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_IA32
  66 #include <sys/procfs.h>
  67 #include <machine/fpu.h>
  68 #include <compat/ia32/ia32_reg.h>
  69
  70 struct ptrace_io_desc32 {
  71         int             piod_op;
  72         u_int32_t       piod_offs;
  73         u_int32_t       piod_addr;
  74         u_int32_t       piod_len;
  75 };
  76
  77 struct ptrace_vm_entry32 {
  78         uint32_t        pve_cookie;
  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         uint32_t        pve_path;
  85 };
  86
  87 #endif
  88
  89 /*
  90  * Functions implemented using PROC_ACTION():
  91  *
  92  * proc_read_regs(proc, regs)
  93  *      Get the current user-visible register set from the process
  94  *      and copy it into the regs structure (<machine/reg.h>).
  95  *      The process is stopped at the time read_regs is called.
  96  *
  97  * proc_write_regs(proc, regs)
  98  *      Update the current register set from the passed in regs
  99  *      structure.  Take care to avoid clobbering special CPU
 100  *      registers or privileged bits in the PSL.
 101  *      Depending on the architecture this may have fix-up work to do,
 102  *      especially if the IAR or PCW are modified.
 103  *      The process is stopped at the time write_regs is called.
 104  *
 105  * proc_read_fpregs, proc_write_fpregs
 106  *      deal with the floating point register set, otherwise as above.
 107  *
 108  * proc_read_dbregs, proc_write_dbregs
 109  *      deal with the processor debug register set, otherwise as above.
 110  *
 111  * proc_sstep(proc)
 112  *      Arrange for the process to trap after executing a single instruction.
 113  */
 114
 115 #define PROC_ACTION(action) do {                                        \
 116         int error;                                                      \
 117                                                                         \
 118         PROC_LOCK_ASSERT(td->td_proc, MA_OWNED);                        \
 119         if ((td->td_proc->p_flag & P_INMEM) == 0)                       \
 120                 error = EIO;                                            \
 121         else                                                            \
 122                 error = (action);                                       \
 123         return (error);                                                 \
 124 } while(0)
 125
 126 int
 127 proc_read_regs(struct thread *td, struct reg *regs)
 128 {
 129
 130         PROC_ACTION(fill_regs(td, regs));
 131 }
 132
 133 int
 134 proc_write_regs(struct thread *td, struct reg *regs)
 135 {
 136
 137         PROC_ACTION(set_regs(td, regs));
 138 }
 139
 140 int
 141 proc_read_dbregs(struct thread *td, struct dbreg *dbregs)
 142 {
 143
 144         PROC_ACTION(fill_dbregs(td, dbregs));
 145 }
 146
 147 int
 148 proc_write_dbregs(struct thread *td, struct dbreg *dbregs)
 149 {
 150
 151         PROC_ACTION(set_dbregs(td, dbregs));
 152 }
 153
 154 /*
 155  * Ptrace doesn't support fpregs at all, and there are no security holes
 156  * or translations for fpregs, so we can just copy them.
 157  */
 158 int
 159 proc_read_fpregs(struct thread *td, struct fpreg *fpregs)
 160 {
 161
 162         PROC_ACTION(fill_fpregs(td, fpregs));
 163 }
 164
 165 int
 166 proc_write_fpregs(struct thread *td, struct fpreg *fpregs)
 167 {
 168
 169         PROC_ACTION(set_fpregs(td, fpregs));
 170 }
 171
 172 #ifdef COMPAT_IA32
 173 /* For 32 bit binaries, we need to expose the 32 bit regs layouts. */
 174 int
 175 proc_read_regs32(struct thread *td, struct reg32 *regs32)
 176 {
 177
 178         PROC_ACTION(fill_regs32(td, regs32));
 179 }
 180
 181 int
 182 proc_write_regs32(struct thread *td, struct reg32 *regs32)
 183 {
 184
 185         PROC_ACTION(set_regs32(td, regs32));
 186 }
 187
 188 int
 189 proc_read_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 190 {
 191
 192         PROC_ACTION(fill_dbregs32(td, dbregs32));
 193 }
 194
 195 int
 196 proc_write_dbregs32(struct thread *td, struct dbreg32 *dbregs32)
 197 {
 198
 199         PROC_ACTION(set_dbregs32(td, dbregs32));
 200 }
 201
 202 int
 203 proc_read_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 204 {
 205
 206         PROC_ACTION(fill_fpregs32(td, fpregs32));
 207 }
 208
 209 int
 210 proc_write_fpregs32(struct thread *td, struct fpreg32 *fpregs32)
 211 {
 212
 213         PROC_ACTION(set_fpregs32(td, fpregs32));
 214 }
 215 #endif
 216
 217 int
 218 proc_sstep(struct thread *td)
 219 {
 220
 221         PROC_ACTION(ptrace_single_step(td));
 222 }
 223
 224 int
 225 proc_rwmem(struct proc *p, struct uio *uio)
 226 {
 227         vm_map_t map;
 228         vm_object_t backing_object, object;
 229         vm_offset_t pageno;             /* page number */
 230         vm_prot_t reqprot;
 231         int error, writing;
 232
 233         /*
 234          * Assert that someone has locked this vmspace.  (Should be
 235          * curthread but we can't assert that.)  This keeps the process
 236          * from exiting out from under us until this operation completes.
 237          */
 238         KASSERT(p->p_lock >= 1, ("%s: process %p (pid %d) not held", __func__,
 239             p, p->p_pid));
 240
 241         /*
 242          * The map we want...
 243          */
 244         map = &p->p_vmspace->vm_map;
 245
 246         writing = uio->uio_rw == UIO_WRITE;
 247         reqprot = writing ? VM_PROT_COPY | VM_PROT_READ : VM_PROT_READ;
 248
 249         /*
 250          * Only map in one page at a time.  We don't have to, but it
 251          * makes things easier.  This way is trivial - right?
 252          */
 253         do {
 254                 vm_map_t tmap;
 255                 vm_offset_t uva;
 256                 int page_offset;                /* offset into page */
 257                 vm_map_entry_t out_entry;
 258                 vm_prot_t out_prot;
 259                 boolean_t wired;
 260                 vm_pindex_t pindex;
 261                 u_int len;
 262                 vm_page_t m;
 263
 264                 object = NULL;
 265
 266                 uva = (vm_offset_t)uio->uio_offset;
 267
 268                 /*
 269                  * Get the page number of this segment.
 270                  */
 271                 pageno = trunc_page(uva);
 272                 page_offset = uva - pageno;
 273
 274                 /*
 275                  * How many bytes to copy
 276                  */
 277                 len = min(PAGE_SIZE - page_offset, uio->uio_resid);
 278
 279                 /*
 280                  * Fault the page on behalf of the process
 281                  */
 282                 error = vm_fault(map, pageno, reqprot, VM_FAULT_NORMAL);
 283                 if (error) {
 284                         if (error == KERN_RESOURCE_SHORTAGE)
 285                                 error = ENOMEM;
 286                         else
 287                                 error = EFAULT;
 288                         break;
 289                 }
 290
 291                 /*
 292                  * Now we need to get the page.  out_entry and wired
 293                  * aren't used.  One would think the vm code
 294                  * would be a *bit* nicer...  We use tmap because
 295                  * vm_map_lookup() can change the map argument.
 296                  */
 297                 tmap = map;
 298                 error = vm_map_lookup(&tmap, pageno, reqprot, &out_entry,
 299                     &object, &pindex, &out_prot, &wired);
 300                 if (error) {
 301                         error = EFAULT;
 302                         break;
 303                 }
 304                 VM_OBJECT_LOCK(object);
 305                 while ((m = vm_page_lookup(object, pindex)) == NULL &&
 306                     !writing &&
 307                     (backing_object = object->backing_object) != NULL) {
 308                         /*
 309                          * Allow fallback to backing objects if we are reading.
 310                          */
 311                         VM_OBJECT_LOCK(backing_object);
 312                         pindex += OFF_TO_IDX(object->backing_object_offset);
 313                         VM_OBJECT_UNLOCK(object);
 314                         object = backing_object;
 315                 }
 316                 if (writing && m != NULL) {
 317                         vm_page_dirty(m);
 318                         vm_pager_page_unswapped(m);
 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         vm_map_t map;
 364         vm_map_entry_t entry;
 365         vm_object_t obj, tobj, lobj;
 366         struct vnode *vp;
 367         char *freepath, *fullpath;
 368         u_int pathlen;
 369         int error, vfslocked;
 370
 371         map = &p->p_vmspace->vm_map;
 372         entry = map->header.next;
 373         if (pve->pve_cookie != NULL) {
 374                 while (entry != &map->header && entry != pve->pve_cookie)
 375                         entry = entry->next;
 376                 if (entry != pve->pve_cookie)
 377                         return (EINVAL);
 378                 entry = entry->next;
 379         }
 380         while (entry != &map->header && (entry->eflags & MAP_ENTRY_IS_SUB_MAP))
 381                 entry = entry->next;
 382         if (entry == &map->header)
 383                 return (ENOENT);
 384
 385         /* We got an entry. */
 386         pve->pve_cookie = entry;
 387         pve->pve_start = entry->start;
 388         pve->pve_end = entry->end - 1;
 389         pve->pve_offset = entry->offset;
 390         pve->pve_prot = entry->protection;
 391
 392         /* Backing object's path needed? */
 393         if (pve->pve_pathlen == 0)
 394                 return (0);
 395
 396         pathlen = pve->pve_pathlen;
 397         pve->pve_pathlen = 0;
 398
 399         obj = entry->object.vm_object;
 400         if (obj == NULL)
 401                 return (0);
 402
 403         VM_OBJECT_LOCK(obj);
 404         for (lobj = tobj = obj; tobj; tobj = tobj->backing_object) {
 405                 if (tobj != obj)
 406                         VM_OBJECT_LOCK(tobj);
 407                 if (lobj != obj)
 408                         VM_OBJECT_UNLOCK(lobj);
 409                 lobj = tobj;
 410                 pve->pve_offset += tobj->backing_object_offset;
 411         }
 412         if (lobj != NULL) {
 413                 vp = (lobj->type == OBJT_VNODE) ? lobj->handle : NULL;
 414                 if (vp != NULL)
 415                         vref(vp);
 416                 if (lobj != obj)
 417                         VM_OBJECT_UNLOCK(lobj);
 418                 VM_OBJECT_UNLOCK(obj);
 419         } else
 420                 vp = NULL;
 421
 422         if (vp == NULL)
 423                 return (0);
 424
 425         freepath = NULL;
 426         fullpath = NULL;
 427         vn_fullpath(td, vp, &fullpath, &freepath);
 428         vfslocked = VFS_LOCK_GIANT(vp->v_mount);
 429         vrele(vp);
 430         VFS_UNLOCK_GIANT(vfslocked);
 431
 432         error = 0;
 433         if (fullpath != NULL) {
 434                 pve->pve_pathlen = strlen(fullpath) + 1;
 435                 if (pve->pve_pathlen <= pathlen) {
 436                         error = copyout(fullpath, pve->pve_path,
 437                             pve->pve_pathlen);
 438                 } else
 439                         error = ENAMETOOLONG;
 440         }
 441         if (freepath != NULL)
 442                 free(freepath, M_TEMP);
 443         return (error);
 444 }
 445
 446 /*
 447  * Process debugging system call.
 448  */
 449 #ifndef _SYS_SYSPROTO_H_
 450 struct ptrace_args {
 451         int     req;
 452         pid_t   pid;
 453         caddr_t addr;
 454         int     data;
 455 };
 456 #endif
 457
 458 #ifdef COMPAT_IA32
 459 /*
 460  * This CPP subterfuge is to try and reduce the number of ifdefs in
 461  * the body of the code.
 462  *   COPYIN(uap->addr, &r.reg, sizeof r.reg);
 463  * becomes either:
 464  *   copyin(uap->addr, &r.reg, sizeof r.reg);
 465  * or
 466  *   copyin(uap->addr, &r.reg32, sizeof r.reg32);
 467  * .. except this is done at runtime.
 468  */
 469 #define COPYIN(u, k, s)         wrap32 ? \
 470         copyin(u, k ## 32, s ## 32) : \
 471         copyin(u, k, s)
 472 #define COPYOUT(k, u, s)        wrap32 ? \
 473         copyout(k ## 32, u, s ## 32) : \
 474         copyout(k, u, s)
 475 #else
 476 #define COPYIN(u, k, s)         copyin(u, k, s)
 477 #define COPYOUT(k, u, s)        copyout(k, u, s)
 478 #endif
 479 int
 480 ptrace(struct thread *td, struct ptrace_args *uap)
 481 {
 482         /*
 483          * XXX this obfuscation is to reduce stack usage, but the register
 484          * structs may be too large to put on the stack anyway.
 485          */
 486         union {
 487                 struct ptrace_io_desc piod;
 488                 struct ptrace_lwpinfo pl;
 489                 struct ptrace_vm_entry pve;
 490                 struct dbreg dbreg;
 491                 struct fpreg fpreg;
 492                 struct reg reg;
 493 #ifdef COMPAT_IA32
 494                 struct dbreg32 dbreg32;
 495                 struct fpreg32 fpreg32;
 496                 struct reg32 reg32;
 497                 struct ptrace_io_desc32 piod32;
 498                 struct ptrace_vm_entry32 pve32;
 499 #endif
 500         } r;
 501         void *addr;
 502         int error = 0;
 503 #ifdef COMPAT_IA32
 504         int wrap32 = 0;
 505
 506         if (SV_CURPROC_FLAG(SV_ILP32))
 507                 wrap32 = 1;
 508 #endif
 509         AUDIT_ARG_PID(uap->pid);
 510         AUDIT_ARG_CMD(uap->req);
 511         AUDIT_ARG_VALUE(uap->data);
 512         addr = &r;
 513         switch (uap->req) {
 514         case PT_GETREGS:
 515         case PT_GETFPREGS:
 516         case PT_GETDBREGS:
 517         case PT_LWPINFO:
 518                 break;
 519         case PT_SETREGS:
 520                 error = COPYIN(uap->addr, &r.reg, sizeof r.reg);
 521                 break;
 522         case PT_SETFPREGS:
 523                 error = COPYIN(uap->addr, &r.fpreg, sizeof r.fpreg);
 524                 break;
 525         case PT_SETDBREGS:
 526                 error = COPYIN(uap->addr, &r.dbreg, sizeof r.dbreg);
 527                 break;
 528         case PT_IO:
 529                 error = COPYIN(uap->addr, &r.piod, sizeof r.piod);
 530                 break;
 531         case PT_VM_ENTRY:
 532                 error = COPYIN(uap->addr, &r.pve, sizeof r.pve);
 533                 break;
 534         default:
 535                 addr = uap->addr;
 536                 break;
 537         }
 538         if (error)
 539                 return (error);
 540
 541         error = kern_ptrace(td, uap->req, uap->pid, addr, uap->data);
 542         if (error)
 543                 return (error);
 544
 545         switch (uap->req) {
 546         case PT_VM_ENTRY:
 547                 error = COPYOUT(&r.pve, uap->addr, sizeof r.pve);
 548                 break;
 549         case PT_IO:
 550                 error = COPYOUT(&r.piod, uap->addr, sizeof r.piod);
 551                 break;
 552         case PT_GETREGS:
 553                 error = COPYOUT(&r.reg, uap->addr, sizeof r.reg);
 554                 break;
 555         case PT_GETFPREGS:
 556                 error = COPYOUT(&r.fpreg, uap->addr, sizeof r.fpreg);
 557                 break;
 558         case PT_GETDBREGS:
 559                 error = COPYOUT(&r.dbreg, uap->addr, sizeof r.dbreg);
 560                 break;
 561         case PT_LWPINFO:
 562                 error = copyout(&r.pl, uap->addr, uap->data);
 563                 break;
 564         }
 565
 566         return (error);
 567 }
 568 #undef COPYIN
 569 #undef COPYOUT
 570
 571 #ifdef COMPAT_IA32
 572 /*
 573  *   PROC_READ(regs, td2, addr);
 574  * becomes either:
 575  *   proc_read_regs(td2, addr);
 576  * or
 577  *   proc_read_regs32(td2, addr);
 578  * .. except this is done at runtime.  There is an additional
 579  * complication in that PROC_WRITE disallows 32 bit consumers
 580  * from writing to 64 bit address space targets.
 581  */
 582 #define PROC_READ(w, t, a)      wrap32 ? \
 583         proc_read_ ## w ## 32(t, a) : \
 584         proc_read_ ## w (t, a)
 585 #define PROC_WRITE(w, t, a)     wrap32 ? \
 586         (safe ? proc_write_ ## w ## 32(t, a) : EINVAL ) : \
 587         proc_write_ ## w (t, a)
 588 #else
 589 #define PROC_READ(w, t, a)      proc_read_ ## w (t, a)
 590 #define PROC_WRITE(w, t, a)     proc_write_ ## w (t, a)
 591 #endif
 592
 593 int
 594 kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
 595 {
 596         struct iovec iov;
 597         struct uio uio;
 598         struct proc *curp, *p, *pp;
 599         struct thread *td2 = NULL;
 600         struct ptrace_io_desc *piod = NULL;
 601         struct ptrace_lwpinfo *pl;
 602         int error, write, tmp, num;
 603         int proctree_locked = 0;
 604         lwpid_t tid = 0, *buf;
 605 #ifdef COMPAT_IA32
 606         int wrap32 = 0, safe = 0;
 607         struct ptrace_io_desc32 *piod32 = NULL;
 608 #endif
 609
 610         curp = td->td_proc;
 611
 612         /* Lock proctree before locking the process. */
 613         switch (req) {
 614         case PT_TRACE_ME:
 615         case PT_ATTACH:
 616         case PT_STEP:
 617         case PT_CONTINUE:
 618         case PT_TO_SCE:
 619         case PT_TO_SCX:
 620         case PT_SYSCALL:
 621         case PT_DETACH:
 622                 sx_xlock(&proctree_lock);
 623                 proctree_locked = 1;
 624                 break;
 625         default:
 626                 break;
 627         }
 628
 629         write = 0;
 630         if (req == PT_TRACE_ME) {
 631                 p = td->td_proc;
 632                 PROC_LOCK(p);
 633         } else {
 634                 if (pid <= PID_MAX) {
 635                         if ((p = pfind(pid)) == NULL) {
 636                                 if (proctree_locked)
 637                                         sx_xunlock(&proctree_lock);
 638                                 return (ESRCH);
 639                         }
 640                 } else {
 641                         /* this is slow, should be optimized */
 642                         sx_slock(&allproc_lock);
 643                         FOREACH_PROC_IN_SYSTEM(p) {
 644                                 PROC_LOCK(p);
 645                                 FOREACH_THREAD_IN_PROC(p, td2) {
 646                                         if (td2->td_tid == pid)
 647                                                 break;
 648                                 }
 649                                 if (td2 != NULL)
 650                                         break; /* proc lock held */
 651                                 PROC_UNLOCK(p);
 652                         }
 653                         sx_sunlock(&allproc_lock);
 654                         if (p == NULL) {
 655                                 if (proctree_locked)
 656                                         sx_xunlock(&proctree_lock);
 657                                 return (ESRCH);
 658                         }
 659                         tid = pid;
 660                         pid = p->p_pid;
 661                 }
 662         }
 663         AUDIT_ARG_PROCESS(p);
 664
 665         if ((p->p_flag & P_WEXIT) != 0) {
 666                 error = ESRCH;
 667                 goto fail;
 668         }
 669         if ((error = p_cansee(td, p)) != 0)
 670                 goto fail;
 671
 672         if ((error = p_candebug(td, p)) != 0)
 673                 goto fail;
 674
 675         /*
 676          * System processes can't be debugged.
 677          */
 678         if ((p->p_flag & P_SYSTEM) != 0) {
 679                 error = EINVAL;
 680                 goto fail;
 681         }
 682
 683         if (tid == 0) {
 684                 if ((p->p_flag & P_STOPPED_TRACE) != 0) {
 685                         KASSERT(p->p_xthread != NULL, ("NULL p_xthread"));
 686                         td2 = p->p_xthread;
 687                 } else {
 688                         td2 = FIRST_THREAD_IN_PROC(p);
 689                 }
 690                 tid = td2->td_tid;
 691         }
 692
 693 #ifdef COMPAT_IA32
 694         /*
 695          * Test if we're a 32 bit client and what the target is.
 696          * Set the wrap controls accordingly.
 697          */
 698         if (SV_CURPROC_FLAG(SV_ILP32)) {
 699                 if (td2->td_proc->p_sysent->sv_flags & SV_ILP32)
 700                         safe = 1;
 701                 wrap32 = 1;
 702         }
 703 #endif
 704         /*
 705          * Permissions check
 706          */
 707         switch (req) {
 708         case PT_TRACE_ME:
 709                 /* Always legal. */
 710                 break;
 711
 712         case PT_ATTACH:
 713                 /* Self */
 714                 if (p->p_pid == td->td_proc->p_pid) {
 715                         error = EINVAL;
 716                         goto fail;
 717                 }
 718
 719                 /* Already traced */
 720                 if (p->p_flag & P_TRACED) {
 721                         error = EBUSY;
 722                         goto fail;
 723                 }
 724
 725                 /* Can't trace an ancestor if you're being traced. */
 726                 if (curp->p_flag & P_TRACED) {
 727                         for (pp = curp->p_pptr; pp != NULL; pp = pp->p_pptr) {
 728                                 if (pp == p) {
 729                                         error = EINVAL;
 730                                         goto fail;
 731                                 }
 732                         }
 733                 }
 734
 735
 736                 /* OK */
 737                 break;
 738
 739         case PT_CLEARSTEP:
 740                 /* Allow thread to clear single step for itself */
 741                 if (td->td_tid == tid)
 742                         break;
 743
 744                 /* FALLTHROUGH */
 745         default:
 746                 /* not being traced... */
 747                 if ((p->p_flag & P_TRACED) == 0) {
 748                         error = EPERM;
 749                         goto fail;
 750                 }
 751
 752                 /* not being traced by YOU */
 753                 if (p->p_pptr != td->td_proc) {
 754                         error = EBUSY;
 755                         goto fail;
 756                 }
 757
 758                 /* not currently stopped */
 759                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) == 0 ||
 760                     p->p_suspcount != p->p_numthreads  ||
 761                     (p->p_flag & P_WAITED) == 0) {
 762                         error = EBUSY;
 763                         goto fail;
 764                 }
 765
 766                 if ((p->p_flag & P_STOPPED_TRACE) == 0) {
 767                         static int count = 0;
 768                         if (count++ == 0)
 769                                 printf("P_STOPPED_TRACE not set.\n");
 770                 }
 771
 772                 /* OK */
 773                 break;
 774         }
 775
 776         /* Keep this process around until we finish this request. */
 777         _PHOLD(p);
 778
 779 #ifdef FIX_SSTEP
 780         /*
 781          * Single step fixup ala procfs
 782          */
 783         FIX_SSTEP(td2);
 784 #endif
 785
 786         /*
 787          * Actually do the requests
 788          */
 789
 790         td->td_retval[0] = 0;
 791
 792         switch (req) {
 793         case PT_TRACE_ME:
 794                 /* set my trace flag and "owner" so it can read/write me */
 795                 p->p_flag |= P_TRACED;
 796                 p->p_oppid = p->p_pptr->p_pid;
 797                 break;
 798
 799         case PT_ATTACH:
 800                 /* security check done above */
 801                 p->p_flag |= P_TRACED;
 802                 p->p_oppid = p->p_pptr->p_pid;
 803                 if (p->p_pptr != td->td_proc)
 804                         proc_reparent(p, td->td_proc);
 805                 data = SIGSTOP;
 806                 goto sendsig;   /* in PT_CONTINUE below */
 807
 808         case PT_CLEARSTEP:
 809                 error = ptrace_clear_single_step(td2);
 810                 break;
 811
 812         case PT_SETSTEP:
 813                 error = ptrace_single_step(td2);
 814                 break;
 815
 816         case PT_SUSPEND:
 817                 td2->td_dbgflags |= TDB_SUSPEND;
 818                 thread_lock(td2);
 819                 td2->td_flags |= TDF_NEEDSUSPCHK;
 820                 thread_unlock(td2);
 821                 break;
 822
 823         case PT_RESUME:
 824                 td2->td_dbgflags &= ~TDB_SUSPEND;
 825                 break;
 826
 827         case PT_STEP:
 828         case PT_CONTINUE:
 829         case PT_TO_SCE:
 830         case PT_TO_SCX:
 831         case PT_SYSCALL:
 832         case PT_DETACH:
 833                 /* Zero means do not send any signal */
 834                 if (data < 0 || data > _SIG_MAXSIG) {
 835                         error = EINVAL;
 836                         break;
 837                 }
 838
 839                 switch (req) {
 840                 case PT_STEP:
 841                         error = ptrace_single_step(td2);
 842                         if (error)
 843                                 goto out;
 844                         break;
 845                 case PT_TO_SCE:
 846                         p->p_stops |= S_PT_SCE;
 847                         break;
 848                 case PT_TO_SCX:
 849                         p->p_stops |= S_PT_SCX;
 850                         break;
 851                 case PT_SYSCALL:
 852                         p->p_stops |= S_PT_SCE | S_PT_SCX;
 853                         break;
 854                 }
 855
 856                 if (addr != (void *)1) {
 857                         error = ptrace_set_pc(td2, (u_long)(uintfptr_t)addr);
 858                         if (error)
 859                                 break;
 860                 }
 861
 862                 if (req == PT_DETACH) {
 863                         /* reset process parent */
 864                         if (p->p_oppid != p->p_pptr->p_pid) {
 865                                 struct proc *pp;
 866
 867                                 PROC_LOCK(p->p_pptr);
 868                                 sigqueue_take(p->p_ksi);
 869                                 PROC_UNLOCK(p->p_pptr);
 870
 871                                 PROC_UNLOCK(p);
 872                                 pp = pfind(p->p_oppid);
 873                                 if (pp == NULL)
 874                                         pp = initproc;
 875                                 else
 876                                         PROC_UNLOCK(pp);
 877                                 PROC_LOCK(p);
 878                                 proc_reparent(p, pp);
 879                                 if (pp == initproc)
 880                                         p->p_sigparent = SIGCHLD;
 881                         }
 882                         p->p_flag &= ~(P_TRACED | P_WAITED);
 883                         p->p_oppid = 0;
 884
 885                         /* should we send SIGCHLD? */
 886                         /* childproc_continued(p); */
 887                 }
 888
 889         sendsig:
 890                 if (proctree_locked) {
 891                         sx_xunlock(&proctree_lock);
 892                         proctree_locked = 0;
 893                 }
 894                 p->p_xstat = data;
 895                 p->p_xthread = NULL;
 896                 if ((p->p_flag & (P_STOPPED_SIG | P_STOPPED_TRACE)) != 0) {
 897                         /* deliver or queue signal */
 898                         td2->td_dbgflags &= ~TDB_XSIG;
 899                         td2->td_xsig = data;
 900
 901                         if (req == PT_DETACH) {
 902                                 struct thread *td3;
 903                                 FOREACH_THREAD_IN_PROC(p, td3) {
 904                                         td3->td_dbgflags &= ~TDB_SUSPEND;
 905                                 }
 906                         }
 907                         /*
 908                          * unsuspend all threads, to not let a thread run,
 909                          * you should use PT_SUSPEND to suspend it before
 910                          * continuing process.
 911                          */
 912                         PROC_SLOCK(p);
 913                         p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
 914                         thread_unsuspend(p);
 915                         PROC_SUNLOCK(p);
 916                 } else {
 917                         if (data)
 918                                 psignal(p, data);
 919                 }
 920                 break;
 921
 922         case PT_WRITE_I:
 923         case PT_WRITE_D:
 924                 td2->td_dbgflags |= TDB_USERWR;
 925                 write = 1;
 926                 /* FALLTHROUGH */
 927         case PT_READ_I:
 928         case PT_READ_D:
 929                 PROC_UNLOCK(p);
 930                 tmp = 0;
 931                 /* write = 0 set above */
 932                 iov.iov_base = write ? (caddr_t)&data : (caddr_t)&tmp;
 933                 iov.iov_len = sizeof(int);
 934                 uio.uio_iov = &iov;
 935                 uio.uio_iovcnt = 1;
 936                 uio.uio_offset = (off_t)(uintptr_t)addr;
 937                 uio.uio_resid = sizeof(int);
 938                 uio.uio_segflg = UIO_SYSSPACE;  /* i.e.: the uap */
 939                 uio.uio_rw = write ? UIO_WRITE : UIO_READ;
 940                 uio.uio_td = td;
 941                 error = proc_rwmem(p, &uio);
 942                 if (uio.uio_resid != 0) {
 943                         /*
 944                          * XXX proc_rwmem() doesn't currently return ENOSPC,
 945                          * so I think write() can bogusly return 0.
 946                          * XXX what happens for short writes?  We don't want
 947                          * to write partial data.
 948                          * XXX proc_rwmem() returns EPERM for other invalid
 949                          * addresses.  Convert this to EINVAL.  Does this
 950                          * clobber returns of EPERM for other reasons?
 951                          */
 952                         if (error == 0 || error == ENOSPC || error == EPERM)
 953                                 error = EINVAL; /* EOF */
 954                 }
 955                 if (!write)
 956                         td->td_retval[0] = tmp;
 957                 PROC_LOCK(p);
 958                 break;
 959
 960         case PT_IO:
 961 #ifdef COMPAT_IA32
 962                 if (wrap32) {
 963                         piod32 = addr;
 964                         iov.iov_base = (void *)(uintptr_t)piod32->piod_addr;
 965                         iov.iov_len = piod32->piod_len;
 966                         uio.uio_offset = (off_t)(uintptr_t)piod32->piod_offs;
 967                         uio.uio_resid = piod32->piod_len;
 968                 } else
 969 #endif
 970                 {
 971                         piod = addr;
 972                         iov.iov_base = piod->piod_addr;
 973                         iov.iov_len = piod->piod_len;
 974                         uio.uio_offset = (off_t)(uintptr_t)piod->piod_offs;
 975                         uio.uio_resid = piod->piod_len;
 976                 }
 977                 uio.uio_iov = &iov;
 978                 uio.uio_iovcnt = 1;
 979                 uio.uio_segflg = UIO_USERSPACE;
 980                 uio.uio_td = td;
 981 #ifdef COMPAT_IA32
 982                 tmp = wrap32 ? piod32->piod_op : piod->piod_op;
 983 #else
 984                 tmp = piod->piod_op;
 985 #endif
 986                 switch (tmp) {
 987                 case PIOD_READ_D:
 988                 case PIOD_READ_I:
 989                         uio.uio_rw = UIO_READ;
 990                         break;
 991                 case PIOD_WRITE_D:
 992                 case PIOD_WRITE_I:
 993                         td2->td_dbgflags |= TDB_USERWR;
 994                         uio.uio_rw = UIO_WRITE;
 995                         break;
 996                 default:
 997                         error = EINVAL;
 998                         goto out;
 999                 }
1000                 PROC_UNLOCK(p);
1001                 error = proc_rwmem(p, &uio);
1002 #ifdef COMPAT_IA32
1003                 if (wrap32)
1004                         piod32->piod_len -= uio.uio_resid;
1005                 else
1006 #endif
1007                         piod->piod_len -= uio.uio_resid;
1008                 PROC_LOCK(p);
1009                 break;
1010
1011         case PT_KILL:
1012                 data = SIGKILL;
1013                 goto sendsig;   /* in PT_CONTINUE above */
1014
1015         case PT_SETREGS:
1016                 td2->td_dbgflags |= TDB_USERWR;
1017                 error = PROC_WRITE(regs, td2, addr);
1018                 break;
1019
1020         case PT_GETREGS:
1021                 error = PROC_READ(regs, td2, addr);
1022                 break;
1023
1024         case PT_SETFPREGS:
1025                 td2->td_dbgflags |= TDB_USERWR;
1026                 error = PROC_WRITE(fpregs, td2, addr);
1027                 break;
1028
1029         case PT_GETFPREGS:
1030                 error = PROC_READ(fpregs, td2, addr);
1031                 break;
1032
1033         case PT_SETDBREGS:
1034                 td2->td_dbgflags |= TDB_USERWR;
1035                 error = PROC_WRITE(dbregs, td2, addr);
1036                 break;
1037
1038         case PT_GETDBREGS:
1039                 error = PROC_READ(dbregs, td2, addr);
1040                 break;
1041
1042         case PT_LWPINFO:
1043                 if (data <= 0 || data > sizeof(*pl)) {
1044                         error = EINVAL;
1045                         break;
1046                 }
1047                 pl = addr;
1048                 pl->pl_lwpid = td2->td_tid;
1049                 if (td2->td_dbgflags & TDB_XSIG)
1050                         pl->pl_event = PL_EVENT_SIGNAL;
1051                 else
1052                         pl->pl_event = 0;
1053                 pl->pl_flags = 0;
1054                 pl->pl_sigmask = td2->td_sigmask;
1055                 pl->pl_siglist = td2->td_siglist;
1056                 break;
1057
1058         case PT_GETNUMLWPS:
1059                 td->td_retval[0] = p->p_numthreads;
1060                 break;
1061
1062         case PT_GETLWPLIST:
1063                 if (data <= 0) {
1064                         error = EINVAL;
1065                         break;
1066                 }
1067                 num = imin(p->p_numthreads, data);
1068                 PROC_UNLOCK(p);
1069                 buf = malloc(num * sizeof(lwpid_t), M_TEMP, M_WAITOK);
1070                 tmp = 0;
1071                 PROC_LOCK(p);
1072                 FOREACH_THREAD_IN_PROC(p, td2) {
1073                         if (tmp >= num)
1074                                 break;
1075                         buf[tmp++] = td2->td_tid;
1076                 }
1077                 PROC_UNLOCK(p);
1078                 error = copyout(buf, addr, tmp * sizeof(lwpid_t));
1079                 free(buf, M_TEMP);
1080                 if (!error)
1081                         td->td_retval[0] = tmp;
1082                 PROC_LOCK(p);
1083                 break;
1084
1085         case PT_VM_TIMESTAMP:
1086                 td->td_retval[0] = p->p_vmspace->vm_map.timestamp;
1087                 break;
1088
1089         case PT_VM_ENTRY:
1090 #ifdef COMPAT_IA32
1091                 /* XXX to be implemented. */
1092                 if (wrap32) {
1093                         error = EDOOFUS;
1094                         break;
1095                 }
1096 #endif
1097                 PROC_UNLOCK(p);
1098                 error = ptrace_vm_entry(td, p, addr);
1099                 PROC_LOCK(p);
1100                 break;
1101
1102         default:
1103 #ifdef __HAVE_PTRACE_MACHDEP
1104                 if (req >= PT_FIRSTMACH) {
1105                         PROC_UNLOCK(p);
1106                         error = cpu_ptrace(td2, req, addr, data);
1107                         PROC_LOCK(p);
1108                 } else
1109 #endif
1110                         /* Unknown request. */
1111                         error = EINVAL;
1112                 break;
1113         }
1114
1115 out:
1116         /* Drop our hold on this process now that the request has completed. */
1117         _PRELE(p);
1118 fail:
1119         PROC_UNLOCK(p);
1120         if (proctree_locked)
1121                 sx_xunlock(&proctree_lock);
1122         return (error);
1123 }
1124 #undef PROC_READ
1125 #undef PROC_WRITE
1126
1127 /*
1128  * Stop a process because of a debugging event;
1129  * stay stopped until p->p_step is cleared
1130  * (cleared by PIOCCONT in procfs).
1131  */
1132 void
1133 stopevent(struct proc *p, unsigned int event, unsigned int val)
1134 {
1135
1136         PROC_LOCK_ASSERT(p, MA_OWNED);
1137         p->p_step = 1;
1138         do {
1139                 p->p_xstat = val;
1140                 p->p_xthread = NULL;
1141                 p->p_stype = event;     /* Which event caused the stop? */
1142                 wakeup(&p->p_stype);    /* Wake up any PIOCWAIT'ing procs */
1143                 msleep(&p->p_step, &p->p_mtx, PWAIT, "stopevent", 0);
1144         } while (p->p_step);
1145 }