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 *interpvplabel = 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 mac_execve_interpreter_enter(binvp, &interpvplabel); 498 #endif 499 if (imgp->opened) { 500 VOP_CLOSE(binvp, FREAD, td->td_ucred, td); 501 imgp->opened = 0; 502 } 503 vput(binvp); 504 vm_object_deallocate(imgp->object); 505 imgp->object = NULL; 506 VFS_UNLOCK_GIANT(vfslocked); 507 vfslocked = 0; 508 /* set new name to that of the interpreter */ 509 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE, 510 UIO_SYSSPACE, imgp->interpreter_name, td); 511 args->fname = imgp->interpreter_name; 512 goto interpret; 513 } 514 515 /* 516 * NB: We unlock the vnode here because it is believed that none 517 * of the sv_copyout_strings/sv_fixup operations require the vnode. 518 */ 519 VOP_UNLOCK(imgp->vp, 0); 520 /* 521 * Copy out strings (args and env) and initialize stack base 522 */ 523 if (p->p_sysent->sv_copyout_strings) 524 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 525 else 526 stack_base = exec_copyout_strings(imgp); 527 528 /* 529 * If custom stack fixup routine present for this process 530 * let it do the stack setup. 531 * Else stuff argument count as first item on stack 532 */ 533 if (p->p_sysent->sv_fixup != NULL) 534 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 535 else 536 suword(--stack_base, imgp->args->argc); 537 538 /* 539 * For security and other reasons, the file descriptor table cannot 540 * be shared after an exec. 541 */ 542 fdunshare(p, td); 543 544 /* 545 * Malloc things before we need locks. 546 */ 547 newcred = crget(); 548 euip = uifind(attr.va_uid); 549 i = imgp->args->begin_envv - imgp->args->begin_argv; 550 /* Cache arguments if they fit inside our allowance */ 551 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 552 newargs = pargs_alloc(i); 553 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 554 } 555 556 /* close files on exec */ 557 fdcloseexec(td); 558 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 559 560 /* Get a reference to the vnode prior to locking the proc */ 561 VREF(binvp); 562 563 /* 564 * For security and other reasons, signal handlers cannot 565 * be shared after an exec. The new process gets a copy of the old 566 * handlers. In execsigs(), the new process will have its signals 567 * reset. 568 */ 569 PROC_LOCK(p); 570 if (sigacts_shared(p->p_sigacts)) { 571 oldsigacts = p->p_sigacts; 572 PROC_UNLOCK(p); 573 newsigacts = sigacts_alloc(); 574 sigacts_copy(newsigacts, oldsigacts); 575 PROC_LOCK(p); 576 p->p_sigacts = newsigacts; 577 } else 578 oldsigacts = NULL; 579 580 /* Stop profiling */ 581 stopprofclock(p); 582 583 /* reset caught signals */ 584 execsigs(p); 585 586 /* name this process - nameiexec(p, ndp) */ 587 if (args->fname) { 588 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN); 589 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len); 590 } else { 591 len = MAXCOMLEN; 592 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0) 593 len = MAXCOMLEN; 594 else { 595 len = sizeof(fexecv_proc_title); 596 bcopy(fexecv_proc_title, p->p_comm, len); 597 } 598 } 599 p->p_comm[len] = 0; 600 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 601 602 /* 603 * mark as execed, wakeup the process that vforked (if any) and tell 604 * it that it now has its own resources back 605 */ 606 p->p_flag |= P_EXEC; 607 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 608 p->p_flag &= ~P_PPWAIT; 609 wakeup(p->p_pptr); 610 } 611 612 /* 613 * Implement image setuid/setgid. 614 * 615 * Don't honor setuid/setgid if the filesystem prohibits it or if 616 * the process is being traced. 617 * 618 * XXXMAC: For the time being, use NOSUID to also prohibit 619 * transitions on the file system. 620 */ 621 oldcred = p->p_ucred; 622 credential_changing = 0; 623 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid != 624 attr.va_uid; 625 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid != 626 attr.va_gid; 627 #ifdef MAC 628 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 629 interpvplabel, imgp); 630 credential_changing |= will_transition; 631 #endif 632 633 if (credential_changing && 634 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 635 (p->p_flag & P_TRACED) == 0) { 636 /* 637 * Turn off syscall tracing for set-id programs, except for 638 * root. Record any set-id flags first to make sure that 639 * we do not regain any tracing during a possible block. 640 */ 641 setsugid(p); 642 643 #ifdef KTRACE 644 if (p->p_tracevp != NULL && 645 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) { 646 mtx_lock(&ktrace_mtx); 647 p->p_traceflag = 0; 648 tracevp = p->p_tracevp; 649 p->p_tracevp = NULL; 650 tracecred = p->p_tracecred; 651 p->p_tracecred = NULL; 652 mtx_unlock(&ktrace_mtx); 653 } 654 #endif 655 /* 656 * Close any file descriptors 0..2 that reference procfs, 657 * then make sure file descriptors 0..2 are in use. 658 * 659 * setugidsafety() may call closef() and then pfind() 660 * which may grab the process lock. 661 * fdcheckstd() may call falloc() which may block to 662 * allocate memory, so temporarily drop the process lock. 663 */ 664 PROC_UNLOCK(p); 665 setugidsafety(td); 666 VOP_UNLOCK(imgp->vp, 0); 667 error = fdcheckstd(td); 668 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 669 if (error != 0) 670 goto done1; 671 PROC_LOCK(p); 672 /* 673 * Set the new credentials. 674 */ 675 crcopy(newcred, oldcred); 676 if (attr.va_mode & VSUID) 677 change_euid(newcred, euip); 678 if (attr.va_mode & VSGID) 679 change_egid(newcred, attr.va_gid); 680 #ifdef MAC 681 if (will_transition) { 682 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 683 interpvplabel, imgp); 684 } 685 #endif 686 /* 687 * Implement correct POSIX saved-id behavior. 688 * 689 * XXXMAC: Note that the current logic will save the 690 * uid and gid if a MAC domain transition occurs, even 691 * though maybe it shouldn't. 692 */ 693 change_svuid(newcred, newcred->cr_uid); 694 change_svgid(newcred, newcred->cr_gid); 695 p->p_ucred = newcred; 696 newcred = NULL; 697 } else { 698 if (oldcred->cr_uid == oldcred->cr_ruid && 699 oldcred->cr_gid == oldcred->cr_rgid) 700 p->p_flag &= ~P_SUGID; 701 /* 702 * Implement correct POSIX saved-id behavior. 703 * 704 * XXX: It's not clear that the existing behavior is 705 * POSIX-compliant. A number of sources indicate that the 706 * saved uid/gid should only be updated if the new ruid is 707 * not equal to the old ruid, or the new euid is not equal 708 * to the old euid and the new euid is not equal to the old 709 * ruid. The FreeBSD code always updates the saved uid/gid. 710 * Also, this code uses the new (replaced) euid and egid as 711 * the source, which may or may not be the right ones to use. 712 */ 713 if (oldcred->cr_svuid != oldcred->cr_uid || 714 oldcred->cr_svgid != oldcred->cr_gid) { 715 crcopy(newcred, oldcred); 716 change_svuid(newcred, newcred->cr_uid); 717 change_svgid(newcred, newcred->cr_gid); 718 p->p_ucred = newcred; 719 newcred = NULL; 720 } 721 } 722 723 /* 724 * Store the vp for use in procfs. This vnode was referenced prior 725 * to locking the proc lock. 726 */ 727 textvp = p->p_textvp; 728 p->p_textvp = binvp; 729 730 #ifdef KDTRACE_HOOKS 731 /* 732 * Tell the DTrace fasttrap provider about the exec if it 733 * has declared an interest. 734 */ 735 if (dtrace_fasttrap_exec) 736 dtrace_fasttrap_exec(p); 737 #endif 738 739 /* 740 * Notify others that we exec'd, and clear the P_INEXEC flag 741 * as we're now a bona fide freshly-execed process. 742 */ 743 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 744 p->p_flag &= ~P_INEXEC; 745 746 /* 747 * If tracing the process, trap to debugger so breakpoints 748 * can be set before the program executes. 749 * Use tdsignal to deliver signal to current thread, use 750 * psignal may cause the signal to be delivered to wrong thread 751 * because that thread will exit, remember we are going to enter 752 * single thread mode. 753 */ 754 if (p->p_flag & P_TRACED) 755 tdsignal(p, td, SIGTRAP, NULL); 756 757 /* clear "fork but no exec" flag, as we _are_ execing */ 758 p->p_acflag &= ~AFORK; 759 760 /* 761 * Free any previous argument cache and replace it with 762 * the new argument cache, if any. 763 */ 764 oldargs = p->p_args; 765 p->p_args = newargs; 766 newargs = NULL; 767 768 #ifdef HWPMC_HOOKS 769 /* 770 * Check if system-wide sampling is in effect or if the 771 * current process is using PMCs. If so, do exec() time 772 * processing. This processing needs to happen AFTER the 773 * P_INEXEC flag is cleared. 774 * 775 * The proc lock needs to be released before taking the PMC 776 * SX. 777 */ 778 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 779 PROC_UNLOCK(p); 780 pe.pm_credentialschanged = credential_changing; 781 pe.pm_entryaddr = imgp->entry_addr; 782 783 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 784 } else 785 PROC_UNLOCK(p); 786 #else /* !HWPMC_HOOKS */ 787 PROC_UNLOCK(p); 788 #endif 789 790 /* Set values passed into the program in registers. */ 791 if (p->p_sysent->sv_setregs) 792 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr, 793 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 794 else 795 exec_setregs(td, imgp->entry_addr, 796 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 797 798 vfs_mark_atime(imgp->vp, td); 799 800 done1: 801 802 /* 803 * Free any resources malloc'd earlier that we didn't use. 804 */ 805 uifree(euip); 806 if (newcred == NULL) 807 crfree(oldcred); 808 else 809 crfree(newcred); 810 VOP_UNLOCK(imgp->vp, 0); 811 812 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0); 813 814 /* 815 * Handle deferred decrement of ref counts. 816 */ 817 if (textvp != NULL) { 818 int tvfslocked; 819 820 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount); 821 vrele(textvp); 822 VFS_UNLOCK_GIANT(tvfslocked); 823 } 824 if (binvp && error != 0) 825 vrele(binvp); 826 #ifdef KTRACE 827 if (tracevp != NULL) { 828 int tvfslocked; 829 830 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 831 vrele(tracevp); 832 VFS_UNLOCK_GIANT(tvfslocked); 833 } 834 if (tracecred != NULL) 835 crfree(tracecred); 836 #endif 837 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 838 pargs_drop(oldargs); 839 pargs_drop(newargs); 840 if (oldsigacts != NULL) 841 sigacts_free(oldsigacts); 842 843 exec_fail_dealloc: 844 845 /* 846 * free various allocated resources 847 */ 848 if (imgp->firstpage != NULL) 849 exec_unmap_first_page(imgp); 850 851 if (imgp->vp != NULL) { 852 if (args->fname) 853 NDFREE(ndp, NDF_ONLY_PNBUF); 854 if (imgp->opened) 855 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 856 vput(imgp->vp); 857 } 858 859 if (imgp->object != NULL) 860 vm_object_deallocate(imgp->object); 861 862 if (error == 0) { 863 /* 864 * Stop the process here if its stop event mask has 865 * the S_EXEC bit set. 866 */ 867 STOPEVENT(p, S_EXEC, 0); 868 goto done2; 869 } 870 871 exec_fail: 872 /* we're done here, clear P_INEXEC */ 873 PROC_LOCK(p); 874 p->p_flag &= ~P_INEXEC; 875 PROC_UNLOCK(p); 876 877 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0); 878 879 done2: 880 #ifdef MAC 881 mac_execve_exit(imgp); 882 mac_execve_interpreter_exit(interpvplabel); 883 #endif 884 VFS_UNLOCK_GIANT(vfslocked); 885 exec_free_args(args); 886 887 if (error && imgp->vmspace_destroyed) { 888 /* sorry, no more process anymore. exit gracefully */ 889 exit1(td, W_EXITCODE(0, SIGABRT)); 890 /* NOT REACHED */ 891 } 892 return (error); 893 } 894 895 int 896 exec_map_first_page(imgp) 897 struct image_params *imgp; 898 { 899 int rv, i; 900 int initial_pagein; 901 vm_page_t ma[VM_INITIAL_PAGEIN]; 902 vm_object_t object; 903 904 if (imgp->firstpage != NULL) 905 exec_unmap_first_page(imgp); 906 907 object = imgp->vp->v_object; 908 if (object == NULL) 909 return (EACCES); 910 VM_OBJECT_LOCK(object); 911 #if VM_NRESERVLEVEL > 0 912 if ((object->flags & OBJ_COLORED) == 0) { 913 object->flags |= OBJ_COLORED; 914 object->pg_color = 0; 915 } 916 #endif 917 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 918 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 919 initial_pagein = VM_INITIAL_PAGEIN; 920 if (initial_pagein > object->size) 921 initial_pagein = object->size; 922 for (i = 1; i < initial_pagein; i++) { 923 if ((ma[i] = vm_page_lookup(object, i)) != NULL) { 924 if (ma[i]->valid) 925 break; 926 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy) 927 break; 928 vm_page_busy(ma[i]); 929 } else { 930 ma[i] = vm_page_alloc(object, i, 931 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 932 if (ma[i] == NULL) 933 break; 934 } 935 } 936 initial_pagein = i; 937 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 938 ma[0] = vm_page_lookup(object, 0); 939 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || 940 (ma[0]->valid == 0)) { 941 if (ma[0]) { 942 vm_page_lock_queues(); 943 vm_page_free(ma[0]); 944 vm_page_unlock_queues(); 945 } 946 VM_OBJECT_UNLOCK(object); 947 return (EIO); 948 } 949 } 950 vm_page_lock_queues(); 951 vm_page_hold(ma[0]); 952 vm_page_unlock_queues(); 953 vm_page_wakeup(ma[0]); 954 VM_OBJECT_UNLOCK(object); 955 956 imgp->firstpage = sf_buf_alloc(ma[0], 0); 957 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 958 959 return (0); 960 } 961 962 void 963 exec_unmap_first_page(imgp) 964 struct image_params *imgp; 965 { 966 vm_page_t m; 967 968 if (imgp->firstpage != NULL) { 969 m = sf_buf_page(imgp->firstpage); 970 sf_buf_free(imgp->firstpage); 971 imgp->firstpage = NULL; 972 vm_page_lock_queues(); 973 vm_page_unhold(m); 974 vm_page_unlock_queues(); 975 } 976 } 977 978 /* 979 * Destroy old address space, and allocate a new stack 980 * The new stack is only SGROWSIZ large because it is grown 981 * automatically in trap.c. 982 */ 983 int 984 exec_new_vmspace(imgp, sv) 985 struct image_params *imgp; 986 struct sysentvec *sv; 987 { 988 int error; 989 struct proc *p = imgp->proc; 990 struct vmspace *vmspace = p->p_vmspace; 991 vm_offset_t stack_addr; 992 vm_map_t map; 993 u_long ssiz; 994 995 imgp->vmspace_destroyed = 1; 996 imgp->sysent = sv; 997 998 /* May be called with Giant held */ 999 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1000 1001 /* 1002 * Blow away entire process VM, if address space not shared, 1003 * otherwise, create a new VM space so that other threads are 1004 * not disrupted 1005 */ 1006 map = &vmspace->vm_map; 1007 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser && 1008 vm_map_max(map) == sv->sv_maxuser) { 1009 shmexit(vmspace); 1010 pmap_remove_pages(vmspace_pmap(vmspace)); 1011 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1012 } else { 1013 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser); 1014 if (error) 1015 return (error); 1016 vmspace = p->p_vmspace; 1017 map = &vmspace->vm_map; 1018 } 1019 1020 /* Allocate a new stack */ 1021 if (sv->sv_maxssiz != NULL) 1022 ssiz = *sv->sv_maxssiz; 1023 else 1024 ssiz = maxssiz; 1025 stack_addr = sv->sv_usrstack - ssiz; 1026 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1027 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1028 if (error) 1029 return (error); 1030 1031 #ifdef __ia64__ 1032 /* Allocate a new register stack */ 1033 stack_addr = IA64_BACKINGSTORE; 1034 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1035 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1036 if (error) 1037 return (error); 1038 #endif 1039 1040 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1041 * VM_STACK case, but they are still used to monitor the size of the 1042 * process stack so we can check the stack rlimit. 1043 */ 1044 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1045 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; 1046 1047 return (0); 1048 } 1049 1050 /* 1051 * Copy out argument and environment strings from the old process address 1052 * space into the temporary string buffer. 1053 */ 1054 int 1055 exec_copyin_args(struct image_args *args, char *fname, 1056 enum uio_seg segflg, char **argv, char **envv) 1057 { 1058 char *argp, *envp; 1059 int error; 1060 size_t length; 1061 1062 error = 0; 1063 1064 bzero(args, sizeof(*args)); 1065 if (argv == NULL) 1066 return (EFAULT); 1067 /* 1068 * Allocate temporary demand zeroed space for argument and 1069 * environment strings: 1070 * 1071 * o ARG_MAX for argument and environment; 1072 * o MAXSHELLCMDLEN for the name of interpreters. 1073 */ 1074 args->buf = (char *) kmem_alloc_wait(exec_map, 1075 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1076 if (args->buf == NULL) 1077 return (ENOMEM); 1078 args->begin_argv = args->buf; 1079 args->endp = args->begin_argv; 1080 args->stringspace = ARG_MAX; 1081 /* 1082 * Copy the file name. 1083 */ 1084 if (fname != NULL) { 1085 args->fname = args->buf + ARG_MAX; 1086 error = (segflg == UIO_SYSSPACE) ? 1087 copystr(fname, args->fname, PATH_MAX, &length) : 1088 copyinstr(fname, args->fname, PATH_MAX, &length); 1089 if (error != 0) 1090 goto err_exit; 1091 } else 1092 args->fname = NULL; 1093 1094 /* 1095 * extract arguments first 1096 */ 1097 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) { 1098 if (argp == (caddr_t) -1) { 1099 error = EFAULT; 1100 goto err_exit; 1101 } 1102 if ((error = copyinstr(argp, args->endp, 1103 args->stringspace, &length))) { 1104 if (error == ENAMETOOLONG) 1105 error = E2BIG; 1106 goto err_exit; 1107 } 1108 args->stringspace -= length; 1109 args->endp += length; 1110 args->argc++; 1111 } 1112 1113 args->begin_envv = args->endp; 1114 1115 /* 1116 * extract environment strings 1117 */ 1118 if (envv) { 1119 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 1120 if (envp == (caddr_t)-1) { 1121 error = EFAULT; 1122 goto err_exit; 1123 } 1124 if ((error = copyinstr(envp, args->endp, 1125 args->stringspace, &length))) { 1126 if (error == ENAMETOOLONG) 1127 error = E2BIG; 1128 goto err_exit; 1129 } 1130 args->stringspace -= length; 1131 args->endp += length; 1132 args->envc++; 1133 } 1134 } 1135 1136 return (0); 1137 1138 err_exit: 1139 exec_free_args(args); 1140 return (error); 1141 } 1142 1143 static void 1144 exec_free_args(struct image_args *args) 1145 { 1146 1147 if (args->buf) { 1148 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, 1149 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1150 args->buf = NULL; 1151 } 1152 } 1153 1154 /* 1155 * Copy strings out to the new process address space, constructing new arg 1156 * and env vector tables. Return a pointer to the base so that it can be used 1157 * as the initial stack pointer. 1158 */ 1159 register_t * 1160 exec_copyout_strings(imgp) 1161 struct image_params *imgp; 1162 { 1163 int argc, envc; 1164 char **vectp; 1165 char *stringp, *destp; 1166 register_t *stack_base; 1167 struct ps_strings *arginfo; 1168 struct proc *p; 1169 int szsigcode; 1170 1171 /* 1172 * Calculate string base and vector table pointers. 1173 * Also deal with signal trampoline code for this exec type. 1174 */ 1175 p = imgp->proc; 1176 szsigcode = 0; 1177 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1178 if (p->p_sysent->sv_szsigcode != NULL) 1179 szsigcode = *(p->p_sysent->sv_szsigcode); 1180 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - 1181 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); 1182 1183 /* 1184 * install sigcode 1185 */ 1186 if (szsigcode) 1187 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - 1188 szsigcode), szsigcode); 1189 1190 /* 1191 * If we have a valid auxargs ptr, prepare some room 1192 * on the stack. 1193 */ 1194 if (imgp->auxargs) { 1195 /* 1196 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1197 * lower compatibility. 1198 */ 1199 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1200 (AT_COUNT * 2); 1201 /* 1202 * The '+ 2' is for the null pointers at the end of each of 1203 * the arg and env vector sets,and imgp->auxarg_size is room 1204 * for argument of Runtime loader. 1205 */ 1206 vectp = (char **)(destp - (imgp->args->argc + 1207 imgp->args->envc + 2 + imgp->auxarg_size) * 1208 sizeof(char *)); 1209 1210 } else { 1211 /* 1212 * The '+ 2' is for the null pointers at the end of each of 1213 * the arg and env vector sets 1214 */ 1215 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * 1216 sizeof(char *)); 1217 } 1218 1219 /* 1220 * vectp also becomes our initial stack base 1221 */ 1222 stack_base = (register_t *)vectp; 1223 1224 stringp = imgp->args->begin_argv; 1225 argc = imgp->args->argc; 1226 envc = imgp->args->envc; 1227 1228 /* 1229 * Copy out strings - arguments and environment. 1230 */ 1231 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 1232 1233 /* 1234 * Fill in "ps_strings" struct for ps, w, etc. 1235 */ 1236 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1237 suword(&arginfo->ps_nargvstr, argc); 1238 1239 /* 1240 * Fill in argument portion of vector table. 1241 */ 1242 for (; argc > 0; --argc) { 1243 suword(vectp++, (long)(intptr_t)destp); 1244 while (*stringp++ != 0) 1245 destp++; 1246 destp++; 1247 } 1248 1249 /* a null vector table pointer separates the argp's from the envp's */ 1250 suword(vectp++, 0); 1251 1252 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1253 suword(&arginfo->ps_nenvstr, envc); 1254 1255 /* 1256 * Fill in environment portion of vector table. 1257 */ 1258 for (; envc > 0; --envc) { 1259 suword(vectp++, (long)(intptr_t)destp); 1260 while (*stringp++ != 0) 1261 destp++; 1262 destp++; 1263 } 1264 1265 /* end of vector table is a null pointer */ 1266 suword(vectp, 0); 1267 1268 return (stack_base); 1269 } 1270 1271 /* 1272 * Check permissions of file to execute. 1273 * Called with imgp->vp locked. 1274 * Return 0 for success or error code on failure. 1275 */ 1276 int 1277 exec_check_permissions(imgp) 1278 struct image_params *imgp; 1279 { 1280 struct vnode *vp = imgp->vp; 1281 struct vattr *attr = imgp->attr; 1282 struct thread *td; 1283 int error; 1284 1285 td = curthread; 1286 1287 /* Get file attributes */ 1288 error = VOP_GETATTR(vp, attr, td->td_ucred, td); 1289 if (error) 1290 return (error); 1291 1292 #ifdef MAC 1293 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1294 if (error) 1295 return (error); 1296 #endif 1297 1298 /* 1299 * 1) Check if file execution is disabled for the filesystem that this 1300 * file resides on. 1301 * 2) Insure that at least one execute bit is on - otherwise root 1302 * will always succeed, and we don't want to happen unless the 1303 * file really is executable. 1304 * 3) Insure that the file is a regular file. 1305 */ 1306 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1307 ((attr->va_mode & 0111) == 0) || 1308 (attr->va_type != VREG)) 1309 return (EACCES); 1310 1311 /* 1312 * Zero length files can't be exec'd 1313 */ 1314 if (attr->va_size == 0) 1315 return (ENOEXEC); 1316 1317 /* 1318 * Check for execute permission to file based on current credentials. 1319 */ 1320 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1321 if (error) 1322 return (error); 1323 1324 /* 1325 * Check number of open-for-writes on the file and deny execution 1326 * if there are any. 1327 */ 1328 if (vp->v_writecount) 1329 return (ETXTBSY); 1330 1331 /* 1332 * Call filesystem specific open routine (which does nothing in the 1333 * general case). 1334 */ 1335 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1336 if (error == 0) 1337 imgp->opened = 1; 1338 return (error); 1339 } 1340 1341 /* 1342 * Exec handler registration 1343 */ 1344 int 1345 exec_register(execsw_arg) 1346 const struct execsw *execsw_arg; 1347 { 1348 const struct execsw **es, **xs, **newexecsw; 1349 int count = 2; /* New slot and trailing NULL */ 1350 1351 if (execsw) 1352 for (es = execsw; *es; es++) 1353 count++; 1354 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1355 if (newexecsw == NULL) 1356 return (ENOMEM); 1357 xs = newexecsw; 1358 if (execsw) 1359 for (es = execsw; *es; es++) 1360 *xs++ = *es; 1361 *xs++ = execsw_arg; 1362 *xs = NULL; 1363 if (execsw) 1364 free(execsw, M_TEMP); 1365 execsw = newexecsw; 1366 return (0); 1367 } 1368 1369 int 1370 exec_unregister(execsw_arg) 1371 const struct execsw *execsw_arg; 1372 { 1373 const struct execsw **es, **xs, **newexecsw; 1374 int count = 1; 1375 1376 if (execsw == NULL) 1377 panic("unregister with no handlers left?\n"); 1378 1379 for (es = execsw; *es; es++) { 1380 if (*es == execsw_arg) 1381 break; 1382 } 1383 if (*es == NULL) 1384 return (ENOENT); 1385 for (es = execsw; *es; es++) 1386 if (*es != execsw_arg) 1387 count++; 1388 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1389 if (newexecsw == NULL) 1390 return (ENOMEM); 1391 xs = newexecsw; 1392 for (es = execsw; *es; es++) 1393 if (*es != execsw_arg) 1394 *xs++ = *es; 1395 *xs = NULL; 1396 if (execsw) 1397 free(execsw, M_TEMP); 1398 execsw = newexecsw; 1399 return (0); 1400 } 1401