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