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 * $FreeBSD: src/sys/kern/kern_exec.c,v 1.107.2.15 2002/07/30 15:40:46 nectar Exp $ 27 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/sysmsg.h> 32 #include <sys/kernel.h> 33 #include <sys/mount.h> 34 #include <sys/filedesc.h> 35 #include <sys/fcntl.h> 36 #include <sys/acct.h> 37 #include <sys/exec.h> 38 #include <sys/imgact.h> 39 #include <sys/imgact_elf.h> 40 #include <sys/kern_syscall.h> 41 #include <sys/wait.h> 42 #include <sys/malloc.h> 43 #include <sys/proc.h> 44 #include <sys/priv.h> 45 #include <sys/ktrace.h> 46 #include <sys/signalvar.h> 47 #include <sys/pioctl.h> 48 #include <sys/nlookup.h> 49 #include <sys/sysent.h> 50 #include <sys/shm.h> 51 #include <sys/sysctl.h> 52 #include <sys/vnode.h> 53 #include <sys/vmmeter.h> 54 #include <sys/libkern.h> 55 56 #include <cpu/lwbuf.h> 57 58 #include <vm/vm.h> 59 #include <vm/vm_param.h> 60 #include <sys/lock.h> 61 #include <vm/pmap.h> 62 #include <vm/vm_page.h> 63 #include <vm/vm_map.h> 64 #include <vm/vm_kern.h> 65 #include <vm/vm_extern.h> 66 #include <vm/vm_object.h> 67 #include <vm/vnode_pager.h> 68 #include <vm/vm_pager.h> 69 70 #include <sys/reg.h> 71 72 #include <sys/objcache.h> 73 #include <sys/refcount.h> 74 #include <sys/thread2.h> 75 #include <vm/vm_page2.h> 76 77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 78 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 79 80 enum exec_path_segflg { 81 PATH_SYSSPACE, 82 PATH_USERSPACE, 83 }; 84 85 static register_t *exec_copyout_strings(struct image_params *); 86 static int exec_copyin_args(struct image_args *, char *, 87 enum exec_path_segflg, char **, char **); 88 static void exec_free_args(struct image_args *); 89 static void print_execve_args(struct image_args *args); 90 91 /* XXX This should be vm_size_t. */ 92 __read_mostly static u_long ps_strings = PS_STRINGS; 93 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 94 95 /* XXX This should be vm_size_t. */ 96 __read_mostly static u_long usrstack = USRSTACK; 97 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 98 99 __read_mostly u_long ps_arg_cache_limit = PAGE_SIZE / 16; 100 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 101 &ps_arg_cache_limit, 0, ""); 102 103 __read_mostly int ps_argsopen = 1; 104 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 105 106 __read_mostly static int ktrace_suid = 0; 107 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, ""); 108 109 __read_mostly static int debug_execve_args = 0; 110 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 111 0, ""); 112 113 /* 114 * Exec arguments object cache 115 */ 116 __read_mostly static struct objcache *exec_objcache; 117 118 static 119 void 120 exec_objcache_init(void *arg __unused) 121 { 122 int cluster_limit; 123 size_t limsize; 124 125 /* 126 * Maximum number of concurrent execs. This can be limiting on 127 * systems with a lot of cpu cores but it also eats a significant 128 * amount of memory. 129 */ 130 cluster_limit = (ncpus < 16) ? 16 : ncpus; 131 limsize = kmem_lim_size(); 132 if (limsize > 7 * 1024) 133 cluster_limit *= 2; 134 if (limsize > 15 * 1024) 135 cluster_limit *= 2; 136 137 exec_objcache = objcache_create_mbacked( 138 M_EXECARGS, PATH_MAX + ARG_MAX, 139 cluster_limit, 8, 140 NULL, NULL, NULL); 141 } 142 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 143 144 /* 145 * stackgap_random specifies if the stackgap should have a random size added 146 * to it. It must be a power of 2. If non-zero, the stack gap will be 147 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 148 */ 149 __read_mostly static int stackgap_random = 1024; 150 151 static int 152 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 153 { 154 int error, new_val; 155 new_val = stackgap_random; 156 error = sysctl_handle_int(oidp, &new_val, 0, req); 157 if (error != 0 || req->newptr == NULL) 158 return (error); 159 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val))) 160 return (EINVAL); 161 stackgap_random = new_val; 162 163 return(0); 164 } 165 166 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT, 167 0, 0, sysctl_kern_stackgap, "I", 168 "Max random stack gap (power of 2), static gap if negative"); 169 170 static void 171 print_execve_args(struct image_args *args) 172 { 173 char *cp; 174 int ndx; 175 176 cp = args->begin_argv; 177 for (ndx = 0; ndx < args->argc; ndx++) { 178 kprintf("\targv[%d]: %s\n", ndx, cp); 179 while (*cp++ != '\0'); 180 } 181 for (ndx = 0; ndx < args->envc; ndx++) { 182 kprintf("\tenvv[%d]: %s\n", ndx, cp); 183 while (*cp++ != '\0'); 184 } 185 } 186 187 /* 188 * Each of the items is a pointer to a `const struct execsw', hence the 189 * double pointer here. 190 */ 191 __read_mostly static const struct execsw **execsw; 192 193 /* 194 * Replace current vmspace with a new binary. 195 * Returns 0 on success, > 0 on recoverable error (use as errno). 196 * Returns -1 on lethal error which demands killing of the current 197 * process! 198 */ 199 int 200 kern_execve(struct nlookupdata *nd, struct file *fp, 201 struct image_args *args) 202 { 203 static const char *proctitle = "(execve)"; 204 register_t *stack_base; 205 struct thread *td = curthread; 206 struct lwp *lp = td->td_lwp; 207 struct proc *p = td->td_proc; 208 struct vnode *ovp; 209 struct pargs *pa; 210 struct sigacts *ops; 211 struct sigacts *nps; 212 struct image_params image_params, *imgp; 213 struct filedesc *fds; 214 struct nchandle *nch; 215 struct nlookupdata nd_interpreter; 216 struct vattr_lite lva; 217 int error, len, i; 218 int (*img_first) (struct image_params *); 219 220 if (debug_execve_args) { 221 kprintf("%s()\n", __func__); 222 print_execve_args(args); 223 } 224 225 KKASSERT(p); 226 lwkt_gettoken(&p->p_token); 227 imgp = &image_params; 228 229 /* 230 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 231 * no modifications to the process at all until we get there. 232 * 233 * Note that multiple threads may be trying to exec at the same 234 * time. exec_new_vmspace() handles that too. 235 */ 236 237 /* 238 * Initialize part of the common data 239 */ 240 imgp->proc = p; 241 imgp->args = args; 242 imgp->lvap = &lva; 243 imgp->entry_addr = 0; 244 imgp->resident = 0; 245 imgp->vmspace_destroyed = 0; 246 imgp->interpreted = 0; 247 imgp->interpreter_name[0] = 0; 248 imgp->auxargs = NULL; 249 imgp->vp = NULL; 250 imgp->firstpage = NULL; 251 imgp->ps_strings = 0; 252 imgp->execpath = imgp->freepath = NULL; 253 imgp->execpathp = 0; 254 imgp->image_header = NULL; 255 256 interpret: 257 258 if (nd) { 259 /* 260 * Translate the file name to a vnode. Unlock the cache 261 * entry to improve parallelism for programs exec'd in 262 * parallel. 263 */ 264 nch = &nd->nl_nch; 265 nd->nl_flags |= NLC_SHAREDLOCK; 266 if ((error = nlookup(nd)) != 0) 267 goto failed; 268 269 error = cache_vget(nch, nd->nl_cred, LK_SHARED, &imgp->vp); 270 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 271 nd->nl_flags &= ~NLC_NCPISLOCKED; 272 cache_unlock(nch); 273 } else { 274 nch = &fp->f_nchandle; 275 imgp->vp = fp->f_data; 276 error = vget(imgp->vp, LK_SHARED); 277 } 278 if (error) { 279 imgp->vp = NULL; 280 goto failed; 281 } 282 283 /* 284 * Check file permissions (also 'opens' file). 285 * Include also the top level mount in the check. 286 */ 287 error = exec_check_permissions(imgp, nch->mount); 288 if (error) { 289 vn_unlock(imgp->vp); 290 goto failed; 291 } 292 293 error = exec_map_first_page(imgp); 294 vn_unlock(imgp->vp); 295 if (error) 296 goto failed; 297 298 imgp->proc->p_osrel = 0; 299 300 if (debug_execve_args && imgp->interpreted) { 301 kprintf(" target is interpreted -- recursive pass\n"); 302 kprintf(" interpreter: %s\n", imgp->interpreter_name); 303 print_execve_args(args); 304 } 305 306 /* 307 * If the current process has a special image activator it 308 * wants to try first, call it. For example, emulating shell 309 * scripts differently. 310 */ 311 error = -1; 312 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 313 error = img_first(imgp); 314 315 /* 316 * If the vnode has a registered vmspace, exec the vmspace 317 */ 318 if (error == -1 && imgp->vp->v_resident) 319 error = exec_resident_imgact(imgp); 320 321 /* 322 * Loop through the list of image activators, calling each one. 323 * An activator returns -1 if there is no match, 0 on success, 324 * and an error otherwise. 325 */ 326 for (i = 0; error == -1 && execsw[i]; ++i) { 327 if (execsw[i]->ex_imgact == NULL || 328 execsw[i]->ex_imgact == img_first) { 329 continue; 330 } 331 error = (*execsw[i]->ex_imgact)(imgp); 332 } 333 334 if (error) { 335 if (error == -1) 336 error = ENOEXEC; 337 goto failed; 338 } 339 340 /* 341 * Special interpreter operation, cleanup and loop up to try to 342 * activate the interpreter. 343 */ 344 if (imgp->interpreted) { 345 exec_unmap_first_page(imgp); 346 vrele(imgp->vp); 347 imgp->vp = NULL; 348 349 nd = &nd_interpreter; 350 error = nlookup_init(nd, imgp->interpreter_name, 351 UIO_SYSSPACE, NLC_FOLLOW); 352 if (error) 353 goto failed; 354 355 goto interpret; 356 } 357 358 /* 359 * Do the best to calculate the full path to the image file 360 */ 361 if (imgp->auxargs != NULL && 362 ((args->fname != NULL && args->fname[0] == '/') || 363 vn_fullpath(imgp->proc, imgp->vp, &imgp->execpath, 364 &imgp->freepath, 0) != 0)) 365 { 366 imgp->execpath = args->fname; 367 } 368 369 /* 370 * Copy out strings (args and env) and initialize stack base 371 */ 372 stack_base = exec_copyout_strings(imgp); 373 p->p_vmspace->vm_minsaddr = (char *)stack_base; 374 375 /* 376 * If custom stack fixup routine present for this process 377 * let it do the stack setup. If we are running a resident 378 * image there is no auxinfo or other image activator context 379 * so don't try to add fixups to the stack. 380 * 381 * Else stuff argument count as first item on stack 382 */ 383 if (p->p_sysent->sv_fixup && imgp->resident == 0) 384 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 385 else 386 suword64(--stack_base, imgp->args->argc); 387 388 /* 389 * For security and other reasons, the file descriptor table cannot 390 * be shared after an exec. 391 */ 392 if (p->p_fd->fd_refcnt > 1) { 393 if ((error = fdcopy(p, &fds)) != 0) 394 goto failed; 395 396 fdfree(p, fds); 397 } 398 399 /* 400 * For security and other reasons, signal handlers cannot 401 * be shared after an exec. The new proces gets a copy of the old 402 * handlers. In execsigs(), the new process will have its signals 403 * reset. 404 */ 405 ops = p->p_sigacts; 406 if (ops->ps_refcnt > 1) { 407 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK); 408 bcopy(ops, nps, sizeof(*nps)); 409 refcount_init(&nps->ps_refcnt, 1); 410 p->p_sigacts = nps; 411 if (refcount_release(&ops->ps_refcnt)) { 412 kfree(ops, M_SUBPROC); 413 ops = NULL; 414 } 415 } 416 417 /* 418 * Clean up shared pages, the new program will allocate fresh 419 * copies as needed. This is also for security purposes and 420 * to ensure (for example) that things like sys_lpmap->blockallsigs 421 * state is properly reset on exec. 422 */ 423 lwp_userunmap(lp); 424 proc_userunmap(p); 425 426 /* 427 * For security and other reasons virtual kernels cannot be 428 * inherited by an exec. This also allows a virtual kernel 429 * to fork/exec unrelated applications. 430 */ 431 if (p->p_vkernel) 432 vkernel_exit(p); 433 434 /* Stop profiling */ 435 stopprofclock(p); 436 437 /* close files on exec */ 438 fdcloseexec(p); 439 440 /* reset caught signals */ 441 execsigs(p); 442 443 /* name this process */ 444 if (nch->ncp) { 445 len = min(nch->ncp->nc_nlen, MAXCOMLEN); 446 bcopy(nch->ncp->nc_name, p->p_comm, len); 447 } else { 448 len = sizeof(proctitle) - 1; 449 bcopy(proctitle, p->p_comm, len); 450 } 451 p->p_comm[len] = 0; 452 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 453 454 /* 455 * mark as execed, wakeup the process that vforked (if any) and tell 456 * it that it now has its own resources back 457 * 458 * We are using the P_PPWAIT as an interlock so an atomic op is 459 * necessary to synchronize with the parent's cpu. 460 */ 461 p->p_flags |= P_EXEC; 462 if (p->p_pptr && (p->p_flags & P_PPWAIT)) { 463 if (p->p_pptr->p_upmap) 464 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 465 atomic_clear_int(&p->p_flags, P_PPWAIT); 466 wakeup(p->p_pptr); 467 } 468 469 /* 470 * Implement image setuid/setgid. 471 * 472 * Don't honor setuid/setgid if the filesystem prohibits it or if 473 * the process is being traced. 474 */ 475 if ((((lva.va_mode & VSUID) && p->p_ucred->cr_uid != lva.va_uid) || 476 ((lva.va_mode & VSGID) && p->p_ucred->cr_gid != lva.va_gid)) && 477 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 478 (p->p_flags & P_TRACED) == 0) { 479 /* 480 * Turn off syscall tracing for set-id programs, except for 481 * root. Record any set-id flags first to make sure that 482 * we do not regain any tracing during a possible block. 483 */ 484 setsugid(); 485 if (p->p_tracenode && ktrace_suid == 0 && 486 priv_check(td, PRIV_ROOT) != 0) { 487 ktrdestroy(&p->p_tracenode); 488 p->p_traceflag = 0; 489 } 490 491 /* Clear any PROC_PDEATHSIG_CTL setting */ 492 p->p_deathsig = 0; 493 494 /* Close any file descriptors 0..2 that reference procfs */ 495 setugidsafety(p); 496 /* Make sure file descriptors 0..2 are in use. */ 497 error = fdcheckstd(lp); 498 if (error != 0) 499 goto failed; 500 501 /* 502 * Set the new credentials. 503 */ 504 cratom_proc(p); 505 if (lva.va_mode & VSUID) 506 change_euid(lva.va_uid); 507 if (lva.va_mode & VSGID) 508 p->p_ucred->cr_gid = lva.va_gid; 509 510 /* Clear local varsym variables */ 511 varsymset_clean(&p->p_varsymset); 512 } else { 513 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 514 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 515 p->p_flags &= ~P_SUGID; 516 } 517 518 /* 519 * Implement correct POSIX saved-id behavior. 520 */ 521 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 522 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 523 cratom_proc(p); 524 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 525 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 526 } 527 528 /* 529 * Store the vp for use in procfs. Be sure to keep p_textvp 530 * consistent if we block during the switch-over. 531 */ 532 ovp = p->p_textvp; 533 vref(imgp->vp); /* ref new vp */ 534 p->p_textvp = imgp->vp; 535 if (ovp) /* release old vp */ 536 vrele(ovp); 537 538 /* Release old namecache handle to text file */ 539 if (p->p_textnch.ncp) 540 cache_drop(&p->p_textnch); 541 if (nch->mount) 542 cache_copy(nch, &p->p_textnch); 543 544 /* 545 * Notify others that we exec'd, and clear the P_INEXEC flag 546 * as we're now a bona fide freshly-execed process. 547 */ 548 KNOTE(&p->p_klist, NOTE_EXEC); 549 p->p_flags &= ~P_INEXEC; 550 if (p->p_stops) 551 wakeup(&p->p_stype); 552 553 /* 554 * If tracing the process, trap to debugger so breakpoints 555 * can be set before the program executes. 556 */ 557 STOPEVENT(p, S_EXEC, 0); 558 559 if (p->p_flags & P_TRACED) 560 ksignal(p, SIGTRAP); 561 562 /* clear "fork but no exec" flag, as we _are_ execing */ 563 p->p_acflag &= ~AFORK; 564 565 /* Set values passed into the program in registers. */ 566 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 567 imgp->ps_strings); 568 569 /* Set the access time on the vnode */ 570 vn_mark_atime(imgp->vp, td); 571 572 /* 573 * Free any previous argument cache 574 */ 575 pa = p->p_args; 576 p->p_args = NULL; 577 if (pa && refcount_release(&pa->ar_ref)) { 578 kfree(pa, M_PARGS); 579 pa = NULL; 580 } 581 582 /* 583 * Cache arguments if they fit inside our allowance 584 */ 585 i = imgp->args->begin_envv - imgp->args->begin_argv; 586 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 587 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 588 refcount_init(&pa->ar_ref, 1); 589 pa->ar_length = i; 590 bcopy(imgp->args->begin_argv, pa->ar_args, i); 591 KKASSERT(p->p_args == NULL); 592 p->p_args = pa; 593 } 594 595 failed: 596 597 /* 598 * free various allocated resources 599 */ 600 if (imgp->firstpage) 601 exec_unmap_first_page(imgp); 602 if (imgp->vp) 603 vrele(imgp->vp); 604 if (imgp->freepath) 605 kfree(imgp->freepath, M_TEMP); 606 if (nd == &nd_interpreter) 607 nlookup_done(nd); 608 609 if (error == 0) { 610 ++mycpu->gd_cnt.v_exec; 611 lwkt_reltoken(&p->p_token); 612 return (0); 613 } 614 615 /* 616 * we're done here, clear P_INEXEC if we were the ones that 617 * set it. Otherwise if vmspace_destroyed is still set we 618 * raced another thread and that thread is responsible for 619 * clearing it. 620 */ 621 if (imgp->vmspace_destroyed & 2) { 622 p->p_flags &= ~P_INEXEC; 623 if (p->p_stops) 624 wakeup(&p->p_stype); 625 } 626 lwkt_reltoken(&p->p_token); 627 if (imgp->vmspace_destroyed) { 628 /* 629 * Sorry, no more process anymore. exit gracefully. 630 * However we can't die right here, because our 631 * caller might have to clean up, so indicate a 632 * lethal error by returning -1. 633 */ 634 return (-1); 635 } else { 636 return (error); 637 } 638 } 639 640 /* 641 * execve() system call. 642 */ 643 int 644 sys_execve(struct sysmsg *sysmsg, const struct execve_args *uap) 645 { 646 struct nlookupdata nd; 647 struct image_args args; 648 int error; 649 650 bzero(&args, sizeof(args)); 651 652 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 653 if (error == 0) { 654 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 655 uap->argv, uap->envv); 656 } 657 if (error == 0) 658 error = kern_execve(&nd, NULL, &args); 659 nlookup_done(&nd); 660 exec_free_args(&args); 661 662 if (error < 0) { 663 /* We hit a lethal error condition. Let's die now. */ 664 exit1(W_EXITCODE(0, SIGABRT)); 665 /* NOTREACHED */ 666 } 667 668 /* 669 * The syscall result is returned in registers to the new program. 670 * Linux will register %edx as an atexit function and we must be 671 * sure to set it to 0. XXX 672 */ 673 if (error == 0) 674 sysmsg->sysmsg_result64 = 0; 675 676 return (error); 677 } 678 679 /* 680 * fexecve() system call. 681 */ 682 int 683 sys_fexecve(struct sysmsg *sysmsg, const struct fexecve_args *uap) 684 { 685 struct image_args args; 686 struct thread *td = curthread; 687 struct file *fp; 688 char fname[32]; /* "/dev/fd/xxx" */ 689 int error; 690 691 if ((error = holdvnode(td, uap->fd, &fp)) != 0) 692 return (error); 693 694 /* 695 * Require a descriptor opened only with O_RDONLY or O_EXEC. 696 * XXX: missing O_EXEC support 697 */ 698 if ((fp->f_flag & FWRITE) != 0 || (fp->f_flag & FREAD) == 0) { 699 error = EBADF; 700 goto done; 701 } 702 703 /* 704 * The 'fname' argument is required when executing an 705 * interpreted program because the interpreter must know 706 * the script path. Supply it with '/dev/fd/xxx'. 707 */ 708 ksnprintf(fname, sizeof(fname), "/dev/fd/%d", uap->fd); 709 bzero(&args, sizeof(args)); 710 error = exec_copyin_args(&args, fname, PATH_SYSSPACE, 711 uap->argv, uap->envv); 712 if (error == 0) 713 error = kern_execve(NULL, fp, &args); 714 exec_free_args(&args); 715 716 if (error < 0) { 717 /* We hit a lethal error condition. Let's die now. */ 718 exit1(W_EXITCODE(0, SIGABRT)); 719 /* NOTREACHED */ 720 } 721 722 /* 723 * The syscall result is returned in registers to the new program. 724 * Linux will register %edx as an atexit function and we must be 725 * sure to set it to 0. XXX 726 */ 727 if (error == 0) 728 sysmsg->sysmsg_result64 = 0; 729 730 done: 731 fdrop(fp); 732 return (error); 733 } 734 735 int 736 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 737 struct lwbuf **plwb, const char **pdata) 738 { 739 int rv; 740 vm_page_t ma; 741 vm_page_t m; 742 vm_object_t object; 743 744 /* 745 * The file has to be mappable. 746 */ 747 if ((object = imgp->vp->v_object) == NULL) 748 return (EIO); 749 750 if (pageno >= object->size) 751 return (EIO); 752 753 /* 754 * Shortcut using shared locks, improve concurrent execs. 755 */ 756 vm_object_hold_shared(object); 757 m = vm_page_lookup(object, pageno); 758 if (m) { 759 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 760 vm_page_hold(m); 761 vm_page_sleep_busy(m, FALSE, "execpg"); 762 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 763 m->object == object && m->pindex == pageno) { 764 vm_object_drop(object); 765 goto done; 766 } 767 vm_page_unhold(m); 768 } 769 } 770 vm_object_drop(object); 771 772 /* 773 * Do it the hard way 774 */ 775 vm_object_hold(object); 776 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 777 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 778 ma = m; 779 780 /* 781 * get_pages unbusies all the requested pages except the 782 * primary page (at index 0 in this case). The primary 783 * page may have been wired during the pagein (e.g. by 784 * the buffer cache) so vnode_pager_freepage() must be 785 * used to properly release it. 786 */ 787 rv = vm_pager_get_page(object, pageno, &ma, 1); 788 m = vm_page_lookup(object, pageno); 789 790 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 791 if (m) { 792 vm_page_protect(m, VM_PROT_NONE); 793 vnode_pager_freepage(m); 794 } 795 vm_object_drop(object); 796 return EIO; 797 } 798 } 799 vm_page_hold(m); 800 vm_page_wakeup(m); /* unbusy the page */ 801 vm_object_drop(object); 802 803 done: 804 *plwb = lwbuf_alloc(m, *plwb); 805 *pdata = (void *)lwbuf_kva(*plwb); 806 807 return (0); 808 } 809 810 /* 811 * Map the first page of an executable image. 812 * 813 * NOTE: If the mapping fails we have to NULL-out firstpage which may 814 * still be pointing to our supplied lwp structure. 815 */ 816 int 817 exec_map_first_page(struct image_params *imgp) 818 { 819 int err; 820 821 if (imgp->firstpage) 822 exec_unmap_first_page(imgp); 823 824 imgp->firstpage = &imgp->firstpage_cache; 825 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 826 827 if (err) { 828 imgp->firstpage = NULL; 829 return err; 830 } 831 832 return 0; 833 } 834 835 void 836 exec_unmap_page(struct lwbuf *lwb) 837 { 838 vm_page_t m; 839 840 crit_enter(); 841 if (lwb != NULL) { 842 m = lwbuf_page(lwb); 843 lwbuf_free(lwb); 844 vm_page_unhold(m); 845 } 846 crit_exit(); 847 } 848 849 void 850 exec_unmap_first_page(struct image_params *imgp) 851 { 852 exec_unmap_page(imgp->firstpage); 853 imgp->firstpage = NULL; 854 imgp->image_header = NULL; 855 } 856 857 /* 858 * Destroy old address space, and allocate a new stack 859 * The new stack is only SGROWSIZ large because it is grown 860 * automatically in trap.c. 861 * 862 * This is the point of no return. 863 */ 864 int 865 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 866 { 867 struct vmspace *vmspace = imgp->proc->p_vmspace; 868 vm_offset_t stack_addr = USRSTACK - maxssiz; 869 struct lwp *lp; 870 struct proc *p; 871 vm_map_t map; 872 int error; 873 874 /* 875 * Indicate that we cannot gracefully error out any more, kill 876 * any other threads present, and set P_INEXEC to indicate that 877 * we are now messing with the process structure proper. 878 * 879 * If killalllwps() races return an error which coupled with 880 * vmspace_destroyed will cause us to exit. This is what we 881 * want since another thread is patiently waiting for us to exit 882 * in that case. 883 */ 884 lp = curthread->td_lwp; 885 p = lp->lwp_proc; 886 imgp->vmspace_destroyed = 1; 887 888 if (curthread->td_proc->p_nthreads > 1) { 889 error = killalllwps(1); 890 if (error) 891 return (error); 892 } 893 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 894 p->p_flags |= P_INEXEC; 895 896 /* 897 * Tell procfs to release its hold on the process. It 898 * will return EAGAIN. 899 */ 900 if (p->p_stops) 901 wakeup(&p->p_stype); 902 903 /* 904 * After setting P_INEXEC wait for any remaining references to 905 * the process (p) to go away. 906 * 907 * In particular, a vfork/exec sequence will replace p->p_vmspace 908 * and we must interlock anyone trying to access the space (aka 909 * procfs or sys_process.c calling procfs_domem()). 910 * 911 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us. 912 */ 913 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0)); 914 915 /* 916 * Blow away entire process VM, if address space not shared, 917 * otherwise, create a new VM space so that other threads are 918 * not disrupted. If we are execing a resident vmspace we 919 * create a duplicate of it and remap the stack. 920 */ 921 map = &vmspace->vm_map; 922 if (vmcopy) { 923 vmspace_exec(imgp->proc, vmcopy); 924 vmspace = imgp->proc->p_vmspace; 925 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 926 map = &vmspace->vm_map; 927 } else if (vmspace_getrefs(vmspace) == 1) { 928 shmexit(vmspace); 929 pmap_remove_pages(vmspace_pmap(vmspace), 930 0, VM_MAX_USER_ADDRESS); 931 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 932 } else { 933 vmspace_exec(imgp->proc, NULL); 934 vmspace = imgp->proc->p_vmspace; 935 map = &vmspace->vm_map; 936 } 937 938 /* 939 * Really make sure lwp-specific and process-specific mappings 940 * are gone. 941 * 942 * Once we've done that, and because we are the only LWP left, with 943 * no TID-dependent mappings, we can reset the TID to 1 (the RB tree 944 * will remain consistent since it has only one entry). This way 945 * the exec'd program gets a nice deterministic tid of 1. 946 */ 947 lwp_userunmap(lp); 948 proc_userunmap(p); 949 lp->lwp_tid = 1; 950 p->p_lasttid = 1; 951 952 /* 953 * Allocate a new stack, generally make the stack non-executable 954 * but allow the program to adjust that (the program may desire to 955 * use areas of the stack for executable code). 956 */ 957 error = vm_map_stack(&vmspace->vm_map, &stack_addr, (vm_size_t)maxssiz, 958 0, 959 VM_PROT_READ|VM_PROT_WRITE, 960 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, 961 0); 962 if (error) 963 return (error); 964 965 /* 966 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 967 * VM_STACK case, but they are still used to monitor the size of the 968 * process stack so we can check the stack rlimit. 969 */ 970 vmspace->vm_ssize = sgrowsiz; /* in bytes */ 971 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 972 973 return(0); 974 } 975 976 /* 977 * Copy out argument and environment strings from the old process 978 * address space into the temporary string buffer. 979 */ 980 static int 981 exec_copyin_args(struct image_args *args, char *fname, 982 enum exec_path_segflg segflg, char **argv, char **envv) 983 { 984 char *argp, *envp; 985 int error = 0; 986 size_t length; 987 988 args->buf = objcache_get(exec_objcache, M_WAITOK); 989 if (args->buf == NULL) 990 return (ENOMEM); 991 992 args->begin_argv = args->buf; 993 args->endp = args->begin_argv; 994 args->space = ARG_MAX; 995 996 args->fname = args->buf + ARG_MAX; 997 998 /* 999 * Copy the file name. 1000 */ 1001 if (segflg == PATH_SYSSPACE) 1002 error = copystr(fname, args->fname, PATH_MAX, &length); 1003 else 1004 error = copyinstr(fname, args->fname, PATH_MAX, &length); 1005 if (error) 1006 return (error); 1007 1008 /* 1009 * Extract argument strings. argv may not be NULL. The argv 1010 * array is terminated by a NULL entry. We special-case the 1011 * situation where argv[0] is NULL by passing { filename, NULL } 1012 * to the new program to guarentee that the interpreter knows what 1013 * file to open in case we exec an interpreted file. Note that 1014 * a NULL argv[0] terminates the argv[] array. 1015 * 1016 * XXX the special-casing of argv[0] is historical and needs to be 1017 * revisited. 1018 */ 1019 if (argv == NULL) 1020 error = EFAULT; 1021 if (error == 0) { 1022 while ((argp = (caddr_t)(intptr_t) 1023 fuword64((uintptr_t *)argv++)) != NULL) { 1024 if (argp == (caddr_t)-1) { 1025 error = EFAULT; 1026 break; 1027 } 1028 error = copyinstr(argp, args->endp, 1029 args->space, &length); 1030 if (error) { 1031 if (error == ENAMETOOLONG) 1032 error = E2BIG; 1033 break; 1034 } 1035 args->space -= length; 1036 args->endp += length; 1037 args->argc++; 1038 } 1039 if (args->argc == 0 && error == 0) { 1040 length = strlen(args->fname) + 1; 1041 if (length > args->space) { 1042 error = E2BIG; 1043 } else { 1044 bcopy(args->fname, args->endp, length); 1045 args->space -= length; 1046 args->endp += length; 1047 args->argc++; 1048 } 1049 } 1050 } 1051 1052 args->begin_envv = args->endp; 1053 1054 /* 1055 * extract environment strings. envv may be NULL. 1056 */ 1057 if (envv && error == 0) { 1058 while ((envp = (caddr_t)(intptr_t) 1059 fuword64((uintptr_t *)envv++))) { 1060 if (envp == (caddr_t) -1) { 1061 error = EFAULT; 1062 break; 1063 } 1064 error = copyinstr(envp, args->endp, 1065 args->space, &length); 1066 if (error) { 1067 if (error == ENAMETOOLONG) 1068 error = E2BIG; 1069 break; 1070 } 1071 args->space -= length; 1072 args->endp += length; 1073 args->envc++; 1074 } 1075 } 1076 1077 return (error); 1078 } 1079 1080 static void 1081 exec_free_args(struct image_args *args) 1082 { 1083 if (args->buf) { 1084 objcache_put(exec_objcache, args->buf); 1085 args->buf = NULL; 1086 } 1087 } 1088 1089 /* 1090 * Copy strings out to the new process address space, constructing 1091 * new arg and env vector tables. Return a pointer to the base 1092 * so that it can be used as the initial stack pointer. 1093 * 1094 * The format is, roughly: 1095 * 1096 * [argv[]] <-- vectp 1097 * [envp[]] 1098 * [ELF_Auxargs] 1099 * 1100 * [args & env] <-- destp 1101 * [sgap] 1102 * [SPARE_USRSPACE] 1103 * [execpath] 1104 * [szsigcode] RO|NX 1105 * [ps_strings] RO|NX Top of user stack 1106 * 1107 */ 1108 static register_t * 1109 exec_copyout_strings(struct image_params *imgp) 1110 { 1111 int argc, envc, sgap; 1112 int gap; 1113 int argsenvspace; 1114 char **vectp; 1115 char *stringp, *destp, *szsigbase; 1116 register_t *stack_base; 1117 struct ps_strings *arginfo; 1118 size_t execpath_len; 1119 int szsigcode; 1120 1121 /* 1122 * Calculate string base and vector table pointers. 1123 * Also deal with signal trampoline code for this exec type. 1124 */ 1125 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1126 execpath_len = strlen(imgp->execpath) + 1; 1127 else 1128 execpath_len = 0; 1129 arginfo = (struct ps_strings *)PS_STRINGS; 1130 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 1131 1132 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 1133 gap = stackgap_random; 1134 cpu_ccfence(); 1135 if (gap != 0) { 1136 if (gap < 0) 1137 sgap = ALIGN(-gap); 1138 else 1139 sgap = ALIGN(karc4random() & (gap - 1)); 1140 } else { 1141 sgap = 0; 1142 } 1143 1144 /* 1145 * Calculate destp, which points to [args & env] and above. 1146 */ 1147 szsigbase = (char *)(intptr_t) 1148 trunc_page64((intptr_t)arginfo - szsigcode); 1149 szsigbase -= SZSIGCODE_EXTRA_BYTES; 1150 destp = szsigbase - 1151 roundup(execpath_len, sizeof(char *)) - 1152 SPARE_USRSPACE - 1153 sgap - 1154 argsenvspace; 1155 1156 /* 1157 * install sigcode 1158 */ 1159 if (szsigcode) 1160 copyout(imgp->proc->p_sysent->sv_sigcode, szsigbase, szsigcode); 1161 1162 /* 1163 * Copy the image path for the rtld 1164 */ 1165 if (execpath_len) { 1166 imgp->execpathp = (uintptr_t)szsigbase - 1167 roundup(execpath_len, sizeof(char *)); 1168 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 1169 } 1170 1171 /* 1172 * Calculate base for argv[], envp[], and ELF_Auxargs. 1173 */ 1174 vectp = (char **)destp - (AT_COUNT * 2); 1175 vectp -= imgp->args->argc + imgp->args->envc + 2; 1176 1177 stack_base = (register_t *)vectp; 1178 1179 stringp = imgp->args->begin_argv; 1180 argc = imgp->args->argc; 1181 envc = imgp->args->envc; 1182 1183 /* 1184 * Copy out strings - arguments and environment (at destp) 1185 */ 1186 copyout(stringp, destp, ARG_MAX - imgp->args->space); 1187 1188 /* 1189 * Fill in "ps_strings" struct for ps, w, etc. 1190 */ 1191 suword64((void *)&arginfo->ps_argvstr, (uint64_t)(intptr_t)vectp); 1192 suword32((void *)&arginfo->ps_nargvstr, argc); 1193 1194 /* 1195 * Fill in argument portion of vector table. 1196 */ 1197 for (; argc > 0; --argc) { 1198 suword64((void *)vectp++, (uintptr_t)destp); 1199 while (*stringp++ != 0) 1200 destp++; 1201 destp++; 1202 } 1203 1204 /* a null vector table pointer separates the argp's from the envp's */ 1205 suword64((void *)vectp++, 0); 1206 1207 suword64((void *)&arginfo->ps_envstr, (uintptr_t)vectp); 1208 suword32((void *)&arginfo->ps_nenvstr, envc); 1209 1210 /* 1211 * Fill in environment portion of vector table. 1212 */ 1213 for (; envc > 0; --envc) { 1214 suword64((void *)vectp++, (uintptr_t)destp); 1215 while (*stringp++ != 0) 1216 destp++; 1217 destp++; 1218 } 1219 1220 /* end of vector table is a null pointer */ 1221 suword64((void *)vectp, 0); 1222 1223 /* 1224 * Make the signal trampoline executable and read-only. 1225 */ 1226 vm_map_protect(&imgp->proc->p_vmspace->vm_map, 1227 (vm_offset_t)szsigbase, 1228 (vm_offset_t)szsigbase + PAGE_SIZE, 1229 VM_PROT_READ|VM_PROT_EXECUTE, FALSE); 1230 1231 return (stack_base); 1232 } 1233 1234 /* 1235 * Check permissions of file to execute. 1236 * Return 0 for success or error code on failure. 1237 */ 1238 int 1239 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 1240 { 1241 struct proc *p = imgp->proc; 1242 struct vnode *vp = imgp->vp; 1243 struct vattr_lite *lvap = imgp->lvap; 1244 int error; 1245 1246 /* Get file attributes */ 1247 error = VOP_GETATTR_LITE(vp, lvap); 1248 if (error) 1249 return (error); 1250 1251 /* 1252 * 1) Check if file execution is disabled for the filesystem that this 1253 * file resides on. 1254 * 2) Insure that at least one execute bit is on - otherwise root 1255 * will always succeed, and we don't want to happen unless the 1256 * file really is executable. 1257 * 3) Insure that the file is a regular file. 1258 */ 1259 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1260 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1261 ((lvap->va_mode & 0111) == 0) || 1262 (lvap->va_type != VREG)) { 1263 return (EACCES); 1264 } 1265 1266 /* 1267 * Zero length files can't be exec'd 1268 */ 1269 if (lvap->va_size == 0) 1270 return (ENOEXEC); 1271 1272 /* 1273 * Check for execute permission to file based on current credentials. 1274 */ 1275 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1276 if (error) 1277 return (error); 1278 1279 /* 1280 * Check number of open-for-writes on the file and deny execution 1281 * if there are any. 1282 */ 1283 if (vp->v_writecount) 1284 return (ETXTBSY); 1285 1286 /* 1287 * Call filesystem specific open routine, which allows us to read, 1288 * write, and mmap the file. Without the VOP_OPEN we can only 1289 * stat the file. 1290 */ 1291 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1292 if (error) 1293 return (error); 1294 1295 return (0); 1296 } 1297 1298 /* 1299 * Exec handler registration 1300 */ 1301 int 1302 exec_register(const struct execsw *execsw_arg) 1303 { 1304 const struct execsw **es, **xs, **newexecsw; 1305 int count = 2; /* New slot and trailing NULL */ 1306 1307 if (execsw) 1308 for (es = execsw; *es; es++) 1309 count++; 1310 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1311 xs = newexecsw; 1312 if (execsw) 1313 for (es = execsw; *es; es++) 1314 *xs++ = *es; 1315 *xs++ = execsw_arg; 1316 *xs = NULL; 1317 if (execsw) 1318 kfree(execsw, M_TEMP); 1319 execsw = newexecsw; 1320 return 0; 1321 } 1322 1323 int 1324 exec_unregister(const struct execsw *execsw_arg) 1325 { 1326 const struct execsw **es, **xs, **newexecsw; 1327 int count = 1; 1328 1329 if (execsw == NULL) 1330 panic("unregister with no handlers left?"); 1331 1332 for (es = execsw; *es; es++) { 1333 if (*es == execsw_arg) 1334 break; 1335 } 1336 if (*es == NULL) 1337 return ENOENT; 1338 for (es = execsw; *es; es++) 1339 if (*es != execsw_arg) 1340 count++; 1341 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1342 xs = newexecsw; 1343 for (es = execsw; *es; es++) 1344 if (*es != execsw_arg) 1345 *xs++ = *es; 1346 *xs = NULL; 1347 if (execsw) 1348 kfree(execsw, M_TEMP); 1349 execsw = newexecsw; 1350 return 0; 1351 } 1352