1 /*- 2 * Copyright (c) 1993, David Greenman 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include "opt_hwpmc_hooks.h" 31 #include "opt_kdtrace.h" 32 #include "opt_ktrace.h" 33 #include "opt_mac.h" 34 #include "opt_vm.h" 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/eventhandler.h> 39 #include <sys/lock.h> 40 #include <sys/mutex.h> 41 #include <sys/sysproto.h> 42 #include <sys/signalvar.h> 43 #include <sys/kernel.h> 44 #include <sys/mount.h> 45 #include <sys/filedesc.h> 46 #include <sys/fcntl.h> 47 #include <sys/acct.h> 48 #include <sys/exec.h> 49 #include <sys/imgact.h> 50 #include <sys/imgact_elf.h> 51 #include <sys/wait.h> 52 #include <sys/malloc.h> 53 #include <sys/priv.h> 54 #include <sys/proc.h> 55 #include <sys/pioctl.h> 56 #include <sys/namei.h> 57 #include <sys/resourcevar.h> 58 #include <sys/sdt.h> 59 #include <sys/sf_buf.h> 60 #include <sys/syscallsubr.h> 61 #include <sys/sysent.h> 62 #include <sys/shm.h> 63 #include <sys/sysctl.h> 64 #include <sys/vnode.h> 65 #ifdef KTRACE 66 #include <sys/ktrace.h> 67 #endif 68 69 #include <vm/vm.h> 70 #include <vm/vm_param.h> 71 #include <vm/pmap.h> 72 #include <vm/vm_page.h> 73 #include <vm/vm_map.h> 74 #include <vm/vm_kern.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_object.h> 77 #include <vm/vm_pager.h> 78 79 #ifdef HWPMC_HOOKS 80 #include <sys/pmckern.h> 81 #endif 82 83 #include <machine/reg.h> 84 85 #include <security/audit/audit.h> 86 #include <security/mac/mac_framework.h> 87 88 #ifdef KDTRACE_HOOKS 89 #include <sys/dtrace_bsd.h> 90 dtrace_execexit_func_t dtrace_fasttrap_exec; 91 #endif 92 93 SDT_PROVIDER_DECLARE(proc); 94 SDT_PROBE_DEFINE(proc, kernel, , exec); 95 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *"); 96 SDT_PROBE_DEFINE(proc, kernel, , exec_failure); 97 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int"); 98 SDT_PROBE_DEFINE(proc, kernel, , exec_success); 99 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *"); 100 101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 102 103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 106 static int do_execve(struct thread *td, struct image_args *args, 107 struct mac *mac_p); 108 static void exec_free_args(struct image_args *); 109 110 /* XXX This should be vm_size_t. */ 111 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 112 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 113 114 /* XXX This should be vm_size_t. */ 115 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD, 116 NULL, 0, sysctl_kern_usrstack, "LU", ""); 117 118 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 119 NULL, 0, sysctl_kern_stackprot, "I", ""); 120 121 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 122 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 123 &ps_arg_cache_limit, 0, ""); 124 125 static int 126 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 127 { 128 struct proc *p; 129 int error; 130 131 p = curproc; 132 #ifdef SCTL_MASK32 133 if (req->flags & SCTL_MASK32) { 134 unsigned int val; 135 val = (unsigned int)p->p_sysent->sv_psstrings; 136 error = SYSCTL_OUT(req, &val, sizeof(val)); 137 } else 138 #endif 139 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 140 sizeof(p->p_sysent->sv_psstrings)); 141 return error; 142 } 143 144 static int 145 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 146 { 147 struct proc *p; 148 int error; 149 150 p = curproc; 151 #ifdef SCTL_MASK32 152 if (req->flags & SCTL_MASK32) { 153 unsigned int val; 154 val = (unsigned int)p->p_sysent->sv_usrstack; 155 error = SYSCTL_OUT(req, &val, sizeof(val)); 156 } else 157 #endif 158 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 159 sizeof(p->p_sysent->sv_usrstack)); 160 return error; 161 } 162 163 static int 164 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 165 { 166 struct proc *p; 167 168 p = curproc; 169 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 170 sizeof(p->p_sysent->sv_stackprot))); 171 } 172 173 /* 174 * Each of the items is a pointer to a `const struct execsw', hence the 175 * double pointer here. 176 */ 177 static const struct execsw **execsw; 178 179 #ifndef _SYS_SYSPROTO_H_ 180 struct execve_args { 181 char *fname; 182 char **argv; 183 char **envv; 184 }; 185 #endif 186 187 int 188 execve(td, uap) 189 struct thread *td; 190 struct execve_args /* { 191 char *fname; 192 char **argv; 193 char **envv; 194 } */ *uap; 195 { 196 int error; 197 struct image_args args; 198 199 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 200 uap->argv, uap->envv); 201 if (error == 0) 202 error = kern_execve(td, &args, NULL); 203 return (error); 204 } 205 206 #ifndef _SYS_SYSPROTO_H_ 207 struct fexecve_args { 208 int fd; 209 char **argv; 210 char **envv; 211 } 212 #endif 213 int 214 fexecve(struct thread *td, struct fexecve_args *uap) 215 { 216 int error; 217 struct image_args args; 218 219 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 220 uap->argv, uap->envv); 221 if (error == 0) { 222 args.fd = uap->fd; 223 error = kern_execve(td, &args, NULL); 224 } 225 return (error); 226 } 227 228 #ifndef _SYS_SYSPROTO_H_ 229 struct __mac_execve_args { 230 char *fname; 231 char **argv; 232 char **envv; 233 struct mac *mac_p; 234 }; 235 #endif 236 237 int 238 __mac_execve(td, uap) 239 struct thread *td; 240 struct __mac_execve_args /* { 241 char *fname; 242 char **argv; 243 char **envv; 244 struct mac *mac_p; 245 } */ *uap; 246 { 247 #ifdef MAC 248 int error; 249 struct image_args args; 250 251 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 252 uap->argv, uap->envv); 253 if (error == 0) 254 error = kern_execve(td, &args, uap->mac_p); 255 return (error); 256 #else 257 return (ENOSYS); 258 #endif 259 } 260 261 /* 262 * XXX: kern_execve has the astonishing property of not always returning to 263 * the caller. If sufficiently bad things happen during the call to 264 * do_execve(), it can end up calling exit1(); as a result, callers must 265 * avoid doing anything which they might need to undo (e.g., allocating 266 * memory). 267 */ 268 int 269 kern_execve(td, args, mac_p) 270 struct thread *td; 271 struct image_args *args; 272 struct mac *mac_p; 273 { 274 struct proc *p = td->td_proc; 275 int error; 276 277 AUDIT_ARG(argv, args->begin_argv, args->argc, 278 args->begin_envv - args->begin_argv); 279 AUDIT_ARG(envv, args->begin_envv, args->envc, 280 args->endp - args->begin_envv); 281 if (p->p_flag & P_HADTHREADS) { 282 PROC_LOCK(p); 283 if (thread_single(SINGLE_BOUNDARY)) { 284 PROC_UNLOCK(p); 285 exec_free_args(args); 286 return (ERESTART); /* Try again later. */ 287 } 288 PROC_UNLOCK(p); 289 } 290 291 error = do_execve(td, args, mac_p); 292 293 if (p->p_flag & P_HADTHREADS) { 294 PROC_LOCK(p); 295 /* 296 * If success, we upgrade to SINGLE_EXIT state to 297 * force other threads to suicide. 298 */ 299 if (error == 0) 300 thread_single(SINGLE_EXIT); 301 else 302 thread_single_end(); 303 PROC_UNLOCK(p); 304 } 305 306 return (error); 307 } 308 309 /* 310 * In-kernel implementation of execve(). All arguments are assumed to be 311 * userspace pointers from the passed thread. 312 */ 313 static int 314 do_execve(td, args, mac_p) 315 struct thread *td; 316 struct image_args *args; 317 struct mac *mac_p; 318 { 319 struct proc *p = td->td_proc; 320 struct nameidata nd, *ndp; 321 struct ucred *newcred = NULL, *oldcred; 322 struct uidinfo *euip; 323 register_t *stack_base; 324 int error, len = 0, i; 325 struct image_params image_params, *imgp; 326 struct vattr attr; 327 int (*img_first)(struct image_params *); 328 struct pargs *oldargs = NULL, *newargs = NULL; 329 struct sigacts *oldsigacts, *newsigacts; 330 #ifdef KTRACE 331 struct vnode *tracevp = NULL; 332 struct ucred *tracecred = NULL; 333 #endif 334 struct vnode *textvp = NULL, *binvp = NULL; 335 int credential_changing; 336 int vfslocked; 337 int textset; 338 #ifdef MAC 339 struct label *interplabel = NULL; 340 int will_transition; 341 #endif 342 #ifdef HWPMC_HOOKS 343 struct pmckern_procexec pe; 344 #endif 345 static const char fexecv_proc_title[] = "(fexecv)"; 346 347 vfslocked = 0; 348 imgp = &image_params; 349 350 /* 351 * Lock the process and set the P_INEXEC flag to indicate that 352 * it should be left alone until we're done here. This is 353 * necessary to avoid race conditions - e.g. in ptrace() - 354 * that might allow a local user to illicitly obtain elevated 355 * privileges. 356 */ 357 PROC_LOCK(p); 358 KASSERT((p->p_flag & P_INEXEC) == 0, 359 ("%s(): process already has P_INEXEC flag", __func__)); 360 p->p_flag |= P_INEXEC; 361 PROC_UNLOCK(p); 362 363 /* 364 * Initialize part of the common data 365 */ 366 imgp->proc = p; 367 imgp->execlabel = NULL; 368 imgp->attr = &attr; 369 imgp->entry_addr = 0; 370 imgp->vmspace_destroyed = 0; 371 imgp->interpreted = 0; 372 imgp->opened = 0; 373 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX; 374 imgp->auxargs = NULL; 375 imgp->vp = NULL; 376 imgp->object = NULL; 377 imgp->firstpage = NULL; 378 imgp->ps_strings = 0; 379 imgp->auxarg_size = 0; 380 imgp->args = args; 381 382 #ifdef MAC 383 error = mac_execve_enter(imgp, mac_p); 384 if (error) 385 goto exec_fail; 386 #endif 387 388 imgp->image_header = NULL; 389 390 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); 391 392 /* 393 * Translate the file name. namei() returns a vnode pointer 394 * in ni_vp amoung other things. 395 * 396 * XXXAUDIT: It would be desirable to also audit the name of the 397 * interpreter if this is an interpreted binary. 398 */ 399 if (args->fname != NULL) { 400 ndp = &nd; 401 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 402 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 403 } 404 405 interpret: 406 if (args->fname != NULL) { 407 error = namei(ndp); 408 if (error) 409 goto exec_fail; 410 411 vfslocked = NDHASGIANT(ndp); 412 binvp = ndp->ni_vp; 413 imgp->vp = binvp; 414 } else { 415 error = fgetvp(td, args->fd, &binvp); 416 if (error) 417 goto exec_fail; 418 vfslocked = VFS_LOCK_GIANT(binvp->v_mount); 419 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY); 420 imgp->vp = binvp; 421 } 422 423 /* 424 * Check file permissions (also 'opens' file) 425 */ 426 error = exec_check_permissions(imgp); 427 if (error) 428 goto exec_fail_dealloc; 429 430 imgp->object = imgp->vp->v_object; 431 if (imgp->object != NULL) 432 vm_object_reference(imgp->object); 433 434 /* 435 * Set VV_TEXT now so no one can write to the executable while we're 436 * activating it. 437 * 438 * Remember if this was set before and unset it in case this is not 439 * actually an executable image. 440 */ 441 textset = imgp->vp->v_vflag & VV_TEXT; 442 imgp->vp->v_vflag |= VV_TEXT; 443 444 error = exec_map_first_page(imgp); 445 if (error) 446 goto exec_fail_dealloc; 447 448 imgp->proc->p_osrel = 0; 449 /* 450 * If the current process has a special image activator it 451 * wants to try first, call it. For example, emulating shell 452 * scripts differently. 453 */ 454 error = -1; 455 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 456 error = img_first(imgp); 457 458 /* 459 * Loop through the list of image activators, calling each one. 460 * An activator returns -1 if there is no match, 0 on success, 461 * and an error otherwise. 462 */ 463 for (i = 0; error == -1 && execsw[i]; ++i) { 464 if (execsw[i]->ex_imgact == NULL || 465 execsw[i]->ex_imgact == img_first) { 466 continue; 467 } 468 error = (*execsw[i]->ex_imgact)(imgp); 469 } 470 471 if (error) { 472 if (error == -1) { 473 if (textset == 0) 474 imgp->vp->v_vflag &= ~VV_TEXT; 475 error = ENOEXEC; 476 } 477 goto exec_fail_dealloc; 478 } 479 480 /* 481 * Special interpreter operation, cleanup and loop up to try to 482 * activate the interpreter. 483 */ 484 if (imgp->interpreted) { 485 exec_unmap_first_page(imgp); 486 /* 487 * VV_TEXT needs to be unset for scripts. There is a short 488 * period before we determine that something is a script where 489 * VV_TEXT will be set. The vnode lock is held over this 490 * entire period so nothing should illegitimately be blocked. 491 */ 492 imgp->vp->v_vflag &= ~VV_TEXT; 493 /* free name buffer and old vnode */ 494 if (args->fname != NULL) 495 NDFREE(ndp, NDF_ONLY_PNBUF); 496 #ifdef MAC 497 interplabel = mac_vnode_label_alloc(); 498 mac_vnode_copy_label(binvp->v_label, interplabel); 499 #endif 500 if (imgp->opened) { 501 VOP_CLOSE(binvp, FREAD, td->td_ucred, td); 502 imgp->opened = 0; 503 } 504 vput(binvp); 505 vm_object_deallocate(imgp->object); 506 imgp->object = NULL; 507 VFS_UNLOCK_GIANT(vfslocked); 508 vfslocked = 0; 509 /* set new name to that of the interpreter */ 510 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE, 511 UIO_SYSSPACE, imgp->interpreter_name, td); 512 args->fname = imgp->interpreter_name; 513 goto interpret; 514 } 515 516 /* 517 * Copy out strings (args and env) and initialize stack base 518 */ 519 if (p->p_sysent->sv_copyout_strings) 520 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 521 else 522 stack_base = exec_copyout_strings(imgp); 523 524 /* 525 * If custom stack fixup routine present for this process 526 * let it do the stack setup. 527 * Else stuff argument count as first item on stack 528 */ 529 if (p->p_sysent->sv_fixup != NULL) 530 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 531 else 532 suword(--stack_base, imgp->args->argc); 533 534 /* 535 * For security and other reasons, the file descriptor table cannot 536 * be shared after an exec. 537 */ 538 fdunshare(p, td); 539 540 /* 541 * Malloc things before we need locks. 542 */ 543 newcred = crget(); 544 euip = uifind(attr.va_uid); 545 i = imgp->args->begin_envv - imgp->args->begin_argv; 546 /* Cache arguments if they fit inside our allowance */ 547 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 548 newargs = pargs_alloc(i); 549 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 550 } 551 552 /* close files on exec */ 553 VOP_UNLOCK(imgp->vp, 0); 554 fdcloseexec(td); 555 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 556 557 /* Get a reference to the vnode prior to locking the proc */ 558 VREF(binvp); 559 560 /* 561 * For security and other reasons, signal handlers cannot 562 * be shared after an exec. The new process gets a copy of the old 563 * handlers. In execsigs(), the new process will have its signals 564 * reset. 565 */ 566 PROC_LOCK(p); 567 if (sigacts_shared(p->p_sigacts)) { 568 oldsigacts = p->p_sigacts; 569 PROC_UNLOCK(p); 570 newsigacts = sigacts_alloc(); 571 sigacts_copy(newsigacts, oldsigacts); 572 PROC_LOCK(p); 573 p->p_sigacts = newsigacts; 574 } else 575 oldsigacts = NULL; 576 577 /* Stop profiling */ 578 stopprofclock(p); 579 580 /* reset caught signals */ 581 execsigs(p); 582 583 /* name this process - nameiexec(p, ndp) */ 584 if (args->fname) { 585 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN); 586 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len); 587 } else { 588 len = MAXCOMLEN; 589 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0) 590 len = MAXCOMLEN; 591 else { 592 len = sizeof(fexecv_proc_title); 593 bcopy(fexecv_proc_title, p->p_comm, len); 594 } 595 } 596 p->p_comm[len] = 0; 597 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 598 599 /* 600 * mark as execed, wakeup the process that vforked (if any) and tell 601 * it that it now has its own resources back 602 */ 603 p->p_flag |= P_EXEC; 604 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 605 p->p_flag &= ~P_PPWAIT; 606 wakeup(p->p_pptr); 607 } 608 609 /* 610 * Implement image setuid/setgid. 611 * 612 * Don't honor setuid/setgid if the filesystem prohibits it or if 613 * the process is being traced. 614 * 615 * XXXMAC: For the time being, use NOSUID to also prohibit 616 * transitions on the file system. 617 */ 618 oldcred = p->p_ucred; 619 credential_changing = 0; 620 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid != 621 attr.va_uid; 622 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid != 623 attr.va_gid; 624 #ifdef MAC 625 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 626 interplabel, imgp); 627 credential_changing |= will_transition; 628 #endif 629 630 if (credential_changing && 631 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 632 (p->p_flag & P_TRACED) == 0) { 633 /* 634 * Turn off syscall tracing for set-id programs, except for 635 * root. Record any set-id flags first to make sure that 636 * we do not regain any tracing during a possible block. 637 */ 638 setsugid(p); 639 640 #ifdef KTRACE 641 if (p->p_tracevp != NULL && 642 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) { 643 mtx_lock(&ktrace_mtx); 644 p->p_traceflag = 0; 645 tracevp = p->p_tracevp; 646 p->p_tracevp = NULL; 647 tracecred = p->p_tracecred; 648 p->p_tracecred = NULL; 649 mtx_unlock(&ktrace_mtx); 650 } 651 #endif 652 /* 653 * Close any file descriptors 0..2 that reference procfs, 654 * then make sure file descriptors 0..2 are in use. 655 * 656 * setugidsafety() may call closef() and then pfind() 657 * which may grab the process lock. 658 * fdcheckstd() may call falloc() which may block to 659 * allocate memory, so temporarily drop the process lock. 660 */ 661 PROC_UNLOCK(p); 662 setugidsafety(td); 663 VOP_UNLOCK(imgp->vp, 0); 664 error = fdcheckstd(td); 665 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 666 if (error != 0) 667 goto done1; 668 PROC_LOCK(p); 669 /* 670 * Set the new credentials. 671 */ 672 crcopy(newcred, oldcred); 673 if (attr.va_mode & VSUID) 674 change_euid(newcred, euip); 675 if (attr.va_mode & VSGID) 676 change_egid(newcred, attr.va_gid); 677 #ifdef MAC 678 if (will_transition) { 679 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 680 interplabel, imgp); 681 } 682 #endif 683 /* 684 * Implement correct POSIX saved-id behavior. 685 * 686 * XXXMAC: Note that the current logic will save the 687 * uid and gid if a MAC domain transition occurs, even 688 * though maybe it shouldn't. 689 */ 690 change_svuid(newcred, newcred->cr_uid); 691 change_svgid(newcred, newcred->cr_gid); 692 p->p_ucred = newcred; 693 newcred = NULL; 694 } else { 695 if (oldcred->cr_uid == oldcred->cr_ruid && 696 oldcred->cr_gid == oldcred->cr_rgid) 697 p->p_flag &= ~P_SUGID; 698 /* 699 * Implement correct POSIX saved-id behavior. 700 * 701 * XXX: It's not clear that the existing behavior is 702 * POSIX-compliant. A number of sources indicate that the 703 * saved uid/gid should only be updated if the new ruid is 704 * not equal to the old ruid, or the new euid is not equal 705 * to the old euid and the new euid is not equal to the old 706 * ruid. The FreeBSD code always updates the saved uid/gid. 707 * Also, this code uses the new (replaced) euid and egid as 708 * the source, which may or may not be the right ones to use. 709 */ 710 if (oldcred->cr_svuid != oldcred->cr_uid || 711 oldcred->cr_svgid != oldcred->cr_gid) { 712 crcopy(newcred, oldcred); 713 change_svuid(newcred, newcred->cr_uid); 714 change_svgid(newcred, newcred->cr_gid); 715 p->p_ucred = newcred; 716 newcred = NULL; 717 } 718 } 719 720 /* 721 * Store the vp for use in procfs. This vnode was referenced prior 722 * to locking the proc lock. 723 */ 724 textvp = p->p_textvp; 725 p->p_textvp = binvp; 726 727 #ifdef KDTRACE_HOOKS 728 /* 729 * Tell the DTrace fasttrap provider about the exec if it 730 * has declared an interest. 731 */ 732 if (dtrace_fasttrap_exec) 733 dtrace_fasttrap_exec(p); 734 #endif 735 736 /* 737 * Notify others that we exec'd, and clear the P_INEXEC flag 738 * as we're now a bona fide freshly-execed process. 739 */ 740 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 741 p->p_flag &= ~P_INEXEC; 742 743 /* 744 * If tracing the process, trap to debugger so breakpoints 745 * can be set before the program executes. 746 * Use tdsignal to deliver signal to current thread, use 747 * psignal may cause the signal to be delivered to wrong thread 748 * because that thread will exit, remember we are going to enter 749 * single thread mode. 750 */ 751 if (p->p_flag & P_TRACED) 752 tdsignal(p, td, SIGTRAP, NULL); 753 754 /* clear "fork but no exec" flag, as we _are_ execing */ 755 p->p_acflag &= ~AFORK; 756 757 /* 758 * Free any previous argument cache and replace it with 759 * the new argument cache, if any. 760 */ 761 oldargs = p->p_args; 762 p->p_args = newargs; 763 newargs = NULL; 764 765 #ifdef HWPMC_HOOKS 766 /* 767 * Check if system-wide sampling is in effect or if the 768 * current process is using PMCs. If so, do exec() time 769 * processing. This processing needs to happen AFTER the 770 * P_INEXEC flag is cleared. 771 * 772 * The proc lock needs to be released before taking the PMC 773 * SX. 774 */ 775 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 776 PROC_UNLOCK(p); 777 pe.pm_credentialschanged = credential_changing; 778 pe.pm_entryaddr = imgp->entry_addr; 779 780 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 781 } else 782 PROC_UNLOCK(p); 783 #else /* !HWPMC_HOOKS */ 784 PROC_UNLOCK(p); 785 #endif 786 787 /* Set values passed into the program in registers. */ 788 if (p->p_sysent->sv_setregs) 789 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr, 790 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 791 else 792 exec_setregs(td, imgp->entry_addr, 793 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 794 795 vfs_mark_atime(imgp->vp, td); 796 797 done1: 798 799 /* 800 * Free any resources malloc'd earlier that we didn't use. 801 */ 802 uifree(euip); 803 if (newcred == NULL) 804 crfree(oldcred); 805 else 806 crfree(newcred); 807 VOP_UNLOCK(imgp->vp, 0); 808 809 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0); 810 811 /* 812 * Handle deferred decrement of ref counts. 813 */ 814 if (textvp != NULL) { 815 int tvfslocked; 816 817 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount); 818 vrele(textvp); 819 VFS_UNLOCK_GIANT(tvfslocked); 820 } 821 if (binvp && error != 0) 822 vrele(binvp); 823 #ifdef KTRACE 824 if (tracevp != NULL) { 825 int tvfslocked; 826 827 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 828 vrele(tracevp); 829 VFS_UNLOCK_GIANT(tvfslocked); 830 } 831 if (tracecred != NULL) 832 crfree(tracecred); 833 #endif 834 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 835 pargs_drop(oldargs); 836 pargs_drop(newargs); 837 if (oldsigacts != NULL) 838 sigacts_free(oldsigacts); 839 840 exec_fail_dealloc: 841 842 /* 843 * free various allocated resources 844 */ 845 if (imgp->firstpage != NULL) 846 exec_unmap_first_page(imgp); 847 848 if (imgp->vp != NULL) { 849 if (args->fname) 850 NDFREE(ndp, NDF_ONLY_PNBUF); 851 if (imgp->opened) 852 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 853 vput(imgp->vp); 854 } 855 856 if (imgp->object != NULL) 857 vm_object_deallocate(imgp->object); 858 859 if (error == 0) { 860 /* 861 * Stop the process here if its stop event mask has 862 * the S_EXEC bit set. 863 */ 864 STOPEVENT(p, S_EXEC, 0); 865 goto done2; 866 } 867 868 exec_fail: 869 /* we're done here, clear P_INEXEC */ 870 PROC_LOCK(p); 871 p->p_flag &= ~P_INEXEC; 872 PROC_UNLOCK(p); 873 874 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0); 875 876 done2: 877 #ifdef MAC 878 mac_execve_exit(imgp); 879 if (interplabel != NULL) 880 mac_vnode_label_free(interplabel); 881 #endif 882 VFS_UNLOCK_GIANT(vfslocked); 883 exec_free_args(args); 884 885 if (error && imgp->vmspace_destroyed) { 886 /* sorry, no more process anymore. exit gracefully */ 887 exit1(td, W_EXITCODE(0, SIGABRT)); 888 /* NOT REACHED */ 889 } 890 return (error); 891 } 892 893 int 894 exec_map_first_page(imgp) 895 struct image_params *imgp; 896 { 897 int rv, i; 898 int initial_pagein; 899 vm_page_t ma[VM_INITIAL_PAGEIN]; 900 vm_object_t object; 901 902 if (imgp->firstpage != NULL) 903 exec_unmap_first_page(imgp); 904 905 object = imgp->vp->v_object; 906 if (object == NULL) 907 return (EACCES); 908 VM_OBJECT_LOCK(object); 909 #if VM_NRESERVLEVEL > 0 910 if ((object->flags & OBJ_COLORED) == 0) { 911 object->flags |= OBJ_COLORED; 912 object->pg_color = 0; 913 } 914 #endif 915 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 916 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 917 initial_pagein = VM_INITIAL_PAGEIN; 918 if (initial_pagein > object->size) 919 initial_pagein = object->size; 920 for (i = 1; i < initial_pagein; i++) { 921 if ((ma[i] = vm_page_lookup(object, i)) != NULL) { 922 if (ma[i]->valid) 923 break; 924 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy) 925 break; 926 vm_page_busy(ma[i]); 927 } else { 928 ma[i] = vm_page_alloc(object, i, 929 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 930 if (ma[i] == NULL) 931 break; 932 } 933 } 934 initial_pagein = i; 935 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 936 ma[0] = vm_page_lookup(object, 0); 937 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || 938 (ma[0]->valid == 0)) { 939 if (ma[0]) { 940 vm_page_lock_queues(); 941 vm_page_free(ma[0]); 942 vm_page_unlock_queues(); 943 } 944 VM_OBJECT_UNLOCK(object); 945 return (EIO); 946 } 947 } 948 vm_page_lock_queues(); 949 vm_page_hold(ma[0]); 950 vm_page_unlock_queues(); 951 vm_page_wakeup(ma[0]); 952 VM_OBJECT_UNLOCK(object); 953 954 imgp->firstpage = sf_buf_alloc(ma[0], 0); 955 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 956 957 return (0); 958 } 959 960 void 961 exec_unmap_first_page(imgp) 962 struct image_params *imgp; 963 { 964 vm_page_t m; 965 966 if (imgp->firstpage != NULL) { 967 m = sf_buf_page(imgp->firstpage); 968 sf_buf_free(imgp->firstpage); 969 imgp->firstpage = NULL; 970 vm_page_lock_queues(); 971 vm_page_unhold(m); 972 vm_page_unlock_queues(); 973 } 974 } 975 976 /* 977 * Destroy old address space, and allocate a new stack 978 * The new stack is only SGROWSIZ large because it is grown 979 * automatically in trap.c. 980 */ 981 int 982 exec_new_vmspace(imgp, sv) 983 struct image_params *imgp; 984 struct sysentvec *sv; 985 { 986 int error; 987 struct proc *p = imgp->proc; 988 struct vmspace *vmspace = p->p_vmspace; 989 vm_offset_t stack_addr; 990 vm_map_t map; 991 u_long ssiz; 992 993 imgp->vmspace_destroyed = 1; 994 imgp->sysent = sv; 995 996 /* May be called with Giant held */ 997 EVENTHANDLER_INVOKE(process_exec, p, imgp); 998 999 /* 1000 * Blow away entire process VM, if address space not shared, 1001 * otherwise, create a new VM space so that other threads are 1002 * not disrupted 1003 */ 1004 map = &vmspace->vm_map; 1005 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser && 1006 vm_map_max(map) == sv->sv_maxuser) { 1007 shmexit(vmspace); 1008 pmap_remove_pages(vmspace_pmap(vmspace)); 1009 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1010 } else { 1011 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser); 1012 if (error) 1013 return (error); 1014 vmspace = p->p_vmspace; 1015 map = &vmspace->vm_map; 1016 } 1017 1018 /* Allocate a new stack */ 1019 if (sv->sv_maxssiz != NULL) 1020 ssiz = *sv->sv_maxssiz; 1021 else 1022 ssiz = maxssiz; 1023 stack_addr = sv->sv_usrstack - ssiz; 1024 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1025 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1026 if (error) 1027 return (error); 1028 1029 #ifdef __ia64__ 1030 /* Allocate a new register stack */ 1031 stack_addr = IA64_BACKINGSTORE; 1032 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1033 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1034 if (error) 1035 return (error); 1036 #endif 1037 1038 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1039 * VM_STACK case, but they are still used to monitor the size of the 1040 * process stack so we can check the stack rlimit. 1041 */ 1042 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1043 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; 1044 1045 return (0); 1046 } 1047 1048 /* 1049 * Copy out argument and environment strings from the old process address 1050 * space into the temporary string buffer. 1051 */ 1052 int 1053 exec_copyin_args(struct image_args *args, char *fname, 1054 enum uio_seg segflg, char **argv, char **envv) 1055 { 1056 char *argp, *envp; 1057 int error; 1058 size_t length; 1059 1060 error = 0; 1061 1062 bzero(args, sizeof(*args)); 1063 if (argv == NULL) 1064 return (EFAULT); 1065 /* 1066 * Allocate temporary demand zeroed space for argument and 1067 * environment strings: 1068 * 1069 * o ARG_MAX for argument and environment; 1070 * o MAXSHELLCMDLEN for the name of interpreters. 1071 */ 1072 args->buf = (char *) kmem_alloc_wait(exec_map, 1073 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1074 if (args->buf == NULL) 1075 return (ENOMEM); 1076 args->begin_argv = args->buf; 1077 args->endp = args->begin_argv; 1078 args->stringspace = ARG_MAX; 1079 /* 1080 * Copy the file name. 1081 */ 1082 if (fname != NULL) { 1083 args->fname = args->buf + ARG_MAX; 1084 error = (segflg == UIO_SYSSPACE) ? 1085 copystr(fname, args->fname, PATH_MAX, &length) : 1086 copyinstr(fname, args->fname, PATH_MAX, &length); 1087 if (error != 0) 1088 goto err_exit; 1089 } else 1090 args->fname = NULL; 1091 1092 /* 1093 * extract arguments first 1094 */ 1095 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) { 1096 if (argp == (caddr_t) -1) { 1097 error = EFAULT; 1098 goto err_exit; 1099 } 1100 if ((error = copyinstr(argp, args->endp, 1101 args->stringspace, &length))) { 1102 if (error == ENAMETOOLONG) 1103 error = E2BIG; 1104 goto err_exit; 1105 } 1106 args->stringspace -= length; 1107 args->endp += length; 1108 args->argc++; 1109 } 1110 1111 args->begin_envv = args->endp; 1112 1113 /* 1114 * extract environment strings 1115 */ 1116 if (envv) { 1117 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 1118 if (envp == (caddr_t)-1) { 1119 error = EFAULT; 1120 goto err_exit; 1121 } 1122 if ((error = copyinstr(envp, args->endp, 1123 args->stringspace, &length))) { 1124 if (error == ENAMETOOLONG) 1125 error = E2BIG; 1126 goto err_exit; 1127 } 1128 args->stringspace -= length; 1129 args->endp += length; 1130 args->envc++; 1131 } 1132 } 1133 1134 return (0); 1135 1136 err_exit: 1137 exec_free_args(args); 1138 return (error); 1139 } 1140 1141 static void 1142 exec_free_args(struct image_args *args) 1143 { 1144 1145 if (args->buf) { 1146 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, 1147 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1148 args->buf = NULL; 1149 } 1150 } 1151 1152 /* 1153 * Copy strings out to the new process address space, constructing new arg 1154 * and env vector tables. Return a pointer to the base so that it can be used 1155 * as the initial stack pointer. 1156 */ 1157 register_t * 1158 exec_copyout_strings(imgp) 1159 struct image_params *imgp; 1160 { 1161 int argc, envc; 1162 char **vectp; 1163 char *stringp, *destp; 1164 register_t *stack_base; 1165 struct ps_strings *arginfo; 1166 struct proc *p; 1167 int szsigcode; 1168 1169 /* 1170 * Calculate string base and vector table pointers. 1171 * Also deal with signal trampoline code for this exec type. 1172 */ 1173 p = imgp->proc; 1174 szsigcode = 0; 1175 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1176 if (p->p_sysent->sv_szsigcode != NULL) 1177 szsigcode = *(p->p_sysent->sv_szsigcode); 1178 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - 1179 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); 1180 1181 /* 1182 * install sigcode 1183 */ 1184 if (szsigcode) 1185 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - 1186 szsigcode), szsigcode); 1187 1188 /* 1189 * If we have a valid auxargs ptr, prepare some room 1190 * on the stack. 1191 */ 1192 if (imgp->auxargs) { 1193 /* 1194 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1195 * lower compatibility. 1196 */ 1197 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1198 (AT_COUNT * 2); 1199 /* 1200 * The '+ 2' is for the null pointers at the end of each of 1201 * the arg and env vector sets,and imgp->auxarg_size is room 1202 * for argument of Runtime loader. 1203 */ 1204 vectp = (char **)(destp - (imgp->args->argc + 1205 imgp->args->envc + 2 + imgp->auxarg_size) * 1206 sizeof(char *)); 1207 1208 } else { 1209 /* 1210 * The '+ 2' is for the null pointers at the end of each of 1211 * the arg and env vector sets 1212 */ 1213 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * 1214 sizeof(char *)); 1215 } 1216 1217 /* 1218 * vectp also becomes our initial stack base 1219 */ 1220 stack_base = (register_t *)vectp; 1221 1222 stringp = imgp->args->begin_argv; 1223 argc = imgp->args->argc; 1224 envc = imgp->args->envc; 1225 1226 /* 1227 * Copy out strings - arguments and environment. 1228 */ 1229 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 1230 1231 /* 1232 * Fill in "ps_strings" struct for ps, w, etc. 1233 */ 1234 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1235 suword(&arginfo->ps_nargvstr, argc); 1236 1237 /* 1238 * Fill in argument portion of vector table. 1239 */ 1240 for (; argc > 0; --argc) { 1241 suword(vectp++, (long)(intptr_t)destp); 1242 while (*stringp++ != 0) 1243 destp++; 1244 destp++; 1245 } 1246 1247 /* a null vector table pointer separates the argp's from the envp's */ 1248 suword(vectp++, 0); 1249 1250 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1251 suword(&arginfo->ps_nenvstr, envc); 1252 1253 /* 1254 * Fill in environment portion of vector table. 1255 */ 1256 for (; envc > 0; --envc) { 1257 suword(vectp++, (long)(intptr_t)destp); 1258 while (*stringp++ != 0) 1259 destp++; 1260 destp++; 1261 } 1262 1263 /* end of vector table is a null pointer */ 1264 suword(vectp, 0); 1265 1266 return (stack_base); 1267 } 1268 1269 /* 1270 * Check permissions of file to execute. 1271 * Called with imgp->vp locked. 1272 * Return 0 for success or error code on failure. 1273 */ 1274 int 1275 exec_check_permissions(imgp) 1276 struct image_params *imgp; 1277 { 1278 struct vnode *vp = imgp->vp; 1279 struct vattr *attr = imgp->attr; 1280 struct thread *td; 1281 int error; 1282 1283 td = curthread; 1284 1285 /* Get file attributes */ 1286 error = VOP_GETATTR(vp, attr, td->td_ucred, td); 1287 if (error) 1288 return (error); 1289 1290 #ifdef MAC 1291 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1292 if (error) 1293 return (error); 1294 #endif 1295 1296 /* 1297 * 1) Check if file execution is disabled for the filesystem that this 1298 * file resides on. 1299 * 2) Insure that at least one execute bit is on - otherwise root 1300 * will always succeed, and we don't want to happen unless the 1301 * file really is executable. 1302 * 3) Insure that the file is a regular file. 1303 */ 1304 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1305 ((attr->va_mode & 0111) == 0) || 1306 (attr->va_type != VREG)) 1307 return (EACCES); 1308 1309 /* 1310 * Zero length files can't be exec'd 1311 */ 1312 if (attr->va_size == 0) 1313 return (ENOEXEC); 1314 1315 /* 1316 * Check for execute permission to file based on current credentials. 1317 */ 1318 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1319 if (error) 1320 return (error); 1321 1322 /* 1323 * Check number of open-for-writes on the file and deny execution 1324 * if there are any. 1325 */ 1326 if (vp->v_writecount) 1327 return (ETXTBSY); 1328 1329 /* 1330 * Call filesystem specific open routine (which does nothing in the 1331 * general case). 1332 */ 1333 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1334 if (error == 0) 1335 imgp->opened = 1; 1336 return (error); 1337 } 1338 1339 /* 1340 * Exec handler registration 1341 */ 1342 int 1343 exec_register(execsw_arg) 1344 const struct execsw *execsw_arg; 1345 { 1346 const struct execsw **es, **xs, **newexecsw; 1347 int count = 2; /* New slot and trailing NULL */ 1348 1349 if (execsw) 1350 for (es = execsw; *es; es++) 1351 count++; 1352 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1353 if (newexecsw == NULL) 1354 return (ENOMEM); 1355 xs = newexecsw; 1356 if (execsw) 1357 for (es = execsw; *es; es++) 1358 *xs++ = *es; 1359 *xs++ = execsw_arg; 1360 *xs = NULL; 1361 if (execsw) 1362 free(execsw, M_TEMP); 1363 execsw = newexecsw; 1364 return (0); 1365 } 1366 1367 int 1368 exec_unregister(execsw_arg) 1369 const struct execsw *execsw_arg; 1370 { 1371 const struct execsw **es, **xs, **newexecsw; 1372 int count = 1; 1373 1374 if (execsw == NULL) 1375 panic("unregister with no handlers left?\n"); 1376 1377 for (es = execsw; *es; es++) { 1378 if (*es == execsw_arg) 1379 break; 1380 } 1381 if (*es == NULL) 1382 return (ENOENT); 1383 for (es = execsw; *es; es++) 1384 if (*es != execsw_arg) 1385 count++; 1386 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1387 if (newexecsw == NULL) 1388 return (ENOMEM); 1389 xs = newexecsw; 1390 for (es = execsw; *es; es++) 1391 if (*es != execsw_arg) 1392 *xs++ = *es; 1393 *xs = NULL; 1394 if (execsw) 1395 free(execsw, M_TEMP); 1396 execsw = newexecsw; 1397 return (0); 1398 } 1399