kern_sharedpage.c revision 146829
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: head/sys/kern/kern_exec.c 146829 2005-05-31 19:39:52Z kensmith $"); 29 30#include "opt_ktrace.h" 31#include "opt_mac.h" 32 33#include <sys/param.h> 34#include <sys/systm.h> 35#include <sys/eventhandler.h> 36#include <sys/lock.h> 37#include <sys/mutex.h> 38#include <sys/sysproto.h> 39#include <sys/signalvar.h> 40#include <sys/kernel.h> 41#include <sys/mac.h> 42#include <sys/mount.h> 43#include <sys/filedesc.h> 44#include <sys/fcntl.h> 45#include <sys/acct.h> 46#include <sys/exec.h> 47#include <sys/imgact.h> 48#include <sys/imgact_elf.h> 49#include <sys/wait.h> 50#include <sys/malloc.h> 51#include <sys/proc.h> 52#include <sys/pioctl.h> 53#include <sys/namei.h> 54#include <sys/resourcevar.h> 55#include <sys/sf_buf.h> 56#include <sys/syscallsubr.h> 57#include <sys/sysent.h> 58#include <sys/shm.h> 59#include <sys/sysctl.h> 60#include <sys/vnode.h> 61#ifdef KTRACE 62#include <sys/ktrace.h> 63#endif 64 65#include <vm/vm.h> 66#include <vm/vm_param.h> 67#include <vm/pmap.h> 68#include <vm/vm_page.h> 69#include <vm/vm_map.h> 70#include <vm/vm_kern.h> 71#include <vm/vm_extern.h> 72#include <vm/vm_object.h> 73#include <vm/vm_pager.h> 74 75#ifdef HWPMC_HOOKS 76#include <sys/pmckern.h> 77#endif 78 79#include <machine/reg.h> 80 81MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 82 83static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 84static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 85static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 86static int do_execve(struct thread *td, struct image_args *args, 87 struct mac *mac_p); 88 89/* XXX This should be vm_size_t. */ 90SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 91 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 92 93/* XXX This should be vm_size_t. */ 94SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD, 95 NULL, 0, sysctl_kern_usrstack, "LU", ""); 96 97SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 98 NULL, 0, sysctl_kern_stackprot, "I", ""); 99 100u_long ps_arg_cache_limit = PAGE_SIZE / 16; 101SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 102 &ps_arg_cache_limit, 0, ""); 103 104static int 105sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 106{ 107 struct proc *p; 108 int error; 109 110 p = curproc; 111#ifdef SCTL_MASK32 112 if (req->flags & SCTL_MASK32) { 113 unsigned int val; 114 val = (unsigned int)p->p_sysent->sv_psstrings; 115 error = SYSCTL_OUT(req, &val, sizeof(val)); 116 } else 117#endif 118 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 119 sizeof(p->p_sysent->sv_psstrings)); 120 return error; 121} 122 123static int 124sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 125{ 126 struct proc *p; 127 int error; 128 129 p = curproc; 130#ifdef SCTL_MASK32 131 if (req->flags & SCTL_MASK32) { 132 unsigned int val; 133 val = (unsigned int)p->p_sysent->sv_usrstack; 134 error = SYSCTL_OUT(req, &val, sizeof(val)); 135 } else 136#endif 137 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 138 sizeof(p->p_sysent->sv_usrstack)); 139 return error; 140} 141 142static int 143sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 144{ 145 struct proc *p; 146 147 p = curproc; 148 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 149 sizeof(p->p_sysent->sv_stackprot))); 150} 151 152/* 153 * Each of the items is a pointer to a `const struct execsw', hence the 154 * double pointer here. 155 */ 156static const struct execsw **execsw; 157 158#ifndef _SYS_SYSPROTO_H_ 159struct execve_args { 160 char *fname; 161 char **argv; 162 char **envv; 163}; 164#endif 165 166/* 167 * MPSAFE 168 */ 169int 170execve(td, uap) 171 struct thread *td; 172 struct execve_args /* { 173 char *fname; 174 char **argv; 175 char **envv; 176 } */ *uap; 177{ 178 int error; 179 struct image_args args; 180 181 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 182 uap->argv, uap->envv); 183 184 if (error == 0) 185 error = kern_execve(td, &args, NULL); 186 187 exec_free_args(&args); 188 189 return (error); 190} 191 192#ifndef _SYS_SYSPROTO_H_ 193struct __mac_execve_args { 194 char *fname; 195 char **argv; 196 char **envv; 197 struct mac *mac_p; 198}; 199#endif 200 201/* 202 * MPSAFE 203 */ 204int 205__mac_execve(td, uap) 206 struct thread *td; 207 struct __mac_execve_args /* { 208 char *fname; 209 char **argv; 210 char **envv; 211 struct mac *mac_p; 212 } */ *uap; 213{ 214#ifdef MAC 215 int error; 216 struct image_args args; 217 218 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 219 uap->argv, uap->envv); 220 221 if (error == 0) 222 error = kern_execve(td, &args, uap->mac_p); 223 224 exec_free_args(&args); 225 226 return (error); 227#else 228 return (ENOSYS); 229#endif 230} 231 232int 233kern_execve(td, args, mac_p) 234 struct thread *td; 235 struct image_args *args; 236 struct mac *mac_p; 237{ 238 struct proc *p = td->td_proc; 239 int error; 240 241 if (p->p_flag & P_HADTHREADS) { 242 PROC_LOCK(p); 243 if (thread_single(SINGLE_BOUNDARY)) { 244 PROC_UNLOCK(p); 245 return (ERESTART); /* Try again later. */ 246 } 247 PROC_UNLOCK(p); 248 } 249 250 error = do_execve(td, args, mac_p); 251 252 if (p->p_flag & P_HADTHREADS) { 253 PROC_LOCK(p); 254 /* 255 * If success, we upgrade to SINGLE_EXIT state to 256 * force other threads to suicide. 257 */ 258 if (error == 0) 259 thread_single(SINGLE_EXIT); 260 else 261 thread_single_end(); 262 PROC_UNLOCK(p); 263 } 264 265 return (error); 266} 267 268/* 269 * In-kernel implementation of execve(). All arguments are assumed to be 270 * userspace pointers from the passed thread. 271 * 272 * MPSAFE 273 */ 274static int 275do_execve(td, args, mac_p) 276 struct thread *td; 277 struct image_args *args; 278 struct mac *mac_p; 279{ 280 struct proc *p = td->td_proc; 281 struct nameidata nd, *ndp; 282 struct ucred *newcred = NULL, *oldcred; 283 struct uidinfo *euip; 284 register_t *stack_base; 285 int error, len, i; 286 struct image_params image_params, *imgp; 287 struct vattr atimeattr, attr; 288 int (*img_first)(struct image_params *); 289 struct pargs *oldargs = NULL, *newargs = NULL; 290 struct sigacts *oldsigacts, *newsigacts; 291#ifdef KTRACE 292 struct vnode *tracevp = NULL; 293 struct ucred *tracecred = NULL; 294#endif 295 struct vnode *textvp = NULL; 296 int credential_changing; 297 int vfslocked; 298 int textset; 299#ifdef MAC 300 struct label *interplabel = NULL; 301 int will_transition; 302#endif 303 304 vfslocked = 0; 305 imgp = &image_params; 306 307 /* 308 * Lock the process and set the P_INEXEC flag to indicate that 309 * it should be left alone until we're done here. This is 310 * necessary to avoid race conditions - e.g. in ptrace() - 311 * that might allow a local user to illicitly obtain elevated 312 * privileges. 313 */ 314 PROC_LOCK(p); 315 KASSERT((p->p_flag & P_INEXEC) == 0, 316 ("%s(): process already has P_INEXEC flag", __func__)); 317 p->p_flag |= P_INEXEC; 318 PROC_UNLOCK(p); 319 320 /* 321 * Initialize part of the common data 322 */ 323 imgp->proc = p; 324 imgp->execlabel = NULL; 325 imgp->attr = &attr; 326 imgp->entry_addr = 0; 327 imgp->vmspace_destroyed = 0; 328 imgp->interpreted = 0; 329 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX; 330 imgp->auxargs = NULL; 331 imgp->vp = NULL; 332 imgp->object = NULL; 333 imgp->firstpage = NULL; 334 imgp->ps_strings = 0; 335 imgp->auxarg_size = 0; 336 imgp->args = args; 337 338#ifdef MAC 339 error = mac_execve_enter(imgp, mac_p); 340 if (error) 341 goto exec_fail; 342#endif 343 344 imgp->image_header = NULL; 345 346 /* 347 * Translate the file name. namei() returns a vnode pointer 348 * in ni_vp amoung other things. 349 */ 350 ndp = &nd; 351 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE, 352 UIO_SYSSPACE, args->fname, td); 353 354interpret: 355 error = namei(ndp); 356 if (error) 357 goto exec_fail; 358 359 vfslocked = NDHASGIANT(ndp); 360 imgp->vp = ndp->ni_vp; 361 362 /* 363 * Check file permissions (also 'opens' file) 364 */ 365 error = exec_check_permissions(imgp); 366 if (error) 367 goto exec_fail_dealloc; 368 369 imgp->object = imgp->vp->v_object; 370 if (imgp->object != NULL) 371 vm_object_reference(imgp->object); 372 373 /* 374 * Set VV_TEXT now so no one can write to the executable while we're 375 * activating it. 376 * 377 * Remember if this was set before and unset it in case this is not 378 * actually an executable image. 379 */ 380 textset = imgp->vp->v_vflag & VV_TEXT; 381 imgp->vp->v_vflag |= VV_TEXT; 382 383 error = exec_map_first_page(imgp); 384 if (error) 385 goto exec_fail_dealloc; 386 387 /* 388 * If the current process has a special image activator it 389 * wants to try first, call it. For example, emulating shell 390 * scripts differently. 391 */ 392 error = -1; 393 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 394 error = img_first(imgp); 395 396 /* 397 * Loop through the list of image activators, calling each one. 398 * An activator returns -1 if there is no match, 0 on success, 399 * and an error otherwise. 400 */ 401 for (i = 0; error == -1 && execsw[i]; ++i) { 402 if (execsw[i]->ex_imgact == NULL || 403 execsw[i]->ex_imgact == img_first) { 404 continue; 405 } 406 error = (*execsw[i]->ex_imgact)(imgp); 407 } 408 409 if (error) { 410 if (error == -1) { 411 if (textset == 0) 412 imgp->vp->v_vflag &= ~VV_TEXT; 413 error = ENOEXEC; 414 } 415 goto exec_fail_dealloc; 416 } 417 418 /* 419 * Special interpreter operation, cleanup and loop up to try to 420 * activate the interpreter. 421 */ 422 if (imgp->interpreted) { 423 exec_unmap_first_page(imgp); 424 /* 425 * VV_TEXT needs to be unset for scripts. There is a short 426 * period before we determine that something is a script where 427 * VV_TEXT will be set. The vnode lock is held over this 428 * entire period so nothing should illegitimately be blocked. 429 */ 430 imgp->vp->v_vflag &= ~VV_TEXT; 431 /* free name buffer and old vnode */ 432 NDFREE(ndp, NDF_ONLY_PNBUF); 433#ifdef MAC 434 interplabel = mac_vnode_label_alloc(); 435 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel); 436#endif 437 vput(ndp->ni_vp); 438 vm_object_deallocate(imgp->object); 439 imgp->object = NULL; 440 VFS_UNLOCK_GIANT(vfslocked); 441 vfslocked = 0; 442 /* set new name to that of the interpreter */ 443 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE, 444 UIO_SYSSPACE, imgp->interpreter_name, td); 445 goto interpret; 446 } 447 448 /* 449 * Copy out strings (args and env) and initialize stack base 450 */ 451 if (p->p_sysent->sv_copyout_strings) 452 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 453 else 454 stack_base = exec_copyout_strings(imgp); 455 456 /* 457 * If custom stack fixup routine present for this process 458 * let it do the stack setup. 459 * Else stuff argument count as first item on stack 460 */ 461 if (p->p_sysent->sv_fixup != NULL) 462 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 463 else 464 suword(--stack_base, imgp->args->argc); 465 466 /* 467 * For security and other reasons, the file descriptor table cannot 468 * be shared after an exec. 469 */ 470 fdunshare(p, td); 471 472 /* 473 * Malloc things before we need locks. 474 */ 475 newcred = crget(); 476 euip = uifind(attr.va_uid); 477 i = imgp->args->begin_envv - imgp->args->begin_argv; 478 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) 479 newargs = pargs_alloc(i); 480 481 /* close files on exec */ 482 fdcloseexec(td); 483 484 /* Get a reference to the vnode prior to locking the proc */ 485 VREF(ndp->ni_vp); 486 487 /* 488 * For security and other reasons, signal handlers cannot 489 * be shared after an exec. The new process gets a copy of the old 490 * handlers. In execsigs(), the new process will have its signals 491 * reset. 492 */ 493 PROC_LOCK(p); 494 if (sigacts_shared(p->p_sigacts)) { 495 oldsigacts = p->p_sigacts; 496 PROC_UNLOCK(p); 497 newsigacts = sigacts_alloc(); 498 sigacts_copy(newsigacts, oldsigacts); 499 PROC_LOCK(p); 500 p->p_sigacts = newsigacts; 501 } else 502 oldsigacts = NULL; 503 504 /* Stop profiling */ 505 stopprofclock(p); 506 507 /* reset caught signals */ 508 execsigs(p); 509 510 /* name this process - nameiexec(p, ndp) */ 511 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN); 512 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len); 513 p->p_comm[len] = 0; 514 515 /* 516 * mark as execed, wakeup the process that vforked (if any) and tell 517 * it that it now has its own resources back 518 */ 519 p->p_flag |= P_EXEC; 520 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 521 p->p_flag &= ~P_PPWAIT; 522 wakeup(p->p_pptr); 523 } 524 525 /* 526 * Implement image setuid/setgid. 527 * 528 * Don't honor setuid/setgid if the filesystem prohibits it or if 529 * the process is being traced. 530 * 531 * XXXMAC: For the time being, use NOSUID to also prohibit 532 * transitions on the file system. 533 */ 534 oldcred = p->p_ucred; 535 credential_changing = 0; 536 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid != 537 attr.va_uid; 538 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid != 539 attr.va_gid; 540#ifdef MAC 541 will_transition = mac_execve_will_transition(oldcred, imgp->vp, 542 interplabel, imgp); 543 credential_changing |= will_transition; 544#endif 545 546 if (credential_changing && 547 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 548 (p->p_flag & P_TRACED) == 0) { 549 /* 550 * Turn off syscall tracing for set-id programs, except for 551 * root. Record any set-id flags first to make sure that 552 * we do not regain any tracing during a possible block. 553 */ 554 setsugid(p); 555#ifdef KTRACE 556 if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) { 557 mtx_lock(&ktrace_mtx); 558 p->p_traceflag = 0; 559 tracevp = p->p_tracevp; 560 p->p_tracevp = NULL; 561 tracecred = p->p_tracecred; 562 p->p_tracecred = NULL; 563 mtx_unlock(&ktrace_mtx); 564 } 565#endif 566 /* 567 * Close any file descriptors 0..2 that reference procfs, 568 * then make sure file descriptors 0..2 are in use. 569 * 570 * setugidsafety() may call closef() and then pfind() 571 * which may grab the process lock. 572 * fdcheckstd() may call falloc() which may block to 573 * allocate memory, so temporarily drop the process lock. 574 */ 575 PROC_UNLOCK(p); 576 setugidsafety(td); 577 error = fdcheckstd(td); 578 if (error != 0) 579 goto done1; 580 PROC_LOCK(p); 581 /* 582 * Set the new credentials. 583 */ 584 crcopy(newcred, oldcred); 585 if (attr.va_mode & VSUID) 586 change_euid(newcred, euip); 587 if (attr.va_mode & VSGID) 588 change_egid(newcred, attr.va_gid); 589#ifdef MAC 590 if (will_transition) { 591 mac_execve_transition(oldcred, newcred, imgp->vp, 592 interplabel, imgp); 593 } 594#endif 595 /* 596 * Implement correct POSIX saved-id behavior. 597 * 598 * XXXMAC: Note that the current logic will save the 599 * uid and gid if a MAC domain transition occurs, even 600 * though maybe it shouldn't. 601 */ 602 change_svuid(newcred, newcred->cr_uid); 603 change_svgid(newcred, newcred->cr_gid); 604 p->p_ucred = newcred; 605 newcred = NULL; 606 } else { 607 if (oldcred->cr_uid == oldcred->cr_ruid && 608 oldcred->cr_gid == oldcred->cr_rgid) 609 p->p_flag &= ~P_SUGID; 610 /* 611 * Implement correct POSIX saved-id behavior. 612 * 613 * XXX: It's not clear that the existing behavior is 614 * POSIX-compliant. A number of sources indicate that the 615 * saved uid/gid should only be updated if the new ruid is 616 * not equal to the old ruid, or the new euid is not equal 617 * to the old euid and the new euid is not equal to the old 618 * ruid. The FreeBSD code always updates the saved uid/gid. 619 * Also, this code uses the new (replaced) euid and egid as 620 * the source, which may or may not be the right ones to use. 621 */ 622 if (oldcred->cr_svuid != oldcred->cr_uid || 623 oldcred->cr_svgid != oldcred->cr_gid) { 624 crcopy(newcred, oldcred); 625 change_svuid(newcred, newcred->cr_uid); 626 change_svgid(newcred, newcred->cr_gid); 627 p->p_ucred = newcred; 628 newcred = NULL; 629 } 630 } 631 632 /* 633 * Store the vp for use in procfs. This vnode was referenced prior 634 * to locking the proc lock. 635 */ 636 textvp = p->p_textvp; 637 p->p_textvp = ndp->ni_vp; 638 639 /* 640 * Notify others that we exec'd, and clear the P_INEXEC flag 641 * as we're now a bona fide freshly-execed process. 642 */ 643 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 644 p->p_flag &= ~P_INEXEC; 645 646 /* 647 * If tracing the process, trap to debugger so breakpoints 648 * can be set before the program executes. 649 * Use tdsignal to deliver signal to current thread, use 650 * psignal may cause the signal to be delivered to wrong thread 651 * because that thread will exit, remember we are going to enter 652 * single thread mode. 653 */ 654 if (p->p_flag & P_TRACED) 655 tdsignal(td, SIGTRAP, SIGTARGET_TD); 656 657 /* clear "fork but no exec" flag, as we _are_ execing */ 658 p->p_acflag &= ~AFORK; 659 660 /* Free any previous argument cache */ 661 oldargs = p->p_args; 662 p->p_args = NULL; 663 664 /* Cache arguments if they fit inside our allowance */ 665 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 666 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 667 p->p_args = newargs; 668 newargs = NULL; 669 } 670 671#ifdef HWPMC_HOOKS 672 /* 673 * Check if the process is using PMCs and if so do exec() time 674 * processing. This processing needs to happen AFTER the 675 * P_INEXEC flag is cleared. 676 * 677 * The proc lock needs to be released before taking the PMC 678 * SX. 679 */ 680 if (PMC_PROC_IS_USING_PMCS(p)) { 681 PROC_UNLOCK(p); 682 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, 683 (void *) &credential_changing); 684 } else 685 PROC_UNLOCK(p); 686#else /* !HWPMC_HOOKS */ 687 PROC_UNLOCK(p); 688#endif 689 690 /* Set values passed into the program in registers. */ 691 if (p->p_sysent->sv_setregs) 692 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr, 693 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 694 else 695 exec_setregs(td, imgp->entry_addr, 696 (u_long)(uintptr_t)stack_base, imgp->ps_strings); 697 698 /* 699 * Here we should update the access time of the file. This must 700 * be implemented by the underlying filesystem in the same way as 701 * access timestamps for a VOP_READ() because we want to avoid 702 * blocking and/or I/O, and have not called vn_start_write(). 703 */ 704 if ((imgp->vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) { 705 VATTR_NULL(&atimeattr); 706 atimeattr.va_vaflags |= VA_EXECVE_ATIME; 707 (void)VOP_SETATTR(imgp->vp, &atimeattr, td->td_ucred, td); 708 } 709 710done1: 711 /* 712 * Free any resources malloc'd earlier that we didn't use. 713 */ 714 uifree(euip); 715 if (newcred == NULL) 716 crfree(oldcred); 717 else 718 crfree(newcred); 719 /* 720 * Handle deferred decrement of ref counts. 721 */ 722 if (textvp != NULL) 723 vrele(textvp); 724 if (ndp->ni_vp && error != 0) 725 vrele(ndp->ni_vp); 726#ifdef KTRACE 727 if (tracevp != NULL) 728 vrele(tracevp); 729 if (tracecred != NULL) 730 crfree(tracecred); 731#endif 732 if (oldargs != NULL) 733 pargs_drop(oldargs); 734 if (newargs != NULL) 735 pargs_drop(newargs); 736 if (oldsigacts != NULL) 737 sigacts_free(oldsigacts); 738 739exec_fail_dealloc: 740 741 /* 742 * free various allocated resources 743 */ 744 if (imgp->firstpage != NULL) 745 exec_unmap_first_page(imgp); 746 747 if (imgp->vp != NULL) { 748 NDFREE(ndp, NDF_ONLY_PNBUF); 749 vput(imgp->vp); 750 } 751 752 if (imgp->object != NULL) 753 vm_object_deallocate(imgp->object); 754 755 if (error == 0) { 756 /* 757 * Stop the process here if its stop event mask has 758 * the S_EXEC bit set. 759 */ 760 STOPEVENT(p, S_EXEC, 0); 761 goto done2; 762 } 763 764exec_fail: 765 /* we're done here, clear P_INEXEC */ 766 PROC_LOCK(p); 767 p->p_flag &= ~P_INEXEC; 768 PROC_UNLOCK(p); 769 770 if (imgp->vmspace_destroyed) { 771 /* sorry, no more process anymore. exit gracefully */ 772#ifdef MAC 773 mac_execve_exit(imgp); 774 if (interplabel != NULL) 775 mac_vnode_label_free(interplabel); 776#endif 777 VFS_UNLOCK_GIANT(vfslocked); 778 exit1(td, W_EXITCODE(0, SIGABRT)); 779 /* NOT REACHED */ 780 error = 0; 781 } 782done2: 783#ifdef MAC 784 mac_execve_exit(imgp); 785 if (interplabel != NULL) 786 mac_vnode_label_free(interplabel); 787#endif 788 VFS_UNLOCK_GIANT(vfslocked); 789 return (error); 790} 791 792int 793exec_map_first_page(imgp) 794 struct image_params *imgp; 795{ 796 int rv, i; 797 int initial_pagein; 798 vm_page_t ma[VM_INITIAL_PAGEIN]; 799 vm_object_t object; 800 801 if (imgp->firstpage != NULL) 802 exec_unmap_first_page(imgp); 803 804 object = imgp->vp->v_object; 805 if (object == NULL) 806 return (EACCES); 807 VM_OBJECT_LOCK(object); 808 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 809 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 810 initial_pagein = VM_INITIAL_PAGEIN; 811 if (initial_pagein > object->size) 812 initial_pagein = object->size; 813 for (i = 1; i < initial_pagein; i++) { 814 if ((ma[i] = vm_page_lookup(object, i)) != NULL) { 815 if (ma[i]->valid) 816 break; 817 vm_page_lock_queues(); 818 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) { 819 vm_page_unlock_queues(); 820 break; 821 } 822 vm_page_busy(ma[i]); 823 vm_page_unlock_queues(); 824 } else { 825 ma[i] = vm_page_alloc(object, i, 826 VM_ALLOC_NORMAL); 827 if (ma[i] == NULL) 828 break; 829 } 830 } 831 initial_pagein = i; 832 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 833 ma[0] = vm_page_lookup(object, 0); 834 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) || 835 (ma[0]->valid == 0)) { 836 if (ma[0]) { 837 vm_page_lock_queues(); 838 pmap_remove_all(ma[0]); 839 vm_page_free(ma[0]); 840 vm_page_unlock_queues(); 841 } 842 VM_OBJECT_UNLOCK(object); 843 return (EIO); 844 } 845 } 846 vm_page_lock_queues(); 847 vm_page_hold(ma[0]); 848 vm_page_wakeup(ma[0]); 849 vm_page_unlock_queues(); 850 VM_OBJECT_UNLOCK(object); 851 852 imgp->firstpage = sf_buf_alloc(ma[0], 0); 853 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 854 855 return (0); 856} 857 858void 859exec_unmap_first_page(imgp) 860 struct image_params *imgp; 861{ 862 vm_page_t m; 863 864 if (imgp->firstpage != NULL) { 865 m = sf_buf_page(imgp->firstpage); 866 sf_buf_free(imgp->firstpage); 867 imgp->firstpage = NULL; 868 vm_page_lock_queues(); 869 vm_page_unhold(m); 870 vm_page_unlock_queues(); 871 } 872} 873 874/* 875 * Destroy old address space, and allocate a new stack 876 * The new stack is only SGROWSIZ large because it is grown 877 * automatically in trap.c. 878 */ 879int 880exec_new_vmspace(imgp, sv) 881 struct image_params *imgp; 882 struct sysentvec *sv; 883{ 884 int error; 885 struct proc *p = imgp->proc; 886 struct vmspace *vmspace = p->p_vmspace; 887 vm_offset_t stack_addr; 888 vm_map_t map; 889 890 imgp->vmspace_destroyed = 1; 891 892 /* Called with Giant held, do not depend on it! */ 893 EVENTHANDLER_INVOKE(process_exec, p); 894 895 /* 896 * Here is as good a place as any to do any resource limit cleanups. 897 * This is needed if a 64 bit binary exec's a 32 bit binary - the 898 * data size limit may need to be changed to a value that makes 899 * sense for the 32 bit binary. 900 */ 901 if (sv->sv_fixlimits != NULL) 902 sv->sv_fixlimits(imgp); 903 904 /* 905 * Blow away entire process VM, if address space not shared, 906 * otherwise, create a new VM space so that other threads are 907 * not disrupted 908 */ 909 map = &vmspace->vm_map; 910 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser && 911 vm_map_max(map) == sv->sv_maxuser) { 912 shmexit(vmspace); 913 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map), 914 vm_map_max(map)); 915 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 916 } else { 917 vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser); 918 vmspace = p->p_vmspace; 919 map = &vmspace->vm_map; 920 } 921 922 /* Allocate a new stack */ 923 stack_addr = sv->sv_usrstack - maxssiz; 924 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz, 925 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 926 if (error) 927 return (error); 928 929#ifdef __ia64__ 930 /* Allocate a new register stack */ 931 stack_addr = IA64_BACKINGSTORE; 932 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz, 933 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 934 if (error) 935 return (error); 936#endif 937 938 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 939 * VM_STACK case, but they are still used to monitor the size of the 940 * process stack so we can check the stack rlimit. 941 */ 942 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 943 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz; 944 945 return (0); 946} 947 948/* 949 * Copy out argument and environment strings from the old process 950 * address space into the temporary string buffer. 951 */ 952int 953exec_copyin_args(struct image_args *args, char *fname, 954 enum uio_seg segflg, char **argv, char **envv) 955{ 956 char *argp, *envp; 957 int error; 958 size_t length; 959 960 error = 0; 961 962 bzero(args, sizeof(*args)); 963 if (argv == NULL) 964 return (EFAULT); 965 /* 966 * Allocate temporary demand zeroed space for argument and 967 * environment strings: 968 * 969 * o ARG_MAX for argument and environment; 970 * o MAXSHELLCMDLEN for the name of interpreters. 971 */ 972 args->buf = (char *) kmem_alloc_wait(exec_map, 973 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 974 if (args->buf == NULL) 975 return (ENOMEM); 976 args->begin_argv = args->buf; 977 args->endp = args->begin_argv; 978 args->stringspace = ARG_MAX; 979 980 args->fname = args->buf + ARG_MAX; 981 982 /* 983 * Copy the file name. 984 */ 985 error = (segflg == UIO_SYSSPACE) ? 986 copystr(fname, args->fname, PATH_MAX, &length) : 987 copyinstr(fname, args->fname, PATH_MAX, &length); 988 if (error != 0) 989 return (error); 990 991 /* 992 * extract arguments first 993 */ 994 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) { 995 if (argp == (caddr_t) -1) 996 return (EFAULT); 997 if ((error = copyinstr(argp, args->endp, 998 args->stringspace, &length))) { 999 if (error == ENAMETOOLONG) 1000 return (E2BIG); 1001 return (error); 1002 } 1003 args->stringspace -= length; 1004 args->endp += length; 1005 args->argc++; 1006 } 1007 1008 args->begin_envv = args->endp; 1009 1010 /* 1011 * extract environment strings 1012 */ 1013 if (envv) { 1014 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 1015 if (envp == (caddr_t)-1) 1016 return (EFAULT); 1017 if ((error = copyinstr(envp, args->endp, 1018 args->stringspace, &length))) { 1019 if (error == ENAMETOOLONG) 1020 return (E2BIG); 1021 return (error); 1022 } 1023 args->stringspace -= length; 1024 args->endp += length; 1025 args->envc++; 1026 } 1027 } 1028 1029 return (0); 1030} 1031 1032void 1033exec_free_args(struct image_args *args) 1034{ 1035 1036 if (args->buf) { 1037 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, 1038 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN); 1039 args->buf = NULL; 1040 } 1041} 1042 1043/* 1044 * Copy strings out to the new process address space, constructing 1045 * new arg and env vector tables. Return a pointer to the base 1046 * so that it can be used as the initial stack pointer. 1047 */ 1048register_t * 1049exec_copyout_strings(imgp) 1050 struct image_params *imgp; 1051{ 1052 int argc, envc; 1053 char **vectp; 1054 char *stringp, *destp; 1055 register_t *stack_base; 1056 struct ps_strings *arginfo; 1057 struct proc *p; 1058 int szsigcode; 1059 1060 /* 1061 * Calculate string base and vector table pointers. 1062 * Also deal with signal trampoline code for this exec type. 1063 */ 1064 p = imgp->proc; 1065 szsigcode = 0; 1066 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1067 if (p->p_sysent->sv_szsigcode != NULL) 1068 szsigcode = *(p->p_sysent->sv_szsigcode); 1069 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - 1070 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); 1071 1072 /* 1073 * install sigcode 1074 */ 1075 if (szsigcode) 1076 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - 1077 szsigcode), szsigcode); 1078 1079 /* 1080 * If we have a valid auxargs ptr, prepare some room 1081 * on the stack. 1082 */ 1083 if (imgp->auxargs) { 1084 /* 1085 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1086 * lower compatibility. 1087 */ 1088 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1089 (AT_COUNT * 2); 1090 /* 1091 * The '+ 2' is for the null pointers at the end of each of 1092 * the arg and env vector sets,and imgp->auxarg_size is room 1093 * for argument of Runtime loader. 1094 */ 1095 vectp = (char **)(destp - (imgp->args->argc + 1096 imgp->args->envc + 2 + imgp->auxarg_size) * 1097 sizeof(char *)); 1098 1099 } else { 1100 /* 1101 * The '+ 2' is for the null pointers at the end of each of 1102 * the arg and env vector sets 1103 */ 1104 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * 1105 sizeof(char *)); 1106 } 1107 1108 /* 1109 * vectp also becomes our initial stack base 1110 */ 1111 stack_base = (register_t *)vectp; 1112 1113 stringp = imgp->args->begin_argv; 1114 argc = imgp->args->argc; 1115 envc = imgp->args->envc; 1116 1117 /* 1118 * Copy out strings - arguments and environment. 1119 */ 1120 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 1121 1122 /* 1123 * Fill in "ps_strings" struct for ps, w, etc. 1124 */ 1125 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1126 suword(&arginfo->ps_nargvstr, argc); 1127 1128 /* 1129 * Fill in argument portion of vector table. 1130 */ 1131 for (; argc > 0; --argc) { 1132 suword(vectp++, (long)(intptr_t)destp); 1133 while (*stringp++ != 0) 1134 destp++; 1135 destp++; 1136 } 1137 1138 /* a null vector table pointer separates the argp's from the envp's */ 1139 suword(vectp++, 0); 1140 1141 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1142 suword(&arginfo->ps_nenvstr, envc); 1143 1144 /* 1145 * Fill in environment portion of vector table. 1146 */ 1147 for (; envc > 0; --envc) { 1148 suword(vectp++, (long)(intptr_t)destp); 1149 while (*stringp++ != 0) 1150 destp++; 1151 destp++; 1152 } 1153 1154 /* end of vector table is a null pointer */ 1155 suword(vectp, 0); 1156 1157 return (stack_base); 1158} 1159 1160/* 1161 * Check permissions of file to execute. 1162 * Called with imgp->vp locked. 1163 * Return 0 for success or error code on failure. 1164 */ 1165int 1166exec_check_permissions(imgp) 1167 struct image_params *imgp; 1168{ 1169 struct vnode *vp = imgp->vp; 1170 struct vattr *attr = imgp->attr; 1171 struct thread *td; 1172 int error; 1173 1174 td = curthread; /* XXXKSE */ 1175 1176 /* Get file attributes */ 1177 error = VOP_GETATTR(vp, attr, td->td_ucred, td); 1178 if (error) 1179 return (error); 1180 1181#ifdef MAC 1182 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp); 1183 if (error) 1184 return (error); 1185#endif 1186 1187 /* 1188 * 1) Check if file execution is disabled for the filesystem that this 1189 * file resides on. 1190 * 2) Insure that at least one execute bit is on - otherwise root 1191 * will always succeed, and we don't want to happen unless the 1192 * file really is executable. 1193 * 3) Insure that the file is a regular file. 1194 */ 1195 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1196 ((attr->va_mode & 0111) == 0) || 1197 (attr->va_type != VREG)) 1198 return (EACCES); 1199 1200 /* 1201 * Zero length files can't be exec'd 1202 */ 1203 if (attr->va_size == 0) 1204 return (ENOEXEC); 1205 1206 /* 1207 * Check for execute permission to file based on current credentials. 1208 */ 1209 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1210 if (error) 1211 return (error); 1212 1213 /* 1214 * Check number of open-for-writes on the file and deny execution 1215 * if there are any. 1216 */ 1217 if (vp->v_writecount) 1218 return (ETXTBSY); 1219 1220 /* 1221 * Call filesystem specific open routine (which does nothing in the 1222 * general case). 1223 */ 1224 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1); 1225 return (error); 1226} 1227 1228/* 1229 * Exec handler registration 1230 */ 1231int 1232exec_register(execsw_arg) 1233 const struct execsw *execsw_arg; 1234{ 1235 const struct execsw **es, **xs, **newexecsw; 1236 int count = 2; /* New slot and trailing NULL */ 1237 1238 if (execsw) 1239 for (es = execsw; *es; es++) 1240 count++; 1241 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1242 if (newexecsw == NULL) 1243 return (ENOMEM); 1244 xs = newexecsw; 1245 if (execsw) 1246 for (es = execsw; *es; es++) 1247 *xs++ = *es; 1248 *xs++ = execsw_arg; 1249 *xs = NULL; 1250 if (execsw) 1251 free(execsw, M_TEMP); 1252 execsw = newexecsw; 1253 return (0); 1254} 1255 1256int 1257exec_unregister(execsw_arg) 1258 const struct execsw *execsw_arg; 1259{ 1260 const struct execsw **es, **xs, **newexecsw; 1261 int count = 1; 1262 1263 if (execsw == NULL) 1264 panic("unregister with no handlers left?\n"); 1265 1266 for (es = execsw; *es; es++) { 1267 if (*es == execsw_arg) 1268 break; 1269 } 1270 if (*es == NULL) 1271 return (ENOENT); 1272 for (es = execsw; *es; es++) 1273 if (*es != execsw_arg) 1274 count++; 1275 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1276 if (newexecsw == NULL) 1277 return (ENOMEM); 1278 xs = newexecsw; 1279 for (es = execsw; *es; es++) 1280 if (*es != execsw_arg) 1281 *xs++ = *es; 1282 *xs = NULL; 1283 if (execsw) 1284 free(execsw, M_TEMP); 1285 execsw = newexecsw; 1286 return (0); 1287} 1288