kern_exec.c revision 302237
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 302237 2016-06-27 22:10:07Z bdrewery $"); 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/capsicum.h> 38#include <sys/systm.h> 39#include <sys/capsicum.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, , , exec, "char *"); 100SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 101SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 102 103MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 104 105int coredump_pack_fileinfo = 1; 106SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 107 &coredump_pack_fileinfo, 0, 108 "Enable file path packing in 'procstat -f' coredump notes"); 109 110int coredump_pack_vmmapinfo = 1; 111SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 112 &coredump_pack_vmmapinfo, 0, 113 "Enable file path packing in 'procstat -v' coredump notes"); 114 115static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 116static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 117static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 118static int do_execve(struct thread *td, struct image_args *args, 119 struct mac *mac_p); 120 121/* XXX This should be vm_size_t. */ 122SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 123 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 124 125/* XXX This should be vm_size_t. */ 126SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 127 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", ""); 128 129SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 130 NULL, 0, sysctl_kern_stackprot, "I", ""); 131 132u_long ps_arg_cache_limit = PAGE_SIZE / 16; 133SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 134 &ps_arg_cache_limit, 0, ""); 135 136static int disallow_high_osrel; 137SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 138 &disallow_high_osrel, 0, 139 "Disallow execution of binaries built for higher version of the world"); 140 141static int map_at_zero = 0; 142TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero); 143SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0, 144 "Permit processes to map an object at virtual address 0."); 145 146static int 147sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 148{ 149 struct proc *p; 150 int error; 151 152 p = curproc; 153#ifdef SCTL_MASK32 154 if (req->flags & SCTL_MASK32) { 155 unsigned int val; 156 val = (unsigned int)p->p_sysent->sv_psstrings; 157 error = SYSCTL_OUT(req, &val, sizeof(val)); 158 } else 159#endif 160 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 161 sizeof(p->p_sysent->sv_psstrings)); 162 return error; 163} 164 165static int 166sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 167{ 168 struct proc *p; 169 int error; 170 171 p = curproc; 172#ifdef SCTL_MASK32 173 if (req->flags & SCTL_MASK32) { 174 unsigned int val; 175 val = (unsigned int)p->p_sysent->sv_usrstack; 176 error = SYSCTL_OUT(req, &val, sizeof(val)); 177 } else 178#endif 179 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 180 sizeof(p->p_sysent->sv_usrstack)); 181 return error; 182} 183 184static int 185sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 186{ 187 struct proc *p; 188 189 p = curproc; 190 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 191 sizeof(p->p_sysent->sv_stackprot))); 192} 193 194/* 195 * Each of the items is a pointer to a `const struct execsw', hence the 196 * double pointer here. 197 */ 198static const struct execsw **execsw; 199 200#ifndef _SYS_SYSPROTO_H_ 201struct execve_args { 202 char *fname; 203 char **argv; 204 char **envv; 205}; 206#endif 207 208int 209sys_execve(struct thread *td, struct execve_args *uap) 210{ 211 struct image_args args; 212 struct vmspace *oldvmspace; 213 int error; 214 215 error = pre_execve(td, &oldvmspace); 216 if (error != 0) 217 return (error); 218 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 219 uap->argv, uap->envv); 220 if (error == 0) 221 error = kern_execve(td, &args, NULL); 222 post_execve(td, error, oldvmspace); 223 return (error); 224} 225 226#ifndef _SYS_SYSPROTO_H_ 227struct fexecve_args { 228 int fd; 229 char **argv; 230 char **envv; 231} 232#endif 233int 234sys_fexecve(struct thread *td, struct fexecve_args *uap) 235{ 236 struct image_args args; 237 struct vmspace *oldvmspace; 238 int error; 239 240 error = pre_execve(td, &oldvmspace); 241 if (error != 0) 242 return (error); 243 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 244 uap->argv, uap->envv); 245 if (error == 0) { 246 args.fd = uap->fd; 247 error = kern_execve(td, &args, NULL); 248 } 249 post_execve(td, error, oldvmspace); 250 return (error); 251} 252 253#ifndef _SYS_SYSPROTO_H_ 254struct __mac_execve_args { 255 char *fname; 256 char **argv; 257 char **envv; 258 struct mac *mac_p; 259}; 260#endif 261 262int 263sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 264{ 265#ifdef MAC 266 struct image_args args; 267 struct vmspace *oldvmspace; 268 int error; 269 270 error = pre_execve(td, &oldvmspace); 271 if (error != 0) 272 return (error); 273 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 274 uap->argv, uap->envv); 275 if (error == 0) 276 error = kern_execve(td, &args, uap->mac_p); 277 post_execve(td, error, oldvmspace); 278 return (error); 279#else 280 return (ENOSYS); 281#endif 282} 283 284int 285pre_execve(struct thread *td, struct vmspace **oldvmspace) 286{ 287 struct proc *p; 288 int error; 289 290 KASSERT(td == curthread, ("non-current thread %p", td)); 291 error = 0; 292 p = td->td_proc; 293 if ((p->p_flag & P_HADTHREADS) != 0) { 294 PROC_LOCK(p); 295 if (thread_single(p, SINGLE_BOUNDARY) != 0) 296 error = ERESTART; 297 PROC_UNLOCK(p); 298 } 299 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 300 ("nested execve")); 301 *oldvmspace = p->p_vmspace; 302 return (error); 303} 304 305void 306post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 307{ 308 struct proc *p; 309 310 KASSERT(td == curthread, ("non-current thread %p", td)); 311 p = td->td_proc; 312 if ((p->p_flag & P_HADTHREADS) != 0) { 313 PROC_LOCK(p); 314 /* 315 * If success, we upgrade to SINGLE_EXIT state to 316 * force other threads to suicide. 317 */ 318 if (error == 0) 319 thread_single(p, SINGLE_EXIT); 320 else 321 thread_single_end(p, SINGLE_BOUNDARY); 322 PROC_UNLOCK(p); 323 } 324 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 325 KASSERT(p->p_vmspace != oldvmspace, 326 ("oldvmspace still used")); 327 vmspace_free(oldvmspace); 328 td->td_pflags &= ~TDP_EXECVMSPC; 329 } 330} 331 332/* 333 * XXX: kern_execve has the astonishing property of not always returning to 334 * the caller. If sufficiently bad things happen during the call to 335 * do_execve(), it can end up calling exit1(); as a result, callers must 336 * avoid doing anything which they might need to undo (e.g., allocating 337 * memory). 338 */ 339int 340kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 341{ 342 343 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 344 args->begin_envv - args->begin_argv); 345 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 346 args->endp - args->begin_envv); 347 return (do_execve(td, args, mac_p)); 348} 349 350/* 351 * In-kernel implementation of execve(). All arguments are assumed to be 352 * userspace pointers from the passed thread. 353 */ 354static int 355do_execve(td, args, mac_p) 356 struct thread *td; 357 struct image_args *args; 358 struct mac *mac_p; 359{ 360 struct proc *p = td->td_proc; 361 struct nameidata nd; 362 struct ucred *newcred = NULL, *oldcred; 363 struct uidinfo *euip = NULL; 364 register_t *stack_base; 365 int error, i; 366 struct image_params image_params, *imgp; 367 struct vattr attr; 368 int (*img_first)(struct image_params *); 369 struct pargs *oldargs = NULL, *newargs = NULL; 370 struct sigacts *oldsigacts, *newsigacts; 371#ifdef KTRACE 372 struct vnode *tracevp = NULL; 373 struct ucred *tracecred = NULL; 374#endif 375 struct vnode *oldtextvp = NULL, *newtextvp; 376 cap_rights_t rights; 377 int credential_changing; 378 int textset; 379#ifdef MAC 380 struct label *interpvplabel = NULL; 381 int will_transition; 382#endif 383#ifdef HWPMC_HOOKS 384 struct pmckern_procexec pe; 385#endif 386 static const char fexecv_proc_title[] = "(fexecv)"; 387 388 imgp = &image_params; 389 390 /* 391 * Lock the process and set the P_INEXEC flag to indicate that 392 * it should be left alone until we're done here. This is 393 * necessary to avoid race conditions - e.g. in ptrace() - 394 * that might allow a local user to illicitly obtain elevated 395 * privileges. 396 */ 397 PROC_LOCK(p); 398 KASSERT((p->p_flag & P_INEXEC) == 0, 399 ("%s(): process already has P_INEXEC flag", __func__)); 400 p->p_flag |= P_INEXEC; 401 PROC_UNLOCK(p); 402 403 /* 404 * Initialize part of the common data 405 */ 406 bzero(imgp, sizeof(*imgp)); 407 imgp->proc = p; 408 imgp->attr = &attr; 409 imgp->args = args; 410 411#ifdef MAC 412 error = mac_execve_enter(imgp, mac_p); 413 if (error) 414 goto exec_fail; 415#endif 416 417 /* 418 * Translate the file name. namei() returns a vnode pointer 419 * in ni_vp among 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_PROBE1(proc, , , exec, args->fname); 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 newtextvp = nd.ni_vp; 450 imgp->vp = newtextvp; 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), &newtextvp); 458 if (error) 459 goto exec_fail; 460 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY); 461 AUDIT_ARG_VNODE1(newtextvp); 462 imgp->vp = newtextvp; 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(newtextvp, &interpvplabel); 540#endif 541 if (imgp->opened) { 542 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 543 imgp->opened = 0; 544 } 545 vput(newtextvp); 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 /* 617 * For security and other reasons, signal handlers cannot 618 * be shared after an exec. The new process gets a copy of the old 619 * handlers. In execsigs(), the new process will have its signals 620 * reset. 621 */ 622 if (sigacts_shared(p->p_sigacts)) { 623 oldsigacts = p->p_sigacts; 624 newsigacts = sigacts_alloc(); 625 sigacts_copy(newsigacts, oldsigacts); 626 } else { 627 oldsigacts = NULL; 628 newsigacts = NULL; /* satisfy gcc */ 629 } 630 631 PROC_LOCK(p); 632 if (oldsigacts) 633 p->p_sigacts = newsigacts; 634 oldcred = p->p_ucred; 635 /* Stop profiling */ 636 stopprofclock(p); 637 638 /* reset caught signals */ 639 execsigs(p); 640 641 /* name this process - nameiexec(p, ndp) */ 642 bzero(p->p_comm, sizeof(p->p_comm)); 643 if (args->fname) 644 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 645 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 646 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 647 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 648 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 649#ifdef KTR 650 sched_clear_tdname(td); 651#endif 652 653 /* 654 * mark as execed, wakeup the process that vforked (if any) and tell 655 * it that it now has its own resources back 656 */ 657 p->p_flag |= P_EXEC; 658 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 659 p->p_flag2 &= ~P2_NOTRACE; 660 if (p->p_flag & P_PPWAIT) { 661 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 662 cv_broadcast(&p->p_pwait); 663 } 664 665 /* 666 * Implement image setuid/setgid. 667 * 668 * Don't honor setuid/setgid if the filesystem prohibits it or if 669 * the process is being traced. 670 * 671 * We disable setuid/setgid/etc in capability mode on the basis 672 * that most setugid applications are not written with that 673 * environment in mind, and will therefore almost certainly operate 674 * incorrectly. In principle there's no reason that setugid 675 * applications might not be useful in capability mode, so we may want 676 * to reconsider this conservative design choice in the future. 677 * 678 * XXXMAC: For the time being, use NOSUID to also prohibit 679 * transitions on the file system. 680 */ 681 credential_changing = 0; 682 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != 683 attr.va_uid; 684 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != 685 attr.va_gid; 686#ifdef MAC 687 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 688 interpvplabel, imgp); 689 credential_changing |= will_transition; 690#endif 691 692 if (credential_changing && 693#ifdef CAPABILITY_MODE 694 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 695#endif 696 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 697 (p->p_flag & P_TRACED) == 0) { 698 /* 699 * Turn off syscall tracing for set-id programs, except for 700 * root. Record any set-id flags first to make sure that 701 * we do not regain any tracing during a possible block. 702 */ 703 setsugid(p); 704 705#ifdef KTRACE 706 if (p->p_tracecred != NULL && 707 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 708 ktrprocexec(p, &tracecred, &tracevp); 709#endif 710 /* 711 * Close any file descriptors 0..2 that reference procfs, 712 * then make sure file descriptors 0..2 are in use. 713 * 714 * setugidsafety() may call closef() and then pfind() 715 * which may grab the process lock. 716 * fdcheckstd() may call falloc() which may block to 717 * allocate memory, so temporarily drop the process lock. 718 */ 719 PROC_UNLOCK(p); 720 VOP_UNLOCK(imgp->vp, 0); 721 setugidsafety(td); 722 error = fdcheckstd(td); 723 if (error != 0) 724 goto done1; 725 newcred = crdup(oldcred); 726 euip = uifind(attr.va_uid); 727 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 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 proc_set_cred(p, newcred); 752 } else { 753 if (oldcred->cr_uid == oldcred->cr_ruid && 754 oldcred->cr_gid == oldcred->cr_rgid) 755 p->p_flag &= ~P_SUGID; 756 /* 757 * Implement correct POSIX saved-id behavior. 758 * 759 * XXX: It's not clear that the existing behavior is 760 * POSIX-compliant. A number of sources indicate that the 761 * saved uid/gid should only be updated if the new ruid is 762 * not equal to the old ruid, or the new euid is not equal 763 * to the old euid and the new euid is not equal to the old 764 * ruid. The FreeBSD code always updates the saved uid/gid. 765 * Also, this code uses the new (replaced) euid and egid as 766 * the source, which may or may not be the right ones to use. 767 */ 768 if (oldcred->cr_svuid != oldcred->cr_uid || 769 oldcred->cr_svgid != oldcred->cr_gid) { 770 PROC_UNLOCK(p); 771 VOP_UNLOCK(imgp->vp, 0); 772 newcred = crdup(oldcred); 773 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 774 PROC_LOCK(p); 775 change_svuid(newcred, newcred->cr_uid); 776 change_svgid(newcred, newcred->cr_gid); 777 proc_set_cred(p, newcred); 778 } 779 } 780 781 /* 782 * Store the vp for use in procfs. This vnode was referenced by namei 783 * or fgetvp_exec. 784 */ 785 oldtextvp = p->p_textvp; 786 p->p_textvp = newtextvp; 787 788#ifdef KDTRACE_HOOKS 789 /* 790 * Tell the DTrace fasttrap provider about the exec if it 791 * has declared an interest. 792 */ 793 if (dtrace_fasttrap_exec) 794 dtrace_fasttrap_exec(p); 795#endif 796 797 /* 798 * Notify others that we exec'd, and clear the P_INEXEC flag 799 * as we're now a bona fide freshly-execed process. 800 */ 801 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 802 p->p_flag &= ~P_INEXEC; 803 804 /* clear "fork but no exec" flag, as we _are_ execing */ 805 p->p_acflag &= ~AFORK; 806 807 /* 808 * Free any previous argument cache and replace it with 809 * the new argument cache, if any. 810 */ 811 oldargs = p->p_args; 812 p->p_args = newargs; 813 newargs = NULL; 814 815#ifdef HWPMC_HOOKS 816 /* 817 * Check if system-wide sampling is in effect or if the 818 * current process is using PMCs. If so, do exec() time 819 * processing. This processing needs to happen AFTER the 820 * P_INEXEC flag is cleared. 821 * 822 * The proc lock needs to be released before taking the PMC 823 * SX. 824 */ 825 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 826 PROC_UNLOCK(p); 827 VOP_UNLOCK(imgp->vp, 0); 828 pe.pm_credentialschanged = credential_changing; 829 pe.pm_entryaddr = imgp->entry_addr; 830 831 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 832 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 833 } else 834 PROC_UNLOCK(p); 835#else /* !HWPMC_HOOKS */ 836 PROC_UNLOCK(p); 837#endif 838 839 /* Set values passed into the program in registers. */ 840 if (p->p_sysent->sv_setregs) 841 (*p->p_sysent->sv_setregs)(td, imgp, 842 (u_long)(uintptr_t)stack_base); 843 else 844 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 845 846 vfs_mark_atime(imgp->vp, td->td_ucred); 847 848 SDT_PROBE1(proc, , , exec__success, args->fname); 849 850 VOP_UNLOCK(imgp->vp, 0); 851done1: 852 /* 853 * Free any resources malloc'd earlier that we didn't use. 854 */ 855 if (euip != NULL) 856 uifree(euip); 857 if (newcred != NULL) 858 crfree(oldcred); 859 860 /* 861 * Handle deferred decrement of ref counts. 862 */ 863 if (oldtextvp != NULL) 864 vrele(oldtextvp); 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 if (error != 0) 891 vput(imgp->vp); 892 else 893 VOP_UNLOCK(imgp->vp, 0); 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_PROBE1(proc, , , exec__failure, error); 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 struct rlimit rlim_stack; 1040 vm_offset_t sv_minuser, stack_addr; 1041 vm_map_t map; 1042 u_long ssiz; 1043 1044 imgp->vmspace_destroyed = 1; 1045 imgp->sysent = sv; 1046 1047 /* May be called with Giant held */ 1048 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1049 1050 /* 1051 * Blow away entire process VM, if address space not shared, 1052 * otherwise, create a new VM space so that other threads are 1053 * not disrupted 1054 */ 1055 map = &vmspace->vm_map; 1056 if (map_at_zero) 1057 sv_minuser = sv->sv_minuser; 1058 else 1059 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1060 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1061 vm_map_max(map) == sv->sv_maxuser) { 1062 shmexit(vmspace); 1063 pmap_remove_pages(vmspace_pmap(vmspace)); 1064 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1065 } else { 1066 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1067 if (error) 1068 return (error); 1069 vmspace = p->p_vmspace; 1070 map = &vmspace->vm_map; 1071 } 1072 1073 /* Map a shared page */ 1074 obj = sv->sv_shared_page_obj; 1075 if (obj != NULL) { 1076 vm_object_reference(obj); 1077 error = vm_map_fixed(map, obj, 0, 1078 sv->sv_shared_page_base, sv->sv_shared_page_len, 1079 VM_PROT_READ | VM_PROT_EXECUTE, 1080 VM_PROT_READ | VM_PROT_EXECUTE, 1081 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1082 if (error) { 1083 vm_object_deallocate(obj); 1084 return (error); 1085 } 1086 } 1087 1088 /* Allocate a new stack */ 1089 if (imgp->stack_sz != 0) { 1090 ssiz = trunc_page(imgp->stack_sz); 1091 PROC_LOCK(p); 1092 lim_rlimit(p, RLIMIT_STACK, &rlim_stack); 1093 PROC_UNLOCK(p); 1094 if (ssiz > rlim_stack.rlim_max) 1095 ssiz = rlim_stack.rlim_max; 1096 if (ssiz > rlim_stack.rlim_cur) { 1097 rlim_stack.rlim_cur = ssiz; 1098 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1099 } 1100 } else if (sv->sv_maxssiz != NULL) { 1101 ssiz = *sv->sv_maxssiz; 1102 } else { 1103 ssiz = maxssiz; 1104 } 1105 stack_addr = sv->sv_usrstack - ssiz; 1106 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1107 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1108 sv->sv_stackprot, 1109 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1110 if (error) 1111 return (error); 1112 1113#ifdef __ia64__ 1114 /* Allocate a new register stack */ 1115 error = vm_map_stack(map, IA64_BACKINGSTORE, (vm_size_t)ssiz, 1116 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1117 if (error) 1118 return (error); 1119#endif 1120 1121 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1122 * VM_STACK case, but they are still used to monitor the size of the 1123 * process stack so we can check the stack rlimit. 1124 */ 1125 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1126 vmspace->vm_maxsaddr = (char *)stack_addr; 1127 1128 return (0); 1129} 1130 1131/* 1132 * Copy out argument and environment strings from the old process address 1133 * space into the temporary string buffer. 1134 */ 1135int 1136exec_copyin_args(struct image_args *args, char *fname, 1137 enum uio_seg segflg, char **argv, char **envv) 1138{ 1139 u_long argp, envp; 1140 int error; 1141 size_t length; 1142 1143 bzero(args, sizeof(*args)); 1144 if (argv == NULL) 1145 return (EFAULT); 1146 1147 /* 1148 * Allocate demand-paged memory for the file name, argument, and 1149 * environment strings. 1150 */ 1151 error = exec_alloc_args(args); 1152 if (error != 0) 1153 return (error); 1154 1155 /* 1156 * Copy the file name. 1157 */ 1158 if (fname != NULL) { 1159 args->fname = args->buf; 1160 error = (segflg == UIO_SYSSPACE) ? 1161 copystr(fname, args->fname, PATH_MAX, &length) : 1162 copyinstr(fname, args->fname, PATH_MAX, &length); 1163 if (error != 0) 1164 goto err_exit; 1165 } else 1166 length = 0; 1167 1168 args->begin_argv = args->buf + length; 1169 args->endp = args->begin_argv; 1170 args->stringspace = ARG_MAX; 1171 1172 /* 1173 * extract arguments first 1174 */ 1175 for (;;) { 1176 error = fueword(argv++, &argp); 1177 if (error == -1) { 1178 error = EFAULT; 1179 goto err_exit; 1180 } 1181 if (argp == 0) 1182 break; 1183 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1184 args->stringspace, &length); 1185 if (error != 0) { 1186 if (error == ENAMETOOLONG) 1187 error = E2BIG; 1188 goto err_exit; 1189 } 1190 args->stringspace -= length; 1191 args->endp += length; 1192 args->argc++; 1193 } 1194 1195 args->begin_envv = args->endp; 1196 1197 /* 1198 * extract environment strings 1199 */ 1200 if (envv) { 1201 for (;;) { 1202 error = fueword(envv++, &envp); 1203 if (error == -1) { 1204 error = EFAULT; 1205 goto err_exit; 1206 } 1207 if (envp == 0) 1208 break; 1209 error = copyinstr((void *)(uintptr_t)envp, 1210 args->endp, args->stringspace, &length); 1211 if (error != 0) { 1212 if (error == ENAMETOOLONG) 1213 error = E2BIG; 1214 goto err_exit; 1215 } 1216 args->stringspace -= length; 1217 args->endp += length; 1218 args->envc++; 1219 } 1220 } 1221 1222 return (0); 1223 1224err_exit: 1225 exec_free_args(args); 1226 return (error); 1227} 1228 1229/* 1230 * Allocate temporary demand-paged, zero-filled memory for the file name, 1231 * argument, and environment strings. Returns zero if the allocation succeeds 1232 * and ENOMEM otherwise. 1233 */ 1234int 1235exec_alloc_args(struct image_args *args) 1236{ 1237 1238 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 1239 return (args->buf != NULL ? 0 : ENOMEM); 1240} 1241 1242void 1243exec_free_args(struct image_args *args) 1244{ 1245 1246 if (args->buf != NULL) { 1247 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf, 1248 PATH_MAX + ARG_MAX); 1249 args->buf = NULL; 1250 } 1251 if (args->fname_buf != NULL) { 1252 free(args->fname_buf, M_TEMP); 1253 args->fname_buf = NULL; 1254 } 1255} 1256 1257/* 1258 * Copy strings out to the new process address space, constructing new arg 1259 * and env vector tables. Return a pointer to the base so that it can be used 1260 * as the initial stack pointer. 1261 */ 1262register_t * 1263exec_copyout_strings(imgp) 1264 struct image_params *imgp; 1265{ 1266 int argc, envc; 1267 char **vectp; 1268 char *stringp; 1269 uintptr_t destp; 1270 register_t *stack_base; 1271 struct ps_strings *arginfo; 1272 struct proc *p; 1273 size_t execpath_len; 1274 int szsigcode, szps; 1275 char canary[sizeof(long) * 8]; 1276 1277 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1278 /* 1279 * Calculate string base and vector table pointers. 1280 * Also deal with signal trampoline code for this exec type. 1281 */ 1282 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1283 execpath_len = strlen(imgp->execpath) + 1; 1284 else 1285 execpath_len = 0; 1286 p = imgp->proc; 1287 szsigcode = 0; 1288 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1289 if (p->p_sysent->sv_sigcode_base == 0) { 1290 if (p->p_sysent->sv_szsigcode != NULL) 1291 szsigcode = *(p->p_sysent->sv_szsigcode); 1292 } 1293 destp = (uintptr_t)arginfo; 1294 1295 /* 1296 * install sigcode 1297 */ 1298 if (szsigcode != 0) { 1299 destp -= szsigcode; 1300 destp = rounddown2(destp, sizeof(void *)); 1301 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1302 } 1303 1304 /* 1305 * Copy the image path for the rtld. 1306 */ 1307 if (execpath_len != 0) { 1308 destp -= execpath_len; 1309 imgp->execpathp = destp; 1310 copyout(imgp->execpath, (void *)destp, execpath_len); 1311 } 1312 1313 /* 1314 * Prepare the canary for SSP. 1315 */ 1316 arc4rand(canary, sizeof(canary), 0); 1317 destp -= sizeof(canary); 1318 imgp->canary = destp; 1319 copyout(canary, (void *)destp, sizeof(canary)); 1320 imgp->canarylen = sizeof(canary); 1321 1322 /* 1323 * Prepare the pagesizes array. 1324 */ 1325 destp -= szps; 1326 destp = rounddown2(destp, sizeof(void *)); 1327 imgp->pagesizes = destp; 1328 copyout(pagesizes, (void *)destp, szps); 1329 imgp->pagesizeslen = szps; 1330 1331 destp -= ARG_MAX - imgp->args->stringspace; 1332 destp = rounddown2(destp, sizeof(void *)); 1333 1334 /* 1335 * If we have a valid auxargs ptr, prepare some room 1336 * on the stack. 1337 */ 1338 if (imgp->auxargs) { 1339 /* 1340 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1341 * lower compatibility. 1342 */ 1343 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1344 (AT_COUNT * 2); 1345 /* 1346 * The '+ 2' is for the null pointers at the end of each of 1347 * the arg and env vector sets,and imgp->auxarg_size is room 1348 * for argument of Runtime loader. 1349 */ 1350 vectp = (char **)(destp - (imgp->args->argc + 1351 imgp->args->envc + 2 + imgp->auxarg_size) 1352 * sizeof(char *)); 1353 } else { 1354 /* 1355 * The '+ 2' is for the null pointers at the end of each of 1356 * the arg and env vector sets 1357 */ 1358 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc 1359 + 2) * sizeof(char *)); 1360 } 1361 1362 /* 1363 * vectp also becomes our initial stack base 1364 */ 1365 stack_base = (register_t *)vectp; 1366 1367 stringp = imgp->args->begin_argv; 1368 argc = imgp->args->argc; 1369 envc = imgp->args->envc; 1370 1371 /* 1372 * Copy out strings - arguments and environment. 1373 */ 1374 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1375 1376 /* 1377 * Fill in "ps_strings" struct for ps, w, etc. 1378 */ 1379 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1380 suword32(&arginfo->ps_nargvstr, argc); 1381 1382 /* 1383 * Fill in argument portion of vector table. 1384 */ 1385 for (; argc > 0; --argc) { 1386 suword(vectp++, (long)(intptr_t)destp); 1387 while (*stringp++ != 0) 1388 destp++; 1389 destp++; 1390 } 1391 1392 /* a null vector table pointer separates the argp's from the envp's */ 1393 suword(vectp++, 0); 1394 1395 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1396 suword32(&arginfo->ps_nenvstr, envc); 1397 1398 /* 1399 * Fill in environment portion of vector table. 1400 */ 1401 for (; envc > 0; --envc) { 1402 suword(vectp++, (long)(intptr_t)destp); 1403 while (*stringp++ != 0) 1404 destp++; 1405 destp++; 1406 } 1407 1408 /* end of vector table is a null pointer */ 1409 suword(vectp, 0); 1410 1411 return (stack_base); 1412} 1413 1414/* 1415 * Check permissions of file to execute. 1416 * Called with imgp->vp locked. 1417 * Return 0 for success or error code on failure. 1418 */ 1419int 1420exec_check_permissions(imgp) 1421 struct image_params *imgp; 1422{ 1423 struct vnode *vp = imgp->vp; 1424 struct vattr *attr = imgp->attr; 1425 struct thread *td; 1426 int error, writecount; 1427 1428 td = curthread; 1429 1430 /* Get file attributes */ 1431 error = VOP_GETATTR(vp, attr, td->td_ucred); 1432 if (error) 1433 return (error); 1434 1435#ifdef MAC 1436 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1437 if (error) 1438 return (error); 1439#endif 1440 1441 /* 1442 * 1) Check if file execution is disabled for the filesystem that 1443 * this file resides on. 1444 * 2) Ensure that at least one execute bit is on. Otherwise, a 1445 * privileged user will always succeed, and we don't want this 1446 * to happen unless the file really is executable. 1447 * 3) Ensure that the file is a regular file. 1448 */ 1449 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1450 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1451 (attr->va_type != VREG)) 1452 return (EACCES); 1453 1454 /* 1455 * Zero length files can't be exec'd 1456 */ 1457 if (attr->va_size == 0) 1458 return (ENOEXEC); 1459 1460 /* 1461 * Check for execute permission to file based on current credentials. 1462 */ 1463 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1464 if (error) 1465 return (error); 1466 1467 /* 1468 * Check number of open-for-writes on the file and deny execution 1469 * if there are any. 1470 */ 1471 error = VOP_GET_WRITECOUNT(vp, &writecount); 1472 if (error != 0) 1473 return (error); 1474 if (writecount != 0) 1475 return (ETXTBSY); 1476 1477 /* 1478 * Call filesystem specific open routine (which does nothing in the 1479 * general case). 1480 */ 1481 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1482 if (error == 0) 1483 imgp->opened = 1; 1484 return (error); 1485} 1486 1487/* 1488 * Exec handler registration 1489 */ 1490int 1491exec_register(execsw_arg) 1492 const struct execsw *execsw_arg; 1493{ 1494 const struct execsw **es, **xs, **newexecsw; 1495 int count = 2; /* New slot and trailing NULL */ 1496 1497 if (execsw) 1498 for (es = execsw; *es; es++) 1499 count++; 1500 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1501 xs = newexecsw; 1502 if (execsw) 1503 for (es = execsw; *es; es++) 1504 *xs++ = *es; 1505 *xs++ = execsw_arg; 1506 *xs = NULL; 1507 if (execsw) 1508 free(execsw, M_TEMP); 1509 execsw = newexecsw; 1510 return (0); 1511} 1512 1513int 1514exec_unregister(execsw_arg) 1515 const struct execsw *execsw_arg; 1516{ 1517 const struct execsw **es, **xs, **newexecsw; 1518 int count = 1; 1519 1520 if (execsw == NULL) 1521 panic("unregister with no handlers left?\n"); 1522 1523 for (es = execsw; *es; es++) { 1524 if (*es == execsw_arg) 1525 break; 1526 } 1527 if (*es == NULL) 1528 return (ENOENT); 1529 for (es = execsw; *es; es++) 1530 if (*es != execsw_arg) 1531 count++; 1532 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1533 xs = newexecsw; 1534 for (es = execsw; *es; es++) 1535 if (*es != execsw_arg) 1536 *xs++ = *es; 1537 *xs = NULL; 1538 if (execsw) 1539 free(execsw, M_TEMP); 1540 execsw = newexecsw; 1541 return (0); 1542} 1543