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