xref: /freebsd-13.1/sys/kern/kern_exec.c (revision 2a454b54)
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/asan.h>
41 #include <sys/capsicum.h>
42 #include <sys/eventhandler.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/filedesc.h>
46 #include <sys/imgact.h>
47 #include <sys/imgact_elf.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mman.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.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/timers.h>
72 #include <sys/umtx.h>
73 #include <sys/vnode.h>
74 #include <sys/wait.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78 
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/pmap.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_pager.h>
88 
89 #ifdef	HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92 
93 #include <machine/reg.h>
94 
95 #include <security/audit/audit.h>
96 #include <security/mac/mac_framework.h>
97 
98 #ifdef KDTRACE_HOOKS
99 #include <sys/dtrace_bsd.h>
100 dtrace_execexit_func_t	dtrace_fasttrap_exec;
101 #endif
102 
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
107 
108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
109 
110 int coredump_pack_fileinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
112     &coredump_pack_fileinfo, 0,
113     "Enable file path packing in 'procstat -f' coredump notes");
114 
115 int coredump_pack_vmmapinfo = 1;
116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
117     &coredump_pack_vmmapinfo, 0,
118     "Enable file path packing in 'procstat -v' coredump notes");
119 
120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
123 static int do_execve(struct thread *td, struct image_args *args,
124     struct mac *mac_p, struct vmspace *oldvmspace);
125 
126 /* XXX This should be vm_size_t. */
127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
128     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
129     "Location of process' ps_strings structure");
130 
131 /* XXX This should be vm_size_t. */
132 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
133     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
134     "Top of process stack");
135 
136 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
137     NULL, 0, sysctl_kern_stackprot, "I",
138     "Stack memory permissions");
139 
140 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
141 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
142     &ps_arg_cache_limit, 0,
143     "Process' command line characters cache limit");
144 
145 static int disallow_high_osrel;
146 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
147     &disallow_high_osrel, 0,
148     "Disallow execution of binaries built for higher version of the world");
149 
150 static int map_at_zero = 0;
151 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
152     "Permit processes to map an object at virtual address 0.");
153 
154 int core_dump_can_intr = 1;
155 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
156     &core_dump_can_intr, 0,
157     "Core dumping interruptible with SIGKILL");
158 
159 static int
sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)160 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
161 {
162 	struct proc *p;
163 	vm_offset_t ps_strings;
164 
165 	p = curproc;
166 #ifdef SCTL_MASK32
167 	if (req->flags & SCTL_MASK32) {
168 		unsigned int val;
169 		val = (unsigned int)PROC_PS_STRINGS(p);
170 		return (SYSCTL_OUT(req, &val, sizeof(val)));
171 	}
172 #endif
173 	ps_strings = PROC_PS_STRINGS(p);
174 	return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
175 }
176 
177 static int
sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)178 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
179 {
180 	struct proc *p;
181 	vm_offset_t val;
182 
183 	p = curproc;
184 #ifdef SCTL_MASK32
185 	if (req->flags & SCTL_MASK32) {
186 		unsigned int val32;
187 
188 		val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
189 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
190 	}
191 #endif
192 	val = round_page(p->p_vmspace->vm_stacktop);
193 	return (SYSCTL_OUT(req, &val, sizeof(val)));
194 }
195 
196 static int
sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)197 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
198 {
199 	struct proc *p;
200 
201 	p = curproc;
202 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
203 	    sizeof(p->p_sysent->sv_stackprot)));
204 }
205 
206 /*
207  * Each of the items is a pointer to a `const struct execsw', hence the
208  * double pointer here.
209  */
210 static const struct execsw **execsw;
211 
212 #ifndef _SYS_SYSPROTO_H_
213 struct execve_args {
214 	char    *fname;
215 	char    **argv;
216 	char    **envv;
217 };
218 #endif
219 
220 int
sys_execve(struct thread * td,struct execve_args * uap)221 sys_execve(struct thread *td, struct execve_args *uap)
222 {
223 	struct image_args args;
224 	struct vmspace *oldvmspace;
225 	int error;
226 
227 	error = pre_execve(td, &oldvmspace);
228 	if (error != 0)
229 		return (error);
230 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
231 	    uap->argv, uap->envv);
232 	if (error == 0)
233 		error = kern_execve(td, &args, NULL, oldvmspace);
234 	post_execve(td, error, oldvmspace);
235 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
236 	return (error);
237 }
238 
239 #ifndef _SYS_SYSPROTO_H_
240 struct fexecve_args {
241 	int	fd;
242 	char	**argv;
243 	char	**envv;
244 };
245 #endif
246 int
sys_fexecve(struct thread * td,struct fexecve_args * uap)247 sys_fexecve(struct thread *td, struct fexecve_args *uap)
248 {
249 	struct image_args args;
250 	struct vmspace *oldvmspace;
251 	int error;
252 
253 	error = pre_execve(td, &oldvmspace);
254 	if (error != 0)
255 		return (error);
256 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
257 	    uap->argv, uap->envv);
258 	if (error == 0) {
259 		args.fd = uap->fd;
260 		error = kern_execve(td, &args, NULL, oldvmspace);
261 	}
262 	post_execve(td, error, oldvmspace);
263 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
264 	return (error);
265 }
266 
267 #ifndef _SYS_SYSPROTO_H_
268 struct __mac_execve_args {
269 	char	*fname;
270 	char	**argv;
271 	char	**envv;
272 	struct mac	*mac_p;
273 };
274 #endif
275 
276 int
sys___mac_execve(struct thread * td,struct __mac_execve_args * uap)277 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
278 {
279 #ifdef MAC
280 	struct image_args args;
281 	struct vmspace *oldvmspace;
282 	int error;
283 
284 	error = pre_execve(td, &oldvmspace);
285 	if (error != 0)
286 		return (error);
287 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
288 	    uap->argv, uap->envv);
289 	if (error == 0)
290 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
291 	post_execve(td, error, oldvmspace);
292 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
293 	return (error);
294 #else
295 	return (ENOSYS);
296 #endif
297 }
298 
299 int
pre_execve(struct thread * td,struct vmspace ** oldvmspace)300 pre_execve(struct thread *td, struct vmspace **oldvmspace)
301 {
302 	struct proc *p;
303 	int error;
304 
305 	KASSERT(td == curthread, ("non-current thread %p", td));
306 	error = 0;
307 	p = td->td_proc;
308 	if ((p->p_flag & P_HADTHREADS) != 0) {
309 		PROC_LOCK(p);
310 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
311 			error = ERESTART;
312 		PROC_UNLOCK(p);
313 	}
314 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
315 	    ("nested execve"));
316 	*oldvmspace = p->p_vmspace;
317 	return (error);
318 }
319 
320 void
post_execve(struct thread * td,int error,struct vmspace * oldvmspace)321 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
322 {
323 	struct proc *p;
324 
325 	KASSERT(td == curthread, ("non-current thread %p", td));
326 	p = td->td_proc;
327 	if ((p->p_flag & P_HADTHREADS) != 0) {
328 		PROC_LOCK(p);
329 		/*
330 		 * If success, we upgrade to SINGLE_EXIT state to
331 		 * force other threads to suicide.
332 		 */
333 		if (error == EJUSTRETURN)
334 			thread_single(p, SINGLE_EXIT);
335 		else
336 			thread_single_end(p, SINGLE_BOUNDARY);
337 		PROC_UNLOCK(p);
338 	}
339 	exec_cleanup(td, oldvmspace);
340 }
341 
342 /*
343  * kern_execve() has the astonishing property of not always returning to
344  * the caller.  If sufficiently bad things happen during the call to
345  * do_execve(), it can end up calling exit1(); as a result, callers must
346  * avoid doing anything which they might need to undo (e.g., allocating
347  * memory).
348  */
349 int
kern_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)350 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
351     struct vmspace *oldvmspace)
352 {
353 
354 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
355 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
356 	    exec_args_get_begin_envv(args) - args->begin_argv);
357 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
358 	    args->endp - exec_args_get_begin_envv(args));
359 
360 	/* Must have at least one argument. */
361 	if (args->argc == 0) {
362 		exec_free_args(args);
363 		return (EINVAL);
364 	}
365 	return (do_execve(td, args, mac_p, oldvmspace));
366 }
367 
368 static void
execve_nosetid(struct image_params * imgp)369 execve_nosetid(struct image_params *imgp)
370 {
371 	imgp->credential_setid = false;
372 	if (imgp->newcred != NULL) {
373 		crfree(imgp->newcred);
374 		imgp->newcred = NULL;
375 	}
376 }
377 
378 /*
379  * In-kernel implementation of execve().  All arguments are assumed to be
380  * userspace pointers from the passed thread.
381  */
382 static int
do_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)383 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
384     struct vmspace *oldvmspace)
385 {
386 	struct proc *p = td->td_proc;
387 	struct nameidata nd;
388 	struct ucred *oldcred;
389 	struct uidinfo *euip = NULL;
390 	uintptr_t stack_base;
391 	struct image_params image_params, *imgp;
392 	struct vattr attr;
393 	int (*img_first)(struct image_params *);
394 	struct pargs *oldargs = NULL, *newargs = NULL;
395 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
396 #ifdef KTRACE
397 	struct ktr_io_params *kiop;
398 #endif
399 	struct vnode *oldtextvp, *newtextvp;
400 	struct vnode *oldtextdvp, *newtextdvp;
401 	char *oldbinname, *newbinname;
402 	bool credential_changing;
403 #ifdef MAC
404 	struct label *interpvplabel = NULL;
405 	bool will_transition;
406 #endif
407 #ifdef HWPMC_HOOKS
408 	struct pmckern_procexec pe;
409 #endif
410 	int error, i, orig_osrel;
411 	uint32_t orig_fctl0;
412 	size_t freepath_size;
413 	static const char fexecv_proc_title[] = "(fexecv)";
414 
415 	imgp = &image_params;
416 	oldtextvp = oldtextdvp = NULL;
417 	newtextvp = newtextdvp = NULL;
418 	newbinname = oldbinname = NULL;
419 #ifdef KTRACE
420 	kiop = NULL;
421 #endif
422 
423 	/*
424 	 * Lock the process and set the P_INEXEC flag to indicate that
425 	 * it should be left alone until we're done here.  This is
426 	 * necessary to avoid race conditions - e.g. in ptrace() -
427 	 * that might allow a local user to illicitly obtain elevated
428 	 * privileges.
429 	 */
430 	PROC_LOCK(p);
431 	KASSERT((p->p_flag & P_INEXEC) == 0,
432 	    ("%s(): process already has P_INEXEC flag", __func__));
433 	p->p_flag |= P_INEXEC;
434 	PROC_UNLOCK(p);
435 
436 	/*
437 	 * Initialize part of the common data
438 	 */
439 	bzero(imgp, sizeof(*imgp));
440 	imgp->proc = p;
441 	imgp->attr = &attr;
442 	imgp->args = args;
443 	oldcred = p->p_ucred;
444 	orig_osrel = p->p_osrel;
445 	orig_fctl0 = p->p_fctl0;
446 
447 #ifdef MAC
448 	error = mac_execve_enter(imgp, mac_p);
449 	if (error)
450 		goto exec_fail;
451 #endif
452 
453 	SDT_PROBE1(proc, , , exec, args->fname);
454 
455 interpret:
456 	if (args->fname != NULL) {
457 #ifdef CAPABILITY_MODE
458 		/*
459 		 * While capability mode can't reach this point via direct
460 		 * path arguments to execve(), we also don't allow
461 		 * interpreters to be used in capability mode (for now).
462 		 * Catch indirect lookups and return a permissions error.
463 		 */
464 		if (IN_CAPABILITY_MODE(td)) {
465 			error = ECAPMODE;
466 			goto exec_fail;
467 		}
468 #endif
469 
470 		/*
471 		 * Translate the file name. namei() returns a vnode
472 		 * pointer in ni_vp among other things.
473 		 */
474 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
475 		    SAVENAME | AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
476 		    args->fname, td);
477 
478 		error = namei(&nd);
479 		if (error)
480 			goto exec_fail;
481 
482 		newtextvp = nd.ni_vp;
483 		newtextdvp = nd.ni_dvp;
484 		nd.ni_dvp = NULL;
485 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
486 		    M_WAITOK);
487 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
488 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
489 		imgp->vp = newtextvp;
490 
491 		/*
492 		 * Do the best to calculate the full path to the image file.
493 		 */
494 		if (args->fname[0] == '/') {
495 			imgp->execpath = args->fname;
496 		} else {
497 			VOP_UNLOCK(imgp->vp);
498 			freepath_size = MAXPATHLEN;
499 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
500 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
501 			    &imgp->freepath, &freepath_size) != 0)
502 				imgp->execpath = args->fname;
503 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
504 		}
505 	} else {
506 		AUDIT_ARG_FD(args->fd);
507 
508 		/*
509 		 * If the descriptors was not opened with O_PATH, then
510 		 * we require that it was opened with O_EXEC or
511 		 * O_RDONLY.  In either case, exec_check_permissions()
512 		 * below checks _current_ file access mode regardless
513 		 * of the permissions additionally checked at the
514 		 * open(2).
515 		 */
516 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
517 		    &newtextvp);
518 		if (error != 0)
519 			goto exec_fail;
520 
521 		if (vn_fullpath(newtextvp, &imgp->execpath,
522 		    &imgp->freepath) != 0)
523 			imgp->execpath = args->fname;
524 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
525 		AUDIT_ARG_VNODE1(newtextvp);
526 		imgp->vp = newtextvp;
527 	}
528 
529 	/*
530 	 * Check file permissions.  Also 'opens' file and sets its vnode to
531 	 * text mode.
532 	 */
533 	error = exec_check_permissions(imgp);
534 	if (error)
535 		goto exec_fail_dealloc;
536 
537 	imgp->object = imgp->vp->v_object;
538 	if (imgp->object != NULL)
539 		vm_object_reference(imgp->object);
540 
541 	error = exec_map_first_page(imgp);
542 	if (error)
543 		goto exec_fail_dealloc;
544 
545 	imgp->proc->p_osrel = 0;
546 	imgp->proc->p_fctl0 = 0;
547 
548 	/*
549 	 * Implement image setuid/setgid.
550 	 *
551 	 * Determine new credentials before attempting image activators
552 	 * so that it can be used by process_exec handlers to determine
553 	 * credential/setid changes.
554 	 *
555 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
556 	 * the process is being traced.
557 	 *
558 	 * We disable setuid/setgid/etc in capability mode on the basis
559 	 * that most setugid applications are not written with that
560 	 * environment in mind, and will therefore almost certainly operate
561 	 * incorrectly. In principle there's no reason that setugid
562 	 * applications might not be useful in capability mode, so we may want
563 	 * to reconsider this conservative design choice in the future.
564 	 *
565 	 * XXXMAC: For the time being, use NOSUID to also prohibit
566 	 * transitions on the file system.
567 	 */
568 	credential_changing = false;
569 	credential_changing |= (attr.va_mode & S_ISUID) &&
570 	    oldcred->cr_uid != attr.va_uid;
571 	credential_changing |= (attr.va_mode & S_ISGID) &&
572 	    oldcred->cr_gid != attr.va_gid;
573 #ifdef MAC
574 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
575 	    interpvplabel, imgp) != 0;
576 	credential_changing |= will_transition;
577 #endif
578 
579 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
580 	if (credential_changing)
581 		imgp->proc->p_pdeathsig = 0;
582 
583 	if (credential_changing &&
584 #ifdef CAPABILITY_MODE
585 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
586 #endif
587 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
588 	    (p->p_flag & P_TRACED) == 0) {
589 		imgp->credential_setid = true;
590 		VOP_UNLOCK(imgp->vp);
591 		imgp->newcred = crdup(oldcred);
592 		if (attr.va_mode & S_ISUID) {
593 			euip = uifind(attr.va_uid);
594 			change_euid(imgp->newcred, euip);
595 		}
596 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
597 		if (attr.va_mode & S_ISGID)
598 			change_egid(imgp->newcred, attr.va_gid);
599 		/*
600 		 * Implement correct POSIX saved-id behavior.
601 		 *
602 		 * XXXMAC: Note that the current logic will save the
603 		 * uid and gid if a MAC domain transition occurs, even
604 		 * though maybe it shouldn't.
605 		 */
606 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
607 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
608 	} else {
609 		/*
610 		 * Implement correct POSIX saved-id behavior.
611 		 *
612 		 * XXX: It's not clear that the existing behavior is
613 		 * POSIX-compliant.  A number of sources indicate that the
614 		 * saved uid/gid should only be updated if the new ruid is
615 		 * not equal to the old ruid, or the new euid is not equal
616 		 * to the old euid and the new euid is not equal to the old
617 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
618 		 * Also, this code uses the new (replaced) euid and egid as
619 		 * the source, which may or may not be the right ones to use.
620 		 */
621 		if (oldcred->cr_svuid != oldcred->cr_uid ||
622 		    oldcred->cr_svgid != oldcred->cr_gid) {
623 			VOP_UNLOCK(imgp->vp);
624 			imgp->newcred = crdup(oldcred);
625 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
626 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
627 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
628 		}
629 	}
630 	/* The new credentials are installed into the process later. */
631 
632 	/*
633 	 *	If the current process has a special image activator it
634 	 *	wants to try first, call it.   For example, emulating shell
635 	 *	scripts differently.
636 	 */
637 	error = -1;
638 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
639 		error = img_first(imgp);
640 
641 	/*
642 	 *	Loop through the list of image activators, calling each one.
643 	 *	An activator returns -1 if there is no match, 0 on success,
644 	 *	and an error otherwise.
645 	 */
646 	for (i = 0; error == -1 && execsw[i]; ++i) {
647 		if (execsw[i]->ex_imgact == NULL ||
648 		    execsw[i]->ex_imgact == img_first) {
649 			continue;
650 		}
651 		error = (*execsw[i]->ex_imgact)(imgp);
652 	}
653 
654 	if (error) {
655 		if (error == -1)
656 			error = ENOEXEC;
657 		goto exec_fail_dealloc;
658 	}
659 
660 	/*
661 	 * Special interpreter operation, cleanup and loop up to try to
662 	 * activate the interpreter.
663 	 */
664 	if (imgp->interpreted) {
665 		exec_unmap_first_page(imgp);
666 		/*
667 		 * The text reference needs to be removed for scripts.
668 		 * There is a short period before we determine that
669 		 * something is a script where text reference is active.
670 		 * The vnode lock is held over this entire period
671 		 * so nothing should illegitimately be blocked.
672 		 */
673 		MPASS(imgp->textset);
674 		VOP_UNSET_TEXT_CHECKED(newtextvp);
675 		imgp->textset = false;
676 		/* free name buffer and old vnode */
677 #ifdef MAC
678 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
679 #endif
680 		if (imgp->opened) {
681 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
682 			imgp->opened = false;
683 		}
684 		vput(newtextvp);
685 		imgp->vp = newtextvp = NULL;
686 		if (args->fname != NULL) {
687 			if (newtextdvp != NULL) {
688 				vrele(newtextdvp);
689 				newtextdvp = NULL;
690 			}
691 			NDFREE(&nd, NDF_ONLY_PNBUF);
692 			free(newbinname, M_PARGS);
693 			newbinname = NULL;
694 		}
695 		vm_object_deallocate(imgp->object);
696 		imgp->object = NULL;
697 		execve_nosetid(imgp);
698 		imgp->execpath = NULL;
699 		free(imgp->freepath, M_TEMP);
700 		imgp->freepath = NULL;
701 		/* set new name to that of the interpreter */
702 		args->fname = imgp->interpreter_name;
703 		goto interpret;
704 	}
705 
706 	/*
707 	 * NB: We unlock the vnode here because it is believed that none
708 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
709 	 */
710 	VOP_UNLOCK(imgp->vp);
711 
712 	if (disallow_high_osrel &&
713 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
714 		error = ENOEXEC;
715 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
716 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
717 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
718 		goto exec_fail_dealloc;
719 	}
720 
721 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
722 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
723 		sys_cap_enter(td, NULL);
724 
725 	/*
726 	 * Copy out strings (args and env) and initialize stack base.
727 	 */
728 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
729 	if (error != 0) {
730 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
731 		goto exec_fail_dealloc;
732 	}
733 
734 	/*
735 	 * Stack setup.
736 	 */
737 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
738 	if (error != 0) {
739 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
740 		goto exec_fail_dealloc;
741 	}
742 
743 	if (args->fdp != NULL) {
744 		/* Install a brand new file descriptor table. */
745 		fdinstall_remapped(td, args->fdp);
746 		args->fdp = NULL;
747 	} else {
748 		/*
749 		 * Keep on using the existing file descriptor table. For
750 		 * security and other reasons, the file descriptor table
751 		 * cannot be shared after an exec.
752 		 */
753 		fdunshare(td);
754 		pdunshare(td);
755 		/* close files on exec */
756 		fdcloseexec(td);
757 	}
758 
759 	/*
760 	 * Malloc things before we need locks.
761 	 */
762 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
763 	/* Cache arguments if they fit inside our allowance */
764 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
765 		newargs = pargs_alloc(i);
766 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
767 	}
768 
769 	/*
770 	 * For security and other reasons, signal handlers cannot
771 	 * be shared after an exec. The new process gets a copy of the old
772 	 * handlers. In execsigs(), the new process will have its signals
773 	 * reset.
774 	 */
775 	if (sigacts_shared(p->p_sigacts)) {
776 		oldsigacts = p->p_sigacts;
777 		newsigacts = sigacts_alloc();
778 		sigacts_copy(newsigacts, oldsigacts);
779 	}
780 
781 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
782 
783 	PROC_LOCK(p);
784 	if (oldsigacts)
785 		p->p_sigacts = newsigacts;
786 	/* Stop profiling */
787 	stopprofclock(p);
788 
789 	/* reset caught signals */
790 	execsigs(p);
791 
792 	/* name this process - nameiexec(p, ndp) */
793 	bzero(p->p_comm, sizeof(p->p_comm));
794 	if (args->fname)
795 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
796 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
797 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
798 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
799 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
800 #ifdef KTR
801 	sched_clear_tdname(td);
802 #endif
803 
804 	/*
805 	 * mark as execed, wakeup the process that vforked (if any) and tell
806 	 * it that it now has its own resources back
807 	 */
808 	p->p_flag |= P_EXEC;
809 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
810 		p->p_flag2 &= ~P2_NOTRACE;
811 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
812 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
813 	if (p->p_flag & P_PPWAIT) {
814 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
815 		cv_broadcast(&p->p_pwait);
816 		/* STOPs are no longer ignored, arrange for AST */
817 		signotify(td);
818 	}
819 
820 	if ((imgp->sysent->sv_setid_allowed != NULL &&
821 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
822 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
823 		execve_nosetid(imgp);
824 
825 	/*
826 	 * Implement image setuid/setgid installation.
827 	 */
828 	if (imgp->credential_setid) {
829 		/*
830 		 * Turn off syscall tracing for set-id programs, except for
831 		 * root.  Record any set-id flags first to make sure that
832 		 * we do not regain any tracing during a possible block.
833 		 */
834 		setsugid(p);
835 #ifdef KTRACE
836 		kiop = ktrprocexec(p);
837 #endif
838 		/*
839 		 * Close any file descriptors 0..2 that reference procfs,
840 		 * then make sure file descriptors 0..2 are in use.
841 		 *
842 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
843 		 * taking sleepable locks, so temporarily drop our locks.
844 		 */
845 		PROC_UNLOCK(p);
846 		VOP_UNLOCK(imgp->vp);
847 		fdsetugidsafety(td);
848 		error = fdcheckstd(td);
849 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
850 		if (error != 0)
851 			goto exec_fail_dealloc;
852 		PROC_LOCK(p);
853 #ifdef MAC
854 		if (will_transition) {
855 			mac_vnode_execve_transition(oldcred, imgp->newcred,
856 			    imgp->vp, interpvplabel, imgp);
857 		}
858 #endif
859 	} else {
860 		if (oldcred->cr_uid == oldcred->cr_ruid &&
861 		    oldcred->cr_gid == oldcred->cr_rgid)
862 			p->p_flag &= ~P_SUGID;
863 	}
864 	/*
865 	 * Set the new credentials.
866 	 */
867 	if (imgp->newcred != NULL) {
868 		proc_set_cred(p, imgp->newcred);
869 		crfree(oldcred);
870 		oldcred = NULL;
871 	}
872 
873 	/*
874 	 * Store the vp for use in kern.proc.pathname.  This vnode was
875 	 * referenced by namei() or by fexecve variant of fname handling.
876 	 */
877 	oldtextvp = p->p_textvp;
878 	p->p_textvp = newtextvp;
879 	oldtextdvp = p->p_textdvp;
880 	p->p_textdvp = newtextdvp;
881 	newtextdvp = NULL;
882 	oldbinname = p->p_binname;
883 	p->p_binname = newbinname;
884 	newbinname = NULL;
885 
886 #ifdef KDTRACE_HOOKS
887 	/*
888 	 * Tell the DTrace fasttrap provider about the exec if it
889 	 * has declared an interest.
890 	 */
891 	if (dtrace_fasttrap_exec)
892 		dtrace_fasttrap_exec(p);
893 #endif
894 
895 	/*
896 	 * Notify others that we exec'd, and clear the P_INEXEC flag
897 	 * as we're now a bona fide freshly-execed process.
898 	 */
899 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
900 	p->p_flag &= ~P_INEXEC;
901 
902 	/* clear "fork but no exec" flag, as we _are_ execing */
903 	p->p_acflag &= ~AFORK;
904 
905 	/*
906 	 * Free any previous argument cache and replace it with
907 	 * the new argument cache, if any.
908 	 */
909 	oldargs = p->p_args;
910 	p->p_args = newargs;
911 	newargs = NULL;
912 
913 	PROC_UNLOCK(p);
914 
915 #ifdef	HWPMC_HOOKS
916 	/*
917 	 * Check if system-wide sampling is in effect or if the
918 	 * current process is using PMCs.  If so, do exec() time
919 	 * processing.  This processing needs to happen AFTER the
920 	 * P_INEXEC flag is cleared.
921 	 */
922 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
923 		VOP_UNLOCK(imgp->vp);
924 		pe.pm_credentialschanged = credential_changing;
925 		pe.pm_entryaddr = imgp->entry_addr;
926 
927 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
928 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
929 	}
930 #endif
931 
932 	/* Set values passed into the program in registers. */
933 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
934 
935 	VOP_MMAPPED(imgp->vp);
936 
937 	SDT_PROBE1(proc, , , exec__success, args->fname);
938 
939 exec_fail_dealloc:
940 	if (error != 0) {
941 		p->p_osrel = orig_osrel;
942 		p->p_fctl0 = orig_fctl0;
943 	}
944 
945 	if (imgp->firstpage != NULL)
946 		exec_unmap_first_page(imgp);
947 
948 	if (imgp->vp != NULL) {
949 		if (imgp->opened)
950 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
951 		if (imgp->textset)
952 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
953 		if (error != 0)
954 			vput(imgp->vp);
955 		else
956 			VOP_UNLOCK(imgp->vp);
957 		if (args->fname != NULL)
958 			NDFREE(&nd, NDF_ONLY_PNBUF);
959 		if (newtextdvp != NULL)
960 			vrele(newtextdvp);
961 		free(newbinname, M_PARGS);
962 	}
963 
964 	if (imgp->object != NULL)
965 		vm_object_deallocate(imgp->object);
966 
967 	free(imgp->freepath, M_TEMP);
968 
969 	if (error == 0) {
970 		if (p->p_ptevents & PTRACE_EXEC) {
971 			PROC_LOCK(p);
972 			if (p->p_ptevents & PTRACE_EXEC)
973 				td->td_dbgflags |= TDB_EXEC;
974 			PROC_UNLOCK(p);
975 		}
976 	} else {
977 exec_fail:
978 		/* we're done here, clear P_INEXEC */
979 		PROC_LOCK(p);
980 		p->p_flag &= ~P_INEXEC;
981 		PROC_UNLOCK(p);
982 
983 		SDT_PROBE1(proc, , , exec__failure, error);
984 	}
985 
986 	if (imgp->newcred != NULL && oldcred != NULL)
987 		crfree(imgp->newcred);
988 
989 #ifdef MAC
990 	mac_execve_exit(imgp);
991 	mac_execve_interpreter_exit(interpvplabel);
992 #endif
993 	exec_free_args(args);
994 
995 	/*
996 	 * Handle deferred decrement of ref counts.
997 	 */
998 	if (oldtextvp != NULL)
999 		vrele(oldtextvp);
1000 	if (oldtextdvp != NULL)
1001 		vrele(oldtextdvp);
1002 	free(oldbinname, M_PARGS);
1003 #ifdef KTRACE
1004 	ktr_io_params_free(kiop);
1005 #endif
1006 	pargs_drop(oldargs);
1007 	pargs_drop(newargs);
1008 	if (oldsigacts != NULL)
1009 		sigacts_free(oldsigacts);
1010 	if (euip != NULL)
1011 		uifree(euip);
1012 
1013 	if (error && imgp->vmspace_destroyed) {
1014 		/* sorry, no more process anymore. exit gracefully */
1015 		exec_cleanup(td, oldvmspace);
1016 		exit1(td, 0, SIGABRT);
1017 		/* NOT REACHED */
1018 	}
1019 
1020 #ifdef KTRACE
1021 	if (error == 0)
1022 		ktrprocctor(p);
1023 #endif
1024 
1025 	/*
1026 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1027 	 * the register values put in place by exec_setregs().
1028 	 * Implementations of cpu_set_syscall_retval() will leave
1029 	 * registers unmodified when returning EJUSTRETURN.
1030 	 */
1031 	return (error == 0 ? EJUSTRETURN : error);
1032 }
1033 
1034 void
exec_cleanup(struct thread * td,struct vmspace * oldvmspace)1035 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1036 {
1037 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1038 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1039 		    ("oldvmspace still used"));
1040 		vmspace_free(oldvmspace);
1041 		td->td_pflags &= ~TDP_EXECVMSPC;
1042 	}
1043 }
1044 
1045 int
exec_map_first_page(struct image_params * imgp)1046 exec_map_first_page(struct image_params *imgp)
1047 {
1048 	vm_object_t object;
1049 	vm_page_t m;
1050 	int error;
1051 
1052 	if (imgp->firstpage != NULL)
1053 		exec_unmap_first_page(imgp);
1054 
1055 	object = imgp->vp->v_object;
1056 	if (object == NULL)
1057 		return (EACCES);
1058 #if VM_NRESERVLEVEL > 0
1059 	if ((object->flags & OBJ_COLORED) == 0) {
1060 		VM_OBJECT_WLOCK(object);
1061 		vm_object_color(object, 0);
1062 		VM_OBJECT_WUNLOCK(object);
1063 	}
1064 #endif
1065 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1066 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1067 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1068 
1069 	if (error != VM_PAGER_OK)
1070 		return (EIO);
1071 	imgp->firstpage = sf_buf_alloc(m, 0);
1072 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1073 
1074 	return (0);
1075 }
1076 
1077 void
exec_unmap_first_page(struct image_params * imgp)1078 exec_unmap_first_page(struct image_params *imgp)
1079 {
1080 	vm_page_t m;
1081 
1082 	if (imgp->firstpage != NULL) {
1083 		m = sf_buf_page(imgp->firstpage);
1084 		sf_buf_free(imgp->firstpage);
1085 		imgp->firstpage = NULL;
1086 		vm_page_unwire(m, PQ_ACTIVE);
1087 	}
1088 }
1089 
1090 void
exec_onexec_old(struct thread * td)1091 exec_onexec_old(struct thread *td)
1092 {
1093 	sigfastblock_clear(td);
1094 	umtx_exec(td->td_proc);
1095 }
1096 
1097 /*
1098  * This is an optimization which removes the unmanaged shared page
1099  * mapping. In combination with pmap_remove_pages(), which cleans all
1100  * managed mappings in the process' vmspace pmap, no work will be left
1101  * for pmap_remove(min, max).
1102  */
1103 void
exec_free_abi_mappings(struct proc * p)1104 exec_free_abi_mappings(struct proc *p)
1105 {
1106 	struct vmspace *vmspace;
1107 	struct sysentvec *sv;
1108 
1109 	vmspace = p->p_vmspace;
1110 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1111 		return;
1112 
1113 	sv = p->p_sysent;
1114 	if (sv->sv_shared_page_obj == NULL)
1115 		return;
1116 
1117 	pmap_remove(vmspace_pmap(vmspace), sv->sv_shared_page_base,
1118 	    sv->sv_shared_page_base + sv->sv_shared_page_len);
1119 }
1120 
1121 /*
1122  * Run down the current address space and install a new one.  Map the shared
1123  * page.
1124  */
1125 int
exec_new_vmspace(struct image_params * imgp,struct sysentvec * sv)1126 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1127 {
1128 	int error;
1129 	struct proc *p = imgp->proc;
1130 	struct vmspace *vmspace = p->p_vmspace;
1131 	struct thread *td = curthread;
1132 	vm_object_t obj;
1133 	vm_offset_t sv_minuser;
1134 	vm_map_t map;
1135 
1136 	imgp->vmspace_destroyed = true;
1137 	imgp->sysent = sv;
1138 
1139 	if (p->p_sysent->sv_onexec_old != NULL)
1140 		p->p_sysent->sv_onexec_old(td);
1141 	itimers_exec(p);
1142 	if (sv->sv_onexec != NULL)
1143 		sv->sv_onexec(p, imgp);
1144 
1145 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1146 
1147 	/*
1148 	 * Blow away entire process VM, if address space not shared,
1149 	 * otherwise, create a new VM space so that other threads are
1150 	 * not disrupted
1151 	 */
1152 	map = &vmspace->vm_map;
1153 	if (map_at_zero)
1154 		sv_minuser = sv->sv_minuser;
1155 	else
1156 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1157 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1158 	    vm_map_min(map) == sv_minuser &&
1159 	    vm_map_max(map) == sv->sv_maxuser &&
1160 	    cpu_exec_vmspace_reuse(p, map)) {
1161 		exec_free_abi_mappings(p);
1162 		shmexit(vmspace);
1163 		pmap_remove_pages(vmspace_pmap(vmspace));
1164 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1165 		/*
1166 		 * An exec terminates mlockall(MCL_FUTURE).
1167 		 * ASLR and W^X states must be re-evaluated.
1168 		 */
1169 		vm_map_lock(map);
1170 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1171 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1172 		vm_map_unlock(map);
1173 	} else {
1174 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1175 		if (error)
1176 			return (error);
1177 		vmspace = p->p_vmspace;
1178 		map = &vmspace->vm_map;
1179 	}
1180 	map->flags |= imgp->map_flags;
1181 
1182 	/* Map a shared page */
1183 	obj = sv->sv_shared_page_obj;
1184 	if (obj != NULL) {
1185 		vm_object_reference(obj);
1186 		error = vm_map_fixed(map, obj, 0,
1187 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1188 		    VM_PROT_READ | VM_PROT_EXECUTE,
1189 		    VM_PROT_READ | VM_PROT_EXECUTE,
1190 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1191 		if (error != KERN_SUCCESS) {
1192 			vm_object_deallocate(obj);
1193 			return (vm_mmap_to_errno(error));
1194 		}
1195 	}
1196 
1197 	return (0);
1198 }
1199 
1200 /*
1201  * Compute the stack size limit and map the main process stack.
1202  */
1203 int
exec_map_stack(struct image_params * imgp)1204 exec_map_stack(struct image_params *imgp)
1205 {
1206 	struct rlimit rlim_stack;
1207 	struct sysentvec *sv;
1208 	struct proc *p;
1209 	vm_map_t map;
1210 	struct vmspace *vmspace;
1211 	vm_offset_t stack_addr, stack_top;
1212 	u_long ssiz;
1213 	int error, find_space, stack_off;
1214 	vm_prot_t stack_prot;
1215 
1216 	p = imgp->proc;
1217 	sv = p->p_sysent;
1218 
1219 	if (imgp->stack_sz != 0) {
1220 		ssiz = trunc_page(imgp->stack_sz);
1221 		PROC_LOCK(p);
1222 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1223 		PROC_UNLOCK(p);
1224 		if (ssiz > rlim_stack.rlim_max)
1225 			ssiz = rlim_stack.rlim_max;
1226 		if (ssiz > rlim_stack.rlim_cur) {
1227 			rlim_stack.rlim_cur = ssiz;
1228 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1229 		}
1230 	} else if (sv->sv_maxssiz != NULL) {
1231 		ssiz = *sv->sv_maxssiz;
1232 	} else {
1233 		ssiz = maxssiz;
1234 	}
1235 
1236 	vmspace = p->p_vmspace;
1237 	map = &vmspace->vm_map;
1238 
1239 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1240 	    imgp->stack_prot : sv->sv_stackprot;
1241 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1242 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1243 		    lim_max(curthread, RLIMIT_DATA));
1244 		find_space = VMFS_ANY_SPACE;
1245 	} else {
1246 		stack_addr = sv->sv_usrstack - ssiz;
1247 		find_space = VMFS_NO_SPACE;
1248 	}
1249 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1250 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1251 	    MAP_STACK_GROWS_DOWN);
1252 	if (error != KERN_SUCCESS) {
1253 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1254 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1255 		    stack_prot, error, vm_mmap_to_errno(error));
1256 		return (vm_mmap_to_errno(error));
1257 	}
1258 
1259 	stack_top = stack_addr + ssiz;
1260 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1261 		/* Randomize within the first page of the stack. */
1262 		arc4rand(&stack_off, sizeof(stack_off), 0);
1263 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1264 	}
1265 
1266 	/*
1267 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1268 	 * are still used to enforce the stack rlimit on the process stack.
1269 	 */
1270 	vmspace->vm_maxsaddr = (char *)stack_addr;
1271 	vmspace->vm_stacktop = stack_top;
1272 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1273 
1274 	return (0);
1275 }
1276 
1277 /*
1278  * Copy out argument and environment strings from the old process address
1279  * space into the temporary string buffer.
1280  */
1281 int
exec_copyin_args(struct image_args * args,const char * fname,enum uio_seg segflg,char ** argv,char ** envv)1282 exec_copyin_args(struct image_args *args, const char *fname,
1283     enum uio_seg segflg, char **argv, char **envv)
1284 {
1285 	u_long arg, env;
1286 	int error;
1287 
1288 	bzero(args, sizeof(*args));
1289 	if (argv == NULL)
1290 		return (EFAULT);
1291 
1292 	/*
1293 	 * Allocate demand-paged memory for the file name, argument, and
1294 	 * environment strings.
1295 	 */
1296 	error = exec_alloc_args(args);
1297 	if (error != 0)
1298 		return (error);
1299 
1300 	/*
1301 	 * Copy the file name.
1302 	 */
1303 	error = exec_args_add_fname(args, fname, segflg);
1304 	if (error != 0)
1305 		goto err_exit;
1306 
1307 	/*
1308 	 * extract arguments first
1309 	 */
1310 	for (;;) {
1311 		error = fueword(argv++, &arg);
1312 		if (error == -1) {
1313 			error = EFAULT;
1314 			goto err_exit;
1315 		}
1316 		if (arg == 0)
1317 			break;
1318 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1319 		    UIO_USERSPACE);
1320 		if (error != 0)
1321 			goto err_exit;
1322 	}
1323 
1324 	/*
1325 	 * extract environment strings
1326 	 */
1327 	if (envv) {
1328 		for (;;) {
1329 			error = fueword(envv++, &env);
1330 			if (error == -1) {
1331 				error = EFAULT;
1332 				goto err_exit;
1333 			}
1334 			if (env == 0)
1335 				break;
1336 			error = exec_args_add_env(args,
1337 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1338 			if (error != 0)
1339 				goto err_exit;
1340 		}
1341 	}
1342 
1343 	return (0);
1344 
1345 err_exit:
1346 	exec_free_args(args);
1347 	return (error);
1348 }
1349 
1350 int
exec_copyin_data_fds(struct thread * td,struct image_args * args,const void * data,size_t datalen,const int * fds,size_t fdslen)1351 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1352     const void *data, size_t datalen, const int *fds, size_t fdslen)
1353 {
1354 	struct filedesc *ofdp;
1355 	const char *p;
1356 	int *kfds;
1357 	int error;
1358 
1359 	memset(args, '\0', sizeof(*args));
1360 	ofdp = td->td_proc->p_fd;
1361 	if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1362 		return (E2BIG);
1363 	error = exec_alloc_args(args);
1364 	if (error != 0)
1365 		return (error);
1366 
1367 	args->begin_argv = args->buf;
1368 	args->stringspace = ARG_MAX;
1369 
1370 	if (datalen > 0) {
1371 		/*
1372 		 * Argument buffer has been provided. Copy it into the
1373 		 * kernel as a single string and add a terminating null
1374 		 * byte.
1375 		 */
1376 		error = copyin(data, args->begin_argv, datalen);
1377 		if (error != 0)
1378 			goto err_exit;
1379 		args->begin_argv[datalen] = '\0';
1380 		args->endp = args->begin_argv + datalen + 1;
1381 		args->stringspace -= datalen + 1;
1382 
1383 		/*
1384 		 * Traditional argument counting. Count the number of
1385 		 * null bytes.
1386 		 */
1387 		for (p = args->begin_argv; p < args->endp; ++p)
1388 			if (*p == '\0')
1389 				++args->argc;
1390 	} else {
1391 		/* No argument buffer provided. */
1392 		args->endp = args->begin_argv;
1393 	}
1394 
1395 	/* Create new file descriptor table. */
1396 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1397 	error = copyin(fds, kfds, fdslen * sizeof(int));
1398 	if (error != 0) {
1399 		free(kfds, M_TEMP);
1400 		goto err_exit;
1401 	}
1402 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1403 	free(kfds, M_TEMP);
1404 	if (error != 0)
1405 		goto err_exit;
1406 
1407 	return (0);
1408 err_exit:
1409 	exec_free_args(args);
1410 	return (error);
1411 }
1412 
1413 struct exec_args_kva {
1414 	vm_offset_t addr;
1415 	u_int gen;
1416 	SLIST_ENTRY(exec_args_kva) next;
1417 };
1418 
1419 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1420 
1421 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1422 static struct mtx exec_args_kva_mtx;
1423 static u_int exec_args_gen;
1424 
1425 static void
exec_prealloc_args_kva(void * arg __unused)1426 exec_prealloc_args_kva(void *arg __unused)
1427 {
1428 	struct exec_args_kva *argkva;
1429 	u_int i;
1430 
1431 	SLIST_INIT(&exec_args_kva_freelist);
1432 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1433 	for (i = 0; i < exec_map_entries; i++) {
1434 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1435 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1436 		argkva->gen = exec_args_gen;
1437 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1438 	}
1439 }
1440 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1441 
1442 static vm_offset_t
exec_alloc_args_kva(void ** cookie)1443 exec_alloc_args_kva(void **cookie)
1444 {
1445 	struct exec_args_kva *argkva;
1446 
1447 	argkva = (void *)atomic_readandclear_ptr(
1448 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1449 	if (argkva == NULL) {
1450 		mtx_lock(&exec_args_kva_mtx);
1451 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1452 			(void)mtx_sleep(&exec_args_kva_freelist,
1453 			    &exec_args_kva_mtx, 0, "execkva", 0);
1454 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1455 		mtx_unlock(&exec_args_kva_mtx);
1456 	}
1457 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1458 	    exec_map_entry_size, 0);
1459 	*(struct exec_args_kva **)cookie = argkva;
1460 	return (argkva->addr);
1461 }
1462 
1463 static void
exec_release_args_kva(struct exec_args_kva * argkva,u_int gen)1464 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1465 {
1466 	vm_offset_t base;
1467 
1468 	base = argkva->addr;
1469 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1470 	    KASAN_EXEC_ARGS_FREED);
1471 	if (argkva->gen != gen) {
1472 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1473 		    MADV_FREE);
1474 		argkva->gen = gen;
1475 	}
1476 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1477 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1478 		mtx_lock(&exec_args_kva_mtx);
1479 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1480 		wakeup_one(&exec_args_kva_freelist);
1481 		mtx_unlock(&exec_args_kva_mtx);
1482 	}
1483 }
1484 
1485 static void
exec_free_args_kva(void * cookie)1486 exec_free_args_kva(void *cookie)
1487 {
1488 
1489 	exec_release_args_kva(cookie, exec_args_gen);
1490 }
1491 
1492 static void
exec_args_kva_lowmem(void * arg __unused)1493 exec_args_kva_lowmem(void *arg __unused)
1494 {
1495 	SLIST_HEAD(, exec_args_kva) head;
1496 	struct exec_args_kva *argkva;
1497 	u_int gen;
1498 	int i;
1499 
1500 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1501 
1502 	/*
1503 	 * Force an madvise of each KVA range. Any currently allocated ranges
1504 	 * will have MADV_FREE applied once they are freed.
1505 	 */
1506 	SLIST_INIT(&head);
1507 	mtx_lock(&exec_args_kva_mtx);
1508 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1509 	mtx_unlock(&exec_args_kva_mtx);
1510 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1511 		SLIST_REMOVE_HEAD(&head, next);
1512 		exec_release_args_kva(argkva, gen);
1513 	}
1514 
1515 	CPU_FOREACH(i) {
1516 		argkva = (void *)atomic_readandclear_ptr(
1517 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1518 		if (argkva != NULL)
1519 			exec_release_args_kva(argkva, gen);
1520 	}
1521 }
1522 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1523     EVENTHANDLER_PRI_ANY);
1524 
1525 /*
1526  * Allocate temporary demand-paged, zero-filled memory for the file name,
1527  * argument, and environment strings.
1528  */
1529 int
exec_alloc_args(struct image_args * args)1530 exec_alloc_args(struct image_args *args)
1531 {
1532 
1533 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1534 	return (0);
1535 }
1536 
1537 void
exec_free_args(struct image_args * args)1538 exec_free_args(struct image_args *args)
1539 {
1540 
1541 	if (args->buf != NULL) {
1542 		exec_free_args_kva(args->bufkva);
1543 		args->buf = NULL;
1544 	}
1545 	if (args->fname_buf != NULL) {
1546 		free(args->fname_buf, M_TEMP);
1547 		args->fname_buf = NULL;
1548 	}
1549 	if (args->fdp != NULL)
1550 		fdescfree_remapped(args->fdp);
1551 }
1552 
1553 /*
1554  * A set to functions to fill struct image args.
1555  *
1556  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1557  * fname) before the other functions.  All exec_args_add_arg() calls must
1558  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1559  * may be called any time after exec_args_add_fname().
1560  *
1561  * exec_args_add_fname() - install path to be executed
1562  * exec_args_add_arg() - append an argument string
1563  * exec_args_add_env() - append an env string
1564  * exec_args_adjust_args() - adjust location of the argument list to
1565  *                           allow new arguments to be prepended
1566  */
1567 int
exec_args_add_fname(struct image_args * args,const char * fname,enum uio_seg segflg)1568 exec_args_add_fname(struct image_args *args, const char *fname,
1569     enum uio_seg segflg)
1570 {
1571 	int error;
1572 	size_t length;
1573 
1574 	KASSERT(args->fname == NULL, ("fname already appended"));
1575 	KASSERT(args->endp == NULL, ("already appending to args"));
1576 
1577 	if (fname != NULL) {
1578 		args->fname = args->buf;
1579 		error = segflg == UIO_SYSSPACE ?
1580 		    copystr(fname, args->fname, PATH_MAX, &length) :
1581 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1582 		if (error != 0)
1583 			return (error == ENAMETOOLONG ? E2BIG : error);
1584 	} else
1585 		length = 0;
1586 
1587 	/* Set up for _arg_*()/_env_*() */
1588 	args->endp = args->buf + length;
1589 	/* begin_argv must be set and kept updated */
1590 	args->begin_argv = args->endp;
1591 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1592 	    ("too little space remaining for arguments %zu < %zu",
1593 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1594 	args->stringspace = ARG_MAX;
1595 
1596 	return (0);
1597 }
1598 
1599 static int
exec_args_add_str(struct image_args * args,const char * str,enum uio_seg segflg,int * countp)1600 exec_args_add_str(struct image_args *args, const char *str,
1601     enum uio_seg segflg, int *countp)
1602 {
1603 	int error;
1604 	size_t length;
1605 
1606 	KASSERT(args->endp != NULL, ("endp not initialized"));
1607 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1608 
1609 	error = (segflg == UIO_SYSSPACE) ?
1610 	    copystr(str, args->endp, args->stringspace, &length) :
1611 	    copyinstr(str, args->endp, args->stringspace, &length);
1612 	if (error != 0)
1613 		return (error == ENAMETOOLONG ? E2BIG : error);
1614 	args->stringspace -= length;
1615 	args->endp += length;
1616 	(*countp)++;
1617 
1618 	return (0);
1619 }
1620 
1621 int
exec_args_add_arg(struct image_args * args,const char * argp,enum uio_seg segflg)1622 exec_args_add_arg(struct image_args *args, const char *argp,
1623     enum uio_seg segflg)
1624 {
1625 
1626 	KASSERT(args->envc == 0, ("appending args after env"));
1627 
1628 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1629 }
1630 
1631 int
exec_args_add_env(struct image_args * args,const char * envp,enum uio_seg segflg)1632 exec_args_add_env(struct image_args *args, const char *envp,
1633     enum uio_seg segflg)
1634 {
1635 
1636 	if (args->envc == 0)
1637 		args->begin_envv = args->endp;
1638 
1639 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1640 }
1641 
1642 int
exec_args_adjust_args(struct image_args * args,size_t consume,ssize_t extend)1643 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1644 {
1645 	ssize_t offset;
1646 
1647 	KASSERT(args->endp != NULL, ("endp not initialized"));
1648 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1649 
1650 	offset = extend - consume;
1651 	if (args->stringspace < offset)
1652 		return (E2BIG);
1653 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1654 	    args->endp - args->begin_argv + consume);
1655 	if (args->envc > 0)
1656 		args->begin_envv += offset;
1657 	args->endp += offset;
1658 	args->stringspace -= offset;
1659 	return (0);
1660 }
1661 
1662 char *
exec_args_get_begin_envv(struct image_args * args)1663 exec_args_get_begin_envv(struct image_args *args)
1664 {
1665 
1666 	KASSERT(args->endp != NULL, ("endp not initialized"));
1667 
1668 	if (args->envc > 0)
1669 		return (args->begin_envv);
1670 	return (args->endp);
1671 }
1672 
1673 /*
1674  * Copy strings out to the new process address space, constructing new arg
1675  * and env vector tables. Return a pointer to the base so that it can be used
1676  * as the initial stack pointer.
1677  */
1678 int
exec_copyout_strings(struct image_params * imgp,uintptr_t * stack_base)1679 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1680 {
1681 	int argc, envc;
1682 	char **vectp;
1683 	char *stringp;
1684 	uintptr_t destp, ustringp;
1685 	struct ps_strings *arginfo;
1686 	struct proc *p;
1687 	struct sysentvec *sysent;
1688 	size_t execpath_len;
1689 	int error, szsigcode;
1690 	char canary[sizeof(long) * 8];
1691 
1692 	p = imgp->proc;
1693 	sysent = p->p_sysent;
1694 
1695 	destp =	PROC_PS_STRINGS(p);
1696 	arginfo = imgp->ps_strings = (void *)destp;
1697 
1698 	/*
1699 	 * Install sigcode.
1700 	 */
1701 	if (sysent->sv_sigcode_base == 0 && sysent->sv_szsigcode != NULL) {
1702 		szsigcode = *(sysent->sv_szsigcode);
1703 		destp -= szsigcode;
1704 		destp = rounddown2(destp, sizeof(void *));
1705 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1706 		if (error != 0)
1707 			return (error);
1708 	}
1709 
1710 	/*
1711 	 * Copy the image path for the rtld.
1712 	 */
1713 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1714 		execpath_len = strlen(imgp->execpath) + 1;
1715 		destp -= execpath_len;
1716 		destp = rounddown2(destp, sizeof(void *));
1717 		imgp->execpathp = (void *)destp;
1718 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1719 		if (error != 0)
1720 			return (error);
1721 	}
1722 
1723 	/*
1724 	 * Prepare the canary for SSP.
1725 	 */
1726 	arc4rand(canary, sizeof(canary), 0);
1727 	destp -= sizeof(canary);
1728 	imgp->canary = (void *)destp;
1729 	error = copyout(canary, imgp->canary, sizeof(canary));
1730 	if (error != 0)
1731 		return (error);
1732 	imgp->canarylen = sizeof(canary);
1733 
1734 	/*
1735 	 * Prepare the pagesizes array.
1736 	 */
1737 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1738 	destp -= imgp->pagesizeslen;
1739 	destp = rounddown2(destp, sizeof(void *));
1740 	imgp->pagesizes = (void *)destp;
1741 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1742 	if (error != 0)
1743 		return (error);
1744 
1745 	/*
1746 	 * Allocate room for the argument and environment strings.
1747 	 */
1748 	destp -= ARG_MAX - imgp->args->stringspace;
1749 	destp = rounddown2(destp, sizeof(void *));
1750 	ustringp = destp;
1751 
1752 	if (imgp->auxargs) {
1753 		/*
1754 		 * Allocate room on the stack for the ELF auxargs
1755 		 * array.  It has up to AT_COUNT entries.
1756 		 */
1757 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1758 		destp = rounddown2(destp, sizeof(void *));
1759 	}
1760 
1761 	vectp = (char **)destp;
1762 
1763 	/*
1764 	 * Allocate room for the argv[] and env vectors including the
1765 	 * terminating NULL pointers.
1766 	 */
1767 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1768 
1769 	/*
1770 	 * vectp also becomes our initial stack base
1771 	 */
1772 	*stack_base = (uintptr_t)vectp;
1773 
1774 	stringp = imgp->args->begin_argv;
1775 	argc = imgp->args->argc;
1776 	envc = imgp->args->envc;
1777 
1778 	/*
1779 	 * Copy out strings - arguments and environment.
1780 	 */
1781 	error = copyout(stringp, (void *)ustringp,
1782 	    ARG_MAX - imgp->args->stringspace);
1783 	if (error != 0)
1784 		return (error);
1785 
1786 	/*
1787 	 * Fill in "ps_strings" struct for ps, w, etc.
1788 	 */
1789 	imgp->argv = vectp;
1790 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1791 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1792 		return (EFAULT);
1793 
1794 	/*
1795 	 * Fill in argument portion of vector table.
1796 	 */
1797 	for (; argc > 0; --argc) {
1798 		if (suword(vectp++, ustringp) != 0)
1799 			return (EFAULT);
1800 		while (*stringp++ != 0)
1801 			ustringp++;
1802 		ustringp++;
1803 	}
1804 
1805 	/* a null vector table pointer separates the argp's from the envp's */
1806 	if (suword(vectp++, 0) != 0)
1807 		return (EFAULT);
1808 
1809 	imgp->envv = vectp;
1810 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1811 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1812 		return (EFAULT);
1813 
1814 	/*
1815 	 * Fill in environment portion of vector table.
1816 	 */
1817 	for (; envc > 0; --envc) {
1818 		if (suword(vectp++, ustringp) != 0)
1819 			return (EFAULT);
1820 		while (*stringp++ != 0)
1821 			ustringp++;
1822 		ustringp++;
1823 	}
1824 
1825 	/* end of vector table is a null pointer */
1826 	if (suword(vectp, 0) != 0)
1827 		return (EFAULT);
1828 
1829 	if (imgp->auxargs) {
1830 		vectp++;
1831 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1832 		    (uintptr_t)vectp);
1833 		if (error != 0)
1834 			return (error);
1835 	}
1836 
1837 	return (0);
1838 }
1839 
1840 /*
1841  * Check permissions of file to execute.
1842  *	Called with imgp->vp locked.
1843  *	Return 0 for success or error code on failure.
1844  */
1845 int
exec_check_permissions(struct image_params * imgp)1846 exec_check_permissions(struct image_params *imgp)
1847 {
1848 	struct vnode *vp = imgp->vp;
1849 	struct vattr *attr = imgp->attr;
1850 	struct thread *td;
1851 	int error;
1852 
1853 	td = curthread;
1854 
1855 	/* Get file attributes */
1856 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1857 	if (error)
1858 		return (error);
1859 
1860 #ifdef MAC
1861 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1862 	if (error)
1863 		return (error);
1864 #endif
1865 
1866 	/*
1867 	 * 1) Check if file execution is disabled for the filesystem that
1868 	 *    this file resides on.
1869 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1870 	 *    privileged user will always succeed, and we don't want this
1871 	 *    to happen unless the file really is executable.
1872 	 * 3) Ensure that the file is a regular file.
1873 	 */
1874 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1875 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1876 	    (attr->va_type != VREG))
1877 		return (EACCES);
1878 
1879 	/*
1880 	 * Zero length files can't be exec'd
1881 	 */
1882 	if (attr->va_size == 0)
1883 		return (ENOEXEC);
1884 
1885 	/*
1886 	 *  Check for execute permission to file based on current credentials.
1887 	 */
1888 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1889 	if (error)
1890 		return (error);
1891 
1892 	/*
1893 	 * Check number of open-for-writes on the file and deny execution
1894 	 * if there are any.
1895 	 *
1896 	 * Add a text reference now so no one can write to the
1897 	 * executable while we're activating it.
1898 	 *
1899 	 * Remember if this was set before and unset it in case this is not
1900 	 * actually an executable image.
1901 	 */
1902 	error = VOP_SET_TEXT(vp);
1903 	if (error != 0)
1904 		return (error);
1905 	imgp->textset = true;
1906 
1907 	/*
1908 	 * Call filesystem specific open routine (which does nothing in the
1909 	 * general case).
1910 	 */
1911 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1912 	if (error == 0)
1913 		imgp->opened = true;
1914 	return (error);
1915 }
1916 
1917 /*
1918  * Exec handler registration
1919  */
1920 int
exec_register(const struct execsw * execsw_arg)1921 exec_register(const struct execsw *execsw_arg)
1922 {
1923 	const struct execsw **es, **xs, **newexecsw;
1924 	u_int count = 2;	/* New slot and trailing NULL */
1925 
1926 	if (execsw)
1927 		for (es = execsw; *es; es++)
1928 			count++;
1929 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1930 	xs = newexecsw;
1931 	if (execsw)
1932 		for (es = execsw; *es; es++)
1933 			*xs++ = *es;
1934 	*xs++ = execsw_arg;
1935 	*xs = NULL;
1936 	if (execsw)
1937 		free(execsw, M_TEMP);
1938 	execsw = newexecsw;
1939 	return (0);
1940 }
1941 
1942 int
exec_unregister(const struct execsw * execsw_arg)1943 exec_unregister(const struct execsw *execsw_arg)
1944 {
1945 	const struct execsw **es, **xs, **newexecsw;
1946 	int count = 1;
1947 
1948 	if (execsw == NULL)
1949 		panic("unregister with no handlers left?\n");
1950 
1951 	for (es = execsw; *es; es++) {
1952 		if (*es == execsw_arg)
1953 			break;
1954 	}
1955 	if (*es == NULL)
1956 		return (ENOENT);
1957 	for (es = execsw; *es; es++)
1958 		if (*es != execsw_arg)
1959 			count++;
1960 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1961 	xs = newexecsw;
1962 	for (es = execsw; *es; es++)
1963 		if (*es != execsw_arg)
1964 			*xs++ = *es;
1965 	*xs = NULL;
1966 	if (execsw)
1967 		free(execsw, M_TEMP);
1968 	execsw = newexecsw;
1969 	return (0);
1970 }
1971