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