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 #include <compat/freebsd32/freebsd32_signal.h>
  67
  68 struct ptrace_io_desc32 {
  69         int             piod_op;
  70         u_int32_t       piod_offs;
  71         u_int32_t       piod_addr;
  72         u_int32_t       piod_len;
  73 };
  74
  75 struct ptrace_vm_entry32 {
  76         int             pve_entry;
  77         int             pve_timestamp;
  78         uint32_t        pve_start;
  79         uint32_t        pve_end;
  80         uint32_t        pve_offset;
  81         u_int           pve_prot;
  82         u_int           pve_pathlen;
  83         int32_t         pve_fileid;
  84         u_int           pve_fsid;
  85         uint32_t        pve_path;
  86 };
  87
  88 struct ptrace_lwpinfo32 {
  89         lwpid_t pl_lwpid;       /* LWP described. */
  90         int     pl_event;       /* Event that stopped the LWP. */
  91         int     pl_flags;       /* LWP flags. */
  92         sigset_t        pl_sigmask;     /* LWP signal mask */
  93         sigset_t        pl_siglist;     /* LWP pending signal */
  94         struct siginfo32 pl_siginfo;    /* siginfo for signal */
  95 };
  96
  97 #endif
  98
  99 /*
 100  * Functions implemented using PROC_ACTION():
 101  *
 102  * proc_read_regs(proc, regs)
 103  *      Get the current user-visible register set from the process
 104  *      and copy it into the regs structure (<machine/reg.h>).
 105  *      The process is stopped at the time read_regs is called.
 106  *
 107  * proc_write_regs(proc, regs)
 108  *      Update the current register set from the passed in regs
 109  *      structure.  Take care to avoid clobbering special CPU
 110  *      registers or privileged bits in the PSL.
 111  *      Depending on the architecture this may have fix-up work to do,
 112  *      especially if the IAR or PCW are modified.
 113  *      The process is stopped at the time write_regs is called.
 114  *
 115  * proc_read_fpregs, proc_write_fpregs
 116  *      deal with the floating point register set, otherwise as above.
 117  *
 118  * proc_read_dbregs, proc_write_dbregs
 119  *      deal with the processor debug register set, otherwise as above.
 120  *
 121  * proc_sstep(proc)
 122  *      Arrange for the process to trap after executing a single instruction.
 123  */
 124
 125 #define PROC_ACTION(action) do {                                        \
 126         int error;                                                      \
 127                                                                         \
 128         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 129         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
 130                 error = EIO;                                            \
 131         else                                                            \
 132                 error = (action);                                       \
 133         return (error);                                                 \
 134 } while(0)
 135
 136 int
 137 proc_read_regs(struct thread *td, struct reg *regs)
 138 {
 139
 140         PROC_ACTION(fill_regs(td, regs));
 141 }
 142
 143 int
 144 proc_write_regs(struct thread *td, struct reg *regs)
 145 {
 146
 147         PROC_ACTION(set_regs(td, regs));
 148 }
 149
 150 int
 151 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 152 {
 153
 154         PROC_ACTION(fill_dbregs(td, dbregs));
 155 }
 156
 157 int
 158 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 159 {
 160
 161         PROC_ACTION(set_dbregs(td, dbregs));
 162 }
 163
 164 /*
 165  * Ptrace doesn't support fpregs at all, and there are no security holes
 166  * or translations for fpregs, so we can just copy them.
 167  */
 168 int
 169 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 170 {
 171
 172         PROC_ACTION(fill_fpregs(td, fpregs));
 173 }
 174
 175 int
 176 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 177 {
 178
 179         PROC_ACTION(set_fpregs(td, fpregs));
 180 }
 181
 182 #ifdef COMPAT_FREEBSD32
 183 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 184 int
 185 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 186 {
 187
 188         PROC_ACTION(fill_regs32(td, regs32));
 189 }
 190
 191 int
 192 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 193 {
 194
 195         PROC_ACTION(set_regs32(td, regs32));
 196 }
 197
 198 int
 199 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 200 {
 201
 202         PROC_ACTION(fill_dbregs32(td, dbregs32));
 203 }
 204
 205 int
 206 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 207 {
 208
 209         PROC_ACTION(set_dbregs32(td, dbregs32));
 210 }
 211
 212 int
 213 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 214 {
 215
 216         PROC_ACTION(fill_fpregs32(td, fpregs32));
 217 }
 218
 219 int
 220 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 221 {
 222
 223         PROC_ACTION(set_fpregs32(td, fpregs32));
 224 }
 225 #endif
 226
 227 int
 228 proc_sstep(struct thread *td)
 229 {
 230
 231         PROC_ACTION(ptrace_single_step(td));
 232 }
 233
 234 int
 235 proc_rwmem(struct proc *p, struct uio *uio)
 236 {
 237         vm_map_t map;
 238         vm_object_t backing_object, object = NULL;
 239         vm_offset_t pageno = 0;         /* page number */
 240         vm_prot_t reqprot;
 241         int error, fault_flags, writing;
 242
 243         /*
 244          * Assert that someone has locked this vmspace.  (Should be
 245          * curthread but we can't assert that.)  This keeps the process
 246          * from exiting out from under us until this operation completes.
 247          */
 248         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 249             p, p->p_pid));
 250
 251         /*
 252          * The map we want...
 253          */
 254         map = &p->p_vmspace->vm_map;
 255
 256         writing = uio->uio_rw == UIO_WRITE;
 257         reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
 258             VM_PROT_READ;
 259         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
 260
 261         /*
 262          * Only map in one page at a time.  We don't have to, but it
 263          * makes things easier.  This way is trivial - right?
 264          */
 265         do {
 266                 vm_map_t tmap;
 267                 vm_offset_t uva;
 268                 int page_offset;                /* offset into page */
 269                 vm_map_entry_t out_entry;
 270                 vm_prot_t out_prot;
 271                 boolean_t wired;
 272                 vm_pindex_t pindex;
 273                 u_int len;
 274                 vm_page_t m;
 275
 276                 object = NULL;
 277
 278                 uva = (vm_offset_t)uio->uio_offset;
 279
 280                 /*
 281                  * Get the page number of this segment.
 282                  */
 283                 pageno = trunc_page(uva);
 284                 page_offset = uva - pageno;
 285
 286                 /*
 287                  * How many bytes to copy
 288                  */
 289                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 290
 291                 /*
 292                  * Fault the page on behalf of the process
 293                  */
 294                 error = vm_fault(map, pageno, reqprot, fault_flags);
 295                 if (error) {
 296                         if (error == KERN_RESOURCE_SHORTAGE)
 297                                 error = ENOMEM;
 298                         else
 299                                 error = EFAULT;
 300                         break;
 301                 }
 302
 303                 /*
 304                  * Now we need to get the page.  out_entry, out_prot, wired,
 305                  * and single_use aren't used.  One would think the vm code
 306                  * would be a *bit* nicer...  We use tmap because
 307                  * vm_map_lookup() can change the map argument.
 308                  */
 309                 tmap = map;
 310                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 311                     &object, &pindex, &out_prot, &wired);
 312                 if (error) {
 313                         error = EFAULT;
 314                         break;
 315                 }
 316                 VM_OBJECT_LOCK(object);
 317                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 318                     !writing &&
 319                     (backing_object = object->backing_object) != NULL) {
 320                         /*
 321                          * Allow fallback to backing objects if we are reading.
 322                          */
 323                         VM_OBJECT_LOCK(backing_object);
 324                         pindex += OFF_TO_IDX(object->backing_object_offset);
 325                         VM_OBJECT_UNLOCK(object);
 326                         object = backing_object;
 327                 }
 328                 VM_OBJECT_UNLOCK(object);
 329                 if (m == NULL) {
 330                         vm_map_lookup_done(tmap, out_entry);
 331                         error = EFAULT;
 332                         break;
 333                 }
 334
 335                 /*
 336                  * Hold the page in memory.
 337                  */
 338                 vm_page_lock_queues();
 339                 vm_page_hold(m);
 340                 vm_page_unlock_queues();
 341
 342                 /*
 343                  * We're done with tmap now.
 344                  */
 345                 vm_map_lookup_done(tmap, out_entry);
 346
 347                 /*
 348                  * Now do the i/o move.
 349                  */
 350                 error = uiomove_fromphys(&m, page_offset, len, uio);
 351
 352                 /* Make the I-cache coherent for breakpoints. */
 353                 if (!error && writing && (out_prot & VM_PROT_EXECUTE))
 354                         vm_sync_icache(map, uva, len);
 355
 356                 /*
 357                  * Release the page.
 358                  */
 359                 vm_page_lock_queues();
 360                 vm_page_unhold(m);
 361                 vm_page_unlock_queues();
 362
 363         } while (error == 0 && uio->uio_resid > 0);
 364
 365         return (error);
 366 }
 367
 368 static int
 369 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
 370 {
 371         struct vattr vattr;
 372         vm_map_t map;
 373         vm_map_entry_t entry;
 374         vm_object_t obj, tobj, lobj;
 375         struct vmspace *vm;
 376         struct vnode *vp;
 377         char *freepath, *fullpath;
 378         u_int pathlen;
 379         int error, index, vfslocked;
 380
 381         error = 0;
 382         obj = NULL;
 383
 384         vm = vmspace_acquire_ref(p);
 385         map = &vm->vm_map;
 386         vm_map_lock_read(map);
 387
 388         do {
 389                 entry = map->header.next;
 390                 index = 0;
 391                 while (index < pve->pve_entry && entry != &map->header) {
 392                         entry = entry->next;
 393                         index++;
 394                 }
 395                 if (index != pve->pve_entry) {
 396                         error = EINVAL;
 397                         break;
 398                 }
 399                 while (entry != &map->header &&
 400                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
 401                         entry = entry->next;
 402                         index++;
 403                 }
 404                 if (entry == &map->header) {
 405                         error = ENOENT;
 406                         break;
 407                 }
 408
 409                 /* We got an entry. */
 410                 pve->pve_entry = index + 1;
 411                 pve->pve_timestamp = map->timestamp;
 412                 pve->pve_start = entry->start;
 413                 pve->pve_end = entry->end - 1;
 414                 pve->pve_offset = entry->offset;
 415                 pve->pve_prot = entry->protection;
 416
 417                 /* Backing object's path needed? */
 418                 if (pve->pve_pathlen == 0)
 419                         break;
 420
 421                 pathlen = pve->pve_pathlen;
 422                 pve->pve_pathlen = 0;
 423
 424                 obj = entry->object.vm_object;
 425                 if (obj != NULL)
 426                         VM_OBJECT_LOCK(obj);
 427         } while (0);
 428
 429         vm_map_unlock_read(map);
 430         vmspace_free(vm);
 431
 432         pve->pve_fsid = VNOVAL;
 433         pve->pve_fileid = VNOVAL;
 434
 435         if (error == 0 && obj != NULL) {
 436                 lobj = obj;
 437                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
 438                         if (tobj != obj)
 439                                 VM_OBJECT_LOCK(tobj);
 440                         if (lobj != obj)
 441                                 VM_OBJECT_UNLOCK(lobj);
 442                         lobj = tobj;
 443                         pve->pve_offset += tobj->backing_object_offset;
 444                 }
 445                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 446                 if (vp != NULL)
 447                         vref(vp);
 448                 if (lobj != obj)
 449                         VM_OBJECT_UNLOCK(lobj);
 450                 VM_OBJECT_UNLOCK(obj);
 451
 452                 if (vp != NULL) {
 453                         freepath = NULL;
 454                         fullpath = NULL;
 455                         vn_fullpath(td, vp, &fullpath, &freepath);
 456                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 457                         vn_lock(vp, LK_SHARED | LK_RETRY);
 458                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 459                                 pve->pve_fileid = vattr.va_fileid;
 460                                 pve->pve_fsid = vattr.va_fsid;
 461                         }
 462                         vput(vp);
 463                         VFS_UNLOCK_GIANT(vfslocked);
 464
 465                         if (fullpath != NULL) {
 466                                 pve->pve_pathlen = strlen(fullpath) + 1;
 467                                 if (pve->pve_pathlen <= pathlen) {
 468                                         error = copyout(fullpath, pve->pve_path,
 469                                             pve->pve_pathlen);
 470                                 } else
 471                                         error = ENAMETOOLONG;
 472                         }
 473                         if (freepath != NULL)
 474                                 free(freepath, M_TEMP);
 475                 }
 476         }
 477
 478         return (error);
 479 }
 480
 481 #ifdef COMPAT_FREEBSD32
 482 static int
 483 ptrace_vm_entry32(struct thread *td, struct proc *p,
 484     struct ptrace_vm_entry32 *pve32)
 485 {
 486         struct ptrace_vm_entry pve;
 487         int error;
 488
 489         pve.pve_entry = pve32->pve_entry;
 490         pve.pve_pathlen = pve32->pve_pathlen;
 491         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 492
 493         error = ptrace_vm_entry(td, p, &pve);
 494         if (error == 0) {
 495                 pve32->pve_entry = pve.pve_entry;
 496                 pve32->pve_timestamp = pve.pve_timestamp;
 497                 pve32->pve_start = pve.pve_start;
 498                 pve32->pve_end = pve.pve_end;
 499                 pve32->pve_offset = pve.pve_offset;
 500                 pve32->pve_prot = pve.pve_prot;
 501                 pve32->pve_fileid = pve.pve_fileid;
 502                 pve32->pve_fsid = pve.pve_fsid;
 503         }
 504
 505         pve32->pve_pathlen = pve.pve_pathlen;
 506         return (error);
 507 }
 508
 509 static void
 510 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
 511     struct ptrace_lwpinfo32 *pl32)
 512 {
 513
 514         pl32->pl_lwpid = pl->pl_lwpid;
 515         pl32->pl_event = pl->pl_event;
 516         pl32->pl_flags = pl->pl_flags;
 517         pl32->pl_sigmask = pl->pl_sigmask;
 518         pl32->pl_siglist = pl->pl_siglist;
 519         siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
 520 }
 521 #endif /* COMPAT_FREEBSD32 */
 522
 523 /*
 524  * Process debugging system call.
 525  */
 526 #ifndef _SYS_SYSPROTO_H_
 527 struct ptrace_args {
 528         int     req;
 529         pid_t   pid;
 530         caddr_t addr;
 531         int     data;
 532 };
 533 #endif
 534
 535 #ifdef COMPAT_FREEBSD32
 536 /*
 537  * This CPP subterfuge is to try and reduce the number of ifdefs in
 538  * the body of the code.
 539  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 540  * becomes either:
 541  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 542  * or
 543  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 544  * .. except this is done at runtime.
 545  */
 546 #define COPYIN(u, k, s)         wrap32 ? \
 547         copyin(u, k ## 32, s ## 32) : \
 548         copyin(u, k, s)
 549 #define COPYOUT(k, u, s)        wrap32 ? \
 550         copyout(k ## 32, u, s ## 32) : \
 551         copyout(k, u, s)
 552 #else
 553 #define COPYIN(u, k, s)         copyin(u, k, s)
 554 #define COPYOUT(k, u, s)        copyout(k, u, s)
 555 #endif
 556 int
 557 ptrace(struct thread *td, struct ptrace_args *uap)
 558 {
 559         /*
 560          * XXX this obfuscation is to reduce stack usage, but the register
 561          * structs may be too large to put on the stack anyway.
 562          */
 563         union {
 564                 struct ptrace_io_desc piod;
 565                 struct ptrace_lwpinfo pl;
 566                 struct ptrace_vm_entry pve;
 567                 struct dbreg dbreg;
 568                 struct fpreg fpreg;
 569                 struct reg reg;
 570 #ifdef COMPAT_FREEBSD32
 571                 struct dbreg32 dbreg32;
 572                 struct fpreg32 fpreg32;
 573                 struct reg32 reg32;
 574                 struct ptrace_io_desc32 piod32;
 575                 struct ptrace_lwpinfo32 pl32;
 576                 struct ptrace_vm_entry32 pve32;
 577 #endif
 578         } r;
 579         void *addr;
 580         int error = 0;
 581 #ifdef COMPAT_FREEBSD32
 582         int wrap32 = 0;
 583
 584         if (SV_CURPROC_FLAG(SV_ILP32))
 585                 wrap32 = 1;
 586 #endif
 587         AUDIT_ARG_PID(uap->pid);
 588         AUDIT_ARG_CMD(uap->req);
 589         AUDIT_ARG_VALUE(uap->data);
 590         addr = &r;
 591         switch (uap->req) {
 592         case PT_GETREGS:
 593         case PT_GETFPREGS:
 594         case PT_GETDBREGS:
 595         case PT_LWPINFO:
 596                 break;
 597         case PT_SETREGS:
 598                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 599                 break;
 600         case PT_SETFPREGS:
 601                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 602                 break;
 603         case PT_SETDBREGS:
 604                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 605                 break;
 606         case PT_IO:
 607                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 608                 break;
 609         case PT_VM_ENTRY:
 610                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 611                 break;
 612         default:
 613                 addr = uap->addr;
 614                 break;
 615         }
 616         if (error)
 617                 return (error);
 618
 619         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 620         if (error)
 621                 return (error);
 622
 623         switch (uap->req) {
 624         case PT_VM_ENTRY:
 625                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 626                 break;
 627         case PT_IO:
 628                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 629                 break;
 630         case PT_GETREGS:
 631                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 632                 break;
 633         case PT_GETFPREGS:
 634                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 635                 break;
 636         case PT_GETDBREGS:
 637                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 638                 break;
 639         case PT_LWPINFO:
 640                 error = copyout(&r.pl, uap->addr, uap->data);
 641                 break;
 642         }
 643
 644         return (error);
 645 }
 646 #undef COPYIN
 647 #undef COPYOUT
 648
 649 #ifdef COMPAT_FREEBSD32
 650 /*
 651  *   PROC_READ(regs, td2, addr);
 652  * becomes either:
 653  *   proc_read_regs(td2, addr);
 654  * or
 655  *   proc_read_regs32(td2, addr);
 656  * .. except this is done at runtime.  There is an additional
 657  * complication in that PROC_WRITE disallows 32 bit consumers
 658  * from writing to 64 bit address space targets.
 659  */
 660 #define PROC_READ(w, t, a)      wrap32 ? \
 661         proc_read_ ## w ## 32(t, a) : \
 662         proc_read_ ## w (t, a)
 663 #define PROC_WRITE(w, t, a)     wrap32 ? \
 664         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 665         proc_write_ ## w (t, a)
 666 #else
 667 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 668 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 669 #endif
 670
 671 int
 672 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 673 {
 674         struct iovec iov;
 675         struct uio uio;
 676         struct proc *curp, *p, *pp;
 677         struct thread *td2 = NULL;
 678         struct ptrace_io_desc *piod = NULL;
 679         struct ptrace_lwpinfo *pl;
 680         int error, write, tmp, num;
 681         int proctree_locked = 0;
 682         lwpid_t tid = 0, *buf;
 683 #ifdef COMPAT_FREEBSD32
 684         int wrap32 = 0, safe = 0;
 685         struct ptrace_io_desc32 *piod32 = NULL;
 686         struct ptrace_lwpinfo32 *pl32 = NULL;
 687         struct ptrace_lwpinfo plr;
 688 #endif
 689
 690         curp = td->td_proc;
 691
 692         /* Lock proctree before locking the process. */
 693         switch (req) {
 694         case PT_TRACE_ME:
 695         case PT_ATTACH:
 696         case PT_STEP:
 697         case PT_CONTINUE:
 698         case PT_TO_SCE:
 699         case PT_TO_SCX:
 700         case PT_SYSCALL:
 701         case PT_DETACH:
 702                 sx_xlock(&proctree_lock);
 703                 proctree_locked = 1;
 704                 break;
 705         default:
 706                 break;
 707         }
 708
 709         write = 0;
 710         if (req == PT_TRACE_ME) {
 711                 p = td->td_proc;
 712                 PROC_LOCK(p);
 713         } else {
 714                 if (pid <= PID_MAX) {
 715                         if ((p = pfind(pid)) == NULL) {
 716                                 if (proctree_locked)
 717                                         sx_xunlock(&proctree_lock);
 718                                 return (ESRCH);
 719                         }
 720                 } else {
 721                         /* this is slow, should be optimized */
 722                         sx_slock(&allproc_lock);
 723                         FOREACH_PROC_IN_SYSTEM(p) {
 724                                 PROC_LOCK(p);
 725                                 FOREACH_THREAD_IN_PROC(p, td2) {
 726                                         if (td2->td_tid == pid)
 727                                                 break;
 728                                 }
 729                                 if (td2 != NULL)
 730                                         break; /* proc lock held */
 731                                 PROC_UNLOCK(p);
 732                         }
 733                         sx_sunlock(&allproc_lock);
 734                         if (p == NULL) {
 735                                 if (proctree_locked)
 736                                         sx_xunlock(&proctree_lock);
 737                                 return (ESRCH);
 738                         }
 739                         tid = pid;
 740                         pid = p->p_pid;
 741                 }
 742         }
 743         AUDIT_ARG_PROCESS(p);
 744
 745         if ((p->p_flag & P_WEXIT) != 0) {
 746                 error = ESRCH;
 747                 goto fail;
 748         }
 749         if ((error = p_cansee(td, p)) != 0)
 750                 goto fail;
 751
 752         if ((error = p_candebug(td, p)) != 0)
 753                 goto fail;
 754
 755         /*
 756          * System processes can't be debugged.
 757          */
 758         if ((p->p_flag & P_SYSTEM) != 0) {
 759                 error = EINVAL;
 760                 goto fail;
 761         }
 762
 763         if (tid == 0) {
 764                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 765                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 766                         td2 = p->p_xthread;
 767                 } else {
 768                         td2 = FIRST_THREAD_IN_PROC(p);
 769                 }
 770                 tid = td2->td_tid;
 771         }
 772
 773 #ifdef COMPAT_FREEBSD32
 774         /*
 775          * Test if we're a 32 bit client and what the target is.
 776          * Set the wrap controls accordingly.
 777          */
 778         if (SV_CURPROC_FLAG(SV_ILP32)) {
 779                 if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
 780                         safe = 1;
 781                 wrap32 = 1;
 782         }
 783 #endif
 784         /*
 785          * Permissions check
 786          */
 787         switch (req) {
 788         case PT_TRACE_ME:
 789                 /* Always legal. */
 790                 break;
 791
 792         case PT_ATTACH:
 793                 /* Self */
 794                 if (p->p_pid == td->td_proc->p_pid) {
 795                         error = EINVAL;
 796                         goto fail;
 797                 }
 798
 799                 /* Already traced */
 800                 if (p->p_flag & P_TRACED) {
 801                         error = EBUSY;
 802                         goto fail;
 803                 }
 804
 805                 /* Can't trace an ancestor if you're being traced. */
 806                 if (curp->p_flag & P_TRACED) {
 807                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 808                                 if (pp == p) {
 809                                         error = EINVAL;
 810                                         goto fail;
 811                                 }
 812                         }
 813                 }
 814
 815
 816                 /* OK */
 817                 break;
 818
 819         case PT_CLEARSTEP:
 820                 /* Allow thread to clear single step for itself */
 821                 if (td->td_tid == tid)
 822                         break;
 823
 824                 /* FALLTHROUGH */
 825         default:
 826                 /* not being traced... */
 827                 if ((p->p_flag & P_TRACED) == 0) {
 828                         error = EPERM;
 829                         goto fail;
 830                 }
 831
 832                 /* not being traced by YOU */
 833                 if (p->p_pptr != td->td_proc) {
 834                         error = EBUSY;
 835                         goto fail;
 836                 }
 837
 838                 /* not currently stopped */
 839                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 840                     p->p_suspcount != p->p_numthreads  ||
 841                     (p->p_flag & P_WAITED) == 0) {
 842                         error = EBUSY;
 843                         goto fail;
 844                 }
 845
 846                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 847                         static int count = 0;
 848                         if (count++ == 0)
 849                                 printf("P_STOPPED_TRACE not set.\n");
 850                 }
 851
 852                 /* OK */
 853                 break;
 854         }
 855
 856         /* Keep this process around until we finish this request. */
 857         _PHOLD(p);
 858
 859 #ifdef FIX_SSTEP
 860         /*
 861          * Single step fixup ala procfs
 862          */
 863         FIX_SSTEP(td2);
 864 #endif
 865
 866         /*
 867          * Actually do the requests
 868          */
 869
 870         td->td_retval[0] = 0;
 871
 872         switch (req) {
 873         case PT_TRACE_ME:
 874                 /* set my trace flag and "owner" so it can read/write me */
 875                 p->p_flag |= P_TRACED;
 876                 p->p_oppid = p->p_pptr->p_pid;
 877                 break;
 878
 879         case PT_ATTACH:
 880                 /* security check done above */
 881                 p->p_flag |= P_TRACED;
 882                 p->p_oppid = p->p_pptr->p_pid;
 883                 if (p->p_pptr != td->td_proc)
 884                         proc_reparent(p, td->td_proc);
 885                 data = SIGSTOP;
 886                 goto sendsig;   /* in PT_CONTINUE below */
 887
 888         case PT_CLEARSTEP:
 889                 error = ptrace_clear_single_step(td2);
 890                 break;
 891
 892         case PT_SETSTEP:
 893                 error = ptrace_single_step(td2);
 894                 break;
 895
 896         case PT_SUSPEND:
 897                 td2->td_dbgflags |= TDB_SUSPEND;
 898                 thread_lock(td2);
 899                 td2->td_flags |= TDF_NEEDSUSPCHK;
 900                 thread_unlock(td2);
 901                 break;
 902
 903         case PT_RESUME:
 904                 td2->td_dbgflags &= ~TDB_SUSPEND;
 905                 break;
 906
 907         case PT_STEP:
 908         case PT_CONTINUE:
 909         case PT_TO_SCE:
 910         case PT_TO_SCX:
 911         case PT_SYSCALL:
 912         case PT_DETACH:
 913                 /* Zero means do not send any signal */
 914                 if (data < 0 || data > _SIG_MAXSIG) {
 915                         error = EINVAL;
 916                         break;
 917                 }
 918
 919                 switch (req) {
 920                 case PT_STEP:
 921                         error = ptrace_single_step(td2);
 922                         if (error)
 923                                 goto out;
 924                         break;
 925                 case PT_CONTINUE:
 926                 case PT_TO_SCE:
 927                 case PT_TO_SCX:
 928                 case PT_SYSCALL:
 929                         if (addr != (void *)1) {
 930                                 error = ptrace_set_pc(td2,
 931                                     (u_long)(uintfptr_t)addr);
 932                                 if (error)
 933                                         goto out;
 934                         }
 935                         switch (req) {
 936                         case PT_TO_SCE:
 937                                 p->p_stops |= S_PT_SCE;
 938                                 break;
 939                         case PT_TO_SCX:
 940                                 p->p_stops |= S_PT_SCX;
 941                                 break;
 942                         case PT_SYSCALL:
 943                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
 944                                 break;
 945                         }
 946                         break;
 947                 case PT_DETACH:
 948                         /* reset process parent */
 949                         if (p->p_oppid != p->p_pptr->p_pid) {
 950                                 struct proc *pp;
 951
 952                                 PROC_LOCK(p->p_pptr);
 953                                 sigqueue_take(p->p_ksi);
 954                                 PROC_UNLOCK(p->p_pptr);
 955
 956                                 PROC_UNLOCK(p);
 957                                 pp = pfind(p->p_oppid);
 958                                 if (pp == NULL)
 959                                         pp = initproc;
 960                                 else
 961                                         PROC_UNLOCK(pp);
 962                                 PROC_LOCK(p);
 963                                 proc_reparent(p, pp);
 964                                 if (pp == initproc)
 965                                         p->p_sigparent = SIGCHLD;
 966                         }
 967                         p->p_flag &= ~(P_TRACED | P_WAITED);
 968                         p->p_oppid = 0;
 969
 970                         /* should we send SIGCHLD? */
 971                         /* childproc_continued(p); */
 972                         break;
 973                 }
 974
 975         sendsig:
 976                 if (proctree_locked) {
 977                         sx_xunlock(&proctree_lock);
 978                         proctree_locked = 0;
 979                 }
 980                 p->p_xstat = data;
 981                 p->p_xthread = NULL;
 982                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 983                         /* deliver or queue signal */
 984                         td2->td_dbgflags &= ~TDB_XSIG;
 985                         td2->td_xsig = data;
 986
 987                         if (req == PT_DETACH) {
 988                                 struct thread *td3;
 989                                 FOREACH_THREAD_IN_PROC(p, td3) {
 990                                         td3->td_dbgflags &= ~TDB_SUSPEND;
 991                                 }
 992                         }
 993                         /*
 994                          * unsuspend all threads, to not let a thread run,
 995                          * you should use PT_SUSPEND to suspend it before
 996                          * continuing process.
 997                          */
 998                         PROC_SLOCK(p);
 999                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
1000                         thread_unsuspend(p);
1001                         PROC_SUNLOCK(p);
1002                 } else {
1003                         if (data)
1004                                 psignal(p, data);
1005                 }
1006                 break;
1007
1008         case PT_WRITE_I:
1009         case PT_WRITE_D:
1010                 td2->td_dbgflags |= TDB_USERWR;
1011                 write = 1;
1012                 /* FALLTHROUGH */
1013         case PT_READ_I:
1014         case PT_READ_D:
1015                 PROC_UNLOCK(p);
1016                 tmp = 0;
1017                 /* write = 0 set above */
1018                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1019                 iov.iov_len = sizeof(int);
1020                 uio.uio_iov = &iov;
1021                 uio.uio_iovcnt = 1;
1022                 uio.uio_offset = (off_t)(uintptr_t)addr;
1023                 uio.uio_resid = sizeof(int);
1024                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
1025                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1026                 uio.uio_td = td;
1027                 error = proc_rwmem(p, &uio);
1028                 if (uio.uio_resid != 0) {
1029                         /*
1030                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1031                          * so I think write() can bogusly return 0.
1032                          * XXX what happens for short writes?  We don't want
1033                          * to write partial data.
1034                          * XXX proc_rwmem() returns EPERM for other invalid
1035                          * addresses.  Convert this to EINVAL.  Does this
1036                          * clobber returns of EPERM for other reasons?
1037                          */
1038                         if (error == 0 || error == ENOSPC || error == EPERM)
1039                                 error = EINVAL; /* EOF */
1040                 }
1041                 if (!write)
1042                         td->td_retval[0] = tmp;
1043                 PROC_LOCK(p);
1044                 break;
1045
1046         case PT_IO:
1047 #ifdef COMPAT_FREEBSD32
1048                 if (wrap32) {
1049                         piod32 = addr;
1050                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1051                         iov.iov_len = piod32->piod_len;
1052                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1053                         uio.uio_resid = piod32->piod_len;
1054                 } else
1055 #endif
1056                 {
1057                         piod = addr;
1058                         iov.iov_base = piod->piod_addr;
1059                         iov.iov_len = piod->piod_len;
1060                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1061                         uio.uio_resid = piod->piod_len;
1062                 }
1063                 uio.uio_iov = &iov;
1064                 uio.uio_iovcnt = 1;
1065                 uio.uio_segflg = UIO_USERSPACE;
1066                 uio.uio_td = td;
1067 #ifdef COMPAT_FREEBSD32
1068                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1069 #else
1070                 tmp = piod->piod_op;
1071 #endif
1072                 switch (tmp) {
1073                 case PIOD_READ_D:
1074                 case PIOD_READ_I:
1075                         uio.uio_rw = UIO_READ;
1076                         break;
1077                 case PIOD_WRITE_D:
1078                 case PIOD_WRITE_I:
1079                         td2->td_dbgflags |= TDB_USERWR;
1080                         uio.uio_rw = UIO_WRITE;
1081                         break;
1082                 default:
1083                         error = EINVAL;
1084                         goto out;
1085                 }
1086                 PROC_UNLOCK(p);
1087                 error = proc_rwmem(p, &uio);
1088 #ifdef COMPAT_FREEBSD32
1089                 if (wrap32)
1090                         piod32->piod_len -= uio.uio_resid;
1091                 else
1092 #endif
1093                         piod->piod_len -= uio.uio_resid;
1094                 PROC_LOCK(p);
1095                 break;
1096
1097         case PT_KILL:
1098                 data = SIGKILL;
1099                 goto sendsig;   /* in PT_CONTINUE above */
1100
1101         case PT_SETREGS:
1102                 td2->td_dbgflags |= TDB_USERWR;
1103                 error = PROC_WRITE(regs, td2, addr);
1104                 break;
1105
1106         case PT_GETREGS:
1107                 error = PROC_READ(regs, td2, addr);
1108                 break;
1109
1110         case PT_SETFPREGS:
1111                 td2->td_dbgflags |= TDB_USERWR;
1112                 error = PROC_WRITE(fpregs, td2, addr);
1113                 break;
1114
1115         case PT_GETFPREGS:
1116                 error = PROC_READ(fpregs, td2, addr);
1117                 break;
1118
1119         case PT_SETDBREGS:
1120                 td2->td_dbgflags |= TDB_USERWR;
1121                 error = PROC_WRITE(dbregs, td2, addr);
1122                 break;
1123
1124         case PT_GETDBREGS:
1125                 error = PROC_READ(dbregs, td2, addr);
1126                 break;
1127
1128         case PT_LWPINFO:
1129                 if (data <= 0 ||
1130 #ifdef COMPAT_FREEBSD32
1131                     (!wrap32 && data > sizeof(*pl)) ||
1132                     (wrap32 && data > sizeof(*pl32))) {
1133 #else
1134                     data > sizeof(*pl)) {
1135 #endif
1136                         error = EINVAL;
1137                         break;
1138                 }
1139 #ifdef COMPAT_FREEBSD32
1140                 if (wrap32) {
1141                         pl = &plr;
1142                         pl32 = addr;
1143                 } else
1144 #endif
1145                 pl = addr;
1146                 pl->pl_lwpid = td2->td_tid;
1147                 pl->pl_flags = 0;
1148                 if (td2->td_dbgflags & TDB_XSIG) {
1149                         pl->pl_event = PL_EVENT_SIGNAL;
1150                         if (td2->td_dbgksi.ksi_signo != 0 &&
1151 #ifdef COMPAT_FREEBSD32
1152                             ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1153                             pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1154                             (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1155                             pl_siginfo) + sizeof(struct siginfo32)))
1156 #else
1157                             data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1158                             + sizeof(pl->pl_siginfo)
1159 #endif
1160                         ){
1161                                 pl->pl_flags |= PL_FLAG_SI;
1162                                 pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1163                         }
1164                 }
1165                 if ((pl->pl_flags & PL_FLAG_SI) == 0)
1166                         bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1167                 if (td2->td_dbgflags & TDB_SCE)
1168                         pl->pl_flags |= PL_FLAG_SCE;
1169                 else if (td2->td_dbgflags & TDB_SCX)
1170                         pl->pl_flags |= PL_FLAG_SCX;
1171                 if (td2->td_dbgflags & TDB_EXEC)
1172                         pl->pl_flags |= PL_FLAG_EXEC;
1173                 pl->pl_sigmask = td2->td_sigmask;
1174                 pl->pl_siglist = td2->td_siglist;
1175 #ifdef COMPAT_FREEBSD32
1176                 if (wrap32)
1177                         ptrace_lwpinfo_to32(pl, pl32);
1178 #endif
1179                 break;
1180
1181         case PT_GETNUMLWPS:
1182                 td->td_retval[0] = p->p_numthreads;
1183                 break;
1184
1185         case PT_GETLWPLIST:
1186                 if (data <= 0) {
1187                         error = EINVAL;
1188                         break;
1189                 }
1190                 num = imin(p->p_numthreads, data);
1191                 PROC_UNLOCK(p);
1192                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1193                 tmp = 0;
1194                 PROC_LOCK(p);
1195                 FOREACH_THREAD_IN_PROC(p, td2) {
1196                         if (tmp >= num)
1197                                 break;
1198                         buf[tmp++] = td2->td_tid;
1199                 }
1200                 PROC_UNLOCK(p);
1201                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1202                 free(buf, M_TEMP);
1203                 if (!error)
1204                         td->td_retval[0] = tmp;
1205                 PROC_LOCK(p);
1206                 break;
1207
1208         case PT_VM_TIMESTAMP:
1209                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1210                 break;
1211
1212         case PT_VM_ENTRY:
1213                 PROC_UNLOCK(p);
1214 #ifdef COMPAT_FREEBSD32
1215                 if (wrap32)
1216                         error = ptrace_vm_entry32(td, p, addr);
1217                 else
1218 #endif
1219                 error = ptrace_vm_entry(td, p, addr);
1220                 PROC_LOCK(p);
1221                 break;
1222
1223         default:
1224 #ifdef __HAVE_PTRACE_MACHDEP
1225                 if (req >= PT_FIRSTMACH) {
1226                         PROC_UNLOCK(p);
1227                         error = cpu_ptrace(td2, req, addr, data);
1228                         PROC_LOCK(p);
1229                 } else
1230 #endif
1231                         /* Unknown request. */
1232                         error = EINVAL;
1233                 break;
1234         }
1235
1236 out:
1237         /* Drop our hold on this process now that the request has completed. */
1238         _PRELE(p);
1239 fail:
1240         PROC_UNLOCK(p);
1241         if (proctree_locked)
1242                 sx_xunlock(&proctree_lock);
1243         return (error);
1244 }
1245 #undef PROC_READ
1246 #undef PROC_WRITE
1247
1248 /*
1249  * Stop a process because of a debugging event;
1250  * stay stopped until p->p_step is cleared
1251  * (cleared by PIOCCONT in procfs).
1252  */
1253 void
1254 stopevent(struct proc *p, unsigned int event, unsigned int val)
1255 {
1256
1257         PROC_LOCK_ASSERT(p, MA_OWNED);
1258         p->p_step = 1;
1259         do {
1260                 p->p_xstat = val;
1261                 p->p_xthread = NULL;
1262                 p->p_stype = event;     /* Which event caused the stop? */
1263                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1264                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1265         } while (p->p_step);
1266 }