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