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/6.1/sys/kern/kern_exec.c 158179 2006-04-30 16:44:43Z cvs2svn $");
29
30 #include "opt_hwpmc_hooks.h"
31 #include "opt_ktrace.h"
32 #include "opt_mac.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/eventhandler.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/sysproto.h>
40 #include <sys/signalvar.h>
41 #include <sys/kernel.h>
42 #include <sys/mac.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/proc.h>
53 #include <sys/pioctl.h>
54 #include <sys/namei.h>
55 #include <sys/resourcevar.h>
56 #include <sys/sf_buf.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
59 #include <sys/shm.h>
60 #include <sys/sysctl.h>
61 #include <sys/vnode.h>
62 #ifdef KTRACE
63 #include <sys/ktrace.h>
64 #endif
65
66 #include <vm/vm.h>
67 #include <vm/vm_param.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_page.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_kern.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_object.h>
74 #include <vm/vm_pager.h>
75
76 #ifdef HWPMC_HOOKS
77 #include <sys/pmckern.h>
78 #endif
79
80 #include <machine/reg.h>
81
82 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
83
84 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
85 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
86 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
87 static int do_execve(struct thread *td, struct image_args *args,
88 struct mac *mac_p);
89
90 /* XXX This should be vm_size_t. */
91 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
92 NULL, 0, sysctl_kern_ps_strings, "LU", "");
93
94 /* XXX This should be vm_size_t. */
95 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
96 NULL, 0, sysctl_kern_usrstack, "LU", "");
97
98 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
99 NULL, 0, sysctl_kern_stackprot, "I", "");
100
101 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
102 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
103 &ps_arg_cache_limit, 0, "");
104
105 static int
106 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
107 {
108 struct proc *p;
109 int error;
110
111 p = curproc;
112 #ifdef SCTL_MASK32
113 if (req->flags & SCTL_MASK32) {
114 unsigned int val;
115 val = (unsigned int)p->p_sysent->sv_psstrings;
116 error = SYSCTL_OUT(req, &val, sizeof(val));
117 } else
118 #endif
119 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
120 sizeof(p->p_sysent->sv_psstrings));
121 return error;
122 }
123
124 static int
125 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
126 {
127 struct proc *p;
128 int error;
129
130 p = curproc;
131 #ifdef SCTL_MASK32
132 if (req->flags & SCTL_MASK32) {
133 unsigned int val;
134 val = (unsigned int)p->p_sysent->sv_usrstack;
135 error = SYSCTL_OUT(req, &val, sizeof(val));
136 } else
137 #endif
138 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
139 sizeof(p->p_sysent->sv_usrstack));
140 return error;
141 }
142
143 static int
144 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
145 {
146 struct proc *p;
147
148 p = curproc;
149 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
150 sizeof(p->p_sysent->sv_stackprot)));
151 }
152
153 /*
154 * Each of the items is a pointer to a `const struct execsw', hence the
155 * double pointer here.
156 */
157 static const struct execsw **execsw;
158
159 #ifndef _SYS_SYSPROTO_H_
160 struct execve_args {
161 char *fname;
162 char **argv;
163 char **envv;
164 };
165 #endif
166
167 /*
168 * MPSAFE
169 */
170 int
171 execve(td, uap)
172 struct thread *td;
173 struct execve_args /* {
174 char *fname;
175 char **argv;
176 char **envv;
177 } */ *uap;
178 {
179 int error;
180 struct image_args args;
181
182 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
183 uap->argv, uap->envv);
184
185 if (error == 0)
186 error = kern_execve(td, &args, NULL);
187
188 exec_free_args(&args);
189
190 return (error);
191 }
192
193 #ifndef _SYS_SYSPROTO_H_
194 struct __mac_execve_args {
195 char *fname;
196 char **argv;
197 char **envv;
198 struct mac *mac_p;
199 };
200 #endif
201
202 /*
203 * MPSAFE
204 */
205 int
206 __mac_execve(td, uap)
207 struct thread *td;
208 struct __mac_execve_args /* {
209 char *fname;
210 char **argv;
211 char **envv;
212 struct mac *mac_p;
213 } */ *uap;
214 {
215 #ifdef MAC
216 int error;
217 struct image_args args;
218
219 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
220 uap->argv, uap->envv);
221
222 if (error == 0)
223 error = kern_execve(td, &args, uap->mac_p);
224
225 exec_free_args(&args);
226
227 return (error);
228 #else
229 return (ENOSYS);
230 #endif
231 }
232
233 /*
234 * XXX: kern_execve has the astonishing property of not always
235 * returning to the caller. If sufficiently bad things happen during
236 * the call to do_execve(), it can end up calling exit1(); as a result,
237 * callers must avoid doing anything which they might need to undo
238 * (e.g., allocating memory).
239 */
240 int
241 kern_execve(td, args, mac_p)
242 struct thread *td;
243 struct image_args *args;
244 struct mac *mac_p;
245 {
246 struct proc *p = td->td_proc;
247 int error;
248
249 if (p->p_flag & P_HADTHREADS) {
250 PROC_LOCK(p);
251 if (thread_single(SINGLE_BOUNDARY)) {
252 PROC_UNLOCK(p);
253 return (ERESTART); /* Try again later. */
254 }
255 PROC_UNLOCK(p);
256 }
257
258 error = do_execve(td, args, mac_p);
259
260 if (p->p_flag & P_HADTHREADS) {
261 PROC_LOCK(p);
262 /*
263 * If success, we upgrade to SINGLE_EXIT state to
264 * force other threads to suicide.
265 */
266 if (error == 0)
267 thread_single(SINGLE_EXIT);
268 else
269 thread_single_end();
270 PROC_UNLOCK(p);
271 }
272
273 return (error);
274 }
275
276 /*
277 * In-kernel implementation of execve(). All arguments are assumed to be
278 * userspace pointers from the passed thread.
279 *
280 * MPSAFE
281 */
282 static int
283 do_execve(td, args, mac_p)
284 struct thread *td;
285 struct image_args *args;
286 struct mac *mac_p;
287 {
288 struct proc *p = td->td_proc;
289 struct nameidata nd, *ndp;
290 struct ucred *newcred = NULL, *oldcred;
291 struct uidinfo *euip;
292 register_t *stack_base;
293 int error, len, i;
294 struct image_params image_params, *imgp;
295 struct vattr attr;
296 int (*img_first)(struct image_params *);
297 struct pargs *oldargs = NULL, *newargs = NULL;
298 struct sigacts *oldsigacts, *newsigacts;
299 #ifdef KTRACE
300 struct vnode *tracevp = NULL;
301 struct ucred *tracecred = NULL;
302 #endif
303 struct vnode *textvp = NULL;
304 int credential_changing;
305 int vfslocked;
306 int textset;
307 #ifdef MAC
308 struct label *interplabel = NULL;
309 int will_transition;
310 #endif
311 #ifdef HWPMC_HOOKS
312 struct pmckern_procexec pe;
313 #endif
314
315 vfslocked = 0;
316 imgp = &image_params;
317
318 /*
319 * Lock the process and set the P_INEXEC flag to indicate that
320 * it should be left alone until we're done here. This is
321 * necessary to avoid race conditions - e.g. in ptrace() -
322 * that might allow a local user to illicitly obtain elevated
323 * privileges.
324 */
325 PROC_LOCK(p);
326 KASSERT((p->p_flag & P_INEXEC) == 0,
327 ("%s(): process already has P_INEXEC flag", __func__));
328 p->p_flag |= P_INEXEC;
329 PROC_UNLOCK(p);
330
331 /*
332 * Initialize part of the common data
333 */
334 imgp->proc = p;
335 imgp->execlabel = NULL;
336 imgp->attr = &attr;
337 imgp->entry_addr = 0;
338 imgp->vmspace_destroyed = 0;
339 imgp->interpreted = 0;
340 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
341 imgp->auxargs = NULL;
342 imgp->vp = NULL;
343 imgp->object = NULL;
344 imgp->firstpage = NULL;
345 imgp->ps_strings = 0;
346 imgp->auxarg_size = 0;
347 imgp->args = args;
348
349 #ifdef MAC
350 error = mac_execve_enter(imgp, mac_p);
351 if (error)
352 goto exec_fail;
353 #endif
354
355 imgp->image_header = NULL;
356
357 /*
358 * Translate the file name. namei() returns a vnode pointer
359 * in ni_vp amoung other things.
360 */
361 ndp = &nd;
362 NDINIT(ndp, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
363 UIO_SYSSPACE, args->fname, td);
364
365 interpret:
366 error = namei(ndp);
367 if (error)
368 goto exec_fail;
369
370 vfslocked = NDHASGIANT(ndp);
371 imgp->vp = ndp->ni_vp;
372
373 /*
374 * Check file permissions (also 'opens' file)
375 */
376 error = exec_check_permissions(imgp);
377 if (error)
378 goto exec_fail_dealloc;
379
380 imgp->object = imgp->vp->v_object;
381 if (imgp->object != NULL)
382 vm_object_reference(imgp->object);
383
384 /*
385 * Set VV_TEXT now so no one can write to the executable while we're
386 * activating it.
387 *
388 * Remember if this was set before and unset it in case this is not
389 * actually an executable image.
390 */
391 textset = imgp->vp->v_vflag & VV_TEXT;
392 imgp->vp->v_vflag |= VV_TEXT;
393
394 error = exec_map_first_page(imgp);
395 if (error)
396 goto exec_fail_dealloc;
397
398 /*
399 * If the current process has a special image activator it
400 * wants to try first, call it. For example, emulating shell
401 * scripts differently.
402 */
403 error = -1;
404 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
405 error = img_first(imgp);
406
407 /*
408 * Loop through the list of image activators, calling each one.
409 * An activator returns -1 if there is no match, 0 on success,
410 * and an error otherwise.
411 */
412 for (i = 0; error == -1 && execsw[i]; ++i) {
413 if (execsw[i]->ex_imgact == NULL ||
414 execsw[i]->ex_imgact == img_first) {
415 continue;
416 }
417 error = (*execsw[i]->ex_imgact)(imgp);
418 }
419
420 if (error) {
421 if (error == -1) {
422 if (textset == 0)
423 imgp->vp->v_vflag &= ~VV_TEXT;
424 error = ENOEXEC;
425 }
426 goto exec_fail_dealloc;
427 }
428
429 /*
430 * Special interpreter operation, cleanup and loop up to try to
431 * activate the interpreter.
432 */
433 if (imgp->interpreted) {
434 exec_unmap_first_page(imgp);
435 /*
436 * VV_TEXT needs to be unset for scripts. There is a short
437 * period before we determine that something is a script where
438 * VV_TEXT will be set. The vnode lock is held over this
439 * entire period so nothing should illegitimately be blocked.
440 */
441 imgp->vp->v_vflag &= ~VV_TEXT;
442 /* free name buffer and old vnode */
443 NDFREE(ndp, NDF_ONLY_PNBUF);
444 #ifdef MAC
445 interplabel = mac_vnode_label_alloc();
446 mac_copy_vnode_label(ndp->ni_vp->v_label, interplabel);
447 #endif
448 vput(ndp->ni_vp);
449 vm_object_deallocate(imgp->object);
450 imgp->object = NULL;
451 VFS_UNLOCK_GIANT(vfslocked);
452 vfslocked = 0;
453 /* set new name to that of the interpreter */
454 NDINIT(ndp, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
455 UIO_SYSSPACE, imgp->interpreter_name, td);
456 goto interpret;
457 }
458
459 /*
460 * Copy out strings (args and env) and initialize stack base
461 */
462 if (p->p_sysent->sv_copyout_strings)
463 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
464 else
465 stack_base = exec_copyout_strings(imgp);
466
467 /*
468 * If custom stack fixup routine present for this process
469 * let it do the stack setup.
470 * Else stuff argument count as first item on stack
471 */
472 if (p->p_sysent->sv_fixup != NULL)
473 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
474 else
475 suword(--stack_base, imgp->args->argc);
476
477 /*
478 * For security and other reasons, the file descriptor table cannot
479 * be shared after an exec.
480 */
481 fdunshare(p, td);
482
483 /*
484 * Malloc things before we need locks.
485 */
486 newcred = crget();
487 euip = uifind(attr.va_uid);
488 i = imgp->args->begin_envv - imgp->args->begin_argv;
489 /* Cache arguments if they fit inside our allowance */
490 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
491 newargs = pargs_alloc(i);
492 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
493 }
494
495 /* close files on exec */
496 fdcloseexec(td);
497
498 /* Get a reference to the vnode prior to locking the proc */
499 VREF(ndp->ni_vp);
500
501 /*
502 * For security and other reasons, signal handlers cannot
503 * be shared after an exec. The new process gets a copy of the old
504 * handlers. In execsigs(), the new process will have its signals
505 * reset.
506 */
507 PROC_LOCK(p);
508 if (sigacts_shared(p->p_sigacts)) {
509 oldsigacts = p->p_sigacts;
510 PROC_UNLOCK(p);
511 newsigacts = sigacts_alloc();
512 sigacts_copy(newsigacts, oldsigacts);
513 PROC_LOCK(p);
514 p->p_sigacts = newsigacts;
515 } else
516 oldsigacts = NULL;
517
518 /* Stop profiling */
519 stopprofclock(p);
520
521 /* reset caught signals */
522 execsigs(p);
523
524 /* name this process - nameiexec(p, ndp) */
525 len = min(ndp->ni_cnd.cn_namelen,MAXCOMLEN);
526 bcopy(ndp->ni_cnd.cn_nameptr, p->p_comm, len);
527 p->p_comm[len] = 0;
528
529 /*
530 * mark as execed, wakeup the process that vforked (if any) and tell
531 * it that it now has its own resources back
532 */
533 p->p_flag |= P_EXEC;
534 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
535 p->p_flag &= ~P_PPWAIT;
536 wakeup(p->p_pptr);
537 }
538
539 /*
540 * Implement image setuid/setgid.
541 *
542 * Don't honor setuid/setgid if the filesystem prohibits it or if
543 * the process is being traced.
544 *
545 * XXXMAC: For the time being, use NOSUID to also prohibit
546 * transitions on the file system.
547 */
548 oldcred = p->p_ucred;
549 credential_changing = 0;
550 credential_changing |= (attr.va_mode & VSUID) && oldcred->cr_uid !=
551 attr.va_uid;
552 credential_changing |= (attr.va_mode & VSGID) && oldcred->cr_gid !=
553 attr.va_gid;
554 #ifdef MAC
555 will_transition = mac_execve_will_transition(oldcred, imgp->vp,
556 interplabel, imgp);
557 credential_changing |= will_transition;
558 #endif
559
560 if (credential_changing &&
561 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
562 (p->p_flag & P_TRACED) == 0) {
563 /*
564 * Turn off syscall tracing for set-id programs, except for
565 * root. Record any set-id flags first to make sure that
566 * we do not regain any tracing during a possible block.
567 */
568 setsugid(p);
569 #ifdef KTRACE
570 if (p->p_tracevp != NULL && suser_cred(oldcred, SUSER_ALLOWJAIL)) {
571 mtx_lock(&ktrace_mtx);
572 p->p_traceflag = 0;
573 tracevp = p->p_tracevp;
574 p->p_tracevp = NULL;
575 tracecred = p->p_tracecred;
576 p->p_tracecred = NULL;
577 mtx_unlock(&ktrace_mtx);
578 }
579 #endif
580 /*
581 * Close any file descriptors 0..2 that reference procfs,
582 * then make sure file descriptors 0..2 are in use.
583 *
584 * setugidsafety() may call closef() and then pfind()
585 * which may grab the process lock.
586 * fdcheckstd() may call falloc() which may block to
587 * allocate memory, so temporarily drop the process lock.
588 */
589 PROC_UNLOCK(p);
590 setugidsafety(td);
591 error = fdcheckstd(td);
592 if (error != 0)
593 goto done1;
594 PROC_LOCK(p);
595 /*
596 * Set the new credentials.
597 */
598 crcopy(newcred, oldcred);
599 if (attr.va_mode & VSUID)
600 change_euid(newcred, euip);
601 if (attr.va_mode & VSGID)
602 change_egid(newcred, attr.va_gid);
603 #ifdef MAC
604 if (will_transition) {
605 mac_execve_transition(oldcred, newcred, imgp->vp,
606 interplabel, imgp);
607 }
608 #endif
609 /*
610 * Implement correct POSIX saved-id behavior.
611 *
612 * XXXMAC: Note that the current logic will save the
613 * uid and gid if a MAC domain transition occurs, even
614 * though maybe it shouldn't.
615 */
616 change_svuid(newcred, newcred->cr_uid);
617 change_svgid(newcred, newcred->cr_gid);
618 p->p_ucred = newcred;
619 newcred = NULL;
620 } else {
621 if (oldcred->cr_uid == oldcred->cr_ruid &&
622 oldcred->cr_gid == oldcred->cr_rgid)
623 p->p_flag &= ~P_SUGID;
624 /*
625 * Implement correct POSIX saved-id behavior.
626 *
627 * XXX: It's not clear that the existing behavior is
628 * POSIX-compliant. A number of sources indicate that the
629 * saved uid/gid should only be updated if the new ruid is
630 * not equal to the old ruid, or the new euid is not equal
631 * to the old euid and the new euid is not equal to the old
632 * ruid. The FreeBSD code always updates the saved uid/gid.
633 * Also, this code uses the new (replaced) euid and egid as
634 * the source, which may or may not be the right ones to use.
635 */
636 if (oldcred->cr_svuid != oldcred->cr_uid ||
637 oldcred->cr_svgid != oldcred->cr_gid) {
638 crcopy(newcred, oldcred);
639 change_svuid(newcred, newcred->cr_uid);
640 change_svgid(newcred, newcred->cr_gid);
641 p->p_ucred = newcred;
642 newcred = NULL;
643 }
644 }
645
646 /*
647 * Store the vp for use in procfs. This vnode was referenced prior
648 * to locking the proc lock.
649 */
650 textvp = p->p_textvp;
651 p->p_textvp = ndp->ni_vp;
652
653 /*
654 * Notify others that we exec'd, and clear the P_INEXEC flag
655 * as we're now a bona fide freshly-execed process.
656 */
657 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
658 p->p_flag &= ~P_INEXEC;
659
660 /*
661 * If tracing the process, trap to debugger so breakpoints
662 * can be set before the program executes.
663 * Use tdsignal to deliver signal to current thread, use
664 * psignal may cause the signal to be delivered to wrong thread
665 * because that thread will exit, remember we are going to enter
666 * single thread mode.
667 */
668 if (p->p_flag & P_TRACED)
669 tdsignal(td, SIGTRAP, SIGTARGET_TD);
670
671 /* clear "fork but no exec" flag, as we _are_ execing */
672 p->p_acflag &= ~AFORK;
673
674 /*
675 * Free any previous argument cache and replace it with
676 * the new argument cache, if any.
677 */
678 oldargs = p->p_args;
679 p->p_args = newargs;
680 newargs = NULL;
681
682 #ifdef HWPMC_HOOKS
683 /*
684 * Check if system-wide sampling is in effect or if the
685 * current process is using PMCs. If so, do exec() time
686 * processing. This processing needs to happen AFTER the
687 * P_INEXEC flag is cleared.
688 *
689 * The proc lock needs to be released before taking the PMC
690 * SX.
691 */
692 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
693 PROC_UNLOCK(p);
694 pe.pm_credentialschanged = credential_changing;
695 pe.pm_entryaddr = imgp->entry_addr;
696
697 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
698 } else
699 PROC_UNLOCK(p);
700 #else /* !HWPMC_HOOKS */
701 PROC_UNLOCK(p);
702 #endif
703
704 /* Set values passed into the program in registers. */
705 if (p->p_sysent->sv_setregs)
706 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
707 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
708 else
709 exec_setregs(td, imgp->entry_addr,
710 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
711
712 vfs_mark_atime(imgp->vp, td);
713
714 done1:
715 /*
716 * Free any resources malloc'd earlier that we didn't use.
717 */
718 uifree(euip);
719 if (newcred == NULL)
720 crfree(oldcred);
721 else
722 crfree(newcred);
723 /*
724 * Handle deferred decrement of ref counts.
725 */
726 if (textvp != NULL) {
727 int tvfslocked;
728
729 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
730 vrele(textvp);
731 VFS_UNLOCK_GIANT(tvfslocked);
732 }
733 if (ndp->ni_vp && error != 0)
734 vrele(ndp->ni_vp);
735 #ifdef KTRACE
736 if (tracevp != NULL)
737 vrele(tracevp);
738 if (tracecred != NULL)
739 crfree(tracecred);
740 #endif
741 if (oldargs != NULL)
742 pargs_drop(oldargs);
743 if (newargs != NULL)
744 pargs_drop(newargs);
745 if (oldsigacts != NULL)
746 sigacts_free(oldsigacts);
747
748 exec_fail_dealloc:
749
750 /*
751 * free various allocated resources
752 */
753 if (imgp->firstpage != NULL)
754 exec_unmap_first_page(imgp);
755
756 if (imgp->vp != NULL) {
757 NDFREE(ndp, NDF_ONLY_PNBUF);
758 vput(imgp->vp);
759 }
760
761 if (imgp->object != NULL)
762 vm_object_deallocate(imgp->object);
763
764 if (error == 0) {
765 /*
766 * Stop the process here if its stop event mask has
767 * the S_EXEC bit set.
768 */
769 STOPEVENT(p, S_EXEC, 0);
770 goto done2;
771 }
772
773 exec_fail:
774 /* we're done here, clear P_INEXEC */
775 PROC_LOCK(p);
776 p->p_flag &= ~P_INEXEC;
777 PROC_UNLOCK(p);
778
779 if (imgp->vmspace_destroyed) {
780 /* sorry, no more process anymore. exit gracefully */
781 #ifdef MAC
782 mac_execve_exit(imgp);
783 if (interplabel != NULL)
784 mac_vnode_label_free(interplabel);
785 #endif
786 VFS_UNLOCK_GIANT(vfslocked);
787 exec_free_args(args);
788 exit1(td, W_EXITCODE(0, SIGABRT));
789 /* NOT REACHED */
790 error = 0;
791 }
792 done2:
793 #ifdef MAC
794 mac_execve_exit(imgp);
795 if (interplabel != NULL)
796 mac_vnode_label_free(interplabel);
797 #endif
798 VFS_UNLOCK_GIANT(vfslocked);
799 return (error);
800 }
801
802 int
803 exec_map_first_page(imgp)
804 struct image_params *imgp;
805 {
806 int rv, i;
807 int initial_pagein;
808 vm_page_t ma[VM_INITIAL_PAGEIN];
809 vm_object_t object;
810
811 if (imgp->firstpage != NULL)
812 exec_unmap_first_page(imgp);
813
814 object = imgp->vp->v_object;
815 if (object == NULL)
816 return (EACCES);
817 VM_OBJECT_LOCK(object);
818 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
819 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
820 initial_pagein = VM_INITIAL_PAGEIN;
821 if (initial_pagein > object->size)
822 initial_pagein = object->size;
823 for (i = 1; i < initial_pagein; i++) {
824 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
825 if (ma[i]->valid)
826 break;
827 vm_page_lock_queues();
828 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
829 vm_page_unlock_queues();
830 break;
831 }
832 vm_page_busy(ma[i]);
833 vm_page_unlock_queues();
834 } else {
835 ma[i] = vm_page_alloc(object, i,
836 VM_ALLOC_NORMAL);
837 if (ma[i] == NULL)
838 break;
839 }
840 }
841 initial_pagein = i;
842 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
843 ma[0] = vm_page_lookup(object, 0);
844 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
845 (ma[0]->valid == 0)) {
846 if (ma[0]) {
847 vm_page_lock_queues();
848 pmap_remove_all(ma[0]);
849 vm_page_free(ma[0]);
850 vm_page_unlock_queues();
851 }
852 VM_OBJECT_UNLOCK(object);
853 return (EIO);
854 }
855 }
856 vm_page_lock_queues();
857 vm_page_hold(ma[0]);
858 vm_page_wakeup(ma[0]);
859 vm_page_unlock_queues();
860 VM_OBJECT_UNLOCK(object);
861
862 imgp->firstpage = sf_buf_alloc(ma[0], 0);
863 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
864
865 return (0);
866 }
867
868 void
869 exec_unmap_first_page(imgp)
870 struct image_params *imgp;
871 {
872 vm_page_t m;
873
874 if (imgp->firstpage != NULL) {
875 m = sf_buf_page(imgp->firstpage);
876 sf_buf_free(imgp->firstpage);
877 imgp->firstpage = NULL;
878 vm_page_lock_queues();
879 vm_page_unhold(m);
880 vm_page_unlock_queues();
881 }
882 }
883
884 /*
885 * Destroy old address space, and allocate a new stack
886 * The new stack is only SGROWSIZ large because it is grown
887 * automatically in trap.c.
888 */
889 int
890 exec_new_vmspace(imgp, sv)
891 struct image_params *imgp;
892 struct sysentvec *sv;
893 {
894 int error;
895 struct proc *p = imgp->proc;
896 struct vmspace *vmspace = p->p_vmspace;
897 vm_offset_t stack_addr;
898 vm_map_t map;
899
900 imgp->vmspace_destroyed = 1;
901
902 /* Called with Giant held, do not depend on it! */
903 EVENTHANDLER_INVOKE(process_exec, p);
904
905 /*
906 * Here is as good a place as any to do any resource limit cleanups.
907 * This is needed if a 64 bit binary exec's a 32 bit binary - the
908 * data size limit may need to be changed to a value that makes
909 * sense for the 32 bit binary.
910 */
911 if (sv->sv_fixlimits != NULL)
912 sv->sv_fixlimits(imgp);
913
914 /*
915 * Blow away entire process VM, if address space not shared,
916 * otherwise, create a new VM space so that other threads are
917 * not disrupted
918 */
919 map = &vmspace->vm_map;
920 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
921 vm_map_max(map) == sv->sv_maxuser) {
922 shmexit(vmspace);
923 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
924 vm_map_max(map));
925 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
926 } else {
927 vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
928 vmspace = p->p_vmspace;
929 map = &vmspace->vm_map;
930 }
931
932 /* Allocate a new stack */
933 stack_addr = sv->sv_usrstack - maxssiz;
934 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
935 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
936 if (error)
937 return (error);
938
939 #ifdef __ia64__
940 /* Allocate a new register stack */
941 stack_addr = IA64_BACKINGSTORE;
942 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
943 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
944 if (error)
945 return (error);
946 #endif
947
948 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
949 * VM_STACK case, but they are still used to monitor the size of the
950 * process stack so we can check the stack rlimit.
951 */
952 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
953 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
954
955 return (0);
956 }
957
958 /*
959 * Copy out argument and environment strings from the old process
960 * address space into the temporary string buffer.
961 */
962 int
963 exec_copyin_args(struct image_args *args, char *fname,
964 enum uio_seg segflg, char **argv, char **envv)
965 {
966 char *argp, *envp;
967 int error;
968 size_t length;
969
970 error = 0;
971
972 bzero(args, sizeof(*args));
973 if (argv == NULL)
974 return (EFAULT);
975 /*
976 * Allocate temporary demand zeroed space for argument and
977 * environment strings:
978 *
979 * o ARG_MAX for argument and environment;
980 * o MAXSHELLCMDLEN for the name of interpreters.
981 */
982 args->buf = (char *) kmem_alloc_wait(exec_map,
983 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
984 if (args->buf == NULL)
985 return (ENOMEM);
986 args->begin_argv = args->buf;
987 args->endp = args->begin_argv;
988 args->stringspace = ARG_MAX;
989
990 args->fname = args->buf + ARG_MAX;
991
992 /*
993 * Copy the file name.
994 */
995 error = (segflg == UIO_SYSSPACE) ?
996 copystr(fname, args->fname, PATH_MAX, &length) :
997 copyinstr(fname, args->fname, PATH_MAX, &length);
998 if (error != 0)
999 return (error);
1000
1001 /*
1002 * extract arguments first
1003 */
1004 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1005 if (argp == (caddr_t) -1)
1006 return (EFAULT);
1007 if ((error = copyinstr(argp, args->endp,
1008 args->stringspace, &length))) {
1009 if (error == ENAMETOOLONG)
1010 return (E2BIG);
1011 return (error);
1012 }
1013 args->stringspace -= length;
1014 args->endp += length;
1015 args->argc++;
1016 }
1017
1018 args->begin_envv = args->endp;
1019
1020 /*
1021 * extract environment strings
1022 */
1023 if (envv) {
1024 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1025 if (envp == (caddr_t)-1)
1026 return (EFAULT);
1027 if ((error = copyinstr(envp, args->endp,
1028 args->stringspace, &length))) {
1029 if (error == ENAMETOOLONG)
1030 return (E2BIG);
1031 return (error);
1032 }
1033 args->stringspace -= length;
1034 args->endp += length;
1035 args->envc++;
1036 }
1037 }
1038
1039 return (0);
1040 }
1041
1042 void
1043 exec_free_args(struct image_args *args)
1044 {
1045
1046 if (args->buf) {
1047 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1048 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1049 args->buf = NULL;
1050 }
1051 }
1052
1053 /*
1054 * Copy strings out to the new process address space, constructing
1055 * new arg and env vector tables. Return a pointer to the base
1056 * so that it can be used as the initial stack pointer.
1057 */
1058 register_t *
1059 exec_copyout_strings(imgp)
1060 struct image_params *imgp;
1061 {
1062 int argc, envc;
1063 char **vectp;
1064 char *stringp, *destp;
1065 register_t *stack_base;
1066 struct ps_strings *arginfo;
1067 struct proc *p;
1068 int szsigcode;
1069
1070 /*
1071 * Calculate string base and vector table pointers.
1072 * Also deal with signal trampoline code for this exec type.
1073 */
1074 p = imgp->proc;
1075 szsigcode = 0;
1076 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1077 if (p->p_sysent->sv_szsigcode != NULL)
1078 szsigcode = *(p->p_sysent->sv_szsigcode);
1079 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1080 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1081
1082 /*
1083 * install sigcode
1084 */
1085 if (szsigcode)
1086 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1087 szsigcode), szsigcode);
1088
1089 /*
1090 * If we have a valid auxargs ptr, prepare some room
1091 * on the stack.
1092 */
1093 if (imgp->auxargs) {
1094 /*
1095 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1096 * lower compatibility.
1097 */
1098 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1099 (AT_COUNT * 2);
1100 /*
1101 * The '+ 2' is for the null pointers at the end of each of
1102 * the arg and env vector sets,and imgp->auxarg_size is room
1103 * for argument of Runtime loader.
1104 */
1105 vectp = (char **)(destp - (imgp->args->argc +
1106 imgp->args->envc + 2 + imgp->auxarg_size) *
1107 sizeof(char *));
1108
1109 } else {
1110 /*
1111 * The '+ 2' is for the null pointers at the end of each of
1112 * the arg and env vector sets
1113 */
1114 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1115 sizeof(char *));
1116 }
1117
1118 /*
1119 * vectp also becomes our initial stack base
1120 */
1121 stack_base = (register_t *)vectp;
1122
1123 stringp = imgp->args->begin_argv;
1124 argc = imgp->args->argc;
1125 envc = imgp->args->envc;
1126
1127 /*
1128 * Copy out strings - arguments and environment.
1129 */
1130 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1131
1132 /*
1133 * Fill in "ps_strings" struct for ps, w, etc.
1134 */
1135 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1136 suword(&arginfo->ps_nargvstr, argc);
1137
1138 /*
1139 * Fill in argument portion of vector table.
1140 */
1141 for (; argc > 0; --argc) {
1142 suword(vectp++, (long)(intptr_t)destp);
1143 while (*stringp++ != 0)
1144 destp++;
1145 destp++;
1146 }
1147
1148 /* a null vector table pointer separates the argp's from the envp's */
1149 suword(vectp++, 0);
1150
1151 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1152 suword(&arginfo->ps_nenvstr, envc);
1153
1154 /*
1155 * Fill in environment portion of vector table.
1156 */
1157 for (; envc > 0; --envc) {
1158 suword(vectp++, (long)(intptr_t)destp);
1159 while (*stringp++ != 0)
1160 destp++;
1161 destp++;
1162 }
1163
1164 /* end of vector table is a null pointer */
1165 suword(vectp, 0);
1166
1167 return (stack_base);
1168 }
1169
1170 /*
1171 * Check permissions of file to execute.
1172 * Called with imgp->vp locked.
1173 * Return 0 for success or error code on failure.
1174 */
1175 int
1176 exec_check_permissions(imgp)
1177 struct image_params *imgp;
1178 {
1179 struct vnode *vp = imgp->vp;
1180 struct vattr *attr = imgp->attr;
1181 struct thread *td;
1182 int error;
1183
1184 td = curthread; /* XXXKSE */
1185
1186 /* Get file attributes */
1187 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1188 if (error)
1189 return (error);
1190
1191 #ifdef MAC
1192 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1193 if (error)
1194 return (error);
1195 #endif
1196
1197 /*
1198 * 1) Check if file execution is disabled for the filesystem that this
1199 * file resides on.
1200 * 2) Insure that at least one execute bit is on - otherwise root
1201 * will always succeed, and we don't want to happen unless the
1202 * file really is executable.
1203 * 3) Insure that the file is a regular file.
1204 */
1205 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1206 ((attr->va_mode & 0111) == 0) ||
1207 (attr->va_type != VREG))
1208 return (EACCES);
1209
1210 /*
1211 * Zero length files can't be exec'd
1212 */
1213 if (attr->va_size == 0)
1214 return (ENOEXEC);
1215
1216 /*
1217 * Check for execute permission to file based on current credentials.
1218 */
1219 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1220 if (error)
1221 return (error);
1222
1223 /*
1224 * Check number of open-for-writes on the file and deny execution
1225 * if there are any.
1226 */
1227 if (vp->v_writecount)
1228 return (ETXTBSY);
1229
1230 /*
1231 * Call filesystem specific open routine (which does nothing in the
1232 * general case).
1233 */
1234 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1235 return (error);
1236 }
1237
1238 /*
1239 * Exec handler registration
1240 */
1241 int
1242 exec_register(execsw_arg)
1243 const struct execsw *execsw_arg;
1244 {
1245 const struct execsw **es, **xs, **newexecsw;
1246 int count = 2; /* New slot and trailing NULL */
1247
1248 if (execsw)
1249 for (es = execsw; *es; es++)
1250 count++;
1251 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1252 if (newexecsw == NULL)
1253 return (ENOMEM);
1254 xs = newexecsw;
1255 if (execsw)
1256 for (es = execsw; *es; es++)
1257 *xs++ = *es;
1258 *xs++ = execsw_arg;
1259 *xs = NULL;
1260 if (execsw)
1261 free(execsw, M_TEMP);
1262 execsw = newexecsw;
1263 return (0);
1264 }
1265
1266 int
1267 exec_unregister(execsw_arg)
1268 const struct execsw *execsw_arg;
1269 {
1270 const struct execsw **es, **xs, **newexecsw;
1271 int count = 1;
1272
1273 if (execsw == NULL)
1274 panic("unregister with no handlers left?\n");
1275
1276 for (es = execsw; *es; es++) {
1277 if (*es == execsw_arg)
1278 break;
1279 }
1280 if (*es == NULL)
1281 return (ENOENT);
1282 for (es = execsw; *es; es++)
1283 if (*es != execsw_arg)
1284 count++;
1285 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1286 if (newexecsw == NULL)
1287 return (ENOMEM);
1288 xs = newexecsw;
1289 for (es = execsw; *es; es++)
1290 if (*es != execsw_arg)
1291 *xs++ = *es;
1292 *xs = NULL;
1293 if (execsw)
1294 free(execsw, M_TEMP);
1295 execsw = newexecsw;
1296 return (0);
1297 }
Cache object: 138e0758541e4f1f9bb26a84748579dc
|