xref: /freebsd/sys/kern/kern_exec.c (revision b4490c6e937e404832d1169ca7dbac3b2e712355)
19454b2d8SWarner Losh /*-
226f9a767SRodney W. Grimes  * Copyright (c) 1993, David Greenman
326f9a767SRodney W. Grimes  * All rights reserved.
4df8bae1dSRodney W. Grimes  *
5df8bae1dSRodney W. Grimes  * Redistribution and use in source and binary forms, with or without
6df8bae1dSRodney W. Grimes  * modification, are permitted provided that the following conditions
7df8bae1dSRodney W. Grimes  * are met:
8df8bae1dSRodney W. Grimes  * 1. Redistributions of source code must retain the above copyright
9df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer.
10df8bae1dSRodney W. Grimes  * 2. Redistributions in binary form must reproduce the above copyright
11df8bae1dSRodney W. Grimes  *    notice, this list of conditions and the following disclaimer in the
12df8bae1dSRodney W. Grimes  *    documentation and/or other materials provided with the distribution.
13df8bae1dSRodney W. Grimes  *
1426f9a767SRodney W. Grimes  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15df8bae1dSRodney W. Grimes  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16df8bae1dSRodney W. Grimes  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
1726f9a767SRodney W. Grimes  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18df8bae1dSRodney W. Grimes  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19df8bae1dSRodney W. Grimes  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20df8bae1dSRodney W. Grimes  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21df8bae1dSRodney W. Grimes  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22df8bae1dSRodney W. Grimes  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23df8bae1dSRodney W. Grimes  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24df8bae1dSRodney W. Grimes  * SUCH DAMAGE.
25df8bae1dSRodney W. Grimes  */
26df8bae1dSRodney W. Grimes 
27677b542eSDavid E. O'Brien #include <sys/cdefs.h>
28677b542eSDavid E. O'Brien __FBSDID("$FreeBSD$");
29677b542eSDavid E. O'Brien 
3012bc222eSJonathan Anderson #include "opt_capsicum.h"
314da0d332SPeter Wemm #include "opt_hwpmc_hooks.h"
321e9b3d91SPeter Wemm #include "opt_ktrace.h"
33f8a47341SAlan Cox #include "opt_vm.h"
341e9b3d91SPeter Wemm 
35df8bae1dSRodney W. Grimes #include <sys/param.h>
364a144410SRobert Watson #include <sys/capsicum.h>
3726f9a767SRodney W. Grimes #include <sys/systm.h>
3875b8b3b2SJohn Baldwin #include <sys/eventhandler.h>
39fb919e4dSMark Murray #include <sys/lock.h>
40fb919e4dSMark Murray #include <sys/mutex.h>
41d2d3e875SBruce Evans #include <sys/sysproto.h>
4226f9a767SRodney W. Grimes #include <sys/signalvar.h>
4326f9a767SRodney W. Grimes #include <sys/kernel.h>
4426f9a767SRodney W. Grimes #include <sys/mount.h>
45797f2d22SPoul-Henning Kamp #include <sys/filedesc.h>
46079cc25bSDavid Greenman #include <sys/fcntl.h>
4726f9a767SRodney W. Grimes #include <sys/acct.h>
4826f9a767SRodney W. Grimes #include <sys/exec.h>
4926f9a767SRodney W. Grimes #include <sys/imgact.h>
50e1743d02SSøren Schmidt #include <sys/imgact_elf.h>
5126f9a767SRodney W. Grimes #include <sys/wait.h>
52333ea485SGuido van Rooij #include <sys/malloc.h>
53acd3428bSRobert Watson #include <sys/priv.h>
54a794e791SBruce Evans #include <sys/proc.h>
552a024a2bSSean Eric Fagan #include <sys/pioctl.h>
56a794e791SBruce Evans #include <sys/namei.h>
576004362eSDavid Schultz #include <sys/resourcevar.h>
5889f6b863SAttilio Rao #include <sys/rwlock.h>
5944ad5475SJohn Baldwin #include <sys/sched.h>
605d217f17SJohn Birrell #include <sys/sdt.h>
6159c8bc40SAlan Cox #include <sys/sf_buf.h>
6284e0b075SDavid Xu #include <sys/syscallsubr.h>
631e1e0b44SSøren Schmidt #include <sys/sysent.h>
64797f2d22SPoul-Henning Kamp #include <sys/shm.h>
6599ac3bc8SPeter Wemm #include <sys/sysctl.h>
66a794e791SBruce Evans #include <sys/vnode.h>
67dfa7fd1dSEdward Tomasz Napierala #include <sys/stat.h>
68c781aea8SPeter Wemm #ifdef KTRACE
69c781aea8SPeter Wemm #include <sys/ktrace.h>
70c781aea8SPeter Wemm #endif
7126f9a767SRodney W. Grimes 
7226f9a767SRodney W. Grimes #include <vm/vm.h>
73efeaf95aSDavid Greenman #include <vm/vm_param.h>
74efeaf95aSDavid Greenman #include <vm/pmap.h>
751616db3cSJohn Dyson #include <vm/vm_page.h>
76efeaf95aSDavid Greenman #include <vm/vm_map.h>
7726f9a767SRodney W. Grimes #include <vm/vm_kern.h>
78efeaf95aSDavid Greenman #include <vm/vm_extern.h>
791ebd0c59SDavid Greenman #include <vm/vm_object.h>
801616db3cSJohn Dyson #include <vm/vm_pager.h>
8126f9a767SRodney W. Grimes 
82ebccf1e3SJoseph Koshy #ifdef	HWPMC_HOOKS
83ebccf1e3SJoseph Koshy #include <sys/pmckern.h>
84ebccf1e3SJoseph Koshy #endif
85ebccf1e3SJoseph Koshy 
8626f9a767SRodney W. Grimes #include <machine/reg.h>
8726f9a767SRodney W. Grimes 
88ae1078d6SWayne Salamon #include <security/audit/audit.h>
89aed55708SRobert Watson #include <security/mac/mac_framework.h>
90ae1078d6SWayne Salamon 
915d217f17SJohn Birrell #ifdef KDTRACE_HOOKS
925d217f17SJohn Birrell #include <sys/dtrace_bsd.h>
935d217f17SJohn Birrell dtrace_execexit_func_t	dtrace_fasttrap_exec;
945d217f17SJohn Birrell #endif
955d217f17SJohn Birrell 
965d217f17SJohn Birrell SDT_PROVIDER_DECLARE(proc);
97d9fae5abSAndriy Gapon SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
98d9fae5abSAndriy Gapon SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
99d9fae5abSAndriy Gapon SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
1005d217f17SJohn Birrell 
101b9df5231SPoul-Henning Kamp MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102b9df5231SPoul-Henning Kamp 
10305ba50f5SJake Burkholder static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
10405ba50f5SJake Burkholder static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
1055dadd17bSJake Burkholder static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106610ecfe0SMaxim Sobolev static int do_execve(struct thread *td, struct image_args *args,
107610ecfe0SMaxim Sobolev     struct mac *mac_p);
10805ba50f5SJake Burkholder 
1099701cd40SJohn Baldwin /* XXX This should be vm_size_t. */
11005ba50f5SJake Burkholder SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
11105ba50f5SJake Burkholder     NULL, 0, sysctl_kern_ps_strings, "LU", "");
112486bddb0SDoug Rabson 
1139701cd40SJohn Baldwin /* XXX This should be vm_size_t. */
114ff66f6a4SRobert Watson SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
115ff66f6a4SRobert Watson     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
11699ac3bc8SPeter Wemm 
1175dadd17bSJake Burkholder SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
1185dadd17bSJake Burkholder     NULL, 0, sysctl_kern_stackprot, "I", "");
1195dadd17bSJake Burkholder 
120b9df5231SPoul-Henning Kamp u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121c3699b5fSPeter Wemm SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
1229701cd40SJohn Baldwin     &ps_arg_cache_limit, 0, "");
123b9df5231SPoul-Henning Kamp 
124eda6009cSKonstantin Belousov static int disallow_high_osrel;
125eda6009cSKonstantin Belousov SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
126eda6009cSKonstantin Belousov     &disallow_high_osrel, 0,
127eda6009cSKonstantin Belousov     "Disallow execution of binaries built for higher version of the world");
128eda6009cSKonstantin Belousov 
129878adb85SBjoern A. Zeeb static int map_at_zero = 0;
130af3b2549SHans Petter Selasky SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
131878adb85SBjoern A. Zeeb     "Permit processes to map an object at virtual address 0.");
132878adb85SBjoern A. Zeeb 
13305ba50f5SJake Burkholder static int
13405ba50f5SJake Burkholder sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
13505ba50f5SJake Burkholder {
13605ba50f5SJake Burkholder 	struct proc *p;
137df7c361eSPeter Wemm 	int error;
13805ba50f5SJake Burkholder 
13905ba50f5SJake Burkholder 	p = curproc;
140a7bc3102SPeter Wemm #ifdef SCTL_MASK32
141a7bc3102SPeter Wemm 	if (req->flags & SCTL_MASK32) {
142df7c361eSPeter Wemm 		unsigned int val;
143df7c361eSPeter Wemm 		val = (unsigned int)p->p_sysent->sv_psstrings;
144df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &val, sizeof(val));
145df7c361eSPeter Wemm 	} else
146df7c361eSPeter Wemm #endif
147df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148df7c361eSPeter Wemm 		   sizeof(p->p_sysent->sv_psstrings));
149df7c361eSPeter Wemm 	return error;
15005ba50f5SJake Burkholder }
15105ba50f5SJake Burkholder 
15205ba50f5SJake Burkholder static int
15305ba50f5SJake Burkholder sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
15405ba50f5SJake Burkholder {
15505ba50f5SJake Burkholder 	struct proc *p;
156df7c361eSPeter Wemm 	int error;
15705ba50f5SJake Burkholder 
15805ba50f5SJake Burkholder 	p = curproc;
159a7bc3102SPeter Wemm #ifdef SCTL_MASK32
160a7bc3102SPeter Wemm 	if (req->flags & SCTL_MASK32) {
161df7c361eSPeter Wemm 		unsigned int val;
162df7c361eSPeter Wemm 		val = (unsigned int)p->p_sysent->sv_usrstack;
163df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &val, sizeof(val));
164df7c361eSPeter Wemm 	} else
165df7c361eSPeter Wemm #endif
166df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167df7c361eSPeter Wemm 		    sizeof(p->p_sysent->sv_usrstack));
168df7c361eSPeter Wemm 	return error;
16905ba50f5SJake Burkholder }
17005ba50f5SJake Burkholder 
1715dadd17bSJake Burkholder static int
1725dadd17bSJake Burkholder sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
1735dadd17bSJake Burkholder {
1745dadd17bSJake Burkholder 	struct proc *p;
1755dadd17bSJake Burkholder 
1765dadd17bSJake Burkholder 	p = curproc;
1775dadd17bSJake Burkholder 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
1785dadd17bSJake Burkholder 	    sizeof(p->p_sysent->sv_stackprot)));
1795dadd17bSJake Burkholder }
1805dadd17bSJake Burkholder 
18199ac3bc8SPeter Wemm /*
182aa855a59SPeter Wemm  * Each of the items is a pointer to a `const struct execsw', hence the
183aa855a59SPeter Wemm  * double pointer here.
184df8bae1dSRodney W. Grimes  */
185aa855a59SPeter Wemm static const struct execsw **execsw;
18626f9a767SRodney W. Grimes 
187ca46e90eSBruce Evans #ifndef _SYS_SYSPROTO_H_
188ca46e90eSBruce Evans struct execve_args {
189ca46e90eSBruce Evans 	char    *fname;
190ca46e90eSBruce Evans 	char    **argv;
191ca46e90eSBruce Evans 	char    **envv;
192ca46e90eSBruce Evans };
193ca46e90eSBruce Evans #endif
194ca46e90eSBruce Evans 
195ca46e90eSBruce Evans int
1967b445033SKonstantin Belousov sys_execve(struct thread *td, struct execve_args *uap)
197ca46e90eSBruce Evans {
198610ecfe0SMaxim Sobolev 	struct image_args args;
1997b445033SKonstantin Belousov 	struct vmspace *oldvmspace;
2007b445033SKonstantin Belousov 	int error;
201ca46e90eSBruce Evans 
2027b445033SKonstantin Belousov 	error = pre_execve(td, &oldvmspace);
2037b445033SKonstantin Belousov 	if (error != 0)
2047b445033SKonstantin Belousov 		return (error);
205610ecfe0SMaxim Sobolev 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
206610ecfe0SMaxim Sobolev 	    uap->argv, uap->envv);
207610ecfe0SMaxim Sobolev 	if (error == 0)
208610ecfe0SMaxim Sobolev 		error = kern_execve(td, &args, NULL);
2097b445033SKonstantin Belousov 	post_execve(td, error, oldvmspace);
210610ecfe0SMaxim Sobolev 	return (error);
211ca46e90eSBruce Evans }
212ca46e90eSBruce Evans 
213ca46e90eSBruce Evans #ifndef _SYS_SYSPROTO_H_
214632dbc19SKonstantin Belousov struct fexecve_args {
215632dbc19SKonstantin Belousov 	int	fd;
216632dbc19SKonstantin Belousov 	char	**argv;
217632dbc19SKonstantin Belousov 	char	**envv;
218632dbc19SKonstantin Belousov }
219632dbc19SKonstantin Belousov #endif
220632dbc19SKonstantin Belousov int
2218451d0ddSKip Macy sys_fexecve(struct thread *td, struct fexecve_args *uap)
222632dbc19SKonstantin Belousov {
223632dbc19SKonstantin Belousov 	struct image_args args;
2247b445033SKonstantin Belousov 	struct vmspace *oldvmspace;
2257b445033SKonstantin Belousov 	int error;
226632dbc19SKonstantin Belousov 
2277b445033SKonstantin Belousov 	error = pre_execve(td, &oldvmspace);
2287b445033SKonstantin Belousov 	if (error != 0)
2297b445033SKonstantin Belousov 		return (error);
230632dbc19SKonstantin Belousov 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
231632dbc19SKonstantin Belousov 	    uap->argv, uap->envv);
232632dbc19SKonstantin Belousov 	if (error == 0) {
233632dbc19SKonstantin Belousov 		args.fd = uap->fd;
234632dbc19SKonstantin Belousov 		error = kern_execve(td, &args, NULL);
235632dbc19SKonstantin Belousov 	}
2367b445033SKonstantin Belousov 	post_execve(td, error, oldvmspace);
237632dbc19SKonstantin Belousov 	return (error);
238632dbc19SKonstantin Belousov }
239632dbc19SKonstantin Belousov 
240632dbc19SKonstantin Belousov #ifndef _SYS_SYSPROTO_H_
241ca46e90eSBruce Evans struct __mac_execve_args {
242ca46e90eSBruce Evans 	char	*fname;
243ca46e90eSBruce Evans 	char	**argv;
244ca46e90eSBruce Evans 	char	**envv;
245ca46e90eSBruce Evans 	struct mac	*mac_p;
246ca46e90eSBruce Evans };
247ca46e90eSBruce Evans #endif
248ca46e90eSBruce Evans 
249ca46e90eSBruce Evans int
2507b445033SKonstantin Belousov sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
251ca46e90eSBruce Evans {
252ca46e90eSBruce Evans #ifdef MAC
253610ecfe0SMaxim Sobolev 	struct image_args args;
2547b445033SKonstantin Belousov 	struct vmspace *oldvmspace;
2557b445033SKonstantin Belousov 	int error;
256610ecfe0SMaxim Sobolev 
2577b445033SKonstantin Belousov 	error = pre_execve(td, &oldvmspace);
2587b445033SKonstantin Belousov 	if (error != 0)
2597b445033SKonstantin Belousov 		return (error);
260610ecfe0SMaxim Sobolev 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
261610ecfe0SMaxim Sobolev 	    uap->argv, uap->envv);
262610ecfe0SMaxim Sobolev 	if (error == 0)
263610ecfe0SMaxim Sobolev 		error = kern_execve(td, &args, uap->mac_p);
2647b445033SKonstantin Belousov 	post_execve(td, error, oldvmspace);
265610ecfe0SMaxim Sobolev 	return (error);
266ca46e90eSBruce Evans #else
267ca46e90eSBruce Evans 	return (ENOSYS);
268ca46e90eSBruce Evans #endif
269ca46e90eSBruce Evans }
270ca46e90eSBruce Evans 
27184e0b075SDavid Xu int
2727b445033SKonstantin Belousov pre_execve(struct thread *td, struct vmspace **oldvmspace)
273906ac69dSDavid Xu {
2747b445033SKonstantin Belousov 	struct proc *p;
275906ac69dSDavid Xu 	int error;
276906ac69dSDavid Xu 
2777b445033SKonstantin Belousov 	KASSERT(td == curthread, ("non-current thread %p", td));
2787b445033SKonstantin Belousov 	error = 0;
2797b445033SKonstantin Belousov 	p = td->td_proc;
2807b445033SKonstantin Belousov 	if ((p->p_flag & P_HADTHREADS) != 0) {
281906ac69dSDavid Xu 		PROC_LOCK(p);
2827b445033SKonstantin Belousov 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
2837b445033SKonstantin Belousov 			error = ERESTART;
284906ac69dSDavid Xu 		PROC_UNLOCK(p);
285906ac69dSDavid Xu 	}
2867b445033SKonstantin Belousov 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
2877b445033SKonstantin Belousov 	    ("nested execve"));
2887b445033SKonstantin Belousov 	*oldvmspace = p->p_vmspace;
2897b445033SKonstantin Belousov 	return (error);
2907b445033SKonstantin Belousov }
291906ac69dSDavid Xu 
2927b445033SKonstantin Belousov void
2937b445033SKonstantin Belousov post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
2947b445033SKonstantin Belousov {
2957b445033SKonstantin Belousov 	struct proc *p;
296906ac69dSDavid Xu 
2977b445033SKonstantin Belousov 	KASSERT(td == curthread, ("non-current thread %p", td));
2987b445033SKonstantin Belousov 	p = td->td_proc;
2997b445033SKonstantin Belousov 	if ((p->p_flag & P_HADTHREADS) != 0) {
300906ac69dSDavid Xu 		PROC_LOCK(p);
301906ac69dSDavid Xu 		/*
302906ac69dSDavid Xu 		 * If success, we upgrade to SINGLE_EXIT state to
303906ac69dSDavid Xu 		 * force other threads to suicide.
304906ac69dSDavid Xu 		 */
305906ac69dSDavid Xu 		if (error == 0)
3066ddcc233SKonstantin Belousov 			thread_single(p, SINGLE_EXIT);
307906ac69dSDavid Xu 		else
3086ddcc233SKonstantin Belousov 			thread_single_end(p, SINGLE_BOUNDARY);
309906ac69dSDavid Xu 		PROC_UNLOCK(p);
310906ac69dSDavid Xu 	}
3117032434eSKonstantin Belousov 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
3127b445033SKonstantin Belousov 		KASSERT(p->p_vmspace != oldvmspace,
3137032434eSKonstantin Belousov 		    ("oldvmspace still used"));
3147032434eSKonstantin Belousov 		vmspace_free(oldvmspace);
3157032434eSKonstantin Belousov 		td->td_pflags &= ~TDP_EXECVMSPC;
3167032434eSKonstantin Belousov 	}
3177b445033SKonstantin Belousov }
318906ac69dSDavid Xu 
3197b445033SKonstantin Belousov /*
3207b445033SKonstantin Belousov  * XXX: kern_execve has the astonishing property of not always returning to
3217b445033SKonstantin Belousov  * the caller.  If sufficiently bad things happen during the call to
3227b445033SKonstantin Belousov  * do_execve(), it can end up calling exit1(); as a result, callers must
3237b445033SKonstantin Belousov  * avoid doing anything which they might need to undo (e.g., allocating
3247b445033SKonstantin Belousov  * memory).
3257b445033SKonstantin Belousov  */
3267b445033SKonstantin Belousov int
3277b445033SKonstantin Belousov kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
3287b445033SKonstantin Belousov {
3297b445033SKonstantin Belousov 
3307b445033SKonstantin Belousov 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
3317b445033SKonstantin Belousov 	    args->begin_envv - args->begin_argv);
3327b445033SKonstantin Belousov 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
3337b445033SKonstantin Belousov 	    args->endp - args->begin_envv);
3347b445033SKonstantin Belousov 	return (do_execve(td, args, mac_p));
335906ac69dSDavid Xu }
336906ac69dSDavid Xu 
33726f9a767SRodney W. Grimes /*
338450ffb44SRobert Watson  * In-kernel implementation of execve().  All arguments are assumed to be
339450ffb44SRobert Watson  * userspace pointers from the passed thread.
34026f9a767SRodney W. Grimes  */
341450ffb44SRobert Watson static int
342610ecfe0SMaxim Sobolev do_execve(td, args, mac_p)
343b40ce416SJulian Elischer 	struct thread *td;
344610ecfe0SMaxim Sobolev 	struct image_args *args;
345670cb89bSRobert Watson 	struct mac *mac_p;
346df8bae1dSRodney W. Grimes {
347b40ce416SJulian Elischer 	struct proc *p = td->td_proc;
34869c9eff8SEd Schouten 	struct nameidata nd;
3499b3b1c5fSJohn Baldwin 	struct ucred *newcred = NULL, *oldcred;
350350d5181SMateusz Guzik 	struct uidinfo *euip = NULL;
351654f6be1SBruce Evans 	register_t *stack_base;
3525ca4819dSJohn Baldwin 	int error, i;
353c52007c2SDavid Greenman 	struct image_params image_params, *imgp;
3549f5c1d19SDiomidis Spinellis 	struct vattr attr;
3554d77a549SAlfred Perlstein 	int (*img_first)(struct image_params *);
35669be5db9SAlfred Perlstein 	struct pargs *oldargs = NULL, *newargs = NULL;
35790af4afaSJohn Baldwin 	struct sigacts *oldsigacts, *newsigacts;
3586c84de02SJohn Baldwin #ifdef KTRACE
3596c84de02SJohn Baldwin 	struct vnode *tracevp = NULL;
360a5881ea5SJohn Baldwin 	struct ucred *tracecred = NULL;
3616c84de02SJohn Baldwin #endif
36261617058SMateusz Guzik 	struct vnode *oldtextvp = NULL, *newtextvp;
3637008be5bSPawel Jakub Dawidek 	cap_rights_t rights;
364d06c0d4dSRobert Watson 	int credential_changing;
365619eb6e5SJeff Roberson 	int textset;
366ccafe7ebSRobert Watson #ifdef MAC
3676356dba0SRobert Watson 	struct label *interpvplabel = NULL;
368eca8a663SRobert Watson 	int will_transition;
369ccafe7ebSRobert Watson #endif
37015139246SJoseph Koshy #ifdef HWPMC_HOOKS
37115139246SJoseph Koshy 	struct pmckern_procexec pe;
37215139246SJoseph Koshy #endif
373632dbc19SKonstantin Belousov 	static const char fexecv_proc_title[] = "(fexecv)";
37426f9a767SRodney W. Grimes 
375c52007c2SDavid Greenman 	imgp = &image_params;
376df8bae1dSRodney W. Grimes 
3779ca45e81SDag-Erling Smørgrav 	/*
3789ca45e81SDag-Erling Smørgrav 	 * Lock the process and set the P_INEXEC flag to indicate that
3799ca45e81SDag-Erling Smørgrav 	 * it should be left alone until we're done here.  This is
3809ca45e81SDag-Erling Smørgrav 	 * necessary to avoid race conditions - e.g. in ptrace() -
3819ca45e81SDag-Erling Smørgrav 	 * that might allow a local user to illicitly obtain elevated
3829ca45e81SDag-Erling Smørgrav 	 * privileges.
3839ca45e81SDag-Erling Smørgrav 	 */
3849ca45e81SDag-Erling Smørgrav 	PROC_LOCK(p);
3859ca45e81SDag-Erling Smørgrav 	KASSERT((p->p_flag & P_INEXEC) == 0,
38691f91617SDavid E. O'Brien 	    ("%s(): process already has P_INEXEC flag", __func__));
3879ca45e81SDag-Erling Smørgrav 	p->p_flag |= P_INEXEC;
3889ca45e81SDag-Erling Smørgrav 	PROC_UNLOCK(p);
3899ca45e81SDag-Erling Smørgrav 
390df8bae1dSRodney W. Grimes 	/*
391c52007c2SDavid Greenman 	 * Initialize part of the common data
392df8bae1dSRodney W. Grimes 	 */
3938e572983SMateusz Guzik 	bzero(imgp, sizeof(*imgp));
394c52007c2SDavid Greenman 	imgp->proc = p;
395c52007c2SDavid Greenman 	imgp->attr = &attr;
396610ecfe0SMaxim Sobolev 	imgp->args = args;
39726f9a767SRodney W. Grimes 
398670cb89bSRobert Watson #ifdef MAC
399eca8a663SRobert Watson 	error = mac_execve_enter(imgp, mac_p);
400532c491bSJeff Roberson 	if (error)
401670cb89bSRobert Watson 		goto exec_fail;
402670cb89bSRobert Watson #endif
403670cb89bSRobert Watson 
40426f9a767SRodney W. Grimes 	/*
40526f9a767SRodney W. Grimes 	 * Translate the file name. namei() returns a vnode pointer
40626f9a767SRodney W. Grimes 	 *	in ni_vp amoung other things.
407ae1078d6SWayne Salamon 	 *
408ae1078d6SWayne Salamon 	 * XXXAUDIT: It would be desirable to also audit the name of the
409ae1078d6SWayne Salamon 	 * interpreter if this is an interpreted binary.
41026f9a767SRodney W. Grimes 	 */
411632dbc19SKonstantin Belousov 	if (args->fname != NULL) {
41269c9eff8SEd Schouten 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
4135050aa86SKonstantin Belousov 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
414632dbc19SKonstantin Belousov 	}
41526f9a767SRodney W. Grimes 
416e506f34bSCraig Rodrigues 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
417e506f34bSCraig Rodrigues 
41826f9a767SRodney W. Grimes interpret:
419632dbc19SKonstantin Belousov 	if (args->fname != NULL) {
42012bc222eSJonathan Anderson #ifdef CAPABILITY_MODE
42112bc222eSJonathan Anderson 		/*
42212bc222eSJonathan Anderson 		 * While capability mode can't reach this point via direct
42312bc222eSJonathan Anderson 		 * path arguments to execve(), we also don't allow
42412bc222eSJonathan Anderson 		 * interpreters to be used in capability mode (for now).
42512bc222eSJonathan Anderson 		 * Catch indirect lookups and return a permissions error.
42612bc222eSJonathan Anderson 		 */
42712bc222eSJonathan Anderson 		if (IN_CAPABILITY_MODE(td)) {
42812bc222eSJonathan Anderson 			error = ECAPMODE;
42912bc222eSJonathan Anderson 			goto exec_fail;
43012bc222eSJonathan Anderson 		}
43112bc222eSJonathan Anderson #endif
43269c9eff8SEd Schouten 		error = namei(&nd);
433610ecfe0SMaxim Sobolev 		if (error)
43426f9a767SRodney W. Grimes 			goto exec_fail;
43526f9a767SRodney W. Grimes 
43661617058SMateusz Guzik 		newtextvp = nd.ni_vp;
43761617058SMateusz Guzik 		imgp->vp = newtextvp;
438632dbc19SKonstantin Belousov 	} else {
43914961ba7SRobert Watson 		AUDIT_ARG_FD(args->fd);
440a9d2f8d8SRobert Watson 		/*
441c8e781f6SPawel Jakub Dawidek 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
442a9d2f8d8SRobert Watson 		 */
4437008be5bSPawel Jakub Dawidek 		error = fgetvp_exec(td, args->fd,
44461617058SMateusz Guzik 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
445632dbc19SKonstantin Belousov 		if (error)
446632dbc19SKonstantin Belousov 			goto exec_fail;
44761617058SMateusz Guzik 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
44861617058SMateusz Guzik 		AUDIT_ARG_VNODE1(newtextvp);
44961617058SMateusz Guzik 		imgp->vp = newtextvp;
450632dbc19SKonstantin Belousov 	}
45126f9a767SRodney W. Grimes 
45226f9a767SRodney W. Grimes 	/*
4539022e4adSDavid Greenman 	 * Check file permissions (also 'opens' file)
4549022e4adSDavid Greenman 	 */
455c52007c2SDavid Greenman 	error = exec_check_permissions(imgp);
456619eb6e5SJeff Roberson 	if (error)
45726f9a767SRodney W. Grimes 		goto exec_fail_dealloc;
458619eb6e5SJeff Roberson 
4598516dd18SPoul-Henning Kamp 	imgp->object = imgp->vp->v_object;
4608516dd18SPoul-Henning Kamp 	if (imgp->object != NULL)
4610b2ed1aeSJeff Roberson 		vm_object_reference(imgp->object);
46226f9a767SRodney W. Grimes 
463619eb6e5SJeff Roberson 	/*
464619eb6e5SJeff Roberson 	 * Set VV_TEXT now so no one can write to the executable while we're
465619eb6e5SJeff Roberson 	 * activating it.
466619eb6e5SJeff Roberson 	 *
467619eb6e5SJeff Roberson 	 * Remember if this was set before and unset it in case this is not
468619eb6e5SJeff Roberson 	 * actually an executable image.
469619eb6e5SJeff Roberson 	 */
470877d24acSKonstantin Belousov 	textset = VOP_IS_TEXT(imgp->vp);
471877d24acSKonstantin Belousov 	VOP_SET_TEXT(imgp->vp);
472619eb6e5SJeff Roberson 
4731616db3cSJohn Dyson 	error = exec_map_first_page(imgp);
4746d5a0a8cSDavid Greenman 	if (error)
47526f9a767SRodney W. Grimes 		goto exec_fail_dealloc;
47626f9a767SRodney W. Grimes 
477f231de47SKonstantin Belousov 	imgp->proc->p_osrel = 0;
47826f9a767SRodney W. Grimes 	/*
479d323ddf3SMatthew Dillon 	 *	If the current process has a special image activator it
480d323ddf3SMatthew Dillon 	 *	wants to try first, call it.   For example, emulating shell
481d323ddf3SMatthew Dillon 	 *	scripts differently.
48226f9a767SRodney W. Grimes 	 */
483d323ddf3SMatthew Dillon 	error = -1;
484d323ddf3SMatthew Dillon 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
485d323ddf3SMatthew Dillon 		error = img_first(imgp);
486d323ddf3SMatthew Dillon 
487d323ddf3SMatthew Dillon 	/*
488d323ddf3SMatthew Dillon 	 *	Loop through the list of image activators, calling each one.
489d323ddf3SMatthew Dillon 	 *	An activator returns -1 if there is no match, 0 on success,
490d323ddf3SMatthew Dillon 	 *	and an error otherwise.
491d323ddf3SMatthew Dillon 	 */
492d323ddf3SMatthew Dillon 	for (i = 0; error == -1 && execsw[i]; ++i) {
493d323ddf3SMatthew Dillon 		if (execsw[i]->ex_imgact == NULL ||
494d323ddf3SMatthew Dillon 		    execsw[i]->ex_imgact == img_first) {
495d323ddf3SMatthew Dillon 			continue;
496d323ddf3SMatthew Dillon 		}
497c52007c2SDavid Greenman 		error = (*execsw[i]->ex_imgact)(imgp);
498d323ddf3SMatthew Dillon 	}
499d323ddf3SMatthew Dillon 
500d323ddf3SMatthew Dillon 	if (error) {
501619eb6e5SJeff Roberson 		if (error == -1) {
502877d24acSKonstantin Belousov 			if (textset == 0)
503877d24acSKonstantin Belousov 				VOP_UNSET_TEXT(imgp->vp);
504d323ddf3SMatthew Dillon 			error = ENOEXEC;
505619eb6e5SJeff Roberson 		}
50626f9a767SRodney W. Grimes 		goto exec_fail_dealloc;
507d323ddf3SMatthew Dillon 	}
508d323ddf3SMatthew Dillon 
509d323ddf3SMatthew Dillon 	/*
510d323ddf3SMatthew Dillon 	 * Special interpreter operation, cleanup and loop up to try to
511d323ddf3SMatthew Dillon 	 * activate the interpreter.
512d323ddf3SMatthew Dillon 	 */
513c52007c2SDavid Greenman 	if (imgp->interpreted) {
5141616db3cSJohn Dyson 		exec_unmap_first_page(imgp);
515619eb6e5SJeff Roberson 		/*
516619eb6e5SJeff Roberson 		 * VV_TEXT needs to be unset for scripts.  There is a short
517619eb6e5SJeff Roberson 		 * period before we determine that something is a script where
518619eb6e5SJeff Roberson 		 * VV_TEXT will be set. The vnode lock is held over this
519619eb6e5SJeff Roberson 		 * entire period so nothing should illegitimately be blocked.
520619eb6e5SJeff Roberson 		 */
521877d24acSKonstantin Belousov 		VOP_UNSET_TEXT(imgp->vp);
522762e6b85SEivind Eklund 		/* free name buffer and old vnode */
523632dbc19SKonstantin Belousov 		if (args->fname != NULL)
52469c9eff8SEd Schouten 			NDFREE(&nd, NDF_ONLY_PNBUF);
525670cb89bSRobert Watson #ifdef MAC
52661617058SMateusz Guzik 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
527670cb89bSRobert Watson #endif
5289a75ea23SKonstantin Belousov 		if (imgp->opened) {
52961617058SMateusz Guzik 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
5309a75ea23SKonstantin Belousov 			imgp->opened = 0;
5319a75ea23SKonstantin Belousov 		}
53261617058SMateusz Guzik 		vput(newtextvp);
5330b2ed1aeSJeff Roberson 		vm_object_deallocate(imgp->object);
5340b2ed1aeSJeff Roberson 		imgp->object = NULL;
53526f9a767SRodney W. Grimes 		/* set new name to that of the interpreter */
5365050aa86SKonstantin Belousov 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
537b40ce416SJulian Elischer 		    UIO_SYSSPACE, imgp->interpreter_name, td);
538632dbc19SKonstantin Belousov 		args->fname = imgp->interpreter_name;
53926f9a767SRodney W. Grimes 		goto interpret;
54026f9a767SRodney W. Grimes 	}
54126f9a767SRodney W. Grimes 
54226f9a767SRodney W. Grimes 	/*
543ded7d39cSChristian S.J. Peron 	 * NB: We unlock the vnode here because it is believed that none
544ded7d39cSChristian S.J. Peron 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
545ded7d39cSChristian S.J. Peron 	 */
546ded7d39cSChristian S.J. Peron 	VOP_UNLOCK(imgp->vp, 0);
5473ff06357SKonstantin Belousov 
5483ff06357SKonstantin Belousov 	/*
5493ff06357SKonstantin Belousov 	 * Do the best to calculate the full path to the image file.
5503ff06357SKonstantin Belousov 	 */
5513ff06357SKonstantin Belousov 	if (imgp->auxargs != NULL &&
5523ff06357SKonstantin Belousov 	    ((args->fname != NULL && args->fname[0] == '/') ||
5533ff06357SKonstantin Belousov 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
5543ff06357SKonstantin Belousov 		imgp->execpath = args->fname;
5553ff06357SKonstantin Belousov 
556eda6009cSKonstantin Belousov 	if (disallow_high_osrel &&
557eda6009cSKonstantin Belousov 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
558eda6009cSKonstantin Belousov 		error = ENOEXEC;
559eda6009cSKonstantin Belousov 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
560eda6009cSKonstantin Belousov 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
561eda6009cSKonstantin Belousov 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
562eda6009cSKonstantin Belousov 		goto exec_fail_dealloc;
563eda6009cSKonstantin Belousov 	}
564eda6009cSKonstantin Belousov 
565ded7d39cSChristian S.J. Peron 	/*
56626f9a767SRodney W. Grimes 	 * Copy out strings (args and env) and initialize stack base
56726f9a767SRodney W. Grimes 	 */
5683ebc1248SPeter Wemm 	if (p->p_sysent->sv_copyout_strings)
5693ebc1248SPeter Wemm 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
5703ebc1248SPeter Wemm 	else
571c52007c2SDavid Greenman 		stack_base = exec_copyout_strings(imgp);
57226f9a767SRodney W. Grimes 
57326f9a767SRodney W. Grimes 	/*
5741e1e0b44SSøren Schmidt 	 * If custom stack fixup routine present for this process
5751e1e0b44SSøren Schmidt 	 * let it do the stack setup.
5761e1e0b44SSøren Schmidt 	 * Else stuff argument count as first item on stack
57726f9a767SRodney W. Grimes 	 */
578d6c847f3SBruce Evans 	if (p->p_sysent->sv_fixup != NULL)
579c52007c2SDavid Greenman 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
5801e1e0b44SSøren Schmidt 	else
581610ecfe0SMaxim Sobolev 		suword(--stack_base, imgp->args->argc);
58226f9a767SRodney W. Grimes 
583457f7e23SEd Schouten 	if (args->fdp != NULL) {
584457f7e23SEd Schouten 		/* Install a brand new file descriptor table. */
585457f7e23SEd Schouten 		fdinstall_remapped(td, args->fdp);
586457f7e23SEd Schouten 		args->fdp = NULL;
587457f7e23SEd Schouten 	} else {
588a78e8d2aSDavid Greenman 		/*
589457f7e23SEd Schouten 		 * Keep on using the existing file descriptor table. For
590457f7e23SEd Schouten 		 * security and other reasons, the file descriptor table
591457f7e23SEd Schouten 		 * cannot be shared after an exec.
592a78e8d2aSDavid Greenman 		 */
593b9d32c36SMateusz Guzik 		fdunshare(td);
594b0bc0cadSMateusz Guzik 		/* close files on exec */
595b0bc0cadSMateusz Guzik 		fdcloseexec(td);
596457f7e23SEd Schouten 	}
597a78e8d2aSDavid Greenman 
598333ea485SGuido van Rooij 	/*
5999b3b1c5fSJohn Baldwin 	 * Malloc things before we need locks.
6009b3b1c5fSJohn Baldwin 	 */
601610ecfe0SMaxim Sobolev 	i = imgp->args->begin_envv - imgp->args->begin_argv;
6025997cae9SDon Lewis 	/* Cache arguments if they fit inside our allowance */
6035997cae9SDon Lewis 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
6049b3b1c5fSJohn Baldwin 		newargs = pargs_alloc(i);
6055997cae9SDon Lewis 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
6065997cae9SDon Lewis 	}
6079b3b1c5fSJohn Baldwin 
6082974cc36SKonstantin Belousov 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
6099b3b1c5fSJohn Baldwin 
6109b3b1c5fSJohn Baldwin 	/*
611333ea485SGuido van Rooij 	 * For security and other reasons, signal handlers cannot
6128688bb93SJohn Baldwin 	 * be shared after an exec. The new process gets a copy of the old
613b2c3fa70SDima Dorfman 	 * handlers. In execsigs(), the new process will have its signals
614333ea485SGuido van Rooij 	 * reset.
615333ea485SGuido van Rooij 	 */
61690af4afaSJohn Baldwin 	if (sigacts_shared(p->p_sigacts)) {
61790af4afaSJohn Baldwin 		oldsigacts = p->p_sigacts;
61890af4afaSJohn Baldwin 		newsigacts = sigacts_alloc();
61990af4afaSJohn Baldwin 		sigacts_copy(newsigacts, oldsigacts);
620a6bad85eSMateusz Guzik 	} else {
62190af4afaSJohn Baldwin 		oldsigacts = NULL;
622a6bad85eSMateusz Guzik 		newsigacts = NULL; /* satisfy gcc */
623a6bad85eSMateusz Guzik 	}
624333ea485SGuido van Rooij 
625d00c8ea4SMateusz Guzik 	PROC_LOCK(p);
626d00c8ea4SMateusz Guzik 	if (oldsigacts)
627d00c8ea4SMateusz Guzik 		p->p_sigacts = newsigacts;
628350d5181SMateusz Guzik 	oldcred = p->p_ucred;
629fdf4e8b3SWarner Losh 	/* Stop profiling */
630fdf4e8b3SWarner Losh 	stopprofclock(p);
631fdf4e8b3SWarner Losh 
63226f9a767SRodney W. Grimes 	/* reset caught signals */
63326f9a767SRodney W. Grimes 	execsigs(p);
63426f9a767SRodney W. Grimes 
63526f9a767SRodney W. Grimes 	/* name this process - nameiexec(p, ndp) */
6365ca4819dSJohn Baldwin 	bzero(p->p_comm, sizeof(p->p_comm));
6375ca4819dSJohn Baldwin 	if (args->fname)
6385ca4819dSJohn Baldwin 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
6395ca4819dSJohn Baldwin 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
64061617058SMateusz Guzik 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
6415ca4819dSJohn Baldwin 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
642ca081fdbSJulian Elischer 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
64344ad5475SJohn Baldwin #ifdef KTR
64444ad5475SJohn Baldwin 	sched_clear_tdname(td);
64544ad5475SJohn Baldwin #endif
64626f9a767SRodney W. Grimes 
64726f9a767SRodney W. Grimes 	/*
64824b34f09SSujal Patel 	 * mark as execed, wakeup the process that vforked (if any) and tell
649dc733423SDag-Erling Smørgrav 	 * it that it now has its own resources back
65026f9a767SRodney W. Grimes 	 */
65126f9a767SRodney W. Grimes 	p->p_flag |= P_EXEC;
652677258f7SKonstantin Belousov 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
653677258f7SKonstantin Belousov 		p->p_flag2 &= ~P2_NOTRACE;
654965d0860SMateusz Guzik 	if (p->p_flag & P_PPWAIT) {
655888d4d4fSKonstantin Belousov 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
656aeb32571SKonstantin Belousov 		cv_broadcast(&p->p_pwait);
65726f9a767SRodney W. Grimes 	}
65826f9a767SRodney W. Grimes 
65926f9a767SRodney W. Grimes 	/*
660a78e8d2aSDavid Greenman 	 * Implement image setuid/setgid.
661a78e8d2aSDavid Greenman 	 *
662a78e8d2aSDavid Greenman 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
663a78e8d2aSDavid Greenman 	 * the process is being traced.
664670cb89bSRobert Watson 	 *
66512bc222eSJonathan Anderson 	 * We disable setuid/setgid/etc in compatibility mode on the basis
66612bc222eSJonathan Anderson 	 * that most setugid applications are not written with that
66712bc222eSJonathan Anderson 	 * environment in mind, and will therefore almost certainly operate
66812bc222eSJonathan Anderson 	 * incorrectly. In principle there's no reason that setugid
66912bc222eSJonathan Anderson 	 * applications might not be useful in capability mode, so we may want
67012bc222eSJonathan Anderson 	 * to reconsider this conservative design choice in the future.
67112bc222eSJonathan Anderson 	 *
672670cb89bSRobert Watson 	 * XXXMAC: For the time being, use NOSUID to also prohibit
673670cb89bSRobert Watson 	 * transitions on the file system.
674c52007c2SDavid Greenman 	 */
675d06c0d4dSRobert Watson 	credential_changing = 0;
676dfa7fd1dSEdward Tomasz Napierala 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
677d06c0d4dSRobert Watson 	    attr.va_uid;
678dfa7fd1dSEdward Tomasz Napierala 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
679d06c0d4dSRobert Watson 	    attr.va_gid;
680ccafe7ebSRobert Watson #ifdef MAC
68130d239bcSRobert Watson 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
6826356dba0SRobert Watson 	    interpvplabel, imgp);
683ccafe7ebSRobert Watson 	credential_changing |= will_transition;
684ccafe7ebSRobert Watson #endif
685d06c0d4dSRobert Watson 
686d06c0d4dSRobert Watson 	if (credential_changing &&
68712bc222eSJonathan Anderson #ifdef CAPABILITY_MODE
68812bc222eSJonathan Anderson 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
68912bc222eSJonathan Anderson #endif
690a78e8d2aSDavid Greenman 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
691c52007c2SDavid Greenman 	    (p->p_flag & P_TRACED) == 0) {
692c52007c2SDavid Greenman 		/*
693c52007c2SDavid Greenman 		 * Turn off syscall tracing for set-id programs, except for
694b85db196SPeter Wemm 		 * root.  Record any set-id flags first to make sure that
695b85db196SPeter Wemm 		 * we do not regain any tracing during a possible block.
69626f9a767SRodney W. Grimes 		 */
697b85db196SPeter Wemm 		setsugid(p);
698acd3428bSRobert Watson 
6996c84de02SJohn Baldwin #ifdef KTRACE
700c331c970SAndriy Gapon 		if (p->p_tracecred != NULL &&
701c331c970SAndriy Gapon 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
702d680caabSJohn Baldwin 			ktrprocexec(p, &tracecred, &tracevp);
7036c84de02SJohn Baldwin #endif
70428b325aaSDon Lewis 		/*
705c1e2d386SNate Lawson 		 * Close any file descriptors 0..2 that reference procfs,
706c1e2d386SNate Lawson 		 * then make sure file descriptors 0..2 are in use.
70728b325aaSDon Lewis 		 *
7088a5177ccSMateusz Guzik 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
7098a5177ccSMateusz Guzik 		 * taking sleepable locks, so temporarily drop our locks.
71028b325aaSDon Lewis 		 */
71128b325aaSDon Lewis 		PROC_UNLOCK(p);
71222db15c0SAttilio Rao 		VOP_UNLOCK(imgp->vp, 0);
71311888da8SMateusz Guzik 		fdsetugidsafety(td);
714e983a376SJacques Vidrine 		error = fdcheckstd(td);
71563c9e754SJohn Baldwin 		if (error != 0)
71669be5db9SAlfred Perlstein 			goto done1;
717350d5181SMateusz Guzik 		newcred = crdup(oldcred);
718350d5181SMateusz Guzik 		euip = uifind(attr.va_uid);
7195e2554b7SMateusz Guzik 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
720e1b1aa3bSJohn Baldwin 		PROC_LOCK(p);
721c52007c2SDavid Greenman 		/*
722c52007c2SDavid Greenman 		 * Set the new credentials.
723c52007c2SDavid Greenman 		 */
724dfa7fd1dSEdward Tomasz Napierala 		if (attr.va_mode & S_ISUID)
7251419eacbSAlfred Perlstein 			change_euid(newcred, euip);
726dfa7fd1dSEdward Tomasz Napierala 		if (attr.va_mode & S_ISGID)
727b1fc0ec1SRobert Watson 			change_egid(newcred, attr.va_gid);
728ccafe7ebSRobert Watson #ifdef MAC
729670cb89bSRobert Watson 		if (will_transition) {
73030d239bcSRobert Watson 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
7316356dba0SRobert Watson 			    interpvplabel, imgp);
732670cb89bSRobert Watson 		}
733ccafe7ebSRobert Watson #endif
7349b3b1c5fSJohn Baldwin 		/*
7359b3b1c5fSJohn Baldwin 		 * Implement correct POSIX saved-id behavior.
736ccafe7ebSRobert Watson 		 *
737ccafe7ebSRobert Watson 		 * XXXMAC: Note that the current logic will save the
738ccafe7ebSRobert Watson 		 * uid and gid if a MAC domain transition occurs, even
739ccafe7ebSRobert Watson 		 * though maybe it shouldn't.
7409b3b1c5fSJohn Baldwin 		 */
7419b3b1c5fSJohn Baldwin 		change_svuid(newcred, newcred->cr_uid);
7429b3b1c5fSJohn Baldwin 		change_svgid(newcred, newcred->cr_gid);
743daf63fd2SMateusz Guzik 		proc_set_cred(p, newcred);
744c52007c2SDavid Greenman 	} else {
745b1fc0ec1SRobert Watson 		if (oldcred->cr_uid == oldcred->cr_ruid &&
746b1fc0ec1SRobert Watson 		    oldcred->cr_gid == oldcred->cr_rgid)
747c52007c2SDavid Greenman 			p->p_flag &= ~P_SUGID;
74826f9a767SRodney W. Grimes 		/*
749c52007c2SDavid Greenman 		 * Implement correct POSIX saved-id behavior.
750b1fc0ec1SRobert Watson 		 *
751b1fc0ec1SRobert Watson 		 * XXX: It's not clear that the existing behavior is
7529b3b1c5fSJohn Baldwin 		 * POSIX-compliant.  A number of sources indicate that the
7539b3b1c5fSJohn Baldwin 		 * saved uid/gid should only be updated if the new ruid is
7549b3b1c5fSJohn Baldwin 		 * not equal to the old ruid, or the new euid is not equal
7559b3b1c5fSJohn Baldwin 		 * to the old euid and the new euid is not equal to the old
7569b3b1c5fSJohn Baldwin 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
7579b3b1c5fSJohn Baldwin 		 * Also, this code uses the new (replaced) euid and egid as
7589b3b1c5fSJohn Baldwin 		 * the source, which may or may not be the right ones to use.
75926f9a767SRodney W. Grimes 		 */
760b1fc0ec1SRobert Watson 		if (oldcred->cr_svuid != oldcred->cr_uid ||
761b1fc0ec1SRobert Watson 		    oldcred->cr_svgid != oldcred->cr_gid) {
762350d5181SMateusz Guzik 			PROC_UNLOCK(p);
7635e2554b7SMateusz Guzik 			VOP_UNLOCK(imgp->vp, 0);
764350d5181SMateusz Guzik 			newcred = crdup(oldcred);
7655e2554b7SMateusz Guzik 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
766350d5181SMateusz Guzik 			PROC_LOCK(p);
767b1fc0ec1SRobert Watson 			change_svuid(newcred, newcred->cr_uid);
768b1fc0ec1SRobert Watson 			change_svgid(newcred, newcred->cr_gid);
769daf63fd2SMateusz Guzik 			proc_set_cred(p, newcred);
7709b3b1c5fSJohn Baldwin 		}
771b1fc0ec1SRobert Watson 	}
77226f9a767SRodney W. Grimes 
77326f9a767SRodney W. Grimes 	/*
774853be5ffSMateusz Guzik 	 * Store the vp for use in procfs.  This vnode was referenced by namei
775853be5ffSMateusz Guzik 	 * or fgetvp_exec.
7762a531c80SDavid Greenman 	 */
77761617058SMateusz Guzik 	oldtextvp = p->p_textvp;
77861617058SMateusz Guzik 	p->p_textvp = newtextvp;
7792a531c80SDavid Greenman 
7805d217f17SJohn Birrell #ifdef KDTRACE_HOOKS
7815d217f17SJohn Birrell 	/*
7825d217f17SJohn Birrell 	 * Tell the DTrace fasttrap provider about the exec if it
7835d217f17SJohn Birrell 	 * has declared an interest.
7845d217f17SJohn Birrell 	 */
7855d217f17SJohn Birrell 	if (dtrace_fasttrap_exec)
7865d217f17SJohn Birrell 		dtrace_fasttrap_exec(p);
7875d217f17SJohn Birrell #endif
7885d217f17SJohn Birrell 
7892a531c80SDavid Greenman 	/*
7909ca45e81SDag-Erling Smørgrav 	 * Notify others that we exec'd, and clear the P_INEXEC flag
7919ca45e81SDag-Erling Smørgrav 	 * as we're now a bona fide freshly-execed process.
792cb679c38SJonathan Lemon 	 */
793ad3b9257SJohn-Mark Gurney 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
7949ca45e81SDag-Erling Smørgrav 	p->p_flag &= ~P_INEXEC;
795cb679c38SJonathan Lemon 
79626f9a767SRodney W. Grimes 	/* clear "fork but no exec" flag, as we _are_ execing */
79726f9a767SRodney W. Grimes 	p->p_acflag &= ~AFORK;
79826f9a767SRodney W. Grimes 
7995997cae9SDon Lewis 	/*
80034ea500bSDon Lewis 	 * Free any previous argument cache and replace it with
8015997cae9SDon Lewis 	 * the new argument cache, if any.
8025997cae9SDon Lewis 	 */
8039b3b1c5fSJohn Baldwin 	oldargs = p->p_args;
804e1b1aa3bSJohn Baldwin 	p->p_args = newargs;
805e1b1aa3bSJohn Baldwin 	newargs = NULL;
806ebccf1e3SJoseph Koshy 
807ebccf1e3SJoseph Koshy #ifdef	HWPMC_HOOKS
808ebccf1e3SJoseph Koshy 	/*
809f263522aSJoseph Koshy 	 * Check if system-wide sampling is in effect or if the
810f263522aSJoseph Koshy 	 * current process is using PMCs.  If so, do exec() time
811ebccf1e3SJoseph Koshy 	 * processing.  This processing needs to happen AFTER the
812ebccf1e3SJoseph Koshy 	 * P_INEXEC flag is cleared.
813ebccf1e3SJoseph Koshy 	 *
814ebccf1e3SJoseph Koshy 	 * The proc lock needs to be released before taking the PMC
815ebccf1e3SJoseph Koshy 	 * SX.
816ebccf1e3SJoseph Koshy 	 */
817f263522aSJoseph Koshy 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
818e1b1aa3bSJohn Baldwin 		PROC_UNLOCK(p);
8191ef6ea9bSKonstantin Belousov 		VOP_UNLOCK(imgp->vp, 0);
82015139246SJoseph Koshy 		pe.pm_credentialschanged = credential_changing;
82115139246SJoseph Koshy 		pe.pm_entryaddr = imgp->entry_addr;
82215139246SJoseph Koshy 
82315139246SJoseph Koshy 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
8242974cc36SKonstantin Belousov 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
825ebccf1e3SJoseph Koshy 	} else
826ebccf1e3SJoseph Koshy 		PROC_UNLOCK(p);
827ebccf1e3SJoseph Koshy #else  /* !HWPMC_HOOKS */
828ebccf1e3SJoseph Koshy 	PROC_UNLOCK(p);
829ebccf1e3SJoseph Koshy #endif
830e1b1aa3bSJohn Baldwin 
831e913ca22SDoug Rabson 	/* Set values passed into the program in registers. */
8323ebc1248SPeter Wemm 	if (p->p_sysent->sv_setregs)
833a107d8aaSNathan Whitehorn 		(*p->p_sysent->sv_setregs)(td, imgp,
834a107d8aaSNathan Whitehorn 		    (u_long)(uintptr_t)stack_base);
8353ebc1248SPeter Wemm 	else
836a107d8aaSNathan Whitehorn 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
837e913ca22SDoug Rabson 
8380359a12eSAttilio Rao 	vfs_mark_atime(imgp->vp, td->td_ucred);
8396341095eSKen Smith 
840d9fae5abSAndriy Gapon 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
8415d217f17SJohn Birrell 
8425e2554b7SMateusz Guzik 	VOP_UNLOCK(imgp->vp, 0);
843e506f34bSCraig Rodrigues done1:
8449b3b1c5fSJohn Baldwin 	/*
8459b3b1c5fSJohn Baldwin 	 * Free any resources malloc'd earlier that we didn't use.
8469b3b1c5fSJohn Baldwin 	 */
847350d5181SMateusz Guzik 	if (euip != NULL)
8481419eacbSAlfred Perlstein 		uifree(euip);
849350d5181SMateusz Guzik 	if (newcred != NULL)
8509b3b1c5fSJohn Baldwin 		crfree(oldcred);
8515d217f17SJohn Birrell 
8529b3b1c5fSJohn Baldwin 	/*
8539b3b1c5fSJohn Baldwin 	 * Handle deferred decrement of ref counts.
8549b3b1c5fSJohn Baldwin 	 */
85561617058SMateusz Guzik 	if (oldtextvp != NULL)
85661617058SMateusz Guzik 		vrele(oldtextvp);
8576c84de02SJohn Baldwin #ifdef KTRACE
8585050aa86SKonstantin Belousov 	if (tracevp != NULL)
8599b3b1c5fSJohn Baldwin 		vrele(tracevp);
860a5881ea5SJohn Baldwin 	if (tracecred != NULL)
861a5881ea5SJohn Baldwin 		crfree(tracecred);
8626c84de02SJohn Baldwin #endif
8632974cc36SKonstantin Belousov 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
8649b3b1c5fSJohn Baldwin 	pargs_drop(oldargs);
86569be5db9SAlfred Perlstein 	pargs_drop(newargs);
86690af4afaSJohn Baldwin 	if (oldsigacts != NULL)
86790af4afaSJohn Baldwin 		sigacts_free(oldsigacts);
868b9df5231SPoul-Henning Kamp 
8691616db3cSJohn Dyson exec_fail_dealloc:
8701616db3cSJohn Dyson 
87126f9a767SRodney W. Grimes 	/*
87226f9a767SRodney W. Grimes 	 * free various allocated resources
87326f9a767SRodney W. Grimes 	 */
874d6c847f3SBruce Evans 	if (imgp->firstpage != NULL)
8751616db3cSJohn Dyson 		exec_unmap_first_page(imgp);
87626f9a767SRodney W. Grimes 
877d6c847f3SBruce Evans 	if (imgp->vp != NULL) {
878632dbc19SKonstantin Belousov 		if (args->fname)
87969c9eff8SEd Schouten 			NDFREE(&nd, NDF_ONLY_PNBUF);
8809a75ea23SKonstantin Belousov 		if (imgp->opened)
8819a75ea23SKonstantin Belousov 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
882853be5ffSMateusz Guzik 		if (error != 0)
883619eb6e5SJeff Roberson 			vput(imgp->vp);
884853be5ffSMateusz Guzik 		else
885853be5ffSMateusz Guzik 			VOP_UNLOCK(imgp->vp, 0);
886a3cf6ebaSDavid Greenman 	}
88726f9a767SRodney W. Grimes 
888d6c847f3SBruce Evans 	if (imgp->object != NULL)
8890b2ed1aeSJeff Roberson 		vm_object_deallocate(imgp->object);
8900b2ed1aeSJeff Roberson 
8913ff06357SKonstantin Belousov 	free(imgp->freepath, M_TEMP);
8923ff06357SKonstantin Belousov 
893d1989db5SRobert Drehmel 	if (error == 0) {
894afe1a688SKonstantin Belousov 		PROC_LOCK(p);
895afe1a688SKonstantin Belousov 		td->td_dbgflags |= TDB_EXEC;
896afe1a688SKonstantin Belousov 		PROC_UNLOCK(p);
897afe1a688SKonstantin Belousov 
898d1989db5SRobert Drehmel 		/*
899d1989db5SRobert Drehmel 		 * Stop the process here if its stop event mask has
900d1989db5SRobert Drehmel 		 * the S_EXEC bit set.
901d1989db5SRobert Drehmel 		 */
902d1989db5SRobert Drehmel 		STOPEVENT(p, S_EXEC, 0);
903116734c4SMatthew Dillon 		goto done2;
904d1989db5SRobert Drehmel 	}
9051616db3cSJohn Dyson 
90626f9a767SRodney W. Grimes exec_fail:
9079ca45e81SDag-Erling Smørgrav 	/* we're done here, clear P_INEXEC */
9089ca45e81SDag-Erling Smørgrav 	PROC_LOCK(p);
9099ca45e81SDag-Erling Smørgrav 	p->p_flag &= ~P_INEXEC;
9109ca45e81SDag-Erling Smørgrav 	PROC_UNLOCK(p);
9119ca45e81SDag-Erling Smørgrav 
912d9fae5abSAndriy Gapon 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
9135d217f17SJohn Birrell 
914116734c4SMatthew Dillon done2:
915670cb89bSRobert Watson #ifdef MAC
916670cb89bSRobert Watson 	mac_execve_exit(imgp);
9176356dba0SRobert Watson 	mac_execve_interpreter_exit(interpvplabel);
918670cb89bSRobert Watson #endif
9198917b8d2SJohn Baldwin 	exec_free_args(args);
9208917b8d2SJohn Baldwin 
9218917b8d2SJohn Baldwin 	if (error && imgp->vmspace_destroyed) {
9228917b8d2SJohn Baldwin 		/* sorry, no more process anymore. exit gracefully */
923*b4490c6eSKonstantin Belousov 		exit1(td, 0, SIGABRT);
9248917b8d2SJohn Baldwin 		/* NOT REACHED */
9258917b8d2SJohn Baldwin 	}
9267705d4b2SDmitry Chagin 
9277705d4b2SDmitry Chagin #ifdef KTRACE
9287705d4b2SDmitry Chagin 	if (error == 0)
9297705d4b2SDmitry Chagin 		ktrprocctor(p);
9307705d4b2SDmitry Chagin #endif
9317705d4b2SDmitry Chagin 
932116734c4SMatthew Dillon 	return (error);
93326f9a767SRodney W. Grimes }
93426f9a767SRodney W. Grimes 
9351616db3cSJohn Dyson int
9361616db3cSJohn Dyson exec_map_first_page(imgp)
9371616db3cSJohn Dyson 	struct image_params *imgp;
9381616db3cSJohn Dyson {
939d8aad40cSJohn Baldwin 	int rv, i;
94095461b45SJohn Dyson 	int initial_pagein;
94195461b45SJohn Dyson 	vm_page_t ma[VM_INITIAL_PAGEIN];
9421616db3cSJohn Dyson 	vm_object_t object;
9431616db3cSJohn Dyson 
944d6c847f3SBruce Evans 	if (imgp->firstpage != NULL)
9451616db3cSJohn Dyson 		exec_unmap_first_page(imgp);
9461616db3cSJohn Dyson 
9478516dd18SPoul-Henning Kamp 	object = imgp->vp->v_object;
9484d7d2993SJeff Roberson 	if (object == NULL)
9494d7d2993SJeff Roberson 		return (EACCES);
95089f6b863SAttilio Rao 	VM_OBJECT_WLOCK(object);
951f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0
9523d653db0SAlan Cox 	vm_object_color(object, 0);
953f8a47341SAlan Cox #endif
9545944de8eSKonstantin Belousov 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
9550a2e596aSAlan Cox 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
95695461b45SJohn Dyson 		initial_pagein = VM_INITIAL_PAGEIN;
95795461b45SJohn Dyson 		if (initial_pagein > object->size)
95895461b45SJohn Dyson 			initial_pagein = object->size;
95995461b45SJohn Dyson 		for (i = 1; i < initial_pagein; i++) {
96041890423SAlan Cox 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
9616ec2fca5SAlan Cox 				if (ma[i]->valid)
9626ec2fca5SAlan Cox 					break;
963c7aebda8SAttilio Rao 				if (vm_page_tryxbusy(ma[i]))
96406fa71cdSAlan Cox 					break;
96595461b45SJohn Dyson 			} else {
966ca0387efSJake Burkholder 				ma[i] = vm_page_alloc(object, i,
9677bfda801SAlan Cox 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
96895461b45SJohn Dyson 				if (ma[i] == NULL)
96995461b45SJohn Dyson 					break;
97095461b45SJohn Dyson 			}
97195461b45SJohn Dyson 		}
97295461b45SJohn Dyson 		initial_pagein = i;
97395461b45SJohn Dyson 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
974093c7f39SGleb Smirnoff 		if (rv != VM_PAGER_OK) {
9755ac59343SAlan Cox 			vm_page_lock(ma[0]);
9768f9110f6SJohn Dyson 			vm_page_free(ma[0]);
9775ac59343SAlan Cox 			vm_page_unlock(ma[0]);
97889f6b863SAttilio Rao 			VM_OBJECT_WUNLOCK(object);
979a7cddfedSJake Burkholder 			return (EIO);
9801616db3cSJohn Dyson 		}
9811616db3cSJohn Dyson 	}
982c7aebda8SAttilio Rao 	vm_page_xunbusy(ma[0]);
9832965a453SKip Macy 	vm_page_lock(ma[0]);
984be996836SAttilio Rao 	vm_page_hold(ma[0]);
98570978c93SKonstantin Belousov 	vm_page_activate(ma[0]);
9862965a453SKip Macy 	vm_page_unlock(ma[0]);
98789f6b863SAttilio Rao 	VM_OBJECT_WUNLOCK(object);
9881616db3cSJohn Dyson 
98959c8bc40SAlan Cox 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
99059c8bc40SAlan Cox 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
9911616db3cSJohn Dyson 
992a7cddfedSJake Burkholder 	return (0);
9931616db3cSJohn Dyson }
9941616db3cSJohn Dyson 
9951616db3cSJohn Dyson void
9961616db3cSJohn Dyson exec_unmap_first_page(imgp)
9971616db3cSJohn Dyson 	struct image_params *imgp;
9981616db3cSJohn Dyson {
99959c8bc40SAlan Cox 	vm_page_t m;
100023955314SAlfred Perlstein 
1001d6c847f3SBruce Evans 	if (imgp->firstpage != NULL) {
100259c8bc40SAlan Cox 		m = sf_buf_page(imgp->firstpage);
100359c8bc40SAlan Cox 		sf_buf_free(imgp->firstpage);
10041616db3cSJohn Dyson 		imgp->firstpage = NULL;
10052965a453SKip Macy 		vm_page_lock(m);
1006be996836SAttilio Rao 		vm_page_unhold(m);
10072965a453SKip Macy 		vm_page_unlock(m);
10081616db3cSJohn Dyson 	}
10091616db3cSJohn Dyson }
10101616db3cSJohn Dyson 
101126f9a767SRodney W. Grimes /*
101226f9a767SRodney W. Grimes  * Destroy old address space, and allocate a new stack
101326f9a767SRodney W. Grimes  *	The new stack is only SGROWSIZ large because it is grown
101426f9a767SRodney W. Grimes  *	automatically in trap.c.
101526f9a767SRodney W. Grimes  */
101626f9a767SRodney W. Grimes int
101705ba50f5SJake Burkholder exec_new_vmspace(imgp, sv)
1018c52007c2SDavid Greenman 	struct image_params *imgp;
101905ba50f5SJake Burkholder 	struct sysentvec *sv;
102026f9a767SRodney W. Grimes {
102126f9a767SRodney W. Grimes 	int error;
10225e20c11fSAlan Cox 	struct proc *p = imgp->proc;
10235e20c11fSAlan Cox 	struct vmspace *vmspace = p->p_vmspace;
10246297a3d8SKonstantin Belousov 	vm_object_t obj;
1025316b3843SKonstantin Belousov 	struct rlimit rlim_stack;
1026878adb85SBjoern A. Zeeb 	vm_offset_t sv_minuser, stack_addr;
102705ba50f5SJake Burkholder 	vm_map_t map;
102859d8f3ffSJohn Baldwin 	u_long ssiz;
102926f9a767SRodney W. Grimes 
1030c52007c2SDavid Greenman 	imgp->vmspace_destroyed = 1;
1031993182e5SAlexander Leidinger 	imgp->sysent = sv;
103226f9a767SRodney W. Grimes 
1033bd37fd72SKris Kennaway 	/* May be called with Giant held */
1034993182e5SAlexander Leidinger 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
10356f5dafeaSAlan Cox 
10366f5dafeaSAlan Cox 	/*
1037af9ec885SJohn Dyson 	 * Blow away entire process VM, if address space not shared,
1038af9ec885SJohn Dyson 	 * otherwise, create a new VM space so that other threads are
1039af9ec885SJohn Dyson 	 * not disrupted
1040af9ec885SJohn Dyson 	 */
104105ba50f5SJake Burkholder 	map = &vmspace->vm_map;
1042878adb85SBjoern A. Zeeb 	if (map_at_zero)
1043878adb85SBjoern A. Zeeb 		sv_minuser = sv->sv_minuser;
1044878adb85SBjoern A. Zeeb 	else
1045878adb85SBjoern A. Zeeb 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1046878adb85SBjoern A. Zeeb 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
104705ba50f5SJake Burkholder 	    vm_map_max(map) == sv->sv_maxuser) {
10483db161e0SMatthew Dillon 		shmexit(vmspace);
1049b9eee07eSPeter Wemm 		pmap_remove_pages(vmspace_pmap(vmspace));
105005ba50f5SJake Burkholder 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1051af9ec885SJohn Dyson 	} else {
1052878adb85SBjoern A. Zeeb 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
105389b57fcfSKonstantin Belousov 		if (error)
105489b57fcfSKonstantin Belousov 			return (error);
10555e20c11fSAlan Cox 		vmspace = p->p_vmspace;
105605ba50f5SJake Burkholder 		map = &vmspace->vm_map;
1057af9ec885SJohn Dyson 	}
105826f9a767SRodney W. Grimes 
10596297a3d8SKonstantin Belousov 	/* Map a shared page */
10606297a3d8SKonstantin Belousov 	obj = sv->sv_shared_page_obj;
10616297a3d8SKonstantin Belousov 	if (obj != NULL) {
10626297a3d8SKonstantin Belousov 		vm_object_reference(obj);
10636297a3d8SKonstantin Belousov 		error = vm_map_fixed(map, obj, 0,
10646297a3d8SKonstantin Belousov 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
10651e65d73cSKonstantin Belousov 		    VM_PROT_READ | VM_PROT_EXECUTE,
10661e65d73cSKonstantin Belousov 		    VM_PROT_READ | VM_PROT_EXECUTE,
10671e65d73cSKonstantin Belousov 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
10686297a3d8SKonstantin Belousov 		if (error) {
10696297a3d8SKonstantin Belousov 			vm_object_deallocate(obj);
10706297a3d8SKonstantin Belousov 			return (error);
10716297a3d8SKonstantin Belousov 		}
10726297a3d8SKonstantin Belousov 	}
10736297a3d8SKonstantin Belousov 
107426f9a767SRodney W. Grimes 	/* Allocate a new stack */
1075316b3843SKonstantin Belousov 	if (imgp->stack_sz != 0) {
1076fe8a824cSKonstantin Belousov 		ssiz = trunc_page(imgp->stack_sz);
1077316b3843SKonstantin Belousov 		PROC_LOCK(p);
1078f6f6d240SMateusz Guzik 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1079316b3843SKonstantin Belousov 		PROC_UNLOCK(p);
1080316b3843SKonstantin Belousov 		if (ssiz > rlim_stack.rlim_max)
1081316b3843SKonstantin Belousov 			ssiz = rlim_stack.rlim_max;
1082316b3843SKonstantin Belousov 		if (ssiz > rlim_stack.rlim_cur) {
1083316b3843SKonstantin Belousov 			rlim_stack.rlim_cur = ssiz;
1084316b3843SKonstantin Belousov 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1085316b3843SKonstantin Belousov 		}
1086316b3843SKonstantin Belousov 	} else if (sv->sv_maxssiz != NULL) {
108759d8f3ffSJohn Baldwin 		ssiz = *sv->sv_maxssiz;
1088316b3843SKonstantin Belousov 	} else {
108959d8f3ffSJohn Baldwin 		ssiz = maxssiz;
1090316b3843SKonstantin Belousov 	}
109159d8f3ffSJohn Baldwin 	stack_addr = sv->sv_usrstack - ssiz;
109259d8f3ffSJohn Baldwin 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
10936297a3d8SKonstantin Belousov 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
10946297a3d8SKonstantin Belousov 		sv->sv_stackprot,
10956297a3d8SKonstantin Belousov 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
10962267af78SJulian Elischer 	if (error)
10972267af78SJulian Elischer 		return (error);
10982267af78SJulian Elischer 
109978960940SAlan Cox 	/*
110078960940SAlan Cox 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
110178960940SAlan Cox 	 * are still used to enforce the stack rlimit on the process stack.
11022267af78SJulian Elischer 	 */
1103cbc89bfbSPaul Saab 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
11046ef12002SKonstantin Belousov 	vmspace->vm_maxsaddr = (char *)stack_addr;
110526f9a767SRodney W. Grimes 
110626f9a767SRodney W. Grimes 	return (0);
110726f9a767SRodney W. Grimes }
110826f9a767SRodney W. Grimes 
110926f9a767SRodney W. Grimes /*
1110873fbcd7SRobert Watson  * Copy out argument and environment strings from the old process address
1111873fbcd7SRobert Watson  * space into the temporary string buffer.
111226f9a767SRodney W. Grimes  */
111326f9a767SRodney W. Grimes int
1114610ecfe0SMaxim Sobolev exec_copyin_args(struct image_args *args, char *fname,
1115610ecfe0SMaxim Sobolev     enum uio_seg segflg, char **argv, char **envv)
111626f9a767SRodney W. Grimes {
11170a2c94b8SKonstantin Belousov 	u_long argp, envp;
1118ecbb00a2SDoug Rabson 	int error;
1119ecbb00a2SDoug Rabson 	size_t length;
112026f9a767SRodney W. Grimes 
1121610ecfe0SMaxim Sobolev 	bzero(args, sizeof(*args));
1122610ecfe0SMaxim Sobolev 	if (argv == NULL)
1123610ecfe0SMaxim Sobolev 		return (EFAULT);
11242af6e14dSAlan Cox 
112526f9a767SRodney W. Grimes 	/*
11262af6e14dSAlan Cox 	 * Allocate demand-paged memory for the file name, argument, and
11272af6e14dSAlan Cox 	 * environment strings.
112826f9a767SRodney W. Grimes 	 */
11292af6e14dSAlan Cox 	error = exec_alloc_args(args);
11302af6e14dSAlan Cox 	if (error != 0)
11312af6e14dSAlan Cox 		return (error);
11329e4e5114SAlan Cox 
1133610ecfe0SMaxim Sobolev 	/*
1134610ecfe0SMaxim Sobolev 	 * Copy the file name.
1135610ecfe0SMaxim Sobolev 	 */
1136632dbc19SKonstantin Belousov 	if (fname != NULL) {
11372af6e14dSAlan Cox 		args->fname = args->buf;
1138610ecfe0SMaxim Sobolev 		error = (segflg == UIO_SYSSPACE) ?
1139610ecfe0SMaxim Sobolev 		    copystr(fname, args->fname, PATH_MAX, &length) :
1140610ecfe0SMaxim Sobolev 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1141c30af532SMaxim Sobolev 		if (error != 0)
114268ff3c24SStephan Uphoff 			goto err_exit;
1143632dbc19SKonstantin Belousov 	} else
11449e4e5114SAlan Cox 		length = 0;
11459e4e5114SAlan Cox 
11462af6e14dSAlan Cox 	args->begin_argv = args->buf + length;
11479e4e5114SAlan Cox 	args->endp = args->begin_argv;
11489e4e5114SAlan Cox 	args->stringspace = ARG_MAX;
114926f9a767SRodney W. Grimes 
1150610ecfe0SMaxim Sobolev 	/*
1151610ecfe0SMaxim Sobolev 	 * extract arguments first
1152610ecfe0SMaxim Sobolev 	 */
11530a2c94b8SKonstantin Belousov 	for (;;) {
11540a2c94b8SKonstantin Belousov 		error = fueword(argv++, &argp);
11550a2c94b8SKonstantin Belousov 		if (error == -1) {
115668ff3c24SStephan Uphoff 			error = EFAULT;
115768ff3c24SStephan Uphoff 			goto err_exit;
115868ff3c24SStephan Uphoff 		}
11590a2c94b8SKonstantin Belousov 		if (argp == 0)
11600a2c94b8SKonstantin Belousov 			break;
11610a2c94b8SKonstantin Belousov 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
11620a2c94b8SKonstantin Belousov 		    args->stringspace, &length);
11630a2c94b8SKonstantin Belousov 		if (error != 0) {
1164610ecfe0SMaxim Sobolev 			if (error == ENAMETOOLONG)
116568ff3c24SStephan Uphoff 				error = E2BIG;
116668ff3c24SStephan Uphoff 			goto err_exit;
1167610ecfe0SMaxim Sobolev 		}
1168610ecfe0SMaxim Sobolev 		args->stringspace -= length;
1169610ecfe0SMaxim Sobolev 		args->endp += length;
1170610ecfe0SMaxim Sobolev 		args->argc++;
1171610ecfe0SMaxim Sobolev 	}
1172610ecfe0SMaxim Sobolev 
1173610ecfe0SMaxim Sobolev 	args->begin_envv = args->endp;
1174b9df5231SPoul-Henning Kamp 
117526f9a767SRodney W. Grimes 	/*
117626f9a767SRodney W. Grimes 	 * extract environment strings
117726f9a767SRodney W. Grimes 	 */
11780fd1a014SDavid Greenman 	if (envv) {
11790a2c94b8SKonstantin Belousov 		for (;;) {
11800a2c94b8SKonstantin Belousov 			error = fueword(envv++, &envp);
11810a2c94b8SKonstantin Belousov 			if (error == -1) {
118268ff3c24SStephan Uphoff 				error = EFAULT;
118368ff3c24SStephan Uphoff 				goto err_exit;
118468ff3c24SStephan Uphoff 			}
11850a2c94b8SKonstantin Belousov 			if (envp == 0)
11860a2c94b8SKonstantin Belousov 				break;
11870a2c94b8SKonstantin Belousov 			error = copyinstr((void *)(uintptr_t)envp,
11880a2c94b8SKonstantin Belousov 			    args->endp, args->stringspace, &length);
11890a2c94b8SKonstantin Belousov 			if (error != 0) {
11900fd1a014SDavid Greenman 				if (error == ENAMETOOLONG)
119168ff3c24SStephan Uphoff 					error = E2BIG;
119268ff3c24SStephan Uphoff 				goto err_exit;
11930fd1a014SDavid Greenman 			}
1194610ecfe0SMaxim Sobolev 			args->stringspace -= length;
1195610ecfe0SMaxim Sobolev 			args->endp += length;
1196610ecfe0SMaxim Sobolev 			args->envc++;
119726f9a767SRodney W. Grimes 		}
11980fd1a014SDavid Greenman 	}
119926f9a767SRodney W. Grimes 
120026f9a767SRodney W. Grimes 	return (0);
120168ff3c24SStephan Uphoff 
120268ff3c24SStephan Uphoff err_exit:
120368ff3c24SStephan Uphoff 	exec_free_args(args);
120468ff3c24SStephan Uphoff 	return (error);
120526f9a767SRodney W. Grimes }
120626f9a767SRodney W. Grimes 
1207457f7e23SEd Schouten int
1208457f7e23SEd Schouten exec_copyin_data_fds(struct thread *td, struct image_args *args,
1209457f7e23SEd Schouten     const void *data, size_t datalen, const int *fds, size_t fdslen)
1210457f7e23SEd Schouten {
1211457f7e23SEd Schouten 	struct filedesc *ofdp;
1212457f7e23SEd Schouten 	const char *p;
1213457f7e23SEd Schouten 	int *kfds;
1214457f7e23SEd Schouten 	int error;
1215457f7e23SEd Schouten 
1216457f7e23SEd Schouten 	memset(args, '\0', sizeof(*args));
1217457f7e23SEd Schouten 	ofdp = td->td_proc->p_fd;
1218457f7e23SEd Schouten 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1219457f7e23SEd Schouten 		return (E2BIG);
1220457f7e23SEd Schouten 	error = exec_alloc_args(args);
1221457f7e23SEd Schouten 	if (error != 0)
1222457f7e23SEd Schouten 		return (error);
1223457f7e23SEd Schouten 
1224457f7e23SEd Schouten 	args->begin_argv = args->buf;
1225457f7e23SEd Schouten 	args->stringspace = ARG_MAX;
1226457f7e23SEd Schouten 
1227457f7e23SEd Schouten 	if (datalen > 0) {
1228457f7e23SEd Schouten 		/*
1229457f7e23SEd Schouten 		 * Argument buffer has been provided. Copy it into the
1230457f7e23SEd Schouten 		 * kernel as a single string and add a terminating null
1231457f7e23SEd Schouten 		 * byte.
1232457f7e23SEd Schouten 		 */
1233457f7e23SEd Schouten 		error = copyin(data, args->begin_argv, datalen);
1234457f7e23SEd Schouten 		if (error != 0)
1235457f7e23SEd Schouten 			goto err_exit;
1236457f7e23SEd Schouten 		args->begin_argv[datalen] = '\0';
1237457f7e23SEd Schouten 		args->endp = args->begin_argv + datalen + 1;
1238457f7e23SEd Schouten 		args->stringspace -= datalen + 1;
1239457f7e23SEd Schouten 
1240457f7e23SEd Schouten 		/*
1241457f7e23SEd Schouten 		 * Traditional argument counting. Count the number of
1242457f7e23SEd Schouten 		 * null bytes.
1243457f7e23SEd Schouten 		 */
1244457f7e23SEd Schouten 		for (p = args->begin_argv; p < args->endp; ++p)
1245457f7e23SEd Schouten 			if (*p == '\0')
1246457f7e23SEd Schouten 				++args->argc;
1247457f7e23SEd Schouten 	} else {
1248457f7e23SEd Schouten 		/* No argument buffer provided. */
1249457f7e23SEd Schouten 		args->endp = args->begin_argv;
1250457f7e23SEd Schouten 	}
1251457f7e23SEd Schouten 	/* There are no environment variables. */
1252457f7e23SEd Schouten 	args->begin_envv = args->endp;
1253457f7e23SEd Schouten 
1254457f7e23SEd Schouten 	/* Create new file descriptor table. */
1255457f7e23SEd Schouten 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1256457f7e23SEd Schouten 	error = copyin(fds, kfds, fdslen * sizeof(int));
1257457f7e23SEd Schouten 	if (error != 0) {
1258457f7e23SEd Schouten 		free(kfds, M_TEMP);
1259457f7e23SEd Schouten 		goto err_exit;
1260457f7e23SEd Schouten 	}
1261457f7e23SEd Schouten 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1262457f7e23SEd Schouten 	free(kfds, M_TEMP);
1263457f7e23SEd Schouten 	if (error != 0)
1264457f7e23SEd Schouten 		goto err_exit;
1265457f7e23SEd Schouten 
1266457f7e23SEd Schouten 	return (0);
1267457f7e23SEd Schouten err_exit:
1268457f7e23SEd Schouten 	exec_free_args(args);
1269457f7e23SEd Schouten 	return (error);
1270457f7e23SEd Schouten }
1271457f7e23SEd Schouten 
12722af6e14dSAlan Cox /*
12732af6e14dSAlan Cox  * Allocate temporary demand-paged, zero-filled memory for the file name,
12742af6e14dSAlan Cox  * argument, and environment strings.  Returns zero if the allocation succeeds
12752af6e14dSAlan Cox  * and ENOMEM otherwise.
12762af6e14dSAlan Cox  */
12772af6e14dSAlan Cox int
12782af6e14dSAlan Cox exec_alloc_args(struct image_args *args)
12792af6e14dSAlan Cox {
12802af6e14dSAlan Cox 
12815df87b21SJeff Roberson 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
12822af6e14dSAlan Cox 	return (args->buf != NULL ? 0 : ENOMEM);
12832af6e14dSAlan Cox }
12842af6e14dSAlan Cox 
128569a8f9e3SAlan Cox void
1286610ecfe0SMaxim Sobolev exec_free_args(struct image_args *args)
1287610ecfe0SMaxim Sobolev {
1288610ecfe0SMaxim Sobolev 
128969a8f9e3SAlan Cox 	if (args->buf != NULL) {
12905df87b21SJeff Roberson 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
12912af6e14dSAlan Cox 		    PATH_MAX + ARG_MAX);
1292610ecfe0SMaxim Sobolev 		args->buf = NULL;
1293610ecfe0SMaxim Sobolev 	}
1294a14a9498SAlan Cox 	if (args->fname_buf != NULL) {
1295a14a9498SAlan Cox 		free(args->fname_buf, M_TEMP);
1296a14a9498SAlan Cox 		args->fname_buf = NULL;
1297a14a9498SAlan Cox 	}
1298457f7e23SEd Schouten 	if (args->fdp != NULL)
1299457f7e23SEd Schouten 		fdescfree_remapped(args->fdp);
1300610ecfe0SMaxim Sobolev }
1301610ecfe0SMaxim Sobolev 
130226f9a767SRodney W. Grimes /*
1303873fbcd7SRobert Watson  * Copy strings out to the new process address space, constructing new arg
1304873fbcd7SRobert Watson  * and env vector tables. Return a pointer to the base so that it can be used
1305873fbcd7SRobert Watson  * as the initial stack pointer.
130626f9a767SRodney W. Grimes  */
1307654f6be1SBruce Evans register_t *
1308c52007c2SDavid Greenman exec_copyout_strings(imgp)
1309c52007c2SDavid Greenman 	struct image_params *imgp;
131026f9a767SRodney W. Grimes {
131126f9a767SRodney W. Grimes 	int argc, envc;
131226f9a767SRodney W. Grimes 	char **vectp;
131388b124ceSKonstantin Belousov 	char *stringp;
131488b124ceSKonstantin Belousov 	uintptr_t destp;
1315654f6be1SBruce Evans 	register_t *stack_base;
131693f6448cSDavid Greenman 	struct ps_strings *arginfo;
1317f3bec5d7SJake Burkholder 	struct proc *p;
13183ff06357SKonstantin Belousov 	size_t execpath_len;
1319ee235befSKonstantin Belousov 	int szsigcode, szps;
1320ee235befSKonstantin Belousov 	char canary[sizeof(long) * 8];
132126f9a767SRodney W. Grimes 
1322ee235befSKonstantin Belousov 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
132326f9a767SRodney W. Grimes 	/*
132426f9a767SRodney W. Grimes 	 * Calculate string base and vector table pointers.
1325d66a5066SPeter Wemm 	 * Also deal with signal trampoline code for this exec type.
132626f9a767SRodney W. Grimes 	 */
13273ff06357SKonstantin Belousov 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
13283ff06357SKonstantin Belousov 		execpath_len = strlen(imgp->execpath) + 1;
13293ff06357SKonstantin Belousov 	else
13303ff06357SKonstantin Belousov 		execpath_len = 0;
1331f3bec5d7SJake Burkholder 	p = imgp->proc;
1332f3bec5d7SJake Burkholder 	szsigcode = 0;
133305ba50f5SJake Burkholder 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
13346297a3d8SKonstantin Belousov 	if (p->p_sysent->sv_sigcode_base == 0) {
1335f3bec5d7SJake Burkholder 		if (p->p_sysent->sv_szsigcode != NULL)
1336f3bec5d7SJake Burkholder 			szsigcode = *(p->p_sysent->sv_szsigcode);
13376297a3d8SKonstantin Belousov 	}
133888b124ceSKonstantin Belousov 	destp =	(uintptr_t)arginfo;
13391ed012f9SPeter Wemm 
134026f9a767SRodney W. Grimes 	/*
1341d66a5066SPeter Wemm 	 * install sigcode
1342d66a5066SPeter Wemm 	 */
134388b124ceSKonstantin Belousov 	if (szsigcode != 0) {
134488b124ceSKonstantin Belousov 		destp -= szsigcode;
134588b124ceSKonstantin Belousov 		destp = rounddown2(destp, sizeof(void *));
134688b124ceSKonstantin Belousov 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
134788b124ceSKonstantin Belousov 	}
1348d66a5066SPeter Wemm 
1349d66a5066SPeter Wemm 	/*
13503ff06357SKonstantin Belousov 	 * Copy the image path for the rtld.
13513ff06357SKonstantin Belousov 	 */
13523ff06357SKonstantin Belousov 	if (execpath_len != 0) {
135388b124ceSKonstantin Belousov 		destp -= execpath_len;
135488b124ceSKonstantin Belousov 		imgp->execpathp = destp;
135588b124ceSKonstantin Belousov 		copyout(imgp->execpath, (void *)destp, execpath_len);
13563ff06357SKonstantin Belousov 	}
13573ff06357SKonstantin Belousov 
13583ff06357SKonstantin Belousov 	/*
1359ee235befSKonstantin Belousov 	 * Prepare the canary for SSP.
1360ee235befSKonstantin Belousov 	 */
1361ee235befSKonstantin Belousov 	arc4rand(canary, sizeof(canary), 0);
136288b124ceSKonstantin Belousov 	destp -= sizeof(canary);
136388b124ceSKonstantin Belousov 	imgp->canary = destp;
136488b124ceSKonstantin Belousov 	copyout(canary, (void *)destp, sizeof(canary));
1365ee235befSKonstantin Belousov 	imgp->canarylen = sizeof(canary);
1366ee235befSKonstantin Belousov 
1367ee235befSKonstantin Belousov 	/*
1368ee235befSKonstantin Belousov 	 * Prepare the pagesizes array.
1369ee235befSKonstantin Belousov 	 */
137088b124ceSKonstantin Belousov 	destp -= szps;
137188b124ceSKonstantin Belousov 	destp = rounddown2(destp, sizeof(void *));
137288b124ceSKonstantin Belousov 	imgp->pagesizes = destp;
137388b124ceSKonstantin Belousov 	copyout(pagesizes, (void *)destp, szps);
1374ee235befSKonstantin Belousov 	imgp->pagesizeslen = szps;
1375ee235befSKonstantin Belousov 
137688b124ceSKonstantin Belousov 	destp -= ARG_MAX - imgp->args->stringspace;
137788b124ceSKonstantin Belousov 	destp = rounddown2(destp, sizeof(void *));
137888b124ceSKonstantin Belousov 
1379ee235befSKonstantin Belousov 	/*
1380e1743d02SSøren Schmidt 	 * If we have a valid auxargs ptr, prepare some room
1381e1743d02SSøren Schmidt 	 * on the stack.
1382e1743d02SSøren Schmidt 	 */
1383b9a22da4STakanori Watanabe 	if (imgp->auxargs) {
1384e1743d02SSøren Schmidt 		/*
1385b9a22da4STakanori Watanabe 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1386b9a22da4STakanori Watanabe 		 * lower compatibility.
1387b9a22da4STakanori Watanabe 		 */
1388ca0387efSJake Burkholder 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1389ca0387efSJake Burkholder 		    (AT_COUNT * 2);
1390b9a22da4STakanori Watanabe 		/*
1391b9a22da4STakanori Watanabe 		 * The '+ 2' is for the null pointers at the end of each of
1392b9a22da4STakanori Watanabe 		 * the arg and env vector sets,and imgp->auxarg_size is room
1393b9a22da4STakanori Watanabe 		 * for argument of Runtime loader.
1394e1743d02SSøren Schmidt 		 */
1395610ecfe0SMaxim Sobolev 		vectp = (char **)(destp - (imgp->args->argc +
1396ee235befSKonstantin Belousov 		    imgp->args->envc + 2 + imgp->auxarg_size)
1397ee235befSKonstantin Belousov 		    * sizeof(char *));
1398610ecfe0SMaxim Sobolev 	} else {
1399e1743d02SSøren Schmidt 		/*
1400b9a22da4STakanori Watanabe 		 * The '+ 2' is for the null pointers at the end of each of
1401b9a22da4STakanori Watanabe 		 * the arg and env vector sets
140226f9a767SRodney W. Grimes 		 */
140388b124ceSKonstantin Belousov 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
140488b124ceSKonstantin Belousov 		    + 2) * sizeof(char *));
1405610ecfe0SMaxim Sobolev 	}
140626f9a767SRodney W. Grimes 
140726f9a767SRodney W. Grimes 	/*
140826f9a767SRodney W. Grimes 	 * vectp also becomes our initial stack base
140926f9a767SRodney W. Grimes 	 */
1410654f6be1SBruce Evans 	stack_base = (register_t *)vectp;
141126f9a767SRodney W. Grimes 
1412610ecfe0SMaxim Sobolev 	stringp = imgp->args->begin_argv;
1413610ecfe0SMaxim Sobolev 	argc = imgp->args->argc;
1414610ecfe0SMaxim Sobolev 	envc = imgp->args->envc;
141526f9a767SRodney W. Grimes 
141693f6448cSDavid Greenman 	/*
14173aed948bSDavid Greenman 	 * Copy out strings - arguments and environment.
141893f6448cSDavid Greenman 	 */
141988b124ceSKonstantin Belousov 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
142093f6448cSDavid Greenman 
142193f6448cSDavid Greenman 	/*
14223aed948bSDavid Greenman 	 * Fill in "ps_strings" struct for ps, w, etc.
14233aed948bSDavid Greenman 	 */
1424aae0aa45SBruce Evans 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1425f4e26adeSNathan Whitehorn 	suword32(&arginfo->ps_nargvstr, argc);
14263aed948bSDavid Greenman 
14273aed948bSDavid Greenman 	/*
14283aed948bSDavid Greenman 	 * Fill in argument portion of vector table.
142993f6448cSDavid Greenman 	 */
143026f9a767SRodney W. Grimes 	for (; argc > 0; --argc) {
1431aae0aa45SBruce Evans 		suword(vectp++, (long)(intptr_t)destp);
14323aed948bSDavid Greenman 		while (*stringp++ != 0)
14333aed948bSDavid Greenman 			destp++;
14343aed948bSDavid Greenman 		destp++;
143526f9a767SRodney W. Grimes 	}
143626f9a767SRodney W. Grimes 
14371a6e52d0SJeroen Ruigrok van der Werven 	/* a null vector table pointer separates the argp's from the envp's */
14386ab46d52SBruce Evans 	suword(vectp++, 0);
143926f9a767SRodney W. Grimes 
1440aae0aa45SBruce Evans 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1441f4e26adeSNathan Whitehorn 	suword32(&arginfo->ps_nenvstr, envc);
144293f6448cSDavid Greenman 
144393f6448cSDavid Greenman 	/*
14443aed948bSDavid Greenman 	 * Fill in environment portion of vector table.
144593f6448cSDavid Greenman 	 */
144626f9a767SRodney W. Grimes 	for (; envc > 0; --envc) {
1447aae0aa45SBruce Evans 		suword(vectp++, (long)(intptr_t)destp);
14483aed948bSDavid Greenman 		while (*stringp++ != 0)
14493aed948bSDavid Greenman 			destp++;
14503aed948bSDavid Greenman 		destp++;
145126f9a767SRodney W. Grimes 	}
145226f9a767SRodney W. Grimes 
145326f9a767SRodney W. Grimes 	/* end of vector table is a null pointer */
14546ab46d52SBruce Evans 	suword(vectp, 0);
145526f9a767SRodney W. Grimes 
145626f9a767SRodney W. Grimes 	return (stack_base);
145726f9a767SRodney W. Grimes }
145826f9a767SRodney W. Grimes 
145926f9a767SRodney W. Grimes /*
146026f9a767SRodney W. Grimes  * Check permissions of file to execute.
1461cf64863aSRobert Watson  *	Called with imgp->vp locked.
146226f9a767SRodney W. Grimes  *	Return 0 for success or error code on failure.
146326f9a767SRodney W. Grimes  */
1464c8a79999SPeter Wemm int
1465c52007c2SDavid Greenman exec_check_permissions(imgp)
1466c52007c2SDavid Greenman 	struct image_params *imgp;
146726f9a767SRodney W. Grimes {
1468c52007c2SDavid Greenman 	struct vnode *vp = imgp->vp;
1469c52007c2SDavid Greenman 	struct vattr *attr = imgp->attr;
1470a854ed98SJohn Baldwin 	struct thread *td;
1471140dedb8SKonstantin Belousov 	int error, writecount;
147226f9a767SRodney W. Grimes 
14736617724cSJeff Roberson 	td = curthread;
1474339b79b9SRobert Watson 
1475ec35c2afSRobert Watson 	/* Get file attributes */
14760359a12eSAttilio Rao 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1477ec35c2afSRobert Watson 	if (error)
1478ec35c2afSRobert Watson 		return (error);
1479ec35c2afSRobert Watson 
1480339b79b9SRobert Watson #ifdef MAC
148130d239bcSRobert Watson 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1482339b79b9SRobert Watson 	if (error)
1483339b79b9SRobert Watson 		return (error);
1484339b79b9SRobert Watson #endif
1485339b79b9SRobert Watson 
148626f9a767SRodney W. Grimes 	/*
1487de478dd4SJaakko Heinonen 	 * 1) Check if file execution is disabled for the filesystem that
1488de478dd4SJaakko Heinonen 	 *    this file resides on.
1489de478dd4SJaakko Heinonen 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1490de478dd4SJaakko Heinonen 	 *    privileged user will always succeed, and we don't want this
1491de478dd4SJaakko Heinonen 	 *    to happen unless the file really is executable.
1492de478dd4SJaakko Heinonen 	 * 3) Ensure that the file is a regular file.
149326f9a767SRodney W. Grimes 	 */
1494c52007c2SDavid Greenman 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1495de478dd4SJaakko Heinonen 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1496a854ed98SJohn Baldwin 	    (attr->va_type != VREG))
149726f9a767SRodney W. Grimes 		return (EACCES);
149826f9a767SRodney W. Grimes 
149926f9a767SRodney W. Grimes 	/*
150026f9a767SRodney W. Grimes 	 * Zero length files can't be exec'd
150126f9a767SRodney W. Grimes 	 */
150226f9a767SRodney W. Grimes 	if (attr->va_size == 0)
150326f9a767SRodney W. Grimes 		return (ENOEXEC);
150426f9a767SRodney W. Grimes 
150526f9a767SRodney W. Grimes 	/*
150626f9a767SRodney W. Grimes 	 *  Check for execute permission to file based on current credentials.
150726f9a767SRodney W. Grimes 	 */
1508a854ed98SJohn Baldwin 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
150926f9a767SRodney W. Grimes 	if (error)
151026f9a767SRodney W. Grimes 		return (error);
151126f9a767SRodney W. Grimes 
15126d5a0a8cSDavid Greenman 	/*
15136d5a0a8cSDavid Greenman 	 * Check number of open-for-writes on the file and deny execution
15146d5a0a8cSDavid Greenman 	 * if there are any.
15156d5a0a8cSDavid Greenman 	 */
1516140dedb8SKonstantin Belousov 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1517140dedb8SKonstantin Belousov 	if (error != 0)
1518140dedb8SKonstantin Belousov 		return (error);
1519140dedb8SKonstantin Belousov 	if (writecount != 0)
15206d5a0a8cSDavid Greenman 		return (ETXTBSY);
15216d5a0a8cSDavid Greenman 
15226d5a0a8cSDavid Greenman 	/*
15236d5a0a8cSDavid Greenman 	 * Call filesystem specific open routine (which does nothing in the
15246d5a0a8cSDavid Greenman 	 * general case).
15256d5a0a8cSDavid Greenman 	 */
15269e223287SKonstantin Belousov 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
15279a75ea23SKonstantin Belousov 	if (error == 0)
15289a75ea23SKonstantin Belousov 		imgp->opened = 1;
152926f9a767SRodney W. Grimes 	return (error);
1530df8bae1dSRodney W. Grimes }
1531aa855a59SPeter Wemm 
1532aa855a59SPeter Wemm /*
1533aa855a59SPeter Wemm  * Exec handler registration
1534aa855a59SPeter Wemm  */
1535aa855a59SPeter Wemm int
1536aa855a59SPeter Wemm exec_register(execsw_arg)
1537aa855a59SPeter Wemm 	const struct execsw *execsw_arg;
1538aa855a59SPeter Wemm {
1539aa855a59SPeter Wemm 	const struct execsw **es, **xs, **newexecsw;
1540aa855a59SPeter Wemm 	int count = 2;	/* New slot and trailing NULL */
1541aa855a59SPeter Wemm 
1542aa855a59SPeter Wemm 	if (execsw)
1543aa855a59SPeter Wemm 		for (es = execsw; *es; es++)
1544aa855a59SPeter Wemm 			count++;
1545a163d034SWarner Losh 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1546aa855a59SPeter Wemm 	if (newexecsw == NULL)
1547a7cddfedSJake Burkholder 		return (ENOMEM);
1548aa855a59SPeter Wemm 	xs = newexecsw;
1549aa855a59SPeter Wemm 	if (execsw)
1550aa855a59SPeter Wemm 		for (es = execsw; *es; es++)
1551aa855a59SPeter Wemm 			*xs++ = *es;
1552aa855a59SPeter Wemm 	*xs++ = execsw_arg;
1553aa855a59SPeter Wemm 	*xs = NULL;
1554aa855a59SPeter Wemm 	if (execsw)
1555aa855a59SPeter Wemm 		free(execsw, M_TEMP);
1556aa855a59SPeter Wemm 	execsw = newexecsw;
1557a7cddfedSJake Burkholder 	return (0);
1558aa855a59SPeter Wemm }
1559aa855a59SPeter Wemm 
1560aa855a59SPeter Wemm int
1561aa855a59SPeter Wemm exec_unregister(execsw_arg)
1562aa855a59SPeter Wemm 	const struct execsw *execsw_arg;
1563aa855a59SPeter Wemm {
1564aa855a59SPeter Wemm 	const struct execsw **es, **xs, **newexecsw;
1565aa855a59SPeter Wemm 	int count = 1;
1566aa855a59SPeter Wemm 
1567aa855a59SPeter Wemm 	if (execsw == NULL)
1568aa855a59SPeter Wemm 		panic("unregister with no handlers left?\n");
1569aa855a59SPeter Wemm 
1570aa855a59SPeter Wemm 	for (es = execsw; *es; es++) {
1571aa855a59SPeter Wemm 		if (*es == execsw_arg)
1572aa855a59SPeter Wemm 			break;
1573aa855a59SPeter Wemm 	}
1574aa855a59SPeter Wemm 	if (*es == NULL)
1575a7cddfedSJake Burkholder 		return (ENOENT);
1576aa855a59SPeter Wemm 	for (es = execsw; *es; es++)
1577aa855a59SPeter Wemm 		if (*es != execsw_arg)
1578aa855a59SPeter Wemm 			count++;
1579a163d034SWarner Losh 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1580aa855a59SPeter Wemm 	if (newexecsw == NULL)
1581a7cddfedSJake Burkholder 		return (ENOMEM);
1582aa855a59SPeter Wemm 	xs = newexecsw;
1583aa855a59SPeter Wemm 	for (es = execsw; *es; es++)
1584aa855a59SPeter Wemm 		if (*es != execsw_arg)
1585aa855a59SPeter Wemm 			*xs++ = *es;
1586aa855a59SPeter Wemm 	*xs = NULL;
1587aa855a59SPeter Wemm 	if (execsw)
1588aa855a59SPeter Wemm 		free(execsw, M_TEMP);
1589aa855a59SPeter Wemm 	execsw = newexecsw;
1590a7cddfedSJake Burkholder 	return (0);
1591aa855a59SPeter Wemm }
1592