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 1.56 2007/04/29 18:25:34 dillon 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/vm_pager.h> 68 69 #include <sys/user.h> 70 #include <sys/reg.h> 71 72 #include <sys/thread2.h> 73 74 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 75 76 static register_t *exec_copyout_strings (struct image_params *); 77 78 /* XXX This should be vm_size_t. */ 79 static u_long ps_strings = PS_STRINGS; 80 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 81 82 /* XXX This should be vm_size_t. */ 83 static u_long usrstack = USRSTACK; 84 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 85 86 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 87 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 88 &ps_arg_cache_limit, 0, ""); 89 90 int ps_argsopen = 1; 91 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 92 93 void print_execve_args(struct image_args *args); 94 int debug_execve_args = 0; 95 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 96 0, ""); 97 98 /* 99 * stackgap_random specifies if the stackgap should have a random size added 100 * to it. It must be a power of 2. If non-zero, the stack gap will be 101 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 102 */ 103 static int stackgap_random = 1024; 104 static int 105 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 106 { 107 int error, new_val; 108 new_val = stackgap_random; 109 error = sysctl_handle_int(oidp, &new_val, 0, req); 110 if (error != 0 || req->newptr == NULL) 111 return (error); 112 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 113 return (EINVAL); 114 stackgap_random = new_val; 115 116 return(0); 117 } 118 119 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 120 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 121 122 void 123 print_execve_args(struct image_args *args) 124 { 125 char *cp; 126 int ndx; 127 128 cp = args->begin_argv; 129 for (ndx = 0; ndx < args->argc; ndx++) { 130 kprintf("\targv[%d]: %s\n", ndx, cp); 131 while (*cp++ != '\0'); 132 } 133 for (ndx = 0; ndx < args->envc; ndx++) { 134 kprintf("\tenvv[%d]: %s\n", ndx, cp); 135 while (*cp++ != '\0'); 136 } 137 } 138 139 /* 140 * Each of the items is a pointer to a `const struct execsw', hence the 141 * double pointer here. 142 */ 143 static const struct execsw **execsw; 144 145 int 146 kern_execve(struct nlookupdata *nd, struct image_args *args) 147 { 148 struct thread *td = curthread; 149 struct lwp *lp = td->td_lwp; 150 struct proc *p = td->td_proc; 151 register_t *stack_base; 152 int error, len, i; 153 struct image_params image_params, *imgp; 154 struct vattr attr; 155 int (*img_first) (struct image_params *); 156 157 if (debug_execve_args) { 158 kprintf("%s()\n", __func__); 159 print_execve_args(args); 160 } 161 162 KKASSERT(p); 163 imgp = &image_params; 164 165 /* 166 * Lock the process and set the P_INEXEC flag to indicate that 167 * it should be left alone until we're done here. This is 168 * necessary to avoid race conditions - e.g. in ptrace() - 169 * that might allow a local user to illicitly obtain elevated 170 * privileges. 171 */ 172 p->p_flag |= P_INEXEC; 173 174 /* 175 * Initialize part of the common data 176 */ 177 imgp->proc = p; 178 imgp->args = args; 179 imgp->attr = &attr; 180 imgp->entry_addr = 0; 181 imgp->resident = 0; 182 imgp->vmspace_destroyed = 0; 183 imgp->interpreted = 0; 184 imgp->interpreter_name[0] = 0; 185 imgp->auxargs = NULL; 186 imgp->vp = NULL; 187 imgp->firstpage = NULL; 188 imgp->ps_strings = 0; 189 imgp->image_header = NULL; 190 191 interpret: 192 193 /* 194 * Translate the file name to a vnode. Unlock the cache entry to 195 * improve parallelism for programs exec'd in parallel. 196 */ 197 if ((error = nlookup(nd)) != 0) 198 goto exec_fail; 199 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 200 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 201 nd->nl_flags &= ~NLC_NCPISLOCKED; 202 cache_unlock(&nd->nl_nch); 203 if (error) 204 goto exec_fail; 205 206 /* 207 * Check file permissions (also 'opens' file) 208 */ 209 error = exec_check_permissions(imgp); 210 if (error) { 211 vn_unlock(imgp->vp); 212 goto exec_fail_dealloc; 213 } 214 215 error = exec_map_first_page(imgp); 216 vn_unlock(imgp->vp); 217 if (error) 218 goto exec_fail_dealloc; 219 220 if (debug_execve_args && imgp->interpreted) { 221 kprintf(" target is interpreted -- recursive pass\n"); 222 kprintf(" interpreter: %s\n", imgp->interpreter_name); 223 print_execve_args(args); 224 } 225 226 /* 227 * If the current process has a special image activator it 228 * wants to try first, call it. For example, emulating shell 229 * scripts differently. 230 */ 231 error = -1; 232 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 233 error = img_first(imgp); 234 235 /* 236 * If the vnode has a registered vmspace, exec the vmspace 237 */ 238 if (error == -1 && imgp->vp->v_resident) { 239 error = exec_resident_imgact(imgp); 240 } 241 242 /* 243 * Loop through the list of image activators, calling each one. 244 * An activator returns -1 if there is no match, 0 on success, 245 * and an error otherwise. 246 */ 247 for (i = 0; error == -1 && execsw[i]; ++i) { 248 if (execsw[i]->ex_imgact == NULL || 249 execsw[i]->ex_imgact == img_first) { 250 continue; 251 } 252 error = (*execsw[i]->ex_imgact)(imgp); 253 } 254 255 if (error) { 256 if (error == -1) 257 error = ENOEXEC; 258 goto exec_fail_dealloc; 259 } 260 261 /* 262 * Special interpreter operation, cleanup and loop up to try to 263 * activate the interpreter. 264 */ 265 if (imgp->interpreted) { 266 exec_unmap_first_page(imgp); 267 nlookup_done(nd); 268 vrele(imgp->vp); 269 imgp->vp = NULL; 270 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 271 NLC_FOLLOW); 272 if (error) 273 goto exec_fail; 274 goto interpret; 275 } 276 277 /* 278 * Copy out strings (args and env) and initialize stack base 279 */ 280 stack_base = exec_copyout_strings(imgp); 281 p->p_vmspace->vm_minsaddr = (char *)stack_base; 282 283 /* 284 * If custom stack fixup routine present for this process 285 * let it do the stack setup. If we are running a resident 286 * image there is no auxinfo or other image activator context 287 * so don't try to add fixups to the stack. 288 * 289 * Else stuff argument count as first item on stack 290 */ 291 if (p->p_sysent->sv_fixup && imgp->resident == 0) 292 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 293 else 294 suword(--stack_base, imgp->args->argc); 295 296 /* 297 * For security and other reasons, the file descriptor table cannot 298 * be shared after an exec. 299 */ 300 if (p->p_fd->fd_refcnt > 1) { 301 struct filedesc *tmp; 302 303 tmp = fdcopy(p); 304 fdfree(p); 305 p->p_fd = tmp; 306 } 307 308 /* 309 * For security and other reasons, signal handlers cannot 310 * be shared after an exec. The new proces gets a copy of the old 311 * handlers. In execsigs(), the new process will have its signals 312 * reset. 313 */ 314 if (p->p_sigacts->ps_refcnt > 1) { 315 struct sigacts *newsigacts; 316 317 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts), 318 M_SUBPROC, M_WAITOK); 319 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts)); 320 p->p_sigacts->ps_refcnt--; 321 p->p_sigacts = newsigacts; 322 p->p_sigacts->ps_refcnt = 1; 323 } 324 325 /* 326 * For security and other reasons virtual kernels cannot be 327 * inherited by an exec. This also allows a virtual kernel 328 * to fork/exec unrelated applications. 329 */ 330 if (p->p_vkernel) 331 vkernel_exit(p); 332 333 /* Stop profiling */ 334 stopprofclock(p); 335 336 /* close files on exec */ 337 fdcloseexec(p); 338 339 /* reset caught signals */ 340 execsigs(p); 341 342 /* name this process - nameiexec(p, ndp) */ 343 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 344 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 345 p->p_comm[len] = 0; 346 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 347 348 /* 349 * mark as execed, wakeup the process that vforked (if any) and tell 350 * it that it now has its own resources back 351 */ 352 p->p_flag |= P_EXEC; 353 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 354 p->p_flag &= ~P_PPWAIT; 355 wakeup((caddr_t)p->p_pptr); 356 } 357 358 /* 359 * Implement image setuid/setgid. 360 * 361 * Don't honor setuid/setgid if the filesystem prohibits it or if 362 * the process is being traced. 363 */ 364 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 365 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 366 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 367 (p->p_flag & P_TRACED) == 0) { 368 /* 369 * Turn off syscall tracing for set-id programs, except for 370 * root. Record any set-id flags first to make sure that 371 * we do not regain any tracing during a possible block. 372 */ 373 setsugid(); 374 if (p->p_tracenode && suser(td) != 0) { 375 ktrdestroy(&p->p_tracenode); 376 p->p_traceflag = 0; 377 } 378 /* Close any file descriptors 0..2 that reference procfs */ 379 setugidsafety(p); 380 /* Make sure file descriptors 0..2 are in use. */ 381 error = fdcheckstd(p); 382 if (error != 0) 383 goto exec_fail_dealloc; 384 /* 385 * Set the new credentials. 386 */ 387 cratom(&p->p_ucred); 388 if (attr.va_mode & VSUID) 389 change_euid(attr.va_uid); 390 if (attr.va_mode & VSGID) 391 p->p_ucred->cr_gid = attr.va_gid; 392 393 /* 394 * Clear local varsym variables 395 */ 396 varsymset_clean(&p->p_varsymset); 397 } else { 398 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 399 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 400 p->p_flag &= ~P_SUGID; 401 } 402 403 /* 404 * Implement correct POSIX saved-id behavior. 405 */ 406 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 407 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 408 cratom(&p->p_ucred); 409 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 410 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 411 } 412 413 /* 414 * Store the vp for use in procfs 415 */ 416 if (p->p_textvp) /* release old reference */ 417 vrele(p->p_textvp); 418 p->p_textvp = imgp->vp; 419 vref(p->p_textvp); 420 421 /* 422 * Notify others that we exec'd, and clear the P_INEXEC flag 423 * as we're now a bona fide freshly-execed process. 424 */ 425 KNOTE(&p->p_klist, NOTE_EXEC); 426 p->p_flag &= ~P_INEXEC; 427 428 /* 429 * If tracing the process, trap to debugger so breakpoints 430 * can be set before the program executes. 431 */ 432 STOPEVENT(p, S_EXEC, 0); 433 434 if (p->p_flag & P_TRACED) 435 ksignal(p, SIGTRAP); 436 437 /* clear "fork but no exec" flag, as we _are_ execing */ 438 p->p_acflag &= ~AFORK; 439 440 /* Set values passed into the program in registers. */ 441 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 442 imgp->ps_strings); 443 444 /* Free any previous argument cache */ 445 if (p->p_args && --p->p_args->ar_ref == 0) 446 FREE(p->p_args, M_PARGS); 447 p->p_args = NULL; 448 449 /* Cache arguments if they fit inside our allowance */ 450 i = imgp->args->begin_envv - imgp->args->begin_argv; 451 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 452 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 453 M_PARGS, M_WAITOK); 454 p->p_args->ar_ref = 1; 455 p->p_args->ar_length = i; 456 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 457 } 458 459 exec_fail_dealloc: 460 461 /* 462 * free various allocated resources 463 */ 464 if (imgp->firstpage) 465 exec_unmap_first_page(imgp); 466 467 if (imgp->vp) { 468 vrele(imgp->vp); 469 imgp->vp = NULL; 470 } 471 472 if (error == 0) { 473 ++mycpu->gd_cnt.v_exec; 474 return (0); 475 } 476 477 exec_fail: 478 /* we're done here, clear P_INEXEC */ 479 p->p_flag &= ~P_INEXEC; 480 if (imgp->vmspace_destroyed) { 481 /* sorry, no more process anymore. exit gracefully */ 482 exit1(W_EXITCODE(0, SIGABRT)); 483 /* NOT REACHED */ 484 return(0); 485 } else { 486 return(error); 487 } 488 } 489 490 /* 491 * execve() system call. 492 */ 493 int 494 sys_execve(struct execve_args *uap) 495 { 496 struct nlookupdata nd; 497 struct image_args args; 498 int error; 499 500 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 501 if (error == 0) { 502 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 503 uap->argv, uap->envv); 504 } 505 if (error == 0) 506 error = kern_execve(&nd, &args); 507 nlookup_done(&nd); 508 exec_free_args(&args); 509 510 /* 511 * The syscall result is returned in registers to the new program. 512 * Linux will register %edx as an atexit function and we must be 513 * sure to set it to 0. XXX 514 */ 515 if (error == 0) 516 uap->sysmsg_result64 = 0; 517 518 return (error); 519 } 520 521 int 522 exec_map_first_page(struct image_params *imgp) 523 { 524 int rv, i; 525 int initial_pagein; 526 vm_page_t ma[VM_INITIAL_PAGEIN]; 527 vm_page_t m; 528 vm_object_t object; 529 530 if (imgp->firstpage) 531 exec_unmap_first_page(imgp); 532 533 /* 534 * The file has to be mappable. 535 */ 536 if ((object = imgp->vp->v_object) == NULL) 537 return (EIO); 538 539 /* 540 * We shouldn't need protection for vm_page_grab() but we certainly 541 * need it for the lookup loop below (lookup/busy race), since 542 * an interrupt can unbusy and free the page before our busy check. 543 */ 544 crit_enter(); 545 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 546 547 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 548 ma[0] = m; 549 initial_pagein = VM_INITIAL_PAGEIN; 550 if (initial_pagein > object->size) 551 initial_pagein = object->size; 552 for (i = 1; i < initial_pagein; i++) { 553 if ((m = vm_page_lookup(object, i)) != NULL) { 554 if ((m->flags & PG_BUSY) || m->busy) 555 break; 556 if (m->valid) 557 break; 558 vm_page_busy(m); 559 } else { 560 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 561 if (m == NULL) 562 break; 563 } 564 ma[i] = m; 565 } 566 initial_pagein = i; 567 568 /* 569 * get_pages unbusies all the requested pages except the 570 * primary page (at index 0 in this case). 571 */ 572 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 573 m = vm_page_lookup(object, 0); 574 575 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 576 if (m) { 577 vm_page_protect(m, VM_PROT_NONE); 578 vm_page_free(m); 579 } 580 crit_exit(); 581 return EIO; 582 } 583 } 584 vm_page_hold(m); 585 vm_page_wakeup(m); /* unbusy the page */ 586 crit_exit(); 587 588 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 589 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 590 591 return 0; 592 } 593 594 void 595 exec_unmap_first_page(struct image_params *imgp) 596 { 597 vm_page_t m; 598 599 crit_enter(); 600 if (imgp->firstpage != NULL) { 601 m = sf_buf_page(imgp->firstpage); 602 sf_buf_free(imgp->firstpage); 603 imgp->firstpage = NULL; 604 imgp->image_header = NULL; 605 vm_page_unhold(m); 606 } 607 crit_exit(); 608 } 609 610 /* 611 * Destroy old address space, and allocate a new stack 612 * The new stack is only SGROWSIZ large because it is grown 613 * automatically in trap.c. 614 */ 615 int 616 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 617 { 618 int error; 619 struct vmspace *vmspace = imgp->proc->p_vmspace; 620 vm_offset_t stack_addr = USRSTACK - maxssiz; 621 vm_map_t map; 622 623 imgp->vmspace_destroyed = 1; 624 625 if (curthread->td_proc->p_nthreads > 1) 626 killlwps(curthread->td_lwp); 627 628 /* 629 * Prevent a pending AIO from modifying the new address space. 630 */ 631 aio_proc_rundown(imgp->proc); 632 633 /* 634 * Blow away entire process VM, if address space not shared, 635 * otherwise, create a new VM space so that other threads are 636 * not disrupted. If we are execing a resident vmspace we 637 * create a duplicate of it and remap the stack. 638 * 639 * The exitingcnt test is not strictly necessary but has been 640 * included for code sanity (to make the code more deterministic). 641 */ 642 map = &vmspace->vm_map; 643 if (vmcopy) { 644 vmspace_exec(imgp->proc, vmcopy); 645 vmspace = imgp->proc->p_vmspace; 646 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 647 map = &vmspace->vm_map; 648 } else if (vmspace->vm_sysref.refcnt == 1 && 649 vmspace->vm_exitingcnt == 0) { 650 shmexit(vmspace); 651 if (vmspace->vm_upcalls) 652 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 653 pmap_remove_pages(vmspace_pmap(vmspace), 654 0, VM_MAX_USER_ADDRESS); 655 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 656 } else { 657 vmspace_exec(imgp->proc, NULL); 658 vmspace = imgp->proc->p_vmspace; 659 map = &vmspace->vm_map; 660 } 661 662 /* Allocate a new stack */ 663 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 664 VM_PROT_ALL, VM_PROT_ALL, 0); 665 if (error) 666 return (error); 667 668 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 669 * VM_STACK case, but they are still used to monitor the size of the 670 * process stack so we can check the stack rlimit. 671 */ 672 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 673 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 674 675 return(0); 676 } 677 678 /* 679 * Copy out argument and environment strings from the old process 680 * address space into the temporary string buffer. 681 */ 682 int 683 exec_copyin_args(struct image_args *args, char *fname, 684 enum exec_path_segflg segflg, char **argv, char **envv) 685 { 686 char *argp, *envp; 687 int error = 0; 688 size_t length; 689 690 bzero(args, sizeof(*args)); 691 args->buf = (char *) kmem_alloc_wait(&exec_map, PATH_MAX + ARG_MAX); 692 if (args->buf == NULL) 693 return (ENOMEM); 694 args->begin_argv = args->buf; 695 args->endp = args->begin_argv; 696 args->space = ARG_MAX; 697 698 args->fname = args->buf + ARG_MAX; 699 700 /* 701 * Copy the file name. 702 */ 703 if (segflg == PATH_SYSSPACE) { 704 error = copystr(fname, args->fname, PATH_MAX, &length); 705 } else if (segflg == PATH_USERSPACE) { 706 error = copyinstr(fname, args->fname, PATH_MAX, &length); 707 } 708 709 /* 710 * Extract argument strings. argv may not be NULL. The argv 711 * array is terminated by a NULL entry. We special-case the 712 * situation where argv[0] is NULL by passing { filename, NULL } 713 * to the new program to guarentee that the interpreter knows what 714 * file to open in case we exec an interpreted file. Note that 715 * a NULL argv[0] terminates the argv[] array. 716 * 717 * XXX the special-casing of argv[0] is historical and needs to be 718 * revisited. 719 */ 720 if (argv == NULL) 721 error = EFAULT; 722 if (error == 0) { 723 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 724 if (argp == (caddr_t)-1) { 725 error = EFAULT; 726 break; 727 } 728 error = copyinstr(argp, args->endp, 729 args->space, &length); 730 if (error) { 731 if (error == ENAMETOOLONG) 732 error = E2BIG; 733 break; 734 } 735 args->space -= length; 736 args->endp += length; 737 args->argc++; 738 } 739 if (args->argc == 0 && error == 0) { 740 length = strlen(args->fname) + 1; 741 if (length > args->space) { 742 error = E2BIG; 743 } else { 744 bcopy(args->fname, args->endp, length); 745 args->space -= length; 746 args->endp += length; 747 args->argc++; 748 } 749 } 750 } 751 752 args->begin_envv = args->endp; 753 754 /* 755 * extract environment strings. envv may be NULL. 756 */ 757 if (envv && error == 0) { 758 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 759 if (envp == (caddr_t) -1) { 760 error = EFAULT; 761 break; 762 } 763 error = copyinstr(envp, args->endp, args->space, 764 &length); 765 if (error) { 766 if (error == ENAMETOOLONG) 767 error = E2BIG; 768 break; 769 } 770 args->space -= length; 771 args->endp += length; 772 args->envc++; 773 } 774 } 775 return (error); 776 } 777 778 void 779 exec_free_args(struct image_args *args) 780 { 781 if (args->buf) { 782 kmem_free_wakeup(&exec_map, 783 (vm_offset_t)args->buf, PATH_MAX + ARG_MAX); 784 args->buf = NULL; 785 } 786 } 787 788 /* 789 * Copy strings out to the new process address space, constructing 790 * new arg and env vector tables. Return a pointer to the base 791 * so that it can be used as the initial stack pointer. 792 */ 793 register_t * 794 exec_copyout_strings(struct image_params *imgp) 795 { 796 int argc, envc, sgap; 797 char **vectp; 798 char *stringp, *destp; 799 register_t *stack_base; 800 struct ps_strings *arginfo; 801 int szsigcode; 802 803 /* 804 * Calculate string base and vector table pointers. 805 * Also deal with signal trampoline code for this exec type. 806 */ 807 arginfo = (struct ps_strings *)PS_STRINGS; 808 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 809 if (stackgap_random != 0) 810 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 811 else 812 sgap = 0; 813 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 814 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 815 816 /* 817 * install sigcode 818 */ 819 if (szsigcode) 820 copyout(imgp->proc->p_sysent->sv_sigcode, 821 ((caddr_t)arginfo - szsigcode), szsigcode); 822 823 /* 824 * If we have a valid auxargs ptr, prepare some room 825 * on the stack. 826 * 827 * The '+ 2' is for the null pointers at the end of each of the 828 * arg and env vector sets, and 'AT_COUNT*2' is room for the 829 * ELF Auxargs data. 830 */ 831 if (imgp->auxargs) { 832 vectp = (char **)(destp - (imgp->args->argc + 833 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 834 } else { 835 vectp = (char **)(destp - (imgp->args->argc + 836 imgp->args->envc + 2) * sizeof(char*)); 837 } 838 839 /* 840 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 841 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 842 */ 843 844 /* 845 * vectp also becomes our initial stack base 846 */ 847 stack_base = (register_t *)vectp; 848 849 stringp = imgp->args->begin_argv; 850 argc = imgp->args->argc; 851 envc = imgp->args->envc; 852 853 /* 854 * Copy out strings - arguments and environment. 855 */ 856 copyout(stringp, destp, ARG_MAX - imgp->args->space); 857 858 /* 859 * Fill in "ps_strings" struct for ps, w, etc. 860 */ 861 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 862 suword(&arginfo->ps_nargvstr, argc); 863 864 /* 865 * Fill in argument portion of vector table. 866 */ 867 for (; argc > 0; --argc) { 868 suword(vectp++, (long)(intptr_t)destp); 869 while (*stringp++ != 0) 870 destp++; 871 destp++; 872 } 873 874 /* a null vector table pointer separates the argp's from the envp's */ 875 suword(vectp++, 0); 876 877 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 878 suword(&arginfo->ps_nenvstr, envc); 879 880 /* 881 * Fill in environment portion of vector table. 882 */ 883 for (; envc > 0; --envc) { 884 suword(vectp++, (long)(intptr_t)destp); 885 while (*stringp++ != 0) 886 destp++; 887 destp++; 888 } 889 890 /* end of vector table is a null pointer */ 891 suword(vectp, 0); 892 893 return (stack_base); 894 } 895 896 /* 897 * Check permissions of file to execute. 898 * Return 0 for success or error code on failure. 899 */ 900 int 901 exec_check_permissions(struct image_params *imgp) 902 { 903 struct proc *p = imgp->proc; 904 struct vnode *vp = imgp->vp; 905 struct vattr *attr = imgp->attr; 906 int error; 907 908 /* Get file attributes */ 909 error = VOP_GETATTR(vp, attr); 910 if (error) 911 return (error); 912 913 /* 914 * 1) Check if file execution is disabled for the filesystem that this 915 * file resides on. 916 * 2) Insure that at least one execute bit is on - otherwise root 917 * will always succeed, and we don't want to happen unless the 918 * file really is executable. 919 * 3) Insure that the file is a regular file. 920 */ 921 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 922 ((attr->va_mode & 0111) == 0) || 923 (attr->va_type != VREG)) { 924 return (EACCES); 925 } 926 927 /* 928 * Zero length files can't be exec'd 929 */ 930 if (attr->va_size == 0) 931 return (ENOEXEC); 932 933 /* 934 * Check for execute permission to file based on current credentials. 935 */ 936 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 937 if (error) 938 return (error); 939 940 /* 941 * Check number of open-for-writes on the file and deny execution 942 * if there are any. 943 */ 944 if (vp->v_writecount) 945 return (ETXTBSY); 946 947 /* 948 * Call filesystem specific open routine, which allows us to read, 949 * write, and mmap the file. Without the VOP_OPEN we can only 950 * stat the file. 951 */ 952 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 953 if (error) 954 return (error); 955 956 return (0); 957 } 958 959 /* 960 * Exec handler registration 961 */ 962 int 963 exec_register(const struct execsw *execsw_arg) 964 { 965 const struct execsw **es, **xs, **newexecsw; 966 int count = 2; /* New slot and trailing NULL */ 967 968 if (execsw) 969 for (es = execsw; *es; es++) 970 count++; 971 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 972 if (newexecsw == NULL) 973 return ENOMEM; 974 xs = newexecsw; 975 if (execsw) 976 for (es = execsw; *es; es++) 977 *xs++ = *es; 978 *xs++ = execsw_arg; 979 *xs = NULL; 980 if (execsw) 981 kfree(execsw, M_TEMP); 982 execsw = newexecsw; 983 return 0; 984 } 985 986 int 987 exec_unregister(const struct execsw *execsw_arg) 988 { 989 const struct execsw **es, **xs, **newexecsw; 990 int count = 1; 991 992 if (execsw == NULL) 993 panic("unregister with no handlers left?"); 994 995 for (es = execsw; *es; es++) { 996 if (*es == execsw_arg) 997 break; 998 } 999 if (*es == NULL) 1000 return ENOENT; 1001 for (es = execsw; *es; es++) 1002 if (*es != execsw_arg) 1003 count++; 1004 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1005 if (newexecsw == NULL) 1006 return ENOMEM; 1007 xs = newexecsw; 1008 for (es = execsw; *es; es++) 1009 if (*es != execsw_arg) 1010 *xs++ = *es; 1011 *xs = NULL; 1012 if (execsw) 1013 kfree(execsw, M_TEMP); 1014 execsw = newexecsw; 1015 return 0; 1016 } 1017