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/sfbuf.h> 51 #include <sys/sysent.h> 52 #include <sys/shm.h> 53 #include <sys/sysctl.h> 54 #include <sys/vnode.h> 55 #include <sys/vmmeter.h> 56 #include <sys/aio.h> 57 #include <sys/libkern.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/thread2.h> 75 76 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 77 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 78 79 static register_t *exec_copyout_strings (struct image_params *); 80 81 /* XXX This should be vm_size_t. */ 82 static u_long ps_strings = PS_STRINGS; 83 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 84 85 /* XXX This should be vm_size_t. */ 86 static u_long usrstack = USRSTACK; 87 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 88 89 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 90 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 91 &ps_arg_cache_limit, 0, ""); 92 93 int ps_argsopen = 1; 94 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 95 96 void print_execve_args(struct image_args *args); 97 int debug_execve_args = 0; 98 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 99 0, ""); 100 101 /* 102 * Exec arguments object cache 103 */ 104 static struct objcache *exec_objcache; 105 106 static 107 void 108 exec_objcache_init(void *arg __unused) 109 { 110 int cluster_limit; 111 112 cluster_limit = 16; /* up to this many objects */ 113 exec_objcache = objcache_create_mbacked( 114 M_EXECARGS, PATH_MAX + ARG_MAX, 115 &cluster_limit, 116 2, /* minimal magazine capacity */ 117 NULL, NULL, NULL); 118 } 119 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 120 121 /* 122 * stackgap_random specifies if the stackgap should have a random size added 123 * to it. It must be a power of 2. If non-zero, the stack gap will be 124 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 125 */ 126 static int stackgap_random = 1024; 127 static int 128 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 129 { 130 int error, new_val; 131 new_val = stackgap_random; 132 error = sysctl_handle_int(oidp, &new_val, 0, req); 133 if (error != 0 || req->newptr == NULL) 134 return (error); 135 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 136 return (EINVAL); 137 stackgap_random = new_val; 138 139 return(0); 140 } 141 142 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 143 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 144 145 void 146 print_execve_args(struct image_args *args) 147 { 148 char *cp; 149 int ndx; 150 151 cp = args->begin_argv; 152 for (ndx = 0; ndx < args->argc; ndx++) { 153 kprintf("\targv[%d]: %s\n", ndx, cp); 154 while (*cp++ != '\0'); 155 } 156 for (ndx = 0; ndx < args->envc; ndx++) { 157 kprintf("\tenvv[%d]: %s\n", ndx, cp); 158 while (*cp++ != '\0'); 159 } 160 } 161 162 /* 163 * Each of the items is a pointer to a `const struct execsw', hence the 164 * double pointer here. 165 */ 166 static const struct execsw **execsw; 167 168 /* 169 * Replace current vmspace with a new binary. 170 * Returns 0 on success, > 0 on recoverable error (use as errno). 171 * Returns -1 on lethal error which demands killing of the current 172 * process! 173 */ 174 int 175 kern_execve(struct nlookupdata *nd, struct image_args *args) 176 { 177 struct thread *td = curthread; 178 struct lwp *lp = td->td_lwp; 179 struct proc *p = td->td_proc; 180 register_t *stack_base; 181 int error, len, i; 182 struct image_params image_params, *imgp; 183 struct vattr attr; 184 int (*img_first) (struct image_params *); 185 186 if (debug_execve_args) { 187 kprintf("%s()\n", __func__); 188 print_execve_args(args); 189 } 190 191 KKASSERT(p); 192 imgp = &image_params; 193 194 /* 195 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 196 * no modifications to the process at all until we get there. 197 * 198 * Note that multiple threads may be trying to exec at the same 199 * time. exec_new_vmspace() handles that too. 200 */ 201 202 /* 203 * Initialize part of the common data 204 */ 205 imgp->proc = p; 206 imgp->args = args; 207 imgp->attr = &attr; 208 imgp->entry_addr = 0; 209 imgp->resident = 0; 210 imgp->vmspace_destroyed = 0; 211 imgp->interpreted = 0; 212 imgp->interpreter_name[0] = 0; 213 imgp->auxargs = NULL; 214 imgp->vp = NULL; 215 imgp->firstpage = NULL; 216 imgp->ps_strings = 0; 217 imgp->image_header = NULL; 218 219 interpret: 220 221 /* 222 * Translate the file name to a vnode. Unlock the cache entry to 223 * improve parallelism for programs exec'd in parallel. 224 */ 225 if ((error = nlookup(nd)) != 0) 226 goto exec_fail; 227 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 228 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 229 nd->nl_flags &= ~NLC_NCPISLOCKED; 230 cache_unlock(&nd->nl_nch); 231 if (error) 232 goto exec_fail; 233 234 /* 235 * Check file permissions (also 'opens' file) 236 */ 237 error = exec_check_permissions(imgp); 238 if (error) { 239 vn_unlock(imgp->vp); 240 goto exec_fail_dealloc; 241 } 242 243 error = exec_map_first_page(imgp); 244 vn_unlock(imgp->vp); 245 if (error) 246 goto exec_fail_dealloc; 247 248 if (debug_execve_args && imgp->interpreted) { 249 kprintf(" target is interpreted -- recursive pass\n"); 250 kprintf(" interpreter: %s\n", imgp->interpreter_name); 251 print_execve_args(args); 252 } 253 254 /* 255 * If the current process has a special image activator it 256 * wants to try first, call it. For example, emulating shell 257 * scripts differently. 258 */ 259 error = -1; 260 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 261 error = img_first(imgp); 262 263 /* 264 * If the vnode has a registered vmspace, exec the vmspace 265 */ 266 if (error == -1 && imgp->vp->v_resident) { 267 error = exec_resident_imgact(imgp); 268 } 269 270 /* 271 * Loop through the list of image activators, calling each one. 272 * An activator returns -1 if there is no match, 0 on success, 273 * and an error otherwise. 274 */ 275 for (i = 0; error == -1 && execsw[i]; ++i) { 276 if (execsw[i]->ex_imgact == NULL || 277 execsw[i]->ex_imgact == img_first) { 278 continue; 279 } 280 error = (*execsw[i]->ex_imgact)(imgp); 281 } 282 283 if (error) { 284 if (error == -1) 285 error = ENOEXEC; 286 goto exec_fail_dealloc; 287 } 288 289 /* 290 * Special interpreter operation, cleanup and loop up to try to 291 * activate the interpreter. 292 */ 293 if (imgp->interpreted) { 294 exec_unmap_first_page(imgp); 295 nlookup_done(nd); 296 vrele(imgp->vp); 297 imgp->vp = NULL; 298 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 299 NLC_FOLLOW); 300 if (error) 301 goto exec_fail; 302 goto interpret; 303 } 304 305 /* 306 * Copy out strings (args and env) and initialize stack base 307 */ 308 stack_base = exec_copyout_strings(imgp); 309 p->p_vmspace->vm_minsaddr = (char *)stack_base; 310 311 /* 312 * If custom stack fixup routine present for this process 313 * let it do the stack setup. If we are running a resident 314 * image there is no auxinfo or other image activator context 315 * so don't try to add fixups to the stack. 316 * 317 * Else stuff argument count as first item on stack 318 */ 319 if (p->p_sysent->sv_fixup && imgp->resident == 0) 320 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 321 else 322 suword(--stack_base, imgp->args->argc); 323 324 /* 325 * For security and other reasons, the file descriptor table cannot 326 * be shared after an exec. 327 */ 328 if (p->p_fd->fd_refcnt > 1) { 329 struct filedesc *tmp; 330 331 tmp = fdcopy(p); 332 fdfree(p, tmp); 333 } 334 335 /* 336 * For security and other reasons, signal handlers cannot 337 * be shared after an exec. The new proces gets a copy of the old 338 * handlers. In execsigs(), the new process will have its signals 339 * reset. 340 */ 341 if (p->p_sigacts->ps_refcnt > 1) { 342 struct sigacts *newsigacts; 343 344 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts), 345 M_SUBPROC, M_WAITOK); 346 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts)); 347 p->p_sigacts->ps_refcnt--; 348 p->p_sigacts = newsigacts; 349 p->p_sigacts->ps_refcnt = 1; 350 } 351 352 /* 353 * For security and other reasons virtual kernels cannot be 354 * inherited by an exec. This also allows a virtual kernel 355 * to fork/exec unrelated applications. 356 */ 357 if (p->p_vkernel) 358 vkernel_exit(p); 359 360 /* Stop profiling */ 361 stopprofclock(p); 362 363 /* close files on exec */ 364 fdcloseexec(p); 365 366 /* reset caught signals */ 367 execsigs(p); 368 369 /* name this process - nameiexec(p, ndp) */ 370 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 371 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 372 p->p_comm[len] = 0; 373 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 374 375 /* 376 * mark as execed, wakeup the process that vforked (if any) and tell 377 * it that it now has its own resources back 378 */ 379 p->p_flag |= P_EXEC; 380 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 381 p->p_flag &= ~P_PPWAIT; 382 wakeup((caddr_t)p->p_pptr); 383 } 384 385 /* 386 * Implement image setuid/setgid. 387 * 388 * Don't honor setuid/setgid if the filesystem prohibits it or if 389 * the process is being traced. 390 */ 391 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 392 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 393 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 394 (p->p_flag & P_TRACED) == 0) { 395 /* 396 * Turn off syscall tracing for set-id programs, except for 397 * root. Record any set-id flags first to make sure that 398 * we do not regain any tracing during a possible block. 399 */ 400 setsugid(); 401 if (p->p_tracenode && priv_check(td, PRIV_ROOT) != 0) { 402 ktrdestroy(&p->p_tracenode); 403 p->p_traceflag = 0; 404 } 405 /* Close any file descriptors 0..2 that reference procfs */ 406 setugidsafety(p); 407 /* Make sure file descriptors 0..2 are in use. */ 408 error = fdcheckstd(p); 409 if (error != 0) 410 goto exec_fail_dealloc; 411 /* 412 * Set the new credentials. 413 */ 414 cratom(&p->p_ucred); 415 if (attr.va_mode & VSUID) 416 change_euid(attr.va_uid); 417 if (attr.va_mode & VSGID) 418 p->p_ucred->cr_gid = attr.va_gid; 419 420 /* 421 * Clear local varsym variables 422 */ 423 varsymset_clean(&p->p_varsymset); 424 } else { 425 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 426 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 427 p->p_flag &= ~P_SUGID; 428 } 429 430 /* 431 * Implement correct POSIX saved-id behavior. 432 */ 433 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 434 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 435 cratom(&p->p_ucred); 436 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 437 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 438 } 439 440 /* 441 * Store the vp for use in procfs 442 */ 443 if (p->p_textvp) /* release old reference */ 444 vrele(p->p_textvp); 445 p->p_textvp = imgp->vp; 446 vref(p->p_textvp); 447 448 /* 449 * Notify others that we exec'd, and clear the P_INEXEC flag 450 * as we're now a bona fide freshly-execed process. 451 */ 452 KNOTE(&p->p_klist, NOTE_EXEC); 453 p->p_flag &= ~P_INEXEC; 454 455 /* 456 * If tracing the process, trap to debugger so breakpoints 457 * can be set before the program executes. 458 */ 459 STOPEVENT(p, S_EXEC, 0); 460 461 if (p->p_flag & P_TRACED) 462 ksignal(p, SIGTRAP); 463 464 /* clear "fork but no exec" flag, as we _are_ execing */ 465 p->p_acflag &= ~AFORK; 466 467 /* Set values passed into the program in registers. */ 468 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 469 imgp->ps_strings); 470 471 /* Set the access time on the vnode */ 472 vn_mark_atime(imgp->vp, td); 473 474 /* Free any previous argument cache */ 475 if (p->p_args && --p->p_args->ar_ref == 0) 476 FREE(p->p_args, M_PARGS); 477 p->p_args = NULL; 478 479 /* Cache arguments if they fit inside our allowance */ 480 i = imgp->args->begin_envv - imgp->args->begin_argv; 481 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 482 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 483 M_PARGS, M_WAITOK); 484 p->p_args->ar_ref = 1; 485 p->p_args->ar_length = i; 486 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 487 } 488 489 exec_fail_dealloc: 490 491 /* 492 * free various allocated resources 493 */ 494 if (imgp->firstpage) 495 exec_unmap_first_page(imgp); 496 497 if (imgp->vp) { 498 vrele(imgp->vp); 499 imgp->vp = NULL; 500 } 501 502 if (error == 0) { 503 ++mycpu->gd_cnt.v_exec; 504 return (0); 505 } 506 507 exec_fail: 508 /* 509 * we're done here, clear P_INEXEC if we were the ones that 510 * set it. Otherwise if vmspace_destroyed is still set we 511 * raced another thread and that thread is responsible for 512 * clearing it. 513 */ 514 if (imgp->vmspace_destroyed & 2) 515 p->p_flag &= ~P_INEXEC; 516 if (imgp->vmspace_destroyed) { 517 /* 518 * Sorry, no more process anymore. exit gracefully. 519 * However we can't die right here, because our 520 * caller might have to clean up, so indicate a 521 * lethal error by returning -1. 522 */ 523 return(-1); 524 } else { 525 return(error); 526 } 527 } 528 529 /* 530 * execve() system call. 531 */ 532 int 533 sys_execve(struct execve_args *uap) 534 { 535 struct nlookupdata nd; 536 struct image_args args; 537 int error; 538 539 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 540 if (error == 0) { 541 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 542 uap->argv, uap->envv); 543 } 544 if (error == 0) 545 error = kern_execve(&nd, &args); 546 nlookup_done(&nd); 547 exec_free_args(&args); 548 549 if (error < 0) { 550 /* We hit a lethal error condition. Let's die now. */ 551 exit1(W_EXITCODE(0, SIGABRT)); 552 /* NOTREACHED */ 553 } 554 555 /* 556 * The syscall result is returned in registers to the new program. 557 * Linux will register %edx as an atexit function and we must be 558 * sure to set it to 0. XXX 559 */ 560 if (error == 0) 561 uap->sysmsg_result64 = 0; 562 563 return (error); 564 } 565 566 int 567 exec_map_first_page(struct image_params *imgp) 568 { 569 int rv, i; 570 int initial_pagein; 571 vm_page_t ma[VM_INITIAL_PAGEIN]; 572 vm_page_t m; 573 vm_object_t object; 574 575 if (imgp->firstpage) 576 exec_unmap_first_page(imgp); 577 578 /* 579 * The file has to be mappable. 580 */ 581 if ((object = imgp->vp->v_object) == NULL) 582 return (EIO); 583 584 /* 585 * We shouldn't need protection for vm_page_grab() but we certainly 586 * need it for the lookup loop below (lookup/busy race), since 587 * an interrupt can unbusy and free the page before our busy check. 588 */ 589 crit_enter(); 590 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 591 592 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 593 ma[0] = m; 594 initial_pagein = VM_INITIAL_PAGEIN; 595 if (initial_pagein > object->size) 596 initial_pagein = object->size; 597 for (i = 1; i < initial_pagein; i++) { 598 if ((m = vm_page_lookup(object, i)) != NULL) { 599 if ((m->flags & PG_BUSY) || m->busy) 600 break; 601 if (m->valid) 602 break; 603 vm_page_busy(m); 604 } else { 605 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 606 if (m == NULL) 607 break; 608 } 609 ma[i] = m; 610 } 611 initial_pagein = i; 612 613 /* 614 * get_pages unbusies all the requested pages except the 615 * primary page (at index 0 in this case). The primary 616 * page may have been wired during the pagein (e.g. by 617 * the buffer cache) so vnode_pager_freepage() must be 618 * used to properly release it. 619 */ 620 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 621 m = vm_page_lookup(object, 0); 622 623 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 624 if (m) { 625 vm_page_protect(m, VM_PROT_NONE); 626 vnode_pager_freepage(m); 627 } 628 crit_exit(); 629 return EIO; 630 } 631 } 632 vm_page_hold(m); 633 vm_page_wakeup(m); /* unbusy the page */ 634 crit_exit(); 635 636 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 637 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 638 639 return 0; 640 } 641 642 void 643 exec_unmap_first_page(struct image_params *imgp) 644 { 645 vm_page_t m; 646 647 crit_enter(); 648 if (imgp->firstpage != NULL) { 649 m = sf_buf_page(imgp->firstpage); 650 sf_buf_free(imgp->firstpage); 651 imgp->firstpage = NULL; 652 imgp->image_header = NULL; 653 vm_page_unhold(m); 654 } 655 crit_exit(); 656 } 657 658 /* 659 * Destroy old address space, and allocate a new stack 660 * The new stack is only SGROWSIZ large because it is grown 661 * automatically in trap.c. 662 * 663 * This is the point of no return. 664 */ 665 int 666 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 667 { 668 struct vmspace *vmspace = imgp->proc->p_vmspace; 669 vm_offset_t stack_addr = USRSTACK - maxssiz; 670 struct proc *p; 671 vm_map_t map; 672 int error; 673 674 /* 675 * Indicate that we cannot gracefully error out any more, kill 676 * any other threads present, and set P_INEXEC to indicate that 677 * we are now messing with the process structure proper. 678 * 679 * If killalllwps() races return an error which coupled with 680 * vmspace_destroyed will cause us to exit. This is what we 681 * want since another thread is patiently waiting for us to exit 682 * in that case. 683 */ 684 p = curproc; 685 imgp->vmspace_destroyed = 1; 686 687 if (curthread->td_proc->p_nthreads > 1) { 688 error = killalllwps(1); 689 if (error) 690 return (error); 691 } 692 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 693 p->p_flag |= P_INEXEC; 694 695 /* 696 * Prevent a pending AIO from modifying the new address space. 697 */ 698 aio_proc_rundown(imgp->proc); 699 700 /* 701 * Blow away entire process VM, if address space not shared, 702 * otherwise, create a new VM space so that other threads are 703 * not disrupted. If we are execing a resident vmspace we 704 * create a duplicate of it and remap the stack. 705 * 706 * The exitingcnt test is not strictly necessary but has been 707 * included for code sanity (to make the code more deterministic). 708 */ 709 map = &vmspace->vm_map; 710 if (vmcopy) { 711 vmspace_exec(imgp->proc, vmcopy); 712 vmspace = imgp->proc->p_vmspace; 713 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 714 map = &vmspace->vm_map; 715 } else if (vmspace->vm_sysref.refcnt == 1 && 716 vmspace->vm_exitingcnt == 0) { 717 shmexit(vmspace); 718 if (vmspace->vm_upcalls) 719 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 720 pmap_remove_pages(vmspace_pmap(vmspace), 721 0, VM_MAX_USER_ADDRESS); 722 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 723 } else { 724 vmspace_exec(imgp->proc, NULL); 725 vmspace = imgp->proc->p_vmspace; 726 map = &vmspace->vm_map; 727 } 728 729 /* Allocate a new stack */ 730 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 731 0, VM_PROT_ALL, VM_PROT_ALL, 0); 732 if (error) 733 return (error); 734 735 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 736 * VM_STACK case, but they are still used to monitor the size of the 737 * process stack so we can check the stack rlimit. 738 */ 739 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 740 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 741 742 return(0); 743 } 744 745 /* 746 * Copy out argument and environment strings from the old process 747 * address space into the temporary string buffer. 748 */ 749 int 750 exec_copyin_args(struct image_args *args, char *fname, 751 enum exec_path_segflg segflg, char **argv, char **envv) 752 { 753 char *argp, *envp; 754 int error = 0; 755 size_t length; 756 757 bzero(args, sizeof(*args)); 758 759 args->buf = objcache_get(exec_objcache, M_WAITOK); 760 if (args->buf == NULL) 761 return (ENOMEM); 762 args->begin_argv = args->buf; 763 args->endp = args->begin_argv; 764 args->space = ARG_MAX; 765 766 args->fname = args->buf + ARG_MAX; 767 768 /* 769 * Copy the file name. 770 */ 771 if (segflg == PATH_SYSSPACE) { 772 error = copystr(fname, args->fname, PATH_MAX, &length); 773 } else if (segflg == PATH_USERSPACE) { 774 error = copyinstr(fname, args->fname, PATH_MAX, &length); 775 } 776 777 /* 778 * Extract argument strings. argv may not be NULL. The argv 779 * array is terminated by a NULL entry. We special-case the 780 * situation where argv[0] is NULL by passing { filename, NULL } 781 * to the new program to guarentee that the interpreter knows what 782 * file to open in case we exec an interpreted file. Note that 783 * a NULL argv[0] terminates the argv[] array. 784 * 785 * XXX the special-casing of argv[0] is historical and needs to be 786 * revisited. 787 */ 788 if (argv == NULL) 789 error = EFAULT; 790 if (error == 0) { 791 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 792 if (argp == (caddr_t)-1) { 793 error = EFAULT; 794 break; 795 } 796 error = copyinstr(argp, args->endp, 797 args->space, &length); 798 if (error) { 799 if (error == ENAMETOOLONG) 800 error = E2BIG; 801 break; 802 } 803 args->space -= length; 804 args->endp += length; 805 args->argc++; 806 } 807 if (args->argc == 0 && error == 0) { 808 length = strlen(args->fname) + 1; 809 if (length > args->space) { 810 error = E2BIG; 811 } else { 812 bcopy(args->fname, args->endp, length); 813 args->space -= length; 814 args->endp += length; 815 args->argc++; 816 } 817 } 818 } 819 820 args->begin_envv = args->endp; 821 822 /* 823 * extract environment strings. envv may be NULL. 824 */ 825 if (envv && error == 0) { 826 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 827 if (envp == (caddr_t) -1) { 828 error = EFAULT; 829 break; 830 } 831 error = copyinstr(envp, args->endp, args->space, 832 &length); 833 if (error) { 834 if (error == ENAMETOOLONG) 835 error = E2BIG; 836 break; 837 } 838 args->space -= length; 839 args->endp += length; 840 args->envc++; 841 } 842 } 843 return (error); 844 } 845 846 void 847 exec_free_args(struct image_args *args) 848 { 849 if (args->buf) { 850 objcache_put(exec_objcache, args->buf); 851 args->buf = NULL; 852 } 853 } 854 855 /* 856 * Copy strings out to the new process address space, constructing 857 * new arg and env vector tables. Return a pointer to the base 858 * so that it can be used as the initial stack pointer. 859 */ 860 register_t * 861 exec_copyout_strings(struct image_params *imgp) 862 { 863 int argc, envc, sgap; 864 char **vectp; 865 char *stringp, *destp; 866 register_t *stack_base; 867 struct ps_strings *arginfo; 868 int szsigcode; 869 870 /* 871 * Calculate string base and vector table pointers. 872 * Also deal with signal trampoline code for this exec type. 873 */ 874 arginfo = (struct ps_strings *)PS_STRINGS; 875 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 876 if (stackgap_random != 0) 877 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 878 else 879 sgap = 0; 880 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 881 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 882 883 /* 884 * install sigcode 885 */ 886 if (szsigcode) 887 copyout(imgp->proc->p_sysent->sv_sigcode, 888 ((caddr_t)arginfo - szsigcode), szsigcode); 889 890 /* 891 * If we have a valid auxargs ptr, prepare some room 892 * on the stack. 893 * 894 * The '+ 2' is for the null pointers at the end of each of the 895 * arg and env vector sets, and 'AT_COUNT*2' is room for the 896 * ELF Auxargs data. 897 */ 898 if (imgp->auxargs) { 899 vectp = (char **)(destp - (imgp->args->argc + 900 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 901 } else { 902 vectp = (char **)(destp - (imgp->args->argc + 903 imgp->args->envc + 2) * sizeof(char*)); 904 } 905 906 /* 907 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 908 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 909 */ 910 911 /* 912 * vectp also becomes our initial stack base 913 */ 914 stack_base = (register_t *)vectp; 915 916 stringp = imgp->args->begin_argv; 917 argc = imgp->args->argc; 918 envc = imgp->args->envc; 919 920 /* 921 * Copy out strings - arguments and environment. 922 */ 923 copyout(stringp, destp, ARG_MAX - imgp->args->space); 924 925 /* 926 * Fill in "ps_strings" struct for ps, w, etc. 927 */ 928 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 929 suword(&arginfo->ps_nargvstr, argc); 930 931 /* 932 * Fill in argument portion of vector table. 933 */ 934 for (; argc > 0; --argc) { 935 suword(vectp++, (long)(intptr_t)destp); 936 while (*stringp++ != 0) 937 destp++; 938 destp++; 939 } 940 941 /* a null vector table pointer separates the argp's from the envp's */ 942 suword(vectp++, 0); 943 944 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 945 suword(&arginfo->ps_nenvstr, envc); 946 947 /* 948 * Fill in environment portion of vector table. 949 */ 950 for (; envc > 0; --envc) { 951 suword(vectp++, (long)(intptr_t)destp); 952 while (*stringp++ != 0) 953 destp++; 954 destp++; 955 } 956 957 /* end of vector table is a null pointer */ 958 suword(vectp, 0); 959 960 return (stack_base); 961 } 962 963 /* 964 * Check permissions of file to execute. 965 * Return 0 for success or error code on failure. 966 */ 967 int 968 exec_check_permissions(struct image_params *imgp) 969 { 970 struct proc *p = imgp->proc; 971 struct vnode *vp = imgp->vp; 972 struct vattr *attr = imgp->attr; 973 int error; 974 975 /* Get file attributes */ 976 error = VOP_GETATTR(vp, attr); 977 if (error) 978 return (error); 979 980 /* 981 * 1) Check if file execution is disabled for the filesystem that this 982 * file resides on. 983 * 2) Insure that at least one execute bit is on - otherwise root 984 * will always succeed, and we don't want to happen unless the 985 * file really is executable. 986 * 3) Insure that the file is a regular file. 987 */ 988 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 989 ((attr->va_mode & 0111) == 0) || 990 (attr->va_type != VREG)) { 991 return (EACCES); 992 } 993 994 /* 995 * Zero length files can't be exec'd 996 */ 997 if (attr->va_size == 0) 998 return (ENOEXEC); 999 1000 /* 1001 * Check for execute permission to file based on current credentials. 1002 */ 1003 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 1004 if (error) 1005 return (error); 1006 1007 /* 1008 * Check number of open-for-writes on the file and deny execution 1009 * if there are any. 1010 */ 1011 if (vp->v_writecount) 1012 return (ETXTBSY); 1013 1014 /* 1015 * Call filesystem specific open routine, which allows us to read, 1016 * write, and mmap the file. Without the VOP_OPEN we can only 1017 * stat the file. 1018 */ 1019 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1020 if (error) 1021 return (error); 1022 1023 return (0); 1024 } 1025 1026 /* 1027 * Exec handler registration 1028 */ 1029 int 1030 exec_register(const struct execsw *execsw_arg) 1031 { 1032 const struct execsw **es, **xs, **newexecsw; 1033 int count = 2; /* New slot and trailing NULL */ 1034 1035 if (execsw) 1036 for (es = execsw; *es; es++) 1037 count++; 1038 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1039 xs = newexecsw; 1040 if (execsw) 1041 for (es = execsw; *es; es++) 1042 *xs++ = *es; 1043 *xs++ = execsw_arg; 1044 *xs = NULL; 1045 if (execsw) 1046 kfree(execsw, M_TEMP); 1047 execsw = newexecsw; 1048 return 0; 1049 } 1050 1051 int 1052 exec_unregister(const struct execsw *execsw_arg) 1053 { 1054 const struct execsw **es, **xs, **newexecsw; 1055 int count = 1; 1056 1057 if (execsw == NULL) 1058 panic("unregister with no handlers left?"); 1059 1060 for (es = execsw; *es; es++) { 1061 if (*es == execsw_arg) 1062 break; 1063 } 1064 if (*es == NULL) 1065 return ENOENT; 1066 for (es = execsw; *es; es++) 1067 if (*es != execsw_arg) 1068 count++; 1069 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1070 xs = newexecsw; 1071 for (es = execsw; *es; es++) 1072 if (*es != execsw_arg) 1073 *xs++ = *es; 1074 *xs = NULL; 1075 if (execsw) 1076 kfree(execsw, M_TEMP); 1077 execsw = newexecsw; 1078 return 0; 1079 } 1080