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 static int ktrace_suid = 0; 97 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, ""); 98 99 void print_execve_args(struct image_args *args); 100 int debug_execve_args = 0; 101 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 102 0, ""); 103 104 /* 105 * Exec arguments object cache 106 */ 107 static struct objcache *exec_objcache; 108 109 static 110 void 111 exec_objcache_init(void *arg __unused) 112 { 113 int cluster_limit; 114 115 cluster_limit = 16; /* up to this many objects */ 116 exec_objcache = objcache_create_mbacked( 117 M_EXECARGS, PATH_MAX + ARG_MAX, 118 &cluster_limit, 119 2, /* minimal magazine capacity */ 120 NULL, NULL, NULL); 121 } 122 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 123 124 /* 125 * stackgap_random specifies if the stackgap should have a random size added 126 * to it. It must be a power of 2. If non-zero, the stack gap will be 127 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 128 */ 129 static int stackgap_random = 1024; 130 static int 131 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 132 { 133 int error, new_val; 134 new_val = stackgap_random; 135 error = sysctl_handle_int(oidp, &new_val, 0, req); 136 if (error != 0 || req->newptr == NULL) 137 return (error); 138 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 139 return (EINVAL); 140 stackgap_random = new_val; 141 142 return(0); 143 } 144 145 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 146 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 147 148 void 149 print_execve_args(struct image_args *args) 150 { 151 char *cp; 152 int ndx; 153 154 cp = args->begin_argv; 155 for (ndx = 0; ndx < args->argc; ndx++) { 156 kprintf("\targv[%d]: %s\n", ndx, cp); 157 while (*cp++ != '\0'); 158 } 159 for (ndx = 0; ndx < args->envc; ndx++) { 160 kprintf("\tenvv[%d]: %s\n", ndx, cp); 161 while (*cp++ != '\0'); 162 } 163 } 164 165 /* 166 * Each of the items is a pointer to a `const struct execsw', hence the 167 * double pointer here. 168 */ 169 static const struct execsw **execsw; 170 171 /* 172 * Replace current vmspace with a new binary. 173 * Returns 0 on success, > 0 on recoverable error (use as errno). 174 * Returns -1 on lethal error which demands killing of the current 175 * process! 176 */ 177 int 178 kern_execve(struct nlookupdata *nd, struct image_args *args) 179 { 180 struct thread *td = curthread; 181 struct lwp *lp = td->td_lwp; 182 struct proc *p = td->td_proc; 183 register_t *stack_base; 184 int error, len, i; 185 struct image_params image_params, *imgp; 186 struct vattr attr; 187 int (*img_first) (struct image_params *); 188 189 if (debug_execve_args) { 190 kprintf("%s()\n", __func__); 191 print_execve_args(args); 192 } 193 194 KKASSERT(p); 195 imgp = &image_params; 196 197 /* 198 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 199 * no modifications to the process at all until we get there. 200 * 201 * Note that multiple threads may be trying to exec at the same 202 * time. exec_new_vmspace() handles that too. 203 */ 204 205 /* 206 * Initialize part of the common data 207 */ 208 imgp->proc = p; 209 imgp->args = args; 210 imgp->attr = &attr; 211 imgp->entry_addr = 0; 212 imgp->resident = 0; 213 imgp->vmspace_destroyed = 0; 214 imgp->interpreted = 0; 215 imgp->interpreter_name[0] = 0; 216 imgp->auxargs = NULL; 217 imgp->vp = NULL; 218 imgp->firstpage = NULL; 219 imgp->ps_strings = 0; 220 imgp->image_header = NULL; 221 222 interpret: 223 224 /* 225 * Translate the file name to a vnode. Unlock the cache entry to 226 * improve parallelism for programs exec'd in parallel. 227 */ 228 if ((error = nlookup(nd)) != 0) 229 goto exec_fail; 230 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 231 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 232 nd->nl_flags &= ~NLC_NCPISLOCKED; 233 cache_unlock(&nd->nl_nch); 234 if (error) 235 goto exec_fail; 236 237 /* 238 * Check file permissions (also 'opens' file) 239 */ 240 error = exec_check_permissions(imgp); 241 if (error) { 242 vn_unlock(imgp->vp); 243 goto exec_fail_dealloc; 244 } 245 246 error = exec_map_first_page(imgp); 247 vn_unlock(imgp->vp); 248 if (error) 249 goto exec_fail_dealloc; 250 251 if (debug_execve_args && imgp->interpreted) { 252 kprintf(" target is interpreted -- recursive pass\n"); 253 kprintf(" interpreter: %s\n", imgp->interpreter_name); 254 print_execve_args(args); 255 } 256 257 /* 258 * If the current process has a special image activator it 259 * wants to try first, call it. For example, emulating shell 260 * scripts differently. 261 */ 262 error = -1; 263 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 264 error = img_first(imgp); 265 266 /* 267 * If the vnode has a registered vmspace, exec the vmspace 268 */ 269 if (error == -1 && imgp->vp->v_resident) { 270 error = exec_resident_imgact(imgp); 271 } 272 273 /* 274 * Loop through the list of image activators, calling each one. 275 * An activator returns -1 if there is no match, 0 on success, 276 * and an error otherwise. 277 */ 278 for (i = 0; error == -1 && execsw[i]; ++i) { 279 if (execsw[i]->ex_imgact == NULL || 280 execsw[i]->ex_imgact == img_first) { 281 continue; 282 } 283 error = (*execsw[i]->ex_imgact)(imgp); 284 } 285 286 if (error) { 287 if (error == -1) 288 error = ENOEXEC; 289 goto exec_fail_dealloc; 290 } 291 292 /* 293 * Special interpreter operation, cleanup and loop up to try to 294 * activate the interpreter. 295 */ 296 if (imgp->interpreted) { 297 exec_unmap_first_page(imgp); 298 nlookup_done(nd); 299 vrele(imgp->vp); 300 imgp->vp = NULL; 301 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 302 NLC_FOLLOW); 303 if (error) 304 goto exec_fail; 305 goto interpret; 306 } 307 308 /* 309 * Copy out strings (args and env) and initialize stack base 310 */ 311 stack_base = exec_copyout_strings(imgp); 312 p->p_vmspace->vm_minsaddr = (char *)stack_base; 313 314 /* 315 * If custom stack fixup routine present for this process 316 * let it do the stack setup. If we are running a resident 317 * image there is no auxinfo or other image activator context 318 * so don't try to add fixups to the stack. 319 * 320 * Else stuff argument count as first item on stack 321 */ 322 if (p->p_sysent->sv_fixup && imgp->resident == 0) 323 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 324 else 325 suword(--stack_base, imgp->args->argc); 326 327 /* 328 * For security and other reasons, the file descriptor table cannot 329 * be shared after an exec. 330 */ 331 if (p->p_fd->fd_refcnt > 1) { 332 struct filedesc *tmp; 333 334 tmp = fdcopy(p); 335 fdfree(p, tmp); 336 } 337 338 /* 339 * For security and other reasons, signal handlers cannot 340 * be shared after an exec. The new proces gets a copy of the old 341 * handlers. In execsigs(), the new process will have its signals 342 * reset. 343 */ 344 if (p->p_sigacts->ps_refcnt > 1) { 345 struct sigacts *newsigacts; 346 347 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts), 348 M_SUBPROC, M_WAITOK); 349 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts)); 350 p->p_sigacts->ps_refcnt--; 351 p->p_sigacts = newsigacts; 352 p->p_sigacts->ps_refcnt = 1; 353 } 354 355 /* 356 * For security and other reasons virtual kernels cannot be 357 * inherited by an exec. This also allows a virtual kernel 358 * to fork/exec unrelated applications. 359 */ 360 if (p->p_vkernel) 361 vkernel_exit(p); 362 363 /* Stop profiling */ 364 stopprofclock(p); 365 366 /* close files on exec */ 367 fdcloseexec(p); 368 369 /* reset caught signals */ 370 execsigs(p); 371 372 /* name this process - nameiexec(p, ndp) */ 373 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 374 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 375 p->p_comm[len] = 0; 376 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 377 378 /* 379 * mark as execed, wakeup the process that vforked (if any) and tell 380 * it that it now has its own resources back 381 */ 382 p->p_flag |= P_EXEC; 383 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 384 p->p_flag &= ~P_PPWAIT; 385 wakeup((caddr_t)p->p_pptr); 386 } 387 388 /* 389 * Implement image setuid/setgid. 390 * 391 * Don't honor setuid/setgid if the filesystem prohibits it or if 392 * the process is being traced. 393 */ 394 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 395 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 396 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 397 (p->p_flag & P_TRACED) == 0) { 398 /* 399 * Turn off syscall tracing for set-id programs, except for 400 * root. Record any set-id flags first to make sure that 401 * we do not regain any tracing during a possible block. 402 */ 403 setsugid(); 404 if (p->p_tracenode && ktrace_suid == 0 && 405 priv_check(td, PRIV_ROOT) != 0) { 406 ktrdestroy(&p->p_tracenode); 407 p->p_traceflag = 0; 408 } 409 /* Close any file descriptors 0..2 that reference procfs */ 410 setugidsafety(p); 411 /* Make sure file descriptors 0..2 are in use. */ 412 error = fdcheckstd(lp); 413 if (error != 0) 414 goto exec_fail_dealloc; 415 /* 416 * Set the new credentials. 417 */ 418 cratom(&p->p_ucred); 419 if (attr.va_mode & VSUID) 420 change_euid(attr.va_uid); 421 if (attr.va_mode & VSGID) 422 p->p_ucred->cr_gid = attr.va_gid; 423 424 /* 425 * Clear local varsym variables 426 */ 427 varsymset_clean(&p->p_varsymset); 428 } else { 429 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 430 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 431 p->p_flag &= ~P_SUGID; 432 } 433 434 /* 435 * Implement correct POSIX saved-id behavior. 436 */ 437 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 438 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 439 cratom(&p->p_ucred); 440 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 441 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 442 } 443 444 /* 445 * Store the vp for use in procfs 446 */ 447 if (p->p_textvp) /* release old reference */ 448 vrele(p->p_textvp); 449 p->p_textvp = imgp->vp; 450 vref(p->p_textvp); 451 452 /* 453 * Notify others that we exec'd, and clear the P_INEXEC flag 454 * as we're now a bona fide freshly-execed process. 455 */ 456 KNOTE(&p->p_klist, NOTE_EXEC); 457 p->p_flag &= ~P_INEXEC; 458 459 /* 460 * If tracing the process, trap to debugger so breakpoints 461 * can be set before the program executes. 462 */ 463 STOPEVENT(p, S_EXEC, 0); 464 465 if (p->p_flag & P_TRACED) 466 ksignal(p, SIGTRAP); 467 468 /* clear "fork but no exec" flag, as we _are_ execing */ 469 p->p_acflag &= ~AFORK; 470 471 /* Set values passed into the program in registers. */ 472 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 473 imgp->ps_strings); 474 475 /* Set the access time on the vnode */ 476 vn_mark_atime(imgp->vp, td); 477 478 /* Free any previous argument cache */ 479 if (p->p_args && --p->p_args->ar_ref == 0) 480 FREE(p->p_args, M_PARGS); 481 p->p_args = NULL; 482 483 /* Cache arguments if they fit inside our allowance */ 484 i = imgp->args->begin_envv - imgp->args->begin_argv; 485 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 486 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 487 M_PARGS, M_WAITOK); 488 p->p_args->ar_ref = 1; 489 p->p_args->ar_length = i; 490 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 491 } 492 493 exec_fail_dealloc: 494 495 /* 496 * free various allocated resources 497 */ 498 if (imgp->firstpage) 499 exec_unmap_first_page(imgp); 500 501 if (imgp->vp) { 502 vrele(imgp->vp); 503 imgp->vp = NULL; 504 } 505 506 if (error == 0) { 507 ++mycpu->gd_cnt.v_exec; 508 return (0); 509 } 510 511 exec_fail: 512 /* 513 * we're done here, clear P_INEXEC if we were the ones that 514 * set it. Otherwise if vmspace_destroyed is still set we 515 * raced another thread and that thread is responsible for 516 * clearing it. 517 */ 518 if (imgp->vmspace_destroyed & 2) 519 p->p_flag &= ~P_INEXEC; 520 if (imgp->vmspace_destroyed) { 521 /* 522 * Sorry, no more process anymore. exit gracefully. 523 * However we can't die right here, because our 524 * caller might have to clean up, so indicate a 525 * lethal error by returning -1. 526 */ 527 return(-1); 528 } else { 529 return(error); 530 } 531 } 532 533 /* 534 * execve() system call. 535 * 536 * MPALMOSTSAFE 537 */ 538 int 539 sys_execve(struct execve_args *uap) 540 { 541 struct nlookupdata nd; 542 struct image_args args; 543 int error; 544 545 bzero(&args, sizeof(args)); 546 547 get_mplock(); 548 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 549 if (error == 0) { 550 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 551 uap->argv, uap->envv); 552 } 553 if (error == 0) 554 error = kern_execve(&nd, &args); 555 nlookup_done(&nd); 556 exec_free_args(&args); 557 558 if (error < 0) { 559 /* We hit a lethal error condition. Let's die now. */ 560 exit1(W_EXITCODE(0, SIGABRT)); 561 /* NOTREACHED */ 562 } 563 rel_mplock(); 564 565 /* 566 * The syscall result is returned in registers to the new program. 567 * Linux will register %edx as an atexit function and we must be 568 * sure to set it to 0. XXX 569 */ 570 if (error == 0) 571 uap->sysmsg_result64 = 0; 572 573 return (error); 574 } 575 576 int 577 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 578 struct sf_buf **psfb, const char **pdata) 579 { 580 int rv, i; 581 int initial_pagein; 582 vm_page_t ma[VM_INITIAL_PAGEIN]; 583 vm_page_t m; 584 vm_object_t object; 585 586 /* 587 * The file has to be mappable. 588 */ 589 if ((object = imgp->vp->v_object) == NULL) 590 return (EIO); 591 592 if (pageno >= object->size) 593 return (EIO); 594 595 /* 596 * We shouldn't need protection for vm_page_grab() but we certainly 597 * need it for the lookup loop below (lookup/busy race), since 598 * an interrupt can unbusy and free the page before our busy check. 599 */ 600 crit_enter(); 601 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 602 603 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 604 ma[0] = m; 605 initial_pagein = VM_INITIAL_PAGEIN; 606 if (initial_pagein + pageno > object->size) 607 initial_pagein = object->size - pageno; 608 for (i = 1; i < initial_pagein; i++) { 609 if ((m = vm_page_lookup(object, i + pageno)) != NULL) { 610 if ((m->flags & PG_BUSY) || m->busy) 611 break; 612 if (m->valid) 613 break; 614 vm_page_busy(m); 615 } else { 616 m = vm_page_alloc(object, i + pageno, VM_ALLOC_NORMAL); 617 if (m == NULL) 618 break; 619 } 620 ma[i] = m; 621 } 622 initial_pagein = i; 623 624 /* 625 * get_pages unbusies all the requested pages except the 626 * primary page (at index 0 in this case). The primary 627 * page may have been wired during the pagein (e.g. by 628 * the buffer cache) so vnode_pager_freepage() must be 629 * used to properly release it. 630 */ 631 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 632 m = vm_page_lookup(object, pageno); 633 634 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 635 if (m) { 636 vm_page_protect(m, VM_PROT_NONE); 637 vnode_pager_freepage(m); 638 } 639 crit_exit(); 640 return EIO; 641 } 642 } 643 vm_page_hold(m); 644 vm_page_wakeup(m); /* unbusy the page */ 645 crit_exit(); 646 647 *psfb = sf_buf_alloc(m, SFB_CPUPRIVATE); 648 *pdata = (void *)sf_buf_kva(*psfb); 649 650 return (0); 651 } 652 653 int 654 exec_map_first_page(struct image_params *imgp) 655 { 656 int err; 657 658 if (imgp->firstpage) 659 exec_unmap_first_page(imgp); 660 661 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 662 663 if (err) 664 return err; 665 666 return 0; 667 } 668 669 void 670 exec_unmap_page(struct sf_buf *sfb) 671 { 672 vm_page_t m; 673 674 crit_enter(); 675 if (sfb != NULL) { 676 m = sf_buf_page(sfb); 677 sf_buf_free(sfb); 678 vm_page_unhold(m); 679 } 680 crit_exit(); 681 } 682 683 void 684 exec_unmap_first_page(struct image_params *imgp) 685 { 686 exec_unmap_page(imgp->firstpage); 687 imgp->firstpage = NULL; 688 imgp->image_header = NULL; 689 } 690 691 /* 692 * Destroy old address space, and allocate a new stack 693 * The new stack is only SGROWSIZ large because it is grown 694 * automatically in trap.c. 695 * 696 * This is the point of no return. 697 */ 698 int 699 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 700 { 701 struct vmspace *vmspace = imgp->proc->p_vmspace; 702 vm_offset_t stack_addr = USRSTACK - maxssiz; 703 struct proc *p; 704 vm_map_t map; 705 int error; 706 707 /* 708 * Indicate that we cannot gracefully error out any more, kill 709 * any other threads present, and set P_INEXEC to indicate that 710 * we are now messing with the process structure proper. 711 * 712 * If killalllwps() races return an error which coupled with 713 * vmspace_destroyed will cause us to exit. This is what we 714 * want since another thread is patiently waiting for us to exit 715 * in that case. 716 */ 717 p = curproc; 718 imgp->vmspace_destroyed = 1; 719 720 if (curthread->td_proc->p_nthreads > 1) { 721 error = killalllwps(1); 722 if (error) 723 return (error); 724 } 725 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 726 p->p_flag |= P_INEXEC; 727 728 /* 729 * Prevent a pending AIO from modifying the new address space. 730 */ 731 aio_proc_rundown(imgp->proc); 732 733 /* 734 * Blow away entire process VM, if address space not shared, 735 * otherwise, create a new VM space so that other threads are 736 * not disrupted. If we are execing a resident vmspace we 737 * create a duplicate of it and remap the stack. 738 * 739 * The exitingcnt test is not strictly necessary but has been 740 * included for code sanity (to make the code more deterministic). 741 */ 742 map = &vmspace->vm_map; 743 if (vmcopy) { 744 vmspace_exec(imgp->proc, vmcopy); 745 vmspace = imgp->proc->p_vmspace; 746 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 747 map = &vmspace->vm_map; 748 } else if (vmspace->vm_sysref.refcnt == 1 && 749 vmspace->vm_exitingcnt == 0) { 750 shmexit(vmspace); 751 if (vmspace->vm_upcalls) 752 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 753 pmap_remove_pages(vmspace_pmap(vmspace), 754 0, VM_MAX_USER_ADDRESS); 755 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 756 } else { 757 vmspace_exec(imgp->proc, NULL); 758 vmspace = imgp->proc->p_vmspace; 759 map = &vmspace->vm_map; 760 } 761 762 /* Allocate a new stack */ 763 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 764 0, VM_PROT_ALL, VM_PROT_ALL, 0); 765 if (error) 766 return (error); 767 768 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 769 * VM_STACK case, but they are still used to monitor the size of the 770 * process stack so we can check the stack rlimit. 771 */ 772 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 773 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 774 775 return(0); 776 } 777 778 /* 779 * Copy out argument and environment strings from the old process 780 * address space into the temporary string buffer. 781 */ 782 int 783 exec_copyin_args(struct image_args *args, char *fname, 784 enum exec_path_segflg segflg, char **argv, char **envv) 785 { 786 char *argp, *envp; 787 int error = 0; 788 size_t length; 789 790 args->buf = objcache_get(exec_objcache, M_WAITOK); 791 if (args->buf == NULL) 792 return (ENOMEM); 793 args->begin_argv = args->buf; 794 args->endp = args->begin_argv; 795 args->space = ARG_MAX; 796 797 args->fname = args->buf + ARG_MAX; 798 799 /* 800 * Copy the file name. 801 */ 802 if (segflg == PATH_SYSSPACE) { 803 error = copystr(fname, args->fname, PATH_MAX, &length); 804 } else if (segflg == PATH_USERSPACE) { 805 error = copyinstr(fname, args->fname, PATH_MAX, &length); 806 } 807 808 /* 809 * Extract argument strings. argv may not be NULL. The argv 810 * array is terminated by a NULL entry. We special-case the 811 * situation where argv[0] is NULL by passing { filename, NULL } 812 * to the new program to guarentee that the interpreter knows what 813 * file to open in case we exec an interpreted file. Note that 814 * a NULL argv[0] terminates the argv[] array. 815 * 816 * XXX the special-casing of argv[0] is historical and needs to be 817 * revisited. 818 */ 819 if (argv == NULL) 820 error = EFAULT; 821 if (error == 0) { 822 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 823 if (argp == (caddr_t)-1) { 824 error = EFAULT; 825 break; 826 } 827 error = copyinstr(argp, args->endp, 828 args->space, &length); 829 if (error) { 830 if (error == ENAMETOOLONG) 831 error = E2BIG; 832 break; 833 } 834 args->space -= length; 835 args->endp += length; 836 args->argc++; 837 } 838 if (args->argc == 0 && error == 0) { 839 length = strlen(args->fname) + 1; 840 if (length > args->space) { 841 error = E2BIG; 842 } else { 843 bcopy(args->fname, args->endp, length); 844 args->space -= length; 845 args->endp += length; 846 args->argc++; 847 } 848 } 849 } 850 851 args->begin_envv = args->endp; 852 853 /* 854 * extract environment strings. envv may be NULL. 855 */ 856 if (envv && error == 0) { 857 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 858 if (envp == (caddr_t) -1) { 859 error = EFAULT; 860 break; 861 } 862 error = copyinstr(envp, args->endp, args->space, 863 &length); 864 if (error) { 865 if (error == ENAMETOOLONG) 866 error = E2BIG; 867 break; 868 } 869 args->space -= length; 870 args->endp += length; 871 args->envc++; 872 } 873 } 874 return (error); 875 } 876 877 void 878 exec_free_args(struct image_args *args) 879 { 880 if (args->buf) { 881 objcache_put(exec_objcache, args->buf); 882 args->buf = NULL; 883 } 884 } 885 886 /* 887 * Copy strings out to the new process address space, constructing 888 * new arg and env vector tables. Return a pointer to the base 889 * so that it can be used as the initial stack pointer. 890 */ 891 register_t * 892 exec_copyout_strings(struct image_params *imgp) 893 { 894 int argc, envc, sgap; 895 char **vectp; 896 char *stringp, *destp; 897 register_t *stack_base; 898 struct ps_strings *arginfo; 899 int szsigcode; 900 901 /* 902 * Calculate string base and vector table pointers. 903 * Also deal with signal trampoline code for this exec type. 904 */ 905 arginfo = (struct ps_strings *)PS_STRINGS; 906 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 907 if (stackgap_random != 0) 908 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 909 else 910 sgap = 0; 911 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 912 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 913 914 /* 915 * install sigcode 916 */ 917 if (szsigcode) 918 copyout(imgp->proc->p_sysent->sv_sigcode, 919 ((caddr_t)arginfo - szsigcode), szsigcode); 920 921 /* 922 * If we have a valid auxargs ptr, prepare some room 923 * on the stack. 924 * 925 * The '+ 2' is for the null pointers at the end of each of the 926 * arg and env vector sets, and 'AT_COUNT*2' is room for the 927 * ELF Auxargs data. 928 */ 929 if (imgp->auxargs) { 930 vectp = (char **)(destp - (imgp->args->argc + 931 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 932 } else { 933 vectp = (char **)(destp - (imgp->args->argc + 934 imgp->args->envc + 2) * sizeof(char*)); 935 } 936 937 /* 938 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 939 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 940 */ 941 942 /* 943 * vectp also becomes our initial stack base 944 */ 945 stack_base = (register_t *)vectp; 946 947 stringp = imgp->args->begin_argv; 948 argc = imgp->args->argc; 949 envc = imgp->args->envc; 950 951 /* 952 * Copy out strings - arguments and environment. 953 */ 954 copyout(stringp, destp, ARG_MAX - imgp->args->space); 955 956 /* 957 * Fill in "ps_strings" struct for ps, w, etc. 958 */ 959 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 960 suword(&arginfo->ps_nargvstr, argc); 961 962 /* 963 * Fill in argument portion of vector table. 964 */ 965 for (; argc > 0; --argc) { 966 suword(vectp++, (long)(intptr_t)destp); 967 while (*stringp++ != 0) 968 destp++; 969 destp++; 970 } 971 972 /* a null vector table pointer separates the argp's from the envp's */ 973 suword(vectp++, 0); 974 975 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 976 suword(&arginfo->ps_nenvstr, envc); 977 978 /* 979 * Fill in environment portion of vector table. 980 */ 981 for (; envc > 0; --envc) { 982 suword(vectp++, (long)(intptr_t)destp); 983 while (*stringp++ != 0) 984 destp++; 985 destp++; 986 } 987 988 /* end of vector table is a null pointer */ 989 suword(vectp, 0); 990 991 return (stack_base); 992 } 993 994 /* 995 * Check permissions of file to execute. 996 * Return 0 for success or error code on failure. 997 */ 998 int 999 exec_check_permissions(struct image_params *imgp) 1000 { 1001 struct proc *p = imgp->proc; 1002 struct vnode *vp = imgp->vp; 1003 struct vattr *attr = imgp->attr; 1004 int error; 1005 1006 /* Get file attributes */ 1007 error = VOP_GETATTR(vp, attr); 1008 if (error) 1009 return (error); 1010 1011 /* 1012 * 1) Check if file execution is disabled for the filesystem that this 1013 * file resides on. 1014 * 2) Insure that at least one execute bit is on - otherwise root 1015 * will always succeed, and we don't want to happen unless the 1016 * file really is executable. 1017 * 3) Insure that the file is a regular file. 1018 */ 1019 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1020 ((attr->va_mode & 0111) == 0) || 1021 (attr->va_type != VREG)) { 1022 return (EACCES); 1023 } 1024 1025 /* 1026 * Zero length files can't be exec'd 1027 */ 1028 if (attr->va_size == 0) 1029 return (ENOEXEC); 1030 1031 /* 1032 * Check for execute permission to file based on current credentials. 1033 */ 1034 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1035 if (error) 1036 return (error); 1037 1038 /* 1039 * Check number of open-for-writes on the file and deny execution 1040 * if there are any. 1041 */ 1042 if (vp->v_writecount) 1043 return (ETXTBSY); 1044 1045 /* 1046 * Call filesystem specific open routine, which allows us to read, 1047 * write, and mmap the file. Without the VOP_OPEN we can only 1048 * stat the file. 1049 */ 1050 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1051 if (error) 1052 return (error); 1053 1054 return (0); 1055 } 1056 1057 /* 1058 * Exec handler registration 1059 */ 1060 int 1061 exec_register(const struct execsw *execsw_arg) 1062 { 1063 const struct execsw **es, **xs, **newexecsw; 1064 int count = 2; /* New slot and trailing NULL */ 1065 1066 if (execsw) 1067 for (es = execsw; *es; es++) 1068 count++; 1069 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1070 xs = newexecsw; 1071 if (execsw) 1072 for (es = execsw; *es; es++) 1073 *xs++ = *es; 1074 *xs++ = execsw_arg; 1075 *xs = NULL; 1076 if (execsw) 1077 kfree(execsw, M_TEMP); 1078 execsw = newexecsw; 1079 return 0; 1080 } 1081 1082 int 1083 exec_unregister(const struct execsw *execsw_arg) 1084 { 1085 const struct execsw **es, **xs, **newexecsw; 1086 int count = 1; 1087 1088 if (execsw == NULL) 1089 panic("unregister with no handlers left?"); 1090 1091 for (es = execsw; *es; es++) { 1092 if (*es == execsw_arg) 1093 break; 1094 } 1095 if (*es == NULL) 1096 return ENOENT; 1097 for (es = execsw; *es; es++) 1098 if (*es != execsw_arg) 1099 count++; 1100 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1101 xs = newexecsw; 1102 for (es = execsw; *es; es++) 1103 if (*es != execsw_arg) 1104 *xs++ = *es; 1105 *xs = NULL; 1106 if (execsw) 1107 kfree(execsw, M_TEMP); 1108 execsw = newexecsw; 1109 return 0; 1110 } 1111