xref: /freebsd/sys/kern/kern_exec.c (revision bcb60d52e65a5a77e9118d040d87e014372b66d9)
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 
103*bcb60d52SConrad Meyer int coredump_pack_fileinfo = 1;
104*bcb60d52SConrad Meyer SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
105*bcb60d52SConrad Meyer     &coredump_pack_fileinfo, 0,
106*bcb60d52SConrad Meyer     "Enable file path packing in 'procstat -f' coredump notes");
107*bcb60d52SConrad Meyer 
10805ba50f5SJake Burkholder static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
10905ba50f5SJake Burkholder static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
1105dadd17bSJake Burkholder static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
111610ecfe0SMaxim Sobolev static int do_execve(struct thread *td, struct image_args *args,
112610ecfe0SMaxim Sobolev     struct mac *mac_p);
11305ba50f5SJake Burkholder 
1149701cd40SJohn Baldwin /* XXX This should be vm_size_t. */
11505ba50f5SJake Burkholder SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
11605ba50f5SJake Burkholder     NULL, 0, sysctl_kern_ps_strings, "LU", "");
117486bddb0SDoug Rabson 
1189701cd40SJohn Baldwin /* XXX This should be vm_size_t. */
119ff66f6a4SRobert Watson SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
120ff66f6a4SRobert Watson     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
12199ac3bc8SPeter Wemm 
1225dadd17bSJake Burkholder SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
1235dadd17bSJake Burkholder     NULL, 0, sysctl_kern_stackprot, "I", "");
1245dadd17bSJake Burkholder 
125b9df5231SPoul-Henning Kamp u_long ps_arg_cache_limit = PAGE_SIZE / 16;
126c3699b5fSPeter Wemm SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
1279701cd40SJohn Baldwin     &ps_arg_cache_limit, 0, "");
128b9df5231SPoul-Henning Kamp 
129eda6009cSKonstantin Belousov static int disallow_high_osrel;
130eda6009cSKonstantin Belousov SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
131eda6009cSKonstantin Belousov     &disallow_high_osrel, 0,
132eda6009cSKonstantin Belousov     "Disallow execution of binaries built for higher version of the world");
133eda6009cSKonstantin Belousov 
134878adb85SBjoern A. Zeeb static int map_at_zero = 0;
135af3b2549SHans Petter Selasky SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
136878adb85SBjoern A. Zeeb     "Permit processes to map an object at virtual address 0.");
137878adb85SBjoern A. Zeeb 
13805ba50f5SJake Burkholder static int
13905ba50f5SJake Burkholder sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
14005ba50f5SJake Burkholder {
14105ba50f5SJake Burkholder 	struct proc *p;
142df7c361eSPeter Wemm 	int error;
14305ba50f5SJake Burkholder 
14405ba50f5SJake Burkholder 	p = curproc;
145a7bc3102SPeter Wemm #ifdef SCTL_MASK32
146a7bc3102SPeter Wemm 	if (req->flags & SCTL_MASK32) {
147df7c361eSPeter Wemm 		unsigned int val;
148df7c361eSPeter Wemm 		val = (unsigned int)p->p_sysent->sv_psstrings;
149df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &val, sizeof(val));
150df7c361eSPeter Wemm 	} else
151df7c361eSPeter Wemm #endif
152df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
153df7c361eSPeter Wemm 		   sizeof(p->p_sysent->sv_psstrings));
154df7c361eSPeter Wemm 	return error;
15505ba50f5SJake Burkholder }
15605ba50f5SJake Burkholder 
15705ba50f5SJake Burkholder static int
15805ba50f5SJake Burkholder sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
15905ba50f5SJake Burkholder {
16005ba50f5SJake Burkholder 	struct proc *p;
161df7c361eSPeter Wemm 	int error;
16205ba50f5SJake Burkholder 
16305ba50f5SJake Burkholder 	p = curproc;
164a7bc3102SPeter Wemm #ifdef SCTL_MASK32
165a7bc3102SPeter Wemm 	if (req->flags & SCTL_MASK32) {
166df7c361eSPeter Wemm 		unsigned int val;
167df7c361eSPeter Wemm 		val = (unsigned int)p->p_sysent->sv_usrstack;
168df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &val, sizeof(val));
169df7c361eSPeter Wemm 	} else
170df7c361eSPeter Wemm #endif
171df7c361eSPeter Wemm 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
172df7c361eSPeter Wemm 		    sizeof(p->p_sysent->sv_usrstack));
173df7c361eSPeter Wemm 	return error;
17405ba50f5SJake Burkholder }
17505ba50f5SJake Burkholder 
1765dadd17bSJake Burkholder static int
1775dadd17bSJake Burkholder sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
1785dadd17bSJake Burkholder {
1795dadd17bSJake Burkholder 	struct proc *p;
1805dadd17bSJake Burkholder 
1815dadd17bSJake Burkholder 	p = curproc;
1825dadd17bSJake Burkholder 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
1835dadd17bSJake Burkholder 	    sizeof(p->p_sysent->sv_stackprot)));
1845dadd17bSJake Burkholder }
1855dadd17bSJake Burkholder 
18699ac3bc8SPeter Wemm /*
187aa855a59SPeter Wemm  * Each of the items is a pointer to a `const struct execsw', hence the
188aa855a59SPeter Wemm  * double pointer here.
189df8bae1dSRodney W. Grimes  */
190aa855a59SPeter Wemm static const struct execsw **execsw;
19126f9a767SRodney W. Grimes 
192ca46e90eSBruce Evans #ifndef _SYS_SYSPROTO_H_
193ca46e90eSBruce Evans struct execve_args {
194ca46e90eSBruce Evans 	char    *fname;
195ca46e90eSBruce Evans 	char    **argv;
196ca46e90eSBruce Evans 	char    **envv;
197ca46e90eSBruce Evans };
198ca46e90eSBruce Evans #endif
199ca46e90eSBruce Evans 
200ca46e90eSBruce Evans int
2017b445033SKonstantin Belousov sys_execve(struct thread *td, struct execve_args *uap)
202ca46e90eSBruce Evans {
203610ecfe0SMaxim Sobolev 	struct image_args args;
2047b445033SKonstantin Belousov 	struct vmspace *oldvmspace;
2057b445033SKonstantin Belousov 	int error;
206ca46e90eSBruce Evans 
2077b445033SKonstantin Belousov 	error = pre_execve(td, &oldvmspace);
2087b445033SKonstantin Belousov 	if (error != 0)
2097b445033SKonstantin Belousov 		return (error);
210610ecfe0SMaxim Sobolev 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
211610ecfe0SMaxim Sobolev 	    uap->argv, uap->envv);
212610ecfe0SMaxim Sobolev 	if (error == 0)
213610ecfe0SMaxim Sobolev 		error = kern_execve(td, &args, NULL);
2147b445033SKonstantin Belousov 	post_execve(td, error, oldvmspace);
215610ecfe0SMaxim Sobolev 	return (error);
216ca46e90eSBruce Evans }
217ca46e90eSBruce Evans 
218ca46e90eSBruce Evans #ifndef _SYS_SYSPROTO_H_
219632dbc19SKonstantin Belousov struct fexecve_args {
220632dbc19SKonstantin Belousov 	int	fd;
221632dbc19SKonstantin Belousov 	char	**argv;
222632dbc19SKonstantin Belousov 	char	**envv;
223632dbc19SKonstantin Belousov }
224632dbc19SKonstantin Belousov #endif
225632dbc19SKonstantin Belousov int
2268451d0ddSKip Macy sys_fexecve(struct thread *td, struct fexecve_args *uap)
227632dbc19SKonstantin Belousov {
228632dbc19SKonstantin Belousov 	struct image_args args;
2297b445033SKonstantin Belousov 	struct vmspace *oldvmspace;
2307b445033SKonstantin Belousov 	int error;
231632dbc19SKonstantin Belousov 
2327b445033SKonstantin Belousov 	error = pre_execve(td, &oldvmspace);
2337b445033SKonstantin Belousov 	if (error != 0)
2347b445033SKonstantin Belousov 		return (error);
235632dbc19SKonstantin Belousov 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
236632dbc19SKonstantin Belousov 	    uap->argv, uap->envv);
237632dbc19SKonstantin Belousov 	if (error == 0) {
238632dbc19SKonstantin Belousov 		args.fd = uap->fd;
239632dbc19SKonstantin Belousov 		error = kern_execve(td, &args, NULL);
240632dbc19SKonstantin Belousov 	}
2417b445033SKonstantin Belousov 	post_execve(td, error, oldvmspace);
242632dbc19SKonstantin Belousov 	return (error);
243632dbc19SKonstantin Belousov }
244632dbc19SKonstantin Belousov 
245632dbc19SKonstantin Belousov #ifndef _SYS_SYSPROTO_H_
246ca46e90eSBruce Evans struct __mac_execve_args {
247ca46e90eSBruce Evans 	char	*fname;
248ca46e90eSBruce Evans 	char	**argv;
249ca46e90eSBruce Evans 	char	**envv;
250ca46e90eSBruce Evans 	struct mac	*mac_p;
251ca46e90eSBruce Evans };
252ca46e90eSBruce Evans #endif
253ca46e90eSBruce Evans 
254ca46e90eSBruce Evans int
2557b445033SKonstantin Belousov sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
256ca46e90eSBruce Evans {
257ca46e90eSBruce Evans #ifdef MAC
258610ecfe0SMaxim Sobolev 	struct image_args args;
2597b445033SKonstantin Belousov 	struct vmspace *oldvmspace;
2607b445033SKonstantin Belousov 	int error;
261610ecfe0SMaxim Sobolev 
2627b445033SKonstantin Belousov 	error = pre_execve(td, &oldvmspace);
2637b445033SKonstantin Belousov 	if (error != 0)
2647b445033SKonstantin Belousov 		return (error);
265610ecfe0SMaxim Sobolev 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
266610ecfe0SMaxim Sobolev 	    uap->argv, uap->envv);
267610ecfe0SMaxim Sobolev 	if (error == 0)
268610ecfe0SMaxim Sobolev 		error = kern_execve(td, &args, uap->mac_p);
2697b445033SKonstantin Belousov 	post_execve(td, error, oldvmspace);
270610ecfe0SMaxim Sobolev 	return (error);
271ca46e90eSBruce Evans #else
272ca46e90eSBruce Evans 	return (ENOSYS);
273ca46e90eSBruce Evans #endif
274ca46e90eSBruce Evans }
275ca46e90eSBruce Evans 
27684e0b075SDavid Xu int
2777b445033SKonstantin Belousov pre_execve(struct thread *td, struct vmspace **oldvmspace)
278906ac69dSDavid Xu {
2797b445033SKonstantin Belousov 	struct proc *p;
280906ac69dSDavid Xu 	int error;
281906ac69dSDavid Xu 
2827b445033SKonstantin Belousov 	KASSERT(td == curthread, ("non-current thread %p", td));
2837b445033SKonstantin Belousov 	error = 0;
2847b445033SKonstantin Belousov 	p = td->td_proc;
2857b445033SKonstantin Belousov 	if ((p->p_flag & P_HADTHREADS) != 0) {
286906ac69dSDavid Xu 		PROC_LOCK(p);
2877b445033SKonstantin Belousov 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
2887b445033SKonstantin Belousov 			error = ERESTART;
289906ac69dSDavid Xu 		PROC_UNLOCK(p);
290906ac69dSDavid Xu 	}
2917b445033SKonstantin Belousov 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
2927b445033SKonstantin Belousov 	    ("nested execve"));
2937b445033SKonstantin Belousov 	*oldvmspace = p->p_vmspace;
2947b445033SKonstantin Belousov 	return (error);
2957b445033SKonstantin Belousov }
296906ac69dSDavid Xu 
2977b445033SKonstantin Belousov void
2987b445033SKonstantin Belousov post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
2997b445033SKonstantin Belousov {
3007b445033SKonstantin Belousov 	struct proc *p;
301906ac69dSDavid Xu 
3027b445033SKonstantin Belousov 	KASSERT(td == curthread, ("non-current thread %p", td));
3037b445033SKonstantin Belousov 	p = td->td_proc;
3047b445033SKonstantin Belousov 	if ((p->p_flag & P_HADTHREADS) != 0) {
305906ac69dSDavid Xu 		PROC_LOCK(p);
306906ac69dSDavid Xu 		/*
307906ac69dSDavid Xu 		 * If success, we upgrade to SINGLE_EXIT state to
308906ac69dSDavid Xu 		 * force other threads to suicide.
309906ac69dSDavid Xu 		 */
310906ac69dSDavid Xu 		if (error == 0)
3116ddcc233SKonstantin Belousov 			thread_single(p, SINGLE_EXIT);
312906ac69dSDavid Xu 		else
3136ddcc233SKonstantin Belousov 			thread_single_end(p, SINGLE_BOUNDARY);
314906ac69dSDavid Xu 		PROC_UNLOCK(p);
315906ac69dSDavid Xu 	}
3167032434eSKonstantin Belousov 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
3177b445033SKonstantin Belousov 		KASSERT(p->p_vmspace != oldvmspace,
3187032434eSKonstantin Belousov 		    ("oldvmspace still used"));
3197032434eSKonstantin Belousov 		vmspace_free(oldvmspace);
3207032434eSKonstantin Belousov 		td->td_pflags &= ~TDP_EXECVMSPC;
3217032434eSKonstantin Belousov 	}
3227b445033SKonstantin Belousov }
323906ac69dSDavid Xu 
3247b445033SKonstantin Belousov /*
3257b445033SKonstantin Belousov  * XXX: kern_execve has the astonishing property of not always returning to
3267b445033SKonstantin Belousov  * the caller.  If sufficiently bad things happen during the call to
3277b445033SKonstantin Belousov  * do_execve(), it can end up calling exit1(); as a result, callers must
3287b445033SKonstantin Belousov  * avoid doing anything which they might need to undo (e.g., allocating
3297b445033SKonstantin Belousov  * memory).
3307b445033SKonstantin Belousov  */
3317b445033SKonstantin Belousov int
3327b445033SKonstantin Belousov kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
3337b445033SKonstantin Belousov {
3347b445033SKonstantin Belousov 
3357b445033SKonstantin Belousov 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
3367b445033SKonstantin Belousov 	    args->begin_envv - args->begin_argv);
3377b445033SKonstantin Belousov 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
3387b445033SKonstantin Belousov 	    args->endp - args->begin_envv);
3397b445033SKonstantin Belousov 	return (do_execve(td, args, mac_p));
340906ac69dSDavid Xu }
341906ac69dSDavid Xu 
34226f9a767SRodney W. Grimes /*
343450ffb44SRobert Watson  * In-kernel implementation of execve().  All arguments are assumed to be
344450ffb44SRobert Watson  * userspace pointers from the passed thread.
34526f9a767SRodney W. Grimes  */
346450ffb44SRobert Watson static int
347610ecfe0SMaxim Sobolev do_execve(td, args, mac_p)
348b40ce416SJulian Elischer 	struct thread *td;
349610ecfe0SMaxim Sobolev 	struct image_args *args;
350670cb89bSRobert Watson 	struct mac *mac_p;
351df8bae1dSRodney W. Grimes {
352b40ce416SJulian Elischer 	struct proc *p = td->td_proc;
35369c9eff8SEd Schouten 	struct nameidata nd;
3549b3b1c5fSJohn Baldwin 	struct ucred *newcred = NULL, *oldcred;
355350d5181SMateusz Guzik 	struct uidinfo *euip = NULL;
356654f6be1SBruce Evans 	register_t *stack_base;
3575ca4819dSJohn Baldwin 	int error, i;
358c52007c2SDavid Greenman 	struct image_params image_params, *imgp;
3599f5c1d19SDiomidis Spinellis 	struct vattr attr;
3604d77a549SAlfred Perlstein 	int (*img_first)(struct image_params *);
36169be5db9SAlfred Perlstein 	struct pargs *oldargs = NULL, *newargs = NULL;
36290af4afaSJohn Baldwin 	struct sigacts *oldsigacts, *newsigacts;
3636c84de02SJohn Baldwin #ifdef KTRACE
3646c84de02SJohn Baldwin 	struct vnode *tracevp = NULL;
365a5881ea5SJohn Baldwin 	struct ucred *tracecred = NULL;
3666c84de02SJohn Baldwin #endif
36761617058SMateusz Guzik 	struct vnode *oldtextvp = NULL, *newtextvp;
3687008be5bSPawel Jakub Dawidek 	cap_rights_t rights;
369d06c0d4dSRobert Watson 	int credential_changing;
370619eb6e5SJeff Roberson 	int textset;
371ccafe7ebSRobert Watson #ifdef MAC
3726356dba0SRobert Watson 	struct label *interpvplabel = NULL;
373eca8a663SRobert Watson 	int will_transition;
374ccafe7ebSRobert Watson #endif
37515139246SJoseph Koshy #ifdef HWPMC_HOOKS
37615139246SJoseph Koshy 	struct pmckern_procexec pe;
37715139246SJoseph Koshy #endif
378632dbc19SKonstantin Belousov 	static const char fexecv_proc_title[] = "(fexecv)";
37926f9a767SRodney W. Grimes 
380c52007c2SDavid Greenman 	imgp = &image_params;
381df8bae1dSRodney W. Grimes 
3829ca45e81SDag-Erling Smørgrav 	/*
3839ca45e81SDag-Erling Smørgrav 	 * Lock the process and set the P_INEXEC flag to indicate that
3849ca45e81SDag-Erling Smørgrav 	 * it should be left alone until we're done here.  This is
3859ca45e81SDag-Erling Smørgrav 	 * necessary to avoid race conditions - e.g. in ptrace() -
3869ca45e81SDag-Erling Smørgrav 	 * that might allow a local user to illicitly obtain elevated
3879ca45e81SDag-Erling Smørgrav 	 * privileges.
3889ca45e81SDag-Erling Smørgrav 	 */
3899ca45e81SDag-Erling Smørgrav 	PROC_LOCK(p);
3909ca45e81SDag-Erling Smørgrav 	KASSERT((p->p_flag & P_INEXEC) == 0,
39191f91617SDavid E. O'Brien 	    ("%s(): process already has P_INEXEC flag", __func__));
3929ca45e81SDag-Erling Smørgrav 	p->p_flag |= P_INEXEC;
3939ca45e81SDag-Erling Smørgrav 	PROC_UNLOCK(p);
3949ca45e81SDag-Erling Smørgrav 
395df8bae1dSRodney W. Grimes 	/*
396c52007c2SDavid Greenman 	 * Initialize part of the common data
397df8bae1dSRodney W. Grimes 	 */
3988e572983SMateusz Guzik 	bzero(imgp, sizeof(*imgp));
399c52007c2SDavid Greenman 	imgp->proc = p;
400c52007c2SDavid Greenman 	imgp->attr = &attr;
401610ecfe0SMaxim Sobolev 	imgp->args = args;
40226f9a767SRodney W. Grimes 
403670cb89bSRobert Watson #ifdef MAC
404eca8a663SRobert Watson 	error = mac_execve_enter(imgp, mac_p);
405532c491bSJeff Roberson 	if (error)
406670cb89bSRobert Watson 		goto exec_fail;
407670cb89bSRobert Watson #endif
408670cb89bSRobert Watson 
40926f9a767SRodney W. Grimes 	/*
41026f9a767SRodney W. Grimes 	 * Translate the file name. namei() returns a vnode pointer
41126f9a767SRodney W. Grimes 	 *	in ni_vp amoung other things.
412ae1078d6SWayne Salamon 	 *
413ae1078d6SWayne Salamon 	 * XXXAUDIT: It would be desirable to also audit the name of the
414ae1078d6SWayne Salamon 	 * interpreter if this is an interpreted binary.
41526f9a767SRodney W. Grimes 	 */
416632dbc19SKonstantin Belousov 	if (args->fname != NULL) {
41769c9eff8SEd Schouten 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
4185050aa86SKonstantin Belousov 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
419632dbc19SKonstantin Belousov 	}
42026f9a767SRodney W. Grimes 
421e506f34bSCraig Rodrigues 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
422e506f34bSCraig Rodrigues 
42326f9a767SRodney W. Grimes interpret:
424632dbc19SKonstantin Belousov 	if (args->fname != NULL) {
42512bc222eSJonathan Anderson #ifdef CAPABILITY_MODE
42612bc222eSJonathan Anderson 		/*
42712bc222eSJonathan Anderson 		 * While capability mode can't reach this point via direct
42812bc222eSJonathan Anderson 		 * path arguments to execve(), we also don't allow
42912bc222eSJonathan Anderson 		 * interpreters to be used in capability mode (for now).
43012bc222eSJonathan Anderson 		 * Catch indirect lookups and return a permissions error.
43112bc222eSJonathan Anderson 		 */
43212bc222eSJonathan Anderson 		if (IN_CAPABILITY_MODE(td)) {
43312bc222eSJonathan Anderson 			error = ECAPMODE;
43412bc222eSJonathan Anderson 			goto exec_fail;
43512bc222eSJonathan Anderson 		}
43612bc222eSJonathan Anderson #endif
43769c9eff8SEd Schouten 		error = namei(&nd);
438610ecfe0SMaxim Sobolev 		if (error)
43926f9a767SRodney W. Grimes 			goto exec_fail;
44026f9a767SRodney W. Grimes 
44161617058SMateusz Guzik 		newtextvp = nd.ni_vp;
44261617058SMateusz Guzik 		imgp->vp = newtextvp;
443632dbc19SKonstantin Belousov 	} else {
44414961ba7SRobert Watson 		AUDIT_ARG_FD(args->fd);
445a9d2f8d8SRobert Watson 		/*
446c8e781f6SPawel Jakub Dawidek 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
447a9d2f8d8SRobert Watson 		 */
4487008be5bSPawel Jakub Dawidek 		error = fgetvp_exec(td, args->fd,
44961617058SMateusz Guzik 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
450632dbc19SKonstantin Belousov 		if (error)
451632dbc19SKonstantin Belousov 			goto exec_fail;
45261617058SMateusz Guzik 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
45361617058SMateusz Guzik 		AUDIT_ARG_VNODE1(newtextvp);
45461617058SMateusz Guzik 		imgp->vp = newtextvp;
455632dbc19SKonstantin Belousov 	}
45626f9a767SRodney W. Grimes 
45726f9a767SRodney W. Grimes 	/*
4589022e4adSDavid Greenman 	 * Check file permissions (also 'opens' file)
4599022e4adSDavid Greenman 	 */
460c52007c2SDavid Greenman 	error = exec_check_permissions(imgp);
461619eb6e5SJeff Roberson 	if (error)
46226f9a767SRodney W. Grimes 		goto exec_fail_dealloc;
463619eb6e5SJeff Roberson 
4648516dd18SPoul-Henning Kamp 	imgp->object = imgp->vp->v_object;
4658516dd18SPoul-Henning Kamp 	if (imgp->object != NULL)
4660b2ed1aeSJeff Roberson 		vm_object_reference(imgp->object);
46726f9a767SRodney W. Grimes 
468619eb6e5SJeff Roberson 	/*
469619eb6e5SJeff Roberson 	 * Set VV_TEXT now so no one can write to the executable while we're
470619eb6e5SJeff Roberson 	 * activating it.
471619eb6e5SJeff Roberson 	 *
472619eb6e5SJeff Roberson 	 * Remember if this was set before and unset it in case this is not
473619eb6e5SJeff Roberson 	 * actually an executable image.
474619eb6e5SJeff Roberson 	 */
475877d24acSKonstantin Belousov 	textset = VOP_IS_TEXT(imgp->vp);
476877d24acSKonstantin Belousov 	VOP_SET_TEXT(imgp->vp);
477619eb6e5SJeff Roberson 
4781616db3cSJohn Dyson 	error = exec_map_first_page(imgp);
4796d5a0a8cSDavid Greenman 	if (error)
48026f9a767SRodney W. Grimes 		goto exec_fail_dealloc;
48126f9a767SRodney W. Grimes 
482f231de47SKonstantin Belousov 	imgp->proc->p_osrel = 0;
48326f9a767SRodney W. Grimes 	/*
484d323ddf3SMatthew Dillon 	 *	If the current process has a special image activator it
485d323ddf3SMatthew Dillon 	 *	wants to try first, call it.   For example, emulating shell
486d323ddf3SMatthew Dillon 	 *	scripts differently.
48726f9a767SRodney W. Grimes 	 */
488d323ddf3SMatthew Dillon 	error = -1;
489d323ddf3SMatthew Dillon 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
490d323ddf3SMatthew Dillon 		error = img_first(imgp);
491d323ddf3SMatthew Dillon 
492d323ddf3SMatthew Dillon 	/*
493d323ddf3SMatthew Dillon 	 *	Loop through the list of image activators, calling each one.
494d323ddf3SMatthew Dillon 	 *	An activator returns -1 if there is no match, 0 on success,
495d323ddf3SMatthew Dillon 	 *	and an error otherwise.
496d323ddf3SMatthew Dillon 	 */
497d323ddf3SMatthew Dillon 	for (i = 0; error == -1 && execsw[i]; ++i) {
498d323ddf3SMatthew Dillon 		if (execsw[i]->ex_imgact == NULL ||
499d323ddf3SMatthew Dillon 		    execsw[i]->ex_imgact == img_first) {
500d323ddf3SMatthew Dillon 			continue;
501d323ddf3SMatthew Dillon 		}
502c52007c2SDavid Greenman 		error = (*execsw[i]->ex_imgact)(imgp);
503d323ddf3SMatthew Dillon 	}
504d323ddf3SMatthew Dillon 
505d323ddf3SMatthew Dillon 	if (error) {
506619eb6e5SJeff Roberson 		if (error == -1) {
507877d24acSKonstantin Belousov 			if (textset == 0)
508877d24acSKonstantin Belousov 				VOP_UNSET_TEXT(imgp->vp);
509d323ddf3SMatthew Dillon 			error = ENOEXEC;
510619eb6e5SJeff Roberson 		}
51126f9a767SRodney W. Grimes 		goto exec_fail_dealloc;
512d323ddf3SMatthew Dillon 	}
513d323ddf3SMatthew Dillon 
514d323ddf3SMatthew Dillon 	/*
515d323ddf3SMatthew Dillon 	 * Special interpreter operation, cleanup and loop up to try to
516d323ddf3SMatthew Dillon 	 * activate the interpreter.
517d323ddf3SMatthew Dillon 	 */
518c52007c2SDavid Greenman 	if (imgp->interpreted) {
5191616db3cSJohn Dyson 		exec_unmap_first_page(imgp);
520619eb6e5SJeff Roberson 		/*
521619eb6e5SJeff Roberson 		 * VV_TEXT needs to be unset for scripts.  There is a short
522619eb6e5SJeff Roberson 		 * period before we determine that something is a script where
523619eb6e5SJeff Roberson 		 * VV_TEXT will be set. The vnode lock is held over this
524619eb6e5SJeff Roberson 		 * entire period so nothing should illegitimately be blocked.
525619eb6e5SJeff Roberson 		 */
526877d24acSKonstantin Belousov 		VOP_UNSET_TEXT(imgp->vp);
527762e6b85SEivind Eklund 		/* free name buffer and old vnode */
528632dbc19SKonstantin Belousov 		if (args->fname != NULL)
52969c9eff8SEd Schouten 			NDFREE(&nd, NDF_ONLY_PNBUF);
530670cb89bSRobert Watson #ifdef MAC
53161617058SMateusz Guzik 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
532670cb89bSRobert Watson #endif
5339a75ea23SKonstantin Belousov 		if (imgp->opened) {
53461617058SMateusz Guzik 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
5359a75ea23SKonstantin Belousov 			imgp->opened = 0;
5369a75ea23SKonstantin Belousov 		}
53761617058SMateusz Guzik 		vput(newtextvp);
5380b2ed1aeSJeff Roberson 		vm_object_deallocate(imgp->object);
5390b2ed1aeSJeff Roberson 		imgp->object = NULL;
54026f9a767SRodney W. Grimes 		/* set new name to that of the interpreter */
5415050aa86SKonstantin Belousov 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
542b40ce416SJulian Elischer 		    UIO_SYSSPACE, imgp->interpreter_name, td);
543632dbc19SKonstantin Belousov 		args->fname = imgp->interpreter_name;
54426f9a767SRodney W. Grimes 		goto interpret;
54526f9a767SRodney W. Grimes 	}
54626f9a767SRodney W. Grimes 
54726f9a767SRodney W. Grimes 	/*
548ded7d39cSChristian S.J. Peron 	 * NB: We unlock the vnode here because it is believed that none
549ded7d39cSChristian S.J. Peron 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
550ded7d39cSChristian S.J. Peron 	 */
551ded7d39cSChristian S.J. Peron 	VOP_UNLOCK(imgp->vp, 0);
5523ff06357SKonstantin Belousov 
5533ff06357SKonstantin Belousov 	/*
5543ff06357SKonstantin Belousov 	 * Do the best to calculate the full path to the image file.
5553ff06357SKonstantin Belousov 	 */
5563ff06357SKonstantin Belousov 	if (imgp->auxargs != NULL &&
5573ff06357SKonstantin Belousov 	    ((args->fname != NULL && args->fname[0] == '/') ||
5583ff06357SKonstantin Belousov 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
5593ff06357SKonstantin Belousov 		imgp->execpath = args->fname;
5603ff06357SKonstantin Belousov 
561eda6009cSKonstantin Belousov 	if (disallow_high_osrel &&
562eda6009cSKonstantin Belousov 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
563eda6009cSKonstantin Belousov 		error = ENOEXEC;
564eda6009cSKonstantin Belousov 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
565eda6009cSKonstantin Belousov 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
566eda6009cSKonstantin Belousov 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
567eda6009cSKonstantin Belousov 		goto exec_fail_dealloc;
568eda6009cSKonstantin Belousov 	}
569eda6009cSKonstantin Belousov 
57039f5ebb7SEd Schouten 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
57139f5ebb7SEd Schouten 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
57239f5ebb7SEd Schouten 		sys_cap_enter(td, NULL);
57339f5ebb7SEd Schouten 
574ded7d39cSChristian S.J. Peron 	/*
57526f9a767SRodney W. Grimes 	 * Copy out strings (args and env) and initialize stack base
57626f9a767SRodney W. Grimes 	 */
5773ebc1248SPeter Wemm 	if (p->p_sysent->sv_copyout_strings)
5783ebc1248SPeter Wemm 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
5793ebc1248SPeter Wemm 	else
580c52007c2SDavid Greenman 		stack_base = exec_copyout_strings(imgp);
58126f9a767SRodney W. Grimes 
58226f9a767SRodney W. Grimes 	/*
5831e1e0b44SSøren Schmidt 	 * If custom stack fixup routine present for this process
5841e1e0b44SSøren Schmidt 	 * let it do the stack setup.
5851e1e0b44SSøren Schmidt 	 * Else stuff argument count as first item on stack
58626f9a767SRodney W. Grimes 	 */
587d6c847f3SBruce Evans 	if (p->p_sysent->sv_fixup != NULL)
588c52007c2SDavid Greenman 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
5891e1e0b44SSøren Schmidt 	else
590610ecfe0SMaxim Sobolev 		suword(--stack_base, imgp->args->argc);
59126f9a767SRodney W. Grimes 
592457f7e23SEd Schouten 	if (args->fdp != NULL) {
593457f7e23SEd Schouten 		/* Install a brand new file descriptor table. */
594457f7e23SEd Schouten 		fdinstall_remapped(td, args->fdp);
595457f7e23SEd Schouten 		args->fdp = NULL;
596457f7e23SEd Schouten 	} else {
597a78e8d2aSDavid Greenman 		/*
598457f7e23SEd Schouten 		 * Keep on using the existing file descriptor table. For
599457f7e23SEd Schouten 		 * security and other reasons, the file descriptor table
600457f7e23SEd Schouten 		 * cannot be shared after an exec.
601a78e8d2aSDavid Greenman 		 */
602b9d32c36SMateusz Guzik 		fdunshare(td);
603b0bc0cadSMateusz Guzik 		/* close files on exec */
604b0bc0cadSMateusz Guzik 		fdcloseexec(td);
605457f7e23SEd Schouten 	}
606a78e8d2aSDavid Greenman 
607333ea485SGuido van Rooij 	/*
6089b3b1c5fSJohn Baldwin 	 * Malloc things before we need locks.
6099b3b1c5fSJohn Baldwin 	 */
610610ecfe0SMaxim Sobolev 	i = imgp->args->begin_envv - imgp->args->begin_argv;
6115997cae9SDon Lewis 	/* Cache arguments if they fit inside our allowance */
6125997cae9SDon Lewis 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
6139b3b1c5fSJohn Baldwin 		newargs = pargs_alloc(i);
6145997cae9SDon Lewis 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
6155997cae9SDon Lewis 	}
6169b3b1c5fSJohn Baldwin 
6172974cc36SKonstantin Belousov 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
6189b3b1c5fSJohn Baldwin 
6199b3b1c5fSJohn Baldwin 	/*
620333ea485SGuido van Rooij 	 * For security and other reasons, signal handlers cannot
6218688bb93SJohn Baldwin 	 * be shared after an exec. The new process gets a copy of the old
622b2c3fa70SDima Dorfman 	 * handlers. In execsigs(), the new process will have its signals
623333ea485SGuido van Rooij 	 * reset.
624333ea485SGuido van Rooij 	 */
62590af4afaSJohn Baldwin 	if (sigacts_shared(p->p_sigacts)) {
62690af4afaSJohn Baldwin 		oldsigacts = p->p_sigacts;
62790af4afaSJohn Baldwin 		newsigacts = sigacts_alloc();
62890af4afaSJohn Baldwin 		sigacts_copy(newsigacts, oldsigacts);
629a6bad85eSMateusz Guzik 	} else {
63090af4afaSJohn Baldwin 		oldsigacts = NULL;
631a6bad85eSMateusz Guzik 		newsigacts = NULL; /* satisfy gcc */
632a6bad85eSMateusz Guzik 	}
633333ea485SGuido van Rooij 
634d00c8ea4SMateusz Guzik 	PROC_LOCK(p);
635d00c8ea4SMateusz Guzik 	if (oldsigacts)
636d00c8ea4SMateusz Guzik 		p->p_sigacts = newsigacts;
637350d5181SMateusz Guzik 	oldcred = p->p_ucred;
638fdf4e8b3SWarner Losh 	/* Stop profiling */
639fdf4e8b3SWarner Losh 	stopprofclock(p);
640fdf4e8b3SWarner Losh 
64126f9a767SRodney W. Grimes 	/* reset caught signals */
64226f9a767SRodney W. Grimes 	execsigs(p);
64326f9a767SRodney W. Grimes 
64426f9a767SRodney W. Grimes 	/* name this process - nameiexec(p, ndp) */
6455ca4819dSJohn Baldwin 	bzero(p->p_comm, sizeof(p->p_comm));
6465ca4819dSJohn Baldwin 	if (args->fname)
6475ca4819dSJohn Baldwin 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
6485ca4819dSJohn Baldwin 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
64961617058SMateusz Guzik 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
6505ca4819dSJohn Baldwin 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
651ca081fdbSJulian Elischer 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
65244ad5475SJohn Baldwin #ifdef KTR
65344ad5475SJohn Baldwin 	sched_clear_tdname(td);
65444ad5475SJohn Baldwin #endif
65526f9a767SRodney W. Grimes 
65626f9a767SRodney W. Grimes 	/*
65724b34f09SSujal Patel 	 * mark as execed, wakeup the process that vforked (if any) and tell
658dc733423SDag-Erling Smørgrav 	 * it that it now has its own resources back
65926f9a767SRodney W. Grimes 	 */
66026f9a767SRodney W. Grimes 	p->p_flag |= P_EXEC;
661677258f7SKonstantin Belousov 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
662677258f7SKonstantin Belousov 		p->p_flag2 &= ~P2_NOTRACE;
663965d0860SMateusz Guzik 	if (p->p_flag & P_PPWAIT) {
664888d4d4fSKonstantin Belousov 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
665aeb32571SKonstantin Belousov 		cv_broadcast(&p->p_pwait);
66626f9a767SRodney W. Grimes 	}
66726f9a767SRodney W. Grimes 
66826f9a767SRodney W. Grimes 	/*
669a78e8d2aSDavid Greenman 	 * Implement image setuid/setgid.
670a78e8d2aSDavid Greenman 	 *
671a78e8d2aSDavid Greenman 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
672a78e8d2aSDavid Greenman 	 * the process is being traced.
673670cb89bSRobert Watson 	 *
67412bc222eSJonathan Anderson 	 * We disable setuid/setgid/etc in compatibility mode on the basis
67512bc222eSJonathan Anderson 	 * that most setugid applications are not written with that
67612bc222eSJonathan Anderson 	 * environment in mind, and will therefore almost certainly operate
67712bc222eSJonathan Anderson 	 * incorrectly. In principle there's no reason that setugid
67812bc222eSJonathan Anderson 	 * applications might not be useful in capability mode, so we may want
67912bc222eSJonathan Anderson 	 * to reconsider this conservative design choice in the future.
68012bc222eSJonathan Anderson 	 *
681670cb89bSRobert Watson 	 * XXXMAC: For the time being, use NOSUID to also prohibit
682670cb89bSRobert Watson 	 * transitions on the file system.
683c52007c2SDavid Greenman 	 */
684d06c0d4dSRobert Watson 	credential_changing = 0;
685dfa7fd1dSEdward Tomasz Napierala 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
686d06c0d4dSRobert Watson 	    attr.va_uid;
687dfa7fd1dSEdward Tomasz Napierala 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
688d06c0d4dSRobert Watson 	    attr.va_gid;
689ccafe7ebSRobert Watson #ifdef MAC
69030d239bcSRobert Watson 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
6916356dba0SRobert Watson 	    interpvplabel, imgp);
692ccafe7ebSRobert Watson 	credential_changing |= will_transition;
693ccafe7ebSRobert Watson #endif
694d06c0d4dSRobert Watson 
695d06c0d4dSRobert Watson 	if (credential_changing &&
69612bc222eSJonathan Anderson #ifdef CAPABILITY_MODE
69712bc222eSJonathan Anderson 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
69812bc222eSJonathan Anderson #endif
699a78e8d2aSDavid Greenman 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
700c52007c2SDavid Greenman 	    (p->p_flag & P_TRACED) == 0) {
701c52007c2SDavid Greenman 		/*
702c52007c2SDavid Greenman 		 * Turn off syscall tracing for set-id programs, except for
703b85db196SPeter Wemm 		 * root.  Record any set-id flags first to make sure that
704b85db196SPeter Wemm 		 * we do not regain any tracing during a possible block.
70526f9a767SRodney W. Grimes 		 */
706b85db196SPeter Wemm 		setsugid(p);
707acd3428bSRobert Watson 
7086c84de02SJohn Baldwin #ifdef KTRACE
709c331c970SAndriy Gapon 		if (p->p_tracecred != NULL &&
710c331c970SAndriy Gapon 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
711d680caabSJohn Baldwin 			ktrprocexec(p, &tracecred, &tracevp);
7126c84de02SJohn Baldwin #endif
71328b325aaSDon Lewis 		/*
714c1e2d386SNate Lawson 		 * Close any file descriptors 0..2 that reference procfs,
715c1e2d386SNate Lawson 		 * then make sure file descriptors 0..2 are in use.
71628b325aaSDon Lewis 		 *
7178a5177ccSMateusz Guzik 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
7188a5177ccSMateusz Guzik 		 * taking sleepable locks, so temporarily drop our locks.
71928b325aaSDon Lewis 		 */
72028b325aaSDon Lewis 		PROC_UNLOCK(p);
72122db15c0SAttilio Rao 		VOP_UNLOCK(imgp->vp, 0);
72211888da8SMateusz Guzik 		fdsetugidsafety(td);
723e983a376SJacques Vidrine 		error = fdcheckstd(td);
72463c9e754SJohn Baldwin 		if (error != 0)
72569be5db9SAlfred Perlstein 			goto done1;
726350d5181SMateusz Guzik 		newcred = crdup(oldcred);
727350d5181SMateusz Guzik 		euip = uifind(attr.va_uid);
7285e2554b7SMateusz Guzik 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
729e1b1aa3bSJohn Baldwin 		PROC_LOCK(p);
730c52007c2SDavid Greenman 		/*
731c52007c2SDavid Greenman 		 * Set the new credentials.
732c52007c2SDavid Greenman 		 */
733dfa7fd1dSEdward Tomasz Napierala 		if (attr.va_mode & S_ISUID)
7341419eacbSAlfred Perlstein 			change_euid(newcred, euip);
735dfa7fd1dSEdward Tomasz Napierala 		if (attr.va_mode & S_ISGID)
736b1fc0ec1SRobert Watson 			change_egid(newcred, attr.va_gid);
737ccafe7ebSRobert Watson #ifdef MAC
738670cb89bSRobert Watson 		if (will_transition) {
73930d239bcSRobert Watson 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
7406356dba0SRobert Watson 			    interpvplabel, imgp);
741670cb89bSRobert Watson 		}
742ccafe7ebSRobert Watson #endif
7439b3b1c5fSJohn Baldwin 		/*
7449b3b1c5fSJohn Baldwin 		 * Implement correct POSIX saved-id behavior.
745ccafe7ebSRobert Watson 		 *
746ccafe7ebSRobert Watson 		 * XXXMAC: Note that the current logic will save the
747ccafe7ebSRobert Watson 		 * uid and gid if a MAC domain transition occurs, even
748ccafe7ebSRobert Watson 		 * though maybe it shouldn't.
7499b3b1c5fSJohn Baldwin 		 */
7509b3b1c5fSJohn Baldwin 		change_svuid(newcred, newcred->cr_uid);
7519b3b1c5fSJohn Baldwin 		change_svgid(newcred, newcred->cr_gid);
752daf63fd2SMateusz Guzik 		proc_set_cred(p, newcred);
753c52007c2SDavid Greenman 	} else {
754b1fc0ec1SRobert Watson 		if (oldcred->cr_uid == oldcred->cr_ruid &&
755b1fc0ec1SRobert Watson 		    oldcred->cr_gid == oldcred->cr_rgid)
756c52007c2SDavid Greenman 			p->p_flag &= ~P_SUGID;
75726f9a767SRodney W. Grimes 		/*
758c52007c2SDavid Greenman 		 * Implement correct POSIX saved-id behavior.
759b1fc0ec1SRobert Watson 		 *
760b1fc0ec1SRobert Watson 		 * XXX: It's not clear that the existing behavior is
7619b3b1c5fSJohn Baldwin 		 * POSIX-compliant.  A number of sources indicate that the
7629b3b1c5fSJohn Baldwin 		 * saved uid/gid should only be updated if the new ruid is
7639b3b1c5fSJohn Baldwin 		 * not equal to the old ruid, or the new euid is not equal
7649b3b1c5fSJohn Baldwin 		 * to the old euid and the new euid is not equal to the old
7659b3b1c5fSJohn Baldwin 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
7669b3b1c5fSJohn Baldwin 		 * Also, this code uses the new (replaced) euid and egid as
7679b3b1c5fSJohn Baldwin 		 * the source, which may or may not be the right ones to use.
76826f9a767SRodney W. Grimes 		 */
769b1fc0ec1SRobert Watson 		if (oldcred->cr_svuid != oldcred->cr_uid ||
770b1fc0ec1SRobert Watson 		    oldcred->cr_svgid != oldcred->cr_gid) {
771350d5181SMateusz Guzik 			PROC_UNLOCK(p);
7725e2554b7SMateusz Guzik 			VOP_UNLOCK(imgp->vp, 0);
773350d5181SMateusz Guzik 			newcred = crdup(oldcred);
7745e2554b7SMateusz Guzik 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
775350d5181SMateusz Guzik 			PROC_LOCK(p);
776b1fc0ec1SRobert Watson 			change_svuid(newcred, newcred->cr_uid);
777b1fc0ec1SRobert Watson 			change_svgid(newcred, newcred->cr_gid);
778daf63fd2SMateusz Guzik 			proc_set_cred(p, newcred);
7799b3b1c5fSJohn Baldwin 		}
780b1fc0ec1SRobert Watson 	}
78126f9a767SRodney W. Grimes 
78226f9a767SRodney W. Grimes 	/*
783853be5ffSMateusz Guzik 	 * Store the vp for use in procfs.  This vnode was referenced by namei
784853be5ffSMateusz Guzik 	 * or fgetvp_exec.
7852a531c80SDavid Greenman 	 */
78661617058SMateusz Guzik 	oldtextvp = p->p_textvp;
78761617058SMateusz Guzik 	p->p_textvp = newtextvp;
7882a531c80SDavid Greenman 
7895d217f17SJohn Birrell #ifdef KDTRACE_HOOKS
7905d217f17SJohn Birrell 	/*
7915d217f17SJohn Birrell 	 * Tell the DTrace fasttrap provider about the exec if it
7925d217f17SJohn Birrell 	 * has declared an interest.
7935d217f17SJohn Birrell 	 */
7945d217f17SJohn Birrell 	if (dtrace_fasttrap_exec)
7955d217f17SJohn Birrell 		dtrace_fasttrap_exec(p);
7965d217f17SJohn Birrell #endif
7975d217f17SJohn Birrell 
7982a531c80SDavid Greenman 	/*
7999ca45e81SDag-Erling Smørgrav 	 * Notify others that we exec'd, and clear the P_INEXEC flag
8009ca45e81SDag-Erling Smørgrav 	 * as we're now a bona fide freshly-execed process.
801cb679c38SJonathan Lemon 	 */
802ad3b9257SJohn-Mark Gurney 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
8039ca45e81SDag-Erling Smørgrav 	p->p_flag &= ~P_INEXEC;
804cb679c38SJonathan Lemon 
80526f9a767SRodney W. Grimes 	/* clear "fork but no exec" flag, as we _are_ execing */
80626f9a767SRodney W. Grimes 	p->p_acflag &= ~AFORK;
80726f9a767SRodney W. Grimes 
8085997cae9SDon Lewis 	/*
80934ea500bSDon Lewis 	 * Free any previous argument cache and replace it with
8105997cae9SDon Lewis 	 * the new argument cache, if any.
8115997cae9SDon Lewis 	 */
8129b3b1c5fSJohn Baldwin 	oldargs = p->p_args;
813e1b1aa3bSJohn Baldwin 	p->p_args = newargs;
814e1b1aa3bSJohn Baldwin 	newargs = NULL;
815ebccf1e3SJoseph Koshy 
816ebccf1e3SJoseph Koshy #ifdef	HWPMC_HOOKS
817ebccf1e3SJoseph Koshy 	/*
818f263522aSJoseph Koshy 	 * Check if system-wide sampling is in effect or if the
819f263522aSJoseph Koshy 	 * current process is using PMCs.  If so, do exec() time
820ebccf1e3SJoseph Koshy 	 * processing.  This processing needs to happen AFTER the
821ebccf1e3SJoseph Koshy 	 * P_INEXEC flag is cleared.
822ebccf1e3SJoseph Koshy 	 *
823ebccf1e3SJoseph Koshy 	 * The proc lock needs to be released before taking the PMC
824ebccf1e3SJoseph Koshy 	 * SX.
825ebccf1e3SJoseph Koshy 	 */
826f263522aSJoseph Koshy 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
827e1b1aa3bSJohn Baldwin 		PROC_UNLOCK(p);
8281ef6ea9bSKonstantin Belousov 		VOP_UNLOCK(imgp->vp, 0);
82915139246SJoseph Koshy 		pe.pm_credentialschanged = credential_changing;
83015139246SJoseph Koshy 		pe.pm_entryaddr = imgp->entry_addr;
83115139246SJoseph Koshy 
83215139246SJoseph Koshy 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
8332974cc36SKonstantin Belousov 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
834ebccf1e3SJoseph Koshy 	} else
835ebccf1e3SJoseph Koshy 		PROC_UNLOCK(p);
836ebccf1e3SJoseph Koshy #else  /* !HWPMC_HOOKS */
837ebccf1e3SJoseph Koshy 	PROC_UNLOCK(p);
838ebccf1e3SJoseph Koshy #endif
839e1b1aa3bSJohn Baldwin 
840e913ca22SDoug Rabson 	/* Set values passed into the program in registers. */
8413ebc1248SPeter Wemm 	if (p->p_sysent->sv_setregs)
842a107d8aaSNathan Whitehorn 		(*p->p_sysent->sv_setregs)(td, imgp,
843a107d8aaSNathan Whitehorn 		    (u_long)(uintptr_t)stack_base);
8443ebc1248SPeter Wemm 	else
845a107d8aaSNathan Whitehorn 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
846e913ca22SDoug Rabson 
8470359a12eSAttilio Rao 	vfs_mark_atime(imgp->vp, td->td_ucred);
8486341095eSKen Smith 
849d9fae5abSAndriy Gapon 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
8505d217f17SJohn Birrell 
8515e2554b7SMateusz Guzik 	VOP_UNLOCK(imgp->vp, 0);
852e506f34bSCraig Rodrigues done1:
8539b3b1c5fSJohn Baldwin 	/*
8549b3b1c5fSJohn Baldwin 	 * Free any resources malloc'd earlier that we didn't use.
8559b3b1c5fSJohn Baldwin 	 */
856350d5181SMateusz Guzik 	if (euip != NULL)
8571419eacbSAlfred Perlstein 		uifree(euip);
858350d5181SMateusz Guzik 	if (newcred != NULL)
8599b3b1c5fSJohn Baldwin 		crfree(oldcred);
8605d217f17SJohn Birrell 
8619b3b1c5fSJohn Baldwin 	/*
8629b3b1c5fSJohn Baldwin 	 * Handle deferred decrement of ref counts.
8639b3b1c5fSJohn Baldwin 	 */
86461617058SMateusz Guzik 	if (oldtextvp != NULL)
86561617058SMateusz Guzik 		vrele(oldtextvp);
8666c84de02SJohn Baldwin #ifdef KTRACE
8675050aa86SKonstantin Belousov 	if (tracevp != NULL)
8689b3b1c5fSJohn Baldwin 		vrele(tracevp);
869a5881ea5SJohn Baldwin 	if (tracecred != NULL)
870a5881ea5SJohn Baldwin 		crfree(tracecred);
8716c84de02SJohn Baldwin #endif
8722974cc36SKonstantin Belousov 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
8739b3b1c5fSJohn Baldwin 	pargs_drop(oldargs);
87469be5db9SAlfred Perlstein 	pargs_drop(newargs);
87590af4afaSJohn Baldwin 	if (oldsigacts != NULL)
87690af4afaSJohn Baldwin 		sigacts_free(oldsigacts);
877b9df5231SPoul-Henning Kamp 
8781616db3cSJohn Dyson exec_fail_dealloc:
8791616db3cSJohn Dyson 
88026f9a767SRodney W. Grimes 	/*
88126f9a767SRodney W. Grimes 	 * free various allocated resources
88226f9a767SRodney W. Grimes 	 */
883d6c847f3SBruce Evans 	if (imgp->firstpage != NULL)
8841616db3cSJohn Dyson 		exec_unmap_first_page(imgp);
88526f9a767SRodney W. Grimes 
886d6c847f3SBruce Evans 	if (imgp->vp != NULL) {
887632dbc19SKonstantin Belousov 		if (args->fname)
88869c9eff8SEd Schouten 			NDFREE(&nd, NDF_ONLY_PNBUF);
8899a75ea23SKonstantin Belousov 		if (imgp->opened)
8909a75ea23SKonstantin Belousov 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
891853be5ffSMateusz Guzik 		if (error != 0)
892619eb6e5SJeff Roberson 			vput(imgp->vp);
893853be5ffSMateusz Guzik 		else
894853be5ffSMateusz Guzik 			VOP_UNLOCK(imgp->vp, 0);
895a3cf6ebaSDavid Greenman 	}
89626f9a767SRodney W. Grimes 
897d6c847f3SBruce Evans 	if (imgp->object != NULL)
8980b2ed1aeSJeff Roberson 		vm_object_deallocate(imgp->object);
8990b2ed1aeSJeff Roberson 
9003ff06357SKonstantin Belousov 	free(imgp->freepath, M_TEMP);
9013ff06357SKonstantin Belousov 
902d1989db5SRobert Drehmel 	if (error == 0) {
903afe1a688SKonstantin Belousov 		PROC_LOCK(p);
904afe1a688SKonstantin Belousov 		td->td_dbgflags |= TDB_EXEC;
905afe1a688SKonstantin Belousov 		PROC_UNLOCK(p);
906afe1a688SKonstantin Belousov 
907d1989db5SRobert Drehmel 		/*
908d1989db5SRobert Drehmel 		 * Stop the process here if its stop event mask has
909d1989db5SRobert Drehmel 		 * the S_EXEC bit set.
910d1989db5SRobert Drehmel 		 */
911d1989db5SRobert Drehmel 		STOPEVENT(p, S_EXEC, 0);
912116734c4SMatthew Dillon 		goto done2;
913d1989db5SRobert Drehmel 	}
9141616db3cSJohn Dyson 
91526f9a767SRodney W. Grimes exec_fail:
9169ca45e81SDag-Erling Smørgrav 	/* we're done here, clear P_INEXEC */
9179ca45e81SDag-Erling Smørgrav 	PROC_LOCK(p);
9189ca45e81SDag-Erling Smørgrav 	p->p_flag &= ~P_INEXEC;
9199ca45e81SDag-Erling Smørgrav 	PROC_UNLOCK(p);
9209ca45e81SDag-Erling Smørgrav 
921d9fae5abSAndriy Gapon 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
9225d217f17SJohn Birrell 
923116734c4SMatthew Dillon done2:
924670cb89bSRobert Watson #ifdef MAC
925670cb89bSRobert Watson 	mac_execve_exit(imgp);
9266356dba0SRobert Watson 	mac_execve_interpreter_exit(interpvplabel);
927670cb89bSRobert Watson #endif
9288917b8d2SJohn Baldwin 	exec_free_args(args);
9298917b8d2SJohn Baldwin 
9308917b8d2SJohn Baldwin 	if (error && imgp->vmspace_destroyed) {
9318917b8d2SJohn Baldwin 		/* sorry, no more process anymore. exit gracefully */
932b4490c6eSKonstantin Belousov 		exit1(td, 0, SIGABRT);
9338917b8d2SJohn Baldwin 		/* NOT REACHED */
9348917b8d2SJohn Baldwin 	}
9357705d4b2SDmitry Chagin 
9367705d4b2SDmitry Chagin #ifdef KTRACE
9377705d4b2SDmitry Chagin 	if (error == 0)
9387705d4b2SDmitry Chagin 		ktrprocctor(p);
9397705d4b2SDmitry Chagin #endif
9407705d4b2SDmitry Chagin 
941116734c4SMatthew Dillon 	return (error);
94226f9a767SRodney W. Grimes }
94326f9a767SRodney W. Grimes 
9441616db3cSJohn Dyson int
9451616db3cSJohn Dyson exec_map_first_page(imgp)
9461616db3cSJohn Dyson 	struct image_params *imgp;
9471616db3cSJohn Dyson {
948d8aad40cSJohn Baldwin 	int rv, i;
94995461b45SJohn Dyson 	int initial_pagein;
95095461b45SJohn Dyson 	vm_page_t ma[VM_INITIAL_PAGEIN];
9511616db3cSJohn Dyson 	vm_object_t object;
9521616db3cSJohn Dyson 
953d6c847f3SBruce Evans 	if (imgp->firstpage != NULL)
9541616db3cSJohn Dyson 		exec_unmap_first_page(imgp);
9551616db3cSJohn Dyson 
9568516dd18SPoul-Henning Kamp 	object = imgp->vp->v_object;
9574d7d2993SJeff Roberson 	if (object == NULL)
9584d7d2993SJeff Roberson 		return (EACCES);
95989f6b863SAttilio Rao 	VM_OBJECT_WLOCK(object);
960f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0
9613d653db0SAlan Cox 	vm_object_color(object, 0);
962f8a47341SAlan Cox #endif
9635944de8eSKonstantin Belousov 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
9640a2e596aSAlan Cox 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
96595461b45SJohn Dyson 		initial_pagein = VM_INITIAL_PAGEIN;
96695461b45SJohn Dyson 		if (initial_pagein > object->size)
96795461b45SJohn Dyson 			initial_pagein = object->size;
96895461b45SJohn Dyson 		for (i = 1; i < initial_pagein; i++) {
96941890423SAlan Cox 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
9706ec2fca5SAlan Cox 				if (ma[i]->valid)
9716ec2fca5SAlan Cox 					break;
972c7aebda8SAttilio Rao 				if (vm_page_tryxbusy(ma[i]))
97306fa71cdSAlan Cox 					break;
97495461b45SJohn Dyson 			} else {
975ca0387efSJake Burkholder 				ma[i] = vm_page_alloc(object, i,
9767bfda801SAlan Cox 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
97795461b45SJohn Dyson 				if (ma[i] == NULL)
97895461b45SJohn Dyson 					break;
97995461b45SJohn Dyson 			}
98095461b45SJohn Dyson 		}
98195461b45SJohn Dyson 		initial_pagein = i;
98295461b45SJohn Dyson 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
983093c7f39SGleb Smirnoff 		if (rv != VM_PAGER_OK) {
9845ac59343SAlan Cox 			vm_page_lock(ma[0]);
9858f9110f6SJohn Dyson 			vm_page_free(ma[0]);
9865ac59343SAlan Cox 			vm_page_unlock(ma[0]);
98789f6b863SAttilio Rao 			VM_OBJECT_WUNLOCK(object);
988a7cddfedSJake Burkholder 			return (EIO);
9891616db3cSJohn Dyson 		}
9901616db3cSJohn Dyson 	}
991c7aebda8SAttilio Rao 	vm_page_xunbusy(ma[0]);
9922965a453SKip Macy 	vm_page_lock(ma[0]);
993be996836SAttilio Rao 	vm_page_hold(ma[0]);
99470978c93SKonstantin Belousov 	vm_page_activate(ma[0]);
9952965a453SKip Macy 	vm_page_unlock(ma[0]);
99689f6b863SAttilio Rao 	VM_OBJECT_WUNLOCK(object);
9971616db3cSJohn Dyson 
99859c8bc40SAlan Cox 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
99959c8bc40SAlan Cox 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
10001616db3cSJohn Dyson 
1001a7cddfedSJake Burkholder 	return (0);
10021616db3cSJohn Dyson }
10031616db3cSJohn Dyson 
10041616db3cSJohn Dyson void
10051616db3cSJohn Dyson exec_unmap_first_page(imgp)
10061616db3cSJohn Dyson 	struct image_params *imgp;
10071616db3cSJohn Dyson {
100859c8bc40SAlan Cox 	vm_page_t m;
100923955314SAlfred Perlstein 
1010d6c847f3SBruce Evans 	if (imgp->firstpage != NULL) {
101159c8bc40SAlan Cox 		m = sf_buf_page(imgp->firstpage);
101259c8bc40SAlan Cox 		sf_buf_free(imgp->firstpage);
10131616db3cSJohn Dyson 		imgp->firstpage = NULL;
10142965a453SKip Macy 		vm_page_lock(m);
1015be996836SAttilio Rao 		vm_page_unhold(m);
10162965a453SKip Macy 		vm_page_unlock(m);
10171616db3cSJohn Dyson 	}
10181616db3cSJohn Dyson }
10191616db3cSJohn Dyson 
102026f9a767SRodney W. Grimes /*
102126f9a767SRodney W. Grimes  * Destroy old address space, and allocate a new stack
102226f9a767SRodney W. Grimes  *	The new stack is only SGROWSIZ large because it is grown
102326f9a767SRodney W. Grimes  *	automatically in trap.c.
102426f9a767SRodney W. Grimes  */
102526f9a767SRodney W. Grimes int
102605ba50f5SJake Burkholder exec_new_vmspace(imgp, sv)
1027c52007c2SDavid Greenman 	struct image_params *imgp;
102805ba50f5SJake Burkholder 	struct sysentvec *sv;
102926f9a767SRodney W. Grimes {
103026f9a767SRodney W. Grimes 	int error;
10315e20c11fSAlan Cox 	struct proc *p = imgp->proc;
10325e20c11fSAlan Cox 	struct vmspace *vmspace = p->p_vmspace;
10336297a3d8SKonstantin Belousov 	vm_object_t obj;
1034316b3843SKonstantin Belousov 	struct rlimit rlim_stack;
1035878adb85SBjoern A. Zeeb 	vm_offset_t sv_minuser, stack_addr;
103605ba50f5SJake Burkholder 	vm_map_t map;
103759d8f3ffSJohn Baldwin 	u_long ssiz;
103826f9a767SRodney W. Grimes 
1039c52007c2SDavid Greenman 	imgp->vmspace_destroyed = 1;
1040993182e5SAlexander Leidinger 	imgp->sysent = sv;
104126f9a767SRodney W. Grimes 
1042bd37fd72SKris Kennaway 	/* May be called with Giant held */
1043993182e5SAlexander Leidinger 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
10446f5dafeaSAlan Cox 
10456f5dafeaSAlan Cox 	/*
1046af9ec885SJohn Dyson 	 * Blow away entire process VM, if address space not shared,
1047af9ec885SJohn Dyson 	 * otherwise, create a new VM space so that other threads are
1048af9ec885SJohn Dyson 	 * not disrupted
1049af9ec885SJohn Dyson 	 */
105005ba50f5SJake Burkholder 	map = &vmspace->vm_map;
1051878adb85SBjoern A. Zeeb 	if (map_at_zero)
1052878adb85SBjoern A. Zeeb 		sv_minuser = sv->sv_minuser;
1053878adb85SBjoern A. Zeeb 	else
1054878adb85SBjoern A. Zeeb 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1055878adb85SBjoern A. Zeeb 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
105605ba50f5SJake Burkholder 	    vm_map_max(map) == sv->sv_maxuser) {
10573db161e0SMatthew Dillon 		shmexit(vmspace);
1058b9eee07eSPeter Wemm 		pmap_remove_pages(vmspace_pmap(vmspace));
105905ba50f5SJake Burkholder 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1060af9ec885SJohn Dyson 	} else {
1061878adb85SBjoern A. Zeeb 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
106289b57fcfSKonstantin Belousov 		if (error)
106389b57fcfSKonstantin Belousov 			return (error);
10645e20c11fSAlan Cox 		vmspace = p->p_vmspace;
106505ba50f5SJake Burkholder 		map = &vmspace->vm_map;
1066af9ec885SJohn Dyson 	}
106726f9a767SRodney W. Grimes 
10686297a3d8SKonstantin Belousov 	/* Map a shared page */
10696297a3d8SKonstantin Belousov 	obj = sv->sv_shared_page_obj;
10706297a3d8SKonstantin Belousov 	if (obj != NULL) {
10716297a3d8SKonstantin Belousov 		vm_object_reference(obj);
10726297a3d8SKonstantin Belousov 		error = vm_map_fixed(map, obj, 0,
10736297a3d8SKonstantin Belousov 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
10741e65d73cSKonstantin Belousov 		    VM_PROT_READ | VM_PROT_EXECUTE,
10751e65d73cSKonstantin Belousov 		    VM_PROT_READ | VM_PROT_EXECUTE,
10761e65d73cSKonstantin Belousov 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
10776297a3d8SKonstantin Belousov 		if (error) {
10786297a3d8SKonstantin Belousov 			vm_object_deallocate(obj);
10796297a3d8SKonstantin Belousov 			return (error);
10806297a3d8SKonstantin Belousov 		}
10816297a3d8SKonstantin Belousov 	}
10826297a3d8SKonstantin Belousov 
108326f9a767SRodney W. Grimes 	/* Allocate a new stack */
1084316b3843SKonstantin Belousov 	if (imgp->stack_sz != 0) {
1085fe8a824cSKonstantin Belousov 		ssiz = trunc_page(imgp->stack_sz);
1086316b3843SKonstantin Belousov 		PROC_LOCK(p);
1087f6f6d240SMateusz Guzik 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1088316b3843SKonstantin Belousov 		PROC_UNLOCK(p);
1089316b3843SKonstantin Belousov 		if (ssiz > rlim_stack.rlim_max)
1090316b3843SKonstantin Belousov 			ssiz = rlim_stack.rlim_max;
1091316b3843SKonstantin Belousov 		if (ssiz > rlim_stack.rlim_cur) {
1092316b3843SKonstantin Belousov 			rlim_stack.rlim_cur = ssiz;
1093316b3843SKonstantin Belousov 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1094316b3843SKonstantin Belousov 		}
1095316b3843SKonstantin Belousov 	} else if (sv->sv_maxssiz != NULL) {
109659d8f3ffSJohn Baldwin 		ssiz = *sv->sv_maxssiz;
1097316b3843SKonstantin Belousov 	} else {
109859d8f3ffSJohn Baldwin 		ssiz = maxssiz;
1099316b3843SKonstantin Belousov 	}
110059d8f3ffSJohn Baldwin 	stack_addr = sv->sv_usrstack - ssiz;
110159d8f3ffSJohn Baldwin 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
11026297a3d8SKonstantin Belousov 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
11036297a3d8SKonstantin Belousov 		sv->sv_stackprot,
11046297a3d8SKonstantin Belousov 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
11052267af78SJulian Elischer 	if (error)
11062267af78SJulian Elischer 		return (error);
11072267af78SJulian Elischer 
110878960940SAlan Cox 	/*
110978960940SAlan Cox 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
111078960940SAlan Cox 	 * are still used to enforce the stack rlimit on the process stack.
11112267af78SJulian Elischer 	 */
1112cbc89bfbSPaul Saab 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
11136ef12002SKonstantin Belousov 	vmspace->vm_maxsaddr = (char *)stack_addr;
111426f9a767SRodney W. Grimes 
111526f9a767SRodney W. Grimes 	return (0);
111626f9a767SRodney W. Grimes }
111726f9a767SRodney W. Grimes 
111826f9a767SRodney W. Grimes /*
1119873fbcd7SRobert Watson  * Copy out argument and environment strings from the old process address
1120873fbcd7SRobert Watson  * space into the temporary string buffer.
112126f9a767SRodney W. Grimes  */
112226f9a767SRodney W. Grimes int
1123610ecfe0SMaxim Sobolev exec_copyin_args(struct image_args *args, char *fname,
1124610ecfe0SMaxim Sobolev     enum uio_seg segflg, char **argv, char **envv)
112526f9a767SRodney W. Grimes {
11260a2c94b8SKonstantin Belousov 	u_long argp, envp;
1127ecbb00a2SDoug Rabson 	int error;
1128ecbb00a2SDoug Rabson 	size_t length;
112926f9a767SRodney W. Grimes 
1130610ecfe0SMaxim Sobolev 	bzero(args, sizeof(*args));
1131610ecfe0SMaxim Sobolev 	if (argv == NULL)
1132610ecfe0SMaxim Sobolev 		return (EFAULT);
11332af6e14dSAlan Cox 
113426f9a767SRodney W. Grimes 	/*
11352af6e14dSAlan Cox 	 * Allocate demand-paged memory for the file name, argument, and
11362af6e14dSAlan Cox 	 * environment strings.
113726f9a767SRodney W. Grimes 	 */
11382af6e14dSAlan Cox 	error = exec_alloc_args(args);
11392af6e14dSAlan Cox 	if (error != 0)
11402af6e14dSAlan Cox 		return (error);
11419e4e5114SAlan Cox 
1142610ecfe0SMaxim Sobolev 	/*
1143610ecfe0SMaxim Sobolev 	 * Copy the file name.
1144610ecfe0SMaxim Sobolev 	 */
1145632dbc19SKonstantin Belousov 	if (fname != NULL) {
11462af6e14dSAlan Cox 		args->fname = args->buf;
1147610ecfe0SMaxim Sobolev 		error = (segflg == UIO_SYSSPACE) ?
1148610ecfe0SMaxim Sobolev 		    copystr(fname, args->fname, PATH_MAX, &length) :
1149610ecfe0SMaxim Sobolev 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1150c30af532SMaxim Sobolev 		if (error != 0)
115168ff3c24SStephan Uphoff 			goto err_exit;
1152632dbc19SKonstantin Belousov 	} else
11539e4e5114SAlan Cox 		length = 0;
11549e4e5114SAlan Cox 
11552af6e14dSAlan Cox 	args->begin_argv = args->buf + length;
11569e4e5114SAlan Cox 	args->endp = args->begin_argv;
11579e4e5114SAlan Cox 	args->stringspace = ARG_MAX;
115826f9a767SRodney W. Grimes 
1159610ecfe0SMaxim Sobolev 	/*
1160610ecfe0SMaxim Sobolev 	 * extract arguments first
1161610ecfe0SMaxim Sobolev 	 */
11620a2c94b8SKonstantin Belousov 	for (;;) {
11630a2c94b8SKonstantin Belousov 		error = fueword(argv++, &argp);
11640a2c94b8SKonstantin Belousov 		if (error == -1) {
116568ff3c24SStephan Uphoff 			error = EFAULT;
116668ff3c24SStephan Uphoff 			goto err_exit;
116768ff3c24SStephan Uphoff 		}
11680a2c94b8SKonstantin Belousov 		if (argp == 0)
11690a2c94b8SKonstantin Belousov 			break;
11700a2c94b8SKonstantin Belousov 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
11710a2c94b8SKonstantin Belousov 		    args->stringspace, &length);
11720a2c94b8SKonstantin Belousov 		if (error != 0) {
1173610ecfe0SMaxim Sobolev 			if (error == ENAMETOOLONG)
117468ff3c24SStephan Uphoff 				error = E2BIG;
117568ff3c24SStephan Uphoff 			goto err_exit;
1176610ecfe0SMaxim Sobolev 		}
1177610ecfe0SMaxim Sobolev 		args->stringspace -= length;
1178610ecfe0SMaxim Sobolev 		args->endp += length;
1179610ecfe0SMaxim Sobolev 		args->argc++;
1180610ecfe0SMaxim Sobolev 	}
1181610ecfe0SMaxim Sobolev 
1182610ecfe0SMaxim Sobolev 	args->begin_envv = args->endp;
1183b9df5231SPoul-Henning Kamp 
118426f9a767SRodney W. Grimes 	/*
118526f9a767SRodney W. Grimes 	 * extract environment strings
118626f9a767SRodney W. Grimes 	 */
11870fd1a014SDavid Greenman 	if (envv) {
11880a2c94b8SKonstantin Belousov 		for (;;) {
11890a2c94b8SKonstantin Belousov 			error = fueword(envv++, &envp);
11900a2c94b8SKonstantin Belousov 			if (error == -1) {
119168ff3c24SStephan Uphoff 				error = EFAULT;
119268ff3c24SStephan Uphoff 				goto err_exit;
119368ff3c24SStephan Uphoff 			}
11940a2c94b8SKonstantin Belousov 			if (envp == 0)
11950a2c94b8SKonstantin Belousov 				break;
11960a2c94b8SKonstantin Belousov 			error = copyinstr((void *)(uintptr_t)envp,
11970a2c94b8SKonstantin Belousov 			    args->endp, args->stringspace, &length);
11980a2c94b8SKonstantin Belousov 			if (error != 0) {
11990fd1a014SDavid Greenman 				if (error == ENAMETOOLONG)
120068ff3c24SStephan Uphoff 					error = E2BIG;
120168ff3c24SStephan Uphoff 				goto err_exit;
12020fd1a014SDavid Greenman 			}
1203610ecfe0SMaxim Sobolev 			args->stringspace -= length;
1204610ecfe0SMaxim Sobolev 			args->endp += length;
1205610ecfe0SMaxim Sobolev 			args->envc++;
120626f9a767SRodney W. Grimes 		}
12070fd1a014SDavid Greenman 	}
120826f9a767SRodney W. Grimes 
120926f9a767SRodney W. Grimes 	return (0);
121068ff3c24SStephan Uphoff 
121168ff3c24SStephan Uphoff err_exit:
121268ff3c24SStephan Uphoff 	exec_free_args(args);
121368ff3c24SStephan Uphoff 	return (error);
121426f9a767SRodney W. Grimes }
121526f9a767SRodney W. Grimes 
1216457f7e23SEd Schouten int
1217457f7e23SEd Schouten exec_copyin_data_fds(struct thread *td, struct image_args *args,
1218457f7e23SEd Schouten     const void *data, size_t datalen, const int *fds, size_t fdslen)
1219457f7e23SEd Schouten {
1220457f7e23SEd Schouten 	struct filedesc *ofdp;
1221457f7e23SEd Schouten 	const char *p;
1222457f7e23SEd Schouten 	int *kfds;
1223457f7e23SEd Schouten 	int error;
1224457f7e23SEd Schouten 
1225457f7e23SEd Schouten 	memset(args, '\0', sizeof(*args));
1226457f7e23SEd Schouten 	ofdp = td->td_proc->p_fd;
1227457f7e23SEd Schouten 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1228457f7e23SEd Schouten 		return (E2BIG);
1229457f7e23SEd Schouten 	error = exec_alloc_args(args);
1230457f7e23SEd Schouten 	if (error != 0)
1231457f7e23SEd Schouten 		return (error);
1232457f7e23SEd Schouten 
1233457f7e23SEd Schouten 	args->begin_argv = args->buf;
1234457f7e23SEd Schouten 	args->stringspace = ARG_MAX;
1235457f7e23SEd Schouten 
1236457f7e23SEd Schouten 	if (datalen > 0) {
1237457f7e23SEd Schouten 		/*
1238457f7e23SEd Schouten 		 * Argument buffer has been provided. Copy it into the
1239457f7e23SEd Schouten 		 * kernel as a single string and add a terminating null
1240457f7e23SEd Schouten 		 * byte.
1241457f7e23SEd Schouten 		 */
1242457f7e23SEd Schouten 		error = copyin(data, args->begin_argv, datalen);
1243457f7e23SEd Schouten 		if (error != 0)
1244457f7e23SEd Schouten 			goto err_exit;
1245457f7e23SEd Schouten 		args->begin_argv[datalen] = '\0';
1246457f7e23SEd Schouten 		args->endp = args->begin_argv + datalen + 1;
1247457f7e23SEd Schouten 		args->stringspace -= datalen + 1;
1248457f7e23SEd Schouten 
1249457f7e23SEd Schouten 		/*
1250457f7e23SEd Schouten 		 * Traditional argument counting. Count the number of
1251457f7e23SEd Schouten 		 * null bytes.
1252457f7e23SEd Schouten 		 */
1253457f7e23SEd Schouten 		for (p = args->begin_argv; p < args->endp; ++p)
1254457f7e23SEd Schouten 			if (*p == '\0')
1255457f7e23SEd Schouten 				++args->argc;
1256457f7e23SEd Schouten 	} else {
1257457f7e23SEd Schouten 		/* No argument buffer provided. */
1258457f7e23SEd Schouten 		args->endp = args->begin_argv;
1259457f7e23SEd Schouten 	}
1260457f7e23SEd Schouten 	/* There are no environment variables. */
1261457f7e23SEd Schouten 	args->begin_envv = args->endp;
1262457f7e23SEd Schouten 
1263457f7e23SEd Schouten 	/* Create new file descriptor table. */
1264457f7e23SEd Schouten 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1265457f7e23SEd Schouten 	error = copyin(fds, kfds, fdslen * sizeof(int));
1266457f7e23SEd Schouten 	if (error != 0) {
1267457f7e23SEd Schouten 		free(kfds, M_TEMP);
1268457f7e23SEd Schouten 		goto err_exit;
1269457f7e23SEd Schouten 	}
1270457f7e23SEd Schouten 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1271457f7e23SEd Schouten 	free(kfds, M_TEMP);
1272457f7e23SEd Schouten 	if (error != 0)
1273457f7e23SEd Schouten 		goto err_exit;
1274457f7e23SEd Schouten 
1275457f7e23SEd Schouten 	return (0);
1276457f7e23SEd Schouten err_exit:
1277457f7e23SEd Schouten 	exec_free_args(args);
1278457f7e23SEd Schouten 	return (error);
1279457f7e23SEd Schouten }
1280457f7e23SEd Schouten 
12812af6e14dSAlan Cox /*
12822af6e14dSAlan Cox  * Allocate temporary demand-paged, zero-filled memory for the file name,
12832af6e14dSAlan Cox  * argument, and environment strings.  Returns zero if the allocation succeeds
12842af6e14dSAlan Cox  * and ENOMEM otherwise.
12852af6e14dSAlan Cox  */
12862af6e14dSAlan Cox int
12872af6e14dSAlan Cox exec_alloc_args(struct image_args *args)
12882af6e14dSAlan Cox {
12892af6e14dSAlan Cox 
12905df87b21SJeff Roberson 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
12912af6e14dSAlan Cox 	return (args->buf != NULL ? 0 : ENOMEM);
12922af6e14dSAlan Cox }
12932af6e14dSAlan Cox 
129469a8f9e3SAlan Cox void
1295610ecfe0SMaxim Sobolev exec_free_args(struct image_args *args)
1296610ecfe0SMaxim Sobolev {
1297610ecfe0SMaxim Sobolev 
129869a8f9e3SAlan Cox 	if (args->buf != NULL) {
12995df87b21SJeff Roberson 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
13002af6e14dSAlan Cox 		    PATH_MAX + ARG_MAX);
1301610ecfe0SMaxim Sobolev 		args->buf = NULL;
1302610ecfe0SMaxim Sobolev 	}
1303a14a9498SAlan Cox 	if (args->fname_buf != NULL) {
1304a14a9498SAlan Cox 		free(args->fname_buf, M_TEMP);
1305a14a9498SAlan Cox 		args->fname_buf = NULL;
1306a14a9498SAlan Cox 	}
1307457f7e23SEd Schouten 	if (args->fdp != NULL)
1308457f7e23SEd Schouten 		fdescfree_remapped(args->fdp);
1309610ecfe0SMaxim Sobolev }
1310610ecfe0SMaxim Sobolev 
131126f9a767SRodney W. Grimes /*
1312873fbcd7SRobert Watson  * Copy strings out to the new process address space, constructing new arg
1313873fbcd7SRobert Watson  * and env vector tables. Return a pointer to the base so that it can be used
1314873fbcd7SRobert Watson  * as the initial stack pointer.
131526f9a767SRodney W. Grimes  */
1316654f6be1SBruce Evans register_t *
1317c52007c2SDavid Greenman exec_copyout_strings(imgp)
1318c52007c2SDavid Greenman 	struct image_params *imgp;
131926f9a767SRodney W. Grimes {
132026f9a767SRodney W. Grimes 	int argc, envc;
132126f9a767SRodney W. Grimes 	char **vectp;
132288b124ceSKonstantin Belousov 	char *stringp;
132388b124ceSKonstantin Belousov 	uintptr_t destp;
1324654f6be1SBruce Evans 	register_t *stack_base;
132593f6448cSDavid Greenman 	struct ps_strings *arginfo;
1326f3bec5d7SJake Burkholder 	struct proc *p;
13273ff06357SKonstantin Belousov 	size_t execpath_len;
1328ee235befSKonstantin Belousov 	int szsigcode, szps;
1329ee235befSKonstantin Belousov 	char canary[sizeof(long) * 8];
133026f9a767SRodney W. Grimes 
1331ee235befSKonstantin Belousov 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
133226f9a767SRodney W. Grimes 	/*
133326f9a767SRodney W. Grimes 	 * Calculate string base and vector table pointers.
1334d66a5066SPeter Wemm 	 * Also deal with signal trampoline code for this exec type.
133526f9a767SRodney W. Grimes 	 */
13363ff06357SKonstantin Belousov 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
13373ff06357SKonstantin Belousov 		execpath_len = strlen(imgp->execpath) + 1;
13383ff06357SKonstantin Belousov 	else
13393ff06357SKonstantin Belousov 		execpath_len = 0;
1340f3bec5d7SJake Burkholder 	p = imgp->proc;
1341f3bec5d7SJake Burkholder 	szsigcode = 0;
134205ba50f5SJake Burkholder 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
13436297a3d8SKonstantin Belousov 	if (p->p_sysent->sv_sigcode_base == 0) {
1344f3bec5d7SJake Burkholder 		if (p->p_sysent->sv_szsigcode != NULL)
1345f3bec5d7SJake Burkholder 			szsigcode = *(p->p_sysent->sv_szsigcode);
13466297a3d8SKonstantin Belousov 	}
134788b124ceSKonstantin Belousov 	destp =	(uintptr_t)arginfo;
13481ed012f9SPeter Wemm 
134926f9a767SRodney W. Grimes 	/*
1350d66a5066SPeter Wemm 	 * install sigcode
1351d66a5066SPeter Wemm 	 */
135288b124ceSKonstantin Belousov 	if (szsigcode != 0) {
135388b124ceSKonstantin Belousov 		destp -= szsigcode;
135488b124ceSKonstantin Belousov 		destp = rounddown2(destp, sizeof(void *));
135588b124ceSKonstantin Belousov 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
135688b124ceSKonstantin Belousov 	}
1357d66a5066SPeter Wemm 
1358d66a5066SPeter Wemm 	/*
13593ff06357SKonstantin Belousov 	 * Copy the image path for the rtld.
13603ff06357SKonstantin Belousov 	 */
13613ff06357SKonstantin Belousov 	if (execpath_len != 0) {
136288b124ceSKonstantin Belousov 		destp -= execpath_len;
136388b124ceSKonstantin Belousov 		imgp->execpathp = destp;
136488b124ceSKonstantin Belousov 		copyout(imgp->execpath, (void *)destp, execpath_len);
13653ff06357SKonstantin Belousov 	}
13663ff06357SKonstantin Belousov 
13673ff06357SKonstantin Belousov 	/*
1368ee235befSKonstantin Belousov 	 * Prepare the canary for SSP.
1369ee235befSKonstantin Belousov 	 */
1370ee235befSKonstantin Belousov 	arc4rand(canary, sizeof(canary), 0);
137188b124ceSKonstantin Belousov 	destp -= sizeof(canary);
137288b124ceSKonstantin Belousov 	imgp->canary = destp;
137388b124ceSKonstantin Belousov 	copyout(canary, (void *)destp, sizeof(canary));
1374ee235befSKonstantin Belousov 	imgp->canarylen = sizeof(canary);
1375ee235befSKonstantin Belousov 
1376ee235befSKonstantin Belousov 	/*
1377ee235befSKonstantin Belousov 	 * Prepare the pagesizes array.
1378ee235befSKonstantin Belousov 	 */
137988b124ceSKonstantin Belousov 	destp -= szps;
138088b124ceSKonstantin Belousov 	destp = rounddown2(destp, sizeof(void *));
138188b124ceSKonstantin Belousov 	imgp->pagesizes = destp;
138288b124ceSKonstantin Belousov 	copyout(pagesizes, (void *)destp, szps);
1383ee235befSKonstantin Belousov 	imgp->pagesizeslen = szps;
1384ee235befSKonstantin Belousov 
138588b124ceSKonstantin Belousov 	destp -= ARG_MAX - imgp->args->stringspace;
138688b124ceSKonstantin Belousov 	destp = rounddown2(destp, sizeof(void *));
138788b124ceSKonstantin Belousov 
1388ee235befSKonstantin Belousov 	/*
1389e1743d02SSøren Schmidt 	 * If we have a valid auxargs ptr, prepare some room
1390e1743d02SSøren Schmidt 	 * on the stack.
1391e1743d02SSøren Schmidt 	 */
1392b9a22da4STakanori Watanabe 	if (imgp->auxargs) {
1393e1743d02SSøren Schmidt 		/*
1394b9a22da4STakanori Watanabe 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1395b9a22da4STakanori Watanabe 		 * lower compatibility.
1396b9a22da4STakanori Watanabe 		 */
1397ca0387efSJake Burkholder 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1398ca0387efSJake Burkholder 		    (AT_COUNT * 2);
1399b9a22da4STakanori Watanabe 		/*
1400b9a22da4STakanori Watanabe 		 * The '+ 2' is for the null pointers at the end of each of
1401b9a22da4STakanori Watanabe 		 * the arg and env vector sets,and imgp->auxarg_size is room
1402b9a22da4STakanori Watanabe 		 * for argument of Runtime loader.
1403e1743d02SSøren Schmidt 		 */
1404610ecfe0SMaxim Sobolev 		vectp = (char **)(destp - (imgp->args->argc +
1405ee235befSKonstantin Belousov 		    imgp->args->envc + 2 + imgp->auxarg_size)
1406ee235befSKonstantin Belousov 		    * sizeof(char *));
1407610ecfe0SMaxim Sobolev 	} else {
1408e1743d02SSøren Schmidt 		/*
1409b9a22da4STakanori Watanabe 		 * The '+ 2' is for the null pointers at the end of each of
1410b9a22da4STakanori Watanabe 		 * the arg and env vector sets
141126f9a767SRodney W. Grimes 		 */
141288b124ceSKonstantin Belousov 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
141388b124ceSKonstantin Belousov 		    + 2) * sizeof(char *));
1414610ecfe0SMaxim Sobolev 	}
141526f9a767SRodney W. Grimes 
141626f9a767SRodney W. Grimes 	/*
141726f9a767SRodney W. Grimes 	 * vectp also becomes our initial stack base
141826f9a767SRodney W. Grimes 	 */
1419654f6be1SBruce Evans 	stack_base = (register_t *)vectp;
142026f9a767SRodney W. Grimes 
1421610ecfe0SMaxim Sobolev 	stringp = imgp->args->begin_argv;
1422610ecfe0SMaxim Sobolev 	argc = imgp->args->argc;
1423610ecfe0SMaxim Sobolev 	envc = imgp->args->envc;
142426f9a767SRodney W. Grimes 
142593f6448cSDavid Greenman 	/*
14263aed948bSDavid Greenman 	 * Copy out strings - arguments and environment.
142793f6448cSDavid Greenman 	 */
142888b124ceSKonstantin Belousov 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
142993f6448cSDavid Greenman 
143093f6448cSDavid Greenman 	/*
14313aed948bSDavid Greenman 	 * Fill in "ps_strings" struct for ps, w, etc.
14323aed948bSDavid Greenman 	 */
1433aae0aa45SBruce Evans 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1434f4e26adeSNathan Whitehorn 	suword32(&arginfo->ps_nargvstr, argc);
14353aed948bSDavid Greenman 
14363aed948bSDavid Greenman 	/*
14373aed948bSDavid Greenman 	 * Fill in argument portion of vector table.
143893f6448cSDavid Greenman 	 */
143926f9a767SRodney W. Grimes 	for (; argc > 0; --argc) {
1440aae0aa45SBruce Evans 		suword(vectp++, (long)(intptr_t)destp);
14413aed948bSDavid Greenman 		while (*stringp++ != 0)
14423aed948bSDavid Greenman 			destp++;
14433aed948bSDavid Greenman 		destp++;
144426f9a767SRodney W. Grimes 	}
144526f9a767SRodney W. Grimes 
14461a6e52d0SJeroen Ruigrok van der Werven 	/* a null vector table pointer separates the argp's from the envp's */
14476ab46d52SBruce Evans 	suword(vectp++, 0);
144826f9a767SRodney W. Grimes 
1449aae0aa45SBruce Evans 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1450f4e26adeSNathan Whitehorn 	suword32(&arginfo->ps_nenvstr, envc);
145193f6448cSDavid Greenman 
145293f6448cSDavid Greenman 	/*
14533aed948bSDavid Greenman 	 * Fill in environment portion of vector table.
145493f6448cSDavid Greenman 	 */
145526f9a767SRodney W. Grimes 	for (; envc > 0; --envc) {
1456aae0aa45SBruce Evans 		suword(vectp++, (long)(intptr_t)destp);
14573aed948bSDavid Greenman 		while (*stringp++ != 0)
14583aed948bSDavid Greenman 			destp++;
14593aed948bSDavid Greenman 		destp++;
146026f9a767SRodney W. Grimes 	}
146126f9a767SRodney W. Grimes 
146226f9a767SRodney W. Grimes 	/* end of vector table is a null pointer */
14636ab46d52SBruce Evans 	suword(vectp, 0);
146426f9a767SRodney W. Grimes 
146526f9a767SRodney W. Grimes 	return (stack_base);
146626f9a767SRodney W. Grimes }
146726f9a767SRodney W. Grimes 
146826f9a767SRodney W. Grimes /*
146926f9a767SRodney W. Grimes  * Check permissions of file to execute.
1470cf64863aSRobert Watson  *	Called with imgp->vp locked.
147126f9a767SRodney W. Grimes  *	Return 0 for success or error code on failure.
147226f9a767SRodney W. Grimes  */
1473c8a79999SPeter Wemm int
1474c52007c2SDavid Greenman exec_check_permissions(imgp)
1475c52007c2SDavid Greenman 	struct image_params *imgp;
147626f9a767SRodney W. Grimes {
1477c52007c2SDavid Greenman 	struct vnode *vp = imgp->vp;
1478c52007c2SDavid Greenman 	struct vattr *attr = imgp->attr;
1479a854ed98SJohn Baldwin 	struct thread *td;
1480140dedb8SKonstantin Belousov 	int error, writecount;
148126f9a767SRodney W. Grimes 
14826617724cSJeff Roberson 	td = curthread;
1483339b79b9SRobert Watson 
1484ec35c2afSRobert Watson 	/* Get file attributes */
14850359a12eSAttilio Rao 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1486ec35c2afSRobert Watson 	if (error)
1487ec35c2afSRobert Watson 		return (error);
1488ec35c2afSRobert Watson 
1489339b79b9SRobert Watson #ifdef MAC
149030d239bcSRobert Watson 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1491339b79b9SRobert Watson 	if (error)
1492339b79b9SRobert Watson 		return (error);
1493339b79b9SRobert Watson #endif
1494339b79b9SRobert Watson 
149526f9a767SRodney W. Grimes 	/*
1496de478dd4SJaakko Heinonen 	 * 1) Check if file execution is disabled for the filesystem that
1497de478dd4SJaakko Heinonen 	 *    this file resides on.
1498de478dd4SJaakko Heinonen 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1499de478dd4SJaakko Heinonen 	 *    privileged user will always succeed, and we don't want this
1500de478dd4SJaakko Heinonen 	 *    to happen unless the file really is executable.
1501de478dd4SJaakko Heinonen 	 * 3) Ensure that the file is a regular file.
150226f9a767SRodney W. Grimes 	 */
1503c52007c2SDavid Greenman 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1504de478dd4SJaakko Heinonen 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1505a854ed98SJohn Baldwin 	    (attr->va_type != VREG))
150626f9a767SRodney W. Grimes 		return (EACCES);
150726f9a767SRodney W. Grimes 
150826f9a767SRodney W. Grimes 	/*
150926f9a767SRodney W. Grimes 	 * Zero length files can't be exec'd
151026f9a767SRodney W. Grimes 	 */
151126f9a767SRodney W. Grimes 	if (attr->va_size == 0)
151226f9a767SRodney W. Grimes 		return (ENOEXEC);
151326f9a767SRodney W. Grimes 
151426f9a767SRodney W. Grimes 	/*
151526f9a767SRodney W. Grimes 	 *  Check for execute permission to file based on current credentials.
151626f9a767SRodney W. Grimes 	 */
1517a854ed98SJohn Baldwin 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
151826f9a767SRodney W. Grimes 	if (error)
151926f9a767SRodney W. Grimes 		return (error);
152026f9a767SRodney W. Grimes 
15216d5a0a8cSDavid Greenman 	/*
15226d5a0a8cSDavid Greenman 	 * Check number of open-for-writes on the file and deny execution
15236d5a0a8cSDavid Greenman 	 * if there are any.
15246d5a0a8cSDavid Greenman 	 */
1525140dedb8SKonstantin Belousov 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1526140dedb8SKonstantin Belousov 	if (error != 0)
1527140dedb8SKonstantin Belousov 		return (error);
1528140dedb8SKonstantin Belousov 	if (writecount != 0)
15296d5a0a8cSDavid Greenman 		return (ETXTBSY);
15306d5a0a8cSDavid Greenman 
15316d5a0a8cSDavid Greenman 	/*
15326d5a0a8cSDavid Greenman 	 * Call filesystem specific open routine (which does nothing in the
15336d5a0a8cSDavid Greenman 	 * general case).
15346d5a0a8cSDavid Greenman 	 */
15359e223287SKonstantin Belousov 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
15369a75ea23SKonstantin Belousov 	if (error == 0)
15379a75ea23SKonstantin Belousov 		imgp->opened = 1;
153826f9a767SRodney W. Grimes 	return (error);
1539df8bae1dSRodney W. Grimes }
1540aa855a59SPeter Wemm 
1541aa855a59SPeter Wemm /*
1542aa855a59SPeter Wemm  * Exec handler registration
1543aa855a59SPeter Wemm  */
1544aa855a59SPeter Wemm int
1545aa855a59SPeter Wemm exec_register(execsw_arg)
1546aa855a59SPeter Wemm 	const struct execsw *execsw_arg;
1547aa855a59SPeter Wemm {
1548aa855a59SPeter Wemm 	const struct execsw **es, **xs, **newexecsw;
1549aa855a59SPeter Wemm 	int count = 2;	/* New slot and trailing NULL */
1550aa855a59SPeter Wemm 
1551aa855a59SPeter Wemm 	if (execsw)
1552aa855a59SPeter Wemm 		for (es = execsw; *es; es++)
1553aa855a59SPeter Wemm 			count++;
1554a163d034SWarner Losh 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1555aa855a59SPeter Wemm 	if (newexecsw == NULL)
1556a7cddfedSJake Burkholder 		return (ENOMEM);
1557aa855a59SPeter Wemm 	xs = newexecsw;
1558aa855a59SPeter Wemm 	if (execsw)
1559aa855a59SPeter Wemm 		for (es = execsw; *es; es++)
1560aa855a59SPeter Wemm 			*xs++ = *es;
1561aa855a59SPeter Wemm 	*xs++ = execsw_arg;
1562aa855a59SPeter Wemm 	*xs = NULL;
1563aa855a59SPeter Wemm 	if (execsw)
1564aa855a59SPeter Wemm 		free(execsw, M_TEMP);
1565aa855a59SPeter Wemm 	execsw = newexecsw;
1566a7cddfedSJake Burkholder 	return (0);
1567aa855a59SPeter Wemm }
1568aa855a59SPeter Wemm 
1569aa855a59SPeter Wemm int
1570aa855a59SPeter Wemm exec_unregister(execsw_arg)
1571aa855a59SPeter Wemm 	const struct execsw *execsw_arg;
1572aa855a59SPeter Wemm {
1573aa855a59SPeter Wemm 	const struct execsw **es, **xs, **newexecsw;
1574aa855a59SPeter Wemm 	int count = 1;
1575aa855a59SPeter Wemm 
1576aa855a59SPeter Wemm 	if (execsw == NULL)
1577aa855a59SPeter Wemm 		panic("unregister with no handlers left?\n");
1578aa855a59SPeter Wemm 
1579aa855a59SPeter Wemm 	for (es = execsw; *es; es++) {
1580aa855a59SPeter Wemm 		if (*es == execsw_arg)
1581aa855a59SPeter Wemm 			break;
1582aa855a59SPeter Wemm 	}
1583aa855a59SPeter Wemm 	if (*es == NULL)
1584a7cddfedSJake Burkholder 		return (ENOENT);
1585aa855a59SPeter Wemm 	for (es = execsw; *es; es++)
1586aa855a59SPeter Wemm 		if (*es != execsw_arg)
1587aa855a59SPeter Wemm 			count++;
1588a163d034SWarner Losh 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1589aa855a59SPeter Wemm 	if (newexecsw == NULL)
1590a7cddfedSJake Burkholder 		return (ENOMEM);
1591aa855a59SPeter Wemm 	xs = newexecsw;
1592aa855a59SPeter Wemm 	for (es = execsw; *es; es++)
1593aa855a59SPeter Wemm 		if (*es != execsw_arg)
1594aa855a59SPeter Wemm 			*xs++ = *es;
1595aa855a59SPeter Wemm 	*xs = NULL;
1596aa855a59SPeter Wemm 	if (execsw)
1597aa855a59SPeter Wemm 		free(execsw, M_TEMP);
1598aa855a59SPeter Wemm 	execsw = newexecsw;
1599a7cddfedSJake Burkholder 	return (0);
1600aa855a59SPeter Wemm }
1601