         void *data;
1416
1417         switch (uap->com) {
1418         case PROC_SPROTECT:
1419                 error = copyin(uap->data, &flags, sizeof(flags));
1420                 if (error)
1421                         return (error);
1422                 data = &flags;
1423                 break;
1424         default:
1425                 return (EINVAL);
1426         }
1427
1428         return (kern_procctl(td, uap->idtype, uap->id, uap->com, data));
1429 }
1430
1431 static int
1432 kern_procctl_single(struct thread *td, struct proc *p, int com, void *data)
1433 {
1434
1435         PROC_LOCK_ASSERT(p, MA_OWNED);
1436         switch (com) {
1437         case PROC_SPROTECT:
1438                 return (protect_set(td, p, *(int *)data));
1439         default:
1440                 return (EINVAL);
1441         }
1442 }
1443
1444 int
1445 kern_procctl(struct thread *td, idtype_t idtype, id_t id, int com, void *data)
1446 {
1447         struct pgrp *pg;
1448         struct proc *p;
1449         int error, first_error, ok;
1450
1451         sx_slock(&proctree_lock);
1452         switch (idtype) {
1453         case P_PID:
1454                 p = pfind(id);
1455                 if (p == NULL) {
1456                         error = ESRCH;
1457                         break;
1458                 }
1459                 if (p->p_state == PRS_NEW)
1460                         error = ESRCH;
1461                 else
1462                         error = p_cansee(td, p);
1463                 if (error == 0)
1464                         error = kern_procctl_single(td, p, com, data);
1465                 PROC_UNLOCK(p);
1466                 break;
1467         case P_PGID:
1468                 /*
1469                  * Attempt to apply the operation to all members of the
1470                  * group.  Ignore processes in the group that can't be
1471                  * seen.  Ignore errors so long as at least one process is
1472                  * able to complete the request successfully.
1473                  */
1474                 pg = pgfind(id);
1475                 if (pg == NULL) {
1476                         error = ESRCH;
1477                         break;
1478                 }
1479                 PGRP_UNLOCK(pg);
1480                 ok = 0;
1481                 first_error = 0;
1482                 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1483                         PROC_LOCK(p);
1484                         if (p->p_state == PRS_NEW || p_cansee(td, p) != 0) {
1485                                 PROC_UNLOCK(p);
1486                                 continue;
1487                         }
1488                         error = kern_procctl_single(td, p, com, data);
1489                         PROC_UNLOCK(p);
1490                         if (error == 0)
1491                                 ok = 1;
1492                         else if (first_error == 0)
1493                                 first_error = error;
1494                 }
1495                 if (ok)
1496                         error = 0;
1497                 else if (first_error != 0)
1498                         error = first_error;
1499                 else
1500                         /*
1501                          * Was not able to see any processes in the
1502                          * process group.
1503                          */
1504                         error = ESRCH;
1505                 break;
1506         default:
1507                 error = EINVAL;
1508                 break;
1509         }
1510         sx_sunlock(&proctree_lock);
1511         return (error);
1512 }