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 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 static u_long usrstack = USRSTACK; 88 SYSCTL_ULONG(_kern, KERN_USRSTACK, usrstack, CTLFLAG_RD, &usrstack, 0, ""); 89 90 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 int ps_argsopen = 1; 95 SYSCTL_INT(_kern, OID_AUTO, ps_argsopen, CTLFLAG_RW, &ps_argsopen, 0, ""); 96 97 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 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 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 static int stackgap_random = 1024; 142 static int 143 sysctl_kern_stackgap(SYSCTL_HANDLER_ARGS) 144 { 145 int error, new_val; 146 new_val = stackgap_random; 147 error = sysctl_handle_int(oidp, &new_val, 0, req); 148 if (error != 0 || req->newptr == NULL) 149 return (error); 150 if (new_val > 0 && ((new_val > 16 * PAGE_SIZE) || !powerof2(new_val))) 151 return (EINVAL); 152 stackgap_random = new_val; 153 154 return(0); 155 } 156 157 SYSCTL_PROC(_kern, OID_AUTO, stackgap_random, CTLFLAG_RW|CTLTYPE_INT, 158 0, 0, sysctl_kern_stackgap, "I", 159 "Max random stack gap (power of 2), static gap if negative"); 160 161 void 162 print_execve_args(struct image_args *args) 163 { 164 char *cp; 165 int ndx; 166 167 cp = args->begin_argv; 168 for (ndx = 0; ndx < args->argc; ndx++) { 169 kprintf("\targv[%d]: %s\n", ndx, cp); 170 while (*cp++ != '\0'); 171 } 172 for (ndx = 0; ndx < args->envc; ndx++) { 173 kprintf("\tenvv[%d]: %s\n", ndx, cp); 174 while (*cp++ != '\0'); 175 } 176 } 177 178 /* 179 * Each of the items is a pointer to a `const struct execsw', hence the 180 * double pointer here. 181 */ 182 static const struct execsw **execsw; 183 184 /* 185 * Replace current vmspace with a new binary. 186 * Returns 0 on success, > 0 on recoverable error (use as errno). 187 * Returns -1 on lethal error which demands killing of the current 188 * process! 189 */ 190 int 191 kern_execve(struct nlookupdata *nd, struct image_args *args) 192 { 193 struct thread *td = curthread; 194 struct lwp *lp = td->td_lwp; 195 struct proc *p = td->td_proc; 196 struct vnode *ovp; 197 register_t *stack_base; 198 struct pargs *pa; 199 struct sigacts *ops; 200 struct sigacts *nps; 201 int error, len, i; 202 struct image_params image_params, *imgp; 203 struct vattr attr; 204 int (*img_first) (struct image_params *); 205 206 if (debug_execve_args) { 207 kprintf("%s()\n", __func__); 208 print_execve_args(args); 209 } 210 211 KKASSERT(p); 212 lwkt_gettoken(&p->p_token); 213 imgp = &image_params; 214 215 /* 216 * NOTE: P_INEXEC is handled by exec_new_vmspace() now. We make 217 * no modifications to the process at all until we get there. 218 * 219 * Note that multiple threads may be trying to exec at the same 220 * time. exec_new_vmspace() handles that too. 221 */ 222 223 /* 224 * Initialize part of the common data 225 */ 226 imgp->proc = p; 227 imgp->args = args; 228 imgp->attr = &attr; 229 imgp->entry_addr = 0; 230 imgp->resident = 0; 231 imgp->vmspace_destroyed = 0; 232 imgp->interpreted = 0; 233 imgp->interpreter_name[0] = 0; 234 imgp->auxargs = NULL; 235 imgp->vp = NULL; 236 imgp->firstpage = NULL; 237 imgp->ps_strings = 0; 238 imgp->execpath = imgp->freepath = NULL; 239 imgp->execpathp = 0; 240 imgp->image_header = NULL; 241 242 interpret: 243 244 /* 245 * Translate the file name to a vnode. Unlock the cache entry to 246 * improve parallelism for programs exec'd in parallel. 247 */ 248 nd->nl_flags |= NLC_SHAREDLOCK; 249 if ((error = nlookup(nd)) != 0) 250 goto exec_fail; 251 error = cache_vget(&nd->nl_nch, nd->nl_cred, LK_SHARED, &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 suword64(--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 * Clean up shared pages, the new program will allocate fresh 396 * copies as needed. This is also for security purposes and 397 * to ensure (for example) that things like sys_lpmap->blockallsigs 398 * state is properly reset on exec. 399 */ 400 lwp_userunmap(lp); 401 proc_userunmap(p); 402 403 /* 404 * For security and other reasons virtual kernels cannot be 405 * inherited by an exec. This also allows a virtual kernel 406 * to fork/exec unrelated applications. 407 */ 408 if (p->p_vkernel) 409 vkernel_exit(p); 410 411 /* Stop profiling */ 412 stopprofclock(p); 413 414 /* close files on exec */ 415 fdcloseexec(p); 416 417 /* reset caught signals */ 418 execsigs(p); 419 420 /* name this process - nameiexec(p, ndp) */ 421 len = min(nd->nl_nch.ncp->nc_nlen, MAXCOMLEN); 422 bcopy(nd->nl_nch.ncp->nc_name, p->p_comm, len); 423 p->p_comm[len] = 0; 424 bcopy(p->p_comm, lp->lwp_thread->td_comm, MAXCOMLEN+1); 425 426 /* 427 * mark as execed, wakeup the process that vforked (if any) and tell 428 * it that it now has its own resources back 429 * 430 * We are using the P_PPWAIT as an interlock so an atomic op is 431 * necessary to synchronize with the parent's cpu. 432 */ 433 p->p_flags |= P_EXEC; 434 if (p->p_pptr && (p->p_flags & P_PPWAIT)) { 435 if (p->p_pptr->p_upmap) 436 atomic_add_int(&p->p_pptr->p_upmap->invfork, -1); 437 atomic_clear_int(&p->p_flags, P_PPWAIT); 438 wakeup(p->p_pptr); 439 } 440 441 /* 442 * Implement image setuid/setgid. 443 * 444 * Don't honor setuid/setgid if the filesystem prohibits it or if 445 * the process is being traced. 446 */ 447 if ((((attr.va_mode & VSUID) && p->p_ucred->cr_uid != attr.va_uid) || 448 ((attr.va_mode & VSGID) && p->p_ucred->cr_gid != attr.va_gid)) && 449 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 450 (p->p_flags & P_TRACED) == 0) { 451 /* 452 * Turn off syscall tracing for set-id programs, except for 453 * root. Record any set-id flags first to make sure that 454 * we do not regain any tracing during a possible block. 455 */ 456 setsugid(); 457 if (p->p_tracenode && ktrace_suid == 0 && 458 priv_check(td, PRIV_ROOT) != 0) { 459 ktrdestroy(&p->p_tracenode); 460 p->p_traceflag = 0; 461 } 462 /* Close any file descriptors 0..2 that reference procfs */ 463 setugidsafety(p); 464 /* Make sure file descriptors 0..2 are in use. */ 465 error = fdcheckstd(lp); 466 if (error != 0) 467 goto exec_fail_dealloc; 468 /* 469 * Set the new credentials. 470 */ 471 cratom_proc(p); 472 if (attr.va_mode & VSUID) 473 change_euid(attr.va_uid); 474 if (attr.va_mode & VSGID) 475 p->p_ucred->cr_gid = attr.va_gid; 476 477 /* 478 * Clear local varsym variables 479 */ 480 varsymset_clean(&p->p_varsymset); 481 } else { 482 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 483 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 484 p->p_flags &= ~P_SUGID; 485 } 486 487 /* 488 * Implement correct POSIX saved-id behavior. 489 */ 490 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 491 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 492 cratom_proc(p); 493 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 494 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 495 } 496 497 /* 498 * Store the vp for use in procfs. Be sure to keep p_textvp 499 * consistent if we block during the switch-over. 500 */ 501 ovp = p->p_textvp; 502 vref(imgp->vp); /* ref new vp */ 503 p->p_textvp = imgp->vp; 504 if (ovp) /* release old vp */ 505 vrele(ovp); 506 507 /* Release old namecache handle to text file */ 508 if (p->p_textnch.ncp) 509 cache_drop(&p->p_textnch); 510 511 if (nd->nl_nch.mount) 512 cache_copy(&nd->nl_nch, &p->p_textnch); 513 514 /* 515 * Notify others that we exec'd, and clear the P_INEXEC flag 516 * as we're now a bona fide freshly-execed process. 517 */ 518 KNOTE(&p->p_klist, NOTE_EXEC); 519 p->p_flags &= ~P_INEXEC; 520 if (p->p_stops) 521 wakeup(&p->p_stype); 522 523 /* 524 * If tracing the process, trap to debugger so breakpoints 525 * can be set before the program executes. 526 */ 527 STOPEVENT(p, S_EXEC, 0); 528 529 if (p->p_flags & P_TRACED) 530 ksignal(p, SIGTRAP); 531 532 /* clear "fork but no exec" flag, as we _are_ execing */ 533 p->p_acflag &= ~AFORK; 534 535 /* Set values passed into the program in registers. */ 536 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 537 imgp->ps_strings); 538 539 /* Set the access time on the vnode */ 540 vn_mark_atime(imgp->vp, td); 541 542 /* 543 * Free any previous argument cache 544 */ 545 pa = p->p_args; 546 p->p_args = NULL; 547 if (pa && refcount_release(&pa->ar_ref)) { 548 kfree(pa, M_PARGS); 549 pa = NULL; 550 } 551 552 /* 553 * Cache arguments if they fit inside our allowance 554 */ 555 i = imgp->args->begin_envv - imgp->args->begin_argv; 556 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 557 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 558 refcount_init(&pa->ar_ref, 1); 559 pa->ar_length = i; 560 bcopy(imgp->args->begin_argv, pa->ar_args, i); 561 KKASSERT(p->p_args == NULL); 562 p->p_args = pa; 563 } 564 565 exec_fail_dealloc: 566 567 /* 568 * free various allocated resources 569 */ 570 if (imgp->firstpage) 571 exec_unmap_first_page(imgp); 572 573 if (imgp->vp) { 574 vrele(imgp->vp); 575 imgp->vp = NULL; 576 } 577 578 if (imgp->freepath) 579 kfree(imgp->freepath, M_TEMP); 580 581 if (error == 0) { 582 ++mycpu->gd_cnt.v_exec; 583 lwkt_reltoken(&p->p_token); 584 return (0); 585 } 586 587 exec_fail: 588 /* 589 * we're done here, clear P_INEXEC if we were the ones that 590 * set it. Otherwise if vmspace_destroyed is still set we 591 * raced another thread and that thread is responsible for 592 * clearing it. 593 */ 594 if (imgp->vmspace_destroyed & 2) { 595 p->p_flags &= ~P_INEXEC; 596 if (p->p_stops) 597 wakeup(&p->p_stype); 598 } 599 lwkt_reltoken(&p->p_token); 600 if (imgp->vmspace_destroyed) { 601 /* 602 * Sorry, no more process anymore. exit gracefully. 603 * However we can't die right here, because our 604 * caller might have to clean up, so indicate a 605 * lethal error by returning -1. 606 */ 607 return(-1); 608 } else { 609 return(error); 610 } 611 } 612 613 /* 614 * execve() system call. 615 */ 616 int 617 sys_execve(struct execve_args *uap) 618 { 619 struct nlookupdata nd; 620 struct image_args args; 621 int error; 622 623 bzero(&args, sizeof(args)); 624 625 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 626 if (error == 0) { 627 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 628 uap->argv, uap->envv); 629 } 630 if (error == 0) 631 error = kern_execve(&nd, &args); 632 nlookup_done(&nd); 633 exec_free_args(&args); 634 635 if (error < 0) { 636 /* We hit a lethal error condition. Let's die now. */ 637 exit1(W_EXITCODE(0, SIGABRT)); 638 /* NOTREACHED */ 639 } 640 641 /* 642 * The syscall result is returned in registers to the new program. 643 * Linux will register %edx as an atexit function and we must be 644 * sure to set it to 0. XXX 645 */ 646 if (error == 0) 647 uap->sysmsg_result64 = 0; 648 649 return (error); 650 } 651 652 int 653 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 654 struct lwbuf **plwb, const char **pdata) 655 { 656 int rv; 657 vm_page_t ma; 658 vm_page_t m; 659 vm_object_t object; 660 661 /* 662 * The file has to be mappable. 663 */ 664 if ((object = imgp->vp->v_object) == NULL) 665 return (EIO); 666 667 if (pageno >= object->size) 668 return (EIO); 669 670 /* 671 * Shortcut using shared locks, improve concurrent execs. 672 */ 673 vm_object_hold_shared(object); 674 m = vm_page_lookup(object, pageno); 675 if (m) { 676 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 677 vm_page_hold(m); 678 vm_page_sleep_busy(m, FALSE, "execpg"); 679 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 680 m->object == object && m->pindex == pageno) { 681 vm_object_drop(object); 682 goto done; 683 } 684 vm_page_unhold(m); 685 } 686 } 687 vm_object_drop(object); 688 689 /* 690 * Do it the hard way 691 */ 692 vm_object_hold(object); 693 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 694 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 695 ma = m; 696 697 /* 698 * get_pages unbusies all the requested pages except the 699 * primary page (at index 0 in this case). The primary 700 * page may have been wired during the pagein (e.g. by 701 * the buffer cache) so vnode_pager_freepage() must be 702 * used to properly release it. 703 */ 704 rv = vm_pager_get_page(object, &ma, 1); 705 m = vm_page_lookup(object, pageno); 706 707 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 708 if (m) { 709 vm_page_protect(m, VM_PROT_NONE); 710 vnode_pager_freepage(m); 711 } 712 vm_object_drop(object); 713 return EIO; 714 } 715 } 716 vm_page_hold(m); 717 vm_page_wakeup(m); /* unbusy the page */ 718 vm_object_drop(object); 719 720 done: 721 *plwb = lwbuf_alloc(m, *plwb); 722 *pdata = (void *)lwbuf_kva(*plwb); 723 724 return (0); 725 } 726 727 /* 728 * Map the first page of an executable image. 729 * 730 * NOTE: If the mapping fails we have to NULL-out firstpage which may 731 * still be pointing to our supplied lwp structure. 732 */ 733 int 734 exec_map_first_page(struct image_params *imgp) 735 { 736 int err; 737 738 if (imgp->firstpage) 739 exec_unmap_first_page(imgp); 740 741 imgp->firstpage = &imgp->firstpage_cache; 742 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 743 744 if (err) { 745 imgp->firstpage = NULL; 746 return err; 747 } 748 749 return 0; 750 } 751 752 void 753 exec_unmap_page(struct lwbuf *lwb) 754 { 755 vm_page_t m; 756 757 crit_enter(); 758 if (lwb != NULL) { 759 m = lwbuf_page(lwb); 760 lwbuf_free(lwb); 761 vm_page_unhold(m); 762 } 763 crit_exit(); 764 } 765 766 void 767 exec_unmap_first_page(struct image_params *imgp) 768 { 769 exec_unmap_page(imgp->firstpage); 770 imgp->firstpage = NULL; 771 imgp->image_header = NULL; 772 } 773 774 /* 775 * Destroy old address space, and allocate a new stack 776 * The new stack is only SGROWSIZ large because it is grown 777 * automatically in trap.c. 778 * 779 * This is the point of no return. 780 */ 781 int 782 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 783 { 784 struct vmspace *vmspace = imgp->proc->p_vmspace; 785 vm_offset_t stack_addr = USRSTACK - maxssiz; 786 struct lwp *lp; 787 struct proc *p; 788 vm_map_t map; 789 int error; 790 791 /* 792 * Indicate that we cannot gracefully error out any more, kill 793 * any other threads present, and set P_INEXEC to indicate that 794 * we are now messing with the process structure proper. 795 * 796 * If killalllwps() races return an error which coupled with 797 * vmspace_destroyed will cause us to exit. This is what we 798 * want since another thread is patiently waiting for us to exit 799 * in that case. 800 */ 801 lp = curthread->td_lwp; 802 p = lp->lwp_proc; 803 imgp->vmspace_destroyed = 1; 804 805 if (curthread->td_proc->p_nthreads > 1) { 806 error = killalllwps(1); 807 if (error) 808 return (error); 809 } 810 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 811 p->p_flags |= P_INEXEC; 812 813 /* 814 * Tell procfs to release its hold on the process. It 815 * will return EAGAIN. 816 */ 817 if (p->p_stops) 818 wakeup(&p->p_stype); 819 820 /* 821 * After setting P_INEXEC wait for any remaining references to 822 * the process (p) to go away. 823 * 824 * In particular, a vfork/exec sequence will replace p->p_vmspace 825 * and we must interlock anyone trying to access the space (aka 826 * procfs or sys_process.c calling procfs_domem()). 827 * 828 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us. 829 */ 830 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0)); 831 832 /* 833 * Blow away entire process VM, if address space not shared, 834 * otherwise, create a new VM space so that other threads are 835 * not disrupted. If we are execing a resident vmspace we 836 * create a duplicate of it and remap the stack. 837 */ 838 map = &vmspace->vm_map; 839 if (vmcopy) { 840 vmspace_exec(imgp->proc, vmcopy); 841 vmspace = imgp->proc->p_vmspace; 842 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 843 map = &vmspace->vm_map; 844 } else if (vmspace_getrefs(vmspace) == 1) { 845 shmexit(vmspace); 846 pmap_remove_pages(vmspace_pmap(vmspace), 847 0, VM_MAX_USER_ADDRESS); 848 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 849 } else { 850 vmspace_exec(imgp->proc, NULL); 851 vmspace = imgp->proc->p_vmspace; 852 map = &vmspace->vm_map; 853 } 854 855 /* 856 * Really make sure lwp-specific and process-specific mappings 857 * are gone. 858 * 859 * Once we've done that, and because we are the only LWP left, with 860 * no TID-dependent mappings, we can reset the TID to 1 (the RB tree 861 * will remain consistent since it has only one entry). This way 862 * the exec'd program gets a nice deterministic tid of 1. 863 */ 864 lwp_userunmap(lp); 865 proc_userunmap(p); 866 lp->lwp_tid = 1; 867 p->p_lasttid = 1; 868 869 /* 870 * Allocate a new stack, generally make the stack non-executable 871 * but allow the program to adjust that (the program may desire to 872 * use areas of the stack for executable code). 873 */ 874 error = vm_map_stack(&vmspace->vm_map, &stack_addr, (vm_size_t)maxssiz, 875 0, 876 VM_PROT_READ|VM_PROT_WRITE, 877 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, 878 0); 879 if (error) 880 return (error); 881 882 /* 883 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 884 * VM_STACK case, but they are still used to monitor the size of the 885 * process stack so we can check the stack rlimit. 886 */ 887 vmspace->vm_ssize = sgrowsiz; /* in bytes */ 888 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 889 890 return(0); 891 } 892 893 /* 894 * Copy out argument and environment strings from the old process 895 * address space into the temporary string buffer. 896 */ 897 int 898 exec_copyin_args(struct image_args *args, char *fname, 899 enum exec_path_segflg segflg, char **argv, char **envv) 900 { 901 char *argp, *envp; 902 int error = 0; 903 size_t length; 904 905 args->buf = objcache_get(exec_objcache, M_WAITOK); 906 if (args->buf == NULL) 907 return (ENOMEM); 908 args->begin_argv = args->buf; 909 args->endp = args->begin_argv; 910 args->space = ARG_MAX; 911 912 args->fname = args->buf + ARG_MAX; 913 914 /* 915 * Copy the file name. 916 */ 917 if (segflg == PATH_SYSSPACE) { 918 error = copystr(fname, args->fname, PATH_MAX, &length); 919 } else if (segflg == PATH_USERSPACE) { 920 error = copyinstr(fname, args->fname, PATH_MAX, &length); 921 } 922 923 /* 924 * Extract argument strings. argv may not be NULL. The argv 925 * array is terminated by a NULL entry. We special-case the 926 * situation where argv[0] is NULL by passing { filename, NULL } 927 * to the new program to guarentee that the interpreter knows what 928 * file to open in case we exec an interpreted file. Note that 929 * a NULL argv[0] terminates the argv[] array. 930 * 931 * XXX the special-casing of argv[0] is historical and needs to be 932 * revisited. 933 */ 934 if (argv == NULL) 935 error = EFAULT; 936 if (error == 0) { 937 while ((argp = (caddr_t)(intptr_t) 938 fuword64((uintptr_t *)argv++)) != NULL) { 939 if (argp == (caddr_t)-1) { 940 error = EFAULT; 941 break; 942 } 943 error = copyinstr(argp, args->endp, 944 args->space, &length); 945 if (error) { 946 if (error == ENAMETOOLONG) 947 error = E2BIG; 948 break; 949 } 950 args->space -= length; 951 args->endp += length; 952 args->argc++; 953 } 954 if (args->argc == 0 && error == 0) { 955 length = strlen(args->fname) + 1; 956 if (length > args->space) { 957 error = E2BIG; 958 } else { 959 bcopy(args->fname, args->endp, length); 960 args->space -= length; 961 args->endp += length; 962 args->argc++; 963 } 964 } 965 } 966 967 args->begin_envv = args->endp; 968 969 /* 970 * extract environment strings. envv may be NULL. 971 */ 972 if (envv && error == 0) { 973 while ((envp = (caddr_t)(intptr_t) 974 fuword64((uintptr_t *)envv++))) { 975 if (envp == (caddr_t) -1) { 976 error = EFAULT; 977 break; 978 } 979 error = copyinstr(envp, args->endp, 980 args->space, &length); 981 if (error) { 982 if (error == ENAMETOOLONG) 983 error = E2BIG; 984 break; 985 } 986 args->space -= length; 987 args->endp += length; 988 args->envc++; 989 } 990 } 991 return (error); 992 } 993 994 void 995 exec_free_args(struct image_args *args) 996 { 997 if (args->buf) { 998 objcache_put(exec_objcache, args->buf); 999 args->buf = NULL; 1000 } 1001 } 1002 1003 /* 1004 * Copy strings out to the new process address space, constructing 1005 * new arg and env vector tables. Return a pointer to the base 1006 * so that it can be used as the initial stack pointer. 1007 * 1008 * The format is, roughly: 1009 * 1010 * [argv[]] <-- vectp 1011 * [envp[]] 1012 * [ELF_Auxargs] 1013 * 1014 * [args & env] <-- destp 1015 * [sgap] 1016 * [SPARE_USRSPACE] 1017 * [execpath] 1018 * [szsigcode] RO|NX 1019 * [ps_strings] RO|NX Top of user stack 1020 * 1021 */ 1022 static register_t * 1023 exec_copyout_strings(struct image_params *imgp) 1024 { 1025 int argc, envc, sgap; 1026 int gap; 1027 int argsenvspace; 1028 char **vectp; 1029 char *stringp, *destp, *szsigbase; 1030 register_t *stack_base; 1031 struct ps_strings *arginfo; 1032 size_t execpath_len; 1033 int szsigcode; 1034 1035 /* 1036 * Calculate string base and vector table pointers. 1037 * Also deal with signal trampoline code for this exec type. 1038 */ 1039 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1040 execpath_len = strlen(imgp->execpath) + 1; 1041 else 1042 execpath_len = 0; 1043 arginfo = (struct ps_strings *)PS_STRINGS; 1044 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 1045 1046 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 1047 gap = stackgap_random; 1048 cpu_ccfence(); 1049 if (gap != 0) { 1050 if (gap < 0) 1051 sgap = ALIGN(-gap); 1052 else 1053 sgap = ALIGN(karc4random() & (gap - 1)); 1054 } else { 1055 sgap = 0; 1056 } 1057 1058 /* 1059 * Calculate destp, which points to [args & env] and above. 1060 */ 1061 szsigbase = (char *)(intptr_t) 1062 trunc_page64((intptr_t)arginfo - szsigcode); 1063 szsigbase -= SZSIGCODE_EXTRA_BYTES; 1064 destp = szsigbase - 1065 roundup(execpath_len, sizeof(char *)) - 1066 SPARE_USRSPACE - 1067 sgap - 1068 argsenvspace; 1069 1070 /* 1071 * install sigcode 1072 */ 1073 if (szsigcode) 1074 copyout(imgp->proc->p_sysent->sv_sigcode, szsigbase, szsigcode); 1075 1076 /* 1077 * Copy the image path for the rtld 1078 */ 1079 if (execpath_len) { 1080 imgp->execpathp = (uintptr_t)szsigbase - 1081 roundup(execpath_len, sizeof(char *)); 1082 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 1083 } 1084 1085 /* 1086 * Calculate base for argv[], envp[], and ELF_Auxargs. 1087 */ 1088 vectp = (char **)destp - (AT_COUNT * 2); 1089 vectp -= imgp->args->argc + imgp->args->envc + 2; 1090 1091 stack_base = (register_t *)vectp; 1092 1093 stringp = imgp->args->begin_argv; 1094 argc = imgp->args->argc; 1095 envc = imgp->args->envc; 1096 1097 /* 1098 * Copy out strings - arguments and environment (at destp) 1099 */ 1100 copyout(stringp, destp, ARG_MAX - imgp->args->space); 1101 1102 /* 1103 * Fill in "ps_strings" struct for ps, w, etc. 1104 */ 1105 suword64((void *)&arginfo->ps_argvstr, (uint64_t)(intptr_t)vectp); 1106 suword32((void *)&arginfo->ps_nargvstr, argc); 1107 1108 /* 1109 * Fill in argument portion of vector table. 1110 */ 1111 for (; argc > 0; --argc) { 1112 suword64((void *)vectp++, (uintptr_t)destp); 1113 while (*stringp++ != 0) 1114 destp++; 1115 destp++; 1116 } 1117 1118 /* a null vector table pointer separates the argp's from the envp's */ 1119 suword64((void *)vectp++, 0); 1120 1121 suword64((void *)&arginfo->ps_envstr, (uintptr_t)vectp); 1122 suword32((void *)&arginfo->ps_nenvstr, envc); 1123 1124 /* 1125 * Fill in environment portion of vector table. 1126 */ 1127 for (; envc > 0; --envc) { 1128 suword64((void *)vectp++, (uintptr_t)destp); 1129 while (*stringp++ != 0) 1130 destp++; 1131 destp++; 1132 } 1133 1134 /* end of vector table is a null pointer */ 1135 suword64((void *)vectp, 0); 1136 1137 /* 1138 * Make the signal trampoline executable and read-only. 1139 */ 1140 vm_map_protect(&imgp->proc->p_vmspace->vm_map, 1141 (vm_offset_t)szsigbase, 1142 (vm_offset_t)szsigbase + PAGE_SIZE, 1143 VM_PROT_READ|VM_PROT_EXECUTE, FALSE); 1144 1145 return (stack_base); 1146 } 1147 1148 /* 1149 * Check permissions of file to execute. 1150 * Return 0 for success or error code on failure. 1151 */ 1152 int 1153 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 1154 { 1155 struct proc *p = imgp->proc; 1156 struct vnode *vp = imgp->vp; 1157 struct vattr *attr = imgp->attr; 1158 int error; 1159 1160 /* Get file attributes */ 1161 error = VOP_GETATTR(vp, attr); 1162 if (error) 1163 return (error); 1164 1165 /* 1166 * 1) Check if file execution is disabled for the filesystem that this 1167 * file resides on. 1168 * 2) Insure that at least one execute bit is on - otherwise root 1169 * will always succeed, and we don't want to happen unless the 1170 * file really is executable. 1171 * 3) Insure that the file is a regular file. 1172 */ 1173 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1174 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1175 ((attr->va_mode & 0111) == 0) || 1176 (attr->va_type != VREG)) { 1177 return (EACCES); 1178 } 1179 1180 /* 1181 * Zero length files can't be exec'd 1182 */ 1183 if (attr->va_size == 0) 1184 return (ENOEXEC); 1185 1186 /* 1187 * Check for execute permission to file based on current credentials. 1188 */ 1189 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1190 if (error) 1191 return (error); 1192 1193 /* 1194 * Check number of open-for-writes on the file and deny execution 1195 * if there are any. 1196 */ 1197 if (vp->v_writecount) 1198 return (ETXTBSY); 1199 1200 /* 1201 * Call filesystem specific open routine, which allows us to read, 1202 * write, and mmap the file. Without the VOP_OPEN we can only 1203 * stat the file. 1204 */ 1205 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1206 if (error) 1207 return (error); 1208 1209 return (0); 1210 } 1211 1212 /* 1213 * Exec handler registration 1214 */ 1215 int 1216 exec_register(const struct execsw *execsw_arg) 1217 { 1218 const struct execsw **es, **xs, **newexecsw; 1219 int count = 2; /* New slot and trailing NULL */ 1220 1221 if (execsw) 1222 for (es = execsw; *es; es++) 1223 count++; 1224 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1225 xs = newexecsw; 1226 if (execsw) 1227 for (es = execsw; *es; es++) 1228 *xs++ = *es; 1229 *xs++ = execsw_arg; 1230 *xs = NULL; 1231 if (execsw) 1232 kfree(execsw, M_TEMP); 1233 execsw = newexecsw; 1234 return 0; 1235 } 1236 1237 int 1238 exec_unregister(const struct execsw *execsw_arg) 1239 { 1240 const struct execsw **es, **xs, **newexecsw; 1241 int count = 1; 1242 1243 if (execsw == NULL) 1244 panic("unregister with no handlers left?"); 1245 1246 for (es = execsw; *es; es++) { 1247 if (*es == execsw_arg) 1248 break; 1249 } 1250 if (*es == NULL) 1251 return ENOENT; 1252 for (es = execsw; *es; es++) 1253 if (*es != execsw_arg) 1254 count++; 1255 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1256 xs = newexecsw; 1257 for (es = execsw; *es; es++) 1258 if (*es != execsw_arg) 1259 *xs++ = *es; 1260 *xs = NULL; 1261 if (execsw) 1262 kfree(execsw, M_TEMP); 1263 execsw = newexecsw; 1264 return 0; 1265 } 1266