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/9.1/sys/kern/kern_exec.c 267018 2014-06-03 19:03:11Z delphij $");
29
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
34 #include "opt_vm.h"
35
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
49 #include <sys/acct.h>
50 #include <sys/exec.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
53 #include <sys/wait.h>
54 #include <sys/malloc.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
65 #include <sys/shm.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/stat.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
82
83 #ifdef HWPMC_HOOKS
84 #include <sys/pmckern.h>
85 #endif
86
87 #include <machine/reg.h>
88
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
91
92 #ifdef KDTRACE_HOOKS
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t dtrace_fasttrap_exec;
95 #endif
96
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
104
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106
107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
110 static int do_execve(struct thread *td, struct image_args *args,
111 struct mac *mac_p);
112
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
115 NULL, 0, sysctl_kern_ps_strings, "LU", "");
116
117 /* XXX This should be vm_size_t. */
118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
119 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
120
121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
122 NULL, 0, sysctl_kern_stackprot, "I", "");
123
124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
126 &ps_arg_cache_limit, 0, "");
127
128 static int map_at_zero = 0;
129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
131 "Permit processes to map an object at virtual address 0.");
132
133 static int
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
135 {
136 struct proc *p;
137 int error;
138
139 p = curproc;
140 #ifdef SCTL_MASK32
141 if (req->flags & SCTL_MASK32) {
142 unsigned int val;
143 val = (unsigned int)p->p_sysent->sv_psstrings;
144 error = SYSCTL_OUT(req, &val, sizeof(val));
145 } else
146 #endif
147 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 sizeof(p->p_sysent->sv_psstrings));
149 return error;
150 }
151
152 static int
153 sysctl_kern_usrstack(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_usrstack;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
164 } else
165 #endif
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 sizeof(p->p_sysent->sv_usrstack));
168 return error;
169 }
170
171 static int
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
173 {
174 struct proc *p;
175
176 p = curproc;
177 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 sizeof(p->p_sysent->sv_stackprot)));
179 }
180
181 /*
182 * Each of the items is a pointer to a `const struct execsw', hence the
183 * double pointer here.
184 */
185 static const struct execsw **execsw;
186
187 #ifndef _SYS_SYSPROTO_H_
188 struct execve_args {
189 char *fname;
190 char **argv;
191 char **envv;
192 };
193 #endif
194
195 int
196 sys_execve(td, uap)
197 struct thread *td;
198 struct execve_args /* {
199 char *fname;
200 char **argv;
201 char **envv;
202 } */ *uap;
203 {
204 int error;
205 struct image_args args;
206
207 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 uap->argv, uap->envv);
209 if (error == 0)
210 error = kern_execve(td, &args, NULL);
211 return (error);
212 }
213
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
216 int fd;
217 char **argv;
218 char **envv;
219 }
220 #endif
221 int
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
223 {
224 int error;
225 struct image_args args;
226
227 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 uap->argv, uap->envv);
229 if (error == 0) {
230 args.fd = uap->fd;
231 error = kern_execve(td, &args, NULL);
232 }
233 return (error);
234 }
235
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
238 char *fname;
239 char **argv;
240 char **envv;
241 struct mac *mac_p;
242 };
243 #endif
244
245 int
246 sys___mac_execve(td, uap)
247 struct thread *td;
248 struct __mac_execve_args /* {
249 char *fname;
250 char **argv;
251 char **envv;
252 struct mac *mac_p;
253 } */ *uap;
254 {
255 #ifdef MAC
256 int error;
257 struct image_args args;
258
259 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 uap->argv, uap->envv);
261 if (error == 0)
262 error = kern_execve(td, &args, uap->mac_p);
263 return (error);
264 #else
265 return (ENOSYS);
266 #endif
267 }
268
269 /*
270 * XXX: kern_execve has the astonishing property of not always returning to
271 * the caller. If sufficiently bad things happen during the call to
272 * do_execve(), it can end up calling exit1(); as a result, callers must
273 * avoid doing anything which they might need to undo (e.g., allocating
274 * memory).
275 */
276 int
277 kern_execve(td, args, mac_p)
278 struct thread *td;
279 struct image_args *args;
280 struct mac *mac_p;
281 {
282 struct proc *p = td->td_proc;
283 struct vmspace *oldvmspace;
284 int error;
285
286 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
287 args->begin_envv - args->begin_argv);
288 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
289 args->endp - args->begin_envv);
290 if (p->p_flag & P_HADTHREADS) {
291 PROC_LOCK(p);
292 if (thread_single(SINGLE_BOUNDARY)) {
293 PROC_UNLOCK(p);
294 exec_free_args(args);
295 return (ERESTART); /* Try again later. */
296 }
297 PROC_UNLOCK(p);
298 }
299
300 KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
301 oldvmspace = td->td_proc->p_vmspace;
302 error = do_execve(td, args, mac_p);
303
304 if (p->p_flag & P_HADTHREADS) {
305 PROC_LOCK(p);
306 /*
307 * If success, we upgrade to SINGLE_EXIT state to
308 * force other threads to suicide.
309 */
310 if (error == 0)
311 thread_single(SINGLE_EXIT);
312 else
313 thread_single_end();
314 PROC_UNLOCK(p);
315 }
316 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
317 KASSERT(td->td_proc->p_vmspace != oldvmspace,
318 ("oldvmspace still used"));
319 vmspace_free(oldvmspace);
320 td->td_pflags &= ~TDP_EXECVMSPC;
321 }
322
323 return (error);
324 }
325
326 /*
327 * In-kernel implementation of execve(). All arguments are assumed to be
328 * userspace pointers from the passed thread.
329 */
330 static int
331 do_execve(td, args, mac_p)
332 struct thread *td;
333 struct image_args *args;
334 struct mac *mac_p;
335 {
336 struct proc *p = td->td_proc;
337 struct nameidata nd;
338 struct ucred *newcred = NULL, *oldcred;
339 struct uidinfo *euip;
340 register_t *stack_base;
341 int error, i;
342 struct image_params image_params, *imgp;
343 struct vattr attr;
344 int (*img_first)(struct image_params *);
345 struct pargs *oldargs = NULL, *newargs = NULL;
346 struct sigacts *oldsigacts, *newsigacts;
347 #ifdef KTRACE
348 struct vnode *tracevp = NULL;
349 struct ucred *tracecred = NULL;
350 #endif
351 struct vnode *textvp = NULL, *binvp = NULL;
352 int credential_changing;
353 int vfslocked;
354 int textset;
355 #ifdef MAC
356 struct label *interpvplabel = NULL;
357 int will_transition;
358 #endif
359 #ifdef HWPMC_HOOKS
360 struct pmckern_procexec pe;
361 #endif
362 static const char fexecv_proc_title[] = "(fexecv)";
363
364 vfslocked = 0;
365 imgp = &image_params;
366
367 /*
368 * Lock the process and set the P_INEXEC flag to indicate that
369 * it should be left alone until we're done here. This is
370 * necessary to avoid race conditions - e.g. in ptrace() -
371 * that might allow a local user to illicitly obtain elevated
372 * privileges.
373 */
374 PROC_LOCK(p);
375 KASSERT((p->p_flag & P_INEXEC) == 0,
376 ("%s(): process already has P_INEXEC flag", __func__));
377 p->p_flag |= P_INEXEC;
378 PROC_UNLOCK(p);
379
380 /*
381 * Initialize part of the common data
382 */
383 imgp->proc = p;
384 imgp->execlabel = NULL;
385 imgp->attr = &attr;
386 imgp->entry_addr = 0;
387 imgp->reloc_base = 0;
388 imgp->vmspace_destroyed = 0;
389 imgp->interpreted = 0;
390 imgp->opened = 0;
391 imgp->interpreter_name = NULL;
392 imgp->auxargs = NULL;
393 imgp->vp = NULL;
394 imgp->object = NULL;
395 imgp->firstpage = NULL;
396 imgp->ps_strings = 0;
397 imgp->auxarg_size = 0;
398 imgp->args = args;
399 imgp->execpath = imgp->freepath = NULL;
400 imgp->execpathp = 0;
401 imgp->canary = 0;
402 imgp->canarylen = 0;
403 imgp->pagesizes = 0;
404 imgp->pagesizeslen = 0;
405 imgp->stack_prot = 0;
406
407 #ifdef MAC
408 error = mac_execve_enter(imgp, mac_p);
409 if (error)
410 goto exec_fail;
411 #endif
412
413 imgp->image_header = NULL;
414
415 /*
416 * Translate the file name. namei() returns a vnode pointer
417 * in ni_vp amoung 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 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
425 }
426
427 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
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 vfslocked = NDHASGIANT(&nd);
448 binvp = nd.ni_vp;
449 imgp->vp = binvp;
450 } else {
451 AUDIT_ARG_FD(args->fd);
452 /*
453 * Some might argue that CAP_READ and/or CAP_MMAP should also
454 * be required here; such arguments will be entertained.
455 *
456 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
457 */
458 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
459 if (error)
460 goto exec_fail;
461 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
462 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
463 AUDIT_ARG_VNODE1(binvp);
464 imgp->vp = binvp;
465 }
466
467 /*
468 * Check file permissions (also 'opens' file)
469 */
470 error = exec_check_permissions(imgp);
471 if (error)
472 goto exec_fail_dealloc;
473
474 imgp->object = imgp->vp->v_object;
475 if (imgp->object != NULL)
476 vm_object_reference(imgp->object);
477
478 /*
479 * Set VV_TEXT now so no one can write to the executable while we're
480 * activating it.
481 *
482 * Remember if this was set before and unset it in case this is not
483 * actually an executable image.
484 */
485 textset = imgp->vp->v_vflag & VV_TEXT;
486 ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
487 imgp->vp->v_vflag |= VV_TEXT;
488
489 error = exec_map_first_page(imgp);
490 if (error)
491 goto exec_fail_dealloc;
492
493 imgp->proc->p_osrel = 0;
494 /*
495 * If the current process has a special image activator it
496 * wants to try first, call it. For example, emulating shell
497 * scripts differently.
498 */
499 error = -1;
500 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
501 error = img_first(imgp);
502
503 /*
504 * Loop through the list of image activators, calling each one.
505 * An activator returns -1 if there is no match, 0 on success,
506 * and an error otherwise.
507 */
508 for (i = 0; error == -1 && execsw[i]; ++i) {
509 if (execsw[i]->ex_imgact == NULL ||
510 execsw[i]->ex_imgact == img_first) {
511 continue;
512 }
513 error = (*execsw[i]->ex_imgact)(imgp);
514 }
515
516 if (error) {
517 if (error == -1) {
518 if (textset == 0) {
519 ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
520 imgp->vp->v_vflag &= ~VV_TEXT;
521 }
522 error = ENOEXEC;
523 }
524 goto exec_fail_dealloc;
525 }
526
527 /*
528 * Special interpreter operation, cleanup and loop up to try to
529 * activate the interpreter.
530 */
531 if (imgp->interpreted) {
532 exec_unmap_first_page(imgp);
533 /*
534 * VV_TEXT needs to be unset for scripts. There is a short
535 * period before we determine that something is a script where
536 * VV_TEXT will be set. The vnode lock is held over this
537 * entire period so nothing should illegitimately be blocked.
538 */
539 imgp->vp->v_vflag &= ~VV_TEXT;
540 /* free name buffer and old vnode */
541 if (args->fname != NULL)
542 NDFREE(&nd, NDF_ONLY_PNBUF);
543 #ifdef MAC
544 mac_execve_interpreter_enter(binvp, &interpvplabel);
545 #endif
546 if (imgp->opened) {
547 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
548 imgp->opened = 0;
549 }
550 vput(binvp);
551 vm_object_deallocate(imgp->object);
552 imgp->object = NULL;
553 VFS_UNLOCK_GIANT(vfslocked);
554 vfslocked = 0;
555 /* set new name to that of the interpreter */
556 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
557 UIO_SYSSPACE, imgp->interpreter_name, td);
558 args->fname = imgp->interpreter_name;
559 goto interpret;
560 }
561
562 /*
563 * NB: We unlock the vnode here because it is believed that none
564 * of the sv_copyout_strings/sv_fixup operations require the vnode.
565 */
566 VOP_UNLOCK(imgp->vp, 0);
567
568 /*
569 * Do the best to calculate the full path to the image file.
570 */
571 if (imgp->auxargs != NULL &&
572 ((args->fname != NULL && args->fname[0] == '/') ||
573 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
574 imgp->execpath = args->fname;
575
576 /*
577 * Copy out strings (args and env) and initialize stack base
578 */
579 if (p->p_sysent->sv_copyout_strings)
580 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
581 else
582 stack_base = exec_copyout_strings(imgp);
583
584 /*
585 * If custom stack fixup routine present for this process
586 * let it do the stack setup.
587 * Else stuff argument count as first item on stack
588 */
589 if (p->p_sysent->sv_fixup != NULL)
590 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
591 else
592 suword(--stack_base, imgp->args->argc);
593
594 /*
595 * For security and other reasons, the file descriptor table cannot
596 * be shared after an exec.
597 */
598 fdunshare(p, td);
599
600 /*
601 * Malloc things before we need locks.
602 */
603 newcred = crget();
604 euip = uifind(attr.va_uid);
605 i = imgp->args->begin_envv - imgp->args->begin_argv;
606 /* Cache arguments if they fit inside our allowance */
607 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
608 newargs = pargs_alloc(i);
609 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
610 }
611
612 /* close files on exec */
613 fdcloseexec(td);
614 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
615
616 /* Get a reference to the vnode prior to locking the proc */
617 VREF(binvp);
618
619 /*
620 * For security and other reasons, signal handlers cannot
621 * be shared after an exec. The new process gets a copy of the old
622 * handlers. In execsigs(), the new process will have its signals
623 * reset.
624 */
625 PROC_LOCK(p);
626 oldcred = crcopysafe(p, newcred);
627 if (sigacts_shared(p->p_sigacts)) {
628 oldsigacts = p->p_sigacts;
629 PROC_UNLOCK(p);
630 newsigacts = sigacts_alloc();
631 sigacts_copy(newsigacts, oldsigacts);
632 PROC_LOCK(p);
633 p->p_sigacts = newsigacts;
634 } else
635 oldsigacts = NULL;
636
637 /* Stop profiling */
638 stopprofclock(p);
639
640 /* reset caught signals */
641 execsigs(p);
642
643 /* name this process - nameiexec(p, ndp) */
644 bzero(p->p_comm, sizeof(p->p_comm));
645 if (args->fname)
646 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
647 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
648 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
649 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
650 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
651 #ifdef KTR
652 sched_clear_tdname(td);
653 #endif
654
655 /*
656 * mark as execed, wakeup the process that vforked (if any) and tell
657 * it that it now has its own resources back
658 */
659 p->p_flag |= P_EXEC;
660 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
661 p->p_flag &= ~P_PPWAIT;
662 cv_broadcast(&p->p_pwait);
663 }
664
665 /*
666 * Implement image setuid/setgid.
667 *
668 * Don't honor setuid/setgid if the filesystem prohibits it or if
669 * the process is being traced.
670 *
671 * We disable setuid/setgid/etc in compatibility mode on the basis
672 * that most setugid applications are not written with that
673 * environment in mind, and will therefore almost certainly operate
674 * incorrectly. In principle there's no reason that setugid
675 * applications might not be useful in capability mode, so we may want
676 * to reconsider this conservative design choice in the future.
677 *
678 * XXXMAC: For the time being, use NOSUID to also prohibit
679 * transitions on the file system.
680 */
681 credential_changing = 0;
682 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
683 attr.va_uid;
684 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
685 attr.va_gid;
686 #ifdef MAC
687 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
688 interpvplabel, imgp);
689 credential_changing |= will_transition;
690 #endif
691
692 if (credential_changing &&
693 #ifdef CAPABILITY_MODE
694 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
695 #endif
696 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
697 (p->p_flag & P_TRACED) == 0) {
698 /*
699 * Turn off syscall tracing for set-id programs, except for
700 * root. Record any set-id flags first to make sure that
701 * we do not regain any tracing during a possible block.
702 */
703 setsugid(p);
704
705 #ifdef KTRACE
706 if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0))
707 ktrprocexec(p, &tracecred, &tracevp);
708 #endif
709 /*
710 * Close any file descriptors 0..2 that reference procfs,
711 * then make sure file descriptors 0..2 are in use.
712 *
713 * setugidsafety() may call closef() and then pfind()
714 * which may grab the process lock.
715 * fdcheckstd() may call falloc() which may block to
716 * allocate memory, so temporarily drop the process lock.
717 */
718 PROC_UNLOCK(p);
719 VOP_UNLOCK(imgp->vp, 0);
720 setugidsafety(td);
721 error = fdcheckstd(td);
722 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
723 if (error != 0)
724 goto done1;
725 PROC_LOCK(p);
726 /*
727 * Set the new credentials.
728 */
729 if (attr.va_mode & S_ISUID)
730 change_euid(newcred, euip);
731 if (attr.va_mode & S_ISGID)
732 change_egid(newcred, attr.va_gid);
733 #ifdef MAC
734 if (will_transition) {
735 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
736 interpvplabel, imgp);
737 }
738 #endif
739 /*
740 * Implement correct POSIX saved-id behavior.
741 *
742 * XXXMAC: Note that the current logic will save the
743 * uid and gid if a MAC domain transition occurs, even
744 * though maybe it shouldn't.
745 */
746 change_svuid(newcred, newcred->cr_uid);
747 change_svgid(newcred, newcred->cr_gid);
748 p->p_ucred = newcred;
749 newcred = NULL;
750 } else {
751 if (oldcred->cr_uid == oldcred->cr_ruid &&
752 oldcred->cr_gid == oldcred->cr_rgid)
753 p->p_flag &= ~P_SUGID;
754 /*
755 * Implement correct POSIX saved-id behavior.
756 *
757 * XXX: It's not clear that the existing behavior is
758 * POSIX-compliant. A number of sources indicate that the
759 * saved uid/gid should only be updated if the new ruid is
760 * not equal to the old ruid, or the new euid is not equal
761 * to the old euid and the new euid is not equal to the old
762 * ruid. The FreeBSD code always updates the saved uid/gid.
763 * Also, this code uses the new (replaced) euid and egid as
764 * the source, which may or may not be the right ones to use.
765 */
766 if (oldcred->cr_svuid != oldcred->cr_uid ||
767 oldcred->cr_svgid != oldcred->cr_gid) {
768 change_svuid(newcred, newcred->cr_uid);
769 change_svgid(newcred, newcred->cr_gid);
770 p->p_ucred = newcred;
771 newcred = NULL;
772 }
773 }
774
775 /*
776 * Store the vp for use in procfs. This vnode was referenced prior
777 * to locking the proc lock.
778 */
779 textvp = p->p_textvp;
780 p->p_textvp = binvp;
781
782 #ifdef KDTRACE_HOOKS
783 /*
784 * Tell the DTrace fasttrap provider about the exec if it
785 * has declared an interest.
786 */
787 if (dtrace_fasttrap_exec)
788 dtrace_fasttrap_exec(p);
789 #endif
790
791 /*
792 * Notify others that we exec'd, and clear the P_INEXEC flag
793 * as we're now a bona fide freshly-execed process.
794 */
795 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
796 p->p_flag &= ~P_INEXEC;
797
798 /* clear "fork but no exec" flag, as we _are_ execing */
799 p->p_acflag &= ~AFORK;
800
801 /*
802 * Free any previous argument cache and replace it with
803 * the new argument cache, if any.
804 */
805 oldargs = p->p_args;
806 p->p_args = newargs;
807 newargs = NULL;
808
809 #ifdef HWPMC_HOOKS
810 /*
811 * Check if system-wide sampling is in effect or if the
812 * current process is using PMCs. If so, do exec() time
813 * processing. This processing needs to happen AFTER the
814 * P_INEXEC flag is cleared.
815 *
816 * The proc lock needs to be released before taking the PMC
817 * SX.
818 */
819 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
820 PROC_UNLOCK(p);
821 VOP_UNLOCK(imgp->vp, 0);
822 pe.pm_credentialschanged = credential_changing;
823 pe.pm_entryaddr = imgp->entry_addr;
824
825 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
826 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
827 } else
828 PROC_UNLOCK(p);
829 #else /* !HWPMC_HOOKS */
830 PROC_UNLOCK(p);
831 #endif
832
833 /* Set values passed into the program in registers. */
834 if (p->p_sysent->sv_setregs)
835 (*p->p_sysent->sv_setregs)(td, imgp,
836 (u_long)(uintptr_t)stack_base);
837 else
838 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
839
840 vfs_mark_atime(imgp->vp, td->td_ucred);
841
842 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
843
844 done1:
845 /*
846 * Free any resources malloc'd earlier that we didn't use.
847 */
848 uifree(euip);
849 if (newcred == NULL)
850 crfree(oldcred);
851 else
852 crfree(newcred);
853 VOP_UNLOCK(imgp->vp, 0);
854
855 /*
856 * Handle deferred decrement of ref counts.
857 */
858 if (textvp != NULL) {
859 int tvfslocked;
860
861 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
862 vrele(textvp);
863 VFS_UNLOCK_GIANT(tvfslocked);
864 }
865 if (binvp && error != 0)
866 vrele(binvp);
867 #ifdef KTRACE
868 if (tracevp != NULL) {
869 int tvfslocked;
870
871 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
872 vrele(tracevp);
873 VFS_UNLOCK_GIANT(tvfslocked);
874 }
875 if (tracecred != NULL)
876 crfree(tracecred);
877 #endif
878 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
879 pargs_drop(oldargs);
880 pargs_drop(newargs);
881 if (oldsigacts != NULL)
882 sigacts_free(oldsigacts);
883
884 exec_fail_dealloc:
885
886 /*
887 * free various allocated resources
888 */
889 if (imgp->firstpage != NULL)
890 exec_unmap_first_page(imgp);
891
892 if (imgp->vp != NULL) {
893 if (args->fname)
894 NDFREE(&nd, NDF_ONLY_PNBUF);
895 if (imgp->opened)
896 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
897 vput(imgp->vp);
898 }
899
900 if (imgp->object != NULL)
901 vm_object_deallocate(imgp->object);
902
903 free(imgp->freepath, M_TEMP);
904
905 if (error == 0) {
906 PROC_LOCK(p);
907 td->td_dbgflags |= TDB_EXEC;
908 PROC_UNLOCK(p);
909
910 /*
911 * Stop the process here if its stop event mask has
912 * the S_EXEC bit set.
913 */
914 STOPEVENT(p, S_EXEC, 0);
915 goto done2;
916 }
917
918 exec_fail:
919 /* we're done here, clear P_INEXEC */
920 PROC_LOCK(p);
921 p->p_flag &= ~P_INEXEC;
922 PROC_UNLOCK(p);
923
924 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
925
926 done2:
927 #ifdef MAC
928 mac_execve_exit(imgp);
929 mac_execve_interpreter_exit(interpvplabel);
930 #endif
931 VFS_UNLOCK_GIANT(vfslocked);
932 exec_free_args(args);
933
934 if (error && imgp->vmspace_destroyed) {
935 /* sorry, no more process anymore. exit gracefully */
936 exit1(td, W_EXITCODE(0, SIGABRT));
937 /* NOT REACHED */
938 }
939
940 #ifdef KTRACE
941 if (error == 0)
942 ktrprocctor(p);
943 #endif
944
945 return (error);
946 }
947
948 int
949 exec_map_first_page(imgp)
950 struct image_params *imgp;
951 {
952 int rv, i;
953 int initial_pagein;
954 vm_page_t ma[VM_INITIAL_PAGEIN];
955 vm_object_t object;
956
957 if (imgp->firstpage != NULL)
958 exec_unmap_first_page(imgp);
959
960 object = imgp->vp->v_object;
961 if (object == NULL)
962 return (EACCES);
963 VM_OBJECT_LOCK(object);
964 #if VM_NRESERVLEVEL > 0
965 if ((object->flags & OBJ_COLORED) == 0) {
966 object->flags |= OBJ_COLORED;
967 object->pg_color = 0;
968 }
969 #endif
970 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
971 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
972 initial_pagein = VM_INITIAL_PAGEIN;
973 if (initial_pagein > object->size)
974 initial_pagein = object->size;
975 for (i = 1; i < initial_pagein; i++) {
976 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
977 if (ma[i]->valid)
978 break;
979 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
980 break;
981 vm_page_busy(ma[i]);
982 } else {
983 ma[i] = vm_page_alloc(object, i,
984 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
985 if (ma[i] == NULL)
986 break;
987 }
988 }
989 initial_pagein = i;
990 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
991 ma[0] = vm_page_lookup(object, 0);
992 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
993 if (ma[0] != NULL) {
994 vm_page_lock(ma[0]);
995 vm_page_free(ma[0]);
996 vm_page_unlock(ma[0]);
997 }
998 VM_OBJECT_UNLOCK(object);
999 return (EIO);
1000 }
1001 }
1002 vm_page_lock(ma[0]);
1003 vm_page_hold(ma[0]);
1004 vm_page_unlock(ma[0]);
1005 vm_page_wakeup(ma[0]);
1006 VM_OBJECT_UNLOCK(object);
1007
1008 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1009 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1010
1011 return (0);
1012 }
1013
1014 void
1015 exec_unmap_first_page(imgp)
1016 struct image_params *imgp;
1017 {
1018 vm_page_t m;
1019
1020 if (imgp->firstpage != NULL) {
1021 m = sf_buf_page(imgp->firstpage);
1022 sf_buf_free(imgp->firstpage);
1023 imgp->firstpage = NULL;
1024 vm_page_lock(m);
1025 vm_page_unhold(m);
1026 vm_page_unlock(m);
1027 }
1028 }
1029
1030 /*
1031 * Destroy old address space, and allocate a new stack
1032 * The new stack is only SGROWSIZ large because it is grown
1033 * automatically in trap.c.
1034 */
1035 int
1036 exec_new_vmspace(imgp, sv)
1037 struct image_params *imgp;
1038 struct sysentvec *sv;
1039 {
1040 int error;
1041 struct proc *p = imgp->proc;
1042 struct vmspace *vmspace = p->p_vmspace;
1043 vm_object_t obj;
1044 vm_offset_t sv_minuser, stack_addr;
1045 vm_map_t map;
1046 u_long ssiz;
1047
1048 imgp->vmspace_destroyed = 1;
1049 imgp->sysent = sv;
1050
1051 /* May be called with Giant held */
1052 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1053
1054 /*
1055 * Blow away entire process VM, if address space not shared,
1056 * otherwise, create a new VM space so that other threads are
1057 * not disrupted
1058 */
1059 map = &vmspace->vm_map;
1060 if (map_at_zero)
1061 sv_minuser = sv->sv_minuser;
1062 else
1063 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1064 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1065 vm_map_max(map) == sv->sv_maxuser) {
1066 shmexit(vmspace);
1067 pmap_remove_pages(vmspace_pmap(vmspace));
1068 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1069 } else {
1070 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1071 if (error)
1072 return (error);
1073 vmspace = p->p_vmspace;
1074 map = &vmspace->vm_map;
1075 }
1076
1077 /* Map a shared page */
1078 obj = sv->sv_shared_page_obj;
1079 if (obj != NULL) {
1080 vm_object_reference(obj);
1081 error = vm_map_fixed(map, obj, 0,
1082 sv->sv_shared_page_base, sv->sv_shared_page_len,
1083 VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL,
1084 MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE);
1085 if (error) {
1086 vm_object_deallocate(obj);
1087 return (error);
1088 }
1089 }
1090
1091 /* Allocate a new stack */
1092 if (sv->sv_maxssiz != NULL)
1093 ssiz = *sv->sv_maxssiz;
1094 else
1095 ssiz = maxssiz;
1096 stack_addr = sv->sv_usrstack - ssiz;
1097 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1098 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1099 sv->sv_stackprot,
1100 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1101 if (error)
1102 return (error);
1103
1104 #ifdef __ia64__
1105 /* Allocate a new register stack */
1106 stack_addr = IA64_BACKINGSTORE;
1107 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1108 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1109 if (error)
1110 return (error);
1111 #endif
1112
1113 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1114 * VM_STACK case, but they are still used to monitor the size of the
1115 * process stack so we can check the stack rlimit.
1116 */
1117 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1118 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1119
1120 return (0);
1121 }
1122
1123 /*
1124 * Copy out argument and environment strings from the old process address
1125 * space into the temporary string buffer.
1126 */
1127 int
1128 exec_copyin_args(struct image_args *args, char *fname,
1129 enum uio_seg segflg, char **argv, char **envv)
1130 {
1131 char *argp, *envp;
1132 int error;
1133 size_t length;
1134
1135 bzero(args, sizeof(*args));
1136 if (argv == NULL)
1137 return (EFAULT);
1138
1139 /*
1140 * Allocate demand-paged memory for the file name, argument, and
1141 * environment strings.
1142 */
1143 error = exec_alloc_args(args);
1144 if (error != 0)
1145 return (error);
1146
1147 /*
1148 * Copy the file name.
1149 */
1150 if (fname != NULL) {
1151 args->fname = args->buf;
1152 error = (segflg == UIO_SYSSPACE) ?
1153 copystr(fname, args->fname, PATH_MAX, &length) :
1154 copyinstr(fname, args->fname, PATH_MAX, &length);
1155 if (error != 0)
1156 goto err_exit;
1157 } else
1158 length = 0;
1159
1160 args->begin_argv = args->buf + length;
1161 args->endp = args->begin_argv;
1162 args->stringspace = ARG_MAX;
1163
1164 /*
1165 * extract arguments first
1166 */
1167 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1168 if (argp == (caddr_t) -1) {
1169 error = EFAULT;
1170 goto err_exit;
1171 }
1172 if ((error = copyinstr(argp, args->endp,
1173 args->stringspace, &length))) {
1174 if (error == ENAMETOOLONG)
1175 error = E2BIG;
1176 goto err_exit;
1177 }
1178 args->stringspace -= length;
1179 args->endp += length;
1180 args->argc++;
1181 }
1182
1183 args->begin_envv = args->endp;
1184
1185 /*
1186 * extract environment strings
1187 */
1188 if (envv) {
1189 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1190 if (envp == (caddr_t)-1) {
1191 error = EFAULT;
1192 goto err_exit;
1193 }
1194 if ((error = copyinstr(envp, args->endp,
1195 args->stringspace, &length))) {
1196 if (error == ENAMETOOLONG)
1197 error = E2BIG;
1198 goto err_exit;
1199 }
1200 args->stringspace -= length;
1201 args->endp += length;
1202 args->envc++;
1203 }
1204 }
1205
1206 return (0);
1207
1208 err_exit:
1209 exec_free_args(args);
1210 return (error);
1211 }
1212
1213 /*
1214 * Allocate temporary demand-paged, zero-filled memory for the file name,
1215 * argument, and environment strings. Returns zero if the allocation succeeds
1216 * and ENOMEM otherwise.
1217 */
1218 int
1219 exec_alloc_args(struct image_args *args)
1220 {
1221
1222 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1223 return (args->buf != NULL ? 0 : ENOMEM);
1224 }
1225
1226 void
1227 exec_free_args(struct image_args *args)
1228 {
1229
1230 if (args->buf != NULL) {
1231 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1232 PATH_MAX + ARG_MAX);
1233 args->buf = NULL;
1234 }
1235 if (args->fname_buf != NULL) {
1236 free(args->fname_buf, M_TEMP);
1237 args->fname_buf = NULL;
1238 }
1239 }
1240
1241 /*
1242 * Copy strings out to the new process address space, constructing new arg
1243 * and env vector tables. Return a pointer to the base so that it can be used
1244 * as the initial stack pointer.
1245 */
1246 register_t *
1247 exec_copyout_strings(imgp)
1248 struct image_params *imgp;
1249 {
1250 int argc, envc;
1251 char **vectp;
1252 char *stringp, *destp;
1253 register_t *stack_base;
1254 struct ps_strings *arginfo;
1255 struct proc *p;
1256 size_t execpath_len;
1257 int szsigcode, szps;
1258 char canary[sizeof(long) * 8];
1259
1260 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1261 /*
1262 * Calculate string base and vector table pointers.
1263 * Also deal with signal trampoline code for this exec type.
1264 */
1265 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1266 execpath_len = strlen(imgp->execpath) + 1;
1267 else
1268 execpath_len = 0;
1269 p = imgp->proc;
1270 szsigcode = 0;
1271 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1272 if (p->p_sysent->sv_sigcode_base == 0) {
1273 if (p->p_sysent->sv_szsigcode != NULL)
1274 szsigcode = *(p->p_sysent->sv_szsigcode);
1275 }
1276 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1277 roundup(execpath_len, sizeof(char *)) -
1278 roundup(sizeof(canary), sizeof(char *)) -
1279 roundup(szps, sizeof(char *)) -
1280 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1281
1282 /*
1283 * install sigcode
1284 */
1285 if (szsigcode != 0)
1286 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1287 szsigcode), szsigcode);
1288
1289 /*
1290 * Copy the image path for the rtld.
1291 */
1292 if (execpath_len != 0) {
1293 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1294 copyout(imgp->execpath, (void *)imgp->execpathp,
1295 execpath_len);
1296 }
1297
1298 /*
1299 * Prepare the canary for SSP.
1300 */
1301 arc4rand(canary, sizeof(canary), 0);
1302 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1303 sizeof(canary);
1304 copyout(canary, (void *)imgp->canary, sizeof(canary));
1305 imgp->canarylen = sizeof(canary);
1306
1307 /*
1308 * Prepare the pagesizes array.
1309 */
1310 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1311 roundup(sizeof(canary), sizeof(char *)) - szps;
1312 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1313 imgp->pagesizeslen = szps;
1314
1315 /*
1316 * If we have a valid auxargs ptr, prepare some room
1317 * on the stack.
1318 */
1319 if (imgp->auxargs) {
1320 /*
1321 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1322 * lower compatibility.
1323 */
1324 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1325 (AT_COUNT * 2);
1326 /*
1327 * The '+ 2' is for the null pointers at the end of each of
1328 * the arg and env vector sets,and imgp->auxarg_size is room
1329 * for argument of Runtime loader.
1330 */
1331 vectp = (char **)(destp - (imgp->args->argc +
1332 imgp->args->envc + 2 + imgp->auxarg_size)
1333 * sizeof(char *));
1334 } else {
1335 /*
1336 * The '+ 2' is for the null pointers at the end of each of
1337 * the arg and env vector sets
1338 */
1339 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1340 sizeof(char *));
1341 }
1342
1343 /*
1344 * vectp also becomes our initial stack base
1345 */
1346 stack_base = (register_t *)vectp;
1347
1348 stringp = imgp->args->begin_argv;
1349 argc = imgp->args->argc;
1350 envc = imgp->args->envc;
1351
1352 /*
1353 * Copy out strings - arguments and environment.
1354 */
1355 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1356
1357 /*
1358 * Fill in "ps_strings" struct for ps, w, etc.
1359 */
1360 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1361 suword32(&arginfo->ps_nargvstr, argc);
1362
1363 /*
1364 * Fill in argument portion of vector table.
1365 */
1366 for (; argc > 0; --argc) {
1367 suword(vectp++, (long)(intptr_t)destp);
1368 while (*stringp++ != 0)
1369 destp++;
1370 destp++;
1371 }
1372
1373 /* a null vector table pointer separates the argp's from the envp's */
1374 suword(vectp++, 0);
1375
1376 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1377 suword32(&arginfo->ps_nenvstr, envc);
1378
1379 /*
1380 * Fill in environment portion of vector table.
1381 */
1382 for (; envc > 0; --envc) {
1383 suword(vectp++, (long)(intptr_t)destp);
1384 while (*stringp++ != 0)
1385 destp++;
1386 destp++;
1387 }
1388
1389 /* end of vector table is a null pointer */
1390 suword(vectp, 0);
1391
1392 return (stack_base);
1393 }
1394
1395 /*
1396 * Check permissions of file to execute.
1397 * Called with imgp->vp locked.
1398 * Return 0 for success or error code on failure.
1399 */
1400 int
1401 exec_check_permissions(imgp)
1402 struct image_params *imgp;
1403 {
1404 struct vnode *vp = imgp->vp;
1405 struct vattr *attr = imgp->attr;
1406 struct thread *td;
1407 int error;
1408
1409 td = curthread;
1410
1411 /* Get file attributes */
1412 error = VOP_GETATTR(vp, attr, td->td_ucred);
1413 if (error)
1414 return (error);
1415
1416 #ifdef MAC
1417 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1418 if (error)
1419 return (error);
1420 #endif
1421
1422 /*
1423 * 1) Check if file execution is disabled for the filesystem that
1424 * this file resides on.
1425 * 2) Ensure that at least one execute bit is on. Otherwise, a
1426 * privileged user will always succeed, and we don't want this
1427 * to happen unless the file really is executable.
1428 * 3) Ensure that the file is a regular file.
1429 */
1430 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1431 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1432 (attr->va_type != VREG))
1433 return (EACCES);
1434
1435 /*
1436 * Zero length files can't be exec'd
1437 */
1438 if (attr->va_size == 0)
1439 return (ENOEXEC);
1440
1441 /*
1442 * Check for execute permission to file based on current credentials.
1443 */
1444 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1445 if (error)
1446 return (error);
1447
1448 /*
1449 * Check number of open-for-writes on the file and deny execution
1450 * if there are any.
1451 */
1452 if (vp->v_writecount)
1453 return (ETXTBSY);
1454
1455 /*
1456 * Call filesystem specific open routine (which does nothing in the
1457 * general case).
1458 */
1459 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1460 if (error == 0)
1461 imgp->opened = 1;
1462 return (error);
1463 }
1464
1465 /*
1466 * Exec handler registration
1467 */
1468 int
1469 exec_register(execsw_arg)
1470 const struct execsw *execsw_arg;
1471 {
1472 const struct execsw **es, **xs, **newexecsw;
1473 int count = 2; /* New slot and trailing NULL */
1474
1475 if (execsw)
1476 for (es = execsw; *es; es++)
1477 count++;
1478 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1479 if (newexecsw == NULL)
1480 return (ENOMEM);
1481 xs = newexecsw;
1482 if (execsw)
1483 for (es = execsw; *es; es++)
1484 *xs++ = *es;
1485 *xs++ = execsw_arg;
1486 *xs = NULL;
1487 if (execsw)
1488 free(execsw, M_TEMP);
1489 execsw = newexecsw;
1490 return (0);
1491 }
1492
1493 int
1494 exec_unregister(execsw_arg)
1495 const struct execsw *execsw_arg;
1496 {
1497 const struct execsw **es, **xs, **newexecsw;
1498 int count = 1;
1499
1500 if (execsw == NULL)
1501 panic("unregister with no handlers left?\n");
1502
1503 for (es = execsw; *es; es++) {
1504 if (*es == execsw_arg)
1505 break;
1506 }
1507 if (*es == NULL)
1508 return (ENOENT);
1509 for (es = execsw; *es; es++)
1510 if (*es != execsw_arg)
1511 count++;
1512 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1513 if (newexecsw == NULL)
1514 return (ENOMEM);
1515 xs = newexecsw;
1516 for (es = execsw; *es; es++)
1517 if (*es != execsw_arg)
1518 *xs++ = *es;
1519 *xs = NULL;
1520 if (execsw)
1521 free(execsw, M_TEMP);
1522 execsw = newexecsw;
1523 return (0);
1524 }
1525
1526 static vm_object_t shared_page_obj;
1527 static int shared_page_free;
1528
1529 int
1530 shared_page_fill(int size, int align, const char *data)
1531 {
1532 vm_page_t m;
1533 struct sf_buf *s;
1534 vm_offset_t sk;
1535 int res;
1536
1537 VM_OBJECT_LOCK(shared_page_obj);
1538 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY);
1539 res = roundup(shared_page_free, align);
1540 if (res + size >= IDX_TO_OFF(shared_page_obj->size))
1541 res = -1;
1542 else {
1543 VM_OBJECT_UNLOCK(shared_page_obj);
1544 s = sf_buf_alloc(m, SFB_DEFAULT);
1545 sk = sf_buf_kva(s);
1546 bcopy(data, (void *)(sk + res), size);
1547 shared_page_free = res + size;
1548 sf_buf_free(s);
1549 VM_OBJECT_LOCK(shared_page_obj);
1550 }
1551 vm_page_wakeup(m);
1552 VM_OBJECT_UNLOCK(shared_page_obj);
1553 return (res);
1554 }
1555
1556 static void
1557 shared_page_init(void *dummy __unused)
1558 {
1559 vm_page_t m;
1560
1561 shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE,
1562 VM_PROT_DEFAULT, 0, NULL);
1563 VM_OBJECT_LOCK(shared_page_obj);
1564 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY | VM_ALLOC_NOBUSY |
1565 VM_ALLOC_ZERO);
1566 m->valid = VM_PAGE_BITS_ALL;
1567 VM_OBJECT_UNLOCK(shared_page_obj);
1568 }
1569
1570 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)shared_page_init,
1571 NULL);
1572
1573 void
1574 exec_sysvec_init(void *param)
1575 {
1576 struct sysentvec *sv;
1577
1578 sv = (struct sysentvec *)param;
1579
1580 if ((sv->sv_flags & SV_SHP) == 0)
1581 return;
1582 sv->sv_shared_page_obj = shared_page_obj;
1583 sv->sv_sigcode_base = sv->sv_shared_page_base +
1584 shared_page_fill(*(sv->sv_szsigcode), 16, sv->sv_sigcode);
1585 }
Cache object: 417d8ae9074083e98e73f53388bfa8e3
|