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