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/reg.h> 71 72 #include <sys/objcache.h> 73 #include <sys/refcount.h> 74 #include <sys/thread2.h> 75 #include <vm/vm_page2.h> 76 77 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 78 MALLOC_DEFINE(M_EXECARGS, "exec-args", "Exec arguments"); 79 80 static register_t *exec_copyout_strings (struct image_params *); 81 82 /* XXX This should be vm_size_t. */ 83 __read_mostly static u_long ps_strings = PS_STRINGS; 84 SYSCTL_ULONG(_kern, KERN_PS_STRINGS, ps_strings, CTLFLAG_RD, &ps_strings, 0, ""); 85 86 /* XXX This should be vm_size_t. */ 87 __read_mostly static u_long usrstack = USRSTACK; 88 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 89 90 __read_mostly u_long ps_arg_cache_limit = PAGE_SIZE / 16; 91 SYSCTL_LONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 92 &ps_arg_cache_limit, 0, ""); 93 94 __read_mostly int ps_argsopen = 1; 95 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 96 97 __read_mostly static int ktrace_suid = 0; 98 SYSCTL_INT(_kern, OID_AUTO, ktrace_suid, CTLFLAG_RW, &ktrace_suid, 0, ""); 99 100 void print_execve_args(struct image_args *args); 101 __read_mostly int debug_execve_args = 0; 102 SYSCTL_INT(_kern, OID_AUTO, debug_execve_args, CTLFLAG_RW, &debug_execve_args, 103 0, ""); 104 105 /* 106 * Exec arguments object cache 107 */ 108 __read_mostly static struct objcache *exec_objcache; 109 110 static 111 void 112 exec_objcache_init(void *arg __unused) 113 { 114 int cluster_limit; 115 size_t limsize; 116 117 /* 118 * Maximum number of concurrent execs. This can be limiting on 119 * systems with a lot of cpu cores but it also eats a significant 120 * amount of memory. 121 */ 122 cluster_limit = (ncpus < 16) ? 16 : ncpus; 123 limsize = kmem_lim_size(); 124 if (limsize > 7 * 1024) 125 cluster_limit *= 2; 126 if (limsize > 15 * 1024) 127 cluster_limit *= 2; 128 129 exec_objcache = objcache_create_mbacked( 130 M_EXECARGS, PATH_MAX + ARG_MAX, 131 cluster_limit, 8, 132 NULL, NULL, NULL); 133 } 134 SYSINIT(exec_objcache, SI_BOOT2_MACHDEP, SI_ORDER_ANY, exec_objcache_init, 0); 135 136 /* 137 * stackgap_random specifies if the stackgap should have a random size added 138 * to it. It must be a power of 2. If non-zero, the stack gap will be 139 * calculated as: ALIGN(karc4random() & (stackgap_random - 1)). 140 */ 141 __read_mostly static int stackgap_random = 1024; 142 143 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 __read_mostly 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 nd->nl_flags |= NLC_SHAREDLOCK; 250 if ((error = nlookup(nd)) != 0) 251 goto exec_fail; 252 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &imgp->vp); 253 KKASSERT(nd->nl_flags & NLC_NCPISLOCKED); 254 nd->nl_flags &= ~NLC_NCPISLOCKED; 255 cache_unlock(&nd->nl_nch); 256 if (error) 257 goto exec_fail; 258 259 /* 260 * Check file permissions (also 'opens' file). 261 * Include also the top level mount in the check. 262 */ 263 error = exec_check_permissions(imgp, nd->nl_nch.mount); 264 if (error) { 265 vn_unlock(imgp->vp); 266 goto exec_fail_dealloc; 267 } 268 269 error = exec_map_first_page(imgp); 270 vn_unlock(imgp->vp); 271 if (error) 272 goto exec_fail_dealloc; 273 274 imgp->proc->p_osrel = 0; 275 276 if (debug_execve_args && imgp->interpreted) { 277 kprintf(" target is interpreted -- recursive pass\n"); 278 kprintf(" interpreter: %s\n", imgp->interpreter_name); 279 print_execve_args(args); 280 } 281 282 /* 283 * If the current process has a special image activator it 284 * wants to try first, call it. For example, emulating shell 285 * scripts differently. 286 */ 287 error = -1; 288 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 289 error = img_first(imgp); 290 291 /* 292 * If the vnode has a registered vmspace, exec the vmspace 293 */ 294 if (error == -1 && imgp->vp->v_resident) { 295 error = exec_resident_imgact(imgp); 296 } 297 298 /* 299 * Loop through the list of image activators, calling each one. 300 * An activator returns -1 if there is no match, 0 on success, 301 * and an error otherwise. 302 */ 303 for (i = 0; error == -1 && execsw[i]; ++i) { 304 if (execsw[i]->ex_imgact == NULL || 305 execsw[i]->ex_imgact == img_first) { 306 continue; 307 } 308 error = (*execsw[i]->ex_imgact)(imgp); 309 } 310 311 if (error) { 312 if (error == -1) 313 error = ENOEXEC; 314 goto exec_fail_dealloc; 315 } 316 317 /* 318 * Special interpreter operation, cleanup and loop up to try to 319 * activate the interpreter. 320 */ 321 if (imgp->interpreted) { 322 exec_unmap_first_page(imgp); 323 nlookup_done(nd); 324 vrele(imgp->vp); 325 imgp->vp = NULL; 326 error = nlookup_init(nd, imgp->interpreter_name, UIO_SYSSPACE, 327 NLC_FOLLOW); 328 if (error) 329 goto exec_fail; 330 goto interpret; 331 } 332 333 /* 334 * Do the best to calculate the full path to the image file 335 */ 336 if (imgp->auxargs != NULL && 337 ((args->fname != NULL && args->fname[0] == '/') || 338 vn_fullpath(imgp->proc, 339 imgp->vp, 340 &imgp->execpath, 341 &imgp->freepath, 342 0) != 0)) 343 imgp->execpath = args->fname; 344 345 /* 346 * Copy out strings (args and env) and initialize stack base 347 */ 348 stack_base = exec_copyout_strings(imgp); 349 p->p_vmspace->vm_minsaddr = (char *)stack_base; 350 351 /* 352 * If custom stack fixup routine present for this process 353 * let it do the stack setup. If we are running a resident 354 * image there is no auxinfo or other image activator context 355 * so don't try to add fixups to the stack. 356 * 357 * Else stuff argument count as first item on stack 358 */ 359 if (p->p_sysent->sv_fixup && imgp->resident == 0) 360 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 361 else 362 suword64(--stack_base, imgp->args->argc); 363 364 /* 365 * For security and other reasons, the file descriptor table cannot 366 * be shared after an exec. 367 */ 368 if (p->p_fd->fd_refcnt > 1) { 369 struct filedesc *tmp; 370 371 error = fdcopy(p, &tmp); 372 if (error != 0) 373 goto exec_fail; 374 fdfree(p, tmp); 375 } 376 377 /* 378 * For security and other reasons, signal handlers cannot 379 * be shared after an exec. The new proces gets a copy of the old 380 * handlers. In execsigs(), the new process will have its signals 381 * reset. 382 */ 383 ops = p->p_sigacts; 384 if (ops->ps_refcnt > 1) { 385 nps = kmalloc(sizeof(*nps), M_SUBPROC, M_WAITOK); 386 bcopy(ops, nps, sizeof(*nps)); 387 refcount_init(&nps->ps_refcnt, 1); 388 p->p_sigacts = nps; 389 if (refcount_release(&ops->ps_refcnt)) { 390 kfree(ops, M_SUBPROC); 391 ops = NULL; 392 } 393 } 394 395 /* 396 * Clean up shared pages, the new program will allocate fresh 397 * copies as needed. This is also for security purposes and 398 * to ensure (for example) that things like sys_lpmap->blockallsigs 399 * state is properly reset on exec. 400 */ 401 lwp_userunmap(lp); 402 proc_userunmap(p); 403 404 /* 405 * For security and other reasons virtual kernels cannot be 406 * inherited by an exec. This also allows a virtual kernel 407 * to fork/exec unrelated applications. 408 */ 409 if (p->p_vkernel) 410 vkernel_exit(p); 411 412 /* Stop profiling */ 413 stopprofclock(p); 414 415 /* close files on exec */ 416 fdcloseexec(p); 417 418 /* reset caught signals */ 419 execsigs(p); 420 421 /* name this process - nameiexec(p, ndp) */ 422 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 423 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 424 p->p_comm[len] = 0; 425 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 426 427 /* 428 * mark as execed, wakeup the process that vforked (if any) and tell 429 * it that it now has its own resources back 430 * 431 * We are using the P_PPWAIT as an interlock so an atomic op is 432 * necessary to synchronize with the parent's cpu. 433 */ 434 p->p_flags |= P_EXEC; 435 if (p->p_pptr && (p->p_flags & P_PPWAIT)) { 436 if (p->p_pptr->p_upmap) 437 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 438 atomic_clear_int(&p->p_flags, P_PPWAIT); 439 wakeup(p->p_pptr); 440 } 441 442 /* 443 * Implement image setuid/setgid. 444 * 445 * Don't honor setuid/setgid if the filesystem prohibits it or if 446 * the process is being traced. 447 */ 448 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 449 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 450 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 451 (p->p_flags & P_TRACED) == 0) { 452 /* 453 * Turn off syscall tracing for set-id programs, except for 454 * root. Record any set-id flags first to make sure that 455 * we do not regain any tracing during a possible block. 456 */ 457 setsugid(); 458 if (p->p_tracenode && ktrace_suid == 0 && 459 priv_check(td, PRIV_ROOT) != 0) { 460 ktrdestroy(&p->p_tracenode); 461 p->p_traceflag = 0; 462 } 463 /* Close any file descriptors 0..2 that reference procfs */ 464 setugidsafety(p); 465 /* Make sure file descriptors 0..2 are in use. */ 466 error = fdcheckstd(lp); 467 if (error != 0) 468 goto exec_fail_dealloc; 469 /* 470 * Set the new credentials. 471 */ 472 cratom_proc(p); 473 if (attr.va_mode & VSUID) 474 change_euid(attr.va_uid); 475 if (attr.va_mode & VSGID) 476 p->p_ucred->cr_gid = attr.va_gid; 477 478 /* 479 * Clear local varsym variables 480 */ 481 varsymset_clean(&p->p_varsymset); 482 } else { 483 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 484 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 485 p->p_flags &= ~P_SUGID; 486 } 487 488 /* 489 * Implement correct POSIX saved-id behavior. 490 */ 491 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 492 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 493 cratom_proc(p); 494 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 495 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 496 } 497 498 /* 499 * Store the vp for use in procfs. Be sure to keep p_textvp 500 * consistent if we block during the switch-over. 501 */ 502 ovp = p->p_textvp; 503 vref(imgp->vp); /* ref new vp */ 504 p->p_textvp = imgp->vp; 505 if (ovp) /* release old vp */ 506 vrele(ovp); 507 508 /* Release old namecache handle to text file */ 509 if (p->p_textnch.ncp) 510 cache_drop(&p->p_textnch); 511 512 if (nd->nl_nch.mount) 513 cache_copy(&nd->nl_nch, &p->p_textnch); 514 515 /* 516 * Notify others that we exec'd, and clear the P_INEXEC flag 517 * as we're now a bona fide freshly-execed process. 518 */ 519 KNOTE(&p->p_klist, NOTE_EXEC); 520 p->p_flags &= ~P_INEXEC; 521 if (p->p_stops) 522 wakeup(&p->p_stype); 523 524 /* 525 * If tracing the process, trap to debugger so breakpoints 526 * can be set before the program executes. 527 */ 528 STOPEVENT(p, S_EXEC, 0); 529 530 if (p->p_flags & P_TRACED) 531 ksignal(p, SIGTRAP); 532 533 /* clear "fork but no exec" flag, as we _are_ execing */ 534 p->p_acflag &= ~AFORK; 535 536 /* Set values passed into the program in registers. */ 537 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 538 imgp->ps_strings); 539 540 /* Set the access time on the vnode */ 541 vn_mark_atime(imgp->vp, td); 542 543 /* 544 * Free any previous argument cache 545 */ 546 pa = p->p_args; 547 p->p_args = NULL; 548 if (pa && refcount_release(&pa->ar_ref)) { 549 kfree(pa, M_PARGS); 550 pa = NULL; 551 } 552 553 /* 554 * Cache arguments if they fit inside our allowance 555 */ 556 i = imgp->args->begin_envv - imgp->args->begin_argv; 557 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 558 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 559 refcount_init(&pa->ar_ref, 1); 560 pa->ar_length = i; 561 bcopy(imgp->args->begin_argv, pa->ar_args, i); 562 KKASSERT(p->p_args == NULL); 563 p->p_args = pa; 564 } 565 566 exec_fail_dealloc: 567 568 /* 569 * free various allocated resources 570 */ 571 if (imgp->firstpage) 572 exec_unmap_first_page(imgp); 573 574 if (imgp->vp) { 575 vrele(imgp->vp); 576 imgp->vp = NULL; 577 } 578 579 if (imgp->freepath) 580 kfree(imgp->freepath, M_TEMP); 581 582 if (error == 0) { 583 ++mycpu->gd_cnt.v_exec; 584 lwkt_reltoken(&p->p_token); 585 return (0); 586 } 587 588 exec_fail: 589 /* 590 * we're done here, clear P_INEXEC if we were the ones that 591 * set it. Otherwise if vmspace_destroyed is still set we 592 * raced another thread and that thread is responsible for 593 * clearing it. 594 */ 595 if (imgp->vmspace_destroyed & 2) { 596 p->p_flags &= ~P_INEXEC; 597 if (p->p_stops) 598 wakeup(&p->p_stype); 599 } 600 lwkt_reltoken(&p->p_token); 601 if (imgp->vmspace_destroyed) { 602 /* 603 * Sorry, no more process anymore. exit gracefully. 604 * However we can't die right here, because our 605 * caller might have to clean up, so indicate a 606 * lethal error by returning -1. 607 */ 608 return(-1); 609 } else { 610 return(error); 611 } 612 } 613 614 /* 615 * execve() system call. 616 */ 617 int 618 sys_execve(struct execve_args *uap) 619 { 620 struct nlookupdata nd; 621 struct image_args args; 622 int error; 623 624 bzero(&args, sizeof(args)); 625 626 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 627 if (error == 0) { 628 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 629 uap->argv, uap->envv); 630 } 631 if (error == 0) 632 error = kern_execve(&nd, &args); 633 nlookup_done(&nd); 634 exec_free_args(&args); 635 636 if (error < 0) { 637 /* We hit a lethal error condition. Let's die now. */ 638 exit1(W_EXITCODE(0, SIGABRT)); 639 /* NOTREACHED */ 640 } 641 642 /* 643 * The syscall result is returned in registers to the new program. 644 * Linux will register %edx as an atexit function and we must be 645 * sure to set it to 0. XXX 646 */ 647 if (error == 0) 648 uap->sysmsg_result64 = 0; 649 650 return (error); 651 } 652 653 int 654 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 655 struct lwbuf **plwb, const char **pdata) 656 { 657 int rv; 658 vm_page_t ma; 659 vm_page_t m; 660 vm_object_t object; 661 662 /* 663 * The file has to be mappable. 664 */ 665 if ((object = imgp->vp->v_object) == NULL) 666 return (EIO); 667 668 if (pageno >= object->size) 669 return (EIO); 670 671 /* 672 * Shortcut using shared locks, improve concurrent execs. 673 */ 674 vm_object_hold_shared(object); 675 m = vm_page_lookup(object, pageno); 676 if (m) { 677 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 678 vm_page_hold(m); 679 vm_page_sleep_busy(m, FALSE, "execpg"); 680 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 681 m->object == object && m->pindex == pageno) { 682 vm_object_drop(object); 683 goto done; 684 } 685 vm_page_unhold(m); 686 } 687 } 688 vm_object_drop(object); 689 690 /* 691 * Do it the hard way 692 */ 693 vm_object_hold(object); 694 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 695 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 696 ma = m; 697 698 /* 699 * get_pages unbusies all the requested pages except the 700 * primary page (at index 0 in this case). The primary 701 * page may have been wired during the pagein (e.g. by 702 * the buffer cache) so vnode_pager_freepage() must be 703 * used to properly release it. 704 */ 705 rv = vm_pager_get_page(object, &ma, 1); 706 m = vm_page_lookup(object, pageno); 707 708 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 709 if (m) { 710 vm_page_protect(m, VM_PROT_NONE); 711 vnode_pager_freepage(m); 712 } 713 vm_object_drop(object); 714 return EIO; 715 } 716 } 717 vm_page_hold(m); 718 vm_page_wakeup(m); /* unbusy the page */ 719 vm_object_drop(object); 720 721 done: 722 *plwb = lwbuf_alloc(m, *plwb); 723 *pdata = (void *)lwbuf_kva(*plwb); 724 725 return (0); 726 } 727 728 /* 729 * Map the first page of an executable image. 730 * 731 * NOTE: If the mapping fails we have to NULL-out firstpage which may 732 * still be pointing to our supplied lwp structure. 733 */ 734 int 735 exec_map_first_page(struct image_params *imgp) 736 { 737 int err; 738 739 if (imgp->firstpage) 740 exec_unmap_first_page(imgp); 741 742 imgp->firstpage = &imgp->firstpage_cache; 743 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 744 745 if (err) { 746 imgp->firstpage = NULL; 747 return err; 748 } 749 750 return 0; 751 } 752 753 void 754 exec_unmap_page(struct lwbuf *lwb) 755 { 756 vm_page_t m; 757 758 crit_enter(); 759 if (lwb != NULL) { 760 m = lwbuf_page(lwb); 761 lwbuf_free(lwb); 762 vm_page_unhold(m); 763 } 764 crit_exit(); 765 } 766 767 void 768 exec_unmap_first_page(struct image_params *imgp) 769 { 770 exec_unmap_page(imgp->firstpage); 771 imgp->firstpage = NULL; 772 imgp->image_header = NULL; 773 } 774 775 /* 776 * Destroy old address space, and allocate a new stack 777 * The new stack is only SGROWSIZ large because it is grown 778 * automatically in trap.c. 779 * 780 * This is the point of no return. 781 */ 782 int 783 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 784 { 785 struct vmspace *vmspace = imgp->proc->p_vmspace; 786 vm_offset_t stack_addr = USRSTACK - maxssiz; 787 struct lwp *lp; 788 struct proc *p; 789 vm_map_t map; 790 int error; 791 792 /* 793 * Indicate that we cannot gracefully error out any more, kill 794 * any other threads present, and set P_INEXEC to indicate that 795 * we are now messing with the process structure proper. 796 * 797 * If killalllwps() races return an error which coupled with 798 * vmspace_destroyed will cause us to exit. This is what we 799 * want since another thread is patiently waiting for us to exit 800 * in that case. 801 */ 802 lp = curthread->td_lwp; 803 p = lp->lwp_proc; 804 imgp->vmspace_destroyed = 1; 805 806 if (curthread->td_proc->p_nthreads > 1) { 807 error = killalllwps(1); 808 if (error) 809 return (error); 810 } 811 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 812 p->p_flags |= P_INEXEC; 813 814 /* 815 * Tell procfs to release its hold on the process. It 816 * will return EAGAIN. 817 */ 818 if (p->p_stops) 819 wakeup(&p->p_stype); 820 821 /* 822 * After setting P_INEXEC wait for any remaining references to 823 * the process (p) to go away. 824 * 825 * In particular, a vfork/exec sequence will replace p->p_vmspace 826 * and we must interlock anyone trying to access the space (aka 827 * procfs or sys_process.c calling procfs_domem()). 828 * 829 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us. 830 */ 831 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0)); 832 833 /* 834 * Blow away entire process VM, if address space not shared, 835 * otherwise, create a new VM space so that other threads are 836 * not disrupted. If we are execing a resident vmspace we 837 * create a duplicate of it and remap the stack. 838 */ 839 map = &vmspace->vm_map; 840 if (vmcopy) { 841 vmspace_exec(imgp->proc, vmcopy); 842 vmspace = imgp->proc->p_vmspace; 843 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 844 map = &vmspace->vm_map; 845 } else if (vmspace_getrefs(vmspace) == 1) { 846 shmexit(vmspace); 847 pmap_remove_pages(vmspace_pmap(vmspace), 848 0, VM_MAX_USER_ADDRESS); 849 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 850 } else { 851 vmspace_exec(imgp->proc, NULL); 852 vmspace = imgp->proc->p_vmspace; 853 map = &vmspace->vm_map; 854 } 855 856 /* 857 * Really make sure lwp-specific and process-specific mappings 858 * are gone. 859 * 860 * Once we've done that, and because we are the only LWP left, with 861 * no TID-dependent mappings, we can reset the TID to 1 (the RB tree 862 * will remain consistent since it has only one entry). This way 863 * the exec'd program gets a nice deterministic tid of 1. 864 */ 865 lwp_userunmap(lp); 866 proc_userunmap(p); 867 lp->lwp_tid = 1; 868 p->p_lasttid = 1; 869 870 /* 871 * Allocate a new stack, generally make the stack non-executable 872 * but allow the program to adjust that (the program may desire to 873 * use areas of the stack for executable code). 874 */ 875 error = vm_map_stack(&vmspace->vm_map, &stack_addr, (vm_size_t)maxssiz, 876 0, 877 VM_PROT_READ|VM_PROT_WRITE, 878 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, 879 0); 880 if (error) 881 return (error); 882 883 /* 884 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 885 * VM_STACK case, but they are still used to monitor the size of the 886 * process stack so we can check the stack rlimit. 887 */ 888 vmspace->vm_ssize = sgrowsiz; /* in bytes */ 889 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 890 891 return(0); 892 } 893 894 /* 895 * Copy out argument and environment strings from the old process 896 * address space into the temporary string buffer. 897 */ 898 int 899 exec_copyin_args(struct image_args *args, char *fname, 900 enum exec_path_segflg segflg, char **argv, char **envv) 901 { 902 char *argp, *envp; 903 int error = 0; 904 size_t length; 905 906 args->buf = objcache_get(exec_objcache, M_WAITOK); 907 if (args->buf == NULL) 908 return (ENOMEM); 909 args->begin_argv = args->buf; 910 args->endp = args->begin_argv; 911 args->space = ARG_MAX; 912 913 args->fname = args->buf + ARG_MAX; 914 915 /* 916 * Copy the file name. 917 */ 918 if (segflg == PATH_SYSSPACE) { 919 error = copystr(fname, args->fname, PATH_MAX, &length); 920 } else if (segflg == PATH_USERSPACE) { 921 error = copyinstr(fname, args->fname, PATH_MAX, &length); 922 } 923 924 /* 925 * Extract argument strings. argv may not be NULL. The argv 926 * array is terminated by a NULL entry. We special-case the 927 * situation where argv[0] is NULL by passing { filename, NULL } 928 * to the new program to guarentee that the interpreter knows what 929 * file to open in case we exec an interpreted file. Note that 930 * a NULL argv[0] terminates the argv[] array. 931 * 932 * XXX the special-casing of argv[0] is historical and needs to be 933 * revisited. 934 */ 935 if (argv == NULL) 936 error = EFAULT; 937 if (error == 0) { 938 while ((argp = (caddr_t)(intptr_t) 939 fuword64((uintptr_t *)argv++)) != NULL) { 940 if (argp == (caddr_t)-1) { 941 error = EFAULT; 942 break; 943 } 944 error = copyinstr(argp, args->endp, 945 args->space, &length); 946 if (error) { 947 if (error == ENAMETOOLONG) 948 error = E2BIG; 949 break; 950 } 951 args->space -= length; 952 args->endp += length; 953 args->argc++; 954 } 955 if (args->argc == 0 && error == 0) { 956 length = strlen(args->fname) + 1; 957 if (length > args->space) { 958 error = E2BIG; 959 } else { 960 bcopy(args->fname, args->endp, length); 961 args->space -= length; 962 args->endp += length; 963 args->argc++; 964 } 965 } 966 } 967 968 args->begin_envv = args->endp; 969 970 /* 971 * extract environment strings. envv may be NULL. 972 */ 973 if (envv && error == 0) { 974 while ((envp = (caddr_t)(intptr_t) 975 fuword64((uintptr_t *)envv++))) { 976 if (envp == (caddr_t) -1) { 977 error = EFAULT; 978 break; 979 } 980 error = copyinstr(envp, args->endp, 981 args->space, &length); 982 if (error) { 983 if (error == ENAMETOOLONG) 984 error = E2BIG; 985 break; 986 } 987 args->space -= length; 988 args->endp += length; 989 args->envc++; 990 } 991 } 992 return (error); 993 } 994 995 void 996 exec_free_args(struct image_args *args) 997 { 998 if (args->buf) { 999 objcache_put(exec_objcache, args->buf); 1000 args->buf = NULL; 1001 } 1002 } 1003 1004 /* 1005 * Copy strings out to the new process address space, constructing 1006 * new arg and env vector tables. Return a pointer to the base 1007 * so that it can be used as the initial stack pointer. 1008 * 1009 * The format is, roughly: 1010 * 1011 * [argv[]] <-- vectp 1012 * [envp[]] 1013 * [ELF_Auxargs] 1014 * 1015 * [args & env] <-- destp 1016 * [sgap] 1017 * [SPARE_USRSPACE] 1018 * [execpath] 1019 * [szsigcode] RO|NX 1020 * [ps_strings] RO|NX Top of user stack 1021 * 1022 */ 1023 static register_t * 1024 exec_copyout_strings(struct image_params *imgp) 1025 { 1026 int argc, envc, sgap; 1027 int gap; 1028 int argsenvspace; 1029 char **vectp; 1030 char *stringp, *destp, *szsigbase; 1031 register_t *stack_base; 1032 struct ps_strings *arginfo; 1033 size_t execpath_len; 1034 int szsigcode; 1035 1036 /* 1037 * Calculate string base and vector table pointers. 1038 * Also deal with signal trampoline code for this exec type. 1039 */ 1040 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1041 execpath_len = strlen(imgp->execpath) + 1; 1042 else 1043 execpath_len = 0; 1044 arginfo = (struct ps_strings *)PS_STRINGS; 1045 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 1046 1047 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 1048 gap = stackgap_random; 1049 cpu_ccfence(); 1050 if (gap != 0) { 1051 if (gap < 0) 1052 sgap = ALIGN(-gap); 1053 else 1054 sgap = ALIGN(karc4random() & (gap - 1)); 1055 } else { 1056 sgap = 0; 1057 } 1058 1059 /* 1060 * Calculate destp, which points to [args & env] and above. 1061 */ 1062 szsigbase = (char *)(intptr_t) 1063 trunc_page64((intptr_t)arginfo - szsigcode); 1064 szsigbase -= SZSIGCODE_EXTRA_BYTES; 1065 destp = szsigbase - 1066 roundup(execpath_len, sizeof(char *)) - 1067 SPARE_USRSPACE - 1068 sgap - 1069 argsenvspace; 1070 1071 /* 1072 * install sigcode 1073 */ 1074 if (szsigcode) 1075 copyout(imgp->proc->p_sysent->sv_sigcode, szsigbase, szsigcode); 1076 1077 /* 1078 * Copy the image path for the rtld 1079 */ 1080 if (execpath_len) { 1081 imgp->execpathp = (uintptr_t)szsigbase - 1082 roundup(execpath_len, sizeof(char *)); 1083 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 1084 } 1085 1086 /* 1087 * Calculate base for argv[], envp[], and ELF_Auxargs. 1088 */ 1089 vectp = (char **)destp - (AT_COUNT * 2); 1090 vectp -= imgp->args->argc + imgp->args->envc + 2; 1091 1092 stack_base = (register_t *)vectp; 1093 1094 stringp = imgp->args->begin_argv; 1095 argc = imgp->args->argc; 1096 envc = imgp->args->envc; 1097 1098 /* 1099 * Copy out strings - arguments and environment (at destp) 1100 */ 1101 copyout(stringp, destp, ARG_MAX - imgp->args->space); 1102 1103 /* 1104 * Fill in "ps_strings" struct for ps, w, etc. 1105 */ 1106 suword64((void *)&arginfo->ps_argvstr, (uint64_t)(intptr_t)vectp); 1107 suword32((void *)&arginfo->ps_nargvstr, argc); 1108 1109 /* 1110 * Fill in argument portion of vector table. 1111 */ 1112 for (; argc > 0; --argc) { 1113 suword64((void *)vectp++, (uintptr_t)destp); 1114 while (*stringp++ != 0) 1115 destp++; 1116 destp++; 1117 } 1118 1119 /* a null vector table pointer separates the argp's from the envp's */ 1120 suword64((void *)vectp++, 0); 1121 1122 suword64((void *)&arginfo->ps_envstr, (uintptr_t)vectp); 1123 suword32((void *)&arginfo->ps_nenvstr, envc); 1124 1125 /* 1126 * Fill in environment portion of vector table. 1127 */ 1128 for (; envc > 0; --envc) { 1129 suword64((void *)vectp++, (uintptr_t)destp); 1130 while (*stringp++ != 0) 1131 destp++; 1132 destp++; 1133 } 1134 1135 /* end of vector table is a null pointer */ 1136 suword64((void *)vectp, 0); 1137 1138 /* 1139 * Make the signal trampoline executable and read-only. 1140 */ 1141 vm_map_protect(&imgp->proc->p_vmspace->vm_map, 1142 (vm_offset_t)szsigbase, 1143 (vm_offset_t)szsigbase + PAGE_SIZE, 1144 VM_PROT_READ|VM_PROT_EXECUTE, FALSE); 1145 1146 return (stack_base); 1147 } 1148 1149 /* 1150 * Check permissions of file to execute. 1151 * Return 0 for success or error code on failure. 1152 */ 1153 int 1154 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 1155 { 1156 struct proc *p = imgp->proc; 1157 struct vnode *vp = imgp->vp; 1158 struct vattr *attr = imgp->attr; 1159 int error; 1160 1161 /* Get file attributes */ 1162 error = VOP_GETATTR(vp, attr); 1163 if (error) 1164 return (error); 1165 1166 /* 1167 * 1) Check if file execution is disabled for the filesystem that this 1168 * file resides on. 1169 * 2) Insure that at least one execute bit is on - otherwise root 1170 * will always succeed, and we don't want to happen unless the 1171 * file really is executable. 1172 * 3) Insure that the file is a regular file. 1173 */ 1174 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1175 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1176 ((attr->va_mode & 0111) == 0) || 1177 (attr->va_type != VREG)) { 1178 return (EACCES); 1179 } 1180 1181 /* 1182 * Zero length files can't be exec'd 1183 */ 1184 if (attr->va_size == 0) 1185 return (ENOEXEC); 1186 1187 /* 1188 * Check for execute permission to file based on current credentials. 1189 */ 1190 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1191 if (error) 1192 return (error); 1193 1194 /* 1195 * Check number of open-for-writes on the file and deny execution 1196 * if there are any. 1197 */ 1198 if (vp->v_writecount) 1199 return (ETXTBSY); 1200 1201 /* 1202 * Call filesystem specific open routine, which allows us to read, 1203 * write, and mmap the file. Without the VOP_OPEN we can only 1204 * stat the file. 1205 */ 1206 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1207 if (error) 1208 return (error); 1209 1210 return (0); 1211 } 1212 1213 /* 1214 * Exec handler registration 1215 */ 1216 int 1217 exec_register(const struct execsw *execsw_arg) 1218 { 1219 const struct execsw **es, **xs, **newexecsw; 1220 int count = 2; /* New slot and trailing NULL */ 1221 1222 if (execsw) 1223 for (es = execsw; *es; es++) 1224 count++; 1225 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1226 xs = newexecsw; 1227 if (execsw) 1228 for (es = execsw; *es; es++) 1229 *xs++ = *es; 1230 *xs++ = execsw_arg; 1231 *xs = NULL; 1232 if (execsw) 1233 kfree(execsw, M_TEMP); 1234 execsw = newexecsw; 1235 return 0; 1236 } 1237 1238 int 1239 exec_unregister(const struct execsw *execsw_arg) 1240 { 1241 const struct execsw **es, **xs, **newexecsw; 1242 int count = 1; 1243 1244 if (execsw == NULL) 1245 panic("unregister with no handlers left?"); 1246 1247 for (es = execsw; *es; es++) { 1248 if (*es == execsw_arg) 1249 break; 1250 } 1251 if (*es == NULL) 1252 return ENOENT; 1253 for (es = execsw; *es; es++) 1254 if (*es != execsw_arg) 1255 count++; 1256 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1257 xs = newexecsw; 1258 for (es = execsw; *es; es++) 1259 if (*es != execsw_arg) 1260 *xs++ = *es; 1261 *xs = NULL; 1262 if (execsw) 1263 kfree(execsw, M_TEMP); 1264 execsw = newexecsw; 1265 return 0; 1266 } 1267