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 * $DragonFly: src/sys/kern/kern_exec.c,v 1.64 2008/10/26 04:29:19 sephe Exp $ 28 */ 29 30 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/sysproto.h> 33 #include <sys/kernel.h> 34 #include <sys/mount.h> 35 #include <sys/filedesc.h> 36 #include <sys/fcntl.h> 37 #include <sys/acct.h> 38 #include <sys/exec.h> 39 #include <sys/imgact.h> 40 #include <sys/imgact_elf.h> 41 #include <sys/kern_syscall.h> 42 #include <sys/wait.h> 43 #include <sys/malloc.h> 44 #include <sys/proc.h> 45 #include <sys/priv.h> 46 #include <sys/ktrace.h> 47 #include <sys/signalvar.h> 48 #include <sys/pioctl.h> 49 #include <sys/nlookup.h> 50 #include <sys/sysent.h> 51 #include <sys/shm.h> 52 #include <sys/sysctl.h> 53 #include <sys/vnode.h> 54 #include <sys/vmmeter.h> 55 #include <sys/libkern.h> 56 57 #include <cpu/lwbuf.h> 58 59 #include <vm/vm.h> 60 #include <vm/vm_param.h> 61 #include <sys/lock.h> 62 #include <vm/pmap.h> 63 #include <vm/vm_page.h> 64 #include <vm/vm_map.h> 65 #include <vm/vm_kern.h> 66 #include <vm/vm_extern.h> 67 #include <vm/vm_object.h> 68 #include <vm/vnode_pager.h> 69 #include <vm/vm_pager.h> 70 71 #include <sys/user.h> 72 #include <sys/reg.h> 73 74 #include <sys/refcount.h> 75 #include <sys/thread2.h> 76 #include <sys/mplock2.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; 116 117 cluster_limit = 16; /* up to this many objects */ 118 exec_objcache = objcache_create_mbacked( 119 M_EXECARGS, PATH_MAX + ARG_MAX, 120 &cluster_limit, 8, 121 NULL, NULL, NULL); 122 } 123 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 124 125 /* 126 * stackgap_random specifies if the stackgap should have a random size added 127 * to it. It must be a power of 2. If non-zero, the stack gap will be 128 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 129 */ 130 static int stackgap_random = 1024; 131 static int 132 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 133 { 134 int error, new_val; 135 new_val = stackgap_random; 136 error = sysctl_handle_int(oidp, &new_val, 0, req); 137 if (error != 0 || req->newptr == NULL) 138 return (error); 139 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 140 return (EINVAL); 141 stackgap_random = new_val; 142 143 return(0); 144 } 145 146 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 147 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 148 149 void 150 print_execve_args(struct image_args *args) 151 { 152 char *cp; 153 int ndx; 154 155 cp = args->begin_argv; 156 for (ndx = 0; ndx < args->argc; ndx++) { 157 kprintf("\targv[%d]: %s\n", ndx, cp); 158 while (*cp++ != '\0'); 159 } 160 for (ndx = 0; ndx < args->envc; ndx++) { 161 kprintf("\tenvv[%d]: %s\n", ndx, cp); 162 while (*cp++ != '\0'); 163 } 164 } 165 166 /* 167 * Each of the items is a pointer to a `const struct execsw', hence the 168 * double pointer here. 169 */ 170 static const struct execsw **execsw; 171 172 /* 173 * Replace current vmspace with a new binary. 174 * Returns 0 on success, > 0 on recoverable error (use as errno). 175 * Returns -1 on lethal error which demands killing of the current 176 * process! 177 */ 178 int 179 kern_execve(struct nlookupdata *nd, struct image_args *args) 180 { 181 struct thread *td = curthread; 182 struct lwp *lp = td->td_lwp; 183 struct proc *p = td->td_proc; 184 register_t *stack_base; 185 struct pargs *pa; 186 struct sigacts *ops; 187 struct sigacts *nps; 188 int error, len, i; 189 struct image_params image_params, *imgp; 190 struct vattr attr; 191 int (*img_first) (struct image_params *); 192 193 if (debug_execve_args) { 194 kprintf("%s()\n", __func__); 195 print_execve_args(args); 196 } 197 198 KKASSERT(p); 199 lwkt_gettoken(&p->p_token); 200 imgp = &image_params; 201 202 /* 203 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 204 * no modifications to the process at all until we get there. 205 * 206 * Note that multiple threads may be trying to exec at the same 207 * time. exec_new_vmspace() handles that too. 208 */ 209 210 /* 211 * Initialize part of the common data 212 */ 213 imgp->proc = p; 214 imgp->args = args; 215 imgp->attr = &attr; 216 imgp->entry_addr = 0; 217 imgp->resident = 0; 218 imgp->vmspace_destroyed = 0; 219 imgp->interpreted = 0; 220 imgp->interpreter_name[0] = 0; 221 imgp->auxargs = NULL; 222 imgp->vp = NULL; 223 imgp->firstpage = NULL; 224 imgp->ps_strings = 0; 225 imgp->image_header = NULL; 226 227 interpret: 228 229 /* 230 * Translate the file name to a vnode. Unlock the cache entry to 231 * improve parallelism for programs exec'd in parallel. 232 */ 233 if ((error = nlookup(nd)) != 0) 234 goto exec_fail; 235 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 236 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 237 nd->nl_flags &= ~NLC_NCPISLOCKED; 238 cache_unlock(&nd->nl_nch); 239 if (error) 240 goto exec_fail; 241 242 /* 243 * Check file permissions (also 'opens' file). 244 * Include also the top level mount in the check. 245 */ 246 error = exec_check_permissions(imgp, nd->nl_nch.mount); 247 if (error) { 248 vn_unlock(imgp->vp); 249 goto exec_fail_dealloc; 250 } 251 252 error = exec_map_first_page(imgp); 253 vn_unlock(imgp->vp); 254 if (error) 255 goto exec_fail_dealloc; 256 257 if (debug_execve_args && imgp->interpreted) { 258 kprintf(" target is interpreted -- recursive pass\n"); 259 kprintf(" interpreter: %s\n", imgp->interpreter_name); 260 print_execve_args(args); 261 } 262 263 /* 264 * If the current process has a special image activator it 265 * wants to try first, call it. For example, emulating shell 266 * scripts differently. 267 */ 268 error = -1; 269 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 270 error = img_first(imgp); 271 272 /* 273 * If the vnode has a registered vmspace, exec the vmspace 274 */ 275 if (error == -1 && imgp->vp->v_resident) { 276 error = exec_resident_imgact(imgp); 277 } 278 279 /* 280 * Loop through the list of image activators, calling each one. 281 * An activator returns -1 if there is no match, 0 on success, 282 * and an error otherwise. 283 */ 284 for (i = 0; error == -1 && execsw[i]; ++i) { 285 if (execsw[i]->ex_imgact == NULL || 286 execsw[i]->ex_imgact == img_first) { 287 continue; 288 } 289 error = (*execsw[i]->ex_imgact)(imgp); 290 } 291 292 if (error) { 293 if (error == -1) 294 error = ENOEXEC; 295 goto exec_fail_dealloc; 296 } 297 298 /* 299 * Special interpreter operation, cleanup and loop up to try to 300 * activate the interpreter. 301 */ 302 if (imgp->interpreted) { 303 exec_unmap_first_page(imgp); 304 nlookup_done(nd); 305 vrele(imgp->vp); 306 imgp->vp = NULL; 307 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 308 NLC_FOLLOW); 309 if (error) 310 goto exec_fail; 311 goto interpret; 312 } 313 314 /* 315 * Copy out strings (args and env) and initialize stack base 316 */ 317 stack_base = exec_copyout_strings(imgp); 318 p->p_vmspace->vm_minsaddr = (char *)stack_base; 319 320 /* 321 * If custom stack fixup routine present for this process 322 * let it do the stack setup. If we are running a resident 323 * image there is no auxinfo or other image activator context 324 * so don't try to add fixups to the stack. 325 * 326 * Else stuff argument count as first item on stack 327 */ 328 if (p->p_sysent->sv_fixup && imgp->resident == 0) 329 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 330 else 331 suword(--stack_base, imgp->args->argc); 332 333 /* 334 * For security and other reasons, the file descriptor table cannot 335 * be shared after an exec. 336 */ 337 if (p->p_fd->fd_refcnt > 1) { 338 struct filedesc *tmp; 339 340 error = fdcopy(p, &tmp); 341 if (error != 0) 342 goto exec_fail; 343 fdfree(p, tmp); 344 } 345 346 /* 347 * For security and other reasons, signal handlers cannot 348 * be shared after an exec. The new proces gets a copy of the old 349 * handlers. In execsigs(), the new process will have its signals 350 * reset. 351 */ 352 ops = p->p_sigacts; 353 if (ops->ps_refcnt > 1) { 354 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK); 355 bcopy(ops, nps, sizeof(*nps)); 356 refcount_init(&nps->ps_refcnt, 1); 357 p->p_sigacts = nps; 358 if (refcount_release(&ops->ps_refcnt)) { 359 kfree(ops, M_SUBPROC); 360 ops = NULL; 361 } 362 } 363 364 /* 365 * For security and other reasons virtual kernels cannot be 366 * inherited by an exec. This also allows a virtual kernel 367 * to fork/exec unrelated applications. 368 */ 369 if (p->p_vkernel) 370 vkernel_exit(p); 371 372 /* Stop profiling */ 373 stopprofclock(p); 374 375 /* close files on exec */ 376 fdcloseexec(p); 377 378 /* reset caught signals */ 379 execsigs(p); 380 381 /* name this process - nameiexec(p, ndp) */ 382 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 383 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 384 p->p_comm[len] = 0; 385 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 386 387 /* 388 * mark as execed, wakeup the process that vforked (if any) and tell 389 * it that it now has its own resources back 390 */ 391 p->p_flag |= P_EXEC; 392 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 393 p->p_flag &= ~P_PPWAIT; 394 wakeup((caddr_t)p->p_pptr); 395 } 396 397 /* 398 * Implement image setuid/setgid. 399 * 400 * Don't honor setuid/setgid if the filesystem prohibits it or if 401 * the process is being traced. 402 */ 403 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 404 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 405 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 406 (p->p_flag & P_TRACED) == 0) { 407 /* 408 * Turn off syscall tracing for set-id programs, except for 409 * root. Record any set-id flags first to make sure that 410 * we do not regain any tracing during a possible block. 411 */ 412 setsugid(); 413 if (p->p_tracenode && ktrace_suid == 0 && 414 priv_check(td, PRIV_ROOT) != 0) { 415 ktrdestroy(&p->p_tracenode); 416 p->p_traceflag = 0; 417 } 418 /* Close any file descriptors 0..2 that reference procfs */ 419 setugidsafety(p); 420 /* Make sure file descriptors 0..2 are in use. */ 421 error = fdcheckstd(lp); 422 if (error != 0) 423 goto exec_fail_dealloc; 424 /* 425 * Set the new credentials. 426 */ 427 cratom(&p->p_ucred); 428 if (attr.va_mode & VSUID) 429 change_euid(attr.va_uid); 430 if (attr.va_mode & VSGID) 431 p->p_ucred->cr_gid = attr.va_gid; 432 433 /* 434 * Clear local varsym variables 435 */ 436 varsymset_clean(&p->p_varsymset); 437 } else { 438 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 439 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 440 p->p_flag &= ~P_SUGID; 441 } 442 443 /* 444 * Implement correct POSIX saved-id behavior. 445 */ 446 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 447 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 448 cratom(&p->p_ucred); 449 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 450 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 451 } 452 453 /* 454 * Store the vp for use in procfs 455 */ 456 if (p->p_textvp) /* release old reference */ 457 vrele(p->p_textvp); 458 p->p_textvp = imgp->vp; 459 vref(p->p_textvp); 460 461 /* Release old namecache handle to text file */ 462 if (p->p_textnch.ncp) 463 cache_drop(&p->p_textnch); 464 465 if (nd->nl_nch.mount) 466 cache_copy(&nd->nl_nch, &p->p_textnch); 467 468 /* 469 * Notify others that we exec'd, and clear the P_INEXEC flag 470 * as we're now a bona fide freshly-execed process. 471 */ 472 KNOTE(&p->p_klist, NOTE_EXEC); 473 p->p_flag &= ~P_INEXEC; 474 475 /* 476 * If tracing the process, trap to debugger so breakpoints 477 * can be set before the program executes. 478 */ 479 STOPEVENT(p, S_EXEC, 0); 480 481 if (p->p_flag & P_TRACED) 482 ksignal(p, SIGTRAP); 483 484 /* clear "fork but no exec" flag, as we _are_ execing */ 485 p->p_acflag &= ~AFORK; 486 487 /* Set values passed into the program in registers. */ 488 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 489 imgp->ps_strings); 490 491 /* Set the access time on the vnode */ 492 vn_mark_atime(imgp->vp, td); 493 494 /* 495 * Free any previous argument cache 496 */ 497 pa = p->p_args; 498 p->p_args = NULL; 499 if (pa && refcount_release(&pa->ar_ref)) { 500 kfree(pa, M_PARGS); 501 pa = NULL; 502 } 503 504 /* 505 * Cache arguments if they fit inside our allowance 506 */ 507 i = imgp->args->begin_envv - imgp->args->begin_argv; 508 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 509 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 510 refcount_init(&pa->ar_ref, 1); 511 pa->ar_length = i; 512 bcopy(imgp->args->begin_argv, pa->ar_args, i); 513 KKASSERT(p->p_args == NULL); 514 p->p_args = pa; 515 } 516 517 exec_fail_dealloc: 518 519 /* 520 * free various allocated resources 521 */ 522 if (imgp->firstpage) 523 exec_unmap_first_page(imgp); 524 525 if (imgp->vp) { 526 vrele(imgp->vp); 527 imgp->vp = NULL; 528 } 529 530 if (error == 0) { 531 ++mycpu->gd_cnt.v_exec; 532 lwkt_reltoken(&p->p_token); 533 return (0); 534 } 535 536 exec_fail: 537 /* 538 * we're done here, clear P_INEXEC if we were the ones that 539 * set it. Otherwise if vmspace_destroyed is still set we 540 * raced another thread and that thread is responsible for 541 * clearing it. 542 */ 543 if (imgp->vmspace_destroyed & 2) 544 p->p_flag &= ~P_INEXEC; 545 lwkt_reltoken(&p->p_token); 546 if (imgp->vmspace_destroyed) { 547 /* 548 * Sorry, no more process anymore. exit gracefully. 549 * However we can't die right here, because our 550 * caller might have to clean up, so indicate a 551 * lethal error by returning -1. 552 */ 553 return(-1); 554 } else { 555 return(error); 556 } 557 } 558 559 /* 560 * execve() system call. 561 * 562 * MPALMOSTSAFE 563 */ 564 int 565 sys_execve(struct execve_args *uap) 566 { 567 struct nlookupdata nd; 568 struct image_args args; 569 int error; 570 571 bzero(&args, sizeof(args)); 572 573 get_mplock(); 574 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 575 if (error == 0) { 576 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 577 uap->argv, uap->envv); 578 } 579 if (error == 0) 580 error = kern_execve(&nd, &args); 581 nlookup_done(&nd); 582 exec_free_args(&args); 583 584 if (error < 0) { 585 /* We hit a lethal error condition. Let's die now. */ 586 exit1(W_EXITCODE(0, SIGABRT)); 587 /* NOTREACHED */ 588 } 589 rel_mplock(); 590 591 /* 592 * The syscall result is returned in registers to the new program. 593 * Linux will register %edx as an atexit function and we must be 594 * sure to set it to 0. XXX 595 */ 596 if (error == 0) 597 uap->sysmsg_result64 = 0; 598 599 return (error); 600 } 601 602 int 603 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 604 struct lwbuf **plwb, const char **pdata) 605 { 606 int rv; 607 vm_page_t ma; 608 vm_page_t m; 609 vm_object_t object; 610 611 /* 612 * The file has to be mappable. 613 */ 614 if ((object = imgp->vp->v_object) == NULL) 615 return (EIO); 616 617 if (pageno >= object->size) 618 return (EIO); 619 620 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 621 622 lwkt_gettoken(&vm_token); 623 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 624 ma = m; 625 626 /* 627 * get_pages unbusies all the requested pages except the 628 * primary page (at index 0 in this case). The primary 629 * page may have been wired during the pagein (e.g. by 630 * the buffer cache) so vnode_pager_freepage() must be 631 * used to properly release it. 632 */ 633 rv = vm_pager_get_page(object, &ma, 1); 634 m = vm_page_lookup(object, pageno); 635 636 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 637 if (m) { 638 vm_page_protect(m, VM_PROT_NONE); 639 vnode_pager_freepage(m); 640 } 641 lwkt_reltoken(&vm_token); 642 return EIO; 643 } 644 } 645 vm_page_hold(m); /* requires vm_token to be held */ 646 vm_page_wakeup(m); /* unbusy the page */ 647 lwkt_reltoken(&vm_token); 648 649 *plwb = lwbuf_alloc(m, *plwb); 650 *pdata = (void *)lwbuf_kva(*plwb); 651 652 return (0); 653 } 654 655 int 656 exec_map_first_page(struct image_params *imgp) 657 { 658 int err; 659 660 if (imgp->firstpage) 661 exec_unmap_first_page(imgp); 662 663 imgp->firstpage = &imgp->firstpage_cache; 664 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 665 666 if (err) 667 return err; 668 669 return 0; 670 } 671 672 void 673 exec_unmap_page(struct lwbuf *lwb) 674 { 675 vm_page_t m; 676 677 crit_enter(); 678 if (lwb != NULL) { 679 m = lwbuf_page(lwb); 680 lwbuf_free(lwb); 681 vm_page_unhold(m); 682 } 683 crit_exit(); 684 } 685 686 void 687 exec_unmap_first_page(struct image_params *imgp) 688 { 689 exec_unmap_page(imgp->firstpage); 690 imgp->firstpage = NULL; 691 imgp->image_header = NULL; 692 } 693 694 /* 695 * Destroy old address space, and allocate a new stack 696 * The new stack is only SGROWSIZ large because it is grown 697 * automatically in trap.c. 698 * 699 * This is the point of no return. 700 */ 701 int 702 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 703 { 704 struct vmspace *vmspace = imgp->proc->p_vmspace; 705 vm_offset_t stack_addr = USRSTACK - maxssiz; 706 struct proc *p; 707 vm_map_t map; 708 int error; 709 710 /* 711 * Indicate that we cannot gracefully error out any more, kill 712 * any other threads present, and set P_INEXEC to indicate that 713 * we are now messing with the process structure proper. 714 * 715 * If killalllwps() races return an error which coupled with 716 * vmspace_destroyed will cause us to exit. This is what we 717 * want since another thread is patiently waiting for us to exit 718 * in that case. 719 */ 720 p = curproc; 721 imgp->vmspace_destroyed = 1; 722 723 if (curthread->td_proc->p_nthreads > 1) { 724 error = killalllwps(1); 725 if (error) 726 return (error); 727 } 728 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 729 p->p_flag |= P_INEXEC; 730 731 /* 732 * Blow away entire process VM, if address space not shared, 733 * otherwise, create a new VM space so that other threads are 734 * not disrupted. If we are execing a resident vmspace we 735 * create a duplicate of it and remap the stack. 736 * 737 * The exitingcnt test is not strictly necessary but has been 738 * included for code sanity (to make the code more deterministic). 739 */ 740 map = &vmspace->vm_map; 741 if (vmcopy) { 742 vmspace_exec(imgp->proc, vmcopy); 743 vmspace = imgp->proc->p_vmspace; 744 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 745 map = &vmspace->vm_map; 746 } else if (vmspace->vm_sysref.refcnt == 1 && 747 vmspace->vm_exitingcnt == 0) { 748 shmexit(vmspace); 749 if (vmspace->vm_upcalls) 750 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 751 pmap_remove_pages(vmspace_pmap(vmspace), 752 0, VM_MAX_USER_ADDRESS); 753 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 754 } else { 755 vmspace_exec(imgp->proc, NULL); 756 vmspace = imgp->proc->p_vmspace; 757 map = &vmspace->vm_map; 758 } 759 760 /* Allocate a new stack */ 761 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 762 0, VM_PROT_ALL, VM_PROT_ALL, 0); 763 if (error) 764 return (error); 765 766 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 767 * VM_STACK case, but they are still used to monitor the size of the 768 * process stack so we can check the stack rlimit. 769 */ 770 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 771 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 772 773 return(0); 774 } 775 776 /* 777 * Copy out argument and environment strings from the old process 778 * address space into the temporary string buffer. 779 */ 780 int 781 exec_copyin_args(struct image_args *args, char *fname, 782 enum exec_path_segflg segflg, char **argv, char **envv) 783 { 784 char *argp, *envp; 785 int error = 0; 786 size_t length; 787 788 args->buf = objcache_get(exec_objcache, M_WAITOK); 789 if (args->buf == NULL) 790 return (ENOMEM); 791 args->begin_argv = args->buf; 792 args->endp = args->begin_argv; 793 args->space = ARG_MAX; 794 795 args->fname = args->buf + ARG_MAX; 796 797 /* 798 * Copy the file name. 799 */ 800 if (segflg == PATH_SYSSPACE) { 801 error = copystr(fname, args->fname, PATH_MAX, &length); 802 } else if (segflg == PATH_USERSPACE) { 803 error = copyinstr(fname, args->fname, PATH_MAX, &length); 804 } 805 806 /* 807 * Extract argument strings. argv may not be NULL. The argv 808 * array is terminated by a NULL entry. We special-case the 809 * situation where argv[0] is NULL by passing { filename, NULL } 810 * to the new program to guarentee that the interpreter knows what 811 * file to open in case we exec an interpreted file. Note that 812 * a NULL argv[0] terminates the argv[] array. 813 * 814 * XXX the special-casing of argv[0] is historical and needs to be 815 * revisited. 816 */ 817 if (argv == NULL) 818 error = EFAULT; 819 if (error == 0) { 820 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 821 if (argp == (caddr_t)-1) { 822 error = EFAULT; 823 break; 824 } 825 error = copyinstr(argp, args->endp, 826 args->space, &length); 827 if (error) { 828 if (error == ENAMETOOLONG) 829 error = E2BIG; 830 break; 831 } 832 args->space -= length; 833 args->endp += length; 834 args->argc++; 835 } 836 if (args->argc == 0 && error == 0) { 837 length = strlen(args->fname) + 1; 838 if (length > args->space) { 839 error = E2BIG; 840 } else { 841 bcopy(args->fname, args->endp, length); 842 args->space -= length; 843 args->endp += length; 844 args->argc++; 845 } 846 } 847 } 848 849 args->begin_envv = args->endp; 850 851 /* 852 * extract environment strings. envv may be NULL. 853 */ 854 if (envv && error == 0) { 855 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 856 if (envp == (caddr_t) -1) { 857 error = EFAULT; 858 break; 859 } 860 error = copyinstr(envp, args->endp, args->space, 861 &length); 862 if (error) { 863 if (error == ENAMETOOLONG) 864 error = E2BIG; 865 break; 866 } 867 args->space -= length; 868 args->endp += length; 869 args->envc++; 870 } 871 } 872 return (error); 873 } 874 875 void 876 exec_free_args(struct image_args *args) 877 { 878 if (args->buf) { 879 objcache_put(exec_objcache, args->buf); 880 args->buf = NULL; 881 } 882 } 883 884 /* 885 * Copy strings out to the new process address space, constructing 886 * new arg and env vector tables. Return a pointer to the base 887 * so that it can be used as the initial stack pointer. 888 */ 889 register_t * 890 exec_copyout_strings(struct image_params *imgp) 891 { 892 int argc, envc, sgap; 893 char **vectp; 894 char *stringp, *destp; 895 register_t *stack_base; 896 struct ps_strings *arginfo; 897 int szsigcode; 898 899 /* 900 * Calculate string base and vector table pointers. 901 * Also deal with signal trampoline code for this exec type. 902 */ 903 arginfo = (struct ps_strings *)PS_STRINGS; 904 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 905 if (stackgap_random != 0) 906 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 907 else 908 sgap = 0; 909 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 910 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 911 912 /* 913 * install sigcode 914 */ 915 if (szsigcode) 916 copyout(imgp->proc->p_sysent->sv_sigcode, 917 ((caddr_t)arginfo - szsigcode), szsigcode); 918 919 /* 920 * If we have a valid auxargs ptr, prepare some room 921 * on the stack. 922 * 923 * The '+ 2' is for the null pointers at the end of each of the 924 * arg and env vector sets, and 'AT_COUNT*2' is room for the 925 * ELF Auxargs data. 926 */ 927 if (imgp->auxargs) { 928 vectp = (char **)(destp - (imgp->args->argc + 929 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 930 } else { 931 vectp = (char **)(destp - (imgp->args->argc + 932 imgp->args->envc + 2) * sizeof(char*)); 933 } 934 935 /* 936 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 937 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 938 */ 939 940 /* 941 * vectp also becomes our initial stack base 942 */ 943 stack_base = (register_t *)vectp; 944 945 stringp = imgp->args->begin_argv; 946 argc = imgp->args->argc; 947 envc = imgp->args->envc; 948 949 /* 950 * Copy out strings - arguments and environment. 951 */ 952 copyout(stringp, destp, ARG_MAX - imgp->args->space); 953 954 /* 955 * Fill in "ps_strings" struct for ps, w, etc. 956 */ 957 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 958 suword(&arginfo->ps_nargvstr, argc); 959 960 /* 961 * Fill in argument portion of vector table. 962 */ 963 for (; argc > 0; --argc) { 964 suword(vectp++, (long)(intptr_t)destp); 965 while (*stringp++ != 0) 966 destp++; 967 destp++; 968 } 969 970 /* a null vector table pointer separates the argp's from the envp's */ 971 suword(vectp++, 0); 972 973 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 974 suword(&arginfo->ps_nenvstr, envc); 975 976 /* 977 * Fill in environment portion of vector table. 978 */ 979 for (; envc > 0; --envc) { 980 suword(vectp++, (long)(intptr_t)destp); 981 while (*stringp++ != 0) 982 destp++; 983 destp++; 984 } 985 986 /* end of vector table is a null pointer */ 987 suword(vectp, 0); 988 989 return (stack_base); 990 } 991 992 /* 993 * Check permissions of file to execute. 994 * Return 0 for success or error code on failure. 995 */ 996 int 997 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 998 { 999 struct proc *p = imgp->proc; 1000 struct vnode *vp = imgp->vp; 1001 struct vattr *attr = imgp->attr; 1002 int error; 1003 1004 /* Get file attributes */ 1005 error = VOP_GETATTR(vp, attr); 1006 if (error) 1007 return (error); 1008 1009 /* 1010 * 1) Check if file execution is disabled for the filesystem that this 1011 * file resides on. 1012 * 2) Insure that at least one execute bit is on - otherwise root 1013 * will always succeed, and we don't want to happen unless the 1014 * file really is executable. 1015 * 3) Insure that the file is a regular file. 1016 */ 1017 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1018 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1019 ((attr->va_mode & 0111) == 0) || 1020 (attr->va_type != VREG)) { 1021 return (EACCES); 1022 } 1023 1024 /* 1025 * Zero length files can't be exec'd 1026 */ 1027 if (attr->va_size == 0) 1028 return (ENOEXEC); 1029 1030 /* 1031 * Check for execute permission to file based on current credentials. 1032 */ 1033 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1034 if (error) 1035 return (error); 1036 1037 /* 1038 * Check number of open-for-writes on the file and deny execution 1039 * if there are any. 1040 */ 1041 if (vp->v_writecount) 1042 return (ETXTBSY); 1043 1044 /* 1045 * Call filesystem specific open routine, which allows us to read, 1046 * write, and mmap the file. Without the VOP_OPEN we can only 1047 * stat the file. 1048 */ 1049 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1050 if (error) 1051 return (error); 1052 1053 return (0); 1054 } 1055 1056 /* 1057 * Exec handler registration 1058 */ 1059 int 1060 exec_register(const struct execsw *execsw_arg) 1061 { 1062 const struct execsw **es, **xs, **newexecsw; 1063 int count = 2; /* New slot and trailing NULL */ 1064 1065 if (execsw) 1066 for (es = execsw; *es; es++) 1067 count++; 1068 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1069 xs = newexecsw; 1070 if (execsw) 1071 for (es = execsw; *es; es++) 1072 *xs++ = *es; 1073 *xs++ = execsw_arg; 1074 *xs = NULL; 1075 if (execsw) 1076 kfree(execsw, M_TEMP); 1077 execsw = newexecsw; 1078 return 0; 1079 } 1080 1081 int 1082 exec_unregister(const struct execsw *execsw_arg) 1083 { 1084 const struct execsw **es, **xs, **newexecsw; 1085 int count = 1; 1086 1087 if (execsw == NULL) 1088 panic("unregister with no handlers left?"); 1089 1090 for (es = execsw; *es; es++) { 1091 if (*es == execsw_arg) 1092 break; 1093 } 1094 if (*es == NULL) 1095 return ENOENT; 1096 for (es = execsw; *es; es++) 1097 if (*es != execsw_arg) 1098 count++; 1099 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1100 xs = newexecsw; 1101 for (es = execsw; *es; es++) 1102 if (*es != execsw_arg) 1103 *xs++ = *es; 1104 *xs = NULL; 1105 if (execsw) 1106 kfree(execsw, M_TEMP); 1107 execsw = newexecsw; 1108 return 0; 1109 } 1110