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 <sys/mplock2.h> 76 #include <vm/vm_page2.h> 77 78 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 79 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 80 81 static register_t *exec_copyout_strings (struct image_params *); 82 83 /* XXX This should be vm_size_t. */ 84 static u_long ps_strings = PS_STRINGS; 85 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 86 87 /* XXX This should be vm_size_t. */ 88 static u_long usrstack = USRSTACK; 89 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 90 91 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 92 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 93 &ps_arg_cache_limit, 0, ""); 94 95 int ps_argsopen = 1; 96 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 97 98 static int ktrace_suid = 0; 99 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, ""); 100 101 void print_execve_args(struct image_args *args); 102 int debug_execve_args = 0; 103 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 104 0, ""); 105 106 /* 107 * Exec arguments object cache 108 */ 109 static struct objcache *exec_objcache; 110 111 static 112 void 113 exec_objcache_init(void *arg __unused) 114 { 115 int cluster_limit; 116 size_t limsize; 117 118 /* 119 * Maximum number of concurrent execs. This can be limiting on 120 * systems with a lot of cpu cores but it also eats a significant 121 * amount of memory. 122 */ 123 cluster_limit = (ncpus < 16) ? 16 : ncpus; 124 limsize = kmem_lim_size(); 125 if (limsize > 7 * 1024) 126 cluster_limit *= 2; 127 if (limsize > 15 * 1024) 128 cluster_limit *= 2; 129 130 exec_objcache = objcache_create_mbacked( 131 M_EXECARGS, PATH_MAX + ARG_MAX, 132 cluster_limit, 8, 133 NULL, NULL, NULL); 134 } 135 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 136 137 /* 138 * stackgap_random specifies if the stackgap should have a random size added 139 * to it. It must be a power of 2. If non-zero, the stack gap will be 140 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 141 */ 142 static int stackgap_random = 1024; 143 static int 144 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 145 { 146 int error, new_val; 147 new_val = stackgap_random; 148 error = sysctl_handle_int(oidp, &new_val, 0, req); 149 if (error != 0 || req->newptr == NULL) 150 return (error); 151 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val))) 152 return (EINVAL); 153 stackgap_random = new_val; 154 155 return(0); 156 } 157 158 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT, 159 0, 0, sysctl_kern_stackgap, "I", 160 "Max random stack gap (power of 2), static gap if negative"); 161 162 void 163 print_execve_args(struct image_args *args) 164 { 165 char *cp; 166 int ndx; 167 168 cp = args->begin_argv; 169 for (ndx = 0; ndx < args->argc; ndx++) { 170 kprintf("\targv[%d]: %s\n", ndx, cp); 171 while (*cp++ != '\0'); 172 } 173 for (ndx = 0; ndx < args->envc; ndx++) { 174 kprintf("\tenvv[%d]: %s\n", ndx, cp); 175 while (*cp++ != '\0'); 176 } 177 } 178 179 /* 180 * Each of the items is a pointer to a `const struct execsw', hence the 181 * double pointer here. 182 */ 183 static const struct execsw **execsw; 184 185 /* 186 * Replace current vmspace with a new binary. 187 * Returns 0 on success, > 0 on recoverable error (use as errno). 188 * Returns -1 on lethal error which demands killing of the current 189 * process! 190 */ 191 int 192 kern_execve(struct nlookupdata *nd, struct image_args *args) 193 { 194 struct thread *td = curthread; 195 struct lwp *lp = td->td_lwp; 196 struct proc *p = td->td_proc; 197 struct vnode *ovp; 198 register_t *stack_base; 199 struct pargs *pa; 200 struct sigacts *ops; 201 struct sigacts *nps; 202 int error, len, i; 203 struct image_params image_params, *imgp; 204 struct vattr attr; 205 int (*img_first) (struct image_params *); 206 207 if (debug_execve_args) { 208 kprintf("%s()\n", __func__); 209 print_execve_args(args); 210 } 211 212 KKASSERT(p); 213 lwkt_gettoken(&p->p_token); 214 imgp = &image_params; 215 216 /* 217 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 218 * no modifications to the process at all until we get there. 219 * 220 * Note that multiple threads may be trying to exec at the same 221 * time. exec_new_vmspace() handles that too. 222 */ 223 224 /* 225 * Initialize part of the common data 226 */ 227 imgp->proc = p; 228 imgp->args = args; 229 imgp->attr = &attr; 230 imgp->entry_addr = 0; 231 imgp->resident = 0; 232 imgp->vmspace_destroyed = 0; 233 imgp->interpreted = 0; 234 imgp->interpreter_name[0] = 0; 235 imgp->auxargs = NULL; 236 imgp->vp = NULL; 237 imgp->firstpage = NULL; 238 imgp->ps_strings = 0; 239 imgp->execpath = imgp->freepath = NULL; 240 imgp->execpathp = 0; 241 imgp->image_header = NULL; 242 243 interpret: 244 245 /* 246 * Translate the file name to a vnode. Unlock the cache entry to 247 * improve parallelism for programs exec'd in parallel. 248 */ 249 nd->nl_flags |= NLC_SHAREDLOCK; 250 if ((error = nlookup(nd)) != 0) 251 goto exec_fail; 252 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &imgp->vp); 253 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 254 nd->nl_flags &= ~NLC_NCPISLOCKED; 255 cache_unlock(&nd->nl_nch); 256 if (error) 257 goto exec_fail; 258 259 /* 260 * Check file permissions (also 'opens' file). 261 * Include also the top level mount in the check. 262 */ 263 error = exec_check_permissions(imgp, nd->nl_nch.mount); 264 if (error) { 265 vn_unlock(imgp->vp); 266 goto exec_fail_dealloc; 267 } 268 269 error = exec_map_first_page(imgp); 270 vn_unlock(imgp->vp); 271 if (error) 272 goto exec_fail_dealloc; 273 274 imgp->proc->p_osrel = 0; 275 276 if (debug_execve_args && imgp->interpreted) { 277 kprintf(" target is interpreted -- recursive pass\n"); 278 kprintf(" interpreter: %s\n", imgp->interpreter_name); 279 print_execve_args(args); 280 } 281 282 /* 283 * If the current process has a special image activator it 284 * wants to try first, call it. For example, emulating shell 285 * scripts differently. 286 */ 287 error = -1; 288 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 289 error = img_first(imgp); 290 291 /* 292 * If the vnode has a registered vmspace, exec the vmspace 293 */ 294 if (error == -1 && imgp->vp->v_resident) { 295 error = exec_resident_imgact(imgp); 296 } 297 298 /* 299 * Loop through the list of image activators, calling each one. 300 * An activator returns -1 if there is no match, 0 on success, 301 * and an error otherwise. 302 */ 303 for (i = 0; error == -1 && execsw[i]; ++i) { 304 if (execsw[i]->ex_imgact == NULL || 305 execsw[i]->ex_imgact == img_first) { 306 continue; 307 } 308 error = (*execsw[i]->ex_imgact)(imgp); 309 } 310 311 if (error) { 312 if (error == -1) 313 error = ENOEXEC; 314 goto exec_fail_dealloc; 315 } 316 317 /* 318 * Special interpreter operation, cleanup and loop up to try to 319 * activate the interpreter. 320 */ 321 if (imgp->interpreted) { 322 exec_unmap_first_page(imgp); 323 nlookup_done(nd); 324 vrele(imgp->vp); 325 imgp->vp = NULL; 326 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 327 NLC_FOLLOW); 328 if (error) 329 goto exec_fail; 330 goto interpret; 331 } 332 333 /* 334 * Do the best to calculate the full path to the image file 335 */ 336 if (imgp->auxargs != NULL && 337 ((args->fname != NULL && args->fname[0] == '/') || 338 vn_fullpath(imgp->proc, 339 imgp->vp, 340 &imgp->execpath, 341 &imgp->freepath, 342 0) != 0)) 343 imgp->execpath = args->fname; 344 345 /* 346 * Copy out strings (args and env) and initialize stack base 347 */ 348 stack_base = exec_copyout_strings(imgp); 349 p->p_vmspace->vm_minsaddr = (char *)stack_base; 350 351 /* 352 * If custom stack fixup routine present for this process 353 * let it do the stack setup. If we are running a resident 354 * image there is no auxinfo or other image activator context 355 * so don't try to add fixups to the stack. 356 * 357 * Else stuff argument count as first item on stack 358 */ 359 if (p->p_sysent->sv_fixup && imgp->resident == 0) 360 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 361 else 362 suword(--stack_base, imgp->args->argc); 363 364 /* 365 * For security and other reasons, the file descriptor table cannot 366 * be shared after an exec. 367 */ 368 if (p->p_fd->fd_refcnt > 1) { 369 struct filedesc *tmp; 370 371 error = fdcopy(p, &tmp); 372 if (error != 0) 373 goto exec_fail; 374 fdfree(p, tmp); 375 } 376 377 /* 378 * For security and other reasons, signal handlers cannot 379 * be shared after an exec. The new proces gets a copy of the old 380 * handlers. In execsigs(), the new process will have its signals 381 * reset. 382 */ 383 ops = p->p_sigacts; 384 if (ops->ps_refcnt > 1) { 385 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK); 386 bcopy(ops, nps, sizeof(*nps)); 387 refcount_init(&nps->ps_refcnt, 1); 388 p->p_sigacts = nps; 389 if (refcount_release(&ops->ps_refcnt)) { 390 kfree(ops, M_SUBPROC); 391 ops = NULL; 392 } 393 } 394 395 /* 396 * For security and other reasons virtual kernels cannot be 397 * inherited by an exec. This also allows a virtual kernel 398 * to fork/exec unrelated applications. 399 */ 400 if (p->p_vkernel) 401 vkernel_exit(p); 402 403 /* Stop profiling */ 404 stopprofclock(p); 405 406 /* close files on exec */ 407 fdcloseexec(p); 408 409 /* reset caught signals */ 410 execsigs(p); 411 412 /* name this process - nameiexec(p, ndp) */ 413 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 414 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 415 p->p_comm[len] = 0; 416 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 417 418 /* 419 * mark as execed, wakeup the process that vforked (if any) and tell 420 * it that it now has its own resources back 421 * 422 * We are using the P_PPWAIT as an interlock so an atomic op is 423 * necessary to synchronize with the parent's cpu. 424 */ 425 p->p_flags |= P_EXEC; 426 if (p->p_pptr && (p->p_flags & P_PPWAIT)) { 427 atomic_clear_int(&p->p_flags, P_PPWAIT); 428 wakeup(p->p_pptr); 429 } 430 431 /* 432 * Implement image setuid/setgid. 433 * 434 * Don't honor setuid/setgid if the filesystem prohibits it or if 435 * the process is being traced. 436 */ 437 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 438 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 439 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 440 (p->p_flags & P_TRACED) == 0) { 441 /* 442 * Turn off syscall tracing for set-id programs, except for 443 * root. Record any set-id flags first to make sure that 444 * we do not regain any tracing during a possible block. 445 */ 446 setsugid(); 447 if (p->p_tracenode && ktrace_suid == 0 && 448 priv_check(td, PRIV_ROOT) != 0) { 449 ktrdestroy(&p->p_tracenode); 450 p->p_traceflag = 0; 451 } 452 /* Close any file descriptors 0..2 that reference procfs */ 453 setugidsafety(p); 454 /* Make sure file descriptors 0..2 are in use. */ 455 error = fdcheckstd(lp); 456 if (error != 0) 457 goto exec_fail_dealloc; 458 /* 459 * Set the new credentials. 460 */ 461 cratom_proc(p); 462 if (attr.va_mode & VSUID) 463 change_euid(attr.va_uid); 464 if (attr.va_mode & VSGID) 465 p->p_ucred->cr_gid = attr.va_gid; 466 467 /* 468 * Clear local varsym variables 469 */ 470 varsymset_clean(&p->p_varsymset); 471 } else { 472 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 473 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 474 p->p_flags &= ~P_SUGID; 475 } 476 477 /* 478 * Implement correct POSIX saved-id behavior. 479 */ 480 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 481 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 482 cratom_proc(p); 483 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 484 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 485 } 486 487 /* 488 * Store the vp for use in procfs. Be sure to keep p_textvp 489 * consistent if we block during the switch-over. 490 */ 491 ovp = p->p_textvp; 492 vref(imgp->vp); /* ref new vp */ 493 p->p_textvp = imgp->vp; 494 if (ovp) /* release old vp */ 495 vrele(ovp); 496 497 /* Release old namecache handle to text file */ 498 if (p->p_textnch.ncp) 499 cache_drop(&p->p_textnch); 500 501 if (nd->nl_nch.mount) 502 cache_copy(&nd->nl_nch, &p->p_textnch); 503 504 /* 505 * Notify others that we exec'd, and clear the P_INEXEC flag 506 * as we're now a bona fide freshly-execed process. 507 */ 508 KNOTE(&p->p_klist, NOTE_EXEC); 509 p->p_flags &= ~P_INEXEC; 510 if (p->p_stops) 511 wakeup(&p->p_stype); 512 513 /* 514 * If tracing the process, trap to debugger so breakpoints 515 * can be set before the program executes. 516 */ 517 STOPEVENT(p, S_EXEC, 0); 518 519 if (p->p_flags & P_TRACED) 520 ksignal(p, SIGTRAP); 521 522 /* clear "fork but no exec" flag, as we _are_ execing */ 523 p->p_acflag &= ~AFORK; 524 525 /* Set values passed into the program in registers. */ 526 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 527 imgp->ps_strings); 528 529 /* Set the access time on the vnode */ 530 vn_mark_atime(imgp->vp, td); 531 532 /* 533 * Free any previous argument cache 534 */ 535 pa = p->p_args; 536 p->p_args = NULL; 537 if (pa && refcount_release(&pa->ar_ref)) { 538 kfree(pa, M_PARGS); 539 pa = NULL; 540 } 541 542 /* 543 * Cache arguments if they fit inside our allowance 544 */ 545 i = imgp->args->begin_envv - imgp->args->begin_argv; 546 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 547 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 548 refcount_init(&pa->ar_ref, 1); 549 pa->ar_length = i; 550 bcopy(imgp->args->begin_argv, pa->ar_args, i); 551 KKASSERT(p->p_args == NULL); 552 p->p_args = pa; 553 } 554 555 exec_fail_dealloc: 556 557 /* 558 * free various allocated resources 559 */ 560 if (imgp->firstpage) 561 exec_unmap_first_page(imgp); 562 563 if (imgp->vp) { 564 vrele(imgp->vp); 565 imgp->vp = NULL; 566 } 567 568 if (imgp->freepath) 569 kfree(imgp->freepath, M_TEMP); 570 571 if (error == 0) { 572 ++mycpu->gd_cnt.v_exec; 573 lwkt_reltoken(&p->p_token); 574 return (0); 575 } 576 577 exec_fail: 578 /* 579 * we're done here, clear P_INEXEC if we were the ones that 580 * set it. Otherwise if vmspace_destroyed is still set we 581 * raced another thread and that thread is responsible for 582 * clearing it. 583 */ 584 if (imgp->vmspace_destroyed & 2) { 585 p->p_flags &= ~P_INEXEC; 586 if (p->p_stops) 587 wakeup(&p->p_stype); 588 } 589 lwkt_reltoken(&p->p_token); 590 if (imgp->vmspace_destroyed) { 591 /* 592 * Sorry, no more process anymore. exit gracefully. 593 * However we can't die right here, because our 594 * caller might have to clean up, so indicate a 595 * lethal error by returning -1. 596 */ 597 return(-1); 598 } else { 599 return(error); 600 } 601 } 602 603 /* 604 * execve() system call. 605 */ 606 int 607 sys_execve(struct execve_args *uap) 608 { 609 struct nlookupdata nd; 610 struct image_args args; 611 int error; 612 613 bzero(&args, sizeof(args)); 614 615 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 616 if (error == 0) { 617 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 618 uap->argv, uap->envv); 619 } 620 if (error == 0) 621 error = kern_execve(&nd, &args); 622 nlookup_done(&nd); 623 exec_free_args(&args); 624 625 if (error < 0) { 626 /* We hit a lethal error condition. Let's die now. */ 627 exit1(W_EXITCODE(0, SIGABRT)); 628 /* NOTREACHED */ 629 } 630 631 /* 632 * The syscall result is returned in registers to the new program. 633 * Linux will register %edx as an atexit function and we must be 634 * sure to set it to 0. XXX 635 */ 636 if (error == 0) 637 uap->sysmsg_result64 = 0; 638 639 return (error); 640 } 641 642 int 643 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 644 struct lwbuf **plwb, const char **pdata) 645 { 646 int rv; 647 vm_page_t ma; 648 vm_page_t m; 649 vm_object_t object; 650 651 /* 652 * The file has to be mappable. 653 */ 654 if ((object = imgp->vp->v_object) == NULL) 655 return (EIO); 656 657 if (pageno >= object->size) 658 return (EIO); 659 660 /* 661 * Shortcut using shared locks, improve concurrent execs. 662 */ 663 vm_object_hold_shared(object); 664 m = vm_page_lookup(object, pageno); 665 if (m) { 666 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 667 vm_page_hold(m); 668 vm_page_sleep_busy(m, FALSE, "execpg"); 669 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 670 vm_object_drop(object); 671 goto done; 672 } 673 } 674 vm_page_unhold(m); 675 } 676 vm_object_drop(object); 677 678 /* 679 * Do it the hard way 680 */ 681 vm_object_hold(object); 682 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 683 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 684 ma = m; 685 686 /* 687 * get_pages unbusies all the requested pages except the 688 * primary page (at index 0 in this case). The primary 689 * page may have been wired during the pagein (e.g. by 690 * the buffer cache) so vnode_pager_freepage() must be 691 * used to properly release it. 692 */ 693 rv = vm_pager_get_page(object, &ma, 1); 694 m = vm_page_lookup(object, pageno); 695 696 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 697 if (m) { 698 vm_page_protect(m, VM_PROT_NONE); 699 vnode_pager_freepage(m); 700 } 701 vm_object_drop(object); 702 return EIO; 703 } 704 } 705 vm_page_hold(m); 706 vm_page_wakeup(m); /* unbusy the page */ 707 vm_object_drop(object); 708 709 done: 710 *plwb = lwbuf_alloc(m, *plwb); 711 *pdata = (void *)lwbuf_kva(*plwb); 712 713 return (0); 714 } 715 716 /* 717 * Map the first page of an executable image. 718 * 719 * NOTE: If the mapping fails we have to NULL-out firstpage which may 720 * still be pointing to our supplied lwp structure. 721 */ 722 int 723 exec_map_first_page(struct image_params *imgp) 724 { 725 int err; 726 727 if (imgp->firstpage) 728 exec_unmap_first_page(imgp); 729 730 imgp->firstpage = &imgp->firstpage_cache; 731 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 732 733 if (err) { 734 imgp->firstpage = NULL; 735 return err; 736 } 737 738 return 0; 739 } 740 741 void 742 exec_unmap_page(struct lwbuf *lwb) 743 { 744 vm_page_t m; 745 746 crit_enter(); 747 if (lwb != NULL) { 748 m = lwbuf_page(lwb); 749 lwbuf_free(lwb); 750 vm_page_unhold(m); 751 } 752 crit_exit(); 753 } 754 755 void 756 exec_unmap_first_page(struct image_params *imgp) 757 { 758 exec_unmap_page(imgp->firstpage); 759 imgp->firstpage = NULL; 760 imgp->image_header = NULL; 761 } 762 763 /* 764 * Destroy old address space, and allocate a new stack 765 * The new stack is only SGROWSIZ large because it is grown 766 * automatically in trap.c. 767 * 768 * This is the point of no return. 769 */ 770 int 771 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 772 { 773 struct vmspace *vmspace = imgp->proc->p_vmspace; 774 vm_offset_t stack_addr = USRSTACK - maxssiz; 775 struct proc *p; 776 vm_map_t map; 777 int error; 778 779 /* 780 * Indicate that we cannot gracefully error out any more, kill 781 * any other threads present, and set P_INEXEC to indicate that 782 * we are now messing with the process structure proper. 783 * 784 * If killalllwps() races return an error which coupled with 785 * vmspace_destroyed will cause us to exit. This is what we 786 * want since another thread is patiently waiting for us to exit 787 * in that case. 788 */ 789 p = curproc; 790 imgp->vmspace_destroyed = 1; 791 792 if (curthread->td_proc->p_nthreads > 1) { 793 error = killalllwps(1); 794 if (error) 795 return (error); 796 } 797 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 798 p->p_flags |= P_INEXEC; 799 800 /* 801 * Tell procfs to release its hold on the process. It 802 * will return EAGAIN. 803 */ 804 if (p->p_stops) 805 wakeup(&p->p_stype); 806 807 /* 808 * After setting P_INEXEC wait for any remaining references to 809 * the process (p) to go away. 810 * 811 * In particular, a vfork/exec sequence will replace p->p_vmspace 812 * and we must interlock anyone trying to access the space (aka 813 * procfs or sys_process.c calling procfs_domem()). 814 * 815 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us. 816 */ 817 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0)); 818 819 /* 820 * Blow away entire process VM, if address space not shared, 821 * otherwise, create a new VM space so that other threads are 822 * not disrupted. If we are execing a resident vmspace we 823 * create a duplicate of it and remap the stack. 824 */ 825 map = &vmspace->vm_map; 826 if (vmcopy) { 827 vmspace_exec(imgp->proc, vmcopy); 828 vmspace = imgp->proc->p_vmspace; 829 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 830 map = &vmspace->vm_map; 831 } else if (vmspace->vm_sysref.refcnt == 1) { 832 shmexit(vmspace); 833 pmap_remove_pages(vmspace_pmap(vmspace), 834 0, VM_MAX_USER_ADDRESS); 835 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 836 } else { 837 vmspace_exec(imgp->proc, NULL); 838 vmspace = imgp->proc->p_vmspace; 839 map = &vmspace->vm_map; 840 } 841 842 /* Allocate a new stack */ 843 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 844 0, VM_PROT_ALL, VM_PROT_ALL, 0); 845 if (error) 846 return (error); 847 848 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 849 * VM_STACK case, but they are still used to monitor the size of the 850 * process stack so we can check the stack rlimit. 851 */ 852 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 853 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 854 855 return(0); 856 } 857 858 /* 859 * Copy out argument and environment strings from the old process 860 * address space into the temporary string buffer. 861 */ 862 int 863 exec_copyin_args(struct image_args *args, char *fname, 864 enum exec_path_segflg segflg, char **argv, char **envv) 865 { 866 char *argp, *envp; 867 int error = 0; 868 size_t length; 869 870 args->buf = objcache_get(exec_objcache, M_WAITOK); 871 if (args->buf == NULL) 872 return (ENOMEM); 873 args->begin_argv = args->buf; 874 args->endp = args->begin_argv; 875 args->space = ARG_MAX; 876 877 args->fname = args->buf + ARG_MAX; 878 879 /* 880 * Copy the file name. 881 */ 882 if (segflg == PATH_SYSSPACE) { 883 error = copystr(fname, args->fname, PATH_MAX, &length); 884 } else if (segflg == PATH_USERSPACE) { 885 error = copyinstr(fname, args->fname, PATH_MAX, &length); 886 } 887 888 /* 889 * Extract argument strings. argv may not be NULL. The argv 890 * array is terminated by a NULL entry. We special-case the 891 * situation where argv[0] is NULL by passing { filename, NULL } 892 * to the new program to guarentee that the interpreter knows what 893 * file to open in case we exec an interpreted file. Note that 894 * a NULL argv[0] terminates the argv[] array. 895 * 896 * XXX the special-casing of argv[0] is historical and needs to be 897 * revisited. 898 */ 899 if (argv == NULL) 900 error = EFAULT; 901 if (error == 0) { 902 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 903 if (argp == (caddr_t)-1) { 904 error = EFAULT; 905 break; 906 } 907 error = copyinstr(argp, args->endp, 908 args->space, &length); 909 if (error) { 910 if (error == ENAMETOOLONG) 911 error = E2BIG; 912 break; 913 } 914 args->space -= length; 915 args->endp += length; 916 args->argc++; 917 } 918 if (args->argc == 0 && error == 0) { 919 length = strlen(args->fname) + 1; 920 if (length > args->space) { 921 error = E2BIG; 922 } else { 923 bcopy(args->fname, args->endp, length); 924 args->space -= length; 925 args->endp += length; 926 args->argc++; 927 } 928 } 929 } 930 931 args->begin_envv = args->endp; 932 933 /* 934 * extract environment strings. envv may be NULL. 935 */ 936 if (envv && error == 0) { 937 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 938 if (envp == (caddr_t) -1) { 939 error = EFAULT; 940 break; 941 } 942 error = copyinstr(envp, args->endp, 943 args->space, &length); 944 if (error) { 945 if (error == ENAMETOOLONG) 946 error = E2BIG; 947 break; 948 } 949 args->space -= length; 950 args->endp += length; 951 args->envc++; 952 } 953 } 954 return (error); 955 } 956 957 void 958 exec_free_args(struct image_args *args) 959 { 960 if (args->buf) { 961 objcache_put(exec_objcache, args->buf); 962 args->buf = NULL; 963 } 964 } 965 966 /* 967 * Copy strings out to the new process address space, constructing 968 * new arg and env vector tables. Return a pointer to the base 969 * so that it can be used as the initial stack pointer. 970 * 971 * The format is, roughly: 972 * 973 * [argv[]] <-- vectp 974 * [envp[]] 975 * [ELF_Auxargs] 976 * 977 * [args & env] <-- destp 978 * [sgap] 979 * [SPARE_USRSPACE] 980 * [execpath] 981 * [szsigcode] 982 * [ps_strings] top of user stack 983 * 984 */ 985 register_t * 986 exec_copyout_strings(struct image_params *imgp) 987 { 988 int argc, envc, sgap; 989 int gap; 990 int argsenvspace; 991 char **vectp; 992 char *stringp, *destp; 993 register_t *stack_base; 994 struct ps_strings *arginfo; 995 size_t execpath_len; 996 int szsigcode; 997 998 /* 999 * Calculate string base and vector table pointers. 1000 * Also deal with signal trampoline code for this exec type. 1001 */ 1002 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1003 execpath_len = strlen(imgp->execpath) + 1; 1004 else 1005 execpath_len = 0; 1006 arginfo = (struct ps_strings *)PS_STRINGS; 1007 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 1008 1009 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 1010 gap = stackgap_random; 1011 cpu_ccfence(); 1012 if (gap != 0) { 1013 if (gap < 0) 1014 sgap = ALIGN(-gap); 1015 else 1016 sgap = ALIGN(karc4random() & (gap - 1)); 1017 } else { 1018 sgap = 0; 1019 } 1020 1021 /* 1022 * Calculate destp, which points to [args & env] and above. 1023 */ 1024 destp = (caddr_t)arginfo - 1025 szsigcode - 1026 roundup(execpath_len, sizeof(char *)) - 1027 SPARE_USRSPACE - 1028 sgap - 1029 argsenvspace; 1030 1031 /* 1032 * install sigcode 1033 */ 1034 if (szsigcode) { 1035 copyout(imgp->proc->p_sysent->sv_sigcode, 1036 ((caddr_t)arginfo - szsigcode), szsigcode); 1037 } 1038 1039 /* 1040 * Copy the image path for the rtld 1041 */ 1042 if (execpath_len) { 1043 imgp->execpathp = (uintptr_t)arginfo 1044 - szsigcode 1045 - roundup(execpath_len, sizeof(char *)); 1046 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 1047 } 1048 1049 /* 1050 * Calculate base for argv[], envp[], and ELF_Auxargs. 1051 */ 1052 vectp = (char **)destp - (AT_COUNT * 2); 1053 vectp -= imgp->args->argc + imgp->args->envc + 2; 1054 1055 stack_base = (register_t *)vectp; 1056 1057 stringp = imgp->args->begin_argv; 1058 argc = imgp->args->argc; 1059 envc = imgp->args->envc; 1060 1061 /* 1062 * Copy out strings - arguments and environment (at destp) 1063 */ 1064 copyout(stringp, destp, ARG_MAX - imgp->args->space); 1065 1066 /* 1067 * Fill in "ps_strings" struct for ps, w, etc. 1068 */ 1069 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1070 suword32(&arginfo->ps_nargvstr, argc); 1071 1072 /* 1073 * Fill in argument portion of vector table. 1074 */ 1075 for (; argc > 0; --argc) { 1076 suword(vectp++, (long)(intptr_t)destp); 1077 while (*stringp++ != 0) 1078 destp++; 1079 destp++; 1080 } 1081 1082 /* a null vector table pointer separates the argp's from the envp's */ 1083 suword(vectp++, 0); 1084 1085 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1086 suword32(&arginfo->ps_nenvstr, envc); 1087 1088 /* 1089 * Fill in environment portion of vector table. 1090 */ 1091 for (; envc > 0; --envc) { 1092 suword(vectp++, (long)(intptr_t)destp); 1093 while (*stringp++ != 0) 1094 destp++; 1095 destp++; 1096 } 1097 1098 /* end of vector table is a null pointer */ 1099 suword(vectp, 0); 1100 1101 return (stack_base); 1102 } 1103 1104 /* 1105 * Check permissions of file to execute. 1106 * Return 0 for success or error code on failure. 1107 */ 1108 int 1109 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 1110 { 1111 struct proc *p = imgp->proc; 1112 struct vnode *vp = imgp->vp; 1113 struct vattr *attr = imgp->attr; 1114 int error; 1115 1116 /* Get file attributes */ 1117 error = VOP_GETATTR(vp, attr); 1118 if (error) 1119 return (error); 1120 1121 /* 1122 * 1) Check if file execution is disabled for the filesystem that this 1123 * file resides on. 1124 * 2) Insure that at least one execute bit is on - otherwise root 1125 * will always succeed, and we don't want to happen unless the 1126 * file really is executable. 1127 * 3) Insure that the file is a regular file. 1128 */ 1129 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1130 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1131 ((attr->va_mode & 0111) == 0) || 1132 (attr->va_type != VREG)) { 1133 return (EACCES); 1134 } 1135 1136 /* 1137 * Zero length files can't be exec'd 1138 */ 1139 if (attr->va_size == 0) 1140 return (ENOEXEC); 1141 1142 /* 1143 * Check for execute permission to file based on current credentials. 1144 */ 1145 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1146 if (error) 1147 return (error); 1148 1149 /* 1150 * Check number of open-for-writes on the file and deny execution 1151 * if there are any. 1152 */ 1153 if (vp->v_writecount) 1154 return (ETXTBSY); 1155 1156 /* 1157 * Call filesystem specific open routine, which allows us to read, 1158 * write, and mmap the file. Without the VOP_OPEN we can only 1159 * stat the file. 1160 */ 1161 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1162 if (error) 1163 return (error); 1164 1165 return (0); 1166 } 1167 1168 /* 1169 * Exec handler registration 1170 */ 1171 int 1172 exec_register(const struct execsw *execsw_arg) 1173 { 1174 const struct execsw **es, **xs, **newexecsw; 1175 int count = 2; /* New slot and trailing NULL */ 1176 1177 if (execsw) 1178 for (es = execsw; *es; es++) 1179 count++; 1180 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1181 xs = newexecsw; 1182 if (execsw) 1183 for (es = execsw; *es; es++) 1184 *xs++ = *es; 1185 *xs++ = execsw_arg; 1186 *xs = NULL; 1187 if (execsw) 1188 kfree(execsw, M_TEMP); 1189 execsw = newexecsw; 1190 return 0; 1191 } 1192 1193 int 1194 exec_unregister(const struct execsw *execsw_arg) 1195 { 1196 const struct execsw **es, **xs, **newexecsw; 1197 int count = 1; 1198 1199 if (execsw == NULL) 1200 panic("unregister with no handlers left?"); 1201 1202 for (es = execsw; *es; es++) { 1203 if (*es == execsw_arg) 1204 break; 1205 } 1206 if (*es == NULL) 1207 return ENOENT; 1208 for (es = execsw; *es; es++) 1209 if (*es != execsw_arg) 1210 count++; 1211 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1212 xs = newexecsw; 1213 for (es = execsw; *es; es++) 1214 if (*es != execsw_arg) 1215 *xs++ = *es; 1216 *xs = NULL; 1217 if (execsw) 1218 kfree(execsw, M_TEMP); 1219 execsw = newexecsw; 1220 return 0; 1221 } 1222