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 1.33 2005/06/22 19:40:35 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/signalvar.h> 46 #include <sys/pioctl.h> 47 #include <sys/nlookup.h> 48 #include <sys/sfbuf.h> 49 #include <sys/sysent.h> 50 #include <sys/shm.h> 51 #include <sys/sysctl.h> 52 #include <sys/vnode.h> 53 #include <sys/vmmeter.h> 54 #include <sys/aio.h> 55 #include <sys/libkern.h> 56 57 #include <vm/vm.h> 58 #include <vm/vm_param.h> 59 #include <sys/lock.h> 60 #include <vm/pmap.h> 61 #include <vm/vm_page.h> 62 #include <vm/vm_map.h> 63 #include <vm/vm_kern.h> 64 #include <vm/vm_extern.h> 65 #include <vm/vm_object.h> 66 #include <vm/vm_pager.h> 67 68 #include <sys/user.h> 69 #include <machine/reg.h> 70 71 #include <sys/thread2.h> 72 73 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 74 75 static register_t *exec_copyout_strings (struct image_params *); 76 77 /* XXX This should be vm_size_t. */ 78 static u_long ps_strings = PS_STRINGS; 79 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 80 81 /* XXX This should be vm_size_t. */ 82 static u_long usrstack = USRSTACK; 83 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 84 85 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 86 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 87 &ps_arg_cache_limit, 0, ""); 88 89 int ps_argsopen = 1; 90 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 91 92 void print_execve_args(struct image_args *args); 93 int debug_execve_args = 0; 94 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 95 0, ""); 96 97 /* 98 * stackgap_random specifies if the stackgap should have a random size added 99 * to it. It must be a power of 2. If non-zero, the stack gap will be 100 * calculated as: ALIGN(arc4random() & (stackgap_random - 1)). 101 */ 102 static int stackgap_random = 1024; 103 static int 104 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 105 { 106 int error, new_val; 107 new_val = stackgap_random; 108 error = sysctl_handle_int(oidp, &new_val, 0, req); 109 if (error != 0 || req->newptr == NULL) 110 return (error); 111 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 112 return (EINVAL); 113 stackgap_random = new_val; 114 115 return(0); 116 } 117 118 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 119 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 120 121 void 122 print_execve_args(struct image_args *args) 123 { 124 char *cp; 125 int ndx; 126 127 cp = args->begin_argv; 128 for (ndx = 0; ndx < args->argc; ndx++) { 129 printf("\targv[%d]: %s\n", ndx, cp); 130 while (*cp++ != '\0'); 131 } 132 for (ndx = 0; ndx < args->envc; ndx++) { 133 printf("\tenvv[%d]: %s\n", ndx, cp); 134 while (*cp++ != '\0'); 135 } 136 } 137 138 /* 139 * Each of the items is a pointer to a `const struct execsw', hence the 140 * double pointer here. 141 */ 142 static const struct execsw **execsw; 143 144 int 145 kern_execve(struct nlookupdata *nd, struct image_args *args) 146 { 147 struct thread *td = curthread; 148 struct proc *p = td->td_proc; 149 register_t *stack_base; 150 int error, len, i; 151 struct image_params image_params, *imgp; 152 struct vattr attr; 153 int (*img_first) (struct image_params *); 154 155 if (debug_execve_args) { 156 printf("%s()\n", __func__); 157 print_execve_args(args); 158 } 159 160 KKASSERT(p); 161 imgp = &image_params; 162 163 /* 164 * Lock the process and set the P_INEXEC flag to indicate that 165 * it should be left alone until we're done here. This is 166 * necessary to avoid race conditions - e.g. in ptrace() - 167 * that might allow a local user to illicitly obtain elevated 168 * privileges. 169 */ 170 p->p_flag |= P_INEXEC; 171 172 /* 173 * Initialize part of the common data 174 */ 175 imgp->proc = p; 176 imgp->args = args; 177 imgp->attr = &attr; 178 imgp->entry_addr = 0; 179 imgp->resident = 0; 180 imgp->vmspace_destroyed = 0; 181 imgp->interpreted = 0; 182 imgp->interpreter_name[0] = 0; 183 imgp->auxargs = NULL; 184 imgp->vp = NULL; 185 imgp->firstpage = NULL; 186 imgp->ps_strings = 0; 187 imgp->image_header = NULL; 188 189 interpret: 190 191 /* 192 * Translate the file name to a vnode. Unlock the cache entry to 193 * improve parallelism for programs exec'd in parallel. 194 */ 195 if ((error = nlookup(nd)) != 0) 196 goto exec_fail; 197 error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 198 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 199 nd->nl_flags &= ~NLC_NCPISLOCKED; 200 cache_unlock(nd->nl_ncp); 201 if (error) 202 goto exec_fail; 203 204 /* 205 * Check file permissions (also 'opens' file) 206 */ 207 error = exec_check_permissions(imgp); 208 if (error) { 209 VOP_UNLOCK(imgp->vp, 0, td); 210 goto exec_fail_dealloc; 211 } 212 213 error = exec_map_first_page(imgp); 214 VOP_UNLOCK(imgp->vp, 0, td); 215 if (error) 216 goto exec_fail_dealloc; 217 218 if (debug_execve_args && imgp->interpreted) { 219 printf(" target is interpreted -- recursive pass\n"); 220 printf(" interpreter: %s\n", imgp->interpreter_name); 221 print_execve_args(args); 222 } 223 224 /* 225 * If the current process has a special image activator it 226 * wants to try first, call it. For example, emulating shell 227 * scripts differently. 228 */ 229 error = -1; 230 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 231 error = img_first(imgp); 232 233 /* 234 * If the vnode has a registered vmspace, exec the vmspace 235 */ 236 if (error == -1 && imgp->vp->v_resident) { 237 error = exec_resident_imgact(imgp); 238 } 239 240 /* 241 * Loop through the list of image activators, calling each one. 242 * An activator returns -1 if there is no match, 0 on success, 243 * and an error otherwise. 244 */ 245 for (i = 0; error == -1 && execsw[i]; ++i) { 246 if (execsw[i]->ex_imgact == NULL || 247 execsw[i]->ex_imgact == img_first) { 248 continue; 249 } 250 error = (*execsw[i]->ex_imgact)(imgp); 251 } 252 253 if (error) { 254 if (error == -1) 255 error = ENOEXEC; 256 goto exec_fail_dealloc; 257 } 258 259 /* 260 * Special interpreter operation, cleanup and loop up to try to 261 * activate the interpreter. 262 */ 263 if (imgp->interpreted) { 264 exec_unmap_first_page(imgp); 265 nlookup_done(nd); 266 vrele(imgp->vp); 267 imgp->vp = NULL; 268 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 269 NLC_FOLLOW); 270 if (error) 271 goto exec_fail; 272 goto interpret; 273 } 274 275 /* 276 * Copy out strings (args and env) and initialize stack base 277 */ 278 stack_base = exec_copyout_strings(imgp); 279 p->p_vmspace->vm_minsaddr = (char *)stack_base; 280 281 /* 282 * If custom stack fixup routine present for this process 283 * let it do the stack setup. If we are running a resident 284 * image there is no auxinfo or other image activator context 285 * so don't try to add fixups to the stack. 286 * 287 * Else stuff argument count as first item on stack 288 */ 289 if (p->p_sysent->sv_fixup && imgp->resident == 0) 290 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 291 else 292 suword(--stack_base, imgp->args->argc); 293 294 /* 295 * For security and other reasons, the file descriptor table cannot 296 * be shared after an exec. 297 */ 298 if (p->p_fd->fd_refcnt > 1) { 299 struct filedesc *tmp; 300 301 tmp = fdcopy(p); 302 fdfree(p); 303 p->p_fd = tmp; 304 } 305 306 /* 307 * For security and other reasons, signal handlers cannot 308 * be shared after an exec. The new proces gets a copy of the old 309 * handlers. In execsigs(), the new process will have its signals 310 * reset. 311 */ 312 if (p->p_procsig->ps_refcnt > 1) { 313 struct procsig *newprocsig; 314 315 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 316 M_SUBPROC, M_WAITOK); 317 bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig)); 318 p->p_procsig->ps_refcnt--; 319 p->p_procsig = newprocsig; 320 p->p_procsig->ps_refcnt = 1; 321 if (p->p_sigacts == &p->p_addr->u_sigacts) 322 panic("shared procsig but private sigacts?"); 323 324 p->p_addr->u_sigacts = *p->p_sigacts; 325 p->p_sigacts = &p->p_addr->u_sigacts; 326 } 327 328 /* Stop profiling */ 329 stopprofclock(p); 330 331 /* close files on exec */ 332 fdcloseexec(p); 333 334 /* reset caught signals */ 335 execsigs(p); 336 337 /* name this process - nameiexec(p, ndp) */ 338 len = min(nd->nl_ncp->nc_nlen, MAXCOMLEN); 339 bcopy(nd->nl_ncp->nc_name, p->p_comm, len); 340 p->p_comm[len] = 0; 341 342 /* 343 * mark as execed, wakeup the process that vforked (if any) and tell 344 * it that it now has its own resources back 345 */ 346 p->p_flag |= P_EXEC; 347 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 348 p->p_flag &= ~P_PPWAIT; 349 wakeup((caddr_t)p->p_pptr); 350 } 351 352 /* 353 * Implement image setuid/setgid. 354 * 355 * Don't honor setuid/setgid if the filesystem prohibits it or if 356 * the process is being traced. 357 */ 358 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 359 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 360 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 361 (p->p_flag & P_TRACED) == 0) { 362 /* 363 * Turn off syscall tracing for set-id programs, except for 364 * root. Record any set-id flags first to make sure that 365 * we do not regain any tracing during a possible block. 366 */ 367 setsugid(); 368 if (p->p_tracep && suser(td)) { 369 struct vnode *vtmp; 370 371 if ((vtmp = p->p_tracep) != NULL) { 372 p->p_tracep = NULL; 373 p->p_traceflag = 0; 374 vrele(vtmp); 375 } 376 } 377 /* Close any file descriptors 0..2 that reference procfs */ 378 setugidsafety(p); 379 /* Make sure file descriptors 0..2 are in use. */ 380 error = fdcheckstd(p); 381 if (error != 0) 382 goto exec_fail_dealloc; 383 /* 384 * Set the new credentials. 385 */ 386 cratom(&p->p_ucred); 387 if (attr.va_mode & VSUID) 388 change_euid(attr.va_uid); 389 if (attr.va_mode & VSGID) 390 p->p_ucred->cr_gid = attr.va_gid; 391 392 /* 393 * Clear local varsym variables 394 */ 395 varsymset_clean(&p->p_varsymset); 396 } else { 397 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 398 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 399 p->p_flag &= ~P_SUGID; 400 } 401 402 /* 403 * Implement correct POSIX saved-id behavior. 404 */ 405 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 406 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 407 cratom(&p->p_ucred); 408 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 409 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 410 } 411 412 /* 413 * Store the vp for use in procfs 414 */ 415 if (p->p_textvp) /* release old reference */ 416 vrele(p->p_textvp); 417 p->p_textvp = imgp->vp; 418 vref(p->p_textvp); 419 420 /* 421 * Notify others that we exec'd, and clear the P_INEXEC flag 422 * as we're now a bona fide freshly-execed process. 423 */ 424 KNOTE(&p->p_klist, NOTE_EXEC); 425 p->p_flag &= ~P_INEXEC; 426 427 /* 428 * If tracing the process, trap to debugger so breakpoints 429 * can be set before the program executes. 430 */ 431 STOPEVENT(p, S_EXEC, 0); 432 433 if (p->p_flag & P_TRACED) 434 psignal(p, SIGTRAP); 435 436 /* clear "fork but no exec" flag, as we _are_ execing */ 437 p->p_acflag &= ~AFORK; 438 439 /* Set values passed into the program in registers. */ 440 setregs(p, imgp->entry_addr, (u_long)(uintptr_t)stack_base, 441 imgp->ps_strings); 442 443 /* Free any previous argument cache */ 444 if (p->p_args && --p->p_args->ar_ref == 0) 445 FREE(p->p_args, M_PARGS); 446 p->p_args = NULL; 447 448 /* Cache arguments if they fit inside our allowance */ 449 i = imgp->args->begin_envv - imgp->args->begin_argv; 450 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 451 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 452 M_PARGS, M_WAITOK); 453 p->p_args->ar_ref = 1; 454 p->p_args->ar_length = i; 455 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 456 } 457 458 exec_fail_dealloc: 459 460 /* 461 * free various allocated resources 462 */ 463 if (imgp->firstpage) 464 exec_unmap_first_page(imgp); 465 466 if (imgp->vp) { 467 vrele(imgp->vp); 468 imgp->vp = NULL; 469 } 470 471 if (error == 0) { 472 ++mycpu->gd_cnt.v_exec; 473 return (0); 474 } 475 476 exec_fail: 477 /* we're done here, clear P_INEXEC */ 478 p->p_flag &= ~P_INEXEC; 479 if (imgp->vmspace_destroyed) { 480 /* sorry, no more process anymore. exit gracefully */ 481 exit1(W_EXITCODE(0, SIGABRT)); 482 /* NOT REACHED */ 483 return(0); 484 } else { 485 return(error); 486 } 487 } 488 489 /* 490 * execve() system call. 491 */ 492 int 493 execve(struct execve_args *uap) 494 { 495 struct nlookupdata nd; 496 struct image_args args; 497 int error; 498 499 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 500 if (error == 0) { 501 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 502 uap->argv, uap->envv); 503 } 504 if (error == 0) 505 error = kern_execve(&nd, &args); 506 nlookup_done(&nd); 507 exec_free_args(&args); 508 509 /* 510 * The syscall result is returned in registers to the new program. 511 * Linux will register %edx as an atexit function and we must be 512 * sure to set it to 0. XXX 513 */ 514 if (error == 0) 515 uap->sysmsg_result64 = 0; 516 517 return (error); 518 } 519 520 int 521 exec_map_first_page(struct image_params *imgp) 522 { 523 int rv, i; 524 int initial_pagein; 525 vm_page_t ma[VM_INITIAL_PAGEIN]; 526 vm_page_t m; 527 vm_object_t object; 528 int error; 529 530 if (imgp->firstpage) 531 exec_unmap_first_page(imgp); 532 533 /* 534 * XXX the callers should really use vn_open so we don't have to 535 * do this junk. 536 */ 537 if ((error = VOP_GETVOBJECT(imgp->vp, &object)) != 0) { 538 if (vn_canvmio(imgp->vp) == TRUE) { 539 error = vfs_object_create(imgp->vp, curthread); 540 if (error == 0) 541 error = VOP_GETVOBJECT(imgp->vp, &object); 542 } 543 } 544 if (error) 545 return (EIO); 546 547 /* 548 * We shouldn't need protection for vm_page_grab() but we certainly 549 * need it for the lookup loop below (lookup/busy race), since 550 * an interrupt can unbusy and free the page before our busy check. 551 */ 552 crit_enter(); 553 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 554 555 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 556 ma[0] = m; 557 initial_pagein = VM_INITIAL_PAGEIN; 558 if (initial_pagein > object->size) 559 initial_pagein = object->size; 560 for (i = 1; i < initial_pagein; i++) { 561 if ((m = vm_page_lookup(object, i)) != NULL) { 562 if ((m->flags & PG_BUSY) || m->busy) 563 break; 564 if (m->valid) 565 break; 566 vm_page_busy(m); 567 } else { 568 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 569 if (m == NULL) 570 break; 571 } 572 ma[i] = m; 573 } 574 initial_pagein = i; 575 576 /* 577 * get_pages unbusies all the requested pages except the 578 * primary page (at index 0 in this case). 579 */ 580 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 581 m = vm_page_lookup(object, 0); 582 583 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 584 if (m) { 585 vm_page_protect(m, VM_PROT_NONE); 586 vm_page_free(m); 587 } 588 crit_exit(); 589 return EIO; 590 } 591 } 592 vm_page_hold(m); 593 vm_page_wakeup(m); /* unbusy the page */ 594 crit_exit(); 595 596 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 597 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 598 599 return 0; 600 } 601 602 void 603 exec_unmap_first_page(struct image_params *imgp) 604 { 605 vm_page_t m; 606 607 crit_enter(); 608 if (imgp->firstpage != NULL) { 609 m = sf_buf_page(imgp->firstpage); 610 sf_buf_free(imgp->firstpage); 611 imgp->firstpage = NULL; 612 imgp->image_header = NULL; 613 vm_page_unhold(m); 614 } 615 crit_exit(); 616 } 617 618 /* 619 * Destroy old address space, and allocate a new stack 620 * The new stack is only SGROWSIZ large because it is grown 621 * automatically in trap.c. 622 */ 623 int 624 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 625 { 626 int error; 627 struct vmspace *vmspace = imgp->proc->p_vmspace; 628 vm_offset_t stack_addr = USRSTACK - maxssiz; 629 vm_map_t map; 630 631 imgp->vmspace_destroyed = 1; 632 633 /* 634 * Prevent a pending AIO from modifying the new address space. 635 */ 636 aio_proc_rundown(imgp->proc); 637 638 /* 639 * Blow away entire process VM, if address space not shared, 640 * otherwise, create a new VM space so that other threads are 641 * not disrupted. If we are execing a resident vmspace we 642 * create a duplicate of it and remap the stack. 643 * 644 * The exitingcnt test is not strictly necessary but has been 645 * included for code sanity (to make the code more deterministic). 646 */ 647 map = &vmspace->vm_map; 648 if (vmcopy) { 649 vmspace_exec(imgp->proc, vmcopy); 650 vmspace = imgp->proc->p_vmspace; 651 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 652 map = &vmspace->vm_map; 653 } else if (vmspace->vm_refcnt == 1 && vmspace->vm_exitingcnt == 0) { 654 shmexit(vmspace); 655 if (vmspace->vm_upcalls) 656 upc_release(vmspace, imgp->proc); 657 pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS); 658 vm_map_remove(map, 0, VM_MAXUSER_ADDRESS); 659 } else { 660 vmspace_exec(imgp->proc, NULL); 661 vmspace = imgp->proc->p_vmspace; 662 map = &vmspace->vm_map; 663 } 664 665 /* Allocate a new stack */ 666 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 667 VM_PROT_ALL, VM_PROT_ALL, 0); 668 if (error) 669 return (error); 670 671 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 672 * VM_STACK case, but they are still used to monitor the size of the 673 * process stack so we can check the stack rlimit. 674 */ 675 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 676 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 677 678 return(0); 679 } 680 681 /* 682 * Copy out argument and environment strings from the old process 683 * address space into the temporary string buffer. 684 */ 685 int 686 exec_copyin_args(struct image_args *args, char *fname, 687 enum exec_path_segflg segflg, char **argv, char **envv) 688 { 689 char *argp, *envp; 690 int error = 0; 691 size_t length; 692 693 bzero(args, sizeof(*args)); 694 args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 695 if (args->buf == NULL) 696 return (ENOMEM); 697 args->begin_argv = args->buf; 698 args->endp = args->begin_argv; 699 args->space = ARG_MAX; 700 701 args->fname = args->buf + ARG_MAX; 702 703 /* 704 * Copy the file name. 705 */ 706 if (segflg == PATH_SYSSPACE) { 707 error = copystr(fname, args->fname, PATH_MAX, &length); 708 } else if (segflg == PATH_USERSPACE) { 709 error = copyinstr(fname, args->fname, PATH_MAX, &length); 710 } 711 712 /* 713 * Extract argument strings. argv may not be NULL. The argv 714 * array is terminated by a NULL entry. We special-case the 715 * situation where argv[0] is NULL by passing { filename, NULL } 716 * to the new program to guarentee that the interpreter knows what 717 * file to open in case we exec an interpreted file. Note that 718 * a NULL argv[0] terminates the argv[] array. 719 * 720 * XXX the special-casing of argv[0] is historical and needs to be 721 * revisited. 722 */ 723 if (argv == NULL) 724 error = EFAULT; 725 if (error == 0) { 726 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 727 if (argp == (caddr_t)-1) { 728 error = EFAULT; 729 break; 730 } 731 error = copyinstr(argp, args->endp, 732 args->space, &length); 733 if (error) { 734 if (error == ENAMETOOLONG) 735 error = E2BIG; 736 break; 737 } 738 args->space -= length; 739 args->endp += length; 740 args->argc++; 741 } 742 if (args->argc == 0 && error == 0) { 743 length = strlen(args->fname) + 1; 744 if (length > args->space) { 745 error = E2BIG; 746 } else { 747 bcopy(args->fname, args->endp, length); 748 args->space -= length; 749 args->endp += length; 750 args->argc++; 751 } 752 } 753 } 754 755 args->begin_envv = args->endp; 756 757 /* 758 * extract environment strings. envv may be NULL. 759 */ 760 if (envv && error == 0) { 761 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 762 if (envp == (caddr_t) -1) { 763 error = EFAULT; 764 break; 765 } 766 error = copyinstr(envp, args->endp, args->space, 767 &length); 768 if (error) { 769 if (error == ENAMETOOLONG) 770 error = E2BIG; 771 break; 772 } 773 args->space -= length; 774 args->endp += length; 775 args->envc++; 776 } 777 } 778 return (error); 779 } 780 781 void 782 exec_free_args(struct image_args *args) 783 { 784 if (args->buf) { 785 kmem_free_wakeup(exec_map, 786 (vm_offset_t)args->buf, PATH_MAX + ARG_MAX); 787 args->buf = NULL; 788 } 789 } 790 791 /* 792 * Copy strings out to the new process address space, constructing 793 * new arg and env vector tables. Return a pointer to the base 794 * so that it can be used as the initial stack pointer. 795 */ 796 register_t * 797 exec_copyout_strings(struct image_params *imgp) 798 { 799 int argc, envc, sgap; 800 char **vectp; 801 char *stringp, *destp; 802 register_t *stack_base; 803 struct ps_strings *arginfo; 804 int szsigcode; 805 806 /* 807 * Calculate string base and vector table pointers. 808 * Also deal with signal trampoline code for this exec type. 809 */ 810 arginfo = (struct ps_strings *)PS_STRINGS; 811 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 812 if (stackgap_random != 0) 813 sgap = ALIGN(arc4random() & (stackgap_random - 1)); 814 else 815 sgap = 0; 816 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 817 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 818 819 /* 820 * install sigcode 821 */ 822 if (szsigcode) 823 copyout(imgp->proc->p_sysent->sv_sigcode, 824 ((caddr_t)arginfo - szsigcode), szsigcode); 825 826 /* 827 * If we have a valid auxargs ptr, prepare some room 828 * on the stack. 829 * 830 * The '+ 2' is for the null pointers at the end of each of the 831 * arg and env vector sets, and 'AT_COUNT*2' is room for the 832 * ELF Auxargs data. 833 */ 834 if (imgp->auxargs) { 835 vectp = (char **)(destp - (imgp->args->argc + 836 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 837 } else { 838 vectp = (char **)(destp - (imgp->args->argc + 839 imgp->args->envc + 2) * sizeof(char*)); 840 } 841 842 /* 843 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 844 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 845 */ 846 847 /* 848 * vectp also becomes our initial stack base 849 */ 850 stack_base = (register_t *)vectp; 851 852 stringp = imgp->args->begin_argv; 853 argc = imgp->args->argc; 854 envc = imgp->args->envc; 855 856 /* 857 * Copy out strings - arguments and environment. 858 */ 859 copyout(stringp, destp, ARG_MAX - imgp->args->space); 860 861 /* 862 * Fill in "ps_strings" struct for ps, w, etc. 863 */ 864 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 865 suword(&arginfo->ps_nargvstr, argc); 866 867 /* 868 * Fill in argument portion of vector table. 869 */ 870 for (; argc > 0; --argc) { 871 suword(vectp++, (long)(intptr_t)destp); 872 while (*stringp++ != 0) 873 destp++; 874 destp++; 875 } 876 877 /* a null vector table pointer separates the argp's from the envp's */ 878 suword(vectp++, 0); 879 880 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 881 suword(&arginfo->ps_nenvstr, envc); 882 883 /* 884 * Fill in environment portion of vector table. 885 */ 886 for (; envc > 0; --envc) { 887 suword(vectp++, (long)(intptr_t)destp); 888 while (*stringp++ != 0) 889 destp++; 890 destp++; 891 } 892 893 /* end of vector table is a null pointer */ 894 suword(vectp, 0); 895 896 return (stack_base); 897 } 898 899 /* 900 * Check permissions of file to execute. 901 * Return 0 for success or error code on failure. 902 */ 903 int 904 exec_check_permissions(struct image_params *imgp) 905 { 906 struct proc *p = imgp->proc; 907 struct vnode *vp = imgp->vp; 908 struct vattr *attr = imgp->attr; 909 struct thread *td = p->p_thread; 910 int error; 911 912 /* Get file attributes */ 913 error = VOP_GETATTR(vp, attr, td); 914 if (error) 915 return (error); 916 917 /* 918 * 1) Check if file execution is disabled for the filesystem that this 919 * file resides on. 920 * 2) Insure that at least one execute bit is on - otherwise root 921 * will always succeed, and we don't want to happen unless the 922 * file really is executable. 923 * 3) Insure that the file is a regular file. 924 */ 925 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 926 ((attr->va_mode & 0111) == 0) || 927 (attr->va_type != VREG)) { 928 return (EACCES); 929 } 930 931 /* 932 * Zero length files can't be exec'd 933 */ 934 if (attr->va_size == 0) 935 return (ENOEXEC); 936 937 /* 938 * Check for execute permission to file based on current credentials. 939 */ 940 error = VOP_ACCESS(vp, VEXEC, p->p_ucred, td); 941 if (error) 942 return (error); 943 944 /* 945 * Check number of open-for-writes on the file and deny execution 946 * if there are any. 947 */ 948 if (vp->v_writecount) 949 return (ETXTBSY); 950 951 /* 952 * Call filesystem specific open routine (which does nothing in the 953 * general case). 954 */ 955 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL, td); 956 if (error) 957 return (error); 958 959 return (0); 960 } 961 962 /* 963 * Exec handler registration 964 */ 965 int 966 exec_register(const struct execsw *execsw_arg) 967 { 968 const struct execsw **es, **xs, **newexecsw; 969 int count = 2; /* New slot and trailing NULL */ 970 971 if (execsw) 972 for (es = execsw; *es; es++) 973 count++; 974 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 975 if (newexecsw == NULL) 976 return ENOMEM; 977 xs = newexecsw; 978 if (execsw) 979 for (es = execsw; *es; es++) 980 *xs++ = *es; 981 *xs++ = execsw_arg; 982 *xs = NULL; 983 if (execsw) 984 free(execsw, M_TEMP); 985 execsw = newexecsw; 986 return 0; 987 } 988 989 int 990 exec_unregister(const struct execsw *execsw_arg) 991 { 992 const struct execsw **es, **xs, **newexecsw; 993 int count = 1; 994 995 if (execsw == NULL) 996 panic("unregister with no handlers left?"); 997 998 for (es = execsw; *es; es++) { 999 if (*es == execsw_arg) 1000 break; 1001 } 1002 if (*es == NULL) 1003 return ENOENT; 1004 for (es = execsw; *es; es++) 1005 if (*es != execsw_arg) 1006 count++; 1007 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1008 if (newexecsw == NULL) 1009 return ENOMEM; 1010 xs = newexecsw; 1011 for (es = execsw; *es; es++) 1012 if (*es != execsw_arg) 1013 *xs++ = *es; 1014 *xs = NULL; 1015 if (execsw) 1016 free(execsw, M_TEMP); 1017 execsw = newexecsw; 1018 return 0; 1019 } 1020