kern_exec.c revision 302229
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 302229 2016-06-27 21:25:01Z 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, 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 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 *textvp = NULL, *binvp = NULL; 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 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_PROBE1(proc, kernel, , 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 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_flag2 & P2_NOTRACE_EXEC) == 0) 662 p->p_flag2 &= ~P2_NOTRACE; 663 if (p->p_flag & P_PPWAIT) { 664 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 665 cv_broadcast(&p->p_pwait); 666 } 667 668 /* 669 * Implement image setuid/setgid. 670 * 671 * Don't honor setuid/setgid if the filesystem prohibits it or if 672 * the process is being traced. 673 * 674 * We disable setuid/setgid/etc in capability mode on the basis 675 * that most setugid applications are not written with that 676 * environment in mind, and will therefore almost certainly operate 677 * incorrectly. In principle there's no reason that setugid 678 * applications might not be useful in capability mode, so we may want 679 * to reconsider this conservative design choice in the future. 680 * 681 * XXXMAC: For the time being, use NOSUID to also prohibit 682 * transitions on the file system. 683 */ 684 credential_changing = 0; 685 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != 686 attr.va_uid; 687 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != 688 attr.va_gid; 689#ifdef MAC 690 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 691 interpvplabel, imgp); 692 credential_changing |= will_transition; 693#endif 694 695 if (credential_changing && 696#ifdef CAPABILITY_MODE 697 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 698#endif 699 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 700 (p->p_flag & P_TRACED) == 0) { 701 /* 702 * Turn off syscall tracing for set-id programs, except for 703 * root. Record any set-id flags first to make sure that 704 * we do not regain any tracing during a possible block. 705 */ 706 setsugid(p); 707 708#ifdef KTRACE 709 if (p->p_tracecred != NULL && 710 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 711 ktrprocexec(p, &tracecred, &tracevp); 712#endif 713 /* 714 * Close any file descriptors 0..2 that reference procfs, 715 * then make sure file descriptors 0..2 are in use. 716 * 717 * setugidsafety() may call closef() and then pfind() 718 * which may grab the process lock. 719 * fdcheckstd() may call falloc() which may block to 720 * allocate memory, so temporarily drop the process lock. 721 */ 722 PROC_UNLOCK(p); 723 VOP_UNLOCK(imgp->vp, 0); 724 setugidsafety(td); 725 error = fdcheckstd(td); 726 if (error != 0) 727 goto done1; 728 newcred = crdup(oldcred); 729 euip = uifind(attr.va_uid); 730 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 731 PROC_LOCK(p); 732 /* 733 * Set the new credentials. 734 */ 735 if (attr.va_mode & S_ISUID) 736 change_euid(newcred, euip); 737 if (attr.va_mode & S_ISGID) 738 change_egid(newcred, attr.va_gid); 739#ifdef MAC 740 if (will_transition) { 741 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 742 interpvplabel, imgp); 743 } 744#endif 745 /* 746 * Implement correct POSIX saved-id behavior. 747 * 748 * XXXMAC: Note that the current logic will save the 749 * uid and gid if a MAC domain transition occurs, even 750 * though maybe it shouldn't. 751 */ 752 change_svuid(newcred, newcred->cr_uid); 753 change_svgid(newcred, newcred->cr_gid); 754 proc_set_cred(p, newcred); 755 } else { 756 if (oldcred->cr_uid == oldcred->cr_ruid && 757 oldcred->cr_gid == oldcred->cr_rgid) 758 p->p_flag &= ~P_SUGID; 759 /* 760 * Implement correct POSIX saved-id behavior. 761 * 762 * XXX: It's not clear that the existing behavior is 763 * POSIX-compliant. A number of sources indicate that the 764 * saved uid/gid should only be updated if the new ruid is 765 * not equal to the old ruid, or the new euid is not equal 766 * to the old euid and the new euid is not equal to the old 767 * ruid. The FreeBSD code always updates the saved uid/gid. 768 * Also, this code uses the new (replaced) euid and egid as 769 * the source, which may or may not be the right ones to use. 770 */ 771 if (oldcred->cr_svuid != oldcred->cr_uid || 772 oldcred->cr_svgid != oldcred->cr_gid) { 773 PROC_UNLOCK(p); 774 VOP_UNLOCK(imgp->vp, 0); 775 newcred = crdup(oldcred); 776 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 777 PROC_LOCK(p); 778 change_svuid(newcred, newcred->cr_uid); 779 change_svgid(newcred, newcred->cr_gid); 780 proc_set_cred(p, newcred); 781 } 782 } 783 784 /* 785 * Store the vp for use in procfs. This vnode was referenced prior 786 * to locking the proc lock. 787 */ 788 textvp = p->p_textvp; 789 p->p_textvp = binvp; 790 791#ifdef KDTRACE_HOOKS 792 /* 793 * Tell the DTrace fasttrap provider about the exec if it 794 * has declared an interest. 795 */ 796 if (dtrace_fasttrap_exec) 797 dtrace_fasttrap_exec(p); 798#endif 799 800 /* 801 * Notify others that we exec'd, and clear the P_INEXEC flag 802 * as we're now a bona fide freshly-execed process. 803 */ 804 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 805 p->p_flag &= ~P_INEXEC; 806 807 /* clear "fork but no exec" flag, as we _are_ execing */ 808 p->p_acflag &= ~AFORK; 809 810 /* 811 * Free any previous argument cache and replace it with 812 * the new argument cache, if any. 813 */ 814 oldargs = p->p_args; 815 p->p_args = newargs; 816 newargs = NULL; 817 818#ifdef HWPMC_HOOKS 819 /* 820 * Check if system-wide sampling is in effect or if the 821 * current process is using PMCs. If so, do exec() time 822 * processing. This processing needs to happen AFTER the 823 * P_INEXEC flag is cleared. 824 * 825 * The proc lock needs to be released before taking the PMC 826 * SX. 827 */ 828 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 829 PROC_UNLOCK(p); 830 VOP_UNLOCK(imgp->vp, 0); 831 pe.pm_credentialschanged = credential_changing; 832 pe.pm_entryaddr = imgp->entry_addr; 833 834 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 835 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 836 } else 837 PROC_UNLOCK(p); 838#else /* !HWPMC_HOOKS */ 839 PROC_UNLOCK(p); 840#endif 841 842 /* Set values passed into the program in registers. */ 843 if (p->p_sysent->sv_setregs) 844 (*p->p_sysent->sv_setregs)(td, imgp, 845 (u_long)(uintptr_t)stack_base); 846 else 847 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 848 849 vfs_mark_atime(imgp->vp, td->td_ucred); 850 851 SDT_PROBE1(proc, kernel, , exec__success, args->fname); 852 853 VOP_UNLOCK(imgp->vp, 0); 854done1: 855 /* 856 * Free any resources malloc'd earlier that we didn't use. 857 */ 858 if (euip != NULL) 859 uifree(euip); 860 if (newcred != NULL) 861 crfree(oldcred); 862 863 /* 864 * Handle deferred decrement of ref counts. 865 */ 866 if (textvp != NULL) 867 vrele(textvp); 868 if (binvp && error != 0) 869 vrele(binvp); 870#ifdef KTRACE 871 if (tracevp != NULL) 872 vrele(tracevp); 873 if (tracecred != NULL) 874 crfree(tracecred); 875#endif 876 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 877 pargs_drop(oldargs); 878 pargs_drop(newargs); 879 if (oldsigacts != NULL) 880 sigacts_free(oldsigacts); 881 882exec_fail_dealloc: 883 884 /* 885 * free various allocated resources 886 */ 887 if (imgp->firstpage != NULL) 888 exec_unmap_first_page(imgp); 889 890 if (imgp->vp != NULL) { 891 if (args->fname) 892 NDFREE(&nd, NDF_ONLY_PNBUF); 893 if (imgp->opened) 894 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 895 vput(imgp->vp); 896 } 897 898 if (imgp->object != NULL) 899 vm_object_deallocate(imgp->object); 900 901 free(imgp->freepath, M_TEMP); 902 903 if (error == 0) { 904 PROC_LOCK(p); 905 td->td_dbgflags |= TDB_EXEC; 906 PROC_UNLOCK(p); 907 908 /* 909 * Stop the process here if its stop event mask has 910 * the S_EXEC bit set. 911 */ 912 STOPEVENT(p, S_EXEC, 0); 913 goto done2; 914 } 915 916exec_fail: 917 /* we're done here, clear P_INEXEC */ 918 PROC_LOCK(p); 919 p->p_flag &= ~P_INEXEC; 920 PROC_UNLOCK(p); 921 922 SDT_PROBE1(proc, kernel, , exec__failure, error); 923 924done2: 925#ifdef MAC 926 mac_execve_exit(imgp); 927 mac_execve_interpreter_exit(interpvplabel); 928#endif 929 exec_free_args(args); 930 931 if (error && imgp->vmspace_destroyed) { 932 /* sorry, no more process anymore. exit gracefully */ 933 exit1(td, W_EXITCODE(0, SIGABRT)); 934 /* NOT REACHED */ 935 } 936 937#ifdef KTRACE 938 if (error == 0) 939 ktrprocctor(p); 940#endif 941 942 return (error); 943} 944 945int 946exec_map_first_page(imgp) 947 struct image_params *imgp; 948{ 949 int rv, i; 950 int initial_pagein; 951 vm_page_t ma[VM_INITIAL_PAGEIN]; 952 vm_object_t object; 953 954 if (imgp->firstpage != NULL) 955 exec_unmap_first_page(imgp); 956 957 object = imgp->vp->v_object; 958 if (object == NULL) 959 return (EACCES); 960 VM_OBJECT_WLOCK(object); 961#if VM_NRESERVLEVEL > 0 962 if ((object->flags & OBJ_COLORED) == 0) { 963 object->flags |= OBJ_COLORED; 964 object->pg_color = 0; 965 } 966#endif 967 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL); 968 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 969 initial_pagein = VM_INITIAL_PAGEIN; 970 if (initial_pagein > object->size) 971 initial_pagein = object->size; 972 for (i = 1; i < initial_pagein; i++) { 973 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 974 if (ma[i]->valid) 975 break; 976 if (vm_page_tryxbusy(ma[i])) 977 break; 978 } else { 979 ma[i] = vm_page_alloc(object, i, 980 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 981 if (ma[i] == NULL) 982 break; 983 } 984 } 985 initial_pagein = i; 986 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 987 ma[0] = vm_page_lookup(object, 0); 988 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) { 989 if (ma[0] != NULL) { 990 vm_page_lock(ma[0]); 991 vm_page_free(ma[0]); 992 vm_page_unlock(ma[0]); 993 } 994 VM_OBJECT_WUNLOCK(object); 995 return (EIO); 996 } 997 } 998 vm_page_xunbusy(ma[0]); 999 vm_page_lock(ma[0]); 1000 vm_page_hold(ma[0]); 1001 vm_page_activate(ma[0]); 1002 vm_page_unlock(ma[0]); 1003 VM_OBJECT_WUNLOCK(object); 1004 1005 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1006 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1007 1008 return (0); 1009} 1010 1011void 1012exec_unmap_first_page(imgp) 1013 struct image_params *imgp; 1014{ 1015 vm_page_t m; 1016 1017 if (imgp->firstpage != NULL) { 1018 m = sf_buf_page(imgp->firstpage); 1019 sf_buf_free(imgp->firstpage); 1020 imgp->firstpage = NULL; 1021 vm_page_lock(m); 1022 vm_page_unhold(m); 1023 vm_page_unlock(m); 1024 } 1025} 1026 1027/* 1028 * Destroy old address space, and allocate a new stack 1029 * The new stack is only SGROWSIZ large because it is grown 1030 * automatically in trap.c. 1031 */ 1032int 1033exec_new_vmspace(imgp, sv) 1034 struct image_params *imgp; 1035 struct sysentvec *sv; 1036{ 1037 int error; 1038 struct proc *p = imgp->proc; 1039 struct vmspace *vmspace = p->p_vmspace; 1040 vm_object_t obj; 1041 struct rlimit rlim_stack; 1042 vm_offset_t sv_minuser, stack_addr; 1043 vm_map_t map; 1044 u_long ssiz; 1045 1046 imgp->vmspace_destroyed = 1; 1047 imgp->sysent = sv; 1048 1049 /* May be called with Giant held */ 1050 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1051 1052 /* 1053 * Blow away entire process VM, if address space not shared, 1054 * otherwise, create a new VM space so that other threads are 1055 * not disrupted 1056 */ 1057 map = &vmspace->vm_map; 1058 if (map_at_zero) 1059 sv_minuser = sv->sv_minuser; 1060 else 1061 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1062 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1063 vm_map_max(map) == sv->sv_maxuser) { 1064 shmexit(vmspace); 1065 pmap_remove_pages(vmspace_pmap(vmspace)); 1066 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1067 } else { 1068 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1069 if (error) 1070 return (error); 1071 vmspace = p->p_vmspace; 1072 map = &vmspace->vm_map; 1073 } 1074 1075 /* Map a shared page */ 1076 obj = sv->sv_shared_page_obj; 1077 if (obj != NULL) { 1078 vm_object_reference(obj); 1079 error = vm_map_fixed(map, obj, 0, 1080 sv->sv_shared_page_base, sv->sv_shared_page_len, 1081 VM_PROT_READ | VM_PROT_EXECUTE, 1082 VM_PROT_READ | VM_PROT_EXECUTE, 1083 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1084 if (error) { 1085 vm_object_deallocate(obj); 1086 return (error); 1087 } 1088 } 1089 1090 /* Allocate a new stack */ 1091 if (imgp->stack_sz != 0) { 1092 ssiz = trunc_page(imgp->stack_sz); 1093 PROC_LOCK(p); 1094 lim_rlimit(p, RLIMIT_STACK, &rlim_stack); 1095 PROC_UNLOCK(p); 1096 if (ssiz > rlim_stack.rlim_max) 1097 ssiz = rlim_stack.rlim_max; 1098 if (ssiz > rlim_stack.rlim_cur) { 1099 rlim_stack.rlim_cur = ssiz; 1100 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1101 } 1102 } else if (sv->sv_maxssiz != NULL) { 1103 ssiz = *sv->sv_maxssiz; 1104 } else { 1105 ssiz = maxssiz; 1106 } 1107 stack_addr = sv->sv_usrstack - ssiz; 1108 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1109 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1110 sv->sv_stackprot, 1111 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1112 if (error) 1113 return (error); 1114 1115#ifdef __ia64__ 1116 /* Allocate a new register stack */ 1117 error = vm_map_stack(map, IA64_BACKINGSTORE, (vm_size_t)ssiz, 1118 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1119 if (error) 1120 return (error); 1121#endif 1122 1123 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1124 * VM_STACK case, but they are still used to monitor the size of the 1125 * process stack so we can check the stack rlimit. 1126 */ 1127 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1128 vmspace->vm_maxsaddr = (char *)stack_addr; 1129 1130 return (0); 1131} 1132 1133/* 1134 * Copy out argument and environment strings from the old process address 1135 * space into the temporary string buffer. 1136 */ 1137int 1138exec_copyin_args(struct image_args *args, char *fname, 1139 enum uio_seg segflg, char **argv, char **envv) 1140{ 1141 u_long argp, envp; 1142 int error; 1143 size_t length; 1144 1145 bzero(args, sizeof(*args)); 1146 if (argv == NULL) 1147 return (EFAULT); 1148 1149 /* 1150 * Allocate demand-paged memory for the file name, argument, and 1151 * environment strings. 1152 */ 1153 error = exec_alloc_args(args); 1154 if (error != 0) 1155 return (error); 1156 1157 /* 1158 * Copy the file name. 1159 */ 1160 if (fname != NULL) { 1161 args->fname = args->buf; 1162 error = (segflg == UIO_SYSSPACE) ? 1163 copystr(fname, args->fname, PATH_MAX, &length) : 1164 copyinstr(fname, args->fname, PATH_MAX, &length); 1165 if (error != 0) 1166 goto err_exit; 1167 } else 1168 length = 0; 1169 1170 args->begin_argv = args->buf + length; 1171 args->endp = args->begin_argv; 1172 args->stringspace = ARG_MAX; 1173 1174 /* 1175 * extract arguments first 1176 */ 1177 for (;;) { 1178 error = fueword(argv++, &argp); 1179 if (error == -1) { 1180 error = EFAULT; 1181 goto err_exit; 1182 } 1183 if (argp == 0) 1184 break; 1185 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1186 args->stringspace, &length); 1187 if (error != 0) { 1188 if (error == ENAMETOOLONG) 1189 error = E2BIG; 1190 goto err_exit; 1191 } 1192 args->stringspace -= length; 1193 args->endp += length; 1194 args->argc++; 1195 } 1196 1197 args->begin_envv = args->endp; 1198 1199 /* 1200 * extract environment strings 1201 */ 1202 if (envv) { 1203 for (;;) { 1204 error = fueword(envv++, &envp); 1205 if (error == -1) { 1206 error = EFAULT; 1207 goto err_exit; 1208 } 1209 if (envp == 0) 1210 break; 1211 error = copyinstr((void *)(uintptr_t)envp, 1212 args->endp, args->stringspace, &length); 1213 if (error != 0) { 1214 if (error == ENAMETOOLONG) 1215 error = E2BIG; 1216 goto err_exit; 1217 } 1218 args->stringspace -= length; 1219 args->endp += length; 1220 args->envc++; 1221 } 1222 } 1223 1224 return (0); 1225 1226err_exit: 1227 exec_free_args(args); 1228 return (error); 1229} 1230 1231/* 1232 * Allocate temporary demand-paged, zero-filled memory for the file name, 1233 * argument, and environment strings. Returns zero if the allocation succeeds 1234 * and ENOMEM otherwise. 1235 */ 1236int 1237exec_alloc_args(struct image_args *args) 1238{ 1239 1240 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 1241 return (args->buf != NULL ? 0 : ENOMEM); 1242} 1243 1244void 1245exec_free_args(struct image_args *args) 1246{ 1247 1248 if (args->buf != NULL) { 1249 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf, 1250 PATH_MAX + ARG_MAX); 1251 args->buf = NULL; 1252 } 1253 if (args->fname_buf != NULL) { 1254 free(args->fname_buf, M_TEMP); 1255 args->fname_buf = NULL; 1256 } 1257} 1258 1259/* 1260 * Copy strings out to the new process address space, constructing new arg 1261 * and env vector tables. Return a pointer to the base so that it can be used 1262 * as the initial stack pointer. 1263 */ 1264register_t * 1265exec_copyout_strings(imgp) 1266 struct image_params *imgp; 1267{ 1268 int argc, envc; 1269 char **vectp; 1270 char *stringp; 1271 uintptr_t destp; 1272 register_t *stack_base; 1273 struct ps_strings *arginfo; 1274 struct proc *p; 1275 size_t execpath_len; 1276 int szsigcode, szps; 1277 char canary[sizeof(long) * 8]; 1278 1279 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1280 /* 1281 * Calculate string base and vector table pointers. 1282 * Also deal with signal trampoline code for this exec type. 1283 */ 1284 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1285 execpath_len = strlen(imgp->execpath) + 1; 1286 else 1287 execpath_len = 0; 1288 p = imgp->proc; 1289 szsigcode = 0; 1290 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1291 if (p->p_sysent->sv_sigcode_base == 0) { 1292 if (p->p_sysent->sv_szsigcode != NULL) 1293 szsigcode = *(p->p_sysent->sv_szsigcode); 1294 } 1295 destp = (uintptr_t)arginfo; 1296 1297 /* 1298 * install sigcode 1299 */ 1300 if (szsigcode != 0) { 1301 destp -= szsigcode; 1302 destp = rounddown2(destp, sizeof(void *)); 1303 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1304 } 1305 1306 /* 1307 * Copy the image path for the rtld. 1308 */ 1309 if (execpath_len != 0) { 1310 destp -= execpath_len; 1311 imgp->execpathp = destp; 1312 copyout(imgp->execpath, (void *)destp, execpath_len); 1313 } 1314 1315 /* 1316 * Prepare the canary for SSP. 1317 */ 1318 arc4rand(canary, sizeof(canary), 0); 1319 destp -= sizeof(canary); 1320 imgp->canary = destp; 1321 copyout(canary, (void *)destp, sizeof(canary)); 1322 imgp->canarylen = sizeof(canary); 1323 1324 /* 1325 * Prepare the pagesizes array. 1326 */ 1327 destp -= szps; 1328 destp = rounddown2(destp, sizeof(void *)); 1329 imgp->pagesizes = destp; 1330 copyout(pagesizes, (void *)destp, szps); 1331 imgp->pagesizeslen = szps; 1332 1333 destp -= ARG_MAX - imgp->args->stringspace; 1334 destp = rounddown2(destp, sizeof(void *)); 1335 1336 /* 1337 * If we have a valid auxargs ptr, prepare some room 1338 * on the stack. 1339 */ 1340 if (imgp->auxargs) { 1341 /* 1342 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1343 * lower compatibility. 1344 */ 1345 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1346 (AT_COUNT * 2); 1347 /* 1348 * The '+ 2' is for the null pointers at the end of each of 1349 * the arg and env vector sets,and imgp->auxarg_size is room 1350 * for argument of Runtime loader. 1351 */ 1352 vectp = (char **)(destp - (imgp->args->argc + 1353 imgp->args->envc + 2 + imgp->auxarg_size) 1354 * sizeof(char *)); 1355 } else { 1356 /* 1357 * The '+ 2' is for the null pointers at the end of each of 1358 * the arg and env vector sets 1359 */ 1360 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc 1361 + 2) * sizeof(char *)); 1362 } 1363 1364 /* 1365 * vectp also becomes our initial stack base 1366 */ 1367 stack_base = (register_t *)vectp; 1368 1369 stringp = imgp->args->begin_argv; 1370 argc = imgp->args->argc; 1371 envc = imgp->args->envc; 1372 1373 /* 1374 * Copy out strings - arguments and environment. 1375 */ 1376 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1377 1378 /* 1379 * Fill in "ps_strings" struct for ps, w, etc. 1380 */ 1381 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1382 suword32(&arginfo->ps_nargvstr, argc); 1383 1384 /* 1385 * Fill in argument portion of vector table. 1386 */ 1387 for (; argc > 0; --argc) { 1388 suword(vectp++, (long)(intptr_t)destp); 1389 while (*stringp++ != 0) 1390 destp++; 1391 destp++; 1392 } 1393 1394 /* a null vector table pointer separates the argp's from the envp's */ 1395 suword(vectp++, 0); 1396 1397 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1398 suword32(&arginfo->ps_nenvstr, envc); 1399 1400 /* 1401 * Fill in environment portion of vector table. 1402 */ 1403 for (; envc > 0; --envc) { 1404 suword(vectp++, (long)(intptr_t)destp); 1405 while (*stringp++ != 0) 1406 destp++; 1407 destp++; 1408 } 1409 1410 /* end of vector table is a null pointer */ 1411 suword(vectp, 0); 1412 1413 return (stack_base); 1414} 1415 1416/* 1417 * Check permissions of file to execute. 1418 * Called with imgp->vp locked. 1419 * Return 0 for success or error code on failure. 1420 */ 1421int 1422exec_check_permissions(imgp) 1423 struct image_params *imgp; 1424{ 1425 struct vnode *vp = imgp->vp; 1426 struct vattr *attr = imgp->attr; 1427 struct thread *td; 1428 int error, writecount; 1429 1430 td = curthread; 1431 1432 /* Get file attributes */ 1433 error = VOP_GETATTR(vp, attr, td->td_ucred); 1434 if (error) 1435 return (error); 1436 1437#ifdef MAC 1438 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1439 if (error) 1440 return (error); 1441#endif 1442 1443 /* 1444 * 1) Check if file execution is disabled for the filesystem that 1445 * this file resides on. 1446 * 2) Ensure that at least one execute bit is on. Otherwise, a 1447 * privileged user will always succeed, and we don't want this 1448 * to happen unless the file really is executable. 1449 * 3) Ensure that the file is a regular file. 1450 */ 1451 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1452 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1453 (attr->va_type != VREG)) 1454 return (EACCES); 1455 1456 /* 1457 * Zero length files can't be exec'd 1458 */ 1459 if (attr->va_size == 0) 1460 return (ENOEXEC); 1461 1462 /* 1463 * Check for execute permission to file based on current credentials. 1464 */ 1465 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1466 if (error) 1467 return (error); 1468 1469 /* 1470 * Check number of open-for-writes on the file and deny execution 1471 * if there are any. 1472 */ 1473 error = VOP_GET_WRITECOUNT(vp, &writecount); 1474 if (error != 0) 1475 return (error); 1476 if (writecount != 0) 1477 return (ETXTBSY); 1478 1479 /* 1480 * Call filesystem specific open routine (which does nothing in the 1481 * general case). 1482 */ 1483 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1484 if (error == 0) 1485 imgp->opened = 1; 1486 return (error); 1487} 1488 1489/* 1490 * Exec handler registration 1491 */ 1492int 1493exec_register(execsw_arg) 1494 const struct execsw *execsw_arg; 1495{ 1496 const struct execsw **es, **xs, **newexecsw; 1497 int count = 2; /* New slot and trailing NULL */ 1498 1499 if (execsw) 1500 for (es = execsw; *es; es++) 1501 count++; 1502 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1503 if (newexecsw == NULL) 1504 return (ENOMEM); 1505 xs = newexecsw; 1506 if (execsw) 1507 for (es = execsw; *es; es++) 1508 *xs++ = *es; 1509 *xs++ = execsw_arg; 1510 *xs = NULL; 1511 if (execsw) 1512 free(execsw, M_TEMP); 1513 execsw = newexecsw; 1514 return (0); 1515} 1516 1517int 1518exec_unregister(execsw_arg) 1519 const struct execsw *execsw_arg; 1520{ 1521 const struct execsw **es, **xs, **newexecsw; 1522 int count = 1; 1523 1524 if (execsw == NULL) 1525 panic("unregister with no handlers left?\n"); 1526 1527 for (es = execsw; *es; es++) { 1528 if (*es == execsw_arg) 1529 break; 1530 } 1531 if (*es == NULL) 1532 return (ENOENT); 1533 for (es = execsw; *es; es++) 1534 if (*es != execsw_arg) 1535 count++; 1536 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1537 if (newexecsw == NULL) 1538 return (ENOMEM); 1539 xs = newexecsw; 1540 for (es = execsw; *es; es++) 1541 if (*es != execsw_arg) 1542 *xs++ = *es; 1543 *xs = NULL; 1544 if (execsw) 1545 free(execsw, M_TEMP); 1546 execsw = newexecsw; 1547 return (0); 1548} 1549