1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 1993, David Greenman
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD: stable/12/sys/kern/kern_exec.c 371599 2022-02-10 20:23:58Z git2svn $");
31
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/shm.h>
64 #include <sys/signalvar.h>
65 #include <sys/smp.h>
66 #include <sys/stat.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/timers.h>
72 #include <sys/umtx.h>
73 #include <sys/vnode.h>
74 #include <sys/wait.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/pmap.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_pager.h>
88
89 #ifdef HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92
93 #include <machine/reg.h>
94
95 #include <security/audit/audit.h>
96 #include <security/mac/mac_framework.h>
97
98 #ifdef KDTRACE_HOOKS
99 #include <sys/dtrace_bsd.h>
100 dtrace_execexit_func_t dtrace_fasttrap_exec;
101 #endif
102
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
107
108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
109
110 int coredump_pack_fileinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
112 &coredump_pack_fileinfo, 0,
113 "Enable file path packing in 'procstat -f' coredump notes");
114
115 int coredump_pack_vmmapinfo = 1;
116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
117 &coredump_pack_vmmapinfo, 0,
118 "Enable file path packing in 'procstat -v' coredump notes");
119
120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
123 static int do_execve(struct thread *td, struct image_args *args,
124 struct mac *mac_p, struct vmspace *oldvmspace);
125
126 /* XXX This should be vm_size_t. */
127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
128 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
129
130 /* XXX This should be vm_size_t. */
131 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
132 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
133
134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
135 NULL, 0, sysctl_kern_stackprot, "I", "");
136
137 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
138 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
139 &ps_arg_cache_limit, 0, "");
140
141 static int disallow_high_osrel;
142 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
143 &disallow_high_osrel, 0,
144 "Disallow execution of binaries built for higher version of the world");
145
146 static int map_at_zero = 0;
147 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
148 "Permit processes to map an object at virtual address 0.");
149
150 EVENTHANDLER_LIST_DECLARE(process_exec);
151
152 static int
sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)153 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
154 {
155 struct proc *p;
156 int error;
157
158 p = curproc;
159 #ifdef SCTL_MASK32
160 if (req->flags & SCTL_MASK32) {
161 unsigned int val;
162 val = (unsigned int)p->p_sysent->sv_psstrings;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
164 } else
165 #endif
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
167 sizeof(p->p_sysent->sv_psstrings));
168 return error;
169 }
170
171 static int
sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)172 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
173 {
174 struct proc *p;
175 int error;
176
177 p = curproc;
178 #ifdef SCTL_MASK32
179 if (req->flags & SCTL_MASK32) {
180 unsigned int val;
181 val = (unsigned int)p->p_sysent->sv_usrstack;
182 error = SYSCTL_OUT(req, &val, sizeof(val));
183 } else
184 #endif
185 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
186 sizeof(p->p_sysent->sv_usrstack));
187 return error;
188 }
189
190 static int
sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)191 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
192 {
193 struct proc *p;
194
195 p = curproc;
196 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
197 sizeof(p->p_sysent->sv_stackprot)));
198 }
199
200 /*
201 * Each of the items is a pointer to a `const struct execsw', hence the
202 * double pointer here.
203 */
204 static const struct execsw **execsw;
205
206 #ifndef _SYS_SYSPROTO_H_
207 struct execve_args {
208 char *fname;
209 char **argv;
210 char **envv;
211 };
212 #endif
213
214 int
sys_execve(struct thread * td,struct execve_args * uap)215 sys_execve(struct thread *td, struct execve_args *uap)
216 {
217 struct image_args args;
218 struct vmspace *oldvmspace;
219 int error;
220
221 error = pre_execve(td, &oldvmspace);
222 if (error != 0)
223 return (error);
224 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
225 uap->argv, uap->envv);
226 if (error == 0)
227 error = kern_execve(td, &args, NULL, oldvmspace);
228 post_execve(td, error, oldvmspace);
229 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
230 return (error);
231 }
232
233 #ifndef _SYS_SYSPROTO_H_
234 struct fexecve_args {
235 int fd;
236 char **argv;
237 char **envv;
238 }
239 #endif
240 int
sys_fexecve(struct thread * td,struct fexecve_args * uap)241 sys_fexecve(struct thread *td, struct fexecve_args *uap)
242 {
243 struct image_args args;
244 struct vmspace *oldvmspace;
245 int error;
246
247 error = pre_execve(td, &oldvmspace);
248 if (error != 0)
249 return (error);
250 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
251 uap->argv, uap->envv);
252 if (error == 0) {
253 args.fd = uap->fd;
254 error = kern_execve(td, &args, NULL, oldvmspace);
255 }
256 post_execve(td, error, oldvmspace);
257 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
258 return (error);
259 }
260
261 #ifndef _SYS_SYSPROTO_H_
262 struct __mac_execve_args {
263 char *fname;
264 char **argv;
265 char **envv;
266 struct mac *mac_p;
267 };
268 #endif
269
270 int
sys___mac_execve(struct thread * td,struct __mac_execve_args * uap)271 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
272 {
273 #ifdef MAC
274 struct image_args args;
275 struct vmspace *oldvmspace;
276 int error;
277
278 error = pre_execve(td, &oldvmspace);
279 if (error != 0)
280 return (error);
281 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
282 uap->argv, uap->envv);
283 if (error == 0)
284 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
285 post_execve(td, error, oldvmspace);
286 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
287 return (error);
288 #else
289 return (ENOSYS);
290 #endif
291 }
292
293 int
pre_execve(struct thread * td,struct vmspace ** oldvmspace)294 pre_execve(struct thread *td, struct vmspace **oldvmspace)
295 {
296 struct proc *p;
297 int error;
298
299 KASSERT(td == curthread, ("non-current thread %p", td));
300 error = 0;
301 p = td->td_proc;
302 if ((p->p_flag & P_HADTHREADS) != 0) {
303 PROC_LOCK(p);
304 if (thread_single(p, SINGLE_BOUNDARY) != 0)
305 error = ERESTART;
306 PROC_UNLOCK(p);
307 }
308 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
309 ("nested execve"));
310 *oldvmspace = p->p_vmspace;
311 return (error);
312 }
313
314 void
post_execve(struct thread * td,int error,struct vmspace * oldvmspace)315 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
316 {
317 struct proc *p;
318
319 KASSERT(td == curthread, ("non-current thread %p", td));
320 p = td->td_proc;
321 if ((p->p_flag & P_HADTHREADS) != 0) {
322 PROC_LOCK(p);
323 /*
324 * If success, we upgrade to SINGLE_EXIT state to
325 * force other threads to suicide.
326 */
327 if (error == EJUSTRETURN)
328 thread_single(p, SINGLE_EXIT);
329 else
330 thread_single_end(p, SINGLE_BOUNDARY);
331 PROC_UNLOCK(p);
332 }
333 exec_cleanup(td, oldvmspace);
334 }
335
336 /*
337 * kern_execve() has the astonishing property of not always returning to
338 * the caller. If sufficiently bad things happen during the call to
339 * do_execve(), it can end up calling exit1(); as a result, callers must
340 * avoid doing anything which they might need to undo (e.g., allocating
341 * memory).
342 */
343 int
kern_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
345 struct vmspace *oldvmspace)
346 {
347
348 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
349 args->begin_envv - args->begin_argv);
350 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
351 args->endp - args->begin_envv);
352
353 /* Must have at least one argument. */
354 if (args->argc == 0) {
355 exec_free_args(args);
356 return (EINVAL);
357 }
358 return (do_execve(td, args, mac_p, oldvmspace));
359 }
360
361 /*
362 * In-kernel implementation of execve(). All arguments are assumed to be
363 * userspace pointers from the passed thread.
364 */
365 static int
do_execve(struct thread * td,struct image_args * args,struct mac * mac_p,struct vmspace * oldvmspace)366 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
367 struct vmspace *oldvmspace)
368 {
369 struct proc *p = td->td_proc;
370 struct nameidata nd;
371 struct ucred *oldcred;
372 struct uidinfo *euip = NULL;
373 register_t *stack_base;
374 struct image_params image_params, *imgp;
375 struct vattr attr;
376 int (*img_first)(struct image_params *);
377 struct pargs *oldargs = NULL, *newargs = NULL;
378 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
379 #ifdef KTRACE
380 struct vnode *tracevp = NULL;
381 struct ucred *tracecred = NULL;
382 #endif
383 struct vnode *oldtextvp = NULL, *newtextvp;
384 int credential_changing;
385 #ifdef MAC
386 struct label *interpvplabel = NULL;
387 int will_transition;
388 #endif
389 #ifdef HWPMC_HOOKS
390 struct pmckern_procexec pe;
391 #endif
392 int error, i, orig_osrel;
393 uint32_t orig_fctl0;
394 static const char fexecv_proc_title[] = "(fexecv)";
395
396 imgp = &image_params;
397
398 /*
399 * Lock the process and set the P_INEXEC flag to indicate that
400 * it should be left alone until we're done here. This is
401 * necessary to avoid race conditions - e.g. in ptrace() -
402 * that might allow a local user to illicitly obtain elevated
403 * privileges.
404 */
405 PROC_LOCK(p);
406 KASSERT((p->p_flag & P_INEXEC) == 0,
407 ("%s(): process already has P_INEXEC flag", __func__));
408 p->p_flag |= P_INEXEC;
409 PROC_UNLOCK(p);
410
411 /*
412 * Initialize part of the common data
413 */
414 bzero(imgp, sizeof(*imgp));
415 imgp->proc = p;
416 imgp->attr = &attr;
417 imgp->args = args;
418 oldcred = p->p_ucred;
419 orig_osrel = p->p_osrel;
420 orig_fctl0 = p->p_fctl0;
421
422 #ifdef MAC
423 error = mac_execve_enter(imgp, mac_p);
424 if (error)
425 goto exec_fail;
426 #endif
427
428 /*
429 * Translate the file name. namei() returns a vnode pointer
430 * in ni_vp among other things.
431 *
432 * XXXAUDIT: It would be desirable to also audit the name of the
433 * interpreter if this is an interpreted binary.
434 */
435 if (args->fname != NULL) {
436 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
437 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
438 }
439
440 SDT_PROBE1(proc, , , exec, args->fname);
441
442 interpret:
443 if (args->fname != NULL) {
444 #ifdef CAPABILITY_MODE
445 /*
446 * While capability mode can't reach this point via direct
447 * path arguments to execve(), we also don't allow
448 * interpreters to be used in capability mode (for now).
449 * Catch indirect lookups and return a permissions error.
450 */
451 if (IN_CAPABILITY_MODE(td)) {
452 error = ECAPMODE;
453 goto exec_fail;
454 }
455 #endif
456 error = namei(&nd);
457 if (error)
458 goto exec_fail;
459
460 newtextvp = nd.ni_vp;
461 imgp->vp = newtextvp;
462 } else {
463 AUDIT_ARG_FD(args->fd);
464 /*
465 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
466 */
467 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
468 if (error)
469 goto exec_fail;
470 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
471 AUDIT_ARG_VNODE1(newtextvp);
472 imgp->vp = newtextvp;
473 }
474
475 /*
476 * Check file permissions. Also 'opens' file and sets its vnode to
477 * text mode.
478 */
479 error = exec_check_permissions(imgp);
480 if (error)
481 goto exec_fail_dealloc;
482
483 imgp->object = imgp->vp->v_object;
484 if (imgp->object != NULL)
485 vm_object_reference(imgp->object);
486
487 error = exec_map_first_page(imgp);
488 if (error)
489 goto exec_fail_dealloc;
490
491 imgp->proc->p_osrel = 0;
492 imgp->proc->p_fctl0 = 0;
493
494 /*
495 * Implement image setuid/setgid.
496 *
497 * Determine new credentials before attempting image activators
498 * so that it can be used by process_exec handlers to determine
499 * credential/setid changes.
500 *
501 * Don't honor setuid/setgid if the filesystem prohibits it or if
502 * the process is being traced.
503 *
504 * We disable setuid/setgid/etc in capability mode on the basis
505 * that most setugid applications are not written with that
506 * environment in mind, and will therefore almost certainly operate
507 * incorrectly. In principle there's no reason that setugid
508 * applications might not be useful in capability mode, so we may want
509 * to reconsider this conservative design choice in the future.
510 *
511 * XXXMAC: For the time being, use NOSUID to also prohibit
512 * transitions on the file system.
513 */
514 credential_changing = 0;
515 credential_changing |= (attr.va_mode & S_ISUID) &&
516 oldcred->cr_uid != attr.va_uid;
517 credential_changing |= (attr.va_mode & S_ISGID) &&
518 oldcred->cr_gid != attr.va_gid;
519 #ifdef MAC
520 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
521 interpvplabel, imgp);
522 credential_changing |= will_transition;
523 #endif
524
525 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
526 if (credential_changing)
527 imgp->proc->p_pdeathsig = 0;
528
529 if (credential_changing &&
530 #ifdef CAPABILITY_MODE
531 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
532 #endif
533 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
534 (p->p_flag & P_TRACED) == 0) {
535 imgp->credential_setid = true;
536 VOP_UNLOCK(imgp->vp, 0);
537 imgp->newcred = crdup(oldcred);
538 if (attr.va_mode & S_ISUID) {
539 euip = uifind(attr.va_uid);
540 change_euid(imgp->newcred, euip);
541 }
542 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
543 if (attr.va_mode & S_ISGID)
544 change_egid(imgp->newcred, attr.va_gid);
545 /*
546 * Implement correct POSIX saved-id behavior.
547 *
548 * XXXMAC: Note that the current logic will save the
549 * uid and gid if a MAC domain transition occurs, even
550 * though maybe it shouldn't.
551 */
552 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
553 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
554 } else {
555 /*
556 * Implement correct POSIX saved-id behavior.
557 *
558 * XXX: It's not clear that the existing behavior is
559 * POSIX-compliant. A number of sources indicate that the
560 * saved uid/gid should only be updated if the new ruid is
561 * not equal to the old ruid, or the new euid is not equal
562 * to the old euid and the new euid is not equal to the old
563 * ruid. The FreeBSD code always updates the saved uid/gid.
564 * Also, this code uses the new (replaced) euid and egid as
565 * the source, which may or may not be the right ones to use.
566 */
567 if (oldcred->cr_svuid != oldcred->cr_uid ||
568 oldcred->cr_svgid != oldcred->cr_gid) {
569 VOP_UNLOCK(imgp->vp, 0);
570 imgp->newcred = crdup(oldcred);
571 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
572 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
573 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
574 }
575 }
576 /* The new credentials are installed into the process later. */
577
578 /*
579 * Do the best to calculate the full path to the image file.
580 */
581 if (args->fname != NULL && args->fname[0] == '/')
582 imgp->execpath = args->fname;
583 else {
584 VOP_UNLOCK(imgp->vp, 0);
585 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
586 &imgp->freepath) != 0)
587 imgp->execpath = args->fname;
588 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
589 }
590
591 /*
592 * If the current process has a special image activator it
593 * wants to try first, call it. For example, emulating shell
594 * scripts differently.
595 */
596 error = -1;
597 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
598 error = img_first(imgp);
599
600 /*
601 * Loop through the list of image activators, calling each one.
602 * An activator returns -1 if there is no match, 0 on success,
603 * and an error otherwise.
604 */
605 for (i = 0; error == -1 && execsw[i]; ++i) {
606 if (execsw[i]->ex_imgact == NULL ||
607 execsw[i]->ex_imgact == img_first) {
608 continue;
609 }
610 error = (*execsw[i]->ex_imgact)(imgp);
611 }
612
613 if (error) {
614 if (error == -1)
615 error = ENOEXEC;
616 goto exec_fail_dealloc;
617 }
618
619 /*
620 * Special interpreter operation, cleanup and loop up to try to
621 * activate the interpreter.
622 */
623 if (imgp->interpreted) {
624 exec_unmap_first_page(imgp);
625 /*
626 * The text reference needs to be removed for scripts.
627 * There is a short period before we determine that
628 * something is a script where text reference is active.
629 * The vnode lock is held over this entire period
630 * so nothing should illegitimately be blocked.
631 */
632 MPASS(imgp->textset);
633 VOP_UNSET_TEXT_CHECKED(newtextvp);
634 imgp->textset = false;
635 /* free name buffer and old vnode */
636 if (args->fname != NULL)
637 NDFREE(&nd, NDF_ONLY_PNBUF);
638 #ifdef MAC
639 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
640 #endif
641 if (imgp->opened) {
642 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
643 imgp->opened = 0;
644 }
645 vput(newtextvp);
646 vm_object_deallocate(imgp->object);
647 imgp->object = NULL;
648 imgp->credential_setid = false;
649 if (imgp->newcred != NULL) {
650 crfree(imgp->newcred);
651 imgp->newcred = NULL;
652 }
653 imgp->execpath = NULL;
654 free(imgp->freepath, M_TEMP);
655 imgp->freepath = NULL;
656 /* set new name to that of the interpreter */
657 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME,
658 UIO_SYSSPACE, imgp->interpreter_name, td);
659 args->fname = imgp->interpreter_name;
660 goto interpret;
661 }
662
663 /*
664 * NB: We unlock the vnode here because it is believed that none
665 * of the sv_copyout_strings/sv_fixup operations require the vnode.
666 */
667 VOP_UNLOCK(imgp->vp, 0);
668
669 if (disallow_high_osrel &&
670 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
671 error = ENOEXEC;
672 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
673 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
674 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
675 goto exec_fail_dealloc;
676 }
677
678 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
679 if (SV_PROC_FLAG(p, SV_CAPSICUM))
680 sys_cap_enter(td, NULL);
681
682 /*
683 * Copy out strings (args and env) and initialize stack base
684 */
685 if (p->p_sysent->sv_copyout_strings)
686 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
687 else
688 stack_base = exec_copyout_strings(imgp);
689
690 /*
691 * If custom stack fixup routine present for this process
692 * let it do the stack setup.
693 * Else stuff argument count as first item on stack
694 */
695 if (p->p_sysent->sv_fixup != NULL)
696 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
697 else
698 error = suword(--stack_base, imgp->args->argc) == 0 ?
699 0 : EFAULT;
700 if (error != 0) {
701 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
702 goto exec_fail_dealloc;
703 }
704
705 if (args->fdp != NULL) {
706 /* Install a brand new file descriptor table. */
707 fdinstall_remapped(td, args->fdp);
708 args->fdp = NULL;
709 } else {
710 /*
711 * Keep on using the existing file descriptor table. For
712 * security and other reasons, the file descriptor table
713 * cannot be shared after an exec.
714 */
715 fdunshare(td);
716 /* close files on exec */
717 fdcloseexec(td);
718 }
719
720 /*
721 * Malloc things before we need locks.
722 */
723 i = imgp->args->begin_envv - imgp->args->begin_argv;
724 /* Cache arguments if they fit inside our allowance */
725 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
726 newargs = pargs_alloc(i);
727 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
728 }
729
730 /*
731 * For security and other reasons, signal handlers cannot
732 * be shared after an exec. The new process gets a copy of the old
733 * handlers. In execsigs(), the new process will have its signals
734 * reset.
735 */
736 if (sigacts_shared(p->p_sigacts)) {
737 oldsigacts = p->p_sigacts;
738 newsigacts = sigacts_alloc();
739 sigacts_copy(newsigacts, oldsigacts);
740 }
741
742 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
743
744 PROC_LOCK(p);
745 if (oldsigacts)
746 p->p_sigacts = newsigacts;
747 /* Stop profiling */
748 stopprofclock(p);
749
750 /* reset caught signals */
751 execsigs(p);
752
753 /* name this process - nameiexec(p, ndp) */
754 bzero(p->p_comm, sizeof(p->p_comm));
755 if (args->fname)
756 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
757 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
758 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
759 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
760 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
761 #ifdef KTR
762 sched_clear_tdname(td);
763 #endif
764
765 /*
766 * mark as execed, wakeup the process that vforked (if any) and tell
767 * it that it now has its own resources back
768 */
769 p->p_flag |= P_EXEC;
770 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
771 p->p_flag2 &= ~P2_NOTRACE;
772 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
773 p->p_flag2 &= ~P2_STKGAP_DISABLE;
774 if (p->p_flag & P_PPWAIT) {
775 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
776 cv_broadcast(&p->p_pwait);
777 /* STOPs are no longer ignored, arrange for AST */
778 signotify(td);
779 }
780
781 /*
782 * Implement image setuid/setgid installation.
783 */
784 if (imgp->credential_setid) {
785 /*
786 * Turn off syscall tracing for set-id programs, except for
787 * root. Record any set-id flags first to make sure that
788 * we do not regain any tracing during a possible block.
789 */
790 setsugid(p);
791
792 #ifdef KTRACE
793 if (p->p_tracecred != NULL &&
794 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
795 ktrprocexec(p, &tracecred, &tracevp);
796 #endif
797 /*
798 * Close any file descriptors 0..2 that reference procfs,
799 * then make sure file descriptors 0..2 are in use.
800 *
801 * Both fdsetugidsafety() and fdcheckstd() may call functions
802 * taking sleepable locks, so temporarily drop our locks.
803 */
804 PROC_UNLOCK(p);
805 VOP_UNLOCK(imgp->vp, 0);
806 fdsetugidsafety(td);
807 error = fdcheckstd(td);
808 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
809 if (error != 0)
810 goto exec_fail_dealloc;
811 PROC_LOCK(p);
812 #ifdef MAC
813 if (will_transition) {
814 mac_vnode_execve_transition(oldcred, imgp->newcred,
815 imgp->vp, interpvplabel, imgp);
816 }
817 #endif
818 } else {
819 if (oldcred->cr_uid == oldcred->cr_ruid &&
820 oldcred->cr_gid == oldcred->cr_rgid)
821 p->p_flag &= ~P_SUGID;
822 }
823 /*
824 * Set the new credentials.
825 */
826 if (imgp->newcred != NULL) {
827 proc_set_cred(p, imgp->newcred);
828 crfree(oldcred);
829 oldcred = NULL;
830 }
831
832 /*
833 * Store the vp for use in procfs. This vnode was referenced by namei
834 * or fgetvp_exec.
835 */
836 oldtextvp = p->p_textvp;
837 p->p_textvp = newtextvp;
838
839 #ifdef KDTRACE_HOOKS
840 /*
841 * Tell the DTrace fasttrap provider about the exec if it
842 * has declared an interest.
843 */
844 if (dtrace_fasttrap_exec)
845 dtrace_fasttrap_exec(p);
846 #endif
847
848 /*
849 * Notify others that we exec'd, and clear the P_INEXEC flag
850 * as we're now a bona fide freshly-execed process.
851 */
852 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
853 p->p_flag &= ~P_INEXEC;
854
855 /* clear "fork but no exec" flag, as we _are_ execing */
856 p->p_acflag &= ~AFORK;
857
858 /*
859 * Free any previous argument cache and replace it with
860 * the new argument cache, if any.
861 */
862 oldargs = p->p_args;
863 p->p_args = newargs;
864 newargs = NULL;
865
866 PROC_UNLOCK(p);
867
868 #ifdef HWPMC_HOOKS
869 /*
870 * Check if system-wide sampling is in effect or if the
871 * current process is using PMCs. If so, do exec() time
872 * processing. This processing needs to happen AFTER the
873 * P_INEXEC flag is cleared.
874 */
875 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
876 VOP_UNLOCK(imgp->vp, 0);
877 pe.pm_credentialschanged = credential_changing;
878 pe.pm_entryaddr = imgp->entry_addr;
879
880 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
881 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
882 }
883 #endif
884
885 /* Set values passed into the program in registers. */
886 if (p->p_sysent->sv_setregs)
887 (*p->p_sysent->sv_setregs)(td, imgp,
888 (u_long)(uintptr_t)stack_base);
889 else
890 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
891
892 vfs_mark_atime(imgp->vp, td->td_ucred);
893
894 SDT_PROBE1(proc, , , exec__success, args->fname);
895
896 exec_fail_dealloc:
897 if (error != 0) {
898 p->p_osrel = orig_osrel;
899 p->p_fctl0 = orig_fctl0;
900 }
901
902 if (imgp->firstpage != NULL)
903 exec_unmap_first_page(imgp);
904
905 if (imgp->vp != NULL) {
906 if (args->fname)
907 NDFREE(&nd, NDF_ONLY_PNBUF);
908 if (imgp->opened)
909 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
910 if (imgp->textset)
911 VOP_UNSET_TEXT_CHECKED(imgp->vp);
912 if (error != 0)
913 vput(imgp->vp);
914 else
915 VOP_UNLOCK(imgp->vp, 0);
916 }
917
918 if (imgp->object != NULL)
919 vm_object_deallocate(imgp->object);
920
921 free(imgp->freepath, M_TEMP);
922
923 if (error == 0) {
924 if (p->p_ptevents & PTRACE_EXEC) {
925 PROC_LOCK(p);
926 if (p->p_ptevents & PTRACE_EXEC)
927 td->td_dbgflags |= TDB_EXEC;
928 PROC_UNLOCK(p);
929 }
930
931 /*
932 * Stop the process here if its stop event mask has
933 * the S_EXEC bit set.
934 */
935 STOPEVENT(p, S_EXEC, 0);
936 } else {
937 exec_fail:
938 /* we're done here, clear P_INEXEC */
939 PROC_LOCK(p);
940 p->p_flag &= ~P_INEXEC;
941 PROC_UNLOCK(p);
942
943 SDT_PROBE1(proc, , , exec__failure, error);
944 }
945
946 if (imgp->newcred != NULL && oldcred != NULL)
947 crfree(imgp->newcred);
948
949 #ifdef MAC
950 mac_execve_exit(imgp);
951 mac_execve_interpreter_exit(interpvplabel);
952 #endif
953 exec_free_args(args);
954
955 /*
956 * Handle deferred decrement of ref counts.
957 */
958 if (oldtextvp != NULL)
959 vrele(oldtextvp);
960 #ifdef KTRACE
961 if (tracevp != NULL)
962 vrele(tracevp);
963 if (tracecred != NULL)
964 crfree(tracecred);
965 #endif
966 pargs_drop(oldargs);
967 pargs_drop(newargs);
968 if (oldsigacts != NULL)
969 sigacts_free(oldsigacts);
970 if (euip != NULL)
971 uifree(euip);
972
973 if (error && imgp->vmspace_destroyed) {
974 /* sorry, no more process anymore. exit gracefully */
975 exec_cleanup(td, oldvmspace);
976 exit1(td, 0, SIGABRT);
977 /* NOT REACHED */
978 }
979
980 #ifdef KTRACE
981 if (error == 0)
982 ktrprocctor(p);
983 #endif
984
985 /*
986 * We don't want cpu_set_syscall_retval() to overwrite any of
987 * the register values put in place by exec_setregs().
988 * Implementations of cpu_set_syscall_retval() will leave
989 * registers unmodified when returning EJUSTRETURN.
990 */
991 return (error == 0 ? EJUSTRETURN : error);
992 }
993
994 void
exec_cleanup(struct thread * td,struct vmspace * oldvmspace)995 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
996 {
997 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
998 KASSERT(td->td_proc->p_vmspace != oldvmspace,
999 ("oldvmspace still used"));
1000 vmspace_free(oldvmspace);
1001 td->td_pflags &= ~TDP_EXECVMSPC;
1002 }
1003 }
1004
1005 int
exec_map_first_page(struct image_params * imgp)1006 exec_map_first_page(struct image_params *imgp)
1007 {
1008 int rv, i, after, initial_pagein;
1009 vm_page_t ma[VM_INITIAL_PAGEIN];
1010 vm_object_t object;
1011
1012 if (imgp->firstpage != NULL)
1013 exec_unmap_first_page(imgp);
1014
1015 object = imgp->vp->v_object;
1016 if (object == NULL)
1017 return (EACCES);
1018 VM_OBJECT_WLOCK(object);
1019 #if VM_NRESERVLEVEL > 0
1020 vm_object_color(object, 0);
1021 #endif
1022 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
1023 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
1024 vm_page_xbusy(ma[0]);
1025 if (!vm_pager_has_page(object, 0, NULL, &after)) {
1026 vm_page_lock(ma[0]);
1027 if (!vm_page_wired(ma[0]))
1028 vm_page_free(ma[0]);
1029 else
1030 vm_page_xunbusy_maybelocked(ma[0]);
1031 vm_page_unlock(ma[0]);
1032 VM_OBJECT_WUNLOCK(object);
1033 return (EIO);
1034 }
1035 initial_pagein = min(after, VM_INITIAL_PAGEIN);
1036 KASSERT(initial_pagein <= object->size,
1037 ("%s: initial_pagein %d object->size %ju",
1038 __func__, initial_pagein, (uintmax_t )object->size));
1039 for (i = 1; i < initial_pagein; i++) {
1040 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1041 if (ma[i]->valid)
1042 break;
1043 if (!vm_page_tryxbusy(ma[i]))
1044 break;
1045 } else {
1046 ma[i] = vm_page_alloc(object, i,
1047 VM_ALLOC_NORMAL);
1048 if (ma[i] == NULL)
1049 break;
1050 }
1051 }
1052 initial_pagein = i;
1053 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1054 if (rv != VM_PAGER_OK) {
1055 for (i = 0; i < initial_pagein; i++) {
1056 vm_page_lock(ma[i]);
1057 if (!vm_page_wired(ma[i]))
1058 vm_page_free(ma[i]);
1059 else
1060 vm_page_xunbusy_maybelocked(ma[i]);
1061 vm_page_unlock(ma[i]);
1062 }
1063 VM_OBJECT_WUNLOCK(object);
1064 return (EIO);
1065 }
1066 vm_page_xunbusy(ma[0]);
1067 for (i = 1; i < initial_pagein; i++)
1068 vm_page_readahead_finish(ma[i]);
1069 }
1070 vm_page_lock(ma[0]);
1071 vm_page_hold(ma[0]);
1072 vm_page_activate(ma[0]);
1073 vm_page_unlock(ma[0]);
1074 VM_OBJECT_WUNLOCK(object);
1075
1076 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1077 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1078
1079 return (0);
1080 }
1081
1082 void
exec_unmap_first_page(struct image_params * imgp)1083 exec_unmap_first_page(struct image_params *imgp)
1084 {
1085 vm_page_t m;
1086
1087 if (imgp->firstpage != NULL) {
1088 m = sf_buf_page(imgp->firstpage);
1089 sf_buf_free(imgp->firstpage);
1090 imgp->firstpage = NULL;
1091 vm_page_lock(m);
1092 vm_page_unhold(m);
1093 vm_page_unlock(m);
1094 }
1095 }
1096
1097 /*
1098 * Destroy old address space, and allocate a new stack.
1099 * The new stack is only sgrowsiz large because it is grown
1100 * automatically on a page fault.
1101 */
1102 int
exec_new_vmspace(struct image_params * imgp,struct sysentvec * sv)1103 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1104 {
1105 int error;
1106 struct proc *p = imgp->proc;
1107 struct vmspace *vmspace = p->p_vmspace;
1108 vm_object_t obj;
1109 struct rlimit rlim_stack;
1110 vm_offset_t sv_minuser, stack_addr;
1111 vm_map_t map;
1112 u_long ssiz;
1113
1114 imgp->vmspace_destroyed = 1;
1115 imgp->sysent = sv;
1116
1117 umtx_exec(p);
1118 itimers_exec(p);
1119
1120 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1121
1122 /*
1123 * Blow away entire process VM, if address space not shared,
1124 * otherwise, create a new VM space so that other threads are
1125 * not disrupted
1126 */
1127 map = &vmspace->vm_map;
1128 if (map_at_zero)
1129 sv_minuser = sv->sv_minuser;
1130 else
1131 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1132 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1133 vm_map_max(map) == sv->sv_maxuser &&
1134 cpu_exec_vmspace_reuse(p, map)) {
1135 shmexit(vmspace);
1136 pmap_remove_pages(vmspace_pmap(vmspace));
1137 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1138 /*
1139 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1140 * must be re-evaluated.
1141 */
1142 vm_map_lock(map);
1143 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1144 MAP_ASLR_IGNSTART);
1145 vm_map_unlock(map);
1146 } else {
1147 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1148 if (error)
1149 return (error);
1150 vmspace = p->p_vmspace;
1151 map = &vmspace->vm_map;
1152 }
1153 map->flags |= imgp->map_flags;
1154
1155 /* Map a shared page */
1156 obj = sv->sv_shared_page_obj;
1157 if (obj != NULL) {
1158 vm_object_reference(obj);
1159 error = vm_map_fixed(map, obj, 0,
1160 sv->sv_shared_page_base, sv->sv_shared_page_len,
1161 VM_PROT_READ | VM_PROT_EXECUTE,
1162 VM_PROT_READ | VM_PROT_EXECUTE,
1163 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1164 if (error != KERN_SUCCESS) {
1165 vm_object_deallocate(obj);
1166 return (vm_mmap_to_errno(error));
1167 }
1168 }
1169
1170 /* Allocate a new stack */
1171 if (imgp->stack_sz != 0) {
1172 ssiz = trunc_page(imgp->stack_sz);
1173 PROC_LOCK(p);
1174 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1175 PROC_UNLOCK(p);
1176 if (ssiz > rlim_stack.rlim_max)
1177 ssiz = rlim_stack.rlim_max;
1178 if (ssiz > rlim_stack.rlim_cur) {
1179 rlim_stack.rlim_cur = ssiz;
1180 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1181 }
1182 } else if (sv->sv_maxssiz != NULL) {
1183 ssiz = *sv->sv_maxssiz;
1184 } else {
1185 ssiz = maxssiz;
1186 }
1187 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1188 if (ssiz < imgp->eff_stack_sz)
1189 imgp->eff_stack_sz = ssiz;
1190 stack_addr = sv->sv_usrstack - ssiz;
1191 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1192 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1193 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1194 if (error != KERN_SUCCESS)
1195 return (vm_mmap_to_errno(error));
1196
1197 /*
1198 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1199 * are still used to enforce the stack rlimit on the process stack.
1200 */
1201 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1202 vmspace->vm_maxsaddr = (char *)stack_addr;
1203
1204 return (0);
1205 }
1206
1207 /*
1208 * Copy out argument and environment strings from the old process address
1209 * space into the temporary string buffer.
1210 */
1211 int
exec_copyin_args(struct image_args * args,char * fname,enum uio_seg segflg,char ** argv,char ** envv)1212 exec_copyin_args(struct image_args *args, char *fname,
1213 enum uio_seg segflg, char **argv, char **envv)
1214 {
1215 u_long argp, envp;
1216 int error;
1217 size_t length;
1218
1219 bzero(args, sizeof(*args));
1220 if (argv == NULL)
1221 return (EFAULT);
1222
1223 /*
1224 * Allocate demand-paged memory for the file name, argument, and
1225 * environment strings.
1226 */
1227 error = exec_alloc_args(args);
1228 if (error != 0)
1229 return (error);
1230
1231 /*
1232 * Copy the file name.
1233 */
1234 if (fname != NULL) {
1235 args->fname = args->buf;
1236 error = (segflg == UIO_SYSSPACE) ?
1237 copystr(fname, args->fname, PATH_MAX, &length) :
1238 copyinstr(fname, args->fname, PATH_MAX, &length);
1239 if (error != 0)
1240 goto err_exit;
1241 } else
1242 length = 0;
1243
1244 args->begin_argv = args->buf + length;
1245 args->endp = args->begin_argv;
1246 args->stringspace = ARG_MAX;
1247
1248 /*
1249 * extract arguments first
1250 */
1251 for (;;) {
1252 error = fueword(argv++, &argp);
1253 if (error == -1) {
1254 error = EFAULT;
1255 goto err_exit;
1256 }
1257 if (argp == 0)
1258 break;
1259 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1260 args->stringspace, &length);
1261 if (error != 0) {
1262 if (error == ENAMETOOLONG)
1263 error = E2BIG;
1264 goto err_exit;
1265 }
1266 args->stringspace -= length;
1267 args->endp += length;
1268 args->argc++;
1269 }
1270
1271 args->begin_envv = args->endp;
1272
1273 /*
1274 * extract environment strings
1275 */
1276 if (envv) {
1277 for (;;) {
1278 error = fueword(envv++, &envp);
1279 if (error == -1) {
1280 error = EFAULT;
1281 goto err_exit;
1282 }
1283 if (envp == 0)
1284 break;
1285 error = copyinstr((void *)(uintptr_t)envp,
1286 args->endp, args->stringspace, &length);
1287 if (error != 0) {
1288 if (error == ENAMETOOLONG)
1289 error = E2BIG;
1290 goto err_exit;
1291 }
1292 args->stringspace -= length;
1293 args->endp += length;
1294 args->envc++;
1295 }
1296 }
1297
1298 return (0);
1299
1300 err_exit:
1301 exec_free_args(args);
1302 return (error);
1303 }
1304
1305 int
exec_copyin_data_fds(struct thread * td,struct image_args * args,const void * data,size_t datalen,const int * fds,size_t fdslen)1306 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1307 const void *data, size_t datalen, const int *fds, size_t fdslen)
1308 {
1309 struct filedesc *ofdp;
1310 const char *p;
1311 int *kfds;
1312 int error;
1313
1314 memset(args, '\0', sizeof(*args));
1315 ofdp = td->td_proc->p_fd;
1316 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1317 return (E2BIG);
1318 error = exec_alloc_args(args);
1319 if (error != 0)
1320 return (error);
1321
1322 args->begin_argv = args->buf;
1323 args->stringspace = ARG_MAX;
1324
1325 if (datalen > 0) {
1326 /*
1327 * Argument buffer has been provided. Copy it into the
1328 * kernel as a single string and add a terminating null
1329 * byte.
1330 */
1331 error = copyin(data, args->begin_argv, datalen);
1332 if (error != 0)
1333 goto err_exit;
1334 args->begin_argv[datalen] = '\0';
1335 args->endp = args->begin_argv + datalen + 1;
1336 args->stringspace -= datalen + 1;
1337
1338 /*
1339 * Traditional argument counting. Count the number of
1340 * null bytes.
1341 */
1342 for (p = args->begin_argv; p < args->endp; ++p)
1343 if (*p == '\0')
1344 ++args->argc;
1345 } else {
1346 /* No argument buffer provided. */
1347 args->endp = args->begin_argv;
1348 }
1349 /* There are no environment variables. */
1350 args->begin_envv = args->endp;
1351
1352 /* Create new file descriptor table. */
1353 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1354 error = copyin(fds, kfds, fdslen * sizeof(int));
1355 if (error != 0) {
1356 free(kfds, M_TEMP);
1357 goto err_exit;
1358 }
1359 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1360 free(kfds, M_TEMP);
1361 if (error != 0)
1362 goto err_exit;
1363
1364 return (0);
1365 err_exit:
1366 exec_free_args(args);
1367 return (error);
1368 }
1369
1370 struct exec_args_kva {
1371 vm_offset_t addr;
1372 u_int gen;
1373 SLIST_ENTRY(exec_args_kva) next;
1374 };
1375
1376 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1377
1378 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1379 static struct mtx exec_args_kva_mtx;
1380 static u_int exec_args_gen;
1381
1382 static void
exec_prealloc_args_kva(void * arg __unused)1383 exec_prealloc_args_kva(void *arg __unused)
1384 {
1385 struct exec_args_kva *argkva;
1386 u_int i;
1387
1388 SLIST_INIT(&exec_args_kva_freelist);
1389 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1390 for (i = 0; i < exec_map_entries; i++) {
1391 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1392 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1393 argkva->gen = exec_args_gen;
1394 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1395 }
1396 }
1397 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1398
1399 static vm_offset_t
exec_alloc_args_kva(void ** cookie)1400 exec_alloc_args_kva(void **cookie)
1401 {
1402 struct exec_args_kva *argkva;
1403
1404 argkva = (void *)atomic_readandclear_ptr(
1405 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1406 if (argkva == NULL) {
1407 mtx_lock(&exec_args_kva_mtx);
1408 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1409 (void)mtx_sleep(&exec_args_kva_freelist,
1410 &exec_args_kva_mtx, 0, "execkva", 0);
1411 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1412 mtx_unlock(&exec_args_kva_mtx);
1413 }
1414 *(struct exec_args_kva **)cookie = argkva;
1415 return (argkva->addr);
1416 }
1417
1418 static void
exec_release_args_kva(struct exec_args_kva * argkva,u_int gen)1419 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1420 {
1421 vm_offset_t base;
1422
1423 base = argkva->addr;
1424 if (argkva->gen != gen) {
1425 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1426 MADV_FREE);
1427 argkva->gen = gen;
1428 }
1429 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1430 (uintptr_t)NULL, (uintptr_t)argkva)) {
1431 mtx_lock(&exec_args_kva_mtx);
1432 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1433 wakeup_one(&exec_args_kva_freelist);
1434 mtx_unlock(&exec_args_kva_mtx);
1435 }
1436 }
1437
1438 static void
exec_free_args_kva(void * cookie)1439 exec_free_args_kva(void *cookie)
1440 {
1441
1442 exec_release_args_kva(cookie, exec_args_gen);
1443 }
1444
1445 static void
exec_args_kva_lowmem(void * arg __unused)1446 exec_args_kva_lowmem(void *arg __unused)
1447 {
1448 SLIST_HEAD(, exec_args_kva) head;
1449 struct exec_args_kva *argkva;
1450 u_int gen;
1451 int i;
1452
1453 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1454
1455 /*
1456 * Force an madvise of each KVA range. Any currently allocated ranges
1457 * will have MADV_FREE applied once they are freed.
1458 */
1459 SLIST_INIT(&head);
1460 mtx_lock(&exec_args_kva_mtx);
1461 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1462 mtx_unlock(&exec_args_kva_mtx);
1463 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1464 SLIST_REMOVE_HEAD(&head, next);
1465 exec_release_args_kva(argkva, gen);
1466 }
1467
1468 CPU_FOREACH(i) {
1469 argkva = (void *)atomic_readandclear_ptr(
1470 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1471 if (argkva != NULL)
1472 exec_release_args_kva(argkva, gen);
1473 }
1474 }
1475 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1476 EVENTHANDLER_PRI_ANY);
1477
1478 /*
1479 * Allocate temporary demand-paged, zero-filled memory for the file name,
1480 * argument, and environment strings.
1481 */
1482 int
exec_alloc_args(struct image_args * args)1483 exec_alloc_args(struct image_args *args)
1484 {
1485
1486 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1487 return (0);
1488 }
1489
1490 void
exec_free_args(struct image_args * args)1491 exec_free_args(struct image_args *args)
1492 {
1493
1494 if (args->buf != NULL) {
1495 exec_free_args_kva(args->bufkva);
1496 args->buf = NULL;
1497 }
1498 if (args->fname_buf != NULL) {
1499 free(args->fname_buf, M_TEMP);
1500 args->fname_buf = NULL;
1501 }
1502 if (args->fdp != NULL)
1503 fdescfree_remapped(args->fdp);
1504 }
1505
1506 void
exec_stackgap(struct image_params * imgp,uintptr_t * dp)1507 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1508 {
1509 if (imgp->sysent->sv_stackgap == NULL ||
1510 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1511 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1512 (imgp->map_flags & MAP_ASLR) == 0)
1513 return;
1514 imgp->sysent->sv_stackgap(imgp, (u_long *)dp);
1515 }
1516
1517 /*
1518 * Copy strings out to the new process address space, constructing new arg
1519 * and env vector tables. Return a pointer to the base so that it can be used
1520 * as the initial stack pointer.
1521 */
1522 register_t *
exec_copyout_strings(struct image_params * imgp)1523 exec_copyout_strings(struct image_params *imgp)
1524 {
1525 int argc, envc;
1526 char **vectp;
1527 char *stringp;
1528 uintptr_t destp;
1529 register_t *stack_base;
1530 struct ps_strings *arginfo;
1531 struct proc *p;
1532 size_t execpath_len;
1533 int szsigcode, szps;
1534 char canary[sizeof(long) * 8];
1535
1536 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1537 /*
1538 * Calculate string base and vector table pointers.
1539 * Also deal with signal trampoline code for this exec type.
1540 */
1541 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1542 execpath_len = strlen(imgp->execpath) + 1;
1543 else
1544 execpath_len = 0;
1545 p = imgp->proc;
1546 szsigcode = 0;
1547 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1548 if (p->p_sysent->sv_sigcode_base == 0) {
1549 if (p->p_sysent->sv_szsigcode != NULL)
1550 szsigcode = *(p->p_sysent->sv_szsigcode);
1551 }
1552 destp = (uintptr_t)arginfo;
1553
1554 /*
1555 * install sigcode
1556 */
1557 if (szsigcode != 0) {
1558 destp -= szsigcode;
1559 destp = rounddown2(destp, sizeof(void *));
1560 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1561 }
1562
1563 /*
1564 * Copy the image path for the rtld.
1565 */
1566 if (execpath_len != 0) {
1567 destp -= execpath_len;
1568 destp = rounddown2(destp, sizeof(void *));
1569 imgp->execpathp = destp;
1570 copyout(imgp->execpath, (void *)destp, execpath_len);
1571 }
1572
1573 /*
1574 * Prepare the canary for SSP.
1575 */
1576 arc4rand(canary, sizeof(canary), 0);
1577 destp -= sizeof(canary);
1578 imgp->canary = destp;
1579 copyout(canary, (void *)destp, sizeof(canary));
1580 imgp->canarylen = sizeof(canary);
1581
1582 /*
1583 * Prepare the pagesizes array.
1584 */
1585 destp -= szps;
1586 destp = rounddown2(destp, sizeof(void *));
1587 imgp->pagesizes = destp;
1588 copyout(pagesizes, (void *)destp, szps);
1589 imgp->pagesizeslen = szps;
1590
1591 destp -= ARG_MAX - imgp->args->stringspace;
1592 destp = rounddown2(destp, sizeof(void *));
1593
1594 exec_stackgap(imgp, &destp);
1595 vectp = (char **)destp;
1596
1597 if (imgp->auxargs) {
1598 /*
1599 * Allocate room on the stack for the ELF auxargs
1600 * array. It has up to AT_COUNT entries.
1601 */
1602 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1603 sizeof(*vectp));
1604 }
1605
1606 /*
1607 * Allocate room for the argv[] and env vectors including the
1608 * terminating NULL pointers.
1609 */
1610 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1611
1612 /*
1613 * vectp also becomes our initial stack base
1614 */
1615 stack_base = (register_t *)vectp;
1616
1617 stringp = imgp->args->begin_argv;
1618 argc = imgp->args->argc;
1619 envc = imgp->args->envc;
1620
1621 /*
1622 * Copy out strings - arguments and environment.
1623 */
1624 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1625
1626 /*
1627 * Fill in "ps_strings" struct for ps, w, etc.
1628 */
1629 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1630 suword32(&arginfo->ps_nargvstr, argc);
1631
1632 /*
1633 * Fill in argument portion of vector table.
1634 */
1635 for (; argc > 0; --argc) {
1636 suword(vectp++, (long)(intptr_t)destp);
1637 while (*stringp++ != 0)
1638 destp++;
1639 destp++;
1640 }
1641
1642 /* a null vector table pointer separates the argp's from the envp's */
1643 suword(vectp++, 0);
1644
1645 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1646 suword32(&arginfo->ps_nenvstr, envc);
1647
1648 /*
1649 * Fill in environment portion of vector table.
1650 */
1651 for (; envc > 0; --envc) {
1652 suword(vectp++, (long)(intptr_t)destp);
1653 while (*stringp++ != 0)
1654 destp++;
1655 destp++;
1656 }
1657
1658 /* end of vector table is a null pointer */
1659 suword(vectp, 0);
1660
1661 return (stack_base);
1662 }
1663
1664 /*
1665 * Check permissions of file to execute.
1666 * Called with imgp->vp locked.
1667 * Return 0 for success or error code on failure.
1668 */
1669 int
exec_check_permissions(struct image_params * imgp)1670 exec_check_permissions(struct image_params *imgp)
1671 {
1672 struct vnode *vp = imgp->vp;
1673 struct vattr *attr = imgp->attr;
1674 struct thread *td;
1675 int error;
1676
1677 td = curthread;
1678
1679 /* Get file attributes */
1680 error = VOP_GETATTR(vp, attr, td->td_ucred);
1681 if (error)
1682 return (error);
1683
1684 #ifdef MAC
1685 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1686 if (error)
1687 return (error);
1688 #endif
1689
1690 /*
1691 * 1) Check if file execution is disabled for the filesystem that
1692 * this file resides on.
1693 * 2) Ensure that at least one execute bit is on. Otherwise, a
1694 * privileged user will always succeed, and we don't want this
1695 * to happen unless the file really is executable.
1696 * 3) Ensure that the file is a regular file.
1697 */
1698 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1699 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1700 (attr->va_type != VREG))
1701 return (EACCES);
1702
1703 /*
1704 * Zero length files can't be exec'd
1705 */
1706 if (attr->va_size == 0)
1707 return (ENOEXEC);
1708
1709 /*
1710 * Check for execute permission to file based on current credentials.
1711 */
1712 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1713 if (error)
1714 return (error);
1715
1716 /*
1717 * Check number of open-for-writes on the file and deny execution
1718 * if there are any.
1719 *
1720 * Add a text reference now so no one can write to the
1721 * executable while we're activating it.
1722 *
1723 * Remember if this was set before and unset it in case this is not
1724 * actually an executable image.
1725 */
1726 error = VOP_SET_TEXT(vp);
1727 if (error != 0)
1728 return (error);
1729 imgp->textset = true;
1730
1731 /*
1732 * Call filesystem specific open routine (which does nothing in the
1733 * general case).
1734 */
1735 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1736 if (error == 0)
1737 imgp->opened = 1;
1738 return (error);
1739 }
1740
1741 /*
1742 * Exec handler registration
1743 */
1744 int
exec_register(const struct execsw * execsw_arg)1745 exec_register(const struct execsw *execsw_arg)
1746 {
1747 const struct execsw **es, **xs, **newexecsw;
1748 u_int count = 2; /* New slot and trailing NULL */
1749
1750 if (execsw)
1751 for (es = execsw; *es; es++)
1752 count++;
1753 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1754 xs = newexecsw;
1755 if (execsw)
1756 for (es = execsw; *es; es++)
1757 *xs++ = *es;
1758 *xs++ = execsw_arg;
1759 *xs = NULL;
1760 if (execsw)
1761 free(execsw, M_TEMP);
1762 execsw = newexecsw;
1763 return (0);
1764 }
1765
1766 int
exec_unregister(const struct execsw * execsw_arg)1767 exec_unregister(const struct execsw *execsw_arg)
1768 {
1769 const struct execsw **es, **xs, **newexecsw;
1770 int count = 1;
1771
1772 if (execsw == NULL)
1773 panic("unregister with no handlers left?\n");
1774
1775 for (es = execsw; *es; es++) {
1776 if (*es == execsw_arg)
1777 break;
1778 }
1779 if (*es == NULL)
1780 return (ENOENT);
1781 for (es = execsw; *es; es++)
1782 if (*es != execsw_arg)
1783 count++;
1784 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1785 xs = newexecsw;
1786 for (es = execsw; *es; es++)
1787 if (*es != execsw_arg)
1788 *xs++ = *es;
1789 *xs = NULL;
1790 if (execsw)
1791 free(execsw, M_TEMP);
1792 execsw = newexecsw;
1793 return (0);
1794 }
1795