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