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/sysproto.h>
  43 #include <sys/proc.h>
  44 #include <sys/vnode.h>
  45 #include <sys/ptrace.h>
  46 #include <sys/sx.h>
  47 #include <sys/malloc.h>
  48 #include <sys/signalvar.h>
  49
  50 #include <machine/reg.h>
  51
  52 #include <security/audit/audit.h>
  53
  54 #include <vm/vm.h>
  55 #include <vm/pmap.h>
  56 #include <vm/vm_extern.h>
  57 #include <vm/vm_map.h>
  58 #include <vm/vm_kern.h>
  59 #include <vm/vm_object.h>
  60 #include <vm/vm_page.h>
  61
  62 #ifdef COMPAT_IA32
  63 #include <sys/procfs.h>
  64 #include <machine/fpu.h>
  65 #include <compat/ia32/ia32_reg.h>
  66
  67 extern struct sysentvec ia32_freebsd_sysvec;
  68
  69 struct ptrace_io_desc32 {
  70         int             piod_op;
  71         u_int32_t       piod_offs;
  72         u_int32_t       piod_addr;
  73         u_int32_t       piod_len;
  74 };
  75 #endif
  76
  77 /*
  78  * Functions implemented using PROC_ACTION():
  79  *
  80  * proc_read_regs(proc, regs)
  81  *      Get the current user-visible register set from the process
  82  *      and copy it into the regs structure (<machine/reg.h>).
  83  *      The process is stopped at the time read_regs is called.
  84  *
  85  * proc_write_regs(proc, regs)
  86  *      Update the current register set from the passed in regs
  87  *      structure.  Take care to avoid clobbering special CPU
  88  *      registers or privileged bits in the PSL.
  89  *      Depending on the architecture this may have fix-up work to do,
  90  *      especially if the IAR or PCW are modified.
  91  *      The process is stopped at the time write_regs is called.
  92  *
  93  * proc_read_fpregs, proc_write_fpregs
  94  *      deal with the floating point register set, otherwise as above.
  95  *
  96  * proc_read_dbregs, proc_write_dbregs
  97  *      deal with the processor debug register set, otherwise as above.
  98  *
  99  * proc_sstep(proc)
 100  *      Arrange for the process to trap after executing a single instruction.
 101  */
 102
 103 #define PROC_ACTION(action) do {                                        \
 104         int error;                                                      \
 105                                                                         \
 106         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 107         if ((td->td_proc->p_sflag & PS_INMEM) == 0)                     \
 108                 error = EIO;                                            \
 109         else                                                            \
 110                 error = (action);                                       \
 111         return (error);                                                 \
 112 } while(0)
 113
 114 int
 115 proc_read_regs(struct thread *td, struct reg *regs)
 116 {
 117
 118         PROC_ACTION(fill_regs(td, regs));
 119 }
 120
 121 int
 122 proc_write_regs(struct thread *td, struct reg *regs)
 123 {
 124
 125         PROC_ACTION(set_regs(td, regs));
 126 }
 127
 128 int
 129 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 130 {
 131
 132         PROC_ACTION(fill_dbregs(td, dbregs));
 133 }
 134
 135 int
 136 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 137 {
 138
 139         PROC_ACTION(set_dbregs(td, dbregs));
 140 }
 141
 142 /*
 143  * Ptrace doesn't support fpregs at all, and there are no security holes
 144  * or translations for fpregs, so we can just copy them.
 145  */
 146 int
 147 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 148 {
 149
 150         PROC_ACTION(fill_fpregs(td, fpregs));
 151 }
 152
 153 int
 154 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 155 {
 156
 157         PROC_ACTION(set_fpregs(td, fpregs));
 158 }
 159
 160 #ifdef COMPAT_IA32
 161 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 162 int
 163 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 164 {
 165
 166         PROC_ACTION(fill_regs32(td, regs32));
 167 }
 168
 169 int
 170 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 171 {
 172
 173         PROC_ACTION(set_regs32(td, regs32));
 174 }
 175
 176 int
 177 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 178 {
 179
 180         PROC_ACTION(fill_dbregs32(td, dbregs32));
 181 }
 182
 183 int
 184 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 185 {
 186
 187         PROC_ACTION(set_dbregs32(td, dbregs32));
 188 }
 189
 190 int
 191 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 192 {
 193
 194         PROC_ACTION(fill_fpregs32(td, fpregs32));
 195 }
 196
 197 int
 198 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 199 {
 200
 201         PROC_ACTION(set_fpregs32(td, fpregs32));
 202 }
 203 #endif
 204
 205 int
 206 proc_sstep(struct thread *td)
 207 {
 208
 209         PROC_ACTION(ptrace_single_step(td));
 210 }
 211
 212 int
 213 proc_rwmem(struct proc *p, struct uio *uio)
 214 {
 215         vm_map_t map;
 216         vm_object_t backing_object, object = NULL;
 217         vm_offset_t pageno = 0;         /* page number */
 218         vm_prot_t reqprot;
 219         int error, writing;
 220
 221         /*
 222          * Assert that someone has locked this vmspace.  (Should be
 223          * curthread but we can't assert that.)  This keeps the process
 224          * from exiting out from under us until this operation completes.
 225          */
 226         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 227             p, p->p_pid));
 228
 229         /*
 230          * The map we want...
 231          */
 232         map = &p->p_vmspace->vm_map;
 233
 234         writing = uio->uio_rw == UIO_WRITE;
 235         reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
 236             VM_PROT_READ;
 237
 238         /*
 239          * Only map in one page at a time.  We don't have to, but it
 240          * makes things easier.  This way is trivial - right?
 241          */
 242         do {
 243                 vm_map_t tmap;
 244                 vm_offset_t uva;
 245                 int page_offset;                /* offset into page */
 246                 vm_map_entry_t out_entry;
 247                 vm_prot_t out_prot;
 248                 boolean_t wired;
 249                 vm_pindex_t pindex;
 250                 u_int len;
 251                 vm_page_t m;
 252
 253                 object = NULL;
 254
 255                 uva = (vm_offset_t)uio->uio_offset;
 256
 257                 /*
 258                  * Get the page number of this segment.
 259                  */
 260                 pageno = trunc_page(uva);
 261                 page_offset = uva - pageno;
 262
 263                 /*
 264                  * How many bytes to copy
 265                  */
 266                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 267
 268                 /*
 269                  * Fault the page on behalf of the process
 270                  */
 271                 error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
 272                 if (error) {
 273                         error = EFAULT;
 274                         break;
 275                 }
 276
 277                 /*
 278                  * Now we need to get the page.  out_entry, out_prot, wired,
 279                  * and single_use aren't used.  One would think the vm code
 280                  * would be a *bit* nicer...  We use tmap because
 281                  * vm_map_lookup() can change the map argument.
 282                  */
 283                 tmap = map;
 284                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 285                     &object, &pindex, &out_prot, &wired);
 286                 if (error) {
 287                         error = EFAULT;
 288                         break;
 289                 }
 290                 VM_OBJECT_LOCK(object);
 291                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 292                     !writing &&
 293                     (backing_object = object->backing_object) != NULL) {
 294                         /*
 295                          * Allow fallback to backing objects if we are reading.
 296                          */
 297                         VM_OBJECT_LOCK(backing_object);
 298                         pindex += OFF_TO_IDX(object->backing_object_offset);
 299                         VM_OBJECT_UNLOCK(object);
 300                         object = backing_object;
 301                 }
 302                 VM_OBJECT_UNLOCK(object);
 303                 if (m == NULL) {
 304                         vm_map_lookup_done(tmap, out_entry);
 305                         error = EFAULT;
 306                         break;
 307                 }
 308
 309                 /*
 310                  * Hold the page in memory.
 311                  */
 312                 vm_page_lock_queues();
 313                 vm_page_hold(m);
 314                 vm_page_unlock_queues();
 315
 316                 /*
 317                  * We're done with tmap now.
 318                  */
 319                 vm_map_lookup_done(tmap, out_entry);
 320
 321                 /*
 322                  * Now do the i/o move.
 323                  */
 324                 error = uiomove_fromphys(&m, page_offset, len, uio);
 325
 326                 /*
 327                  * Release the page.
 328                  */
 329                 vm_page_lock_queues();
 330                 vm_page_unhold(m);
 331                 vm_page_unlock_queues();
 332
 333         } while (error == 0 && uio->uio_resid > 0);
 334
 335         return (error);
 336 }
 337
 338 /*
 339  * Process debugging system call.
 340  */
 341 #ifndef _SYS_SYSPROTO_H_
 342 struct ptrace_args {
 343         int     req;
 344         pid_t   pid;
 345         caddr_t addr;
 346         int     data;
 347 };
 348 #endif
 349
 350 #ifdef COMPAT_IA32
 351 /*
 352  * This CPP subterfuge is to try and reduce the number of ifdefs in
 353  * the body of the code.
 354  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 355  * becomes either:
 356  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 357  * or
 358  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 359  * .. except this is done at runtime.
 360  */
 361 #define COPYIN(u, k, s)         wrap32 ? \
 362         copyin(u, k ## 32, s ## 32) : \
 363         copyin(u, k, s)
 364 #define COPYOUT(k, u, s)        wrap32 ? \
 365         copyout(k ## 32, u, s ## 32) : \
 366         copyout(k, u, s)
 367 #else
 368 #define COPYIN(u, k, s)         copyin(u, k, s)
 369 #define COPYOUT(k, u, s)        copyout(k, u, s)
 370 #endif
 371 /*
 372  * MPSAFE
 373  */
 374 int
 375 ptrace(struct thread *td, struct ptrace_args *uap)
 376 {
 377         /*
 378          * XXX this obfuscation is to reduce stack usage, but the register
 379          * structs may be too large to put on the stack anyway.
 380          */
 381         union {
 382                 struct ptrace_io_desc piod;
 383                 struct ptrace_lwpinfo pl;
 384                 struct dbreg dbreg;
 385                 struct fpreg fpreg;
 386                 struct reg reg;
 387 #ifdef COMPAT_IA32
 388                 struct dbreg32 dbreg32;
 389                 struct fpreg32 fpreg32;
 390                 struct reg32 reg32;
 391                 struct ptrace_io_desc32 piod32;
 392 #endif
 393         } r;
 394         void *addr;
 395         int error = 0;
 396 #ifdef COMPAT_IA32
 397         int wrap32 = 0;
 398
 399         if (td->td_proc->p_sysent == &ia32_freebsd_sysvec)
 400                 wrap32 = 1;
 401 #endif
 402         AUDIT_ARG(pid, uap->pid);
 403         AUDIT_ARG(cmd, uap->req);
 404         AUDIT_ARG(addr, uap->addr);
 405         AUDIT_ARG(value, uap->data);
 406         addr = &r;
 407         switch (uap->req) {
 408         case PT_GETREGS:
 409         case PT_GETFPREGS:
 410         case PT_GETDBREGS:
 411         case PT_LWPINFO:
 412                 break;
 413         case PT_SETREGS:
 414                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 415                 break;
 416         case PT_SETFPREGS:
 417                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 418                 break;
 419         case PT_SETDBREGS:
 420                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 421                 break;
 422         case PT_IO:
 423                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 424                 break;
 425         default:
 426                 addr = uap->addr;
 427                 break;
 428         }
 429         if (error)
 430                 return (error);
 431
 432         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 433         if (error)
 434                 return (error);
 435
 436         switch (uap->req) {
 437         case PT_IO:
 438                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 439                 break;
 440         case PT_GETREGS:
 441                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 442                 break;
 443         case PT_GETFPREGS:
 444                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 445                 break;
 446         case PT_GETDBREGS:
 447                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 448                 break;
 449         case PT_LWPINFO:
 450                 error = copyout(&r.pl, uap->addr, uap->data);
 451                 break;
 452         }
 453
 454         return (error);
 455 }
 456 #undef COPYIN
 457 #undef COPYOUT
 458
 459 #ifdef COMPAT_IA32
 460 /*
 461  *   PROC_READ(regs, td2, addr);
 462  * becomes either:
 463  *   proc_read_regs(td2, addr);
 464  * or
 465  *   proc_read_regs32(td2, addr);
 466  * .. except this is done at runtime.  There is an additional
 467  * complication in that PROC_WRITE disallows 32 bit consumers
 468  * from writing to 64 bit address space targets.
 469  */
 470 #define PROC_READ(w, t, a)      wrap32 ? \
 471         proc_read_ ## w ## 32(t, a) : \
 472         proc_read_ ## w (t, a)
 473 #define PROC_WRITE(w, t, a)     wrap32 ? \
 474         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 475         proc_write_ ## w (t, a)
 476 #else
 477 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 478 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 479 #endif
 480
 481 int
 482 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 483 {
 484         struct iovec iov;
 485         struct uio uio;
 486         struct proc *curp, *p, *pp;
 487         struct thread *td2 = NULL;
 488         struct ptrace_io_desc *piod = NULL;
 489         struct ptrace_lwpinfo *pl;
 490         int error, write, tmp, num;
 491         int proctree_locked = 0;
 492         lwpid_t tid = 0, *buf;
 493 #ifdef COMPAT_IA32
 494         int wrap32 = 0, safe = 0;
 495         struct ptrace_io_desc32 *piod32 = NULL;
 496 #endif
 497
 498         curp = td->td_proc;
 499
 500         /* Lock proctree before locking the process. */
 501         switch (req) {
 502         case PT_TRACE_ME:
 503         case PT_ATTACH:
 504         case PT_STEP:
 505         case PT_CONTINUE:
 506         case PT_TO_SCE:
 507         case PT_TO_SCX:
 508         case PT_SYSCALL:
 509         case PT_DETACH:
 510                 sx_xlock(&proctree_lock);
 511                 proctree_locked = 1;
 512                 break;
 513         default:
 514                 break;
 515         }
 516
 517         write = 0;
 518         if (req == PT_TRACE_ME) {
 519                 p = td->td_proc;
 520                 PROC_LOCK(p);
 521         } else {
 522                 if (pid <= PID_MAX) {
 523                         if ((p = pfind(pid)) == NULL) {
 524                                 if (proctree_locked)
 525                                         sx_xunlock(&proctree_lock);
 526                                 return (ESRCH);
 527                         }
 528                 } else {
 529                         /* this is slow, should be optimized */
 530                         sx_slock(&allproc_lock);
 531                         FOREACH_PROC_IN_SYSTEM(p) {
 532                                 PROC_LOCK(p);
 533                                 mtx_lock_spin(&sched_lock);
 534                                 FOREACH_THREAD_IN_PROC(p, td2) {
 535                                         if (td2->td_tid == pid)
 536                                                 break;
 537                                 }
 538                                 mtx_unlock_spin(&sched_lock);
 539                                 if (td2 != NULL)
 540                                         break; /* proc lock held */
 541                                 PROC_UNLOCK(p);
 542                         }
 543                         sx_sunlock(&allproc_lock);
 544                         if (p == NULL) {
 545                                 if (proctree_locked)
 546                                         sx_xunlock(&proctree_lock);
 547                                 return (ESRCH);
 548                         }
 549                         tid = pid;
 550                         pid = p->p_pid;
 551                 }
 552         }
 553         AUDIT_ARG(process, p);
 554
 555         if ((p->p_flag & P_WEXIT) != 0) {
 556                 error = ESRCH;
 557                 goto fail;
 558         }
 559         if ((error = p_cansee(td, p)) != 0)
 560                 goto fail;
 561
 562         if ((error = p_candebug(td, p)) != 0)
 563                 goto fail;
 564
 565         /*
 566          * System processes can't be debugged.
 567          */
 568         if ((p->p_flag & P_SYSTEM) != 0) {
 569                 error = EINVAL;
 570                 goto fail;
 571         }
 572
 573         if (tid == 0) {
 574                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 575                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 576                         td2 = p->p_xthread;
 577                 } else {
 578                         td2 = FIRST_THREAD_IN_PROC(p);
 579                 }
 580                 tid = td2->td_tid;
 581         }
 582
 583 #ifdef COMPAT_IA32
 584         /*
 585          * Test if we're a 32 bit client and what the target is.
 586          * Set the wrap controls accordingly.
 587          */
 588         if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) {
 589                 if (td2->td_proc->p_sysent == &ia32_freebsd_sysvec)
 590                         safe = 1;
 591                 wrap32 = 1;
 592         }
 593 #endif
 594         /*
 595          * Permissions check
 596          */
 597         switch (req) {
 598         case PT_TRACE_ME:
 599                 /* Always legal. */
 600                 break;
 601
 602         case PT_ATTACH:
 603                 /* Self */
 604                 if (p->p_pid == td->td_proc->p_pid) {
 605                         error = EINVAL;
 606                         goto fail;
 607                 }
 608
 609                 /* Already traced */
 610                 if (p->p_flag & P_TRACED) {
 611                         error = EBUSY;
 612                         goto fail;
 613                 }
 614
 615                 /* Can't trace an ancestor if you're being traced. */
 616                 if (curp->p_flag & P_TRACED) {
 617                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 618                                 if (pp == p) {
 619                                         error = EINVAL;
 620                                         goto fail;
 621                                 }
 622                         }
 623                 }
 624
 625
 626                 /* OK */
 627                 break;
 628
 629         case PT_CLEARSTEP:
 630                 /* Allow thread to clear single step for itself */
 631                 if (td->td_tid == tid)
 632                         break;
 633
 634                 /* FALLTHROUGH */
 635         default:
 636                 /* not being traced... */
 637                 if ((p->p_flag & P_TRACED) == 0) {
 638                         error = EPERM;
 639                         goto fail;
 640                 }
 641
 642                 /* not being traced by YOU */
 643                 if (p->p_pptr != td->td_proc) {
 644                         error = EBUSY;
 645                         goto fail;
 646                 }
 647
 648                 /* not currently stopped */
 649                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 650                     p->p_suspcount != p->p_numthreads  ||
 651                     (p->p_flag & P_WAITED) == 0) {
 652                         error = EBUSY;
 653                         goto fail;
 654                 }
 655
 656                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 657                         static int count = 0;
 658                         if (count++ == 0)
 659                                 printf("P_STOPPED_TRACE not set.\n");
 660                 }
 661
 662                 /* OK */
 663                 break;
 664         }
 665
 666         /* Keep this process around until we finish this request. */
 667         _PHOLD(p);
 668
 669 #ifdef FIX_SSTEP
 670         /*
 671          * Single step fixup ala procfs
 672          */
 673         FIX_SSTEP(td2);
 674 #endif
 675
 676         /*
 677          * Actually do the requests
 678          */
 679
 680         td->td_retval[0] = 0;
 681
 682         switch (req) {
 683         case PT_TRACE_ME:
 684                 /* set my trace flag and "owner" so it can read/write me */
 685                 p->p_flag |= P_TRACED;
 686                 p->p_oppid = p->p_pptr->p_pid;
 687                 break;
 688
 689         case PT_ATTACH:
 690                 /* security check done above */
 691                 p->p_flag |= P_TRACED;
 692                 p->p_oppid = p->p_pptr->p_pid;
 693                 if (p->p_pptr != td->td_proc)
 694                         proc_reparent(p, td->td_proc);
 695                 data = SIGSTOP;
 696                 goto sendsig;   /* in PT_CONTINUE below */
 697
 698         case PT_CLEARSTEP:
 699                 error = ptrace_clear_single_step(td2);
 700                 break;
 701
 702         case PT_SETSTEP:
 703                 error = ptrace_single_step(td2);
 704                 break;
 705
 706         case PT_SUSPEND:
 707                 mtx_lock_spin(&sched_lock);
 708                 td2->td_flags |= TDF_DBSUSPEND;
 709                 mtx_unlock_spin(&sched_lock);
 710                 break;
 711
 712         case PT_RESUME:
 713                 mtx_lock_spin(&sched_lock);
 714                 td2->td_flags &= ~TDF_DBSUSPEND;
 715                 mtx_unlock_spin(&sched_lock);
 716                 break;
 717
 718         case PT_STEP:
 719         case PT_CONTINUE:
 720         case PT_TO_SCE:
 721         case PT_TO_SCX:
 722         case PT_SYSCALL:
 723         case PT_DETACH:
 724                 /* Zero means do not send any signal */
 725                 if (data < 0 || data > _SIG_MAXSIG) {
 726                         error = EINVAL;
 727                         break;
 728                 }
 729
 730                 switch (req) {
 731                 case PT_STEP:
 732                         error = ptrace_single_step(td2);
 733                         if (error)
 734                                 goto out;
 735                         break;
 736                 case PT_TO_SCE:
 737                         p->p_stops |= S_PT_SCE;
 738                         break;
 739                 case PT_TO_SCX:
 740                         p->p_stops |= S_PT_SCX;
 741                         break;
 742                 case PT_SYSCALL:
 743                         p->p_stops |= S_PT_SCE | S_PT_SCX;
 744                         break;
 745                 }
 746
 747                 if (addr != (void *)1) {
 748                         error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
 749                         if (error)
 750                                 break;
 751                 }
 752
 753                 if (req == PT_DETACH) {
 754                         /* reset process parent */
 755                         if (p->p_oppid != p->p_pptr->p_pid) {
 756                                 struct proc *pp;
 757
 758                                 PROC_LOCK(p->p_pptr);
 759                                 sigqueue_take(p->p_ksi);
 760                                 PROC_UNLOCK(p->p_pptr);
 761
 762                                 PROC_UNLOCK(p);
 763                                 pp = pfind(p->p_oppid);
 764                                 if (pp == NULL)
 765                                         pp = initproc;
 766                                 else
 767                                         PROC_UNLOCK(pp);
 768                                 PROC_LOCK(p);
 769                                 proc_reparent(p, pp);
 770                                 if (pp == initproc)
 771                                         p->p_sigparent = SIGCHLD;
 772                         }
 773                         p->p_flag &= ~(P_TRACED | P_WAITED);
 774                         p->p_oppid = 0;
 775
 776                         /* should we send SIGCHLD? */
 777                         /* childproc_continued(p); */
 778                 }
 779
 780         sendsig:
 781                 if (proctree_locked) {
 782                         sx_xunlock(&proctree_lock);
 783                         proctree_locked = 0;
 784                 }
 785                 /* deliver or queue signal */
 786                 mtx_lock_spin(&sched_lock);
 787                 td2->td_flags &= ~TDF_XSIG;
 788                 mtx_unlock_spin(&sched_lock);
 789                 td2->td_xsig = data;
 790                 p->p_xstat = data;
 791                 p->p_xthread = NULL;
 792                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 793                         mtx_lock_spin(&sched_lock);
 794                         if (req == PT_DETACH) {
 795                                 struct thread *td3;
 796                                 FOREACH_THREAD_IN_PROC(p, td3)
 797                                         td3->td_flags &= ~TDF_DBSUSPEND;
 798                         }
 799                         /*
 800                          * unsuspend all threads, to not let a thread run,
 801                          * you should use PT_SUSPEND to suspend it before
 802                          * continuing process.
 803                          */
 804                         mtx_unlock_spin(&sched_lock);
 805 #ifdef KSE
 806                         thread_continued(p);
 807 #endif
 808                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
 809                         mtx_lock_spin(&sched_lock);
 810                         thread_unsuspend(p);
 811                         mtx_unlock_spin(&sched_lock);
 812                 }
 813
 814                 if (data)
 815                         psignal(p, data);
 816
 817                 break;
 818
 819         case PT_WRITE_I:
 820         case PT_WRITE_D:
 821                 write = 1;
 822                 /* FALLTHROUGH */
 823         case PT_READ_I:
 824         case PT_READ_D:
 825                 PROC_UNLOCK(p);
 826                 tmp = 0;
 827                 /* write = 0 set above */
 828                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
 829                 iov.iov_len = sizeof(int);
 830                 uio.uio_iov = &iov;
 831                 uio.uio_iovcnt = 1;
 832                 uio.uio_offset = (off_t)(uintptr_t)addr;
 833                 uio.uio_resid = sizeof(int);
 834                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
 835                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
 836                 uio.uio_td = td;
 837                 error = proc_rwmem(p, &uio);
 838                 if (uio.uio_resid != 0) {
 839                         /*
 840                          * XXX proc_rwmem() doesn't currently return ENOSPC,
 841                          * so I think write() can bogusly return 0.
 842                          * XXX what happens for short writes?  We don't want
 843                          * to write partial data.
 844                          * XXX proc_rwmem() returns EPERM for other invalid
 845                          * addresses.  Convert this to EINVAL.  Does this
 846                          * clobber returns of EPERM for other reasons?
 847                          */
 848                         if (error == 0 || error == ENOSPC || error == EPERM)
 849                                 error = EINVAL; /* EOF */
 850                 }
 851                 if (!write)
 852                         td->td_retval[0] = tmp;
 853                 PROC_LOCK(p);
 854                 break;
 855
 856         case PT_IO:
 857 #ifdef COMPAT_IA32
 858                 if (wrap32) {
 859                         piod32 = addr;
 860                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
 861                         iov.iov_len = piod32->piod_len;
 862                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
 863                         uio.uio_resid = piod32->piod_len;
 864                 } else
 865 #endif
 866                 {
 867                         piod = addr;
 868                         iov.iov_base = piod->piod_addr;
 869                         iov.iov_len = piod->piod_len;
 870                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
 871                         uio.uio_resid = piod->piod_len;
 872                 }
 873                 uio.uio_iov = &iov;
 874                 uio.uio_iovcnt = 1;
 875                 uio.uio_segflg = UIO_USERSPACE;
 876                 uio.uio_td = td;
 877 #ifdef COMPAT_IA32
 878                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
 879 #else
 880                 tmp = piod->piod_op;
 881 #endif
 882                 switch (tmp) {
 883                 case PIOD_READ_D:
 884                 case PIOD_READ_I:
 885                         uio.uio_rw = UIO_READ;
 886                         break;
 887                 case PIOD_WRITE_D:
 888                 case PIOD_WRITE_I:
 889                         uio.uio_rw = UIO_WRITE;
 890                         break;
 891                 default:
 892                         error = EINVAL;
 893                         goto out;
 894                 }
 895                 PROC_UNLOCK(p);
 896                 error = proc_rwmem(p, &uio);
 897 #ifdef COMPAT_IA32
 898                 if (wrap32)
 899                         piod32->piod_len -= uio.uio_resid;
 900                 else
 901 #endif
 902                         piod->piod_len -= uio.uio_resid;
 903                 PROC_LOCK(p);
 904                 break;
 905
 906         case PT_KILL:
 907                 data = SIGKILL;
 908                 goto sendsig;   /* in PT_CONTINUE above */
 909
 910         case PT_SETREGS:
 911                 error = PROC_WRITE(regs, td2, addr);
 912                 break;
 913
 914         case PT_GETREGS:
 915                 error = PROC_READ(regs, td2, addr);
 916                 break;
 917
 918         case PT_SETFPREGS:
 919                 error = PROC_WRITE(fpregs, td2, addr);
 920                 break;
 921
 922         case PT_GETFPREGS:
 923                 error = PROC_READ(fpregs, td2, addr);
 924                 break;
 925
 926         case PT_SETDBREGS:
 927                 error = PROC_WRITE(dbregs, td2, addr);
 928                 break;
 929
 930         case PT_GETDBREGS:
 931                 error = PROC_READ(dbregs, td2, addr);
 932                 break;
 933
 934         case PT_LWPINFO:
 935                 if (data <= 0 || data > sizeof(*pl)) {
 936                         error = EINVAL;
 937                         break;
 938                 }
 939                 pl = addr;
 940                 pl->pl_lwpid = td2->td_tid;
 941                 if (td2->td_flags & TDF_XSIG)
 942                         pl->pl_event = PL_EVENT_SIGNAL;
 943                 else
 944                         pl->pl_event = 0;
 945 #ifdef KSE
 946                 if (td2->td_pflags & TDP_SA) {
 947                         pl->pl_flags = PL_FLAG_SA;
 948                         if (td2->td_upcall && !TD_CAN_UNBIND(td2))
 949                                 pl->pl_flags |= PL_FLAG_BOUND;
 950                 } else {
 951                         pl->pl_flags = 0;
 952                 }
 953 #else
 954                 pl->pl_flags = 0;
 955 #endif
 956                 pl->pl_sigmask = td2->td_sigmask;
 957                 pl->pl_siglist = td2->td_siglist;
 958                 break;
 959
 960         case PT_GETNUMLWPS:
 961                 td->td_retval[0] = p->p_numthreads;
 962                 break;
 963
 964         case PT_GETLWPLIST:
 965                 if (data <= 0) {
 966                         error = EINVAL;
 967                         break;
 968                 }
 969                 num = imin(p->p_numthreads, data);
 970                 PROC_UNLOCK(p);
 971                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
 972                 tmp = 0;
 973                 PROC_LOCK(p);
 974                 mtx_lock_spin(&sched_lock);
 975                 FOREACH_THREAD_IN_PROC(p, td2) {
 976                         if (tmp >= num)
 977                                 break;
 978                         buf[tmp++] = td2->td_tid;
 979                 }
 980                 mtx_unlock_spin(&sched_lock);
 981                 PROC_UNLOCK(p);
 982                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
 983                 free(buf, M_TEMP);
 984                 if (!error)
 985                         td->td_retval[0] = tmp;
 986                 PROC_LOCK(p);
 987                 break;
 988
 989         default:
 990 #ifdef __HAVE_PTRACE_MACHDEP
 991                 if (req >= PT_FIRSTMACH) {
 992                         PROC_UNLOCK(p);
 993                         error = cpu_ptrace(td2, req, addr, data);
 994                         PROC_LOCK(p);
 995                 } else
 996 #endif
 997                         /* Unknown request. */
 998                         error = EINVAL;
 999                 break;
1000         }
1001
1002 out:
1003         /* Drop our hold on this process now that the request has completed. */
1004         _PRELE(p);
1005 fail:
1006         PROC_UNLOCK(p);
1007         if (proctree_locked)
1008                 sx_xunlock(&proctree_lock);
1009         return (error);
1010 }
1011 #undef PROC_READ
1012 #undef PROC_WRITE
1013
1014 /*
1015  * Stop a process because of a debugging event;
1016  * stay stopped until p->p_step is cleared
1017  * (cleared by PIOCCONT in procfs).
1018  */
1019 void
1020 stopevent(struct proc *p, unsigned int event, unsigned int val)
1021 {
1022
1023         PROC_LOCK_ASSERT(p, MA_OWNED);
1024         p->p_step = 1;
1025         do {
1026                 p->p_xstat = val;
1027                 p->p_xthread = NULL;
1028                 p->p_stype = event;     /* Which event caused the stop? */
1029                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1030                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1031         } while (p->p_step);
1032 }