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 1.46 2006/09/19 11:47:35 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 <machine/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 printf("\targv[%d]: %s\n", ndx, cp); 131 while (*cp++ != '\0'); 132 } 133 for (ndx = 0; ndx < args->envc; ndx++) { 134 printf("\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 proc *p = td->td_proc; 150 register_t *stack_base; 151 int error, len, i; 152 struct image_params image_params, *imgp; 153 struct vattr attr; 154 int (*img_first) (struct image_params *); 155 156 if (debug_execve_args) { 157 printf("%s()\n", __func__); 158 print_execve_args(args); 159 } 160 161 KKASSERT(p); 162 imgp = &image_params; 163 164 /* 165 * Lock the process and set the P_INEXEC flag to indicate that 166 * it should be left alone until we're done here. This is 167 * necessary to avoid race conditions - e.g. in ptrace() - 168 * that might allow a local user to illicitly obtain elevated 169 * privileges. 170 */ 171 p->p_flag |= P_INEXEC; 172 173 /* 174 * Initialize part of the common data 175 */ 176 imgp->proc = p; 177 imgp->args = args; 178 imgp->attr = &attr; 179 imgp->entry_addr = 0; 180 imgp->resident = 0; 181 imgp->vmspace_destroyed = 0; 182 imgp->interpreted = 0; 183 imgp->interpreter_name[0] = 0; 184 imgp->auxargs = NULL; 185 imgp->vp = NULL; 186 imgp->firstpage = NULL; 187 imgp->ps_strings = 0; 188 imgp->image_header = NULL; 189 190 interpret: 191 192 /* 193 * Translate the file name to a vnode. Unlock the cache entry to 194 * improve parallelism for programs exec'd in parallel. 195 */ 196 if ((error = nlookup(nd)) != 0) 197 goto exec_fail; 198 error = cache_vget(nd->nl_ncp, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 199 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 200 nd->nl_flags &= ~NLC_NCPISLOCKED; 201 cache_unlock(nd->nl_ncp); 202 if (error) 203 goto exec_fail; 204 205 /* 206 * Check file permissions (also 'opens' file) 207 */ 208 error = exec_check_permissions(imgp); 209 if (error) { 210 vn_unlock(imgp->vp); 211 goto exec_fail_dealloc; 212 } 213 214 error = exec_map_first_page(imgp); 215 vn_unlock(imgp->vp); 216 if (error) 217 goto exec_fail_dealloc; 218 219 if (debug_execve_args && imgp->interpreted) { 220 printf(" target is interpreted -- recursive pass\n"); 221 printf(" interpreter: %s\n", imgp->interpreter_name); 222 print_execve_args(args); 223 } 224 225 /* 226 * If the current process has a special image activator it 227 * wants to try first, call it. For example, emulating shell 228 * scripts differently. 229 */ 230 error = -1; 231 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 232 error = img_first(imgp); 233 234 /* 235 * If the vnode has a registered vmspace, exec the vmspace 236 */ 237 if (error == -1 && imgp->vp->v_resident) { 238 error = exec_resident_imgact(imgp); 239 } 240 241 /* 242 * Loop through the list of image activators, calling each one. 243 * An activator returns -1 if there is no match, 0 on success, 244 * and an error otherwise. 245 */ 246 for (i = 0; error == -1 && execsw[i]; ++i) { 247 if (execsw[i]->ex_imgact == NULL || 248 execsw[i]->ex_imgact == img_first) { 249 continue; 250 } 251 error = (*execsw[i]->ex_imgact)(imgp); 252 } 253 254 if (error) { 255 if (error == -1) 256 error = ENOEXEC; 257 goto exec_fail_dealloc; 258 } 259 260 /* 261 * Special interpreter operation, cleanup and loop up to try to 262 * activate the interpreter. 263 */ 264 if (imgp->interpreted) { 265 exec_unmap_first_page(imgp); 266 nlookup_done(nd); 267 vrele(imgp->vp); 268 imgp->vp = NULL; 269 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 270 NLC_FOLLOW); 271 if (error) 272 goto exec_fail; 273 goto interpret; 274 } 275 276 /* 277 * Copy out strings (args and env) and initialize stack base 278 */ 279 stack_base = exec_copyout_strings(imgp); 280 p->p_vmspace->vm_minsaddr = (char *)stack_base; 281 282 /* 283 * If custom stack fixup routine present for this process 284 * let it do the stack setup. If we are running a resident 285 * image there is no auxinfo or other image activator context 286 * so don't try to add fixups to the stack. 287 * 288 * Else stuff argument count as first item on stack 289 */ 290 if (p->p_sysent->sv_fixup && imgp->resident == 0) 291 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 292 else 293 suword(--stack_base, imgp->args->argc); 294 295 /* 296 * For security and other reasons, the file descriptor table cannot 297 * be shared after an exec. 298 */ 299 if (p->p_fd->fd_refcnt > 1) { 300 struct filedesc *tmp; 301 302 tmp = fdcopy(p); 303 fdfree(p); 304 p->p_fd = tmp; 305 } 306 307 /* 308 * For security and other reasons, signal handlers cannot 309 * be shared after an exec. The new proces gets a copy of the old 310 * handlers. In execsigs(), the new process will have its signals 311 * reset. 312 */ 313 if (p->p_procsig->ps_refcnt > 1) { 314 struct procsig *newprocsig; 315 316 MALLOC(newprocsig, struct procsig *, sizeof(struct procsig), 317 M_SUBPROC, M_WAITOK); 318 bcopy(p->p_procsig, newprocsig, sizeof(*newprocsig)); 319 p->p_procsig->ps_refcnt--; 320 p->p_procsig = newprocsig; 321 p->p_procsig->ps_refcnt = 1; 322 if (p->p_sigacts == &p->p_addr->u_sigacts) 323 panic("shared procsig but private sigacts?"); 324 325 p->p_addr->u_sigacts = *p->p_sigacts; 326 p->p_sigacts = &p->p_addr->u_sigacts; 327 } 328 329 /* 330 * For security and other reasons virtual kernels cannot be 331 * inherited by an exec. This also allows a virtual kernel 332 * to fork/exec unrelated applications. 333 */ 334 if (p->p_vkernel) { 335 vkernel_drop(p->p_vkernel); 336 p->p_vkernel = NULL; 337 } 338 339 /* Stop profiling */ 340 stopprofclock(p); 341 342 /* close files on exec */ 343 fdcloseexec(p); 344 345 /* reset caught signals */ 346 execsigs(p); 347 348 /* name this process - nameiexec(p, ndp) */ 349 len = min(nd->nl_ncp->nc_nlen, MAXCOMLEN); 350 bcopy(nd->nl_ncp->nc_name, p->p_comm, len); 351 p->p_comm[len] = 0; 352 bcopy(p->p_comm, p->p_lwp.lwp_thread->td_comm, MAXCOMLEN+1); 353 354 /* 355 * mark as execed, wakeup the process that vforked (if any) and tell 356 * it that it now has its own resources back 357 */ 358 p->p_flag |= P_EXEC; 359 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 360 p->p_flag &= ~P_PPWAIT; 361 wakeup((caddr_t)p->p_pptr); 362 } 363 364 /* 365 * Implement image setuid/setgid. 366 * 367 * Don't honor setuid/setgid if the filesystem prohibits it or if 368 * the process is being traced. 369 */ 370 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 371 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 372 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 373 (p->p_flag & P_TRACED) == 0) { 374 /* 375 * Turn off syscall tracing for set-id programs, except for 376 * root. Record any set-id flags first to make sure that 377 * we do not regain any tracing during a possible block. 378 */ 379 setsugid(); 380 if (p->p_tracenode && suser(td) != 0) { 381 ktrdestroy(&p->p_tracenode); 382 p->p_traceflag = 0; 383 } 384 /* Close any file descriptors 0..2 that reference procfs */ 385 setugidsafety(p); 386 /* Make sure file descriptors 0..2 are in use. */ 387 error = fdcheckstd(p); 388 if (error != 0) 389 goto exec_fail_dealloc; 390 /* 391 * Set the new credentials. 392 */ 393 cratom(&p->p_ucred); 394 if (attr.va_mode & VSUID) 395 change_euid(attr.va_uid); 396 if (attr.va_mode & VSGID) 397 p->p_ucred->cr_gid = attr.va_gid; 398 399 /* 400 * Clear local varsym variables 401 */ 402 varsymset_clean(&p->p_varsymset); 403 } else { 404 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 405 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 406 p->p_flag &= ~P_SUGID; 407 } 408 409 /* 410 * Implement correct POSIX saved-id behavior. 411 */ 412 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 413 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 414 cratom(&p->p_ucred); 415 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 416 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 417 } 418 419 /* 420 * Store the vp for use in procfs 421 */ 422 if (p->p_textvp) /* release old reference */ 423 vrele(p->p_textvp); 424 p->p_textvp = imgp->vp; 425 vref(p->p_textvp); 426 427 /* 428 * Notify others that we exec'd, and clear the P_INEXEC flag 429 * as we're now a bona fide freshly-execed process. 430 */ 431 KNOTE(&p->p_klist, NOTE_EXEC); 432 p->p_flag &= ~P_INEXEC; 433 434 /* 435 * If tracing the process, trap to debugger so breakpoints 436 * can be set before the program executes. 437 */ 438 STOPEVENT(p, S_EXEC, 0); 439 440 if (p->p_flag & P_TRACED) 441 ksignal(p, SIGTRAP); 442 443 /* clear "fork but no exec" flag, as we _are_ execing */ 444 p->p_acflag &= ~AFORK; 445 446 /* Set values passed into the program in registers. */ 447 setregs(td->td_lwp, imgp->entry_addr, (u_long)(uintptr_t)stack_base, 448 imgp->ps_strings); 449 450 /* Free any previous argument cache */ 451 if (p->p_args && --p->p_args->ar_ref == 0) 452 FREE(p->p_args, M_PARGS); 453 p->p_args = NULL; 454 455 /* Cache arguments if they fit inside our allowance */ 456 i = imgp->args->begin_envv - imgp->args->begin_argv; 457 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 458 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 459 M_PARGS, M_WAITOK); 460 p->p_args->ar_ref = 1; 461 p->p_args->ar_length = i; 462 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 463 } 464 465 exec_fail_dealloc: 466 467 /* 468 * free various allocated resources 469 */ 470 if (imgp->firstpage) 471 exec_unmap_first_page(imgp); 472 473 if (imgp->vp) { 474 vrele(imgp->vp); 475 imgp->vp = NULL; 476 } 477 478 if (error == 0) { 479 ++mycpu->gd_cnt.v_exec; 480 return (0); 481 } 482 483 exec_fail: 484 /* we're done here, clear P_INEXEC */ 485 p->p_flag &= ~P_INEXEC; 486 if (imgp->vmspace_destroyed) { 487 /* sorry, no more process anymore. exit gracefully */ 488 exit1(W_EXITCODE(0, SIGABRT)); 489 /* NOT REACHED */ 490 return(0); 491 } else { 492 return(error); 493 } 494 } 495 496 /* 497 * execve() system call. 498 */ 499 int 500 sys_execve(struct execve_args *uap) 501 { 502 struct nlookupdata nd; 503 struct image_args args; 504 int error; 505 506 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 507 if (error == 0) { 508 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 509 uap->argv, uap->envv); 510 } 511 if (error == 0) 512 error = kern_execve(&nd, &args); 513 nlookup_done(&nd); 514 exec_free_args(&args); 515 516 /* 517 * The syscall result is returned in registers to the new program. 518 * Linux will register %edx as an atexit function and we must be 519 * sure to set it to 0. XXX 520 */ 521 if (error == 0) 522 uap->sysmsg_result64 = 0; 523 524 return (error); 525 } 526 527 int 528 exec_map_first_page(struct image_params *imgp) 529 { 530 int rv, i; 531 int initial_pagein; 532 vm_page_t ma[VM_INITIAL_PAGEIN]; 533 vm_page_t m; 534 vm_object_t object; 535 536 if (imgp->firstpage) 537 exec_unmap_first_page(imgp); 538 539 /* 540 * The file has to be mappable. 541 */ 542 if ((object = imgp->vp->v_object) == NULL) 543 return (EIO); 544 545 /* 546 * We shouldn't need protection for vm_page_grab() but we certainly 547 * need it for the lookup loop below (lookup/busy race), since 548 * an interrupt can unbusy and free the page before our busy check. 549 */ 550 crit_enter(); 551 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 552 553 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 554 ma[0] = m; 555 initial_pagein = VM_INITIAL_PAGEIN; 556 if (initial_pagein > object->size) 557 initial_pagein = object->size; 558 for (i = 1; i < initial_pagein; i++) { 559 if ((m = vm_page_lookup(object, i)) != NULL) { 560 if ((m->flags & PG_BUSY) || m->busy) 561 break; 562 if (m->valid) 563 break; 564 vm_page_busy(m); 565 } else { 566 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 567 if (m == NULL) 568 break; 569 } 570 ma[i] = m; 571 } 572 initial_pagein = i; 573 574 /* 575 * get_pages unbusies all the requested pages except the 576 * primary page (at index 0 in this case). 577 */ 578 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 579 m = vm_page_lookup(object, 0); 580 581 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 582 if (m) { 583 vm_page_protect(m, VM_PROT_NONE); 584 vm_page_free(m); 585 } 586 crit_exit(); 587 return EIO; 588 } 589 } 590 vm_page_hold(m); 591 vm_page_wakeup(m); /* unbusy the page */ 592 crit_exit(); 593 594 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 595 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 596 597 return 0; 598 } 599 600 void 601 exec_unmap_first_page(struct image_params *imgp) 602 { 603 vm_page_t m; 604 605 crit_enter(); 606 if (imgp->firstpage != NULL) { 607 m = sf_buf_page(imgp->firstpage); 608 sf_buf_free(imgp->firstpage); 609 imgp->firstpage = NULL; 610 imgp->image_header = NULL; 611 vm_page_unhold(m); 612 } 613 crit_exit(); 614 } 615 616 /* 617 * Destroy old address space, and allocate a new stack 618 * The new stack is only SGROWSIZ large because it is grown 619 * automatically in trap.c. 620 */ 621 int 622 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 623 { 624 int error; 625 struct vmspace *vmspace = imgp->proc->p_vmspace; 626 vm_offset_t stack_addr = USRSTACK - maxssiz; 627 vm_map_t map; 628 629 imgp->vmspace_destroyed = 1; 630 631 /* 632 * Prevent a pending AIO from modifying the new address space. 633 */ 634 aio_proc_rundown(imgp->proc); 635 636 /* 637 * Blow away entire process VM, if address space not shared, 638 * otherwise, create a new VM space so that other threads are 639 * not disrupted. If we are execing a resident vmspace we 640 * create a duplicate of it and remap the stack. 641 * 642 * The exitingcnt test is not strictly necessary but has been 643 * included for code sanity (to make the code more deterministic). 644 */ 645 map = &vmspace->vm_map; 646 if (vmcopy) { 647 vmspace_exec(imgp->proc, vmcopy); 648 vmspace = imgp->proc->p_vmspace; 649 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 650 map = &vmspace->vm_map; 651 } else if (vmspace->vm_refcnt == 1 && vmspace->vm_exitingcnt == 0) { 652 shmexit(vmspace); 653 if (vmspace->vm_upcalls) 654 upc_release(vmspace, &imgp->proc->p_lwp); 655 pmap_remove_pages(vmspace_pmap(vmspace), 0, VM_MAXUSER_ADDRESS); 656 vm_map_remove(map, 0, VM_MAXUSER_ADDRESS); 657 } else { 658 vmspace_exec(imgp->proc, NULL); 659 vmspace = imgp->proc->p_vmspace; 660 map = &vmspace->vm_map; 661 } 662 663 /* Allocate a new stack */ 664 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 665 VM_PROT_ALL, VM_PROT_ALL, 0); 666 if (error) 667 return (error); 668 669 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 670 * VM_STACK case, but they are still used to monitor the size of the 671 * process stack so we can check the stack rlimit. 672 */ 673 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 674 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 675 676 return(0); 677 } 678 679 /* 680 * Copy out argument and environment strings from the old process 681 * address space into the temporary string buffer. 682 */ 683 int 684 exec_copyin_args(struct image_args *args, char *fname, 685 enum exec_path_segflg segflg, char **argv, char **envv) 686 { 687 char *argp, *envp; 688 int error = 0; 689 size_t length; 690 691 bzero(args, sizeof(*args)); 692 args->buf = (char *) kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 693 if (args->buf == NULL) 694 return (ENOMEM); 695 args->begin_argv = args->buf; 696 args->endp = args->begin_argv; 697 args->space = ARG_MAX; 698 699 args->fname = args->buf + ARG_MAX; 700 701 /* 702 * Copy the file name. 703 */ 704 if (segflg == PATH_SYSSPACE) { 705 error = copystr(fname, args->fname, PATH_MAX, &length); 706 } else if (segflg == PATH_USERSPACE) { 707 error = copyinstr(fname, args->fname, PATH_MAX, &length); 708 } 709 710 /* 711 * Extract argument strings. argv may not be NULL. The argv 712 * array is terminated by a NULL entry. We special-case the 713 * situation where argv[0] is NULL by passing { filename, NULL } 714 * to the new program to guarentee that the interpreter knows what 715 * file to open in case we exec an interpreted file. Note that 716 * a NULL argv[0] terminates the argv[] array. 717 * 718 * XXX the special-casing of argv[0] is historical and needs to be 719 * revisited. 720 */ 721 if (argv == NULL) 722 error = EFAULT; 723 if (error == 0) { 724 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 725 if (argp == (caddr_t)-1) { 726 error = EFAULT; 727 break; 728 } 729 error = copyinstr(argp, args->endp, 730 args->space, &length); 731 if (error) { 732 if (error == ENAMETOOLONG) 733 error = E2BIG; 734 break; 735 } 736 args->space -= length; 737 args->endp += length; 738 args->argc++; 739 } 740 if (args->argc == 0 && error == 0) { 741 length = strlen(args->fname) + 1; 742 if (length > args->space) { 743 error = E2BIG; 744 } else { 745 bcopy(args->fname, args->endp, length); 746 args->space -= length; 747 args->endp += length; 748 args->argc++; 749 } 750 } 751 } 752 753 args->begin_envv = args->endp; 754 755 /* 756 * extract environment strings. envv may be NULL. 757 */ 758 if (envv && error == 0) { 759 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 760 if (envp == (caddr_t) -1) { 761 error = EFAULT; 762 break; 763 } 764 error = copyinstr(envp, args->endp, args->space, 765 &length); 766 if (error) { 767 if (error == ENAMETOOLONG) 768 error = E2BIG; 769 break; 770 } 771 args->space -= length; 772 args->endp += length; 773 args->envc++; 774 } 775 } 776 return (error); 777 } 778 779 void 780 exec_free_args(struct image_args *args) 781 { 782 if (args->buf) { 783 kmem_free_wakeup(exec_map, 784 (vm_offset_t)args->buf, PATH_MAX + ARG_MAX); 785 args->buf = NULL; 786 } 787 } 788 789 /* 790 * Copy strings out to the new process address space, constructing 791 * new arg and env vector tables. Return a pointer to the base 792 * so that it can be used as the initial stack pointer. 793 */ 794 register_t * 795 exec_copyout_strings(struct image_params *imgp) 796 { 797 int argc, envc, sgap; 798 char **vectp; 799 char *stringp, *destp; 800 register_t *stack_base; 801 struct ps_strings *arginfo; 802 int szsigcode; 803 804 /* 805 * Calculate string base and vector table pointers. 806 * Also deal with signal trampoline code for this exec type. 807 */ 808 arginfo = (struct ps_strings *)PS_STRINGS; 809 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 810 if (stackgap_random != 0) 811 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 812 else 813 sgap = 0; 814 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 815 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 816 817 /* 818 * install sigcode 819 */ 820 if (szsigcode) 821 copyout(imgp->proc->p_sysent->sv_sigcode, 822 ((caddr_t)arginfo - szsigcode), szsigcode); 823 824 /* 825 * If we have a valid auxargs ptr, prepare some room 826 * on the stack. 827 * 828 * The '+ 2' is for the null pointers at the end of each of the 829 * arg and env vector sets, and 'AT_COUNT*2' is room for the 830 * ELF Auxargs data. 831 */ 832 if (imgp->auxargs) { 833 vectp = (char **)(destp - (imgp->args->argc + 834 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 835 } else { 836 vectp = (char **)(destp - (imgp->args->argc + 837 imgp->args->envc + 2) * sizeof(char*)); 838 } 839 840 /* 841 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 842 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 843 */ 844 845 /* 846 * vectp also becomes our initial stack base 847 */ 848 stack_base = (register_t *)vectp; 849 850 stringp = imgp->args->begin_argv; 851 argc = imgp->args->argc; 852 envc = imgp->args->envc; 853 854 /* 855 * Copy out strings - arguments and environment. 856 */ 857 copyout(stringp, destp, ARG_MAX - imgp->args->space); 858 859 /* 860 * Fill in "ps_strings" struct for ps, w, etc. 861 */ 862 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 863 suword(&arginfo->ps_nargvstr, argc); 864 865 /* 866 * Fill in argument portion of vector table. 867 */ 868 for (; argc > 0; --argc) { 869 suword(vectp++, (long)(intptr_t)destp); 870 while (*stringp++ != 0) 871 destp++; 872 destp++; 873 } 874 875 /* a null vector table pointer separates the argp's from the envp's */ 876 suword(vectp++, 0); 877 878 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 879 suword(&arginfo->ps_nenvstr, envc); 880 881 /* 882 * Fill in environment portion of vector table. 883 */ 884 for (; envc > 0; --envc) { 885 suword(vectp++, (long)(intptr_t)destp); 886 while (*stringp++ != 0) 887 destp++; 888 destp++; 889 } 890 891 /* end of vector table is a null pointer */ 892 suword(vectp, 0); 893 894 return (stack_base); 895 } 896 897 /* 898 * Check permissions of file to execute. 899 * Return 0 for success or error code on failure. 900 */ 901 int 902 exec_check_permissions(struct image_params *imgp) 903 { 904 struct proc *p = imgp->proc; 905 struct vnode *vp = imgp->vp; 906 struct vattr *attr = imgp->attr; 907 int error; 908 909 /* Get file attributes */ 910 error = VOP_GETATTR(vp, attr); 911 if (error) 912 return (error); 913 914 /* 915 * 1) Check if file execution is disabled for the filesystem that this 916 * file resides on. 917 * 2) Insure that at least one execute bit is on - otherwise root 918 * will always succeed, and we don't want to happen unless the 919 * file really is executable. 920 * 3) Insure that the file is a regular file. 921 */ 922 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 923 ((attr->va_mode & 0111) == 0) || 924 (attr->va_type != VREG)) { 925 return (EACCES); 926 } 927 928 /* 929 * Zero length files can't be exec'd 930 */ 931 if (attr->va_size == 0) 932 return (ENOEXEC); 933 934 /* 935 * Check for execute permission to file based on current credentials. 936 */ 937 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 938 if (error) 939 return (error); 940 941 /* 942 * Check number of open-for-writes on the file and deny execution 943 * if there are any. 944 */ 945 if (vp->v_writecount) 946 return (ETXTBSY); 947 948 /* 949 * Call filesystem specific open routine, which allows us to read, 950 * write, and mmap the file. Without the VOP_OPEN we can only 951 * stat the file. 952 */ 953 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 954 if (error) 955 return (error); 956 957 return (0); 958 } 959 960 /* 961 * Exec handler registration 962 */ 963 int 964 exec_register(const struct execsw *execsw_arg) 965 { 966 const struct execsw **es, **xs, **newexecsw; 967 int count = 2; /* New slot and trailing NULL */ 968 969 if (execsw) 970 for (es = execsw; *es; es++) 971 count++; 972 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 973 if (newexecsw == NULL) 974 return ENOMEM; 975 xs = newexecsw; 976 if (execsw) 977 for (es = execsw; *es; es++) 978 *xs++ = *es; 979 *xs++ = execsw_arg; 980 *xs = NULL; 981 if (execsw) 982 kfree(execsw, M_TEMP); 983 execsw = newexecsw; 984 return 0; 985 } 986 987 int 988 exec_unregister(const struct execsw *execsw_arg) 989 { 990 const struct execsw **es, **xs, **newexecsw; 991 int count = 1; 992 993 if (execsw == NULL) 994 panic("unregister with no handlers left?"); 995 996 for (es = execsw; *es; es++) { 997 if (*es == execsw_arg) 998 break; 999 } 1000 if (*es == NULL) 1001 return ENOENT; 1002 for (es = execsw; *es; es++) 1003 if (*es != execsw_arg) 1004 count++; 1005 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1006 if (newexecsw == NULL) 1007 return ENOMEM; 1008 xs = newexecsw; 1009 for (es = execsw; *es; es++) 1010 if (*es != execsw_arg) 1011 *xs++ = *es; 1012 *xs = NULL; 1013 if (execsw) 1014 kfree(execsw, M_TEMP); 1015 execsw = newexecsw; 1016 return 0; 1017 } 1018