Home | History | Annotate | Line # | Download | only in kern
kern_exec.c revision 1.209
      1 /*	$NetBSD: kern_exec.c,v 1.209 2005/08/19 02:04:03 christos 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 1.209 2005/08/19 02:04:03 christos Exp $");
     37 
     38 #include "opt_ktrace.h"
     39 #include "opt_syscall_debug.h"
     40 #include "opt_compat_netbsd.h"
     41 #include "opt_verified_exec.h"
     42 
     43 #include <sys/param.h>
     44 #include <sys/systm.h>
     45 #include <sys/filedesc.h>
     46 #include <sys/kernel.h>
     47 #include <sys/proc.h>
     48 #include <sys/mount.h>
     49 #include <sys/malloc.h>
     50 #include <sys/namei.h>
     51 #include <sys/vnode.h>
     52 #include <sys/file.h>
     53 #include <sys/acct.h>
     54 #include <sys/exec.h>
     55 #include <sys/ktrace.h>
     56 #include <sys/resourcevar.h>
     57 #include <sys/wait.h>
     58 #include <sys/mman.h>
     59 #include <sys/ras.h>
     60 #include <sys/signalvar.h>
     61 #include <sys/stat.h>
     62 #include <sys/syscall.h>
     63 
     64 #include <sys/sa.h>
     65 #include <sys/savar.h>
     66 #include <sys/syscallargs.h>
     67 #ifdef VERIFIED_EXEC
     68 #include <sys/verified_exec.h>
     69 #endif
     70 
     71 #ifdef SYSTRACE
     72 #include <sys/systrace.h>
     73 #endif /* SYSTRACE */
     74 
     75 #include <uvm/uvm_extern.h>
     76 
     77 #include <machine/cpu.h>
     78 #include <machine/reg.h>
     79 
     80 static int exec_sigcode_map(struct proc *, const struct emul *);
     81 
     82 #ifdef DEBUG_EXEC
     83 #define DPRINTF(a) uprintf a
     84 #else
     85 #define DPRINTF(a)
     86 #endif /* DEBUG_EXEC */
     87 
     88 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec");
     89 
     90 /*
     91  * Exec function switch:
     92  *
     93  * Note that each makecmds function is responsible for loading the
     94  * exec package with the necessary functions for any exec-type-specific
     95  * handling.
     96  *
     97  * Functions for specific exec types should be defined in their own
     98  * header file.
     99  */
    100 extern const struct execsw	execsw_builtin[];
    101 extern int			nexecs_builtin;
    102 static const struct execsw	**execsw = NULL;
    103 static int			nexecs;
    104 
    105 u_int	exec_maxhdrsz;		/* must not be static - netbsd32 needs it */
    106 
    107 #ifdef LKM
    108 /* list of supported emulations */
    109 static
    110 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head);
    111 struct emul_entry {
    112 	LIST_ENTRY(emul_entry)	el_list;
    113 	const struct emul	*el_emul;
    114 	int			ro_entry;
    115 };
    116 
    117 /* list of dynamically loaded execsw entries */
    118 static
    119 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head);
    120 struct exec_entry {
    121 	LIST_ENTRY(exec_entry)	ex_list;
    122 	const struct execsw	*es;
    123 };
    124 
    125 /* structure used for building execw[] */
    126 struct execsw_entry {
    127 	struct execsw_entry	*next;
    128 	const struct execsw	*es;
    129 };
    130 #endif /* LKM */
    131 
    132 #ifdef SYSCALL_DEBUG
    133 extern const char * const syscallnames[];
    134 #endif
    135 
    136 #ifdef COMPAT_16
    137 extern char	sigcode[], esigcode[];
    138 struct uvm_object *emul_netbsd_object;
    139 #endif
    140 
    141 #ifndef __HAVE_SYSCALL_INTERN
    142 void	syscall(void);
    143 #endif
    144 
    145 /* NetBSD emul struct */
    146 const struct emul emul_netbsd = {
    147 	"netbsd",
    148 	NULL,		/* emulation path */
    149 #ifndef __HAVE_MINIMAL_EMUL
    150 	EMUL_HAS_SYS___syscall,
    151 	NULL,
    152 	SYS_syscall,
    153 	SYS_NSYSENT,
    154 #endif
    155 	sysent,
    156 #ifdef SYSCALL_DEBUG
    157 	syscallnames,
    158 #else
    159 	NULL,
    160 #endif
    161 	sendsig,
    162 	trapsignal,
    163 	NULL,
    164 #ifdef COMPAT_16
    165 	sigcode,
    166 	esigcode,
    167 	&emul_netbsd_object,
    168 #else
    169 	NULL,
    170 	NULL,
    171 	NULL,
    172 #endif
    173 	setregs,
    174 	NULL,
    175 	NULL,
    176 	NULL,
    177 	NULL,
    178 	NULL,
    179 #ifdef __HAVE_SYSCALL_INTERN
    180 	syscall_intern,
    181 #else
    182 	syscall,
    183 #endif
    184 	NULL,
    185 	NULL,
    186 
    187 	uvm_default_mapaddr,
    188 };
    189 
    190 #ifdef LKM
    191 /*
    192  * Exec lock. Used to control access to execsw[] structures.
    193  * This must not be static so that netbsd32 can access it, too.
    194  */
    195 struct lock exec_lock;
    196 
    197 static void link_es(struct execsw_entry **, const struct execsw *);
    198 #endif /* LKM */
    199 
    200 /*
    201  * check exec:
    202  * given an "executable" described in the exec package's namei info,
    203  * see what we can do with it.
    204  *
    205  * ON ENTRY:
    206  *	exec package with appropriate namei info
    207  *	proc pointer of exec'ing proc
    208  *      if verified exec enabled then flag indicating a direct exec or
    209  *        an indirect exec (i.e. for a shell script interpreter)
    210  *	NO SELF-LOCKED VNODES
    211  *
    212  * ON EXIT:
    213  *	error:	nothing held, etc.  exec header still allocated.
    214  *	ok:	filled exec package, executable's vnode (unlocked).
    215  *
    216  * EXEC SWITCH ENTRY:
    217  * 	Locked vnode to check, exec package, proc.
    218  *
    219  * EXEC SWITCH EXIT:
    220  *	ok:	return 0, filled exec package, executable's vnode (unlocked).
    221  *	error:	destructive:
    222  *			everything deallocated execept exec header.
    223  *		non-destructive:
    224  *			error code, executable's vnode (unlocked),
    225  *			exec header unmodified.
    226  */
    227 int
    228 /*ARGSUSED*/
    229 check_exec(struct proc *p, struct exec_package *epp, int flag)
    230 {
    231 	int		error, i;
    232 	struct vnode	*vp;
    233 	struct nameidata *ndp;
    234 	size_t		resid;
    235 
    236 	ndp = epp->ep_ndp;
    237 	ndp->ni_cnd.cn_nameiop = LOOKUP;
    238 	ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME;
    239 	/* first get the vnode */
    240 	if ((error = namei(ndp)) != 0)
    241 		return error;
    242 	epp->ep_vp = vp = ndp->ni_vp;
    243 
    244 	/* check access and type */
    245 	if (vp->v_type != VREG) {
    246 		error = EACCES;
    247 		goto bad1;
    248 	}
    249 	if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0)
    250 		goto bad1;
    251 
    252 	/* get attributes */
    253 	if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0)
    254 		goto bad1;
    255 
    256 	/* Check mount point */
    257 	if (vp->v_mount->mnt_flag & MNT_NOEXEC) {
    258 		error = EACCES;
    259 		goto bad1;
    260 	}
    261 	if (vp->v_mount->mnt_flag & MNT_NOSUID)
    262 		epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID);
    263 
    264 	/* try to open it */
    265 	if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0)
    266 		goto bad1;
    267 
    268 	/* unlock vp, since we need it unlocked from here on out. */
    269 	VOP_UNLOCK(vp, 0);
    270 
    271 
    272 #ifdef VERIFIED_EXEC
    273         if ((error = veriexec_verify(p, vp, epp->ep_vap, epp->ep_ndp->ni_dirp,
    274 				     flag, NULL)) != 0)
    275                 goto bad2;
    276 #endif
    277 
    278 	/* now we have the file, get the exec header */
    279 	uvn_attach(vp, VM_PROT_READ);
    280 	error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0,
    281 			UIO_SYSSPACE, 0, p->p_ucred, &resid, NULL);
    282 	if (error)
    283 		goto bad2;
    284 	epp->ep_hdrvalid = epp->ep_hdrlen - resid;
    285 
    286 	/*
    287 	 * Set up default address space limits.  Can be overridden
    288 	 * by individual exec packages.
    289 	 *
    290 	 * XXX probably should be all done in the exec pakages.
    291 	 */
    292 	epp->ep_vm_minaddr = VM_MIN_ADDRESS;
    293 	epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS;
    294 	/*
    295 	 * set up the vmcmds for creation of the process
    296 	 * address space
    297 	 */
    298 	error = ENOEXEC;
    299 	for (i = 0; i < nexecs && error != 0; i++) {
    300 		int newerror;
    301 
    302 		epp->ep_esch = execsw[i];
    303 		newerror = (*execsw[i]->es_makecmds)(p, epp);
    304 		/* make sure the first "interesting" error code is saved. */
    305 		if (!newerror || error == ENOEXEC)
    306 			error = newerror;
    307 
    308 		/* if es_makecmds call was successful, update epp->ep_es */
    309 		if (!newerror && (epp->ep_flags & EXEC_HASES) == 0)
    310 			epp->ep_es = execsw[i];
    311 
    312 		if (epp->ep_flags & EXEC_DESTR && error != 0)
    313 			return error;
    314 	}
    315 	if (!error) {
    316 		/* check that entry point is sane */
    317 		if (epp->ep_entry > VM_MAXUSER_ADDRESS)
    318 			error = ENOEXEC;
    319 
    320 		/* check limits */
    321 		if ((epp->ep_tsize > MAXTSIZ) ||
    322 		    (epp->ep_dsize >
    323 		     (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur))
    324 			error = ENOMEM;
    325 
    326 		if (!error)
    327 			return (0);
    328 	}
    329 
    330 	/*
    331 	 * free any vmspace-creation commands,
    332 	 * and release their references
    333 	 */
    334 	kill_vmcmds(&epp->ep_vmcmds);
    335 
    336 bad2:
    337 	/*
    338 	 * close and release the vnode, restore the old one, free the
    339 	 * pathname buf, and punt.
    340 	 */
    341 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
    342 	VOP_CLOSE(vp, FREAD, p->p_ucred, p);
    343 	vput(vp);
    344 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
    345 	return error;
    346 
    347 bad1:
    348 	/*
    349 	 * free the namei pathname buffer, and put the vnode
    350 	 * (which we don't yet have open).
    351 	 */
    352 	vput(vp);				/* was still locked */
    353 	PNBUF_PUT(ndp->ni_cnd.cn_pnbuf);
    354 	return error;
    355 }
    356 
    357 #ifdef __MACHINE_STACK_GROWS_UP
    358 #define STACK_PTHREADSPACE NBPG
    359 #else
    360 #define STACK_PTHREADSPACE 0
    361 #endif
    362 
    363 static int
    364 execve_fetch_element(char * const *array, size_t index, char **value)
    365 {
    366 	return copyin(array + index, value, sizeof(*value));
    367 }
    368 
    369 /*
    370  * exec system call
    371  */
    372 /* ARGSUSED */
    373 int
    374 sys_execve(struct lwp *l, void *v, register_t *retval)
    375 {
    376 	struct sys_execve_args /* {
    377 		syscallarg(const char *)	path;
    378 		syscallarg(char * const *)	argp;
    379 		syscallarg(char * const *)	envp;
    380 	} */ *uap = v;
    381 
    382 	return execve1(l, SCARG(uap, path), SCARG(uap, argp),
    383 	    SCARG(uap, envp), execve_fetch_element);
    384 }
    385 
    386 int
    387 execve1(struct lwp *l, const char *path, char * const *args,
    388     char * const *envs, execve_fetch_element_t fetch_element)
    389 {
    390 	int			error;
    391 	u_int			i;
    392 	struct exec_package	pack;
    393 	struct nameidata	nid;
    394 	struct vattr		attr;
    395 	struct proc		*p;
    396 	struct ucred		*cred;
    397 	char			*argp;
    398 	char			*dp, *sp;
    399 	long			argc, envc;
    400 	size_t			len;
    401 	char			*stack;
    402 	struct ps_strings	arginfo;
    403 	struct vmspace		*vm;
    404 	char			**tmpfap;
    405 	int			szsigcode;
    406 	struct exec_vmcmd	*base_vcp;
    407 	int			oldlwpflags;
    408 #ifdef SYSTRACE
    409 	int			wassugid = ISSET(p->p_flag, P_SUGID);
    410 	char			pathbuf[MAXPATHLEN];
    411 	size_t			pathbuflen;
    412 #endif /* SYSTRACE */
    413 
    414 	/* Disable scheduler activation upcalls. */
    415 	oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL);
    416 	if (l->l_flag & L_SA)
    417 		l->l_flag &= ~(L_SA | L_SA_UPCALL);
    418 
    419 	p = l->l_proc;
    420 	/*
    421 	 * Lock the process and set the P_INEXEC flag to indicate that
    422 	 * it should be left alone until we're done here.  This is
    423 	 * necessary to avoid race conditions - e.g. in ptrace() -
    424 	 * that might allow a local user to illicitly obtain elevated
    425 	 * privileges.
    426 	 */
    427 	p->p_flag |= P_INEXEC;
    428 
    429 	cred = p->p_ucred;
    430 	base_vcp = NULL;
    431 	/*
    432 	 * Init the namei data to point the file user's program name.
    433 	 * This is done here rather than in check_exec(), so that it's
    434 	 * possible to override this settings if any of makecmd/probe
    435 	 * functions call check_exec() recursively - for example,
    436 	 * see exec_script_makecmds().
    437 	 */
    438 #ifdef SYSTRACE
    439 	if (ISSET(p->p_flag, P_SYSTRACE))
    440 		systrace_execve0(p);
    441 
    442 	error = copyinstr(path, pathbuf, sizeof(pathbuf),
    443 			  &pathbuflen);
    444 	if (error)
    445 		goto clrflg;
    446 
    447 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_SYSSPACE, pathbuf, p);
    448 #else
    449 	NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, path, p);
    450 #endif /* SYSTRACE */
    451 
    452 	/*
    453 	 * initialize the fields of the exec package.
    454 	 */
    455 #ifdef SYSTRACE
    456 	pack.ep_name = pathbuf;
    457 #else
    458 	pack.ep_name = path;
    459 #endif /* SYSTRACE */
    460 	pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK);
    461 	pack.ep_hdrlen = exec_maxhdrsz;
    462 	pack.ep_hdrvalid = 0;
    463 	pack.ep_ndp = &nid;
    464 	pack.ep_emul_arg = NULL;
    465 	pack.ep_vmcmds.evs_cnt = 0;
    466 	pack.ep_vmcmds.evs_used = 0;
    467 	pack.ep_vap = &attr;
    468 	pack.ep_flags = 0;
    469 
    470 #ifdef LKM
    471 	lockmgr(&exec_lock, LK_SHARED, NULL);
    472 #endif
    473 
    474 	/* see if we can run it. */
    475 #ifdef VERIFIED_EXEC
    476         if ((error = check_exec(p, &pack, VERIEXEC_DIRECT)) != 0)
    477 #else
    478         if ((error = check_exec(p, &pack, 0)) != 0)
    479 #endif
    480 		goto freehdr;
    481 
    482 	/* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */
    483 
    484 	/* allocate an argument buffer */
    485 	argp = (char *) uvm_km_alloc(exec_map, NCARGS, 0,
    486 	    UVM_KMF_PAGEABLE|UVM_KMF_WAITVA);
    487 #ifdef DIAGNOSTIC
    488 	if (argp == NULL)
    489 		panic("execve: argp == NULL");
    490 #endif
    491 	dp = argp;
    492 	argc = 0;
    493 
    494 	/* copy the fake args list, if there's one, freeing it as we go */
    495 	if (pack.ep_flags & EXEC_HASARGL) {
    496 		tmpfap = pack.ep_fa;
    497 		while (*tmpfap != NULL) {
    498 			char *cp;
    499 
    500 			cp = *tmpfap;
    501 			while (*cp)
    502 				*dp++ = *cp++;
    503 			dp++;
    504 
    505 			FREE(*tmpfap, M_EXEC);
    506 			tmpfap++; argc++;
    507 		}
    508 		FREE(pack.ep_fa, M_EXEC);
    509 		pack.ep_flags &= ~EXEC_HASARGL;
    510 	}
    511 
    512 	/* Now get argv & environment */
    513 	if (args == NULL) {
    514 		error = EINVAL;
    515 		goto bad;
    516 	}
    517 	/* 'i' will index the argp/envp element to be retrieved */
    518 	i = 0;
    519 	if (pack.ep_flags & EXEC_SKIPARG)
    520 		i++;
    521 
    522 	while (1) {
    523 		len = argp + ARG_MAX - dp;
    524 		if ((error = (*fetch_element)(args, i, &sp)) != 0)
    525 			goto bad;
    526 		if (!sp)
    527 			break;
    528 		if ((error = copyinstr(sp, dp, len, &len)) != 0) {
    529 			if (error == ENAMETOOLONG)
    530 				error = E2BIG;
    531 			goto bad;
    532 		}
    533 #ifdef KTRACE
    534 		if (KTRPOINT(p, KTR_EXEC_ARG))
    535 			ktrkmem(p, KTR_EXEC_ARG, dp, len - 1);
    536 #endif
    537 		dp += len;
    538 		i++;
    539 		argc++;
    540 	}
    541 
    542 	envc = 0;
    543 	/* environment need not be there */
    544 	if (envs != NULL) {
    545 		i = 0;
    546 		while (1) {
    547 			len = argp + ARG_MAX - dp;
    548 			if ((error = (*fetch_element)(envs, i, &sp)) != 0)
    549 				goto bad;
    550 			if (!sp)
    551 				break;
    552 			if ((error = copyinstr(sp, dp, len, &len)) != 0) {
    553 				if (error == ENAMETOOLONG)
    554 					error = E2BIG;
    555 				goto bad;
    556 			}
    557 #ifdef KTRACE
    558 			if (KTRPOINT(p, KTR_EXEC_ENV))
    559 				ktrkmem(p, KTR_EXEC_ENV, dp, len - 1);
    560 #endif
    561 			dp += len;
    562 			i++;
    563 			envc++;
    564 		}
    565 	}
    566 
    567 	dp = (char *) ALIGN(dp);
    568 
    569 	szsigcode = pack.ep_es->es_emul->e_esigcode -
    570 	    pack.ep_es->es_emul->e_sigcode;
    571 
    572 	/* Now check if args & environ fit into new stack */
    573 	if (pack.ep_flags & EXEC_32)
    574 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
    575 		    sizeof(int) + sizeof(int) + dp + STACKGAPLEN +
    576 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
    577 		    - argp;
    578 	else
    579 		len = ((argc + envc + 2 + pack.ep_es->es_arglen) *
    580 		    sizeof(char *) + sizeof(int) + dp + STACKGAPLEN +
    581 		    szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE)
    582 		    - argp;
    583 
    584 	len = ALIGN(len);	/* make the stack "safely" aligned */
    585 
    586 	if (len > pack.ep_ssize) { /* in effect, compare to initial limit */
    587 		error = ENOMEM;
    588 		goto bad;
    589 	}
    590 
    591 	/* Get rid of other LWPs/ */
    592 	p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */
    593 	exit_lwps(l);
    594 	p->p_flag &= ~P_WEXIT;
    595 	KDASSERT(p->p_nlwps == 1);
    596 
    597 	/* This is now LWP 1 */
    598 	l->l_lid = 1;
    599 	p->p_nlwpid = 1;
    600 
    601 	/* Release any SA state. */
    602 	if (p->p_sa)
    603 		sa_release(p);
    604 
    605 	/* Remove POSIX timers */
    606 	timers_free(p, TIMERS_POSIX);
    607 
    608 	/* adjust "active stack depth" for process VSZ */
    609 	pack.ep_ssize = len;	/* maybe should go elsewhere, but... */
    610 
    611 	/*
    612 	 * Do whatever is necessary to prepare the address space
    613 	 * for remapping.  Note that this might replace the current
    614 	 * vmspace with another!
    615 	 */
    616 	uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr);
    617 
    618 	/* record proc's vnode, for use by procfs and others */
    619         if (p->p_textvp)
    620                 vrele(p->p_textvp);
    621 	VREF(pack.ep_vp);
    622 	p->p_textvp = pack.ep_vp;
    623 
    624 	/* Now map address space */
    625 	vm = p->p_vmspace;
    626 	vm->vm_taddr = (caddr_t) pack.ep_taddr;
    627 	vm->vm_tsize = btoc(pack.ep_tsize);
    628 	vm->vm_daddr = (caddr_t) pack.ep_daddr;
    629 	vm->vm_dsize = btoc(pack.ep_dsize);
    630 	vm->vm_ssize = btoc(pack.ep_ssize);
    631 	vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr;
    632 	vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr;
    633 
    634 	/* create the new process's VM space by running the vmcmds */
    635 #ifdef DIAGNOSTIC
    636 	if (pack.ep_vmcmds.evs_used == 0)
    637 		panic("execve: no vmcmds");
    638 #endif
    639 	for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) {
    640 		struct exec_vmcmd *vcp;
    641 
    642 		vcp = &pack.ep_vmcmds.evs_cmds[i];
    643 		if (vcp->ev_flags & VMCMD_RELATIVE) {
    644 #ifdef DIAGNOSTIC
    645 			if (base_vcp == NULL)
    646 				panic("execve: relative vmcmd with no base");
    647 			if (vcp->ev_flags & VMCMD_BASE)
    648 				panic("execve: illegal base & relative vmcmd");
    649 #endif
    650 			vcp->ev_addr += base_vcp->ev_addr;
    651 		}
    652 		error = (*vcp->ev_proc)(p, vcp);
    653 #ifdef DEBUG_EXEC
    654 		if (error) {
    655 			int j;
    656 			struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0];
    657 			for (j = 0; j <= i; j++)
    658 				uprintf(
    659 			    "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n",
    660 				    j, vp[j].ev_addr, vp[j].ev_len,
    661 				    vp[j].ev_offset, vp[j].ev_prot,
    662 				    vp[j].ev_flags);
    663 		}
    664 #endif /* DEBUG_EXEC */
    665 		if (vcp->ev_flags & VMCMD_BASE)
    666 			base_vcp = vcp;
    667 	}
    668 
    669 	/* free the vmspace-creation commands, and release their references */
    670 	kill_vmcmds(&pack.ep_vmcmds);
    671 
    672 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
    673 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
    674 	vput(pack.ep_vp);
    675 
    676 	/* if an error happened, deallocate and punt */
    677 	if (error) {
    678 		DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error));
    679 		goto exec_abort;
    680 	}
    681 
    682 	/* remember information about the process */
    683 	arginfo.ps_nargvstr = argc;
    684 	arginfo.ps_nenvstr = envc;
    685 
    686 	stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr,
    687 		STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode),
    688 		len - (sizeof(struct ps_strings) + szsigcode));
    689 #ifdef __MACHINE_STACK_GROWS_UP
    690 	/*
    691 	 * The copyargs call always copies into lower addresses
    692 	 * first, moving towards higher addresses, starting with
    693 	 * the stack pointer that we give.  When the stack grows
    694 	 * down, this puts argc/argv/envp very shallow on the
    695 	 * stack, right at the first user stack pointer, and puts
    696 	 * STACKGAPLEN very deep in the stack.  When the stack
    697 	 * grows up, the situation is reversed.
    698 	 *
    699 	 * Normally, this is no big deal.  But the ld_elf.so _rtld()
    700 	 * function expects to be called with a single pointer to
    701 	 * a region that has a few words it can stash values into,
    702 	 * followed by argc/argv/envp.  When the stack grows down,
    703 	 * it's easy to decrement the stack pointer a little bit to
    704 	 * allocate the space for these few words and pass the new
    705 	 * stack pointer to _rtld.  When the stack grows up, however,
    706 	 * a few words before argc is part of the signal trampoline, XXX
    707 	 * so we have a problem.
    708 	 *
    709 	 * Instead of changing how _rtld works, we take the easy way
    710 	 * out and steal 32 bytes before we call copyargs.  This
    711 	 * space is effectively stolen from STACKGAPLEN.
    712 	 */
    713 	stack += 32;
    714 #endif /* __MACHINE_STACK_GROWS_UP */
    715 
    716 	/* Now copy argc, args & environ to new stack */
    717 	error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp);
    718 	if (error) {
    719 		DPRINTF(("execve: copyargs failed %d\n", error));
    720 		goto exec_abort;
    721 	}
    722 	/* Move the stack back to original point */
    723 	stack = (char *)STACK_GROW(vm->vm_minsaddr, len);
    724 
    725 	/* fill process ps_strings info */
    726 	p->p_psstr = (struct ps_strings *)
    727 	    STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE),
    728 	    sizeof(struct ps_strings));
    729 	p->p_psargv = offsetof(struct ps_strings, ps_argvstr);
    730 	p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr);
    731 	p->p_psenv = offsetof(struct ps_strings, ps_envstr);
    732 	p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr);
    733 
    734 	/* copy out the process's ps_strings structure */
    735 	if ((error = copyout(&arginfo, (char *)p->p_psstr,
    736 	    sizeof(arginfo))) != 0) {
    737 		DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n",
    738 		       &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo)));
    739 		goto exec_abort;
    740 	}
    741 
    742 	stopprofclock(p);	/* stop profiling */
    743 	fdcloseexec(p);		/* handle close on exec */
    744 	execsigs(p);		/* reset catched signals */
    745 
    746 	l->l_ctxlink = NULL;	/* reset ucontext link */
    747 
    748 	/* set command name & other accounting info */
    749 	len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN);
    750 	memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len);
    751 	p->p_comm[len] = 0;
    752 	p->p_acflag &= ~AFORK;
    753 
    754 	p->p_flag |= P_EXEC;
    755 	if (p->p_flag & P_PPWAIT) {
    756 		p->p_flag &= ~P_PPWAIT;
    757 		wakeup((caddr_t) p->p_pptr);
    758 	}
    759 
    760 	/*
    761 	 * deal with set[ug]id.
    762 	 * MNT_NOSUID has already been used to disable s[ug]id.
    763 	 */
    764 	if ((p->p_flag & P_TRACED) == 0 &&
    765 
    766 	    (((attr.va_mode & S_ISUID) != 0 &&
    767 	      p->p_ucred->cr_uid != attr.va_uid) ||
    768 
    769 	     ((attr.va_mode & S_ISGID) != 0 &&
    770 	      p->p_ucred->cr_gid != attr.va_gid))) {
    771 		/*
    772 		 * Mark the process as SUGID before we do
    773 		 * anything that might block.
    774 		 */
    775 		p_sugid(p);
    776 
    777 		/* Make sure file descriptors 0..2 are in use. */
    778 		if ((error = fdcheckstd(p)) != 0) {
    779 			DPRINTF(("execve: fdcheckstd failed %d\n", error));
    780 			goto exec_abort;
    781 		}
    782 
    783 		p->p_ucred = crcopy(cred);
    784 #ifdef KTRACE
    785 		/*
    786 		 * If process is being ktraced, turn off - unless
    787 		 * root set it.
    788 		 */
    789 		if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT))
    790 			ktrderef(p);
    791 #endif
    792 		if (attr.va_mode & S_ISUID)
    793 			p->p_ucred->cr_uid = attr.va_uid;
    794 		if (attr.va_mode & S_ISGID)
    795 			p->p_ucred->cr_gid = attr.va_gid;
    796 	} else
    797 		p->p_flag &= ~P_SUGID;
    798 	p->p_cred->p_svuid = p->p_ucred->cr_uid;
    799 	p->p_cred->p_svgid = p->p_ucred->cr_gid;
    800 
    801 #if defined(__HAVE_RAS)
    802 	/*
    803 	 * Remove all RASs from the address space.
    804 	 */
    805 	ras_purgeall(p);
    806 #endif
    807 
    808 	doexechooks(p);
    809 
    810 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
    811 
    812 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
    813 
    814 	/* notify others that we exec'd */
    815 	KNOTE(&p->p_klist, NOTE_EXEC);
    816 
    817 	/* setup new registers and do misc. setup. */
    818 	(*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack);
    819 	if (pack.ep_es->es_setregs)
    820 		(*pack.ep_es->es_setregs)(l, &pack, (u_long) stack);
    821 
    822 	/* map the process's signal trampoline code */
    823 	if (exec_sigcode_map(p, pack.ep_es->es_emul)) {
    824 		DPRINTF(("execve: map sigcode failed %d\n", error));
    825 		goto exec_abort;
    826 	}
    827 
    828 	if (p->p_flag & P_TRACED)
    829 		psignal(p, SIGTRAP);
    830 
    831 	free(pack.ep_hdr, M_EXEC);
    832 
    833 	/*
    834 	 * Call emulation specific exec hook. This can setup per-process
    835 	 * p->p_emuldata or do any other per-process stuff an emulation needs.
    836 	 *
    837 	 * If we are executing process of different emulation than the
    838 	 * original forked process, call e_proc_exit() of the old emulation
    839 	 * first, then e_proc_exec() of new emulation. If the emulation is
    840 	 * same, the exec hook code should deallocate any old emulation
    841 	 * resources held previously by this process.
    842 	 */
    843 	if (p->p_emul && p->p_emul->e_proc_exit
    844 	    && p->p_emul != pack.ep_es->es_emul)
    845 		(*p->p_emul->e_proc_exit)(p);
    846 
    847 	/*
    848 	 * Call exec hook. Emulation code may NOT store reference to anything
    849 	 * from &pack.
    850 	 */
    851         if (pack.ep_es->es_emul->e_proc_exec)
    852                 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack);
    853 
    854 	/* update p_emul, the old value is no longer needed */
    855 	p->p_emul = pack.ep_es->es_emul;
    856 
    857 	/* ...and the same for p_execsw */
    858 	p->p_execsw = pack.ep_es;
    859 
    860 #ifdef __HAVE_SYSCALL_INTERN
    861 	(*p->p_emul->e_syscall_intern)(p);
    862 #endif
    863 #ifdef KTRACE
    864 	if (KTRPOINT(p, KTR_EMUL))
    865 		ktremul(p);
    866 #endif
    867 
    868 #ifdef LKM
    869 	lockmgr(&exec_lock, LK_RELEASE, NULL);
    870 #endif
    871 	p->p_flag &= ~P_INEXEC;
    872 
    873 	if (p->p_flag & P_STOPEXEC) {
    874 		int s;
    875 
    876 		sigminusset(&contsigmask, &p->p_sigctx.ps_siglist);
    877 		SCHED_LOCK(s);
    878 		p->p_pptr->p_nstopchild++;
    879 		p->p_stat = SSTOP;
    880 		l->l_stat = LSSTOP;
    881 		p->p_nrlwps--;
    882 		mi_switch(l, NULL);
    883 		SCHED_ASSERT_UNLOCKED();
    884 		splx(s);
    885 	}
    886 
    887 #ifdef SYSTRACE
    888 	if (ISSET(p->p_flag, P_SYSTRACE) &&
    889 	    wassugid && !ISSET(p->p_flag, P_SUGID))
    890 		systrace_execve1(pathbuf, p);
    891 #endif /* SYSTRACE */
    892 
    893 	return (EJUSTRETURN);
    894 
    895  bad:
    896 	p->p_flag &= ~P_INEXEC;
    897 	/* free the vmspace-creation commands, and release their references */
    898 	kill_vmcmds(&pack.ep_vmcmds);
    899 	/* kill any opened file descriptor, if necessary */
    900 	if (pack.ep_flags & EXEC_HASFD) {
    901 		pack.ep_flags &= ~EXEC_HASFD;
    902 		(void) fdrelease(p, pack.ep_fd);
    903 	}
    904 	/* close and put the exec'd file */
    905 	vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY);
    906 	VOP_CLOSE(pack.ep_vp, FREAD, cred, p);
    907 	vput(pack.ep_vp);
    908 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
    909 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
    910 
    911  freehdr:
    912 	free(pack.ep_hdr, M_EXEC);
    913 
    914 #ifdef SYSTRACE
    915  clrflg:
    916 #endif /* SYSTRACE */
    917 	l->l_flag |= oldlwpflags;
    918 	p->p_flag &= ~P_INEXEC;
    919 #ifdef LKM
    920 	lockmgr(&exec_lock, LK_RELEASE, NULL);
    921 #endif
    922 
    923 	return error;
    924 
    925  exec_abort:
    926 	p->p_flag &= ~P_INEXEC;
    927 #ifdef LKM
    928 	lockmgr(&exec_lock, LK_RELEASE, NULL);
    929 #endif
    930 
    931 	/*
    932 	 * the old process doesn't exist anymore.  exit gracefully.
    933 	 * get rid of the (new) address space we have created, if any, get rid
    934 	 * of our namei data and vnode, and exit noting failure
    935 	 */
    936 	uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS,
    937 		VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS);
    938 	if (pack.ep_emul_arg)
    939 		FREE(pack.ep_emul_arg, M_TEMP);
    940 	PNBUF_PUT(nid.ni_cnd.cn_pnbuf);
    941 	uvm_km_free(exec_map, (vaddr_t) argp, NCARGS, UVM_KMF_PAGEABLE);
    942 	free(pack.ep_hdr, M_EXEC);
    943 	exit1(l, W_EXITCODE(error, SIGABRT));
    944 
    945 	/* NOTREACHED */
    946 	return 0;
    947 }
    948 
    949 
    950 int
    951 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo,
    952     char **stackp, void *argp)
    953 {
    954 	char	**cpp, *dp, *sp;
    955 	size_t	len;
    956 	void	*nullp;
    957 	long	argc, envc;
    958 	int	error;
    959 
    960 	cpp = (char **)*stackp;
    961 	nullp = NULL;
    962 	argc = arginfo->ps_nargvstr;
    963 	envc = arginfo->ps_nenvstr;
    964 	if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0)
    965 		return error;
    966 
    967 	dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen);
    968 	sp = argp;
    969 
    970 	/* XXX don't copy them out, remap them! */
    971 	arginfo->ps_argvstr = cpp; /* remember location of argv for later */
    972 
    973 	for (; --argc >= 0; sp += len, dp += len)
    974 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
    975 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
    976 			return error;
    977 
    978 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
    979 		return error;
    980 
    981 	arginfo->ps_envstr = cpp; /* remember location of envp for later */
    982 
    983 	for (; --envc >= 0; sp += len, dp += len)
    984 		if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 ||
    985 		    (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0)
    986 			return error;
    987 
    988 	if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0)
    989 		return error;
    990 
    991 	*stackp = (char *)cpp;
    992 	return 0;
    993 }
    994 
    995 #ifdef LKM
    996 /*
    997  * Find an emulation of given name in list of emulations.
    998  * Needs to be called with the exec_lock held.
    999  */
   1000 const struct emul *
   1001 emul_search(const char *name)
   1002 {
   1003 	struct emul_entry *it;
   1004 
   1005 	LIST_FOREACH(it, &el_head, el_list) {
   1006 		if (strcmp(name, it->el_emul->e_name) == 0)
   1007 			return it->el_emul;
   1008 	}
   1009 
   1010 	return NULL;
   1011 }
   1012 
   1013 /*
   1014  * Add an emulation to list, if it's not there already.
   1015  */
   1016 int
   1017 emul_register(const struct emul *emul, int ro_entry)
   1018 {
   1019 	struct emul_entry	*ee;
   1020 	int			error;
   1021 
   1022 	error = 0;
   1023 	lockmgr(&exec_lock, LK_SHARED, NULL);
   1024 
   1025 	if (emul_search(emul->e_name)) {
   1026 		error = EEXIST;
   1027 		goto out;
   1028 	}
   1029 
   1030 	MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry),
   1031 		M_EXEC, M_WAITOK);
   1032 	ee->el_emul = emul;
   1033 	ee->ro_entry = ro_entry;
   1034 	LIST_INSERT_HEAD(&el_head, ee, el_list);
   1035 
   1036  out:
   1037 	lockmgr(&exec_lock, LK_RELEASE, NULL);
   1038 	return error;
   1039 }
   1040 
   1041 /*
   1042  * Remove emulation with name 'name' from list of supported emulations.
   1043  */
   1044 int
   1045 emul_unregister(const char *name)
   1046 {
   1047 	const struct proclist_desc *pd;
   1048 	struct emul_entry	*it;
   1049 	int			i, error;
   1050 	struct proc		*ptmp;
   1051 
   1052 	error = 0;
   1053 	lockmgr(&exec_lock, LK_SHARED, NULL);
   1054 
   1055 	LIST_FOREACH(it, &el_head, el_list) {
   1056 		if (strcmp(it->el_emul->e_name, name) == 0)
   1057 			break;
   1058 	}
   1059 
   1060 	if (!it) {
   1061 		error = ENOENT;
   1062 		goto out;
   1063 	}
   1064 
   1065 	if (it->ro_entry) {
   1066 		error = EBUSY;
   1067 		goto out;
   1068 	}
   1069 
   1070 	/* test if any execw[] entry is still using this */
   1071 	for(i=0; i < nexecs; i++) {
   1072 		if (execsw[i]->es_emul == it->el_emul) {
   1073 			error = EBUSY;
   1074 			goto out;
   1075 		}
   1076 	}
   1077 
   1078 	/*
   1079 	 * Test if any process is running under this emulation - since
   1080 	 * emul_unregister() is running quite sendomly, it's better
   1081 	 * to do expensive check here than to use any locking.
   1082 	 */
   1083 	proclist_lock_read();
   1084 	for (pd = proclists; pd->pd_list != NULL && !error; pd++) {
   1085 		PROCLIST_FOREACH(ptmp, pd->pd_list) {
   1086 			if (ptmp->p_emul == it->el_emul) {
   1087 				error = EBUSY;
   1088 				break;
   1089 			}
   1090 		}
   1091 	}
   1092 	proclist_unlock_read();
   1093 
   1094 	if (error)
   1095 		goto out;
   1096 
   1097 
   1098 	/* entry is not used, remove it */
   1099 	LIST_REMOVE(it, el_list);
   1100 	FREE(it, M_EXEC);
   1101 
   1102  out:
   1103 	lockmgr(&exec_lock, LK_RELEASE, NULL);
   1104 	return error;
   1105 }
   1106 
   1107 /*
   1108  * Add execsw[] entry.
   1109  */
   1110 int
   1111 exec_add(struct execsw *esp, const char *e_name)
   1112 {
   1113 	struct exec_entry	*it;
   1114 	int			error;
   1115 
   1116 	error = 0;
   1117 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
   1118 
   1119 	if (!esp->es_emul) {
   1120 		esp->es_emul = emul_search(e_name);
   1121 		if (!esp->es_emul) {
   1122 			error = ENOENT;
   1123 			goto out;
   1124 		}
   1125 	}
   1126 
   1127 	LIST_FOREACH(it, &ex_head, ex_list) {
   1128 		/* assume tuple (makecmds, probe_func, emulation) is unique */
   1129 		if (it->es->es_makecmds == esp->es_makecmds
   1130 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
   1131 		    && it->es->es_emul == esp->es_emul) {
   1132 			error = EEXIST;
   1133 			goto out;
   1134 		}
   1135 	}
   1136 
   1137 	/* if we got here, the entry doesn't exist yet */
   1138 	MALLOC(it, struct exec_entry *, sizeof(struct exec_entry),
   1139 		M_EXEC, M_WAITOK);
   1140 	it->es = esp;
   1141 	LIST_INSERT_HEAD(&ex_head, it, ex_list);
   1142 
   1143 	/* update execsw[] */
   1144 	exec_init(0);
   1145 
   1146  out:
   1147 	lockmgr(&exec_lock, LK_RELEASE, NULL);
   1148 	return error;
   1149 }
   1150 
   1151 /*
   1152  * Remove execsw[] entry.
   1153  */
   1154 int
   1155 exec_remove(const struct execsw *esp)
   1156 {
   1157 	struct exec_entry	*it;
   1158 	int			error;
   1159 
   1160 	error = 0;
   1161 	lockmgr(&exec_lock, LK_EXCLUSIVE, NULL);
   1162 
   1163 	LIST_FOREACH(it, &ex_head, ex_list) {
   1164 		/* assume tuple (makecmds, probe_func, emulation) is unique */
   1165 		if (it->es->es_makecmds == esp->es_makecmds
   1166 		    && it->es->u.elf_probe_func == esp->u.elf_probe_func
   1167 		    && it->es->es_emul == esp->es_emul)
   1168 			break;
   1169 	}
   1170 	if (!it) {
   1171 		error = ENOENT;
   1172 		goto out;
   1173 	}
   1174 
   1175 	/* remove item from list and free resources */
   1176 	LIST_REMOVE(it, ex_list);
   1177 	FREE(it, M_EXEC);
   1178 
   1179 	/* update execsw[] */
   1180 	exec_init(0);
   1181 
   1182  out:
   1183 	lockmgr(&exec_lock, LK_RELEASE, NULL);
   1184 	return error;
   1185 }
   1186 
   1187 static void
   1188 link_es(struct execsw_entry **listp, const struct execsw *esp)
   1189 {
   1190 	struct execsw_entry *et, *e1;
   1191 
   1192 	MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry),
   1193 			M_TEMP, M_WAITOK);
   1194 	et->next = NULL;
   1195 	et->es = esp;
   1196 	if (*listp == NULL) {
   1197 		*listp = et;
   1198 		return;
   1199 	}
   1200 
   1201 	switch(et->es->es_prio) {
   1202 	case EXECSW_PRIO_FIRST:
   1203 		/* put new entry as the first */
   1204 		et->next = *listp;
   1205 		*listp = et;
   1206 		break;
   1207 	case EXECSW_PRIO_ANY:
   1208 		/* put new entry after all *_FIRST and *_ANY entries */
   1209 		for(e1 = *listp; e1->next
   1210 			&& e1->next->es->es_prio != EXECSW_PRIO_LAST;
   1211 			e1 = e1->next);
   1212 		et->next = e1->next;
   1213 		e1->next = et;
   1214 		break;
   1215 	case EXECSW_PRIO_LAST:
   1216 		/* put new entry as the last one */
   1217 		for(e1 = *listp; e1->next; e1 = e1->next);
   1218 		e1->next = et;
   1219 		break;
   1220 	default:
   1221 #ifdef DIAGNOSTIC
   1222 		panic("execw[] entry with unknown priority %d found",
   1223 			et->es->es_prio);
   1224 #endif
   1225 		break;
   1226 	}
   1227 }
   1228 
   1229 /*
   1230  * Initialize exec structures. If init_boot is true, also does necessary
   1231  * one-time initialization (it's called from main() that way).
   1232  * Once system is multiuser, this should be called with exec_lock held,
   1233  * i.e. via exec_{add|remove}().
   1234  */
   1235 int
   1236 exec_init(int init_boot)
   1237 {
   1238 	const struct execsw	**new_es, * const *old_es;
   1239 	struct execsw_entry	*list, *e1;
   1240 	struct exec_entry	*e2;
   1241 	int			i, es_sz;
   1242 
   1243 	if (init_boot) {
   1244 		/* do one-time initializations */
   1245 		lockinit(&exec_lock, PWAIT, "execlck", 0, 0);
   1246 
   1247 		/* register compiled-in emulations */
   1248 		for(i=0; i < nexecs_builtin; i++) {
   1249 			if (execsw_builtin[i].es_emul)
   1250 				emul_register(execsw_builtin[i].es_emul, 1);
   1251 		}
   1252 #ifdef DIAGNOSTIC
   1253 		if (i == 0)
   1254 			panic("no emulations found in execsw_builtin[]");
   1255 #endif
   1256 	}
   1257 
   1258 	/*
   1259 	 * Build execsw[] array from builtin entries and entries added
   1260 	 * at runtime.
   1261 	 */
   1262 	list = NULL;
   1263 	for(i=0; i < nexecs_builtin; i++)
   1264 		link_es(&list, &execsw_builtin[i]);
   1265 
   1266 	/* Add dynamically loaded entries */
   1267 	es_sz = nexecs_builtin;
   1268 	LIST_FOREACH(e2, &ex_head, ex_list) {
   1269 		link_es(&list, e2->es);
   1270 		es_sz++;
   1271 	}
   1272 
   1273 	/*
   1274 	 * Now that we have sorted all execw entries, create new execsw[]
   1275 	 * and free no longer needed memory in the process.
   1276 	 */
   1277 	new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK);
   1278 	for(i=0; list; i++) {
   1279 		new_es[i] = list->es;
   1280 		e1 = list->next;
   1281 		FREE(list, M_TEMP);
   1282 		list = e1;
   1283 	}
   1284 
   1285 	/*
   1286 	 * New execsw[] array built, now replace old execsw[] and free
   1287 	 * used memory.
   1288 	 */
   1289 	old_es = execsw;
   1290 	execsw = new_es;
   1291 	nexecs = es_sz;
   1292 	if (old_es)
   1293 		/*XXXUNCONST*/
   1294 		free(__UNCONST(old_es), M_EXEC);
   1295 
   1296 	/*
   1297 	 * Figure out the maximum size of an exec header.
   1298 	 */
   1299 	exec_maxhdrsz = 0;
   1300 	for (i = 0; i < nexecs; i++) {
   1301 		if (execsw[i]->es_hdrsz > exec_maxhdrsz)
   1302 			exec_maxhdrsz = execsw[i]->es_hdrsz;
   1303 	}
   1304 
   1305 	return 0;
   1306 }
   1307 #endif
   1308 
   1309 #ifndef LKM
   1310 /*
   1311  * Simplified exec_init() for kernels without LKMs. Only initialize
   1312  * exec_maxhdrsz and execsw[].
   1313  */
   1314 int
   1315 exec_init(int init_boot)
   1316 {
   1317 	int i;
   1318 
   1319 #ifdef DIAGNOSTIC
   1320 	if (!init_boot)
   1321 		panic("exec_init(): called with init_boot == 0");
   1322 #endif
   1323 
   1324 	/* do one-time initializations */
   1325 	nexecs = nexecs_builtin;
   1326 	execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK);
   1327 
   1328 	/*
   1329 	 * Fill in execsw[] and figure out the maximum size of an exec header.
   1330 	 */
   1331 	exec_maxhdrsz = 0;
   1332 	for(i=0; i < nexecs; i++) {
   1333 		execsw[i] = &execsw_builtin[i];
   1334 		if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz)
   1335 			exec_maxhdrsz = execsw_builtin[i].es_hdrsz;
   1336 	}
   1337 
   1338 	return 0;
   1339 
   1340 }
   1341 #endif /* !LKM */
   1342 
   1343 static int
   1344 exec_sigcode_map(struct proc *p, const struct emul *e)
   1345 {
   1346 	vaddr_t va;
   1347 	vsize_t sz;
   1348 	int error;
   1349 	struct uvm_object *uobj;
   1350 
   1351 	sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode;
   1352 
   1353 	if (e->e_sigobject == NULL || sz == 0) {
   1354 		return 0;
   1355 	}
   1356 
   1357 	/*
   1358 	 * If we don't have a sigobject for this emulation, create one.
   1359 	 *
   1360 	 * sigobject is an anonymous memory object (just like SYSV shared
   1361 	 * memory) that we keep a permanent reference to and that we map
   1362 	 * in all processes that need this sigcode. The creation is simple,
   1363 	 * we create an object, add a permanent reference to it, map it in
   1364 	 * kernel space, copy out the sigcode to it and unmap it.
   1365 	 * We map it with PROT_READ|PROT_EXEC into the process just
   1366 	 * the way sys_mmap() would map it.
   1367 	 */
   1368 
   1369 	uobj = *e->e_sigobject;
   1370 	if (uobj == NULL) {
   1371 		uobj = uao_create(sz, 0);
   1372 		(*uobj->pgops->pgo_reference)(uobj);
   1373 		va = vm_map_min(kernel_map);
   1374 		if ((error = uvm_map(kernel_map, &va, round_page(sz),
   1375 		    uobj, 0, 0,
   1376 		    UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW,
   1377 		    UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) {
   1378 			printf("kernel mapping failed %d\n", error);
   1379 			(*uobj->pgops->pgo_detach)(uobj);
   1380 			return (error);
   1381 		}
   1382 		memcpy((void *)va, e->e_sigcode, sz);
   1383 #ifdef PMAP_NEED_PROCWR
   1384 		pmap_procwr(&proc0, va, sz);
   1385 #endif
   1386 		uvm_unmap(kernel_map, va, va + round_page(sz));
   1387 		*e->e_sigobject = uobj;
   1388 	}
   1389 
   1390 	/* Just a hint to uvm_map where to put it. */
   1391 	va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr,
   1392 	    round_page(sz));
   1393 
   1394 #ifdef __alpha__
   1395 	/*
   1396 	 * Tru64 puts /sbin/loader at the end of user virtual memory,
   1397 	 * which causes the above calculation to put the sigcode at
   1398 	 * an invalid address.  Put it just below the text instead.
   1399 	 */
   1400 	if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) {
   1401 		va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz);
   1402 	}
   1403 #endif
   1404 
   1405 	(*uobj->pgops->pgo_reference)(uobj);
   1406 	error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz),
   1407 			uobj, 0, 0,
   1408 			UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE,
   1409 				    UVM_ADV_RANDOM, 0));
   1410 	if (error) {
   1411 		(*uobj->pgops->pgo_detach)(uobj);
   1412 		return (error);
   1413 	}
   1414 	p->p_sigctx.ps_sigcode = (void *)va;
   1415 	return (0);
   1416 }
   1417