sys/kern/sys_process.c

   1 /*-
   2  * Copyright (c) 1994, Sean Eric Fagan
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by Sean Eric Fagan.
  16  * 4. The name of the author may not be used to endorse or promote products
  17  *    derived from this software without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __FBSDID("$FreeBSD$");
  34
  35 #include "opt_compat.h"
  36
  37 #include <sys/param.h>
  38 #include <sys/systm.h>
  39 #include <sys/lock.h>
  40 #include <sys/mutex.h>
  41 #include <sys/syscallsubr.h>
  42 #include <sys/sysent.h>
  43 #include <sys/sysproto.h>
  44 #include <sys/proc.h>
  45 #include <sys/vnode.h>
  46 #include <sys/ptrace.h>
  47 #include <sys/sx.h>
  48 #include <sys/malloc.h>
  49 #include <sys/signalvar.h>
  50
  51 #include <machine/reg.h>
  52
  53 #include <security/audit/audit.h>
  54
  55 #include <vm/vm.h>
  56 #include <vm/pmap.h>
  57 #include <vm/vm_extern.h>
  58 #include <vm/vm_map.h>
  59 #include <vm/vm_kern.h>
  60 #include <vm/vm_object.h>
  61 #include <vm/vm_page.h>
  62 #include <vm/vm_param.h>
  63
  64 #ifdef COMPAT_FREEBSD32
  65 #include <sys/procfs.h>
  66
  67 struct ptrace_io_desc32 {
  68         int             piod_op;
  69         u_int32_t       piod_offs;
  70         u_int32_t       piod_addr;
  71         u_int32_t       piod_len;
  72 };
  73
  74 struct ptrace_vm_entry32 {
  75         int             pve_entry;
  76         int             pve_timestamp;
  77         uint32_t        pve_start;
  78         uint32_t        pve_end;
  79         uint32_t        pve_offset;
  80         u_int           pve_prot;
  81         u_int           pve_pathlen;
  82         int32_t         pve_fileid;
  83         u_int           pve_fsid;
  84         uint32_t        pve_path;
  85 };
  86
  87 #endif
  88
  89 /*
  90  * Functions implemented using PROC_ACTION():
  91  *
  92  * proc_read_regs(proc, regs)
  93  *      Get the current user-visible register set from the process
  94  *      and copy it into the regs structure (<machine/reg.h>).
  95  *      The process is stopped at the time read_regs is called.
  96  *
  97  * proc_write_regs(proc, regs)
  98  *      Update the current register set from the passed in regs
  99  *      structure.  Take care to avoid clobbering special CPU
 100  *      registers or privileged bits in the PSL.
 101  *      Depending on the architecture this may have fix-up work to do,
 102  *      especially if the IAR or PCW are modified.
 103  *      The process is stopped at the time write_regs is called.
 104  *
 105  * proc_read_fpregs, proc_write_fpregs
 106  *      deal with the floating point register set, otherwise as above.
 107  *
 108  * proc_read_dbregs, proc_write_dbregs
 109  *      deal with the processor debug register set, otherwise as above.
 110  *
 111  * proc_sstep(proc)
 112  *      Arrange for the process to trap after executing a single instruction.
 113  */
 114
 115 #define PROC_ACTION(action) do {                                        \
 116         int error;                                                      \
 117                                                                         \
 118         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 119         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
 120                 error = EIO;                                            \
 121         else                                                            \
 122                 error = (action);                                       \
 123         return (error);                                                 \
 124 } while(0)
 125
 126 int
 127 proc_read_regs(struct thread *td, struct reg *regs)
 128 {
 129
 130         PROC_ACTION(fill_regs(td, regs));
 131 }
 132
 133 int
 134 proc_write_regs(struct thread *td, struct reg *regs)
 135 {
 136
 137         PROC_ACTION(set_regs(td, regs));
 138 }
 139
 140 int
 141 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 142 {
 143
 144         PROC_ACTION(fill_dbregs(td, dbregs));
 145 }
 146
 147 int
 148 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 149 {
 150
 151         PROC_ACTION(set_dbregs(td, dbregs));
 152 }
 153
 154 /*
 155  * Ptrace doesn't support fpregs at all, and there are no security holes
 156  * or translations for fpregs, so we can just copy them.
 157  */
 158 int
 159 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 160 {
 161
 162         PROC_ACTION(fill_fpregs(td, fpregs));
 163 }
 164
 165 int
 166 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 167 {
 168
 169         PROC_ACTION(set_fpregs(td, fpregs));
 170 }
 171
 172 #ifdef COMPAT_FREEBSD32
 173 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 174 int
 175 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 176 {
 177
 178         PROC_ACTION(fill_regs32(td, regs32));
 179 }
 180
 181 int
 182 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 183 {
 184
 185         PROC_ACTION(set_regs32(td, regs32));
 186 }
 187
 188 int
 189 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 190 {
 191
 192         PROC_ACTION(fill_dbregs32(td, dbregs32));
 193 }
 194
 195 int
 196 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 197 {
 198
 199         PROC_ACTION(set_dbregs32(td, dbregs32));
 200 }
 201
 202 int
 203 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 204 {
 205
 206         PROC_ACTION(fill_fpregs32(td, fpregs32));
 207 }
 208
 209 int
 210 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 211 {
 212
 213         PROC_ACTION(set_fpregs32(td, fpregs32));
 214 }
 215 #endif
 216
 217 int
 218 proc_sstep(struct thread *td)
 219 {
 220
 221         PROC_ACTION(ptrace_single_step(td));
 222 }
 223
 224 int
 225 proc_rwmem(struct proc *p, struct uio *uio)
 226 {
 227         vm_map_t map;
 228         vm_object_t backing_object, object = NULL;
 229         vm_offset_t pageno = 0;         /* page number */
 230         vm_prot_t reqprot;
 231         int error, fault_flags, writing;
 232
 233         /*
 234          * Assert that someone has locked this vmspace.  (Should be
 235          * curthread but we can't assert that.)  This keeps the process
 236          * from exiting out from under us until this operation completes.
 237          */
 238         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 239             p, p->p_pid));
 240
 241         /*
 242          * The map we want...
 243          */
 244         map = &p->p_vmspace->vm_map;
 245
 246         writing = uio->uio_rw == UIO_WRITE;
 247         reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
 248             VM_PROT_READ;
 249         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
 250
 251         /*
 252          * Only map in one page at a time.  We don't have to, but it
 253          * makes things easier.  This way is trivial - right?
 254          */
 255         do {
 256                 vm_map_t tmap;
 257                 vm_offset_t uva;
 258                 int page_offset;                /* offset into page */
 259                 vm_map_entry_t out_entry;
 260                 vm_prot_t out_prot;
 261                 boolean_t wired;
 262                 vm_pindex_t pindex;
 263                 u_int len;
 264                 vm_page_t m;
 265
 266                 object = NULL;
 267
 268                 uva = (vm_offset_t)uio->uio_offset;
 269
 270                 /*
 271                  * Get the page number of this segment.
 272                  */
 273                 pageno = trunc_page(uva);
 274                 page_offset = uva - pageno;
 275
 276                 /*
 277                  * How many bytes to copy
 278                  */
 279                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 280
 281                 /*
 282                  * Fault the page on behalf of the process
 283                  */
 284                 error = vm_fault(map, pageno, reqprot, fault_flags);
 285                 if (error) {
 286                         if (error == KERN_RESOURCE_SHORTAGE)
 287                                 error = ENOMEM;
 288                         else
 289                                 error = EFAULT;
 290                         break;
 291                 }
 292
 293                 /*
 294                  * Now we need to get the page.  out_entry, out_prot, wired,
 295                  * and single_use aren't used.  One would think the vm code
 296                  * would be a *bit* nicer...  We use tmap because
 297                  * vm_map_lookup() can change the map argument.
 298                  */
 299                 tmap = map;
 300                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 301                     &object, &pindex, &out_prot, &wired);
 302                 if (error) {
 303                         error = EFAULT;
 304                         break;
 305                 }
 306                 VM_OBJECT_LOCK(object);
 307                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 308                     !writing &&
 309                     (backing_object = object->backing_object) != NULL) {
 310                         /*
 311                          * Allow fallback to backing objects if we are reading.
 312                          */
 313                         VM_OBJECT_LOCK(backing_object);
 314                         pindex += OFF_TO_IDX(object->backing_object_offset);
 315                         VM_OBJECT_UNLOCK(object);
 316                         object = backing_object;
 317                 }
 318                 VM_OBJECT_UNLOCK(object);
 319                 if (m == NULL) {
 320                         vm_map_lookup_done(tmap, out_entry);
 321                         error = EFAULT;
 322                         break;
 323                 }
 324
 325                 /*
 326                  * Hold the page in memory.
 327                  */
 328                 vm_page_lock_queues();
 329                 vm_page_hold(m);
 330                 vm_page_unlock_queues();
 331
 332                 /*
 333                  * We're done with tmap now.
 334                  */
 335                 vm_map_lookup_done(tmap, out_entry);
 336
 337                 /*
 338                  * Now do the i/o move.
 339                  */
 340                 error = uiomove_fromphys(&m, page_offset, len, uio);
 341
 342                 /* Make the I-cache coherent for breakpoints. */
 343                 if (!error && writing && (out_prot & VM_PROT_EXECUTE))
 344                         vm_sync_icache(map, uva, len);
 345
 346                 /*
 347                  * Release the page.
 348                  */
 349                 vm_page_lock_queues();
 350                 vm_page_unhold(m);
 351                 vm_page_unlock_queues();
 352
 353         } while (error == 0 && uio->uio_resid > 0);
 354
 355         return (error);
 356 }
 357
 358 static int
 359 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
 360 {
 361         struct vattr vattr;
 362         vm_map_t map;
 363         vm_map_entry_t entry;
 364         vm_object_t obj, tobj, lobj;
 365         struct vmspace *vm;
 366         struct vnode *vp;
 367         char *freepath, *fullpath;
 368         u_int pathlen;
 369         int error, index, vfslocked;
 370
 371         error = 0;
 372         obj = NULL;
 373
 374         vm = vmspace_acquire_ref(p);
 375         map = &vm->vm_map;
 376         vm_map_lock_read(map);
 377
 378         do {
 379                 entry = map->header.next;
 380                 index = 0;
 381                 while (index < pve->pve_entry && entry != &map->header) {
 382                         entry = entry->next;
 383                         index++;
 384                 }
 385                 if (index != pve->pve_entry) {
 386                         error = EINVAL;
 387                         break;
 388                 }
 389                 while (entry != &map->header &&
 390                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
 391                         entry = entry->next;
 392                         index++;
 393                 }
 394                 if (entry == &map->header) {
 395                         error = ENOENT;
 396                         break;
 397                 }
 398
 399                 /* We got an entry. */
 400                 pve->pve_entry = index + 1;
 401                 pve->pve_timestamp = map->timestamp;
 402                 pve->pve_start = entry->start;
 403                 pve->pve_end = entry->end - 1;
 404                 pve->pve_offset = entry->offset;
 405                 pve->pve_prot = entry->protection;
 406
 407                 /* Backing object's path needed? */
 408                 if (pve->pve_pathlen == 0)
 409                         break;
 410
 411                 pathlen = pve->pve_pathlen;
 412                 pve->pve_pathlen = 0;
 413
 414                 obj = entry->object.vm_object;
 415                 if (obj != NULL)
 416                         VM_OBJECT_LOCK(obj);
 417         } while (0);
 418
 419         vm_map_unlock_read(map);
 420         vmspace_free(vm);
 421
 422         pve->pve_fsid = VNOVAL;
 423         pve->pve_fileid = VNOVAL;
 424
 425         if (error == 0 && obj != NULL) {
 426                 lobj = obj;
 427                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
 428                         if (tobj != obj)
 429                                 VM_OBJECT_LOCK(tobj);
 430                         if (lobj != obj)
 431                                 VM_OBJECT_UNLOCK(lobj);
 432                         lobj = tobj;
 433                         pve->pve_offset += tobj->backing_object_offset;
 434                 }
 435                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 436                 if (vp != NULL)
 437                         vref(vp);
 438                 if (lobj != obj)
 439                         VM_OBJECT_UNLOCK(lobj);
 440                 VM_OBJECT_UNLOCK(obj);
 441
 442                 if (vp != NULL) {
 443                         freepath = NULL;
 444                         fullpath = NULL;
 445                         vn_fullpath(td, vp, &fullpath, &freepath);
 446                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 447                         vn_lock(vp, LK_SHARED | LK_RETRY);
 448                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 449                                 pve->pve_fileid = vattr.va_fileid;
 450                                 pve->pve_fsid = vattr.va_fsid;
 451                         }
 452                         vput(vp);
 453                         VFS_UNLOCK_GIANT(vfslocked);
 454
 455                         if (fullpath != NULL) {
 456                                 pve->pve_pathlen = strlen(fullpath) + 1;
 457                                 if (pve->pve_pathlen <= pathlen) {
 458                                         error = copyout(fullpath, pve->pve_path,
 459                                             pve->pve_pathlen);
 460                                 } else
 461                                         error = ENAMETOOLONG;
 462                         }
 463                         if (freepath != NULL)
 464                                 free(freepath, M_TEMP);
 465                 }
 466         }
 467
 468         return (error);
 469 }
 470
 471 #ifdef COMPAT_FREEBSD32
 472 static int
 473 ptrace_vm_entry32(struct thread *td, struct proc *p,
 474     struct ptrace_vm_entry32 *pve32)
 475 {
 476         struct ptrace_vm_entry pve;
 477         int error;
 478
 479         pve.pve_entry = pve32->pve_entry;
 480         pve.pve_pathlen = pve32->pve_pathlen;
 481         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 482
 483         error = ptrace_vm_entry(td, p, &pve);
 484         if (error == 0) {
 485                 pve32->pve_entry = pve.pve_entry;
 486                 pve32->pve_timestamp = pve.pve_timestamp;
 487                 pve32->pve_start = pve.pve_start;
 488                 pve32->pve_end = pve.pve_end;
 489                 pve32->pve_offset = pve.pve_offset;
 490                 pve32->pve_prot = pve.pve_prot;
 491                 pve32->pve_fileid = pve.pve_fileid;
 492                 pve32->pve_fsid = pve.pve_fsid;
 493         }
 494
 495         pve32->pve_pathlen = pve.pve_pathlen;
 496         return (error);
 497 }
 498 #endif /* COMPAT_FREEBSD32 */
 499
 500 /*
 501  * Process debugging system call.
 502  */
 503 #ifndef _SYS_SYSPROTO_H_
 504 struct ptrace_args {
 505         int     req;
 506         pid_t   pid;
 507         caddr_t addr;
 508         int     data;
 509 };
 510 #endif
 511
 512 #ifdef COMPAT_FREEBSD32
 513 /*
 514  * This CPP subterfuge is to try and reduce the number of ifdefs in
 515  * the body of the code.
 516  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 517  * becomes either:
 518  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 519  * or
 520  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 521  * .. except this is done at runtime.
 522  */
 523 #define COPYIN(u, k, s)         wrap32 ? \
 524         copyin(u, k ## 32, s ## 32) : \
 525         copyin(u, k, s)
 526 #define COPYOUT(k, u, s)        wrap32 ? \
 527         copyout(k ## 32, u, s ## 32) : \
 528         copyout(k, u, s)
 529 #else
 530 #define COPYIN(u, k, s)         copyin(u, k, s)
 531 #define COPYOUT(k, u, s)        copyout(k, u, s)
 532 #endif
 533 int
 534 ptrace(struct thread *td, struct ptrace_args *uap)
 535 {
 536         /*
 537          * XXX this obfuscation is to reduce stack usage, but the register
 538          * structs may be too large to put on the stack anyway.
 539          */
 540         union {
 541                 struct ptrace_io_desc piod;
 542                 struct ptrace_lwpinfo pl;
 543                 struct ptrace_vm_entry pve;
 544                 struct dbreg dbreg;
 545                 struct fpreg fpreg;
 546                 struct reg reg;
 547 #ifdef COMPAT_FREEBSD32
 548                 struct dbreg32 dbreg32;
 549                 struct fpreg32 fpreg32;
 550                 struct reg32 reg32;
 551                 struct ptrace_io_desc32 piod32;
 552                 struct ptrace_vm_entry32 pve32;
 553 #endif
 554         } r;
 555         void *addr;
 556         int error = 0;
 557 #ifdef COMPAT_FREEBSD32
 558         int wrap32 = 0;
 559
 560         if (SV_CURPROC_FLAG(SV_ILP32))
 561                 wrap32 = 1;
 562 #endif
 563         AUDIT_ARG_PID(uap->pid);
 564         AUDIT_ARG_CMD(uap->req);
 565         AUDIT_ARG_VALUE(uap->data);
 566         addr = &r;
 567         switch (uap->req) {
 568         case PT_GETREGS:
 569         case PT_GETFPREGS:
 570         case PT_GETDBREGS:
 571         case PT_LWPINFO:
 572                 break;
 573         case PT_SETREGS:
 574                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 575                 break;
 576         case PT_SETFPREGS:
 577                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 578                 break;
 579         case PT_SETDBREGS:
 580                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 581                 break;
 582         case PT_IO:
 583                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 584                 break;
 585         case PT_VM_ENTRY:
 586                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 587                 break;
 588         default:
 589                 addr = uap->addr;
 590                 break;
 591         }
 592         if (error)
 593                 return (error);
 594
 595         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 596         if (error)
 597                 return (error);
 598
 599         switch (uap->req) {
 600         case PT_VM_ENTRY:
 601                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 602                 break;
 603         case PT_IO:
 604                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 605                 break;
 606         case PT_GETREGS:
 607                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 608                 break;
 609         case PT_GETFPREGS:
 610                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 611                 break;
 612         case PT_GETDBREGS:
 613                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 614                 break;
 615         case PT_LWPINFO:
 616                 error = copyout(&r.pl, uap->addr, uap->data);
 617                 break;
 618         }
 619
 620         return (error);
 621 }
 622 #undef COPYIN
 623 #undef COPYOUT
 624
 625 #ifdef COMPAT_FREEBSD32
 626 /*
 627  *   PROC_READ(regs, td2, addr);
 628  * becomes either:
 629  *   proc_read_regs(td2, addr);
 630  * or
 631  *   proc_read_regs32(td2, addr);
 632  * .. except this is done at runtime.  There is an additional
 633  * complication in that PROC_WRITE disallows 32 bit consumers
 634  * from writing to 64 bit address space targets.
 635  */
 636 #define PROC_READ(w, t, a)      wrap32 ? \
 637         proc_read_ ## w ## 32(t, a) : \
 638         proc_read_ ## w (t, a)
 639 #define PROC_WRITE(w, t, a)     wrap32 ? \
 640         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 641         proc_write_ ## w (t, a)
 642 #else
 643 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 644 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 645 #endif
 646
 647 int
 648 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 649 {
 650         struct iovec iov;
 651         struct uio uio;
 652         struct proc *curp, *p, *pp;
 653         struct thread *td2 = NULL;
 654         struct ptrace_io_desc *piod = NULL;
 655         struct ptrace_lwpinfo *pl;
 656         int error, write, tmp, num;
 657         int proctree_locked = 0;
 658         lwpid_t tid = 0, *buf;
 659 #ifdef COMPAT_FREEBSD32
 660         int wrap32 = 0, safe = 0;
 661         struct ptrace_io_desc32 *piod32 = NULL;
 662 #endif
 663
 664         curp = td->td_proc;
 665
 666         /* Lock proctree before locking the process. */
 667         switch (req) {
 668         case PT_TRACE_ME:
 669         case PT_ATTACH:
 670         case PT_STEP:
 671         case PT_CONTINUE:
 672         case PT_TO_SCE:
 673         case PT_TO_SCX:
 674         case PT_SYSCALL:
 675         case PT_DETACH:
 676                 sx_xlock(&proctree_lock);
 677                 proctree_locked = 1;
 678                 break;
 679         default:
 680                 break;
 681         }
 682
 683         write = 0;
 684         if (req == PT_TRACE_ME) {
 685                 p = td->td_proc;
 686                 PROC_LOCK(p);
 687         } else {
 688                 if (pid <= PID_MAX) {
 689                         if ((p = pfind(pid)) == NULL) {
 690                                 if (proctree_locked)
 691                                         sx_xunlock(&proctree_lock);
 692                                 return (ESRCH);
 693                         }
 694                 } else {
 695                         /* this is slow, should be optimized */
 696                         sx_slock(&allproc_lock);
 697                         FOREACH_PROC_IN_SYSTEM(p) {
 698                                 PROC_LOCK(p);
 699                                 FOREACH_THREAD_IN_PROC(p, td2) {
 700                                         if (td2->td_tid == pid)
 701                                                 break;
 702                                 }
 703                                 if (td2 != NULL)
 704                                         break; /* proc lock held */
 705                                 PROC_UNLOCK(p);
 706                         }
 707                         sx_sunlock(&allproc_lock);
 708                         if (p == NULL) {
 709                                 if (proctree_locked)
 710                                         sx_xunlock(&proctree_lock);
 711                                 return (ESRCH);
 712                         }
 713                         tid = pid;
 714                         pid = p->p_pid;
 715                 }
 716         }
 717         AUDIT_ARG_PROCESS(p);
 718
 719         if ((p->p_flag & P_WEXIT) != 0) {
 720                 error = ESRCH;
 721                 goto fail;
 722         }
 723         if ((error = p_cansee(td, p)) != 0)
 724                 goto fail;
 725
 726         if ((error = p_candebug(td, p)) != 0)
 727                 goto fail;
 728
 729         /*
 730          * System processes can't be debugged.
 731          */
 732         if ((p->p_flag & P_SYSTEM) != 0) {
 733                 error = EINVAL;
 734                 goto fail;
 735         }
 736
 737         if (tid == 0) {
 738                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 739                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 740                         td2 = p->p_xthread;
 741                 } else {
 742                         td2 = FIRST_THREAD_IN_PROC(p);
 743                 }
 744                 tid = td2->td_tid;
 745         }
 746
 747 #ifdef COMPAT_FREEBSD32
 748         /*
 749          * Test if we're a 32 bit client and what the target is.
 750          * Set the wrap controls accordingly.
 751          */
 752         if (SV_CURPROC_FLAG(SV_ILP32)) {
 753                 if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
 754                         safe = 1;
 755                 wrap32 = 1;
 756         }
 757 #endif
 758         /*
 759          * Permissions check
 760          */
 761         switch (req) {
 762         case PT_TRACE_ME:
 763                 /* Always legal. */
 764                 break;
 765
 766         case PT_ATTACH:
 767                 /* Self */
 768                 if (p->p_pid == td->td_proc->p_pid) {
 769                         error = EINVAL;
 770                         goto fail;
 771                 }
 772
 773                 /* Already traced */
 774                 if (p->p_flag & P_TRACED) {
 775                         error = EBUSY;
 776                         goto fail;
 777                 }
 778
 779                 /* Can't trace an ancestor if you're being traced. */
 780                 if (curp->p_flag & P_TRACED) {
 781                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 782                                 if (pp == p) {
 783                                         error = EINVAL;
 784                                         goto fail;
 785                                 }
 786                         }
 787                 }
 788
 789
 790                 /* OK */
 791                 break;
 792
 793         case PT_CLEARSTEP:
 794                 /* Allow thread to clear single step for itself */
 795                 if (td->td_tid == tid)
 796                         break;
 797
 798                 /* FALLTHROUGH */
 799         default:
 800                 /* not being traced... */
 801                 if ((p->p_flag & P_TRACED) == 0) {
 802                         error = EPERM;
 803                         goto fail;
 804                 }
 805
 806                 /* not being traced by YOU */
 807                 if (p->p_pptr != td->td_proc) {
 808                         error = EBUSY;
 809                         goto fail;
 810                 }
 811
 812                 /* not currently stopped */
 813                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 814                     p->p_suspcount != p->p_numthreads  ||
 815                     (p->p_flag & P_WAITED) == 0) {
 816                         error = EBUSY;
 817                         goto fail;
 818                 }
 819
 820                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 821                         static int count = 0;
 822                         if (count++ == 0)
 823                                 printf("P_STOPPED_TRACE not set.\n");
 824                 }
 825
 826                 /* OK */
 827                 break;
 828         }
 829
 830         /* Keep this process around until we finish this request. */
 831         _PHOLD(p);
 832
 833 #ifdef FIX_SSTEP
 834         /*
 835          * Single step fixup ala procfs
 836          */
 837         FIX_SSTEP(td2);
 838 #endif
 839
 840         /*
 841          * Actually do the requests
 842          */
 843
 844         td->td_retval[0] = 0;
 845
 846         switch (req) {
 847         case PT_TRACE_ME:
 848                 /* set my trace flag and "owner" so it can read/write me */
 849                 p->p_flag |= P_TRACED;
 850                 p->p_oppid = p->p_pptr->p_pid;
 851                 break;
 852
 853         case PT_ATTACH:
 854                 /* security check done above */
 855                 p->p_flag |= P_TRACED;
 856                 p->p_oppid = p->p_pptr->p_pid;
 857                 if (p->p_pptr != td->td_proc)
 858                         proc_reparent(p, td->td_proc);
 859                 data = SIGSTOP;
 860                 goto sendsig;   /* in PT_CONTINUE below */
 861
 862         case PT_CLEARSTEP:
 863                 error = ptrace_clear_single_step(td2);
 864                 break;
 865
 866         case PT_SETSTEP:
 867                 error = ptrace_single_step(td2);
 868                 break;
 869
 870         case PT_SUSPEND:
 871                 td2->td_dbgflags |= TDB_SUSPEND;
 872                 thread_lock(td2);
 873                 td2->td_flags |= TDF_NEEDSUSPCHK;
 874                 thread_unlock(td2);
 875                 break;
 876
 877         case PT_RESUME:
 878                 td2->td_dbgflags &= ~TDB_SUSPEND;
 879                 break;
 880
 881         case PT_STEP:
 882         case PT_CONTINUE:
 883         case PT_TO_SCE:
 884         case PT_TO_SCX:
 885         case PT_SYSCALL:
 886         case PT_DETACH:
 887                 /* Zero means do not send any signal */
 888                 if (data < 0 || data > _SIG_MAXSIG) {
 889                         error = EINVAL;
 890                         break;
 891                 }
 892
 893                 switch (req) {
 894                 case PT_STEP:
 895                         error = ptrace_single_step(td2);
 896                         if (error)
 897                                 goto out;
 898                         break;
 899                 case PT_CONTINUE:
 900                 case PT_TO_SCE:
 901                 case PT_TO_SCX:
 902                 case PT_SYSCALL:
 903                         if (addr != (void *)1) {
 904                                 error = ptrace_set_pc(td2,
 905                                     (u_long)(uintfptr_t)addr);
 906                                 if (error)
 907                                         goto out;
 908                         }
 909                         switch (req) {
 910                         case PT_TO_SCE:
 911                                 p->p_stops |= S_PT_SCE;
 912                                 break;
 913                         case PT_TO_SCX:
 914                                 p->p_stops |= S_PT_SCX;
 915                                 break;
 916                         case PT_SYSCALL:
 917                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
 918                                 break;
 919                         }
 920                         break;
 921                 case PT_DETACH:
 922                         /* reset process parent */
 923                         if (p->p_oppid != p->p_pptr->p_pid) {
 924                                 struct proc *pp;
 925
 926                                 PROC_LOCK(p->p_pptr);
 927                                 sigqueue_take(p->p_ksi);
 928                                 PROC_UNLOCK(p->p_pptr);
 929
 930                                 PROC_UNLOCK(p);
 931                                 pp = pfind(p->p_oppid);
 932                                 if (pp == NULL)
 933                                         pp = initproc;
 934                                 else
 935                                         PROC_UNLOCK(pp);
 936                                 PROC_LOCK(p);
 937                                 proc_reparent(p, pp);
 938                                 if (pp == initproc)
 939                                         p->p_sigparent = SIGCHLD;
 940                         }
 941                         p->p_flag &= ~(P_TRACED | P_WAITED);
 942                         p->p_oppid = 0;
 943
 944                         /* should we send SIGCHLD? */
 945                         /* childproc_continued(p); */
 946                         break;
 947                 }
 948
 949         sendsig:
 950                 if (proctree_locked) {
 951                         sx_xunlock(&proctree_lock);
 952                         proctree_locked = 0;
 953                 }
 954                 p->p_xstat = data;
 955                 p->p_xthread = NULL;
 956                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 957                         /* deliver or queue signal */
 958                         td2->td_dbgflags &= ~TDB_XSIG;
 959                         td2->td_xsig = data;
 960
 961                         if (req == PT_DETACH) {
 962                                 struct thread *td3;
 963                                 FOREACH_THREAD_IN_PROC(p, td3) {
 964                                         td3->td_dbgflags &= ~TDB_SUSPEND;
 965                                 }
 966                         }
 967                         /*
 968                          * unsuspend all threads, to not let a thread run,
 969                          * you should use PT_SUSPEND to suspend it before
 970                          * continuing process.
 971                          */
 972                         PROC_SLOCK(p);
 973                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
 974                         thread_unsuspend(p);
 975                         PROC_SUNLOCK(p);
 976                 } else {
 977                         if (data)
 978                                 psignal(p, data);
 979                 }
 980                 break;
 981
 982         case PT_WRITE_I:
 983         case PT_WRITE_D:
 984                 td2->td_dbgflags |= TDB_USERWR;
 985                 write = 1;
 986                 /* FALLTHROUGH */
 987         case PT_READ_I:
 988         case PT_READ_D:
 989                 PROC_UNLOCK(p);
 990                 tmp = 0;
 991                 /* write = 0 set above */
 992                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
 993                 iov.iov_len = sizeof(int);
 994                 uio.uio_iov = &iov;
 995                 uio.uio_iovcnt = 1;
 996                 uio.uio_offset = (off_t)(uintptr_t)addr;
 997                 uio.uio_resid = sizeof(int);
 998                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
 999                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1000                 uio.uio_td = td;
1001                 error = proc_rwmem(p, &uio);
1002                 if (uio.uio_resid != 0) {
1003                         /*
1004                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1005                          * so I think write() can bogusly return 0.
1006                          * XXX what happens for short writes?  We don't want
1007                          * to write partial data.
1008                          * XXX proc_rwmem() returns EPERM for other invalid
1009                          * addresses.  Convert this to EINVAL.  Does this
1010                          * clobber returns of EPERM for other reasons?
1011                          */
1012                         if (error == 0 || error == ENOSPC || error == EPERM)
1013                                 error = EINVAL; /* EOF */
1014                 }
1015                 if (!write)
1016                         td->td_retval[0] = tmp;
1017                 PROC_LOCK(p);
1018                 break;
1019
1020         case PT_IO:
1021 #ifdef COMPAT_FREEBSD32
1022                 if (wrap32) {
1023                         piod32 = addr;
1024                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1025                         iov.iov_len = piod32->piod_len;
1026                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1027                         uio.uio_resid = piod32->piod_len;
1028                 } else
1029 #endif
1030                 {
1031                         piod = addr;
1032                         iov.iov_base = piod->piod_addr;
1033                         iov.iov_len = piod->piod_len;
1034                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1035                         uio.uio_resid = piod->piod_len;
1036                 }
1037                 uio.uio_iov = &iov;
1038                 uio.uio_iovcnt = 1;
1039                 uio.uio_segflg = UIO_USERSPACE;
1040                 uio.uio_td = td;
1041 #ifdef COMPAT_FREEBSD32
1042                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1043 #else
1044                 tmp = piod->piod_op;
1045 #endif
1046                 switch (tmp) {
1047                 case PIOD_READ_D:
1048                 case PIOD_READ_I:
1049                         uio.uio_rw = UIO_READ;
1050                         break;
1051                 case PIOD_WRITE_D:
1052                 case PIOD_WRITE_I:
1053                         td2->td_dbgflags |= TDB_USERWR;
1054                         uio.uio_rw = UIO_WRITE;
1055                         break;
1056                 default:
1057                         error = EINVAL;
1058                         goto out;
1059                 }
1060                 PROC_UNLOCK(p);
1061                 error = proc_rwmem(p, &uio);
1062 #ifdef COMPAT_FREEBSD32
1063                 if (wrap32)
1064                         piod32->piod_len -= uio.uio_resid;
1065                 else
1066 #endif
1067                         piod->piod_len -= uio.uio_resid;
1068                 PROC_LOCK(p);
1069                 break;
1070
1071         case PT_KILL:
1072                 data = SIGKILL;
1073                 goto sendsig;   /* in PT_CONTINUE above */
1074
1075         case PT_SETREGS:
1076                 td2->td_dbgflags |= TDB_USERWR;
1077                 error = PROC_WRITE(regs, td2, addr);
1078                 break;
1079
1080         case PT_GETREGS:
1081                 error = PROC_READ(regs, td2, addr);
1082                 break;
1083
1084         case PT_SETFPREGS:
1085                 td2->td_dbgflags |= TDB_USERWR;
1086                 error = PROC_WRITE(fpregs, td2, addr);
1087                 break;
1088
1089         case PT_GETFPREGS:
1090                 error = PROC_READ(fpregs, td2, addr);
1091                 break;
1092
1093         case PT_SETDBREGS:
1094                 td2->td_dbgflags |= TDB_USERWR;
1095                 error = PROC_WRITE(dbregs, td2, addr);
1096                 break;
1097
1098         case PT_GETDBREGS:
1099                 error = PROC_READ(dbregs, td2, addr);
1100                 break;
1101
1102         case PT_LWPINFO:
1103                 if (data <= 0 || data > sizeof(*pl)) {
1104                         error = EINVAL;
1105                         break;
1106                 }
1107                 pl = addr;
1108                 pl->pl_lwpid = td2->td_tid;
1109                 if (td2->td_dbgflags & TDB_XSIG)
1110                         pl->pl_event = PL_EVENT_SIGNAL;
1111                 else
1112                         pl->pl_event = 0;
1113                 pl->pl_flags = 0;
1114                 pl->pl_sigmask = td2->td_sigmask;
1115                 pl->pl_siglist = td2->td_siglist;
1116                 break;
1117
1118         case PT_GETNUMLWPS:
1119                 td->td_retval[0] = p->p_numthreads;
1120                 break;
1121
1122         case PT_GETLWPLIST:
1123                 if (data <= 0) {
1124                         error = EINVAL;
1125                         break;
1126                 }
1127                 num = imin(p->p_numthreads, data);
1128                 PROC_UNLOCK(p);
1129                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1130                 tmp = 0;
1131                 PROC_LOCK(p);
1132                 FOREACH_THREAD_IN_PROC(p, td2) {
1133                         if (tmp >= num)
1134                                 break;
1135                         buf[tmp++] = td2->td_tid;
1136                 }
1137                 PROC_UNLOCK(p);
1138                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1139                 free(buf, M_TEMP);
1140                 if (!error)
1141                         td->td_retval[0] = tmp;
1142                 PROC_LOCK(p);
1143                 break;
1144
1145         case PT_VM_TIMESTAMP:
1146                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1147                 break;
1148
1149         case PT_VM_ENTRY:
1150                 PROC_UNLOCK(p);
1151 #ifdef COMPAT_FREEBSD32
1152                 if (wrap32)
1153                         error = ptrace_vm_entry32(td, p, addr);
1154                 else
1155 #endif
1156                 error = ptrace_vm_entry(td, p, addr);
1157                 PROC_LOCK(p);
1158                 break;
1159
1160         default:
1161 #ifdef __HAVE_PTRACE_MACHDEP
1162                 if (req >= PT_FIRSTMACH) {
1163                         PROC_UNLOCK(p);
1164                         error = cpu_ptrace(td2, req, addr, data);
1165                         PROC_LOCK(p);
1166                 } else
1167 #endif
1168                         /* Unknown request. */
1169                         error = EINVAL;
1170                 break;
1171         }
1172
1173 out:
1174         /* Drop our hold on this process now that the request has completed. */
1175         _PRELE(p);
1176 fail:
1177         PROC_UNLOCK(p);
1178         if (proctree_locked)
1179                 sx_xunlock(&proctree_lock);
1180         return (error);
1181 }
1182 #undef PROC_READ
1183 #undef PROC_WRITE
1184
1185 /*
1186  * Stop a process because of a debugging event;
1187  * stay stopped until p->p_step is cleared
1188  * (cleared by PIOCCONT in procfs).
1189  */
1190 void
1191 stopevent(struct proc *p, unsigned int event, unsigned int val)
1192 {
1193
1194         PROC_LOCK_ASSERT(p, MA_OWNED);
1195         p->p_step = 1;
1196         do {
1197                 p->p_xstat = val;
1198                 p->p_xthread = NULL;
1199                 p->p_stype = event;     /* Which event caused the stop? */
1200                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1201                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1202         } while (p->p_step);
1203 }