FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_exec.c
1 /*-
2 * Copyright (c) 1993, David Greenman
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD: releng/11.0/sys/kern/kern_exec.c 303606 2016-08-01 06:34:55Z kib $");
29
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
33 #include "opt_vm.h"
34
35 #include <sys/param.h>
36 #include <sys/capsicum.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/wait.h>
52 #include <sys/malloc.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
64 #include <sys/shm.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/stat.h>
68 #ifdef KTRACE
69 #include <sys/ktrace.h>
70 #endif
71
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
81
82 #ifdef HWPMC_HOOKS
83 #include <sys/pmckern.h>
84 #endif
85
86 #include <machine/reg.h>
87
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
90
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t dtrace_fasttrap_exec;
94 #endif
95
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
98 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
99 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
100
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102
103 int coredump_pack_fileinfo = 1;
104 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
105 &coredump_pack_fileinfo, 0,
106 "Enable file path packing in 'procstat -f' coredump notes");
107
108 int coredump_pack_vmmapinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
110 &coredump_pack_vmmapinfo, 0,
111 "Enable file path packing in 'procstat -v' coredump notes");
112
113 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
114 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
115 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
116 static int do_execve(struct thread *td, struct image_args *args,
117 struct mac *mac_p);
118
119 /* XXX This should be vm_size_t. */
120 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
121 NULL, 0, sysctl_kern_ps_strings, "LU", "");
122
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
125 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
126
127 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
128 NULL, 0, sysctl_kern_stackprot, "I", "");
129
130 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
131 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
132 &ps_arg_cache_limit, 0, "");
133
134 static int disallow_high_osrel;
135 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
136 &disallow_high_osrel, 0,
137 "Disallow execution of binaries built for higher version of the world");
138
139 static int map_at_zero = 0;
140 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
141 "Permit processes to map an object at virtual address 0.");
142
143 static int
144 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
145 {
146 struct proc *p;
147 int error;
148
149 p = curproc;
150 #ifdef SCTL_MASK32
151 if (req->flags & SCTL_MASK32) {
152 unsigned int val;
153 val = (unsigned int)p->p_sysent->sv_psstrings;
154 error = SYSCTL_OUT(req, &val, sizeof(val));
155 } else
156 #endif
157 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
158 sizeof(p->p_sysent->sv_psstrings));
159 return error;
160 }
161
162 static int
163 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
164 {
165 struct proc *p;
166 int error;
167
168 p = curproc;
169 #ifdef SCTL_MASK32
170 if (req->flags & SCTL_MASK32) {
171 unsigned int val;
172 val = (unsigned int)p->p_sysent->sv_usrstack;
173 error = SYSCTL_OUT(req, &val, sizeof(val));
174 } else
175 #endif
176 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
177 sizeof(p->p_sysent->sv_usrstack));
178 return error;
179 }
180
181 static int
182 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
183 {
184 struct proc *p;
185
186 p = curproc;
187 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
188 sizeof(p->p_sysent->sv_stackprot)));
189 }
190
191 /*
192 * Each of the items is a pointer to a `const struct execsw', hence the
193 * double pointer here.
194 */
195 static const struct execsw **execsw;
196
197 #ifndef _SYS_SYSPROTO_H_
198 struct execve_args {
199 char *fname;
200 char **argv;
201 char **envv;
202 };
203 #endif
204
205 int
206 sys_execve(struct thread *td, struct execve_args *uap)
207 {
208 struct image_args args;
209 struct vmspace *oldvmspace;
210 int error;
211
212 error = pre_execve(td, &oldvmspace);
213 if (error != 0)
214 return (error);
215 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
216 uap->argv, uap->envv);
217 if (error == 0)
218 error = kern_execve(td, &args, NULL);
219 post_execve(td, error, oldvmspace);
220 return (error);
221 }
222
223 #ifndef _SYS_SYSPROTO_H_
224 struct fexecve_args {
225 int fd;
226 char **argv;
227 char **envv;
228 }
229 #endif
230 int
231 sys_fexecve(struct thread *td, struct fexecve_args *uap)
232 {
233 struct image_args args;
234 struct vmspace *oldvmspace;
235 int error;
236
237 error = pre_execve(td, &oldvmspace);
238 if (error != 0)
239 return (error);
240 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
241 uap->argv, uap->envv);
242 if (error == 0) {
243 args.fd = uap->fd;
244 error = kern_execve(td, &args, NULL);
245 }
246 post_execve(td, error, oldvmspace);
247 return (error);
248 }
249
250 #ifndef _SYS_SYSPROTO_H_
251 struct __mac_execve_args {
252 char *fname;
253 char **argv;
254 char **envv;
255 struct mac *mac_p;
256 };
257 #endif
258
259 int
260 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
261 {
262 #ifdef MAC
263 struct image_args args;
264 struct vmspace *oldvmspace;
265 int error;
266
267 error = pre_execve(td, &oldvmspace);
268 if (error != 0)
269 return (error);
270 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
271 uap->argv, uap->envv);
272 if (error == 0)
273 error = kern_execve(td, &args, uap->mac_p);
274 post_execve(td, error, oldvmspace);
275 return (error);
276 #else
277 return (ENOSYS);
278 #endif
279 }
280
281 int
282 pre_execve(struct thread *td, struct vmspace **oldvmspace)
283 {
284 struct proc *p;
285 int error;
286
287 KASSERT(td == curthread, ("non-current thread %p", td));
288 error = 0;
289 p = td->td_proc;
290 if ((p->p_flag & P_HADTHREADS) != 0) {
291 PROC_LOCK(p);
292 if (thread_single(p, SINGLE_BOUNDARY) != 0)
293 error = ERESTART;
294 PROC_UNLOCK(p);
295 }
296 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
297 ("nested execve"));
298 *oldvmspace = p->p_vmspace;
299 return (error);
300 }
301
302 void
303 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
304 {
305 struct proc *p;
306
307 KASSERT(td == curthread, ("non-current thread %p", td));
308 p = td->td_proc;
309 if ((p->p_flag & P_HADTHREADS) != 0) {
310 PROC_LOCK(p);
311 /*
312 * If success, we upgrade to SINGLE_EXIT state to
313 * force other threads to suicide.
314 */
315 if (error == 0)
316 thread_single(p, SINGLE_EXIT);
317 else
318 thread_single_end(p, SINGLE_BOUNDARY);
319 PROC_UNLOCK(p);
320 }
321 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
322 KASSERT(p->p_vmspace != oldvmspace,
323 ("oldvmspace still used"));
324 vmspace_free(oldvmspace);
325 td->td_pflags &= ~TDP_EXECVMSPC;
326 }
327 }
328
329 /*
330 * XXX: kern_execve has the astonishing property of not always returning to
331 * the caller. If sufficiently bad things happen during the call to
332 * do_execve(), it can end up calling exit1(); as a result, callers must
333 * avoid doing anything which they might need to undo (e.g., allocating
334 * memory).
335 */
336 int
337 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
338 {
339
340 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
341 args->begin_envv - args->begin_argv);
342 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
343 args->endp - args->begin_envv);
344 return (do_execve(td, args, mac_p));
345 }
346
347 /*
348 * In-kernel implementation of execve(). All arguments are assumed to be
349 * userspace pointers from the passed thread.
350 */
351 static int
352 do_execve(td, args, mac_p)
353 struct thread *td;
354 struct image_args *args;
355 struct mac *mac_p;
356 {
357 struct proc *p = td->td_proc;
358 struct nameidata nd;
359 struct ucred *oldcred;
360 struct uidinfo *euip = NULL;
361 register_t *stack_base;
362 int error, i;
363 struct image_params image_params, *imgp;
364 struct vattr attr;
365 int (*img_first)(struct image_params *);
366 struct pargs *oldargs = NULL, *newargs = NULL;
367 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
368 #ifdef KTRACE
369 struct vnode *tracevp = NULL;
370 struct ucred *tracecred = NULL;
371 #endif
372 struct vnode *oldtextvp = NULL, *newtextvp;
373 cap_rights_t rights;
374 int credential_changing;
375 int textset;
376 #ifdef MAC
377 struct label *interpvplabel = NULL;
378 int will_transition;
379 #endif
380 #ifdef HWPMC_HOOKS
381 struct pmckern_procexec pe;
382 #endif
383 static const char fexecv_proc_title[] = "(fexecv)";
384
385 imgp = &image_params;
386
387 /*
388 * Lock the process and set the P_INEXEC flag to indicate that
389 * it should be left alone until we're done here. This is
390 * necessary to avoid race conditions - e.g. in ptrace() -
391 * that might allow a local user to illicitly obtain elevated
392 * privileges.
393 */
394 PROC_LOCK(p);
395 KASSERT((p->p_flag & P_INEXEC) == 0,
396 ("%s(): process already has P_INEXEC flag", __func__));
397 p->p_flag |= P_INEXEC;
398 PROC_UNLOCK(p);
399
400 /*
401 * Initialize part of the common data
402 */
403 bzero(imgp, sizeof(*imgp));
404 imgp->proc = p;
405 imgp->attr = &attr;
406 imgp->args = args;
407 oldcred = p->p_ucred;
408
409 #ifdef MAC
410 error = mac_execve_enter(imgp, mac_p);
411 if (error)
412 goto exec_fail;
413 #endif
414
415 /*
416 * Translate the file name. namei() returns a vnode pointer
417 * in ni_vp among other things.
418 *
419 * XXXAUDIT: It would be desirable to also audit the name of the
420 * interpreter if this is an interpreted binary.
421 */
422 if (args->fname != NULL) {
423 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
424 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
425 }
426
427 SDT_PROBE1(proc, , , exec, args->fname);
428
429 interpret:
430 if (args->fname != NULL) {
431 #ifdef CAPABILITY_MODE
432 /*
433 * While capability mode can't reach this point via direct
434 * path arguments to execve(), we also don't allow
435 * interpreters to be used in capability mode (for now).
436 * Catch indirect lookups and return a permissions error.
437 */
438 if (IN_CAPABILITY_MODE(td)) {
439 error = ECAPMODE;
440 goto exec_fail;
441 }
442 #endif
443 error = namei(&nd);
444 if (error)
445 goto exec_fail;
446
447 newtextvp = nd.ni_vp;
448 imgp->vp = newtextvp;
449 } else {
450 AUDIT_ARG_FD(args->fd);
451 /*
452 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
453 */
454 error = fgetvp_exec(td, args->fd,
455 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
456 if (error)
457 goto exec_fail;
458 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
459 AUDIT_ARG_VNODE1(newtextvp);
460 imgp->vp = newtextvp;
461 }
462
463 /*
464 * Check file permissions (also 'opens' file)
465 */
466 error = exec_check_permissions(imgp);
467 if (error)
468 goto exec_fail_dealloc;
469
470 imgp->object = imgp->vp->v_object;
471 if (imgp->object != NULL)
472 vm_object_reference(imgp->object);
473
474 /*
475 * Set VV_TEXT now so no one can write to the executable while we're
476 * activating it.
477 *
478 * Remember if this was set before and unset it in case this is not
479 * actually an executable image.
480 */
481 textset = VOP_IS_TEXT(imgp->vp);
482 VOP_SET_TEXT(imgp->vp);
483
484 error = exec_map_first_page(imgp);
485 if (error)
486 goto exec_fail_dealloc;
487
488 imgp->proc->p_osrel = 0;
489
490 /*
491 * Implement image setuid/setgid.
492 *
493 * Determine new credentials before attempting image activators
494 * so that it can be used by process_exec handlers to determine
495 * credential/setid changes.
496 *
497 * Don't honor setuid/setgid if the filesystem prohibits it or if
498 * the process is being traced.
499 *
500 * We disable setuid/setgid/etc in capability mode on the basis
501 * that most setugid applications are not written with that
502 * environment in mind, and will therefore almost certainly operate
503 * incorrectly. In principle there's no reason that setugid
504 * applications might not be useful in capability mode, so we may want
505 * to reconsider this conservative design choice in the future.
506 *
507 * XXXMAC: For the time being, use NOSUID to also prohibit
508 * transitions on the file system.
509 */
510 credential_changing = 0;
511 credential_changing |= (attr.va_mode & S_ISUID) &&
512 oldcred->cr_uid != attr.va_uid;
513 credential_changing |= (attr.va_mode & S_ISGID) &&
514 oldcred->cr_gid != attr.va_gid;
515 #ifdef MAC
516 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
517 interpvplabel, imgp);
518 credential_changing |= will_transition;
519 #endif
520
521 if (credential_changing &&
522 #ifdef CAPABILITY_MODE
523 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
524 #endif
525 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
526 (p->p_flag & P_TRACED) == 0) {
527 imgp->credential_setid = true;
528 VOP_UNLOCK(imgp->vp, 0);
529 imgp->newcred = crdup(oldcred);
530 if (attr.va_mode & S_ISUID) {
531 euip = uifind(attr.va_uid);
532 change_euid(imgp->newcred, euip);
533 }
534 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
535 if (attr.va_mode & S_ISGID)
536 change_egid(imgp->newcred, attr.va_gid);
537 /*
538 * Implement correct POSIX saved-id behavior.
539 *
540 * XXXMAC: Note that the current logic will save the
541 * uid and gid if a MAC domain transition occurs, even
542 * though maybe it shouldn't.
543 */
544 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
545 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
546 } else {
547 /*
548 * Implement correct POSIX saved-id behavior.
549 *
550 * XXX: It's not clear that the existing behavior is
551 * POSIX-compliant. A number of sources indicate that the
552 * saved uid/gid should only be updated if the new ruid is
553 * not equal to the old ruid, or the new euid is not equal
554 * to the old euid and the new euid is not equal to the old
555 * ruid. The FreeBSD code always updates the saved uid/gid.
556 * Also, this code uses the new (replaced) euid and egid as
557 * the source, which may or may not be the right ones to use.
558 */
559 if (oldcred->cr_svuid != oldcred->cr_uid ||
560 oldcred->cr_svgid != oldcred->cr_gid) {
561 VOP_UNLOCK(imgp->vp, 0);
562 imgp->newcred = crdup(oldcred);
563 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
564 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
565 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
566 }
567 }
568 /* The new credentials are installed into the process later. */
569
570 /*
571 * Do the best to calculate the full path to the image file.
572 */
573 if (args->fname != NULL && args->fname[0] == '/')
574 imgp->execpath = args->fname;
575 else {
576 VOP_UNLOCK(imgp->vp, 0);
577 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
578 &imgp->freepath) != 0)
579 imgp->execpath = args->fname;
580 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
581 }
582
583 /*
584 * If the current process has a special image activator it
585 * wants to try first, call it. For example, emulating shell
586 * scripts differently.
587 */
588 error = -1;
589 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
590 error = img_first(imgp);
591
592 /*
593 * Loop through the list of image activators, calling each one.
594 * An activator returns -1 if there is no match, 0 on success,
595 * and an error otherwise.
596 */
597 for (i = 0; error == -1 && execsw[i]; ++i) {
598 if (execsw[i]->ex_imgact == NULL ||
599 execsw[i]->ex_imgact == img_first) {
600 continue;
601 }
602 error = (*execsw[i]->ex_imgact)(imgp);
603 }
604
605 if (error) {
606 if (error == -1) {
607 if (textset == 0)
608 VOP_UNSET_TEXT(imgp->vp);
609 error = ENOEXEC;
610 }
611 goto exec_fail_dealloc;
612 }
613
614 /*
615 * Special interpreter operation, cleanup and loop up to try to
616 * activate the interpreter.
617 */
618 if (imgp->interpreted) {
619 exec_unmap_first_page(imgp);
620 /*
621 * VV_TEXT needs to be unset for scripts. There is a short
622 * period before we determine that something is a script where
623 * VV_TEXT will be set. The vnode lock is held over this
624 * entire period so nothing should illegitimately be blocked.
625 */
626 VOP_UNSET_TEXT(imgp->vp);
627 /* free name buffer and old vnode */
628 if (args->fname != NULL)
629 NDFREE(&nd, NDF_ONLY_PNBUF);
630 #ifdef MAC
631 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
632 #endif
633 if (imgp->opened) {
634 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
635 imgp->opened = 0;
636 }
637 vput(newtextvp);
638 vm_object_deallocate(imgp->object);
639 imgp->object = NULL;
640 imgp->credential_setid = false;
641 if (imgp->newcred != NULL) {
642 crfree(imgp->newcred);
643 imgp->newcred = NULL;
644 }
645 imgp->execpath = NULL;
646 free(imgp->freepath, M_TEMP);
647 imgp->freepath = NULL;
648 /* set new name to that of the interpreter */
649 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
650 UIO_SYSSPACE, imgp->interpreter_name, td);
651 args->fname = imgp->interpreter_name;
652 goto interpret;
653 }
654
655 /*
656 * NB: We unlock the vnode here because it is believed that none
657 * of the sv_copyout_strings/sv_fixup operations require the vnode.
658 */
659 VOP_UNLOCK(imgp->vp, 0);
660
661 if (disallow_high_osrel &&
662 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
663 error = ENOEXEC;
664 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
665 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
666 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
667 goto exec_fail_dealloc;
668 }
669
670 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
671 if (SV_PROC_FLAG(p, SV_CAPSICUM))
672 sys_cap_enter(td, NULL);
673
674 /*
675 * Copy out strings (args and env) and initialize stack base
676 */
677 if (p->p_sysent->sv_copyout_strings)
678 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
679 else
680 stack_base = exec_copyout_strings(imgp);
681
682 /*
683 * If custom stack fixup routine present for this process
684 * let it do the stack setup.
685 * Else stuff argument count as first item on stack
686 */
687 if (p->p_sysent->sv_fixup != NULL)
688 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
689 else
690 suword(--stack_base, imgp->args->argc);
691
692 if (args->fdp != NULL) {
693 /* Install a brand new file descriptor table. */
694 fdinstall_remapped(td, args->fdp);
695 args->fdp = NULL;
696 } else {
697 /*
698 * Keep on using the existing file descriptor table. For
699 * security and other reasons, the file descriptor table
700 * cannot be shared after an exec.
701 */
702 fdunshare(td);
703 /* close files on exec */
704 fdcloseexec(td);
705 }
706
707 /*
708 * Malloc things before we need locks.
709 */
710 i = imgp->args->begin_envv - imgp->args->begin_argv;
711 /* Cache arguments if they fit inside our allowance */
712 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
713 newargs = pargs_alloc(i);
714 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
715 }
716
717 /*
718 * For security and other reasons, signal handlers cannot
719 * be shared after an exec. The new process gets a copy of the old
720 * handlers. In execsigs(), the new process will have its signals
721 * reset.
722 */
723 if (sigacts_shared(p->p_sigacts)) {
724 oldsigacts = p->p_sigacts;
725 newsigacts = sigacts_alloc();
726 sigacts_copy(newsigacts, oldsigacts);
727 }
728
729 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
730
731 PROC_LOCK(p);
732 if (oldsigacts)
733 p->p_sigacts = newsigacts;
734 /* Stop profiling */
735 stopprofclock(p);
736
737 /* reset caught signals */
738 execsigs(p);
739
740 /* name this process - nameiexec(p, ndp) */
741 bzero(p->p_comm, sizeof(p->p_comm));
742 if (args->fname)
743 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
744 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
745 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
746 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
747 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
748 #ifdef KTR
749 sched_clear_tdname(td);
750 #endif
751
752 /*
753 * mark as execed, wakeup the process that vforked (if any) and tell
754 * it that it now has its own resources back
755 */
756 p->p_flag |= P_EXEC;
757 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
758 p->p_flag2 &= ~P2_NOTRACE;
759 if (p->p_flag & P_PPWAIT) {
760 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
761 cv_broadcast(&p->p_pwait);
762 /* STOPs are no longer ignored, arrange for AST */
763 signotify(td);
764 }
765
766 /*
767 * Implement image setuid/setgid installation.
768 */
769 if (imgp->credential_setid) {
770 /*
771 * Turn off syscall tracing for set-id programs, except for
772 * root. Record any set-id flags first to make sure that
773 * we do not regain any tracing during a possible block.
774 */
775 setsugid(p);
776
777 #ifdef KTRACE
778 if (p->p_tracecred != NULL &&
779 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
780 ktrprocexec(p, &tracecred, &tracevp);
781 #endif
782 /*
783 * Close any file descriptors 0..2 that reference procfs,
784 * then make sure file descriptors 0..2 are in use.
785 *
786 * Both fdsetugidsafety() and fdcheckstd() may call functions
787 * taking sleepable locks, so temporarily drop our locks.
788 */
789 PROC_UNLOCK(p);
790 VOP_UNLOCK(imgp->vp, 0);
791 fdsetugidsafety(td);
792 error = fdcheckstd(td);
793 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
794 if (error != 0)
795 goto exec_fail_dealloc;
796 PROC_LOCK(p);
797 #ifdef MAC
798 if (will_transition) {
799 mac_vnode_execve_transition(oldcred, imgp->newcred,
800 imgp->vp, interpvplabel, imgp);
801 }
802 #endif
803 } else {
804 if (oldcred->cr_uid == oldcred->cr_ruid &&
805 oldcred->cr_gid == oldcred->cr_rgid)
806 p->p_flag &= ~P_SUGID;
807 }
808 /*
809 * Set the new credentials.
810 */
811 if (imgp->newcred != NULL) {
812 proc_set_cred(p, imgp->newcred);
813 crfree(oldcred);
814 oldcred = NULL;
815 }
816
817 /*
818 * Store the vp for use in procfs. This vnode was referenced by namei
819 * or fgetvp_exec.
820 */
821 oldtextvp = p->p_textvp;
822 p->p_textvp = newtextvp;
823
824 #ifdef KDTRACE_HOOKS
825 /*
826 * Tell the DTrace fasttrap provider about the exec if it
827 * has declared an interest.
828 */
829 if (dtrace_fasttrap_exec)
830 dtrace_fasttrap_exec(p);
831 #endif
832
833 /*
834 * Notify others that we exec'd, and clear the P_INEXEC flag
835 * as we're now a bona fide freshly-execed process.
836 */
837 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
838 p->p_flag &= ~P_INEXEC;
839
840 /* clear "fork but no exec" flag, as we _are_ execing */
841 p->p_acflag &= ~AFORK;
842
843 /*
844 * Free any previous argument cache and replace it with
845 * the new argument cache, if any.
846 */
847 oldargs = p->p_args;
848 p->p_args = newargs;
849 newargs = NULL;
850
851 #ifdef HWPMC_HOOKS
852 /*
853 * Check if system-wide sampling is in effect or if the
854 * current process is using PMCs. If so, do exec() time
855 * processing. This processing needs to happen AFTER the
856 * P_INEXEC flag is cleared.
857 *
858 * The proc lock needs to be released before taking the PMC
859 * SX.
860 */
861 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
862 PROC_UNLOCK(p);
863 VOP_UNLOCK(imgp->vp, 0);
864 pe.pm_credentialschanged = credential_changing;
865 pe.pm_entryaddr = imgp->entry_addr;
866
867 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
868 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
869 } else
870 PROC_UNLOCK(p);
871 #else /* !HWPMC_HOOKS */
872 PROC_UNLOCK(p);
873 #endif
874
875 /* Set values passed into the program in registers. */
876 if (p->p_sysent->sv_setregs)
877 (*p->p_sysent->sv_setregs)(td, imgp,
878 (u_long)(uintptr_t)stack_base);
879 else
880 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
881
882 vfs_mark_atime(imgp->vp, td->td_ucred);
883
884 SDT_PROBE1(proc, , , exec__success, args->fname);
885
886 exec_fail_dealloc:
887 if (imgp->firstpage != NULL)
888 exec_unmap_first_page(imgp);
889
890 if (imgp->vp != NULL) {
891 if (args->fname)
892 NDFREE(&nd, NDF_ONLY_PNBUF);
893 if (imgp->opened)
894 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
895 if (error != 0)
896 vput(imgp->vp);
897 else
898 VOP_UNLOCK(imgp->vp, 0);
899 }
900
901 if (imgp->object != NULL)
902 vm_object_deallocate(imgp->object);
903
904 free(imgp->freepath, M_TEMP);
905
906 if (error == 0) {
907 PROC_LOCK(p);
908 td->td_dbgflags |= TDB_EXEC;
909 PROC_UNLOCK(p);
910
911 /*
912 * Stop the process here if its stop event mask has
913 * the S_EXEC bit set.
914 */
915 STOPEVENT(p, S_EXEC, 0);
916 } else {
917 exec_fail:
918 /* we're done here, clear P_INEXEC */
919 PROC_LOCK(p);
920 p->p_flag &= ~P_INEXEC;
921 PROC_UNLOCK(p);
922
923 SDT_PROBE1(proc, , , exec__failure, error);
924 }
925
926 if (imgp->newcred != NULL && oldcred != NULL)
927 crfree(imgp->newcred);
928
929 #ifdef MAC
930 mac_execve_exit(imgp);
931 mac_execve_interpreter_exit(interpvplabel);
932 #endif
933 exec_free_args(args);
934
935 /*
936 * Handle deferred decrement of ref counts.
937 */
938 if (oldtextvp != NULL)
939 vrele(oldtextvp);
940 #ifdef KTRACE
941 if (tracevp != NULL)
942 vrele(tracevp);
943 if (tracecred != NULL)
944 crfree(tracecred);
945 #endif
946 pargs_drop(oldargs);
947 pargs_drop(newargs);
948 if (oldsigacts != NULL)
949 sigacts_free(oldsigacts);
950 if (euip != NULL)
951 uifree(euip);
952
953 if (error && imgp->vmspace_destroyed) {
954 /* sorry, no more process anymore. exit gracefully */
955 exit1(td, 0, SIGABRT);
956 /* NOT REACHED */
957 }
958
959 #ifdef KTRACE
960 if (error == 0)
961 ktrprocctor(p);
962 #endif
963
964 return (error);
965 }
966
967 int
968 exec_map_first_page(imgp)
969 struct image_params *imgp;
970 {
971 int rv, i, after, initial_pagein;
972 vm_page_t ma[VM_INITIAL_PAGEIN];
973 vm_object_t object;
974
975 if (imgp->firstpage != NULL)
976 exec_unmap_first_page(imgp);
977
978 object = imgp->vp->v_object;
979 if (object == NULL)
980 return (EACCES);
981 VM_OBJECT_WLOCK(object);
982 #if VM_NRESERVLEVEL > 0
983 vm_object_color(object, 0);
984 #endif
985 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
986 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
987 if (!vm_pager_has_page(object, 0, NULL, &after)) {
988 vm_page_lock(ma[0]);
989 vm_page_free(ma[0]);
990 vm_page_unlock(ma[0]);
991 vm_page_xunbusy(ma[0]);
992 VM_OBJECT_WUNLOCK(object);
993 return (EIO);
994 }
995 initial_pagein = min(after, VM_INITIAL_PAGEIN);
996 KASSERT(initial_pagein <= object->size,
997 ("%s: initial_pagein %d object->size %ju",
998 __func__, initial_pagein, (uintmax_t )object->size));
999 for (i = 1; i < initial_pagein; i++) {
1000 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1001 if (ma[i]->valid)
1002 break;
1003 if (vm_page_tryxbusy(ma[i]))
1004 break;
1005 } else {
1006 ma[i] = vm_page_alloc(object, i,
1007 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
1008 if (ma[i] == NULL)
1009 break;
1010 }
1011 }
1012 initial_pagein = i;
1013 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1014 if (rv != VM_PAGER_OK) {
1015 for (i = 0; i < initial_pagein; i++) {
1016 vm_page_lock(ma[i]);
1017 vm_page_free(ma[i]);
1018 vm_page_unlock(ma[i]);
1019 vm_page_xunbusy(ma[i]);
1020 }
1021 VM_OBJECT_WUNLOCK(object);
1022 return (EIO);
1023 }
1024 for (i = 1; i < initial_pagein; i++)
1025 vm_page_readahead_finish(ma[i]);
1026 }
1027 vm_page_xunbusy(ma[0]);
1028 vm_page_lock(ma[0]);
1029 vm_page_hold(ma[0]);
1030 vm_page_activate(ma[0]);
1031 vm_page_unlock(ma[0]);
1032 VM_OBJECT_WUNLOCK(object);
1033
1034 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1035 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1036
1037 return (0);
1038 }
1039
1040 void
1041 exec_unmap_first_page(imgp)
1042 struct image_params *imgp;
1043 {
1044 vm_page_t m;
1045
1046 if (imgp->firstpage != NULL) {
1047 m = sf_buf_page(imgp->firstpage);
1048 sf_buf_free(imgp->firstpage);
1049 imgp->firstpage = NULL;
1050 vm_page_lock(m);
1051 vm_page_unhold(m);
1052 vm_page_unlock(m);
1053 }
1054 }
1055
1056 /*
1057 * Destroy old address space, and allocate a new stack
1058 * The new stack is only SGROWSIZ large because it is grown
1059 * automatically in trap.c.
1060 */
1061 int
1062 exec_new_vmspace(imgp, sv)
1063 struct image_params *imgp;
1064 struct sysentvec *sv;
1065 {
1066 int error;
1067 struct proc *p = imgp->proc;
1068 struct vmspace *vmspace = p->p_vmspace;
1069 vm_object_t obj;
1070 struct rlimit rlim_stack;
1071 vm_offset_t sv_minuser, stack_addr;
1072 vm_map_t map;
1073 u_long ssiz;
1074
1075 imgp->vmspace_destroyed = 1;
1076 imgp->sysent = sv;
1077
1078 /* May be called with Giant held */
1079 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1080
1081 /*
1082 * Blow away entire process VM, if address space not shared,
1083 * otherwise, create a new VM space so that other threads are
1084 * not disrupted
1085 */
1086 map = &vmspace->vm_map;
1087 if (map_at_zero)
1088 sv_minuser = sv->sv_minuser;
1089 else
1090 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1091 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1092 vm_map_max(map) == sv->sv_maxuser) {
1093 shmexit(vmspace);
1094 pmap_remove_pages(vmspace_pmap(vmspace));
1095 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1096 } else {
1097 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1098 if (error)
1099 return (error);
1100 vmspace = p->p_vmspace;
1101 map = &vmspace->vm_map;
1102 }
1103
1104 /* Map a shared page */
1105 obj = sv->sv_shared_page_obj;
1106 if (obj != NULL) {
1107 vm_object_reference(obj);
1108 error = vm_map_fixed(map, obj, 0,
1109 sv->sv_shared_page_base, sv->sv_shared_page_len,
1110 VM_PROT_READ | VM_PROT_EXECUTE,
1111 VM_PROT_READ | VM_PROT_EXECUTE,
1112 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1113 if (error) {
1114 vm_object_deallocate(obj);
1115 return (error);
1116 }
1117 }
1118
1119 /* Allocate a new stack */
1120 if (imgp->stack_sz != 0) {
1121 ssiz = trunc_page(imgp->stack_sz);
1122 PROC_LOCK(p);
1123 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1124 PROC_UNLOCK(p);
1125 if (ssiz > rlim_stack.rlim_max)
1126 ssiz = rlim_stack.rlim_max;
1127 if (ssiz > rlim_stack.rlim_cur) {
1128 rlim_stack.rlim_cur = ssiz;
1129 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1130 }
1131 } else if (sv->sv_maxssiz != NULL) {
1132 ssiz = *sv->sv_maxssiz;
1133 } else {
1134 ssiz = maxssiz;
1135 }
1136 stack_addr = sv->sv_usrstack - ssiz;
1137 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1138 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1139 sv->sv_stackprot,
1140 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1141 if (error)
1142 return (error);
1143
1144 /*
1145 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1146 * are still used to enforce the stack rlimit on the process stack.
1147 */
1148 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1149 vmspace->vm_maxsaddr = (char *)stack_addr;
1150
1151 return (0);
1152 }
1153
1154 /*
1155 * Copy out argument and environment strings from the old process address
1156 * space into the temporary string buffer.
1157 */
1158 int
1159 exec_copyin_args(struct image_args *args, char *fname,
1160 enum uio_seg segflg, char **argv, char **envv)
1161 {
1162 u_long argp, envp;
1163 int error;
1164 size_t length;
1165
1166 bzero(args, sizeof(*args));
1167 if (argv == NULL)
1168 return (EFAULT);
1169
1170 /*
1171 * Allocate demand-paged memory for the file name, argument, and
1172 * environment strings.
1173 */
1174 error = exec_alloc_args(args);
1175 if (error != 0)
1176 return (error);
1177
1178 /*
1179 * Copy the file name.
1180 */
1181 if (fname != NULL) {
1182 args->fname = args->buf;
1183 error = (segflg == UIO_SYSSPACE) ?
1184 copystr(fname, args->fname, PATH_MAX, &length) :
1185 copyinstr(fname, args->fname, PATH_MAX, &length);
1186 if (error != 0)
1187 goto err_exit;
1188 } else
1189 length = 0;
1190
1191 args->begin_argv = args->buf + length;
1192 args->endp = args->begin_argv;
1193 args->stringspace = ARG_MAX;
1194
1195 /*
1196 * extract arguments first
1197 */
1198 for (;;) {
1199 error = fueword(argv++, &argp);
1200 if (error == -1) {
1201 error = EFAULT;
1202 goto err_exit;
1203 }
1204 if (argp == 0)
1205 break;
1206 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1207 args->stringspace, &length);
1208 if (error != 0) {
1209 if (error == ENAMETOOLONG)
1210 error = E2BIG;
1211 goto err_exit;
1212 }
1213 args->stringspace -= length;
1214 args->endp += length;
1215 args->argc++;
1216 }
1217
1218 args->begin_envv = args->endp;
1219
1220 /*
1221 * extract environment strings
1222 */
1223 if (envv) {
1224 for (;;) {
1225 error = fueword(envv++, &envp);
1226 if (error == -1) {
1227 error = EFAULT;
1228 goto err_exit;
1229 }
1230 if (envp == 0)
1231 break;
1232 error = copyinstr((void *)(uintptr_t)envp,
1233 args->endp, args->stringspace, &length);
1234 if (error != 0) {
1235 if (error == ENAMETOOLONG)
1236 error = E2BIG;
1237 goto err_exit;
1238 }
1239 args->stringspace -= length;
1240 args->endp += length;
1241 args->envc++;
1242 }
1243 }
1244
1245 return (0);
1246
1247 err_exit:
1248 exec_free_args(args);
1249 return (error);
1250 }
1251
1252 int
1253 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1254 const void *data, size_t datalen, const int *fds, size_t fdslen)
1255 {
1256 struct filedesc *ofdp;
1257 const char *p;
1258 int *kfds;
1259 int error;
1260
1261 memset(args, '\0', sizeof(*args));
1262 ofdp = td->td_proc->p_fd;
1263 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1264 return (E2BIG);
1265 error = exec_alloc_args(args);
1266 if (error != 0)
1267 return (error);
1268
1269 args->begin_argv = args->buf;
1270 args->stringspace = ARG_MAX;
1271
1272 if (datalen > 0) {
1273 /*
1274 * Argument buffer has been provided. Copy it into the
1275 * kernel as a single string and add a terminating null
1276 * byte.
1277 */
1278 error = copyin(data, args->begin_argv, datalen);
1279 if (error != 0)
1280 goto err_exit;
1281 args->begin_argv[datalen] = '\0';
1282 args->endp = args->begin_argv + datalen + 1;
1283 args->stringspace -= datalen + 1;
1284
1285 /*
1286 * Traditional argument counting. Count the number of
1287 * null bytes.
1288 */
1289 for (p = args->begin_argv; p < args->endp; ++p)
1290 if (*p == '\0')
1291 ++args->argc;
1292 } else {
1293 /* No argument buffer provided. */
1294 args->endp = args->begin_argv;
1295 }
1296 /* There are no environment variables. */
1297 args->begin_envv = args->endp;
1298
1299 /* Create new file descriptor table. */
1300 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1301 error = copyin(fds, kfds, fdslen * sizeof(int));
1302 if (error != 0) {
1303 free(kfds, M_TEMP);
1304 goto err_exit;
1305 }
1306 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1307 free(kfds, M_TEMP);
1308 if (error != 0)
1309 goto err_exit;
1310
1311 return (0);
1312 err_exit:
1313 exec_free_args(args);
1314 return (error);
1315 }
1316
1317 /*
1318 * Allocate temporary demand-paged, zero-filled memory for the file name,
1319 * argument, and environment strings. Returns zero if the allocation succeeds
1320 * and ENOMEM otherwise.
1321 */
1322 int
1323 exec_alloc_args(struct image_args *args)
1324 {
1325
1326 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1327 return (args->buf != NULL ? 0 : ENOMEM);
1328 }
1329
1330 void
1331 exec_free_args(struct image_args *args)
1332 {
1333
1334 if (args->buf != NULL) {
1335 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1336 PATH_MAX + ARG_MAX);
1337 args->buf = NULL;
1338 }
1339 if (args->fname_buf != NULL) {
1340 free(args->fname_buf, M_TEMP);
1341 args->fname_buf = NULL;
1342 }
1343 if (args->fdp != NULL)
1344 fdescfree_remapped(args->fdp);
1345 }
1346
1347 /*
1348 * Copy strings out to the new process address space, constructing new arg
1349 * and env vector tables. Return a pointer to the base so that it can be used
1350 * as the initial stack pointer.
1351 */
1352 register_t *
1353 exec_copyout_strings(imgp)
1354 struct image_params *imgp;
1355 {
1356 int argc, envc;
1357 char **vectp;
1358 char *stringp;
1359 uintptr_t destp;
1360 register_t *stack_base;
1361 struct ps_strings *arginfo;
1362 struct proc *p;
1363 size_t execpath_len;
1364 int szsigcode, szps;
1365 char canary[sizeof(long) * 8];
1366
1367 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1368 /*
1369 * Calculate string base and vector table pointers.
1370 * Also deal with signal trampoline code for this exec type.
1371 */
1372 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1373 execpath_len = strlen(imgp->execpath) + 1;
1374 else
1375 execpath_len = 0;
1376 p = imgp->proc;
1377 szsigcode = 0;
1378 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1379 if (p->p_sysent->sv_sigcode_base == 0) {
1380 if (p->p_sysent->sv_szsigcode != NULL)
1381 szsigcode = *(p->p_sysent->sv_szsigcode);
1382 }
1383 destp = (uintptr_t)arginfo;
1384
1385 /*
1386 * install sigcode
1387 */
1388 if (szsigcode != 0) {
1389 destp -= szsigcode;
1390 destp = rounddown2(destp, sizeof(void *));
1391 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1392 }
1393
1394 /*
1395 * Copy the image path for the rtld.
1396 */
1397 if (execpath_len != 0) {
1398 destp -= execpath_len;
1399 imgp->execpathp = destp;
1400 copyout(imgp->execpath, (void *)destp, execpath_len);
1401 }
1402
1403 /*
1404 * Prepare the canary for SSP.
1405 */
1406 arc4rand(canary, sizeof(canary), 0);
1407 destp -= sizeof(canary);
1408 imgp->canary = destp;
1409 copyout(canary, (void *)destp, sizeof(canary));
1410 imgp->canarylen = sizeof(canary);
1411
1412 /*
1413 * Prepare the pagesizes array.
1414 */
1415 destp -= szps;
1416 destp = rounddown2(destp, sizeof(void *));
1417 imgp->pagesizes = destp;
1418 copyout(pagesizes, (void *)destp, szps);
1419 imgp->pagesizeslen = szps;
1420
1421 destp -= ARG_MAX - imgp->args->stringspace;
1422 destp = rounddown2(destp, sizeof(void *));
1423
1424 /*
1425 * If we have a valid auxargs ptr, prepare some room
1426 * on the stack.
1427 */
1428 if (imgp->auxargs) {
1429 /*
1430 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1431 * lower compatibility.
1432 */
1433 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1434 (AT_COUNT * 2);
1435 /*
1436 * The '+ 2' is for the null pointers at the end of each of
1437 * the arg and env vector sets,and imgp->auxarg_size is room
1438 * for argument of Runtime loader.
1439 */
1440 vectp = (char **)(destp - (imgp->args->argc +
1441 imgp->args->envc + 2 + imgp->auxarg_size)
1442 * sizeof(char *));
1443 } else {
1444 /*
1445 * The '+ 2' is for the null pointers at the end of each of
1446 * the arg and env vector sets
1447 */
1448 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1449 + 2) * sizeof(char *));
1450 }
1451
1452 /*
1453 * vectp also becomes our initial stack base
1454 */
1455 stack_base = (register_t *)vectp;
1456
1457 stringp = imgp->args->begin_argv;
1458 argc = imgp->args->argc;
1459 envc = imgp->args->envc;
1460
1461 /*
1462 * Copy out strings - arguments and environment.
1463 */
1464 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1465
1466 /*
1467 * Fill in "ps_strings" struct for ps, w, etc.
1468 */
1469 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1470 suword32(&arginfo->ps_nargvstr, argc);
1471
1472 /*
1473 * Fill in argument portion of vector table.
1474 */
1475 for (; argc > 0; --argc) {
1476 suword(vectp++, (long)(intptr_t)destp);
1477 while (*stringp++ != 0)
1478 destp++;
1479 destp++;
1480 }
1481
1482 /* a null vector table pointer separates the argp's from the envp's */
1483 suword(vectp++, 0);
1484
1485 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1486 suword32(&arginfo->ps_nenvstr, envc);
1487
1488 /*
1489 * Fill in environment portion of vector table.
1490 */
1491 for (; envc > 0; --envc) {
1492 suword(vectp++, (long)(intptr_t)destp);
1493 while (*stringp++ != 0)
1494 destp++;
1495 destp++;
1496 }
1497
1498 /* end of vector table is a null pointer */
1499 suword(vectp, 0);
1500
1501 return (stack_base);
1502 }
1503
1504 /*
1505 * Check permissions of file to execute.
1506 * Called with imgp->vp locked.
1507 * Return 0 for success or error code on failure.
1508 */
1509 int
1510 exec_check_permissions(imgp)
1511 struct image_params *imgp;
1512 {
1513 struct vnode *vp = imgp->vp;
1514 struct vattr *attr = imgp->attr;
1515 struct thread *td;
1516 int error, writecount;
1517
1518 td = curthread;
1519
1520 /* Get file attributes */
1521 error = VOP_GETATTR(vp, attr, td->td_ucred);
1522 if (error)
1523 return (error);
1524
1525 #ifdef MAC
1526 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1527 if (error)
1528 return (error);
1529 #endif
1530
1531 /*
1532 * 1) Check if file execution is disabled for the filesystem that
1533 * this file resides on.
1534 * 2) Ensure that at least one execute bit is on. Otherwise, a
1535 * privileged user will always succeed, and we don't want this
1536 * to happen unless the file really is executable.
1537 * 3) Ensure that the file is a regular file.
1538 */
1539 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1540 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1541 (attr->va_type != VREG))
1542 return (EACCES);
1543
1544 /*
1545 * Zero length files can't be exec'd
1546 */
1547 if (attr->va_size == 0)
1548 return (ENOEXEC);
1549
1550 /*
1551 * Check for execute permission to file based on current credentials.
1552 */
1553 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1554 if (error)
1555 return (error);
1556
1557 /*
1558 * Check number of open-for-writes on the file and deny execution
1559 * if there are any.
1560 */
1561 error = VOP_GET_WRITECOUNT(vp, &writecount);
1562 if (error != 0)
1563 return (error);
1564 if (writecount != 0)
1565 return (ETXTBSY);
1566
1567 /*
1568 * Call filesystem specific open routine (which does nothing in the
1569 * general case).
1570 */
1571 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1572 if (error == 0)
1573 imgp->opened = 1;
1574 return (error);
1575 }
1576
1577 /*
1578 * Exec handler registration
1579 */
1580 int
1581 exec_register(execsw_arg)
1582 const struct execsw *execsw_arg;
1583 {
1584 const struct execsw **es, **xs, **newexecsw;
1585 int count = 2; /* New slot and trailing NULL */
1586
1587 if (execsw)
1588 for (es = execsw; *es; es++)
1589 count++;
1590 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1591 xs = newexecsw;
1592 if (execsw)
1593 for (es = execsw; *es; es++)
1594 *xs++ = *es;
1595 *xs++ = execsw_arg;
1596 *xs = NULL;
1597 if (execsw)
1598 free(execsw, M_TEMP);
1599 execsw = newexecsw;
1600 return (0);
1601 }
1602
1603 int
1604 exec_unregister(execsw_arg)
1605 const struct execsw *execsw_arg;
1606 {
1607 const struct execsw **es, **xs, **newexecsw;
1608 int count = 1;
1609
1610 if (execsw == NULL)
1611 panic("unregister with no handlers left?\n");
1612
1613 for (es = execsw; *es; es++) {
1614 if (*es == execsw_arg)
1615 break;
1616 }
1617 if (*es == NULL)
1618 return (ENOENT);
1619 for (es = execsw; *es; es++)
1620 if (*es != execsw_arg)
1621 count++;
1622 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1623 xs = newexecsw;
1624 for (es = execsw; *es; es++)
1625 if (*es != execsw_arg)
1626 *xs++ = *es;
1627 *xs = NULL;
1628 if (execsw)
1629 free(execsw, M_TEMP);
1630 execsw = newexecsw;
1631 return (0);
1632 }
Cache object: b79fb12239fd7398aaf68866edcd1c65
|