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->interpreter_name = args->buf + PATH_MAX + ARG_MAX; 373 imgp->auxargs = NULL; 374 imgp->vp = NULL; 375 imgp->object = NULL; 376 imgp->firstpage = NULL; 377 imgp->ps_strings = 0; 378 imgp->auxarg_size = 0; 379 imgp->args = args; 380 381 #ifdef MAC 382 error = mac_execve_enter(imgp, mac_p); 383 if (error) 384 goto exec_fail; 385 #endif 386 387 imgp->image_header = NULL; 388 389 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); 390 391 /* 392 * Translate the file name. namei() returns a vnode pointer 393 * in ni_vp amoung other things. 394 * 395 * XXXAUDIT: It would be desirable to also audit the name of the 396 * interpreter if this is an interpreted binary. 397 */ 398 if (args->fname != NULL) { 399 ndp = &nd; 400 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 401 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 402 } 403 404 interpret: 405 if (args->fname != NULL) { 406 error = namei(ndp); 407 if (error) 408 goto exec_fail; 409 410 vfslocked = NDHASGIANT(ndp); 411 binvp = ndp->ni_vp; 412 imgp->vp = binvp; 413 } else { 414 error = fgetvp(td, args->fd, &binvp); 415 if (error) 416 goto exec_fail; 417 vfslocked = VFS_LOCK_GIANT(binvp->v_mount); 418 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY); 419 imgp->vp = binvp; 420 } 421 422 /* 423 * Check file permissions (also 'opens' file) 424 */ 425 error = exec_check_permissions(imgp); 426 if (error) 427 goto exec_fail_dealloc; 428 429 imgp->object = imgp->vp->v_object; 430 if (imgp->object != NULL) 431 vm_object_reference(imgp->object); 432 433 /* 434 * Set VV_TEXT now so no one can write to the executable while we're 435 * activating it. 436 * 437 * Remember if this was set before and unset it in case this is not 438 * actually an executable image. 439 */ 440 textset = imgp->vp->v_vflag & VV_TEXT; 441 imgp->vp->v_vflag |= VV_TEXT; 442 443 error = exec_map_first_page(imgp); 444 if (error) 445 goto exec_fail_dealloc; 446 447 imgp->proc->p_osrel = 0; 448 /* 449 * If the current process has a special image activator it 450 * wants to try first, call it. For example, emulating shell 451 * scripts differently. 452 */ 453 error = -1; 454 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 455 error = img_first(imgp); 456 457 /* 458 * Loop through the list of image activators, calling each one. 459 * An activator returns -1 if there is no match, 0 on success, 460 * and an error otherwise. 461 */ 462 for (i = 0; error == -1 && execsw[i]; ++i) { 463 if (execsw[i]->ex_imgact == NULL || 464 execsw[i]->ex_imgact == img_first) { 465 continue; 466 } 467 error = (*execsw[i]->ex_imgact)(imgp); 468 } 469 470 if (error) { 471 if (error == -1) { 472 if (textset == 0) 473 imgp->vp->v_vflag &= ~VV_TEXT; 474 error = ENOEXEC; 475 } 476 goto exec_fail_dealloc; 477 } 478 479 /* 480 * Special interpreter operation, cleanup and loop up to try to 481 * activate the interpreter. 482 */ 483 if (imgp->interpreted) { 484 exec_unmap_first_page(imgp); 485 /* 486 * VV_TEXT needs to be unset for scripts. There is a short 487 * period before we determine that something is a script where 488 * VV_TEXT will be set. The vnode lock is held over this 489 * entire period so nothing should illegitimately be blocked. 490 */ 491 imgp->vp->v_vflag &= ~VV_TEXT; 492 /* free name buffer and old vnode */ 493 if (args->fname != NULL) 494 NDFREE(ndp, NDF_ONLY_PNBUF); 495 #ifdef MAC 496 interplabel = mac_vnode_label_alloc(); 497 mac_vnode_copy_label(binvp->v_label, interplabel); 498 #endif 499 vput(binvp); 500 vm_object_deallocate(imgp->object); 501 imgp->object = NULL; 502 VFS_UNLOCK_GIANT(vfslocked); 503 vfslocked = 0; 504 /* set new name to that of the interpreter */ 505 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE, 506 UIO_SYSSPACE, imgp->interpreter_name, td); 507 args->fname = imgp->interpreter_name; 508 goto interpret; 509 } 510 511 /* 512 * Copy out strings (args and env) and initialize stack base 513 */ 514 if (p->p_sysent->sv_copyout_strings) 515 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 516 else 517 stack_base = exec_copyout_strings(imgp); 518 519 /* 520 * If custom stack fixup routine present for this process 521 * let it do the stack setup. 522 * Else stuff argument count as first item on stack 523 */ 524 if (p->p_sysent->sv_fixup != NULL) 525 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 526 else 527 suword(--stack_base, imgp->args->argc); 528 529 /* 530 * For security and other reasons, the file descriptor table cannot 531 * be shared after an exec. 532 */ 533 fdunshare(p, td); 534 535 /* 536 * Malloc things before we need locks. 537 */ 538 newcred = crget(); 539 euip = uifind(attr.va_uid); 540 i = imgp->args->begin_envv - imgp->args->begin_argv; 541 /* Cache arguments if they fit inside our allowance */ 542 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 543 newargs = pargs_alloc(i); 544 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 545 } 546 547 /* close files on exec */ 548 VOP_UNLOCK(imgp->vp, 0); 549 fdcloseexec(td); 550 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 551 552 /* Get a reference to the vnode prior to locking the proc */ 553 VREF(binvp); 554 555 /* 556 * For security and other reasons, signal handlers cannot 557 * be shared after an exec. The new process gets a copy of the old 558 * handlers. In execsigs(), the new process will have its signals 559 * reset. 560 */ 561 PROC_LOCK(p); 562 if (sigacts_shared(p->p_sigacts)) { 563 oldsigacts = p->p_sigacts; 564 PROC_UNLOCK(p); 565 newsigacts = sigacts_alloc(); 566 sigacts_copy(newsigacts, oldsigacts); 567 PROC_LOCK(p); 568 p->p_sigacts = newsigacts; 569 } else 570 oldsigacts = NULL; 571 572 /* Stop profiling */ 573 stopprofclock(p); 574 575 /* reset caught signals */ 576 execsigs(p); 577 578 /* name this process - nameiexec(p, ndp) */ 579 if (args->fname) { 580 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN); 581 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len); 582 } else { 583 len = MAXCOMLEN; 584 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0) 585 len = MAXCOMLEN; 586 else { 587 len = sizeof(fexecv_proc_title); 588 bcopy(fexecv_proc_title, p->p_comm, len); 589 } 590 } 591 p->p_comm[len] = 0; 592 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 593 594 /* 595 * mark as execed, wakeup the process that vforked (if any) and tell 596 * it that it now has its own resources back 597 */ 598 p->p_flag |= P_EXEC; 599 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 600 p->p_flag &= ~P_PPWAIT; 601 wakeup(p->p_pptr); 602 } 603 604 /* 605 * Implement image setuid/setgid. 606 * 607 * Don't honor setuid/setgid if the filesystem prohibits it or if 608 * the process is being traced. 609 * 610 * XXXMAC: For the time being, use NOSUID to also prohibit 611 * transitions on the file system. 612 */ 613 oldcred = p->p_ucred; 614 credential_changing = 0; 615 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid != 616 attr.va_uid; 617 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid != 618 attr.va_gid; 619 #ifdef MAC 620 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 621 interplabel, imgp); 622 credential_changing |= will_transition; 623 #endif 624 625 if (credential_changing && 626 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 627 (p->p_flag & P_TRACED) == 0) { 628 /* 629 * Turn off syscall tracing for set-id programs, except for 630 * root. Record any set-id flags first to make sure that 631 * we do not regain any tracing during a possible block. 632 */ 633 setsugid(p); 634 635 #ifdef KTRACE 636 if (p->p_tracevp != NULL && 637 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) { 638 mtx_lock(&ktrace_mtx); 639 p->p_traceflag = 0; 640 tracevp = p->p_tracevp; 641 p->p_tracevp = NULL; 642 tracecred = p->p_tracecred; 643 p->p_tracecred = NULL; 644 mtx_unlock(&ktrace_mtx); 645 } 646 #endif 647 /* 648 * Close any file descriptors 0..2 that reference procfs, 649 * then make sure file descriptors 0..2 are in use. 650 * 651 * setugidsafety() may call closef() and then pfind() 652 * which may grab the process lock. 653 * fdcheckstd() may call falloc() which may block to 654 * allocate memory, so temporarily drop the process lock. 655 */ 656 PROC_UNLOCK(p); 657 setugidsafety(td); 658 VOP_UNLOCK(imgp->vp, 0); 659 error = fdcheckstd(td); 660 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 661 if (error != 0) 662 goto done1; 663 PROC_LOCK(p); 664 /* 665 * Set the new credentials. 666 */ 667 crcopy(newcred, oldcred); 668 if (attr.va_mode & VSUID) 669 change_euid(newcred, euip); 670 if (attr.va_mode & VSGID) 671 change_egid(newcred, attr.va_gid); 672 #ifdef MAC 673 if (will_transition) { 674 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 675 interplabel, imgp); 676 } 677 #endif 678 /* 679 * Implement correct POSIX saved-id behavior. 680 * 681 * XXXMAC: Note that the current logic will save the 682 * uid and gid if a MAC domain transition occurs, even 683 * though maybe it shouldn't. 684 */ 685 change_svuid(newcred, newcred->cr_uid); 686 change_svgid(newcred, newcred->cr_gid); 687 p->p_ucred = newcred; 688 newcred = NULL; 689 } else { 690 if (oldcred->cr_uid == oldcred->cr_ruid && 691 oldcred->cr_gid == oldcred->cr_rgid) 692 p->p_flag &= ~P_SUGID; 693 /* 694 * Implement correct POSIX saved-id behavior. 695 * 696 * XXX: It's not clear that the existing behavior is 697 * POSIX-compliant. A number of sources indicate that the 698 * saved uid/gid should only be updated if the new ruid is 699 * not equal to the old ruid, or the new euid is not equal 700 * to the old euid and the new euid is not equal to the old 701 * ruid. The FreeBSD code always updates the saved uid/gid. 702 * Also, this code uses the new (replaced) euid and egid as 703 * the source, which may or may not be the right ones to use. 704 */ 705 if (oldcred->cr_svuid != oldcred->cr_uid || 706 oldcred->cr_svgid != oldcred->cr_gid) { 707 crcopy(newcred, oldcred); 708 change_svuid(newcred, newcred->cr_uid); 709 change_svgid(newcred, newcred->cr_gid); 710 p->p_ucred = newcred; 711 newcred = NULL; 712 } 713 } 714 715 /* 716 * Store the vp for use in procfs. This vnode was referenced prior 717 * to locking the proc lock. 718 */ 719 textvp = p->p_textvp; 720 p->p_textvp = binvp; 721 722 #ifdef KDTRACE_HOOKS 723 /* 724 * Tell the DTrace fasttrap provider about the exec if it 725 * has declared an interest. 726 */ 727 if (dtrace_fasttrap_exec) 728 dtrace_fasttrap_exec(p); 729 #endif 730 731 /* 732 * Notify others that we exec'd, and clear the P_INEXEC flag 733 * as we're now a bona fide freshly-execed process. 734 */ 735 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 736 p->p_flag &= ~P_INEXEC; 737 738 /* 739 * If tracing the process, trap to debugger so breakpoints 740 * can be set before the program executes. 741 * Use tdsignal to deliver signal to current thread, use 742 * psignal may cause the signal to be delivered to wrong thread 743 * because that thread will exit, remember we are going to enter 744 * single thread mode. 745 */ 746 if (p->p_flag & P_TRACED) 747 tdsignal(p, td, SIGTRAP, NULL); 748 749 /* clear "fork but no exec" flag, as we _are_ execing */ 750 p->p_acflag &= ~AFORK; 751 752 /* 753 * Free any previous argument cache and replace it with 754 * the new argument cache, if any. 755 */ 756 oldargs = p->p_args; 757 p->p_args = newargs; 758 newargs = NULL; 759 760 #ifdef HWPMC_HOOKS 761 /* 762 * Check if system-wide sampling is in effect or if the 763 * current process is using PMCs. If so, do exec() time 764 * processing. This processing needs to happen AFTER the 765 * P_INEXEC flag is cleared. 766 * 767 * The proc lock needs to be released before taking the PMC 768 * SX. 769 */ 770 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 771 PROC_UNLOCK(p); 772 pe.pm_credentialschanged = credential_changing; 773 pe.pm_entryaddr = imgp->entry_addr; 774 775 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 776 } else 777 PROC_UNLOCK(p); 778 #else /* !HWPMC_HOOKS */ 779 PROC_UNLOCK(p); 780 #endif 781 782 /* Set values passed into the program in registers. */ 783 if (p->p_sysent->sv_setregs) 784 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr, 785 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 786 else 787 exec_setregs(td, imgp->entry_addr, 788 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 789 790 vfs_mark_atime(imgp->vp, td); 791 792 done1: 793 794 /* 795 * Free any resources malloc'd earlier that we didn't use. 796 */ 797 uifree(euip); 798 if (newcred == NULL) 799 crfree(oldcred); 800 else 801 crfree(newcred); 802 VOP_UNLOCK(imgp->vp, 0); 803 804 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0); 805 806 /* 807 * Handle deferred decrement of ref counts. 808 */ 809 if (textvp != NULL) { 810 int tvfslocked; 811 812 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount); 813 vrele(textvp); 814 VFS_UNLOCK_GIANT(tvfslocked); 815 } 816 if (binvp && error != 0) 817 vrele(binvp); 818 #ifdef KTRACE 819 if (tracevp != NULL) { 820 int tvfslocked; 821 822 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 823 vrele(tracevp); 824 VFS_UNLOCK_GIANT(tvfslocked); 825 } 826 if (tracecred != NULL) 827 crfree(tracecred); 828 #endif 829 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 830 if (oldargs != NULL) 831 pargs_drop(oldargs); 832 if (newargs != NULL) 833 pargs_drop(newargs); 834 if (oldsigacts != NULL) 835 sigacts_free(oldsigacts); 836 837 exec_fail_dealloc: 838 839 /* 840 * free various allocated resources 841 */ 842 if (imgp->firstpage != NULL) 843 exec_unmap_first_page(imgp); 844 845 if (imgp->vp != NULL) { 846 if (args->fname) 847 NDFREE(ndp, NDF_ONLY_PNBUF); 848 vput(imgp->vp); 849 } 850 851 if (imgp->object != NULL) 852 vm_object_deallocate(imgp->object); 853 854 if (error == 0) { 855 /* 856 * Stop the process here if its stop event mask has 857 * the S_EXEC bit set. 858 */ 859 STOPEVENT(p, S_EXEC, 0); 860 goto done2; 861 } 862 863 exec_fail: 864 /* we're done here, clear P_INEXEC */ 865 PROC_LOCK(p); 866 p->p_flag &= ~P_INEXEC; 867 PROC_UNLOCK(p); 868 869 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0); 870 871 done2: 872 #ifdef MAC 873 mac_execve_exit(imgp); 874 if (interplabel != NULL) 875 mac_vnode_label_free(interplabel); 876 #endif 877 VFS_UNLOCK_GIANT(vfslocked); 878 exec_free_args(args); 879 880 if (error && imgp->vmspace_destroyed) { 881 /* sorry, no more process anymore. exit gracefully */ 882 exit1(td, W_EXITCODE(0, SIGABRT)); 883 /* NOT REACHED */ 884 } 885 return (error); 886 } 887 888 int 889 exec_map_first_page(imgp) 890 struct image_params *imgp; 891 { 892 int rv, i; 893 int initial_pagein; 894 vm_page_t ma[VM_INITIAL_PAGEIN]; 895 vm_object_t object; 896 897 if (imgp->firstpage != NULL) 898 exec_unmap_first_page(imgp); 899 900 object = imgp->vp->v_object; 901 if (object == NULL) 902 return (EACCES); 903 VM_OBJECT_LOCK(object); 904 #if VM_NRESERVLEVEL > 0 905 if ((object->flags & OBJ_COLORED) == 0) { 906 object->flags |= OBJ_COLORED; 907 object->pg_color = 0; 908 } 909 #endif 910 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 911 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 912 initial_pagein = VM_INITIAL_PAGEIN; 913 if (initial_pagein > object->size) 914 initial_pagein = object->size; 915 for (i = 1; i < initial_pagein; i++) { 916 if ((ma[i] = vm_page_lookup(object, i)) != NULL) { 917 if (ma[i]->valid) 918 break; 919 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy) 920 break; 921 vm_page_busy(ma[i]); 922 } else { 923 ma[i] = vm_page_alloc(object, i, 924 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 925 if (ma[i] == NULL) 926 break; 927 } 928 } 929 initial_pagein = i; 930 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 931 ma[0] = vm_page_lookup(object, 0); 932 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || 933 (ma[0]->valid == 0)) { 934 if (ma[0]) { 935 vm_page_lock_queues(); 936 vm_page_free(ma[0]); 937 vm_page_unlock_queues(); 938 } 939 VM_OBJECT_UNLOCK(object); 940 return (EIO); 941 } 942 } 943 vm_page_lock_queues(); 944 vm_page_hold(ma[0]); 945 vm_page_unlock_queues(); 946 vm_page_wakeup(ma[0]); 947 VM_OBJECT_UNLOCK(object); 948 949 imgp->firstpage = sf_buf_alloc(ma[0], 0); 950 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 951 952 return (0); 953 } 954 955 void 956 exec_unmap_first_page(imgp) 957 struct image_params *imgp; 958 { 959 vm_page_t m; 960 961 if (imgp->firstpage != NULL) { 962 m = sf_buf_page(imgp->firstpage); 963 sf_buf_free(imgp->firstpage); 964 imgp->firstpage = NULL; 965 vm_page_lock_queues(); 966 vm_page_unhold(m); 967 vm_page_unlock_queues(); 968 } 969 } 970 971 /* 972 * Destroy old address space, and allocate a new stack 973 * The new stack is only SGROWSIZ large because it is grown 974 * automatically in trap.c. 975 */ 976 int 977 exec_new_vmspace(imgp, sv) 978 struct image_params *imgp; 979 struct sysentvec *sv; 980 { 981 int error; 982 struct proc *p = imgp->proc; 983 struct vmspace *vmspace = p->p_vmspace; 984 vm_offset_t stack_addr; 985 vm_map_t map; 986 u_long ssiz; 987 988 imgp->vmspace_destroyed = 1; 989 imgp->sysent = sv; 990 991 /* May be called with Giant held */ 992 EVENTHANDLER_INVOKE(process_exec, p, imgp); 993 994 /* 995 * Blow away entire process VM, if address space not shared, 996 * otherwise, create a new VM space so that other threads are 997 * not disrupted 998 */ 999 map = &vmspace->vm_map; 1000 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser && 1001 vm_map_max(map) == sv->sv_maxuser) { 1002 shmexit(vmspace); 1003 pmap_remove_pages(vmspace_pmap(vmspace)); 1004 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1005 } else { 1006 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser); 1007 if (error) 1008 return (error); 1009 vmspace = p->p_vmspace; 1010 map = &vmspace->vm_map; 1011 } 1012 1013 /* Allocate a new stack */ 1014 if (sv->sv_maxssiz != NULL) 1015 ssiz = *sv->sv_maxssiz; 1016 else 1017 ssiz = maxssiz; 1018 stack_addr = sv->sv_usrstack - ssiz; 1019 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1020 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1021 if (error) 1022 return (error); 1023 1024 #ifdef __ia64__ 1025 /* Allocate a new register stack */ 1026 stack_addr = IA64_BACKINGSTORE; 1027 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1028 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1029 if (error) 1030 return (error); 1031 #endif 1032 1033 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1034 * VM_STACK case, but they are still used to monitor the size of the 1035 * process stack so we can check the stack rlimit. 1036 */ 1037 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1038 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; 1039 1040 return (0); 1041 } 1042 1043 /* 1044 * Copy out argument and environment strings from the old process address 1045 * space into the temporary string buffer. 1046 */ 1047 int 1048 exec_copyin_args(struct image_args *args, char *fname, 1049 enum uio_seg segflg, char **argv, char **envv) 1050 { 1051 char *argp, *envp; 1052 int error; 1053 size_t length; 1054 1055 error = 0; 1056 1057 bzero(args, sizeof(*args)); 1058 if (argv == NULL) 1059 return (EFAULT); 1060 /* 1061 * Allocate temporary demand zeroed space for argument and 1062 * environment strings: 1063 * 1064 * o ARG_MAX for argument and environment; 1065 * o MAXSHELLCMDLEN for the name of interpreters. 1066 */ 1067 args->buf = (char *) kmem_alloc_wait(exec_map, 1068 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1069 if (args->buf == NULL) 1070 return (ENOMEM); 1071 args->begin_argv = args->buf; 1072 args->endp = args->begin_argv; 1073 args->stringspace = ARG_MAX; 1074 /* 1075 * Copy the file name. 1076 */ 1077 if (fname != NULL) { 1078 args->fname = args->buf + ARG_MAX; 1079 error = (segflg == UIO_SYSSPACE) ? 1080 copystr(fname, args->fname, PATH_MAX, &length) : 1081 copyinstr(fname, args->fname, PATH_MAX, &length); 1082 if (error != 0) 1083 goto err_exit; 1084 } else 1085 args->fname = NULL; 1086 1087 /* 1088 * extract arguments first 1089 */ 1090 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) { 1091 if (argp == (caddr_t) -1) { 1092 error = EFAULT; 1093 goto err_exit; 1094 } 1095 if ((error = copyinstr(argp, args->endp, 1096 args->stringspace, &length))) { 1097 if (error == ENAMETOOLONG) 1098 error = E2BIG; 1099 goto err_exit; 1100 } 1101 args->stringspace -= length; 1102 args->endp += length; 1103 args->argc++; 1104 } 1105 1106 args->begin_envv = args->endp; 1107 1108 /* 1109 * extract environment strings 1110 */ 1111 if (envv) { 1112 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 1113 if (envp == (caddr_t)-1) { 1114 error = EFAULT; 1115 goto err_exit; 1116 } 1117 if ((error = copyinstr(envp, args->endp, 1118 args->stringspace, &length))) { 1119 if (error == ENAMETOOLONG) 1120 error = E2BIG; 1121 goto err_exit; 1122 } 1123 args->stringspace -= length; 1124 args->endp += length; 1125 args->envc++; 1126 } 1127 } 1128 1129 return (0); 1130 1131 err_exit: 1132 exec_free_args(args); 1133 return (error); 1134 } 1135 1136 static void 1137 exec_free_args(struct image_args *args) 1138 { 1139 1140 if (args->buf) { 1141 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, 1142 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1143 args->buf = NULL; 1144 } 1145 } 1146 1147 /* 1148 * Copy strings out to the new process address space, constructing new arg 1149 * and env vector tables. Return a pointer to the base so that it can be used 1150 * as the initial stack pointer. 1151 */ 1152 register_t * 1153 exec_copyout_strings(imgp) 1154 struct image_params *imgp; 1155 { 1156 int argc, envc; 1157 char **vectp; 1158 char *stringp, *destp; 1159 register_t *stack_base; 1160 struct ps_strings *arginfo; 1161 struct proc *p; 1162 int szsigcode; 1163 1164 /* 1165 * Calculate string base and vector table pointers. 1166 * Also deal with signal trampoline code for this exec type. 1167 */ 1168 p = imgp->proc; 1169 szsigcode = 0; 1170 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1171 if (p->p_sysent->sv_szsigcode != NULL) 1172 szsigcode = *(p->p_sysent->sv_szsigcode); 1173 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - 1174 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); 1175 1176 /* 1177 * install sigcode 1178 */ 1179 if (szsigcode) 1180 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - 1181 szsigcode), szsigcode); 1182 1183 /* 1184 * If we have a valid auxargs ptr, prepare some room 1185 * on the stack. 1186 */ 1187 if (imgp->auxargs) { 1188 /* 1189 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1190 * lower compatibility. 1191 */ 1192 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1193 (AT_COUNT * 2); 1194 /* 1195 * The '+ 2' is for the null pointers at the end of each of 1196 * the arg and env vector sets,and imgp->auxarg_size is room 1197 * for argument of Runtime loader. 1198 */ 1199 vectp = (char **)(destp - (imgp->args->argc + 1200 imgp->args->envc + 2 + imgp->auxarg_size) * 1201 sizeof(char *)); 1202 1203 } else { 1204 /* 1205 * The '+ 2' is for the null pointers at the end of each of 1206 * the arg and env vector sets 1207 */ 1208 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * 1209 sizeof(char *)); 1210 } 1211 1212 /* 1213 * vectp also becomes our initial stack base 1214 */ 1215 stack_base = (register_t *)vectp; 1216 1217 stringp = imgp->args->begin_argv; 1218 argc = imgp->args->argc; 1219 envc = imgp->args->envc; 1220 1221 /* 1222 * Copy out strings - arguments and environment. 1223 */ 1224 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 1225 1226 /* 1227 * Fill in "ps_strings" struct for ps, w, etc. 1228 */ 1229 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1230 suword(&arginfo->ps_nargvstr, argc); 1231 1232 /* 1233 * Fill in argument portion of vector table. 1234 */ 1235 for (; argc > 0; --argc) { 1236 suword(vectp++, (long)(intptr_t)destp); 1237 while (*stringp++ != 0) 1238 destp++; 1239 destp++; 1240 } 1241 1242 /* a null vector table pointer separates the argp's from the envp's */ 1243 suword(vectp++, 0); 1244 1245 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1246 suword(&arginfo->ps_nenvstr, envc); 1247 1248 /* 1249 * Fill in environment portion of vector table. 1250 */ 1251 for (; envc > 0; --envc) { 1252 suword(vectp++, (long)(intptr_t)destp); 1253 while (*stringp++ != 0) 1254 destp++; 1255 destp++; 1256 } 1257 1258 /* end of vector table is a null pointer */ 1259 suword(vectp, 0); 1260 1261 return (stack_base); 1262 } 1263 1264 /* 1265 * Check permissions of file to execute. 1266 * Called with imgp->vp locked. 1267 * Return 0 for success or error code on failure. 1268 */ 1269 int 1270 exec_check_permissions(imgp) 1271 struct image_params *imgp; 1272 { 1273 struct vnode *vp = imgp->vp; 1274 struct vattr *attr = imgp->attr; 1275 struct thread *td; 1276 int error; 1277 1278 td = curthread; 1279 1280 /* Get file attributes */ 1281 error = VOP_GETATTR(vp, attr, td->td_ucred, td); 1282 if (error) 1283 return (error); 1284 1285 #ifdef MAC 1286 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1287 if (error) 1288 return (error); 1289 #endif 1290 1291 /* 1292 * 1) Check if file execution is disabled for the filesystem that this 1293 * file resides on. 1294 * 2) Insure that at least one execute bit is on - otherwise root 1295 * will always succeed, and we don't want to happen unless the 1296 * file really is executable. 1297 * 3) Insure that the file is a regular file. 1298 */ 1299 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1300 ((attr->va_mode & 0111) == 0) || 1301 (attr->va_type != VREG)) 1302 return (EACCES); 1303 1304 /* 1305 * Zero length files can't be exec'd 1306 */ 1307 if (attr->va_size == 0) 1308 return (ENOEXEC); 1309 1310 /* 1311 * Check for execute permission to file based on current credentials. 1312 */ 1313 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1314 if (error) 1315 return (error); 1316 1317 /* 1318 * Check number of open-for-writes on the file and deny execution 1319 * if there are any. 1320 */ 1321 if (vp->v_writecount) 1322 return (ETXTBSY); 1323 1324 /* 1325 * Call filesystem specific open routine (which does nothing in the 1326 * general case). 1327 */ 1328 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1329 return (error); 1330 } 1331 1332 /* 1333 * Exec handler registration 1334 */ 1335 int 1336 exec_register(execsw_arg) 1337 const struct execsw *execsw_arg; 1338 { 1339 const struct execsw **es, **xs, **newexecsw; 1340 int count = 2; /* New slot and trailing NULL */ 1341 1342 if (execsw) 1343 for (es = execsw; *es; es++) 1344 count++; 1345 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1346 if (newexecsw == NULL) 1347 return (ENOMEM); 1348 xs = newexecsw; 1349 if (execsw) 1350 for (es = execsw; *es; es++) 1351 *xs++ = *es; 1352 *xs++ = execsw_arg; 1353 *xs = NULL; 1354 if (execsw) 1355 free(execsw, M_TEMP); 1356 execsw = newexecsw; 1357 return (0); 1358 } 1359 1360 int 1361 exec_unregister(execsw_arg) 1362 const struct execsw *execsw_arg; 1363 { 1364 const struct execsw **es, **xs, **newexecsw; 1365 int count = 1; 1366 1367 if (execsw == NULL) 1368 panic("unregister with no handlers left?\n"); 1369 1370 for (es = execsw; *es; es++) { 1371 if (*es == execsw_arg) 1372 break; 1373 } 1374 if (*es == NULL) 1375 return (ENOENT); 1376 for (es = execsw; *es; es++) 1377 if (*es != execsw_arg) 1378 count++; 1379 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1380 if (newexecsw == NULL) 1381 return (ENOMEM); 1382 xs = newexecsw; 1383 for (es = execsw; *es; es++) 1384 if (*es != execsw_arg) 1385 *xs++ = *es; 1386 *xs = NULL; 1387 if (execsw) 1388 free(execsw, M_TEMP); 1389 execsw = newexecsw; 1390 return (0); 1391 } 1392