1 /* $NetBSD: kern_exec.c,v 1.247 2007/08/15 12:07:32 ad Exp $ */ 2 3 /*- 4 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou 5 * Copyright (C) 1992 Wolfgang Solfrank. 6 * Copyright (C) 1992 TooLs GmbH. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by TooLs GmbH. 20 * 4. The name of TooLs GmbH may not be used to endorse or promote products 21 * derived from this software without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 26 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 28 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 29 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 30 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 31 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 32 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.247 2007/08/15 12:07:32 ad Exp $"); 37 38 #include "opt_ktrace.h" 39 #include "opt_syscall_debug.h" 40 #include "opt_compat_netbsd.h" 41 #include "veriexec.h" 42 #include "opt_pax.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/filedesc.h> 47 #include <sys/kernel.h> 48 #include <sys/proc.h> 49 #include <sys/mount.h> 50 #include <sys/malloc.h> 51 #include <sys/namei.h> 52 #include <sys/vnode.h> 53 #include <sys/file.h> 54 #include <sys/acct.h> 55 #include <sys/exec.h> 56 #include <sys/ktrace.h> 57 #include <sys/resourcevar.h> 58 #include <sys/wait.h> 59 #include <sys/mman.h> 60 #include <sys/ras.h> 61 #include <sys/signalvar.h> 62 #include <sys/stat.h> 63 #include <sys/syscall.h> 64 #include <sys/kauth.h> 65 66 #include <sys/syscallargs.h> 67 #if NVERIEXEC > 0 68 #include <sys/verified_exec.h> 69 #endif /* NVERIEXEC > 0 */ 70 71 #ifdef SYSTRACE 72 #include <sys/systrace.h> 73 #endif /* SYSTRACE */ 74 75 #ifdef PAX_SEGVGUARD 76 #include <sys/pax.h> 77 #endif /* PAX_SEGVGUARD */ 78 79 #include <uvm/uvm_extern.h> 80 81 #include <machine/cpu.h> 82 #include <machine/reg.h> 83 84 #include <compat/common/compat_util.h> 85 86 static int exec_sigcode_map(struct proc *, const struct emul *); 87 88 #ifdef DEBUG_EXEC 89 #define DPRINTF(a) uprintf a 90 #else 91 #define DPRINTF(a) 92 #endif /* DEBUG_EXEC */ 93 94 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec"); 95 96 /* 97 * Exec function switch: 98 * 99 * Note that each makecmds function is responsible for loading the 100 * exec package with the necessary functions for any exec-type-specific 101 * handling. 102 * 103 * Functions for specific exec types should be defined in their own 104 * header file. 105 */ 106 extern const struct execsw execsw_builtin[]; 107 extern int nexecs_builtin; 108 static const struct execsw **execsw = NULL; 109 static int nexecs; 110 111 u_int exec_maxhdrsz; /* must not be static - netbsd32 needs it */ 112 113 #ifdef LKM 114 /* list of supported emulations */ 115 static 116 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head); 117 struct emul_entry { 118 LIST_ENTRY(emul_entry) el_list; 119 const struct emul *el_emul; 120 int ro_entry; 121 }; 122 123 /* list of dynamically loaded execsw entries */ 124 static 125 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head); 126 struct exec_entry { 127 LIST_ENTRY(exec_entry) ex_list; 128 const struct execsw *es; 129 }; 130 131 /* structure used for building execw[] */ 132 struct execsw_entry { 133 struct execsw_entry *next; 134 const struct execsw *es; 135 }; 136 #endif /* LKM */ 137 138 #ifdef SYSCALL_DEBUG 139 extern const char * const syscallnames[]; 140 #endif 141 142 #ifdef COMPAT_16 143 extern char sigcode[], esigcode[]; 144 struct uvm_object *emul_netbsd_object; 145 #endif 146 147 #ifndef __HAVE_SYSCALL_INTERN 148 void syscall(void); 149 #endif 150 151 /* NetBSD emul struct */ 152 const struct emul emul_netbsd = { 153 "netbsd", 154 NULL, /* emulation path */ 155 #ifndef __HAVE_MINIMAL_EMUL 156 EMUL_HAS_SYS___syscall, 157 NULL, 158 SYS_syscall, 159 SYS_NSYSENT, 160 #endif 161 sysent, 162 #ifdef SYSCALL_DEBUG 163 syscallnames, 164 #else 165 NULL, 166 #endif 167 sendsig, 168 trapsignal, 169 NULL, 170 #ifdef COMPAT_16 171 sigcode, 172 esigcode, 173 &emul_netbsd_object, 174 #else 175 NULL, 176 NULL, 177 NULL, 178 #endif 179 setregs, 180 NULL, 181 NULL, 182 NULL, 183 NULL, 184 NULL, 185 #ifdef __HAVE_SYSCALL_INTERN 186 syscall_intern, 187 #else 188 syscall, 189 #endif 190 NULL, 191 NULL, 192 193 uvm_default_mapaddr, 194 NULL, 195 sizeof(ucontext_t), 196 startlwp, 197 }; 198 199 #ifdef LKM 200 /* 201 * Exec lock. Used to control access to execsw[] structures. 202 * This must not be static so that netbsd32 can access it, too. 203 */ 204 krwlock_t exec_lock; 205 206 static void link_es(struct execsw_entry **, const struct execsw *); 207 #endif /* LKM */ 208 209 /* 210 * check exec: 211 * given an "executable" described in the exec package's namei info, 212 * see what we can do with it. 213 * 214 * ON ENTRY: 215 * exec package with appropriate namei info 216 * lwp pointer of exec'ing lwp 217 * NO SELF-LOCKED VNODES 218 * 219 * ON EXIT: 220 * error: nothing held, etc. exec header still allocated. 221 * ok: filled exec package, executable's vnode (unlocked). 222 * 223 * EXEC SWITCH ENTRY: 224 * Locked vnode to check, exec package, proc. 225 * 226 * EXEC SWITCH EXIT: 227 * ok: return 0, filled exec package, executable's vnode (unlocked). 228 * error: destructive: 229 * everything deallocated execept exec header. 230 * non-destructive: 231 * error code, executable's vnode (unlocked), 232 * exec header unmodified. 233 */ 234 int 235 /*ARGSUSED*/ 236 check_exec(struct lwp *l, struct exec_package *epp) 237 { 238 int error, i; 239 struct vnode *vp; 240 struct nameidata *ndp; 241 size_t resid; 242 243 ndp = epp->ep_ndp; 244 ndp->ni_cnd.cn_nameiop = LOOKUP; 245 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME | TRYEMULROOT; 246 /* first get the vnode */ 247 if ((error = namei(ndp)) != 0) 248 return error; 249 epp->ep_vp = vp = ndp->ni_vp; 250 251 /* check access and type */ 252 if (vp->v_type != VREG) { 253 error = EACCES; 254 goto bad1; 255 } 256 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred, l)) != 0) 257 goto bad1; 258 259 /* get attributes */ 260 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred, l)) != 0) 261 goto bad1; 262 263 /* Check mount point */ 264 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 265 error = EACCES; 266 goto bad1; 267 } 268 if (vp->v_mount->mnt_flag & MNT_NOSUID) 269 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 270 271 /* try to open it */ 272 if ((error = VOP_OPEN(vp, FREAD, l->l_cred, l)) != 0) 273 goto bad1; 274 275 /* unlock vp, since we need it unlocked from here on out. */ 276 VOP_UNLOCK(vp, 0); 277 278 #if NVERIEXEC > 0 279 error = veriexec_verify(l, vp, ndp->ni_cnd.cn_pnbuf, 280 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT, 281 NULL); 282 if (error) 283 goto bad2; 284 #endif /* NVERIEXEC > 0 */ 285 286 #ifdef PAX_SEGVGUARD 287 error = pax_segvguard(l, vp, ndp->ni_cnd.cn_pnbuf, false); 288 if (error) 289 goto bad2; 290 #endif /* PAX_SEGVGUARD */ 291 292 /* now we have the file, get the exec header */ 293 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 294 UIO_SYSSPACE, 0, l->l_cred, &resid, NULL); 295 if (error) 296 goto bad2; 297 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 298 299 /* 300 * Set up default address space limits. Can be overridden 301 * by individual exec packages. 302 * 303 * XXX probably should be all done in the exec packages. 304 */ 305 epp->ep_vm_minaddr = VM_MIN_ADDRESS; 306 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 307 /* 308 * set up the vmcmds for creation of the process 309 * address space 310 */ 311 error = ENOEXEC; 312 for (i = 0; i < nexecs; i++) { 313 int newerror; 314 315 epp->ep_esch = execsw[i]; 316 newerror = (*execsw[i]->es_makecmds)(l, epp); 317 318 if (!newerror) { 319 /* Seems ok: check that entry point is sane */ 320 if (epp->ep_entry > VM_MAXUSER_ADDRESS) { 321 error = ENOEXEC; 322 break; 323 } 324 325 /* check limits */ 326 if ((epp->ep_tsize > MAXTSIZ) || 327 (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit 328 [RLIMIT_DATA].rlim_cur)) { 329 error = ENOMEM; 330 break; 331 } 332 return 0; 333 } 334 335 if (epp->ep_emul_root != NULL) { 336 vrele(epp->ep_emul_root); 337 epp->ep_emul_root = NULL; 338 } 339 if (epp->ep_interp != NULL) { 340 vrele(epp->ep_interp); 341 epp->ep_interp = NULL; 342 } 343 344 /* make sure the first "interesting" error code is saved. */ 345 if (error == ENOEXEC) 346 error = newerror; 347 348 if (epp->ep_flags & EXEC_DESTR) 349 /* Error from "#!" code, tidied up by recursive call */ 350 return error; 351 } 352 353 /* 354 * free any vmspace-creation commands, 355 * and release their references 356 */ 357 kill_vmcmds(&epp->ep_vmcmds); 358 359 bad2: 360 /* 361 * close and release the vnode, restore the old one, free the 362 * pathname buf, and punt. 363 */ 364 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 365 VOP_CLOSE(vp, FREAD, l->l_cred, l); 366 vput(vp); 367 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 368 return error; 369 370 bad1: 371 /* 372 * free the namei pathname buffer, and put the vnode 373 * (which we don't yet have open). 374 */ 375 vput(vp); /* was still locked */ 376 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 377 return error; 378 } 379 380 #ifdef __MACHINE_STACK_GROWS_UP 381 #define STACK_PTHREADSPACE NBPG 382 #else 383 #define STACK_PTHREADSPACE 0 384 #endif 385 386 static int 387 execve_fetch_element(char * const *array, size_t index, char **value) 388 { 389 return copyin(array + index, value, sizeof(*value)); 390 } 391 392 /* 393 * exec system call 394 */ 395 /* ARGSUSED */ 396 int 397 sys_execve(struct lwp *l, void *v, register_t *retval) 398 { 399 struct sys_execve_args /* { 400 syscallarg(const char *) path; 401 syscallarg(char * const *) argp; 402 syscallarg(char * const *) envp; 403 } */ *uap = v; 404 405 return execve1(l, SCARG(uap, path), SCARG(uap, argp), 406 SCARG(uap, envp), execve_fetch_element); 407 } 408 409 int 410 execve1(struct lwp *l, const char *path, char * const *args, 411 char * const *envs, execve_fetch_element_t fetch_element) 412 { 413 int error; 414 u_int i; 415 struct exec_package pack; 416 struct nameidata nid; 417 struct vattr attr; 418 struct proc *p; 419 char *argp; 420 char *dp, *sp; 421 long argc, envc; 422 size_t len; 423 char *stack; 424 struct ps_strings arginfo; 425 struct ps_strings *aip = &arginfo; 426 struct vmspace *vm; 427 char **tmpfap; 428 int szsigcode; 429 struct exec_vmcmd *base_vcp; 430 ksiginfo_t ksi; 431 ksiginfoq_t kq; 432 #ifdef SYSTRACE 433 int wassugid = ISSET(p->p_flag, PK_SUGID); 434 char pathbuf[MAXPATHLEN]; 435 size_t pathbuflen; 436 #endif /* SYSTRACE */ 437 438 p = l->l_proc; 439 440 /* 441 * Drain existing references and forbid new ones. The process 442 * should be left alone until we're done here. This is necessary 443 * to avoid race conditions - e.g. in ptrace() - that might allow 444 * a local user to illicitly obtain elevated privileges. 445 */ 446 mutex_enter(&p->p_mutex); 447 proc_drainrefs(p); 448 mutex_exit(&p->p_mutex); 449 450 base_vcp = NULL; 451 /* 452 * Init the namei data to point the file user's program name. 453 * This is done here rather than in check_exec(), so that it's 454 * possible to override this settings if any of makecmd/probe 455 * functions call check_exec() recursively - for example, 456 * see exec_script_makecmds(). 457 */ 458 #ifdef SYSTRACE 459 if (ISSET(p->p_flag, PK_SYSTRACE)) 460 systrace_execve0(p); 461 462 error = copyinstr(path, pathbuf, sizeof(pathbuf), 463 &pathbuflen); 464 if (error) 465 goto clrflg; 466 467 NDINIT(&nid, LOOKUP, NOFOLLOW | TRYEMULROOT, UIO_SYSSPACE, pathbuf, l); 468 #else 469 NDINIT(&nid, LOOKUP, NOFOLLOW | TRYEMULROOT, UIO_USERSPACE, path, l); 470 #endif /* SYSTRACE */ 471 472 /* 473 * initialize the fields of the exec package. 474 */ 475 #ifdef SYSTRACE 476 pack.ep_name = pathbuf; 477 #else 478 pack.ep_name = path; 479 #endif /* SYSTRACE */ 480 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 481 pack.ep_hdrlen = exec_maxhdrsz; 482 pack.ep_hdrvalid = 0; 483 pack.ep_ndp = &nid; 484 pack.ep_emul_arg = NULL; 485 pack.ep_vmcmds.evs_cnt = 0; 486 pack.ep_vmcmds.evs_used = 0; 487 pack.ep_vap = &attr; 488 pack.ep_flags = 0; 489 pack.ep_emul_root = NULL; 490 pack.ep_interp = NULL; 491 pack.ep_esch = NULL; 492 493 #ifdef LKM 494 rw_enter(&exec_lock, RW_READER); 495 #endif 496 497 /* see if we can run it. */ 498 if ((error = check_exec(l, &pack)) != 0) 499 goto freehdr; 500 501 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 502 503 /* allocate an argument buffer */ 504 argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0, 505 UVM_KMF_PAGEABLE|UVM_KMF_WAITVA); 506 #ifdef DIAGNOSTIC 507 if (argp == NULL) 508 panic("execve: argp == NULL"); 509 #endif 510 dp = argp; 511 argc = 0; 512 513 /* copy the fake args list, if there's one, freeing it as we go */ 514 if (pack.ep_flags & EXEC_HASARGL) { 515 tmpfap = pack.ep_fa; 516 while (*tmpfap != NULL) { 517 char *cp; 518 519 cp = *tmpfap; 520 while (*cp) 521 *dp++ = *cp++; 522 dp++; 523 524 FREE(*tmpfap, M_EXEC); 525 tmpfap++; argc++; 526 } 527 FREE(pack.ep_fa, M_EXEC); 528 pack.ep_flags &= ~EXEC_HASARGL; 529 } 530 531 /* Now get argv & environment */ 532 if (args == NULL) { 533 error = EINVAL; 534 goto bad; 535 } 536 /* 'i' will index the argp/envp element to be retrieved */ 537 i = 0; 538 if (pack.ep_flags & EXEC_SKIPARG) 539 i++; 540 541 while (1) { 542 len = argp + ARG_MAX - dp; 543 if ((error = (*fetch_element)(args, i, &sp)) != 0) 544 goto bad; 545 if (!sp) 546 break; 547 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 548 if (error == ENAMETOOLONG) 549 error = E2BIG; 550 goto bad; 551 } 552 ktrexecarg(dp, len - 1); 553 dp += len; 554 i++; 555 argc++; 556 } 557 558 envc = 0; 559 /* environment need not be there */ 560 if (envs != NULL) { 561 i = 0; 562 while (1) { 563 len = argp + ARG_MAX - dp; 564 if ((error = (*fetch_element)(envs, i, &sp)) != 0) 565 goto bad; 566 if (!sp) 567 break; 568 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 569 if (error == ENAMETOOLONG) 570 error = E2BIG; 571 goto bad; 572 } 573 ktrexecenv(dp, len - 1); 574 dp += len; 575 i++; 576 envc++; 577 } 578 } 579 580 dp = (char *) ALIGN(dp); 581 582 szsigcode = pack.ep_esch->es_emul->e_esigcode - 583 pack.ep_esch->es_emul->e_sigcode; 584 585 /* Now check if args & environ fit into new stack */ 586 if (pack.ep_flags & EXEC_32) 587 len = ((argc + envc + 2 + pack.ep_esch->es_arglen) * 588 sizeof(int) + sizeof(int) + dp + STACKGAPLEN + 589 szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE) 590 - argp; 591 else 592 len = ((argc + envc + 2 + pack.ep_esch->es_arglen) * 593 sizeof(char *) + sizeof(int) + dp + STACKGAPLEN + 594 szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE) 595 - argp; 596 597 #ifdef STACKLALIGN /* arm, etc. */ 598 len = STACKALIGN(len); /* make the stack "safely" aligned */ 599 #else 600 len = ALIGN(len); /* make the stack "safely" aligned */ 601 #endif 602 603 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 604 error = ENOMEM; 605 goto bad; 606 } 607 608 /* Get rid of other LWPs. */ 609 if (p->p_nlwps > 1) { 610 mutex_enter(&p->p_smutex); 611 exit_lwps(l); 612 mutex_exit(&p->p_smutex); 613 } 614 KDASSERT(p->p_nlwps == 1); 615 616 /* This is now LWP 1 */ 617 l->l_lid = 1; 618 p->p_nlwpid = 1; 619 620 /* Remove POSIX timers */ 621 timers_free(p, TIMERS_POSIX); 622 623 /* adjust "active stack depth" for process VSZ */ 624 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 625 626 /* 627 * Do whatever is necessary to prepare the address space 628 * for remapping. Note that this might replace the current 629 * vmspace with another! 630 */ 631 uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 632 633 /* record proc's vnode, for use by procfs and others */ 634 if (p->p_textvp) 635 vrele(p->p_textvp); 636 VREF(pack.ep_vp); 637 p->p_textvp = pack.ep_vp; 638 639 /* Now map address space */ 640 vm = p->p_vmspace; 641 vm->vm_taddr = (void *)pack.ep_taddr; 642 vm->vm_tsize = btoc(pack.ep_tsize); 643 vm->vm_daddr = (void*)pack.ep_daddr; 644 vm->vm_dsize = btoc(pack.ep_dsize); 645 vm->vm_ssize = btoc(pack.ep_ssize); 646 vm->vm_maxsaddr = (void *)pack.ep_maxsaddr; 647 vm->vm_minsaddr = (void *)pack.ep_minsaddr; 648 649 /* create the new process's VM space by running the vmcmds */ 650 #ifdef DIAGNOSTIC 651 if (pack.ep_vmcmds.evs_used == 0) 652 panic("execve: no vmcmds"); 653 #endif 654 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 655 struct exec_vmcmd *vcp; 656 657 vcp = &pack.ep_vmcmds.evs_cmds[i]; 658 if (vcp->ev_flags & VMCMD_RELATIVE) { 659 #ifdef DIAGNOSTIC 660 if (base_vcp == NULL) 661 panic("execve: relative vmcmd with no base"); 662 if (vcp->ev_flags & VMCMD_BASE) 663 panic("execve: illegal base & relative vmcmd"); 664 #endif 665 vcp->ev_addr += base_vcp->ev_addr; 666 } 667 error = (*vcp->ev_proc)(l, vcp); 668 #ifdef DEBUG_EXEC 669 if (error) { 670 int j; 671 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 672 for (j = 0; j <= i; j++) 673 uprintf( 674 "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n", 675 j, vp[j].ev_addr, vp[j].ev_len, 676 vp[j].ev_offset, vp[j].ev_prot, 677 vp[j].ev_flags); 678 } 679 #endif /* DEBUG_EXEC */ 680 if (vcp->ev_flags & VMCMD_BASE) 681 base_vcp = vcp; 682 } 683 684 /* free the vmspace-creation commands, and release their references */ 685 kill_vmcmds(&pack.ep_vmcmds); 686 687 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 688 VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred, l); 689 vput(pack.ep_vp); 690 691 /* if an error happened, deallocate and punt */ 692 if (error) { 693 DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error)); 694 goto exec_abort; 695 } 696 697 /* remember information about the process */ 698 arginfo.ps_nargvstr = argc; 699 arginfo.ps_nenvstr = envc; 700 701 stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 702 STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode), 703 len - (sizeof(struct ps_strings) + szsigcode)); 704 #ifdef __MACHINE_STACK_GROWS_UP 705 /* 706 * The copyargs call always copies into lower addresses 707 * first, moving towards higher addresses, starting with 708 * the stack pointer that we give. When the stack grows 709 * down, this puts argc/argv/envp very shallow on the 710 * stack, right at the first user stack pointer, and puts 711 * STACKGAPLEN very deep in the stack. When the stack 712 * grows up, the situation is reversed. 713 * 714 * Normally, this is no big deal. But the ld_elf.so _rtld() 715 * function expects to be called with a single pointer to 716 * a region that has a few words it can stash values into, 717 * followed by argc/argv/envp. When the stack grows down, 718 * it's easy to decrement the stack pointer a little bit to 719 * allocate the space for these few words and pass the new 720 * stack pointer to _rtld. When the stack grows up, however, 721 * a few words before argc is part of the signal trampoline, XXX 722 * so we have a problem. 723 * 724 * Instead of changing how _rtld works, we take the easy way 725 * out and steal 32 bytes before we call copyargs. This 726 * space is effectively stolen from STACKGAPLEN. 727 */ 728 stack += 32; 729 #endif /* __MACHINE_STACK_GROWS_UP */ 730 731 /* Now copy argc, args & environ to new stack */ 732 error = (*pack.ep_esch->es_copyargs)(l, &pack, &arginfo, &stack, argp); 733 if (error) { 734 DPRINTF(("execve: copyargs failed %d\n", error)); 735 goto exec_abort; 736 } 737 /* Move the stack back to original point */ 738 stack = (char *)STACK_GROW(vm->vm_minsaddr, len); 739 740 /* fill process ps_strings info */ 741 p->p_psstr = (struct ps_strings *) 742 STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE), 743 sizeof(struct ps_strings)); 744 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 745 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 746 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 747 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 748 749 /* copy out the process's ps_strings structure */ 750 if ((error = copyout(aip, (char *)p->p_psstr, 751 sizeof(arginfo))) != 0) { 752 DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n", 753 aip, (char *)p->p_psstr, (long)sizeof(arginfo))); 754 goto exec_abort; 755 } 756 757 fdcloseexec(l); /* handle close on exec */ 758 execsigs(p); /* reset catched signals */ 759 760 l->l_ctxlink = NULL; /* reset ucontext link */ 761 762 /* set command name & other accounting info */ 763 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 764 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 765 p->p_comm[len] = 0; 766 p->p_acflag &= ~AFORK; 767 768 p->p_flag |= PK_EXEC; 769 770 /* 771 * Stop profiling. 772 */ 773 if ((p->p_stflag & PST_PROFIL) != 0) { 774 mutex_spin_enter(&p->p_stmutex); 775 stopprofclock(p); 776 mutex_spin_exit(&p->p_stmutex); 777 } 778 779 /* 780 * It's OK to test PS_PPWAIT unlocked here, as other LWPs have 781 * exited and exec()/exit() are the only places it will be cleared. 782 */ 783 if ((p->p_sflag & PS_PPWAIT) != 0) { 784 mutex_enter(&proclist_lock); 785 mutex_enter(&p->p_smutex); 786 p->p_sflag &= ~PS_PPWAIT; 787 cv_broadcast(&p->p_pptr->p_waitcv); 788 mutex_exit(&p->p_smutex); 789 mutex_exit(&proclist_lock); 790 } 791 792 /* 793 * Deal with set[ug]id. MNT_NOSUID has already been used to disable 794 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked 795 * out additional references on the process for the moment. 796 */ 797 if ((p->p_slflag & PSL_TRACED) == 0 && 798 799 (((attr.va_mode & S_ISUID) != 0 && 800 kauth_cred_geteuid(l->l_cred) != attr.va_uid) || 801 802 ((attr.va_mode & S_ISGID) != 0 && 803 kauth_cred_getegid(l->l_cred) != attr.va_gid))) { 804 /* 805 * Mark the process as SUGID before we do 806 * anything that might block. 807 */ 808 proc_crmod_enter(); 809 proc_crmod_leave(NULL, NULL, true); 810 811 /* Make sure file descriptors 0..2 are in use. */ 812 if ((error = fdcheckstd(l)) != 0) { 813 DPRINTF(("execve: fdcheckstd failed %d\n", error)); 814 goto exec_abort; 815 } 816 817 /* 818 * Copy the credential so other references don't see our 819 * changes. 820 */ 821 l->l_cred = kauth_cred_copy(l->l_cred); 822 #ifdef KTRACE 823 /* 824 * If process is being ktraced, turn off - unless 825 * root set it. 826 */ 827 if (p->p_tracep) { 828 mutex_enter(&ktrace_lock); 829 if (!(p->p_traceflag & KTRFAC_ROOT)) 830 ktrderef(p); 831 mutex_exit(&ktrace_lock); 832 } 833 #endif 834 if (attr.va_mode & S_ISUID) 835 kauth_cred_seteuid(l->l_cred, attr.va_uid); 836 if (attr.va_mode & S_ISGID) 837 kauth_cred_setegid(l->l_cred, attr.va_gid); 838 } else { 839 if (kauth_cred_geteuid(l->l_cred) == 840 kauth_cred_getuid(l->l_cred) && 841 kauth_cred_getegid(l->l_cred) == 842 kauth_cred_getgid(l->l_cred)) 843 p->p_flag &= ~PK_SUGID; 844 } 845 846 /* 847 * Copy the credential so other references don't see our changes. 848 * Test to see if this is necessary first, since in the common case 849 * we won't need a private reference. 850 */ 851 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) || 852 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) { 853 l->l_cred = kauth_cred_copy(l->l_cred); 854 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred)); 855 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred)); 856 } 857 858 /* Update the master credentials. */ 859 if (l->l_cred != p->p_cred) { 860 kauth_cred_t ocred; 861 862 kauth_cred_hold(l->l_cred); 863 mutex_enter(&p->p_mutex); 864 ocred = p->p_cred; 865 p->p_cred = l->l_cred; 866 mutex_exit(&p->p_mutex); 867 kauth_cred_free(ocred); 868 } 869 870 #if defined(__HAVE_RAS) 871 /* 872 * Remove all RASs from the address space. 873 */ 874 ras_purgeall(p); 875 #endif 876 877 doexechooks(p); 878 879 uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE); 880 881 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 882 883 /* notify others that we exec'd */ 884 KNOTE(&p->p_klist, NOTE_EXEC); 885 886 /* setup new registers and do misc. setup. */ 887 (*pack.ep_esch->es_emul->e_setregs)(l, &pack, (u_long) stack); 888 if (pack.ep_esch->es_setregs) 889 (*pack.ep_esch->es_setregs)(l, &pack, (u_long) stack); 890 891 /* map the process's signal trampoline code */ 892 if (exec_sigcode_map(p, pack.ep_esch->es_emul)) { 893 DPRINTF(("execve: map sigcode failed %d\n", error)); 894 goto exec_abort; 895 } 896 897 free(pack.ep_hdr, M_EXEC); 898 899 /* The emulation root will usually have been found when we looked 900 * for the elf interpreter (or similar), if not look now. */ 901 if (pack.ep_esch->es_emul->e_path != NULL && pack.ep_emul_root == NULL) 902 emul_find_root(l, &pack); 903 904 /* Any old emulation root got removed by fdcloseexec */ 905 p->p_cwdi->cwdi_edir = pack.ep_emul_root; 906 pack.ep_emul_root = NULL; 907 if (pack.ep_interp != NULL) 908 vrele(pack.ep_interp); 909 910 /* 911 * Call emulation specific exec hook. This can setup per-process 912 * p->p_emuldata or do any other per-process stuff an emulation needs. 913 * 914 * If we are executing process of different emulation than the 915 * original forked process, call e_proc_exit() of the old emulation 916 * first, then e_proc_exec() of new emulation. If the emulation is 917 * same, the exec hook code should deallocate any old emulation 918 * resources held previously by this process. 919 */ 920 if (p->p_emul && p->p_emul->e_proc_exit 921 && p->p_emul != pack.ep_esch->es_emul) 922 (*p->p_emul->e_proc_exit)(p); 923 924 /* 925 * Call exec hook. Emulation code may NOT store reference to anything 926 * from &pack. 927 */ 928 if (pack.ep_esch->es_emul->e_proc_exec) 929 (*pack.ep_esch->es_emul->e_proc_exec)(p, &pack); 930 931 /* update p_emul, the old value is no longer needed */ 932 p->p_emul = pack.ep_esch->es_emul; 933 934 /* ...and the same for p_execsw */ 935 p->p_execsw = pack.ep_esch; 936 937 #ifdef __HAVE_SYSCALL_INTERN 938 (*p->p_emul->e_syscall_intern)(p); 939 #endif 940 ktremul(); 941 942 #ifdef LKM 943 rw_exit(&exec_lock); 944 #endif 945 946 mutex_enter(&proclist_mutex); 947 948 if ((p->p_slflag & (PSL_TRACED|PSL_SYSCALL)) == PSL_TRACED) { 949 KSI_INIT_EMPTY(&ksi); 950 ksi.ksi_signo = SIGTRAP; 951 ksi.ksi_lid = l->l_lid; 952 kpsignal(p, &ksi, NULL); 953 } 954 955 if (p->p_sflag & PS_STOPEXEC) { 956 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 957 p->p_pptr->p_nstopchild++; 958 p->p_pptr->p_waited = 0; 959 mutex_enter(&p->p_smutex); 960 ksiginfo_queue_init(&kq); 961 sigclearall(p, &contsigmask, &kq); 962 lwp_lock(l); 963 p->p_refcnt = 1; 964 l->l_stat = LSSTOP; 965 p->p_stat = SSTOP; 966 p->p_nrlwps--; 967 mutex_exit(&p->p_smutex); 968 mutex_exit(&proclist_mutex); 969 mi_switch(l); 970 ksiginfo_queue_drain(&kq); 971 KERNEL_LOCK(l->l_biglocks, l); 972 } else { 973 mutex_exit(&proclist_mutex); 974 975 /* Unlock the process. */ 976 mb_write(); 977 p->p_refcnt = 1; 978 } 979 980 #ifdef SYSTRACE 981 /* XXXSMP */ 982 if (ISSET(p->p_flag, PK_SYSTRACE) && 983 wassugid && !ISSET(p->p_flag, PK_SUGID)) 984 systrace_execve1(pathbuf, p); 985 #endif /* SYSTRACE */ 986 987 return (EJUSTRETURN); 988 989 bad: 990 /* free the vmspace-creation commands, and release their references */ 991 kill_vmcmds(&pack.ep_vmcmds); 992 /* kill any opened file descriptor, if necessary */ 993 if (pack.ep_flags & EXEC_HASFD) { 994 pack.ep_flags &= ~EXEC_HASFD; 995 (void) fdrelease(l, pack.ep_fd); 996 } 997 /* close and put the exec'd file */ 998 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 999 VOP_CLOSE(pack.ep_vp, FREAD, l->l_cred, l); 1000 vput(pack.ep_vp); 1001 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 1002 uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE); 1003 1004 freehdr: 1005 free(pack.ep_hdr, M_EXEC); 1006 if (pack.ep_emul_root != NULL) 1007 vrele(pack.ep_emul_root); 1008 if (pack.ep_interp != NULL) 1009 vrele(pack.ep_interp); 1010 1011 #ifdef SYSTRACE 1012 clrflg: 1013 #endif /* SYSTRACE */ 1014 /* Unlock the process. */ 1015 mb_write(); 1016 p->p_refcnt = 1; 1017 1018 #ifdef LKM 1019 rw_exit(&exec_lock); 1020 #endif 1021 1022 return error; 1023 1024 exec_abort: 1025 #ifdef LKM 1026 rw_exit(&exec_lock); 1027 #endif 1028 1029 /* 1030 * the old process doesn't exist anymore. exit gracefully. 1031 * get rid of the (new) address space we have created, if any, get rid 1032 * of our namei data and vnode, and exit noting failure 1033 */ 1034 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 1035 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 1036 if (pack.ep_emul_arg) 1037 FREE(pack.ep_emul_arg, M_TEMP); 1038 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 1039 uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE); 1040 free(pack.ep_hdr, M_EXEC); 1041 if (pack.ep_emul_root != NULL) 1042 vrele(pack.ep_emul_root); 1043 if (pack.ep_interp != NULL) 1044 vrele(pack.ep_interp); 1045 1046 /* 1047 * Acquire the sched-state mutex (exit1() will release it). Since 1048 * this is a failed exec and we are exiting, keep the process locked 1049 * (p->p_refcnt == 0) through exit1(). 1050 */ 1051 mutex_enter(&p->p_smutex); 1052 exit1(l, W_EXITCODE(error, SIGABRT)); 1053 1054 /* NOTREACHED */ 1055 return 0; 1056 } 1057 1058 1059 int 1060 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo, 1061 char **stackp, void *argp) 1062 { 1063 char **cpp, *dp, *sp; 1064 size_t len; 1065 void *nullp; 1066 long argc, envc; 1067 int error; 1068 1069 cpp = (char **)*stackp; 1070 nullp = NULL; 1071 argc = arginfo->ps_nargvstr; 1072 envc = arginfo->ps_nenvstr; 1073 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) 1074 return error; 1075 1076 dp = (char *) (cpp + argc + envc + 2 + pack->ep_esch->es_arglen); 1077 sp = argp; 1078 1079 /* XXX don't copy them out, remap them! */ 1080 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 1081 1082 for (; --argc >= 0; sp += len, dp += len) 1083 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 1084 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 1085 return error; 1086 1087 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 1088 return error; 1089 1090 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 1091 1092 for (; --envc >= 0; sp += len, dp += len) 1093 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 1094 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 1095 return error; 1096 1097 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 1098 return error; 1099 1100 *stackp = (char *)cpp; 1101 return 0; 1102 } 1103 1104 #ifdef LKM 1105 /* 1106 * Find an emulation of given name in list of emulations. 1107 * Needs to be called with the exec_lock held. 1108 */ 1109 const struct emul * 1110 emul_search(const char *name) 1111 { 1112 struct emul_entry *it; 1113 1114 LIST_FOREACH(it, &el_head, el_list) { 1115 if (strcmp(name, it->el_emul->e_name) == 0) 1116 return it->el_emul; 1117 } 1118 1119 return NULL; 1120 } 1121 1122 /* 1123 * Add an emulation to list, if it's not there already. 1124 */ 1125 int 1126 emul_register(const struct emul *emul, int ro_entry) 1127 { 1128 struct emul_entry *ee; 1129 int error; 1130 1131 error = 0; 1132 rw_enter(&exec_lock, RW_WRITER); 1133 1134 if (emul_search(emul->e_name)) { 1135 error = EEXIST; 1136 goto out; 1137 } 1138 1139 MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry), 1140 M_EXEC, M_WAITOK); 1141 ee->el_emul = emul; 1142 ee->ro_entry = ro_entry; 1143 LIST_INSERT_HEAD(&el_head, ee, el_list); 1144 1145 out: 1146 rw_exit(&exec_lock); 1147 return error; 1148 } 1149 1150 /* 1151 * Remove emulation with name 'name' from list of supported emulations. 1152 */ 1153 int 1154 emul_unregister(const char *name) 1155 { 1156 const struct proclist_desc *pd; 1157 struct emul_entry *it; 1158 int i, error; 1159 struct proc *ptmp; 1160 1161 error = 0; 1162 rw_enter(&exec_lock, RW_WRITER); 1163 1164 LIST_FOREACH(it, &el_head, el_list) { 1165 if (strcmp(it->el_emul->e_name, name) == 0) 1166 break; 1167 } 1168 1169 if (!it) { 1170 error = ENOENT; 1171 goto out; 1172 } 1173 1174 if (it->ro_entry) { 1175 error = EBUSY; 1176 goto out; 1177 } 1178 1179 /* test if any execw[] entry is still using this */ 1180 for(i=0; i < nexecs; i++) { 1181 if (execsw[i]->es_emul == it->el_emul) { 1182 error = EBUSY; 1183 goto out; 1184 } 1185 } 1186 1187 /* 1188 * Test if any process is running under this emulation - since 1189 * emul_unregister() is running quite sendomly, it's better 1190 * to do expensive check here than to use any locking. 1191 */ 1192 mutex_enter(&proclist_lock); 1193 for (pd = proclists; pd->pd_list != NULL && !error; pd++) { 1194 PROCLIST_FOREACH(ptmp, pd->pd_list) { 1195 if (ptmp->p_emul == it->el_emul) { 1196 error = EBUSY; 1197 break; 1198 } 1199 } 1200 } 1201 mutex_exit(&proclist_lock); 1202 1203 if (error) 1204 goto out; 1205 1206 1207 /* entry is not used, remove it */ 1208 LIST_REMOVE(it, el_list); 1209 FREE(it, M_EXEC); 1210 1211 out: 1212 rw_exit(&exec_lock); 1213 return error; 1214 } 1215 1216 /* 1217 * Add execsw[] entry. 1218 */ 1219 int 1220 exec_add(struct execsw *esp, const char *e_name) 1221 { 1222 struct exec_entry *it; 1223 int error; 1224 1225 error = 0; 1226 rw_enter(&exec_lock, RW_WRITER); 1227 1228 if (!esp->es_emul) { 1229 esp->es_emul = emul_search(e_name); 1230 if (!esp->es_emul) { 1231 error = ENOENT; 1232 goto out; 1233 } 1234 } 1235 1236 LIST_FOREACH(it, &ex_head, ex_list) { 1237 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1238 if (it->es->es_makecmds == esp->es_makecmds 1239 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1240 && it->es->es_emul == esp->es_emul) { 1241 error = EEXIST; 1242 goto out; 1243 } 1244 } 1245 1246 /* if we got here, the entry doesn't exist yet */ 1247 MALLOC(it, struct exec_entry *, sizeof(struct exec_entry), 1248 M_EXEC, M_WAITOK); 1249 it->es = esp; 1250 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1251 1252 /* update execsw[] */ 1253 exec_init(0); 1254 1255 out: 1256 rw_exit(&exec_lock); 1257 return error; 1258 } 1259 1260 /* 1261 * Remove execsw[] entry. 1262 */ 1263 int 1264 exec_remove(const struct execsw *esp) 1265 { 1266 struct exec_entry *it; 1267 int error; 1268 1269 error = 0; 1270 rw_enter(&exec_lock, RW_WRITER); 1271 1272 LIST_FOREACH(it, &ex_head, ex_list) { 1273 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1274 if (it->es->es_makecmds == esp->es_makecmds 1275 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1276 && it->es->es_emul == esp->es_emul) 1277 break; 1278 } 1279 if (!it) { 1280 error = ENOENT; 1281 goto out; 1282 } 1283 1284 /* remove item from list and free resources */ 1285 LIST_REMOVE(it, ex_list); 1286 FREE(it, M_EXEC); 1287 1288 /* update execsw[] */ 1289 exec_init(0); 1290 1291 out: 1292 rw_exit(&exec_lock); 1293 return error; 1294 } 1295 1296 static void 1297 link_es(struct execsw_entry **listp, const struct execsw *esp) 1298 { 1299 struct execsw_entry *et, *e1; 1300 1301 et = (struct execsw_entry *) malloc(sizeof(struct execsw_entry), 1302 M_TEMP, M_WAITOK); 1303 et->next = NULL; 1304 et->es = esp; 1305 if (*listp == NULL) { 1306 *listp = et; 1307 return; 1308 } 1309 1310 switch(et->es->es_prio) { 1311 case EXECSW_PRIO_FIRST: 1312 /* put new entry as the first */ 1313 et->next = *listp; 1314 *listp = et; 1315 break; 1316 case EXECSW_PRIO_ANY: 1317 /* put new entry after all *_FIRST and *_ANY entries */ 1318 for(e1 = *listp; e1->next 1319 && e1->next->es->es_prio != EXECSW_PRIO_LAST; 1320 e1 = e1->next); 1321 et->next = e1->next; 1322 e1->next = et; 1323 break; 1324 case EXECSW_PRIO_LAST: 1325 /* put new entry as the last one */ 1326 for(e1 = *listp; e1->next; e1 = e1->next); 1327 e1->next = et; 1328 break; 1329 default: 1330 #ifdef DIAGNOSTIC 1331 panic("execw[] entry with unknown priority %d found", 1332 et->es->es_prio); 1333 #else 1334 free(et, M_TEMP); 1335 #endif 1336 break; 1337 } 1338 } 1339 1340 /* 1341 * Initialize exec structures. If init_boot is true, also does necessary 1342 * one-time initialization (it's called from main() that way). 1343 * Once system is multiuser, this should be called with exec_lock held, 1344 * i.e. via exec_{add|remove}(). 1345 */ 1346 int 1347 exec_init(int init_boot) 1348 { 1349 const struct execsw **new_es, * const *old_es; 1350 struct execsw_entry *list, *e1; 1351 struct exec_entry *e2; 1352 int i, es_sz; 1353 1354 if (init_boot) { 1355 /* do one-time initializations */ 1356 rw_init(&exec_lock); 1357 1358 /* register compiled-in emulations */ 1359 for(i=0; i < nexecs_builtin; i++) { 1360 if (execsw_builtin[i].es_emul) 1361 emul_register(execsw_builtin[i].es_emul, 1); 1362 } 1363 #ifdef DIAGNOSTIC 1364 if (i == 0) 1365 panic("no emulations found in execsw_builtin[]"); 1366 #endif 1367 } 1368 1369 /* 1370 * Build execsw[] array from builtin entries and entries added 1371 * at runtime. 1372 */ 1373 list = NULL; 1374 for(i=0; i < nexecs_builtin; i++) 1375 link_es(&list, &execsw_builtin[i]); 1376 1377 /* Add dynamically loaded entries */ 1378 es_sz = nexecs_builtin; 1379 LIST_FOREACH(e2, &ex_head, ex_list) { 1380 link_es(&list, e2->es); 1381 es_sz++; 1382 } 1383 1384 /* 1385 * Now that we have sorted all execw entries, create new execsw[] 1386 * and free no longer needed memory in the process. 1387 */ 1388 new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK); 1389 for(i=0; list; i++) { 1390 new_es[i] = list->es; 1391 e1 = list->next; 1392 free(list, M_TEMP); 1393 list = e1; 1394 } 1395 1396 /* 1397 * New execsw[] array built, now replace old execsw[] and free 1398 * used memory. 1399 */ 1400 old_es = execsw; 1401 execsw = new_es; 1402 nexecs = es_sz; 1403 if (old_es) 1404 /*XXXUNCONST*/ 1405 free(__UNCONST(old_es), M_EXEC); 1406 1407 /* 1408 * Figure out the maximum size of an exec header. 1409 */ 1410 exec_maxhdrsz = 0; 1411 for (i = 0; i < nexecs; i++) { 1412 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1413 exec_maxhdrsz = execsw[i]->es_hdrsz; 1414 } 1415 1416 return 0; 1417 } 1418 #endif 1419 1420 #ifndef LKM 1421 /* 1422 * Simplified exec_init() for kernels without LKMs. Only initialize 1423 * exec_maxhdrsz and execsw[]. 1424 */ 1425 int 1426 exec_init(int init_boot) 1427 { 1428 int i; 1429 1430 #ifdef DIAGNOSTIC 1431 if (!init_boot) 1432 panic("exec_init(): called with init_boot == 0"); 1433 #endif 1434 1435 /* do one-time initializations */ 1436 nexecs = nexecs_builtin; 1437 execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK); 1438 1439 /* 1440 * Fill in execsw[] and figure out the maximum size of an exec header. 1441 */ 1442 exec_maxhdrsz = 0; 1443 for(i=0; i < nexecs; i++) { 1444 execsw[i] = &execsw_builtin[i]; 1445 if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz) 1446 exec_maxhdrsz = execsw_builtin[i].es_hdrsz; 1447 } 1448 1449 return 0; 1450 1451 } 1452 #endif /* !LKM */ 1453 1454 static int 1455 exec_sigcode_map(struct proc *p, const struct emul *e) 1456 { 1457 vaddr_t va; 1458 vsize_t sz; 1459 int error; 1460 struct uvm_object *uobj; 1461 1462 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1463 1464 if (e->e_sigobject == NULL || sz == 0) { 1465 return 0; 1466 } 1467 1468 /* 1469 * If we don't have a sigobject for this emulation, create one. 1470 * 1471 * sigobject is an anonymous memory object (just like SYSV shared 1472 * memory) that we keep a permanent reference to and that we map 1473 * in all processes that need this sigcode. The creation is simple, 1474 * we create an object, add a permanent reference to it, map it in 1475 * kernel space, copy out the sigcode to it and unmap it. 1476 * We map it with PROT_READ|PROT_EXEC into the process just 1477 * the way sys_mmap() would map it. 1478 */ 1479 1480 uobj = *e->e_sigobject; 1481 if (uobj == NULL) { 1482 uobj = uao_create(sz, 0); 1483 (*uobj->pgops->pgo_reference)(uobj); 1484 va = vm_map_min(kernel_map); 1485 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1486 uobj, 0, 0, 1487 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1488 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 1489 printf("kernel mapping failed %d\n", error); 1490 (*uobj->pgops->pgo_detach)(uobj); 1491 return (error); 1492 } 1493 memcpy((void *)va, e->e_sigcode, sz); 1494 #ifdef PMAP_NEED_PROCWR 1495 pmap_procwr(&proc0, va, sz); 1496 #endif 1497 uvm_unmap(kernel_map, va, va + round_page(sz)); 1498 *e->e_sigobject = uobj; 1499 } 1500 1501 /* Just a hint to uvm_map where to put it. */ 1502 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr, 1503 round_page(sz)); 1504 1505 #ifdef __alpha__ 1506 /* 1507 * Tru64 puts /sbin/loader at the end of user virtual memory, 1508 * which causes the above calculation to put the sigcode at 1509 * an invalid address. Put it just below the text instead. 1510 */ 1511 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) { 1512 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 1513 } 1514 #endif 1515 1516 (*uobj->pgops->pgo_reference)(uobj); 1517 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 1518 uobj, 0, 0, 1519 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 1520 UVM_ADV_RANDOM, 0)); 1521 if (error) { 1522 (*uobj->pgops->pgo_detach)(uobj); 1523 return (error); 1524 } 1525 p->p_sigctx.ps_sigcode = (void *)va; 1526 return (0); 1527 } 1528