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 */ 28 29 #include <sys/param.h> 30 #include <sys/systm.h> 31 #include <sys/sysproto.h> 32 #include <sys/kernel.h> 33 #include <sys/mount.h> 34 #include <sys/filedesc.h> 35 #include <sys/fcntl.h> 36 #include <sys/acct.h> 37 #include <sys/exec.h> 38 #include <sys/imgact.h> 39 #include <sys/imgact_elf.h> 40 #include <sys/kern_syscall.h> 41 #include <sys/wait.h> 42 #include <sys/malloc.h> 43 #include <sys/proc.h> 44 #include <sys/priv.h> 45 #include <sys/ktrace.h> 46 #include <sys/signalvar.h> 47 #include <sys/pioctl.h> 48 #include <sys/nlookup.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/libkern.h> 55 56 #include <cpu/lwbuf.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/refcount.h> 74 #include <sys/thread2.h> 75 #include <sys/mplock2.h> 76 #include <vm/vm_page2.h> 77 78 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 79 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 80 81 static register_t *exec_copyout_strings (struct image_params *); 82 83 /* XXX This should be vm_size_t. */ 84 static u_long ps_strings = PS_STRINGS; 85 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 86 87 /* XXX This should be vm_size_t. */ 88 static u_long usrstack = USRSTACK; 89 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 90 91 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 92 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 93 &ps_arg_cache_limit, 0, ""); 94 95 int ps_argsopen = 1; 96 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 97 98 static int ktrace_suid = 0; 99 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, ""); 100 101 void print_execve_args(struct image_args *args); 102 int debug_execve_args = 0; 103 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 104 0, ""); 105 106 /* 107 * Exec arguments object cache 108 */ 109 static struct objcache *exec_objcache; 110 111 static 112 void 113 exec_objcache_init(void *arg __unused) 114 { 115 int cluster_limit; 116 size_t limsize; 117 118 /* 119 * Maximum number of concurrent execs. This can be limiting on 120 * systems with a lot of cpu cores but it also eats a significant 121 * amount of memory. 122 */ 123 cluster_limit = (ncpus < 16) ? 16 : ncpus; 124 limsize = kmem_lim_size(); 125 if (limsize > 7 * 1024) 126 cluster_limit *= 2; 127 if (limsize > 15 * 1024) 128 cluster_limit *= 2; 129 130 exec_objcache = objcache_create_mbacked( 131 M_EXECARGS, PATH_MAX + ARG_MAX, 132 cluster_limit, 8, 133 NULL, NULL, NULL); 134 } 135 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 136 137 /* 138 * stackgap_random specifies if the stackgap should have a random size added 139 * to it. It must be a power of 2. If non-zero, the stack gap will be 140 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 141 */ 142 static int stackgap_random = 1024; 143 static int 144 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 145 { 146 int error, new_val; 147 new_val = stackgap_random; 148 error = sysctl_handle_int(oidp, &new_val, 0, req); 149 if (error != 0 || req->newptr == NULL) 150 return (error); 151 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val))) 152 return (EINVAL); 153 stackgap_random = new_val; 154 155 return(0); 156 } 157 158 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT, 159 0, 0, sysctl_kern_stackgap, "I", 160 "Max random stack gap (power of 2), static gap if negative"); 161 162 void 163 print_execve_args(struct image_args *args) 164 { 165 char *cp; 166 int ndx; 167 168 cp = args->begin_argv; 169 for (ndx = 0; ndx < args->argc; ndx++) { 170 kprintf("\targv[%d]: %s\n", ndx, cp); 171 while (*cp++ != '\0'); 172 } 173 for (ndx = 0; ndx < args->envc; ndx++) { 174 kprintf("\tenvv[%d]: %s\n", ndx, cp); 175 while (*cp++ != '\0'); 176 } 177 } 178 179 /* 180 * Each of the items is a pointer to a `const struct execsw', hence the 181 * double pointer here. 182 */ 183 static const struct execsw **execsw; 184 185 /* 186 * Replace current vmspace with a new binary. 187 * Returns 0 on success, > 0 on recoverable error (use as errno). 188 * Returns -1 on lethal error which demands killing of the current 189 * process! 190 */ 191 int 192 kern_execve(struct nlookupdata *nd, struct image_args *args) 193 { 194 struct thread *td = curthread; 195 struct lwp *lp = td->td_lwp; 196 struct proc *p = td->td_proc; 197 struct vnode *ovp; 198 register_t *stack_base; 199 struct pargs *pa; 200 struct sigacts *ops; 201 struct sigacts *nps; 202 int error, len, i; 203 struct image_params image_params, *imgp; 204 struct vattr attr; 205 int (*img_first) (struct image_params *); 206 207 if (debug_execve_args) { 208 kprintf("%s()\n", __func__); 209 print_execve_args(args); 210 } 211 212 KKASSERT(p); 213 lwkt_gettoken(&p->p_token); 214 imgp = &image_params; 215 216 /* 217 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 218 * no modifications to the process at all until we get there. 219 * 220 * Note that multiple threads may be trying to exec at the same 221 * time. exec_new_vmspace() handles that too. 222 */ 223 224 /* 225 * Initialize part of the common data 226 */ 227 imgp->proc = p; 228 imgp->args = args; 229 imgp->attr = &attr; 230 imgp->entry_addr = 0; 231 imgp->resident = 0; 232 imgp->vmspace_destroyed = 0; 233 imgp->interpreted = 0; 234 imgp->interpreter_name[0] = 0; 235 imgp->auxargs = NULL; 236 imgp->vp = NULL; 237 imgp->firstpage = NULL; 238 imgp->ps_strings = 0; 239 imgp->execpath = imgp->freepath = NULL; 240 imgp->execpathp = 0; 241 imgp->image_header = NULL; 242 243 interpret: 244 245 /* 246 * Translate the file name to a vnode. Unlock the cache entry to 247 * improve parallelism for programs exec'd in parallel. 248 */ 249 if ((error = nlookup(nd)) != 0) 250 goto exec_fail; 251 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_EXCLUSIVE, &imgp->vp); 252 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 253 nd->nl_flags &= ~NLC_NCPISLOCKED; 254 cache_unlock(&nd->nl_nch); 255 if (error) 256 goto exec_fail; 257 258 /* 259 * Check file permissions (also 'opens' file). 260 * Include also the top level mount in the check. 261 */ 262 error = exec_check_permissions(imgp, nd->nl_nch.mount); 263 if (error) { 264 vn_unlock(imgp->vp); 265 goto exec_fail_dealloc; 266 } 267 268 error = exec_map_first_page(imgp); 269 vn_unlock(imgp->vp); 270 if (error) 271 goto exec_fail_dealloc; 272 273 imgp->proc->p_osrel = 0; 274 275 if (debug_execve_args && imgp->interpreted) { 276 kprintf(" target is interpreted -- recursive pass\n"); 277 kprintf(" interpreter: %s\n", imgp->interpreter_name); 278 print_execve_args(args); 279 } 280 281 /* 282 * If the current process has a special image activator it 283 * wants to try first, call it. For example, emulating shell 284 * scripts differently. 285 */ 286 error = -1; 287 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 288 error = img_first(imgp); 289 290 /* 291 * If the vnode has a registered vmspace, exec the vmspace 292 */ 293 if (error == -1 && imgp->vp->v_resident) { 294 error = exec_resident_imgact(imgp); 295 } 296 297 /* 298 * Loop through the list of image activators, calling each one. 299 * An activator returns -1 if there is no match, 0 on success, 300 * and an error otherwise. 301 */ 302 for (i = 0; error == -1 && execsw[i]; ++i) { 303 if (execsw[i]->ex_imgact == NULL || 304 execsw[i]->ex_imgact == img_first) { 305 continue; 306 } 307 error = (*execsw[i]->ex_imgact)(imgp); 308 } 309 310 if (error) { 311 if (error == -1) 312 error = ENOEXEC; 313 goto exec_fail_dealloc; 314 } 315 316 /* 317 * Special interpreter operation, cleanup and loop up to try to 318 * activate the interpreter. 319 */ 320 if (imgp->interpreted) { 321 exec_unmap_first_page(imgp); 322 nlookup_done(nd); 323 vrele(imgp->vp); 324 imgp->vp = NULL; 325 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 326 NLC_FOLLOW); 327 if (error) 328 goto exec_fail; 329 goto interpret; 330 } 331 332 /* 333 * Do the best to calculate the full path to the image file 334 */ 335 if (imgp->auxargs != NULL && 336 ((args->fname != NULL && args->fname[0] == '/') || 337 vn_fullpath(imgp->proc, 338 imgp->vp, 339 &imgp->execpath, 340 &imgp->freepath, 341 0) != 0)) 342 imgp->execpath = args->fname; 343 344 /* 345 * Copy out strings (args and env) and initialize stack base 346 */ 347 stack_base = exec_copyout_strings(imgp); 348 p->p_vmspace->vm_minsaddr = (char *)stack_base; 349 350 /* 351 * If custom stack fixup routine present for this process 352 * let it do the stack setup. If we are running a resident 353 * image there is no auxinfo or other image activator context 354 * so don't try to add fixups to the stack. 355 * 356 * Else stuff argument count as first item on stack 357 */ 358 if (p->p_sysent->sv_fixup && imgp->resident == 0) 359 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 360 else 361 suword(--stack_base, imgp->args->argc); 362 363 /* 364 * For security and other reasons, the file descriptor table cannot 365 * be shared after an exec. 366 */ 367 if (p->p_fd->fd_refcnt > 1) { 368 struct filedesc *tmp; 369 370 error = fdcopy(p, &tmp); 371 if (error != 0) 372 goto exec_fail; 373 fdfree(p, tmp); 374 } 375 376 /* 377 * For security and other reasons, signal handlers cannot 378 * be shared after an exec. The new proces gets a copy of the old 379 * handlers. In execsigs(), the new process will have its signals 380 * reset. 381 */ 382 ops = p->p_sigacts; 383 if (ops->ps_refcnt > 1) { 384 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK); 385 bcopy(ops, nps, sizeof(*nps)); 386 refcount_init(&nps->ps_refcnt, 1); 387 p->p_sigacts = nps; 388 if (refcount_release(&ops->ps_refcnt)) { 389 kfree(ops, M_SUBPROC); 390 ops = NULL; 391 } 392 } 393 394 /* 395 * For security and other reasons virtual kernels cannot be 396 * inherited by an exec. This also allows a virtual kernel 397 * to fork/exec unrelated applications. 398 */ 399 if (p->p_vkernel) 400 vkernel_exit(p); 401 402 /* Stop profiling */ 403 stopprofclock(p); 404 405 /* close files on exec */ 406 fdcloseexec(p); 407 408 /* reset caught signals */ 409 execsigs(p); 410 411 /* name this process - nameiexec(p, ndp) */ 412 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 413 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 414 p->p_comm[len] = 0; 415 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 416 417 /* 418 * mark as execed, wakeup the process that vforked (if any) and tell 419 * it that it now has its own resources back 420 * 421 * We are using the P_PPWAIT as an interlock so an atomic op is 422 * necessary to synchronize with the parent's cpu. 423 */ 424 p->p_flags |= P_EXEC; 425 if (p->p_pptr && (p->p_flags & P_PPWAIT)) { 426 atomic_clear_int(&p->p_flags, P_PPWAIT); 427 wakeup(p->p_pptr); 428 } 429 430 /* 431 * Implement image setuid/setgid. 432 * 433 * Don't honor setuid/setgid if the filesystem prohibits it or if 434 * the process is being traced. 435 */ 436 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 437 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 438 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 439 (p->p_flags & P_TRACED) == 0) { 440 /* 441 * Turn off syscall tracing for set-id programs, except for 442 * root. Record any set-id flags first to make sure that 443 * we do not regain any tracing during a possible block. 444 */ 445 setsugid(); 446 if (p->p_tracenode && ktrace_suid == 0 && 447 priv_check(td, PRIV_ROOT) != 0) { 448 ktrdestroy(&p->p_tracenode); 449 p->p_traceflag = 0; 450 } 451 /* Close any file descriptors 0..2 that reference procfs */ 452 setugidsafety(p); 453 /* Make sure file descriptors 0..2 are in use. */ 454 error = fdcheckstd(lp); 455 if (error != 0) 456 goto exec_fail_dealloc; 457 /* 458 * Set the new credentials. 459 */ 460 cratom(&p->p_ucred); 461 if (attr.va_mode & VSUID) 462 change_euid(attr.va_uid); 463 if (attr.va_mode & VSGID) 464 p->p_ucred->cr_gid = attr.va_gid; 465 466 /* 467 * Clear local varsym variables 468 */ 469 varsymset_clean(&p->p_varsymset); 470 } else { 471 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 472 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 473 p->p_flags &= ~P_SUGID; 474 } 475 476 /* 477 * Implement correct POSIX saved-id behavior. 478 */ 479 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 480 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 481 cratom(&p->p_ucred); 482 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 483 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 484 } 485 486 /* 487 * Store the vp for use in procfs. Be sure to keep p_textvp 488 * consistent if we block during the switch-over. 489 */ 490 ovp = p->p_textvp; 491 vref(imgp->vp); /* ref new vp */ 492 p->p_textvp = imgp->vp; 493 if (ovp) /* release old vp */ 494 vrele(ovp); 495 496 /* Release old namecache handle to text file */ 497 if (p->p_textnch.ncp) 498 cache_drop(&p->p_textnch); 499 500 if (nd->nl_nch.mount) 501 cache_copy(&nd->nl_nch, &p->p_textnch); 502 503 /* 504 * Notify others that we exec'd, and clear the P_INEXEC flag 505 * as we're now a bona fide freshly-execed process. 506 */ 507 KNOTE(&p->p_klist, NOTE_EXEC); 508 p->p_flags &= ~P_INEXEC; 509 if (p->p_stops) 510 wakeup(&p->p_stype); 511 512 /* 513 * If tracing the process, trap to debugger so breakpoints 514 * can be set before the program executes. 515 */ 516 STOPEVENT(p, S_EXEC, 0); 517 518 if (p->p_flags & P_TRACED) 519 ksignal(p, SIGTRAP); 520 521 /* clear "fork but no exec" flag, as we _are_ execing */ 522 p->p_acflag &= ~AFORK; 523 524 /* Set values passed into the program in registers. */ 525 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 526 imgp->ps_strings); 527 528 /* Set the access time on the vnode */ 529 vn_mark_atime(imgp->vp, td); 530 531 /* 532 * Free any previous argument cache 533 */ 534 pa = p->p_args; 535 p->p_args = NULL; 536 if (pa && refcount_release(&pa->ar_ref)) { 537 kfree(pa, M_PARGS); 538 pa = NULL; 539 } 540 541 /* 542 * Cache arguments if they fit inside our allowance 543 */ 544 i = imgp->args->begin_envv - imgp->args->begin_argv; 545 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 546 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 547 refcount_init(&pa->ar_ref, 1); 548 pa->ar_length = i; 549 bcopy(imgp->args->begin_argv, pa->ar_args, i); 550 KKASSERT(p->p_args == NULL); 551 p->p_args = pa; 552 } 553 554 exec_fail_dealloc: 555 556 /* 557 * free various allocated resources 558 */ 559 if (imgp->firstpage) 560 exec_unmap_first_page(imgp); 561 562 if (imgp->vp) { 563 vrele(imgp->vp); 564 imgp->vp = NULL; 565 } 566 567 if (imgp->freepath) 568 kfree(imgp->freepath, M_TEMP); 569 570 if (error == 0) { 571 ++mycpu->gd_cnt.v_exec; 572 lwkt_reltoken(&p->p_token); 573 return (0); 574 } 575 576 exec_fail: 577 /* 578 * we're done here, clear P_INEXEC if we were the ones that 579 * set it. Otherwise if vmspace_destroyed is still set we 580 * raced another thread and that thread is responsible for 581 * clearing it. 582 */ 583 if (imgp->vmspace_destroyed & 2) { 584 p->p_flags &= ~P_INEXEC; 585 if (p->p_stops) 586 wakeup(&p->p_stype); 587 } 588 lwkt_reltoken(&p->p_token); 589 if (imgp->vmspace_destroyed) { 590 /* 591 * Sorry, no more process anymore. exit gracefully. 592 * However we can't die right here, because our 593 * caller might have to clean up, so indicate a 594 * lethal error by returning -1. 595 */ 596 return(-1); 597 } else { 598 return(error); 599 } 600 } 601 602 /* 603 * execve() system call. 604 */ 605 int 606 sys_execve(struct execve_args *uap) 607 { 608 struct nlookupdata nd; 609 struct image_args args; 610 int error; 611 612 bzero(&args, sizeof(args)); 613 614 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 615 if (error == 0) { 616 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 617 uap->argv, uap->envv); 618 } 619 if (error == 0) 620 error = kern_execve(&nd, &args); 621 nlookup_done(&nd); 622 exec_free_args(&args); 623 624 if (error < 0) { 625 /* We hit a lethal error condition. Let's die now. */ 626 exit1(W_EXITCODE(0, SIGABRT)); 627 /* NOTREACHED */ 628 } 629 630 /* 631 * The syscall result is returned in registers to the new program. 632 * Linux will register %edx as an atexit function and we must be 633 * sure to set it to 0. XXX 634 */ 635 if (error == 0) 636 uap->sysmsg_result64 = 0; 637 638 return (error); 639 } 640 641 int 642 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 643 struct lwbuf **plwb, const char **pdata) 644 { 645 int rv; 646 vm_page_t ma; 647 vm_page_t m; 648 vm_object_t object; 649 650 /* 651 * The file has to be mappable. 652 */ 653 if ((object = imgp->vp->v_object) == NULL) 654 return (EIO); 655 656 if (pageno >= object->size) 657 return (EIO); 658 659 /* 660 * Shortcut using shared locks, improve concurrent execs. 661 */ 662 vm_object_hold_shared(object); 663 m = vm_page_lookup(object, pageno); 664 if (m) { 665 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 666 vm_page_hold(m); 667 vm_page_sleep_busy(m, FALSE, "execpg"); 668 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 669 vm_object_drop(object); 670 goto done; 671 } 672 } 673 vm_page_unhold(m); 674 } 675 vm_object_drop(object); 676 677 /* 678 * Do it the hard way 679 */ 680 vm_object_hold(object); 681 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 682 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 683 ma = m; 684 685 /* 686 * get_pages unbusies all the requested pages except the 687 * primary page (at index 0 in this case). The primary 688 * page may have been wired during the pagein (e.g. by 689 * the buffer cache) so vnode_pager_freepage() must be 690 * used to properly release it. 691 */ 692 rv = vm_pager_get_page(object, &ma, 1); 693 m = vm_page_lookup(object, pageno); 694 695 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 696 if (m) { 697 vm_page_protect(m, VM_PROT_NONE); 698 vnode_pager_freepage(m); 699 } 700 vm_object_drop(object); 701 return EIO; 702 } 703 } 704 vm_page_hold(m); 705 vm_page_wakeup(m); /* unbusy the page */ 706 vm_object_drop(object); 707 708 done: 709 *plwb = lwbuf_alloc(m, *plwb); 710 *pdata = (void *)lwbuf_kva(*plwb); 711 712 return (0); 713 } 714 715 /* 716 * Map the first page of an executable image. 717 * 718 * NOTE: If the mapping fails we have to NULL-out firstpage which may 719 * still be pointing to our supplied lwp structure. 720 */ 721 int 722 exec_map_first_page(struct image_params *imgp) 723 { 724 int err; 725 726 if (imgp->firstpage) 727 exec_unmap_first_page(imgp); 728 729 imgp->firstpage = &imgp->firstpage_cache; 730 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 731 732 if (err) { 733 imgp->firstpage = NULL; 734 return err; 735 } 736 737 return 0; 738 } 739 740 void 741 exec_unmap_page(struct lwbuf *lwb) 742 { 743 vm_page_t m; 744 745 crit_enter(); 746 if (lwb != NULL) { 747 m = lwbuf_page(lwb); 748 lwbuf_free(lwb); 749 vm_page_unhold(m); 750 } 751 crit_exit(); 752 } 753 754 void 755 exec_unmap_first_page(struct image_params *imgp) 756 { 757 exec_unmap_page(imgp->firstpage); 758 imgp->firstpage = NULL; 759 imgp->image_header = NULL; 760 } 761 762 /* 763 * Destroy old address space, and allocate a new stack 764 * The new stack is only SGROWSIZ large because it is grown 765 * automatically in trap.c. 766 * 767 * This is the point of no return. 768 */ 769 int 770 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 771 { 772 struct vmspace *vmspace = imgp->proc->p_vmspace; 773 vm_offset_t stack_addr = USRSTACK - maxssiz; 774 struct proc *p; 775 vm_map_t map; 776 int error; 777 778 /* 779 * Indicate that we cannot gracefully error out any more, kill 780 * any other threads present, and set P_INEXEC to indicate that 781 * we are now messing with the process structure proper. 782 * 783 * If killalllwps() races return an error which coupled with 784 * vmspace_destroyed will cause us to exit. This is what we 785 * want since another thread is patiently waiting for us to exit 786 * in that case. 787 */ 788 p = curproc; 789 imgp->vmspace_destroyed = 1; 790 791 if (curthread->td_proc->p_nthreads > 1) { 792 error = killalllwps(1); 793 if (error) 794 return (error); 795 } 796 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 797 p->p_flags |= P_INEXEC; 798 799 /* 800 * Tell procfs to release its hold on the process. It 801 * will return EAGAIN. 802 */ 803 if (p->p_stops) 804 wakeup(&p->p_stype); 805 806 /* 807 * After setting P_INEXEC wait for any remaining references to 808 * the process (p) to go away. 809 * 810 * In particular, a vfork/exec sequence will replace p->p_vmspace 811 * and we must interlock anyone trying to access the space (aka 812 * procfs or sys_process.c calling procfs_domem()). 813 * 814 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us. 815 */ 816 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0)); 817 818 /* 819 * Blow away entire process VM, if address space not shared, 820 * otherwise, create a new VM space so that other threads are 821 * not disrupted. If we are execing a resident vmspace we 822 * create a duplicate of it and remap the stack. 823 */ 824 map = &vmspace->vm_map; 825 if (vmcopy) { 826 vmspace_exec(imgp->proc, vmcopy); 827 vmspace = imgp->proc->p_vmspace; 828 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 829 map = &vmspace->vm_map; 830 } else if (vmspace->vm_sysref.refcnt == 1) { 831 shmexit(vmspace); 832 pmap_remove_pages(vmspace_pmap(vmspace), 833 0, VM_MAX_USER_ADDRESS); 834 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 835 } else { 836 vmspace_exec(imgp->proc, NULL); 837 vmspace = imgp->proc->p_vmspace; 838 map = &vmspace->vm_map; 839 } 840 841 /* Allocate a new stack */ 842 error = vm_map_stack(&vmspace->vm_map, stack_addr, (vm_size_t)maxssiz, 843 0, VM_PROT_ALL, VM_PROT_ALL, 0); 844 if (error) 845 return (error); 846 847 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 848 * VM_STACK case, but they are still used to monitor the size of the 849 * process stack so we can check the stack rlimit. 850 */ 851 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 852 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 853 854 return(0); 855 } 856 857 /* 858 * Copy out argument and environment strings from the old process 859 * address space into the temporary string buffer. 860 */ 861 int 862 exec_copyin_args(struct image_args *args, char *fname, 863 enum exec_path_segflg segflg, char **argv, char **envv) 864 { 865 char *argp, *envp; 866 int error = 0; 867 size_t length; 868 869 args->buf = objcache_get(exec_objcache, M_WAITOK); 870 if (args->buf == NULL) 871 return (ENOMEM); 872 args->begin_argv = args->buf; 873 args->endp = args->begin_argv; 874 args->space = ARG_MAX; 875 876 args->fname = args->buf + ARG_MAX; 877 878 /* 879 * Copy the file name. 880 */ 881 if (segflg == PATH_SYSSPACE) { 882 error = copystr(fname, args->fname, PATH_MAX, &length); 883 } else if (segflg == PATH_USERSPACE) { 884 error = copyinstr(fname, args->fname, PATH_MAX, &length); 885 } 886 887 /* 888 * Extract argument strings. argv may not be NULL. The argv 889 * array is terminated by a NULL entry. We special-case the 890 * situation where argv[0] is NULL by passing { filename, NULL } 891 * to the new program to guarentee that the interpreter knows what 892 * file to open in case we exec an interpreted file. Note that 893 * a NULL argv[0] terminates the argv[] array. 894 * 895 * XXX the special-casing of argv[0] is historical and needs to be 896 * revisited. 897 */ 898 if (argv == NULL) 899 error = EFAULT; 900 if (error == 0) { 901 while ((argp = (caddr_t)(intptr_t)fuword(argv++)) != NULL) { 902 if (argp == (caddr_t)-1) { 903 error = EFAULT; 904 break; 905 } 906 error = copyinstr(argp, args->endp, 907 args->space, &length); 908 if (error) { 909 if (error == ENAMETOOLONG) 910 error = E2BIG; 911 break; 912 } 913 args->space -= length; 914 args->endp += length; 915 args->argc++; 916 } 917 if (args->argc == 0 && error == 0) { 918 length = strlen(args->fname) + 1; 919 if (length > args->space) { 920 error = E2BIG; 921 } else { 922 bcopy(args->fname, args->endp, length); 923 args->space -= length; 924 args->endp += length; 925 args->argc++; 926 } 927 } 928 } 929 930 args->begin_envv = args->endp; 931 932 /* 933 * extract environment strings. envv may be NULL. 934 */ 935 if (envv && error == 0) { 936 while ((envp = (caddr_t) (intptr_t) fuword(envv++))) { 937 if (envp == (caddr_t) -1) { 938 error = EFAULT; 939 break; 940 } 941 error = copyinstr(envp, args->endp, 942 args->space, &length); 943 if (error) { 944 if (error == ENAMETOOLONG) 945 error = E2BIG; 946 break; 947 } 948 args->space -= length; 949 args->endp += length; 950 args->envc++; 951 } 952 } 953 return (error); 954 } 955 956 void 957 exec_free_args(struct image_args *args) 958 { 959 if (args->buf) { 960 objcache_put(exec_objcache, args->buf); 961 args->buf = NULL; 962 } 963 } 964 965 /* 966 * Copy strings out to the new process address space, constructing 967 * new arg and env vector tables. Return a pointer to the base 968 * so that it can be used as the initial stack pointer. 969 * 970 * The format is, roughly: 971 * 972 * [argv[]] <-- vectp 973 * [envp[]] 974 * [ELF_Auxargs] 975 * 976 * [args & env] <-- destp 977 * [sgap] 978 * [SPARE_USRSPACE] 979 * [execpath] 980 * [szsigcode] 981 * [ps_strings] top of user stack 982 * 983 */ 984 register_t * 985 exec_copyout_strings(struct image_params *imgp) 986 { 987 int argc, envc, sgap; 988 int gap; 989 int argsenvspace; 990 char **vectp; 991 char *stringp, *destp; 992 register_t *stack_base; 993 struct ps_strings *arginfo; 994 size_t execpath_len; 995 int szsigcode; 996 997 /* 998 * Calculate string base and vector table pointers. 999 * Also deal with signal trampoline code for this exec type. 1000 */ 1001 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1002 execpath_len = strlen(imgp->execpath) + 1; 1003 else 1004 execpath_len = 0; 1005 arginfo = (struct ps_strings *)PS_STRINGS; 1006 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 1007 1008 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 1009 gap = stackgap_random; 1010 cpu_ccfence(); 1011 if (gap != 0) { 1012 if (gap < 0) 1013 sgap = ALIGN(-gap); 1014 else 1015 sgap = ALIGN(karc4random() & (gap - 1)); 1016 } else { 1017 sgap = 0; 1018 } 1019 1020 /* 1021 * Calculate destp, which points to [args & env] and above. 1022 */ 1023 destp = (caddr_t)arginfo - 1024 szsigcode - 1025 roundup(execpath_len, sizeof(char *)) - 1026 SPARE_USRSPACE - 1027 sgap - 1028 argsenvspace; 1029 1030 /* 1031 * install sigcode 1032 */ 1033 if (szsigcode) { 1034 copyout(imgp->proc->p_sysent->sv_sigcode, 1035 ((caddr_t)arginfo - szsigcode), szsigcode); 1036 } 1037 1038 /* 1039 * Copy the image path for the rtld 1040 */ 1041 if (execpath_len) { 1042 imgp->execpathp = (uintptr_t)arginfo 1043 - szsigcode 1044 - roundup(execpath_len, sizeof(char *)); 1045 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 1046 } 1047 1048 /* 1049 * Calculate base for argv[], envp[], and ELF_Auxargs. 1050 */ 1051 vectp = (char **)destp - (AT_COUNT * 2); 1052 vectp -= imgp->args->argc + imgp->args->envc + 2; 1053 1054 stack_base = (register_t *)vectp; 1055 1056 stringp = imgp->args->begin_argv; 1057 argc = imgp->args->argc; 1058 envc = imgp->args->envc; 1059 1060 /* 1061 * Copy out strings - arguments and environment (at destp) 1062 */ 1063 copyout(stringp, destp, ARG_MAX - imgp->args->space); 1064 1065 /* 1066 * Fill in "ps_strings" struct for ps, w, etc. 1067 */ 1068 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1069 suword32(&arginfo->ps_nargvstr, argc); 1070 1071 /* 1072 * Fill in argument portion of vector table. 1073 */ 1074 for (; argc > 0; --argc) { 1075 suword(vectp++, (long)(intptr_t)destp); 1076 while (*stringp++ != 0) 1077 destp++; 1078 destp++; 1079 } 1080 1081 /* a null vector table pointer separates the argp's from the envp's */ 1082 suword(vectp++, 0); 1083 1084 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1085 suword32(&arginfo->ps_nenvstr, envc); 1086 1087 /* 1088 * Fill in environment portion of vector table. 1089 */ 1090 for (; envc > 0; --envc) { 1091 suword(vectp++, (long)(intptr_t)destp); 1092 while (*stringp++ != 0) 1093 destp++; 1094 destp++; 1095 } 1096 1097 /* end of vector table is a null pointer */ 1098 suword(vectp, 0); 1099 1100 return (stack_base); 1101 } 1102 1103 /* 1104 * Check permissions of file to execute. 1105 * Return 0 for success or error code on failure. 1106 */ 1107 int 1108 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 1109 { 1110 struct proc *p = imgp->proc; 1111 struct vnode *vp = imgp->vp; 1112 struct vattr *attr = imgp->attr; 1113 int error; 1114 1115 /* Get file attributes */ 1116 error = VOP_GETATTR(vp, attr); 1117 if (error) 1118 return (error); 1119 1120 /* 1121 * 1) Check if file execution is disabled for the filesystem that this 1122 * file resides on. 1123 * 2) Insure that at least one execute bit is on - otherwise root 1124 * will always succeed, and we don't want to happen unless the 1125 * file really is executable. 1126 * 3) Insure that the file is a regular file. 1127 */ 1128 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1129 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1130 ((attr->va_mode & 0111) == 0) || 1131 (attr->va_type != VREG)) { 1132 return (EACCES); 1133 } 1134 1135 /* 1136 * Zero length files can't be exec'd 1137 */ 1138 if (attr->va_size == 0) 1139 return (ENOEXEC); 1140 1141 /* 1142 * Check for execute permission to file based on current credentials. 1143 */ 1144 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1145 if (error) 1146 return (error); 1147 1148 /* 1149 * Check number of open-for-writes on the file and deny execution 1150 * if there are any. 1151 */ 1152 if (vp->v_writecount) 1153 return (ETXTBSY); 1154 1155 /* 1156 * Call filesystem specific open routine, which allows us to read, 1157 * write, and mmap the file. Without the VOP_OPEN we can only 1158 * stat the file. 1159 */ 1160 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1161 if (error) 1162 return (error); 1163 1164 return (0); 1165 } 1166 1167 /* 1168 * Exec handler registration 1169 */ 1170 int 1171 exec_register(const struct execsw *execsw_arg) 1172 { 1173 const struct execsw **es, **xs, **newexecsw; 1174 int count = 2; /* New slot and trailing NULL */ 1175 1176 if (execsw) 1177 for (es = execsw; *es; es++) 1178 count++; 1179 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1180 xs = newexecsw; 1181 if (execsw) 1182 for (es = execsw; *es; es++) 1183 *xs++ = *es; 1184 *xs++ = execsw_arg; 1185 *xs = NULL; 1186 if (execsw) 1187 kfree(execsw, M_TEMP); 1188 execsw = newexecsw; 1189 return 0; 1190 } 1191 1192 int 1193 exec_unregister(const struct execsw *execsw_arg) 1194 { 1195 const struct execsw **es, **xs, **newexecsw; 1196 int count = 1; 1197 1198 if (execsw == NULL) 1199 panic("unregister with no handlers left?"); 1200 1201 for (es = execsw; *es; es++) { 1202 if (*es == execsw_arg) 1203 break; 1204 } 1205 if (*es == NULL) 1206 return ENOENT; 1207 for (es = execsw; *es; es++) 1208 if (*es != execsw_arg) 1209 count++; 1210 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1211 xs = newexecsw; 1212 for (es = execsw; *es; es++) 1213 if (*es != execsw_arg) 1214 *xs++ = *es; 1215 *xs = NULL; 1216 if (execsw) 1217 kfree(execsw, M_TEMP); 1218 execsw = newexecsw; 1219 return 0; 1220 } 1221