xref: /freebsd/sys/kern/kern_exec.c (revision d302c0539307d1b7ae9a0206d83a1fb9ccc44900)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
33 #include "opt_vm.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/acct.h>
38 #include <sys/asan.h>
39 #include <sys/capsicum.h>
40 #include <sys/compressor.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/priv.h>
55 #include <sys/proc.h>
56 #include <sys/ptrace.h>
57 #include <sys/reg.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/umtxvar.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 <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, struct vmspace *oldvmspace);
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     "Location of process' ps_strings structure");
128 
129 /* XXX This should be vm_size_t. */
130 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
131     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
132     "Top of process stack");
133 
134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
135     NULL, 0, sysctl_kern_stackprot, "I",
136     "Stack memory permissions");
137 
138 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
139 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
140     &ps_arg_cache_limit, 0,
141     "Process' command line characters cache limit");
142 
143 static int disallow_high_osrel;
144 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
145     &disallow_high_osrel, 0,
146     "Disallow execution of binaries built for higher version of the world");
147 
148 static int map_at_zero = 0;
149 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
150     "Permit processes to map an object at virtual address 0.");
151 
152 static int core_dump_can_intr = 1;
153 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
154     &core_dump_can_intr, 0,
155     "Core dumping interruptible with SIGKILL");
156 
157 static int
sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)158 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
159 {
160 	struct proc *p;
161 	vm_offset_t ps_strings;
162 
163 	p = curproc;
164 #ifdef SCTL_MASK32
165 	if (req->flags & SCTL_MASK32) {
166 		unsigned int val;
167 		val = (unsigned int)PROC_PS_STRINGS(p);
168 		return (SYSCTL_OUT(req, &val, sizeof(val)));
169 	}
170 #endif
171 	ps_strings = PROC_PS_STRINGS(p);
172 	return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
173 }
174 
175 static int
sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)176 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
177 {
178 	struct proc *p;
179 	vm_offset_t val;
180 
181 	p = curproc;
182 #ifdef SCTL_MASK32
183 	if (req->flags & SCTL_MASK32) {
184 		unsigned int val32;
185 
186 		val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
187 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
188 	}
189 #endif
190 	val = round_page(p->p_vmspace->vm_stacktop);
191 	return (SYSCTL_OUT(req, &val, sizeof(val)));
192 }
193 
194 static int
sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)195 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
196 {
197 	struct proc *p;
198 
199 	p = curproc;
200 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
201 	    sizeof(p->p_sysent->sv_stackprot)));
202 }
203 
204 /*
205  * Each of the items is a pointer to a `const struct execsw', hence the
206  * double pointer here.
207  */
208 static const struct execsw **execsw;
209 
210 #ifndef _SYS_SYSPROTO_H_
211 struct execve_args {
212 	char    *fname;
213 	char    **argv;
214 	char    **envv;
215 };
216 #endif
217 
218 int
sys_execve(struct thread * td,struct execve_args * uap)219 sys_execve(struct thread *td, struct execve_args *uap)
220 {
221 	struct image_args args;
222 	struct vmspace *oldvmspace;
223 	int error;
224 
225 	error = pre_execve(td, &oldvmspace);
226 	if (error != 0)
227 		return (error);
228 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
229 	    uap->argv, uap->envv);
230 	if (error == 0)
231 		error = kern_execve(td, &args, NULL, oldvmspace);
232 	post_execve(td, error, oldvmspace);
233 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
234 	return (error);
235 }
236 
237 #ifndef _SYS_SYSPROTO_H_
238 struct fexecve_args {
239 	int	fd;
240 	char	**argv;
241 	char	**envv;
242 };
243 #endif
244 int
sys_fexecve(struct thread * td,struct fexecve_args * uap)245 sys_fexecve(struct thread *td, struct fexecve_args *uap)
246 {
247 	struct image_args args;
248 	struct vmspace *oldvmspace;
249 	int error;
250 
251 	error = pre_execve(td, &oldvmspace);
252 	if (error != 0)
253 		return (error);
254 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
255 	    uap->argv, uap->envv);
256 	if (error == 0) {
257 		args.fd = uap->fd;
258 		error = kern_execve(td, &args, NULL, oldvmspace);
259 	}
260 	post_execve(td, error, oldvmspace);
261 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
262 	return (error);
263 }
264 
265 #ifndef _SYS_SYSPROTO_H_
266 struct __mac_execve_args {
267 	char	*fname;
268 	char	**argv;
269 	char	**envv;
270 	struct mac	*mac_p;
271 };
272 #endif
273 
274 int
sys___mac_execve(struct thread * td,struct __mac_execve_args * uap)275 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
276 {
277 #ifdef MAC
278 	struct image_args args;
279 	struct vmspace *oldvmspace;
280 	int error;
281 
282 	error = pre_execve(td, &oldvmspace);
283 	if (error != 0)
284 		return (error);
285 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
286 	    uap->argv, uap->envv);
287 	if (error == 0)
288 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
289 	post_execve(td, error, oldvmspace);
290 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
291 	return (error);
292 #else
293 	return (ENOSYS);
294 #endif
295 }
296 
297 int
pre_execve(struct thread * td,struct vmspace ** oldvmspace)298 pre_execve(struct thread *td, struct vmspace **oldvmspace)
299 {
300 	struct proc *p;
301 	int error;
302 
303 	KASSERT(td == curthread, ("non-current thread %p", td));
304 	error = 0;
305 	p = td->td_proc;
306 	if ((p->p_flag & P_HADTHREADS) != 0) {
307 		PROC_LOCK(p);
308 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
309 			error = ERESTART;
310 		PROC_UNLOCK(p);
311 	}
312 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
313 	    ("nested execve"));
314 	*oldvmspace = p->p_vmspace;
315 	return (error);
316 }
317 
318 void
post_execve(struct thread * td,int error,struct vmspace * oldvmspace)319 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
320 {
321 	struct proc *p;
322 
323 	KASSERT(td == curthread, ("non-current thread %p", td));
324 	p = td->td_proc;
325 	if ((p->p_flag & P_HADTHREADS) != 0) {
326 		PROC_LOCK(p);
327 		/*
328 		 * If success, we upgrade to SINGLE_EXIT state to
329 		 * force other threads to suicide.
330 		 */
331 		if (error == EJUSTRETURN)
332 			thread_single(p, SINGLE_EXIT);
333 		else
334 			thread_single_end(p, SINGLE_BOUNDARY);
335 		PROC_UNLOCK(p);
336 	}
337 	exec_cleanup(td, oldvmspace);
338 }
339 
340 /*
341  * kern_execve() has the astonishing property of not always returning to
342  * the caller.  If sufficiently bad things happen during the call to
343  * do_execve(), it can end up calling exit1(); as a result, callers must
344  * avoid doing anything which they might need to undo (e.g., allocating
345  * memory).
346  */
347 int
kern_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)348 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
349     struct vmspace *oldvmspace)
350 {
351 
352 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
353 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
354 	    exec_args_get_begin_envv(args) - args->begin_argv);
355 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
356 	    args->endp - exec_args_get_begin_envv(args));
357 #ifdef KTRACE
358 	if (KTRPOINT(td, KTR_ARGS)) {
359 		ktrdata(KTR_ARGS, args->begin_argv,
360 		    exec_args_get_begin_envv(args) - args->begin_argv);
361         }
362 	if (KTRPOINT(td, KTR_ENVS)) {
363 		ktrdata(KTR_ENVS, exec_args_get_begin_envv(args),
364 		    args->endp - exec_args_get_begin_envv(args));
365         }
366 #endif
367 	/* Must have at least one argument. */
368 	if (args->argc == 0) {
369 		exec_free_args(args);
370 		return (EINVAL);
371 	}
372 	return (do_execve(td, args, mac_p, oldvmspace));
373 }
374 
375 static void
execve_nosetid(struct image_params * imgp)376 execve_nosetid(struct image_params *imgp)
377 {
378 	imgp->credential_setid = false;
379 	if (imgp->newcred != NULL) {
380 		crfree(imgp->newcred);
381 		imgp->newcred = NULL;
382 	}
383 }
384 
385 /*
386  * In-kernel implementation of execve().  All arguments are assumed to be
387  * userspace pointers from the passed thread.
388  */
389 static int
do_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)390 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
391     struct vmspace *oldvmspace)
392 {
393 	struct proc *p = td->td_proc;
394 	struct nameidata nd;
395 	struct ucred *oldcred;
396 	struct uidinfo *euip = NULL;
397 	uintptr_t stack_base;
398 	struct image_params image_params, *imgp;
399 	struct vattr attr;
400 	struct pargs *oldargs = NULL, *newargs = NULL;
401 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
402 #ifdef KTRACE
403 	struct ktr_io_params *kiop;
404 #endif
405 	struct vnode *oldtextvp, *newtextvp;
406 	struct vnode *oldtextdvp, *newtextdvp;
407 	char *oldbinname, *newbinname;
408 	bool credential_changing;
409 #ifdef MAC
410 	struct label *interpvplabel = NULL;
411 	bool will_transition;
412 #endif
413 #ifdef HWPMC_HOOKS
414 	struct pmckern_procexec pe;
415 #endif
416 	int error, i, orig_osrel;
417 	uint32_t orig_fctl0;
418 	Elf_Brandinfo *orig_brandinfo;
419 	size_t freepath_size;
420 	static const char fexecv_proc_title[] = "(fexecv)";
421 
422 	imgp = &image_params;
423 	oldtextvp = oldtextdvp = NULL;
424 	newtextvp = newtextdvp = NULL;
425 	newbinname = oldbinname = NULL;
426 #ifdef KTRACE
427 	kiop = NULL;
428 #endif
429 
430 	/*
431 	 * Lock the process and set the P_INEXEC flag to indicate that
432 	 * it should be left alone until we're done here.  This is
433 	 * necessary to avoid race conditions - e.g. in ptrace() -
434 	 * that might allow a local user to illicitly obtain elevated
435 	 * privileges.
436 	 */
437 	PROC_LOCK(p);
438 	KASSERT((p->p_flag & P_INEXEC) == 0,
439 	    ("%s(): process already has P_INEXEC flag", __func__));
440 	p->p_flag |= P_INEXEC;
441 	PROC_UNLOCK(p);
442 
443 	/*
444 	 * Initialize part of the common data
445 	 */
446 	bzero(imgp, sizeof(*imgp));
447 	imgp->proc = p;
448 	imgp->attr = &attr;
449 	imgp->args = args;
450 	oldcred = p->p_ucred;
451 	orig_osrel = p->p_osrel;
452 	orig_fctl0 = p->p_fctl0;
453 	orig_brandinfo = p->p_elf_brandinfo;
454 
455 #ifdef MAC
456 	error = mac_execve_enter(imgp, mac_p);
457 	if (error)
458 		goto exec_fail;
459 #endif
460 
461 	SDT_PROBE1(proc, , , exec, args->fname);
462 
463 interpret:
464 	if (args->fname != NULL) {
465 #ifdef CAPABILITY_MODE
466 		if (CAP_TRACING(td))
467 			ktrcapfail(CAPFAIL_NAMEI, args->fname);
468 		/*
469 		 * While capability mode can't reach this point via direct
470 		 * path arguments to execve(), we also don't allow
471 		 * interpreters to be used in capability mode (for now).
472 		 * Catch indirect lookups and return a permissions error.
473 		 */
474 		if (IN_CAPABILITY_MODE(td)) {
475 			error = ECAPMODE;
476 			goto exec_fail;
477 		}
478 #endif
479 
480 		/*
481 		 * Translate the file name. namei() returns a vnode
482 		 * pointer in ni_vp among other things.
483 		 */
484 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
485 		    AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
486 		    args->fname);
487 
488 		error = namei(&nd);
489 		if (error)
490 			goto exec_fail;
491 
492 		newtextvp = nd.ni_vp;
493 		newtextdvp = nd.ni_dvp;
494 		nd.ni_dvp = NULL;
495 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
496 		    M_WAITOK);
497 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
498 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
499 		imgp->vp = newtextvp;
500 
501 		/*
502 		 * Do the best to calculate the full path to the image file.
503 		 */
504 		if (args->fname[0] == '/') {
505 			imgp->execpath = args->fname;
506 		} else {
507 			VOP_UNLOCK(imgp->vp);
508 			freepath_size = MAXPATHLEN;
509 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
510 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
511 			    &imgp->freepath, &freepath_size) != 0)
512 				imgp->execpath = args->fname;
513 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
514 		}
515 	} else if (imgp->interpreter_vp) {
516 		/*
517 		 * An image activator has already provided an open vnode
518 		 */
519 		newtextvp = imgp->interpreter_vp;
520 		imgp->interpreter_vp = NULL;
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 	} else {
528 		AUDIT_ARG_FD(args->fd);
529 
530 		/*
531 		 * If the descriptors was not opened with O_PATH, then
532 		 * we require that it was opened with O_EXEC or
533 		 * O_RDONLY.  In either case, exec_check_permissions()
534 		 * below checks _current_ file access mode regardless
535 		 * of the permissions additionally checked at the
536 		 * open(2).
537 		 */
538 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
539 		    &newtextvp);
540 		if (error != 0)
541 			goto exec_fail;
542 
543 		if (vn_fullpath(newtextvp, &imgp->execpath,
544 		    &imgp->freepath) != 0)
545 			imgp->execpath = args->fname;
546 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
547 		AUDIT_ARG_VNODE1(newtextvp);
548 		imgp->vp = newtextvp;
549 	}
550 
551 	/*
552 	 * Check file permissions.  Also 'opens' file and sets its vnode to
553 	 * text mode.
554 	 */
555 	error = exec_check_permissions(imgp);
556 	if (error)
557 		goto exec_fail_dealloc;
558 
559 	imgp->object = imgp->vp->v_object;
560 	if (imgp->object != NULL)
561 		vm_object_reference(imgp->object);
562 
563 	error = exec_map_first_page(imgp);
564 	if (error)
565 		goto exec_fail_dealloc;
566 
567 	imgp->proc->p_osrel = 0;
568 	imgp->proc->p_fctl0 = 0;
569 	imgp->proc->p_elf_brandinfo = NULL;
570 
571 	/*
572 	 * Implement image setuid/setgid.
573 	 *
574 	 * Determine new credentials before attempting image activators
575 	 * so that it can be used by process_exec handlers to determine
576 	 * credential/setid changes.
577 	 *
578 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
579 	 * the process is being traced.
580 	 *
581 	 * We disable setuid/setgid/etc in capability mode on the basis
582 	 * that most setugid applications are not written with that
583 	 * environment in mind, and will therefore almost certainly operate
584 	 * incorrectly. In principle there's no reason that setugid
585 	 * applications might not be useful in capability mode, so we may want
586 	 * to reconsider this conservative design choice in the future.
587 	 *
588 	 * XXXMAC: For the time being, use NOSUID to also prohibit
589 	 * transitions on the file system.
590 	 */
591 	credential_changing = false;
592 	credential_changing |= (attr.va_mode & S_ISUID) &&
593 	    oldcred->cr_uid != attr.va_uid;
594 	credential_changing |= (attr.va_mode & S_ISGID) &&
595 	    oldcred->cr_gid != attr.va_gid;
596 #ifdef MAC
597 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
598 	    interpvplabel, imgp) != 0;
599 	credential_changing |= will_transition;
600 #endif
601 
602 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
603 	if (credential_changing)
604 		imgp->proc->p_pdeathsig = 0;
605 
606 	if (credential_changing &&
607 #ifdef CAPABILITY_MODE
608 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
609 #endif
610 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
611 	    (p->p_flag & P_TRACED) == 0) {
612 		imgp->credential_setid = true;
613 		VOP_UNLOCK(imgp->vp);
614 		imgp->newcred = crdup(oldcred);
615 		if (attr.va_mode & S_ISUID) {
616 			euip = uifind(attr.va_uid);
617 			change_euid(imgp->newcred, euip);
618 		}
619 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
620 		if (attr.va_mode & S_ISGID)
621 			change_egid(imgp->newcred, attr.va_gid);
622 		/*
623 		 * Implement correct POSIX saved-id behavior.
624 		 *
625 		 * XXXMAC: Note that the current logic will save the
626 		 * uid and gid if a MAC domain transition occurs, even
627 		 * though maybe it shouldn't.
628 		 */
629 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
630 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
631 	} else {
632 		/*
633 		 * Implement correct POSIX saved-id behavior.
634 		 *
635 		 * XXX: It's not clear that the existing behavior is
636 		 * POSIX-compliant.  A number of sources indicate that the
637 		 * saved uid/gid should only be updated if the new ruid is
638 		 * not equal to the old ruid, or the new euid is not equal
639 		 * to the old euid and the new euid is not equal to the old
640 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
641 		 * Also, this code uses the new (replaced) euid and egid as
642 		 * the source, which may or may not be the right ones to use.
643 		 */
644 		if (oldcred->cr_svuid != oldcred->cr_uid ||
645 		    oldcred->cr_svgid != oldcred->cr_gid) {
646 			VOP_UNLOCK(imgp->vp);
647 			imgp->newcred = crdup(oldcred);
648 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
649 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
650 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
651 		}
652 	}
653 	/* The new credentials are installed into the process later. */
654 
655 	/*
656 	 *	Loop through the list of image activators, calling each one.
657 	 *	An activator returns -1 if there is no match, 0 on success,
658 	 *	and an error otherwise.
659 	 */
660 	error = -1;
661 	for (i = 0; error == -1 && execsw[i]; ++i) {
662 		if (execsw[i]->ex_imgact == NULL)
663 			continue;
664 		error = (*execsw[i]->ex_imgact)(imgp);
665 	}
666 
667 	if (error) {
668 		if (error == -1)
669 			error = ENOEXEC;
670 		goto exec_fail_dealloc;
671 	}
672 
673 	/*
674 	 * Special interpreter operation, cleanup and loop up to try to
675 	 * activate the interpreter.
676 	 */
677 	if (imgp->interpreted) {
678 		exec_unmap_first_page(imgp);
679 		/*
680 		 * The text reference needs to be removed for scripts.
681 		 * There is a short period before we determine that
682 		 * something is a script where text reference is active.
683 		 * The vnode lock is held over this entire period
684 		 * so nothing should illegitimately be blocked.
685 		 */
686 		MPASS(imgp->textset);
687 		VOP_UNSET_TEXT_CHECKED(newtextvp);
688 		imgp->textset = false;
689 		/* free name buffer and old vnode */
690 #ifdef MAC
691 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
692 #endif
693 		if (imgp->opened) {
694 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
695 			imgp->opened = false;
696 		}
697 		vput(newtextvp);
698 		imgp->vp = newtextvp = NULL;
699 		if (args->fname != NULL) {
700 			if (newtextdvp != NULL) {
701 				vrele(newtextdvp);
702 				newtextdvp = NULL;
703 			}
704 			NDFREE_PNBUF(&nd);
705 			free(newbinname, M_PARGS);
706 			newbinname = NULL;
707 		}
708 		vm_object_deallocate(imgp->object);
709 		imgp->object = NULL;
710 		execve_nosetid(imgp);
711 		imgp->execpath = NULL;
712 		free(imgp->freepath, M_TEMP);
713 		imgp->freepath = NULL;
714 		/* set new name to that of the interpreter */
715 		if (imgp->interpreter_vp) {
716 			args->fname = NULL;
717 		} else {
718 			args->fname = imgp->interpreter_name;
719 		}
720 		goto interpret;
721 	}
722 
723 	/*
724 	 * NB: We unlock the vnode here because it is believed that none
725 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
726 	 */
727 	VOP_UNLOCK(imgp->vp);
728 
729 	if (disallow_high_osrel &&
730 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
731 		error = ENOEXEC;
732 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
733 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
734 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
735 		goto exec_fail_dealloc;
736 	}
737 
738 	/*
739 	 * Copy out strings (args and env) and initialize stack base.
740 	 */
741 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
742 	if (error != 0) {
743 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
744 		goto exec_fail_dealloc;
745 	}
746 
747 	/*
748 	 * Stack setup.
749 	 */
750 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
751 	if (error != 0) {
752 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
753 		goto exec_fail_dealloc;
754 	}
755 
756 	/*
757 	 * For security and other reasons, the file descriptor table cannot be
758 	 * shared after an exec.
759 	 */
760 	fdunshare(td);
761 	pdunshare(td);
762 	/* close files on exec */
763 	fdcloseexec(td);
764 
765 	/*
766 	 * Malloc things before we need locks.
767 	 */
768 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
769 	/* Cache arguments if they fit inside our allowance */
770 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
771 		newargs = pargs_alloc(i);
772 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
773 	}
774 
775 	/*
776 	 * For security and other reasons, signal handlers cannot
777 	 * be shared after an exec. The new process gets a copy of the old
778 	 * handlers. In execsigs(), the new process will have its signals
779 	 * reset.
780 	 */
781 	if (sigacts_shared(p->p_sigacts)) {
782 		oldsigacts = p->p_sigacts;
783 		newsigacts = sigacts_alloc();
784 		sigacts_copy(newsigacts, oldsigacts);
785 	}
786 
787 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
788 
789 	PROC_LOCK(p);
790 	if (oldsigacts)
791 		p->p_sigacts = newsigacts;
792 	/* Stop profiling */
793 	stopprofclock(p);
794 
795 	/* reset caught signals */
796 	execsigs(p);
797 
798 	/* name this process - nameiexec(p, ndp) */
799 	bzero(p->p_comm, sizeof(p->p_comm));
800 	if (args->fname)
801 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
802 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
803 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
804 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
805 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
806 #ifdef KTR
807 	sched_clear_tdname(td);
808 #endif
809 
810 	/*
811 	 * mark as execed, wakeup the process that vforked (if any) and tell
812 	 * it that it now has its own resources back
813 	 */
814 	p->p_flag |= P_EXEC;
815 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
816 		p->p_flag2 &= ~P2_NOTRACE;
817 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
818 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
819 	p->p_flag2 &= ~(P2_MEMBAR_PRIVE | P2_MEMBAR_PRIVE_SYNCORE |
820 	    P2_MEMBAR_GLOBE);
821 	if (p->p_flag & P_PPWAIT) {
822 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
823 		cv_broadcast(&p->p_pwait);
824 		/* STOPs are no longer ignored, arrange for AST */
825 		signotify(td);
826 	}
827 
828 	if ((imgp->sysent->sv_setid_allowed != NULL &&
829 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
830 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
831 		execve_nosetid(imgp);
832 
833 	/*
834 	 * Implement image setuid/setgid installation.
835 	 */
836 	if (imgp->credential_setid) {
837 		/*
838 		 * Turn off syscall tracing for set-id programs, except for
839 		 * root.  Record any set-id flags first to make sure that
840 		 * we do not regain any tracing during a possible block.
841 		 */
842 		setsugid(p);
843 #ifdef KTRACE
844 		kiop = ktrprocexec(p);
845 #endif
846 		/*
847 		 * Close any file descriptors 0..2 that reference procfs,
848 		 * then make sure file descriptors 0..2 are in use.
849 		 *
850 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
851 		 * taking sleepable locks, so temporarily drop our locks.
852 		 */
853 		PROC_UNLOCK(p);
854 		VOP_UNLOCK(imgp->vp);
855 		fdsetugidsafety(td);
856 		error = fdcheckstd(td);
857 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
858 		if (error != 0)
859 			goto exec_fail_dealloc;
860 		PROC_LOCK(p);
861 #ifdef MAC
862 		if (will_transition) {
863 			mac_vnode_execve_transition(oldcred, imgp->newcred,
864 			    imgp->vp, interpvplabel, imgp);
865 		}
866 #endif
867 	} else {
868 		if (oldcred->cr_uid == oldcred->cr_ruid &&
869 		    oldcred->cr_gid == oldcred->cr_rgid)
870 			p->p_flag &= ~P_SUGID;
871 	}
872 	/*
873 	 * Set the new credentials.
874 	 */
875 	if (imgp->newcred != NULL) {
876 		proc_set_cred(p, imgp->newcred);
877 		crfree(oldcred);
878 		oldcred = NULL;
879 	}
880 
881 	/*
882 	 * Store the vp for use in kern.proc.pathname.  This vnode was
883 	 * referenced by namei() or by fexecve variant of fname handling.
884 	 */
885 	oldtextvp = p->p_textvp;
886 	p->p_textvp = newtextvp;
887 	oldtextdvp = p->p_textdvp;
888 	p->p_textdvp = newtextdvp;
889 	newtextdvp = NULL;
890 	oldbinname = p->p_binname;
891 	p->p_binname = newbinname;
892 	newbinname = NULL;
893 
894 #ifdef KDTRACE_HOOKS
895 	/*
896 	 * Tell the DTrace fasttrap provider about the exec if it
897 	 * has declared an interest.
898 	 */
899 	if (dtrace_fasttrap_exec)
900 		dtrace_fasttrap_exec(p);
901 #endif
902 
903 	/*
904 	 * Notify others that we exec'd, and clear the P_INEXEC flag
905 	 * as we're now a bona fide freshly-execed process.
906 	 */
907 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
908 	p->p_flag &= ~P_INEXEC;
909 
910 	/* clear "fork but no exec" flag, as we _are_ execing */
911 	p->p_acflag &= ~AFORK;
912 
913 	/*
914 	 * Free any previous argument cache and replace it with
915 	 * the new argument cache, if any.
916 	 */
917 	oldargs = p->p_args;
918 	p->p_args = newargs;
919 	newargs = NULL;
920 
921 	PROC_UNLOCK(p);
922 
923 #ifdef	HWPMC_HOOKS
924 	/*
925 	 * Check if system-wide sampling is in effect or if the
926 	 * current process is using PMCs.  If so, do exec() time
927 	 * processing.  This processing needs to happen AFTER the
928 	 * P_INEXEC flag is cleared.
929 	 */
930 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
931 		VOP_UNLOCK(imgp->vp);
932 		pe.pm_credentialschanged = credential_changing;
933 		pe.pm_baseaddr = imgp->reloc_base;
934 		pe.pm_dynaddr = imgp->et_dyn_addr;
935 
936 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
937 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
938 	}
939 #endif
940 
941 	/* Set values passed into the program in registers. */
942 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
943 
944 	VOP_MMAPPED(imgp->vp);
945 
946 	SDT_PROBE1(proc, , , exec__success, args->fname);
947 
948 exec_fail_dealloc:
949 	if (error != 0) {
950 		p->p_osrel = orig_osrel;
951 		p->p_fctl0 = orig_fctl0;
952 		p->p_elf_brandinfo = orig_brandinfo;
953 	}
954 
955 	if (imgp->firstpage != NULL)
956 		exec_unmap_first_page(imgp);
957 
958 	if (imgp->vp != NULL) {
959 		if (imgp->opened)
960 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
961 		if (imgp->textset)
962 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
963 		if (error != 0)
964 			vput(imgp->vp);
965 		else
966 			VOP_UNLOCK(imgp->vp);
967 		if (args->fname != NULL)
968 			NDFREE_PNBUF(&nd);
969 		if (newtextdvp != NULL)
970 			vrele(newtextdvp);
971 		free(newbinname, M_PARGS);
972 	}
973 
974 	if (imgp->object != NULL)
975 		vm_object_deallocate(imgp->object);
976 
977 	free(imgp->freepath, M_TEMP);
978 
979 	if (error == 0) {
980 		if (p->p_ptevents & PTRACE_EXEC) {
981 			PROC_LOCK(p);
982 			if (p->p_ptevents & PTRACE_EXEC)
983 				td->td_dbgflags |= TDB_EXEC;
984 			PROC_UNLOCK(p);
985 		}
986 	} else {
987 exec_fail:
988 		/* we're done here, clear P_INEXEC */
989 		PROC_LOCK(p);
990 		p->p_flag &= ~P_INEXEC;
991 		PROC_UNLOCK(p);
992 
993 		SDT_PROBE1(proc, , , exec__failure, error);
994 	}
995 
996 	if (imgp->newcred != NULL && oldcred != NULL)
997 		crfree(imgp->newcred);
998 
999 #ifdef MAC
1000 	mac_execve_exit(imgp);
1001 	mac_execve_interpreter_exit(interpvplabel);
1002 #endif
1003 	exec_free_args(args);
1004 
1005 	/*
1006 	 * Handle deferred decrement of ref counts.
1007 	 */
1008 	if (oldtextvp != NULL)
1009 		vrele(oldtextvp);
1010 	if (oldtextdvp != NULL)
1011 		vrele(oldtextdvp);
1012 	free(oldbinname, M_PARGS);
1013 #ifdef KTRACE
1014 	ktr_io_params_free(kiop);
1015 #endif
1016 	pargs_drop(oldargs);
1017 	pargs_drop(newargs);
1018 	if (oldsigacts != NULL)
1019 		sigacts_free(oldsigacts);
1020 	if (euip != NULL)
1021 		uifree(euip);
1022 
1023 	if (error && imgp->vmspace_destroyed) {
1024 		/* sorry, no more process anymore. exit gracefully */
1025 		exec_cleanup(td, oldvmspace);
1026 		exit1(td, 0, SIGABRT);
1027 		/* NOT REACHED */
1028 	}
1029 
1030 #ifdef KTRACE
1031 	if (error == 0)
1032 		ktrprocctor(p);
1033 #endif
1034 
1035 	/*
1036 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1037 	 * the register values put in place by exec_setregs().
1038 	 * Implementations of cpu_set_syscall_retval() will leave
1039 	 * registers unmodified when returning EJUSTRETURN.
1040 	 */
1041 	return (error == 0 ? EJUSTRETURN : error);
1042 }
1043 
1044 void
exec_cleanup(struct thread * td,struct vmspace * oldvmspace)1045 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1046 {
1047 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1048 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1049 		    ("oldvmspace still used"));
1050 		vmspace_free(oldvmspace);
1051 		td->td_pflags &= ~TDP_EXECVMSPC;
1052 	}
1053 }
1054 
1055 int
exec_map_first_page(struct image_params * imgp)1056 exec_map_first_page(struct image_params *imgp)
1057 {
1058 	vm_object_t object;
1059 	vm_page_t m;
1060 	int error;
1061 
1062 	if (imgp->firstpage != NULL)
1063 		exec_unmap_first_page(imgp);
1064 
1065 	object = imgp->vp->v_object;
1066 	if (object == NULL)
1067 		return (EACCES);
1068 #if VM_NRESERVLEVEL > 0
1069 	if ((object->flags & OBJ_COLORED) == 0) {
1070 		VM_OBJECT_WLOCK(object);
1071 		vm_object_color(object, 0);
1072 		VM_OBJECT_WUNLOCK(object);
1073 	}
1074 #endif
1075 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1076 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1077 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1078 
1079 	if (error != VM_PAGER_OK)
1080 		return (EIO);
1081 	imgp->firstpage = sf_buf_alloc(m, 0);
1082 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1083 
1084 	return (0);
1085 }
1086 
1087 void
exec_unmap_first_page(struct image_params * imgp)1088 exec_unmap_first_page(struct image_params *imgp)
1089 {
1090 	vm_page_t m;
1091 
1092 	if (imgp->firstpage != NULL) {
1093 		m = sf_buf_page(imgp->firstpage);
1094 		sf_buf_free(imgp->firstpage);
1095 		imgp->firstpage = NULL;
1096 		vm_page_unwire(m, PQ_ACTIVE);
1097 	}
1098 }
1099 
1100 void
exec_onexec_old(struct thread * td)1101 exec_onexec_old(struct thread *td)
1102 {
1103 	sigfastblock_clear(td);
1104 	umtx_exec(td->td_proc);
1105 }
1106 
1107 /*
1108  * This is an optimization which removes the unmanaged shared page
1109  * mapping. In combination with pmap_remove_pages(), which cleans all
1110  * managed mappings in the process' vmspace pmap, no work will be left
1111  * for pmap_remove(min, max).
1112  */
1113 void
exec_free_abi_mappings(struct proc * p)1114 exec_free_abi_mappings(struct proc *p)
1115 {
1116 	struct vmspace *vmspace;
1117 
1118 	vmspace = p->p_vmspace;
1119 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1120 		return;
1121 
1122 	if (!PROC_HAS_SHP(p))
1123 		return;
1124 
1125 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1126 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1127 }
1128 
1129 /*
1130  * Run down the current address space and install a new one.
1131  */
1132 int
exec_new_vmspace(struct image_params * imgp,struct sysentvec * sv)1133 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1134 {
1135 	int error;
1136 	struct proc *p = imgp->proc;
1137 	struct vmspace *vmspace = p->p_vmspace;
1138 	struct thread *td = curthread;
1139 	vm_offset_t sv_minuser;
1140 	vm_map_t map;
1141 
1142 	imgp->vmspace_destroyed = true;
1143 	imgp->sysent = sv;
1144 
1145 	if (p->p_sysent->sv_onexec_old != NULL)
1146 		p->p_sysent->sv_onexec_old(td);
1147 	itimers_exec(p);
1148 
1149 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1150 
1151 	/*
1152 	 * Blow away entire process VM, if address space not shared,
1153 	 * otherwise, create a new VM space so that other threads are
1154 	 * not disrupted
1155 	 */
1156 	map = &vmspace->vm_map;
1157 	if (map_at_zero)
1158 		sv_minuser = sv->sv_minuser;
1159 	else
1160 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1161 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1162 	    vm_map_min(map) == sv_minuser &&
1163 	    vm_map_max(map) == sv->sv_maxuser &&
1164 	    cpu_exec_vmspace_reuse(p, map)) {
1165 		exec_free_abi_mappings(p);
1166 		shmexit(vmspace);
1167 		pmap_remove_pages(vmspace_pmap(vmspace));
1168 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1169 		/*
1170 		 * An exec terminates mlockall(MCL_FUTURE).
1171 		 * ASLR and W^X states must be re-evaluated.
1172 		 */
1173 		vm_map_lock(map);
1174 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1175 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1176 		vm_map_unlock(map);
1177 	} else {
1178 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1179 		if (error)
1180 			return (error);
1181 		vmspace = p->p_vmspace;
1182 		map = &vmspace->vm_map;
1183 	}
1184 	map->flags |= imgp->map_flags;
1185 
1186 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1187 }
1188 
1189 /*
1190  * Compute the stack size limit and map the main process stack.
1191  * Map the shared page.
1192  */
1193 int
exec_map_stack(struct image_params * imgp)1194 exec_map_stack(struct image_params *imgp)
1195 {
1196 	struct rlimit rlim_stack;
1197 	struct sysentvec *sv;
1198 	struct proc *p;
1199 	vm_map_t map;
1200 	struct vmspace *vmspace;
1201 	vm_offset_t stack_addr, stack_top;
1202 	vm_offset_t sharedpage_addr;
1203 	u_long ssiz;
1204 	int error, find_space, stack_off;
1205 	vm_prot_t stack_prot;
1206 	vm_object_t obj;
1207 
1208 	p = imgp->proc;
1209 	sv = p->p_sysent;
1210 
1211 	if (imgp->stack_sz != 0) {
1212 		ssiz = trunc_page(imgp->stack_sz);
1213 		PROC_LOCK(p);
1214 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1215 		PROC_UNLOCK(p);
1216 		if (ssiz > rlim_stack.rlim_max)
1217 			ssiz = rlim_stack.rlim_max;
1218 		if (ssiz > rlim_stack.rlim_cur) {
1219 			rlim_stack.rlim_cur = ssiz;
1220 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1221 		}
1222 	} else if (sv->sv_maxssiz != NULL) {
1223 		ssiz = *sv->sv_maxssiz;
1224 	} else {
1225 		ssiz = maxssiz;
1226 	}
1227 
1228 	vmspace = p->p_vmspace;
1229 	map = &vmspace->vm_map;
1230 
1231 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1232 	    imgp->stack_prot : sv->sv_stackprot;
1233 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1234 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1235 		    lim_max(curthread, RLIMIT_DATA));
1236 		find_space = VMFS_ANY_SPACE;
1237 	} else {
1238 		stack_addr = sv->sv_usrstack - ssiz;
1239 		find_space = VMFS_NO_SPACE;
1240 	}
1241 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1242 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1243 	    MAP_STACK_AREA);
1244 	if (error != KERN_SUCCESS) {
1245 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1246 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1247 		    stack_prot, error, vm_mmap_to_errno(error));
1248 		return (vm_mmap_to_errno(error));
1249 	}
1250 
1251 	stack_top = stack_addr + ssiz;
1252 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1253 		/* Randomize within the first page of the stack. */
1254 		arc4rand(&stack_off, sizeof(stack_off), 0);
1255 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1256 	}
1257 
1258 	/* Map a shared page */
1259 	obj = sv->sv_shared_page_obj;
1260 	if (obj == NULL) {
1261 		sharedpage_addr = 0;
1262 		goto out;
1263 	}
1264 
1265 	/*
1266 	 * If randomization is disabled then the shared page will
1267 	 * be mapped at address specified in sysentvec.
1268 	 * Otherwise any address above .data section can be selected.
1269 	 * Same logic is used for stack address randomization.
1270 	 * If the address randomization is applied map a guard page
1271 	 * at the top of UVA.
1272 	 */
1273 	vm_object_reference(obj);
1274 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1275 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1276 		    lim_max(curthread, RLIMIT_DATA));
1277 
1278 		error = vm_map_fixed(map, NULL, 0,
1279 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1280 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1281 		if (error != KERN_SUCCESS) {
1282 			/*
1283 			 * This is not fatal, so let's just print a warning
1284 			 * and continue.
1285 			 */
1286 			uprintf("%s: Mapping guard page at the top of UVA failed"
1287 			    " mach error %d errno %d",
1288 			    __func__, error, vm_mmap_to_errno(error));
1289 		}
1290 
1291 		error = vm_map_find(map, obj, 0,
1292 		    &sharedpage_addr, sv->sv_shared_page_len,
1293 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1294 		    VM_PROT_READ | VM_PROT_EXECUTE,
1295 		    VM_PROT_READ | VM_PROT_EXECUTE,
1296 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1297 	} else {
1298 		sharedpage_addr = sv->sv_shared_page_base;
1299 		vm_map_fixed(map, obj, 0,
1300 		    sharedpage_addr, sv->sv_shared_page_len,
1301 		    VM_PROT_READ | VM_PROT_EXECUTE,
1302 		    VM_PROT_READ | VM_PROT_EXECUTE,
1303 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1304 	}
1305 	if (error != KERN_SUCCESS) {
1306 		uprintf("%s: mapping shared page at addr: %p"
1307 		    "failed, mach error %d errno %d\n", __func__,
1308 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1309 		vm_object_deallocate(obj);
1310 		return (vm_mmap_to_errno(error));
1311 	}
1312 out:
1313 	/*
1314 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1315 	 * are still used to enforce the stack rlimit on the process stack.
1316 	 */
1317 	vmspace->vm_maxsaddr = (char *)stack_addr;
1318 	vmspace->vm_stacktop = stack_top;
1319 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1320 	vmspace->vm_shp_base = sharedpage_addr;
1321 
1322 	return (0);
1323 }
1324 
1325 /*
1326  * Copy out argument and environment strings from the old process address
1327  * space into the temporary string buffer.
1328  */
1329 int
exec_copyin_args(struct image_args * args,const char * fname,enum uio_seg segflg,char ** argv,char ** envv)1330 exec_copyin_args(struct image_args *args, const char *fname,
1331     enum uio_seg segflg, char **argv, char **envv)
1332 {
1333 	u_long arg, env;
1334 	int error;
1335 
1336 	bzero(args, sizeof(*args));
1337 	if (argv == NULL)
1338 		return (EFAULT);
1339 
1340 	/*
1341 	 * Allocate demand-paged memory for the file name, argument, and
1342 	 * environment strings.
1343 	 */
1344 	error = exec_alloc_args(args);
1345 	if (error != 0)
1346 		return (error);
1347 
1348 	/*
1349 	 * Copy the file name.
1350 	 */
1351 	error = exec_args_add_fname(args, fname, segflg);
1352 	if (error != 0)
1353 		goto err_exit;
1354 
1355 	/*
1356 	 * extract arguments first
1357 	 */
1358 	for (;;) {
1359 		error = fueword(argv++, &arg);
1360 		if (error == -1) {
1361 			error = EFAULT;
1362 			goto err_exit;
1363 		}
1364 		if (arg == 0)
1365 			break;
1366 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1367 		    UIO_USERSPACE);
1368 		if (error != 0)
1369 			goto err_exit;
1370 	}
1371 
1372 	/*
1373 	 * extract environment strings
1374 	 */
1375 	if (envv) {
1376 		for (;;) {
1377 			error = fueword(envv++, &env);
1378 			if (error == -1) {
1379 				error = EFAULT;
1380 				goto err_exit;
1381 			}
1382 			if (env == 0)
1383 				break;
1384 			error = exec_args_add_env(args,
1385 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1386 			if (error != 0)
1387 				goto err_exit;
1388 		}
1389 	}
1390 
1391 	return (0);
1392 
1393 err_exit:
1394 	exec_free_args(args);
1395 	return (error);
1396 }
1397 
1398 struct exec_args_kva {
1399 	vm_offset_t addr;
1400 	u_int gen;
1401 	SLIST_ENTRY(exec_args_kva) next;
1402 };
1403 
1404 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1405 
1406 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1407 static struct mtx exec_args_kva_mtx;
1408 static u_int exec_args_gen;
1409 
1410 static void
exec_prealloc_args_kva(void * arg __unused)1411 exec_prealloc_args_kva(void *arg __unused)
1412 {
1413 	struct exec_args_kva *argkva;
1414 	u_int i;
1415 
1416 	SLIST_INIT(&exec_args_kva_freelist);
1417 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1418 	for (i = 0; i < exec_map_entries; i++) {
1419 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1420 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1421 		argkva->gen = exec_args_gen;
1422 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1423 	}
1424 }
1425 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1426 
1427 static vm_offset_t
exec_alloc_args_kva(void ** cookie)1428 exec_alloc_args_kva(void **cookie)
1429 {
1430 	struct exec_args_kva *argkva;
1431 
1432 	argkva = (void *)atomic_readandclear_ptr(
1433 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1434 	if (argkva == NULL) {
1435 		mtx_lock(&exec_args_kva_mtx);
1436 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1437 			(void)mtx_sleep(&exec_args_kva_freelist,
1438 			    &exec_args_kva_mtx, 0, "execkva", 0);
1439 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1440 		mtx_unlock(&exec_args_kva_mtx);
1441 	}
1442 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1443 	    exec_map_entry_size, 0);
1444 	*(struct exec_args_kva **)cookie = argkva;
1445 	return (argkva->addr);
1446 }
1447 
1448 static void
exec_release_args_kva(struct exec_args_kva * argkva,u_int gen)1449 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1450 {
1451 	vm_offset_t base;
1452 
1453 	base = argkva->addr;
1454 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1455 	    KASAN_EXEC_ARGS_FREED);
1456 	if (argkva->gen != gen) {
1457 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1458 		    MADV_FREE);
1459 		argkva->gen = gen;
1460 	}
1461 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1462 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1463 		mtx_lock(&exec_args_kva_mtx);
1464 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1465 		wakeup_one(&exec_args_kva_freelist);
1466 		mtx_unlock(&exec_args_kva_mtx);
1467 	}
1468 }
1469 
1470 static void
exec_free_args_kva(void * cookie)1471 exec_free_args_kva(void *cookie)
1472 {
1473 
1474 	exec_release_args_kva(cookie, exec_args_gen);
1475 }
1476 
1477 static void
exec_args_kva_lowmem(void * arg __unused)1478 exec_args_kva_lowmem(void *arg __unused)
1479 {
1480 	SLIST_HEAD(, exec_args_kva) head;
1481 	struct exec_args_kva *argkva;
1482 	u_int gen;
1483 	int i;
1484 
1485 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1486 
1487 	/*
1488 	 * Force an madvise of each KVA range. Any currently allocated ranges
1489 	 * will have MADV_FREE applied once they are freed.
1490 	 */
1491 	SLIST_INIT(&head);
1492 	mtx_lock(&exec_args_kva_mtx);
1493 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1494 	mtx_unlock(&exec_args_kva_mtx);
1495 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1496 		SLIST_REMOVE_HEAD(&head, next);
1497 		exec_release_args_kva(argkva, gen);
1498 	}
1499 
1500 	CPU_FOREACH(i) {
1501 		argkva = (void *)atomic_readandclear_ptr(
1502 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1503 		if (argkva != NULL)
1504 			exec_release_args_kva(argkva, gen);
1505 	}
1506 }
1507 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1508     EVENTHANDLER_PRI_ANY);
1509 
1510 /*
1511  * Allocate temporary demand-paged, zero-filled memory for the file name,
1512  * argument, and environment strings.
1513  */
1514 int
exec_alloc_args(struct image_args * args)1515 exec_alloc_args(struct image_args *args)
1516 {
1517 
1518 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1519 	return (0);
1520 }
1521 
1522 void
exec_free_args(struct image_args * args)1523 exec_free_args(struct image_args *args)
1524 {
1525 
1526 	if (args->buf != NULL) {
1527 		exec_free_args_kva(args->bufkva);
1528 		args->buf = NULL;
1529 	}
1530 	if (args->fname_buf != NULL) {
1531 		free(args->fname_buf, M_TEMP);
1532 		args->fname_buf = NULL;
1533 	}
1534 }
1535 
1536 /*
1537  * A set to functions to fill struct image args.
1538  *
1539  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1540  * fname) before the other functions.  All exec_args_add_arg() calls must
1541  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1542  * may be called any time after exec_args_add_fname().
1543  *
1544  * exec_args_add_fname() - install path to be executed
1545  * exec_args_add_arg() - append an argument string
1546  * exec_args_add_env() - append an env string
1547  * exec_args_adjust_args() - adjust location of the argument list to
1548  *                           allow new arguments to be prepended
1549  */
1550 int
exec_args_add_fname(struct image_args * args,const char * fname,enum uio_seg segflg)1551 exec_args_add_fname(struct image_args *args, const char *fname,
1552     enum uio_seg segflg)
1553 {
1554 	int error;
1555 	size_t length;
1556 
1557 	KASSERT(args->fname == NULL, ("fname already appended"));
1558 	KASSERT(args->endp == NULL, ("already appending to args"));
1559 
1560 	if (fname != NULL) {
1561 		args->fname = args->buf;
1562 		error = segflg == UIO_SYSSPACE ?
1563 		    copystr(fname, args->fname, PATH_MAX, &length) :
1564 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1565 		if (error != 0)
1566 			return (error == ENAMETOOLONG ? E2BIG : error);
1567 	} else
1568 		length = 0;
1569 
1570 	/* Set up for _arg_*()/_env_*() */
1571 	args->endp = args->buf + length;
1572 	/* begin_argv must be set and kept updated */
1573 	args->begin_argv = args->endp;
1574 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1575 	    ("too little space remaining for arguments %zu < %zu",
1576 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1577 	args->stringspace = ARG_MAX;
1578 
1579 	return (0);
1580 }
1581 
1582 static int
exec_args_add_str(struct image_args * args,const char * str,enum uio_seg segflg,int * countp)1583 exec_args_add_str(struct image_args *args, const char *str,
1584     enum uio_seg segflg, int *countp)
1585 {
1586 	int error;
1587 	size_t length;
1588 
1589 	KASSERT(args->endp != NULL, ("endp not initialized"));
1590 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1591 
1592 	error = (segflg == UIO_SYSSPACE) ?
1593 	    copystr(str, args->endp, args->stringspace, &length) :
1594 	    copyinstr(str, args->endp, args->stringspace, &length);
1595 	if (error != 0)
1596 		return (error == ENAMETOOLONG ? E2BIG : error);
1597 	args->stringspace -= length;
1598 	args->endp += length;
1599 	(*countp)++;
1600 
1601 	return (0);
1602 }
1603 
1604 int
exec_args_add_arg(struct image_args * args,const char * argp,enum uio_seg segflg)1605 exec_args_add_arg(struct image_args *args, const char *argp,
1606     enum uio_seg segflg)
1607 {
1608 
1609 	KASSERT(args->envc == 0, ("appending args after env"));
1610 
1611 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1612 }
1613 
1614 int
exec_args_add_env(struct image_args * args,const char * envp,enum uio_seg segflg)1615 exec_args_add_env(struct image_args *args, const char *envp,
1616     enum uio_seg segflg)
1617 {
1618 
1619 	if (args->envc == 0)
1620 		args->begin_envv = args->endp;
1621 
1622 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1623 }
1624 
1625 int
exec_args_adjust_args(struct image_args * args,size_t consume,ssize_t extend)1626 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1627 {
1628 	ssize_t offset;
1629 
1630 	KASSERT(args->endp != NULL, ("endp not initialized"));
1631 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1632 
1633 	offset = extend - consume;
1634 	if (args->stringspace < offset)
1635 		return (E2BIG);
1636 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1637 	    args->endp - args->begin_argv + consume);
1638 	if (args->envc > 0)
1639 		args->begin_envv += offset;
1640 	args->endp += offset;
1641 	args->stringspace -= offset;
1642 	return (0);
1643 }
1644 
1645 char *
exec_args_get_begin_envv(struct image_args * args)1646 exec_args_get_begin_envv(struct image_args *args)
1647 {
1648 
1649 	KASSERT(args->endp != NULL, ("endp not initialized"));
1650 
1651 	if (args->envc > 0)
1652 		return (args->begin_envv);
1653 	return (args->endp);
1654 }
1655 
1656 /*
1657  * Copy strings out to the new process address space, constructing new arg
1658  * and env vector tables. Return a pointer to the base so that it can be used
1659  * as the initial stack pointer.
1660  */
1661 int
exec_copyout_strings(struct image_params * imgp,uintptr_t * stack_base)1662 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1663 {
1664 	int argc, envc;
1665 	char **vectp;
1666 	char *stringp;
1667 	uintptr_t destp, ustringp;
1668 	struct ps_strings *arginfo;
1669 	struct proc *p;
1670 	struct sysentvec *sysent;
1671 	size_t execpath_len;
1672 	int error, szsigcode;
1673 	char canary[sizeof(long) * 8];
1674 
1675 	p = imgp->proc;
1676 	sysent = p->p_sysent;
1677 
1678 	destp =	PROC_PS_STRINGS(p);
1679 	arginfo = imgp->ps_strings = (void *)destp;
1680 
1681 	/*
1682 	 * Install sigcode.
1683 	 */
1684 	if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) {
1685 		szsigcode = *(sysent->sv_szsigcode);
1686 		destp -= szsigcode;
1687 		destp = rounddown2(destp, sizeof(void *));
1688 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1689 		if (error != 0)
1690 			return (error);
1691 	}
1692 
1693 	/*
1694 	 * Copy the image path for the rtld.
1695 	 */
1696 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1697 		execpath_len = strlen(imgp->execpath) + 1;
1698 		destp -= execpath_len;
1699 		destp = rounddown2(destp, sizeof(void *));
1700 		imgp->execpathp = (void *)destp;
1701 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1702 		if (error != 0)
1703 			return (error);
1704 	}
1705 
1706 	/*
1707 	 * Prepare the canary for SSP.
1708 	 */
1709 	arc4rand(canary, sizeof(canary), 0);
1710 	destp -= sizeof(canary);
1711 	imgp->canary = (void *)destp;
1712 	error = copyout(canary, imgp->canary, sizeof(canary));
1713 	if (error != 0)
1714 		return (error);
1715 	imgp->canarylen = sizeof(canary);
1716 
1717 	/*
1718 	 * Prepare the pagesizes array.
1719 	 */
1720 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1721 	destp -= imgp->pagesizeslen;
1722 	destp = rounddown2(destp, sizeof(void *));
1723 	imgp->pagesizes = (void *)destp;
1724 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1725 	if (error != 0)
1726 		return (error);
1727 
1728 	/*
1729 	 * Allocate room for the argument and environment strings.
1730 	 */
1731 	destp -= ARG_MAX - imgp->args->stringspace;
1732 	destp = rounddown2(destp, sizeof(void *));
1733 	ustringp = destp;
1734 
1735 	if (imgp->auxargs) {
1736 		/*
1737 		 * Allocate room on the stack for the ELF auxargs
1738 		 * array.  It has up to AT_COUNT entries.
1739 		 */
1740 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1741 		destp = rounddown2(destp, sizeof(void *));
1742 	}
1743 
1744 	vectp = (char **)destp;
1745 
1746 	/*
1747 	 * Allocate room for the argv[] and env vectors including the
1748 	 * terminating NULL pointers.
1749 	 */
1750 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1751 
1752 	/*
1753 	 * vectp also becomes our initial stack base
1754 	 */
1755 	*stack_base = (uintptr_t)vectp;
1756 
1757 	stringp = imgp->args->begin_argv;
1758 	argc = imgp->args->argc;
1759 	envc = imgp->args->envc;
1760 
1761 	/*
1762 	 * Copy out strings - arguments and environment.
1763 	 */
1764 	error = copyout(stringp, (void *)ustringp,
1765 	    ARG_MAX - imgp->args->stringspace);
1766 	if (error != 0)
1767 		return (error);
1768 
1769 	/*
1770 	 * Fill in "ps_strings" struct for ps, w, etc.
1771 	 */
1772 	imgp->argv = vectp;
1773 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1774 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1775 		return (EFAULT);
1776 
1777 	/*
1778 	 * Fill in argument portion of vector table.
1779 	 */
1780 	for (; argc > 0; --argc) {
1781 		if (suword(vectp++, ustringp) != 0)
1782 			return (EFAULT);
1783 		while (*stringp++ != 0)
1784 			ustringp++;
1785 		ustringp++;
1786 	}
1787 
1788 	/* a null vector table pointer separates the argp's from the envp's */
1789 	if (suword(vectp++, 0) != 0)
1790 		return (EFAULT);
1791 
1792 	imgp->envv = vectp;
1793 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1794 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1795 		return (EFAULT);
1796 
1797 	/*
1798 	 * Fill in environment portion of vector table.
1799 	 */
1800 	for (; envc > 0; --envc) {
1801 		if (suword(vectp++, ustringp) != 0)
1802 			return (EFAULT);
1803 		while (*stringp++ != 0)
1804 			ustringp++;
1805 		ustringp++;
1806 	}
1807 
1808 	/* end of vector table is a null pointer */
1809 	if (suword(vectp, 0) != 0)
1810 		return (EFAULT);
1811 
1812 	if (imgp->auxargs) {
1813 		vectp++;
1814 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1815 		    (uintptr_t)vectp);
1816 		if (error != 0)
1817 			return (error);
1818 	}
1819 
1820 	return (0);
1821 }
1822 
1823 /*
1824  * Check permissions of file to execute.
1825  *	Called with imgp->vp locked.
1826  *	Return 0 for success or error code on failure.
1827  */
1828 int
exec_check_permissions(struct image_params * imgp)1829 exec_check_permissions(struct image_params *imgp)
1830 {
1831 	struct vnode *vp = imgp->vp;
1832 	struct vattr *attr = imgp->attr;
1833 	struct thread *td;
1834 	int error;
1835 
1836 	td = curthread;
1837 
1838 	/* Get file attributes */
1839 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1840 	if (error)
1841 		return (error);
1842 
1843 #ifdef MAC
1844 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1845 	if (error)
1846 		return (error);
1847 #endif
1848 
1849 	/*
1850 	 * 1) Check if file execution is disabled for the filesystem that
1851 	 *    this file resides on.
1852 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1853 	 *    privileged user will always succeed, and we don't want this
1854 	 *    to happen unless the file really is executable.
1855 	 * 3) Ensure that the file is a regular file.
1856 	 */
1857 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1858 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1859 	    (attr->va_type != VREG))
1860 		return (EACCES);
1861 
1862 	/*
1863 	 * Zero length files can't be exec'd
1864 	 */
1865 	if (attr->va_size == 0)
1866 		return (ENOEXEC);
1867 
1868 	/*
1869 	 *  Check for execute permission to file based on current credentials.
1870 	 */
1871 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1872 	if (error)
1873 		return (error);
1874 
1875 	/*
1876 	 * Check number of open-for-writes on the file and deny execution
1877 	 * if there are any.
1878 	 *
1879 	 * Add a text reference now so no one can write to the
1880 	 * executable while we're activating it.
1881 	 *
1882 	 * Remember if this was set before and unset it in case this is not
1883 	 * actually an executable image.
1884 	 */
1885 	error = VOP_SET_TEXT(vp);
1886 	if (error != 0)
1887 		return (error);
1888 	imgp->textset = true;
1889 
1890 	/*
1891 	 * Call filesystem specific open routine (which does nothing in the
1892 	 * general case).
1893 	 */
1894 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1895 	if (error == 0)
1896 		imgp->opened = true;
1897 	return (error);
1898 }
1899 
1900 /*
1901  * Exec handler registration
1902  */
1903 int
exec_register(const struct execsw * execsw_arg)1904 exec_register(const struct execsw *execsw_arg)
1905 {
1906 	const struct execsw **es, **xs, **newexecsw;
1907 	u_int count = 2;	/* New slot and trailing NULL */
1908 
1909 	if (execsw)
1910 		for (es = execsw; *es; es++)
1911 			count++;
1912 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1913 	xs = newexecsw;
1914 	if (execsw)
1915 		for (es = execsw; *es; es++)
1916 			*xs++ = *es;
1917 	*xs++ = execsw_arg;
1918 	*xs = NULL;
1919 	if (execsw)
1920 		free(execsw, M_TEMP);
1921 	execsw = newexecsw;
1922 	return (0);
1923 }
1924 
1925 int
exec_unregister(const struct execsw * execsw_arg)1926 exec_unregister(const struct execsw *execsw_arg)
1927 {
1928 	const struct execsw **es, **xs, **newexecsw;
1929 	int count = 1;
1930 
1931 	if (execsw == NULL)
1932 		panic("unregister with no handlers left?\n");
1933 
1934 	for (es = execsw; *es; es++) {
1935 		if (*es == execsw_arg)
1936 			break;
1937 	}
1938 	if (*es == NULL)
1939 		return (ENOENT);
1940 	for (es = execsw; *es; es++)
1941 		if (*es != execsw_arg)
1942 			count++;
1943 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1944 	xs = newexecsw;
1945 	for (es = execsw; *es; es++)
1946 		if (*es != execsw_arg)
1947 			*xs++ = *es;
1948 	*xs = NULL;
1949 	if (execsw)
1950 		free(execsw, M_TEMP);
1951 	execsw = newexecsw;
1952 	return (0);
1953 }
1954 
1955 /*
1956  * Write out a core segment to the compression stream.
1957  */
1958 static int
compress_chunk(struct coredump_params * cp,char * base,char * buf,size_t len)1959 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1960 {
1961 	size_t chunk_len;
1962 	int error;
1963 
1964 	error = 0;
1965 	while (len > 0) {
1966 		chunk_len = MIN(len, CORE_BUF_SIZE);
1967 
1968 		/*
1969 		 * We can get EFAULT error here.
1970 		 * In that case zero out the current chunk of the segment.
1971 		 */
1972 		error = copyin(base, buf, chunk_len);
1973 		if (error != 0)
1974 			bzero(buf, chunk_len);
1975 		error = compressor_write(cp->comp, buf, chunk_len);
1976 		if (error != 0)
1977 			break;
1978 		base += chunk_len;
1979 		len -= chunk_len;
1980 	}
1981 	return (error);
1982 }
1983 
1984 int
core_write(struct coredump_params * cp,const void * base,size_t len,off_t offset,enum uio_seg seg,size_t * resid)1985 core_write(struct coredump_params *cp, const void *base, size_t len,
1986     off_t offset, enum uio_seg seg, size_t *resid)
1987 {
1988 
1989 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1990 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1991 	    cp->active_cred, cp->file_cred, resid, cp->td));
1992 }
1993 
1994 int
core_output(char * base,size_t len,off_t offset,struct coredump_params * cp,void * tmpbuf)1995 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1996     void *tmpbuf)
1997 {
1998 	vm_map_t map;
1999 	struct mount *mp;
2000 	size_t resid, runlen;
2001 	int error;
2002 	bool success;
2003 
2004 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
2005 	    ("%s: user address %p is not page-aligned", __func__, base));
2006 
2007 	if (cp->comp != NULL)
2008 		return (compress_chunk(cp, base, tmpbuf, len));
2009 
2010 	error = 0;
2011 	map = &cp->td->td_proc->p_vmspace->vm_map;
2012 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2013 		/*
2014 		 * Attempt to page in all virtual pages in the range.  If a
2015 		 * virtual page is not backed by the pager, it is represented as
2016 		 * a hole in the file.  This can occur with zero-filled
2017 		 * anonymous memory or truncated files, for example.
2018 		 */
2019 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2020 			if (core_dump_can_intr && curproc_sigkilled())
2021 				return (EINTR);
2022 			error = vm_fault(map, (uintptr_t)base + runlen,
2023 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2024 			if (runlen == 0)
2025 				success = error == KERN_SUCCESS;
2026 			else if ((error == KERN_SUCCESS) != success)
2027 				break;
2028 		}
2029 
2030 		if (success) {
2031 			error = core_write(cp, base, runlen, offset,
2032 			    UIO_USERSPACE, &resid);
2033 			if (error != 0) {
2034 				if (error != EFAULT)
2035 					break;
2036 
2037 				/*
2038 				 * EFAULT may be returned if the user mapping
2039 				 * could not be accessed, e.g., because a mapped
2040 				 * file has been truncated.  Skip the page if no
2041 				 * progress was made, to protect against a
2042 				 * hypothetical scenario where vm_fault() was
2043 				 * successful but core_write() returns EFAULT
2044 				 * anyway.
2045 				 */
2046 				runlen -= resid;
2047 				if (runlen == 0) {
2048 					success = false;
2049 					runlen = PAGE_SIZE;
2050 				}
2051 			}
2052 		}
2053 		if (!success) {
2054 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2055 			if (error != 0)
2056 				break;
2057 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2058 			error = vn_truncate_locked(cp->vp, offset + runlen,
2059 			    false, cp->td->td_ucred);
2060 			VOP_UNLOCK(cp->vp);
2061 			vn_finished_write(mp);
2062 			if (error != 0)
2063 				break;
2064 		}
2065 	}
2066 	return (error);
2067 }
2068 
2069 /*
2070  * Drain into a core file.
2071  */
2072 int
sbuf_drain_core_output(void * arg,const char * data,int len)2073 sbuf_drain_core_output(void *arg, const char *data, int len)
2074 {
2075 	struct coredump_params *cp;
2076 	struct proc *p;
2077 	int error, locked;
2078 
2079 	cp = arg;
2080 	p = cp->td->td_proc;
2081 
2082 	/*
2083 	 * Some kern_proc out routines that print to this sbuf may
2084 	 * call us with the process lock held. Draining with the
2085 	 * non-sleepable lock held is unsafe. The lock is needed for
2086 	 * those routines when dumping a live process. In our case we
2087 	 * can safely release the lock before draining and acquire
2088 	 * again after.
2089 	 */
2090 	locked = PROC_LOCKED(p);
2091 	if (locked)
2092 		PROC_UNLOCK(p);
2093 	if (cp->comp != NULL)
2094 		error = compressor_write(cp->comp, __DECONST(char *, data),
2095 		    len);
2096 	else
2097 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2098 		    UIO_SYSSPACE, NULL);
2099 	if (locked)
2100 		PROC_LOCK(p);
2101 	if (error != 0)
2102 		return (-error);
2103 	cp->offset += len;
2104 	return (len);
2105 }
2106