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