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 1.53 2007/02/03 17:05:57 corecode 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_procsig->ps_refcnt > 1) { 315 struct procsig *newprocsig; 316 317 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 318 M_SUBPROC, M_WAITOK); 319 bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig)); 320 p->p_procsig->ps_refcnt--; 321 p->p_procsig = newprocsig; 322 p->p_procsig->ps_refcnt = 1; 323 if (p->p_sigacts == &p->p_addr->u_sigacts) 324 panic("shared procsig but private sigacts?"); 325 326 p->p_addr->u_sigacts = *p->p_sigacts; 327 p->p_sigacts = &p->p_addr->u_sigacts; 328 } 329 330 /* 331 * For security and other reasons virtual kernels cannot be 332 * inherited by an exec. This also allows a virtual kernel 333 * to fork/exec unrelated applications. 334 */ 335 if (p->p_vkernel) 336 vkernel_exit(p); 337 338 /* Stop profiling */ 339 stopprofclock(p); 340 341 /* close files on exec */ 342 fdcloseexec(p); 343 344 /* reset caught signals */ 345 execsigs(p); 346 347 /* name this process - nameiexec(p, ndp) */ 348 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 349 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 350 p->p_comm[len] = 0; 351 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 352 353 /* 354 * mark as execed, wakeup the process that vforked (if any) and tell 355 * it that it now has its own resources back 356 */ 357 p->p_flag |= P_EXEC; 358 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 359 p->p_flag &= ~P_PPWAIT; 360 wakeup((caddr_t)p->p_pptr); 361 } 362 363 /* 364 * Implement image setuid/setgid. 365 * 366 * Don't honor setuid/setgid if the filesystem prohibits it or if 367 * the process is being traced. 368 */ 369 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 370 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 371 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 372 (p->p_flag & P_TRACED) == 0) { 373 /* 374 * Turn off syscall tracing for set-id programs, except for 375 * root. Record any set-id flags first to make sure that 376 * we do not regain any tracing during a possible block. 377 */ 378 setsugid(); 379 if (p->p_tracenode && suser(td) != 0) { 380 ktrdestroy(&p->p_tracenode); 381 p->p_traceflag = 0; 382 } 383 /* Close any file descriptors 0..2 that reference procfs */ 384 setugidsafety(p); 385 /* Make sure file descriptors 0..2 are in use. */ 386 error = fdcheckstd(p); 387 if (error != 0) 388 goto exec_fail_dealloc; 389 /* 390 * Set the new credentials. 391 */ 392 cratom(&p->p_ucred); 393 if (attr.va_mode & VSUID) 394 change_euid(attr.va_uid); 395 if (attr.va_mode & VSGID) 396 p->p_ucred->cr_gid = attr.va_gid; 397 398 /* 399 * Clear local varsym variables 400 */ 401 varsymset_clean(&p->p_varsymset); 402 } else { 403 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 404 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 405 p->p_flag &= ~P_SUGID; 406 } 407 408 /* 409 * Implement correct POSIX saved-id behavior. 410 */ 411 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 412 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 413 cratom(&p->p_ucred); 414 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 415 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 416 } 417 418 /* 419 * Store the vp for use in procfs 420 */ 421 if (p->p_textvp) /* release old reference */ 422 vrele(p->p_textvp); 423 p->p_textvp = imgp->vp; 424 vref(p->p_textvp); 425 426 /* 427 * Notify others that we exec'd, and clear the P_INEXEC flag 428 * as we're now a bona fide freshly-execed process. 429 */ 430 KNOTE(&p->p_klist, NOTE_EXEC); 431 p->p_flag &= ~P_INEXEC; 432 433 /* 434 * If tracing the process, trap to debugger so breakpoints 435 * can be set before the program executes. 436 */ 437 STOPEVENT(p, S_EXEC, 0); 438 439 if (p->p_flag & P_TRACED) 440 ksignal(p, SIGTRAP); 441 442 /* clear "fork but no exec" flag, as we _are_ execing */ 443 p->p_acflag &= ~AFORK; 444 445 /* Set values passed into the program in registers. */ 446 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 447 imgp->ps_strings); 448 449 /* Free any previous argument cache */ 450 if (p->p_args && --p->p_args->ar_ref == 0) 451 FREE(p->p_args, M_PARGS); 452 p->p_args = NULL; 453 454 /* Cache arguments if they fit inside our allowance */ 455 i = imgp->args->begin_envv - imgp->args->begin_argv; 456 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 457 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 458 M_PARGS, M_WAITOK); 459 p->p_args->ar_ref = 1; 460 p->p_args->ar_length = i; 461 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 462 } 463 464 exec_fail_dealloc: 465 466 /* 467 * free various allocated resources 468 */ 469 if (imgp->firstpage) 470 exec_unmap_first_page(imgp); 471 472 if (imgp->vp) { 473 vrele(imgp->vp); 474 imgp->vp = NULL; 475 } 476 477 if (error == 0) { 478 ++mycpu->gd_cnt.v_exec; 479 return (0); 480 } 481 482 exec_fail: 483 /* we're done here, clear P_INEXEC */ 484 p->p_flag &= ~P_INEXEC; 485 if (imgp->vmspace_destroyed) { 486 /* sorry, no more process anymore. exit gracefully */ 487 exit1(W_EXITCODE(0, SIGABRT)); 488 /* NOT REACHED */ 489 return(0); 490 } else { 491 return(error); 492 } 493 } 494 495 /* 496 * execve() system call. 497 */ 498 int 499 sys_execve(struct execve_args *uap) 500 { 501 struct nlookupdata nd; 502 struct image_args args; 503 int error; 504 505 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 506 if (error == 0) { 507 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 508 uap->argv, uap->envv); 509 } 510 if (error == 0) 511 error = kern_execve(&nd, &args); 512 nlookup_done(&nd); 513 exec_free_args(&args); 514 515 /* 516 * The syscall result is returned in registers to the new program. 517 * Linux will register %edx as an atexit function and we must be 518 * sure to set it to 0. XXX 519 */ 520 if (error == 0) 521 uap->sysmsg_result64 = 0; 522 523 return (error); 524 } 525 526 int 527 exec_map_first_page(struct image_params *imgp) 528 { 529 int rv, i; 530 int initial_pagein; 531 vm_page_t ma[VM_INITIAL_PAGEIN]; 532 vm_page_t m; 533 vm_object_t object; 534 535 if (imgp->firstpage) 536 exec_unmap_first_page(imgp); 537 538 /* 539 * The file has to be mappable. 540 */ 541 if ((object = imgp->vp->v_object) == NULL) 542 return (EIO); 543 544 /* 545 * We shouldn't need protection for vm_page_grab() but we certainly 546 * need it for the lookup loop below (lookup/busy race), since 547 * an interrupt can unbusy and free the page before our busy check. 548 */ 549 crit_enter(); 550 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 551 552 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 553 ma[0] = m; 554 initial_pagein = VM_INITIAL_PAGEIN; 555 if (initial_pagein > object->size) 556 initial_pagein = object->size; 557 for (i = 1; i < initial_pagein; i++) { 558 if ((m = vm_page_lookup(object, i)) != NULL) { 559 if ((m->flags & PG_BUSY) || m->busy) 560 break; 561 if (m->valid) 562 break; 563 vm_page_busy(m); 564 } else { 565 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 566 if (m == NULL) 567 break; 568 } 569 ma[i] = m; 570 } 571 initial_pagein = i; 572 573 /* 574 * get_pages unbusies all the requested pages except the 575 * primary page (at index 0 in this case). 576 */ 577 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 578 m = vm_page_lookup(object, 0); 579 580 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 581 if (m) { 582 vm_page_protect(m, VM_PROT_NONE); 583 vm_page_free(m); 584 } 585 crit_exit(); 586 return EIO; 587 } 588 } 589 vm_page_hold(m); 590 vm_page_wakeup(m); /* unbusy the page */ 591 crit_exit(); 592 593 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 594 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 595 596 return 0; 597 } 598 599 void 600 exec_unmap_first_page(struct image_params *imgp) 601 { 602 vm_page_t m; 603 604 crit_enter(); 605 if (imgp->firstpage != NULL) { 606 m = sf_buf_page(imgp->firstpage); 607 sf_buf_free(imgp->firstpage); 608 imgp->firstpage = NULL; 609 imgp->image_header = NULL; 610 vm_page_unhold(m); 611 } 612 crit_exit(); 613 } 614 615 /* 616 * Destroy old address space, and allocate a new stack 617 * The new stack is only SGROWSIZ large because it is grown 618 * automatically in trap.c. 619 */ 620 int 621 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 622 { 623 int error; 624 struct vmspace *vmspace = imgp->proc->p_vmspace; 625 vm_offset_t stack_addr = USRSTACK - maxssiz; 626 vm_map_t map; 627 628 imgp->vmspace_destroyed = 1; 629 630 /* 631 * XXX lwp here would be a good place to kill sibling lwps 632 */ 633 634 /* 635 * Prevent a pending AIO from modifying the new address space. 636 */ 637 aio_proc_rundown(imgp->proc); 638 639 /* 640 * Blow away entire process VM, if address space not shared, 641 * otherwise, create a new VM space so that other threads are 642 * not disrupted. If we are execing a resident vmspace we 643 * create a duplicate of it and remap the stack. 644 * 645 * The exitingcnt test is not strictly necessary but has been 646 * included for code sanity (to make the code more deterministic). 647 */ 648 map = &vmspace->vm_map; 649 if (vmcopy) { 650 vmspace_exec(imgp->proc, vmcopy); 651 vmspace = imgp->proc->p_vmspace; 652 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 653 map = &vmspace->vm_map; 654 } else if (vmspace->vm_refcnt == 1 && vmspace->vm_exitingcnt == 0) { 655 shmexit(vmspace); 656 if (vmspace->vm_upcalls) 657 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 658 pmap_remove_pages(vmspace_pmap(vmspace), 659 0, VM_MAX_USER_ADDRESS); 660 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 661 } else { 662 vmspace_exec(imgp->proc, NULL); 663 vmspace = imgp->proc->p_vmspace; 664 map = &vmspace->vm_map; 665 } 666 667 /* Allocate a new stack */ 668 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 669 VM_PROT_ALL, VM_PROT_ALL, 0); 670 if (error) 671 return (error); 672 673 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 674 * VM_STACK case, but they are still used to monitor the size of the 675 * process stack so we can check the stack rlimit. 676 */ 677 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 678 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 679 680 return(0); 681 } 682 683 /* 684 * Copy out argument and environment strings from the old process 685 * address space into the temporary string buffer. 686 */ 687 int 688 exec_copyin_args(struct image_args *args, char *fname, 689 enum exec_path_segflg segflg, char **argv, char **envv) 690 { 691 char *argp, *envp; 692 int error = 0; 693 size_t length; 694 695 bzero(args, sizeof(*args)); 696 args->buf = (char *) kmem_alloc_wait(&exec_map, PATH_MAX + ARG_MAX); 697 if (args->buf == NULL) 698 return (ENOMEM); 699 args->begin_argv = args->buf; 700 args->endp = args->begin_argv; 701 args->space = ARG_MAX; 702 703 args->fname = args->buf + ARG_MAX; 704 705 /* 706 * Copy the file name. 707 */ 708 if (segflg == PATH_SYSSPACE) { 709 error = copystr(fname, args->fname, PATH_MAX, &length); 710 } else if (segflg == PATH_USERSPACE) { 711 error = copyinstr(fname, args->fname, PATH_MAX, &length); 712 } 713 714 /* 715 * Extract argument strings. argv may not be NULL. The argv 716 * array is terminated by a NULL entry. We special-case the 717 * situation where argv[0] is NULL by passing { filename, NULL } 718 * to the new program to guarentee that the interpreter knows what 719 * file to open in case we exec an interpreted file. Note that 720 * a NULL argv[0] terminates the argv[] array. 721 * 722 * XXX the special-casing of argv[0] is historical and needs to be 723 * revisited. 724 */ 725 if (argv == NULL) 726 error = EFAULT; 727 if (error == 0) { 728 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 729 if (argp == (caddr_t)-1) { 730 error = EFAULT; 731 break; 732 } 733 error = copyinstr(argp, args->endp, 734 args->space, &length); 735 if (error) { 736 if (error == ENAMETOOLONG) 737 error = E2BIG; 738 break; 739 } 740 args->space -= length; 741 args->endp += length; 742 args->argc++; 743 } 744 if (args->argc == 0 && error == 0) { 745 length = strlen(args->fname) + 1; 746 if (length > args->space) { 747 error = E2BIG; 748 } else { 749 bcopy(args->fname, args->endp, length); 750 args->space -= length; 751 args->endp += length; 752 args->argc++; 753 } 754 } 755 } 756 757 args->begin_envv = args->endp; 758 759 /* 760 * extract environment strings. envv may be NULL. 761 */ 762 if (envv && error == 0) { 763 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 764 if (envp == (caddr_t) -1) { 765 error = EFAULT; 766 break; 767 } 768 error = copyinstr(envp, args->endp, args->space, 769 &length); 770 if (error) { 771 if (error == ENAMETOOLONG) 772 error = E2BIG; 773 break; 774 } 775 args->space -= length; 776 args->endp += length; 777 args->envc++; 778 } 779 } 780 return (error); 781 } 782 783 void 784 exec_free_args(struct image_args *args) 785 { 786 if (args->buf) { 787 kmem_free_wakeup(&exec_map, 788 (vm_offset_t)args->buf, PATH_MAX + ARG_MAX); 789 args->buf = NULL; 790 } 791 } 792 793 /* 794 * Copy strings out to the new process address space, constructing 795 * new arg and env vector tables. Return a pointer to the base 796 * so that it can be used as the initial stack pointer. 797 */ 798 register_t * 799 exec_copyout_strings(struct image_params *imgp) 800 { 801 int argc, envc, sgap; 802 char **vectp; 803 char *stringp, *destp; 804 register_t *stack_base; 805 struct ps_strings *arginfo; 806 int szsigcode; 807 808 /* 809 * Calculate string base and vector table pointers. 810 * Also deal with signal trampoline code for this exec type. 811 */ 812 arginfo = (struct ps_strings *)PS_STRINGS; 813 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 814 if (stackgap_random != 0) 815 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 816 else 817 sgap = 0; 818 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 819 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 820 821 /* 822 * install sigcode 823 */ 824 if (szsigcode) 825 copyout(imgp->proc->p_sysent->sv_sigcode, 826 ((caddr_t)arginfo - szsigcode), szsigcode); 827 828 /* 829 * If we have a valid auxargs ptr, prepare some room 830 * on the stack. 831 * 832 * The '+ 2' is for the null pointers at the end of each of the 833 * arg and env vector sets, and 'AT_COUNT*2' is room for the 834 * ELF Auxargs data. 835 */ 836 if (imgp->auxargs) { 837 vectp = (char **)(destp - (imgp->args->argc + 838 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 839 } else { 840 vectp = (char **)(destp - (imgp->args->argc + 841 imgp->args->envc + 2) * sizeof(char*)); 842 } 843 844 /* 845 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 846 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 847 */ 848 849 /* 850 * vectp also becomes our initial stack base 851 */ 852 stack_base = (register_t *)vectp; 853 854 stringp = imgp->args->begin_argv; 855 argc = imgp->args->argc; 856 envc = imgp->args->envc; 857 858 /* 859 * Copy out strings - arguments and environment. 860 */ 861 copyout(stringp, destp, ARG_MAX - imgp->args->space); 862 863 /* 864 * Fill in "ps_strings" struct for ps, w, etc. 865 */ 866 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 867 suword(&arginfo->ps_nargvstr, argc); 868 869 /* 870 * Fill in argument portion of vector table. 871 */ 872 for (; argc > 0; --argc) { 873 suword(vectp++, (long)(intptr_t)destp); 874 while (*stringp++ != 0) 875 destp++; 876 destp++; 877 } 878 879 /* a null vector table pointer separates the argp's from the envp's */ 880 suword(vectp++, 0); 881 882 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 883 suword(&arginfo->ps_nenvstr, envc); 884 885 /* 886 * Fill in environment portion of vector table. 887 */ 888 for (; envc > 0; --envc) { 889 suword(vectp++, (long)(intptr_t)destp); 890 while (*stringp++ != 0) 891 destp++; 892 destp++; 893 } 894 895 /* end of vector table is a null pointer */ 896 suword(vectp, 0); 897 898 return (stack_base); 899 } 900 901 /* 902 * Check permissions of file to execute. 903 * Return 0 for success or error code on failure. 904 */ 905 int 906 exec_check_permissions(struct image_params *imgp) 907 { 908 struct proc *p = imgp->proc; 909 struct vnode *vp = imgp->vp; 910 struct vattr *attr = imgp->attr; 911 int error; 912 913 /* Get file attributes */ 914 error = VOP_GETATTR(vp, attr); 915 if (error) 916 return (error); 917 918 /* 919 * 1) Check if file execution is disabled for the filesystem that this 920 * file resides on. 921 * 2) Insure that at least one execute bit is on - otherwise root 922 * will always succeed, and we don't want to happen unless the 923 * file really is executable. 924 * 3) Insure that the file is a regular file. 925 */ 926 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 927 ((attr->va_mode & 0111) == 0) || 928 (attr->va_type != VREG)) { 929 return (EACCES); 930 } 931 932 /* 933 * Zero length files can't be exec'd 934 */ 935 if (attr->va_size == 0) 936 return (ENOEXEC); 937 938 /* 939 * Check for execute permission to file based on current credentials. 940 */ 941 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 942 if (error) 943 return (error); 944 945 /* 946 * Check number of open-for-writes on the file and deny execution 947 * if there are any. 948 */ 949 if (vp->v_writecount) 950 return (ETXTBSY); 951 952 /* 953 * Call filesystem specific open routine, which allows us to read, 954 * write, and mmap the file. Without the VOP_OPEN we can only 955 * stat the file. 956 */ 957 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 958 if (error) 959 return (error); 960 961 return (0); 962 } 963 964 /* 965 * Exec handler registration 966 */ 967 int 968 exec_register(const struct execsw *execsw_arg) 969 { 970 const struct execsw **es, **xs, **newexecsw; 971 int count = 2; /* New slot and trailing NULL */ 972 973 if (execsw) 974 for (es = execsw; *es; es++) 975 count++; 976 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 977 if (newexecsw == NULL) 978 return ENOMEM; 979 xs = newexecsw; 980 if (execsw) 981 for (es = execsw; *es; es++) 982 *xs++ = *es; 983 *xs++ = execsw_arg; 984 *xs = NULL; 985 if (execsw) 986 kfree(execsw, M_TEMP); 987 execsw = newexecsw; 988 return 0; 989 } 990 991 int 992 exec_unregister(const struct execsw *execsw_arg) 993 { 994 const struct execsw **es, **xs, **newexecsw; 995 int count = 1; 996 997 if (execsw == NULL) 998 panic("unregister with no handlers left?"); 999 1000 for (es = execsw; *es; es++) { 1001 if (*es == execsw_arg) 1002 break; 1003 } 1004 if (*es == NULL) 1005 return ENOENT; 1006 for (es = execsw; *es; es++) 1007 if (*es != execsw_arg) 1008 count++; 1009 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1010 if (newexecsw == NULL) 1011 return ENOMEM; 1012 xs = newexecsw; 1013 for (es = execsw; *es; es++) 1014 if (*es != execsw_arg) 1015 *xs++ = *es; 1016 *xs = NULL; 1017 if (execsw) 1018 kfree(execsw, M_TEMP); 1019 execsw = newexecsw; 1020 return 0; 1021 } 1022