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 1.62 2007/08/28 01:09:24 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/vnode_pager.h> 68 #include <vm/vm_pager.h> 69 70 #include <sys/user.h> 71 #include <sys/reg.h> 72 73 #include <sys/thread2.h> 74 75 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 76 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 77 78 static register_t *exec_copyout_strings (struct image_params *); 79 80 /* XXX This should be vm_size_t. */ 81 static u_long ps_strings = PS_STRINGS; 82 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 83 84 /* XXX This should be vm_size_t. */ 85 static u_long usrstack = USRSTACK; 86 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 87 88 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 89 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 90 &ps_arg_cache_limit, 0, ""); 91 92 int ps_argsopen = 1; 93 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 94 95 void print_execve_args(struct image_args *args); 96 int debug_execve_args = 0; 97 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 98 0, ""); 99 100 /* 101 * Exec arguments object cache 102 */ 103 static struct objcache *exec_objcache; 104 105 static 106 void 107 exec_objcache_init(void *arg __unused) 108 { 109 exec_objcache = objcache_create_mbacked( 110 M_EXECARGS, PATH_MAX + ARG_MAX, 111 16, /* up to this many objects */ 112 2, /* minimal magazine capacity */ 113 NULL, NULL, NULL); 114 } 115 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 116 117 /* 118 * stackgap_random specifies if the stackgap should have a random size added 119 * to it. It must be a power of 2. If non-zero, the stack gap will be 120 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 121 */ 122 static int stackgap_random = 1024; 123 static int 124 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 125 { 126 int error, new_val; 127 new_val = stackgap_random; 128 error = sysctl_handle_int(oidp, &new_val, 0, req); 129 if (error != 0 || req->newptr == NULL) 130 return (error); 131 if ((new_val < 0) || (new_val > 16 * PAGE_SIZE) || ! powerof2(new_val)) 132 return (EINVAL); 133 stackgap_random = new_val; 134 135 return(0); 136 } 137 138 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_UINT, 139 0, 0, sysctl_kern_stackgap, "IU", "Max random stack gap (power of 2)"); 140 141 void 142 print_execve_args(struct image_args *args) 143 { 144 char *cp; 145 int ndx; 146 147 cp = args->begin_argv; 148 for (ndx = 0; ndx < args->argc; ndx++) { 149 kprintf("\targv[%d]: %s\n", ndx, cp); 150 while (*cp++ != '\0'); 151 } 152 for (ndx = 0; ndx < args->envc; ndx++) { 153 kprintf("\tenvv[%d]: %s\n", ndx, cp); 154 while (*cp++ != '\0'); 155 } 156 } 157 158 /* 159 * Each of the items is a pointer to a `const struct execsw', hence the 160 * double pointer here. 161 */ 162 static const struct execsw **execsw; 163 164 /* 165 * Replace current vmspace with a new binary. 166 * Returns 0 on success, > 0 on recoverable error (use as errno). 167 * Returns -1 on lethal error which demands killing of the current 168 * process! 169 */ 170 int 171 kern_execve(struct nlookupdata *nd, struct image_args *args) 172 { 173 struct thread *td = curthread; 174 struct lwp *lp = td->td_lwp; 175 struct proc *p = td->td_proc; 176 register_t *stack_base; 177 int error, len, i; 178 struct image_params image_params, *imgp; 179 struct vattr attr; 180 int (*img_first) (struct image_params *); 181 182 if (debug_execve_args) { 183 kprintf("%s()\n", __func__); 184 print_execve_args(args); 185 } 186 187 KKASSERT(p); 188 imgp = &image_params; 189 190 /* 191 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 192 * no modifications to the process at all until we get there. 193 * 194 * Note that multiple threads may be trying to exec at the same 195 * time. exec_new_vmspace() handles that too. 196 */ 197 198 /* 199 * Initialize part of the common data 200 */ 201 imgp->proc = p; 202 imgp->args = args; 203 imgp->attr = &attr; 204 imgp->entry_addr = 0; 205 imgp->resident = 0; 206 imgp->vmspace_destroyed = 0; 207 imgp->interpreted = 0; 208 imgp->interpreter_name[0] = 0; 209 imgp->auxargs = NULL; 210 imgp->vp = NULL; 211 imgp->firstpage = NULL; 212 imgp->ps_strings = 0; 213 imgp->image_header = NULL; 214 215 interpret: 216 217 /* 218 * Translate the file name to a vnode. Unlock the cache entry to 219 * improve parallelism for programs exec'd in parallel. 220 */ 221 if ((error = nlookup(nd)) != 0) 222 goto exec_fail; 223 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 224 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 225 nd->nl_flags &= ~NLC_NCPISLOCKED; 226 cache_unlock(&nd->nl_nch); 227 if (error) 228 goto exec_fail; 229 230 /* 231 * Check file permissions (also 'opens' file) 232 */ 233 error = exec_check_permissions(imgp); 234 if (error) { 235 vn_unlock(imgp->vp); 236 goto exec_fail_dealloc; 237 } 238 239 error = exec_map_first_page(imgp); 240 vn_unlock(imgp->vp); 241 if (error) 242 goto exec_fail_dealloc; 243 244 if (debug_execve_args && imgp->interpreted) { 245 kprintf(" target is interpreted -- recursive pass\n"); 246 kprintf(" interpreter: %s\n", imgp->interpreter_name); 247 print_execve_args(args); 248 } 249 250 /* 251 * If the current process has a special image activator it 252 * wants to try first, call it. For example, emulating shell 253 * scripts differently. 254 */ 255 error = -1; 256 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 257 error = img_first(imgp); 258 259 /* 260 * If the vnode has a registered vmspace, exec the vmspace 261 */ 262 if (error == -1 && imgp->vp->v_resident) { 263 error = exec_resident_imgact(imgp); 264 } 265 266 /* 267 * Loop through the list of image activators, calling each one. 268 * An activator returns -1 if there is no match, 0 on success, 269 * and an error otherwise. 270 */ 271 for (i = 0; error == -1 && execsw[i]; ++i) { 272 if (execsw[i]->ex_imgact == NULL || 273 execsw[i]->ex_imgact == img_first) { 274 continue; 275 } 276 error = (*execsw[i]->ex_imgact)(imgp); 277 } 278 279 if (error) { 280 if (error == -1) 281 error = ENOEXEC; 282 goto exec_fail_dealloc; 283 } 284 285 /* 286 * Special interpreter operation, cleanup and loop up to try to 287 * activate the interpreter. 288 */ 289 if (imgp->interpreted) { 290 exec_unmap_first_page(imgp); 291 nlookup_done(nd); 292 vrele(imgp->vp); 293 imgp->vp = NULL; 294 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 295 NLC_FOLLOW); 296 if (error) 297 goto exec_fail; 298 goto interpret; 299 } 300 301 /* 302 * Copy out strings (args and env) and initialize stack base 303 */ 304 stack_base = exec_copyout_strings(imgp); 305 p->p_vmspace->vm_minsaddr = (char *)stack_base; 306 307 /* 308 * If custom stack fixup routine present for this process 309 * let it do the stack setup. If we are running a resident 310 * image there is no auxinfo or other image activator context 311 * so don't try to add fixups to the stack. 312 * 313 * Else stuff argument count as first item on stack 314 */ 315 if (p->p_sysent->sv_fixup && imgp->resident == 0) 316 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 317 else 318 suword(--stack_base, imgp->args->argc); 319 320 /* 321 * For security and other reasons, the file descriptor table cannot 322 * be shared after an exec. 323 */ 324 if (p->p_fd->fd_refcnt > 1) { 325 struct filedesc *tmp; 326 327 tmp = fdcopy(p); 328 fdfree(p); 329 p->p_fd = tmp; 330 } 331 332 /* 333 * For security and other reasons, signal handlers cannot 334 * be shared after an exec. The new proces gets a copy of the old 335 * handlers. In execsigs(), the new process will have its signals 336 * reset. 337 */ 338 if (p->p_sigacts->ps_refcnt > 1) { 339 struct sigacts *newsigacts; 340 341 newsigacts = (struct sigacts *)kmalloc(sizeof(*newsigacts), 342 M_SUBPROC, M_WAITOK); 343 bcopy(p->p_sigacts, newsigacts, sizeof(*newsigacts)); 344 p->p_sigacts->ps_refcnt--; 345 p->p_sigacts = newsigacts; 346 p->p_sigacts->ps_refcnt = 1; 347 } 348 349 /* 350 * For security and other reasons virtual kernels cannot be 351 * inherited by an exec. This also allows a virtual kernel 352 * to fork/exec unrelated applications. 353 */ 354 if (p->p_vkernel) 355 vkernel_exit(p); 356 357 /* Stop profiling */ 358 stopprofclock(p); 359 360 /* close files on exec */ 361 fdcloseexec(p); 362 363 /* reset caught signals */ 364 execsigs(p); 365 366 /* name this process - nameiexec(p, ndp) */ 367 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 368 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 369 p->p_comm[len] = 0; 370 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 371 372 /* 373 * mark as execed, wakeup the process that vforked (if any) and tell 374 * it that it now has its own resources back 375 */ 376 p->p_flag |= P_EXEC; 377 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 378 p->p_flag &= ~P_PPWAIT; 379 wakeup((caddr_t)p->p_pptr); 380 } 381 382 /* 383 * Implement image setuid/setgid. 384 * 385 * Don't honor setuid/setgid if the filesystem prohibits it or if 386 * the process is being traced. 387 */ 388 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 389 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 390 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 391 (p->p_flag & P_TRACED) == 0) { 392 /* 393 * Turn off syscall tracing for set-id programs, except for 394 * root. Record any set-id flags first to make sure that 395 * we do not regain any tracing during a possible block. 396 */ 397 setsugid(); 398 if (p->p_tracenode && suser(td) != 0) { 399 ktrdestroy(&p->p_tracenode); 400 p->p_traceflag = 0; 401 } 402 /* Close any file descriptors 0..2 that reference procfs */ 403 setugidsafety(p); 404 /* Make sure file descriptors 0..2 are in use. */ 405 error = fdcheckstd(p); 406 if (error != 0) 407 goto exec_fail_dealloc; 408 /* 409 * Set the new credentials. 410 */ 411 cratom(&p->p_ucred); 412 if (attr.va_mode & VSUID) 413 change_euid(attr.va_uid); 414 if (attr.va_mode & VSGID) 415 p->p_ucred->cr_gid = attr.va_gid; 416 417 /* 418 * Clear local varsym variables 419 */ 420 varsymset_clean(&p->p_varsymset); 421 } else { 422 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 423 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 424 p->p_flag &= ~P_SUGID; 425 } 426 427 /* 428 * Implement correct POSIX saved-id behavior. 429 */ 430 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 431 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 432 cratom(&p->p_ucred); 433 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 434 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 435 } 436 437 /* 438 * Store the vp for use in procfs 439 */ 440 if (p->p_textvp) /* release old reference */ 441 vrele(p->p_textvp); 442 p->p_textvp = imgp->vp; 443 vref(p->p_textvp); 444 445 /* 446 * Notify others that we exec'd, and clear the P_INEXEC flag 447 * as we're now a bona fide freshly-execed process. 448 */ 449 KNOTE(&p->p_klist, NOTE_EXEC); 450 p->p_flag &= ~P_INEXEC; 451 452 /* 453 * If tracing the process, trap to debugger so breakpoints 454 * can be set before the program executes. 455 */ 456 STOPEVENT(p, S_EXEC, 0); 457 458 if (p->p_flag & P_TRACED) 459 ksignal(p, SIGTRAP); 460 461 /* clear "fork but no exec" flag, as we _are_ execing */ 462 p->p_acflag &= ~AFORK; 463 464 /* Set values passed into the program in registers. */ 465 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 466 imgp->ps_strings); 467 468 /* Free any previous argument cache */ 469 if (p->p_args && --p->p_args->ar_ref == 0) 470 FREE(p->p_args, M_PARGS); 471 p->p_args = NULL; 472 473 /* Cache arguments if they fit inside our allowance */ 474 i = imgp->args->begin_envv - imgp->args->begin_argv; 475 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 476 MALLOC(p->p_args, struct pargs *, sizeof(struct pargs) + i, 477 M_PARGS, M_WAITOK); 478 p->p_args->ar_ref = 1; 479 p->p_args->ar_length = i; 480 bcopy(imgp->args->begin_argv, p->p_args->ar_args, i); 481 } 482 483 exec_fail_dealloc: 484 485 /* 486 * free various allocated resources 487 */ 488 if (imgp->firstpage) 489 exec_unmap_first_page(imgp); 490 491 if (imgp->vp) { 492 vrele(imgp->vp); 493 imgp->vp = NULL; 494 } 495 496 if (error == 0) { 497 ++mycpu->gd_cnt.v_exec; 498 return (0); 499 } 500 501 exec_fail: 502 /* 503 * we're done here, clear P_INEXEC if we were the ones that 504 * set it. Otherwise if vmspace_destroyed is still set we 505 * raced another thread and that thread is responsible for 506 * clearing it. 507 */ 508 if (imgp->vmspace_destroyed & 2) 509 p->p_flag &= ~P_INEXEC; 510 if (imgp->vmspace_destroyed) { 511 /* 512 * Sorry, no more process anymore. exit gracefully. 513 * However we can't die right here, because our 514 * caller might have to clean up, so indicate a 515 * lethal error by returning -1. 516 */ 517 return(-1); 518 } else { 519 return(error); 520 } 521 } 522 523 /* 524 * execve() system call. 525 */ 526 int 527 sys_execve(struct execve_args *uap) 528 { 529 struct nlookupdata nd; 530 struct image_args args; 531 int error; 532 533 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 534 if (error == 0) { 535 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 536 uap->argv, uap->envv); 537 } 538 if (error == 0) 539 error = kern_execve(&nd, &args); 540 nlookup_done(&nd); 541 exec_free_args(&args); 542 543 if (error < 0) { 544 /* We hit a lethal error condition. Let's die now. */ 545 exit1(W_EXITCODE(0, SIGABRT)); 546 /* NOTREACHED */ 547 } 548 549 /* 550 * The syscall result is returned in registers to the new program. 551 * Linux will register %edx as an atexit function and we must be 552 * sure to set it to 0. XXX 553 */ 554 if (error == 0) 555 uap->sysmsg_result64 = 0; 556 557 return (error); 558 } 559 560 int 561 exec_map_first_page(struct image_params *imgp) 562 { 563 int rv, i; 564 int initial_pagein; 565 vm_page_t ma[VM_INITIAL_PAGEIN]; 566 vm_page_t m; 567 vm_object_t object; 568 569 if (imgp->firstpage) 570 exec_unmap_first_page(imgp); 571 572 /* 573 * The file has to be mappable. 574 */ 575 if ((object = imgp->vp->v_object) == NULL) 576 return (EIO); 577 578 /* 579 * We shouldn't need protection for vm_page_grab() but we certainly 580 * need it for the lookup loop below (lookup/busy race), since 581 * an interrupt can unbusy and free the page before our busy check. 582 */ 583 crit_enter(); 584 m = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 585 586 if ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 587 ma[0] = m; 588 initial_pagein = VM_INITIAL_PAGEIN; 589 if (initial_pagein > object->size) 590 initial_pagein = object->size; 591 for (i = 1; i < initial_pagein; i++) { 592 if ((m = vm_page_lookup(object, i)) != NULL) { 593 if ((m->flags & PG_BUSY) || m->busy) 594 break; 595 if (m->valid) 596 break; 597 vm_page_busy(m); 598 } else { 599 m = vm_page_alloc(object, i, VM_ALLOC_NORMAL); 600 if (m == NULL) 601 break; 602 } 603 ma[i] = m; 604 } 605 initial_pagein = i; 606 607 /* 608 * get_pages unbusies all the requested pages except the 609 * primary page (at index 0 in this case). The primary 610 * page may have been wired during the pagein (e.g. by 611 * the buffer cache) so vnode_pager_freepage() must be 612 * used to properly release it. 613 */ 614 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 615 m = vm_page_lookup(object, 0); 616 617 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 618 if (m) { 619 vm_page_protect(m, VM_PROT_NONE); 620 vnode_pager_freepage(m); 621 } 622 crit_exit(); 623 return EIO; 624 } 625 } 626 vm_page_hold(m); 627 vm_page_wakeup(m); /* unbusy the page */ 628 crit_exit(); 629 630 imgp->firstpage = sf_buf_alloc(m, SFB_CPUPRIVATE); 631 imgp->image_header = (void *)sf_buf_kva(imgp->firstpage); 632 633 return 0; 634 } 635 636 void 637 exec_unmap_first_page(struct image_params *imgp) 638 { 639 vm_page_t m; 640 641 crit_enter(); 642 if (imgp->firstpage != NULL) { 643 m = sf_buf_page(imgp->firstpage); 644 sf_buf_free(imgp->firstpage); 645 imgp->firstpage = NULL; 646 imgp->image_header = NULL; 647 vm_page_unhold(m); 648 } 649 crit_exit(); 650 } 651 652 /* 653 * Destroy old address space, and allocate a new stack 654 * The new stack is only SGROWSIZ large because it is grown 655 * automatically in trap.c. 656 * 657 * This is the point of no return. 658 */ 659 int 660 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 661 { 662 struct vmspace *vmspace = imgp->proc->p_vmspace; 663 vm_offset_t stack_addr = USRSTACK - maxssiz; 664 struct proc *p; 665 vm_map_t map; 666 int error; 667 668 /* 669 * Indicate that we cannot gracefully error out any more, kill 670 * any other threads present, and set P_INEXEC to indicate that 671 * we are now messing with the process structure proper. 672 * 673 * If killalllwps() races return an error which coupled with 674 * vmspace_destroyed will cause us to exit. This is what we 675 * want since another thread is patiently waiting for us to exit 676 * in that case. 677 */ 678 p = curproc; 679 imgp->vmspace_destroyed = 1; 680 681 if (curthread->td_proc->p_nthreads > 1) { 682 error = killalllwps(1); 683 if (error) 684 return (error); 685 } 686 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 687 p->p_flag |= P_INEXEC; 688 689 /* 690 * Prevent a pending AIO from modifying the new address space. 691 */ 692 aio_proc_rundown(imgp->proc); 693 694 /* 695 * Blow away entire process VM, if address space not shared, 696 * otherwise, create a new VM space so that other threads are 697 * not disrupted. If we are execing a resident vmspace we 698 * create a duplicate of it and remap the stack. 699 * 700 * The exitingcnt test is not strictly necessary but has been 701 * included for code sanity (to make the code more deterministic). 702 */ 703 map = &vmspace->vm_map; 704 if (vmcopy) { 705 vmspace_exec(imgp->proc, vmcopy); 706 vmspace = imgp->proc->p_vmspace; 707 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 708 map = &vmspace->vm_map; 709 } else if (vmspace->vm_sysref.refcnt == 1 && 710 vmspace->vm_exitingcnt == 0) { 711 shmexit(vmspace); 712 if (vmspace->vm_upcalls) 713 upc_release(vmspace, ONLY_LWP_IN_PROC(imgp->proc)); 714 pmap_remove_pages(vmspace_pmap(vmspace), 715 0, VM_MAX_USER_ADDRESS); 716 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 717 } else { 718 vmspace_exec(imgp->proc, NULL); 719 vmspace = imgp->proc->p_vmspace; 720 map = &vmspace->vm_map; 721 } 722 723 /* Allocate a new stack */ 724 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 725 VM_PROT_ALL, VM_PROT_ALL, 0); 726 if (error) 727 return (error); 728 729 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 730 * VM_STACK case, but they are still used to monitor the size of the 731 * process stack so we can check the stack rlimit. 732 */ 733 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 734 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 735 736 return(0); 737 } 738 739 /* 740 * Copy out argument and environment strings from the old process 741 * address space into the temporary string buffer. 742 */ 743 int 744 exec_copyin_args(struct image_args *args, char *fname, 745 enum exec_path_segflg segflg, char **argv, char **envv) 746 { 747 char *argp, *envp; 748 int error = 0; 749 size_t length; 750 751 bzero(args, sizeof(*args)); 752 753 args->buf = objcache_get(exec_objcache, M_WAITOK); 754 if (args->buf == NULL) 755 return (ENOMEM); 756 args->begin_argv = args->buf; 757 args->endp = args->begin_argv; 758 args->space = ARG_MAX; 759 760 args->fname = args->buf + ARG_MAX; 761 762 /* 763 * Copy the file name. 764 */ 765 if (segflg == PATH_SYSSPACE) { 766 error = copystr(fname, args->fname, PATH_MAX, &length); 767 } else if (segflg == PATH_USERSPACE) { 768 error = copyinstr(fname, args->fname, PATH_MAX, &length); 769 } 770 771 /* 772 * Extract argument strings. argv may not be NULL. The argv 773 * array is terminated by a NULL entry. We special-case the 774 * situation where argv[0] is NULL by passing { filename, NULL } 775 * to the new program to guarentee that the interpreter knows what 776 * file to open in case we exec an interpreted file. Note that 777 * a NULL argv[0] terminates the argv[] array. 778 * 779 * XXX the special-casing of argv[0] is historical and needs to be 780 * revisited. 781 */ 782 if (argv == NULL) 783 error = EFAULT; 784 if (error == 0) { 785 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 786 if (argp == (caddr_t)-1) { 787 error = EFAULT; 788 break; 789 } 790 error = copyinstr(argp, args->endp, 791 args->space, &length); 792 if (error) { 793 if (error == ENAMETOOLONG) 794 error = E2BIG; 795 break; 796 } 797 args->space -= length; 798 args->endp += length; 799 args->argc++; 800 } 801 if (args->argc == 0 && error == 0) { 802 length = strlen(args->fname) + 1; 803 if (length > args->space) { 804 error = E2BIG; 805 } else { 806 bcopy(args->fname, args->endp, length); 807 args->space -= length; 808 args->endp += length; 809 args->argc++; 810 } 811 } 812 } 813 814 args->begin_envv = args->endp; 815 816 /* 817 * extract environment strings. envv may be NULL. 818 */ 819 if (envv && error == 0) { 820 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 821 if (envp == (caddr_t) -1) { 822 error = EFAULT; 823 break; 824 } 825 error = copyinstr(envp, args->endp, args->space, 826 &length); 827 if (error) { 828 if (error == ENAMETOOLONG) 829 error = E2BIG; 830 break; 831 } 832 args->space -= length; 833 args->endp += length; 834 args->envc++; 835 } 836 } 837 return (error); 838 } 839 840 void 841 exec_free_args(struct image_args *args) 842 { 843 if (args->buf) { 844 objcache_put(exec_objcache, args->buf); 845 args->buf = NULL; 846 } 847 } 848 849 /* 850 * Copy strings out to the new process address space, constructing 851 * new arg and env vector tables. Return a pointer to the base 852 * so that it can be used as the initial stack pointer. 853 */ 854 register_t * 855 exec_copyout_strings(struct image_params *imgp) 856 { 857 int argc, envc, sgap; 858 char **vectp; 859 char *stringp, *destp; 860 register_t *stack_base; 861 struct ps_strings *arginfo; 862 int szsigcode; 863 864 /* 865 * Calculate string base and vector table pointers. 866 * Also deal with signal trampoline code for this exec type. 867 */ 868 arginfo = (struct ps_strings *)PS_STRINGS; 869 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 870 if (stackgap_random != 0) 871 sgap = ALIGN(karc4random() & (stackgap_random - 1)); 872 else 873 sgap = 0; 874 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - sgap - 875 roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 876 877 /* 878 * install sigcode 879 */ 880 if (szsigcode) 881 copyout(imgp->proc->p_sysent->sv_sigcode, 882 ((caddr_t)arginfo - szsigcode), szsigcode); 883 884 /* 885 * If we have a valid auxargs ptr, prepare some room 886 * on the stack. 887 * 888 * The '+ 2' is for the null pointers at the end of each of the 889 * arg and env vector sets, and 'AT_COUNT*2' is room for the 890 * ELF Auxargs data. 891 */ 892 if (imgp->auxargs) { 893 vectp = (char **)(destp - (imgp->args->argc + 894 imgp->args->envc + 2 + AT_COUNT * 2) * sizeof(char*)); 895 } else { 896 vectp = (char **)(destp - (imgp->args->argc + 897 imgp->args->envc + 2) * sizeof(char*)); 898 } 899 900 /* 901 * NOTE: don't bother aligning the stack here for GCC 2.x, it will 902 * be done in crt1.o. Note that GCC 3.x aligns the stack in main. 903 */ 904 905 /* 906 * vectp also becomes our initial stack base 907 */ 908 stack_base = (register_t *)vectp; 909 910 stringp = imgp->args->begin_argv; 911 argc = imgp->args->argc; 912 envc = imgp->args->envc; 913 914 /* 915 * Copy out strings - arguments and environment. 916 */ 917 copyout(stringp, destp, ARG_MAX - imgp->args->space); 918 919 /* 920 * Fill in "ps_strings" struct for ps, w, etc. 921 */ 922 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 923 suword(&arginfo->ps_nargvstr, argc); 924 925 /* 926 * Fill in argument portion of vector table. 927 */ 928 for (; argc > 0; --argc) { 929 suword(vectp++, (long)(intptr_t)destp); 930 while (*stringp++ != 0) 931 destp++; 932 destp++; 933 } 934 935 /* a null vector table pointer separates the argp's from the envp's */ 936 suword(vectp++, 0); 937 938 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 939 suword(&arginfo->ps_nenvstr, envc); 940 941 /* 942 * Fill in environment portion of vector table. 943 */ 944 for (; envc > 0; --envc) { 945 suword(vectp++, (long)(intptr_t)destp); 946 while (*stringp++ != 0) 947 destp++; 948 destp++; 949 } 950 951 /* end of vector table is a null pointer */ 952 suword(vectp, 0); 953 954 return (stack_base); 955 } 956 957 /* 958 * Check permissions of file to execute. 959 * Return 0 for success or error code on failure. 960 */ 961 int 962 exec_check_permissions(struct image_params *imgp) 963 { 964 struct proc *p = imgp->proc; 965 struct vnode *vp = imgp->vp; 966 struct vattr *attr = imgp->attr; 967 int error; 968 969 /* Get file attributes */ 970 error = VOP_GETATTR(vp, attr); 971 if (error) 972 return (error); 973 974 /* 975 * 1) Check if file execution is disabled for the filesystem that this 976 * file resides on. 977 * 2) Insure that at least one execute bit is on - otherwise root 978 * will always succeed, and we don't want to happen unless the 979 * file really is executable. 980 * 3) Insure that the file is a regular file. 981 */ 982 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 983 ((attr->va_mode & 0111) == 0) || 984 (attr->va_type != VREG)) { 985 return (EACCES); 986 } 987 988 /* 989 * Zero length files can't be exec'd 990 */ 991 if (attr->va_size == 0) 992 return (ENOEXEC); 993 994 /* 995 * Check for execute permission to file based on current credentials. 996 */ 997 error = VOP_ACCESS(vp, VEXEC, p->p_ucred); 998 if (error) 999 return (error); 1000 1001 /* 1002 * Check number of open-for-writes on the file and deny execution 1003 * if there are any. 1004 */ 1005 if (vp->v_writecount) 1006 return (ETXTBSY); 1007 1008 /* 1009 * Call filesystem specific open routine, which allows us to read, 1010 * write, and mmap the file. Without the VOP_OPEN we can only 1011 * stat the file. 1012 */ 1013 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1014 if (error) 1015 return (error); 1016 1017 return (0); 1018 } 1019 1020 /* 1021 * Exec handler registration 1022 */ 1023 int 1024 exec_register(const struct execsw *execsw_arg) 1025 { 1026 const struct execsw **es, **xs, **newexecsw; 1027 int count = 2; /* New slot and trailing NULL */ 1028 1029 if (execsw) 1030 for (es = execsw; *es; es++) 1031 count++; 1032 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1033 if (newexecsw == NULL) 1034 return ENOMEM; 1035 xs = newexecsw; 1036 if (execsw) 1037 for (es = execsw; *es; es++) 1038 *xs++ = *es; 1039 *xs++ = execsw_arg; 1040 *xs = NULL; 1041 if (execsw) 1042 kfree(execsw, M_TEMP); 1043 execsw = newexecsw; 1044 return 0; 1045 } 1046 1047 int 1048 exec_unregister(const struct execsw *execsw_arg) 1049 { 1050 const struct execsw **es, **xs, **newexecsw; 1051 int count = 1; 1052 1053 if (execsw == NULL) 1054 panic("unregister with no handlers left?"); 1055 1056 for (es = execsw; *es; es++) { 1057 if (*es == execsw_arg) 1058 break; 1059 } 1060 if (*es == NULL) 1061 return ENOENT; 1062 for (es = execsw; *es; es++) 1063 if (*es != execsw_arg) 1064 count++; 1065 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1066 if (newexecsw == NULL) 1067 return ENOMEM; 1068 xs = newexecsw; 1069 for (es = execsw; *es; es++) 1070 if (*es != execsw_arg) 1071 *xs++ = *es; 1072 *xs = NULL; 1073 if (execsw) 1074 kfree(execsw, M_TEMP); 1075 execsw = newexecsw; 1076 return 0; 1077 } 1078