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/sysmsg.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_lite lva; 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->lvap = &lva; 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 ((((lva.va_mode & VSUID) && p->p_ucred->cr_uid != lva.va_uid) || 449 ((lva.va_mode & VSGID) && p->p_ucred->cr_gid != lva.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 } 463 464 /* Clear any PROC_PDEATHSIG_CTL setting */ 465 p->p_deathsig = 0; 466 467 /* Close any file descriptors 0..2 that reference procfs */ 468 setugidsafety(p); 469 /* Make sure file descriptors 0..2 are in use. */ 470 error = fdcheckstd(lp); 471 if (error != 0) 472 goto exec_fail_dealloc; 473 /* 474 * Set the new credentials. 475 */ 476 cratom_proc(p); 477 if (lva.va_mode & VSUID) 478 change_euid(lva.va_uid); 479 if (lva.va_mode & VSGID) 480 p->p_ucred->cr_gid = lva.va_gid; 481 482 /* 483 * Clear local varsym variables 484 */ 485 varsymset_clean(&p->p_varsymset); 486 } else { 487 if (p->p_ucred->cr_uid == p->p_ucred->cr_ruid && 488 p->p_ucred->cr_gid == p->p_ucred->cr_rgid) 489 p->p_flags &= ~P_SUGID; 490 } 491 492 /* 493 * Implement correct POSIX saved-id behavior. 494 */ 495 if (p->p_ucred->cr_svuid != p->p_ucred->cr_uid || 496 p->p_ucred->cr_svgid != p->p_ucred->cr_gid) { 497 cratom_proc(p); 498 p->p_ucred->cr_svuid = p->p_ucred->cr_uid; 499 p->p_ucred->cr_svgid = p->p_ucred->cr_gid; 500 } 501 502 /* 503 * Store the vp for use in procfs. Be sure to keep p_textvp 504 * consistent if we block during the switch-over. 505 */ 506 ovp = p->p_textvp; 507 vref(imgp->vp); /* ref new vp */ 508 p->p_textvp = imgp->vp; 509 if (ovp) /* release old vp */ 510 vrele(ovp); 511 512 /* Release old namecache handle to text file */ 513 if (p->p_textnch.ncp) 514 cache_drop(&p->p_textnch); 515 516 if (nd->nl_nch.mount) 517 cache_copy(&nd->nl_nch, &p->p_textnch); 518 519 /* 520 * Notify others that we exec'd, and clear the P_INEXEC flag 521 * as we're now a bona fide freshly-execed process. 522 */ 523 KNOTE(&p->p_klist, NOTE_EXEC); 524 p->p_flags &= ~P_INEXEC; 525 if (p->p_stops) 526 wakeup(&p->p_stype); 527 528 /* 529 * If tracing the process, trap to debugger so breakpoints 530 * can be set before the program executes. 531 */ 532 STOPEVENT(p, S_EXEC, 0); 533 534 if (p->p_flags & P_TRACED) 535 ksignal(p, SIGTRAP); 536 537 /* clear "fork but no exec" flag, as we _are_ execing */ 538 p->p_acflag &= ~AFORK; 539 540 /* Set values passed into the program in registers. */ 541 exec_setregs(imgp->entry_addr, (u_long)(uintptr_t)stack_base, 542 imgp->ps_strings); 543 544 /* Set the access time on the vnode */ 545 vn_mark_atime(imgp->vp, td); 546 547 /* 548 * Free any previous argument cache 549 */ 550 pa = p->p_args; 551 p->p_args = NULL; 552 if (pa && refcount_release(&pa->ar_ref)) { 553 kfree(pa, M_PARGS); 554 pa = NULL; 555 } 556 557 /* 558 * Cache arguments if they fit inside our allowance 559 */ 560 i = imgp->args->begin_envv - imgp->args->begin_argv; 561 if (sizeof(struct pargs) + i <= ps_arg_cache_limit) { 562 pa = kmalloc(sizeof(struct pargs) + i, M_PARGS, M_WAITOK); 563 refcount_init(&pa->ar_ref, 1); 564 pa->ar_length = i; 565 bcopy(imgp->args->begin_argv, pa->ar_args, i); 566 KKASSERT(p->p_args == NULL); 567 p->p_args = pa; 568 } 569 570 exec_fail_dealloc: 571 572 /* 573 * free various allocated resources 574 */ 575 if (imgp->firstpage) 576 exec_unmap_first_page(imgp); 577 578 if (imgp->vp) { 579 vrele(imgp->vp); 580 imgp->vp = NULL; 581 } 582 583 if (imgp->freepath) 584 kfree(imgp->freepath, M_TEMP); 585 586 if (error == 0) { 587 ++mycpu->gd_cnt.v_exec; 588 lwkt_reltoken(&p->p_token); 589 return (0); 590 } 591 592 exec_fail: 593 /* 594 * we're done here, clear P_INEXEC if we were the ones that 595 * set it. Otherwise if vmspace_destroyed is still set we 596 * raced another thread and that thread is responsible for 597 * clearing it. 598 */ 599 if (imgp->vmspace_destroyed & 2) { 600 p->p_flags &= ~P_INEXEC; 601 if (p->p_stops) 602 wakeup(&p->p_stype); 603 } 604 lwkt_reltoken(&p->p_token); 605 if (imgp->vmspace_destroyed) { 606 /* 607 * Sorry, no more process anymore. exit gracefully. 608 * However we can't die right here, because our 609 * caller might have to clean up, so indicate a 610 * lethal error by returning -1. 611 */ 612 return(-1); 613 } else { 614 return(error); 615 } 616 } 617 618 /* 619 * execve() system call. 620 */ 621 int 622 sys_execve(struct sysmsg *sysmsg, const struct execve_args *uap) 623 { 624 struct nlookupdata nd; 625 struct image_args args; 626 int error; 627 628 bzero(&args, sizeof(args)); 629 630 error = nlookup_init(&nd, uap->fname, UIO_USERSPACE, NLC_FOLLOW); 631 if (error == 0) { 632 error = exec_copyin_args(&args, uap->fname, PATH_USERSPACE, 633 uap->argv, uap->envv); 634 } 635 if (error == 0) 636 error = kern_execve(&nd, &args); 637 nlookup_done(&nd); 638 exec_free_args(&args); 639 640 if (error < 0) { 641 /* We hit a lethal error condition. Let's die now. */ 642 exit1(W_EXITCODE(0, SIGABRT)); 643 /* NOTREACHED */ 644 } 645 646 /* 647 * The syscall result is returned in registers to the new program. 648 * Linux will register %edx as an atexit function and we must be 649 * sure to set it to 0. XXX 650 */ 651 if (error == 0) 652 sysmsg->sysmsg_result64 = 0; 653 654 return (error); 655 } 656 657 int 658 exec_map_page(struct image_params *imgp, vm_pindex_t pageno, 659 struct lwbuf **plwb, const char **pdata) 660 { 661 int rv; 662 vm_page_t ma; 663 vm_page_t m; 664 vm_object_t object; 665 666 /* 667 * The file has to be mappable. 668 */ 669 if ((object = imgp->vp->v_object) == NULL) 670 return (EIO); 671 672 if (pageno >= object->size) 673 return (EIO); 674 675 /* 676 * Shortcut using shared locks, improve concurrent execs. 677 */ 678 vm_object_hold_shared(object); 679 m = vm_page_lookup(object, pageno); 680 if (m) { 681 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL) { 682 vm_page_hold(m); 683 vm_page_sleep_busy(m, FALSE, "execpg"); 684 if ((m->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL && 685 m->object == object && m->pindex == pageno) { 686 vm_object_drop(object); 687 goto done; 688 } 689 vm_page_unhold(m); 690 } 691 } 692 vm_object_drop(object); 693 694 /* 695 * Do it the hard way 696 */ 697 vm_object_hold(object); 698 m = vm_page_grab(object, pageno, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 699 while ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) { 700 ma = m; 701 702 /* 703 * get_pages unbusies all the requested pages except the 704 * primary page (at index 0 in this case). The primary 705 * page may have been wired during the pagein (e.g. by 706 * the buffer cache) so vnode_pager_freepage() must be 707 * used to properly release it. 708 */ 709 rv = vm_pager_get_page(object, pageno, &ma, 1); 710 m = vm_page_lookup(object, pageno); 711 712 if (rv != VM_PAGER_OK || m == NULL || m->valid == 0) { 713 if (m) { 714 vm_page_protect(m, VM_PROT_NONE); 715 vnode_pager_freepage(m); 716 } 717 vm_object_drop(object); 718 return EIO; 719 } 720 } 721 vm_page_hold(m); 722 vm_page_wakeup(m); /* unbusy the page */ 723 vm_object_drop(object); 724 725 done: 726 *plwb = lwbuf_alloc(m, *plwb); 727 *pdata = (void *)lwbuf_kva(*plwb); 728 729 return (0); 730 } 731 732 /* 733 * Map the first page of an executable image. 734 * 735 * NOTE: If the mapping fails we have to NULL-out firstpage which may 736 * still be pointing to our supplied lwp structure. 737 */ 738 int 739 exec_map_first_page(struct image_params *imgp) 740 { 741 int err; 742 743 if (imgp->firstpage) 744 exec_unmap_first_page(imgp); 745 746 imgp->firstpage = &imgp->firstpage_cache; 747 err = exec_map_page(imgp, 0, &imgp->firstpage, &imgp->image_header); 748 749 if (err) { 750 imgp->firstpage = NULL; 751 return err; 752 } 753 754 return 0; 755 } 756 757 void 758 exec_unmap_page(struct lwbuf *lwb) 759 { 760 vm_page_t m; 761 762 crit_enter(); 763 if (lwb != NULL) { 764 m = lwbuf_page(lwb); 765 lwbuf_free(lwb); 766 vm_page_unhold(m); 767 } 768 crit_exit(); 769 } 770 771 void 772 exec_unmap_first_page(struct image_params *imgp) 773 { 774 exec_unmap_page(imgp->firstpage); 775 imgp->firstpage = NULL; 776 imgp->image_header = NULL; 777 } 778 779 /* 780 * Destroy old address space, and allocate a new stack 781 * The new stack is only SGROWSIZ large because it is grown 782 * automatically in trap.c. 783 * 784 * This is the point of no return. 785 */ 786 int 787 exec_new_vmspace(struct image_params *imgp, struct vmspace *vmcopy) 788 { 789 struct vmspace *vmspace = imgp->proc->p_vmspace; 790 vm_offset_t stack_addr = USRSTACK - maxssiz; 791 struct lwp *lp; 792 struct proc *p; 793 vm_map_t map; 794 int error; 795 796 /* 797 * Indicate that we cannot gracefully error out any more, kill 798 * any other threads present, and set P_INEXEC to indicate that 799 * we are now messing with the process structure proper. 800 * 801 * If killalllwps() races return an error which coupled with 802 * vmspace_destroyed will cause us to exit. This is what we 803 * want since another thread is patiently waiting for us to exit 804 * in that case. 805 */ 806 lp = curthread->td_lwp; 807 p = lp->lwp_proc; 808 imgp->vmspace_destroyed = 1; 809 810 if (curthread->td_proc->p_nthreads > 1) { 811 error = killalllwps(1); 812 if (error) 813 return (error); 814 } 815 imgp->vmspace_destroyed |= 2; /* we are responsible for P_INEXEC */ 816 p->p_flags |= P_INEXEC; 817 818 /* 819 * Tell procfs to release its hold on the process. It 820 * will return EAGAIN. 821 */ 822 if (p->p_stops) 823 wakeup(&p->p_stype); 824 825 /* 826 * After setting P_INEXEC wait for any remaining references to 827 * the process (p) to go away. 828 * 829 * In particular, a vfork/exec sequence will replace p->p_vmspace 830 * and we must interlock anyone trying to access the space (aka 831 * procfs or sys_process.c calling procfs_domem()). 832 * 833 * If P_PPWAIT is set the parent vfork()'d and has a PHOLD() on us. 834 */ 835 PSTALL(p, "exec1", ((p->p_flags & P_PPWAIT) ? 1 : 0)); 836 837 /* 838 * Blow away entire process VM, if address space not shared, 839 * otherwise, create a new VM space so that other threads are 840 * not disrupted. If we are execing a resident vmspace we 841 * create a duplicate of it and remap the stack. 842 */ 843 map = &vmspace->vm_map; 844 if (vmcopy) { 845 vmspace_exec(imgp->proc, vmcopy); 846 vmspace = imgp->proc->p_vmspace; 847 pmap_remove_pages(vmspace_pmap(vmspace), stack_addr, USRSTACK); 848 map = &vmspace->vm_map; 849 } else if (vmspace_getrefs(vmspace) == 1) { 850 shmexit(vmspace); 851 pmap_remove_pages(vmspace_pmap(vmspace), 852 0, VM_MAX_USER_ADDRESS); 853 vm_map_remove(map, 0, VM_MAX_USER_ADDRESS); 854 } else { 855 vmspace_exec(imgp->proc, NULL); 856 vmspace = imgp->proc->p_vmspace; 857 map = &vmspace->vm_map; 858 } 859 860 /* 861 * Really make sure lwp-specific and process-specific mappings 862 * are gone. 863 * 864 * Once we've done that, and because we are the only LWP left, with 865 * no TID-dependent mappings, we can reset the TID to 1 (the RB tree 866 * will remain consistent since it has only one entry). This way 867 * the exec'd program gets a nice deterministic tid of 1. 868 */ 869 lwp_userunmap(lp); 870 proc_userunmap(p); 871 lp->lwp_tid = 1; 872 p->p_lasttid = 1; 873 874 /* 875 * Allocate a new stack, generally make the stack non-executable 876 * but allow the program to adjust that (the program may desire to 877 * use areas of the stack for executable code). 878 */ 879 error = vm_map_stack(&vmspace->vm_map, &stack_addr, (vm_size_t)maxssiz, 880 0, 881 VM_PROT_READ|VM_PROT_WRITE, 882 VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE, 883 0); 884 if (error) 885 return (error); 886 887 /* 888 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 889 * VM_STACK case, but they are still used to monitor the size of the 890 * process stack so we can check the stack rlimit. 891 */ 892 vmspace->vm_ssize = sgrowsiz; /* in bytes */ 893 vmspace->vm_maxsaddr = (char *)USRSTACK - maxssiz; 894 895 return(0); 896 } 897 898 /* 899 * Copy out argument and environment strings from the old process 900 * address space into the temporary string buffer. 901 */ 902 int 903 exec_copyin_args(struct image_args *args, char *fname, 904 enum exec_path_segflg segflg, char **argv, char **envv) 905 { 906 char *argp, *envp; 907 int error = 0; 908 size_t length; 909 910 args->buf = objcache_get(exec_objcache, M_WAITOK); 911 if (args->buf == NULL) 912 return (ENOMEM); 913 args->begin_argv = args->buf; 914 args->endp = args->begin_argv; 915 args->space = ARG_MAX; 916 917 args->fname = args->buf + ARG_MAX; 918 919 /* 920 * Copy the file name. 921 */ 922 if (segflg == PATH_SYSSPACE) { 923 error = copystr(fname, args->fname, PATH_MAX, &length); 924 } else if (segflg == PATH_USERSPACE) { 925 error = copyinstr(fname, args->fname, PATH_MAX, &length); 926 } 927 928 /* 929 * Extract argument strings. argv may not be NULL. The argv 930 * array is terminated by a NULL entry. We special-case the 931 * situation where argv[0] is NULL by passing { filename, NULL } 932 * to the new program to guarentee that the interpreter knows what 933 * file to open in case we exec an interpreted file. Note that 934 * a NULL argv[0] terminates the argv[] array. 935 * 936 * XXX the special-casing of argv[0] is historical and needs to be 937 * revisited. 938 */ 939 if (argv == NULL) 940 error = EFAULT; 941 if (error == 0) { 942 while ((argp = (caddr_t)(intptr_t) 943 fuword64((uintptr_t *)argv++)) != NULL) { 944 if (argp == (caddr_t)-1) { 945 error = EFAULT; 946 break; 947 } 948 error = copyinstr(argp, args->endp, 949 args->space, &length); 950 if (error) { 951 if (error == ENAMETOOLONG) 952 error = E2BIG; 953 break; 954 } 955 args->space -= length; 956 args->endp += length; 957 args->argc++; 958 } 959 if (args->argc == 0 && error == 0) { 960 length = strlen(args->fname) + 1; 961 if (length > args->space) { 962 error = E2BIG; 963 } else { 964 bcopy(args->fname, args->endp, length); 965 args->space -= length; 966 args->endp += length; 967 args->argc++; 968 } 969 } 970 } 971 972 args->begin_envv = args->endp; 973 974 /* 975 * extract environment strings. envv may be NULL. 976 */ 977 if (envv && error == 0) { 978 while ((envp = (caddr_t)(intptr_t) 979 fuword64((uintptr_t *)envv++))) { 980 if (envp == (caddr_t) -1) { 981 error = EFAULT; 982 break; 983 } 984 error = copyinstr(envp, args->endp, 985 args->space, &length); 986 if (error) { 987 if (error == ENAMETOOLONG) 988 error = E2BIG; 989 break; 990 } 991 args->space -= length; 992 args->endp += length; 993 args->envc++; 994 } 995 } 996 return (error); 997 } 998 999 void 1000 exec_free_args(struct image_args *args) 1001 { 1002 if (args->buf) { 1003 objcache_put(exec_objcache, args->buf); 1004 args->buf = NULL; 1005 } 1006 } 1007 1008 /* 1009 * Copy strings out to the new process address space, constructing 1010 * new arg and env vector tables. Return a pointer to the base 1011 * so that it can be used as the initial stack pointer. 1012 * 1013 * The format is, roughly: 1014 * 1015 * [argv[]] <-- vectp 1016 * [envp[]] 1017 * [ELF_Auxargs] 1018 * 1019 * [args & env] <-- destp 1020 * [sgap] 1021 * [SPARE_USRSPACE] 1022 * [execpath] 1023 * [szsigcode] RO|NX 1024 * [ps_strings] RO|NX Top of user stack 1025 * 1026 */ 1027 static register_t * 1028 exec_copyout_strings(struct image_params *imgp) 1029 { 1030 int argc, envc, sgap; 1031 int gap; 1032 int argsenvspace; 1033 char **vectp; 1034 char *stringp, *destp, *szsigbase; 1035 register_t *stack_base; 1036 struct ps_strings *arginfo; 1037 size_t execpath_len; 1038 int szsigcode; 1039 1040 /* 1041 * Calculate string base and vector table pointers. 1042 * Also deal with signal trampoline code for this exec type. 1043 */ 1044 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1045 execpath_len = strlen(imgp->execpath) + 1; 1046 else 1047 execpath_len = 0; 1048 arginfo = (struct ps_strings *)PS_STRINGS; 1049 szsigcode = *(imgp->proc->p_sysent->sv_szsigcode); 1050 1051 argsenvspace = roundup((ARG_MAX - imgp->args->space), sizeof(char *)); 1052 gap = stackgap_random; 1053 cpu_ccfence(); 1054 if (gap != 0) { 1055 if (gap < 0) 1056 sgap = ALIGN(-gap); 1057 else 1058 sgap = ALIGN(karc4random() & (gap - 1)); 1059 } else { 1060 sgap = 0; 1061 } 1062 1063 /* 1064 * Calculate destp, which points to [args & env] and above. 1065 */ 1066 szsigbase = (char *)(intptr_t) 1067 trunc_page64((intptr_t)arginfo - szsigcode); 1068 szsigbase -= SZSIGCODE_EXTRA_BYTES; 1069 destp = szsigbase - 1070 roundup(execpath_len, sizeof(char *)) - 1071 SPARE_USRSPACE - 1072 sgap - 1073 argsenvspace; 1074 1075 /* 1076 * install sigcode 1077 */ 1078 if (szsigcode) 1079 copyout(imgp->proc->p_sysent->sv_sigcode, szsigbase, szsigcode); 1080 1081 /* 1082 * Copy the image path for the rtld 1083 */ 1084 if (execpath_len) { 1085 imgp->execpathp = (uintptr_t)szsigbase - 1086 roundup(execpath_len, sizeof(char *)); 1087 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len); 1088 } 1089 1090 /* 1091 * Calculate base for argv[], envp[], and ELF_Auxargs. 1092 */ 1093 vectp = (char **)destp - (AT_COUNT * 2); 1094 vectp -= imgp->args->argc + imgp->args->envc + 2; 1095 1096 stack_base = (register_t *)vectp; 1097 1098 stringp = imgp->args->begin_argv; 1099 argc = imgp->args->argc; 1100 envc = imgp->args->envc; 1101 1102 /* 1103 * Copy out strings - arguments and environment (at destp) 1104 */ 1105 copyout(stringp, destp, ARG_MAX - imgp->args->space); 1106 1107 /* 1108 * Fill in "ps_strings" struct for ps, w, etc. 1109 */ 1110 suword64((void *)&arginfo->ps_argvstr, (uint64_t)(intptr_t)vectp); 1111 suword32((void *)&arginfo->ps_nargvstr, argc); 1112 1113 /* 1114 * Fill in argument portion of vector table. 1115 */ 1116 for (; argc > 0; --argc) { 1117 suword64((void *)vectp++, (uintptr_t)destp); 1118 while (*stringp++ != 0) 1119 destp++; 1120 destp++; 1121 } 1122 1123 /* a null vector table pointer separates the argp's from the envp's */ 1124 suword64((void *)vectp++, 0); 1125 1126 suword64((void *)&arginfo->ps_envstr, (uintptr_t)vectp); 1127 suword32((void *)&arginfo->ps_nenvstr, envc); 1128 1129 /* 1130 * Fill in environment portion of vector table. 1131 */ 1132 for (; envc > 0; --envc) { 1133 suword64((void *)vectp++, (uintptr_t)destp); 1134 while (*stringp++ != 0) 1135 destp++; 1136 destp++; 1137 } 1138 1139 /* end of vector table is a null pointer */ 1140 suword64((void *)vectp, 0); 1141 1142 /* 1143 * Make the signal trampoline executable and read-only. 1144 */ 1145 vm_map_protect(&imgp->proc->p_vmspace->vm_map, 1146 (vm_offset_t)szsigbase, 1147 (vm_offset_t)szsigbase + PAGE_SIZE, 1148 VM_PROT_READ|VM_PROT_EXECUTE, FALSE); 1149 1150 return (stack_base); 1151 } 1152 1153 /* 1154 * Check permissions of file to execute. 1155 * Return 0 for success or error code on failure. 1156 */ 1157 int 1158 exec_check_permissions(struct image_params *imgp, struct mount *topmnt) 1159 { 1160 struct proc *p = imgp->proc; 1161 struct vnode *vp = imgp->vp; 1162 struct vattr_lite *lvap = imgp->lvap; 1163 int error; 1164 1165 /* Get file attributes */ 1166 error = VOP_GETATTR_LITE(vp, lvap); 1167 if (error) 1168 return (error); 1169 1170 /* 1171 * 1) Check if file execution is disabled for the filesystem that this 1172 * file resides on. 1173 * 2) Insure that at least one execute bit is on - otherwise root 1174 * will always succeed, and we don't want to happen unless the 1175 * file really is executable. 1176 * 3) Insure that the file is a regular file. 1177 */ 1178 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1179 ((topmnt != NULL) && (topmnt->mnt_flag & MNT_NOEXEC)) || 1180 ((lvap->va_mode & 0111) == 0) || 1181 (lvap->va_type != VREG)) { 1182 return (EACCES); 1183 } 1184 1185 /* 1186 * Zero length files can't be exec'd 1187 */ 1188 if (lvap->va_size == 0) 1189 return (ENOEXEC); 1190 1191 /* 1192 * Check for execute permission to file based on current credentials. 1193 */ 1194 error = VOP_EACCESS(vp, VEXEC, p->p_ucred); 1195 if (error) 1196 return (error); 1197 1198 /* 1199 * Check number of open-for-writes on the file and deny execution 1200 * if there are any. 1201 */ 1202 if (vp->v_writecount) 1203 return (ETXTBSY); 1204 1205 /* 1206 * Call filesystem specific open routine, which allows us to read, 1207 * write, and mmap the file. Without the VOP_OPEN we can only 1208 * stat the file. 1209 */ 1210 error = VOP_OPEN(vp, FREAD, p->p_ucred, NULL); 1211 if (error) 1212 return (error); 1213 1214 return (0); 1215 } 1216 1217 /* 1218 * Exec handler registration 1219 */ 1220 int 1221 exec_register(const struct execsw *execsw_arg) 1222 { 1223 const struct execsw **es, **xs, **newexecsw; 1224 int count = 2; /* New slot and trailing NULL */ 1225 1226 if (execsw) 1227 for (es = execsw; *es; es++) 1228 count++; 1229 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1230 xs = newexecsw; 1231 if (execsw) 1232 for (es = execsw; *es; es++) 1233 *xs++ = *es; 1234 *xs++ = execsw_arg; 1235 *xs = NULL; 1236 if (execsw) 1237 kfree(execsw, M_TEMP); 1238 execsw = newexecsw; 1239 return 0; 1240 } 1241 1242 int 1243 exec_unregister(const struct execsw *execsw_arg) 1244 { 1245 const struct execsw **es, **xs, **newexecsw; 1246 int count = 1; 1247 1248 if (execsw == NULL) 1249 panic("unregister with no handlers left?"); 1250 1251 for (es = execsw; *es; es++) { 1252 if (*es == execsw_arg) 1253 break; 1254 } 1255 if (*es == NULL) 1256 return ENOENT; 1257 for (es = execsw; *es; es++) 1258 if (*es != execsw_arg) 1259 count++; 1260 newexecsw = kmalloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1261 xs = newexecsw; 1262 for (es = execsw; *es; es++) 1263 if (*es != execsw_arg) 1264 *xs++ = *es; 1265 *xs = NULL; 1266 if (execsw) 1267 kfree(execsw, M_TEMP); 1268 execsw = newexecsw; 1269 return 0; 1270 } 1271