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_capsicum.h" 31 #include "opt_hwpmc_hooks.h" 32 #include "opt_ktrace.h" 33 #include "opt_vm.h" 34 35 #include <sys/param.h> 36 #include <sys/capsicum.h> 37 #include <sys/systm.h> 38 #include <sys/eventhandler.h> 39 #include <sys/lock.h> 40 #include <sys/mutex.h> 41 #include <sys/sysproto.h> 42 #include <sys/signalvar.h> 43 #include <sys/kernel.h> 44 #include <sys/mount.h> 45 #include <sys/filedesc.h> 46 #include <sys/fcntl.h> 47 #include <sys/acct.h> 48 #include <sys/exec.h> 49 #include <sys/imgact.h> 50 #include <sys/imgact_elf.h> 51 #include <sys/wait.h> 52 #include <sys/malloc.h> 53 #include <sys/priv.h> 54 #include <sys/proc.h> 55 #include <sys/pioctl.h> 56 #include <sys/namei.h> 57 #include <sys/resourcevar.h> 58 #include <sys/rwlock.h> 59 #include <sys/sched.h> 60 #include <sys/sdt.h> 61 #include <sys/sf_buf.h> 62 #include <sys/syscallsubr.h> 63 #include <sys/sysent.h> 64 #include <sys/shm.h> 65 #include <sys/sysctl.h> 66 #include <sys/vnode.h> 67 #include <sys/stat.h> 68 #ifdef KTRACE 69 #include <sys/ktrace.h> 70 #endif 71 72 #include <vm/vm.h> 73 #include <vm/vm_param.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_page.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_kern.h> 78 #include <vm/vm_extern.h> 79 #include <vm/vm_object.h> 80 #include <vm/vm_pager.h> 81 82 #ifdef HWPMC_HOOKS 83 #include <sys/pmckern.h> 84 #endif 85 86 #include <machine/reg.h> 87 88 #include <security/audit/audit.h> 89 #include <security/mac/mac_framework.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 dtrace_execexit_func_t dtrace_fasttrap_exec; 94 #endif 95 96 SDT_PROVIDER_DECLARE(proc); 97 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *"); 98 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int"); 99 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *"); 100 101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 102 103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 106 static int do_execve(struct thread *td, struct image_args *args, 107 struct mac *mac_p); 108 109 /* XXX This should be vm_size_t. */ 110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 111 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 112 113 /* XXX This should be vm_size_t. */ 114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 115 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", ""); 116 117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 118 NULL, 0, sysctl_kern_stackprot, "I", ""); 119 120 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 122 &ps_arg_cache_limit, 0, ""); 123 124 static int disallow_high_osrel; 125 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 126 &disallow_high_osrel, 0, 127 "Disallow execution of binaries built for higher version of the world"); 128 129 static int map_at_zero = 0; 130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 131 "Permit processes to map an object at virtual address 0."); 132 133 static int 134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 135 { 136 struct proc *p; 137 int error; 138 139 p = curproc; 140 #ifdef SCTL_MASK32 141 if (req->flags & SCTL_MASK32) { 142 unsigned int val; 143 val = (unsigned int)p->p_sysent->sv_psstrings; 144 error = SYSCTL_OUT(req, &val, sizeof(val)); 145 } else 146 #endif 147 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 148 sizeof(p->p_sysent->sv_psstrings)); 149 return error; 150 } 151 152 static int 153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 154 { 155 struct proc *p; 156 int error; 157 158 p = curproc; 159 #ifdef SCTL_MASK32 160 if (req->flags & SCTL_MASK32) { 161 unsigned int val; 162 val = (unsigned int)p->p_sysent->sv_usrstack; 163 error = SYSCTL_OUT(req, &val, sizeof(val)); 164 } else 165 #endif 166 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 167 sizeof(p->p_sysent->sv_usrstack)); 168 return error; 169 } 170 171 static int 172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 173 { 174 struct proc *p; 175 176 p = curproc; 177 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 178 sizeof(p->p_sysent->sv_stackprot))); 179 } 180 181 /* 182 * Each of the items is a pointer to a `const struct execsw', hence the 183 * double pointer here. 184 */ 185 static const struct execsw **execsw; 186 187 #ifndef _SYS_SYSPROTO_H_ 188 struct execve_args { 189 char *fname; 190 char **argv; 191 char **envv; 192 }; 193 #endif 194 195 int 196 sys_execve(struct thread *td, struct execve_args *uap) 197 { 198 struct image_args args; 199 struct vmspace *oldvmspace; 200 int error; 201 202 error = pre_execve(td, &oldvmspace); 203 if (error != 0) 204 return (error); 205 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 206 uap->argv, uap->envv); 207 if (error == 0) 208 error = kern_execve(td, &args, NULL); 209 post_execve(td, error, oldvmspace); 210 return (error); 211 } 212 213 #ifndef _SYS_SYSPROTO_H_ 214 struct fexecve_args { 215 int fd; 216 char **argv; 217 char **envv; 218 } 219 #endif 220 int 221 sys_fexecve(struct thread *td, struct fexecve_args *uap) 222 { 223 struct image_args args; 224 struct vmspace *oldvmspace; 225 int error; 226 227 error = pre_execve(td, &oldvmspace); 228 if (error != 0) 229 return (error); 230 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 231 uap->argv, uap->envv); 232 if (error == 0) { 233 args.fd = uap->fd; 234 error = kern_execve(td, &args, NULL); 235 } 236 post_execve(td, error, oldvmspace); 237 return (error); 238 } 239 240 #ifndef _SYS_SYSPROTO_H_ 241 struct __mac_execve_args { 242 char *fname; 243 char **argv; 244 char **envv; 245 struct mac *mac_p; 246 }; 247 #endif 248 249 int 250 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 251 { 252 #ifdef MAC 253 struct image_args args; 254 struct vmspace *oldvmspace; 255 int error; 256 257 error = pre_execve(td, &oldvmspace); 258 if (error != 0) 259 return (error); 260 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 261 uap->argv, uap->envv); 262 if (error == 0) 263 error = kern_execve(td, &args, uap->mac_p); 264 post_execve(td, error, oldvmspace); 265 return (error); 266 #else 267 return (ENOSYS); 268 #endif 269 } 270 271 int 272 pre_execve(struct thread *td, struct vmspace **oldvmspace) 273 { 274 struct proc *p; 275 int error; 276 277 KASSERT(td == curthread, ("non-current thread %p", td)); 278 error = 0; 279 p = td->td_proc; 280 if ((p->p_flag & P_HADTHREADS) != 0) { 281 PROC_LOCK(p); 282 if (thread_single(p, SINGLE_BOUNDARY) != 0) 283 error = ERESTART; 284 PROC_UNLOCK(p); 285 } 286 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 287 ("nested execve")); 288 *oldvmspace = p->p_vmspace; 289 return (error); 290 } 291 292 void 293 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 294 { 295 struct proc *p; 296 297 KASSERT(td == curthread, ("non-current thread %p", td)); 298 p = td->td_proc; 299 if ((p->p_flag & P_HADTHREADS) != 0) { 300 PROC_LOCK(p); 301 /* 302 * If success, we upgrade to SINGLE_EXIT state to 303 * force other threads to suicide. 304 */ 305 if (error == 0) 306 thread_single(p, SINGLE_EXIT); 307 else 308 thread_single_end(p, SINGLE_BOUNDARY); 309 PROC_UNLOCK(p); 310 } 311 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 312 KASSERT(p->p_vmspace != oldvmspace, 313 ("oldvmspace still used")); 314 vmspace_free(oldvmspace); 315 td->td_pflags &= ~TDP_EXECVMSPC; 316 } 317 } 318 319 /* 320 * XXX: kern_execve has the astonishing property of not always returning to 321 * the caller. If sufficiently bad things happen during the call to 322 * do_execve(), it can end up calling exit1(); as a result, callers must 323 * avoid doing anything which they might need to undo (e.g., allocating 324 * memory). 325 */ 326 int 327 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 328 { 329 330 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 331 args->begin_envv - args->begin_argv); 332 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 333 args->endp - args->begin_envv); 334 return (do_execve(td, args, mac_p)); 335 } 336 337 /* 338 * In-kernel implementation of execve(). All arguments are assumed to be 339 * userspace pointers from the passed thread. 340 */ 341 static int 342 do_execve(td, args, mac_p) 343 struct thread *td; 344 struct image_args *args; 345 struct mac *mac_p; 346 { 347 struct proc *p = td->td_proc; 348 struct nameidata nd; 349 struct ucred *newcred = NULL, *oldcred; 350 struct uidinfo *euip = NULL; 351 register_t *stack_base; 352 int error, i; 353 struct image_params image_params, *imgp; 354 struct vattr attr; 355 int (*img_first)(struct image_params *); 356 struct pargs *oldargs = NULL, *newargs = NULL; 357 struct sigacts *oldsigacts, *newsigacts; 358 #ifdef KTRACE 359 struct vnode *tracevp = NULL; 360 struct ucred *tracecred = NULL; 361 #endif 362 struct vnode *oldtextvp = NULL, *newtextvp; 363 cap_rights_t rights; 364 int credential_changing; 365 int textset; 366 #ifdef MAC 367 struct label *interpvplabel = NULL; 368 int will_transition; 369 #endif 370 #ifdef HWPMC_HOOKS 371 struct pmckern_procexec pe; 372 #endif 373 static const char fexecv_proc_title[] = "(fexecv)"; 374 375 imgp = &image_params; 376 377 /* 378 * Lock the process and set the P_INEXEC flag to indicate that 379 * it should be left alone until we're done here. This is 380 * necessary to avoid race conditions - e.g. in ptrace() - 381 * that might allow a local user to illicitly obtain elevated 382 * privileges. 383 */ 384 PROC_LOCK(p); 385 KASSERT((p->p_flag & P_INEXEC) == 0, 386 ("%s(): process already has P_INEXEC flag", __func__)); 387 p->p_flag |= P_INEXEC; 388 PROC_UNLOCK(p); 389 390 /* 391 * Initialize part of the common data 392 */ 393 bzero(imgp, sizeof(*imgp)); 394 imgp->proc = p; 395 imgp->attr = &attr; 396 imgp->args = args; 397 398 #ifdef MAC 399 error = mac_execve_enter(imgp, mac_p); 400 if (error) 401 goto exec_fail; 402 #endif 403 404 /* 405 * Translate the file name. namei() returns a vnode pointer 406 * in ni_vp amoung other things. 407 * 408 * XXXAUDIT: It would be desirable to also audit the name of the 409 * interpreter if this is an interpreted binary. 410 */ 411 if (args->fname != NULL) { 412 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 413 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 414 } 415 416 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); 417 418 interpret: 419 if (args->fname != NULL) { 420 #ifdef CAPABILITY_MODE 421 /* 422 * While capability mode can't reach this point via direct 423 * path arguments to execve(), we also don't allow 424 * interpreters to be used in capability mode (for now). 425 * Catch indirect lookups and return a permissions error. 426 */ 427 if (IN_CAPABILITY_MODE(td)) { 428 error = ECAPMODE; 429 goto exec_fail; 430 } 431 #endif 432 error = namei(&nd); 433 if (error) 434 goto exec_fail; 435 436 newtextvp = nd.ni_vp; 437 imgp->vp = newtextvp; 438 } else { 439 AUDIT_ARG_FD(args->fd); 440 /* 441 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 442 */ 443 error = fgetvp_exec(td, args->fd, 444 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp); 445 if (error) 446 goto exec_fail; 447 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY); 448 AUDIT_ARG_VNODE1(newtextvp); 449 imgp->vp = newtextvp; 450 } 451 452 /* 453 * Check file permissions (also 'opens' file) 454 */ 455 error = exec_check_permissions(imgp); 456 if (error) 457 goto exec_fail_dealloc; 458 459 imgp->object = imgp->vp->v_object; 460 if (imgp->object != NULL) 461 vm_object_reference(imgp->object); 462 463 /* 464 * Set VV_TEXT now so no one can write to the executable while we're 465 * activating it. 466 * 467 * Remember if this was set before and unset it in case this is not 468 * actually an executable image. 469 */ 470 textset = VOP_IS_TEXT(imgp->vp); 471 VOP_SET_TEXT(imgp->vp); 472 473 error = exec_map_first_page(imgp); 474 if (error) 475 goto exec_fail_dealloc; 476 477 imgp->proc->p_osrel = 0; 478 /* 479 * If the current process has a special image activator it 480 * wants to try first, call it. For example, emulating shell 481 * scripts differently. 482 */ 483 error = -1; 484 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 485 error = img_first(imgp); 486 487 /* 488 * Loop through the list of image activators, calling each one. 489 * An activator returns -1 if there is no match, 0 on success, 490 * and an error otherwise. 491 */ 492 for (i = 0; error == -1 && execsw[i]; ++i) { 493 if (execsw[i]->ex_imgact == NULL || 494 execsw[i]->ex_imgact == img_first) { 495 continue; 496 } 497 error = (*execsw[i]->ex_imgact)(imgp); 498 } 499 500 if (error) { 501 if (error == -1) { 502 if (textset == 0) 503 VOP_UNSET_TEXT(imgp->vp); 504 error = ENOEXEC; 505 } 506 goto exec_fail_dealloc; 507 } 508 509 /* 510 * Special interpreter operation, cleanup and loop up to try to 511 * activate the interpreter. 512 */ 513 if (imgp->interpreted) { 514 exec_unmap_first_page(imgp); 515 /* 516 * VV_TEXT needs to be unset for scripts. There is a short 517 * period before we determine that something is a script where 518 * VV_TEXT will be set. The vnode lock is held over this 519 * entire period so nothing should illegitimately be blocked. 520 */ 521 VOP_UNSET_TEXT(imgp->vp); 522 /* free name buffer and old vnode */ 523 if (args->fname != NULL) 524 NDFREE(&nd, NDF_ONLY_PNBUF); 525 #ifdef MAC 526 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 527 #endif 528 if (imgp->opened) { 529 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 530 imgp->opened = 0; 531 } 532 vput(newtextvp); 533 vm_object_deallocate(imgp->object); 534 imgp->object = NULL; 535 /* set new name to that of the interpreter */ 536 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 537 UIO_SYSSPACE, imgp->interpreter_name, td); 538 args->fname = imgp->interpreter_name; 539 goto interpret; 540 } 541 542 /* 543 * NB: We unlock the vnode here because it is believed that none 544 * of the sv_copyout_strings/sv_fixup operations require the vnode. 545 */ 546 VOP_UNLOCK(imgp->vp, 0); 547 548 /* 549 * Do the best to calculate the full path to the image file. 550 */ 551 if (imgp->auxargs != NULL && 552 ((args->fname != NULL && args->fname[0] == '/') || 553 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0)) 554 imgp->execpath = args->fname; 555 556 if (disallow_high_osrel && 557 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 558 error = ENOEXEC; 559 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 560 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 561 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 562 goto exec_fail_dealloc; 563 } 564 565 /* 566 * Copy out strings (args and env) and initialize stack base 567 */ 568 if (p->p_sysent->sv_copyout_strings) 569 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 570 else 571 stack_base = exec_copyout_strings(imgp); 572 573 /* 574 * If custom stack fixup routine present for this process 575 * let it do the stack setup. 576 * Else stuff argument count as first item on stack 577 */ 578 if (p->p_sysent->sv_fixup != NULL) 579 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 580 else 581 suword(--stack_base, imgp->args->argc); 582 583 if (args->fdp != NULL) { 584 /* Install a brand new file descriptor table. */ 585 fdinstall_remapped(td, args->fdp); 586 args->fdp = NULL; 587 } else { 588 /* 589 * Keep on using the existing file descriptor table. For 590 * security and other reasons, the file descriptor table 591 * cannot be shared after an exec. 592 */ 593 fdunshare(td); 594 /* close files on exec */ 595 fdcloseexec(td); 596 } 597 598 /* 599 * Malloc things before we need locks. 600 */ 601 i = imgp->args->begin_envv - imgp->args->begin_argv; 602 /* Cache arguments if they fit inside our allowance */ 603 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 604 newargs = pargs_alloc(i); 605 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 606 } 607 608 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 609 610 /* 611 * For security and other reasons, signal handlers cannot 612 * be shared after an exec. The new process gets a copy of the old 613 * handlers. In execsigs(), the new process will have its signals 614 * reset. 615 */ 616 if (sigacts_shared(p->p_sigacts)) { 617 oldsigacts = p->p_sigacts; 618 newsigacts = sigacts_alloc(); 619 sigacts_copy(newsigacts, oldsigacts); 620 } else { 621 oldsigacts = NULL; 622 newsigacts = NULL; /* satisfy gcc */ 623 } 624 625 PROC_LOCK(p); 626 if (oldsigacts) 627 p->p_sigacts = newsigacts; 628 oldcred = p->p_ucred; 629 /* Stop profiling */ 630 stopprofclock(p); 631 632 /* reset caught signals */ 633 execsigs(p); 634 635 /* name this process - nameiexec(p, ndp) */ 636 bzero(p->p_comm, sizeof(p->p_comm)); 637 if (args->fname) 638 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 639 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 640 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 641 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 642 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 643 #ifdef KTR 644 sched_clear_tdname(td); 645 #endif 646 647 /* 648 * mark as execed, wakeup the process that vforked (if any) and tell 649 * it that it now has its own resources back 650 */ 651 p->p_flag |= P_EXEC; 652 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 653 p->p_flag2 &= ~P2_NOTRACE; 654 if (p->p_flag & P_PPWAIT) { 655 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 656 cv_broadcast(&p->p_pwait); 657 } 658 659 /* 660 * Implement image setuid/setgid. 661 * 662 * Don't honor setuid/setgid if the filesystem prohibits it or if 663 * the process is being traced. 664 * 665 * We disable setuid/setgid/etc in compatibility mode on the basis 666 * that most setugid applications are not written with that 667 * environment in mind, and will therefore almost certainly operate 668 * incorrectly. In principle there's no reason that setugid 669 * applications might not be useful in capability mode, so we may want 670 * to reconsider this conservative design choice in the future. 671 * 672 * XXXMAC: For the time being, use NOSUID to also prohibit 673 * transitions on the file system. 674 */ 675 credential_changing = 0; 676 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != 677 attr.va_uid; 678 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != 679 attr.va_gid; 680 #ifdef MAC 681 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 682 interpvplabel, imgp); 683 credential_changing |= will_transition; 684 #endif 685 686 if (credential_changing && 687 #ifdef CAPABILITY_MODE 688 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 689 #endif 690 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 691 (p->p_flag & P_TRACED) == 0) { 692 /* 693 * Turn off syscall tracing for set-id programs, except for 694 * root. Record any set-id flags first to make sure that 695 * we do not regain any tracing during a possible block. 696 */ 697 setsugid(p); 698 699 #ifdef KTRACE 700 if (p->p_tracecred != NULL && 701 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 702 ktrprocexec(p, &tracecred, &tracevp); 703 #endif 704 /* 705 * Close any file descriptors 0..2 that reference procfs, 706 * then make sure file descriptors 0..2 are in use. 707 * 708 * Both fdsetugidsafety() and fdcheckstd() may call functions 709 * taking sleepable locks, so temporarily drop our locks. 710 */ 711 PROC_UNLOCK(p); 712 VOP_UNLOCK(imgp->vp, 0); 713 fdsetugidsafety(td); 714 error = fdcheckstd(td); 715 if (error != 0) 716 goto done1; 717 newcred = crdup(oldcred); 718 euip = uifind(attr.va_uid); 719 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 720 PROC_LOCK(p); 721 /* 722 * Set the new credentials. 723 */ 724 if (attr.va_mode & S_ISUID) 725 change_euid(newcred, euip); 726 if (attr.va_mode & S_ISGID) 727 change_egid(newcred, attr.va_gid); 728 #ifdef MAC 729 if (will_transition) { 730 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 731 interpvplabel, imgp); 732 } 733 #endif 734 /* 735 * Implement correct POSIX saved-id behavior. 736 * 737 * XXXMAC: Note that the current logic will save the 738 * uid and gid if a MAC domain transition occurs, even 739 * though maybe it shouldn't. 740 */ 741 change_svuid(newcred, newcred->cr_uid); 742 change_svgid(newcred, newcred->cr_gid); 743 proc_set_cred(p, newcred); 744 } else { 745 if (oldcred->cr_uid == oldcred->cr_ruid && 746 oldcred->cr_gid == oldcred->cr_rgid) 747 p->p_flag &= ~P_SUGID; 748 /* 749 * Implement correct POSIX saved-id behavior. 750 * 751 * XXX: It's not clear that the existing behavior is 752 * POSIX-compliant. A number of sources indicate that the 753 * saved uid/gid should only be updated if the new ruid is 754 * not equal to the old ruid, or the new euid is not equal 755 * to the old euid and the new euid is not equal to the old 756 * ruid. The FreeBSD code always updates the saved uid/gid. 757 * Also, this code uses the new (replaced) euid and egid as 758 * the source, which may or may not be the right ones to use. 759 */ 760 if (oldcred->cr_svuid != oldcred->cr_uid || 761 oldcred->cr_svgid != oldcred->cr_gid) { 762 PROC_UNLOCK(p); 763 VOP_UNLOCK(imgp->vp, 0); 764 newcred = crdup(oldcred); 765 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 766 PROC_LOCK(p); 767 change_svuid(newcred, newcred->cr_uid); 768 change_svgid(newcred, newcred->cr_gid); 769 proc_set_cred(p, newcred); 770 } 771 } 772 773 /* 774 * Store the vp for use in procfs. This vnode was referenced by namei 775 * or fgetvp_exec. 776 */ 777 oldtextvp = p->p_textvp; 778 p->p_textvp = newtextvp; 779 780 #ifdef KDTRACE_HOOKS 781 /* 782 * Tell the DTrace fasttrap provider about the exec if it 783 * has declared an interest. 784 */ 785 if (dtrace_fasttrap_exec) 786 dtrace_fasttrap_exec(p); 787 #endif 788 789 /* 790 * Notify others that we exec'd, and clear the P_INEXEC flag 791 * as we're now a bona fide freshly-execed process. 792 */ 793 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 794 p->p_flag &= ~P_INEXEC; 795 796 /* clear "fork but no exec" flag, as we _are_ execing */ 797 p->p_acflag &= ~AFORK; 798 799 /* 800 * Free any previous argument cache and replace it with 801 * the new argument cache, if any. 802 */ 803 oldargs = p->p_args; 804 p->p_args = newargs; 805 newargs = NULL; 806 807 #ifdef HWPMC_HOOKS 808 /* 809 * Check if system-wide sampling is in effect or if the 810 * current process is using PMCs. If so, do exec() time 811 * processing. This processing needs to happen AFTER the 812 * P_INEXEC flag is cleared. 813 * 814 * The proc lock needs to be released before taking the PMC 815 * SX. 816 */ 817 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 818 PROC_UNLOCK(p); 819 VOP_UNLOCK(imgp->vp, 0); 820 pe.pm_credentialschanged = credential_changing; 821 pe.pm_entryaddr = imgp->entry_addr; 822 823 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 824 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 825 } else 826 PROC_UNLOCK(p); 827 #else /* !HWPMC_HOOKS */ 828 PROC_UNLOCK(p); 829 #endif 830 831 /* Set values passed into the program in registers. */ 832 if (p->p_sysent->sv_setregs) 833 (*p->p_sysent->sv_setregs)(td, imgp, 834 (u_long)(uintptr_t)stack_base); 835 else 836 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 837 838 vfs_mark_atime(imgp->vp, td->td_ucred); 839 840 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0); 841 842 VOP_UNLOCK(imgp->vp, 0); 843 done1: 844 /* 845 * Free any resources malloc'd earlier that we didn't use. 846 */ 847 if (euip != NULL) 848 uifree(euip); 849 if (newcred != NULL) 850 crfree(oldcred); 851 852 /* 853 * Handle deferred decrement of ref counts. 854 */ 855 if (oldtextvp != NULL) 856 vrele(oldtextvp); 857 #ifdef KTRACE 858 if (tracevp != NULL) 859 vrele(tracevp); 860 if (tracecred != NULL) 861 crfree(tracecred); 862 #endif 863 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 864 pargs_drop(oldargs); 865 pargs_drop(newargs); 866 if (oldsigacts != NULL) 867 sigacts_free(oldsigacts); 868 869 exec_fail_dealloc: 870 871 /* 872 * free various allocated resources 873 */ 874 if (imgp->firstpage != NULL) 875 exec_unmap_first_page(imgp); 876 877 if (imgp->vp != NULL) { 878 if (args->fname) 879 NDFREE(&nd, NDF_ONLY_PNBUF); 880 if (imgp->opened) 881 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 882 if (error != 0) 883 vput(imgp->vp); 884 else 885 VOP_UNLOCK(imgp->vp, 0); 886 } 887 888 if (imgp->object != NULL) 889 vm_object_deallocate(imgp->object); 890 891 free(imgp->freepath, M_TEMP); 892 893 if (error == 0) { 894 PROC_LOCK(p); 895 td->td_dbgflags |= TDB_EXEC; 896 PROC_UNLOCK(p); 897 898 /* 899 * Stop the process here if its stop event mask has 900 * the S_EXEC bit set. 901 */ 902 STOPEVENT(p, S_EXEC, 0); 903 goto done2; 904 } 905 906 exec_fail: 907 /* we're done here, clear P_INEXEC */ 908 PROC_LOCK(p); 909 p->p_flag &= ~P_INEXEC; 910 PROC_UNLOCK(p); 911 912 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 913 914 done2: 915 #ifdef MAC 916 mac_execve_exit(imgp); 917 mac_execve_interpreter_exit(interpvplabel); 918 #endif 919 exec_free_args(args); 920 921 if (error && imgp->vmspace_destroyed) { 922 /* sorry, no more process anymore. exit gracefully */ 923 exit1(td, 0, SIGABRT); 924 /* NOT REACHED */ 925 } 926 927 #ifdef KTRACE 928 if (error == 0) 929 ktrprocctor(p); 930 #endif 931 932 return (error); 933 } 934 935 int 936 exec_map_first_page(imgp) 937 struct image_params *imgp; 938 { 939 int rv, i; 940 int initial_pagein; 941 vm_page_t ma[VM_INITIAL_PAGEIN]; 942 vm_object_t object; 943 944 if (imgp->firstpage != NULL) 945 exec_unmap_first_page(imgp); 946 947 object = imgp->vp->v_object; 948 if (object == NULL) 949 return (EACCES); 950 VM_OBJECT_WLOCK(object); 951 #if VM_NRESERVLEVEL > 0 952 vm_object_color(object, 0); 953 #endif 954 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL); 955 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 956 initial_pagein = VM_INITIAL_PAGEIN; 957 if (initial_pagein > object->size) 958 initial_pagein = object->size; 959 for (i = 1; i < initial_pagein; i++) { 960 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 961 if (ma[i]->valid) 962 break; 963 if (vm_page_tryxbusy(ma[i])) 964 break; 965 } else { 966 ma[i] = vm_page_alloc(object, i, 967 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 968 if (ma[i] == NULL) 969 break; 970 } 971 } 972 initial_pagein = i; 973 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 974 if (rv != VM_PAGER_OK) { 975 vm_page_lock(ma[0]); 976 vm_page_free(ma[0]); 977 vm_page_unlock(ma[0]); 978 VM_OBJECT_WUNLOCK(object); 979 return (EIO); 980 } 981 } 982 vm_page_xunbusy(ma[0]); 983 vm_page_lock(ma[0]); 984 vm_page_hold(ma[0]); 985 vm_page_activate(ma[0]); 986 vm_page_unlock(ma[0]); 987 VM_OBJECT_WUNLOCK(object); 988 989 imgp->firstpage = sf_buf_alloc(ma[0], 0); 990 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 991 992 return (0); 993 } 994 995 void 996 exec_unmap_first_page(imgp) 997 struct image_params *imgp; 998 { 999 vm_page_t m; 1000 1001 if (imgp->firstpage != NULL) { 1002 m = sf_buf_page(imgp->firstpage); 1003 sf_buf_free(imgp->firstpage); 1004 imgp->firstpage = NULL; 1005 vm_page_lock(m); 1006 vm_page_unhold(m); 1007 vm_page_unlock(m); 1008 } 1009 } 1010 1011 /* 1012 * Destroy old address space, and allocate a new stack 1013 * The new stack is only SGROWSIZ large because it is grown 1014 * automatically in trap.c. 1015 */ 1016 int 1017 exec_new_vmspace(imgp, sv) 1018 struct image_params *imgp; 1019 struct sysentvec *sv; 1020 { 1021 int error; 1022 struct proc *p = imgp->proc; 1023 struct vmspace *vmspace = p->p_vmspace; 1024 vm_object_t obj; 1025 struct rlimit rlim_stack; 1026 vm_offset_t sv_minuser, stack_addr; 1027 vm_map_t map; 1028 u_long ssiz; 1029 1030 imgp->vmspace_destroyed = 1; 1031 imgp->sysent = sv; 1032 1033 /* May be called with Giant held */ 1034 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1035 1036 /* 1037 * Blow away entire process VM, if address space not shared, 1038 * otherwise, create a new VM space so that other threads are 1039 * not disrupted 1040 */ 1041 map = &vmspace->vm_map; 1042 if (map_at_zero) 1043 sv_minuser = sv->sv_minuser; 1044 else 1045 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1046 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1047 vm_map_max(map) == sv->sv_maxuser) { 1048 shmexit(vmspace); 1049 pmap_remove_pages(vmspace_pmap(vmspace)); 1050 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1051 } else { 1052 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1053 if (error) 1054 return (error); 1055 vmspace = p->p_vmspace; 1056 map = &vmspace->vm_map; 1057 } 1058 1059 /* Map a shared page */ 1060 obj = sv->sv_shared_page_obj; 1061 if (obj != NULL) { 1062 vm_object_reference(obj); 1063 error = vm_map_fixed(map, obj, 0, 1064 sv->sv_shared_page_base, sv->sv_shared_page_len, 1065 VM_PROT_READ | VM_PROT_EXECUTE, 1066 VM_PROT_READ | VM_PROT_EXECUTE, 1067 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1068 if (error) { 1069 vm_object_deallocate(obj); 1070 return (error); 1071 } 1072 } 1073 1074 /* Allocate a new stack */ 1075 if (imgp->stack_sz != 0) { 1076 ssiz = trunc_page(imgp->stack_sz); 1077 PROC_LOCK(p); 1078 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1079 PROC_UNLOCK(p); 1080 if (ssiz > rlim_stack.rlim_max) 1081 ssiz = rlim_stack.rlim_max; 1082 if (ssiz > rlim_stack.rlim_cur) { 1083 rlim_stack.rlim_cur = ssiz; 1084 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1085 } 1086 } else if (sv->sv_maxssiz != NULL) { 1087 ssiz = *sv->sv_maxssiz; 1088 } else { 1089 ssiz = maxssiz; 1090 } 1091 stack_addr = sv->sv_usrstack - ssiz; 1092 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1093 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1094 sv->sv_stackprot, 1095 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1096 if (error) 1097 return (error); 1098 1099 /* 1100 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1101 * are still used to enforce the stack rlimit on the process stack. 1102 */ 1103 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1104 vmspace->vm_maxsaddr = (char *)stack_addr; 1105 1106 return (0); 1107 } 1108 1109 /* 1110 * Copy out argument and environment strings from the old process address 1111 * space into the temporary string buffer. 1112 */ 1113 int 1114 exec_copyin_args(struct image_args *args, char *fname, 1115 enum uio_seg segflg, char **argv, char **envv) 1116 { 1117 u_long argp, envp; 1118 int error; 1119 size_t length; 1120 1121 bzero(args, sizeof(*args)); 1122 if (argv == NULL) 1123 return (EFAULT); 1124 1125 /* 1126 * Allocate demand-paged memory for the file name, argument, and 1127 * environment strings. 1128 */ 1129 error = exec_alloc_args(args); 1130 if (error != 0) 1131 return (error); 1132 1133 /* 1134 * Copy the file name. 1135 */ 1136 if (fname != NULL) { 1137 args->fname = args->buf; 1138 error = (segflg == UIO_SYSSPACE) ? 1139 copystr(fname, args->fname, PATH_MAX, &length) : 1140 copyinstr(fname, args->fname, PATH_MAX, &length); 1141 if (error != 0) 1142 goto err_exit; 1143 } else 1144 length = 0; 1145 1146 args->begin_argv = args->buf + length; 1147 args->endp = args->begin_argv; 1148 args->stringspace = ARG_MAX; 1149 1150 /* 1151 * extract arguments first 1152 */ 1153 for (;;) { 1154 error = fueword(argv++, &argp); 1155 if (error == -1) { 1156 error = EFAULT; 1157 goto err_exit; 1158 } 1159 if (argp == 0) 1160 break; 1161 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1162 args->stringspace, &length); 1163 if (error != 0) { 1164 if (error == ENAMETOOLONG) 1165 error = E2BIG; 1166 goto err_exit; 1167 } 1168 args->stringspace -= length; 1169 args->endp += length; 1170 args->argc++; 1171 } 1172 1173 args->begin_envv = args->endp; 1174 1175 /* 1176 * extract environment strings 1177 */ 1178 if (envv) { 1179 for (;;) { 1180 error = fueword(envv++, &envp); 1181 if (error == -1) { 1182 error = EFAULT; 1183 goto err_exit; 1184 } 1185 if (envp == 0) 1186 break; 1187 error = copyinstr((void *)(uintptr_t)envp, 1188 args->endp, args->stringspace, &length); 1189 if (error != 0) { 1190 if (error == ENAMETOOLONG) 1191 error = E2BIG; 1192 goto err_exit; 1193 } 1194 args->stringspace -= length; 1195 args->endp += length; 1196 args->envc++; 1197 } 1198 } 1199 1200 return (0); 1201 1202 err_exit: 1203 exec_free_args(args); 1204 return (error); 1205 } 1206 1207 int 1208 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1209 const void *data, size_t datalen, const int *fds, size_t fdslen) 1210 { 1211 struct filedesc *ofdp; 1212 const char *p; 1213 int *kfds; 1214 int error; 1215 1216 memset(args, '\0', sizeof(*args)); 1217 ofdp = td->td_proc->p_fd; 1218 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1219 return (E2BIG); 1220 error = exec_alloc_args(args); 1221 if (error != 0) 1222 return (error); 1223 1224 args->begin_argv = args->buf; 1225 args->stringspace = ARG_MAX; 1226 1227 if (datalen > 0) { 1228 /* 1229 * Argument buffer has been provided. Copy it into the 1230 * kernel as a single string and add a terminating null 1231 * byte. 1232 */ 1233 error = copyin(data, args->begin_argv, datalen); 1234 if (error != 0) 1235 goto err_exit; 1236 args->begin_argv[datalen] = '\0'; 1237 args->endp = args->begin_argv + datalen + 1; 1238 args->stringspace -= datalen + 1; 1239 1240 /* 1241 * Traditional argument counting. Count the number of 1242 * null bytes. 1243 */ 1244 for (p = args->begin_argv; p < args->endp; ++p) 1245 if (*p == '\0') 1246 ++args->argc; 1247 } else { 1248 /* No argument buffer provided. */ 1249 args->endp = args->begin_argv; 1250 } 1251 /* There are no environment variables. */ 1252 args->begin_envv = args->endp; 1253 1254 /* Create new file descriptor table. */ 1255 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1256 error = copyin(fds, kfds, fdslen * sizeof(int)); 1257 if (error != 0) { 1258 free(kfds, M_TEMP); 1259 goto err_exit; 1260 } 1261 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1262 free(kfds, M_TEMP); 1263 if (error != 0) 1264 goto err_exit; 1265 1266 return (0); 1267 err_exit: 1268 exec_free_args(args); 1269 return (error); 1270 } 1271 1272 /* 1273 * Allocate temporary demand-paged, zero-filled memory for the file name, 1274 * argument, and environment strings. Returns zero if the allocation succeeds 1275 * and ENOMEM otherwise. 1276 */ 1277 int 1278 exec_alloc_args(struct image_args *args) 1279 { 1280 1281 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 1282 return (args->buf != NULL ? 0 : ENOMEM); 1283 } 1284 1285 void 1286 exec_free_args(struct image_args *args) 1287 { 1288 1289 if (args->buf != NULL) { 1290 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf, 1291 PATH_MAX + ARG_MAX); 1292 args->buf = NULL; 1293 } 1294 if (args->fname_buf != NULL) { 1295 free(args->fname_buf, M_TEMP); 1296 args->fname_buf = NULL; 1297 } 1298 if (args->fdp != NULL) 1299 fdescfree_remapped(args->fdp); 1300 } 1301 1302 /* 1303 * Copy strings out to the new process address space, constructing new arg 1304 * and env vector tables. Return a pointer to the base so that it can be used 1305 * as the initial stack pointer. 1306 */ 1307 register_t * 1308 exec_copyout_strings(imgp) 1309 struct image_params *imgp; 1310 { 1311 int argc, envc; 1312 char **vectp; 1313 char *stringp; 1314 uintptr_t destp; 1315 register_t *stack_base; 1316 struct ps_strings *arginfo; 1317 struct proc *p; 1318 size_t execpath_len; 1319 int szsigcode, szps; 1320 char canary[sizeof(long) * 8]; 1321 1322 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1323 /* 1324 * Calculate string base and vector table pointers. 1325 * Also deal with signal trampoline code for this exec type. 1326 */ 1327 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1328 execpath_len = strlen(imgp->execpath) + 1; 1329 else 1330 execpath_len = 0; 1331 p = imgp->proc; 1332 szsigcode = 0; 1333 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1334 if (p->p_sysent->sv_sigcode_base == 0) { 1335 if (p->p_sysent->sv_szsigcode != NULL) 1336 szsigcode = *(p->p_sysent->sv_szsigcode); 1337 } 1338 destp = (uintptr_t)arginfo; 1339 1340 /* 1341 * install sigcode 1342 */ 1343 if (szsigcode != 0) { 1344 destp -= szsigcode; 1345 destp = rounddown2(destp, sizeof(void *)); 1346 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1347 } 1348 1349 /* 1350 * Copy the image path for the rtld. 1351 */ 1352 if (execpath_len != 0) { 1353 destp -= execpath_len; 1354 imgp->execpathp = destp; 1355 copyout(imgp->execpath, (void *)destp, execpath_len); 1356 } 1357 1358 /* 1359 * Prepare the canary for SSP. 1360 */ 1361 arc4rand(canary, sizeof(canary), 0); 1362 destp -= sizeof(canary); 1363 imgp->canary = destp; 1364 copyout(canary, (void *)destp, sizeof(canary)); 1365 imgp->canarylen = sizeof(canary); 1366 1367 /* 1368 * Prepare the pagesizes array. 1369 */ 1370 destp -= szps; 1371 destp = rounddown2(destp, sizeof(void *)); 1372 imgp->pagesizes = destp; 1373 copyout(pagesizes, (void *)destp, szps); 1374 imgp->pagesizeslen = szps; 1375 1376 destp -= ARG_MAX - imgp->args->stringspace; 1377 destp = rounddown2(destp, sizeof(void *)); 1378 1379 /* 1380 * If we have a valid auxargs ptr, prepare some room 1381 * on the stack. 1382 */ 1383 if (imgp->auxargs) { 1384 /* 1385 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1386 * lower compatibility. 1387 */ 1388 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1389 (AT_COUNT * 2); 1390 /* 1391 * The '+ 2' is for the null pointers at the end of each of 1392 * the arg and env vector sets,and imgp->auxarg_size is room 1393 * for argument of Runtime loader. 1394 */ 1395 vectp = (char **)(destp - (imgp->args->argc + 1396 imgp->args->envc + 2 + imgp->auxarg_size) 1397 * sizeof(char *)); 1398 } else { 1399 /* 1400 * The '+ 2' is for the null pointers at the end of each of 1401 * the arg and env vector sets 1402 */ 1403 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc 1404 + 2) * sizeof(char *)); 1405 } 1406 1407 /* 1408 * vectp also becomes our initial stack base 1409 */ 1410 stack_base = (register_t *)vectp; 1411 1412 stringp = imgp->args->begin_argv; 1413 argc = imgp->args->argc; 1414 envc = imgp->args->envc; 1415 1416 /* 1417 * Copy out strings - arguments and environment. 1418 */ 1419 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1420 1421 /* 1422 * Fill in "ps_strings" struct for ps, w, etc. 1423 */ 1424 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1425 suword32(&arginfo->ps_nargvstr, argc); 1426 1427 /* 1428 * Fill in argument portion of vector table. 1429 */ 1430 for (; argc > 0; --argc) { 1431 suword(vectp++, (long)(intptr_t)destp); 1432 while (*stringp++ != 0) 1433 destp++; 1434 destp++; 1435 } 1436 1437 /* a null vector table pointer separates the argp's from the envp's */ 1438 suword(vectp++, 0); 1439 1440 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1441 suword32(&arginfo->ps_nenvstr, envc); 1442 1443 /* 1444 * Fill in environment portion of vector table. 1445 */ 1446 for (; envc > 0; --envc) { 1447 suword(vectp++, (long)(intptr_t)destp); 1448 while (*stringp++ != 0) 1449 destp++; 1450 destp++; 1451 } 1452 1453 /* end of vector table is a null pointer */ 1454 suword(vectp, 0); 1455 1456 return (stack_base); 1457 } 1458 1459 /* 1460 * Check permissions of file to execute. 1461 * Called with imgp->vp locked. 1462 * Return 0 for success or error code on failure. 1463 */ 1464 int 1465 exec_check_permissions(imgp) 1466 struct image_params *imgp; 1467 { 1468 struct vnode *vp = imgp->vp; 1469 struct vattr *attr = imgp->attr; 1470 struct thread *td; 1471 int error, writecount; 1472 1473 td = curthread; 1474 1475 /* Get file attributes */ 1476 error = VOP_GETATTR(vp, attr, td->td_ucred); 1477 if (error) 1478 return (error); 1479 1480 #ifdef MAC 1481 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1482 if (error) 1483 return (error); 1484 #endif 1485 1486 /* 1487 * 1) Check if file execution is disabled for the filesystem that 1488 * this file resides on. 1489 * 2) Ensure that at least one execute bit is on. Otherwise, a 1490 * privileged user will always succeed, and we don't want this 1491 * to happen unless the file really is executable. 1492 * 3) Ensure that the file is a regular file. 1493 */ 1494 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1495 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1496 (attr->va_type != VREG)) 1497 return (EACCES); 1498 1499 /* 1500 * Zero length files can't be exec'd 1501 */ 1502 if (attr->va_size == 0) 1503 return (ENOEXEC); 1504 1505 /* 1506 * Check for execute permission to file based on current credentials. 1507 */ 1508 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1509 if (error) 1510 return (error); 1511 1512 /* 1513 * Check number of open-for-writes on the file and deny execution 1514 * if there are any. 1515 */ 1516 error = VOP_GET_WRITECOUNT(vp, &writecount); 1517 if (error != 0) 1518 return (error); 1519 if (writecount != 0) 1520 return (ETXTBSY); 1521 1522 /* 1523 * Call filesystem specific open routine (which does nothing in the 1524 * general case). 1525 */ 1526 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1527 if (error == 0) 1528 imgp->opened = 1; 1529 return (error); 1530 } 1531 1532 /* 1533 * Exec handler registration 1534 */ 1535 int 1536 exec_register(execsw_arg) 1537 const struct execsw *execsw_arg; 1538 { 1539 const struct execsw **es, **xs, **newexecsw; 1540 int count = 2; /* New slot and trailing NULL */ 1541 1542 if (execsw) 1543 for (es = execsw; *es; es++) 1544 count++; 1545 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1546 if (newexecsw == NULL) 1547 return (ENOMEM); 1548 xs = newexecsw; 1549 if (execsw) 1550 for (es = execsw; *es; es++) 1551 *xs++ = *es; 1552 *xs++ = execsw_arg; 1553 *xs = NULL; 1554 if (execsw) 1555 free(execsw, M_TEMP); 1556 execsw = newexecsw; 1557 return (0); 1558 } 1559 1560 int 1561 exec_unregister(execsw_arg) 1562 const struct execsw *execsw_arg; 1563 { 1564 const struct execsw **es, **xs, **newexecsw; 1565 int count = 1; 1566 1567 if (execsw == NULL) 1568 panic("unregister with no handlers left?\n"); 1569 1570 for (es = execsw; *es; es++) { 1571 if (*es == execsw_arg) 1572 break; 1573 } 1574 if (*es == NULL) 1575 return (ENOENT); 1576 for (es = execsw; *es; es++) 1577 if (*es != execsw_arg) 1578 count++; 1579 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1580 if (newexecsw == NULL) 1581 return (ENOMEM); 1582 xs = newexecsw; 1583 for (es = execsw; *es; es++) 1584 if (*es != execsw_arg) 1585 *xs++ = *es; 1586 *xs = NULL; 1587 if (execsw) 1588 free(execsw, M_TEMP); 1589 execsw = newexecsw; 1590 return (0); 1591 } 1592