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         char    pl_tdname[MAXCOMLEN + 1];       /* LWP name. */
  96         int     pl_child_pid;           /* New child pid */
  97 };
  98
  99 #endif
 100
 101 /*
 102  * Functions implemented using PROC_ACTION():
 103  *
 104  * proc_read_regs(proc, regs)
 105  *      Get the current user-visible register set from the process
 106  *      and copy it into the regs structure (<machine/reg.h>).
 107  *      The process is stopped at the time read_regs is called.
 108  *
 109  * proc_write_regs(proc, regs)
 110  *      Update the current register set from the passed in regs
 111  *      structure.  Take care to avoid clobbering special CPU
 112  *      registers or privileged bits in the PSL.
 113  *      Depending on the architecture this may have fix-up work to do,
 114  *      especially if the IAR or PCW are modified.
 115  *      The process is stopped at the time write_regs is called.
 116  *
 117  * proc_read_fpregs, proc_write_fpregs
 118  *      deal with the floating point register set, otherwise as above.
 119  *
 120  * proc_read_dbregs, proc_write_dbregs
 121  *      deal with the processor debug register set, otherwise as above.
 122  *
 123  * proc_sstep(proc)
 124  *      Arrange for the process to trap after executing a single instruction.
 125  */
 126
 127 #define PROC_ACTION(action) do {                                        \
 128         int error;                                                      \
 129                                                                         \
 130         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 131         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
 132                 error = EIO;                                            \
 133         else                                                            \
 134                 error = (action);                                       \
 135         return (error);                                                 \
 136 } while(0)
 137
 138 int
 139 proc_read_regs(struct thread *td, struct reg *regs)
 140 {
 141
 142         PROC_ACTION(fill_regs(td, regs));
 143 }
 144
 145 int
 146 proc_write_regs(struct thread *td, struct reg *regs)
 147 {
 148
 149         PROC_ACTION(set_regs(td, regs));
 150 }
 151
 152 int
 153 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 154 {
 155
 156         PROC_ACTION(fill_dbregs(td, dbregs));
 157 }
 158
 159 int
 160 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 161 {
 162
 163         PROC_ACTION(set_dbregs(td, dbregs));
 164 }
 165
 166 /*
 167  * Ptrace doesn't support fpregs at all, and there are no security holes
 168  * or translations for fpregs, so we can just copy them.
 169  */
 170 int
 171 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 172 {
 173
 174         PROC_ACTION(fill_fpregs(td, fpregs));
 175 }
 176
 177 int
 178 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 179 {
 180
 181         PROC_ACTION(set_fpregs(td, fpregs));
 182 }
 183
 184 #ifdef COMPAT_FREEBSD32
 185 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 186 int
 187 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 188 {
 189
 190         PROC_ACTION(fill_regs32(td, regs32));
 191 }
 192
 193 int
 194 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 195 {
 196
 197         PROC_ACTION(set_regs32(td, regs32));
 198 }
 199
 200 int
 201 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 202 {
 203
 204         PROC_ACTION(fill_dbregs32(td, dbregs32));
 205 }
 206
 207 int
 208 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 209 {
 210
 211         PROC_ACTION(set_dbregs32(td, dbregs32));
 212 }
 213
 214 int
 215 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 216 {
 217
 218         PROC_ACTION(fill_fpregs32(td, fpregs32));
 219 }
 220
 221 int
 222 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 223 {
 224
 225         PROC_ACTION(set_fpregs32(td, fpregs32));
 226 }
 227 #endif
 228
 229 int
 230 proc_sstep(struct thread *td)
 231 {
 232
 233         PROC_ACTION(ptrace_single_step(td));
 234 }
 235
 236 int
 237 proc_rwmem(struct proc *p, struct uio *uio)
 238 {
 239         vm_map_t map;
 240         vm_object_t backing_object, object = NULL;
 241         vm_offset_t pageno = 0;         /* page number */
 242         vm_prot_t reqprot;
 243         int error, fault_flags, writing;
 244
 245         /*
 246          * Assert that someone has locked this vmspace.  (Should be
 247          * curthread but we can't assert that.)  This keeps the process
 248          * from exiting out from under us until this operation completes.
 249          */
 250         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 251             p, p->p_pid));
 252
 253         /*
 254          * The map we want...
 255          */
 256         map = &p->p_vmspace->vm_map;
 257
 258         writing = uio->uio_rw == UIO_WRITE;
 259         reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
 260             VM_PROT_READ;
 261         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
 262
 263         /*
 264          * Only map in one page at a time.  We don't have to, but it
 265          * makes things easier.  This way is trivial - right?
 266          */
 267         do {
 268                 vm_map_t tmap;
 269                 vm_offset_t uva;
 270                 int page_offset;                /* offset into page */
 271                 vm_map_entry_t out_entry;
 272                 vm_prot_t out_prot;
 273                 boolean_t wired;
 274                 vm_pindex_t pindex;
 275                 u_int len;
 276                 vm_page_t m;
 277
 278                 object = NULL;
 279
 280                 uva = (vm_offset_t)uio->uio_offset;
 281
 282                 /*
 283                  * Get the page number of this segment.
 284                  */
 285                 pageno = trunc_page(uva);
 286                 page_offset = uva - pageno;
 287
 288                 /*
 289                  * How many bytes to copy
 290                  */
 291                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 292
 293                 /*
 294                  * Fault the page on behalf of the process
 295                  */
 296                 error = vm_fault(map, pageno, reqprot, fault_flags);
 297                 if (error) {
 298                         if (error == KERN_RESOURCE_SHORTAGE)
 299                                 error = ENOMEM;
 300                         else
 301                                 error = EFAULT;
 302                         break;
 303                 }
 304
 305                 /*
 306                  * Now we need to get the page.  out_entry, out_prot, wired,
 307                  * and single_use aren't used.  One would think the vm code
 308                  * would be a *bit* nicer...  We use tmap because
 309                  * vm_map_lookup() can change the map argument.
 310                  */
 311                 tmap = map;
 312                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 313                     &object, &pindex, &out_prot, &wired);
 314                 if (error) {
 315                         error = EFAULT;
 316                         break;
 317                 }
 318                 VM_OBJECT_LOCK(object);
 319                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 320                     !writing &&
 321                     (backing_object = object->backing_object) != NULL) {
 322                         /*
 323                          * Allow fallback to backing objects if we are reading.
 324                          */
 325                         VM_OBJECT_LOCK(backing_object);
 326                         pindex += OFF_TO_IDX(object->backing_object_offset);
 327                         VM_OBJECT_UNLOCK(object);
 328                         object = backing_object;
 329                 }
 330                 VM_OBJECT_UNLOCK(object);
 331                 if (m == NULL) {
 332                         vm_map_lookup_done(tmap, out_entry);
 333                         error = EFAULT;
 334                         break;
 335                 }
 336
 337                 /*
 338                  * Hold the page in memory.
 339                  */
 340                 vm_page_lock_queues();
 341                 vm_page_hold(m);
 342                 vm_page_unlock_queues();
 343
 344                 /*
 345                  * We're done with tmap now.
 346                  */
 347                 vm_map_lookup_done(tmap, out_entry);
 348
 349                 /*
 350                  * Now do the i/o move.
 351                  */
 352                 error = uiomove_fromphys(&m, page_offset, len, uio);
 353
 354                 /* Make the I-cache coherent for breakpoints. */
 355                 if (!error && writing && (out_prot & VM_PROT_EXECUTE))
 356                         vm_sync_icache(map, uva, len);
 357
 358                 /*
 359                  * Release the page.
 360                  */
 361                 vm_page_lock_queues();
 362                 vm_page_unhold(m);
 363                 vm_page_unlock_queues();
 364
 365         } while (error == 0 && uio->uio_resid > 0);
 366
 367         return (error);
 368 }
 369
 370 static int
 371 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
 372 {
 373         struct vattr vattr;
 374         vm_map_t map;
 375         vm_map_entry_t entry;
 376         vm_object_t obj, tobj, lobj;
 377         struct vmspace *vm;
 378         struct vnode *vp;
 379         char *freepath, *fullpath;
 380         u_int pathlen;
 381         int error, index, vfslocked;
 382
 383         error = 0;
 384         obj = NULL;
 385
 386         vm = vmspace_acquire_ref(p);
 387         map = &vm->vm_map;
 388         vm_map_lock_read(map);
 389
 390         do {
 391                 entry = map->header.next;
 392                 index = 0;
 393                 while (index < pve->pve_entry && entry != &map->header) {
 394                         entry = entry->next;
 395                         index++;
 396                 }
 397                 if (index != pve->pve_entry) {
 398                         error = EINVAL;
 399                         break;
 400                 }
 401                 while (entry != &map->header &&
 402                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
 403                         entry = entry->next;
 404                         index++;
 405                 }
 406                 if (entry == &map->header) {
 407                         error = ENOENT;
 408                         break;
 409                 }
 410
 411                 /* We got an entry. */
 412                 pve->pve_entry = index + 1;
 413                 pve->pve_timestamp = map->timestamp;
 414                 pve->pve_start = entry->start;
 415                 pve->pve_end = entry->end - 1;
 416                 pve->pve_offset = entry->offset;
 417                 pve->pve_prot = entry->protection;
 418
 419                 /* Backing object's path needed? */
 420                 if (pve->pve_pathlen == 0)
 421                         break;
 422
 423                 pathlen = pve->pve_pathlen;
 424                 pve->pve_pathlen = 0;
 425
 426                 obj = entry->object.vm_object;
 427                 if (obj != NULL)
 428                         VM_OBJECT_LOCK(obj);
 429         } while (0);
 430
 431         vm_map_unlock_read(map);
 432         vmspace_free(vm);
 433
 434         pve->pve_fsid = VNOVAL;
 435         pve->pve_fileid = VNOVAL;
 436
 437         if (error == 0 && obj != NULL) {
 438                 lobj = obj;
 439                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
 440                         if (tobj != obj)
 441                                 VM_OBJECT_LOCK(tobj);
 442                         if (lobj != obj)
 443                                 VM_OBJECT_UNLOCK(lobj);
 444                         lobj = tobj;
 445                         pve->pve_offset += tobj->backing_object_offset;
 446                 }
 447                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 448                 if (vp != NULL)
 449                         vref(vp);
 450                 if (lobj != obj)
 451                         VM_OBJECT_UNLOCK(lobj);
 452                 VM_OBJECT_UNLOCK(obj);
 453
 454                 if (vp != NULL) {
 455                         freepath = NULL;
 456                         fullpath = NULL;
 457                         vn_fullpath(td, vp, &fullpath, &freepath);
 458                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 459                         vn_lock(vp, LK_SHARED | LK_RETRY);
 460                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 461                                 pve->pve_fileid = vattr.va_fileid;
 462                                 pve->pve_fsid = vattr.va_fsid;
 463                         }
 464                         vput(vp);
 465                         VFS_UNLOCK_GIANT(vfslocked);
 466
 467                         if (fullpath != NULL) {
 468                                 pve->pve_pathlen = strlen(fullpath) + 1;
 469                                 if (pve->pve_pathlen <= pathlen) {
 470                                         error = copyout(fullpath, pve->pve_path,
 471                                             pve->pve_pathlen);
 472                                 } else
 473                                         error = ENAMETOOLONG;
 474                         }
 475                         if (freepath != NULL)
 476                                 free(freepath, M_TEMP);
 477                 }
 478         }
 479
 480         return (error);
 481 }
 482
 483 #ifdef COMPAT_FREEBSD32
 484 static int
 485 ptrace_vm_entry32(struct thread *td, struct proc *p,
 486     struct ptrace_vm_entry32 *pve32)
 487 {
 488         struct ptrace_vm_entry pve;
 489         int error;
 490
 491         pve.pve_entry = pve32->pve_entry;
 492         pve.pve_pathlen = pve32->pve_pathlen;
 493         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 494
 495         error = ptrace_vm_entry(td, p, &pve);
 496         if (error == 0) {
 497                 pve32->pve_entry = pve.pve_entry;
 498                 pve32->pve_timestamp = pve.pve_timestamp;
 499                 pve32->pve_start = pve.pve_start;
 500                 pve32->pve_end = pve.pve_end;
 501                 pve32->pve_offset = pve.pve_offset;
 502                 pve32->pve_prot = pve.pve_prot;
 503                 pve32->pve_fileid = pve.pve_fileid;
 504                 pve32->pve_fsid = pve.pve_fsid;
 505         }
 506
 507         pve32->pve_pathlen = pve.pve_pathlen;
 508         return (error);
 509 }
 510
 511 static void
 512 ptrace_lwpinfo_to32(const struct ptrace_lwpinfo *pl,
 513     struct ptrace_lwpinfo32 *pl32)
 514 {
 515
 516         pl32->pl_lwpid = pl->pl_lwpid;
 517         pl32->pl_event = pl->pl_event;
 518         pl32->pl_flags = pl->pl_flags;
 519         pl32->pl_sigmask = pl->pl_sigmask;
 520         pl32->pl_siglist = pl->pl_siglist;
 521         siginfo_to_siginfo32(&pl->pl_siginfo, &pl32->pl_siginfo);
 522         strcpy(pl32->pl_tdname, pl->pl_tdname);
 523         pl32->pl_child_pid = pl->pl_child_pid;
 524 }
 525 #endif /* COMPAT_FREEBSD32 */
 526
 527 /*
 528  * Process debugging system call.
 529  */
 530 #ifndef _SYS_SYSPROTO_H_
 531 struct ptrace_args {
 532         int     req;
 533         pid_t   pid;
 534         caddr_t addr;
 535         int     data;
 536 };
 537 #endif
 538
 539 #ifdef COMPAT_FREEBSD32
 540 /*
 541  * This CPP subterfuge is to try and reduce the number of ifdefs in
 542  * the body of the code.
 543  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 544  * becomes either:
 545  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 546  * or
 547  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 548  * .. except this is done at runtime.
 549  */
 550 #define COPYIN(u, k, s)         wrap32 ? \
 551         copyin(u, k ## 32, s ## 32) : \
 552         copyin(u, k, s)
 553 #define COPYOUT(k, u, s)        wrap32 ? \
 554         copyout(k ## 32, u, s ## 32) : \
 555         copyout(k, u, s)
 556 #else
 557 #define COPYIN(u, k, s)         copyin(u, k, s)
 558 #define COPYOUT(k, u, s)        copyout(k, u, s)
 559 #endif
 560 int
 561 ptrace(struct thread *td, struct ptrace_args *uap)
 562 {
 563         /*
 564          * XXX this obfuscation is to reduce stack usage, but the register
 565          * structs may be too large to put on the stack anyway.
 566          */
 567         union {
 568                 struct ptrace_io_desc piod;
 569                 struct ptrace_lwpinfo pl;
 570                 struct ptrace_vm_entry pve;
 571                 struct dbreg dbreg;
 572                 struct fpreg fpreg;
 573                 struct reg reg;
 574 #ifdef COMPAT_FREEBSD32
 575                 struct dbreg32 dbreg32;
 576                 struct fpreg32 fpreg32;
 577                 struct reg32 reg32;
 578                 struct ptrace_io_desc32 piod32;
 579                 struct ptrace_lwpinfo32 pl32;
 580                 struct ptrace_vm_entry32 pve32;
 581 #endif
 582         } r;
 583         void *addr;
 584         int error = 0;
 585 #ifdef COMPAT_FREEBSD32
 586         int wrap32 = 0;
 587
 588         if (SV_CURPROC_FLAG(SV_ILP32))
 589                 wrap32 = 1;
 590 #endif
 591         AUDIT_ARG_PID(uap->pid);
 592         AUDIT_ARG_CMD(uap->req);
 593         AUDIT_ARG_VALUE(uap->data);
 594         addr = &r;
 595         switch (uap->req) {
 596         case PT_GETREGS:
 597         case PT_GETFPREGS:
 598         case PT_GETDBREGS:
 599         case PT_LWPINFO:
 600                 break;
 601         case PT_SETREGS:
 602                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 603                 break;
 604         case PT_SETFPREGS:
 605                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 606                 break;
 607         case PT_SETDBREGS:
 608                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 609                 break;
 610         case PT_IO:
 611                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 612                 break;
 613         case PT_VM_ENTRY:
 614                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 615                 break;
 616         default:
 617                 addr = uap->addr;
 618                 break;
 619         }
 620         if (error)
 621                 return (error);
 622
 623         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 624         if (error)
 625                 return (error);
 626
 627         switch (uap->req) {
 628         case PT_VM_ENTRY:
 629                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 630                 break;
 631         case PT_IO:
 632                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 633                 break;
 634         case PT_GETREGS:
 635                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 636                 break;
 637         case PT_GETFPREGS:
 638                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 639                 break;
 640         case PT_GETDBREGS:
 641                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 642                 break;
 643         case PT_LWPINFO:
 644                 error = copyout(&r.pl, uap->addr, uap->data);
 645                 break;
 646         }
 647
 648         return (error);
 649 }
 650 #undef COPYIN
 651 #undef COPYOUT
 652
 653 #ifdef COMPAT_FREEBSD32
 654 /*
 655  *   PROC_READ(regs, td2, addr);
 656  * becomes either:
 657  *   proc_read_regs(td2, addr);
 658  * or
 659  *   proc_read_regs32(td2, addr);
 660  * .. except this is done at runtime.  There is an additional
 661  * complication in that PROC_WRITE disallows 32 bit consumers
 662  * from writing to 64 bit address space targets.
 663  */
 664 #define PROC_READ(w, t, a)      wrap32 ? \
 665         proc_read_ ## w ## 32(t, a) : \
 666         proc_read_ ## w (t, a)
 667 #define PROC_WRITE(w, t, a)     wrap32 ? \
 668         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 669         proc_write_ ## w (t, a)
 670 #else
 671 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 672 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 673 #endif
 674
 675 int
 676 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 677 {
 678         struct iovec iov;
 679         struct uio uio;
 680         struct proc *curp, *p, *pp;
 681         struct thread *td2 = NULL;
 682         struct ptrace_io_desc *piod = NULL;
 683         struct ptrace_lwpinfo *pl;
 684         int error, write, tmp, num;
 685         int proctree_locked = 0;
 686         lwpid_t tid = 0, *buf;
 687 #ifdef COMPAT_FREEBSD32
 688         int wrap32 = 0, safe = 0;
 689         struct ptrace_io_desc32 *piod32 = NULL;
 690         struct ptrace_lwpinfo32 *pl32 = NULL;
 691         struct ptrace_lwpinfo plr;
 692 #endif
 693
 694         curp = td->td_proc;
 695
 696         /* Lock proctree before locking the process. */
 697         switch (req) {
 698         case PT_TRACE_ME:
 699         case PT_ATTACH:
 700         case PT_STEP:
 701         case PT_CONTINUE:
 702         case PT_TO_SCE:
 703         case PT_TO_SCX:
 704         case PT_SYSCALL:
 705         case PT_FOLLOW_FORK:
 706         case PT_DETACH:
 707                 sx_xlock(&proctree_lock);
 708                 proctree_locked = 1;
 709                 break;
 710         default:
 711                 break;
 712         }
 713
 714         write = 0;
 715         if (req == PT_TRACE_ME) {
 716                 p = td->td_proc;
 717                 PROC_LOCK(p);
 718         } else {
 719                 if (pid <= PID_MAX) {
 720                         if ((p = pfind(pid)) == NULL) {
 721                                 if (proctree_locked)
 722                                         sx_xunlock(&proctree_lock);
 723                                 return (ESRCH);
 724                         }
 725                 } else {
 726                         /* this is slow, should be optimized */
 727                         sx_slock(&allproc_lock);
 728                         FOREACH_PROC_IN_SYSTEM(p) {
 729                                 PROC_LOCK(p);
 730                                 FOREACH_THREAD_IN_PROC(p, td2) {
 731                                         if (td2->td_tid == pid)
 732                                                 break;
 733                                 }
 734                                 if (td2 != NULL)
 735                                         break; /* proc lock held */
 736                                 PROC_UNLOCK(p);
 737                         }
 738                         sx_sunlock(&allproc_lock);
 739                         if (p == NULL) {
 740                                 if (proctree_locked)
 741                                         sx_xunlock(&proctree_lock);
 742                                 return (ESRCH);
 743                         }
 744                         tid = pid;
 745                         pid = p->p_pid;
 746                 }
 747         }
 748         AUDIT_ARG_PROCESS(p);
 749
 750         if ((p->p_flag & P_WEXIT) != 0) {
 751                 error = ESRCH;
 752                 goto fail;
 753         }
 754         if ((error = p_cansee(td, p)) != 0)
 755                 goto fail;
 756
 757         if ((error = p_candebug(td, p)) != 0)
 758                 goto fail;
 759
 760         /*
 761          * System processes can't be debugged.
 762          */
 763         if ((p->p_flag & P_SYSTEM) != 0) {
 764                 error = EINVAL;
 765                 goto fail;
 766         }
 767
 768         if (tid == 0) {
 769                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 770                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 771                         td2 = p->p_xthread;
 772                 } else {
 773                         td2 = FIRST_THREAD_IN_PROC(p);
 774                 }
 775                 tid = td2->td_tid;
 776         }
 777
 778 #ifdef COMPAT_FREEBSD32
 779         /*
 780          * Test if we're a 32 bit client and what the target is.
 781          * Set the wrap controls accordingly.
 782          */
 783         if (SV_CURPROC_FLAG(SV_ILP32)) {
 784                 if (SV_PROC_FLAG(td2->td_proc, SV_ILP32))
 785                         safe = 1;
 786                 wrap32 = 1;
 787         }
 788 #endif
 789         /*
 790          * Permissions check
 791          */
 792         switch (req) {
 793         case PT_TRACE_ME:
 794                 /* Always legal. */
 795                 break;
 796
 797         case PT_ATTACH:
 798                 /* Self */
 799                 if (p->p_pid == td->td_proc->p_pid) {
 800                         error = EINVAL;
 801                         goto fail;
 802                 }
 803
 804                 /* Already traced */
 805                 if (p->p_flag & P_TRACED) {
 806                         error = EBUSY;
 807                         goto fail;
 808                 }
 809
 810                 /* Can't trace an ancestor if you're being traced. */
 811                 if (curp->p_flag & P_TRACED) {
 812                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 813                                 if (pp == p) {
 814                                         error = EINVAL;
 815                                         goto fail;
 816                                 }
 817                         }
 818                 }
 819
 820
 821                 /* OK */
 822                 break;
 823
 824         case PT_CLEARSTEP:
 825                 /* Allow thread to clear single step for itself */
 826                 if (td->td_tid == tid)
 827                         break;
 828
 829                 /* FALLTHROUGH */
 830         default:
 831                 /* not being traced... */
 832                 if ((p->p_flag & P_TRACED) == 0) {
 833                         error = EPERM;
 834                         goto fail;
 835                 }
 836
 837                 /* not being traced by YOU */
 838                 if (p->p_pptr != td->td_proc) {
 839                         error = EBUSY;
 840                         goto fail;
 841                 }
 842
 843                 /* not currently stopped */
 844                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 845                     p->p_suspcount != p->p_numthreads  ||
 846                     (p->p_flag & P_WAITED) == 0) {
 847                         error = EBUSY;
 848                         goto fail;
 849                 }
 850
 851                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 852                         static int count = 0;
 853                         if (count++ == 0)
 854                                 printf("P_STOPPED_TRACE not set.\n");
 855                 }
 856
 857                 /* OK */
 858                 break;
 859         }
 860
 861         /* Keep this process around until we finish this request. */
 862         _PHOLD(p);
 863
 864 #ifdef FIX_SSTEP
 865         /*
 866          * Single step fixup ala procfs
 867          */
 868         FIX_SSTEP(td2);
 869 #endif
 870
 871         /*
 872          * Actually do the requests
 873          */
 874
 875         td->td_retval[0] = 0;
 876
 877         switch (req) {
 878         case PT_TRACE_ME:
 879                 /* set my trace flag and "owner" so it can read/write me */
 880                 p->p_flag |= P_TRACED;
 881                 p->p_oppid = p->p_pptr->p_pid;
 882                 break;
 883
 884         case PT_ATTACH:
 885                 /* security check done above */
 886                 p->p_flag |= P_TRACED;
 887                 p->p_oppid = p->p_pptr->p_pid;
 888                 if (p->p_pptr != td->td_proc)
 889                         proc_reparent(p, td->td_proc);
 890                 data = SIGSTOP;
 891                 goto sendsig;   /* in PT_CONTINUE below */
 892
 893         case PT_CLEARSTEP:
 894                 error = ptrace_clear_single_step(td2);
 895                 break;
 896
 897         case PT_SETSTEP:
 898                 error = ptrace_single_step(td2);
 899                 break;
 900
 901         case PT_SUSPEND:
 902                 td2->td_dbgflags |= TDB_SUSPEND;
 903                 thread_lock(td2);
 904                 td2->td_flags |= TDF_NEEDSUSPCHK;
 905                 thread_unlock(td2);
 906                 break;
 907
 908         case PT_RESUME:
 909                 td2->td_dbgflags &= ~TDB_SUSPEND;
 910                 break;
 911
 912         case PT_FOLLOW_FORK:
 913                 if (data)
 914                         p->p_flag |= P_FOLLOWFORK;
 915                 else
 916                         p->p_flag &= ~P_FOLLOWFORK;
 917                 break;
 918
 919         case PT_STEP:
 920         case PT_CONTINUE:
 921         case PT_TO_SCE:
 922         case PT_TO_SCX:
 923         case PT_SYSCALL:
 924         case PT_DETACH:
 925                 /* Zero means do not send any signal */
 926                 if (data < 0 || data > _SIG_MAXSIG) {
 927                         error = EINVAL;
 928                         break;
 929                 }
 930
 931                 switch (req) {
 932                 case PT_STEP:
 933                         error = ptrace_single_step(td2);
 934                         if (error)
 935                                 goto out;
 936                         break;
 937                 case PT_CONTINUE:
 938                 case PT_TO_SCE:
 939                 case PT_TO_SCX:
 940                 case PT_SYSCALL:
 941                         if (addr != (void *)1) {
 942                                 error = ptrace_set_pc(td2,
 943                                     (u_long)(uintfptr_t)addr);
 944                                 if (error)
 945                                         goto out;
 946                         }
 947                         switch (req) {
 948                         case PT_TO_SCE:
 949                                 p->p_stops |= S_PT_SCE;
 950                                 break;
 951                         case PT_TO_SCX:
 952                                 p->p_stops |= S_PT_SCX;
 953                                 break;
 954                         case PT_SYSCALL:
 955                                 p->p_stops |= S_PT_SCE | S_PT_SCX;
 956                                 break;
 957                         }
 958                         break;
 959                 case PT_DETACH:
 960                         /* reset process parent */
 961                         if (p->p_oppid != p->p_pptr->p_pid) {
 962                                 struct proc *pp;
 963
 964                                 PROC_LOCK(p->p_pptr);
 965                                 sigqueue_take(p->p_ksi);
 966                                 PROC_UNLOCK(p->p_pptr);
 967
 968                                 PROC_UNLOCK(p);
 969                                 pp = pfind(p->p_oppid);
 970                                 if (pp == NULL)
 971                                         pp = initproc;
 972                                 else
 973                                         PROC_UNLOCK(pp);
 974                                 PROC_LOCK(p);
 975                                 proc_reparent(p, pp);
 976                                 if (pp == initproc)
 977                                         p->p_sigparent = SIGCHLD;
 978                         }
 979                         p->p_flag &= ~(P_TRACED | P_WAITED | P_FOLLOWFORK);
 980                         p->p_oppid = 0;
 981
 982                         /* should we send SIGCHLD? */
 983                         /* childproc_continued(p); */
 984                         break;
 985                 }
 986
 987         sendsig:
 988                 if (proctree_locked) {
 989                         sx_xunlock(&proctree_lock);
 990                         proctree_locked = 0;
 991                 }
 992                 p->p_xstat = data;
 993                 p->p_xthread = NULL;
 994                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 995                         /* deliver or queue signal */
 996                         td2->td_dbgflags &= ~TDB_XSIG;
 997                         td2->td_xsig = data;
 998
 999                         if (req == PT_DETACH) {
1000                                 struct thread *td3;
1001                                 FOREACH_THREAD_IN_PROC(p, td3) {
1002                                         td3->td_dbgflags &= ~TDB_SUSPEND;
1003                                 }
1004                         }
1005                         /*
1006                          * unsuspend all threads, to not let a thread run,
1007                          * you should use PT_SUSPEND to suspend it before
1008                          * continuing process.
1009                          */
1010                         PROC_SLOCK(p);
1011                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
1012                         thread_unsuspend(p);
1013                         PROC_SUNLOCK(p);
1014                 } else {
1015                         if (data)
1016                                 psignal(p, data);
1017                 }
1018                 break;
1019
1020         case PT_WRITE_I:
1021         case PT_WRITE_D:
1022                 td2->td_dbgflags |= TDB_USERWR;
1023                 write = 1;
1024                 /* FALLTHROUGH */
1025         case PT_READ_I:
1026         case PT_READ_D:
1027                 PROC_UNLOCK(p);
1028                 tmp = 0;
1029                 /* write = 0 set above */
1030                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
1031                 iov.iov_len = sizeof(int);
1032                 uio.uio_iov = &iov;
1033                 uio.uio_iovcnt = 1;
1034                 uio.uio_offset = (off_t)(uintptr_t)addr;
1035                 uio.uio_resid = sizeof(int);
1036                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
1037                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
1038                 uio.uio_td = td;
1039                 error = proc_rwmem(p, &uio);
1040                 if (uio.uio_resid != 0) {
1041                         /*
1042                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1043                          * so I think write() can bogusly return 0.
1044                          * XXX what happens for short writes?  We don't want
1045                          * to write partial data.
1046                          * XXX proc_rwmem() returns EPERM for other invalid
1047                          * addresses.  Convert this to EINVAL.  Does this
1048                          * clobber returns of EPERM for other reasons?
1049                          */
1050                         if (error == 0 || error == ENOSPC || error == EPERM)
1051                                 error = EINVAL; /* EOF */
1052                 }
1053                 if (!write)
1054                         td->td_retval[0] = tmp;
1055                 PROC_LOCK(p);
1056                 break;
1057
1058         case PT_IO:
1059 #ifdef COMPAT_FREEBSD32
1060                 if (wrap32) {
1061                         piod32 = addr;
1062                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1063                         iov.iov_len = piod32->piod_len;
1064                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1065                         uio.uio_resid = piod32->piod_len;
1066                 } else
1067 #endif
1068                 {
1069                         piod = addr;
1070                         iov.iov_base = piod->piod_addr;
1071                         iov.iov_len = piod->piod_len;
1072                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1073                         uio.uio_resid = piod->piod_len;
1074                 }
1075                 uio.uio_iov = &iov;
1076                 uio.uio_iovcnt = 1;
1077                 uio.uio_segflg = UIO_USERSPACE;
1078                 uio.uio_td = td;
1079 #ifdef COMPAT_FREEBSD32
1080                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1081 #else
1082                 tmp = piod->piod_op;
1083 #endif
1084                 switch (tmp) {
1085                 case PIOD_READ_D:
1086                 case PIOD_READ_I:
1087                         uio.uio_rw = UIO_READ;
1088                         break;
1089                 case PIOD_WRITE_D:
1090                 case PIOD_WRITE_I:
1091                         td2->td_dbgflags |= TDB_USERWR;
1092                         uio.uio_rw = UIO_WRITE;
1093                         break;
1094                 default:
1095                         error = EINVAL;
1096                         goto out;
1097                 }
1098                 PROC_UNLOCK(p);
1099                 error = proc_rwmem(p, &uio);
1100 #ifdef COMPAT_FREEBSD32
1101                 if (wrap32)
1102                         piod32->piod_len -= uio.uio_resid;
1103                 else
1104 #endif
1105                         piod->piod_len -= uio.uio_resid;
1106                 PROC_LOCK(p);
1107                 break;
1108
1109         case PT_KILL:
1110                 data = SIGKILL;
1111                 goto sendsig;   /* in PT_CONTINUE above */
1112
1113         case PT_SETREGS:
1114                 td2->td_dbgflags |= TDB_USERWR;
1115                 error = PROC_WRITE(regs, td2, addr);
1116                 break;
1117
1118         case PT_GETREGS:
1119                 error = PROC_READ(regs, td2, addr);
1120                 break;
1121
1122         case PT_SETFPREGS:
1123                 td2->td_dbgflags |= TDB_USERWR;
1124                 error = PROC_WRITE(fpregs, td2, addr);
1125                 break;
1126
1127         case PT_GETFPREGS:
1128                 error = PROC_READ(fpregs, td2, addr);
1129                 break;
1130
1131         case PT_SETDBREGS:
1132                 td2->td_dbgflags |= TDB_USERWR;
1133                 error = PROC_WRITE(dbregs, td2, addr);
1134                 break;
1135
1136         case PT_GETDBREGS:
1137                 error = PROC_READ(dbregs, td2, addr);
1138                 break;
1139
1140         case PT_LWPINFO:
1141                 if (data <= 0 ||
1142 #ifdef COMPAT_FREEBSD32
1143                     (!wrap32 && data > sizeof(*pl)) ||
1144                     (wrap32 && data > sizeof(*pl32))) {
1145 #else
1146                     data > sizeof(*pl)) {
1147 #endif
1148                         error = EINVAL;
1149                         break;
1150                 }
1151 #ifdef COMPAT_FREEBSD32
1152                 if (wrap32) {
1153                         pl = &plr;
1154                         pl32 = addr;
1155                 } else
1156 #endif
1157                 pl = addr;
1158                 pl->pl_lwpid = td2->td_tid;
1159                 pl->pl_event = PL_EVENT_NONE;
1160                 pl->pl_flags = 0;
1161                 if (td2->td_dbgflags & TDB_XSIG) {
1162                         pl->pl_event = PL_EVENT_SIGNAL;
1163                         if (td2->td_dbgksi.ksi_signo != 0 &&
1164 #ifdef COMPAT_FREEBSD32
1165                             ((!wrap32 && data >= offsetof(struct ptrace_lwpinfo,
1166                             pl_siginfo) + sizeof(pl->pl_siginfo)) ||
1167                             (wrap32 && data >= offsetof(struct ptrace_lwpinfo32,
1168                             pl_siginfo) + sizeof(struct siginfo32)))
1169 #else
1170                             data >= offsetof(struct ptrace_lwpinfo, pl_siginfo)
1171                             + sizeof(pl->pl_siginfo)
1172 #endif
1173                         ){
1174                                 pl->pl_flags |= PL_FLAG_SI;
1175                                 pl->pl_siginfo = td2->td_dbgksi.ksi_info;
1176                         }
1177                 }
1178                 if ((pl->pl_flags & PL_FLAG_SI) == 0)
1179                         bzero(&pl->pl_siginfo, sizeof(pl->pl_siginfo));
1180                 if (td2->td_dbgflags & TDB_SCE)
1181                         pl->pl_flags |= PL_FLAG_SCE;
1182                 else if (td2->td_dbgflags & TDB_SCX)
1183                         pl->pl_flags |= PL_FLAG_SCX;
1184                 if (td2->td_dbgflags & TDB_EXEC)
1185                         pl->pl_flags |= PL_FLAG_EXEC;
1186                 if (td2->td_dbgflags & TDB_FORK) {
1187                         pl->pl_flags |= PL_FLAG_FORKED;
1188                         pl->pl_child_pid = td2->td_dbg_forked;
1189                 }
1190                 pl->pl_sigmask = td2->td_sigmask;
1191                 pl->pl_siglist = td2->td_siglist;
1192                 strcpy(pl->pl_tdname, td2->td_name);
1193 #ifdef COMPAT_FREEBSD32
1194                 if (wrap32)
1195                         ptrace_lwpinfo_to32(pl, pl32);
1196 #endif
1197                 break;
1198
1199         case PT_GETNUMLWPS:
1200                 td->td_retval[0] = p->p_numthreads;
1201                 break;
1202
1203         case PT_GETLWPLIST:
1204                 if (data <= 0) {
1205                         error = EINVAL;
1206                         break;
1207                 }
1208                 num = imin(p->p_numthreads, data);
1209                 PROC_UNLOCK(p);
1210                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1211                 tmp = 0;
1212                 PROC_LOCK(p);
1213                 FOREACH_THREAD_IN_PROC(p, td2) {
1214                         if (tmp >= num)
1215                                 break;
1216                         buf[tmp++] = td2->td_tid;
1217                 }
1218                 PROC_UNLOCK(p);
1219                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1220                 free(buf, M_TEMP);
1221                 if (!error)
1222                         td->td_retval[0] = tmp;
1223                 PROC_LOCK(p);
1224                 break;
1225
1226         case PT_VM_TIMESTAMP:
1227                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1228                 break;
1229
1230         case PT_VM_ENTRY:
1231                 PROC_UNLOCK(p);
1232 #ifdef COMPAT_FREEBSD32
1233                 if (wrap32)
1234                         error = ptrace_vm_entry32(td, p, addr);
1235                 else
1236 #endif
1237                 error = ptrace_vm_entry(td, p, addr);
1238                 PROC_LOCK(p);
1239                 break;
1240
1241         default:
1242 #ifdef __HAVE_PTRACE_MACHDEP
1243                 if (req >= PT_FIRSTMACH) {
1244                         PROC_UNLOCK(p);
1245                         error = cpu_ptrace(td2, req, addr, data);
1246                         PROC_LOCK(p);
1247                 } else
1248 #endif
1249                         /* Unknown request. */
1250                         error = EINVAL;
1251                 break;
1252         }
1253
1254 out:
1255         /* Drop our hold on this process now that the request has completed. */
1256         _PRELE(p);
1257 fail:
1258         PROC_UNLOCK(p);
1259         if (proctree_locked)
1260                 sx_xunlock(&proctree_lock);
1261         return (error);
1262 }
1263 #undef PROC_READ
1264 #undef PROC_WRITE
1265
1266 /*
1267  * Stop a process because of a debugging event;
1268  * stay stopped until p->p_step is cleared
1269  * (cleared by PIOCCONT in procfs).
1270  */
1271 void
1272 stopevent(struct proc *p, unsigned int event, unsigned int val)
1273 {
1274
1275         PROC_LOCK_ASSERT(p, MA_OWNED);
1276         p->p_step = 1;
1277         do {
1278                 p->p_xstat = val;
1279                 p->p_xthread = NULL;
1280                 p->p_stype = event;     /* Which event caused the stop? */
1281                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1282                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1283         } while (p->p_step);
1284 }