kern_exec.c revision 266582
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 266582 2014-05-23 09:29:04Z 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(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(SINGLE_EXIT); 315 else 316 thread_single_end(); 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; 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(p, td); 601 602 /* 603 * Malloc things before we need locks. 604 */ 605 newcred = crget(); 606 euip = uifind(attr.va_uid); 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 /* close files on exec */ 615 fdcloseexec(td); 616 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 617 618 /* Get a reference to the vnode prior to locking the proc */ 619 VREF(binvp); 620 621 /* 622 * For security and other reasons, signal handlers cannot 623 * be shared after an exec. The new process gets a copy of the old 624 * handlers. In execsigs(), the new process will have its signals 625 * reset. 626 */ 627 PROC_LOCK(p); 628 oldcred = crcopysafe(p, newcred); 629 if (sigacts_shared(p->p_sigacts)) { 630 oldsigacts = p->p_sigacts; 631 PROC_UNLOCK(p); 632 newsigacts = sigacts_alloc(); 633 sigacts_copy(newsigacts, oldsigacts); 634 PROC_LOCK(p); 635 p->p_sigacts = newsigacts; 636 } else 637 oldsigacts = NULL; 638 639 /* Stop profiling */ 640 stopprofclock(p); 641 642 /* reset caught signals */ 643 execsigs(p); 644 645 /* name this process - nameiexec(p, ndp) */ 646 bzero(p->p_comm, sizeof(p->p_comm)); 647 if (args->fname) 648 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 649 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 650 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0) 651 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 652 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 653#ifdef KTR 654 sched_clear_tdname(td); 655#endif 656 657 /* 658 * mark as execed, wakeup the process that vforked (if any) and tell 659 * it that it now has its own resources back 660 */ 661 p->p_flag |= P_EXEC; 662 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 663 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 664 cv_broadcast(&p->p_pwait); 665 } 666 667 /* 668 * Implement image setuid/setgid. 669 * 670 * Don't honor setuid/setgid if the filesystem prohibits it or if 671 * the process is being traced. 672 * 673 * We disable setuid/setgid/etc in compatibility mode on the basis 674 * that most setugid applications are not written with that 675 * environment in mind, and will therefore almost certainly operate 676 * incorrectly. In principle there's no reason that setugid 677 * applications might not be useful in capability mode, so we may want 678 * to reconsider this conservative design choice in the future. 679 * 680 * XXXMAC: For the time being, use NOSUID to also prohibit 681 * transitions on the file system. 682 */ 683 credential_changing = 0; 684 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != 685 attr.va_uid; 686 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != 687 attr.va_gid; 688#ifdef MAC 689 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 690 interpvplabel, imgp); 691 credential_changing |= will_transition; 692#endif 693 694 if (credential_changing && 695#ifdef CAPABILITY_MODE 696 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 697#endif 698 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 699 (p->p_flag & P_TRACED) == 0) { 700 /* 701 * Turn off syscall tracing for set-id programs, except for 702 * root. Record any set-id flags first to make sure that 703 * we do not regain any tracing during a possible block. 704 */ 705 setsugid(p); 706 707#ifdef KTRACE 708 if (p->p_tracecred != NULL && 709 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 710 ktrprocexec(p, &tracecred, &tracevp); 711#endif 712 /* 713 * Close any file descriptors 0..2 that reference procfs, 714 * then make sure file descriptors 0..2 are in use. 715 * 716 * setugidsafety() may call closef() and then pfind() 717 * which may grab the process lock. 718 * fdcheckstd() may call falloc() which may block to 719 * allocate memory, so temporarily drop the process lock. 720 */ 721 PROC_UNLOCK(p); 722 VOP_UNLOCK(imgp->vp, 0); 723 setugidsafety(td); 724 error = fdcheckstd(td); 725 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 726 if (error != 0) 727 goto done1; 728 PROC_LOCK(p); 729 /* 730 * Set the new credentials. 731 */ 732 if (attr.va_mode & S_ISUID) 733 change_euid(newcred, euip); 734 if (attr.va_mode & S_ISGID) 735 change_egid(newcred, attr.va_gid); 736#ifdef MAC 737 if (will_transition) { 738 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 739 interpvplabel, imgp); 740 } 741#endif 742 /* 743 * Implement correct POSIX saved-id behavior. 744 * 745 * XXXMAC: Note that the current logic will save the 746 * uid and gid if a MAC domain transition occurs, even 747 * though maybe it shouldn't. 748 */ 749 change_svuid(newcred, newcred->cr_uid); 750 change_svgid(newcred, newcred->cr_gid); 751 p->p_ucred = newcred; 752 newcred = NULL; 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 change_svuid(newcred, newcred->cr_uid); 772 change_svgid(newcred, newcred->cr_gid); 773 p->p_ucred = newcred; 774 newcred = NULL; 775 } 776 } 777 778 /* 779 * Store the vp for use in procfs. This vnode was referenced prior 780 * to locking the proc lock. 781 */ 782 textvp = p->p_textvp; 783 p->p_textvp = binvp; 784 785#ifdef KDTRACE_HOOKS 786 /* 787 * Tell the DTrace fasttrap provider about the exec if it 788 * has declared an interest. 789 */ 790 if (dtrace_fasttrap_exec) 791 dtrace_fasttrap_exec(p); 792#endif 793 794 /* 795 * Notify others that we exec'd, and clear the P_INEXEC flag 796 * as we're now a bona fide freshly-execed process. 797 */ 798 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 799 p->p_flag &= ~P_INEXEC; 800 801 /* clear "fork but no exec" flag, as we _are_ execing */ 802 p->p_acflag &= ~AFORK; 803 804 /* 805 * Free any previous argument cache and replace it with 806 * the new argument cache, if any. 807 */ 808 oldargs = p->p_args; 809 p->p_args = newargs; 810 newargs = NULL; 811 812#ifdef HWPMC_HOOKS 813 /* 814 * Check if system-wide sampling is in effect or if the 815 * current process is using PMCs. If so, do exec() time 816 * processing. This processing needs to happen AFTER the 817 * P_INEXEC flag is cleared. 818 * 819 * The proc lock needs to be released before taking the PMC 820 * SX. 821 */ 822 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 823 PROC_UNLOCK(p); 824 VOP_UNLOCK(imgp->vp, 0); 825 pe.pm_credentialschanged = credential_changing; 826 pe.pm_entryaddr = imgp->entry_addr; 827 828 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 829 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 830 } else 831 PROC_UNLOCK(p); 832#else /* !HWPMC_HOOKS */ 833 PROC_UNLOCK(p); 834#endif 835 836 /* Set values passed into the program in registers. */ 837 if (p->p_sysent->sv_setregs) 838 (*p->p_sysent->sv_setregs)(td, imgp, 839 (u_long)(uintptr_t)stack_base); 840 else 841 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 842 843 vfs_mark_atime(imgp->vp, td->td_ucred); 844 845 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0); 846 847done1: 848 /* 849 * Free any resources malloc'd earlier that we didn't use. 850 */ 851 uifree(euip); 852 if (newcred == NULL) 853 crfree(oldcred); 854 else 855 crfree(newcred); 856 VOP_UNLOCK(imgp->vp, 0); 857 858 /* 859 * Handle deferred decrement of ref counts. 860 */ 861 if (textvp != NULL) 862 vrele(textvp); 863 if (binvp && error != 0) 864 vrele(binvp); 865#ifdef KTRACE 866 if (tracevp != NULL) 867 vrele(tracevp); 868 if (tracecred != NULL) 869 crfree(tracecred); 870#endif 871 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 872 pargs_drop(oldargs); 873 pargs_drop(newargs); 874 if (oldsigacts != NULL) 875 sigacts_free(oldsigacts); 876 877exec_fail_dealloc: 878 879 /* 880 * free various allocated resources 881 */ 882 if (imgp->firstpage != NULL) 883 exec_unmap_first_page(imgp); 884 885 if (imgp->vp != NULL) { 886 if (args->fname) 887 NDFREE(&nd, NDF_ONLY_PNBUF); 888 if (imgp->opened) 889 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 890 vput(imgp->vp); 891 } 892 893 if (imgp->object != NULL) 894 vm_object_deallocate(imgp->object); 895 896 free(imgp->freepath, M_TEMP); 897 898 if (error == 0) { 899 PROC_LOCK(p); 900 td->td_dbgflags |= TDB_EXEC; 901 PROC_UNLOCK(p); 902 903 /* 904 * Stop the process here if its stop event mask has 905 * the S_EXEC bit set. 906 */ 907 STOPEVENT(p, S_EXEC, 0); 908 goto done2; 909 } 910 911exec_fail: 912 /* we're done here, clear P_INEXEC */ 913 PROC_LOCK(p); 914 p->p_flag &= ~P_INEXEC; 915 PROC_UNLOCK(p); 916 917 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 918 919done2: 920#ifdef MAC 921 mac_execve_exit(imgp); 922 mac_execve_interpreter_exit(interpvplabel); 923#endif 924 exec_free_args(args); 925 926 if (error && imgp->vmspace_destroyed) { 927 /* sorry, no more process anymore. exit gracefully */ 928 exit1(td, W_EXITCODE(0, SIGABRT)); 929 /* NOT REACHED */ 930 } 931 932#ifdef KTRACE 933 if (error == 0) 934 ktrprocctor(p); 935#endif 936 937 return (error); 938} 939 940int 941exec_map_first_page(imgp) 942 struct image_params *imgp; 943{ 944 int rv, i; 945 int initial_pagein; 946 vm_page_t ma[VM_INITIAL_PAGEIN]; 947 vm_object_t object; 948 949 if (imgp->firstpage != NULL) 950 exec_unmap_first_page(imgp); 951 952 object = imgp->vp->v_object; 953 if (object == NULL) 954 return (EACCES); 955 VM_OBJECT_WLOCK(object); 956#if VM_NRESERVLEVEL > 0 957 if ((object->flags & OBJ_COLORED) == 0) { 958 object->flags |= OBJ_COLORED; 959 object->pg_color = 0; 960 } 961#endif 962 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL); 963 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 964 initial_pagein = VM_INITIAL_PAGEIN; 965 if (initial_pagein > object->size) 966 initial_pagein = object->size; 967 for (i = 1; i < initial_pagein; i++) { 968 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 969 if (ma[i]->valid) 970 break; 971 if (vm_page_tryxbusy(ma[i])) 972 break; 973 } else { 974 ma[i] = vm_page_alloc(object, i, 975 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 976 if (ma[i] == NULL) 977 break; 978 } 979 } 980 initial_pagein = i; 981 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 982 ma[0] = vm_page_lookup(object, 0); 983 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) { 984 if (ma[0] != NULL) { 985 vm_page_lock(ma[0]); 986 vm_page_free(ma[0]); 987 vm_page_unlock(ma[0]); 988 } 989 VM_OBJECT_WUNLOCK(object); 990 return (EIO); 991 } 992 } 993 vm_page_xunbusy(ma[0]); 994 vm_page_lock(ma[0]); 995 vm_page_hold(ma[0]); 996 vm_page_unlock(ma[0]); 997 VM_OBJECT_WUNLOCK(object); 998 999 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1000 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1001 1002 return (0); 1003} 1004 1005void 1006exec_unmap_first_page(imgp) 1007 struct image_params *imgp; 1008{ 1009 vm_page_t m; 1010 1011 if (imgp->firstpage != NULL) { 1012 m = sf_buf_page(imgp->firstpage); 1013 sf_buf_free(imgp->firstpage); 1014 imgp->firstpage = NULL; 1015 vm_page_lock(m); 1016 vm_page_unhold(m); 1017 vm_page_unlock(m); 1018 } 1019} 1020 1021/* 1022 * Destroy old address space, and allocate a new stack 1023 * The new stack is only SGROWSIZ large because it is grown 1024 * automatically in trap.c. 1025 */ 1026int 1027exec_new_vmspace(imgp, sv) 1028 struct image_params *imgp; 1029 struct sysentvec *sv; 1030{ 1031 int error; 1032 struct proc *p = imgp->proc; 1033 struct vmspace *vmspace = p->p_vmspace; 1034 vm_object_t obj; 1035 vm_offset_t sv_minuser, stack_addr; 1036 vm_map_t map; 1037 u_long ssiz; 1038 1039 imgp->vmspace_destroyed = 1; 1040 imgp->sysent = sv; 1041 1042 /* May be called with Giant held */ 1043 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1044 1045 /* 1046 * Blow away entire process VM, if address space not shared, 1047 * otherwise, create a new VM space so that other threads are 1048 * not disrupted 1049 */ 1050 map = &vmspace->vm_map; 1051 if (map_at_zero) 1052 sv_minuser = sv->sv_minuser; 1053 else 1054 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1055 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1056 vm_map_max(map) == sv->sv_maxuser) { 1057 shmexit(vmspace); 1058 pmap_remove_pages(vmspace_pmap(vmspace)); 1059 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1060 } else { 1061 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1062 if (error) 1063 return (error); 1064 vmspace = p->p_vmspace; 1065 map = &vmspace->vm_map; 1066 } 1067 1068 /* Map a shared page */ 1069 obj = sv->sv_shared_page_obj; 1070 if (obj != NULL) { 1071 vm_object_reference(obj); 1072 error = vm_map_fixed(map, obj, 0, 1073 sv->sv_shared_page_base, sv->sv_shared_page_len, 1074 VM_PROT_READ | VM_PROT_EXECUTE, 1075 VM_PROT_READ | VM_PROT_EXECUTE, 1076 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1077 if (error) { 1078 vm_object_deallocate(obj); 1079 return (error); 1080 } 1081 } 1082 1083 /* Allocate a new stack */ 1084 if (sv->sv_maxssiz != NULL) 1085 ssiz = *sv->sv_maxssiz; 1086 else 1087 ssiz = maxssiz; 1088 stack_addr = sv->sv_usrstack - ssiz; 1089 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1090 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1091 sv->sv_stackprot, 1092 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1093 if (error) 1094 return (error); 1095 1096#ifdef __ia64__ 1097 /* Allocate a new register stack */ 1098 stack_addr = IA64_BACKINGSTORE; 1099 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1100 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1101 if (error) 1102 return (error); 1103#endif 1104 1105 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1106 * VM_STACK case, but they are still used to monitor the size of the 1107 * process stack so we can check the stack rlimit. 1108 */ 1109 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1110 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; 1111 1112 return (0); 1113} 1114 1115/* 1116 * Copy out argument and environment strings from the old process address 1117 * space into the temporary string buffer. 1118 */ 1119int 1120exec_copyin_args(struct image_args *args, char *fname, 1121 enum uio_seg segflg, char **argv, char **envv) 1122{ 1123 char *argp, *envp; 1124 int error; 1125 size_t length; 1126 1127 bzero(args, sizeof(*args)); 1128 if (argv == NULL) 1129 return (EFAULT); 1130 1131 /* 1132 * Allocate demand-paged memory for the file name, argument, and 1133 * environment strings. 1134 */ 1135 error = exec_alloc_args(args); 1136 if (error != 0) 1137 return (error); 1138 1139 /* 1140 * Copy the file name. 1141 */ 1142 if (fname != NULL) { 1143 args->fname = args->buf; 1144 error = (segflg == UIO_SYSSPACE) ? 1145 copystr(fname, args->fname, PATH_MAX, &length) : 1146 copyinstr(fname, args->fname, PATH_MAX, &length); 1147 if (error != 0) 1148 goto err_exit; 1149 } else 1150 length = 0; 1151 1152 args->begin_argv = args->buf + length; 1153 args->endp = args->begin_argv; 1154 args->stringspace = ARG_MAX; 1155 1156 /* 1157 * extract arguments first 1158 */ 1159 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) { 1160 if (argp == (caddr_t) -1) { 1161 error = EFAULT; 1162 goto err_exit; 1163 } 1164 if ((error = copyinstr(argp, args->endp, 1165 args->stringspace, &length))) { 1166 if (error == ENAMETOOLONG) 1167 error = E2BIG; 1168 goto err_exit; 1169 } 1170 args->stringspace -= length; 1171 args->endp += length; 1172 args->argc++; 1173 } 1174 1175 args->begin_envv = args->endp; 1176 1177 /* 1178 * extract environment strings 1179 */ 1180 if (envv) { 1181 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 1182 if (envp == (caddr_t)-1) { 1183 error = EFAULT; 1184 goto err_exit; 1185 } 1186 if ((error = copyinstr(envp, args->endp, 1187 args->stringspace, &length))) { 1188 if (error == ENAMETOOLONG) 1189 error = E2BIG; 1190 goto err_exit; 1191 } 1192 args->stringspace -= length; 1193 args->endp += length; 1194 args->envc++; 1195 } 1196 } 1197 1198 return (0); 1199 1200err_exit: 1201 exec_free_args(args); 1202 return (error); 1203} 1204 1205/* 1206 * Allocate temporary demand-paged, zero-filled memory for the file name, 1207 * argument, and environment strings. Returns zero if the allocation succeeds 1208 * and ENOMEM otherwise. 1209 */ 1210int 1211exec_alloc_args(struct image_args *args) 1212{ 1213 1214 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 1215 return (args->buf != NULL ? 0 : ENOMEM); 1216} 1217 1218void 1219exec_free_args(struct image_args *args) 1220{ 1221 1222 if (args->buf != NULL) { 1223 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf, 1224 PATH_MAX + ARG_MAX); 1225 args->buf = NULL; 1226 } 1227 if (args->fname_buf != NULL) { 1228 free(args->fname_buf, M_TEMP); 1229 args->fname_buf = NULL; 1230 } 1231} 1232 1233/* 1234 * Copy strings out to the new process address space, constructing new arg 1235 * and env vector tables. Return a pointer to the base so that it can be used 1236 * as the initial stack pointer. 1237 */ 1238register_t * 1239exec_copyout_strings(imgp) 1240 struct image_params *imgp; 1241{ 1242 int argc, envc; 1243 char **vectp; 1244 char *stringp; 1245 uintptr_t destp; 1246 register_t *stack_base; 1247 struct ps_strings *arginfo; 1248 struct proc *p; 1249 size_t execpath_len; 1250 int szsigcode, szps; 1251 char canary[sizeof(long) * 8]; 1252 1253 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1254 /* 1255 * Calculate string base and vector table pointers. 1256 * Also deal with signal trampoline code for this exec type. 1257 */ 1258 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1259 execpath_len = strlen(imgp->execpath) + 1; 1260 else 1261 execpath_len = 0; 1262 p = imgp->proc; 1263 szsigcode = 0; 1264 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1265 if (p->p_sysent->sv_sigcode_base == 0) { 1266 if (p->p_sysent->sv_szsigcode != NULL) 1267 szsigcode = *(p->p_sysent->sv_szsigcode); 1268 } 1269 destp = (uintptr_t)arginfo; 1270 1271 /* 1272 * install sigcode 1273 */ 1274 if (szsigcode != 0) { 1275 destp -= szsigcode; 1276 destp = rounddown2(destp, sizeof(void *)); 1277 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1278 } 1279 1280 /* 1281 * Copy the image path for the rtld. 1282 */ 1283 if (execpath_len != 0) { 1284 destp -= execpath_len; 1285 imgp->execpathp = destp; 1286 copyout(imgp->execpath, (void *)destp, execpath_len); 1287 } 1288 1289 /* 1290 * Prepare the canary for SSP. 1291 */ 1292 arc4rand(canary, sizeof(canary), 0); 1293 destp -= sizeof(canary); 1294 imgp->canary = destp; 1295 copyout(canary, (void *)destp, sizeof(canary)); 1296 imgp->canarylen = sizeof(canary); 1297 1298 /* 1299 * Prepare the pagesizes array. 1300 */ 1301 destp -= szps; 1302 destp = rounddown2(destp, sizeof(void *)); 1303 imgp->pagesizes = destp; 1304 copyout(pagesizes, (void *)destp, szps); 1305 imgp->pagesizeslen = szps; 1306 1307 destp -= ARG_MAX - imgp->args->stringspace; 1308 destp = rounddown2(destp, sizeof(void *)); 1309 1310 /* 1311 * If we have a valid auxargs ptr, prepare some room 1312 * on the stack. 1313 */ 1314 if (imgp->auxargs) { 1315 /* 1316 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1317 * lower compatibility. 1318 */ 1319 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1320 (AT_COUNT * 2); 1321 /* 1322 * The '+ 2' is for the null pointers at the end of each of 1323 * the arg and env vector sets,and imgp->auxarg_size is room 1324 * for argument of Runtime loader. 1325 */ 1326 vectp = (char **)(destp - (imgp->args->argc + 1327 imgp->args->envc + 2 + imgp->auxarg_size) 1328 * sizeof(char *)); 1329 } else { 1330 /* 1331 * The '+ 2' is for the null pointers at the end of each of 1332 * the arg and env vector sets 1333 */ 1334 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc 1335 + 2) * sizeof(char *)); 1336 } 1337 1338 /* 1339 * vectp also becomes our initial stack base 1340 */ 1341 stack_base = (register_t *)vectp; 1342 1343 stringp = imgp->args->begin_argv; 1344 argc = imgp->args->argc; 1345 envc = imgp->args->envc; 1346 1347 /* 1348 * Copy out strings - arguments and environment. 1349 */ 1350 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1351 1352 /* 1353 * Fill in "ps_strings" struct for ps, w, etc. 1354 */ 1355 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1356 suword32(&arginfo->ps_nargvstr, argc); 1357 1358 /* 1359 * Fill in argument portion of vector table. 1360 */ 1361 for (; argc > 0; --argc) { 1362 suword(vectp++, (long)(intptr_t)destp); 1363 while (*stringp++ != 0) 1364 destp++; 1365 destp++; 1366 } 1367 1368 /* a null vector table pointer separates the argp's from the envp's */ 1369 suword(vectp++, 0); 1370 1371 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1372 suword32(&arginfo->ps_nenvstr, envc); 1373 1374 /* 1375 * Fill in environment portion of vector table. 1376 */ 1377 for (; envc > 0; --envc) { 1378 suword(vectp++, (long)(intptr_t)destp); 1379 while (*stringp++ != 0) 1380 destp++; 1381 destp++; 1382 } 1383 1384 /* end of vector table is a null pointer */ 1385 suword(vectp, 0); 1386 1387 return (stack_base); 1388} 1389 1390/* 1391 * Check permissions of file to execute. 1392 * Called with imgp->vp locked. 1393 * Return 0 for success or error code on failure. 1394 */ 1395int 1396exec_check_permissions(imgp) 1397 struct image_params *imgp; 1398{ 1399 struct vnode *vp = imgp->vp; 1400 struct vattr *attr = imgp->attr; 1401 struct thread *td; 1402 int error, writecount; 1403 1404 td = curthread; 1405 1406 /* Get file attributes */ 1407 error = VOP_GETATTR(vp, attr, td->td_ucred); 1408 if (error) 1409 return (error); 1410 1411#ifdef MAC 1412 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1413 if (error) 1414 return (error); 1415#endif 1416 1417 /* 1418 * 1) Check if file execution is disabled for the filesystem that 1419 * this file resides on. 1420 * 2) Ensure that at least one execute bit is on. Otherwise, a 1421 * privileged user will always succeed, and we don't want this 1422 * to happen unless the file really is executable. 1423 * 3) Ensure that the file is a regular file. 1424 */ 1425 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1426 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1427 (attr->va_type != VREG)) 1428 return (EACCES); 1429 1430 /* 1431 * Zero length files can't be exec'd 1432 */ 1433 if (attr->va_size == 0) 1434 return (ENOEXEC); 1435 1436 /* 1437 * Check for execute permission to file based on current credentials. 1438 */ 1439 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1440 if (error) 1441 return (error); 1442 1443 /* 1444 * Check number of open-for-writes on the file and deny execution 1445 * if there are any. 1446 */ 1447 error = VOP_GET_WRITECOUNT(vp, &writecount); 1448 if (error != 0) 1449 return (error); 1450 if (writecount != 0) 1451 return (ETXTBSY); 1452 1453 /* 1454 * Call filesystem specific open routine (which does nothing in the 1455 * general case). 1456 */ 1457 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1458 if (error == 0) 1459 imgp->opened = 1; 1460 return (error); 1461} 1462 1463/* 1464 * Exec handler registration 1465 */ 1466int 1467exec_register(execsw_arg) 1468 const struct execsw *execsw_arg; 1469{ 1470 const struct execsw **es, **xs, **newexecsw; 1471 int count = 2; /* New slot and trailing NULL */ 1472 1473 if (execsw) 1474 for (es = execsw; *es; es++) 1475 count++; 1476 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1477 if (newexecsw == NULL) 1478 return (ENOMEM); 1479 xs = newexecsw; 1480 if (execsw) 1481 for (es = execsw; *es; es++) 1482 *xs++ = *es; 1483 *xs++ = execsw_arg; 1484 *xs = NULL; 1485 if (execsw) 1486 free(execsw, M_TEMP); 1487 execsw = newexecsw; 1488 return (0); 1489} 1490 1491int 1492exec_unregister(execsw_arg) 1493 const struct execsw *execsw_arg; 1494{ 1495 const struct execsw **es, **xs, **newexecsw; 1496 int count = 1; 1497 1498 if (execsw == NULL) 1499 panic("unregister with no handlers left?\n"); 1500 1501 for (es = execsw; *es; es++) { 1502 if (*es == execsw_arg) 1503 break; 1504 } 1505 if (*es == NULL) 1506 return (ENOENT); 1507 for (es = execsw; *es; es++) 1508 if (*es != execsw_arg) 1509 count++; 1510 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1511 if (newexecsw == NULL) 1512 return (ENOMEM); 1513 xs = newexecsw; 1514 for (es = execsw; *es; es++) 1515 if (*es != execsw_arg) 1516 *xs++ = *es; 1517 *xs = NULL; 1518 if (execsw) 1519 free(execsw, M_TEMP); 1520 execsw = newexecsw; 1521 return (0); 1522} 1523