xref: /freebsd-12.1/sys/kern/kern_exec.c (revision 8f817ae5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1993, David Greenman
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/shm.h>
64 #include <sys/signalvar.h>
65 #include <sys/smp.h>
66 #include <sys/stat.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/vnode.h>
72 #include <sys/wait.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_pager.h>
86 
87 #ifdef	HWPMC_HOOKS
88 #include <sys/pmckern.h>
89 #endif
90 
91 #include <machine/reg.h>
92 
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
95 
96 #ifdef KDTRACE_HOOKS
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t	dtrace_fasttrap_exec;
99 #endif
100 
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110     &coredump_pack_fileinfo, 0,
111     "Enable file path packing in 'procstat -f' coredump notes");
112 
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115     &coredump_pack_vmmapinfo, 0,
116     "Enable file path packing in 'procstat -v' coredump notes");
117 
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
122     struct mac *mac_p);
123 
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
127 
128 /* XXX This should be vm_size_t. */
129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
130     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
131 
132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
133     NULL, 0, sysctl_kern_stackprot, "I", "");
134 
135 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
137     &ps_arg_cache_limit, 0, "");
138 
139 static int disallow_high_osrel;
140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
141     &disallow_high_osrel, 0,
142     "Disallow execution of binaries built for higher version of the world");
143 
144 static int map_at_zero = 0;
145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
146     "Permit processes to map an object at virtual address 0.");
147 
148 EVENTHANDLER_LIST_DECLARE(process_exec);
149 
150 static int
sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)151 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
152 {
153 	struct proc *p;
154 	int error;
155 
156 	p = curproc;
157 #ifdef SCTL_MASK32
158 	if (req->flags & SCTL_MASK32) {
159 		unsigned int val;
160 		val = (unsigned int)p->p_sysent->sv_psstrings;
161 		error = SYSCTL_OUT(req, &val, sizeof(val));
162 	} else
163 #endif
164 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
165 		   sizeof(p->p_sysent->sv_psstrings));
166 	return error;
167 }
168 
169 static int
sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
171 {
172 	struct proc *p;
173 	int error;
174 
175 	p = curproc;
176 #ifdef SCTL_MASK32
177 	if (req->flags & SCTL_MASK32) {
178 		unsigned int val;
179 		val = (unsigned int)p->p_sysent->sv_usrstack;
180 		error = SYSCTL_OUT(req, &val, sizeof(val));
181 	} else
182 #endif
183 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
184 		    sizeof(p->p_sysent->sv_usrstack));
185 	return error;
186 }
187 
188 static int
sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
190 {
191 	struct proc *p;
192 
193 	p = curproc;
194 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
195 	    sizeof(p->p_sysent->sv_stackprot)));
196 }
197 
198 /*
199  * Each of the items is a pointer to a `const struct execsw', hence the
200  * double pointer here.
201  */
202 static const struct execsw **execsw;
203 
204 #ifndef _SYS_SYSPROTO_H_
205 struct execve_args {
206 	char    *fname;
207 	char    **argv;
208 	char    **envv;
209 };
210 #endif
211 
212 int
sys_execve(struct thread * td,struct execve_args * uap)213 sys_execve(struct thread *td, struct execve_args *uap)
214 {
215 	struct image_args args;
216 	struct vmspace *oldvmspace;
217 	int error;
218 
219 	error = pre_execve(td, &oldvmspace);
220 	if (error != 0)
221 		return (error);
222 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
223 	    uap->argv, uap->envv);
224 	if (error == 0)
225 		error = kern_execve(td, &args, NULL);
226 	post_execve(td, error, oldvmspace);
227 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
228 	return (error);
229 }
230 
231 #ifndef _SYS_SYSPROTO_H_
232 struct fexecve_args {
233 	int	fd;
234 	char	**argv;
235 	char	**envv;
236 }
237 #endif
238 int
sys_fexecve(struct thread * td,struct fexecve_args * uap)239 sys_fexecve(struct thread *td, struct fexecve_args *uap)
240 {
241 	struct image_args args;
242 	struct vmspace *oldvmspace;
243 	int error;
244 
245 	error = pre_execve(td, &oldvmspace);
246 	if (error != 0)
247 		return (error);
248 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
249 	    uap->argv, uap->envv);
250 	if (error == 0) {
251 		args.fd = uap->fd;
252 		error = kern_execve(td, &args, NULL);
253 	}
254 	post_execve(td, error, oldvmspace);
255 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
256 	return (error);
257 }
258 
259 #ifndef _SYS_SYSPROTO_H_
260 struct __mac_execve_args {
261 	char	*fname;
262 	char	**argv;
263 	char	**envv;
264 	struct mac	*mac_p;
265 };
266 #endif
267 
268 int
sys___mac_execve(struct thread * td,struct __mac_execve_args * uap)269 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
270 {
271 #ifdef MAC
272 	struct image_args args;
273 	struct vmspace *oldvmspace;
274 	int error;
275 
276 	error = pre_execve(td, &oldvmspace);
277 	if (error != 0)
278 		return (error);
279 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
280 	    uap->argv, uap->envv);
281 	if (error == 0)
282 		error = kern_execve(td, &args, uap->mac_p);
283 	post_execve(td, error, oldvmspace);
284 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
285 	return (error);
286 #else
287 	return (ENOSYS);
288 #endif
289 }
290 
291 int
pre_execve(struct thread * td,struct vmspace ** oldvmspace)292 pre_execve(struct thread *td, struct vmspace **oldvmspace)
293 {
294 	struct proc *p;
295 	int error;
296 
297 	KASSERT(td == curthread, ("non-current thread %p", td));
298 	error = 0;
299 	p = td->td_proc;
300 	if ((p->p_flag & P_HADTHREADS) != 0) {
301 		PROC_LOCK(p);
302 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
303 			error = ERESTART;
304 		PROC_UNLOCK(p);
305 	}
306 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
307 	    ("nested execve"));
308 	*oldvmspace = p->p_vmspace;
309 	return (error);
310 }
311 
312 void
post_execve(struct thread * td,int error,struct vmspace * oldvmspace)313 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
314 {
315 	struct proc *p;
316 
317 	KASSERT(td == curthread, ("non-current thread %p", td));
318 	p = td->td_proc;
319 	if ((p->p_flag & P_HADTHREADS) != 0) {
320 		PROC_LOCK(p);
321 		/*
322 		 * If success, we upgrade to SINGLE_EXIT state to
323 		 * force other threads to suicide.
324 		 */
325 		if (error == EJUSTRETURN)
326 			thread_single(p, SINGLE_EXIT);
327 		else
328 			thread_single_end(p, SINGLE_BOUNDARY);
329 		PROC_UNLOCK(p);
330 	}
331 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
332 		KASSERT(p->p_vmspace != oldvmspace,
333 		    ("oldvmspace still used"));
334 		vmspace_free(oldvmspace);
335 		td->td_pflags &= ~TDP_EXECVMSPC;
336 	}
337 }
338 
339 /*
340  * XXX: kern_execve has the astonishing property of not always returning to
341  * the caller.  If sufficiently bad things happen during the call to
342  * do_execve(), it can end up calling exit1(); as a result, callers must
343  * avoid doing anything which they might need to undo (e.g., allocating
344  * memory).
345  */
346 int
kern_execve(struct thread * td,struct image_args * args,struct mac * mac_p)347 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
348 {
349 
350 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
351 	    args->begin_envv - args->begin_argv);
352 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
353 	    args->endp - args->begin_envv);
354 	return (do_execve(td, args, mac_p));
355 }
356 
357 /*
358  * In-kernel implementation of execve().  All arguments are assumed to be
359  * userspace pointers from the passed thread.
360  */
361 static int
do_execve(struct thread * td,struct image_args * args,struct mac * mac_p)362 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
363 {
364 	struct proc *p = td->td_proc;
365 	struct nameidata nd;
366 	struct ucred *oldcred;
367 	struct uidinfo *euip = NULL;
368 	register_t *stack_base;
369 	int error, i;
370 	struct image_params image_params, *imgp;
371 	struct vattr attr;
372 	int (*img_first)(struct image_params *);
373 	struct pargs *oldargs = NULL, *newargs = NULL;
374 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
375 #ifdef KTRACE
376 	struct vnode *tracevp = NULL;
377 	struct ucred *tracecred = NULL;
378 #endif
379 	struct vnode *oldtextvp = NULL, *newtextvp;
380 	int credential_changing;
381 #ifdef MAC
382 	struct label *interpvplabel = NULL;
383 	int will_transition;
384 #endif
385 #ifdef HWPMC_HOOKS
386 	struct pmckern_procexec pe;
387 #endif
388 	static const char fexecv_proc_title[] = "(fexecv)";
389 
390 	imgp = &image_params;
391 
392 	/*
393 	 * Lock the process and set the P_INEXEC flag to indicate that
394 	 * it should be left alone until we're done here.  This is
395 	 * necessary to avoid race conditions - e.g. in ptrace() -
396 	 * that might allow a local user to illicitly obtain elevated
397 	 * privileges.
398 	 */
399 	PROC_LOCK(p);
400 	KASSERT((p->p_flag & P_INEXEC) == 0,
401 	    ("%s(): process already has P_INEXEC flag", __func__));
402 	p->p_flag |= P_INEXEC;
403 	PROC_UNLOCK(p);
404 
405 	/*
406 	 * Initialize part of the common data
407 	 */
408 	bzero(imgp, sizeof(*imgp));
409 	imgp->proc = p;
410 	imgp->attr = &attr;
411 	imgp->args = args;
412 	oldcred = p->p_ucred;
413 
414 #ifdef MAC
415 	error = mac_execve_enter(imgp, mac_p);
416 	if (error)
417 		goto exec_fail;
418 #endif
419 
420 	/*
421 	 * Translate the file name. namei() returns a vnode pointer
422 	 *	in ni_vp among other things.
423 	 *
424 	 * XXXAUDIT: It would be desirable to also audit the name of the
425 	 * interpreter if this is an interpreted binary.
426 	 */
427 	if (args->fname != NULL) {
428 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
429 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
430 	}
431 
432 	SDT_PROBE1(proc, , , exec, args->fname);
433 
434 interpret:
435 	if (args->fname != NULL) {
436 #ifdef CAPABILITY_MODE
437 		/*
438 		 * While capability mode can't reach this point via direct
439 		 * path arguments to execve(), we also don't allow
440 		 * interpreters to be used in capability mode (for now).
441 		 * Catch indirect lookups and return a permissions error.
442 		 */
443 		if (IN_CAPABILITY_MODE(td)) {
444 			error = ECAPMODE;
445 			goto exec_fail;
446 		}
447 #endif
448 		error = namei(&nd);
449 		if (error)
450 			goto exec_fail;
451 
452 		newtextvp = nd.ni_vp;
453 		imgp->vp = newtextvp;
454 	} else {
455 		AUDIT_ARG_FD(args->fd);
456 		/*
457 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
458 		 */
459 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
460 		if (error)
461 			goto exec_fail;
462 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
463 		AUDIT_ARG_VNODE1(newtextvp);
464 		imgp->vp = newtextvp;
465 	}
466 
467 	/*
468 	 * Check file permissions.  Also 'opens' file and sets its vnode to
469 	 * text mode.
470 	 */
471 	error = exec_check_permissions(imgp);
472 	if (error)
473 		goto exec_fail_dealloc;
474 
475 	imgp->object = imgp->vp->v_object;
476 	if (imgp->object != NULL)
477 		vm_object_reference(imgp->object);
478 
479 	error = exec_map_first_page(imgp);
480 	if (error)
481 		goto exec_fail_dealloc;
482 
483 	imgp->proc->p_osrel = 0;
484 	imgp->proc->p_fctl0 = 0;
485 
486 	/*
487 	 * Implement image setuid/setgid.
488 	 *
489 	 * Determine new credentials before attempting image activators
490 	 * so that it can be used by process_exec handlers to determine
491 	 * credential/setid changes.
492 	 *
493 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
494 	 * the process is being traced.
495 	 *
496 	 * We disable setuid/setgid/etc in capability mode on the basis
497 	 * that most setugid applications are not written with that
498 	 * environment in mind, and will therefore almost certainly operate
499 	 * incorrectly. In principle there's no reason that setugid
500 	 * applications might not be useful in capability mode, so we may want
501 	 * to reconsider this conservative design choice in the future.
502 	 *
503 	 * XXXMAC: For the time being, use NOSUID to also prohibit
504 	 * transitions on the file system.
505 	 */
506 	credential_changing = 0;
507 	credential_changing |= (attr.va_mode & S_ISUID) &&
508 	    oldcred->cr_uid != attr.va_uid;
509 	credential_changing |= (attr.va_mode & S_ISGID) &&
510 	    oldcred->cr_gid != attr.va_gid;
511 #ifdef MAC
512 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
513 	    interpvplabel, imgp);
514 	credential_changing |= will_transition;
515 #endif
516 
517 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
518 	if (credential_changing)
519 		imgp->proc->p_pdeathsig = 0;
520 
521 	if (credential_changing &&
522 #ifdef CAPABILITY_MODE
523 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
524 #endif
525 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
526 	    (p->p_flag & P_TRACED) == 0) {
527 		imgp->credential_setid = true;
528 		VOP_UNLOCK(imgp->vp, 0);
529 		imgp->newcred = crdup(oldcred);
530 		if (attr.va_mode & S_ISUID) {
531 			euip = uifind(attr.va_uid);
532 			change_euid(imgp->newcred, euip);
533 		}
534 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
535 		if (attr.va_mode & S_ISGID)
536 			change_egid(imgp->newcred, attr.va_gid);
537 		/*
538 		 * Implement correct POSIX saved-id behavior.
539 		 *
540 		 * XXXMAC: Note that the current logic will save the
541 		 * uid and gid if a MAC domain transition occurs, even
542 		 * though maybe it shouldn't.
543 		 */
544 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
545 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
546 	} else {
547 		/*
548 		 * Implement correct POSIX saved-id behavior.
549 		 *
550 		 * XXX: It's not clear that the existing behavior is
551 		 * POSIX-compliant.  A number of sources indicate that the
552 		 * saved uid/gid should only be updated if the new ruid is
553 		 * not equal to the old ruid, or the new euid is not equal
554 		 * to the old euid and the new euid is not equal to the old
555 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
556 		 * Also, this code uses the new (replaced) euid and egid as
557 		 * the source, which may or may not be the right ones to use.
558 		 */
559 		if (oldcred->cr_svuid != oldcred->cr_uid ||
560 		    oldcred->cr_svgid != oldcred->cr_gid) {
561 			VOP_UNLOCK(imgp->vp, 0);
562 			imgp->newcred = crdup(oldcred);
563 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
564 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
565 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
566 		}
567 	}
568 	/* The new credentials are installed into the process later. */
569 
570 	/*
571 	 * Do the best to calculate the full path to the image file.
572 	 */
573 	if (args->fname != NULL && args->fname[0] == '/')
574 		imgp->execpath = args->fname;
575 	else {
576 		VOP_UNLOCK(imgp->vp, 0);
577 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
578 		    &imgp->freepath) != 0)
579 			imgp->execpath = args->fname;
580 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
581 	}
582 
583 	/*
584 	 *	If the current process has a special image activator it
585 	 *	wants to try first, call it.   For example, emulating shell
586 	 *	scripts differently.
587 	 */
588 	error = -1;
589 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
590 		error = img_first(imgp);
591 
592 	/*
593 	 *	Loop through the list of image activators, calling each one.
594 	 *	An activator returns -1 if there is no match, 0 on success,
595 	 *	and an error otherwise.
596 	 */
597 	for (i = 0; error == -1 && execsw[i]; ++i) {
598 		if (execsw[i]->ex_imgact == NULL ||
599 		    execsw[i]->ex_imgact == img_first) {
600 			continue;
601 		}
602 		error = (*execsw[i]->ex_imgact)(imgp);
603 	}
604 
605 	if (error) {
606 		if (error == -1)
607 			error = ENOEXEC;
608 		goto exec_fail_dealloc;
609 	}
610 
611 	/*
612 	 * Special interpreter operation, cleanup and loop up to try to
613 	 * activate the interpreter.
614 	 */
615 	if (imgp->interpreted) {
616 		exec_unmap_first_page(imgp);
617 		/*
618 		 * The text reference needs to be removed for scripts.
619 		 * There is a short period before we determine that
620 		 * something is a script where text reference is active.
621 		 * The vnode lock is held over this entire period
622 		 * so nothing should illegitimately be blocked.
623 		 */
624 		MPASS(imgp->textset);
625 		VOP_UNSET_TEXT_CHECKED(newtextvp);
626 		imgp->textset = false;
627 		/* free name buffer and old vnode */
628 		if (args->fname != NULL)
629 			NDFREE(&nd, NDF_ONLY_PNBUF);
630 #ifdef MAC
631 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
632 #endif
633 		if (imgp->opened) {
634 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
635 			imgp->opened = 0;
636 		}
637 		vput(newtextvp);
638 		vm_object_deallocate(imgp->object);
639 		imgp->object = NULL;
640 		imgp->credential_setid = false;
641 		if (imgp->newcred != NULL) {
642 			crfree(imgp->newcred);
643 			imgp->newcred = NULL;
644 		}
645 		imgp->execpath = NULL;
646 		free(imgp->freepath, M_TEMP);
647 		imgp->freepath = NULL;
648 		/* set new name to that of the interpreter */
649 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
650 		    UIO_SYSSPACE, imgp->interpreter_name, td);
651 		args->fname = imgp->interpreter_name;
652 		goto interpret;
653 	}
654 
655 	/*
656 	 * NB: We unlock the vnode here because it is believed that none
657 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
658 	 */
659 	VOP_UNLOCK(imgp->vp, 0);
660 
661 	if (disallow_high_osrel &&
662 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
663 		error = ENOEXEC;
664 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
665 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
666 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
667 		goto exec_fail_dealloc;
668 	}
669 
670 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
671 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
672 		sys_cap_enter(td, NULL);
673 
674 	/*
675 	 * Copy out strings (args and env) and initialize stack base
676 	 */
677 	if (p->p_sysent->sv_copyout_strings)
678 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
679 	else
680 		stack_base = exec_copyout_strings(imgp);
681 
682 	/*
683 	 * If custom stack fixup routine present for this process
684 	 * let it do the stack setup.
685 	 * Else stuff argument count as first item on stack
686 	 */
687 	if (p->p_sysent->sv_fixup != NULL)
688 		error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
689 	else
690 		error = suword(--stack_base, imgp->args->argc) == 0 ?
691 		    0 : EFAULT;
692 	if (error != 0) {
693 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
694 		goto exec_fail_dealloc;
695 	}
696 
697 	if (args->fdp != NULL) {
698 		/* Install a brand new file descriptor table. */
699 		fdinstall_remapped(td, args->fdp);
700 		args->fdp = NULL;
701 	} else {
702 		/*
703 		 * Keep on using the existing file descriptor table. For
704 		 * security and other reasons, the file descriptor table
705 		 * cannot be shared after an exec.
706 		 */
707 		fdunshare(td);
708 		/* close files on exec */
709 		fdcloseexec(td);
710 	}
711 
712 	/*
713 	 * Malloc things before we need locks.
714 	 */
715 	i = imgp->args->begin_envv - imgp->args->begin_argv;
716 	/* Cache arguments if they fit inside our allowance */
717 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
718 		newargs = pargs_alloc(i);
719 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
720 	}
721 
722 	/*
723 	 * For security and other reasons, signal handlers cannot
724 	 * be shared after an exec. The new process gets a copy of the old
725 	 * handlers. In execsigs(), the new process will have its signals
726 	 * reset.
727 	 */
728 	if (sigacts_shared(p->p_sigacts)) {
729 		oldsigacts = p->p_sigacts;
730 		newsigacts = sigacts_alloc();
731 		sigacts_copy(newsigacts, oldsigacts);
732 	}
733 
734 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
735 
736 	PROC_LOCK(p);
737 	if (oldsigacts)
738 		p->p_sigacts = newsigacts;
739 	/* Stop profiling */
740 	stopprofclock(p);
741 
742 	/* reset caught signals */
743 	execsigs(p);
744 
745 	/* name this process - nameiexec(p, ndp) */
746 	bzero(p->p_comm, sizeof(p->p_comm));
747 	if (args->fname)
748 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
749 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
750 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
751 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
752 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
753 #ifdef KTR
754 	sched_clear_tdname(td);
755 #endif
756 
757 	/*
758 	 * mark as execed, wakeup the process that vforked (if any) and tell
759 	 * it that it now has its own resources back
760 	 */
761 	p->p_flag |= P_EXEC;
762 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
763 		p->p_flag2 &= ~P2_NOTRACE;
764 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
765 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
766 	if (p->p_flag & P_PPWAIT) {
767 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
768 		cv_broadcast(&p->p_pwait);
769 		/* STOPs are no longer ignored, arrange for AST */
770 		signotify(td);
771 	}
772 
773 	/*
774 	 * Implement image setuid/setgid installation.
775 	 */
776 	if (imgp->credential_setid) {
777 		/*
778 		 * Turn off syscall tracing for set-id programs, except for
779 		 * root.  Record any set-id flags first to make sure that
780 		 * we do not regain any tracing during a possible block.
781 		 */
782 		setsugid(p);
783 
784 #ifdef KTRACE
785 		if (p->p_tracecred != NULL &&
786 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
787 			ktrprocexec(p, &tracecred, &tracevp);
788 #endif
789 		/*
790 		 * Close any file descriptors 0..2 that reference procfs,
791 		 * then make sure file descriptors 0..2 are in use.
792 		 *
793 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
794 		 * taking sleepable locks, so temporarily drop our locks.
795 		 */
796 		PROC_UNLOCK(p);
797 		VOP_UNLOCK(imgp->vp, 0);
798 		fdsetugidsafety(td);
799 		error = fdcheckstd(td);
800 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
801 		if (error != 0)
802 			goto exec_fail_dealloc;
803 		PROC_LOCK(p);
804 #ifdef MAC
805 		if (will_transition) {
806 			mac_vnode_execve_transition(oldcred, imgp->newcred,
807 			    imgp->vp, interpvplabel, imgp);
808 		}
809 #endif
810 	} else {
811 		if (oldcred->cr_uid == oldcred->cr_ruid &&
812 		    oldcred->cr_gid == oldcred->cr_rgid)
813 			p->p_flag &= ~P_SUGID;
814 	}
815 	/*
816 	 * Set the new credentials.
817 	 */
818 	if (imgp->newcred != NULL) {
819 		proc_set_cred(p, imgp->newcred);
820 		crfree(oldcred);
821 		oldcred = NULL;
822 	}
823 
824 	/*
825 	 * Store the vp for use in procfs.  This vnode was referenced by namei
826 	 * or fgetvp_exec.
827 	 */
828 	oldtextvp = p->p_textvp;
829 	p->p_textvp = newtextvp;
830 
831 #ifdef KDTRACE_HOOKS
832 	/*
833 	 * Tell the DTrace fasttrap provider about the exec if it
834 	 * has declared an interest.
835 	 */
836 	if (dtrace_fasttrap_exec)
837 		dtrace_fasttrap_exec(p);
838 #endif
839 
840 	/*
841 	 * Notify others that we exec'd, and clear the P_INEXEC flag
842 	 * as we're now a bona fide freshly-execed process.
843 	 */
844 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
845 	p->p_flag &= ~P_INEXEC;
846 
847 	/* clear "fork but no exec" flag, as we _are_ execing */
848 	p->p_acflag &= ~AFORK;
849 
850 	/*
851 	 * Free any previous argument cache and replace it with
852 	 * the new argument cache, if any.
853 	 */
854 	oldargs = p->p_args;
855 	p->p_args = newargs;
856 	newargs = NULL;
857 
858 	PROC_UNLOCK(p);
859 
860 #ifdef	HWPMC_HOOKS
861 	/*
862 	 * Check if system-wide sampling is in effect or if the
863 	 * current process is using PMCs.  If so, do exec() time
864 	 * processing.  This processing needs to happen AFTER the
865 	 * P_INEXEC flag is cleared.
866 	 */
867 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
868 		VOP_UNLOCK(imgp->vp, 0);
869 		pe.pm_credentialschanged = credential_changing;
870 		pe.pm_entryaddr = imgp->entry_addr;
871 
872 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
873 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
874 	}
875 #endif
876 
877 	/* Set values passed into the program in registers. */
878 	if (p->p_sysent->sv_setregs)
879 		(*p->p_sysent->sv_setregs)(td, imgp,
880 		    (u_long)(uintptr_t)stack_base);
881 	else
882 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
883 
884 	vfs_mark_atime(imgp->vp, td->td_ucred);
885 
886 	SDT_PROBE1(proc, , , exec__success, args->fname);
887 
888 exec_fail_dealloc:
889 	if (imgp->firstpage != NULL)
890 		exec_unmap_first_page(imgp);
891 
892 	if (imgp->vp != NULL) {
893 		if (args->fname)
894 			NDFREE(&nd, NDF_ONLY_PNBUF);
895 		if (imgp->opened)
896 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
897 		if (imgp->textset)
898 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
899 		if (error != 0)
900 			vput(imgp->vp);
901 		else
902 			VOP_UNLOCK(imgp->vp, 0);
903 	}
904 
905 	if (imgp->object != NULL)
906 		vm_object_deallocate(imgp->object);
907 
908 	free(imgp->freepath, M_TEMP);
909 
910 	if (error == 0) {
911 		if (p->p_ptevents & PTRACE_EXEC) {
912 			PROC_LOCK(p);
913 			if (p->p_ptevents & PTRACE_EXEC)
914 				td->td_dbgflags |= TDB_EXEC;
915 			PROC_UNLOCK(p);
916 		}
917 
918 		/*
919 		 * Stop the process here if its stop event mask has
920 		 * the S_EXEC bit set.
921 		 */
922 		STOPEVENT(p, S_EXEC, 0);
923 	} else {
924 exec_fail:
925 		/* we're done here, clear P_INEXEC */
926 		PROC_LOCK(p);
927 		p->p_flag &= ~P_INEXEC;
928 		PROC_UNLOCK(p);
929 
930 		SDT_PROBE1(proc, , , exec__failure, error);
931 	}
932 
933 	if (imgp->newcred != NULL && oldcred != NULL)
934 		crfree(imgp->newcred);
935 
936 #ifdef MAC
937 	mac_execve_exit(imgp);
938 	mac_execve_interpreter_exit(interpvplabel);
939 #endif
940 	exec_free_args(args);
941 
942 	/*
943 	 * Handle deferred decrement of ref counts.
944 	 */
945 	if (oldtextvp != NULL)
946 		vrele(oldtextvp);
947 #ifdef KTRACE
948 	if (tracevp != NULL)
949 		vrele(tracevp);
950 	if (tracecred != NULL)
951 		crfree(tracecred);
952 #endif
953 	pargs_drop(oldargs);
954 	pargs_drop(newargs);
955 	if (oldsigacts != NULL)
956 		sigacts_free(oldsigacts);
957 	if (euip != NULL)
958 		uifree(euip);
959 
960 	if (error && imgp->vmspace_destroyed) {
961 		/* sorry, no more process anymore. exit gracefully */
962 		exit1(td, 0, SIGABRT);
963 		/* NOT REACHED */
964 	}
965 
966 #ifdef KTRACE
967 	if (error == 0)
968 		ktrprocctor(p);
969 #endif
970 
971 	/*
972 	 * We don't want cpu_set_syscall_retval() to overwrite any of
973 	 * the register values put in place by exec_setregs().
974 	 * Implementations of cpu_set_syscall_retval() will leave
975 	 * registers unmodified when returning EJUSTRETURN.
976 	 */
977 	return (error == 0 ? EJUSTRETURN : error);
978 }
979 
980 int
exec_map_first_page(struct image_params * imgp)981 exec_map_first_page(struct image_params *imgp)
982 {
983 	int rv, i, after, initial_pagein;
984 	vm_page_t ma[VM_INITIAL_PAGEIN];
985 	vm_object_t object;
986 
987 	if (imgp->firstpage != NULL)
988 		exec_unmap_first_page(imgp);
989 
990 	object = imgp->vp->v_object;
991 	if (object == NULL)
992 		return (EACCES);
993 	VM_OBJECT_WLOCK(object);
994 #if VM_NRESERVLEVEL > 0
995 	vm_object_color(object, 0);
996 #endif
997 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
998 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
999 		vm_page_xbusy(ma[0]);
1000 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
1001 			vm_page_lock(ma[0]);
1002 			vm_page_free(ma[0]);
1003 			vm_page_unlock(ma[0]);
1004 			VM_OBJECT_WUNLOCK(object);
1005 			return (EIO);
1006 		}
1007 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
1008 		KASSERT(initial_pagein <= object->size,
1009 		    ("%s: initial_pagein %d object->size %ju",
1010 		    __func__, initial_pagein, (uintmax_t )object->size));
1011 		for (i = 1; i < initial_pagein; i++) {
1012 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1013 				if (ma[i]->valid)
1014 					break;
1015 				if (!vm_page_tryxbusy(ma[i]))
1016 					break;
1017 			} else {
1018 				ma[i] = vm_page_alloc(object, i,
1019 				    VM_ALLOC_NORMAL);
1020 				if (ma[i] == NULL)
1021 					break;
1022 			}
1023 		}
1024 		initial_pagein = i;
1025 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1026 		if (rv != VM_PAGER_OK) {
1027 			for (i = 0; i < initial_pagein; i++) {
1028 				vm_page_lock(ma[i]);
1029 				vm_page_free(ma[i]);
1030 				vm_page_unlock(ma[i]);
1031 			}
1032 			VM_OBJECT_WUNLOCK(object);
1033 			return (EIO);
1034 		}
1035 		vm_page_xunbusy(ma[0]);
1036 		for (i = 1; i < initial_pagein; i++)
1037 			vm_page_readahead_finish(ma[i]);
1038 	}
1039 	vm_page_lock(ma[0]);
1040 	vm_page_hold(ma[0]);
1041 	vm_page_activate(ma[0]);
1042 	vm_page_unlock(ma[0]);
1043 	VM_OBJECT_WUNLOCK(object);
1044 
1045 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1046 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1047 
1048 	return (0);
1049 }
1050 
1051 void
exec_unmap_first_page(struct image_params * imgp)1052 exec_unmap_first_page(struct image_params *imgp)
1053 {
1054 	vm_page_t m;
1055 
1056 	if (imgp->firstpage != NULL) {
1057 		m = sf_buf_page(imgp->firstpage);
1058 		sf_buf_free(imgp->firstpage);
1059 		imgp->firstpage = NULL;
1060 		vm_page_lock(m);
1061 		vm_page_unhold(m);
1062 		vm_page_unlock(m);
1063 	}
1064 }
1065 
1066 /*
1067  * Destroy old address space, and allocate a new stack.
1068  *	The new stack is only sgrowsiz large because it is grown
1069  *	automatically on a page fault.
1070  */
1071 int
exec_new_vmspace(struct image_params * imgp,struct sysentvec * sv)1072 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1073 {
1074 	int error;
1075 	struct proc *p = imgp->proc;
1076 	struct vmspace *vmspace = p->p_vmspace;
1077 	vm_object_t obj;
1078 	struct rlimit rlim_stack;
1079 	vm_offset_t sv_minuser, stack_addr;
1080 	vm_map_t map;
1081 	u_long ssiz;
1082 
1083 	imgp->vmspace_destroyed = 1;
1084 	imgp->sysent = sv;
1085 
1086 	/* May be called with Giant held */
1087 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1088 
1089 	/*
1090 	 * Blow away entire process VM, if address space not shared,
1091 	 * otherwise, create a new VM space so that other threads are
1092 	 * not disrupted
1093 	 */
1094 	map = &vmspace->vm_map;
1095 	if (map_at_zero)
1096 		sv_minuser = sv->sv_minuser;
1097 	else
1098 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1099 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1100 	    vm_map_max(map) == sv->sv_maxuser &&
1101 	    cpu_exec_vmspace_reuse(p, map)) {
1102 		shmexit(vmspace);
1103 		pmap_remove_pages(vmspace_pmap(vmspace));
1104 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1105 		/*
1106 		 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1107 		 * must be re-evaluated.
1108 		 */
1109 		vm_map_lock(map);
1110 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1111 		    MAP_ASLR_IGNSTART);
1112 		vm_map_unlock(map);
1113 	} else {
1114 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1115 		if (error)
1116 			return (error);
1117 		vmspace = p->p_vmspace;
1118 		map = &vmspace->vm_map;
1119 	}
1120 	map->flags |= imgp->map_flags;
1121 
1122 	/* Map a shared page */
1123 	obj = sv->sv_shared_page_obj;
1124 	if (obj != NULL) {
1125 		vm_object_reference(obj);
1126 		error = vm_map_fixed(map, obj, 0,
1127 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1128 		    VM_PROT_READ | VM_PROT_EXECUTE,
1129 		    VM_PROT_READ | VM_PROT_EXECUTE,
1130 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1131 		if (error != KERN_SUCCESS) {
1132 			vm_object_deallocate(obj);
1133 			return (vm_mmap_to_errno(error));
1134 		}
1135 	}
1136 
1137 	/* Allocate a new stack */
1138 	if (imgp->stack_sz != 0) {
1139 		ssiz = trunc_page(imgp->stack_sz);
1140 		PROC_LOCK(p);
1141 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1142 		PROC_UNLOCK(p);
1143 		if (ssiz > rlim_stack.rlim_max)
1144 			ssiz = rlim_stack.rlim_max;
1145 		if (ssiz > rlim_stack.rlim_cur) {
1146 			rlim_stack.rlim_cur = ssiz;
1147 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1148 		}
1149 	} else if (sv->sv_maxssiz != NULL) {
1150 		ssiz = *sv->sv_maxssiz;
1151 	} else {
1152 		ssiz = maxssiz;
1153 	}
1154 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1155 	if (ssiz < imgp->eff_stack_sz)
1156 		imgp->eff_stack_sz = ssiz;
1157 	stack_addr = sv->sv_usrstack - ssiz;
1158 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1159 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1160 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1161 	if (error != KERN_SUCCESS)
1162 		return (vm_mmap_to_errno(error));
1163 
1164 	/*
1165 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1166 	 * are still used to enforce the stack rlimit on the process stack.
1167 	 */
1168 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1169 	vmspace->vm_maxsaddr = (char *)stack_addr;
1170 
1171 	return (0);
1172 }
1173 
1174 /*
1175  * Copy out argument and environment strings from the old process address
1176  * space into the temporary string buffer.
1177  */
1178 int
exec_copyin_args(struct image_args * args,char * fname,enum uio_seg segflg,char ** argv,char ** envv)1179 exec_copyin_args(struct image_args *args, char *fname,
1180     enum uio_seg segflg, char **argv, char **envv)
1181 {
1182 	u_long argp, envp;
1183 	int error;
1184 	size_t length;
1185 
1186 	bzero(args, sizeof(*args));
1187 	if (argv == NULL)
1188 		return (EFAULT);
1189 
1190 	/*
1191 	 * Allocate demand-paged memory for the file name, argument, and
1192 	 * environment strings.
1193 	 */
1194 	error = exec_alloc_args(args);
1195 	if (error != 0)
1196 		return (error);
1197 
1198 	/*
1199 	 * Copy the file name.
1200 	 */
1201 	if (fname != NULL) {
1202 		args->fname = args->buf;
1203 		error = (segflg == UIO_SYSSPACE) ?
1204 		    copystr(fname, args->fname, PATH_MAX, &length) :
1205 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1206 		if (error != 0)
1207 			goto err_exit;
1208 	} else
1209 		length = 0;
1210 
1211 	args->begin_argv = args->buf + length;
1212 	args->endp = args->begin_argv;
1213 	args->stringspace = ARG_MAX;
1214 
1215 	/*
1216 	 * extract arguments first
1217 	 */
1218 	for (;;) {
1219 		error = fueword(argv++, &argp);
1220 		if (error == -1) {
1221 			error = EFAULT;
1222 			goto err_exit;
1223 		}
1224 		if (argp == 0)
1225 			break;
1226 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1227 		    args->stringspace, &length);
1228 		if (error != 0) {
1229 			if (error == ENAMETOOLONG)
1230 				error = E2BIG;
1231 			goto err_exit;
1232 		}
1233 		args->stringspace -= length;
1234 		args->endp += length;
1235 		args->argc++;
1236 	}
1237 
1238 	args->begin_envv = args->endp;
1239 
1240 	/*
1241 	 * extract environment strings
1242 	 */
1243 	if (envv) {
1244 		for (;;) {
1245 			error = fueword(envv++, &envp);
1246 			if (error == -1) {
1247 				error = EFAULT;
1248 				goto err_exit;
1249 			}
1250 			if (envp == 0)
1251 				break;
1252 			error = copyinstr((void *)(uintptr_t)envp,
1253 			    args->endp, args->stringspace, &length);
1254 			if (error != 0) {
1255 				if (error == ENAMETOOLONG)
1256 					error = E2BIG;
1257 				goto err_exit;
1258 			}
1259 			args->stringspace -= length;
1260 			args->endp += length;
1261 			args->envc++;
1262 		}
1263 	}
1264 
1265 	return (0);
1266 
1267 err_exit:
1268 	exec_free_args(args);
1269 	return (error);
1270 }
1271 
1272 int
exec_copyin_data_fds(struct thread * td,struct image_args * args,const void * data,size_t datalen,const int * fds,size_t fdslen)1273 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1274     const void *data, size_t datalen, const int *fds, size_t fdslen)
1275 {
1276 	struct filedesc *ofdp;
1277 	const char *p;
1278 	int *kfds;
1279 	int error;
1280 
1281 	memset(args, '\0', sizeof(*args));
1282 	ofdp = td->td_proc->p_fd;
1283 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1284 		return (E2BIG);
1285 	error = exec_alloc_args(args);
1286 	if (error != 0)
1287 		return (error);
1288 
1289 	args->begin_argv = args->buf;
1290 	args->stringspace = ARG_MAX;
1291 
1292 	if (datalen > 0) {
1293 		/*
1294 		 * Argument buffer has been provided. Copy it into the
1295 		 * kernel as a single string and add a terminating null
1296 		 * byte.
1297 		 */
1298 		error = copyin(data, args->begin_argv, datalen);
1299 		if (error != 0)
1300 			goto err_exit;
1301 		args->begin_argv[datalen] = '\0';
1302 		args->endp = args->begin_argv + datalen + 1;
1303 		args->stringspace -= datalen + 1;
1304 
1305 		/*
1306 		 * Traditional argument counting. Count the number of
1307 		 * null bytes.
1308 		 */
1309 		for (p = args->begin_argv; p < args->endp; ++p)
1310 			if (*p == '\0')
1311 				++args->argc;
1312 	} else {
1313 		/* No argument buffer provided. */
1314 		args->endp = args->begin_argv;
1315 	}
1316 	/* There are no environment variables. */
1317 	args->begin_envv = args->endp;
1318 
1319 	/* Create new file descriptor table. */
1320 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1321 	error = copyin(fds, kfds, fdslen * sizeof(int));
1322 	if (error != 0) {
1323 		free(kfds, M_TEMP);
1324 		goto err_exit;
1325 	}
1326 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1327 	free(kfds, M_TEMP);
1328 	if (error != 0)
1329 		goto err_exit;
1330 
1331 	return (0);
1332 err_exit:
1333 	exec_free_args(args);
1334 	return (error);
1335 }
1336 
1337 struct exec_args_kva {
1338 	vm_offset_t addr;
1339 	u_int gen;
1340 	SLIST_ENTRY(exec_args_kva) next;
1341 };
1342 
1343 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1344 
1345 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1346 static struct mtx exec_args_kva_mtx;
1347 static u_int exec_args_gen;
1348 
1349 static void
exec_prealloc_args_kva(void * arg __unused)1350 exec_prealloc_args_kva(void *arg __unused)
1351 {
1352 	struct exec_args_kva *argkva;
1353 	u_int i;
1354 
1355 	SLIST_INIT(&exec_args_kva_freelist);
1356 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1357 	for (i = 0; i < exec_map_entries; i++) {
1358 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1359 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1360 		argkva->gen = exec_args_gen;
1361 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1362 	}
1363 }
1364 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1365 
1366 static vm_offset_t
exec_alloc_args_kva(void ** cookie)1367 exec_alloc_args_kva(void **cookie)
1368 {
1369 	struct exec_args_kva *argkva;
1370 
1371 	argkva = (void *)atomic_readandclear_ptr(
1372 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1373 	if (argkva == NULL) {
1374 		mtx_lock(&exec_args_kva_mtx);
1375 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1376 			(void)mtx_sleep(&exec_args_kva_freelist,
1377 			    &exec_args_kva_mtx, 0, "execkva", 0);
1378 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1379 		mtx_unlock(&exec_args_kva_mtx);
1380 	}
1381 	*(struct exec_args_kva **)cookie = argkva;
1382 	return (argkva->addr);
1383 }
1384 
1385 static void
exec_release_args_kva(struct exec_args_kva * argkva,u_int gen)1386 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1387 {
1388 	vm_offset_t base;
1389 
1390 	base = argkva->addr;
1391 	if (argkva->gen != gen) {
1392 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1393 		    MADV_FREE);
1394 		argkva->gen = gen;
1395 	}
1396 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1397 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1398 		mtx_lock(&exec_args_kva_mtx);
1399 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1400 		wakeup_one(&exec_args_kva_freelist);
1401 		mtx_unlock(&exec_args_kva_mtx);
1402 	}
1403 }
1404 
1405 static void
exec_free_args_kva(void * cookie)1406 exec_free_args_kva(void *cookie)
1407 {
1408 
1409 	exec_release_args_kva(cookie, exec_args_gen);
1410 }
1411 
1412 static void
exec_args_kva_lowmem(void * arg __unused)1413 exec_args_kva_lowmem(void *arg __unused)
1414 {
1415 	SLIST_HEAD(, exec_args_kva) head;
1416 	struct exec_args_kva *argkva;
1417 	u_int gen;
1418 	int i;
1419 
1420 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1421 
1422 	/*
1423 	 * Force an madvise of each KVA range. Any currently allocated ranges
1424 	 * will have MADV_FREE applied once they are freed.
1425 	 */
1426 	SLIST_INIT(&head);
1427 	mtx_lock(&exec_args_kva_mtx);
1428 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1429 	mtx_unlock(&exec_args_kva_mtx);
1430 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1431 		SLIST_REMOVE_HEAD(&head, next);
1432 		exec_release_args_kva(argkva, gen);
1433 	}
1434 
1435 	CPU_FOREACH(i) {
1436 		argkva = (void *)atomic_readandclear_ptr(
1437 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1438 		if (argkva != NULL)
1439 			exec_release_args_kva(argkva, gen);
1440 	}
1441 }
1442 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1443     EVENTHANDLER_PRI_ANY);
1444 
1445 /*
1446  * Allocate temporary demand-paged, zero-filled memory for the file name,
1447  * argument, and environment strings.
1448  */
1449 int
exec_alloc_args(struct image_args * args)1450 exec_alloc_args(struct image_args *args)
1451 {
1452 
1453 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1454 	return (0);
1455 }
1456 
1457 void
exec_free_args(struct image_args * args)1458 exec_free_args(struct image_args *args)
1459 {
1460 
1461 	if (args->buf != NULL) {
1462 		exec_free_args_kva(args->bufkva);
1463 		args->buf = NULL;
1464 	}
1465 	if (args->fname_buf != NULL) {
1466 		free(args->fname_buf, M_TEMP);
1467 		args->fname_buf = NULL;
1468 	}
1469 	if (args->fdp != NULL)
1470 		fdescfree_remapped(args->fdp);
1471 }
1472 
1473 /*
1474  * Copy strings out to the new process address space, constructing new arg
1475  * and env vector tables. Return a pointer to the base so that it can be used
1476  * as the initial stack pointer.
1477  */
1478 register_t *
exec_copyout_strings(struct image_params * imgp)1479 exec_copyout_strings(struct image_params *imgp)
1480 {
1481 	int argc, envc;
1482 	char **vectp;
1483 	char *stringp;
1484 	uintptr_t destp;
1485 	register_t *stack_base;
1486 	struct ps_strings *arginfo;
1487 	struct proc *p;
1488 	size_t execpath_len;
1489 	int szsigcode, szps;
1490 	char canary[sizeof(long) * 8];
1491 
1492 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1493 	/*
1494 	 * Calculate string base and vector table pointers.
1495 	 * Also deal with signal trampoline code for this exec type.
1496 	 */
1497 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1498 		execpath_len = strlen(imgp->execpath) + 1;
1499 	else
1500 		execpath_len = 0;
1501 	p = imgp->proc;
1502 	szsigcode = 0;
1503 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1504 	if (p->p_sysent->sv_sigcode_base == 0) {
1505 		if (p->p_sysent->sv_szsigcode != NULL)
1506 			szsigcode = *(p->p_sysent->sv_szsigcode);
1507 	}
1508 	destp =	(uintptr_t)arginfo;
1509 
1510 	/*
1511 	 * install sigcode
1512 	 */
1513 	if (szsigcode != 0) {
1514 		destp -= szsigcode;
1515 		destp = rounddown2(destp, sizeof(void *));
1516 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1517 	}
1518 
1519 	/*
1520 	 * Copy the image path for the rtld.
1521 	 */
1522 	if (execpath_len != 0) {
1523 		destp -= execpath_len;
1524 		destp = rounddown2(destp, sizeof(void *));
1525 		imgp->execpathp = destp;
1526 		copyout(imgp->execpath, (void *)destp, execpath_len);
1527 	}
1528 
1529 	/*
1530 	 * Prepare the canary for SSP.
1531 	 */
1532 	arc4rand(canary, sizeof(canary), 0);
1533 	destp -= sizeof(canary);
1534 	imgp->canary = destp;
1535 	copyout(canary, (void *)destp, sizeof(canary));
1536 	imgp->canarylen = sizeof(canary);
1537 
1538 	/*
1539 	 * Prepare the pagesizes array.
1540 	 */
1541 	destp -= szps;
1542 	destp = rounddown2(destp, sizeof(void *));
1543 	imgp->pagesizes = destp;
1544 	copyout(pagesizes, (void *)destp, szps);
1545 	imgp->pagesizeslen = szps;
1546 
1547 	destp -= ARG_MAX - imgp->args->stringspace;
1548 	destp = rounddown2(destp, sizeof(void *));
1549 
1550 	vectp = (char **)destp;
1551 	if (imgp->sysent->sv_stackgap != NULL)
1552 		imgp->sysent->sv_stackgap(imgp, (u_long *)&vectp);
1553 
1554 	if (imgp->auxargs) {
1555 		/*
1556 		 * Allocate room on the stack for the ELF auxargs
1557 		 * array.  It has up to AT_COUNT entries.
1558 		 */
1559 		vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1560 		    sizeof(*vectp));
1561 	}
1562 
1563 	/*
1564 	 * Allocate room for the argv[] and env vectors including the
1565 	 * terminating NULL pointers.
1566 	 */
1567 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1568 
1569 	/*
1570 	 * vectp also becomes our initial stack base
1571 	 */
1572 	stack_base = (register_t *)vectp;
1573 
1574 	stringp = imgp->args->begin_argv;
1575 	argc = imgp->args->argc;
1576 	envc = imgp->args->envc;
1577 
1578 	/*
1579 	 * Copy out strings - arguments and environment.
1580 	 */
1581 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1582 
1583 	/*
1584 	 * Fill in "ps_strings" struct for ps, w, etc.
1585 	 */
1586 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1587 	suword32(&arginfo->ps_nargvstr, argc);
1588 
1589 	/*
1590 	 * Fill in argument portion of vector table.
1591 	 */
1592 	for (; argc > 0; --argc) {
1593 		suword(vectp++, (long)(intptr_t)destp);
1594 		while (*stringp++ != 0)
1595 			destp++;
1596 		destp++;
1597 	}
1598 
1599 	/* a null vector table pointer separates the argp's from the envp's */
1600 	suword(vectp++, 0);
1601 
1602 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1603 	suword32(&arginfo->ps_nenvstr, envc);
1604 
1605 	/*
1606 	 * Fill in environment portion of vector table.
1607 	 */
1608 	for (; envc > 0; --envc) {
1609 		suword(vectp++, (long)(intptr_t)destp);
1610 		while (*stringp++ != 0)
1611 			destp++;
1612 		destp++;
1613 	}
1614 
1615 	/* end of vector table is a null pointer */
1616 	suword(vectp, 0);
1617 
1618 	return (stack_base);
1619 }
1620 
1621 /*
1622  * Check permissions of file to execute.
1623  *	Called with imgp->vp locked.
1624  *	Return 0 for success or error code on failure.
1625  */
1626 int
exec_check_permissions(struct image_params * imgp)1627 exec_check_permissions(struct image_params *imgp)
1628 {
1629 	struct vnode *vp = imgp->vp;
1630 	struct vattr *attr = imgp->attr;
1631 	struct thread *td;
1632 	int error;
1633 
1634 	td = curthread;
1635 
1636 	/* Get file attributes */
1637 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1638 	if (error)
1639 		return (error);
1640 
1641 #ifdef MAC
1642 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1643 	if (error)
1644 		return (error);
1645 #endif
1646 
1647 	/*
1648 	 * 1) Check if file execution is disabled for the filesystem that
1649 	 *    this file resides on.
1650 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1651 	 *    privileged user will always succeed, and we don't want this
1652 	 *    to happen unless the file really is executable.
1653 	 * 3) Ensure that the file is a regular file.
1654 	 */
1655 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1656 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1657 	    (attr->va_type != VREG))
1658 		return (EACCES);
1659 
1660 	/*
1661 	 * Zero length files can't be exec'd
1662 	 */
1663 	if (attr->va_size == 0)
1664 		return (ENOEXEC);
1665 
1666 	/*
1667 	 *  Check for execute permission to file based on current credentials.
1668 	 */
1669 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1670 	if (error)
1671 		return (error);
1672 
1673 	/*
1674 	 * Check number of open-for-writes on the file and deny execution
1675 	 * if there are any.
1676 	 *
1677 	 * Add a text reference now so no one can write to the
1678 	 * executable while we're activating it.
1679 	 *
1680 	 * Remember if this was set before and unset it in case this is not
1681 	 * actually an executable image.
1682 	 */
1683 	error = VOP_SET_TEXT(vp);
1684 	if (error != 0)
1685 		return (error);
1686 	imgp->textset = true;
1687 
1688 	/*
1689 	 * Call filesystem specific open routine (which does nothing in the
1690 	 * general case).
1691 	 */
1692 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1693 	if (error == 0)
1694 		imgp->opened = 1;
1695 	return (error);
1696 }
1697 
1698 /*
1699  * Exec handler registration
1700  */
1701 int
exec_register(const struct execsw * execsw_arg)1702 exec_register(const struct execsw *execsw_arg)
1703 {
1704 	const struct execsw **es, **xs, **newexecsw;
1705 	u_int count = 2;	/* New slot and trailing NULL */
1706 
1707 	if (execsw)
1708 		for (es = execsw; *es; es++)
1709 			count++;
1710 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1711 	xs = newexecsw;
1712 	if (execsw)
1713 		for (es = execsw; *es; es++)
1714 			*xs++ = *es;
1715 	*xs++ = execsw_arg;
1716 	*xs = NULL;
1717 	if (execsw)
1718 		free(execsw, M_TEMP);
1719 	execsw = newexecsw;
1720 	return (0);
1721 }
1722 
1723 int
exec_unregister(const struct execsw * execsw_arg)1724 exec_unregister(const struct execsw *execsw_arg)
1725 {
1726 	const struct execsw **es, **xs, **newexecsw;
1727 	int count = 1;
1728 
1729 	if (execsw == NULL)
1730 		panic("unregister with no handlers left?\n");
1731 
1732 	for (es = execsw; *es; es++) {
1733 		if (*es == execsw_arg)
1734 			break;
1735 	}
1736 	if (*es == NULL)
1737 		return (ENOENT);
1738 	for (es = execsw; *es; es++)
1739 		if (*es != execsw_arg)
1740 			count++;
1741 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1742 	xs = newexecsw;
1743 	for (es = execsw; *es; es++)
1744 		if (*es != execsw_arg)
1745 			*xs++ = *es;
1746 	*xs = NULL;
1747 	if (execsw)
1748 		free(execsw, M_TEMP);
1749 	execsw = newexecsw;
1750 	return (0);
1751 }
1752