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/ktrace.h> 46 #include <sys/signalvar.h> 47 #include <sys/pioctl.h> 48 #include <sys/nlookup.h> 49 #include <sys/sfbuf.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 <vm/vm.h> 59 #include <vm/vm_param.h> 60 #include <sys/lock.h> 61 #include <vm/pmap.h> 62 #include <vm/vm_page.h> 63 #include <vm/vm_map.h> 64 #include <vm/vm_kern.h> 65 #include <vm/vm_extern.h> 66 #include <vm/vm_object.h> 67 #include <vm/vnode_pager.h> 68 #include <vm/vm_pager.h> 69 70 #include <sys/user.h> 71 #include <sys/reg.h> 72 73 #include <sys/thread2.h> 74 75 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 76 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 77 78 static register_t *exec_copyout_strings (struct image_params *); 79 80 /* XXX This should be vm_size_t. */ 81 static u_long ps_strings = PS_STRINGS; 82 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 83 84 /* XXX This should be vm_size_t. */ 85 static u_long usrstack = USRSTACK; 86 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 87 88 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 89 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 90 &ps_arg_cache_limit, 0, ""); 91 92 int ps_argsopen = 1; 93 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 94 95 void print_execve_args(struct image_args *args); 96 int debug_execve_args = 0; 97 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 98 0, ""); 99 100 /* 101 * Exec arguments object cache 102 */ 103 static struct objcache *exec_objcache; 104 105 static 106 void 107 exec_objcache_init(void *arg __unused) 108 { 109 int cluster_limit; 110 111 cluster_limit = 16; /* up to this many objects */ 112 exec_objcache = objcache_create_mbacked( 113 M_EXECARGS, PATH_MAX + ARG_MAX, 114 &cluster_limit, 115 2, /* minimal magazine capacity */ 116 NULL, NULL, NULL); 117 } 118 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 119 120 /* 121 * stackgap_random specifies if the stackgap should have a random size added 122 * to it. It must be a power of 2. If non-zero, the stack gap will be 123 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 124 */ 125 static int stackgap_random = 1024; 126 static int 127 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 128 { 129 int error, new_val; 130 new_val = stackgap_random; 131 error = sysctl_handle_int(oidp, &new_val, 0, req); 132 if (error != 0 || req->newptr == NULL) 133 return (error); 134 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 135 return (EINVAL); 136 stackgap_random = new_val; 137 138 return(0); 139 } 140 141 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 142 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 143 144 void 145 print_execve_args(struct image_args *args) 146 { 147 char *cp; 148 int ndx; 149 150 cp = args->begin_argv; 151 for (ndx = 0; ndx < args->argc; ndx++) { 152 kprintf("\targv[%d]: %s\n", ndx, cp); 153 while (*cp++ != '\0'); 154 } 155 for (ndx = 0; ndx < args->envc; ndx++) { 156 kprintf("\tenvv[%d]: %s\n", ndx, cp); 157 while (*cp++ != '\0'); 158 } 159 } 160 161 /* 162 * Each of the items is a pointer to a `const struct execsw', hence the 163 * double pointer here. 164 */ 165 static const struct execsw **execsw; 166 167 /* 168 * Replace current vmspace with a new binary. 169 * Returns 0 on success, > 0 on recoverable error (use as errno). 170 * Returns -1 on lethal error which demands killing of the current 171 * process! 172 */ 173 int 174 kern_execve(struct nlookupdata *nd, struct image_args *args) 175 { 176 struct thread *td = curthread; 177 struct lwp *lp = td->td_lwp; 178 struct proc *p = td->td_proc; 179 register_t *stack_base; 180 int error, len, i; 181 struct image_params image_params, *imgp; 182 struct vattr attr; 183 int (*img_first) (struct image_params *); 184 185 if (debug_execve_args) { 186 kprintf("%s()\n", __func__); 187 print_execve_args(args); 188 } 189 190 KKASSERT(p); 191 imgp = &image_params; 192 193 /* 194 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 195 * no modifications to the process at all until we get there. 196 * 197 * Note that multiple threads may be trying to exec at the same 198 * time. exec_new_vmspace() handles that too. 199 */ 200 201 /* 202 * Initialize part of the common data 203 */ 204 imgp->proc = p; 205 imgp->args = args; 206 imgp->attr = &attr; 207 imgp->entry_addr = 0; 208 imgp->resident = 0; 209 imgp->vmspace_destroyed = 0; 210 imgp->interpreted = 0; 211 imgp->interpreter_name[0] = 0; 212 imgp->auxargs = NULL; 213 imgp->vp = NULL; 214 imgp->firstpage = NULL; 215 imgp->ps_strings = 0; 216 imgp->image_header = NULL; 217 218 interpret: 219 220 /* 221 * Translate the file name to a vnode. Unlock the cache entry to 222 * improve parallelism for programs exec'd in parallel. 223 */ 224 if ((error = nlookup(nd)) != 0) 225 goto exec_fail; 226 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 227 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 228 nd->nl_flags &= ~NLC_NCPISLOCKED; 229 cache_unlock(&nd->nl_nch); 230 if (error) 231 goto exec_fail; 232 233 /* 234 * Check file permissions (also 'opens' file) 235 */ 236 error = exec_check_permissions(imgp); 237 if (error) { 238 vn_unlock(imgp->vp); 239 goto exec_fail_dealloc; 240 } 241 242 error = exec_map_first_page(imgp); 243 vn_unlock(imgp->vp); 244 if (error) 245 goto exec_fail_dealloc; 246 247 if (debug_execve_args && imgp->interpreted) { 248 kprintf(" target is interpreted -- recursive pass\n"); 249 kprintf(" interpreter: %s\n", imgp->interpreter_name); 250 print_execve_args(args); 251 } 252 253 /* 254 * If the current process has a special image activator it 255 * wants to try first, call it. For example, emulating shell 256 * scripts differently. 257 */ 258 error = -1; 259 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 260 error = img_first(imgp); 261 262 /* 263 * If the vnode has a registered vmspace, exec the vmspace 264 */ 265 if (error == -1 && imgp->vp->v_resident) { 266 error = exec_resident_imgact(imgp); 267 } 268 269 /* 270 * Loop through the list of image activators, calling each one. 271 * An activator returns -1 if there is no match, 0 on success, 272 * and an error otherwise. 273 */ 274 for (i = 0; error == -1 && execsw[i]; ++i) { 275 if (execsw[i]->ex_imgact == NULL || 276 execsw[i]->ex_imgact == img_first) { 277 continue; 278 } 279 error = (*execsw[i]->ex_imgact)(imgp); 280 } 281 282 if (error) { 283 if (error == -1) 284 error = ENOEXEC; 285 goto exec_fail_dealloc; 286 } 287 288 /* 289 * Special interpreter operation, cleanup and loop up to try to 290 * activate the interpreter. 291 */ 292 if (imgp->interpreted) { 293 exec_unmap_first_page(imgp); 294 nlookup_done(nd); 295 vrele(imgp->vp); 296 imgp->vp = NULL; 297 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 298 NLC_FOLLOW); 299 if (error) 300 goto exec_fail; 301 goto interpret; 302 } 303 304 /* 305 * Copy out strings (args and env) and initialize stack base 306 */ 307 stack_base = exec_copyout_strings(imgp); 308 p->p_vmspace->vm_minsaddr = (char *)stack_base; 309 310 /* 311 * If custom stack fixup routine present for this process 312 * let it do the stack setup. If we are running a resident 313 * image there is no auxinfo or other image activator context 314 * so don't try to add fixups to the stack. 315 * 316 * Else stuff argument count as first item on stack 317 */ 318 if (p->p_sysent->sv_fixup && imgp->resident == 0) 319 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 320 else 321 suword(--stack_base, imgp->args->argc); 322 323 /* 324 * For security and other reasons, the file descriptor table cannot 325 * be shared after an exec. 326 */ 327 if (p->p_fd->fd_refcnt > 1) { 328 struct filedesc *tmp; 329 330 tmp = fdcopy(p); 331 fdfree(p); 332 p->p_fd = 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 && suser(td) != 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 /* Free any previous argument cache */ 472 if (p->p_args && --p->p_args->ar_ref == 0) 473 FREE(p->p_args, M_PARGS); 474 p->p_args = NULL; 475 476 /* Cache arguments if they fit inside our allowance */ 477 i = imgp->args->begin_envv - imgp->args->begin_argv; 478 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 479 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 480 M_PARGS, M_WAITOK); 481 p->p_args->ar_ref = 1; 482 p->p_args->ar_length = i; 483 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 484 } 485 486 exec_fail_dealloc: 487 488 /* 489 * free various allocated resources 490 */ 491 if (imgp->firstpage) 492 exec_unmap_first_page(imgp); 493 494 if (imgp->vp) { 495 vrele(imgp->vp); 496 imgp->vp = NULL; 497 } 498 499 if (error == 0) { 500 ++mycpu->gd_cnt.v_exec; 501 return (0); 502 } 503 504 exec_fail: 505 /* 506 * we're done here, clear P_INEXEC if we were the ones that 507 * set it. Otherwise if vmspace_destroyed is still set we 508 * raced another thread and that thread is responsible for 509 * clearing it. 510 */ 511 if (imgp->vmspace_destroyed & 2) 512 p->p_flag &= ~P_INEXEC; 513 if (imgp->vmspace_destroyed) { 514 /* 515 * Sorry, no more process anymore. exit gracefully. 516 * However we can't die right here, because our 517 * caller might have to clean up, so indicate a 518 * lethal error by returning -1. 519 */ 520 return(-1); 521 } else { 522 return(error); 523 } 524 } 525 526 /* 527 * execve() system call. 528 */ 529 int 530 sys_execve(struct execve_args *uap) 531 { 532 struct nlookupdata nd; 533 struct image_args args; 534 int error; 535 536 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 537 if (error == 0) { 538 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 539 uap->argv, uap->envv); 540 } 541 if (error == 0) 542 error = kern_execve(&nd, &args); 543 nlookup_done(&nd); 544 exec_free_args(&args); 545 546 if (error < 0) { 547 /* We hit a lethal error condition. Let's die now. */ 548 exit1(W_EXITCODE(0, SIGABRT)); 549 /* NOTREACHED */ 550 } 551 552 /* 553 * The syscall result is returned in registers to the new program. 554 * Linux will register %edx as an atexit function and we must be 555 * sure to set it to 0. XXX 556 */ 557 if (error == 0) 558 uap->sysmsg_result64 = 0; 559 560 return (error); 561 } 562 563 int 564 exec_map_first_page(struct image_params *imgp) 565 { 566 int rv, i; 567 int initial_pagein; 568 vm_page_t ma[VM_INITIAL_PAGEIN]; 569 vm_page_t m; 570 vm_object_t object; 571 572 if (imgp->firstpage) 573 exec_unmap_first_page(imgp); 574 575 /* 576 * The file has to be mappable. 577 */ 578 if ((object = imgp->vp->v_object) == NULL) 579 return (EIO); 580 581 /* 582 * We shouldn't need protection for vm_page_grab() but we certainly 583 * need it for the lookup loop below (lookup/busy race), since 584 * an interrupt can unbusy and free the page before our busy check. 585 */ 586 crit_enter(); 587 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 588 589 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 590 ma[0] = m; 591 initial_pagein = VM_INITIAL_PAGEIN; 592 if (initial_pagein > object->size) 593 initial_pagein = object->size; 594 for (i = 1; i < initial_pagein; i++) { 595 if ((m = vm_page_lookup(object, i)) != NULL) { 596 if ((m->flags & PG_BUSY) || m->busy) 597 break; 598 if (m->valid) 599 break; 600 vm_page_busy(m); 601 } else { 602 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 603 if (m == NULL) 604 break; 605 } 606 ma[i] = m; 607 } 608 initial_pagein = i; 609 610 /* 611 * get_pages unbusies all the requested pages except the 612 * primary page (at index 0 in this case). The primary 613 * page may have been wired during the pagein (e.g. by 614 * the buffer cache) so vnode_pager_freepage() must be 615 * used to properly release it. 616 */ 617 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 618 m = vm_page_lookup(object, 0); 619 620 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 621 if (m) { 622 vm_page_protect(m, VM_PROT_NONE); 623 vnode_pager_freepage(m); 624 } 625 crit_exit(); 626 return EIO; 627 } 628 } 629 vm_page_hold(m); 630 vm_page_wakeup(m); /* unbusy the page */ 631 crit_exit(); 632 633 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 634 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 635 636 return 0; 637 } 638 639 void 640 exec_unmap_first_page(struct image_params *imgp) 641 { 642 vm_page_t m; 643 644 crit_enter(); 645 if (imgp->firstpage != NULL) { 646 m = sf_buf_page(imgp->firstpage); 647 sf_buf_free(imgp->firstpage); 648 imgp->firstpage = NULL; 649 imgp->image_header = NULL; 650 vm_page_unhold(m); 651 } 652 crit_exit(); 653 } 654 655 /* 656 * Destroy old address space, and allocate a new stack 657 * The new stack is only SGROWSIZ large because it is grown 658 * automatically in trap.c. 659 * 660 * This is the point of no return. 661 */ 662 int 663 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 664 { 665 struct vmspace *vmspace = imgp->proc->p_vmspace; 666 vm_offset_t stack_addr = USRSTACK - maxssiz; 667 struct proc *p; 668 vm_map_t map; 669 int error; 670 671 /* 672 * Indicate that we cannot gracefully error out any more, kill 673 * any other threads present, and set P_INEXEC to indicate that 674 * we are now messing with the process structure proper. 675 * 676 * If killalllwps() races return an error which coupled with 677 * vmspace_destroyed will cause us to exit. This is what we 678 * want since another thread is patiently waiting for us to exit 679 * in that case. 680 */ 681 p = curproc; 682 imgp->vmspace_destroyed = 1; 683 684 if (curthread->td_proc->p_nthreads > 1) { 685 error = killalllwps(1); 686 if (error) 687 return (error); 688 } 689 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 690 p->p_flag |= P_INEXEC; 691 692 /* 693 * Prevent a pending AIO from modifying the new address space. 694 */ 695 aio_proc_rundown(imgp->proc); 696 697 /* 698 * Blow away entire process VM, if address space not shared, 699 * otherwise, create a new VM space so that other threads are 700 * not disrupted. If we are execing a resident vmspace we 701 * create a duplicate of it and remap the stack. 702 * 703 * The exitingcnt test is not strictly necessary but has been 704 * included for code sanity (to make the code more deterministic). 705 */ 706 map = &vmspace->vm_map; 707 if (vmcopy) { 708 vmspace_exec(imgp->proc, vmcopy); 709 vmspace = imgp->proc->p_vmspace; 710 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 711 map = &vmspace->vm_map; 712 } else if (vmspace->vm_sysref.refcnt == 1 && 713 vmspace->vm_exitingcnt == 0) { 714 shmexit(vmspace); 715 if (vmspace->vm_upcalls) 716 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 717 pmap_remove_pages(vmspace_pmap(vmspace), 718 0, VM_MAX_USER_ADDRESS); 719 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 720 } else { 721 vmspace_exec(imgp->proc, NULL); 722 vmspace = imgp->proc->p_vmspace; 723 map = &vmspace->vm_map; 724 } 725 726 /* Allocate a new stack */ 727 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 728 VM_PROT_ALL, VM_PROT_ALL, 0); 729 if (error) 730 return (error); 731 732 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 733 * VM_STACK case, but they are still used to monitor the size of the 734 * process stack so we can check the stack rlimit. 735 */ 736 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 737 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 738 739 return(0); 740 } 741 742 /* 743 * Copy out argument and environment strings from the old process 744 * address space into the temporary string buffer. 745 */ 746 int 747 exec_copyin_args(struct image_args *args, char *fname, 748 enum exec_path_segflg segflg, char **argv, char **envv) 749 { 750 char *argp, *envp; 751 int error = 0; 752 size_t length; 753 754 bzero(args, sizeof(*args)); 755 756 args->buf = objcache_get(exec_objcache, M_WAITOK); 757 if (args->buf == NULL) 758 return (ENOMEM); 759 args->begin_argv = args->buf; 760 args->endp = args->begin_argv; 761 args->space = ARG_MAX; 762 763 args->fname = args->buf + ARG_MAX; 764 765 /* 766 * Copy the file name. 767 */ 768 if (segflg == PATH_SYSSPACE) { 769 error = copystr(fname, args->fname, PATH_MAX, &length); 770 } else if (segflg == PATH_USERSPACE) { 771 error = copyinstr(fname, args->fname, PATH_MAX, &length); 772 } 773 774 /* 775 * Extract argument strings. argv may not be NULL. The argv 776 * array is terminated by a NULL entry. We special-case the 777 * situation where argv[0] is NULL by passing { filename, NULL } 778 * to the new program to guarentee that the interpreter knows what 779 * file to open in case we exec an interpreted file. Note that 780 * a NULL argv[0] terminates the argv[] array. 781 * 782 * XXX the special-casing of argv[0] is historical and needs to be 783 * revisited. 784 */ 785 if (argv == NULL) 786 error = EFAULT; 787 if (error == 0) { 788 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 789 if (argp == (caddr_t)-1) { 790 error = EFAULT; 791 break; 792 } 793 error = copyinstr(argp, args->endp, 794 args->space, &length); 795 if (error) { 796 if (error == ENAMETOOLONG) 797 error = E2BIG; 798 break; 799 } 800 args->space -= length; 801 args->endp += length; 802 args->argc++; 803 } 804 if (args->argc == 0 && error == 0) { 805 length = strlen(args->fname) + 1; 806 if (length > args->space) { 807 error = E2BIG; 808 } else { 809 bcopy(args->fname, args->endp, length); 810 args->space -= length; 811 args->endp += length; 812 args->argc++; 813 } 814 } 815 } 816 817 args->begin_envv = args->endp; 818 819 /* 820 * extract environment strings. envv may be NULL. 821 */ 822 if (envv && error == 0) { 823 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 824 if (envp == (caddr_t) -1) { 825 error = EFAULT; 826 break; 827 } 828 error = copyinstr(envp, args->endp, args->space, 829 &length); 830 if (error) { 831 if (error == ENAMETOOLONG) 832 error = E2BIG; 833 break; 834 } 835 args->space -= length; 836 args->endp += length; 837 args->envc++; 838 } 839 } 840 return (error); 841 } 842 843 void 844 exec_free_args(struct image_args *args) 845 { 846 if (args->buf) { 847 objcache_put(exec_objcache, args->buf); 848 args->buf = NULL; 849 } 850 } 851 852 /* 853 * Copy strings out to the new process address space, constructing 854 * new arg and env vector tables. Return a pointer to the base 855 * so that it can be used as the initial stack pointer. 856 */ 857 register_t * 858 exec_copyout_strings(struct image_params *imgp) 859 { 860 int argc, envc, sgap; 861 char **vectp; 862 char *stringp, *destp; 863 register_t *stack_base; 864 struct ps_strings *arginfo; 865 int szsigcode; 866 867 /* 868 * Calculate string base and vector table pointers. 869 * Also deal with signal trampoline code for this exec type. 870 */ 871 arginfo = (struct ps_strings *)PS_STRINGS; 872 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 873 if (stackgap_random != 0) 874 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 875 else 876 sgap = 0; 877 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 878 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 879 880 /* 881 * install sigcode 882 */ 883 if (szsigcode) 884 copyout(imgp->proc->p_sysent->sv_sigcode, 885 ((caddr_t)arginfo - szsigcode), szsigcode); 886 887 /* 888 * If we have a valid auxargs ptr, prepare some room 889 * on the stack. 890 * 891 * The '+ 2' is for the null pointers at the end of each of the 892 * arg and env vector sets, and 'AT_COUNT*2' is room for the 893 * ELF Auxargs data. 894 */ 895 if (imgp->auxargs) { 896 vectp = (char **)(destp - (imgp->args->argc + 897 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 898 } else { 899 vectp = (char **)(destp - (imgp->args->argc + 900 imgp->args->envc + 2) * sizeof(char*)); 901 } 902 903 /* 904 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 905 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 906 */ 907 908 /* 909 * vectp also becomes our initial stack base 910 */ 911 stack_base = (register_t *)vectp; 912 913 stringp = imgp->args->begin_argv; 914 argc = imgp->args->argc; 915 envc = imgp->args->envc; 916 917 /* 918 * Copy out strings - arguments and environment. 919 */ 920 copyout(stringp, destp, ARG_MAX - imgp->args->space); 921 922 /* 923 * Fill in "ps_strings" struct for ps, w, etc. 924 */ 925 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 926 suword(&arginfo->ps_nargvstr, argc); 927 928 /* 929 * Fill in argument portion of vector table. 930 */ 931 for (; argc > 0; --argc) { 932 suword(vectp++, (long)(intptr_t)destp); 933 while (*stringp++ != 0) 934 destp++; 935 destp++; 936 } 937 938 /* a null vector table pointer separates the argp's from the envp's */ 939 suword(vectp++, 0); 940 941 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 942 suword(&arginfo->ps_nenvstr, envc); 943 944 /* 945 * Fill in environment portion of vector table. 946 */ 947 for (; envc > 0; --envc) { 948 suword(vectp++, (long)(intptr_t)destp); 949 while (*stringp++ != 0) 950 destp++; 951 destp++; 952 } 953 954 /* end of vector table is a null pointer */ 955 suword(vectp, 0); 956 957 return (stack_base); 958 } 959 960 /* 961 * Check permissions of file to execute. 962 * Return 0 for success or error code on failure. 963 */ 964 int 965 exec_check_permissions(struct image_params *imgp) 966 { 967 struct proc *p = imgp->proc; 968 struct vnode *vp = imgp->vp; 969 struct vattr *attr = imgp->attr; 970 int error; 971 972 /* Get file attributes */ 973 error = VOP_GETATTR(vp, attr); 974 if (error) 975 return (error); 976 977 /* 978 * 1) Check if file execution is disabled for the filesystem that this 979 * file resides on. 980 * 2) Insure that at least one execute bit is on - otherwise root 981 * will always succeed, and we don't want to happen unless the 982 * file really is executable. 983 * 3) Insure that the file is a regular file. 984 */ 985 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 986 ((attr->va_mode & 0111) == 0) || 987 (attr->va_type != VREG)) { 988 return (EACCES); 989 } 990 991 /* 992 * Zero length files can't be exec'd 993 */ 994 if (attr->va_size == 0) 995 return (ENOEXEC); 996 997 /* 998 * Check for execute permission to file based on current credentials. 999 */ 1000 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 1001 if (error) 1002 return (error); 1003 1004 /* 1005 * Check number of open-for-writes on the file and deny execution 1006 * if there are any. 1007 */ 1008 if (vp->v_writecount) 1009 return (ETXTBSY); 1010 1011 /* 1012 * Call filesystem specific open routine, which allows us to read, 1013 * write, and mmap the file. Without the VOP_OPEN we can only 1014 * stat the file. 1015 */ 1016 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1017 if (error) 1018 return (error); 1019 1020 return (0); 1021 } 1022 1023 /* 1024 * Exec handler registration 1025 */ 1026 int 1027 exec_register(const struct execsw *execsw_arg) 1028 { 1029 const struct execsw **es, **xs, **newexecsw; 1030 int count = 2; /* New slot and trailing NULL */ 1031 1032 if (execsw) 1033 for (es = execsw; *es; es++) 1034 count++; 1035 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1036 xs = newexecsw; 1037 if (execsw) 1038 for (es = execsw; *es; es++) 1039 *xs++ = *es; 1040 *xs++ = execsw_arg; 1041 *xs = NULL; 1042 if (execsw) 1043 kfree(execsw, M_TEMP); 1044 execsw = newexecsw; 1045 return 0; 1046 } 1047 1048 int 1049 exec_unregister(const struct execsw *execsw_arg) 1050 { 1051 const struct execsw **es, **xs, **newexecsw; 1052 int count = 1; 1053 1054 if (execsw == NULL) 1055 panic("unregister with no handlers left?"); 1056 1057 for (es = execsw; *es; es++) { 1058 if (*es == execsw_arg) 1059 break; 1060 } 1061 if (*es == NULL) 1062 return ENOENT; 1063 for (es = execsw; *es; es++) 1064 if (*es != execsw_arg) 1065 count++; 1066 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1067 xs = newexecsw; 1068 for (es = execsw; *es; es++) 1069 if (*es != execsw_arg) 1070 *xs++ = *es; 1071 *xs = NULL; 1072 if (execsw) 1073 kfree(execsw, M_TEMP); 1074 execsw = newexecsw; 1075 return 0; 1076 } 1077