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         struct vmspace *vm;
 216         vm_map_t map;
 217         vm_object_t backing_object, object = NULL;
 218         vm_offset_t pageno = 0;         /* page number */
 219         vm_prot_t reqprot;
 220         int error, refcnt, writing;
 221
 222         /*
 223          * if the vmspace is in the midst of being deallocated or the
 224          * process is exiting, don't try to grab anything.  The page table
 225          * usage in that process can be messed up.
 226          */
 227         vm = p->p_vmspace;
 228         if ((p->p_flag & P_WEXIT))
 229                 return (EFAULT);
 230         do {
 231                 if ((refcnt = vm->vm_refcnt) < 1)
 232                         return (EFAULT);
 233         } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
 234
 235         /*
 236          * The map we want...
 237          */
 238         map = &vm->vm_map;
 239
 240         writing = uio->uio_rw == UIO_WRITE;
 241         reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
 242             VM_PROT_READ;
 243
 244         /*
 245          * Only map in one page at a time.  We don't have to, but it
 246          * makes things easier.  This way is trivial - right?
 247          */
 248         do {
 249                 vm_map_t tmap;
 250                 vm_offset_t uva;
 251                 int page_offset;                /* offset into page */
 252                 vm_map_entry_t out_entry;
 253                 vm_prot_t out_prot;
 254                 boolean_t wired;
 255                 vm_pindex_t pindex;
 256                 u_int len;
 257                 vm_page_t m;
 258
 259                 object = NULL;
 260
 261                 uva = (vm_offset_t)uio->uio_offset;
 262
 263                 /*
 264                  * Get the page number of this segment.
 265                  */
 266                 pageno = trunc_page(uva);
 267                 page_offset = uva - pageno;
 268
 269                 /*
 270                  * How many bytes to copy
 271                  */
 272                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 273
 274                 /*
 275                  * Fault the page on behalf of the process
 276                  */
 277                 error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
 278                 if (error) {
 279                         error = EFAULT;
 280                         break;
 281                 }
 282
 283                 /*
 284                  * Now we need to get the page.  out_entry, out_prot, wired,
 285                  * and single_use aren't used.  One would think the vm code
 286                  * would be a *bit* nicer...  We use tmap because
 287                  * vm_map_lookup() can change the map argument.
 288                  */
 289                 tmap = map;
 290                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 291                     &object, &pindex, &out_prot, &wired);
 292                 if (error) {
 293                         error = EFAULT;
 294                         break;
 295                 }
 296                 VM_OBJECT_LOCK(object);
 297                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 298                     !writing &&
 299                     (backing_object = object->backing_object) != NULL) {
 300                         /*
 301                          * Allow fallback to backing objects if we are reading.
 302                          */
 303                         VM_OBJECT_LOCK(backing_object);
 304                         pindex += OFF_TO_IDX(object->backing_object_offset);
 305                         VM_OBJECT_UNLOCK(object);
 306                         object = backing_object;
 307                 }
 308                 VM_OBJECT_UNLOCK(object);
 309                 if (m == NULL) {
 310                         vm_map_lookup_done(tmap, out_entry);
 311                         error = EFAULT;
 312                         break;
 313                 }
 314
 315                 /*
 316                  * Hold the page in memory.
 317                  */
 318                 vm_page_lock_queues();
 319                 vm_page_hold(m);
 320                 vm_page_unlock_queues();
 321
 322                 /*
 323                  * We're done with tmap now.
 324                  */
 325                 vm_map_lookup_done(tmap, out_entry);
 326
 327                 /*
 328                  * Now do the i/o move.
 329                  */
 330                 error = uiomove_fromphys(&m, page_offset, len, uio);
 331
 332                 /*
 333                  * Release the page.
 334                  */
 335                 vm_page_lock_queues();
 336                 vm_page_unhold(m);
 337                 vm_page_unlock_queues();
 338
 339         } while (error == 0 && uio->uio_resid > 0);
 340
 341         vmspace_free(vm);
 342         return (error);
 343 }
 344
 345 /*
 346  * Process debugging system call.
 347  */
 348 #ifndef _SYS_SYSPROTO_H_
 349 struct ptrace_args {
 350         int     req;
 351         pid_t   pid;
 352         caddr_t addr;
 353         int     data;
 354 };
 355 #endif
 356
 357 #ifdef COMPAT_IA32
 358 /*
 359  * This CPP subterfuge is to try and reduce the number of ifdefs in
 360  * the body of the code.
 361  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 362  * becomes either:
 363  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 364  * or
 365  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 366  * .. except this is done at runtime.
 367  */
 368 #define COPYIN(u, k, s)         wrap32 ? \
 369         copyin(u, k ## 32, s ## 32) : \
 370         copyin(u, k, s)
 371 #define COPYOUT(k, u, s)        wrap32 ? \
 372         copyout(k ## 32, u, s ## 32) : \
 373         copyout(k, u, s)
 374 #else
 375 #define COPYIN(u, k, s)         copyin(u, k, s)
 376 #define COPYOUT(k, u, s)        copyout(k, u, s)
 377 #endif
 378 /*
 379  * MPSAFE
 380  */
 381 int
 382 ptrace(struct thread *td, struct ptrace_args *uap)
 383 {
 384         /*
 385          * XXX this obfuscation is to reduce stack usage, but the register
 386          * structs may be too large to put on the stack anyway.
 387          */
 388         union {
 389                 struct ptrace_io_desc piod;
 390                 struct ptrace_lwpinfo pl;
 391                 struct dbreg dbreg;
 392                 struct fpreg fpreg;
 393                 struct reg reg;
 394 #ifdef COMPAT_IA32
 395                 struct dbreg32 dbreg32;
 396                 struct fpreg32 fpreg32;
 397                 struct reg32 reg32;
 398                 struct ptrace_io_desc32 piod32;
 399 #endif
 400         } r;
 401         void *addr;
 402         int error = 0;
 403 #ifdef COMPAT_IA32
 404         int wrap32 = 0;
 405
 406         if (td->td_proc->p_sysent == &ia32_freebsd_sysvec)
 407                 wrap32 = 1;
 408 #endif
 409         AUDIT_ARG(pid, uap->pid);
 410         AUDIT_ARG(cmd, uap->req);
 411         AUDIT_ARG(addr, uap->addr);
 412         AUDIT_ARG(value, uap->data);
 413         addr = &r;
 414         switch (uap->req) {
 415         case PT_GETREGS:
 416         case PT_GETFPREGS:
 417         case PT_GETDBREGS:
 418         case PT_LWPINFO:
 419                 break;
 420         case PT_SETREGS:
 421                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 422                 break;
 423         case PT_SETFPREGS:
 424                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 425                 break;
 426         case PT_SETDBREGS:
 427                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 428                 break;
 429         case PT_IO:
 430                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 431                 break;
 432         default:
 433                 addr = uap->addr;
 434                 break;
 435         }
 436         if (error)
 437                 return (error);
 438
 439         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 440         if (error)
 441                 return (error);
 442
 443         switch (uap->req) {
 444         case PT_IO:
 445                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 446                 break;
 447         case PT_GETREGS:
 448                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 449                 break;
 450         case PT_GETFPREGS:
 451                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 452                 break;
 453         case PT_GETDBREGS:
 454                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 455                 break;
 456         case PT_LWPINFO:
 457                 error = copyout(&r.pl, uap->addr, uap->data);
 458                 break;
 459         }
 460
 461         return (error);
 462 }
 463 #undef COPYIN
 464 #undef COPYOUT
 465
 466 #ifdef COMPAT_IA32
 467 /*
 468  *   PROC_READ(regs, td2, addr);
 469  * becomes either:
 470  *   proc_read_regs(td2, addr);
 471  * or
 472  *   proc_read_regs32(td2, addr);
 473  * .. except this is done at runtime.  There is an additional
 474  * complication in that PROC_WRITE disallows 32 bit consumers
 475  * from writing to 64 bit address space targets.
 476  */
 477 #define PROC_READ(w, t, a)      wrap32 ? \
 478         proc_read_ ## w ## 32(t, a) : \
 479         proc_read_ ## w (t, a)
 480 #define PROC_WRITE(w, t, a)     wrap32 ? \
 481         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 482         proc_write_ ## w (t, a)
 483 #else
 484 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 485 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 486 #endif
 487
 488 int
 489 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 490 {
 491         struct iovec iov;
 492         struct uio uio;
 493         struct proc *curp, *p, *pp;
 494         struct thread *td2 = NULL;
 495         struct ptrace_io_desc *piod = NULL;
 496         struct ptrace_lwpinfo *pl;
 497         int error, write, tmp, num;
 498         int proctree_locked = 0;
 499         lwpid_t tid = 0, *buf;
 500 #ifdef COMPAT_IA32
 501         int wrap32 = 0, safe = 0;
 502         struct ptrace_io_desc32 *piod32 = NULL;
 503 #endif
 504
 505         curp = td->td_proc;
 506
 507         /* Lock proctree before locking the process. */
 508         switch (req) {
 509         case PT_TRACE_ME:
 510         case PT_ATTACH:
 511         case PT_STEP:
 512         case PT_CONTINUE:
 513         case PT_TO_SCE:
 514         case PT_TO_SCX:
 515         case PT_SYSCALL:
 516         case PT_DETACH:
 517                 sx_xlock(&proctree_lock);
 518                 proctree_locked = 1;
 519                 break;
 520         default:
 521                 break;
 522         }
 523
 524         write = 0;
 525         if (req == PT_TRACE_ME) {
 526                 p = td->td_proc;
 527                 PROC_LOCK(p);
 528         } else {
 529                 if (pid <= PID_MAX) {
 530                         if ((p = pfind(pid)) == NULL) {
 531                                 if (proctree_locked)
 532                                         sx_xunlock(&proctree_lock);
 533                                 return (ESRCH);
 534                         }
 535                 } else {
 536                         /* this is slow, should be optimized */
 537                         sx_slock(&allproc_lock);
 538                         FOREACH_PROC_IN_SYSTEM(p) {
 539                                 PROC_LOCK(p);
 540                                 mtx_lock_spin(&sched_lock);
 541                                 FOREACH_THREAD_IN_PROC(p, td2) {
 542                                         if (td2->td_tid == pid)
 543                                                 break;
 544                                 }
 545                                 mtx_unlock_spin(&sched_lock);
 546                                 if (td2 != NULL)
 547                                         break; /* proc lock held */
 548                                 PROC_UNLOCK(p);
 549                         }
 550                         sx_sunlock(&allproc_lock);
 551                         if (p == NULL) {
 552                                 if (proctree_locked)
 553                                         sx_xunlock(&proctree_lock);
 554                                 return (ESRCH);
 555                         }
 556                         tid = pid;
 557                         pid = p->p_pid;
 558                 }
 559         }
 560         AUDIT_ARG(process, p);
 561         if ((error = p_cansee(td, p)) != 0)
 562                 goto fail;
 563
 564         if ((error = p_candebug(td, p)) != 0)
 565                 goto fail;
 566
 567         /*
 568          * System processes can't be debugged.
 569          */
 570         if ((p->p_flag & P_SYSTEM) != 0) {
 571                 error = EINVAL;
 572                 goto fail;
 573         }
 574
 575         if (tid == 0) {
 576                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 577                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 578                         td2 = p->p_xthread;
 579                 } else {
 580                         td2 = FIRST_THREAD_IN_PROC(p);
 581                 }
 582                 tid = td2->td_tid;
 583         }
 584
 585 #ifdef COMPAT_IA32
 586         /*
 587          * Test if we're a 32 bit client and what the target is.
 588          * Set the wrap controls accordingly.
 589          */
 590         if (td->td_proc->p_sysent == &ia32_freebsd_sysvec) {
 591                 if (td2->td_proc->p_sysent == &ia32_freebsd_sysvec)
 592                         safe = 1;
 593                 wrap32 = 1;
 594         }
 595 #endif
 596         /*
 597          * Permissions check
 598          */
 599         switch (req) {
 600         case PT_TRACE_ME:
 601                 /* Always legal. */
 602                 break;
 603
 604         case PT_ATTACH:
 605                 /* Self */
 606                 if (p->p_pid == td->td_proc->p_pid) {
 607                         error = EINVAL;
 608                         goto fail;
 609                 }
 610
 611                 /* Already traced */
 612                 if (p->p_flag & P_TRACED) {
 613                         error = EBUSY;
 614                         goto fail;
 615                 }
 616
 617                 /* Can't trace an ancestor if you're being traced. */
 618                 if (curp->p_flag & P_TRACED) {
 619                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 620                                 if (pp == p) {
 621                                         error = EINVAL;
 622                                         goto fail;
 623                                 }
 624                         }
 625                 }
 626
 627
 628                 /* OK */
 629                 break;
 630
 631         case PT_CLEARSTEP:
 632                 /* Allow thread to clear single step for itself */
 633                 if (td->td_tid == tid)
 634                         break;
 635
 636                 /* FALLTHROUGH */
 637         default:
 638                 /* not being traced... */
 639                 if ((p->p_flag & P_TRACED) == 0) {
 640                         error = EPERM;
 641                         goto fail;
 642                 }
 643
 644                 /* not being traced by YOU */
 645                 if (p->p_pptr != td->td_proc) {
 646                         error = EBUSY;
 647                         goto fail;
 648                 }
 649
 650                 /* not currently stopped */
 651                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 652                     p->p_suspcount != p->p_numthreads  ||
 653                     (p->p_flag & P_WAITED) == 0) {
 654                         error = EBUSY;
 655                         goto fail;
 656                 }
 657
 658                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 659                         static int count = 0;
 660                         if (count++ == 0)
 661                                 printf("P_STOPPED_TRACE not set.\n");
 662                 }
 663
 664                 /* OK */
 665                 break;
 666         }
 667
 668 #ifdef FIX_SSTEP
 669         /*
 670          * Single step fixup ala procfs
 671          */
 672         FIX_SSTEP(td2);                 /* XXXKSE */
 673 #endif
 674
 675         /*
 676          * Actually do the requests
 677          */
 678
 679         td->td_retval[0] = 0;
 680
 681         switch (req) {
 682         case PT_TRACE_ME:
 683                 /* set my trace flag and "owner" so it can read/write me */
 684                 p->p_flag |= P_TRACED;
 685                 p->p_oppid = p->p_pptr->p_pid;
 686                 PROC_UNLOCK(p);
 687                 sx_xunlock(&proctree_lock);
 688                 return (0);
 689
 690         case PT_ATTACH:
 691                 /* security check done above */
 692                 p->p_flag |= P_TRACED;
 693                 p->p_oppid = p->p_pptr->p_pid;
 694                 if (p->p_pptr != td->td_proc) {
 695                         PROC_LOCK(p->p_pptr);
 696                         sigqueue_take(p->p_ksi);
 697                         PROC_UNLOCK(p->p_pptr);
 698                         proc_reparent(p, td->td_proc);
 699                 }
 700                 data = SIGSTOP;
 701                 goto sendsig;   /* in PT_CONTINUE below */
 702
 703         case PT_CLEARSTEP:
 704                 _PHOLD(p);
 705                 error = ptrace_clear_single_step(td2);
 706                 _PRELE(p);
 707                 if (error)
 708                         goto fail;
 709                 PROC_UNLOCK(p);
 710                 return (0);
 711
 712         case PT_SETSTEP:
 713                 _PHOLD(p);
 714                 error = ptrace_single_step(td2);
 715                 _PRELE(p);
 716                 if (error)
 717                         goto fail;
 718                 PROC_UNLOCK(p);
 719                 return (0);
 720
 721         case PT_SUSPEND:
 722                 _PHOLD(p);
 723                 mtx_lock_spin(&sched_lock);
 724                 td2->td_flags |= TDF_DBSUSPEND;
 725                 mtx_unlock_spin(&sched_lock);
 726                 _PRELE(p);
 727                 PROC_UNLOCK(p);
 728                 return (0);
 729
 730         case PT_RESUME:
 731                 _PHOLD(p);
 732                 mtx_lock_spin(&sched_lock);
 733                 td2->td_flags &= ~TDF_DBSUSPEND;
 734                 mtx_unlock_spin(&sched_lock);
 735                 _PRELE(p);
 736                 PROC_UNLOCK(p);
 737                 return (0);
 738
 739         case PT_STEP:
 740         case PT_CONTINUE:
 741         case PT_TO_SCE:
 742         case PT_TO_SCX:
 743         case PT_SYSCALL:
 744         case PT_DETACH:
 745                 /* Zero means do not send any signal */
 746                 if (data < 0 || data > _SIG_MAXSIG) {
 747                         error = EINVAL;
 748                         goto fail;
 749                 }
 750
 751                 _PHOLD(p);
 752
 753                 switch (req) {
 754                 case PT_STEP:
 755                         PROC_UNLOCK(p);
 756                         error = ptrace_single_step(td2);
 757                         if (error) {
 758                                 PRELE(p);
 759                                 goto fail_noproc;
 760                         }
 761                         PROC_LOCK(p);
 762                         break;
 763                 case PT_TO_SCE:
 764                         p->p_stops |= S_PT_SCE;
 765                         break;
 766                 case PT_TO_SCX:
 767                         p->p_stops |= S_PT_SCX;
 768                         break;
 769                 case PT_SYSCALL:
 770                         p->p_stops |= S_PT_SCE | S_PT_SCX;
 771                         break;
 772                 }
 773
 774                 if (addr != (void *)1) {
 775                         PROC_UNLOCK(p);
 776                         error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
 777                         if (error) {
 778                                 PRELE(p);
 779                                 goto fail_noproc;
 780                         }
 781                         PROC_LOCK(p);
 782                 }
 783                 _PRELE(p);
 784
 785                 if (req == PT_DETACH) {
 786                         /* reset process parent */
 787                         if (p->p_oppid != p->p_pptr->p_pid) {
 788                                 struct proc *pp;
 789
 790                                 PROC_LOCK(p->p_pptr);
 791                                 sigqueue_take(p->p_ksi);
 792                                 PROC_UNLOCK(p->p_pptr);
 793
 794                                 PROC_UNLOCK(p);
 795                                 pp = pfind(p->p_oppid);
 796                                 if (pp == NULL)
 797                                         pp = initproc;
 798                                 else
 799                                         PROC_UNLOCK(pp);
 800                                 PROC_LOCK(p);
 801                                 proc_reparent(p, pp);
 802                                 if (pp == initproc)
 803                                         p->p_sigparent = SIGCHLD;
 804                         }
 805                         p->p_flag &= ~(P_TRACED | P_WAITED);
 806                         p->p_oppid = 0;
 807
 808                         /* should we send SIGCHLD? */
 809                         /* childproc_continued(p); */
 810                 }
 811
 812         sendsig:
 813                 if (proctree_locked)
 814                         sx_xunlock(&proctree_lock);
 815                 /* deliver or queue signal */
 816                 mtx_lock_spin(&sched_lock);
 817                 td2->td_flags &= ~TDF_XSIG;
 818                 mtx_unlock_spin(&sched_lock);
 819                 td2->td_xsig = data;
 820                 p->p_xstat = data;
 821                 p->p_xthread = NULL;
 822                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 823                         mtx_lock_spin(&sched_lock);
 824                         if (req == PT_DETACH) {
 825                                 struct thread *td3;
 826                                 FOREACH_THREAD_IN_PROC(p, td3)
 827                                         td3->td_flags &= ~TDF_DBSUSPEND;
 828                         }
 829                         /*
 830                          * unsuspend all threads, to not let a thread run,
 831                          * you should use PT_SUSPEND to suspend it before
 832                          * continuing process.
 833                          */
 834                         mtx_unlock_spin(&sched_lock);
 835                         thread_continued(p);
 836                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
 837                         mtx_lock_spin(&sched_lock);
 838                         thread_unsuspend(p);
 839                         mtx_unlock_spin(&sched_lock);
 840                 }
 841
 842                 if (data)
 843                         psignal(p, data);
 844
 845                 PROC_UNLOCK(p);
 846                 return (0);
 847
 848         case PT_WRITE_I:
 849         case PT_WRITE_D:
 850                 write = 1;
 851                 /* FALLTHROUGH */
 852         case PT_READ_I:
 853         case PT_READ_D:
 854                 PROC_UNLOCK(p);
 855                 tmp = 0;
 856                 /* write = 0 set above */
 857                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
 858                 iov.iov_len = sizeof(int);
 859                 uio.uio_iov = &iov;
 860                 uio.uio_iovcnt = 1;
 861                 uio.uio_offset = (off_t)(uintptr_t)addr;
 862                 uio.uio_resid = sizeof(int);
 863                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
 864                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
 865                 uio.uio_td = td;
 866                 error = proc_rwmem(p, &uio);
 867                 if (uio.uio_resid != 0) {
 868                         /*
 869                          * XXX proc_rwmem() doesn't currently return ENOSPC,
 870                          * so I think write() can bogusly return 0.
 871                          * XXX what happens for short writes?  We don't want
 872                          * to write partial data.
 873                          * XXX proc_rwmem() returns EPERM for other invalid
 874                          * addresses.  Convert this to EINVAL.  Does this
 875                          * clobber returns of EPERM for other reasons?
 876                          */
 877                         if (error == 0 || error == ENOSPC || error == EPERM)
 878                                 error = EINVAL; /* EOF */
 879                 }
 880                 if (!write)
 881                         td->td_retval[0] = tmp;
 882                 return (error);
 883
 884         case PT_IO:
 885                 PROC_UNLOCK(p);
 886 #ifdef COMPAT_IA32
 887                 if (wrap32) {
 888                         piod32 = addr;
 889                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
 890                         iov.iov_len = piod32->piod_len;
 891                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
 892                         uio.uio_resid = piod32->piod_len;
 893                 } else
 894 #endif
 895                 {
 896                         piod = addr;
 897                         iov.iov_base = piod->piod_addr;
 898                         iov.iov_len = piod->piod_len;
 899                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
 900                         uio.uio_resid = piod->piod_len;
 901                 }
 902                 uio.uio_iov = &iov;
 903                 uio.uio_iovcnt = 1;
 904                 uio.uio_segflg = UIO_USERSPACE;
 905                 uio.uio_td = td;
 906 #ifdef COMPAT_IA32
 907                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
 908 #else
 909                 tmp = piod->piod_op;
 910 #endif
 911                 switch (tmp) {
 912                 case PIOD_READ_D:
 913                 case PIOD_READ_I:
 914                         uio.uio_rw = UIO_READ;
 915                         break;
 916                 case PIOD_WRITE_D:
 917                 case PIOD_WRITE_I:
 918                         uio.uio_rw = UIO_WRITE;
 919                         break;
 920                 default:
 921                         return (EINVAL);
 922                 }
 923                 error = proc_rwmem(p, &uio);
 924 #ifdef COMPAT_IA32
 925                 if (wrap32)
 926                         piod32->piod_len -= uio.uio_resid;
 927                 else
 928 #endif
 929                         piod->piod_len -= uio.uio_resid;
 930                 return (error);
 931
 932         case PT_KILL:
 933                 data = SIGKILL;
 934                 goto sendsig;   /* in PT_CONTINUE above */
 935
 936         case PT_SETREGS:
 937                 _PHOLD(p);
 938                 error = PROC_WRITE(regs, td2, addr);
 939                 _PRELE(p);
 940                 PROC_UNLOCK(p);
 941                 return (error);
 942
 943         case PT_GETREGS:
 944                 _PHOLD(p);
 945                 error = PROC_READ(regs, td2, addr);
 946                 _PRELE(p);
 947                 PROC_UNLOCK(p);
 948                 return (error);
 949
 950         case PT_SETFPREGS:
 951                 _PHOLD(p);
 952                 error = PROC_WRITE(fpregs, td2, addr);
 953                 _PRELE(p);
 954                 PROC_UNLOCK(p);
 955                 return (error);
 956
 957         case PT_GETFPREGS:
 958                 _PHOLD(p);
 959                 error = PROC_READ(fpregs, td2, addr);
 960                 _PRELE(p);
 961                 PROC_UNLOCK(p);
 962                 return (error);
 963
 964         case PT_SETDBREGS:
 965                 _PHOLD(p);
 966                 error = PROC_WRITE(dbregs, td2, addr);
 967                 _PRELE(p);
 968                 PROC_UNLOCK(p);
 969                 return (error);
 970
 971         case PT_GETDBREGS:
 972                 _PHOLD(p);
 973                 error = PROC_READ(dbregs, td2, addr);
 974                 _PRELE(p);
 975                 PROC_UNLOCK(p);
 976                 return (error);
 977
 978         case PT_LWPINFO:
 979                 if (data == 0 || data > sizeof(*pl))
 980                         return (EINVAL);
 981                 pl = addr;
 982                 _PHOLD(p);
 983                 pl->pl_lwpid = td2->td_tid;
 984                 if (td2->td_flags & TDF_XSIG)
 985                         pl->pl_event = PL_EVENT_SIGNAL;
 986                 else
 987                         pl->pl_event = 0;
 988                 if (td2->td_pflags & TDP_SA) {
 989                         pl->pl_flags = PL_FLAG_SA;
 990                         if (td2->td_upcall && !TD_CAN_UNBIND(td2))
 991                                 pl->pl_flags |= PL_FLAG_BOUND;
 992                 } else {
 993                         pl->pl_flags = 0;
 994                 }
 995                 pl->pl_sigmask = td2->td_sigmask;
 996                 pl->pl_siglist = td2->td_siglist;
 997                 _PRELE(p);
 998                 PROC_UNLOCK(p);
 999                 return (0);
1000
1001         case PT_GETNUMLWPS:
1002                 td->td_retval[0] = p->p_numthreads;
1003                 PROC_UNLOCK(p);
1004                 return (0);
1005
1006         case PT_GETLWPLIST:
1007                 if (data <= 0) {
1008                         PROC_UNLOCK(p);
1009                         return (EINVAL);
1010                 }
1011                 num = imin(p->p_numthreads, data);
1012                 PROC_UNLOCK(p);
1013                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1014                 tmp = 0;
1015                 PROC_LOCK(p);
1016                 mtx_lock_spin(&sched_lock);
1017                 FOREACH_THREAD_IN_PROC(p, td2) {
1018                         if (tmp >= num)
1019                                 break;
1020                         buf[tmp++] = td2->td_tid;
1021                 }
1022                 mtx_unlock_spin(&sched_lock);
1023                 PROC_UNLOCK(p);
1024                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1025                 free(buf, M_TEMP);
1026                 if (!error)
1027                         td->td_retval[0] = num;
1028                 return (error);
1029
1030         default:
1031 #ifdef __HAVE_PTRACE_MACHDEP
1032                 if (req >= PT_FIRSTMACH) {
1033                         _PHOLD(p);
1034                         PROC_UNLOCK(p);
1035                         error = cpu_ptrace(td2, req, addr, data);
1036                         PRELE(p);
1037                         return (error);
1038                 }
1039 #endif
1040                 break;
1041         }
1042
1043         /* Unknown request. */
1044         error = EINVAL;
1045
1046 fail:
1047         PROC_UNLOCK(p);
1048 fail_noproc:
1049         if (proctree_locked)
1050                 sx_xunlock(&proctree_lock);
1051         return (error);
1052 }
1053 #undef PROC_READ
1054 #undef PROC_WRITE
1055
1056 /*
1057  * Stop a process because of a debugging event;
1058  * stay stopped until p->p_step is cleared
1059  * (cleared by PIOCCONT in procfs).
1060  */
1061 void
1062 stopevent(struct proc *p, unsigned int event, unsigned int val)
1063 {
1064
1065         PROC_LOCK_ASSERT(p, MA_OWNED);
1066         p->p_step = 1;
1067         do {
1068                 p->p_xstat = val;
1069                 p->p_xthread = NULL;
1070                 p->p_stype = event;     /* Which event caused the stop? */
1071                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1072                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1073         } while (p->p_step);
1074 }