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/7.1/sys/kern/kern_exec.c 197715 2009-10-02 18:09:56Z simon $");
29
30 #include "opt_hwpmc_hooks.h"
31 #include "opt_kdtrace.h"
32 #include "opt_ktrace.h"
33 #include "opt_mac.h"
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/eventhandler.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/sysproto.h>
41 #include <sys/signalvar.h>
42 #include <sys/kernel.h>
43 #include <sys/mount.h>
44 #include <sys/filedesc.h>
45 #include <sys/fcntl.h>
46 #include <sys/acct.h>
47 #include <sys/exec.h>
48 #include <sys/imgact.h>
49 #include <sys/imgact_elf.h>
50 #include <sys/wait.h>
51 #include <sys/malloc.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/pioctl.h>
55 #include <sys/namei.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sdt.h>
58 #include <sys/sf_buf.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysent.h>
61 #include <sys/shm.h>
62 #include <sys/sysctl.h>
63 #include <sys/vnode.h>
64 #ifdef KTRACE
65 #include <sys/ktrace.h>
66 #endif
67
68 #include <vm/vm.h>
69 #include <vm/vm_param.h>
70 #include <vm/pmap.h>
71 #include <vm/vm_page.h>
72 #include <vm/vm_map.h>
73 #include <vm/vm_kern.h>
74 #include <vm/vm_extern.h>
75 #include <vm/vm_object.h>
76 #include <vm/vm_pager.h>
77
78 #ifdef HWPMC_HOOKS
79 #include <sys/pmckern.h>
80 #endif
81
82 #include <machine/reg.h>
83
84 #include <security/audit/audit.h>
85 #include <security/mac/mac_framework.h>
86
87 #ifdef KDTRACE_HOOKS
88 #include <sys/dtrace_bsd.h>
89 dtrace_execexit_func_t dtrace_fasttrap_exec;
90 #endif
91
92 SDT_PROVIDER_DECLARE(proc);
93 SDT_PROBE_DEFINE(proc, kernel, , exec);
94 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
95 SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
96 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
97 SDT_PROBE_DEFINE(proc, kernel, , exec_success);
98 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
99
100 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
101
102 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
103 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
105 static int do_execve(struct thread *td, struct image_args *args,
106 struct mac *mac_p);
107 static void exec_free_args(struct image_args *);
108
109 /* XXX This should be vm_size_t. */
110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
111 NULL, 0, sysctl_kern_ps_strings, "LU", "");
112
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
115 NULL, 0, sysctl_kern_usrstack, "LU", "");
116
117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
118 NULL, 0, sysctl_kern_stackprot, "I", "");
119
120 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
122 &ps_arg_cache_limit, 0, "");
123
124 static int map_at_zero = 1;
125 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
126 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
127 "Permit processes to map an object at virtual address 0.");
128
129 static int
130 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
131 {
132 struct proc *p;
133 int error;
134
135 p = curproc;
136 #ifdef SCTL_MASK32
137 if (req->flags & SCTL_MASK32) {
138 unsigned int val;
139 val = (unsigned int)p->p_sysent->sv_psstrings;
140 error = SYSCTL_OUT(req, &val, sizeof(val));
141 } else
142 #endif
143 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
144 sizeof(p->p_sysent->sv_psstrings));
145 return error;
146 }
147
148 static int
149 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
150 {
151 struct proc *p;
152 int error;
153
154 p = curproc;
155 #ifdef SCTL_MASK32
156 if (req->flags & SCTL_MASK32) {
157 unsigned int val;
158 val = (unsigned int)p->p_sysent->sv_usrstack;
159 error = SYSCTL_OUT(req, &val, sizeof(val));
160 } else
161 #endif
162 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
163 sizeof(p->p_sysent->sv_usrstack));
164 return error;
165 }
166
167 static int
168 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
169 {
170 struct proc *p;
171
172 p = curproc;
173 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
174 sizeof(p->p_sysent->sv_stackprot)));
175 }
176
177 /*
178 * Each of the items is a pointer to a `const struct execsw', hence the
179 * double pointer here.
180 */
181 static const struct execsw **execsw;
182
183 #ifndef _SYS_SYSPROTO_H_
184 struct execve_args {
185 char *fname;
186 char **argv;
187 char **envv;
188 };
189 #endif
190
191 int
192 execve(td, uap)
193 struct thread *td;
194 struct execve_args /* {
195 char *fname;
196 char **argv;
197 char **envv;
198 } */ *uap;
199 {
200 int error;
201 struct image_args args;
202
203 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
204 uap->argv, uap->envv);
205 if (error == 0)
206 error = kern_execve(td, &args, NULL);
207 return (error);
208 }
209
210 #ifndef _SYS_SYSPROTO_H_
211 struct __mac_execve_args {
212 char *fname;
213 char **argv;
214 char **envv;
215 struct mac *mac_p;
216 };
217 #endif
218
219 int
220 __mac_execve(td, uap)
221 struct thread *td;
222 struct __mac_execve_args /* {
223 char *fname;
224 char **argv;
225 char **envv;
226 struct mac *mac_p;
227 } */ *uap;
228 {
229 #ifdef MAC
230 int error;
231 struct image_args args;
232
233 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
234 uap->argv, uap->envv);
235 if (error == 0)
236 error = kern_execve(td, &args, uap->mac_p);
237 return (error);
238 #else
239 return (ENOSYS);
240 #endif
241 }
242
243 /*
244 * XXX: kern_execve has the astonishing property of not always returning to
245 * the caller. If sufficiently bad things happen during the call to
246 * do_execve(), it can end up calling exit1(); as a result, callers must
247 * avoid doing anything which they might need to undo (e.g., allocating
248 * memory).
249 */
250 int
251 kern_execve(td, args, mac_p)
252 struct thread *td;
253 struct image_args *args;
254 struct mac *mac_p;
255 {
256 struct proc *p = td->td_proc;
257 int error;
258
259 AUDIT_ARG(argv, args->begin_argv, args->argc,
260 args->begin_envv - args->begin_argv);
261 AUDIT_ARG(envv, args->begin_envv, args->envc,
262 args->endp - args->begin_envv);
263 if (p->p_flag & P_HADTHREADS) {
264 PROC_LOCK(p);
265 if (thread_single(SINGLE_BOUNDARY)) {
266 PROC_UNLOCK(p);
267 exec_free_args(args);
268 return (ERESTART); /* Try again later. */
269 }
270 PROC_UNLOCK(p);
271 }
272
273 error = do_execve(td, args, mac_p);
274
275 if (p->p_flag & P_HADTHREADS) {
276 PROC_LOCK(p);
277 /*
278 * If success, we upgrade to SINGLE_EXIT state to
279 * force other threads to suicide.
280 */
281 if (error == 0)
282 thread_single(SINGLE_EXIT);
283 else
284 thread_single_end();
285 PROC_UNLOCK(p);
286 }
287
288 return (error);
289 }
290
291 /*
292 * In-kernel implementation of execve(). All arguments are assumed to be
293 * userspace pointers from the passed thread.
294 */
295 static int
296 do_execve(td, args, mac_p)
297 struct thread *td;
298 struct image_args *args;
299 struct mac *mac_p;
300 {
301 struct proc *p = td->td_proc;
302 struct nameidata nd, *ndp;
303 struct ucred *newcred = NULL, *oldcred;
304 struct uidinfo *euip;
305 register_t *stack_base;
306 int error, len, i;
307 struct image_params image_params, *imgp;
308 struct vattr attr;
309 int (*img_first)(struct image_params *);
310 struct pargs *oldargs = NULL, *newargs = NULL;
311 struct sigacts *oldsigacts, *newsigacts;
312 #ifdef KTRACE
313 struct vnode *tracevp = NULL;
314 struct ucred *tracecred = NULL;
315 #endif
316 struct vnode *textvp = NULL;
317 int credential_changing;
318 int vfslocked;
319 int textset;
320 #ifdef MAC
321 struct label *interplabel = NULL;
322 int will_transition;
323 #endif
324 #ifdef HWPMC_HOOKS
325 struct pmckern_procexec pe;
326 #endif
327
328 vfslocked = 0;
329 imgp = &image_params;
330
331 /*
332 * Lock the process and set the P_INEXEC flag to indicate that
333 * it should be left alone until we're done here. This is
334 * necessary to avoid race conditions - e.g. in ptrace() -
335 * that might allow a local user to illicitly obtain elevated
336 * privileges.
337 */
338 PROC_LOCK(p);
339 KASSERT((p->p_flag & P_INEXEC) == 0,
340 ("%s(): process already has P_INEXEC flag", __func__));
341 p->p_flag |= P_INEXEC;
342 PROC_UNLOCK(p);
343
344 /*
345 * Initialize part of the common data
346 */
347 imgp->proc = p;
348 imgp->execlabel = NULL;
349 imgp->attr = &attr;
350 imgp->entry_addr = 0;
351 imgp->vmspace_destroyed = 0;
352 imgp->interpreted = 0;
353 imgp->opened = 0;
354 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
355 imgp->auxargs = NULL;
356 imgp->vp = NULL;
357 imgp->object = NULL;
358 imgp->firstpage = NULL;
359 imgp->ps_strings = 0;
360 imgp->auxarg_size = 0;
361 imgp->args = args;
362
363 #ifdef MAC
364 error = mac_execve_enter(imgp, mac_p);
365 if (error)
366 goto exec_fail;
367 #endif
368
369 imgp->image_header = NULL;
370
371 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
372
373 /*
374 * Translate the file name. namei() returns a vnode pointer
375 * in ni_vp amoung other things.
376 *
377 * XXXAUDIT: It would be desirable to also audit the name of the
378 * interpreter if this is an interpreted binary.
379 */
380 ndp = &nd;
381 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE |
382 AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
383
384 interpret:
385 error = namei(ndp);
386 if (error)
387 goto exec_fail;
388
389 vfslocked = NDHASGIANT(ndp);
390 imgp->vp = ndp->ni_vp;
391
392 /*
393 * Check file permissions (also 'opens' file)
394 */
395 error = exec_check_permissions(imgp);
396 if (error)
397 goto exec_fail_dealloc;
398
399 imgp->object = imgp->vp->v_object;
400 if (imgp->object != NULL)
401 vm_object_reference(imgp->object);
402
403 /*
404 * Set VV_TEXT now so no one can write to the executable while we're
405 * activating it.
406 *
407 * Remember if this was set before and unset it in case this is not
408 * actually an executable image.
409 */
410 textset = imgp->vp->v_vflag & VV_TEXT;
411 imgp->vp->v_vflag |= VV_TEXT;
412
413 error = exec_map_first_page(imgp);
414 if (error)
415 goto exec_fail_dealloc;
416
417 imgp->proc->p_osrel = 0;
418 /*
419 * If the current process has a special image activator it
420 * wants to try first, call it. For example, emulating shell
421 * scripts differently.
422 */
423 error = -1;
424 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
425 error = img_first(imgp);
426
427 /*
428 * Loop through the list of image activators, calling each one.
429 * An activator returns -1 if there is no match, 0 on success,
430 * and an error otherwise.
431 */
432 for (i = 0; error == -1 && execsw[i]; ++i) {
433 if (execsw[i]->ex_imgact == NULL ||
434 execsw[i]->ex_imgact == img_first) {
435 continue;
436 }
437 error = (*execsw[i]->ex_imgact)(imgp);
438 }
439
440 if (error) {
441 if (error == -1) {
442 if (textset == 0)
443 imgp->vp->v_vflag &= ~VV_TEXT;
444 error = ENOEXEC;
445 }
446 goto exec_fail_dealloc;
447 }
448
449 /*
450 * Special interpreter operation, cleanup and loop up to try to
451 * activate the interpreter.
452 */
453 if (imgp->interpreted) {
454 exec_unmap_first_page(imgp);
455 /*
456 * VV_TEXT needs to be unset for scripts. There is a short
457 * period before we determine that something is a script where
458 * VV_TEXT will be set. The vnode lock is held over this
459 * entire period so nothing should illegitimately be blocked.
460 */
461 imgp->vp->v_vflag &= ~VV_TEXT;
462 /* free name buffer and old vnode */
463 NDFREE(ndp, NDF_ONLY_PNBUF);
464 #ifdef MAC
465 interplabel = mac_vnode_label_alloc();
466 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
467 #endif
468 if (imgp->opened) {
469 VOP_CLOSE(ndp->ni_vp, FREAD, td->td_ucred, td);
470 imgp->opened = 0;
471 }
472 vput(ndp->ni_vp);
473 vm_object_deallocate(imgp->object);
474 imgp->object = NULL;
475 VFS_UNLOCK_GIANT(vfslocked);
476 vfslocked = 0;
477 /* set new name to that of the interpreter */
478 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
479 UIO_SYSSPACE, imgp->interpreter_name, td);
480 goto interpret;
481 }
482
483 /*
484 * NB: We unlock the vnode here because it is believed that none
485 * of the sv_copyout_strings/sv_fixup operations require the vnode.
486 */
487 VOP_UNLOCK(imgp->vp, 0, td);
488 /*
489 * Copy out strings (args and env) and initialize stack base
490 */
491 if (p->p_sysent->sv_copyout_strings)
492 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
493 else
494 stack_base = exec_copyout_strings(imgp);
495
496 /*
497 * If custom stack fixup routine present for this process
498 * let it do the stack setup.
499 * Else stuff argument count as first item on stack
500 */
501 if (p->p_sysent->sv_fixup != NULL)
502 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
503 else
504 suword(--stack_base, imgp->args->argc);
505
506 /*
507 * For security and other reasons, the file descriptor table cannot
508 * be shared after an exec.
509 */
510 fdunshare(p, td);
511
512 /*
513 * Malloc things before we need locks.
514 */
515 newcred = crget();
516 euip = uifind(attr.va_uid);
517 i = imgp->args->begin_envv - imgp->args->begin_argv;
518 /* Cache arguments if they fit inside our allowance */
519 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
520 newargs = pargs_alloc(i);
521 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
522 }
523
524 /* close files on exec */
525 fdcloseexec(td);
526 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
527
528 /* Get a reference to the vnode prior to locking the proc */
529 VREF(ndp->ni_vp);
530
531 /*
532 * For security and other reasons, signal handlers cannot
533 * be shared after an exec. The new process gets a copy of the old
534 * handlers. In execsigs(), the new process will have its signals
535 * reset.
536 */
537 PROC_LOCK(p);
538 if (sigacts_shared(p->p_sigacts)) {
539 oldsigacts = p->p_sigacts;
540 PROC_UNLOCK(p);
541 newsigacts = sigacts_alloc();
542 sigacts_copy(newsigacts, oldsigacts);
543 PROC_LOCK(p);
544 p->p_sigacts = newsigacts;
545 } else
546 oldsigacts = NULL;
547
548 /* Stop profiling */
549 stopprofclock(p);
550
551 /* reset caught signals */
552 execsigs(p);
553
554 /* name this process - nameiexec(p, ndp) */
555 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
556 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
557 p->p_comm[len] = 0;
558
559 /*
560 * mark as execed, wakeup the process that vforked (if any) and tell
561 * it that it now has its own resources back
562 */
563 p->p_flag |= P_EXEC;
564 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
565 p->p_flag &= ~P_PPWAIT;
566 wakeup(p->p_pptr);
567 }
568
569 /*
570 * Implement image setuid/setgid.
571 *
572 * Don't honor setuid/setgid if the filesystem prohibits it or if
573 * the process is being traced.
574 *
575 * XXXMAC: For the time being, use NOSUID to also prohibit
576 * transitions on the file system.
577 */
578 oldcred = p->p_ucred;
579 credential_changing = 0;
580 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
581 attr.va_uid;
582 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
583 attr.va_gid;
584 #ifdef MAC
585 will_transition = mac_execve_will_transition(oldcred, imgp->vp,
586 interplabel, imgp);
587 credential_changing |= will_transition;
588 #endif
589
590 if (credential_changing &&
591 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
592 (p->p_flag & P_TRACED) == 0) {
593 /*
594 * Turn off syscall tracing for set-id programs, except for
595 * root. Record any set-id flags first to make sure that
596 * we do not regain any tracing during a possible block.
597 */
598 setsugid(p);
599
600 #ifdef KTRACE
601 if (p->p_tracevp != NULL &&
602 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
603 mtx_lock(&ktrace_mtx);
604 p->p_traceflag = 0;
605 tracevp = p->p_tracevp;
606 p->p_tracevp = NULL;
607 tracecred = p->p_tracecred;
608 p->p_tracecred = NULL;
609 mtx_unlock(&ktrace_mtx);
610 }
611 #endif
612 /*
613 * Close any file descriptors 0..2 that reference procfs,
614 * then make sure file descriptors 0..2 are in use.
615 *
616 * setugidsafety() may call closef() and then pfind()
617 * which may grab the process lock.
618 * fdcheckstd() may call falloc() which may block to
619 * allocate memory, so temporarily drop the process lock.
620 */
621 PROC_UNLOCK(p);
622 setugidsafety(td);
623 VOP_UNLOCK(imgp->vp, 0, td);
624 error = fdcheckstd(td);
625 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
626 if (error != 0)
627 goto done1;
628 PROC_LOCK(p);
629 /*
630 * Set the new credentials.
631 */
632 crcopy(newcred, oldcred);
633 if (attr.va_mode & VSUID)
634 change_euid(newcred, euip);
635 if (attr.va_mode & VSGID)
636 change_egid(newcred, attr.va_gid);
637 #ifdef MAC
638 if (will_transition) {
639 mac_execve_transition(oldcred, newcred, imgp->vp,
640 interplabel, imgp);
641 }
642 #endif
643 /*
644 * Implement correct POSIX saved-id behavior.
645 *
646 * XXXMAC: Note that the current logic will save the
647 * uid and gid if a MAC domain transition occurs, even
648 * though maybe it shouldn't.
649 */
650 change_svuid(newcred, newcred->cr_uid);
651 change_svgid(newcred, newcred->cr_gid);
652 p->p_ucred = newcred;
653 newcred = NULL;
654 } else {
655 if (oldcred->cr_uid == oldcred->cr_ruid &&
656 oldcred->cr_gid == oldcred->cr_rgid)
657 p->p_flag &= ~P_SUGID;
658 /*
659 * Implement correct POSIX saved-id behavior.
660 *
661 * XXX: It's not clear that the existing behavior is
662 * POSIX-compliant. A number of sources indicate that the
663 * saved uid/gid should only be updated if the new ruid is
664 * not equal to the old ruid, or the new euid is not equal
665 * to the old euid and the new euid is not equal to the old
666 * ruid. The FreeBSD code always updates the saved uid/gid.
667 * Also, this code uses the new (replaced) euid and egid as
668 * the source, which may or may not be the right ones to use.
669 */
670 if (oldcred->cr_svuid != oldcred->cr_uid ||
671 oldcred->cr_svgid != oldcred->cr_gid) {
672 crcopy(newcred, oldcred);
673 change_svuid(newcred, newcred->cr_uid);
674 change_svgid(newcred, newcred->cr_gid);
675 p->p_ucred = newcred;
676 newcred = NULL;
677 }
678 }
679
680 /*
681 * Store the vp for use in procfs. This vnode was referenced prior
682 * to locking the proc lock.
683 */
684 textvp = p->p_textvp;
685 p->p_textvp = ndp->ni_vp;
686
687 #ifdef KDTRACE_HOOKS
688 /*
689 * Tell the DTrace fasttrap provider about the exec if it
690 * has declared an interest.
691 */
692 if (dtrace_fasttrap_exec)
693 dtrace_fasttrap_exec(p);
694 #endif
695
696 /*
697 * Notify others that we exec'd, and clear the P_INEXEC flag
698 * as we're now a bona fide freshly-execed process.
699 */
700 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
701 p->p_flag &= ~P_INEXEC;
702
703 /*
704 * If tracing the process, trap to debugger so breakpoints
705 * can be set before the program executes.
706 * Use tdsignal to deliver signal to current thread, use
707 * psignal may cause the signal to be delivered to wrong thread
708 * because that thread will exit, remember we are going to enter
709 * single thread mode.
710 */
711 if (p->p_flag & P_TRACED)
712 tdsignal(p, td, SIGTRAP, NULL);
713
714 /* clear "fork but no exec" flag, as we _are_ execing */
715 p->p_acflag &= ~AFORK;
716
717 /*
718 * Free any previous argument cache and replace it with
719 * the new argument cache, if any.
720 */
721 oldargs = p->p_args;
722 p->p_args = newargs;
723 newargs = NULL;
724
725 #ifdef HWPMC_HOOKS
726 /*
727 * Check if system-wide sampling is in effect or if the
728 * current process is using PMCs. If so, do exec() time
729 * processing. This processing needs to happen AFTER the
730 * P_INEXEC flag is cleared.
731 *
732 * The proc lock needs to be released before taking the PMC
733 * SX.
734 */
735 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
736 PROC_UNLOCK(p);
737 pe.pm_credentialschanged = credential_changing;
738 pe.pm_entryaddr = imgp->entry_addr;
739
740 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
741 } else
742 PROC_UNLOCK(p);
743 #else /* !HWPMC_HOOKS */
744 PROC_UNLOCK(p);
745 #endif
746
747 /* Set values passed into the program in registers. */
748 if (p->p_sysent->sv_setregs)
749 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
750 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
751 else
752 exec_setregs(td, imgp->entry_addr,
753 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
754
755 vfs_mark_atime(imgp->vp, td);
756
757 done1:
758 /*
759 * Free any resources malloc'd earlier that we didn't use.
760 */
761 uifree(euip);
762 if (newcred == NULL)
763 crfree(oldcred);
764 else
765 crfree(newcred);
766 VOP_UNLOCK(imgp->vp, 0, td);
767
768 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
769
770 /*
771 * Handle deferred decrement of ref counts.
772 */
773 if (textvp != NULL) {
774 int tvfslocked;
775
776 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
777 vrele(textvp);
778 VFS_UNLOCK_GIANT(tvfslocked);
779 }
780 if (ndp->ni_vp && error != 0)
781 vrele(ndp->ni_vp);
782 #ifdef KTRACE
783 if (tracevp != NULL) {
784 int tvfslocked;
785
786 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
787 vrele(tracevp);
788 VFS_UNLOCK_GIANT(tvfslocked);
789 }
790 if (tracecred != NULL)
791 crfree(tracecred);
792 #endif
793 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
794 pargs_drop(oldargs);
795 pargs_drop(newargs);
796 if (oldsigacts != NULL)
797 sigacts_free(oldsigacts);
798
799 exec_fail_dealloc:
800
801 /*
802 * free various allocated resources
803 */
804 if (imgp->firstpage != NULL)
805 exec_unmap_first_page(imgp);
806
807 if (imgp->vp != NULL) {
808 NDFREE(ndp, NDF_ONLY_PNBUF);
809 if (imgp->opened)
810 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
811 vput(imgp->vp);
812 }
813
814 if (imgp->object != NULL)
815 vm_object_deallocate(imgp->object);
816
817 if (error == 0) {
818 /*
819 * Stop the process here if its stop event mask has
820 * the S_EXEC bit set.
821 */
822 STOPEVENT(p, S_EXEC, 0);
823 goto done2;
824 }
825
826 exec_fail:
827 /* we're done here, clear P_INEXEC */
828 PROC_LOCK(p);
829 p->p_flag &= ~P_INEXEC;
830 PROC_UNLOCK(p);
831
832 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
833
834 done2:
835 #ifdef MAC
836 mac_execve_exit(imgp);
837 if (interplabel != NULL)
838 mac_vnode_label_free(interplabel);
839 #endif
840 VFS_UNLOCK_GIANT(vfslocked);
841 exec_free_args(args);
842
843 if (error && imgp->vmspace_destroyed) {
844 /* sorry, no more process anymore. exit gracefully */
845 exit1(td, W_EXITCODE(0, SIGABRT));
846 /* NOT REACHED */
847 }
848 return (error);
849 }
850
851 int
852 exec_map_first_page(imgp)
853 struct image_params *imgp;
854 {
855 int rv, i;
856 int initial_pagein;
857 vm_page_t ma[VM_INITIAL_PAGEIN];
858 vm_object_t object;
859
860 if (imgp->firstpage != NULL)
861 exec_unmap_first_page(imgp);
862
863 object = imgp->vp->v_object;
864 if (object == NULL)
865 return (EACCES);
866 VM_OBJECT_LOCK(object);
867 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
868 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
869 initial_pagein = VM_INITIAL_PAGEIN;
870 if (initial_pagein > object->size)
871 initial_pagein = object->size;
872 for (i = 1; i < initial_pagein; i++) {
873 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
874 if (ma[i]->valid)
875 break;
876 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
877 break;
878 vm_page_busy(ma[i]);
879 } else {
880 ma[i] = vm_page_alloc(object, i,
881 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
882 if (ma[i] == NULL)
883 break;
884 }
885 }
886 initial_pagein = i;
887 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
888 ma[0] = vm_page_lookup(object, 0);
889 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
890 (ma[0]->valid == 0)) {
891 if (ma[0]) {
892 vm_page_lock_queues();
893 vm_page_free(ma[0]);
894 vm_page_unlock_queues();
895 }
896 VM_OBJECT_UNLOCK(object);
897 return (EIO);
898 }
899 }
900 vm_page_lock_queues();
901 vm_page_hold(ma[0]);
902 vm_page_unlock_queues();
903 vm_page_wakeup(ma[0]);
904 VM_OBJECT_UNLOCK(object);
905
906 imgp->firstpage = sf_buf_alloc(ma[0], 0);
907 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
908
909 return (0);
910 }
911
912 void
913 exec_unmap_first_page(imgp)
914 struct image_params *imgp;
915 {
916 vm_page_t m;
917
918 if (imgp->firstpage != NULL) {
919 m = sf_buf_page(imgp->firstpage);
920 sf_buf_free(imgp->firstpage);
921 imgp->firstpage = NULL;
922 vm_page_lock_queues();
923 vm_page_unhold(m);
924 vm_page_unlock_queues();
925 }
926 }
927
928 /*
929 * Destroy old address space, and allocate a new stack
930 * The new stack is only SGROWSIZ large because it is grown
931 * automatically in trap.c.
932 */
933 int
934 exec_new_vmspace(imgp, sv)
935 struct image_params *imgp;
936 struct sysentvec *sv;
937 {
938 int error;
939 struct proc *p = imgp->proc;
940 struct vmspace *vmspace = p->p_vmspace;
941 vm_offset_t sv_minuser, stack_addr;
942 vm_map_t map;
943 u_long ssiz;
944
945 imgp->vmspace_destroyed = 1;
946 imgp->sysent = sv;
947
948 /* May be called with Giant held */
949 EVENTHANDLER_INVOKE(process_exec, p, imgp);
950
951 /*
952 * Blow away entire process VM, if address space not shared,
953 * otherwise, create a new VM space so that other threads are
954 * not disrupted
955 */
956 map = &vmspace->vm_map;
957 if (map_at_zero)
958 sv_minuser = sv->sv_minuser;
959 else
960 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
961 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
962 vm_map_max(map) == sv->sv_maxuser) {
963 shmexit(vmspace);
964 pmap_remove_pages(vmspace_pmap(vmspace));
965 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
966 } else {
967 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
968 if (error)
969 return (error);
970 vmspace = p->p_vmspace;
971 map = &vmspace->vm_map;
972 }
973
974 /* Allocate a new stack */
975 if (sv->sv_maxssiz != NULL)
976 ssiz = *sv->sv_maxssiz;
977 else
978 ssiz = maxssiz;
979 stack_addr = sv->sv_usrstack - ssiz;
980 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
981 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
982 if (error)
983 return (error);
984
985 #ifdef __ia64__
986 /* Allocate a new register stack */
987 stack_addr = IA64_BACKINGSTORE;
988 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
989 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
990 if (error)
991 return (error);
992 #endif
993
994 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
995 * VM_STACK case, but they are still used to monitor the size of the
996 * process stack so we can check the stack rlimit.
997 */
998 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
999 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1000
1001 return (0);
1002 }
1003
1004 /*
1005 * Copy out argument and environment strings from the old process address
1006 * space into the temporary string buffer.
1007 */
1008 int
1009 exec_copyin_args(struct image_args *args, char *fname,
1010 enum uio_seg segflg, char **argv, char **envv)
1011 {
1012 char *argp, *envp;
1013 int error;
1014 size_t length;
1015
1016 error = 0;
1017
1018 bzero(args, sizeof(*args));
1019 if (argv == NULL)
1020 return (EFAULT);
1021 /*
1022 * Allocate temporary demand zeroed space for argument and
1023 * environment strings:
1024 *
1025 * o ARG_MAX for argument and environment;
1026 * o MAXSHELLCMDLEN for the name of interpreters.
1027 */
1028 args->buf = (char *) kmem_alloc_wait(exec_map,
1029 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1030 if (args->buf == NULL)
1031 return (ENOMEM);
1032 args->begin_argv = args->buf;
1033 args->endp = args->begin_argv;
1034 args->stringspace = ARG_MAX;
1035
1036 args->fname = args->buf + ARG_MAX;
1037
1038 /*
1039 * Copy the file name.
1040 */
1041 error = (segflg == UIO_SYSSPACE) ?
1042 copystr(fname, args->fname, PATH_MAX, &length) :
1043 copyinstr(fname, args->fname, PATH_MAX, &length);
1044 if (error != 0)
1045 goto err_exit;
1046
1047 /*
1048 * extract arguments first
1049 */
1050 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1051 if (argp == (caddr_t) -1) {
1052 error = EFAULT;
1053 goto err_exit;
1054 }
1055 if ((error = copyinstr(argp, args->endp,
1056 args->stringspace, &length))) {
1057 if (error == ENAMETOOLONG)
1058 error = E2BIG;
1059 goto err_exit;
1060 }
1061 args->stringspace -= length;
1062 args->endp += length;
1063 args->argc++;
1064 }
1065
1066 args->begin_envv = args->endp;
1067
1068 /*
1069 * extract environment strings
1070 */
1071 if (envv) {
1072 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1073 if (envp == (caddr_t)-1) {
1074 error = EFAULT;
1075 goto err_exit;
1076 }
1077 if ((error = copyinstr(envp, args->endp,
1078 args->stringspace, &length))) {
1079 if (error == ENAMETOOLONG)
1080 error = E2BIG;
1081 goto err_exit;
1082 }
1083 args->stringspace -= length;
1084 args->endp += length;
1085 args->envc++;
1086 }
1087 }
1088
1089 return (0);
1090
1091 err_exit:
1092 exec_free_args(args);
1093 return (error);
1094 }
1095
1096 static void
1097 exec_free_args(struct image_args *args)
1098 {
1099
1100 if (args->buf) {
1101 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1102 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1103 args->buf = NULL;
1104 }
1105 }
1106
1107 /*
1108 * Copy strings out to the new process address space, constructing new arg
1109 * and env vector tables. Return a pointer to the base so that it can be used
1110 * as the initial stack pointer.
1111 */
1112 register_t *
1113 exec_copyout_strings(imgp)
1114 struct image_params *imgp;
1115 {
1116 int argc, envc;
1117 char **vectp;
1118 char *stringp, *destp;
1119 register_t *stack_base;
1120 struct ps_strings *arginfo;
1121 struct proc *p;
1122 int szsigcode;
1123
1124 /*
1125 * Calculate string base and vector table pointers.
1126 * Also deal with signal trampoline code for this exec type.
1127 */
1128 p = imgp->proc;
1129 szsigcode = 0;
1130 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1131 if (p->p_sysent->sv_szsigcode != NULL)
1132 szsigcode = *(p->p_sysent->sv_szsigcode);
1133 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1134 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1135
1136 /*
1137 * install sigcode
1138 */
1139 if (szsigcode)
1140 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1141 szsigcode), szsigcode);
1142
1143 /*
1144 * If we have a valid auxargs ptr, prepare some room
1145 * on the stack.
1146 */
1147 if (imgp->auxargs) {
1148 /*
1149 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1150 * lower compatibility.
1151 */
1152 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1153 (AT_COUNT * 2);
1154 /*
1155 * The '+ 2' is for the null pointers at the end of each of
1156 * the arg and env vector sets,and imgp->auxarg_size is room
1157 * for argument of Runtime loader.
1158 */
1159 vectp = (char **)(destp - (imgp->args->argc +
1160 imgp->args->envc + 2 + imgp->auxarg_size) *
1161 sizeof(char *));
1162
1163 } else {
1164 /*
1165 * The '+ 2' is for the null pointers at the end of each of
1166 * the arg and env vector sets
1167 */
1168 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1169 sizeof(char *));
1170 }
1171
1172 /*
1173 * vectp also becomes our initial stack base
1174 */
1175 stack_base = (register_t *)vectp;
1176
1177 stringp = imgp->args->begin_argv;
1178 argc = imgp->args->argc;
1179 envc = imgp->args->envc;
1180
1181 /*
1182 * Copy out strings - arguments and environment.
1183 */
1184 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1185
1186 /*
1187 * Fill in "ps_strings" struct for ps, w, etc.
1188 */
1189 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1190 suword(&arginfo->ps_nargvstr, argc);
1191
1192 /*
1193 * Fill in argument portion of vector table.
1194 */
1195 for (; argc > 0; --argc) {
1196 suword(vectp++, (long)(intptr_t)destp);
1197 while (*stringp++ != 0)
1198 destp++;
1199 destp++;
1200 }
1201
1202 /* a null vector table pointer separates the argp's from the envp's */
1203 suword(vectp++, 0);
1204
1205 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1206 suword(&arginfo->ps_nenvstr, envc);
1207
1208 /*
1209 * Fill in environment portion of vector table.
1210 */
1211 for (; envc > 0; --envc) {
1212 suword(vectp++, (long)(intptr_t)destp);
1213 while (*stringp++ != 0)
1214 destp++;
1215 destp++;
1216 }
1217
1218 /* end of vector table is a null pointer */
1219 suword(vectp, 0);
1220
1221 return (stack_base);
1222 }
1223
1224 /*
1225 * Check permissions of file to execute.
1226 * Called with imgp->vp locked.
1227 * Return 0 for success or error code on failure.
1228 */
1229 int
1230 exec_check_permissions(imgp)
1231 struct image_params *imgp;
1232 {
1233 struct vnode *vp = imgp->vp;
1234 struct vattr *attr = imgp->attr;
1235 struct thread *td;
1236 int error;
1237
1238 td = curthread; /* XXXKSE */
1239
1240 /* Get file attributes */
1241 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1242 if (error)
1243 return (error);
1244
1245 #ifdef MAC
1246 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1247 if (error)
1248 return (error);
1249 #endif
1250
1251 /*
1252 * 1) Check if file execution is disabled for the filesystem that this
1253 * file resides on.
1254 * 2) Insure that at least one execute bit is on - otherwise root
1255 * will always succeed, and we don't want to happen unless the
1256 * file really is executable.
1257 * 3) Insure that the file is a regular file.
1258 */
1259 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1260 ((attr->va_mode & 0111) == 0) ||
1261 (attr->va_type != VREG))
1262 return (EACCES);
1263
1264 /*
1265 * Zero length files can't be exec'd
1266 */
1267 if (attr->va_size == 0)
1268 return (ENOEXEC);
1269
1270 /*
1271 * Check for execute permission to file based on current credentials.
1272 */
1273 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1274 if (error)
1275 return (error);
1276
1277 /*
1278 * Check number of open-for-writes on the file and deny execution
1279 * if there are any.
1280 */
1281 if (vp->v_writecount)
1282 return (ETXTBSY);
1283
1284 /*
1285 * Call filesystem specific open routine (which does nothing in the
1286 * general case).
1287 */
1288 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1289 if (error == 0)
1290 imgp->opened = 1;
1291 return (error);
1292 }
1293
1294 /*
1295 * Exec handler registration
1296 */
1297 int
1298 exec_register(execsw_arg)
1299 const struct execsw *execsw_arg;
1300 {
1301 const struct execsw **es, **xs, **newexecsw;
1302 int count = 2; /* New slot and trailing NULL */
1303
1304 if (execsw)
1305 for (es = execsw; *es; es++)
1306 count++;
1307 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1308 if (newexecsw == NULL)
1309 return (ENOMEM);
1310 xs = newexecsw;
1311 if (execsw)
1312 for (es = execsw; *es; es++)
1313 *xs++ = *es;
1314 *xs++ = execsw_arg;
1315 *xs = NULL;
1316 if (execsw)
1317 free(execsw, M_TEMP);
1318 execsw = newexecsw;
1319 return (0);
1320 }
1321
1322 int
1323 exec_unregister(execsw_arg)
1324 const struct execsw *execsw_arg;
1325 {
1326 const struct execsw **es, **xs, **newexecsw;
1327 int count = 1;
1328
1329 if (execsw == NULL)
1330 panic("unregister with no handlers left?\n");
1331
1332 for (es = execsw; *es; es++) {
1333 if (*es == execsw_arg)
1334 break;
1335 }
1336 if (*es == NULL)
1337 return (ENOENT);
1338 for (es = execsw; *es; es++)
1339 if (*es != execsw_arg)
1340 count++;
1341 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1342 if (newexecsw == NULL)
1343 return (ENOMEM);
1344 xs = newexecsw;
1345 for (es = execsw; *es; es++)
1346 if (*es != execsw_arg)
1347 *xs++ = *es;
1348 *xs = NULL;
1349 if (execsw)
1350 free(execsw, M_TEMP);
1351 execsw = newexecsw;
1352 return (0);
1353 }
Cache object: d2c9da958e37c9cb693496ae9a70a4c2
|