xref: /freebsd/sys/kern/kern_exec.c (revision df114daef4c48548c3c2b86717612761185ae18f)
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_hwt_hooks.h"
33 #include "opt_ktrace.h"
34 #include "opt_vm.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/acct.h>
39 #include <sys/asan.h>
40 #include <sys/capsicum.h>
41 #include <sys/compressor.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/reg.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sched.h>
62 #include <sys/sdt.h>
63 #include <sys/sf_buf.h>
64 #include <sys/shm.h>
65 #include <sys/signalvar.h>
66 #include <sys/smp.h>
67 #include <sys/stat.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/sysproto.h>
72 #include <sys/timers.h>
73 #include <sys/umtxvar.h>
74 #include <sys/vnode.h>
75 #include <sys/wait.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_kern.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_pager.h>
89 
90 #ifdef	HWPMC_HOOKS
91 #include <sys/pmckern.h>
92 #endif
93 
94 #ifdef HWT_HOOKS
95 #include <dev/hwt/hwt_hook.h>
96 #endif
97 
98 #include <security/audit/audit.h>
99 #include <security/mac/mac_framework.h>
100 
101 #ifdef KDTRACE_HOOKS
102 #include <sys/dtrace_bsd.h>
103 dtrace_execexit_func_t	dtrace_fasttrap_exec;
104 #endif
105 
106 SDT_PROVIDER_DECLARE(proc);
107 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
108 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
109 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
110 
111 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
112 
113 int coredump_pack_fileinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
115     &coredump_pack_fileinfo, 0,
116     "Enable file path packing in 'procstat -f' coredump notes");
117 
118 int coredump_pack_vmmapinfo = 1;
119 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
120     &coredump_pack_vmmapinfo, 0,
121     "Enable file path packing in 'procstat -v' coredump notes");
122 
123 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
124 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
125 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
126 static int do_execve(struct thread *td, struct image_args *args,
127     struct mac *mac_p, struct vmspace *oldvmspace);
128 
129 /* XXX This should be vm_size_t. */
130 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
131     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
132     "Location of process' ps_strings structure");
133 
134 /* XXX This should be vm_size_t. */
135 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
136     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
137     "Top of process stack");
138 
139 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
140     NULL, 0, sysctl_kern_stackprot, "I",
141     "Stack memory permissions");
142 
143 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
144 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
145     &ps_arg_cache_limit, 0,
146     "Process' command line characters cache limit");
147 
148 static int disallow_high_osrel;
149 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
150     &disallow_high_osrel, 0,
151     "Disallow execution of binaries built for higher version of the world");
152 
153 static int map_at_zero = 0;
154 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
155     "Permit processes to map an object at virtual address 0.");
156 
157 static int core_dump_can_intr = 1;
158 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
159     &core_dump_can_intr, 0,
160     "Core dumping interruptible with SIGKILL");
161 
162 static int
sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)163 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
164 {
165 	struct proc *p;
166 	vm_offset_t ps_strings;
167 
168 	p = curproc;
169 #ifdef SCTL_MASK32
170 	if (req->flags & SCTL_MASK32) {
171 		unsigned int val;
172 		val = (unsigned int)PROC_PS_STRINGS(p);
173 		return (SYSCTL_OUT(req, &val, sizeof(val)));
174 	}
175 #endif
176 	ps_strings = PROC_PS_STRINGS(p);
177 	return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
178 }
179 
180 static int
sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)181 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
182 {
183 	struct proc *p;
184 	vm_offset_t val;
185 
186 	p = curproc;
187 #ifdef SCTL_MASK32
188 	if (req->flags & SCTL_MASK32) {
189 		unsigned int val32;
190 
191 		val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
192 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
193 	}
194 #endif
195 	val = round_page(p->p_vmspace->vm_stacktop);
196 	return (SYSCTL_OUT(req, &val, sizeof(val)));
197 }
198 
199 static int
sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)200 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
201 {
202 	struct proc *p;
203 
204 	p = curproc;
205 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
206 	    sizeof(p->p_sysent->sv_stackprot)));
207 }
208 
209 /*
210  * Each of the items is a pointer to a `const struct execsw', hence the
211  * double pointer here.
212  */
213 static const struct execsw **execsw;
214 
215 #ifndef _SYS_SYSPROTO_H_
216 struct execve_args {
217 	char    *fname;
218 	char    **argv;
219 	char    **envv;
220 };
221 #endif
222 
223 int
sys_execve(struct thread * td,struct execve_args * uap)224 sys_execve(struct thread *td, struct execve_args *uap)
225 {
226 	struct image_args args;
227 	struct vmspace *oldvmspace;
228 	int error;
229 
230 	error = pre_execve(td, &oldvmspace);
231 	if (error != 0)
232 		return (error);
233 	error = exec_copyin_args(&args, uap->fname, uap->argv, uap->envv);
234 	if (error == 0)
235 		error = kern_execve(td, &args, NULL, oldvmspace);
236 	post_execve(td, error, oldvmspace);
237 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
238 	return (error);
239 }
240 
241 #ifndef _SYS_SYSPROTO_H_
242 struct fexecve_args {
243 	int	fd;
244 	char	**argv;
245 	char	**envv;
246 };
247 #endif
248 int
sys_fexecve(struct thread * td,struct fexecve_args * uap)249 sys_fexecve(struct thread *td, struct fexecve_args *uap)
250 {
251 	struct image_args args;
252 	struct vmspace *oldvmspace;
253 	int error;
254 
255 	error = pre_execve(td, &oldvmspace);
256 	if (error != 0)
257 		return (error);
258 	error = exec_copyin_args(&args, NULL, uap->argv, uap->envv);
259 	if (error == 0) {
260 		args.fd = uap->fd;
261 		error = kern_execve(td, &args, NULL, oldvmspace);
262 	}
263 	post_execve(td, error, oldvmspace);
264 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
265 	return (error);
266 }
267 
268 #ifndef _SYS_SYSPROTO_H_
269 struct __mac_execve_args {
270 	char	*fname;
271 	char	**argv;
272 	char	**envv;
273 	struct mac	*mac_p;
274 };
275 #endif
276 
277 int
sys___mac_execve(struct thread * td,struct __mac_execve_args * uap)278 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
279 {
280 #ifdef MAC
281 	struct image_args args;
282 	struct vmspace *oldvmspace;
283 	int error;
284 
285 	error = pre_execve(td, &oldvmspace);
286 	if (error != 0)
287 		return (error);
288 	error = exec_copyin_args(&args, uap->fname, 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 #ifdef KTRACE
360 	if (KTRPOINT(td, KTR_ARGS)) {
361 		ktrdata(KTR_ARGS, args->begin_argv,
362 		    exec_args_get_begin_envv(args) - args->begin_argv);
363         }
364 	if (KTRPOINT(td, KTR_ENVS)) {
365 		ktrdata(KTR_ENVS, exec_args_get_begin_envv(args),
366 		    args->endp - exec_args_get_begin_envv(args));
367         }
368 #endif
369 	/* Must have at least one argument. */
370 	if (args->argc == 0) {
371 		exec_free_args(args);
372 		return (EINVAL);
373 	}
374 	return (do_execve(td, args, mac_p, oldvmspace));
375 }
376 
377 static void
execve_nosetid(struct image_params * imgp)378 execve_nosetid(struct image_params *imgp)
379 {
380 	imgp->credential_setid = false;
381 	if (imgp->newcred != NULL) {
382 		crfree(imgp->newcred);
383 		imgp->newcred = NULL;
384 	}
385 }
386 
387 /*
388  * In-kernel implementation of execve().  All arguments are assumed to be
389  * userspace pointers from the passed thread.
390  */
391 static int
do_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)392 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
393     struct vmspace *oldvmspace)
394 {
395 	struct proc *p = td->td_proc;
396 	struct nameidata nd;
397 	struct ucred *oldcred;
398 	struct uidinfo *euip = NULL;
399 	uintptr_t stack_base;
400 	struct image_params image_params, *imgp;
401 	struct vattr attr;
402 	struct pargs *oldargs = NULL, *newargs = NULL;
403 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
404 #ifdef KTRACE
405 	struct ktr_io_params *kiop;
406 #endif
407 	struct vnode *oldtextvp, *newtextvp;
408 	struct vnode *oldtextdvp, *newtextdvp;
409 	char *oldbinname, *newbinname;
410 	bool credential_changing;
411 #ifdef MAC
412 	struct label *interpvplabel = NULL;
413 	bool will_transition;
414 #endif
415 #ifdef HWPMC_HOOKS
416 	struct pmckern_procexec pe;
417 #endif
418 	int error, i, orig_osrel;
419 	uint32_t orig_fctl0;
420 	Elf_Brandinfo *orig_brandinfo;
421 	size_t freepath_size;
422 	static const char fexecv_proc_title[] = "(fexecv)";
423 
424 	imgp = &image_params;
425 	oldtextvp = oldtextdvp = NULL;
426 	newtextvp = newtextdvp = NULL;
427 	newbinname = oldbinname = NULL;
428 #ifdef KTRACE
429 	kiop = NULL;
430 #endif
431 
432 	/*
433 	 * Lock the process and set the P_INEXEC flag to indicate that
434 	 * it should be left alone until we're done here.  This is
435 	 * necessary to avoid race conditions - e.g. in ptrace() -
436 	 * that might allow a local user to illicitly obtain elevated
437 	 * privileges.
438 	 */
439 	PROC_LOCK(p);
440 	KASSERT((p->p_flag & P_INEXEC) == 0,
441 	    ("%s(): process already has P_INEXEC flag", __func__));
442 	p->p_flag |= P_INEXEC;
443 	PROC_UNLOCK(p);
444 
445 	/*
446 	 * Initialize part of the common data
447 	 */
448 	bzero(imgp, sizeof(*imgp));
449 	imgp->proc = p;
450 	imgp->attr = &attr;
451 	imgp->args = args;
452 	oldcred = p->p_ucred;
453 	orig_osrel = p->p_osrel;
454 	orig_fctl0 = p->p_fctl0;
455 	orig_brandinfo = p->p_elf_brandinfo;
456 
457 #ifdef MAC
458 	error = mac_execve_enter(imgp, mac_p);
459 	if (error)
460 		goto exec_fail;
461 #endif
462 
463 	SDT_PROBE1(proc, , , exec, args->fname);
464 
465 interpret:
466 	if (args->fname != NULL) {
467 #ifdef CAPABILITY_MODE
468 		if (CAP_TRACING(td))
469 			ktrcapfail(CAPFAIL_NAMEI, args->fname);
470 		/*
471 		 * While capability mode can't reach this point via direct
472 		 * path arguments to execve(), we also don't allow
473 		 * interpreters to be used in capability mode (for now).
474 		 * Catch indirect lookups and return a permissions error.
475 		 */
476 		if (IN_CAPABILITY_MODE(td)) {
477 			error = ECAPMODE;
478 			goto exec_fail;
479 		}
480 #endif
481 
482 		/*
483 		 * Translate the file name. namei() returns a vnode
484 		 * pointer in ni_vp among other things.
485 		 */
486 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
487 		    AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
488 		    args->fname);
489 
490 		error = namei(&nd);
491 		if (error)
492 			goto exec_fail;
493 
494 		newtextvp = nd.ni_vp;
495 		newtextdvp = nd.ni_dvp;
496 		nd.ni_dvp = NULL;
497 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
498 		    M_WAITOK);
499 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
500 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
501 		imgp->vp = newtextvp;
502 
503 		/*
504 		 * Do the best to calculate the full path to the image file.
505 		 */
506 		if (args->fname[0] == '/') {
507 			imgp->execpath = args->fname;
508 		} else {
509 			VOP_UNLOCK(imgp->vp);
510 			freepath_size = MAXPATHLEN;
511 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
512 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
513 			    &imgp->freepath, &freepath_size) != 0)
514 				imgp->execpath = args->fname;
515 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
516 		}
517 	} else if (imgp->interpreter_vp) {
518 		/*
519 		 * An image activator has already provided an open vnode
520 		 */
521 		newtextvp = imgp->interpreter_vp;
522 		imgp->interpreter_vp = NULL;
523 		if (vn_fullpath(newtextvp, &imgp->execpath,
524 		    &imgp->freepath) != 0)
525 			imgp->execpath = args->fname;
526 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
527 		AUDIT_ARG_VNODE1(newtextvp);
528 		imgp->vp = newtextvp;
529 	} else {
530 		AUDIT_ARG_FD(args->fd);
531 
532 		/*
533 		 * If the descriptors was not opened with O_PATH, then
534 		 * we require that it was opened with O_EXEC or
535 		 * O_RDONLY.  In either case, exec_check_permissions()
536 		 * below checks _current_ file access mode regardless
537 		 * of the permissions additionally checked at the
538 		 * open(2).
539 		 */
540 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
541 		    &newtextvp);
542 		if (error != 0)
543 			goto exec_fail;
544 
545 		if (vn_fullpath(newtextvp, &imgp->execpath,
546 		    &imgp->freepath) != 0)
547 			imgp->execpath = args->fname;
548 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
549 		AUDIT_ARG_VNODE1(newtextvp);
550 		imgp->vp = newtextvp;
551 	}
552 
553 	/*
554 	 * Check file permissions.  Also 'opens' file and sets its vnode to
555 	 * text mode.
556 	 */
557 	error = exec_check_permissions(imgp);
558 	if (error)
559 		goto exec_fail_dealloc;
560 
561 	imgp->object = imgp->vp->v_object;
562 	if (imgp->object != NULL)
563 		vm_object_reference(imgp->object);
564 
565 	error = exec_map_first_page(imgp);
566 	if (error)
567 		goto exec_fail_dealloc;
568 
569 	imgp->proc->p_osrel = 0;
570 	imgp->proc->p_fctl0 = 0;
571 	imgp->proc->p_elf_brandinfo = NULL;
572 
573 	/*
574 	 * Implement image setuid/setgid.
575 	 *
576 	 * Determine new credentials before attempting image activators
577 	 * so that it can be used by process_exec handlers to determine
578 	 * credential/setid changes.
579 	 *
580 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
581 	 * the process is being traced.
582 	 *
583 	 * We disable setuid/setgid/etc in capability mode on the basis
584 	 * that most setugid applications are not written with that
585 	 * environment in mind, and will therefore almost certainly operate
586 	 * incorrectly. In principle there's no reason that setugid
587 	 * applications might not be useful in capability mode, so we may want
588 	 * to reconsider this conservative design choice in the future.
589 	 *
590 	 * XXXMAC: For the time being, use NOSUID to also prohibit
591 	 * transitions on the file system.
592 	 */
593 	credential_changing = false;
594 	credential_changing |= (attr.va_mode & S_ISUID) &&
595 	    oldcred->cr_uid != attr.va_uid;
596 	credential_changing |= (attr.va_mode & S_ISGID) &&
597 	    oldcred->cr_gid != attr.va_gid;
598 #ifdef MAC
599 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
600 	    interpvplabel, imgp) != 0;
601 	credential_changing |= will_transition;
602 #endif
603 
604 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
605 	if (credential_changing)
606 		imgp->proc->p_pdeathsig = 0;
607 
608 	if (credential_changing &&
609 #ifdef CAPABILITY_MODE
610 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
611 #endif
612 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
613 	    (p->p_flag & P_TRACED) == 0) {
614 		imgp->credential_setid = true;
615 		VOP_UNLOCK(imgp->vp);
616 		imgp->newcred = crdup(oldcred);
617 		if (attr.va_mode & S_ISUID) {
618 			euip = uifind(attr.va_uid);
619 			change_euid(imgp->newcred, euip);
620 		}
621 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
622 		if (attr.va_mode & S_ISGID)
623 			change_egid(imgp->newcred, attr.va_gid);
624 		/*
625 		 * Implement correct POSIX saved-id behavior.
626 		 *
627 		 * XXXMAC: Note that the current logic will save the
628 		 * uid and gid if a MAC domain transition occurs, even
629 		 * though maybe it shouldn't.
630 		 */
631 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
632 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
633 	} else {
634 		/*
635 		 * Implement correct POSIX saved-id behavior.
636 		 *
637 		 * XXX: It's not clear that the existing behavior is
638 		 * POSIX-compliant.  A number of sources indicate that the
639 		 * saved uid/gid should only be updated if the new ruid is
640 		 * not equal to the old ruid, or the new euid is not equal
641 		 * to the old euid and the new euid is not equal to the old
642 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
643 		 * Also, this code uses the new (replaced) euid and egid as
644 		 * the source, which may or may not be the right ones to use.
645 		 */
646 		if (oldcred->cr_svuid != oldcred->cr_uid ||
647 		    oldcred->cr_svgid != oldcred->cr_gid) {
648 			VOP_UNLOCK(imgp->vp);
649 			imgp->newcred = crdup(oldcred);
650 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
651 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
652 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
653 		}
654 	}
655 	/* The new credentials are installed into the process later. */
656 
657 	/*
658 	 *	Loop through the list of image activators, calling each one.
659 	 *	An activator returns -1 if there is no match, 0 on success,
660 	 *	and an error otherwise.
661 	 */
662 	error = -1;
663 	for (i = 0; error == -1 && execsw[i]; ++i) {
664 		if (execsw[i]->ex_imgact == NULL)
665 			continue;
666 		error = (*execsw[i]->ex_imgact)(imgp);
667 	}
668 
669 	if (error) {
670 		if (error == -1)
671 			error = ENOEXEC;
672 		goto exec_fail_dealloc;
673 	}
674 
675 	/*
676 	 * Special interpreter operation, cleanup and loop up to try to
677 	 * activate the interpreter.
678 	 */
679 	if (imgp->interpreted) {
680 		exec_unmap_first_page(imgp);
681 		/*
682 		 * The text reference needs to be removed for scripts.
683 		 * There is a short period before we determine that
684 		 * something is a script where text reference is active.
685 		 * The vnode lock is held over this entire period
686 		 * so nothing should illegitimately be blocked.
687 		 */
688 		MPASS(imgp->textset);
689 		VOP_UNSET_TEXT_CHECKED(newtextvp);
690 		imgp->textset = false;
691 		/* free name buffer and old vnode */
692 #ifdef MAC
693 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
694 #endif
695 		if (imgp->opened) {
696 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
697 			imgp->opened = false;
698 		}
699 		vput(newtextvp);
700 		imgp->vp = newtextvp = NULL;
701 		if (args->fname != NULL) {
702 			if (newtextdvp != NULL) {
703 				vrele(newtextdvp);
704 				newtextdvp = NULL;
705 			}
706 			NDFREE_PNBUF(&nd);
707 			free(newbinname, M_PARGS);
708 			newbinname = NULL;
709 		}
710 		vm_object_deallocate(imgp->object);
711 		imgp->object = NULL;
712 		execve_nosetid(imgp);
713 		imgp->execpath = NULL;
714 		free(imgp->freepath, M_TEMP);
715 		imgp->freepath = NULL;
716 		/* set new name to that of the interpreter */
717 		if (imgp->interpreter_vp) {
718 			args->fname = NULL;
719 		} else {
720 			args->fname = imgp->interpreter_name;
721 		}
722 		goto interpret;
723 	}
724 
725 	/*
726 	 * NB: We unlock the vnode here because it is believed that none
727 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
728 	 */
729 	VOP_UNLOCK(imgp->vp);
730 
731 	if (disallow_high_osrel &&
732 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
733 		error = ENOEXEC;
734 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
735 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
736 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
737 		goto exec_fail_dealloc;
738 	}
739 
740 	/*
741 	 * Copy out strings (args and env) and initialize stack base.
742 	 */
743 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
744 	if (error != 0) {
745 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
746 		goto exec_fail_dealloc;
747 	}
748 
749 	/*
750 	 * Stack setup.
751 	 */
752 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
753 	if (error != 0) {
754 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
755 		goto exec_fail_dealloc;
756 	}
757 
758 	/*
759 	 * For security and other reasons, the file descriptor table cannot be
760 	 * shared after an exec.
761 	 */
762 	fdunshare(td);
763 	pdunshare(td);
764 	/* close files on exec */
765 	fdcloseexec(td);
766 
767 	/*
768 	 * Malloc things before we need locks.
769 	 */
770 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
771 	/* Cache arguments if they fit inside our allowance */
772 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
773 		newargs = pargs_alloc(i);
774 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
775 	}
776 
777 	/*
778 	 * For security and other reasons, signal handlers cannot
779 	 * be shared after an exec. The new process gets a copy of the old
780 	 * handlers. In execsigs(), the new process will have its signals
781 	 * reset.
782 	 */
783 	if (sigacts_shared(p->p_sigacts)) {
784 		oldsigacts = p->p_sigacts;
785 		newsigacts = sigacts_alloc();
786 		sigacts_copy(newsigacts, oldsigacts);
787 	}
788 
789 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
790 
791 	PROC_LOCK(p);
792 	if (oldsigacts)
793 		p->p_sigacts = newsigacts;
794 	/* Stop profiling */
795 	stopprofclock(p);
796 
797 	/* reset caught signals */
798 	execsigs(p);
799 
800 	/* name this process - nameiexec(p, ndp) */
801 	bzero(p->p_comm, sizeof(p->p_comm));
802 	if (args->fname)
803 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
804 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
805 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
806 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
807 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
808 #ifdef KTR
809 	sched_clear_tdname(td);
810 #endif
811 
812 	/*
813 	 * mark as execed, wakeup the process that vforked (if any) and tell
814 	 * it that it now has its own resources back
815 	 */
816 	p->p_flag |= P_EXEC;
817 	td->td_pflags2 &= ~TDP2_UEXTERR;
818 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
819 		p->p_flag2 &= ~P2_NOTRACE;
820 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
821 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
822 	p->p_flag2 &= ~(P2_MEMBAR_PRIVE | P2_MEMBAR_PRIVE_SYNCORE |
823 	    P2_MEMBAR_GLOBE);
824 	if (p->p_flag & P_PPWAIT) {
825 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
826 		cv_broadcast(&p->p_pwait);
827 		/* STOPs are no longer ignored, arrange for AST */
828 		signotify(td);
829 	}
830 
831 	if ((imgp->sysent->sv_setid_allowed != NULL &&
832 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
833 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
834 		execve_nosetid(imgp);
835 
836 	/*
837 	 * Implement image setuid/setgid installation.
838 	 */
839 	if (imgp->credential_setid) {
840 		/*
841 		 * Turn off syscall tracing for set-id programs, except for
842 		 * root.  Record any set-id flags first to make sure that
843 		 * we do not regain any tracing during a possible block.
844 		 */
845 		setsugid(p);
846 #ifdef KTRACE
847 		kiop = ktrprocexec(p);
848 #endif
849 		/*
850 		 * Close any file descriptors 0..2 that reference procfs,
851 		 * then make sure file descriptors 0..2 are in use.
852 		 *
853 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
854 		 * taking sleepable locks, so temporarily drop our locks.
855 		 */
856 		PROC_UNLOCK(p);
857 		VOP_UNLOCK(imgp->vp);
858 		fdsetugidsafety(td);
859 		error = fdcheckstd(td);
860 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
861 		if (error != 0)
862 			goto exec_fail_dealloc;
863 		PROC_LOCK(p);
864 #ifdef MAC
865 		if (will_transition) {
866 			mac_vnode_execve_transition(oldcred, imgp->newcred,
867 			    imgp->vp, interpvplabel, imgp);
868 		}
869 #endif
870 	} else {
871 		if (oldcred->cr_uid == oldcred->cr_ruid &&
872 		    oldcred->cr_gid == oldcred->cr_rgid)
873 			p->p_flag &= ~P_SUGID;
874 	}
875 	/*
876 	 * Set the new credentials.
877 	 */
878 	if (imgp->newcred != NULL) {
879 		proc_set_cred(p, imgp->newcred);
880 		crfree(oldcred);
881 		oldcred = NULL;
882 	}
883 
884 	/*
885 	 * Store the vp for use in kern.proc.pathname.  This vnode was
886 	 * referenced by namei() or by fexecve variant of fname handling.
887 	 */
888 	oldtextvp = p->p_textvp;
889 	p->p_textvp = newtextvp;
890 	oldtextdvp = p->p_textdvp;
891 	p->p_textdvp = newtextdvp;
892 	newtextdvp = NULL;
893 	oldbinname = p->p_binname;
894 	p->p_binname = newbinname;
895 	newbinname = NULL;
896 
897 #ifdef KDTRACE_HOOKS
898 	/*
899 	 * Tell the DTrace fasttrap provider about the exec if it
900 	 * has declared an interest.
901 	 */
902 	if (dtrace_fasttrap_exec)
903 		dtrace_fasttrap_exec(p);
904 #endif
905 
906 	/*
907 	 * Notify others that we exec'd, and clear the P_INEXEC flag
908 	 * as we're now a bona fide freshly-execed process.
909 	 */
910 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
911 	p->p_flag &= ~P_INEXEC;
912 
913 	/* clear "fork but no exec" flag, as we _are_ execing */
914 	p->p_acflag &= ~AFORK;
915 
916 	/*
917 	 * Free any previous argument cache and replace it with
918 	 * the new argument cache, if any.
919 	 */
920 	oldargs = p->p_args;
921 	p->p_args = newargs;
922 	newargs = NULL;
923 
924 	PROC_UNLOCK(p);
925 
926 #ifdef	HWPMC_HOOKS
927 	/*
928 	 * Check if system-wide sampling is in effect or if the
929 	 * current process is using PMCs.  If so, do exec() time
930 	 * processing.  This processing needs to happen AFTER the
931 	 * P_INEXEC flag is cleared.
932 	 */
933 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
934 		VOP_UNLOCK(imgp->vp);
935 		pe.pm_credentialschanged = credential_changing;
936 		pe.pm_baseaddr = imgp->reloc_base;
937 		pe.pm_dynaddr = imgp->et_dyn_addr;
938 
939 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
940 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
941 	}
942 #endif
943 
944 #ifdef HWT_HOOKS
945 	if ((td->td_proc->p_flag2 & P2_HWT) != 0) {
946 		struct hwt_record_entry ent;
947 
948 		VOP_UNLOCK(imgp->vp);
949 		ent.fullpath = imgp->execpath;
950 		ent.addr = imgp->et_dyn_addr;
951 		ent.baseaddr = imgp->reloc_base;
952 		ent.record_type = HWT_RECORD_EXECUTABLE;
953 		HWT_CALL_HOOK(td, HWT_EXEC, &ent);
954 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
955 	}
956 #endif
957 
958 	/* Set values passed into the program in registers. */
959 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
960 
961 	VOP_MMAPPED(imgp->vp);
962 
963 	SDT_PROBE1(proc, , , exec__success, args->fname);
964 
965 exec_fail_dealloc:
966 	if (error != 0) {
967 		p->p_osrel = orig_osrel;
968 		p->p_fctl0 = orig_fctl0;
969 		p->p_elf_brandinfo = orig_brandinfo;
970 	}
971 
972 	if (imgp->firstpage != NULL)
973 		exec_unmap_first_page(imgp);
974 
975 	if (imgp->vp != NULL) {
976 		if (imgp->opened)
977 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
978 		if (imgp->textset)
979 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
980 		if (error != 0)
981 			vput(imgp->vp);
982 		else
983 			VOP_UNLOCK(imgp->vp);
984 		if (args->fname != NULL)
985 			NDFREE_PNBUF(&nd);
986 		if (newtextdvp != NULL)
987 			vrele(newtextdvp);
988 		free(newbinname, M_PARGS);
989 	}
990 
991 	if (imgp->object != NULL)
992 		vm_object_deallocate(imgp->object);
993 
994 	free(imgp->freepath, M_TEMP);
995 
996 	if (error == 0) {
997 		if (p->p_ptevents & PTRACE_EXEC) {
998 			PROC_LOCK(p);
999 			if (p->p_ptevents & PTRACE_EXEC)
1000 				td->td_dbgflags |= TDB_EXEC;
1001 			PROC_UNLOCK(p);
1002 		}
1003 	} else {
1004 exec_fail:
1005 		/* we're done here, clear P_INEXEC */
1006 		PROC_LOCK(p);
1007 		p->p_flag &= ~P_INEXEC;
1008 		PROC_UNLOCK(p);
1009 
1010 		SDT_PROBE1(proc, , , exec__failure, error);
1011 	}
1012 
1013 	if (imgp->newcred != NULL && oldcred != NULL)
1014 		crfree(imgp->newcred);
1015 
1016 #ifdef MAC
1017 	mac_execve_exit(imgp);
1018 	mac_execve_interpreter_exit(interpvplabel);
1019 #endif
1020 	exec_free_args(args);
1021 
1022 	/*
1023 	 * Handle deferred decrement of ref counts.
1024 	 */
1025 	if (oldtextvp != NULL)
1026 		vrele(oldtextvp);
1027 	if (oldtextdvp != NULL)
1028 		vrele(oldtextdvp);
1029 	free(oldbinname, M_PARGS);
1030 #ifdef KTRACE
1031 	ktr_io_params_free(kiop);
1032 #endif
1033 	pargs_drop(oldargs);
1034 	pargs_drop(newargs);
1035 	if (oldsigacts != NULL)
1036 		sigacts_free(oldsigacts);
1037 	if (euip != NULL)
1038 		uifree(euip);
1039 
1040 	if (error && imgp->vmspace_destroyed) {
1041 		/* sorry, no more process anymore. exit gracefully */
1042 		exec_cleanup(td, oldvmspace);
1043 		exit1(td, 0, SIGABRT);
1044 		/* NOT REACHED */
1045 	}
1046 
1047 #ifdef KTRACE
1048 	if (error == 0)
1049 		ktrprocctor(p);
1050 #endif
1051 
1052 	/*
1053 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1054 	 * the register values put in place by exec_setregs().
1055 	 * Implementations of cpu_set_syscall_retval() will leave
1056 	 * registers unmodified when returning EJUSTRETURN.
1057 	 */
1058 	return (error == 0 ? EJUSTRETURN : error);
1059 }
1060 
1061 void
exec_cleanup(struct thread * td,struct vmspace * oldvmspace)1062 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1063 {
1064 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1065 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1066 		    ("oldvmspace still used"));
1067 		vmspace_free(oldvmspace);
1068 		td->td_pflags &= ~TDP_EXECVMSPC;
1069 	}
1070 }
1071 
1072 int
exec_map_first_page(struct image_params * imgp)1073 exec_map_first_page(struct image_params *imgp)
1074 {
1075 	vm_object_t object;
1076 	vm_page_t m;
1077 	int error;
1078 
1079 	if (imgp->firstpage != NULL)
1080 		exec_unmap_first_page(imgp);
1081 
1082 	object = imgp->vp->v_object;
1083 	if (object == NULL)
1084 		return (EACCES);
1085 #if VM_NRESERVLEVEL > 0
1086 	if ((object->flags & OBJ_COLORED) == 0) {
1087 		VM_OBJECT_WLOCK(object);
1088 		vm_object_color(object, 0);
1089 		VM_OBJECT_WUNLOCK(object);
1090 	}
1091 #endif
1092 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1093 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1094 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1095 
1096 	if (error != VM_PAGER_OK)
1097 		return (EIO);
1098 	imgp->firstpage = sf_buf_alloc(m, 0);
1099 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1100 
1101 	return (0);
1102 }
1103 
1104 void
exec_unmap_first_page(struct image_params * imgp)1105 exec_unmap_first_page(struct image_params *imgp)
1106 {
1107 	vm_page_t m;
1108 
1109 	if (imgp->firstpage != NULL) {
1110 		m = sf_buf_page(imgp->firstpage);
1111 		sf_buf_free(imgp->firstpage);
1112 		imgp->firstpage = NULL;
1113 		vm_page_unwire(m, PQ_ACTIVE);
1114 	}
1115 }
1116 
1117 void
exec_onexec_old(struct thread * td)1118 exec_onexec_old(struct thread *td)
1119 {
1120 	sigfastblock_clear(td);
1121 	umtx_exec(td->td_proc);
1122 }
1123 
1124 /*
1125  * This is an optimization which removes the unmanaged shared page
1126  * mapping. In combination with pmap_remove_pages(), which cleans all
1127  * managed mappings in the process' vmspace pmap, no work will be left
1128  * for pmap_remove(min, max).
1129  */
1130 void
exec_free_abi_mappings(struct proc * p)1131 exec_free_abi_mappings(struct proc *p)
1132 {
1133 	struct vmspace *vmspace;
1134 
1135 	vmspace = p->p_vmspace;
1136 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1137 		return;
1138 
1139 	if (!PROC_HAS_SHP(p))
1140 		return;
1141 
1142 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1143 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1144 }
1145 
1146 /*
1147  * Run down the current address space and install a new one.
1148  */
1149 int
exec_new_vmspace(struct image_params * imgp,struct sysentvec * sv)1150 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1151 {
1152 	int error;
1153 	struct proc *p = imgp->proc;
1154 	struct vmspace *vmspace = p->p_vmspace;
1155 	struct thread *td = curthread;
1156 	vm_offset_t sv_minuser;
1157 	vm_map_t map;
1158 
1159 	imgp->vmspace_destroyed = true;
1160 	imgp->sysent = sv;
1161 
1162 	if (p->p_sysent->sv_onexec_old != NULL)
1163 		p->p_sysent->sv_onexec_old(td);
1164 	itimers_exec(p);
1165 
1166 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1167 
1168 	/*
1169 	 * Blow away entire process VM, if address space not shared,
1170 	 * otherwise, create a new VM space so that other threads are
1171 	 * not disrupted
1172 	 */
1173 	map = &vmspace->vm_map;
1174 	if (map_at_zero)
1175 		sv_minuser = sv->sv_minuser;
1176 	else
1177 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1178 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1179 	    vm_map_min(map) == sv_minuser &&
1180 	    vm_map_max(map) == sv->sv_maxuser &&
1181 	    cpu_exec_vmspace_reuse(p, map)) {
1182 		exec_free_abi_mappings(p);
1183 		shmexit(vmspace);
1184 		pmap_remove_pages(vmspace_pmap(vmspace));
1185 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1186 		/*
1187 		 * An exec terminates mlockall(MCL_FUTURE).
1188 		 * ASLR and W^X states must be re-evaluated.
1189 		 */
1190 		vm_map_lock(map);
1191 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1192 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1193 		vm_map_unlock(map);
1194 	} else {
1195 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1196 		if (error)
1197 			return (error);
1198 		vmspace = p->p_vmspace;
1199 		map = &vmspace->vm_map;
1200 	}
1201 	map->flags |= imgp->map_flags;
1202 
1203 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1204 }
1205 
1206 /*
1207  * Compute the stack size limit and map the main process stack.
1208  * Map the shared page.
1209  */
1210 int
exec_map_stack(struct image_params * imgp)1211 exec_map_stack(struct image_params *imgp)
1212 {
1213 	struct rlimit rlim_stack;
1214 	struct sysentvec *sv;
1215 	struct proc *p;
1216 	vm_map_t map;
1217 	struct vmspace *vmspace;
1218 	vm_offset_t stack_addr, stack_top;
1219 	vm_offset_t sharedpage_addr;
1220 	u_long ssiz;
1221 	int error, find_space, stack_off;
1222 	vm_prot_t stack_prot;
1223 	vm_object_t obj;
1224 
1225 	p = imgp->proc;
1226 	sv = p->p_sysent;
1227 
1228 	if (imgp->stack_sz != 0) {
1229 		ssiz = trunc_page(imgp->stack_sz);
1230 		PROC_LOCK(p);
1231 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1232 		PROC_UNLOCK(p);
1233 		if (ssiz > rlim_stack.rlim_max)
1234 			ssiz = rlim_stack.rlim_max;
1235 		if (ssiz > rlim_stack.rlim_cur) {
1236 			rlim_stack.rlim_cur = ssiz;
1237 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1238 		}
1239 	} else if (sv->sv_maxssiz != NULL) {
1240 		ssiz = *sv->sv_maxssiz;
1241 	} else {
1242 		ssiz = maxssiz;
1243 	}
1244 
1245 	vmspace = p->p_vmspace;
1246 	map = &vmspace->vm_map;
1247 
1248 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1249 	    imgp->stack_prot : sv->sv_stackprot;
1250 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1251 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1252 		    lim_max(curthread, RLIMIT_DATA));
1253 		find_space = VMFS_ANY_SPACE;
1254 	} else {
1255 		stack_addr = sv->sv_usrstack - ssiz;
1256 		find_space = VMFS_NO_SPACE;
1257 	}
1258 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1259 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1260 	    MAP_STACK_AREA);
1261 	if (error != KERN_SUCCESS) {
1262 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1263 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1264 		    stack_prot, error, vm_mmap_to_errno(error));
1265 		return (vm_mmap_to_errno(error));
1266 	}
1267 
1268 	stack_top = stack_addr + ssiz;
1269 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1270 		/* Randomize within the first page of the stack. */
1271 		arc4rand(&stack_off, sizeof(stack_off), 0);
1272 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1273 	}
1274 
1275 	/* Map a shared page */
1276 	obj = sv->sv_shared_page_obj;
1277 	if (obj == NULL) {
1278 		sharedpage_addr = 0;
1279 		goto out;
1280 	}
1281 
1282 	/*
1283 	 * If randomization is disabled then the shared page will
1284 	 * be mapped at address specified in sysentvec.
1285 	 * Otherwise any address above .data section can be selected.
1286 	 * Same logic is used for stack address randomization.
1287 	 * If the address randomization is applied map a guard page
1288 	 * at the top of UVA.
1289 	 */
1290 	vm_object_reference(obj);
1291 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1292 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1293 		    lim_max(curthread, RLIMIT_DATA));
1294 
1295 		error = vm_map_fixed(map, NULL, 0,
1296 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1297 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1298 		if (error != KERN_SUCCESS) {
1299 			/*
1300 			 * This is not fatal, so let's just print a warning
1301 			 * and continue.
1302 			 */
1303 			uprintf("%s: Mapping guard page at the top of UVA failed"
1304 			    " mach error %d errno %d",
1305 			    __func__, error, vm_mmap_to_errno(error));
1306 		}
1307 
1308 		error = vm_map_find(map, obj, 0,
1309 		    &sharedpage_addr, sv->sv_shared_page_len,
1310 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1311 		    VM_PROT_READ | VM_PROT_EXECUTE,
1312 		    VM_PROT_READ | VM_PROT_EXECUTE,
1313 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1314 	} else {
1315 		sharedpage_addr = sv->sv_shared_page_base;
1316 		vm_map_fixed(map, obj, 0,
1317 		    sharedpage_addr, sv->sv_shared_page_len,
1318 		    VM_PROT_READ | VM_PROT_EXECUTE,
1319 		    VM_PROT_READ | VM_PROT_EXECUTE,
1320 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1321 	}
1322 	if (error != KERN_SUCCESS) {
1323 		uprintf("%s: mapping shared page at addr: %p"
1324 		    "failed, mach error %d errno %d\n", __func__,
1325 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1326 		vm_object_deallocate(obj);
1327 		return (vm_mmap_to_errno(error));
1328 	}
1329 out:
1330 	/*
1331 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1332 	 * are still used to enforce the stack rlimit on the process stack.
1333 	 */
1334 	vmspace->vm_maxsaddr = (char *)stack_addr;
1335 	vmspace->vm_stacktop = stack_top;
1336 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1337 	vmspace->vm_shp_base = sharedpage_addr;
1338 
1339 	return (0);
1340 }
1341 
1342 /*
1343  * Copy out argument and environment strings from the old process address
1344  * space into the temporary string buffer.
1345  */
1346 int
exec_copyin_args(struct image_args * args,const char * fname,char ** argv,char ** envv)1347 exec_copyin_args(struct image_args *args, const char *fname,
1348     char **argv, char **envv)
1349 {
1350 	u_long arg, env;
1351 	int error;
1352 
1353 	bzero(args, sizeof(*args));
1354 	if (argv == NULL)
1355 		return (EFAULT);
1356 
1357 	/*
1358 	 * Allocate demand-paged memory for the file name, argument, and
1359 	 * environment strings.
1360 	 */
1361 	error = exec_alloc_args(args);
1362 	if (error != 0)
1363 		return (error);
1364 
1365 	/*
1366 	 * Copy the file name.
1367 	 */
1368 	error = exec_args_add_fname(args, fname, UIO_USERSPACE);
1369 	if (error != 0)
1370 		goto err_exit;
1371 
1372 	/*
1373 	 * extract arguments first
1374 	 */
1375 	for (;;) {
1376 		error = fueword(argv++, &arg);
1377 		if (error == -1) {
1378 			error = EFAULT;
1379 			goto err_exit;
1380 		}
1381 		if (arg == 0)
1382 			break;
1383 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1384 		    UIO_USERSPACE);
1385 		if (error != 0)
1386 			goto err_exit;
1387 	}
1388 
1389 	/*
1390 	 * extract environment strings
1391 	 */
1392 	if (envv) {
1393 		for (;;) {
1394 			error = fueword(envv++, &env);
1395 			if (error == -1) {
1396 				error = EFAULT;
1397 				goto err_exit;
1398 			}
1399 			if (env == 0)
1400 				break;
1401 			error = exec_args_add_env(args,
1402 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1403 			if (error != 0)
1404 				goto err_exit;
1405 		}
1406 	}
1407 
1408 	return (0);
1409 
1410 err_exit:
1411 	exec_free_args(args);
1412 	return (error);
1413 }
1414 
1415 struct exec_args_kva {
1416 	vm_offset_t addr;
1417 	u_int gen;
1418 	SLIST_ENTRY(exec_args_kva) next;
1419 };
1420 
1421 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1422 
1423 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1424 static struct mtx exec_args_kva_mtx;
1425 static u_int exec_args_gen;
1426 
1427 static void
exec_prealloc_args_kva(void * arg __unused)1428 exec_prealloc_args_kva(void *arg __unused)
1429 {
1430 	struct exec_args_kva *argkva;
1431 	u_int i;
1432 
1433 	SLIST_INIT(&exec_args_kva_freelist);
1434 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1435 	for (i = 0; i < exec_map_entries; i++) {
1436 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1437 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1438 		argkva->gen = exec_args_gen;
1439 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1440 	}
1441 }
1442 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1443 
1444 static vm_offset_t
exec_alloc_args_kva(void ** cookie)1445 exec_alloc_args_kva(void **cookie)
1446 {
1447 	struct exec_args_kva *argkva;
1448 
1449 	argkva = (void *)atomic_readandclear_ptr(
1450 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1451 	if (argkva == NULL) {
1452 		mtx_lock(&exec_args_kva_mtx);
1453 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1454 			(void)mtx_sleep(&exec_args_kva_freelist,
1455 			    &exec_args_kva_mtx, 0, "execkva", 0);
1456 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1457 		mtx_unlock(&exec_args_kva_mtx);
1458 	}
1459 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1460 	    exec_map_entry_size, 0);
1461 	*(struct exec_args_kva **)cookie = argkva;
1462 	return (argkva->addr);
1463 }
1464 
1465 static void
exec_release_args_kva(struct exec_args_kva * argkva,u_int gen)1466 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1467 {
1468 	vm_offset_t base;
1469 
1470 	base = argkva->addr;
1471 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1472 	    KASAN_EXEC_ARGS_FREED);
1473 	if (argkva->gen != gen) {
1474 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1475 		    MADV_FREE);
1476 		argkva->gen = gen;
1477 	}
1478 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1479 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1480 		mtx_lock(&exec_args_kva_mtx);
1481 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1482 		wakeup_one(&exec_args_kva_freelist);
1483 		mtx_unlock(&exec_args_kva_mtx);
1484 	}
1485 }
1486 
1487 static void
exec_free_args_kva(void * cookie)1488 exec_free_args_kva(void *cookie)
1489 {
1490 
1491 	exec_release_args_kva(cookie, exec_args_gen);
1492 }
1493 
1494 static void
exec_args_kva_lowmem(void * arg __unused,int flags __unused)1495 exec_args_kva_lowmem(void *arg __unused, int flags __unused)
1496 {
1497 	SLIST_HEAD(, exec_args_kva) head;
1498 	struct exec_args_kva *argkva;
1499 	u_int gen;
1500 	int i;
1501 
1502 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1503 
1504 	/*
1505 	 * Force an madvise of each KVA range. Any currently allocated ranges
1506 	 * will have MADV_FREE applied once they are freed.
1507 	 */
1508 	SLIST_INIT(&head);
1509 	mtx_lock(&exec_args_kva_mtx);
1510 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1511 	mtx_unlock(&exec_args_kva_mtx);
1512 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1513 		SLIST_REMOVE_HEAD(&head, next);
1514 		exec_release_args_kva(argkva, gen);
1515 	}
1516 
1517 	CPU_FOREACH(i) {
1518 		argkva = (void *)atomic_readandclear_ptr(
1519 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1520 		if (argkva != NULL)
1521 			exec_release_args_kva(argkva, gen);
1522 	}
1523 }
1524 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1525     EVENTHANDLER_PRI_ANY);
1526 
1527 /*
1528  * Allocate temporary demand-paged, zero-filled memory for the file name,
1529  * argument, and environment strings.
1530  */
1531 int
exec_alloc_args(struct image_args * args)1532 exec_alloc_args(struct image_args *args)
1533 {
1534 
1535 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1536 	return (0);
1537 }
1538 
1539 void
exec_free_args(struct image_args * args)1540 exec_free_args(struct image_args *args)
1541 {
1542 
1543 	if (args->buf != NULL) {
1544 		exec_free_args_kva(args->bufkva);
1545 		args->buf = NULL;
1546 	}
1547 	if (args->fname_buf != NULL) {
1548 		free(args->fname_buf, M_TEMP);
1549 		args->fname_buf = NULL;
1550 	}
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_shared_page_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 
1972 /*
1973  * Write out a core segment to the compression stream.
1974  */
1975 static int
compress_chunk(struct coredump_params * cp,char * base,char * buf,size_t len)1976 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1977 {
1978 	size_t chunk_len;
1979 	int error;
1980 
1981 	error = 0;
1982 	while (len > 0) {
1983 		chunk_len = MIN(len, CORE_BUF_SIZE);
1984 
1985 		/*
1986 		 * We can get EFAULT error here.
1987 		 * In that case zero out the current chunk of the segment.
1988 		 */
1989 		error = copyin(base, buf, chunk_len);
1990 		if (error != 0)
1991 			bzero(buf, chunk_len);
1992 		error = compressor_write(cp->comp, buf, chunk_len);
1993 		if (error != 0)
1994 			break;
1995 		base += chunk_len;
1996 		len -= chunk_len;
1997 	}
1998 	return (error);
1999 }
2000 
2001 int
core_write(struct coredump_params * cp,const void * base,size_t len,off_t offset,enum uio_seg seg,size_t * resid)2002 core_write(struct coredump_params *cp, const void *base, size_t len,
2003     off_t offset, enum uio_seg seg, size_t *resid)
2004 {
2005 
2006 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
2007 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
2008 	    cp->active_cred, cp->file_cred, resid, cp->td));
2009 }
2010 
2011 int
core_output(char * base,size_t len,off_t offset,struct coredump_params * cp,void * tmpbuf)2012 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
2013     void *tmpbuf)
2014 {
2015 	vm_map_t map;
2016 	struct mount *mp;
2017 	size_t resid, runlen;
2018 	int error;
2019 	bool success;
2020 
2021 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
2022 	    ("%s: user address %p is not page-aligned", __func__, base));
2023 
2024 	if (cp->comp != NULL)
2025 		return (compress_chunk(cp, base, tmpbuf, len));
2026 
2027 	error = 0;
2028 	map = &cp->td->td_proc->p_vmspace->vm_map;
2029 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2030 		/*
2031 		 * Attempt to page in all virtual pages in the range.  If a
2032 		 * virtual page is not backed by the pager, it is represented as
2033 		 * a hole in the file.  This can occur with zero-filled
2034 		 * anonymous memory or truncated files, for example.
2035 		 */
2036 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2037 			if (core_dump_can_intr && curproc_sigkilled())
2038 				return (EINTR);
2039 			error = vm_fault(map, (uintptr_t)base + runlen,
2040 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2041 			if (runlen == 0)
2042 				success = error == KERN_SUCCESS;
2043 			else if ((error == KERN_SUCCESS) != success)
2044 				break;
2045 		}
2046 
2047 		if (success) {
2048 			error = core_write(cp, base, runlen, offset,
2049 			    UIO_USERSPACE, &resid);
2050 			if (error != 0) {
2051 				if (error != EFAULT)
2052 					break;
2053 
2054 				/*
2055 				 * EFAULT may be returned if the user mapping
2056 				 * could not be accessed, e.g., because a mapped
2057 				 * file has been truncated.  Skip the page if no
2058 				 * progress was made, to protect against a
2059 				 * hypothetical scenario where vm_fault() was
2060 				 * successful but core_write() returns EFAULT
2061 				 * anyway.
2062 				 */
2063 				runlen -= resid;
2064 				if (runlen == 0) {
2065 					success = false;
2066 					runlen = PAGE_SIZE;
2067 				}
2068 			}
2069 		}
2070 		if (!success) {
2071 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2072 			if (error != 0)
2073 				break;
2074 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2075 			error = vn_truncate_locked(cp->vp, offset + runlen,
2076 			    false, cp->td->td_ucred);
2077 			VOP_UNLOCK(cp->vp);
2078 			vn_finished_write(mp);
2079 			if (error != 0)
2080 				break;
2081 		}
2082 	}
2083 	return (error);
2084 }
2085 
2086 /*
2087  * Drain into a core file.
2088  */
2089 int
sbuf_drain_core_output(void * arg,const char * data,int len)2090 sbuf_drain_core_output(void *arg, const char *data, int len)
2091 {
2092 	struct coredump_params *cp;
2093 	struct proc *p;
2094 	int error, locked;
2095 
2096 	cp = arg;
2097 	p = cp->td->td_proc;
2098 
2099 	/*
2100 	 * Some kern_proc out routines that print to this sbuf may
2101 	 * call us with the process lock held. Draining with the
2102 	 * non-sleepable lock held is unsafe. The lock is needed for
2103 	 * those routines when dumping a live process. In our case we
2104 	 * can safely release the lock before draining and acquire
2105 	 * again after.
2106 	 */
2107 	locked = PROC_LOCKED(p);
2108 	if (locked)
2109 		PROC_UNLOCK(p);
2110 	if (cp->comp != NULL)
2111 		error = compressor_write(cp->comp, __DECONST(char *, data),
2112 		    len);
2113 	else
2114 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2115 		    UIO_SYSSPACE, NULL);
2116 	if (locked)
2117 		PROC_LOCK(p);
2118 	if (error != 0)
2119 		return (-error);
2120 	cp->offset += len;
2121 	return (len);
2122 }
2123