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                 /*
 345                  * Release the page.
 346                  */
 347                 vm_page_lock_queues();
 348                 vm_page_unhold(m);
 349                 vm_page_unlock_queues();
 350
 351         } while (error == 0 && uio->uio_resid > 0);
 352
 353         return (error);
 354 }
 355
 356 static int
 357 ptrace_vm_entry(struct thread *td, struct proc *p, struct ptrace_vm_entry *pve)
 358 {
 359         struct vattr vattr;
 360         vm_map_t map;
 361         vm_map_entry_t entry;
 362         vm_object_t obj, tobj, lobj;
 363         struct vmspace *vm;
 364         struct vnode *vp;
 365         char *freepath, *fullpath;
 366         u_int pathlen;
 367         int error, index, vfslocked;
 368
 369         error = 0;
 370         obj = NULL;
 371
 372         vm = vmspace_acquire_ref(p);
 373         map = &vm->vm_map;
 374         vm_map_lock_read(map);
 375
 376         do {
 377                 entry = map->header.next;
 378                 index = 0;
 379                 while (index < pve->pve_entry && entry != &map->header) {
 380                         entry = entry->next;
 381                         index++;
 382                 }
 383                 if (index != pve->pve_entry) {
 384                         error = EINVAL;
 385                         break;
 386                 }
 387                 while (entry != &map->header &&
 388                     (entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
 389                         entry = entry->next;
 390                         index++;
 391                 }
 392                 if (entry == &map->header) {
 393                         error = ENOENT;
 394                         break;
 395                 }
 396
 397                 /* We got an entry. */
 398                 pve->pve_entry = index + 1;
 399                 pve->pve_timestamp = map->timestamp;
 400                 pve->pve_start = entry->start;
 401                 pve->pve_end = entry->end - 1;
 402                 pve->pve_offset = entry->offset;
 403                 pve->pve_prot = entry->protection;
 404
 405                 /* Backing object's path needed? */
 406                 if (pve->pve_pathlen == 0)
 407                         break;
 408
 409                 pathlen = pve->pve_pathlen;
 410                 pve->pve_pathlen = 0;
 411
 412                 obj = entry->object.vm_object;
 413                 if (obj != NULL)
 414                         VM_OBJECT_LOCK(obj);
 415         } while (0);
 416
 417         vm_map_unlock_read(map);
 418         vmspace_free(vm);
 419
 420         pve->pve_fsid = VNOVAL;
 421         pve->pve_fileid = VNOVAL;
 422
 423         if (error == 0 && obj != NULL) {
 424                 lobj = obj;
 425                 for (tobj = obj; tobj != NULL; tobj = tobj->backing_object) {
 426                         if (tobj != obj)
 427                                 VM_OBJECT_LOCK(tobj);
 428                         if (lobj != obj)
 429                                 VM_OBJECT_UNLOCK(lobj);
 430                         lobj = tobj;
 431                         pve->pve_offset += tobj->backing_object_offset;
 432                 }
 433                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 434                 if (vp != NULL)
 435                         vref(vp);
 436                 if (lobj != obj)
 437                         VM_OBJECT_UNLOCK(lobj);
 438                 VM_OBJECT_UNLOCK(obj);
 439
 440                 if (vp != NULL) {
 441                         freepath = NULL;
 442                         fullpath = NULL;
 443                         vn_fullpath(td, vp, &fullpath, &freepath);
 444                         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 445                         vn_lock(vp, LK_SHARED | LK_RETRY);
 446                         if (VOP_GETATTR(vp, &vattr, td->td_ucred) == 0) {
 447                                 pve->pve_fileid = vattr.va_fileid;
 448                                 pve->pve_fsid = vattr.va_fsid;
 449                         }
 450                         vput(vp);
 451                         VFS_UNLOCK_GIANT(vfslocked);
 452
 453                         if (fullpath != NULL) {
 454                                 pve->pve_pathlen = strlen(fullpath) + 1;
 455                                 if (pve->pve_pathlen <= pathlen) {
 456                                         error = copyout(fullpath, pve->pve_path,
 457                                             pve->pve_pathlen);
 458                                 } else
 459                                         error = ENAMETOOLONG;
 460                         }
 461                         if (freepath != NULL)
 462                                 free(freepath, M_TEMP);
 463                 }
 464         }
 465
 466         return (error);
 467 }
 468
 469 #ifdef COMPAT_IA32
 470 static int
 471 ptrace_vm_entry32(struct thread *td, struct proc *p,
 472     struct ptrace_vm_entry32 *pve32)
 473 {
 474         struct ptrace_vm_entry pve;
 475         int error;
 476
 477         pve.pve_entry = pve32->pve_entry;
 478         pve.pve_pathlen = pve32->pve_pathlen;
 479         pve.pve_path = (void *)(uintptr_t)pve32->pve_path;
 480
 481         error = ptrace_vm_entry(td, p, &pve);
 482         if (error == 0) {
 483                 pve32->pve_entry = pve.pve_entry;
 484                 pve32->pve_timestamp = pve.pve_timestamp;
 485                 pve32->pve_start = pve.pve_start;
 486                 pve32->pve_end = pve.pve_end;
 487                 pve32->pve_offset = pve.pve_offset;
 488                 pve32->pve_prot = pve.pve_prot;
 489                 pve32->pve_fileid = pve.pve_fileid;
 490                 pve32->pve_fsid = pve.pve_fsid;
 491         }
 492
 493         pve32->pve_pathlen = pve.pve_pathlen;
 494         return (error);
 495 }
 496 #endif /* COMPAT_IA32 */
 497
 498 /*
 499  * Process debugging system call.
 500  */
 501 #ifndef _SYS_SYSPROTO_H_
 502 struct ptrace_args {
 503         int     req;
 504         pid_t   pid;
 505         caddr_t addr;
 506         int     data;
 507 };
 508 #endif
 509
 510 #ifdef COMPAT_IA32
 511 /*
 512  * This CPP subterfuge is to try and reduce the number of ifdefs in
 513  * the body of the code.
 514  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 515  * becomes either:
 516  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 517  * or
 518  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 519  * .. except this is done at runtime.
 520  */
 521 #define COPYIN(u, k, s)         wrap32 ? \
 522         copyin(u, k ## 32, s ## 32) : \
 523         copyin(u, k, s)
 524 #define COPYOUT(k, u, s)        wrap32 ? \
 525         copyout(k ## 32, u, s ## 32) : \
 526         copyout(k, u, s)
 527 #else
 528 #define COPYIN(u, k, s)         copyin(u, k, s)
 529 #define COPYOUT(k, u, s)        copyout(k, u, s)
 530 #endif
 531 int
 532 ptrace(struct thread *td, struct ptrace_args *uap)
 533 {
 534         /*
 535          * XXX this obfuscation is to reduce stack usage, but the register
 536          * structs may be too large to put on the stack anyway.
 537          */
 538         union {
 539                 struct ptrace_io_desc piod;
 540                 struct ptrace_lwpinfo pl;
 541                 struct ptrace_vm_entry pve;
 542                 struct dbreg dbreg;
 543                 struct fpreg fpreg;
 544                 struct reg reg;
 545 #ifdef COMPAT_IA32
 546                 struct dbreg32 dbreg32;
 547                 struct fpreg32 fpreg32;
 548                 struct reg32 reg32;
 549                 struct ptrace_io_desc32 piod32;
 550                 struct ptrace_vm_entry32 pve32;
 551 #endif
 552         } r;
 553         void *addr;
 554         int error = 0;
 555 #ifdef COMPAT_IA32
 556         int wrap32 = 0;
 557
 558         if (SV_CURPROC_FLAG(SV_ILP32))
 559                 wrap32 = 1;
 560 #endif
 561         AUDIT_ARG_PID(uap->pid);
 562         AUDIT_ARG_CMD(uap->req);
 563         AUDIT_ARG_VALUE(uap->data);
 564         addr = &r;
 565         switch (uap->req) {
 566         case PT_GETREGS:
 567         case PT_GETFPREGS:
 568         case PT_GETDBREGS:
 569         case PT_LWPINFO:
 570                 break;
 571         case PT_SETREGS:
 572                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 573                 break;
 574         case PT_SETFPREGS:
 575                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 576                 break;
 577         case PT_SETDBREGS:
 578                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 579                 break;
 580         case PT_IO:
 581                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 582                 break;
 583         case PT_VM_ENTRY:
 584                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 585                 break;
 586         default:
 587                 addr = uap->addr;
 588                 break;
 589         }
 590         if (error)
 591                 return (error);
 592
 593         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 594         if (error)
 595                 return (error);
 596
 597         switch (uap->req) {
 598         case PT_VM_ENTRY:
 599                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 600                 break;
 601         case PT_IO:
 602                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 603                 break;
 604         case PT_GETREGS:
 605                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 606                 break;
 607         case PT_GETFPREGS:
 608                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 609                 break;
 610         case PT_GETDBREGS:
 611                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 612                 break;
 613         case PT_LWPINFO:
 614                 error = copyout(&r.pl, uap->addr, uap->data);
 615                 break;
 616         }
 617
 618         return (error);
 619 }
 620 #undef COPYIN
 621 #undef COPYOUT
 622
 623 #ifdef COMPAT_IA32
 624 /*
 625  *   PROC_READ(regs, td2, addr);
 626  * becomes either:
 627  *   proc_read_regs(td2, addr);
 628  * or
 629  *   proc_read_regs32(td2, addr);
 630  * .. except this is done at runtime.  There is an additional
 631  * complication in that PROC_WRITE disallows 32 bit consumers
 632  * from writing to 64 bit address space targets.
 633  */
 634 #define PROC_READ(w, t, a)      wrap32 ? \
 635         proc_read_ ## w ## 32(t, a) : \
 636         proc_read_ ## w (t, a)
 637 #define PROC_WRITE(w, t, a)     wrap32 ? \
 638         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 639         proc_write_ ## w (t, a)
 640 #else
 641 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 642 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 643 #endif
 644
 645 int
 646 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 647 {
 648         struct iovec iov;
 649         struct uio uio;
 650         struct proc *curp, *p, *pp;
 651         struct thread *td2 = NULL;
 652         struct ptrace_io_desc *piod = NULL;
 653         struct ptrace_lwpinfo *pl;
 654         int error, write, tmp, num;
 655         int proctree_locked = 0;
 656         lwpid_t tid = 0, *buf;
 657 #ifdef COMPAT_IA32
 658         int wrap32 = 0, safe = 0;
 659         struct ptrace_io_desc32 *piod32 = NULL;
 660 #endif
 661
 662         curp = td->td_proc;
 663
 664         /* Lock proctree before locking the process. */
 665         switch (req) {
 666         case PT_TRACE_ME:
 667         case PT_ATTACH:
 668         case PT_STEP:
 669         case PT_CONTINUE:
 670         case PT_TO_SCE:
 671         case PT_TO_SCX:
 672         case PT_SYSCALL:
 673         case PT_DETACH:
 674                 sx_xlock(&proctree_lock);
 675                 proctree_locked = 1;
 676                 break;
 677         default:
 678                 break;
 679         }
 680
 681         write = 0;
 682         if (req == PT_TRACE_ME) {
 683                 p = td->td_proc;
 684                 PROC_LOCK(p);
 685         } else {
 686                 if (pid <= PID_MAX) {
 687                         if ((p = pfind(pid)) == NULL) {
 688                                 if (proctree_locked)
 689                                         sx_xunlock(&proctree_lock);
 690                                 return (ESRCH);
 691                         }
 692                 } else {
 693                         /* this is slow, should be optimized */
 694                         sx_slock(&allproc_lock);
 695                         FOREACH_PROC_IN_SYSTEM(p) {
 696                                 PROC_LOCK(p);
 697                                 FOREACH_THREAD_IN_PROC(p, td2) {
 698                                         if (td2->td_tid == pid)
 699                                                 break;
 700                                 }
 701                                 if (td2 != NULL)
 702                                         break; /* proc lock held */
 703                                 PROC_UNLOCK(p);
 704                         }
 705                         sx_sunlock(&allproc_lock);
 706                         if (p == NULL) {
 707                                 if (proctree_locked)
 708                                         sx_xunlock(&proctree_lock);
 709                                 return (ESRCH);
 710                         }
 711                         tid = pid;
 712                         pid = p->p_pid;
 713                 }
 714         }
 715         AUDIT_ARG_PROCESS(p);
 716
 717         if ((p->p_flag & P_WEXIT) != 0) {
 718                 error = ESRCH;
 719                 goto fail;
 720         }
 721         if ((error = p_cansee(td, p)) != 0)
 722                 goto fail;
 723
 724         if ((error = p_candebug(td, p)) != 0)
 725                 goto fail;
 726
 727         /*
 728          * System processes can't be debugged.
 729          */
 730         if ((p->p_flag & P_SYSTEM) != 0) {
 731                 error = EINVAL;
 732                 goto fail;
 733         }
 734
 735         if (tid == 0) {
 736                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 737                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 738                         td2 = p->p_xthread;
 739                 } else {
 740                         td2 = FIRST_THREAD_IN_PROC(p);
 741                 }
 742                 tid = td2->td_tid;
 743         }
 744
 745 #ifdef COMPAT_IA32
 746         /*
 747          * Test if we're a 32 bit client and what the target is.
 748          * Set the wrap controls accordingly.
 749          */
 750         if (SV_CURPROC_FLAG(SV_ILP32)) {
 751                 if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
 752                         safe = 1;
 753                 wrap32 = 1;
 754         }
 755 #endif
 756         /*
 757          * Permissions check
 758          */
 759         switch (req) {
 760         case PT_TRACE_ME:
 761                 /* Always legal. */
 762                 break;
 763
 764         case PT_ATTACH:
 765                 /* Self */
 766                 if (p->p_pid == td->td_proc->p_pid) {
 767                         error = EINVAL;
 768                         goto fail;
 769                 }
 770
 771                 /* Already traced */
 772                 if (p->p_flag & P_TRACED) {
 773                         error = EBUSY;
 774                         goto fail;
 775                 }
 776
 777                 /* Can't trace an ancestor if you're being traced. */
 778                 if (curp->p_flag & P_TRACED) {
 779                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 780                                 if (pp == p) {
 781                                         error = EINVAL;
 782                                         goto fail;
 783                                 }
 784                         }
 785                 }
 786
 787
 788                 /* OK */
 789                 break;
 790
 791         case PT_CLEARSTEP:
 792                 /* Allow thread to clear single step for itself */
 793                 if (td->td_tid == tid)
 794                         break;
 795
 796                 /* FALLTHROUGH */
 797         default:
 798                 /* not being traced... */
 799                 if ((p->p_flag & P_TRACED) == 0) {
 800                         error = EPERM;
 801                         goto fail;
 802                 }
 803
 804                 /* not being traced by YOU */
 805                 if (p->p_pptr != td->td_proc) {
 806                         error = EBUSY;
 807                         goto fail;
 808                 }
 809
 810                 /* not currently stopped */
 811                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 812                     p->p_suspcount != p->p_numthreads  ||
 813                     (p->p_flag & P_WAITED) == 0) {
 814                         error = EBUSY;
 815                         goto fail;
 816                 }
 817
 818                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 819                         static int count = 0;
 820                         if (count++ == 0)
 821                                 printf("P_STOPPED_TRACE not set.\n");
 822                 }
 823
 824                 /* OK */
 825                 break;
 826         }
 827
 828         /* Keep this process around until we finish this request. */
 829         _PHOLD(p);
 830
 831 #ifdef FIX_SSTEP
 832         /*
 833          * Single step fixup ala procfs
 834          */
 835         FIX_SSTEP(td2);
 836 #endif
 837
 838         /*
 839          * Actually do the requests
 840          */
 841
 842         td->td_retval[0] = 0;
 843
 844         switch (req) {
 845         case PT_TRACE_ME:
 846                 /* set my trace flag and "owner" so it can read/write me */
 847                 p->p_flag |= P_TRACED;
 848                 p->p_oppid = p->p_pptr->p_pid;
 849                 break;
 850
 851         case PT_ATTACH:
 852                 /* security check done above */
 853                 p->p_flag |= P_TRACED;
 854                 p->p_oppid = p->p_pptr->p_pid;
 855                 if (p->p_pptr != td->td_proc)
 856                         proc_reparent(p, td->td_proc);
 857                 data = SIGSTOP;
 858                 goto sendsig;   /* in PT_CONTINUE below */
 859
 860         case PT_CLEARSTEP:
 861                 error = ptrace_clear_single_step(td2);
 862                 break;
 863
 864         case PT_SETSTEP:
 865                 error = ptrace_single_step(td2);
 866                 break;
 867
 868         case PT_SUSPEND:
 869                 td2->td_dbgflags |= TDB_SUSPEND;
 870                 thread_lock(td2);
 871                 td2->td_flags |= TDF_NEEDSUSPCHK;
 872                 thread_unlock(td2);
 873                 break;
 874
 875         case PT_RESUME:
 876                 td2->td_dbgflags &= ~TDB_SUSPEND;
 877                 break;
 878
 879         case PT_STEP:
 880         case PT_CONTINUE:
 881         case PT_TO_SCE:
 882         case PT_TO_SCX:
 883         case PT_SYSCALL:
 884         case PT_DETACH:
 885                 /* Zero means do not send any signal */
 886                 if (data < 0 || data > _SIG_MAXSIG) {
 887                         error = EINVAL;
 888                         break;
 889                 }
 890
 891                 switch (req) {
 892                 case PT_STEP:
 893                         error = ptrace_single_step(td2);
 894                         if (error)
 895                                 goto out;
 896                         break;
 897                 case PT_TO_SCE:
 898                         p->p_stops |= S_PT_SCE;
 899                         break;
 900                 case PT_TO_SCX:
 901                         p->p_stops |= S_PT_SCX;
 902                         break;
 903                 case PT_SYSCALL:
 904                         p->p_stops |= S_PT_SCE | S_PT_SCX;
 905                         break;
 906                 }
 907
 908                 if (addr != (void *)1) {
 909                         error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
 910                         if (error)
 911                                 break;
 912                 }
 913
 914                 if (req == PT_DETACH) {
 915                         /* reset process parent */
 916                         if (p->p_oppid != p->p_pptr->p_pid) {
 917                                 struct proc *pp;
 918
 919                                 PROC_LOCK(p->p_pptr);
 920                                 sigqueue_take(p->p_ksi);
 921                                 PROC_UNLOCK(p->p_pptr);
 922
 923                                 PROC_UNLOCK(p);
 924                                 pp = pfind(p->p_oppid);
 925                                 if (pp == NULL)
 926                                         pp = initproc;
 927                                 else
 928                                         PROC_UNLOCK(pp);
 929                                 PROC_LOCK(p);
 930                                 proc_reparent(p, pp);
 931                                 if (pp == initproc)
 932                                         p->p_sigparent = SIGCHLD;
 933                         }
 934                         p->p_flag &= ~(P_TRACED | P_WAITED);
 935                         p->p_oppid = 0;
 936
 937                         /* should we send SIGCHLD? */
 938                         /* childproc_continued(p); */
 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                                 struct thread *td3;
 955                                 FOREACH_THREAD_IN_PROC(p, td3) {
 956                                         td3->td_dbgflags &= ~TDB_SUSPEND;
 957                                 }
 958                         }
 959                         /*
 960                          * unsuspend all threads, to not let a thread run,
 961                          * you should use PT_SUSPEND to suspend it before
 962                          * continuing process.
 963                          */
 964                         PROC_SLOCK(p);
 965                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
 966                         thread_unsuspend(p);
 967                         PROC_SUNLOCK(p);
 968                 } else {
 969                         if (data)
 970                                 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_IA32
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_IA32
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_IA32
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 || data > sizeof(*pl)) {
1096                         error = EINVAL;
1097                         break;
1098                 }
1099                 pl = addr;
1100                 pl->pl_lwpid = td2->td_tid;
1101                 if (td2->td_dbgflags & TDB_XSIG)
1102                         pl->pl_event = PL_EVENT_SIGNAL;
1103                 else
1104                         pl->pl_event = 0;
1105                 pl->pl_flags = 0;
1106                 pl->pl_sigmask = td2->td_sigmask;
1107                 pl->pl_siglist = td2->td_siglist;
1108                 break;
1109
1110         case PT_GETNUMLWPS:
1111                 td->td_retval[0] = p->p_numthreads;
1112                 break;
1113
1114         case PT_GETLWPLIST:
1115                 if (data <= 0) {
1116                         error = EINVAL;
1117                         break;
1118                 }
1119                 num = imin(p->p_numthreads, data);
1120                 PROC_UNLOCK(p);
1121                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1122                 tmp = 0;
1123                 PROC_LOCK(p);
1124                 FOREACH_THREAD_IN_PROC(p, td2) {
1125                         if (tmp >= num)
1126                                 break;
1127                         buf[tmp++] = td2->td_tid;
1128                 }
1129                 PROC_UNLOCK(p);
1130                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1131                 free(buf, M_TEMP);
1132                 if (!error)
1133                         td->td_retval[0] = tmp;
1134                 PROC_LOCK(p);
1135                 break;
1136
1137         case PT_VM_TIMESTAMP:
1138                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1139                 break;
1140
1141         case PT_VM_ENTRY:
1142                 PROC_UNLOCK(p);
1143 #ifdef COMPAT_IA32
1144                 if (wrap32)
1145                         error = ptrace_vm_entry32(td, p, addr);
1146                 else
1147 #endif
1148                 error = ptrace_vm_entry(td, p, addr);
1149                 PROC_LOCK(p);
1150                 break;
1151
1152         default:
1153 #ifdef __HAVE_PTRACE_MACHDEP
1154                 if (req >= PT_FIRSTMACH) {
1155                         PROC_UNLOCK(p);
1156                         error = cpu_ptrace(td2, req, addr, data);
1157                         PROC_LOCK(p);
1158                 } else
1159 #endif
1160                         /* Unknown request. */
1161                         error = EINVAL;
1162                 break;
1163         }
1164
1165 out:
1166         /* Drop our hold on this process now that the request has completed. */
1167         _PRELE(p);
1168 fail:
1169         PROC_UNLOCK(p);
1170         if (proctree_locked)
1171                 sx_xunlock(&proctree_lock);
1172         return (error);
1173 }
1174 #undef PROC_READ
1175 #undef PROC_WRITE
1176
1177 /*
1178  * Stop a process because of a debugging event;
1179  * stay stopped until p->p_step is cleared
1180  * (cleared by PIOCCONT in procfs).
1181  */
1182 void
1183 stopevent(struct proc *p, unsigned int event, unsigned int val)
1184 {
1185
1186         PROC_LOCK_ASSERT(p, MA_OWNED);
1187         p->p_step = 1;
1188         do {
1189                 p->p_xstat = val;
1190                 p->p_xthread = NULL;
1191                 p->p_stype = event;     /* Which event caused the stop? */
1192                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1193                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1194         } while (p->p_step);
1195 }