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_IA32
  65 #include <sys/procfs.h>
  66 #include <machine/fpu.h>
  67 #include <compat/ia32/ia32_reg.h>
  68
  69 struct ptrace_io_desc32 {
  70         int             piod_op;
  71         u_int32_t       piod_offs;
  72         u_int32_t       piod_addr;
  73         u_int32_t       piod_len;
  74 };
  75
  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 #endif
  90
  91 /*
  92  * Functions implemented using PROC_ACTION():
  93  *
  94  * proc_read_regs(proc, regs)
  95  *      Get the current user-visible register set from the process
  96  *      and copy it into the regs structure (<machine/reg.h>).
  97  *      The process is stopped at the time read_regs is called.
  98  *
  99  * proc_write_regs(proc, regs)
 100  *      Update the current register set from the passed in regs
 101  *      structure.  Take care to avoid clobbering special CPU
 102  *      registers or privileged bits in the PSL.
 103  *      Depending on the architecture this may have fix-up work to do,
 104  *      especially if the IAR or PCW are modified.
 105  *      The process is stopped at the time write_regs is called.
 106  *
 107  * proc_read_fpregs, proc_write_fpregs
 108  *      deal with the floating point register set, otherwise as above.
 109  *
 110  * proc_read_dbregs, proc_write_dbregs
 111  *      deal with the processor debug register set, otherwise as above.
 112  *
 113  * proc_sstep(proc)
 114  *      Arrange for the process to trap after executing a single instruction.
 115  */
 116
 117 #define PROC_ACTION(action) do {                                        \
 118         int error;                                                      \
 119                                                                         \
 120         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 121         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
 122                 error = EIO;                                            \
 123         else                                                            \
 124                 error = (action);                                       \
 125         return (error);                                                 \
 126 } while(0)
 127
 128 int
 129 proc_read_regs(struct thread *td, struct reg *regs)
 130 {
 131
 132         PROC_ACTION(fill_regs(td, regs));
 133 }
 134
 135 int
 136 proc_write_regs(struct thread *td, struct reg *regs)
 137 {
 138
 139         PROC_ACTION(set_regs(td, regs));
 140 }
 141
 142 int
 143 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 144 {
 145
 146         PROC_ACTION(fill_dbregs(td, dbregs));
 147 }
 148
 149 int
 150 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 151 {
 152
 153         PROC_ACTION(set_dbregs(td, dbregs));
 154 }
 155
 156 /*
 157  * Ptrace doesn't support fpregs at all, and there are no security holes
 158  * or translations for fpregs, so we can just copy them.
 159  */
 160 int
 161 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 162 {
 163
 164         PROC_ACTION(fill_fpregs(td, fpregs));
 165 }
 166
 167 int
 168 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 169 {
 170
 171         PROC_ACTION(set_fpregs(td, fpregs));
 172 }
 173
 174 #ifdef COMPAT_IA32
 175 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 176 int
 177 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 178 {
 179
 180         PROC_ACTION(fill_regs32(td, regs32));
 181 }
 182
 183 int
 184 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 185 {
 186
 187         PROC_ACTION(set_regs32(td, regs32));
 188 }
 189
 190 int
 191 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 192 {
 193
 194         PROC_ACTION(fill_dbregs32(td, dbregs32));
 195 }
 196
 197 int
 198 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 199 {
 200
 201         PROC_ACTION(set_dbregs32(td, dbregs32));
 202 }
 203
 204 int
 205 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 206 {
 207
 208         PROC_ACTION(fill_fpregs32(td, fpregs32));
 209 }
 210
 211 int
 212 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 213 {
 214
 215         PROC_ACTION(set_fpregs32(td, fpregs32));
 216 }
 217 #endif
 218
 219 int
 220 proc_sstep(struct thread *td)
 221 {
 222
 223         PROC_ACTION(ptrace_single_step(td));
 224 }
 225
 226 int
 227 proc_rwmem(struct proc *p, struct uio *uio)
 228 {
 229         vm_map_t map;
 230         vm_object_t backing_object, object = NULL;
 231         vm_offset_t pageno = 0;         /* page number */
 232         vm_prot_t reqprot;
 233         int error, fault_flags, writing;
 234
 235         /*
 236          * Assert that someone has locked this vmspace.  (Should be
 237          * curthread but we can't assert that.)  This keeps the process
 238          * from exiting out from under us until this operation completes.
 239          */
 240         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 241             p, p->p_pid));
 242
 243         /*
 244          * The map we want...
 245          */
 246         map = &p->p_vmspace->vm_map;
 247
 248         writing = uio->uio_rw == UIO_WRITE;
 249         reqprot = writing ? (VM_PROT_WRITE | VM_PROT_OVERRIDE_WRITE) :
 250             VM_PROT_READ;
 251         fault_flags = writing ? VM_FAULT_DIRTY : VM_FAULT_NORMAL;
 252
 253         /*
 254          * Only map in one page at a time.  We don't have to, but it
 255          * makes things easier.  This way is trivial - right?
 256          */
 257         do {
 258                 vm_map_t tmap;
 259                 vm_offset_t uva;
 260                 int page_offset;                /* offset into page */
 261                 vm_map_entry_t out_entry;
 262                 vm_prot_t out_prot;
 263                 boolean_t wired;
 264                 vm_pindex_t pindex;
 265                 u_int len;
 266                 vm_page_t m;
 267
 268                 object = NULL;
 269
 270                 uva = (vm_offset_t)uio->uio_offset;
 271
 272                 /*
 273                  * Get the page number of this segment.
 274                  */
 275                 pageno = trunc_page(uva);
 276                 page_offset = uva - pageno;
 277
 278                 /*
 279                  * How many bytes to copy
 280                  */
 281                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 282
 283                 /*
 284                  * Fault the page on behalf of the process
 285                  */
 286                 error = vm_fault(map, pageno, reqprot, fault_flags);
 287                 if (error) {
 288                         if (error == KERN_RESOURCE_SHORTAGE)
 289                                 error = ENOMEM;
 290                         else
 291                                 error = EFAULT;
 292                         break;
 293                 }
 294
 295                 /*
 296                  * Now we need to get the page.  out_entry, out_prot, wired,
 297                  * and single_use aren't used.  One would think the vm code
 298                  * would be a *bit* nicer...  We use tmap because
 299                  * vm_map_lookup() can change the map argument.
 300                  */
 301                 tmap = map;
 302                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 303                     &object, &pindex, &out_prot, &wired);
 304                 if (error) {
 305                         error = EFAULT;
 306                         break;
 307                 }
 308                 VM_OBJECT_LOCK(object);
 309                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 310                     !writing &&
 311                     (backing_object = object->backing_object) != NULL) {
 312                         /*
 313                          * Allow fallback to backing objects if we are reading.
 314                          */
 315                         VM_OBJECT_LOCK(backing_object);
 316                         pindex += OFF_TO_IDX(object->backing_object_offset);
 317                         VM_OBJECT_UNLOCK(object);
 318                         object = backing_object;
 319                 }
 320                 VM_OBJECT_UNLOCK(object);
 321                 if (m == NULL) {
 322                         vm_map_lookup_done(tmap, out_entry);
 323                         error = EFAULT;
 324                         break;
 325                 }
 326
 327                 /*
 328                  * Hold the page in memory.
 329                  */
 330                 vm_page_lock_queues();
 331                 vm_page_hold(m);
 332                 vm_page_unlock_queues();
 333
 334                 /*
 335                  * We're done with tmap now.
 336                  */
 337                 vm_map_lookup_done(tmap, out_entry);
 338
 339                 /*
 340                  * Now do the i/o move.
 341                  */
 342                 error = uiomove_fromphys(&m, page_offset, len, uio);
 343
 344                 /* Make the I-cache coherent for breakpoints. */
 345                 if (!error && writing && (out_prot & VM_PROT_EXECUTE))
 346                         vm_sync_icache(map, uva, len);
 347
 348                 /*
 349                  * Release the page.
 350                  */
 351                 vm_page_lock_queues();
 352                 vm_page_unhold(m);
 353                 vm_page_unlock_queues();
 354
 355         } while (error == 0 && uio->uio_resid > 0);
 356
 357         return (error);
 358 }
 359
 360 static int
 361 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
 362 {
 363         struct vattr vattr;
 364         vm_map_t map;
 365         vm_map_entry_t entry;
 366         vm_object_t obj, tobj, lobj;
 367         struct vmspace *vm;
 368         struct vnode *vp;
 369         char *freepath, *fullpath;
 370         u_int pathlen;
 371         int error, index, vfslocked;
 372
 373         error = 0;
 374         obj = NULL;
 375
 376         vm = vmspace_acquire_ref(p);
 377         map = &vm->vm_map;
 378         vm_map_lock_read(map);
 379
 380         do {
 381                 entry = map->header.next;
 382                 index = 0;
 383                 while (index < pve->pve_entry && entry != &map->header) {
 384                         entry = entry->next;
 385                         index++;
 386                 }
 387                 if (index != pve->pve_entry) {
 388                         error = EINVAL;
 389                         break;
 390                 }
 391                 while (entry != &map->header &&
 392                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
 393                         entry = entry->next;
 394                         index++;
 395                 }
 396                 if (entry == &map->header) {
 397                         error = ENOENT;
 398                         break;
 399                 }
 400
 401                 /* We got an entry. */
 402                 pve->pve_entry = index + 1;
 403                 pve->pve_timestamp = map->timestamp;
 404                 pve->pve_start = entry->start;
 405                 pve->pve_end = entry->end - 1;
 406                 pve->pve_offset = entry->offset;
 407                 pve->pve_prot = entry->protection;
 408
 409                 /* Backing object's path needed? */
 410                 if (pve->pve_pathlen == 0)
 411                         break;
 412
 413                 pathlen = pve->pve_pathlen;
 414                 pve->pve_pathlen = 0;
 415
 416                 obj = entry->object.vm_object;
 417                 if (obj != NULL)
 418                         VM_OBJECT_LOCK(obj);
 419         } while (0);
 420
 421         vm_map_unlock_read(map);
 422         vmspace_free(vm);
 423
 424         pve->pve_fsid = VNOVAL;
 425         pve->pve_fileid = VNOVAL;
 426
 427         if (error == 0 && obj != NULL) {
 428                 lobj = obj;
 429                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
 430                         if (tobj != obj)
 431                                 VM_OBJECT_LOCK(tobj);
 432                         if (lobj != obj)
 433                                 VM_OBJECT_UNLOCK(lobj);
 434                         lobj = tobj;
 435                         pve->pve_offset += tobj->backing_object_offset;
 436                 }
 437                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 438                 if (vp != NULL)
 439                         vref(vp);
 440                 if (lobj != obj)
 441                         VM_OBJECT_UNLOCK(lobj);
 442                 VM_OBJECT_UNLOCK(obj);
 443
 444                 if (vp != NULL) {
 445                         freepath = NULL;
 446                         fullpath = NULL;
 447                         vn_fullpath(td, vp, &fullpath, &freepath);
 448                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 449                         vn_lock(vp, LK_SHARED | LK_RETRY);
 450                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 451                                 pve->pve_fileid = vattr.va_fileid;
 452                                 pve->pve_fsid = vattr.va_fsid;
 453                         }
 454                         vput(vp);
 455                         VFS_UNLOCK_GIANT(vfslocked);
 456
 457                         if (fullpath != NULL) {
 458                                 pve->pve_pathlen = strlen(fullpath) + 1;
 459                                 if (pve->pve_pathlen <= pathlen) {
 460                                         error = copyout(fullpath, pve->pve_path,
 461                                             pve->pve_pathlen);
 462                                 } else
 463                                         error = ENAMETOOLONG;
 464                         }
 465                         if (freepath != NULL)
 466                                 free(freepath, M_TEMP);
 467                 }
 468         }
 469
 470         return (error);
 471 }
 472
 473 #ifdef COMPAT_IA32
 474 static int
 475 ptrace_vm_entry32(struct thread *td, struct proc *p,
 476     struct ptrace_vm_entry32 *pve32)
 477 {
 478         struct ptrace_vm_entry pve;
 479         int error;
 480
 481         pve.pve_entry = pve32->pve_entry;
 482         pve.pve_pathlen = pve32->pve_pathlen;
 483         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 484
 485         error = ptrace_vm_entry(td, p, &pve);
 486         if (error == 0) {
 487                 pve32->pve_entry = pve.pve_entry;
 488                 pve32->pve_timestamp = pve.pve_timestamp;
 489                 pve32->pve_start = pve.pve_start;
 490                 pve32->pve_end = pve.pve_end;
 491                 pve32->pve_offset = pve.pve_offset;
 492                 pve32->pve_prot = pve.pve_prot;
 493                 pve32->pve_fileid = pve.pve_fileid;
 494                 pve32->pve_fsid = pve.pve_fsid;
 495         }
 496
 497         pve32->pve_pathlen = pve.pve_pathlen;
 498         return (error);
 499 }
 500 #endif /* COMPAT_IA32 */
 501
 502 /*
 503  * Process debugging system call.
 504  */
 505 #ifndef _SYS_SYSPROTO_H_
 506 struct ptrace_args {
 507         int     req;
 508         pid_t   pid;
 509         caddr_t addr;
 510         int     data;
 511 };
 512 #endif
 513
 514 #ifdef COMPAT_IA32
 515 /*
 516  * This CPP subterfuge is to try and reduce the number of ifdefs in
 517  * the body of the code.
 518  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 519  * becomes either:
 520  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 521  * or
 522  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 523  * .. except this is done at runtime.
 524  */
 525 #define COPYIN(u, k, s)         wrap32 ? \
 526         copyin(u, k ## 32, s ## 32) : \
 527         copyin(u, k, s)
 528 #define COPYOUT(k, u, s)        wrap32 ? \
 529         copyout(k ## 32, u, s ## 32) : \
 530         copyout(k, u, s)
 531 #else
 532 #define COPYIN(u, k, s)         copyin(u, k, s)
 533 #define COPYOUT(k, u, s)        copyout(k, u, s)
 534 #endif
 535 int
 536 ptrace(struct thread *td, struct ptrace_args *uap)
 537 {
 538         /*
 539          * XXX this obfuscation is to reduce stack usage, but the register
 540          * structs may be too large to put on the stack anyway.
 541          */
 542         union {
 543                 struct ptrace_io_desc piod;
 544                 struct ptrace_lwpinfo pl;
 545                 struct ptrace_vm_entry pve;
 546                 struct dbreg dbreg;
 547                 struct fpreg fpreg;
 548                 struct reg reg;
 549 #ifdef COMPAT_IA32
 550                 struct dbreg32 dbreg32;
 551                 struct fpreg32 fpreg32;
 552                 struct reg32 reg32;
 553                 struct ptrace_io_desc32 piod32;
 554                 struct ptrace_vm_entry32 pve32;
 555 #endif
 556         } r;
 557         void *addr;
 558         int error = 0;
 559 #ifdef COMPAT_IA32
 560         int wrap32 = 0;
 561
 562         if (SV_CURPROC_FLAG(SV_ILP32))
 563                 wrap32 = 1;
 564 #endif
 565         AUDIT_ARG_PID(uap->pid);
 566         AUDIT_ARG_CMD(uap->req);
 567         AUDIT_ARG_VALUE(uap->data);
 568         addr = &r;
 569         switch (uap->req) {
 570         case PT_GETREGS:
 571         case PT_GETFPREGS:
 572         case PT_GETDBREGS:
 573         case PT_LWPINFO:
 574                 break;
 575         case PT_SETREGS:
 576                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 577                 break;
 578         case PT_SETFPREGS:
 579                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 580                 break;
 581         case PT_SETDBREGS:
 582                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 583                 break;
 584         case PT_IO:
 585                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 586                 break;
 587         case PT_VM_ENTRY:
 588                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 589                 break;
 590         default:
 591                 addr = uap->addr;
 592                 break;
 593         }
 594         if (error)
 595                 return (error);
 596
 597         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 598         if (error)
 599                 return (error);
 600
 601         switch (uap->req) {
 602         case PT_VM_ENTRY:
 603                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 604                 break;
 605         case PT_IO:
 606                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 607                 break;
 608         case PT_GETREGS:
 609                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 610                 break;
 611         case PT_GETFPREGS:
 612                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 613                 break;
 614         case PT_GETDBREGS:
 615                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 616                 break;
 617         case PT_LWPINFO:
 618                 error = copyout(&r.pl, uap->addr, uap->data);
 619                 break;
 620         }
 621
 622         return (error);
 623 }
 624 #undef COPYIN
 625 #undef COPYOUT
 626
 627 #ifdef COMPAT_IA32
 628 /*
 629  *   PROC_READ(regs, td2, addr);
 630  * becomes either:
 631  *   proc_read_regs(td2, addr);
 632  * or
 633  *   proc_read_regs32(td2, addr);
 634  * .. except this is done at runtime.  There is an additional
 635  * complication in that PROC_WRITE disallows 32 bit consumers
 636  * from writing to 64 bit address space targets.
 637  */
 638 #define PROC_READ(w, t, a)      wrap32 ? \
 639         proc_read_ ## w ## 32(t, a) : \
 640         proc_read_ ## w (t, a)
 641 #define PROC_WRITE(w, t, a)     wrap32 ? \
 642         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 643         proc_write_ ## w (t, a)
 644 #else
 645 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 646 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 647 #endif
 648
 649 int
 650 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 651 {
 652         struct iovec iov;
 653         struct uio uio;
 654         struct proc *curp, *p, *pp;
 655         struct thread *td2 = NULL;
 656         struct ptrace_io_desc *piod = NULL;
 657         struct ptrace_lwpinfo *pl;
 658         int error, write, tmp, num;
 659         int proctree_locked = 0;
 660         lwpid_t tid = 0, *buf;
 661 #ifdef COMPAT_IA32
 662         int wrap32 = 0, safe = 0;
 663         struct ptrace_io_desc32 *piod32 = NULL;
 664 #endif
 665
 666         curp = td->td_proc;
 667
 668         /* Lock proctree before locking the process. */
 669         switch (req) {
 670         case PT_TRACE_ME:
 671         case PT_ATTACH:
 672         case PT_STEP:
 673         case PT_CONTINUE:
 674         case PT_TO_SCE:
 675         case PT_TO_SCX:
 676         case PT_SYSCALL:
 677         case PT_DETACH:
 678                 sx_xlock(&proctree_lock);
 679                 proctree_locked = 1;
 680                 break;
 681         default:
 682                 break;
 683         }
 684
 685         write = 0;
 686         if (req == PT_TRACE_ME) {
 687                 p = td->td_proc;
 688                 PROC_LOCK(p);
 689         } else {
 690                 if (pid <= PID_MAX) {
 691                         if ((p = pfind(pid)) == NULL) {
 692                                 if (proctree_locked)
 693                                         sx_xunlock(&proctree_lock);
 694                                 return (ESRCH);
 695                         }
 696                 } else {
 697                         /* this is slow, should be optimized */
 698                         sx_slock(&allproc_lock);
 699                         FOREACH_PROC_IN_SYSTEM(p) {
 700                                 PROC_LOCK(p);
 701                                 FOREACH_THREAD_IN_PROC(p, td2) {
 702                                         if (td2->td_tid == pid)
 703                                                 break;
 704                                 }
 705                                 if (td2 != NULL)
 706                                         break; /* proc lock held */
 707                                 PROC_UNLOCK(p);
 708                         }
 709                         sx_sunlock(&allproc_lock);
 710                         if (p == NULL) {
 711                                 if (proctree_locked)
 712                                         sx_xunlock(&proctree_lock);
 713                                 return (ESRCH);
 714                         }
 715                         tid = pid;
 716                         pid = p->p_pid;
 717                 }
 718         }
 719         AUDIT_ARG_PROCESS(p);
 720
 721         if ((p->p_flag & P_WEXIT) != 0) {
 722                 error = ESRCH;
 723                 goto fail;
 724         }
 725         if ((error = p_cansee(td, p)) != 0)
 726                 goto fail;
 727
 728         if ((error = p_candebug(td, p)) != 0)
 729                 goto fail;
 730
 731         /*
 732          * System processes can't be debugged.
 733          */
 734         if ((p->p_flag & P_SYSTEM) != 0) {
 735                 error = EINVAL;
 736                 goto fail;
 737         }
 738
 739         if (tid == 0) {
 740                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 741                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 742                         td2 = p->p_xthread;
 743                 } else {
 744                         td2 = FIRST_THREAD_IN_PROC(p);
 745                 }
 746                 tid = td2->td_tid;
 747         }
 748
 749 #ifdef COMPAT_IA32
 750         /*
 751          * Test if we're a 32 bit client and what the target is.
 752          * Set the wrap controls accordingly.
 753          */
 754         if (SV_CURPROC_FLAG(SV_ILP32)) {
 755                 if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
 756                         safe = 1;
 757                 wrap32 = 1;
 758         }
 759 #endif
 760         /*
 761          * Permissions check
 762          */
 763         switch (req) {
 764         case PT_TRACE_ME:
 765                 /* Always legal. */
 766                 break;
 767
 768         case PT_ATTACH:
 769                 /* Self */
 770                 if (p->p_pid == td->td_proc->p_pid) {
 771                         error = EINVAL;
 772                         goto fail;
 773                 }
 774
 775                 /* Already traced */
 776                 if (p->p_flag & P_TRACED) {
 777                         error = EBUSY;
 778                         goto fail;
 779                 }
 780
 781                 /* Can't trace an ancestor if you're being traced. */
 782                 if (curp->p_flag & P_TRACED) {
 783                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 784                                 if (pp == p) {
 785                                         error = EINVAL;
 786                                         goto fail;
 787                                 }
 788                         }
 789                 }
 790
 791
 792                 /* OK */
 793                 break;
 794
 795         case PT_CLEARSTEP:
 796                 /* Allow thread to clear single step for itself */
 797                 if (td->td_tid == tid)
 798                         break;
 799
 800                 /* FALLTHROUGH */
 801         default:
 802                 /* not being traced... */
 803                 if ((p->p_flag & P_TRACED) == 0) {
 804                         error = EPERM;
 805                         goto fail;
 806                 }
 807
 808                 /* not being traced by YOU */
 809                 if (p->p_pptr != td->td_proc) {
 810                         error = EBUSY;
 811                         goto fail;
 812                 }
 813
 814                 /* not currently stopped */
 815                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 816                     p->p_suspcount != p->p_numthreads  ||
 817                     (p->p_flag & P_WAITED) == 0) {
 818                         error = EBUSY;
 819                         goto fail;
 820                 }
 821
 822                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 823                         static int count = 0;
 824                         if (count++ == 0)
 825                                 printf("P_STOPPED_TRACE not set.\n");
 826                 }
 827
 828                 /* OK */
 829                 break;
 830         }
 831
 832         /* Keep this process around until we finish this request. */
 833         _PHOLD(p);
 834
 835 #ifdef FIX_SSTEP
 836         /*
 837          * Single step fixup ala procfs
 838          */
 839         FIX_SSTEP(td2);
 840 #endif
 841
 842         /*
 843          * Actually do the requests
 844          */
 845
 846         td->td_retval[0] = 0;
 847
 848         switch (req) {
 849         case PT_TRACE_ME:
 850                 /* set my trace flag and "owner" so it can read/write me */
 851                 p->p_flag |= P_TRACED;
 852                 p->p_oppid = p->p_pptr->p_pid;
 853                 break;
 854
 855         case PT_ATTACH:
 856                 /* security check done above */
 857                 p->p_flag |= P_TRACED;
 858                 p->p_oppid = p->p_pptr->p_pid;
 859                 if (p->p_pptr != td->td_proc)
 860                         proc_reparent(p, td->td_proc);
 861                 data = SIGSTOP;
 862                 goto sendsig;   /* in PT_CONTINUE below */
 863
 864         case PT_CLEARSTEP:
 865                 error = ptrace_clear_single_step(td2);
 866                 break;
 867
 868         case PT_SETSTEP:
 869                 error = ptrace_single_step(td2);
 870                 break;
 871
 872         case PT_SUSPEND:
 873                 td2->td_dbgflags |= TDB_SUSPEND;
 874                 thread_lock(td2);
 875                 td2->td_flags |= TDF_NEEDSUSPCHK;
 876                 thread_unlock(td2);
 877                 break;
 878
 879         case PT_RESUME:
 880                 td2->td_dbgflags &= ~TDB_SUSPEND;
 881                 break;
 882
 883         case PT_STEP:
 884         case PT_CONTINUE:
 885         case PT_TO_SCE:
 886         case PT_TO_SCX:
 887         case PT_SYSCALL:
 888         case PT_DETACH:
 889                 /* Zero means do not send any signal */
 890                 if (data < 0 || data > _SIG_MAXSIG) {
 891                         error = EINVAL;
 892                         break;
 893                 }
 894
 895                 switch (req) {
 896                 case PT_STEP:
 897                         error = ptrace_single_step(td2);
 898                         if (error)
 899                                 goto out;
 900                         break;
 901                 case PT_TO_SCE:
 902                         p->p_stops |= S_PT_SCE;
 903                         break;
 904                 case PT_TO_SCX:
 905                         p->p_stops |= S_PT_SCX;
 906                         break;
 907                 case PT_SYSCALL:
 908                         p->p_stops |= S_PT_SCE | S_PT_SCX;
 909                         break;
 910                 }
 911
 912                 if (addr != (void *)1) {
 913                         error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
 914                         if (error)
 915                                 break;
 916                 }
 917
 918                 if (req == PT_DETACH) {
 919                         /* reset process parent */
 920                         if (p->p_oppid != p->p_pptr->p_pid) {
 921                                 struct proc *pp;
 922
 923                                 PROC_LOCK(p->p_pptr);
 924                                 sigqueue_take(p->p_ksi);
 925                                 PROC_UNLOCK(p->p_pptr);
 926
 927                                 PROC_UNLOCK(p);
 928                                 pp = pfind(p->p_oppid);
 929                                 if (pp == NULL)
 930                                         pp = initproc;
 931                                 else
 932                                         PROC_UNLOCK(pp);
 933                                 PROC_LOCK(p);
 934                                 proc_reparent(p, pp);
 935                                 if (pp == initproc)
 936                                         p->p_sigparent = SIGCHLD;
 937                         }
 938                         p->p_flag &= ~(P_TRACED | P_WAITED);
 939                         p->p_oppid = 0;
 940
 941                         /* should we send SIGCHLD? */
 942                         /* childproc_continued(p); */
 943                 }
 944
 945         sendsig:
 946                 if (proctree_locked) {
 947                         sx_xunlock(&proctree_lock);
 948                         proctree_locked = 0;
 949                 }
 950                 p->p_xstat = data;
 951                 p->p_xthread = NULL;
 952                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 953                         /* deliver or queue signal */
 954                         td2->td_dbgflags &= ~TDB_XSIG;
 955                         td2->td_xsig = data;
 956
 957                         if (req == PT_DETACH) {
 958                                 struct thread *td3;
 959                                 FOREACH_THREAD_IN_PROC(p, td3) {
 960                                         td3->td_dbgflags &= ~TDB_SUSPEND;
 961                                 }
 962                         }
 963                         /*
 964                          * unsuspend all threads, to not let a thread run,
 965                          * you should use PT_SUSPEND to suspend it before
 966                          * continuing process.
 967                          */
 968                         PROC_SLOCK(p);
 969                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
 970                         thread_unsuspend(p);
 971                         PROC_SUNLOCK(p);
 972                 } else {
 973                         if (data)
 974                                 psignal(p, data);
 975                 }
 976                 break;
 977
 978         case PT_WRITE_I:
 979         case PT_WRITE_D:
 980                 td2->td_dbgflags |= TDB_USERWR;
 981                 write = 1;
 982                 /* FALLTHROUGH */
 983         case PT_READ_I:
 984         case PT_READ_D:
 985                 PROC_UNLOCK(p);
 986                 tmp = 0;
 987                 /* write = 0 set above */
 988                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
 989                 iov.iov_len = sizeof(int);
 990                 uio.uio_iov = &iov;
 991                 uio.uio_iovcnt = 1;
 992                 uio.uio_offset = (off_t)(uintptr_t)addr;
 993                 uio.uio_resid = sizeof(int);
 994                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
 995                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
 996                 uio.uio_td = td;
 997                 error = proc_rwmem(p, &uio);
 998                 if (uio.uio_resid != 0) {
 999                         /*
1000                          * XXX proc_rwmem() doesn't currently return ENOSPC,
1001                          * so I think write() can bogusly return 0.
1002                          * XXX what happens for short writes?  We don't want
1003                          * to write partial data.
1004                          * XXX proc_rwmem() returns EPERM for other invalid
1005                          * addresses.  Convert this to EINVAL.  Does this
1006                          * clobber returns of EPERM for other reasons?
1007                          */
1008                         if (error == 0 || error == ENOSPC || error == EPERM)
1009                                 error = EINVAL; /* EOF */
1010                 }
1011                 if (!write)
1012                         td->td_retval[0] = tmp;
1013                 PROC_LOCK(p);
1014                 break;
1015
1016         case PT_IO:
1017 #ifdef COMPAT_IA32
1018                 if (wrap32) {
1019                         piod32 = addr;
1020                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
1021                         iov.iov_len = piod32->piod_len;
1022                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
1023                         uio.uio_resid = piod32->piod_len;
1024                 } else
1025 #endif
1026                 {
1027                         piod = addr;
1028                         iov.iov_base = piod->piod_addr;
1029                         iov.iov_len = piod->piod_len;
1030                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
1031                         uio.uio_resid = piod->piod_len;
1032                 }
1033                 uio.uio_iov = &iov;
1034                 uio.uio_iovcnt = 1;
1035                 uio.uio_segflg = UIO_USERSPACE;
1036                 uio.uio_td = td;
1037 #ifdef COMPAT_IA32
1038                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
1039 #else
1040                 tmp = piod->piod_op;
1041 #endif
1042                 switch (tmp) {
1043                 case PIOD_READ_D:
1044                 case PIOD_READ_I:
1045                         uio.uio_rw = UIO_READ;
1046                         break;
1047                 case PIOD_WRITE_D:
1048                 case PIOD_WRITE_I:
1049                         td2->td_dbgflags |= TDB_USERWR;
1050                         uio.uio_rw = UIO_WRITE;
1051                         break;
1052                 default:
1053                         error = EINVAL;
1054                         goto out;
1055                 }
1056                 PROC_UNLOCK(p);
1057                 error = proc_rwmem(p, &uio);
1058 #ifdef COMPAT_IA32
1059                 if (wrap32)
1060                         piod32->piod_len -= uio.uio_resid;
1061                 else
1062 #endif
1063                         piod->piod_len -= uio.uio_resid;
1064                 PROC_LOCK(p);
1065                 break;
1066
1067         case PT_KILL:
1068                 data = SIGKILL;
1069                 goto sendsig;   /* in PT_CONTINUE above */
1070
1071         case PT_SETREGS:
1072                 td2->td_dbgflags |= TDB_USERWR;
1073                 error = PROC_WRITE(regs, td2, addr);
1074                 break;
1075
1076         case PT_GETREGS:
1077                 error = PROC_READ(regs, td2, addr);
1078                 break;
1079
1080         case PT_SETFPREGS:
1081                 td2->td_dbgflags |= TDB_USERWR;
1082                 error = PROC_WRITE(fpregs, td2, addr);
1083                 break;
1084
1085         case PT_GETFPREGS:
1086                 error = PROC_READ(fpregs, td2, addr);
1087                 break;
1088
1089         case PT_SETDBREGS:
1090                 td2->td_dbgflags |= TDB_USERWR;
1091                 error = PROC_WRITE(dbregs, td2, addr);
1092                 break;
1093
1094         case PT_GETDBREGS:
1095                 error = PROC_READ(dbregs, td2, addr);
1096                 break;
1097
1098         case PT_LWPINFO:
1099                 if (data <= 0 || data > sizeof(*pl)) {
1100                         error = EINVAL;
1101                         break;
1102                 }
1103                 pl = addr;
1104                 pl->pl_lwpid = td2->td_tid;
1105                 if (td2->td_dbgflags & TDB_XSIG)
1106                         pl->pl_event = PL_EVENT_SIGNAL;
1107                 else
1108                         pl->pl_event = 0;
1109                 pl->pl_flags = 0;
1110                 pl->pl_sigmask = td2->td_sigmask;
1111                 pl->pl_siglist = td2->td_siglist;
1112                 break;
1113
1114         case PT_GETNUMLWPS:
1115                 td->td_retval[0] = p->p_numthreads;
1116                 break;
1117
1118         case PT_GETLWPLIST:
1119                 if (data <= 0) {
1120                         error = EINVAL;
1121                         break;
1122                 }
1123                 num = imin(p->p_numthreads, data);
1124                 PROC_UNLOCK(p);
1125                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1126                 tmp = 0;
1127                 PROC_LOCK(p);
1128                 FOREACH_THREAD_IN_PROC(p, td2) {
1129                         if (tmp >= num)
1130                                 break;
1131                         buf[tmp++] = td2->td_tid;
1132                 }
1133                 PROC_UNLOCK(p);
1134                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1135                 free(buf, M_TEMP);
1136                 if (!error)
1137                         td->td_retval[0] = tmp;
1138                 PROC_LOCK(p);
1139                 break;
1140
1141         case PT_VM_TIMESTAMP:
1142                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1143                 break;
1144
1145         case PT_VM_ENTRY:
1146                 PROC_UNLOCK(p);
1147 #ifdef COMPAT_IA32
1148                 if (wrap32)
1149                         error = ptrace_vm_entry32(td, p, addr);
1150                 else
1151 #endif
1152                 error = ptrace_vm_entry(td, p, addr);
1153                 PROC_LOCK(p);
1154                 break;
1155
1156         default:
1157 #ifdef __HAVE_PTRACE_MACHDEP
1158                 if (req >= PT_FIRSTMACH) {
1159                         PROC_UNLOCK(p);
1160                         error = cpu_ptrace(td2, req, addr, data);
1161                         PROC_LOCK(p);
1162                 } else
1163 #endif
1164                         /* Unknown request. */
1165                         error = EINVAL;
1166                 break;
1167         }
1168
1169 out:
1170         /* Drop our hold on this process now that the request has completed. */
1171         _PRELE(p);
1172 fail:
1173         PROC_UNLOCK(p);
1174         if (proctree_locked)
1175                 sx_xunlock(&proctree_lock);
1176         return (error);
1177 }
1178 #undef PROC_READ
1179 #undef PROC_WRITE
1180
1181 /*
1182  * Stop a process because of a debugging event;
1183  * stay stopped until p->p_step is cleared
1184  * (cleared by PIOCCONT in procfs).
1185  */
1186 void
1187 stopevent(struct proc *p, unsigned int event, unsigned int val)
1188 {
1189
1190         PROC_LOCK_ASSERT(p, MA_OWNED);
1191         p->p_step = 1;
1192         do {
1193                 p->p_xstat = val;
1194                 p->p_xthread = NULL;
1195                 p->p_stype = event;     /* Which event caused the stop? */
1196                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1197                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1198         } while (p->p_step);
1199 }