1 /* $NetBSD: kern_exec.c,v 1.314 2011/04/26 16:36:42 joerg Exp $ */ 2 3 /*- 4 * Copyright (c) 2008 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 26 * POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 /*- 30 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou 31 * Copyright (C) 1992 Wolfgang Solfrank. 32 * Copyright (C) 1992 TooLs GmbH. 33 * All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. All advertising materials mentioning features or use of this software 44 * must display the following acknowledgement: 45 * This product includes software developed by TooLs GmbH. 46 * 4. The name of TooLs GmbH may not be used to endorse or promote products 47 * derived from this software without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 50 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 51 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 52 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 53 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 54 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 55 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 56 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 57 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 58 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 59 */ 60 61 #include <sys/cdefs.h> 62 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.314 2011/04/26 16:36:42 joerg Exp $"); 63 64 #include "opt_ktrace.h" 65 #include "opt_modular.h" 66 #include "opt_syscall_debug.h" 67 #include "veriexec.h" 68 #include "opt_pax.h" 69 #include "opt_sa.h" 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/filedesc.h> 74 #include <sys/kernel.h> 75 #include <sys/proc.h> 76 #include <sys/mount.h> 77 #include <sys/malloc.h> 78 #include <sys/kmem.h> 79 #include <sys/namei.h> 80 #include <sys/vnode.h> 81 #include <sys/file.h> 82 #include <sys/acct.h> 83 #include <sys/exec.h> 84 #include <sys/ktrace.h> 85 #include <sys/uidinfo.h> 86 #include <sys/wait.h> 87 #include <sys/mman.h> 88 #include <sys/ras.h> 89 #include <sys/signalvar.h> 90 #include <sys/stat.h> 91 #include <sys/syscall.h> 92 #include <sys/kauth.h> 93 #include <sys/lwpctl.h> 94 #include <sys/pax.h> 95 #include <sys/cpu.h> 96 #include <sys/module.h> 97 #include <sys/sa.h> 98 #include <sys/savar.h> 99 #include <sys/syscallvar.h> 100 #include <sys/syscallargs.h> 101 #if NVERIEXEC > 0 102 #include <sys/verified_exec.h> 103 #endif /* NVERIEXEC > 0 */ 104 #include <sys/sdt.h> 105 106 #include <uvm/uvm_extern.h> 107 108 #include <machine/reg.h> 109 110 #include <compat/common/compat_util.h> 111 112 static int exec_sigcode_map(struct proc *, const struct emul *); 113 114 #ifdef DEBUG_EXEC 115 #define DPRINTF(a) printf a 116 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \ 117 __LINE__, (s), (a), (b)) 118 #else 119 #define DPRINTF(a) 120 #define COPYPRINTF(s, a, b) 121 #endif /* DEBUG_EXEC */ 122 123 /* 124 * DTrace SDT provider definitions 125 */ 126 SDT_PROBE_DEFINE(proc,,,exec, 127 "char *", NULL, 128 NULL, NULL, NULL, NULL, 129 NULL, NULL, NULL, NULL); 130 SDT_PROBE_DEFINE(proc,,,exec_success, 131 "char *", NULL, 132 NULL, NULL, NULL, NULL, 133 NULL, NULL, NULL, NULL); 134 SDT_PROBE_DEFINE(proc,,,exec_failure, 135 "int", NULL, 136 NULL, NULL, NULL, NULL, 137 NULL, NULL, NULL, NULL); 138 139 /* 140 * Exec function switch: 141 * 142 * Note that each makecmds function is responsible for loading the 143 * exec package with the necessary functions for any exec-type-specific 144 * handling. 145 * 146 * Functions for specific exec types should be defined in their own 147 * header file. 148 */ 149 static const struct execsw **execsw = NULL; 150 static int nexecs; 151 152 u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */ 153 154 /* list of dynamically loaded execsw entries */ 155 static LIST_HEAD(execlist_head, exec_entry) ex_head = 156 LIST_HEAD_INITIALIZER(ex_head); 157 struct exec_entry { 158 LIST_ENTRY(exec_entry) ex_list; 159 SLIST_ENTRY(exec_entry) ex_slist; 160 const struct execsw *ex_sw; 161 }; 162 163 #ifndef __HAVE_SYSCALL_INTERN 164 void syscall(void); 165 #endif 166 167 #ifdef KERN_SA 168 static struct sa_emul saemul_netbsd = { 169 sizeof(ucontext_t), 170 sizeof(struct sa_t), 171 sizeof(struct sa_t *), 172 NULL, 173 NULL, 174 cpu_upcall, 175 (void (*)(struct lwp *, void *))getucontext_sa, 176 sa_ucsp 177 }; 178 #endif /* KERN_SA */ 179 180 /* NetBSD emul struct */ 181 struct emul emul_netbsd = { 182 .e_name = "netbsd", 183 .e_path = NULL, 184 #ifndef __HAVE_MINIMAL_EMUL 185 .e_flags = EMUL_HAS_SYS___syscall, 186 .e_errno = NULL, 187 .e_nosys = SYS_syscall, 188 .e_nsysent = SYS_NSYSENT, 189 #endif 190 .e_sysent = sysent, 191 #ifdef SYSCALL_DEBUG 192 .e_syscallnames = syscallnames, 193 #else 194 .e_syscallnames = NULL, 195 #endif 196 .e_sendsig = sendsig, 197 .e_trapsignal = trapsignal, 198 .e_tracesig = NULL, 199 .e_sigcode = NULL, 200 .e_esigcode = NULL, 201 .e_sigobject = NULL, 202 .e_setregs = setregs, 203 .e_proc_exec = NULL, 204 .e_proc_fork = NULL, 205 .e_proc_exit = NULL, 206 .e_lwp_fork = NULL, 207 .e_lwp_exit = NULL, 208 #ifdef __HAVE_SYSCALL_INTERN 209 .e_syscall_intern = syscall_intern, 210 #else 211 .e_syscall = syscall, 212 #endif 213 .e_sysctlovly = NULL, 214 .e_fault = NULL, 215 .e_vm_default_addr = uvm_default_mapaddr, 216 .e_usertrap = NULL, 217 #ifdef KERN_SA 218 .e_sa = &saemul_netbsd, 219 #else 220 .e_sa = NULL, 221 #endif 222 .e_ucsize = sizeof(ucontext_t), 223 .e_startlwp = startlwp 224 }; 225 226 /* 227 * Exec lock. Used to control access to execsw[] structures. 228 * This must not be static so that netbsd32 can access it, too. 229 */ 230 krwlock_t exec_lock; 231 232 static kmutex_t sigobject_lock; 233 234 static void * 235 exec_pool_alloc(struct pool *pp, int flags) 236 { 237 238 return (void *)uvm_km_alloc(kernel_map, NCARGS, 0, 239 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA); 240 } 241 242 static void 243 exec_pool_free(struct pool *pp, void *addr) 244 { 245 246 uvm_km_free(kernel_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE); 247 } 248 249 static struct pool exec_pool; 250 251 static struct pool_allocator exec_palloc = { 252 .pa_alloc = exec_pool_alloc, 253 .pa_free = exec_pool_free, 254 .pa_pagesz = NCARGS 255 }; 256 257 /* 258 * check exec: 259 * given an "executable" described in the exec package's namei info, 260 * see what we can do with it. 261 * 262 * ON ENTRY: 263 * exec package with appropriate namei info 264 * lwp pointer of exec'ing lwp 265 * NO SELF-LOCKED VNODES 266 * 267 * ON EXIT: 268 * error: nothing held, etc. exec header still allocated. 269 * ok: filled exec package, executable's vnode (unlocked). 270 * 271 * EXEC SWITCH ENTRY: 272 * Locked vnode to check, exec package, proc. 273 * 274 * EXEC SWITCH EXIT: 275 * ok: return 0, filled exec package, executable's vnode (unlocked). 276 * error: destructive: 277 * everything deallocated execept exec header. 278 * non-destructive: 279 * error code, executable's vnode (unlocked), 280 * exec header unmodified. 281 */ 282 int 283 /*ARGSUSED*/ 284 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb) 285 { 286 int error, i; 287 struct vnode *vp; 288 struct nameidata nd; 289 size_t resid; 290 291 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 292 293 /* first get the vnode */ 294 if ((error = namei(&nd)) != 0) 295 return error; 296 epp->ep_vp = vp = nd.ni_vp; 297 /* this cannot overflow as both are size PATH_MAX */ 298 strcpy(epp->ep_resolvedname, nd.ni_pnbuf); 299 300 #ifdef DIAGNOSTIC 301 /* paranoia (take this out once namei stuff stabilizes) */ 302 memset(nd.ni_pnbuf, '~', PATH_MAX); 303 #endif 304 305 /* check access and type */ 306 if (vp->v_type != VREG) { 307 error = EACCES; 308 goto bad1; 309 } 310 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0) 311 goto bad1; 312 313 /* get attributes */ 314 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0) 315 goto bad1; 316 317 /* Check mount point */ 318 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 319 error = EACCES; 320 goto bad1; 321 } 322 if (vp->v_mount->mnt_flag & MNT_NOSUID) 323 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 324 325 /* try to open it */ 326 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0) 327 goto bad1; 328 329 /* unlock vp, since we need it unlocked from here on out. */ 330 VOP_UNLOCK(vp); 331 332 #if NVERIEXEC > 0 333 error = veriexec_verify(l, vp, epp->ep_resolvedname, 334 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT, 335 NULL); 336 if (error) 337 goto bad2; 338 #endif /* NVERIEXEC > 0 */ 339 340 #ifdef PAX_SEGVGUARD 341 error = pax_segvguard(l, vp, epp->ep_resolvedname, false); 342 if (error) 343 goto bad2; 344 #endif /* PAX_SEGVGUARD */ 345 346 /* now we have the file, get the exec header */ 347 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 348 UIO_SYSSPACE, 0, l->l_cred, &resid, NULL); 349 if (error) 350 goto bad2; 351 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 352 353 /* 354 * Set up default address space limits. Can be overridden 355 * by individual exec packages. 356 * 357 * XXX probably should be all done in the exec packages. 358 */ 359 epp->ep_vm_minaddr = VM_MIN_ADDRESS; 360 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 361 /* 362 * set up the vmcmds for creation of the process 363 * address space 364 */ 365 error = ENOEXEC; 366 for (i = 0; i < nexecs; i++) { 367 int newerror; 368 369 epp->ep_esch = execsw[i]; 370 newerror = (*execsw[i]->es_makecmds)(l, epp); 371 372 if (!newerror) { 373 /* Seems ok: check that entry point is sane */ 374 if (epp->ep_entry > VM_MAXUSER_ADDRESS) { 375 error = ENOEXEC; 376 break; 377 } 378 379 /* check limits */ 380 if ((epp->ep_tsize > MAXTSIZ) || 381 (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit 382 [RLIMIT_DATA].rlim_cur)) { 383 error = ENOMEM; 384 break; 385 } 386 return 0; 387 } 388 389 if (epp->ep_emul_root != NULL) { 390 vrele(epp->ep_emul_root); 391 epp->ep_emul_root = NULL; 392 } 393 if (epp->ep_interp != NULL) { 394 vrele(epp->ep_interp); 395 epp->ep_interp = NULL; 396 } 397 398 /* make sure the first "interesting" error code is saved. */ 399 if (error == ENOEXEC) 400 error = newerror; 401 402 if (epp->ep_flags & EXEC_DESTR) 403 /* Error from "#!" code, tidied up by recursive call */ 404 return error; 405 } 406 407 /* not found, error */ 408 409 /* 410 * free any vmspace-creation commands, 411 * and release their references 412 */ 413 kill_vmcmds(&epp->ep_vmcmds); 414 415 bad2: 416 /* 417 * close and release the vnode, restore the old one, free the 418 * pathname buf, and punt. 419 */ 420 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 421 VOP_CLOSE(vp, FREAD, l->l_cred); 422 vput(vp); 423 return error; 424 425 bad1: 426 /* 427 * free the namei pathname buffer, and put the vnode 428 * (which we don't yet have open). 429 */ 430 vput(vp); /* was still locked */ 431 return error; 432 } 433 434 #ifdef __MACHINE_STACK_GROWS_UP 435 #define STACK_PTHREADSPACE NBPG 436 #else 437 #define STACK_PTHREADSPACE 0 438 #endif 439 440 static int 441 execve_fetch_element(char * const *array, size_t index, char **value) 442 { 443 return copyin(array + index, value, sizeof(*value)); 444 } 445 446 /* 447 * exec system call 448 */ 449 /* ARGSUSED */ 450 int 451 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval) 452 { 453 /* { 454 syscallarg(const char *) path; 455 syscallarg(char * const *) argp; 456 syscallarg(char * const *) envp; 457 } */ 458 459 return execve1(l, SCARG(uap, path), SCARG(uap, argp), 460 SCARG(uap, envp), execve_fetch_element); 461 } 462 463 /* 464 * Load modules to try and execute an image that we do not understand. 465 * If no execsw entries are present, we load those likely to be needed 466 * in order to run native images only. Otherwise, we autoload all 467 * possible modules that could let us run the binary. XXX lame 468 */ 469 static void 470 exec_autoload(void) 471 { 472 #ifdef MODULAR 473 static const char * const native[] = { 474 "exec_elf32", 475 "exec_elf64", 476 "exec_script", 477 NULL 478 }; 479 static const char * const compat[] = { 480 "exec_elf32", 481 "exec_elf64", 482 "exec_script", 483 "exec_aout", 484 "exec_coff", 485 "exec_ecoff", 486 "compat_aoutm68k", 487 "compat_freebsd", 488 "compat_ibcs2", 489 "compat_linux", 490 "compat_linux32", 491 "compat_netbsd32", 492 "compat_sunos", 493 "compat_sunos32", 494 "compat_svr4", 495 "compat_svr4_32", 496 "compat_ultrix", 497 NULL 498 }; 499 char const * const *list; 500 int i; 501 502 list = (nexecs == 0 ? native : compat); 503 for (i = 0; list[i] != NULL; i++) { 504 if (module_autoload(list[i], MODULE_CLASS_MISC) != 0) { 505 continue; 506 } 507 yield(); 508 } 509 #endif 510 } 511 512 int 513 execve1(struct lwp *l, const char *path, char * const *args, 514 char * const *envs, execve_fetch_element_t fetch_element) 515 { 516 int error; 517 struct exec_package pack; 518 struct pathbuf *pb; 519 struct vattr attr; 520 struct proc *p; 521 char *argp; 522 char *dp, *sp; 523 long argc, envc; 524 size_t i, len; 525 char *stack; 526 struct ps_strings arginfo; 527 struct ps_strings32 arginfo32; 528 void *aip; 529 struct vmspace *vm; 530 struct exec_fakearg *tmpfap; 531 int szsigcode; 532 struct exec_vmcmd *base_vcp; 533 int oldlwpflags; 534 ksiginfo_t ksi; 535 ksiginfoq_t kq; 536 const char *pathstring; 537 char *resolvedpathbuf; 538 const char *commandname; 539 u_int modgen; 540 size_t ps_strings_sz; 541 542 p = l->l_proc; 543 modgen = 0; 544 545 SDT_PROBE(proc,,,exec, path, 0, 0, 0, 0); 546 547 /* 548 * Check if we have exceeded our number of processes limit. 549 * This is so that we handle the case where a root daemon 550 * forked, ran setuid to become the desired user and is trying 551 * to exec. The obvious place to do the reference counting check 552 * is setuid(), but we don't do the reference counting check there 553 * like other OS's do because then all the programs that use setuid() 554 * must be modified to check the return code of setuid() and exit(). 555 * It is dangerous to make setuid() fail, because it fails open and 556 * the program will continue to run as root. If we make it succeed 557 * and return an error code, again we are not enforcing the limit. 558 * The best place to enforce the limit is here, when the process tries 559 * to execute a new image, because eventually the process will need 560 * to call exec in order to do something useful. 561 */ 562 retry: 563 if ((p->p_flag & PK_SUGID) && kauth_authorize_generic(l->l_cred, 564 KAUTH_GENERIC_ISSUSER, NULL) != 0 && chgproccnt(kauth_cred_getuid( 565 l->l_cred), 0) > p->p_rlimit[RLIMIT_NPROC].rlim_cur) 566 return EAGAIN; 567 568 oldlwpflags = l->l_flag & (LW_SA | LW_SA_UPCALL); 569 if (l->l_flag & LW_SA) { 570 lwp_lock(l); 571 l->l_flag &= ~(LW_SA | LW_SA_UPCALL); 572 lwp_unlock(l); 573 } 574 575 /* 576 * Drain existing references and forbid new ones. The process 577 * should be left alone until we're done here. This is necessary 578 * to avoid race conditions - e.g. in ptrace() - that might allow 579 * a local user to illicitly obtain elevated privileges. 580 */ 581 rw_enter(&p->p_reflock, RW_WRITER); 582 583 base_vcp = NULL; 584 /* 585 * Init the namei data to point the file user's program name. 586 * This is done here rather than in check_exec(), so that it's 587 * possible to override this settings if any of makecmd/probe 588 * functions call check_exec() recursively - for example, 589 * see exec_script_makecmds(). 590 */ 591 error = pathbuf_copyin(path, &pb); 592 if (error) { 593 DPRINTF(("%s: pathbuf_copyin path @%p %d\n", __func__, 594 path, error)); 595 goto clrflg; 596 } 597 pathstring = pathbuf_stringcopy_get(pb); 598 resolvedpathbuf = PNBUF_GET(); 599 #ifdef DIAGNOSTIC 600 strcpy(resolvedpathbuf, "/wrong"); 601 #endif 602 603 /* 604 * initialize the fields of the exec package. 605 */ 606 pack.ep_name = path; 607 pack.ep_kname = pathstring; 608 pack.ep_resolvedname = resolvedpathbuf; 609 pack.ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP); 610 pack.ep_hdrlen = exec_maxhdrsz; 611 pack.ep_hdrvalid = 0; 612 pack.ep_emul_arg = NULL; 613 pack.ep_vmcmds.evs_cnt = 0; 614 pack.ep_vmcmds.evs_used = 0; 615 pack.ep_vap = &attr; 616 pack.ep_flags = 0; 617 pack.ep_emul_root = NULL; 618 pack.ep_interp = NULL; 619 pack.ep_esch = NULL; 620 pack.ep_pax_flags = 0; 621 622 rw_enter(&exec_lock, RW_READER); 623 624 /* see if we can run it. */ 625 if ((error = check_exec(l, &pack, pb)) != 0) { 626 if (error != ENOENT) { 627 DPRINTF(("%s: check exec failed %d\n", 628 __func__, error)); 629 } 630 goto freehdr; 631 } 632 633 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 634 635 /* allocate an argument buffer */ 636 argp = pool_get(&exec_pool, PR_WAITOK); 637 KASSERT(argp != NULL); 638 dp = argp; 639 argc = 0; 640 641 /* copy the fake args list, if there's one, freeing it as we go */ 642 if (pack.ep_flags & EXEC_HASARGL) { 643 tmpfap = pack.ep_fa; 644 while (tmpfap->fa_arg != NULL) { 645 const char *cp; 646 647 cp = tmpfap->fa_arg; 648 while (*cp) 649 *dp++ = *cp++; 650 *dp++ = '\0'; 651 ktrexecarg(tmpfap->fa_arg, cp - tmpfap->fa_arg); 652 653 kmem_free(tmpfap->fa_arg, tmpfap->fa_len); 654 tmpfap++; argc++; 655 } 656 kmem_free(pack.ep_fa, pack.ep_fa_len); 657 pack.ep_flags &= ~EXEC_HASARGL; 658 } 659 660 /* Now get argv & environment */ 661 if (args == NULL) { 662 DPRINTF(("%s: null args\n", __func__)); 663 error = EINVAL; 664 goto bad; 665 } 666 /* 'i' will index the argp/envp element to be retrieved */ 667 i = 0; 668 if (pack.ep_flags & EXEC_SKIPARG) 669 i++; 670 671 while (1) { 672 len = argp + ARG_MAX - dp; 673 if ((error = (*fetch_element)(args, i, &sp)) != 0) { 674 DPRINTF(("%s: fetch_element args %d\n", 675 __func__, error)); 676 goto bad; 677 } 678 if (!sp) 679 break; 680 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 681 DPRINTF(("%s: copyinstr args %d\n", __func__, error)); 682 if (error == ENAMETOOLONG) 683 error = E2BIG; 684 goto bad; 685 } 686 ktrexecarg(dp, len - 1); 687 dp += len; 688 i++; 689 argc++; 690 } 691 692 envc = 0; 693 /* environment need not be there */ 694 if (envs != NULL) { 695 i = 0; 696 while (1) { 697 len = argp + ARG_MAX - dp; 698 if ((error = (*fetch_element)(envs, i, &sp)) != 0) { 699 DPRINTF(("%s: fetch_element env %d\n", 700 __func__, error)); 701 goto bad; 702 } 703 if (!sp) 704 break; 705 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 706 DPRINTF(("%s: copyinstr env %d\n", 707 __func__, error)); 708 if (error == ENAMETOOLONG) 709 error = E2BIG; 710 goto bad; 711 } 712 ktrexecenv(dp, len - 1); 713 dp += len; 714 i++; 715 envc++; 716 } 717 } 718 719 dp = (char *) ALIGN(dp); 720 721 szsigcode = pack.ep_esch->es_emul->e_esigcode - 722 pack.ep_esch->es_emul->e_sigcode; 723 724 #ifdef __MACHINE_STACK_GROWS_UP 725 /* See big comment lower down */ 726 #define RTLD_GAP 32 727 #else 728 #define RTLD_GAP 0 729 #endif 730 731 /* Now check if args & environ fit into new stack */ 732 if (pack.ep_flags & EXEC_32) { 733 aip = &arginfo32; 734 ps_strings_sz = sizeof(struct ps_strings32); 735 len = ((argc + envc + 2 + pack.ep_esch->es_arglen) * 736 sizeof(int) + sizeof(int) + dp + RTLD_GAP + 737 szsigcode + ps_strings_sz + STACK_PTHREADSPACE) 738 - argp; 739 } else { 740 aip = &arginfo; 741 ps_strings_sz = sizeof(struct ps_strings); 742 len = ((argc + envc + 2 + pack.ep_esch->es_arglen) * 743 sizeof(char *) + sizeof(int) + dp + RTLD_GAP + 744 szsigcode + ps_strings_sz + STACK_PTHREADSPACE) 745 - argp; 746 } 747 748 #ifdef PAX_ASLR 749 if (pax_aslr_active(l)) 750 len += (arc4random() % PAGE_SIZE); 751 #endif /* PAX_ASLR */ 752 753 #ifdef STACKLALIGN /* arm, etc. */ 754 len = STACKALIGN(len); /* make the stack "safely" aligned */ 755 #else 756 len = ALIGN(len); /* make the stack "safely" aligned */ 757 #endif 758 759 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 760 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len)); 761 error = ENOMEM; 762 goto bad; 763 } 764 765 /* Get rid of other LWPs. */ 766 if (p->p_sa || p->p_nlwps > 1) { 767 mutex_enter(p->p_lock); 768 exit_lwps(l); 769 mutex_exit(p->p_lock); 770 } 771 KDASSERT(p->p_nlwps == 1); 772 773 /* Destroy any lwpctl info. */ 774 if (p->p_lwpctl != NULL) 775 lwp_ctl_exit(); 776 777 #ifdef KERN_SA 778 /* Release any SA state. */ 779 if (p->p_sa) 780 sa_release(p); 781 #endif /* KERN_SA */ 782 783 /* Remove POSIX timers */ 784 timers_free(p, TIMERS_POSIX); 785 786 /* adjust "active stack depth" for process VSZ */ 787 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 788 789 /* 790 * Do whatever is necessary to prepare the address space 791 * for remapping. Note that this might replace the current 792 * vmspace with another! 793 */ 794 uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 795 796 /* record proc's vnode, for use by procfs and others */ 797 if (p->p_textvp) 798 vrele(p->p_textvp); 799 vref(pack.ep_vp); 800 p->p_textvp = pack.ep_vp; 801 802 /* Now map address space */ 803 vm = p->p_vmspace; 804 vm->vm_taddr = (void *)pack.ep_taddr; 805 vm->vm_tsize = btoc(pack.ep_tsize); 806 vm->vm_daddr = (void*)pack.ep_daddr; 807 vm->vm_dsize = btoc(pack.ep_dsize); 808 vm->vm_ssize = btoc(pack.ep_ssize); 809 vm->vm_issize = 0; 810 vm->vm_maxsaddr = (void *)pack.ep_maxsaddr; 811 vm->vm_minsaddr = (void *)pack.ep_minsaddr; 812 813 #ifdef PAX_ASLR 814 pax_aslr_init(l, vm); 815 #endif /* PAX_ASLR */ 816 817 /* create the new process's VM space by running the vmcmds */ 818 #ifdef DIAGNOSTIC 819 if (pack.ep_vmcmds.evs_used == 0) 820 panic("%s: no vmcmds", __func__); 821 #endif 822 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 823 struct exec_vmcmd *vcp; 824 825 vcp = &pack.ep_vmcmds.evs_cmds[i]; 826 if (vcp->ev_flags & VMCMD_RELATIVE) { 827 #ifdef DIAGNOSTIC 828 if (base_vcp == NULL) 829 panic("%s: relative vmcmd with no base", 830 __func__); 831 if (vcp->ev_flags & VMCMD_BASE) 832 panic("%s: illegal base & relative vmcmd", 833 __func__); 834 #endif 835 vcp->ev_addr += base_vcp->ev_addr; 836 } 837 error = (*vcp->ev_proc)(l, vcp); 838 #ifdef DEBUG_EXEC 839 if (error) { 840 size_t j; 841 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 842 uprintf("vmcmds %zu/%u, error %d\n", i, 843 pack.ep_vmcmds.evs_used, error); 844 for (j = 0; j <= i; j++) 845 uprintf("vmcmd[%zu] = vmcmd_map_%s %#" 846 PRIxVADDR"/%#"PRIxVSIZE" fd@%#" 847 PRIxVSIZE" prot=0%o flags=%d\n", j, 848 vp[j].ev_proc == vmcmd_map_pagedvn ? 849 "pagedvn" : 850 vp[j].ev_proc == vmcmd_map_readvn ? 851 "readvn" : 852 vp[j].ev_proc == vmcmd_map_zero ? 853 "zero" : "*unknown*", 854 vp[j].ev_addr, vp[j].ev_len, 855 vp[j].ev_offset, vp[j].ev_prot, 856 vp[j].ev_flags); 857 } 858 #endif /* DEBUG_EXEC */ 859 if (vcp->ev_flags & VMCMD_BASE) 860 base_vcp = vcp; 861 } 862 863 /* free the vmspace-creation commands, and release their references */ 864 kill_vmcmds(&pack.ep_vmcmds); 865 866 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 867 VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred); 868 vput(pack.ep_vp); 869 870 /* if an error happened, deallocate and punt */ 871 if (error) { 872 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error)); 873 goto exec_abort; 874 } 875 876 /* remember information about the process */ 877 arginfo.ps_nargvstr = argc; 878 arginfo.ps_nenvstr = envc; 879 880 /* set command name & other accounting info */ 881 commandname = strrchr(pack.ep_resolvedname, '/'); 882 if (commandname != NULL) { 883 commandname++; 884 } else { 885 commandname = pack.ep_resolvedname; 886 } 887 i = min(strlen(commandname), MAXCOMLEN); 888 (void)memcpy(p->p_comm, commandname, i); 889 p->p_comm[i] = '\0'; 890 891 dp = PNBUF_GET(); 892 /* 893 * If the path starts with /, we don't need to do any work. 894 * This handles the majority of the cases. 895 * In the future perhaps we could canonicalize it? 896 */ 897 if (pathstring[0] == '/') 898 (void)strlcpy(pack.ep_path = dp, pathstring, MAXPATHLEN); 899 #ifdef notyet 900 /* 901 * Although this works most of the time [since the entry was just 902 * entered in the cache] we don't use it because it theoretically 903 * can fail and it is not the cleanest interface, because there 904 * could be races. When the namei cache is re-written, this can 905 * be changed to use the appropriate function. 906 */ 907 else if (!(error = vnode_to_path(dp, MAXPATHLEN, p->p_textvp, l, p))) 908 pack.ep_path = dp; 909 #endif 910 else { 911 #ifdef notyet 912 printf("Cannot get path for pid %d [%s] (error %d)", 913 (int)p->p_pid, p->p_comm, error); 914 #endif 915 pack.ep_path = NULL; 916 PNBUF_PUT(dp); 917 } 918 919 stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 920 STACK_PTHREADSPACE + ps_strings_sz + szsigcode), 921 len - (ps_strings_sz + szsigcode)); 922 923 #ifdef __MACHINE_STACK_GROWS_UP 924 /* 925 * The copyargs call always copies into lower addresses 926 * first, moving towards higher addresses, starting with 927 * the stack pointer that we give. When the stack grows 928 * down, this puts argc/argv/envp very shallow on the 929 * stack, right at the first user stack pointer. 930 * When the stack grows up, the situation is reversed. 931 * 932 * Normally, this is no big deal. But the ld_elf.so _rtld() 933 * function expects to be called with a single pointer to 934 * a region that has a few words it can stash values into, 935 * followed by argc/argv/envp. When the stack grows down, 936 * it's easy to decrement the stack pointer a little bit to 937 * allocate the space for these few words and pass the new 938 * stack pointer to _rtld. When the stack grows up, however, 939 * a few words before argc is part of the signal trampoline, XXX 940 * so we have a problem. 941 * 942 * Instead of changing how _rtld works, we take the easy way 943 * out and steal 32 bytes before we call copyargs. 944 * This extra space was allowed for when 'len' was calculated. 945 */ 946 stack += RTLD_GAP; 947 #endif /* __MACHINE_STACK_GROWS_UP */ 948 949 /* Now copy argc, args & environ to new stack */ 950 error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp); 951 if (pack.ep_path) { 952 PNBUF_PUT(pack.ep_path); 953 pack.ep_path = NULL; 954 } 955 if (error) { 956 DPRINTF(("%s: copyargs failed %d\n", __func__, error)); 957 goto exec_abort; 958 } 959 /* Move the stack back to original point */ 960 stack = (char *)STACK_GROW(vm->vm_minsaddr, len); 961 962 /* fill process ps_strings info */ 963 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 964 STACK_PTHREADSPACE), ps_strings_sz); 965 966 if (pack.ep_flags & EXEC_32) { 967 arginfo32.ps_argvstr = (vaddr_t)arginfo.ps_argvstr; 968 arginfo32.ps_nargvstr = arginfo.ps_nargvstr; 969 arginfo32.ps_envstr = (vaddr_t)arginfo.ps_envstr; 970 arginfo32.ps_nenvstr = arginfo.ps_nenvstr; 971 } 972 973 /* copy out the process's ps_strings structure */ 974 if ((error = copyout(aip, (void *)p->p_psstrp, ps_strings_sz)) != 0) { 975 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n", 976 __func__, aip, (void *)p->p_psstrp, ps_strings_sz)); 977 goto exec_abort; 978 } 979 980 cwdexec(p); 981 fd_closeexec(); /* handle close on exec */ 982 execsigs(p); /* reset catched signals */ 983 984 l->l_ctxlink = NULL; /* reset ucontext link */ 985 986 987 p->p_acflag &= ~AFORK; 988 mutex_enter(p->p_lock); 989 p->p_flag |= PK_EXEC; 990 mutex_exit(p->p_lock); 991 992 /* 993 * Stop profiling. 994 */ 995 if ((p->p_stflag & PST_PROFIL) != 0) { 996 mutex_spin_enter(&p->p_stmutex); 997 stopprofclock(p); 998 mutex_spin_exit(&p->p_stmutex); 999 } 1000 1001 /* 1002 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have 1003 * exited and exec()/exit() are the only places it will be cleared. 1004 */ 1005 if ((p->p_lflag & PL_PPWAIT) != 0) { 1006 mutex_enter(proc_lock); 1007 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 1008 p->p_lflag &= ~PL_PPWAIT; 1009 cv_broadcast(&p->p_pptr->p_waitcv); 1010 mutex_exit(proc_lock); 1011 } 1012 1013 /* 1014 * Deal with set[ug]id. MNT_NOSUID has already been used to disable 1015 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked 1016 * out additional references on the process for the moment. 1017 */ 1018 if ((p->p_slflag & PSL_TRACED) == 0 && 1019 1020 (((attr.va_mode & S_ISUID) != 0 && 1021 kauth_cred_geteuid(l->l_cred) != attr.va_uid) || 1022 1023 ((attr.va_mode & S_ISGID) != 0 && 1024 kauth_cred_getegid(l->l_cred) != attr.va_gid))) { 1025 /* 1026 * Mark the process as SUGID before we do 1027 * anything that might block. 1028 */ 1029 proc_crmod_enter(); 1030 proc_crmod_leave(NULL, NULL, true); 1031 1032 /* Make sure file descriptors 0..2 are in use. */ 1033 if ((error = fd_checkstd()) != 0) { 1034 DPRINTF(("%s: fdcheckstd failed %d\n", 1035 __func__, error)); 1036 goto exec_abort; 1037 } 1038 1039 /* 1040 * Copy the credential so other references don't see our 1041 * changes. 1042 */ 1043 l->l_cred = kauth_cred_copy(l->l_cred); 1044 #ifdef KTRACE 1045 /* 1046 * If the persistent trace flag isn't set, turn off. 1047 */ 1048 if (p->p_tracep) { 1049 mutex_enter(&ktrace_lock); 1050 if (!(p->p_traceflag & KTRFAC_PERSISTENT)) 1051 ktrderef(p); 1052 mutex_exit(&ktrace_lock); 1053 } 1054 #endif 1055 if (attr.va_mode & S_ISUID) 1056 kauth_cred_seteuid(l->l_cred, attr.va_uid); 1057 if (attr.va_mode & S_ISGID) 1058 kauth_cred_setegid(l->l_cred, attr.va_gid); 1059 } else { 1060 if (kauth_cred_geteuid(l->l_cred) == 1061 kauth_cred_getuid(l->l_cred) && 1062 kauth_cred_getegid(l->l_cred) == 1063 kauth_cred_getgid(l->l_cred)) 1064 p->p_flag &= ~PK_SUGID; 1065 } 1066 1067 /* 1068 * Copy the credential so other references don't see our changes. 1069 * Test to see if this is necessary first, since in the common case 1070 * we won't need a private reference. 1071 */ 1072 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) || 1073 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) { 1074 l->l_cred = kauth_cred_copy(l->l_cred); 1075 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred)); 1076 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred)); 1077 } 1078 1079 /* Update the master credentials. */ 1080 if (l->l_cred != p->p_cred) { 1081 kauth_cred_t ocred; 1082 1083 kauth_cred_hold(l->l_cred); 1084 mutex_enter(p->p_lock); 1085 ocred = p->p_cred; 1086 p->p_cred = l->l_cred; 1087 mutex_exit(p->p_lock); 1088 kauth_cred_free(ocred); 1089 } 1090 1091 #if defined(__HAVE_RAS) 1092 /* 1093 * Remove all RASs from the address space. 1094 */ 1095 ras_purgeall(); 1096 #endif 1097 1098 doexechooks(p); 1099 1100 /* setup new registers and do misc. setup. */ 1101 (*pack.ep_esch->es_emul->e_setregs)(l, &pack, (vaddr_t)stack); 1102 if (pack.ep_esch->es_setregs) 1103 (*pack.ep_esch->es_setregs)(l, &pack, (vaddr_t)stack); 1104 1105 /* Provide a consistent LWP private setting */ 1106 (void)lwp_setprivate(l, NULL); 1107 1108 /* map the process's signal trampoline code */ 1109 if ((error = exec_sigcode_map(p, pack.ep_esch->es_emul)) != 0) { 1110 DPRINTF(("%s: map sigcode failed %d\n", __func__, error)); 1111 goto exec_abort; 1112 } 1113 1114 pool_put(&exec_pool, argp); 1115 1116 /* notify others that we exec'd */ 1117 KNOTE(&p->p_klist, NOTE_EXEC); 1118 1119 kmem_free(pack.ep_hdr, pack.ep_hdrlen); 1120 1121 SDT_PROBE(proc,,,exec_success, path, 0, 0, 0, 0); 1122 1123 /* The emulation root will usually have been found when we looked 1124 * for the elf interpreter (or similar), if not look now. */ 1125 if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL) 1126 emul_find_root(l, &pack); 1127 1128 /* Any old emulation root got removed by fdcloseexec */ 1129 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER); 1130 p->p_cwdi->cwdi_edir = pack.ep_emul_root; 1131 rw_exit(&p->p_cwdi->cwdi_lock); 1132 pack.ep_emul_root = NULL; 1133 if (pack.ep_interp != NULL) 1134 vrele(pack.ep_interp); 1135 1136 /* 1137 * Call emulation specific exec hook. This can setup per-process 1138 * p->p_emuldata or do any other per-process stuff an emulation needs. 1139 * 1140 * If we are executing process of different emulation than the 1141 * original forked process, call e_proc_exit() of the old emulation 1142 * first, then e_proc_exec() of new emulation. If the emulation is 1143 * same, the exec hook code should deallocate any old emulation 1144 * resources held previously by this process. 1145 */ 1146 if (p->p_emul && p->p_emul->e_proc_exit 1147 && p->p_emul != pack.ep_esch->es_emul) 1148 (*p->p_emul->e_proc_exit)(p); 1149 1150 /* 1151 * This is now LWP 1. 1152 */ 1153 mutex_enter(p->p_lock); 1154 p->p_nlwpid = 1; 1155 l->l_lid = 1; 1156 mutex_exit(p->p_lock); 1157 1158 /* 1159 * Call exec hook. Emulation code may NOT store reference to anything 1160 * from &pack. 1161 */ 1162 if (pack.ep_esch->es_emul->e_proc_exec) 1163 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack); 1164 1165 /* update p_emul, the old value is no longer needed */ 1166 p->p_emul = pack.ep_esch->es_emul; 1167 1168 /* ...and the same for p_execsw */ 1169 p->p_execsw = pack.ep_esch; 1170 1171 #ifdef __HAVE_SYSCALL_INTERN 1172 (*p->p_emul->e_syscall_intern)(p); 1173 #endif 1174 ktremul(); 1175 1176 /* Allow new references from the debugger/procfs. */ 1177 rw_exit(&p->p_reflock); 1178 rw_exit(&exec_lock); 1179 1180 mutex_enter(proc_lock); 1181 1182 if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) { 1183 KSI_INIT_EMPTY(&ksi); 1184 ksi.ksi_signo = SIGTRAP; 1185 ksi.ksi_lid = l->l_lid; 1186 kpsignal(p, &ksi, NULL); 1187 } 1188 1189 if (p->p_sflag & PS_STOPEXEC) { 1190 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 1191 p->p_pptr->p_nstopchild++; 1192 p->p_pptr->p_waited = 0; 1193 mutex_enter(p->p_lock); 1194 ksiginfo_queue_init(&kq); 1195 sigclearall(p, &contsigmask, &kq); 1196 lwp_lock(l); 1197 l->l_stat = LSSTOP; 1198 p->p_stat = SSTOP; 1199 p->p_nrlwps--; 1200 lwp_unlock(l); 1201 mutex_exit(p->p_lock); 1202 mutex_exit(proc_lock); 1203 lwp_lock(l); 1204 mi_switch(l); 1205 ksiginfo_queue_drain(&kq); 1206 KERNEL_LOCK(l->l_biglocks, l); 1207 } else { 1208 mutex_exit(proc_lock); 1209 } 1210 1211 pathbuf_stringcopy_put(pb, pathstring); 1212 pathbuf_destroy(pb); 1213 PNBUF_PUT(resolvedpathbuf); 1214 return (EJUSTRETURN); 1215 1216 bad: 1217 /* free the vmspace-creation commands, and release their references */ 1218 kill_vmcmds(&pack.ep_vmcmds); 1219 /* kill any opened file descriptor, if necessary */ 1220 if (pack.ep_flags & EXEC_HASFD) { 1221 pack.ep_flags &= ~EXEC_HASFD; 1222 fd_close(pack.ep_fd); 1223 } 1224 /* close and put the exec'd file */ 1225 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 1226 VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred); 1227 vput(pack.ep_vp); 1228 pool_put(&exec_pool, argp); 1229 1230 freehdr: 1231 kmem_free(pack.ep_hdr, pack.ep_hdrlen); 1232 if (pack.ep_emul_root != NULL) 1233 vrele(pack.ep_emul_root); 1234 if (pack.ep_interp != NULL) 1235 vrele(pack.ep_interp); 1236 1237 rw_exit(&exec_lock); 1238 1239 pathbuf_stringcopy_put(pb, pathstring); 1240 pathbuf_destroy(pb); 1241 PNBUF_PUT(resolvedpathbuf); 1242 1243 clrflg: 1244 lwp_lock(l); 1245 l->l_flag |= oldlwpflags; 1246 lwp_unlock(l); 1247 rw_exit(&p->p_reflock); 1248 1249 if (modgen != module_gen && error == ENOEXEC) { 1250 modgen = module_gen; 1251 exec_autoload(); 1252 goto retry; 1253 } 1254 1255 SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0); 1256 return error; 1257 1258 exec_abort: 1259 SDT_PROBE(proc,,,exec_failure, error, 0, 0, 0, 0); 1260 rw_exit(&p->p_reflock); 1261 rw_exit(&exec_lock); 1262 1263 pathbuf_stringcopy_put(pb, pathstring); 1264 pathbuf_destroy(pb); 1265 PNBUF_PUT(resolvedpathbuf); 1266 1267 /* 1268 * the old process doesn't exist anymore. exit gracefully. 1269 * get rid of the (new) address space we have created, if any, get rid 1270 * of our namei data and vnode, and exit noting failure 1271 */ 1272 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 1273 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 1274 if (pack.ep_emul_arg) 1275 free(pack.ep_emul_arg, M_TEMP); 1276 pool_put(&exec_pool, argp); 1277 kmem_free(pack.ep_hdr, pack.ep_hdrlen); 1278 if (pack.ep_emul_root != NULL) 1279 vrele(pack.ep_emul_root); 1280 if (pack.ep_interp != NULL) 1281 vrele(pack.ep_interp); 1282 1283 /* Acquire the sched-state mutex (exit1() will release it). */ 1284 mutex_enter(p->p_lock); 1285 exit1(l, W_EXITCODE(error, SIGABRT)); 1286 1287 /* NOTREACHED */ 1288 return 0; 1289 } 1290 1291 int 1292 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo, 1293 char **stackp, void *argp) 1294 { 1295 char **cpp, *dp, *sp; 1296 size_t len; 1297 void *nullp; 1298 long argc, envc; 1299 int error; 1300 1301 cpp = (char **)*stackp; 1302 nullp = NULL; 1303 argc = arginfo->ps_nargvstr; 1304 envc = arginfo->ps_nenvstr; 1305 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) { 1306 COPYPRINTF("", cpp - 1, sizeof(argc)); 1307 return error; 1308 } 1309 1310 dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen); 1311 sp = argp; 1312 1313 /* XXX don't copy them out, remap them! */ 1314 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 1315 1316 for (; --argc >= 0; sp += len, dp += len) { 1317 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1318 COPYPRINTF("", cpp - 1, sizeof(dp)); 1319 return error; 1320 } 1321 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1322 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1323 return error; 1324 } 1325 } 1326 1327 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1328 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1329 return error; 1330 } 1331 1332 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 1333 1334 for (; --envc >= 0; sp += len, dp += len) { 1335 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1336 COPYPRINTF("", cpp - 1, sizeof(dp)); 1337 return error; 1338 } 1339 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1340 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1341 return error; 1342 } 1343 } 1344 1345 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1346 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1347 return error; 1348 } 1349 1350 *stackp = (char *)cpp; 1351 return 0; 1352 } 1353 1354 1355 /* 1356 * Add execsw[] entries. 1357 */ 1358 int 1359 exec_add(struct execsw *esp, int count) 1360 { 1361 struct exec_entry *it; 1362 int i; 1363 1364 if (count == 0) { 1365 return 0; 1366 } 1367 1368 /* Check for duplicates. */ 1369 rw_enter(&exec_lock, RW_WRITER); 1370 for (i = 0; i < count; i++) { 1371 LIST_FOREACH(it, &ex_head, ex_list) { 1372 /* assume unique (makecmds, probe_func, emulation) */ 1373 if (it->ex_sw->es_makecmds == esp[i].es_makecmds && 1374 it->ex_sw->u.elf_probe_func == 1375 esp[i].u.elf_probe_func && 1376 it->ex_sw->es_emul == esp[i].es_emul) { 1377 rw_exit(&exec_lock); 1378 return EEXIST; 1379 } 1380 } 1381 } 1382 1383 /* Allocate new entries. */ 1384 for (i = 0; i < count; i++) { 1385 it = kmem_alloc(sizeof(*it), KM_SLEEP); 1386 it->ex_sw = &esp[i]; 1387 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1388 } 1389 1390 /* update execsw[] */ 1391 exec_init(0); 1392 rw_exit(&exec_lock); 1393 return 0; 1394 } 1395 1396 /* 1397 * Remove execsw[] entry. 1398 */ 1399 int 1400 exec_remove(struct execsw *esp, int count) 1401 { 1402 struct exec_entry *it, *next; 1403 int i; 1404 const struct proclist_desc *pd; 1405 proc_t *p; 1406 1407 if (count == 0) { 1408 return 0; 1409 } 1410 1411 /* Abort if any are busy. */ 1412 rw_enter(&exec_lock, RW_WRITER); 1413 for (i = 0; i < count; i++) { 1414 mutex_enter(proc_lock); 1415 for (pd = proclists; pd->pd_list != NULL; pd++) { 1416 PROCLIST_FOREACH(p, pd->pd_list) { 1417 if (p->p_execsw == &esp[i]) { 1418 mutex_exit(proc_lock); 1419 rw_exit(&exec_lock); 1420 return EBUSY; 1421 } 1422 } 1423 } 1424 mutex_exit(proc_lock); 1425 } 1426 1427 /* None are busy, so remove them all. */ 1428 for (i = 0; i < count; i++) { 1429 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) { 1430 next = LIST_NEXT(it, ex_list); 1431 if (it->ex_sw == &esp[i]) { 1432 LIST_REMOVE(it, ex_list); 1433 kmem_free(it, sizeof(*it)); 1434 break; 1435 } 1436 } 1437 } 1438 1439 /* update execsw[] */ 1440 exec_init(0); 1441 rw_exit(&exec_lock); 1442 return 0; 1443 } 1444 1445 /* 1446 * Initialize exec structures. If init_boot is true, also does necessary 1447 * one-time initialization (it's called from main() that way). 1448 * Once system is multiuser, this should be called with exec_lock held, 1449 * i.e. via exec_{add|remove}(). 1450 */ 1451 int 1452 exec_init(int init_boot) 1453 { 1454 const struct execsw **sw; 1455 struct exec_entry *ex; 1456 SLIST_HEAD(,exec_entry) first; 1457 SLIST_HEAD(,exec_entry) any; 1458 SLIST_HEAD(,exec_entry) last; 1459 int i, sz; 1460 1461 if (init_boot) { 1462 /* do one-time initializations */ 1463 rw_init(&exec_lock); 1464 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE); 1465 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH, 1466 "execargs", &exec_palloc, IPL_NONE); 1467 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0); 1468 } else { 1469 KASSERT(rw_write_held(&exec_lock)); 1470 } 1471 1472 /* Sort each entry onto the appropriate queue. */ 1473 SLIST_INIT(&first); 1474 SLIST_INIT(&any); 1475 SLIST_INIT(&last); 1476 sz = 0; 1477 LIST_FOREACH(ex, &ex_head, ex_list) { 1478 switch(ex->ex_sw->es_prio) { 1479 case EXECSW_PRIO_FIRST: 1480 SLIST_INSERT_HEAD(&first, ex, ex_slist); 1481 break; 1482 case EXECSW_PRIO_ANY: 1483 SLIST_INSERT_HEAD(&any, ex, ex_slist); 1484 break; 1485 case EXECSW_PRIO_LAST: 1486 SLIST_INSERT_HEAD(&last, ex, ex_slist); 1487 break; 1488 default: 1489 panic("%s", __func__); 1490 break; 1491 } 1492 sz++; 1493 } 1494 1495 /* 1496 * Create new execsw[]. Ensure we do not try a zero-sized 1497 * allocation. 1498 */ 1499 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP); 1500 i = 0; 1501 SLIST_FOREACH(ex, &first, ex_slist) { 1502 sw[i++] = ex->ex_sw; 1503 } 1504 SLIST_FOREACH(ex, &any, ex_slist) { 1505 sw[i++] = ex->ex_sw; 1506 } 1507 SLIST_FOREACH(ex, &last, ex_slist) { 1508 sw[i++] = ex->ex_sw; 1509 } 1510 1511 /* Replace old execsw[] and free used memory. */ 1512 if (execsw != NULL) { 1513 kmem_free(__UNCONST(execsw), 1514 nexecs * sizeof(struct execsw *) + 1); 1515 } 1516 execsw = sw; 1517 nexecs = sz; 1518 1519 /* Figure out the maximum size of an exec header. */ 1520 exec_maxhdrsz = sizeof(int); 1521 for (i = 0; i < nexecs; i++) { 1522 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1523 exec_maxhdrsz = execsw[i]->es_hdrsz; 1524 } 1525 1526 return 0; 1527 } 1528 1529 static int 1530 exec_sigcode_map(struct proc *p, const struct emul *e) 1531 { 1532 vaddr_t va; 1533 vsize_t sz; 1534 int error; 1535 struct uvm_object *uobj; 1536 1537 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1538 1539 if (e->e_sigobject == NULL || sz == 0) { 1540 return 0; 1541 } 1542 1543 /* 1544 * If we don't have a sigobject for this emulation, create one. 1545 * 1546 * sigobject is an anonymous memory object (just like SYSV shared 1547 * memory) that we keep a permanent reference to and that we map 1548 * in all processes that need this sigcode. The creation is simple, 1549 * we create an object, add a permanent reference to it, map it in 1550 * kernel space, copy out the sigcode to it and unmap it. 1551 * We map it with PROT_READ|PROT_EXEC into the process just 1552 * the way sys_mmap() would map it. 1553 */ 1554 1555 uobj = *e->e_sigobject; 1556 if (uobj == NULL) { 1557 mutex_enter(&sigobject_lock); 1558 if ((uobj = *e->e_sigobject) == NULL) { 1559 uobj = uao_create(sz, 0); 1560 (*uobj->pgops->pgo_reference)(uobj); 1561 va = vm_map_min(kernel_map); 1562 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1563 uobj, 0, 0, 1564 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1565 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 1566 printf("kernel mapping failed %d\n", error); 1567 (*uobj->pgops->pgo_detach)(uobj); 1568 mutex_exit(&sigobject_lock); 1569 return (error); 1570 } 1571 memcpy((void *)va, e->e_sigcode, sz); 1572 #ifdef PMAP_NEED_PROCWR 1573 pmap_procwr(&proc0, va, sz); 1574 #endif 1575 uvm_unmap(kernel_map, va, va + round_page(sz)); 1576 *e->e_sigobject = uobj; 1577 } 1578 mutex_exit(&sigobject_lock); 1579 } 1580 1581 /* Just a hint to uvm_map where to put it. */ 1582 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr, 1583 round_page(sz)); 1584 1585 #ifdef __alpha__ 1586 /* 1587 * Tru64 puts /sbin/loader at the end of user virtual memory, 1588 * which causes the above calculation to put the sigcode at 1589 * an invalid address. Put it just below the text instead. 1590 */ 1591 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) { 1592 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 1593 } 1594 #endif 1595 1596 (*uobj->pgops->pgo_reference)(uobj); 1597 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 1598 uobj, 0, 0, 1599 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 1600 UVM_ADV_RANDOM, 0)); 1601 if (error) { 1602 DPRINTF(("%s, %d: map %p " 1603 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n", 1604 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz), 1605 va, error)); 1606 (*uobj->pgops->pgo_detach)(uobj); 1607 return (error); 1608 } 1609 p->p_sigctx.ps_sigcode = (void *)va; 1610 return (0); 1611 } 1612