kern_exec.c revision 276272
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: stable/10/sys/kern/kern_exec.c 276272 2014-12-27 00:55:14Z kib $"); 29 30#include "opt_capsicum.h" 31#include "opt_hwpmc_hooks.h" 32#include "opt_kdtrace.h" 33#include "opt_ktrace.h" 34#include "opt_vm.h" 35 36#include <sys/param.h> 37#include <sys/capability.h> 38#include <sys/systm.h> 39#include <sys/capability.h> 40#include <sys/eventhandler.h> 41#include <sys/lock.h> 42#include <sys/mutex.h> 43#include <sys/sysproto.h> 44#include <sys/signalvar.h> 45#include <sys/kernel.h> 46#include <sys/mount.h> 47#include <sys/filedesc.h> 48#include <sys/fcntl.h> 49#include <sys/acct.h> 50#include <sys/exec.h> 51#include <sys/imgact.h> 52#include <sys/imgact_elf.h> 53#include <sys/wait.h> 54#include <sys/malloc.h> 55#include <sys/priv.h> 56#include <sys/proc.h> 57#include <sys/pioctl.h> 58#include <sys/namei.h> 59#include <sys/resourcevar.h> 60#include <sys/rwlock.h> 61#include <sys/sched.h> 62#include <sys/sdt.h> 63#include <sys/sf_buf.h> 64#include <sys/syscallsubr.h> 65#include <sys/sysent.h> 66#include <sys/shm.h> 67#include <sys/sysctl.h> 68#include <sys/vnode.h> 69#include <sys/stat.h> 70#ifdef KTRACE 71#include <sys/ktrace.h> 72#endif 73 74#include <vm/vm.h> 75#include <vm/vm_param.h> 76#include <vm/pmap.h> 77#include <vm/vm_page.h> 78#include <vm/vm_map.h> 79#include <vm/vm_kern.h> 80#include <vm/vm_extern.h> 81#include <vm/vm_object.h> 82#include <vm/vm_pager.h> 83 84#ifdef HWPMC_HOOKS 85#include <sys/pmckern.h> 86#endif 87 88#include <machine/reg.h> 89 90#include <security/audit/audit.h> 91#include <security/mac/mac_framework.h> 92 93#ifdef KDTRACE_HOOKS 94#include <sys/dtrace_bsd.h> 95dtrace_execexit_func_t dtrace_fasttrap_exec; 96#endif 97 98SDT_PROVIDER_DECLARE(proc); 99SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *"); 100SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int"); 101SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *"); 102 103MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 104 105static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 106static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 107static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 108static int do_execve(struct thread *td, struct image_args *args, 109 struct mac *mac_p); 110 111/* XXX This should be vm_size_t. */ 112SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 113 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 114 115/* XXX This should be vm_size_t. */ 116SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 117 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", ""); 118 119SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 120 NULL, 0, sysctl_kern_stackprot, "I", ""); 121 122u_long ps_arg_cache_limit = PAGE_SIZE / 16; 123SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 124 &ps_arg_cache_limit, 0, ""); 125 126static int disallow_high_osrel; 127SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 128 &disallow_high_osrel, 0, 129 "Disallow execution of binaries built for higher version of the world"); 130 131static int map_at_zero = 0; 132TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero); 133SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0, 134 "Permit processes to map an object at virtual address 0."); 135 136static int 137sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 138{ 139 struct proc *p; 140 int error; 141 142 p = curproc; 143#ifdef SCTL_MASK32 144 if (req->flags & SCTL_MASK32) { 145 unsigned int val; 146 val = (unsigned int)p->p_sysent->sv_psstrings; 147 error = SYSCTL_OUT(req, &val, sizeof(val)); 148 } else 149#endif 150 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 151 sizeof(p->p_sysent->sv_psstrings)); 152 return error; 153} 154 155static int 156sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 157{ 158 struct proc *p; 159 int error; 160 161 p = curproc; 162#ifdef SCTL_MASK32 163 if (req->flags & SCTL_MASK32) { 164 unsigned int val; 165 val = (unsigned int)p->p_sysent->sv_usrstack; 166 error = SYSCTL_OUT(req, &val, sizeof(val)); 167 } else 168#endif 169 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 170 sizeof(p->p_sysent->sv_usrstack)); 171 return error; 172} 173 174static int 175sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 176{ 177 struct proc *p; 178 179 p = curproc; 180 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 181 sizeof(p->p_sysent->sv_stackprot))); 182} 183 184/* 185 * Each of the items is a pointer to a `const struct execsw', hence the 186 * double pointer here. 187 */ 188static const struct execsw **execsw; 189 190#ifndef _SYS_SYSPROTO_H_ 191struct execve_args { 192 char *fname; 193 char **argv; 194 char **envv; 195}; 196#endif 197 198int 199sys_execve(td, uap) 200 struct thread *td; 201 struct execve_args /* { 202 char *fname; 203 char **argv; 204 char **envv; 205 } */ *uap; 206{ 207 int error; 208 struct image_args args; 209 210 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 211 uap->argv, uap->envv); 212 if (error == 0) 213 error = kern_execve(td, &args, NULL); 214 return (error); 215} 216 217#ifndef _SYS_SYSPROTO_H_ 218struct fexecve_args { 219 int fd; 220 char **argv; 221 char **envv; 222} 223#endif 224int 225sys_fexecve(struct thread *td, struct fexecve_args *uap) 226{ 227 int error; 228 struct image_args args; 229 230 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 return (error); 237} 238 239#ifndef _SYS_SYSPROTO_H_ 240struct __mac_execve_args { 241 char *fname; 242 char **argv; 243 char **envv; 244 struct mac *mac_p; 245}; 246#endif 247 248int 249sys___mac_execve(td, uap) 250 struct thread *td; 251 struct __mac_execve_args /* { 252 char *fname; 253 char **argv; 254 char **envv; 255 struct mac *mac_p; 256 } */ *uap; 257{ 258#ifdef MAC 259 int error; 260 struct image_args args; 261 262 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 263 uap->argv, uap->envv); 264 if (error == 0) 265 error = kern_execve(td, &args, uap->mac_p); 266 return (error); 267#else 268 return (ENOSYS); 269#endif 270} 271 272/* 273 * XXX: kern_execve has the astonishing property of not always returning to 274 * the caller. If sufficiently bad things happen during the call to 275 * do_execve(), it can end up calling exit1(); as a result, callers must 276 * avoid doing anything which they might need to undo (e.g., allocating 277 * memory). 278 */ 279int 280kern_execve(td, args, mac_p) 281 struct thread *td; 282 struct image_args *args; 283 struct mac *mac_p; 284{ 285 struct proc *p = td->td_proc; 286 struct vmspace *oldvmspace; 287 int error; 288 289 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 290 args->begin_envv - args->begin_argv); 291 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 292 args->endp - args->begin_envv); 293 if (p->p_flag & P_HADTHREADS) { 294 PROC_LOCK(p); 295 if (thread_single(p, SINGLE_BOUNDARY)) { 296 PROC_UNLOCK(p); 297 exec_free_args(args); 298 return (ERESTART); /* Try again later. */ 299 } 300 PROC_UNLOCK(p); 301 } 302 303 KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve")); 304 oldvmspace = td->td_proc->p_vmspace; 305 error = do_execve(td, args, mac_p); 306 307 if (p->p_flag & P_HADTHREADS) { 308 PROC_LOCK(p); 309 /* 310 * If success, we upgrade to SINGLE_EXIT state to 311 * force other threads to suicide. 312 */ 313 if (error == 0) 314 thread_single(p, SINGLE_EXIT); 315 else 316 thread_single_end(p, SINGLE_BOUNDARY); 317 PROC_UNLOCK(p); 318 } 319 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 320 KASSERT(td->td_proc->p_vmspace != oldvmspace, 321 ("oldvmspace still used")); 322 vmspace_free(oldvmspace); 323 td->td_pflags &= ~TDP_EXECVMSPC; 324 } 325 326 return (error); 327} 328 329/* 330 * In-kernel implementation of execve(). All arguments are assumed to be 331 * userspace pointers from the passed thread. 332 */ 333static int 334do_execve(td, args, mac_p) 335 struct thread *td; 336 struct image_args *args; 337 struct mac *mac_p; 338{ 339 struct proc *p = td->td_proc; 340 struct nameidata nd; 341 struct ucred *newcred = NULL, *oldcred; 342 struct uidinfo *euip = NULL; 343 register_t *stack_base; 344 int error, i; 345 struct image_params image_params, *imgp; 346 struct vattr attr; 347 int (*img_first)(struct image_params *); 348 struct pargs *oldargs = NULL, *newargs = NULL; 349 struct sigacts *oldsigacts, *newsigacts; 350#ifdef KTRACE 351 struct vnode *tracevp = NULL; 352 struct ucred *tracecred = NULL; 353#endif 354 struct vnode *textvp = NULL, *binvp = NULL; 355 cap_rights_t rights; 356 int credential_changing; 357 int textset; 358#ifdef MAC 359 struct label *interpvplabel = NULL; 360 int will_transition; 361#endif 362#ifdef HWPMC_HOOKS 363 struct pmckern_procexec pe; 364#endif 365 static const char fexecv_proc_title[] = "(fexecv)"; 366 367 imgp = &image_params; 368 369 /* 370 * Lock the process and set the P_INEXEC flag to indicate that 371 * it should be left alone until we're done here. This is 372 * necessary to avoid race conditions - e.g. in ptrace() - 373 * that might allow a local user to illicitly obtain elevated 374 * privileges. 375 */ 376 PROC_LOCK(p); 377 KASSERT((p->p_flag & P_INEXEC) == 0, 378 ("%s(): process already has P_INEXEC flag", __func__)); 379 p->p_flag |= P_INEXEC; 380 PROC_UNLOCK(p); 381 382 /* 383 * Initialize part of the common data 384 */ 385 imgp->proc = p; 386 imgp->execlabel = NULL; 387 imgp->attr = &attr; 388 imgp->entry_addr = 0; 389 imgp->reloc_base = 0; 390 imgp->vmspace_destroyed = 0; 391 imgp->interpreted = 0; 392 imgp->opened = 0; 393 imgp->interpreter_name = NULL; 394 imgp->auxargs = NULL; 395 imgp->vp = NULL; 396 imgp->object = NULL; 397 imgp->firstpage = NULL; 398 imgp->ps_strings = 0; 399 imgp->auxarg_size = 0; 400 imgp->args = args; 401 imgp->execpath = imgp->freepath = NULL; 402 imgp->execpathp = 0; 403 imgp->canary = 0; 404 imgp->canarylen = 0; 405 imgp->pagesizes = 0; 406 imgp->pagesizeslen = 0; 407 imgp->stack_prot = 0; 408 409#ifdef MAC 410 error = mac_execve_enter(imgp, mac_p); 411 if (error) 412 goto exec_fail; 413#endif 414 415 imgp->image_header = NULL; 416 417 /* 418 * Translate the file name. namei() returns a vnode pointer 419 * in ni_vp amoung other things. 420 * 421 * XXXAUDIT: It would be desirable to also audit the name of the 422 * interpreter if this is an interpreted binary. 423 */ 424 if (args->fname != NULL) { 425 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 426 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 427 } 428 429 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); 430 431interpret: 432 if (args->fname != NULL) { 433#ifdef CAPABILITY_MODE 434 /* 435 * While capability mode can't reach this point via direct 436 * path arguments to execve(), we also don't allow 437 * interpreters to be used in capability mode (for now). 438 * Catch indirect lookups and return a permissions error. 439 */ 440 if (IN_CAPABILITY_MODE(td)) { 441 error = ECAPMODE; 442 goto exec_fail; 443 } 444#endif 445 error = namei(&nd); 446 if (error) 447 goto exec_fail; 448 449 binvp = nd.ni_vp; 450 imgp->vp = binvp; 451 } else { 452 AUDIT_ARG_FD(args->fd); 453 /* 454 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 455 */ 456 error = fgetvp_exec(td, args->fd, 457 cap_rights_init(&rights, CAP_FEXECVE), &binvp); 458 if (error) 459 goto exec_fail; 460 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY); 461 AUDIT_ARG_VNODE1(binvp); 462 imgp->vp = binvp; 463 } 464 465 /* 466 * Check file permissions (also 'opens' file) 467 */ 468 error = exec_check_permissions(imgp); 469 if (error) 470 goto exec_fail_dealloc; 471 472 imgp->object = imgp->vp->v_object; 473 if (imgp->object != NULL) 474 vm_object_reference(imgp->object); 475 476 /* 477 * Set VV_TEXT now so no one can write to the executable while we're 478 * activating it. 479 * 480 * Remember if this was set before and unset it in case this is not 481 * actually an executable image. 482 */ 483 textset = VOP_IS_TEXT(imgp->vp); 484 VOP_SET_TEXT(imgp->vp); 485 486 error = exec_map_first_page(imgp); 487 if (error) 488 goto exec_fail_dealloc; 489 490 imgp->proc->p_osrel = 0; 491 /* 492 * If the current process has a special image activator it 493 * wants to try first, call it. For example, emulating shell 494 * scripts differently. 495 */ 496 error = -1; 497 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 498 error = img_first(imgp); 499 500 /* 501 * Loop through the list of image activators, calling each one. 502 * An activator returns -1 if there is no match, 0 on success, 503 * and an error otherwise. 504 */ 505 for (i = 0; error == -1 && execsw[i]; ++i) { 506 if (execsw[i]->ex_imgact == NULL || 507 execsw[i]->ex_imgact == img_first) { 508 continue; 509 } 510 error = (*execsw[i]->ex_imgact)(imgp); 511 } 512 513 if (error) { 514 if (error == -1) { 515 if (textset == 0) 516 VOP_UNSET_TEXT(imgp->vp); 517 error = ENOEXEC; 518 } 519 goto exec_fail_dealloc; 520 } 521 522 /* 523 * Special interpreter operation, cleanup and loop up to try to 524 * activate the interpreter. 525 */ 526 if (imgp->interpreted) { 527 exec_unmap_first_page(imgp); 528 /* 529 * VV_TEXT needs to be unset for scripts. There is a short 530 * period before we determine that something is a script where 531 * VV_TEXT will be set. The vnode lock is held over this 532 * entire period so nothing should illegitimately be blocked. 533 */ 534 VOP_UNSET_TEXT(imgp->vp); 535 /* free name buffer and old vnode */ 536 if (args->fname != NULL) 537 NDFREE(&nd, NDF_ONLY_PNBUF); 538#ifdef MAC 539 mac_execve_interpreter_enter(binvp, &interpvplabel); 540#endif 541 if (imgp->opened) { 542 VOP_CLOSE(binvp, FREAD, td->td_ucred, td); 543 imgp->opened = 0; 544 } 545 vput(binvp); 546 vm_object_deallocate(imgp->object); 547 imgp->object = NULL; 548 /* set new name to that of the interpreter */ 549 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 550 UIO_SYSSPACE, imgp->interpreter_name, td); 551 args->fname = imgp->interpreter_name; 552 goto interpret; 553 } 554 555 /* 556 * NB: We unlock the vnode here because it is believed that none 557 * of the sv_copyout_strings/sv_fixup operations require the vnode. 558 */ 559 VOP_UNLOCK(imgp->vp, 0); 560 561 /* 562 * Do the best to calculate the full path to the image file. 563 */ 564 if (imgp->auxargs != NULL && 565 ((args->fname != NULL && args->fname[0] == '/') || 566 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0)) 567 imgp->execpath = args->fname; 568 569 if (disallow_high_osrel && 570 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 571 error = ENOEXEC; 572 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 573 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 574 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 575 goto exec_fail_dealloc; 576 } 577 578 /* 579 * Copy out strings (args and env) and initialize stack base 580 */ 581 if (p->p_sysent->sv_copyout_strings) 582 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 583 else 584 stack_base = exec_copyout_strings(imgp); 585 586 /* 587 * If custom stack fixup routine present for this process 588 * let it do the stack setup. 589 * Else stuff argument count as first item on stack 590 */ 591 if (p->p_sysent->sv_fixup != NULL) 592 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 593 else 594 suword(--stack_base, imgp->args->argc); 595 596 /* 597 * For security and other reasons, the file descriptor table cannot 598 * be shared after an exec. 599 */ 600 fdunshare(td); 601 /* close files on exec */ 602 fdcloseexec(td); 603 604 /* 605 * Malloc things before we need locks. 606 */ 607 i = imgp->args->begin_envv - imgp->args->begin_argv; 608 /* Cache arguments if they fit inside our allowance */ 609 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 610 newargs = pargs_alloc(i); 611 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 612 } 613 614 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 615 616 /* Get a reference to the vnode prior to locking the proc */ 617 VREF(binvp); 618 619 /* 620 * For security and other reasons, signal handlers cannot 621 * be shared after an exec. The new process gets a copy of the old 622 * handlers. In execsigs(), the new process will have its signals 623 * reset. 624 */ 625 if (sigacts_shared(p->p_sigacts)) { 626 oldsigacts = p->p_sigacts; 627 newsigacts = sigacts_alloc(); 628 sigacts_copy(newsigacts, oldsigacts); 629 } else { 630 oldsigacts = NULL; 631 newsigacts = NULL; /* satisfy gcc */ 632 } 633 634 PROC_LOCK(p); 635 if (oldsigacts) 636 p->p_sigacts = newsigacts; 637 oldcred = p->p_ucred; 638 /* Stop profiling */ 639 stopprofclock(p); 640 641 /* reset caught signals */ 642 execsigs(p); 643 644 /* name this process - nameiexec(p, ndp) */ 645 bzero(p->p_comm, sizeof(p->p_comm)); 646 if (args->fname) 647 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 648 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 649 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0) 650 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 651 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 652#ifdef KTR 653 sched_clear_tdname(td); 654#endif 655 656 /* 657 * mark as execed, wakeup the process that vforked (if any) and tell 658 * it that it now has its own resources back 659 */ 660 p->p_flag |= P_EXEC; 661 if (p->p_flag & P_PPWAIT) { 662 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 663 cv_broadcast(&p->p_pwait); 664 } 665 666 /* 667 * Implement image setuid/setgid. 668 * 669 * Don't honor setuid/setgid if the filesystem prohibits it or if 670 * the process is being traced. 671 * 672 * We disable setuid/setgid/etc in compatibility mode on the basis 673 * that most setugid applications are not written with that 674 * environment in mind, and will therefore almost certainly operate 675 * incorrectly. In principle there's no reason that setugid 676 * applications might not be useful in capability mode, so we may want 677 * to reconsider this conservative design choice in the future. 678 * 679 * XXXMAC: For the time being, use NOSUID to also prohibit 680 * transitions on the file system. 681 */ 682 credential_changing = 0; 683 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != 684 attr.va_uid; 685 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != 686 attr.va_gid; 687#ifdef MAC 688 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 689 interpvplabel, imgp); 690 credential_changing |= will_transition; 691#endif 692 693 if (credential_changing && 694#ifdef CAPABILITY_MODE 695 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 696#endif 697 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 698 (p->p_flag & P_TRACED) == 0) { 699 /* 700 * Turn off syscall tracing for set-id programs, except for 701 * root. Record any set-id flags first to make sure that 702 * we do not regain any tracing during a possible block. 703 */ 704 setsugid(p); 705 706#ifdef KTRACE 707 if (p->p_tracecred != NULL && 708 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 709 ktrprocexec(p, &tracecred, &tracevp); 710#endif 711 /* 712 * Close any file descriptors 0..2 that reference procfs, 713 * then make sure file descriptors 0..2 are in use. 714 * 715 * setugidsafety() may call closef() and then pfind() 716 * which may grab the process lock. 717 * fdcheckstd() may call falloc() which may block to 718 * allocate memory, so temporarily drop the process lock. 719 */ 720 PROC_UNLOCK(p); 721 VOP_UNLOCK(imgp->vp, 0); 722 setugidsafety(td); 723 error = fdcheckstd(td); 724 if (error != 0) 725 goto done1; 726 newcred = crdup(oldcred); 727 euip = uifind(attr.va_uid); 728 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 729 PROC_LOCK(p); 730 /* 731 * Set the new credentials. 732 */ 733 if (attr.va_mode & S_ISUID) 734 change_euid(newcred, euip); 735 if (attr.va_mode & S_ISGID) 736 change_egid(newcred, attr.va_gid); 737#ifdef MAC 738 if (will_transition) { 739 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 740 interpvplabel, imgp); 741 } 742#endif 743 /* 744 * Implement correct POSIX saved-id behavior. 745 * 746 * XXXMAC: Note that the current logic will save the 747 * uid and gid if a MAC domain transition occurs, even 748 * though maybe it shouldn't. 749 */ 750 change_svuid(newcred, newcred->cr_uid); 751 change_svgid(newcred, newcred->cr_gid); 752 p->p_ucred = newcred; 753 } else { 754 if (oldcred->cr_uid == oldcred->cr_ruid && 755 oldcred->cr_gid == oldcred->cr_rgid) 756 p->p_flag &= ~P_SUGID; 757 /* 758 * Implement correct POSIX saved-id behavior. 759 * 760 * XXX: It's not clear that the existing behavior is 761 * POSIX-compliant. A number of sources indicate that the 762 * saved uid/gid should only be updated if the new ruid is 763 * not equal to the old ruid, or the new euid is not equal 764 * to the old euid and the new euid is not equal to the old 765 * ruid. The FreeBSD code always updates the saved uid/gid. 766 * Also, this code uses the new (replaced) euid and egid as 767 * the source, which may or may not be the right ones to use. 768 */ 769 if (oldcred->cr_svuid != oldcred->cr_uid || 770 oldcred->cr_svgid != oldcred->cr_gid) { 771 PROC_UNLOCK(p); 772 VOP_UNLOCK(imgp->vp, 0); 773 newcred = crdup(oldcred); 774 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 775 PROC_LOCK(p); 776 change_svuid(newcred, newcred->cr_uid); 777 change_svgid(newcred, newcred->cr_gid); 778 p->p_ucred = newcred; 779 } 780 } 781 782 /* 783 * Store the vp for use in procfs. This vnode was referenced prior 784 * to locking the proc lock. 785 */ 786 textvp = p->p_textvp; 787 p->p_textvp = binvp; 788 789#ifdef KDTRACE_HOOKS 790 /* 791 * Tell the DTrace fasttrap provider about the exec if it 792 * has declared an interest. 793 */ 794 if (dtrace_fasttrap_exec) 795 dtrace_fasttrap_exec(p); 796#endif 797 798 /* 799 * Notify others that we exec'd, and clear the P_INEXEC flag 800 * as we're now a bona fide freshly-execed process. 801 */ 802 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 803 p->p_flag &= ~P_INEXEC; 804 805 /* clear "fork but no exec" flag, as we _are_ execing */ 806 p->p_acflag &= ~AFORK; 807 808 /* 809 * Free any previous argument cache and replace it with 810 * the new argument cache, if any. 811 */ 812 oldargs = p->p_args; 813 p->p_args = newargs; 814 newargs = NULL; 815 816#ifdef HWPMC_HOOKS 817 /* 818 * Check if system-wide sampling is in effect or if the 819 * current process is using PMCs. If so, do exec() time 820 * processing. This processing needs to happen AFTER the 821 * P_INEXEC flag is cleared. 822 * 823 * The proc lock needs to be released before taking the PMC 824 * SX. 825 */ 826 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 827 PROC_UNLOCK(p); 828 VOP_UNLOCK(imgp->vp, 0); 829 pe.pm_credentialschanged = credential_changing; 830 pe.pm_entryaddr = imgp->entry_addr; 831 832 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 833 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 834 } else 835 PROC_UNLOCK(p); 836#else /* !HWPMC_HOOKS */ 837 PROC_UNLOCK(p); 838#endif 839 840 /* Set values passed into the program in registers. */ 841 if (p->p_sysent->sv_setregs) 842 (*p->p_sysent->sv_setregs)(td, imgp, 843 (u_long)(uintptr_t)stack_base); 844 else 845 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 846 847 vfs_mark_atime(imgp->vp, td->td_ucred); 848 849 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0); 850 851 VOP_UNLOCK(imgp->vp, 0); 852done1: 853 /* 854 * Free any resources malloc'd earlier that we didn't use. 855 */ 856 if (euip != NULL) 857 uifree(euip); 858 if (newcred != NULL) 859 crfree(oldcred); 860 861 /* 862 * Handle deferred decrement of ref counts. 863 */ 864 if (textvp != NULL) 865 vrele(textvp); 866 if (binvp && error != 0) 867 vrele(binvp); 868#ifdef KTRACE 869 if (tracevp != NULL) 870 vrele(tracevp); 871 if (tracecred != NULL) 872 crfree(tracecred); 873#endif 874 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 875 pargs_drop(oldargs); 876 pargs_drop(newargs); 877 if (oldsigacts != NULL) 878 sigacts_free(oldsigacts); 879 880exec_fail_dealloc: 881 882 /* 883 * free various allocated resources 884 */ 885 if (imgp->firstpage != NULL) 886 exec_unmap_first_page(imgp); 887 888 if (imgp->vp != NULL) { 889 if (args->fname) 890 NDFREE(&nd, NDF_ONLY_PNBUF); 891 if (imgp->opened) 892 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 893 vput(imgp->vp); 894 } 895 896 if (imgp->object != NULL) 897 vm_object_deallocate(imgp->object); 898 899 free(imgp->freepath, M_TEMP); 900 901 if (error == 0) { 902 PROC_LOCK(p); 903 td->td_dbgflags |= TDB_EXEC; 904 PROC_UNLOCK(p); 905 906 /* 907 * Stop the process here if its stop event mask has 908 * the S_EXEC bit set. 909 */ 910 STOPEVENT(p, S_EXEC, 0); 911 goto done2; 912 } 913 914exec_fail: 915 /* we're done here, clear P_INEXEC */ 916 PROC_LOCK(p); 917 p->p_flag &= ~P_INEXEC; 918 PROC_UNLOCK(p); 919 920 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 921 922done2: 923#ifdef MAC 924 mac_execve_exit(imgp); 925 mac_execve_interpreter_exit(interpvplabel); 926#endif 927 exec_free_args(args); 928 929 if (error && imgp->vmspace_destroyed) { 930 /* sorry, no more process anymore. exit gracefully */ 931 exit1(td, W_EXITCODE(0, SIGABRT)); 932 /* NOT REACHED */ 933 } 934 935#ifdef KTRACE 936 if (error == 0) 937 ktrprocctor(p); 938#endif 939 940 return (error); 941} 942 943int 944exec_map_first_page(imgp) 945 struct image_params *imgp; 946{ 947 int rv, i; 948 int initial_pagein; 949 vm_page_t ma[VM_INITIAL_PAGEIN]; 950 vm_object_t object; 951 952 if (imgp->firstpage != NULL) 953 exec_unmap_first_page(imgp); 954 955 object = imgp->vp->v_object; 956 if (object == NULL) 957 return (EACCES); 958 VM_OBJECT_WLOCK(object); 959#if VM_NRESERVLEVEL > 0 960 if ((object->flags & OBJ_COLORED) == 0) { 961 object->flags |= OBJ_COLORED; 962 object->pg_color = 0; 963 } 964#endif 965 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL); 966 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 967 initial_pagein = VM_INITIAL_PAGEIN; 968 if (initial_pagein > object->size) 969 initial_pagein = object->size; 970 for (i = 1; i < initial_pagein; i++) { 971 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 972 if (ma[i]->valid) 973 break; 974 if (vm_page_tryxbusy(ma[i])) 975 break; 976 } else { 977 ma[i] = vm_page_alloc(object, i, 978 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 979 if (ma[i] == NULL) 980 break; 981 } 982 } 983 initial_pagein = i; 984 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 985 ma[0] = vm_page_lookup(object, 0); 986 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) { 987 if (ma[0] != NULL) { 988 vm_page_lock(ma[0]); 989 vm_page_free(ma[0]); 990 vm_page_unlock(ma[0]); 991 } 992 VM_OBJECT_WUNLOCK(object); 993 return (EIO); 994 } 995 } 996 vm_page_xunbusy(ma[0]); 997 vm_page_lock(ma[0]); 998 vm_page_hold(ma[0]); 999 vm_page_activate(ma[0]); 1000 vm_page_unlock(ma[0]); 1001 VM_OBJECT_WUNLOCK(object); 1002 1003 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1004 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1005 1006 return (0); 1007} 1008 1009void 1010exec_unmap_first_page(imgp) 1011 struct image_params *imgp; 1012{ 1013 vm_page_t m; 1014 1015 if (imgp->firstpage != NULL) { 1016 m = sf_buf_page(imgp->firstpage); 1017 sf_buf_free(imgp->firstpage); 1018 imgp->firstpage = NULL; 1019 vm_page_lock(m); 1020 vm_page_unhold(m); 1021 vm_page_unlock(m); 1022 } 1023} 1024 1025/* 1026 * Destroy old address space, and allocate a new stack 1027 * The new stack is only SGROWSIZ large because it is grown 1028 * automatically in trap.c. 1029 */ 1030int 1031exec_new_vmspace(imgp, sv) 1032 struct image_params *imgp; 1033 struct sysentvec *sv; 1034{ 1035 int error; 1036 struct proc *p = imgp->proc; 1037 struct vmspace *vmspace = p->p_vmspace; 1038 vm_object_t obj; 1039 vm_offset_t sv_minuser, stack_addr; 1040 vm_map_t map; 1041 u_long ssiz; 1042 1043 imgp->vmspace_destroyed = 1; 1044 imgp->sysent = sv; 1045 1046 /* May be called with Giant held */ 1047 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1048 1049 /* 1050 * Blow away entire process VM, if address space not shared, 1051 * otherwise, create a new VM space so that other threads are 1052 * not disrupted 1053 */ 1054 map = &vmspace->vm_map; 1055 if (map_at_zero) 1056 sv_minuser = sv->sv_minuser; 1057 else 1058 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1059 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1060 vm_map_max(map) == sv->sv_maxuser) { 1061 shmexit(vmspace); 1062 pmap_remove_pages(vmspace_pmap(vmspace)); 1063 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1064 } else { 1065 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1066 if (error) 1067 return (error); 1068 vmspace = p->p_vmspace; 1069 map = &vmspace->vm_map; 1070 } 1071 1072 /* Map a shared page */ 1073 obj = sv->sv_shared_page_obj; 1074 if (obj != NULL) { 1075 vm_object_reference(obj); 1076 error = vm_map_fixed(map, obj, 0, 1077 sv->sv_shared_page_base, sv->sv_shared_page_len, 1078 VM_PROT_READ | VM_PROT_EXECUTE, 1079 VM_PROT_READ | VM_PROT_EXECUTE, 1080 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1081 if (error) { 1082 vm_object_deallocate(obj); 1083 return (error); 1084 } 1085 } 1086 1087 /* Allocate a new stack */ 1088 if (sv->sv_maxssiz != NULL) 1089 ssiz = *sv->sv_maxssiz; 1090 else 1091 ssiz = maxssiz; 1092 stack_addr = sv->sv_usrstack - ssiz; 1093 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1094 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1095 sv->sv_stackprot, 1096 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1097 if (error) 1098 return (error); 1099 1100#ifdef __ia64__ 1101 /* Allocate a new register stack */ 1102 stack_addr = IA64_BACKINGSTORE; 1103 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1104 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1105 if (error) 1106 return (error); 1107#endif 1108 1109 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1110 * VM_STACK case, but they are still used to monitor the size of the 1111 * process stack so we can check the stack rlimit. 1112 */ 1113 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1114 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; 1115 1116 return (0); 1117} 1118 1119/* 1120 * Copy out argument and environment strings from the old process address 1121 * space into the temporary string buffer. 1122 */ 1123int 1124exec_copyin_args(struct image_args *args, char *fname, 1125 enum uio_seg segflg, char **argv, char **envv) 1126{ 1127 u_long argp, envp; 1128 int error; 1129 size_t length; 1130 1131 bzero(args, sizeof(*args)); 1132 if (argv == NULL) 1133 return (EFAULT); 1134 1135 /* 1136 * Allocate demand-paged memory for the file name, argument, and 1137 * environment strings. 1138 */ 1139 error = exec_alloc_args(args); 1140 if (error != 0) 1141 return (error); 1142 1143 /* 1144 * Copy the file name. 1145 */ 1146 if (fname != NULL) { 1147 args->fname = args->buf; 1148 error = (segflg == UIO_SYSSPACE) ? 1149 copystr(fname, args->fname, PATH_MAX, &length) : 1150 copyinstr(fname, args->fname, PATH_MAX, &length); 1151 if (error != 0) 1152 goto err_exit; 1153 } else 1154 length = 0; 1155 1156 args->begin_argv = args->buf + length; 1157 args->endp = args->begin_argv; 1158 args->stringspace = ARG_MAX; 1159 1160 /* 1161 * extract arguments first 1162 */ 1163 for (;;) { 1164 error = fueword(argv++, &argp); 1165 if (error == -1) { 1166 error = EFAULT; 1167 goto err_exit; 1168 } 1169 if (argp == 0) 1170 break; 1171 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1172 args->stringspace, &length); 1173 if (error != 0) { 1174 if (error == ENAMETOOLONG) 1175 error = E2BIG; 1176 goto err_exit; 1177 } 1178 args->stringspace -= length; 1179 args->endp += length; 1180 args->argc++; 1181 } 1182 1183 args->begin_envv = args->endp; 1184 1185 /* 1186 * extract environment strings 1187 */ 1188 if (envv) { 1189 for (;;) { 1190 error = fueword(envv++, &envp); 1191 if (error == -1) { 1192 error = EFAULT; 1193 goto err_exit; 1194 } 1195 if (envp == 0) 1196 break; 1197 error = copyinstr((void *)(uintptr_t)envp, 1198 args->endp, args->stringspace, &length); 1199 if (error != 0) { 1200 if (error == ENAMETOOLONG) 1201 error = E2BIG; 1202 goto err_exit; 1203 } 1204 args->stringspace -= length; 1205 args->endp += length; 1206 args->envc++; 1207 } 1208 } 1209 1210 return (0); 1211 1212err_exit: 1213 exec_free_args(args); 1214 return (error); 1215} 1216 1217/* 1218 * Allocate temporary demand-paged, zero-filled memory for the file name, 1219 * argument, and environment strings. Returns zero if the allocation succeeds 1220 * and ENOMEM otherwise. 1221 */ 1222int 1223exec_alloc_args(struct image_args *args) 1224{ 1225 1226 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 1227 return (args->buf != NULL ? 0 : ENOMEM); 1228} 1229 1230void 1231exec_free_args(struct image_args *args) 1232{ 1233 1234 if (args->buf != NULL) { 1235 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf, 1236 PATH_MAX + ARG_MAX); 1237 args->buf = NULL; 1238 } 1239 if (args->fname_buf != NULL) { 1240 free(args->fname_buf, M_TEMP); 1241 args->fname_buf = NULL; 1242 } 1243} 1244 1245/* 1246 * Copy strings out to the new process address space, constructing new arg 1247 * and env vector tables. Return a pointer to the base so that it can be used 1248 * as the initial stack pointer. 1249 */ 1250register_t * 1251exec_copyout_strings(imgp) 1252 struct image_params *imgp; 1253{ 1254 int argc, envc; 1255 char **vectp; 1256 char *stringp; 1257 uintptr_t destp; 1258 register_t *stack_base; 1259 struct ps_strings *arginfo; 1260 struct proc *p; 1261 size_t execpath_len; 1262 int szsigcode, szps; 1263 char canary[sizeof(long) * 8]; 1264 1265 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1266 /* 1267 * Calculate string base and vector table pointers. 1268 * Also deal with signal trampoline code for this exec type. 1269 */ 1270 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1271 execpath_len = strlen(imgp->execpath) + 1; 1272 else 1273 execpath_len = 0; 1274 p = imgp->proc; 1275 szsigcode = 0; 1276 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1277 if (p->p_sysent->sv_sigcode_base == 0) { 1278 if (p->p_sysent->sv_szsigcode != NULL) 1279 szsigcode = *(p->p_sysent->sv_szsigcode); 1280 } 1281 destp = (uintptr_t)arginfo; 1282 1283 /* 1284 * install sigcode 1285 */ 1286 if (szsigcode != 0) { 1287 destp -= szsigcode; 1288 destp = rounddown2(destp, sizeof(void *)); 1289 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1290 } 1291 1292 /* 1293 * Copy the image path for the rtld. 1294 */ 1295 if (execpath_len != 0) { 1296 destp -= execpath_len; 1297 imgp->execpathp = destp; 1298 copyout(imgp->execpath, (void *)destp, execpath_len); 1299 } 1300 1301 /* 1302 * Prepare the canary for SSP. 1303 */ 1304 arc4rand(canary, sizeof(canary), 0); 1305 destp -= sizeof(canary); 1306 imgp->canary = destp; 1307 copyout(canary, (void *)destp, sizeof(canary)); 1308 imgp->canarylen = sizeof(canary); 1309 1310 /* 1311 * Prepare the pagesizes array. 1312 */ 1313 destp -= szps; 1314 destp = rounddown2(destp, sizeof(void *)); 1315 imgp->pagesizes = destp; 1316 copyout(pagesizes, (void *)destp, szps); 1317 imgp->pagesizeslen = szps; 1318 1319 destp -= ARG_MAX - imgp->args->stringspace; 1320 destp = rounddown2(destp, sizeof(void *)); 1321 1322 /* 1323 * If we have a valid auxargs ptr, prepare some room 1324 * on the stack. 1325 */ 1326 if (imgp->auxargs) { 1327 /* 1328 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1329 * lower compatibility. 1330 */ 1331 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1332 (AT_COUNT * 2); 1333 /* 1334 * The '+ 2' is for the null pointers at the end of each of 1335 * the arg and env vector sets,and imgp->auxarg_size is room 1336 * for argument of Runtime loader. 1337 */ 1338 vectp = (char **)(destp - (imgp->args->argc + 1339 imgp->args->envc + 2 + imgp->auxarg_size) 1340 * sizeof(char *)); 1341 } else { 1342 /* 1343 * The '+ 2' is for the null pointers at the end of each of 1344 * the arg and env vector sets 1345 */ 1346 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc 1347 + 2) * sizeof(char *)); 1348 } 1349 1350 /* 1351 * vectp also becomes our initial stack base 1352 */ 1353 stack_base = (register_t *)vectp; 1354 1355 stringp = imgp->args->begin_argv; 1356 argc = imgp->args->argc; 1357 envc = imgp->args->envc; 1358 1359 /* 1360 * Copy out strings - arguments and environment. 1361 */ 1362 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1363 1364 /* 1365 * Fill in "ps_strings" struct for ps, w, etc. 1366 */ 1367 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1368 suword32(&arginfo->ps_nargvstr, argc); 1369 1370 /* 1371 * Fill in argument portion of vector table. 1372 */ 1373 for (; argc > 0; --argc) { 1374 suword(vectp++, (long)(intptr_t)destp); 1375 while (*stringp++ != 0) 1376 destp++; 1377 destp++; 1378 } 1379 1380 /* a null vector table pointer separates the argp's from the envp's */ 1381 suword(vectp++, 0); 1382 1383 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1384 suword32(&arginfo->ps_nenvstr, envc); 1385 1386 /* 1387 * Fill in environment portion of vector table. 1388 */ 1389 for (; envc > 0; --envc) { 1390 suword(vectp++, (long)(intptr_t)destp); 1391 while (*stringp++ != 0) 1392 destp++; 1393 destp++; 1394 } 1395 1396 /* end of vector table is a null pointer */ 1397 suword(vectp, 0); 1398 1399 return (stack_base); 1400} 1401 1402/* 1403 * Check permissions of file to execute. 1404 * Called with imgp->vp locked. 1405 * Return 0 for success or error code on failure. 1406 */ 1407int 1408exec_check_permissions(imgp) 1409 struct image_params *imgp; 1410{ 1411 struct vnode *vp = imgp->vp; 1412 struct vattr *attr = imgp->attr; 1413 struct thread *td; 1414 int error, writecount; 1415 1416 td = curthread; 1417 1418 /* Get file attributes */ 1419 error = VOP_GETATTR(vp, attr, td->td_ucred); 1420 if (error) 1421 return (error); 1422 1423#ifdef MAC 1424 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1425 if (error) 1426 return (error); 1427#endif 1428 1429 /* 1430 * 1) Check if file execution is disabled for the filesystem that 1431 * this file resides on. 1432 * 2) Ensure that at least one execute bit is on. Otherwise, a 1433 * privileged user will always succeed, and we don't want this 1434 * to happen unless the file really is executable. 1435 * 3) Ensure that the file is a regular file. 1436 */ 1437 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1438 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1439 (attr->va_type != VREG)) 1440 return (EACCES); 1441 1442 /* 1443 * Zero length files can't be exec'd 1444 */ 1445 if (attr->va_size == 0) 1446 return (ENOEXEC); 1447 1448 /* 1449 * Check for execute permission to file based on current credentials. 1450 */ 1451 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1452 if (error) 1453 return (error); 1454 1455 /* 1456 * Check number of open-for-writes on the file and deny execution 1457 * if there are any. 1458 */ 1459 error = VOP_GET_WRITECOUNT(vp, &writecount); 1460 if (error != 0) 1461 return (error); 1462 if (writecount != 0) 1463 return (ETXTBSY); 1464 1465 /* 1466 * Call filesystem specific open routine (which does nothing in the 1467 * general case). 1468 */ 1469 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1470 if (error == 0) 1471 imgp->opened = 1; 1472 return (error); 1473} 1474 1475/* 1476 * Exec handler registration 1477 */ 1478int 1479exec_register(execsw_arg) 1480 const struct execsw *execsw_arg; 1481{ 1482 const struct execsw **es, **xs, **newexecsw; 1483 int count = 2; /* New slot and trailing NULL */ 1484 1485 if (execsw) 1486 for (es = execsw; *es; es++) 1487 count++; 1488 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1489 if (newexecsw == NULL) 1490 return (ENOMEM); 1491 xs = newexecsw; 1492 if (execsw) 1493 for (es = execsw; *es; es++) 1494 *xs++ = *es; 1495 *xs++ = execsw_arg; 1496 *xs = NULL; 1497 if (execsw) 1498 free(execsw, M_TEMP); 1499 execsw = newexecsw; 1500 return (0); 1501} 1502 1503int 1504exec_unregister(execsw_arg) 1505 const struct execsw *execsw_arg; 1506{ 1507 const struct execsw **es, **xs, **newexecsw; 1508 int count = 1; 1509 1510 if (execsw == NULL) 1511 panic("unregister with no handlers left?\n"); 1512 1513 for (es = execsw; *es; es++) { 1514 if (*es == execsw_arg) 1515 break; 1516 } 1517 if (*es == NULL) 1518 return (ENOENT); 1519 for (es = execsw; *es; es++) 1520 if (*es != execsw_arg) 1521 count++; 1522 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1523 if (newexecsw == NULL) 1524 return (ENOMEM); 1525 xs = newexecsw; 1526 for (es = execsw; *es; es++) 1527 if (*es != execsw_arg) 1528 *xs++ = *es; 1529 *xs = NULL; 1530 if (execsw) 1531 free(execsw, M_TEMP); 1532 execsw = newexecsw; 1533 return (0); 1534} 1535