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 bzero(&args, sizeof(args)); 541 if (error == 0) { 542 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 543 uap->argv, uap->envv); 544 } 545 if (error == 0) 546 error = kern_execve(&nd, &args); 547 nlookup_done(&nd); 548 exec_free_args(&args); 549 550 if (error < 0) { 551 /* We hit a lethal error condition. Let's die now. */ 552 exit1(W_EXITCODE(0, SIGABRT)); 553 /* NOTREACHED */ 554 } 555 556 /* 557 * The syscall result is returned in registers to the new program. 558 * Linux will register %edx as an atexit function and we must be 559 * sure to set it to 0. XXX 560 */ 561 if (error == 0) 562 uap->sysmsg_result64 = 0; 563 564 return (error); 565 } 566 567 int 568 exec_map_first_page(struct image_params *imgp) 569 { 570 int rv, i; 571 int initial_pagein; 572 vm_page_t ma[VM_INITIAL_PAGEIN]; 573 vm_page_t m; 574 vm_object_t object; 575 576 if (imgp->firstpage) 577 exec_unmap_first_page(imgp); 578 579 /* 580 * The file has to be mappable. 581 */ 582 if ((object = imgp->vp->v_object) == NULL) 583 return (EIO); 584 585 /* 586 * We shouldn't need protection for vm_page_grab() but we certainly 587 * need it for the lookup loop below (lookup/busy race), since 588 * an interrupt can unbusy and free the page before our busy check. 589 */ 590 crit_enter(); 591 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 592 593 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 594 ma[0] = m; 595 initial_pagein = VM_INITIAL_PAGEIN; 596 if (initial_pagein > object->size) 597 initial_pagein = object->size; 598 for (i = 1; i < initial_pagein; i++) { 599 if ((m = vm_page_lookup(object, i)) != NULL) { 600 if ((m->flags & PG_BUSY) || m->busy) 601 break; 602 if (m->valid) 603 break; 604 vm_page_busy(m); 605 } else { 606 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 607 if (m == NULL) 608 break; 609 } 610 ma[i] = m; 611 } 612 initial_pagein = i; 613 614 /* 615 * get_pages unbusies all the requested pages except the 616 * primary page (at index 0 in this case). The primary 617 * page may have been wired during the pagein (e.g. by 618 * the buffer cache) so vnode_pager_freepage() must be 619 * used to properly release it. 620 */ 621 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 622 m = vm_page_lookup(object, 0); 623 624 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 625 if (m) { 626 vm_page_protect(m, VM_PROT_NONE); 627 vnode_pager_freepage(m); 628 } 629 crit_exit(); 630 return EIO; 631 } 632 } 633 vm_page_hold(m); 634 vm_page_wakeup(m); /* unbusy the page */ 635 crit_exit(); 636 637 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 638 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 639 640 return 0; 641 } 642 643 void 644 exec_unmap_first_page(struct image_params *imgp) 645 { 646 vm_page_t m; 647 648 crit_enter(); 649 if (imgp->firstpage != NULL) { 650 m = sf_buf_page(imgp->firstpage); 651 sf_buf_free(imgp->firstpage); 652 imgp->firstpage = NULL; 653 imgp->image_header = NULL; 654 vm_page_unhold(m); 655 } 656 crit_exit(); 657 } 658 659 /* 660 * Destroy old address space, and allocate a new stack 661 * The new stack is only SGROWSIZ large because it is grown 662 * automatically in trap.c. 663 * 664 * This is the point of no return. 665 */ 666 int 667 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 668 { 669 struct vmspace *vmspace = imgp->proc->p_vmspace; 670 vm_offset_t stack_addr = USRSTACK - maxssiz; 671 struct proc *p; 672 vm_map_t map; 673 int error; 674 675 /* 676 * Indicate that we cannot gracefully error out any more, kill 677 * any other threads present, and set P_INEXEC to indicate that 678 * we are now messing with the process structure proper. 679 * 680 * If killalllwps() races return an error which coupled with 681 * vmspace_destroyed will cause us to exit. This is what we 682 * want since another thread is patiently waiting for us to exit 683 * in that case. 684 */ 685 p = curproc; 686 imgp->vmspace_destroyed = 1; 687 688 if (curthread->td_proc->p_nthreads > 1) { 689 error = killalllwps(1); 690 if (error) 691 return (error); 692 } 693 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 694 p->p_flag |= P_INEXEC; 695 696 /* 697 * Prevent a pending AIO from modifying the new address space. 698 */ 699 aio_proc_rundown(imgp->proc); 700 701 /* 702 * Blow away entire process VM, if address space not shared, 703 * otherwise, create a new VM space so that other threads are 704 * not disrupted. If we are execing a resident vmspace we 705 * create a duplicate of it and remap the stack. 706 * 707 * The exitingcnt test is not strictly necessary but has been 708 * included for code sanity (to make the code more deterministic). 709 */ 710 map = &vmspace->vm_map; 711 if (vmcopy) { 712 vmspace_exec(imgp->proc, vmcopy); 713 vmspace = imgp->proc->p_vmspace; 714 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 715 map = &vmspace->vm_map; 716 } else if (vmspace->vm_sysref.refcnt == 1 && 717 vmspace->vm_exitingcnt == 0) { 718 shmexit(vmspace); 719 if (vmspace->vm_upcalls) 720 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 721 pmap_remove_pages(vmspace_pmap(vmspace), 722 0, VM_MAX_USER_ADDRESS); 723 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 724 } else { 725 vmspace_exec(imgp->proc, NULL); 726 vmspace = imgp->proc->p_vmspace; 727 map = &vmspace->vm_map; 728 } 729 730 /* Allocate a new stack */ 731 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 732 0, VM_PROT_ALL, VM_PROT_ALL, 0); 733 if (error) 734 return (error); 735 736 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 737 * VM_STACK case, but they are still used to monitor the size of the 738 * process stack so we can check the stack rlimit. 739 */ 740 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 741 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 742 743 return(0); 744 } 745 746 /* 747 * Copy out argument and environment strings from the old process 748 * address space into the temporary string buffer. 749 */ 750 int 751 exec_copyin_args(struct image_args *args, char *fname, 752 enum exec_path_segflg segflg, char **argv, char **envv) 753 { 754 char *argp, *envp; 755 int error = 0; 756 size_t length; 757 758 args->buf = objcache_get(exec_objcache, M_WAITOK); 759 if (args->buf == NULL) 760 return (ENOMEM); 761 args->begin_argv = args->buf; 762 args->endp = args->begin_argv; 763 args->space = ARG_MAX; 764 765 args->fname = args->buf + ARG_MAX; 766 767 /* 768 * Copy the file name. 769 */ 770 if (segflg == PATH_SYSSPACE) { 771 error = copystr(fname, args->fname, PATH_MAX, &length); 772 } else if (segflg == PATH_USERSPACE) { 773 error = copyinstr(fname, args->fname, PATH_MAX, &length); 774 } 775 776 /* 777 * Extract argument strings. argv may not be NULL. The argv 778 * array is terminated by a NULL entry. We special-case the 779 * situation where argv[0] is NULL by passing { filename, NULL } 780 * to the new program to guarentee that the interpreter knows what 781 * file to open in case we exec an interpreted file. Note that 782 * a NULL argv[0] terminates the argv[] array. 783 * 784 * XXX the special-casing of argv[0] is historical and needs to be 785 * revisited. 786 */ 787 if (argv == NULL) 788 error = EFAULT; 789 if (error == 0) { 790 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 791 if (argp == (caddr_t)-1) { 792 error = EFAULT; 793 break; 794 } 795 error = copyinstr(argp, args->endp, 796 args->space, &length); 797 if (error) { 798 if (error == ENAMETOOLONG) 799 error = E2BIG; 800 break; 801 } 802 args->space -= length; 803 args->endp += length; 804 args->argc++; 805 } 806 if (args->argc == 0 && error == 0) { 807 length = strlen(args->fname) + 1; 808 if (length > args->space) { 809 error = E2BIG; 810 } else { 811 bcopy(args->fname, args->endp, length); 812 args->space -= length; 813 args->endp += length; 814 args->argc++; 815 } 816 } 817 } 818 819 args->begin_envv = args->endp; 820 821 /* 822 * extract environment strings. envv may be NULL. 823 */ 824 if (envv && error == 0) { 825 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 826 if (envp == (caddr_t) -1) { 827 error = EFAULT; 828 break; 829 } 830 error = copyinstr(envp, args->endp, args->space, 831 &length); 832 if (error) { 833 if (error == ENAMETOOLONG) 834 error = E2BIG; 835 break; 836 } 837 args->space -= length; 838 args->endp += length; 839 args->envc++; 840 } 841 } 842 return (error); 843 } 844 845 void 846 exec_free_args(struct image_args *args) 847 { 848 if (args->buf) { 849 objcache_put(exec_objcache, args->buf); 850 args->buf = NULL; 851 } 852 } 853 854 /* 855 * Copy strings out to the new process address space, constructing 856 * new arg and env vector tables. Return a pointer to the base 857 * so that it can be used as the initial stack pointer. 858 */ 859 register_t * 860 exec_copyout_strings(struct image_params *imgp) 861 { 862 int argc, envc, sgap; 863 char **vectp; 864 char *stringp, *destp; 865 register_t *stack_base; 866 struct ps_strings *arginfo; 867 int szsigcode; 868 869 /* 870 * Calculate string base and vector table pointers. 871 * Also deal with signal trampoline code for this exec type. 872 */ 873 arginfo = (struct ps_strings *)PS_STRINGS; 874 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 875 if (stackgap_random != 0) 876 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 877 else 878 sgap = 0; 879 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 880 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 881 882 /* 883 * install sigcode 884 */ 885 if (szsigcode) 886 copyout(imgp->proc->p_sysent->sv_sigcode, 887 ((caddr_t)arginfo - szsigcode), szsigcode); 888 889 /* 890 * If we have a valid auxargs ptr, prepare some room 891 * on the stack. 892 * 893 * The '+ 2' is for the null pointers at the end of each of the 894 * arg and env vector sets, and 'AT_COUNT*2' is room for the 895 * ELF Auxargs data. 896 */ 897 if (imgp->auxargs) { 898 vectp = (char **)(destp - (imgp->args->argc + 899 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 900 } else { 901 vectp = (char **)(destp - (imgp->args->argc + 902 imgp->args->envc + 2) * sizeof(char*)); 903 } 904 905 /* 906 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 907 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 908 */ 909 910 /* 911 * vectp also becomes our initial stack base 912 */ 913 stack_base = (register_t *)vectp; 914 915 stringp = imgp->args->begin_argv; 916 argc = imgp->args->argc; 917 envc = imgp->args->envc; 918 919 /* 920 * Copy out strings - arguments and environment. 921 */ 922 copyout(stringp, destp, ARG_MAX - imgp->args->space); 923 924 /* 925 * Fill in "ps_strings" struct for ps, w, etc. 926 */ 927 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 928 suword(&arginfo->ps_nargvstr, argc); 929 930 /* 931 * Fill in argument portion of vector table. 932 */ 933 for (; argc > 0; --argc) { 934 suword(vectp++, (long)(intptr_t)destp); 935 while (*stringp++ != 0) 936 destp++; 937 destp++; 938 } 939 940 /* a null vector table pointer separates the argp's from the envp's */ 941 suword(vectp++, 0); 942 943 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 944 suword(&arginfo->ps_nenvstr, envc); 945 946 /* 947 * Fill in environment portion of vector table. 948 */ 949 for (; envc > 0; --envc) { 950 suword(vectp++, (long)(intptr_t)destp); 951 while (*stringp++ != 0) 952 destp++; 953 destp++; 954 } 955 956 /* end of vector table is a null pointer */ 957 suword(vectp, 0); 958 959 return (stack_base); 960 } 961 962 /* 963 * Check permissions of file to execute. 964 * Return 0 for success or error code on failure. 965 */ 966 int 967 exec_check_permissions(struct image_params *imgp) 968 { 969 struct proc *p = imgp->proc; 970 struct vnode *vp = imgp->vp; 971 struct vattr *attr = imgp->attr; 972 int error; 973 974 /* Get file attributes */ 975 error = VOP_GETATTR(vp, attr); 976 if (error) 977 return (error); 978 979 /* 980 * 1) Check if file execution is disabled for the filesystem that this 981 * file resides on. 982 * 2) Insure that at least one execute bit is on - otherwise root 983 * will always succeed, and we don't want to happen unless the 984 * file really is executable. 985 * 3) Insure that the file is a regular file. 986 */ 987 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 988 ((attr->va_mode & 0111) == 0) || 989 (attr->va_type != VREG)) { 990 return (EACCES); 991 } 992 993 /* 994 * Zero length files can't be exec'd 995 */ 996 if (attr->va_size == 0) 997 return (ENOEXEC); 998 999 /* 1000 * Check for execute permission to file based on current credentials. 1001 */ 1002 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 1003 if (error) 1004 return (error); 1005 1006 /* 1007 * Check number of open-for-writes on the file and deny execution 1008 * if there are any. 1009 */ 1010 if (vp->v_writecount) 1011 return (ETXTBSY); 1012 1013 /* 1014 * Call filesystem specific open routine, which allows us to read, 1015 * write, and mmap the file. Without the VOP_OPEN we can only 1016 * stat the file. 1017 */ 1018 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1019 if (error) 1020 return (error); 1021 1022 return (0); 1023 } 1024 1025 /* 1026 * Exec handler registration 1027 */ 1028 int 1029 exec_register(const struct execsw *execsw_arg) 1030 { 1031 const struct execsw **es, **xs, **newexecsw; 1032 int count = 2; /* New slot and trailing NULL */ 1033 1034 if (execsw) 1035 for (es = execsw; *es; es++) 1036 count++; 1037 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1038 xs = newexecsw; 1039 if (execsw) 1040 for (es = execsw; *es; es++) 1041 *xs++ = *es; 1042 *xs++ = execsw_arg; 1043 *xs = NULL; 1044 if (execsw) 1045 kfree(execsw, M_TEMP); 1046 execsw = newexecsw; 1047 return 0; 1048 } 1049 1050 int 1051 exec_unregister(const struct execsw *execsw_arg) 1052 { 1053 const struct execsw **es, **xs, **newexecsw; 1054 int count = 1; 1055 1056 if (execsw == NULL) 1057 panic("unregister with no handlers left?"); 1058 1059 for (es = execsw; *es; es++) { 1060 if (*es == execsw_arg) 1061 break; 1062 } 1063 if (*es == NULL) 1064 return ENOENT; 1065 for (es = execsw; *es; es++) 1066 if (*es != execsw_arg) 1067 count++; 1068 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1069 xs = newexecsw; 1070 for (es = execsw; *es; es++) 1071 if (*es != execsw_arg) 1072 *xs++ = *es; 1073 *xs = NULL; 1074 if (execsw) 1075 kfree(execsw, M_TEMP); 1076 execsw = newexecsw; 1077 return 0; 1078 } 1079