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("$FreeBSD$");
29
30 #include "opt_hwpmc_hooks.h"
31 #include "opt_kdtrace.h"
32 #include "opt_ktrace.h"
33 #include "opt_mac.h"
34 #include "opt_vm.h"
35
36 #include <sys/param.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/sdt.h>
59 #include <sys/sf_buf.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysent.h>
62 #include <sys/shm.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 #ifdef KTRACE
66 #include <sys/ktrace.h>
67 #endif
68
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_pager.h>
78
79 #ifdef HWPMC_HOOKS
80 #include <sys/pmckern.h>
81 #endif
82
83 #include <machine/reg.h>
84
85 #include <security/audit/audit.h>
86 #include <security/mac/mac_framework.h>
87
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90 dtrace_execexit_func_t dtrace_fasttrap_exec;
91 #endif
92
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE(proc, kernel, , exec);
95 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
96 SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
97 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
98 SDT_PROBE_DEFINE(proc, kernel, , exec_success);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
100
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
107 struct mac *mac_p);
108 static void exec_free_args(struct image_args *);
109
110 /* XXX This should be vm_size_t. */
111 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112 NULL, 0, sysctl_kern_ps_strings, "LU", "");
113
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
116 NULL, 0, sysctl_kern_usrstack, "LU", "");
117
118 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119 NULL, 0, sysctl_kern_stackprot, "I", "");
120
121 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123 &ps_arg_cache_limit, 0, "");
124
125 static int map_at_zero = 1;
126 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
127 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
128 "Permit processes to map an object at virtual address 0.");
129
130 static int
131 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
132 {
133 struct proc *p;
134 int error;
135
136 p = curproc;
137 #ifdef SCTL_MASK32
138 if (req->flags & SCTL_MASK32) {
139 unsigned int val;
140 val = (unsigned int)p->p_sysent->sv_psstrings;
141 error = SYSCTL_OUT(req, &val, sizeof(val));
142 } else
143 #endif
144 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
145 sizeof(p->p_sysent->sv_psstrings));
146 return error;
147 }
148
149 static int
150 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
151 {
152 struct proc *p;
153 int error;
154
155 p = curproc;
156 #ifdef SCTL_MASK32
157 if (req->flags & SCTL_MASK32) {
158 unsigned int val;
159 val = (unsigned int)p->p_sysent->sv_usrstack;
160 error = SYSCTL_OUT(req, &val, sizeof(val));
161 } else
162 #endif
163 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
164 sizeof(p->p_sysent->sv_usrstack));
165 return error;
166 }
167
168 static int
169 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
170 {
171 struct proc *p;
172
173 p = curproc;
174 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
175 sizeof(p->p_sysent->sv_stackprot)));
176 }
177
178 /*
179 * Each of the items is a pointer to a `const struct execsw', hence the
180 * double pointer here.
181 */
182 static const struct execsw **execsw;
183
184 #ifndef _SYS_SYSPROTO_H_
185 struct execve_args {
186 char *fname;
187 char **argv;
188 char **envv;
189 };
190 #endif
191
192 int
193 execve(td, uap)
194 struct thread *td;
195 struct execve_args /* {
196 char *fname;
197 char **argv;
198 char **envv;
199 } */ *uap;
200 {
201 int error;
202 struct image_args args;
203
204 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
205 uap->argv, uap->envv);
206 if (error == 0)
207 error = kern_execve(td, &args, NULL);
208 return (error);
209 }
210
211 #ifndef _SYS_SYSPROTO_H_
212 struct __mac_execve_args {
213 char *fname;
214 char **argv;
215 char **envv;
216 struct mac *mac_p;
217 };
218 #endif
219
220 int
221 __mac_execve(td, uap)
222 struct thread *td;
223 struct __mac_execve_args /* {
224 char *fname;
225 char **argv;
226 char **envv;
227 struct mac *mac_p;
228 } */ *uap;
229 {
230 #ifdef MAC
231 int error;
232 struct image_args args;
233
234 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
235 uap->argv, uap->envv);
236 if (error == 0)
237 error = kern_execve(td, &args, uap->mac_p);
238 return (error);
239 #else
240 return (ENOSYS);
241 #endif
242 }
243
244 /*
245 * XXX: kern_execve has the astonishing property of not always returning to
246 * the caller. If sufficiently bad things happen during the call to
247 * do_execve(), it can end up calling exit1(); as a result, callers must
248 * avoid doing anything which they might need to undo (e.g., allocating
249 * memory).
250 */
251 int
252 kern_execve(td, args, mac_p)
253 struct thread *td;
254 struct image_args *args;
255 struct mac *mac_p;
256 {
257 struct proc *p = td->td_proc;
258 int error;
259
260 AUDIT_ARG(argv, args->begin_argv, args->argc,
261 args->begin_envv - args->begin_argv);
262 AUDIT_ARG(envv, args->begin_envv, args->envc,
263 args->endp - args->begin_envv);
264 if (p->p_flag & P_HADTHREADS) {
265 PROC_LOCK(p);
266 if (thread_single(SINGLE_BOUNDARY)) {
267 PROC_UNLOCK(p);
268 exec_free_args(args);
269 return (ERESTART); /* Try again later. */
270 }
271 PROC_UNLOCK(p);
272 }
273
274 error = do_execve(td, args, mac_p);
275
276 if (p->p_flag & P_HADTHREADS) {
277 PROC_LOCK(p);
278 /*
279 * If success, we upgrade to SINGLE_EXIT state to
280 * force other threads to suicide.
281 */
282 if (error == 0)
283 thread_single(SINGLE_EXIT);
284 else
285 thread_single_end();
286 PROC_UNLOCK(p);
287 }
288
289 return (error);
290 }
291
292 /*
293 * In-kernel implementation of execve(). All arguments are assumed to be
294 * userspace pointers from the passed thread.
295 */
296 static int
297 do_execve(td, args, mac_p)
298 struct thread *td;
299 struct image_args *args;
300 struct mac *mac_p;
301 {
302 struct proc *p = td->td_proc;
303 struct nameidata nd, *ndp;
304 struct ucred *newcred = NULL, *oldcred;
305 struct uidinfo *euip;
306 register_t *stack_base;
307 int error, len, i;
308 struct image_params image_params, *imgp;
309 struct vattr attr;
310 int (*img_first)(struct image_params *);
311 struct pargs *oldargs = NULL, *newargs = NULL;
312 struct sigacts *oldsigacts, *newsigacts;
313 #ifdef KTRACE
314 struct vnode *tracevp = NULL;
315 struct ucred *tracecred = NULL;
316 #endif
317 struct vnode *textvp = NULL;
318 int credential_changing;
319 int vfslocked;
320 int textset;
321 #ifdef MAC
322 struct label *interplabel = NULL;
323 int will_transition;
324 #endif
325 #ifdef HWPMC_HOOKS
326 struct pmckern_procexec pe;
327 #endif
328
329 vfslocked = 0;
330 imgp = &image_params;
331
332 /*
333 * Lock the process and set the P_INEXEC flag to indicate that
334 * it should be left alone until we're done here. This is
335 * necessary to avoid race conditions - e.g. in ptrace() -
336 * that might allow a local user to illicitly obtain elevated
337 * privileges.
338 */
339 PROC_LOCK(p);
340 KASSERT((p->p_flag & P_INEXEC) == 0,
341 ("%s(): process already has P_INEXEC flag", __func__));
342 p->p_flag |= P_INEXEC;
343 PROC_UNLOCK(p);
344
345 /*
346 * Initialize part of the common data
347 */
348 imgp->proc = p;
349 imgp->execlabel = NULL;
350 imgp->attr = &attr;
351 imgp->entry_addr = 0;
352 imgp->vmspace_destroyed = 0;
353 imgp->interpreted = 0;
354 imgp->opened = 0;
355 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
356 imgp->auxargs = NULL;
357 imgp->vp = NULL;
358 imgp->object = NULL;
359 imgp->firstpage = NULL;
360 imgp->ps_strings = 0;
361 imgp->auxarg_size = 0;
362 imgp->args = args;
363 imgp->execpath = imgp->freepath = NULL;
364 imgp->execpathp = 0;
365
366 #ifdef MAC
367 error = mac_execve_enter(imgp, mac_p);
368 if (error)
369 goto exec_fail;
370 #endif
371
372 imgp->image_header = NULL;
373
374 /*
375 * Translate the file name. namei() returns a vnode pointer
376 * in ni_vp amoung other things.
377 *
378 * XXXAUDIT: It would be desirable to also audit the name of the
379 * interpreter if this is an interpreted binary.
380 */
381 ndp = &nd;
382 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE |
383 AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
384
385 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
386
387 interpret:
388 error = namei(ndp);
389 if (error)
390 goto exec_fail;
391
392 vfslocked = NDHASGIANT(ndp);
393 imgp->vp = ndp->ni_vp;
394
395 /*
396 * Check file permissions (also 'opens' file)
397 */
398 error = exec_check_permissions(imgp);
399 if (error)
400 goto exec_fail_dealloc;
401
402 imgp->object = imgp->vp->v_object;
403 if (imgp->object != NULL)
404 vm_object_reference(imgp->object);
405
406 /*
407 * Set VV_TEXT now so no one can write to the executable while we're
408 * activating it.
409 *
410 * Remember if this was set before and unset it in case this is not
411 * actually an executable image.
412 */
413 textset = imgp->vp->v_vflag & VV_TEXT;
414 imgp->vp->v_vflag |= VV_TEXT;
415
416 error = exec_map_first_page(imgp);
417 if (error)
418 goto exec_fail_dealloc;
419
420 imgp->proc->p_osrel = 0;
421 /*
422 * If the current process has a special image activator it
423 * wants to try first, call it. For example, emulating shell
424 * scripts differently.
425 */
426 error = -1;
427 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
428 error = img_first(imgp);
429
430 /*
431 * Loop through the list of image activators, calling each one.
432 * An activator returns -1 if there is no match, 0 on success,
433 * and an error otherwise.
434 */
435 for (i = 0; error == -1 && execsw[i]; ++i) {
436 if (execsw[i]->ex_imgact == NULL ||
437 execsw[i]->ex_imgact == img_first) {
438 continue;
439 }
440 error = (*execsw[i]->ex_imgact)(imgp);
441 }
442
443 if (error) {
444 if (error == -1) {
445 if (textset == 0)
446 imgp->vp->v_vflag &= ~VV_TEXT;
447 error = ENOEXEC;
448 }
449 goto exec_fail_dealloc;
450 }
451
452 /*
453 * Special interpreter operation, cleanup and loop up to try to
454 * activate the interpreter.
455 */
456 if (imgp->interpreted) {
457 exec_unmap_first_page(imgp);
458 /*
459 * VV_TEXT needs to be unset for scripts. There is a short
460 * period before we determine that something is a script where
461 * VV_TEXT will be set. The vnode lock is held over this
462 * entire period so nothing should illegitimately be blocked.
463 */
464 imgp->vp->v_vflag &= ~VV_TEXT;
465 /* free name buffer and old vnode */
466 NDFREE(ndp, NDF_ONLY_PNBUF);
467 #ifdef MAC
468 interplabel = mac_vnode_label_alloc();
469 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
470 #endif
471 if (imgp->opened) {
472 VOP_CLOSE(ndp->ni_vp, FREAD, td->td_ucred, td);
473 imgp->opened = 0;
474 }
475 vput(ndp->ni_vp);
476 vm_object_deallocate(imgp->object);
477 imgp->object = NULL;
478 VFS_UNLOCK_GIANT(vfslocked);
479 vfslocked = 0;
480 /* set new name to that of the interpreter */
481 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
482 UIO_SYSSPACE, imgp->interpreter_name, td);
483 goto interpret;
484 }
485
486 /*
487 * NB: We unlock the vnode here because it is believed that none
488 * of the sv_copyout_strings/sv_fixup operations require the vnode.
489 */
490 VOP_UNLOCK(imgp->vp, 0, td);
491
492 if (imgp->auxargs != NULL)
493 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath);
494
495 /*
496 * Copy out strings (args and env) and initialize stack base
497 */
498 if (p->p_sysent->sv_copyout_strings)
499 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
500 else
501 stack_base = exec_copyout_strings(imgp);
502
503 /*
504 * If custom stack fixup routine present for this process
505 * let it do the stack setup.
506 * Else stuff argument count as first item on stack
507 */
508 if (p->p_sysent->sv_fixup != NULL)
509 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
510 else
511 suword(--stack_base, imgp->args->argc);
512
513 /*
514 * For security and other reasons, the file descriptor table cannot
515 * be shared after an exec.
516 */
517 fdunshare(p, td);
518
519 /*
520 * Malloc things before we need locks.
521 */
522 newcred = crget();
523 euip = uifind(attr.va_uid);
524 i = imgp->args->begin_envv - imgp->args->begin_argv;
525 /* Cache arguments if they fit inside our allowance */
526 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
527 newargs = pargs_alloc(i);
528 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
529 }
530
531 /* close files on exec */
532 fdcloseexec(td);
533 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
534
535 /* Get a reference to the vnode prior to locking the proc */
536 VREF(ndp->ni_vp);
537
538 /*
539 * For security and other reasons, signal handlers cannot
540 * be shared after an exec. The new process gets a copy of the old
541 * handlers. In execsigs(), the new process will have its signals
542 * reset.
543 */
544 PROC_LOCK(p);
545 if (sigacts_shared(p->p_sigacts)) {
546 oldsigacts = p->p_sigacts;
547 PROC_UNLOCK(p);
548 newsigacts = sigacts_alloc();
549 sigacts_copy(newsigacts, oldsigacts);
550 PROC_LOCK(p);
551 p->p_sigacts = newsigacts;
552 } else
553 oldsigacts = NULL;
554
555 /* Stop profiling */
556 stopprofclock(p);
557
558 /* reset caught signals */
559 execsigs(p);
560
561 /* name this process - nameiexec(p, ndp) */
562 bzero(p->p_comm, sizeof(p->p_comm));
563 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
564 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
565
566 /*
567 * mark as execed, wakeup the process that vforked (if any) and tell
568 * it that it now has its own resources back
569 */
570 p->p_flag |= P_EXEC;
571 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
572 p->p_flag &= ~P_PPWAIT;
573 cv_broadcast(&p->p_pwait);
574 }
575
576 /*
577 * Implement image setuid/setgid.
578 *
579 * Don't honor setuid/setgid if the filesystem prohibits it or if
580 * the process is being traced.
581 *
582 * XXXMAC: For the time being, use NOSUID to also prohibit
583 * transitions on the file system.
584 */
585 oldcred = p->p_ucred;
586 credential_changing = 0;
587 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
588 attr.va_uid;
589 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
590 attr.va_gid;
591 #ifdef MAC
592 will_transition = mac_execve_will_transition(oldcred, imgp->vp,
593 interplabel, imgp);
594 credential_changing |= will_transition;
595 #endif
596
597 if (credential_changing &&
598 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
599 (p->p_flag & P_TRACED) == 0) {
600 /*
601 * Turn off syscall tracing for set-id programs, except for
602 * root. Record any set-id flags first to make sure that
603 * we do not regain any tracing during a possible block.
604 */
605 setsugid(p);
606
607 #ifdef KTRACE
608 if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0))
609 ktrprocexec(p, &tracecred, &tracevp);
610 #endif
611 /*
612 * Close any file descriptors 0..2 that reference procfs,
613 * then make sure file descriptors 0..2 are in use.
614 *
615 * setugidsafety() may call closef() and then pfind()
616 * which may grab the process lock.
617 * fdcheckstd() may call falloc() which may block to
618 * allocate memory, so temporarily drop the process lock.
619 */
620 PROC_UNLOCK(p);
621 VOP_UNLOCK(imgp->vp, 0, td);
622 setugidsafety(td);
623 error = fdcheckstd(td);
624 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
625 if (error != 0)
626 goto done1;
627 PROC_LOCK(p);
628 /*
629 * Set the new credentials.
630 */
631 crcopy(newcred, oldcred);
632 if (attr.va_mode & VSUID)
633 change_euid(newcred, euip);
634 if (attr.va_mode & VSGID)
635 change_egid(newcred, attr.va_gid);
636 #ifdef MAC
637 if (will_transition) {
638 mac_execve_transition(oldcred, newcred, imgp->vp,
639 interplabel, imgp);
640 }
641 #endif
642 /*
643 * Implement correct POSIX saved-id behavior.
644 *
645 * XXXMAC: Note that the current logic will save the
646 * uid and gid if a MAC domain transition occurs, even
647 * though maybe it shouldn't.
648 */
649 change_svuid(newcred, newcred->cr_uid);
650 change_svgid(newcred, newcred->cr_gid);
651 p->p_ucred = newcred;
652 newcred = NULL;
653 } else {
654 if (oldcred->cr_uid == oldcred->cr_ruid &&
655 oldcred->cr_gid == oldcred->cr_rgid)
656 p->p_flag &= ~P_SUGID;
657 /*
658 * Implement correct POSIX saved-id behavior.
659 *
660 * XXX: It's not clear that the existing behavior is
661 * POSIX-compliant. A number of sources indicate that the
662 * saved uid/gid should only be updated if the new ruid is
663 * not equal to the old ruid, or the new euid is not equal
664 * to the old euid and the new euid is not equal to the old
665 * ruid. The FreeBSD code always updates the saved uid/gid.
666 * Also, this code uses the new (replaced) euid and egid as
667 * the source, which may or may not be the right ones to use.
668 */
669 if (oldcred->cr_svuid != oldcred->cr_uid ||
670 oldcred->cr_svgid != oldcred->cr_gid) {
671 crcopy(newcred, oldcred);
672 change_svuid(newcred, newcred->cr_uid);
673 change_svgid(newcred, newcred->cr_gid);
674 p->p_ucred = newcred;
675 newcred = NULL;
676 }
677 }
678
679 /*
680 * Store the vp for use in procfs. This vnode was referenced prior
681 * to locking the proc lock.
682 */
683 textvp = p->p_textvp;
684 p->p_textvp = ndp->ni_vp;
685
686 #ifdef KDTRACE_HOOKS
687 /*
688 * Tell the DTrace fasttrap provider about the exec if it
689 * has declared an interest.
690 */
691 if (dtrace_fasttrap_exec)
692 dtrace_fasttrap_exec(p);
693 #endif
694
695 /*
696 * Notify others that we exec'd, and clear the P_INEXEC flag
697 * as we're now a bona fide freshly-execed process.
698 */
699 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
700 p->p_flag &= ~P_INEXEC;
701
702 /*
703 * If tracing the process, trap to debugger so breakpoints
704 * can be set before the program executes.
705 * Use tdsignal to deliver signal to current thread, use
706 * psignal may cause the signal to be delivered to wrong thread
707 * because that thread will exit, remember we are going to enter
708 * single thread mode.
709 */
710 if (p->p_flag & P_TRACED)
711 tdsignal(p, td, SIGTRAP, NULL);
712
713 /* clear "fork but no exec" flag, as we _are_ execing */
714 p->p_acflag &= ~AFORK;
715
716 /*
717 * Free any previous argument cache and replace it with
718 * the new argument cache, if any.
719 */
720 oldargs = p->p_args;
721 p->p_args = newargs;
722 newargs = NULL;
723
724 #ifdef HWPMC_HOOKS
725 /*
726 * Check if system-wide sampling is in effect or if the
727 * current process is using PMCs. If so, do exec() time
728 * processing. This processing needs to happen AFTER the
729 * P_INEXEC flag is cleared.
730 *
731 * The proc lock needs to be released before taking the PMC
732 * SX.
733 */
734 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
735 PROC_UNLOCK(p);
736 pe.pm_credentialschanged = credential_changing;
737 pe.pm_entryaddr = imgp->entry_addr;
738
739 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
740 } else
741 PROC_UNLOCK(p);
742 #else /* !HWPMC_HOOKS */
743 PROC_UNLOCK(p);
744 #endif
745
746 /* Set values passed into the program in registers. */
747 if (p->p_sysent->sv_setregs)
748 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
749 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
750 else
751 exec_setregs(td, imgp->entry_addr,
752 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
753
754 vfs_mark_atime(imgp->vp, td);
755
756 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
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 /*
769 * Handle deferred decrement of ref counts.
770 */
771 if (textvp != NULL) {
772 int tvfslocked;
773
774 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
775 vrele(textvp);
776 VFS_UNLOCK_GIANT(tvfslocked);
777 }
778 if (ndp->ni_vp && error != 0)
779 vrele(ndp->ni_vp);
780 #ifdef KTRACE
781 if (tracevp != NULL) {
782 int tvfslocked;
783
784 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
785 vrele(tracevp);
786 VFS_UNLOCK_GIANT(tvfslocked);
787 }
788 if (tracecred != NULL)
789 crfree(tracecred);
790 #endif
791 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY, td);
792 pargs_drop(oldargs);
793 pargs_drop(newargs);
794 if (oldsigacts != NULL)
795 sigacts_free(oldsigacts);
796
797 exec_fail_dealloc:
798
799 /*
800 * free various allocated resources
801 */
802 if (imgp->firstpage != NULL)
803 exec_unmap_first_page(imgp);
804
805 if (imgp->vp != NULL) {
806 NDFREE(ndp, NDF_ONLY_PNBUF);
807 if (imgp->opened)
808 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
809 vput(imgp->vp);
810 }
811
812 if (imgp->object != NULL)
813 vm_object_deallocate(imgp->object);
814
815 free(imgp->freepath, M_TEMP);
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 #if VM_NRESERVLEVEL > 0
868 if ((object->flags & OBJ_COLORED) == 0) {
869 object->flags |= OBJ_COLORED;
870 object->pg_color = 0;
871 }
872 #endif
873 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
874 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
875 initial_pagein = VM_INITIAL_PAGEIN;
876 if (initial_pagein > object->size)
877 initial_pagein = object->size;
878 for (i = 1; i < initial_pagein; i++) {
879 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
880 if (ma[i]->valid)
881 break;
882 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
883 break;
884 vm_page_busy(ma[i]);
885 } else {
886 ma[i] = vm_page_alloc(object, i,
887 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
888 if (ma[i] == NULL)
889 break;
890 }
891 }
892 initial_pagein = i;
893 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
894 ma[0] = vm_page_lookup(object, 0);
895 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
896 (ma[0]->valid == 0)) {
897 if (ma[0]) {
898 vm_page_lock_queues();
899 vm_page_free(ma[0]);
900 vm_page_unlock_queues();
901 }
902 VM_OBJECT_UNLOCK(object);
903 return (EIO);
904 }
905 }
906 vm_page_lock_queues();
907 vm_page_hold(ma[0]);
908 vm_page_unlock_queues();
909 vm_page_wakeup(ma[0]);
910 VM_OBJECT_UNLOCK(object);
911
912 imgp->firstpage = sf_buf_alloc(ma[0], 0);
913 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
914
915 return (0);
916 }
917
918 void
919 exec_unmap_first_page(imgp)
920 struct image_params *imgp;
921 {
922 vm_page_t m;
923
924 if (imgp->firstpage != NULL) {
925 m = sf_buf_page(imgp->firstpage);
926 sf_buf_free(imgp->firstpage);
927 imgp->firstpage = NULL;
928 vm_page_lock_queues();
929 vm_page_unhold(m);
930 vm_page_unlock_queues();
931 }
932 }
933
934 /*
935 * Destroy old address space, and allocate a new stack
936 * The new stack is only SGROWSIZ large because it is grown
937 * automatically in trap.c.
938 */
939 int
940 exec_new_vmspace(imgp, sv)
941 struct image_params *imgp;
942 struct sysentvec *sv;
943 {
944 int error;
945 struct proc *p = imgp->proc;
946 struct vmspace *vmspace = p->p_vmspace;
947 vm_offset_t sv_minuser, stack_addr;
948 vm_map_t map;
949 u_long ssiz;
950
951 imgp->vmspace_destroyed = 1;
952 imgp->sysent = sv;
953
954 /* May be called with Giant held */
955 EVENTHANDLER_INVOKE(process_exec, p, imgp);
956
957 /*
958 * Blow away entire process VM, if address space not shared,
959 * otherwise, create a new VM space so that other threads are
960 * not disrupted
961 */
962 map = &vmspace->vm_map;
963 if (map_at_zero)
964 sv_minuser = sv->sv_minuser;
965 else
966 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
967 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
968 vm_map_max(map) == sv->sv_maxuser) {
969 shmexit(vmspace);
970 pmap_remove_pages(vmspace_pmap(vmspace));
971 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
972 } else {
973 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
974 if (error)
975 return (error);
976 vmspace = p->p_vmspace;
977 map = &vmspace->vm_map;
978 }
979
980 /* Allocate a new stack */
981 if (sv->sv_maxssiz != NULL)
982 ssiz = *sv->sv_maxssiz;
983 else
984 ssiz = maxssiz;
985 stack_addr = sv->sv_usrstack - ssiz;
986 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
987 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
988 if (error)
989 return (error);
990
991 #ifdef __ia64__
992 /* Allocate a new register stack */
993 stack_addr = IA64_BACKINGSTORE;
994 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
995 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
996 if (error)
997 return (error);
998 #endif
999
1000 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1001 * VM_STACK case, but they are still used to monitor the size of the
1002 * process stack so we can check the stack rlimit.
1003 */
1004 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1005 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1006
1007 return (0);
1008 }
1009
1010 /*
1011 * Copy out argument and environment strings from the old process address
1012 * space into the temporary string buffer.
1013 */
1014 int
1015 exec_copyin_args(struct image_args *args, char *fname,
1016 enum uio_seg segflg, char **argv, char **envv)
1017 {
1018 char *argp, *envp;
1019 int error;
1020 size_t length;
1021
1022 error = 0;
1023
1024 bzero(args, sizeof(*args));
1025 if (argv == NULL)
1026 return (EFAULT);
1027 /*
1028 * Allocate temporary demand zeroed space for argument and
1029 * environment strings:
1030 *
1031 * o ARG_MAX for argument and environment;
1032 * o MAXSHELLCMDLEN for the name of interpreters.
1033 */
1034 args->buf = (char *) kmem_alloc_wait(exec_map,
1035 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1036 if (args->buf == NULL)
1037 return (ENOMEM);
1038 args->begin_argv = args->buf;
1039 args->endp = args->begin_argv;
1040 args->stringspace = ARG_MAX;
1041
1042 args->fname = args->buf + ARG_MAX;
1043
1044 /*
1045 * Copy the file name.
1046 */
1047 error = (segflg == UIO_SYSSPACE) ?
1048 copystr(fname, args->fname, PATH_MAX, &length) :
1049 copyinstr(fname, args->fname, PATH_MAX, &length);
1050 if (error != 0)
1051 goto err_exit;
1052
1053 /*
1054 * extract arguments first
1055 */
1056 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1057 if (argp == (caddr_t) -1) {
1058 error = EFAULT;
1059 goto err_exit;
1060 }
1061 if ((error = copyinstr(argp, args->endp,
1062 args->stringspace, &length))) {
1063 if (error == ENAMETOOLONG)
1064 error = E2BIG;
1065 goto err_exit;
1066 }
1067 args->stringspace -= length;
1068 args->endp += length;
1069 args->argc++;
1070 }
1071
1072 args->begin_envv = args->endp;
1073
1074 /*
1075 * extract environment strings
1076 */
1077 if (envv) {
1078 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1079 if (envp == (caddr_t)-1) {
1080 error = EFAULT;
1081 goto err_exit;
1082 }
1083 if ((error = copyinstr(envp, args->endp,
1084 args->stringspace, &length))) {
1085 if (error == ENAMETOOLONG)
1086 error = E2BIG;
1087 goto err_exit;
1088 }
1089 args->stringspace -= length;
1090 args->endp += length;
1091 args->envc++;
1092 }
1093 }
1094
1095 return (0);
1096
1097 err_exit:
1098 exec_free_args(args);
1099 return (error);
1100 }
1101
1102 static void
1103 exec_free_args(struct image_args *args)
1104 {
1105
1106 if (args->buf) {
1107 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1108 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1109 args->buf = NULL;
1110 }
1111 }
1112
1113 /*
1114 * Copy strings out to the new process address space, constructing new arg
1115 * and env vector tables. Return a pointer to the base so that it can be used
1116 * as the initial stack pointer.
1117 */
1118 register_t *
1119 exec_copyout_strings(imgp)
1120 struct image_params *imgp;
1121 {
1122 int argc, envc;
1123 char **vectp;
1124 char *stringp, *destp;
1125 register_t *stack_base;
1126 struct ps_strings *arginfo;
1127 struct proc *p;
1128 size_t execpath_len;
1129 int szsigcode;
1130
1131 /*
1132 * Calculate string base and vector table pointers.
1133 * Also deal with signal trampoline code for this exec type.
1134 */
1135 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1136 execpath_len = strlen(imgp->execpath) + 1;
1137 else
1138 execpath_len = 0;
1139 p = imgp->proc;
1140 szsigcode = 0;
1141 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1142 if (p->p_sysent->sv_szsigcode != NULL)
1143 szsigcode = *(p->p_sysent->sv_szsigcode);
1144 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1145 roundup(execpath_len, sizeof(char *)) -
1146 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1147
1148 /*
1149 * install sigcode
1150 */
1151 if (szsigcode)
1152 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1153 szsigcode), szsigcode);
1154
1155 /*
1156 * Copy the image path for the rtld.
1157 */
1158 if (execpath_len != 0) {
1159 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1160 copyout(imgp->execpath, (void *)imgp->execpathp,
1161 execpath_len);
1162 }
1163
1164 /*
1165 * If we have a valid auxargs ptr, prepare some room
1166 * on the stack.
1167 */
1168 if (imgp->auxargs) {
1169 /*
1170 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1171 * lower compatibility.
1172 */
1173 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1174 (AT_COUNT * 2);
1175 /*
1176 * The '+ 2' is for the null pointers at the end of each of
1177 * the arg and env vector sets,and imgp->auxarg_size is room
1178 * for argument of Runtime loader.
1179 */
1180 vectp = (char **)(destp - (imgp->args->argc +
1181 imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1182 sizeof(char *));
1183 } else {
1184 /*
1185 * The '+ 2' is for the null pointers at the end of each of
1186 * the arg and env vector sets
1187 */
1188 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1189 sizeof(char *));
1190 }
1191
1192 /*
1193 * vectp also becomes our initial stack base
1194 */
1195 stack_base = (register_t *)vectp;
1196
1197 stringp = imgp->args->begin_argv;
1198 argc = imgp->args->argc;
1199 envc = imgp->args->envc;
1200
1201 /*
1202 * Copy out strings - arguments and environment.
1203 */
1204 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1205
1206 /*
1207 * Fill in "ps_strings" struct for ps, w, etc.
1208 */
1209 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1210 suword(&arginfo->ps_nargvstr, argc);
1211
1212 /*
1213 * Fill in argument portion of vector table.
1214 */
1215 for (; argc > 0; --argc) {
1216 suword(vectp++, (long)(intptr_t)destp);
1217 while (*stringp++ != 0)
1218 destp++;
1219 destp++;
1220 }
1221
1222 /* a null vector table pointer separates the argp's from the envp's */
1223 suword(vectp++, 0);
1224
1225 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1226 suword(&arginfo->ps_nenvstr, envc);
1227
1228 /*
1229 * Fill in environment portion of vector table.
1230 */
1231 for (; envc > 0; --envc) {
1232 suword(vectp++, (long)(intptr_t)destp);
1233 while (*stringp++ != 0)
1234 destp++;
1235 destp++;
1236 }
1237
1238 /* end of vector table is a null pointer */
1239 suword(vectp, 0);
1240
1241 return (stack_base);
1242 }
1243
1244 /*
1245 * Check permissions of file to execute.
1246 * Called with imgp->vp locked.
1247 * Return 0 for success or error code on failure.
1248 */
1249 int
1250 exec_check_permissions(imgp)
1251 struct image_params *imgp;
1252 {
1253 struct vnode *vp = imgp->vp;
1254 struct vattr *attr = imgp->attr;
1255 struct thread *td;
1256 int error;
1257
1258 td = curthread; /* XXXKSE */
1259
1260 /* Get file attributes */
1261 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1262 if (error)
1263 return (error);
1264
1265 #ifdef MAC
1266 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1267 if (error)
1268 return (error);
1269 #endif
1270
1271 /*
1272 * 1) Check if file execution is disabled for the filesystem that this
1273 * file resides on.
1274 * 2) Insure that at least one execute bit is on - otherwise root
1275 * will always succeed, and we don't want to happen unless the
1276 * file really is executable.
1277 * 3) Insure that the file is a regular file.
1278 */
1279 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1280 ((attr->va_mode & 0111) == 0) ||
1281 (attr->va_type != VREG))
1282 return (EACCES);
1283
1284 /*
1285 * Zero length files can't be exec'd
1286 */
1287 if (attr->va_size == 0)
1288 return (ENOEXEC);
1289
1290 /*
1291 * Check for execute permission to file based on current credentials.
1292 */
1293 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1294 if (error)
1295 return (error);
1296
1297 /*
1298 * Check number of open-for-writes on the file and deny execution
1299 * if there are any.
1300 */
1301 if (vp->v_writecount)
1302 return (ETXTBSY);
1303
1304 /*
1305 * Call filesystem specific open routine (which does nothing in the
1306 * general case).
1307 */
1308 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1309 if (error == 0)
1310 imgp->opened = 1;
1311 return (error);
1312 }
1313
1314 /*
1315 * Exec handler registration
1316 */
1317 int
1318 exec_register(execsw_arg)
1319 const struct execsw *execsw_arg;
1320 {
1321 const struct execsw **es, **xs, **newexecsw;
1322 int count = 2; /* New slot and trailing NULL */
1323
1324 if (execsw)
1325 for (es = execsw; *es; es++)
1326 count++;
1327 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1328 if (newexecsw == NULL)
1329 return (ENOMEM);
1330 xs = newexecsw;
1331 if (execsw)
1332 for (es = execsw; *es; es++)
1333 *xs++ = *es;
1334 *xs++ = execsw_arg;
1335 *xs = NULL;
1336 if (execsw)
1337 free(execsw, M_TEMP);
1338 execsw = newexecsw;
1339 return (0);
1340 }
1341
1342 int
1343 exec_unregister(execsw_arg)
1344 const struct execsw *execsw_arg;
1345 {
1346 const struct execsw **es, **xs, **newexecsw;
1347 int count = 1;
1348
1349 if (execsw == NULL)
1350 panic("unregister with no handlers left?\n");
1351
1352 for (es = execsw; *es; es++) {
1353 if (*es == execsw_arg)
1354 break;
1355 }
1356 if (*es == NULL)
1357 return (ENOENT);
1358 for (es = execsw; *es; es++)
1359 if (*es != execsw_arg)
1360 count++;
1361 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1362 if (newexecsw == NULL)
1363 return (ENOMEM);
1364 xs = newexecsw;
1365 for (es = execsw; *es; es++)
1366 if (*es != execsw_arg)
1367 *xs++ = *es;
1368 *xs = NULL;
1369 if (execsw)
1370 free(execsw, M_TEMP);
1371 execsw = newexecsw;
1372 return (0);
1373 }
Cache object: 89c4e8c104905fe1b6ad1b43439377e9
|