1 /* 2 * Copyright (c) 1993, David Greenman 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 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include "opt_ktrace.h" 31 #include "opt_mac.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/eventhandler.h> 36 #include <sys/lock.h> 37 #include <sys/mutex.h> 38 #include <sys/sysproto.h> 39 #include <sys/signalvar.h> 40 #include <sys/kernel.h> 41 #include <sys/mac.h> 42 #include <sys/mount.h> 43 #include <sys/filedesc.h> 44 #include <sys/fcntl.h> 45 #include <sys/acct.h> 46 #include <sys/exec.h> 47 #include <sys/imgact.h> 48 #include <sys/imgact_elf.h> 49 #include <sys/wait.h> 50 #include <sys/malloc.h> 51 #include <sys/proc.h> 52 #include <sys/pioctl.h> 53 #include <sys/namei.h> 54 #include <sys/sf_buf.h> 55 #include <sys/sysent.h> 56 #include <sys/shm.h> 57 #include <sys/sysctl.h> 58 #include <sys/user.h> 59 #include <sys/vnode.h> 60 #ifdef KTRACE 61 #include <sys/ktrace.h> 62 #endif 63 64 #include <vm/vm.h> 65 #include <vm/vm_param.h> 66 #include <vm/pmap.h> 67 #include <vm/vm_page.h> 68 #include <vm/vm_map.h> 69 #include <vm/vm_kern.h> 70 #include <vm/vm_extern.h> 71 #include <vm/vm_object.h> 72 #include <vm/vm_pager.h> 73 74 #include <machine/reg.h> 75 76 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 77 78 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 79 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 80 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 81 static int kern_execve(struct thread *td, char *fname, char **argv, 82 char **envv, struct mac *mac_p); 83 84 /* XXX This should be vm_size_t. */ 85 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 86 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 87 88 /* XXX This should be vm_size_t. */ 89 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD, 90 NULL, 0, sysctl_kern_usrstack, "LU", ""); 91 92 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 93 NULL, 0, sysctl_kern_stackprot, "I", ""); 94 95 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 96 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 97 &ps_arg_cache_limit, 0, ""); 98 99 static int 100 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 101 { 102 struct proc *p; 103 int error; 104 105 p = curproc; 106 #if defined(__amd64__) || defined(__ia64__) 107 if (req->oldlen == sizeof(unsigned int)) { 108 unsigned int val; 109 val = (unsigned int)p->p_sysent->sv_psstrings; 110 error = SYSCTL_OUT(req, &val, sizeof(val)); 111 } else 112 #endif 113 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 114 sizeof(p->p_sysent->sv_psstrings)); 115 return error; 116 } 117 118 static int 119 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 120 { 121 struct proc *p; 122 int error; 123 124 p = curproc; 125 #if defined(__amd64__) || defined(__ia64__) 126 if (req->oldlen == sizeof(unsigned int)) { 127 unsigned int val; 128 val = (unsigned int)p->p_sysent->sv_usrstack; 129 error = SYSCTL_OUT(req, &val, sizeof(val)); 130 } else 131 #endif 132 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 133 sizeof(p->p_sysent->sv_usrstack)); 134 return error; 135 } 136 137 static int 138 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 139 { 140 struct proc *p; 141 142 p = curproc; 143 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 144 sizeof(p->p_sysent->sv_stackprot))); 145 } 146 147 /* 148 * Each of the items is a pointer to a `const struct execsw', hence the 149 * double pointer here. 150 */ 151 static const struct execsw **execsw; 152 153 #ifndef _SYS_SYSPROTO_H_ 154 struct execve_args { 155 char *fname; 156 char **argv; 157 char **envv; 158 }; 159 #endif 160 161 /* 162 * MPSAFE 163 */ 164 int 165 execve(td, uap) 166 struct thread *td; 167 struct execve_args /* { 168 char *fname; 169 char **argv; 170 char **envv; 171 } */ *uap; 172 { 173 174 return (kern_execve(td, uap->fname, uap->argv, uap->envv, NULL)); 175 } 176 177 #ifndef _SYS_SYSPROTO_H_ 178 struct __mac_execve_args { 179 char *fname; 180 char **argv; 181 char **envv; 182 struct mac *mac_p; 183 }; 184 #endif 185 186 /* 187 * MPSAFE 188 */ 189 int 190 __mac_execve(td, uap) 191 struct thread *td; 192 struct __mac_execve_args /* { 193 char *fname; 194 char **argv; 195 char **envv; 196 struct mac *mac_p; 197 } */ *uap; 198 { 199 200 #ifdef MAC 201 return (kern_execve(td, uap->fname, uap->argv, uap->envv, 202 uap->mac_p)); 203 #else 204 return (ENOSYS); 205 #endif 206 } 207 208 /* 209 * In-kernel implementation of execve(). All arguments are assumed to be 210 * userspace pointers from the passed thread. 211 * 212 * MPSAFE 213 */ 214 static int 215 kern_execve(td, fname, argv, envv, mac_p) 216 struct thread *td; 217 char *fname; 218 char **argv; 219 char **envv; 220 struct mac *mac_p; 221 { 222 struct proc *p = td->td_proc; 223 struct nameidata nd, *ndp; 224 struct ucred *newcred = NULL, *oldcred; 225 struct uidinfo *euip; 226 register_t *stack_base; 227 int error, len, i; 228 struct image_params image_params, *imgp; 229 struct vattr attr; 230 int (*img_first)(struct image_params *); 231 struct pargs *oldargs = NULL, *newargs = NULL; 232 struct sigacts *oldsigacts, *newsigacts; 233 #ifdef KTRACE 234 struct vnode *tracevp = NULL; 235 struct ucred *tracecred = NULL; 236 #endif 237 struct vnode *textvp = NULL; 238 int credential_changing; 239 int textset; 240 #ifdef MAC 241 struct label *interplabel = NULL; 242 int will_transition; 243 #endif 244 245 imgp = &image_params; 246 247 /* 248 * Lock the process and set the P_INEXEC flag to indicate that 249 * it should be left alone until we're done here. This is 250 * necessary to avoid race conditions - e.g. in ptrace() - 251 * that might allow a local user to illicitly obtain elevated 252 * privileges. 253 */ 254 PROC_LOCK(p); 255 KASSERT((p->p_flag & P_INEXEC) == 0, 256 ("%s(): process already has P_INEXEC flag", __func__)); 257 if (p->p_flag & P_SA || p->p_numthreads > 1) { 258 if (thread_single(SINGLE_EXIT)) { 259 PROC_UNLOCK(p); 260 mtx_unlock(&Giant); 261 return (ERESTART); /* Try again later. */ 262 } 263 /* 264 * If we get here all other threads are dead, 265 * so unset the associated flags and lose KSE mode. 266 */ 267 p->p_flag &= ~P_SA; 268 td->td_mailbox = NULL; 269 mtx_lock_spin(&sched_lock); 270 td->td_flags &= ~TDF_SA; 271 mtx_unlock_spin(&sched_lock); 272 thread_single_end(); 273 } 274 p->p_flag |= P_INEXEC; 275 PROC_UNLOCK(p); 276 277 /* 278 * Initialize part of the common data 279 */ 280 imgp->proc = p; 281 imgp->userspace_argv = argv; 282 imgp->userspace_envv = envv; 283 imgp->execlabel = NULL; 284 imgp->attr = &attr; 285 imgp->argc = imgp->envc = 0; 286 imgp->argv0 = NULL; 287 imgp->entry_addr = 0; 288 imgp->vmspace_destroyed = 0; 289 imgp->interpreted = 0; 290 imgp->interpreter_name[0] = '\0'; 291 imgp->auxargs = NULL; 292 imgp->vp = NULL; 293 imgp->object = NULL; 294 imgp->firstpage = NULL; 295 imgp->ps_strings = 0; 296 imgp->auxarg_size = 0; 297 298 #ifdef MAC 299 error = mac_execve_enter(imgp, mac_p); 300 if (error) { 301 mtx_lock(&Giant); 302 goto exec_fail; 303 } 304 #endif 305 306 /* 307 * Allocate temporary demand zeroed space for argument and 308 * environment strings 309 */ 310 imgp->stringbase = (char *)kmem_alloc_wait(exec_map, ARG_MAX); 311 if (imgp->stringbase == NULL) { 312 error = ENOMEM; 313 mtx_lock(&Giant); 314 goto exec_fail; 315 } 316 imgp->stringp = imgp->stringbase; 317 imgp->stringspace = ARG_MAX; 318 imgp->image_header = NULL; 319 320 /* 321 * Translate the file name. namei() returns a vnode pointer 322 * in ni_vp amoung other things. 323 */ 324 ndp = &nd; 325 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 326 UIO_USERSPACE, fname, td); 327 328 mtx_lock(&Giant); 329 interpret: 330 331 error = namei(ndp); 332 if (error) { 333 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase, 334 ARG_MAX); 335 goto exec_fail; 336 } 337 338 imgp->vp = ndp->ni_vp; 339 imgp->fname = fname; 340 341 /* 342 * Check file permissions (also 'opens' file) 343 */ 344 error = exec_check_permissions(imgp); 345 if (error) 346 goto exec_fail_dealloc; 347 348 if (VOP_GETVOBJECT(imgp->vp, &imgp->object) == 0) 349 vm_object_reference(imgp->object); 350 351 /* 352 * Set VV_TEXT now so no one can write to the executable while we're 353 * activating it. 354 * 355 * Remember if this was set before and unset it in case this is not 356 * actually an executable image. 357 */ 358 textset = imgp->vp->v_vflag & VV_TEXT; 359 imgp->vp->v_vflag |= VV_TEXT; 360 361 error = exec_map_first_page(imgp); 362 if (error) 363 goto exec_fail_dealloc; 364 365 /* 366 * If the current process has a special image activator it 367 * wants to try first, call it. For example, emulating shell 368 * scripts differently. 369 */ 370 error = -1; 371 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 372 error = img_first(imgp); 373 374 /* 375 * Loop through the list of image activators, calling each one. 376 * An activator returns -1 if there is no match, 0 on success, 377 * and an error otherwise. 378 */ 379 for (i = 0; error == -1 && execsw[i]; ++i) { 380 if (execsw[i]->ex_imgact == NULL || 381 execsw[i]->ex_imgact == img_first) { 382 continue; 383 } 384 error = (*execsw[i]->ex_imgact)(imgp); 385 } 386 387 if (error) { 388 if (error == -1) { 389 if (textset == 0) 390 imgp->vp->v_vflag &= ~VV_TEXT; 391 error = ENOEXEC; 392 } 393 goto exec_fail_dealloc; 394 } 395 396 /* 397 * Special interpreter operation, cleanup and loop up to try to 398 * activate the interpreter. 399 */ 400 if (imgp->interpreted) { 401 exec_unmap_first_page(imgp); 402 /* 403 * VV_TEXT needs to be unset for scripts. There is a short 404 * period before we determine that something is a script where 405 * VV_TEXT will be set. The vnode lock is held over this 406 * entire period so nothing should illegitimately be blocked. 407 */ 408 imgp->vp->v_vflag &= ~VV_TEXT; 409 /* free name buffer and old vnode */ 410 NDFREE(ndp, NDF_ONLY_PNBUF); 411 #ifdef MAC 412 interplabel = mac_vnode_label_alloc(); 413 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel); 414 #endif 415 vput(ndp->ni_vp); 416 vm_object_deallocate(imgp->object); 417 imgp->object = NULL; 418 /* set new name to that of the interpreter */ 419 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 420 UIO_SYSSPACE, imgp->interpreter_name, td); 421 goto interpret; 422 } 423 424 /* 425 * Copy out strings (args and env) and initialize stack base 426 */ 427 if (p->p_sysent->sv_copyout_strings) 428 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 429 else 430 stack_base = exec_copyout_strings(imgp); 431 432 /* 433 * If custom stack fixup routine present for this process 434 * let it do the stack setup. 435 * Else stuff argument count as first item on stack 436 */ 437 if (p->p_sysent->sv_fixup != NULL) 438 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 439 else 440 suword(--stack_base, imgp->argc); 441 442 /* 443 * For security and other reasons, the file descriptor table cannot 444 * be shared after an exec. 445 */ 446 FILEDESC_LOCK(p->p_fd); 447 if (p->p_fd->fd_refcnt > 1) { 448 struct filedesc *tmp; 449 450 tmp = fdcopy(td->td_proc->p_fd); 451 FILEDESC_UNLOCK(p->p_fd); 452 fdfree(td); 453 p->p_fd = tmp; 454 } else 455 FILEDESC_UNLOCK(p->p_fd); 456 457 /* 458 * Malloc things before we need locks. 459 */ 460 newcred = crget(); 461 euip = uifind(attr.va_uid); 462 i = imgp->endargs - imgp->stringbase; 463 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) 464 newargs = pargs_alloc(i); 465 466 /* close files on exec */ 467 fdcloseexec(td); 468 469 /* Get a reference to the vnode prior to locking the proc */ 470 VREF(ndp->ni_vp); 471 472 /* 473 * For security and other reasons, signal handlers cannot 474 * be shared after an exec. The new process gets a copy of the old 475 * handlers. In execsigs(), the new process will have its signals 476 * reset. 477 */ 478 PROC_LOCK(p); 479 if (sigacts_shared(p->p_sigacts)) { 480 oldsigacts = p->p_sigacts; 481 PROC_UNLOCK(p); 482 newsigacts = sigacts_alloc(); 483 sigacts_copy(newsigacts, oldsigacts); 484 PROC_LOCK(p); 485 p->p_sigacts = newsigacts; 486 } else 487 oldsigacts = NULL; 488 489 /* Stop profiling */ 490 stopprofclock(p); 491 492 /* reset caught signals */ 493 execsigs(p); 494 495 /* name this process - nameiexec(p, ndp) */ 496 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN); 497 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len); 498 p->p_comm[len] = 0; 499 500 /* 501 * mark as execed, wakeup the process that vforked (if any) and tell 502 * it that it now has its own resources back 503 */ 504 p->p_flag |= P_EXEC; 505 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 506 p->p_flag &= ~P_PPWAIT; 507 wakeup(p->p_pptr); 508 } 509 510 /* 511 * Implement image setuid/setgid. 512 * 513 * Don't honor setuid/setgid if the filesystem prohibits it or if 514 * the process is being traced. 515 * 516 * XXXMAC: For the time being, use NOSUID to also prohibit 517 * transitions on the file system. 518 */ 519 oldcred = p->p_ucred; 520 credential_changing = 0; 521 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid != 522 attr.va_uid; 523 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid != 524 attr.va_gid; 525 #ifdef MAC 526 will_transition = mac_execve_will_transition(oldcred, imgp->vp, 527 interplabel, imgp); 528 credential_changing |= will_transition; 529 #endif 530 531 if (credential_changing && 532 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 533 (p->p_flag & P_TRACED) == 0) { 534 /* 535 * Turn off syscall tracing for set-id programs, except for 536 * root. Record any set-id flags first to make sure that 537 * we do not regain any tracing during a possible block. 538 */ 539 setsugid(p); 540 #ifdef KTRACE 541 if (p->p_tracevp != NULL && suser_cred(oldcred, PRISON_ROOT)) { 542 mtx_lock(&ktrace_mtx); 543 p->p_traceflag = 0; 544 tracevp = p->p_tracevp; 545 p->p_tracevp = NULL; 546 tracecred = p->p_tracecred; 547 p->p_tracecred = NULL; 548 mtx_unlock(&ktrace_mtx); 549 } 550 #endif 551 /* 552 * Close any file descriptors 0..2 that reference procfs, 553 * then make sure file descriptors 0..2 are in use. 554 * 555 * setugidsafety() may call closef() and then pfind() 556 * which may grab the process lock. 557 * fdcheckstd() may call falloc() which may block to 558 * allocate memory, so temporarily drop the process lock. 559 */ 560 PROC_UNLOCK(p); 561 setugidsafety(td); 562 error = fdcheckstd(td); 563 if (error != 0) 564 goto done1; 565 PROC_LOCK(p); 566 /* 567 * Set the new credentials. 568 */ 569 crcopy(newcred, oldcred); 570 if (attr.va_mode & VSUID) 571 change_euid(newcred, euip); 572 if (attr.va_mode & VSGID) 573 change_egid(newcred, attr.va_gid); 574 #ifdef MAC 575 if (will_transition) { 576 mac_execve_transition(oldcred, newcred, imgp->vp, 577 interplabel, imgp); 578 } 579 #endif 580 /* 581 * Implement correct POSIX saved-id behavior. 582 * 583 * XXXMAC: Note that the current logic will save the 584 * uid and gid if a MAC domain transition occurs, even 585 * though maybe it shouldn't. 586 */ 587 change_svuid(newcred, newcred->cr_uid); 588 change_svgid(newcred, newcred->cr_gid); 589 p->p_ucred = newcred; 590 newcred = NULL; 591 } else { 592 if (oldcred->cr_uid == oldcred->cr_ruid && 593 oldcred->cr_gid == oldcred->cr_rgid) 594 p->p_flag &= ~P_SUGID; 595 /* 596 * Implement correct POSIX saved-id behavior. 597 * 598 * XXX: It's not clear that the existing behavior is 599 * POSIX-compliant. A number of sources indicate that the 600 * saved uid/gid should only be updated if the new ruid is 601 * not equal to the old ruid, or the new euid is not equal 602 * to the old euid and the new euid is not equal to the old 603 * ruid. The FreeBSD code always updates the saved uid/gid. 604 * Also, this code uses the new (replaced) euid and egid as 605 * the source, which may or may not be the right ones to use. 606 */ 607 if (oldcred->cr_svuid != oldcred->cr_uid || 608 oldcred->cr_svgid != oldcred->cr_gid) { 609 crcopy(newcred, oldcred); 610 change_svuid(newcred, newcred->cr_uid); 611 change_svgid(newcred, newcred->cr_gid); 612 p->p_ucred = newcred; 613 newcred = NULL; 614 } 615 } 616 617 /* 618 * Store the vp for use in procfs. This vnode was referenced prior 619 * to locking the proc lock. 620 */ 621 textvp = p->p_textvp; 622 p->p_textvp = ndp->ni_vp; 623 624 /* 625 * Notify others that we exec'd, and clear the P_INEXEC flag 626 * as we're now a bona fide freshly-execed process. 627 */ 628 KNOTE(&p->p_klist, NOTE_EXEC); 629 p->p_flag &= ~P_INEXEC; 630 631 /* 632 * If tracing the process, trap to debugger so breakpoints 633 * can be set before the program executes. 634 */ 635 if (p->p_flag & P_TRACED) 636 psignal(p, SIGTRAP); 637 638 /* clear "fork but no exec" flag, as we _are_ execing */ 639 p->p_acflag &= ~AFORK; 640 641 /* Free any previous argument cache */ 642 oldargs = p->p_args; 643 p->p_args = NULL; 644 645 /* Cache arguments if they fit inside our allowance */ 646 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 647 bcopy(imgp->stringbase, newargs->ar_args, i); 648 p->p_args = newargs; 649 newargs = NULL; 650 } 651 PROC_UNLOCK(p); 652 653 /* Set values passed into the program in registers. */ 654 if (p->p_sysent->sv_setregs) 655 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr, 656 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 657 else 658 exec_setregs(td, imgp->entry_addr, 659 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 660 661 done1: 662 /* 663 * Free any resources malloc'd earlier that we didn't use. 664 */ 665 uifree(euip); 666 if (newcred == NULL) 667 crfree(oldcred); 668 else 669 crfree(newcred); 670 /* 671 * Handle deferred decrement of ref counts. 672 */ 673 if (textvp != NULL) 674 vrele(textvp); 675 if (ndp->ni_vp && error != 0) 676 vrele(ndp->ni_vp); 677 #ifdef KTRACE 678 if (tracevp != NULL) 679 vrele(tracevp); 680 if (tracecred != NULL) 681 crfree(tracecred); 682 #endif 683 if (oldargs != NULL) 684 pargs_drop(oldargs); 685 if (newargs != NULL) 686 pargs_drop(newargs); 687 if (oldsigacts != NULL) 688 sigacts_free(oldsigacts); 689 690 exec_fail_dealloc: 691 692 /* 693 * free various allocated resources 694 */ 695 if (imgp->firstpage != NULL) 696 exec_unmap_first_page(imgp); 697 698 if (imgp->vp != NULL) { 699 NDFREE(ndp, NDF_ONLY_PNBUF); 700 vput(imgp->vp); 701 } 702 703 if (imgp->stringbase != NULL) 704 kmem_free_wakeup(exec_map, (vm_offset_t)imgp->stringbase, 705 ARG_MAX); 706 707 if (imgp->object != NULL) 708 vm_object_deallocate(imgp->object); 709 710 if (error == 0) { 711 /* 712 * Stop the process here if its stop event mask has 713 * the S_EXEC bit set. 714 */ 715 STOPEVENT(p, S_EXEC, 0); 716 goto done2; 717 } 718 719 exec_fail: 720 /* we're done here, clear P_INEXEC */ 721 PROC_LOCK(p); 722 p->p_flag &= ~P_INEXEC; 723 PROC_UNLOCK(p); 724 725 if (imgp->vmspace_destroyed) { 726 /* sorry, no more process anymore. exit gracefully */ 727 #ifdef MAC 728 mac_execve_exit(imgp); 729 if (interplabel != NULL) 730 mac_vnode_label_free(interplabel); 731 #endif 732 exit1(td, W_EXITCODE(0, SIGABRT)); 733 /* NOT REACHED */ 734 error = 0; 735 } 736 done2: 737 #ifdef MAC 738 mac_execve_exit(imgp); 739 if (interplabel != NULL) 740 mac_vnode_label_free(interplabel); 741 #endif 742 mtx_unlock(&Giant); 743 return (error); 744 } 745 746 int 747 exec_map_first_page(imgp) 748 struct image_params *imgp; 749 { 750 int rv, i; 751 int initial_pagein; 752 vm_page_t ma[VM_INITIAL_PAGEIN]; 753 vm_object_t object; 754 755 GIANT_REQUIRED; 756 757 if (imgp->firstpage != NULL) 758 exec_unmap_first_page(imgp); 759 760 VOP_GETVOBJECT(imgp->vp, &object); 761 VM_OBJECT_LOCK(object); 762 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 763 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 764 initial_pagein = VM_INITIAL_PAGEIN; 765 if (initial_pagein > object->size) 766 initial_pagein = object->size; 767 for (i = 1; i < initial_pagein; i++) { 768 if ((ma[i] = vm_page_lookup(object, i)) != NULL) { 769 if (ma[i]->valid) 770 break; 771 vm_page_lock_queues(); 772 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) { 773 vm_page_unlock_queues(); 774 break; 775 } 776 vm_page_busy(ma[i]); 777 vm_page_unlock_queues(); 778 } else { 779 ma[i] = vm_page_alloc(object, i, 780 VM_ALLOC_NORMAL); 781 if (ma[i] == NULL) 782 break; 783 } 784 } 785 initial_pagein = i; 786 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 787 ma[0] = vm_page_lookup(object, 0); 788 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || 789 (ma[0]->valid == 0)) { 790 if (ma[0]) { 791 vm_page_lock_queues(); 792 pmap_remove_all(ma[0]); 793 vm_page_free(ma[0]); 794 vm_page_unlock_queues(); 795 } 796 VM_OBJECT_UNLOCK(object); 797 return (EIO); 798 } 799 } 800 vm_page_lock_queues(); 801 vm_page_hold(ma[0]); 802 vm_page_wakeup(ma[0]); 803 vm_page_unlock_queues(); 804 VM_OBJECT_UNLOCK(object); 805 806 imgp->firstpage = sf_buf_alloc(ma[0], 0); 807 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 808 809 return (0); 810 } 811 812 void 813 exec_unmap_first_page(imgp) 814 struct image_params *imgp; 815 { 816 vm_page_t m; 817 818 if (imgp->firstpage != NULL) { 819 m = sf_buf_page(imgp->firstpage); 820 sf_buf_free(imgp->firstpage); 821 imgp->firstpage = NULL; 822 vm_page_lock_queues(); 823 vm_page_unhold(m); 824 vm_page_unlock_queues(); 825 } 826 } 827 828 /* 829 * Destroy old address space, and allocate a new stack 830 * The new stack is only SGROWSIZ large because it is grown 831 * automatically in trap.c. 832 */ 833 int 834 exec_new_vmspace(imgp, sv) 835 struct image_params *imgp; 836 struct sysentvec *sv; 837 { 838 int error; 839 struct proc *p = imgp->proc; 840 struct vmspace *vmspace = p->p_vmspace; 841 vm_offset_t stack_addr; 842 vm_map_t map; 843 844 GIANT_REQUIRED; 845 846 imgp->vmspace_destroyed = 1; 847 848 /* Called with Giant held, do not depend on it! */ 849 EVENTHANDLER_INVOKE(process_exec, p); 850 851 /* 852 * Here is as good a place as any to do any resource limit cleanups. 853 * This is needed if a 64 bit binary exec's a 32 bit binary - the 854 * data size limit may need to be changed to a value that makes 855 * sense for the 32 bit binary. 856 */ 857 if (sv->sv_fixlimits != NULL) 858 sv->sv_fixlimits(imgp); 859 860 /* 861 * Blow away entire process VM, if address space not shared, 862 * otherwise, create a new VM space so that other threads are 863 * not disrupted 864 */ 865 map = &vmspace->vm_map; 866 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser && 867 vm_map_max(map) == sv->sv_maxuser) { 868 shmexit(vmspace); 869 vm_page_lock_queues(); 870 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map), 871 vm_map_max(map)); 872 vm_page_unlock_queues(); 873 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 874 } else { 875 vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser); 876 vmspace = p->p_vmspace; 877 map = &vmspace->vm_map; 878 } 879 880 /* Allocate a new stack */ 881 stack_addr = sv->sv_usrstack - maxssiz; 882 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz, 883 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 884 if (error) 885 return (error); 886 887 #ifdef __ia64__ 888 /* Allocate a new register stack */ 889 stack_addr = IA64_BACKINGSTORE; 890 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz, 891 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 892 if (error) 893 return (error); 894 #endif 895 896 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 897 * VM_STACK case, but they are still used to monitor the size of the 898 * process stack so we can check the stack rlimit. 899 */ 900 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 901 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz; 902 903 return (0); 904 } 905 906 /* 907 * Copy out argument and environment strings from the old process 908 * address space into the temporary string buffer. 909 */ 910 int 911 exec_extract_strings(imgp) 912 struct image_params *imgp; 913 { 914 char **argv, **envv; 915 char *argp, *envp; 916 int error; 917 size_t length; 918 919 /* 920 * extract arguments first 921 */ 922 923 argv = imgp->userspace_argv; 924 925 if (argv) { 926 argp = (caddr_t)(intptr_t)fuword(argv); 927 if (argp == (caddr_t)-1) 928 return (EFAULT); 929 if (argp) 930 argv++; 931 if (imgp->argv0) 932 argp = imgp->argv0; 933 if (argp) { 934 do { 935 if (argp == (caddr_t)-1) 936 return (EFAULT); 937 if ((error = copyinstr(argp, imgp->stringp, 938 imgp->stringspace, &length))) { 939 if (error == ENAMETOOLONG) 940 return (E2BIG); 941 return (error); 942 } 943 imgp->stringspace -= length; 944 imgp->stringp += length; 945 imgp->argc++; 946 } while ((argp = (caddr_t)(intptr_t)fuword(argv++))); 947 } 948 } else 949 return (EFAULT); 950 951 imgp->endargs = imgp->stringp; 952 953 /* 954 * extract environment strings 955 */ 956 957 envv = imgp->userspace_envv; 958 959 if (envv) { 960 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 961 if (envp == (caddr_t)-1) 962 return (EFAULT); 963 if ((error = copyinstr(envp, imgp->stringp, 964 imgp->stringspace, &length))) { 965 if (error == ENAMETOOLONG) 966 return (E2BIG); 967 return (error); 968 } 969 imgp->stringspace -= length; 970 imgp->stringp += length; 971 imgp->envc++; 972 } 973 } 974 975 return (0); 976 } 977 978 /* 979 * Copy strings out to the new process address space, constructing 980 * new arg and env vector tables. Return a pointer to the base 981 * so that it can be used as the initial stack pointer. 982 */ 983 register_t * 984 exec_copyout_strings(imgp) 985 struct image_params *imgp; 986 { 987 int argc, envc; 988 char **vectp; 989 char *stringp, *destp; 990 register_t *stack_base; 991 struct ps_strings *arginfo; 992 struct proc *p; 993 int szsigcode; 994 995 /* 996 * Calculate string base and vector table pointers. 997 * Also deal with signal trampoline code for this exec type. 998 */ 999 p = imgp->proc; 1000 szsigcode = 0; 1001 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1002 if (p->p_sysent->sv_szsigcode != NULL) 1003 szsigcode = *(p->p_sysent->sv_szsigcode); 1004 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - 1005 roundup((ARG_MAX - imgp->stringspace), sizeof(char *)); 1006 1007 /* 1008 * install sigcode 1009 */ 1010 if (szsigcode) 1011 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - 1012 szsigcode), szsigcode); 1013 1014 /* 1015 * If we have a valid auxargs ptr, prepare some room 1016 * on the stack. 1017 */ 1018 if (imgp->auxargs) { 1019 /* 1020 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1021 * lower compatibility. 1022 */ 1023 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1024 (AT_COUNT * 2); 1025 /* 1026 * The '+ 2' is for the null pointers at the end of each of 1027 * the arg and env vector sets,and imgp->auxarg_size is room 1028 * for argument of Runtime loader. 1029 */ 1030 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2 + 1031 imgp->auxarg_size) * sizeof(char *)); 1032 1033 } else 1034 /* 1035 * The '+ 2' is for the null pointers at the end of each of 1036 * the arg and env vector sets 1037 */ 1038 vectp = (char **)(destp - (imgp->argc + imgp->envc + 2) * 1039 sizeof(char *)); 1040 1041 /* 1042 * vectp also becomes our initial stack base 1043 */ 1044 stack_base = (register_t *)vectp; 1045 1046 stringp = imgp->stringbase; 1047 argc = imgp->argc; 1048 envc = imgp->envc; 1049 1050 /* 1051 * Copy out strings - arguments and environment. 1052 */ 1053 copyout(stringp, destp, ARG_MAX - imgp->stringspace); 1054 1055 /* 1056 * Fill in "ps_strings" struct for ps, w, etc. 1057 */ 1058 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1059 suword(&arginfo->ps_nargvstr, argc); 1060 1061 /* 1062 * Fill in argument portion of vector table. 1063 */ 1064 for (; argc > 0; --argc) { 1065 suword(vectp++, (long)(intptr_t)destp); 1066 while (*stringp++ != 0) 1067 destp++; 1068 destp++; 1069 } 1070 1071 /* a null vector table pointer separates the argp's from the envp's */ 1072 suword(vectp++, 0); 1073 1074 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1075 suword(&arginfo->ps_nenvstr, envc); 1076 1077 /* 1078 * Fill in environment portion of vector table. 1079 */ 1080 for (; envc > 0; --envc) { 1081 suword(vectp++, (long)(intptr_t)destp); 1082 while (*stringp++ != 0) 1083 destp++; 1084 destp++; 1085 } 1086 1087 /* end of vector table is a null pointer */ 1088 suword(vectp, 0); 1089 1090 return (stack_base); 1091 } 1092 1093 /* 1094 * Check permissions of file to execute. 1095 * Called with imgp->vp locked. 1096 * Return 0 for success or error code on failure. 1097 */ 1098 int 1099 exec_check_permissions(imgp) 1100 struct image_params *imgp; 1101 { 1102 struct vnode *vp = imgp->vp; 1103 struct vattr *attr = imgp->attr; 1104 struct thread *td; 1105 int error; 1106 1107 td = curthread; /* XXXKSE */ 1108 1109 /* Get file attributes */ 1110 error = VOP_GETATTR(vp, attr, td->td_ucred, td); 1111 if (error) 1112 return (error); 1113 1114 #ifdef MAC 1115 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp); 1116 if (error) 1117 return (error); 1118 #endif 1119 1120 /* 1121 * 1) Check if file execution is disabled for the filesystem that this 1122 * file resides on. 1123 * 2) Insure that at least one execute bit is on - otherwise root 1124 * will always succeed, and we don't want to happen unless the 1125 * file really is executable. 1126 * 3) Insure that the file is a regular file. 1127 */ 1128 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1129 ((attr->va_mode & 0111) == 0) || 1130 (attr->va_type != VREG)) 1131 return (EACCES); 1132 1133 /* 1134 * Zero length files can't be exec'd 1135 */ 1136 if (attr->va_size == 0) 1137 return (ENOEXEC); 1138 1139 /* 1140 * Check for execute permission to file based on current credentials. 1141 */ 1142 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1143 if (error) 1144 return (error); 1145 1146 /* 1147 * Check number of open-for-writes on the file and deny execution 1148 * if there are any. 1149 */ 1150 if (vp->v_writecount) 1151 return (ETXTBSY); 1152 1153 /* 1154 * Call filesystem specific open routine (which does nothing in the 1155 * general case). 1156 */ 1157 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1); 1158 return (error); 1159 } 1160 1161 /* 1162 * Exec handler registration 1163 */ 1164 int 1165 exec_register(execsw_arg) 1166 const struct execsw *execsw_arg; 1167 { 1168 const struct execsw **es, **xs, **newexecsw; 1169 int count = 2; /* New slot and trailing NULL */ 1170 1171 if (execsw) 1172 for (es = execsw; *es; es++) 1173 count++; 1174 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1175 if (newexecsw == NULL) 1176 return (ENOMEM); 1177 xs = newexecsw; 1178 if (execsw) 1179 for (es = execsw; *es; es++) 1180 *xs++ = *es; 1181 *xs++ = execsw_arg; 1182 *xs = NULL; 1183 if (execsw) 1184 free(execsw, M_TEMP); 1185 execsw = newexecsw; 1186 return (0); 1187 } 1188 1189 int 1190 exec_unregister(execsw_arg) 1191 const struct execsw *execsw_arg; 1192 { 1193 const struct execsw **es, **xs, **newexecsw; 1194 int count = 1; 1195 1196 if (execsw == NULL) 1197 panic("unregister with no handlers left?\n"); 1198 1199 for (es = execsw; *es; es++) { 1200 if (*es == execsw_arg) 1201 break; 1202 } 1203 if (*es == NULL) 1204 return (ENOENT); 1205 for (es = execsw; *es; es++) 1206 if (*es != execsw_arg) 1207 count++; 1208 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1209 if (newexecsw == NULL) 1210 return (ENOMEM); 1211 xs = newexecsw; 1212 for (es = execsw; *es; es++) 1213 if (*es != execsw_arg) 1214 *xs++ = *es; 1215 *xs = NULL; 1216 if (execsw) 1217 free(execsw, M_TEMP); 1218 execsw = newexecsw; 1219 return (0); 1220 } 1221