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/6.0/sys/kern/kern_exec.c 151173 2005-10-09 17:28:12Z ps $");
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 atimeattr, 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 /*
713 * Here we should update the access time of the file. This must
714 * be implemented by the underlying filesystem in the same way as
715 * access timestamps for a VOP_READ() because we want to avoid
716 * blocking and/or I/O, and have not called vn_start_write().
717 */
718 if ((imgp->vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
719 VATTR_NULL(&atimeattr);
720 atimeattr.va_vaflags |= VA_EXECVE_ATIME;
721 (void)VOP_SETATTR(imgp->vp, &atimeattr, td->td_ucred, td);
722 }
723
724 done1:
725 /*
726 * Free any resources malloc'd earlier that we didn't use.
727 */
728 uifree(euip);
729 if (newcred == NULL)
730 crfree(oldcred);
731 else
732 crfree(newcred);
733 /*
734 * Handle deferred decrement of ref counts.
735 */
736 if (textvp != NULL)
737 vrele(textvp);
738 if (ndp->ni_vp && error != 0)
739 vrele(ndp->ni_vp);
740 #ifdef KTRACE
741 if (tracevp != NULL)
742 vrele(tracevp);
743 if (tracecred != NULL)
744 crfree(tracecred);
745 #endif
746 if (oldargs != NULL)
747 pargs_drop(oldargs);
748 if (newargs != NULL)
749 pargs_drop(newargs);
750 if (oldsigacts != NULL)
751 sigacts_free(oldsigacts);
752
753 exec_fail_dealloc:
754
755 /*
756 * free various allocated resources
757 */
758 if (imgp->firstpage != NULL)
759 exec_unmap_first_page(imgp);
760
761 if (imgp->vp != NULL) {
762 NDFREE(ndp, NDF_ONLY_PNBUF);
763 vput(imgp->vp);
764 }
765
766 if (imgp->object != NULL)
767 vm_object_deallocate(imgp->object);
768
769 if (error == 0) {
770 /*
771 * Stop the process here if its stop event mask has
772 * the S_EXEC bit set.
773 */
774 STOPEVENT(p, S_EXEC, 0);
775 goto done2;
776 }
777
778 exec_fail:
779 /* we're done here, clear P_INEXEC */
780 PROC_LOCK(p);
781 p->p_flag &= ~P_INEXEC;
782 PROC_UNLOCK(p);
783
784 if (imgp->vmspace_destroyed) {
785 /* sorry, no more process anymore. exit gracefully */
786 #ifdef MAC
787 mac_execve_exit(imgp);
788 if (interplabel != NULL)
789 mac_vnode_label_free(interplabel);
790 #endif
791 VFS_UNLOCK_GIANT(vfslocked);
792 exec_free_args(args);
793 exit1(td, W_EXITCODE(0, SIGABRT));
794 /* NOT REACHED */
795 error = 0;
796 }
797 done2:
798 #ifdef MAC
799 mac_execve_exit(imgp);
800 if (interplabel != NULL)
801 mac_vnode_label_free(interplabel);
802 #endif
803 VFS_UNLOCK_GIANT(vfslocked);
804 return (error);
805 }
806
807 int
808 exec_map_first_page(imgp)
809 struct image_params *imgp;
810 {
811 int rv, i;
812 int initial_pagein;
813 vm_page_t ma[VM_INITIAL_PAGEIN];
814 vm_object_t object;
815
816 if (imgp->firstpage != NULL)
817 exec_unmap_first_page(imgp);
818
819 object = imgp->vp->v_object;
820 if (object == NULL)
821 return (EACCES);
822 VM_OBJECT_LOCK(object);
823 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
824 if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
825 initial_pagein = VM_INITIAL_PAGEIN;
826 if (initial_pagein > object->size)
827 initial_pagein = object->size;
828 for (i = 1; i < initial_pagein; i++) {
829 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
830 if (ma[i]->valid)
831 break;
832 vm_page_lock_queues();
833 if ((ma[i]->flags & PG_BUSY) || ma[i]->busy) {
834 vm_page_unlock_queues();
835 break;
836 }
837 vm_page_busy(ma[i]);
838 vm_page_unlock_queues();
839 } else {
840 ma[i] = vm_page_alloc(object, i,
841 VM_ALLOC_NORMAL);
842 if (ma[i] == NULL)
843 break;
844 }
845 }
846 initial_pagein = i;
847 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
848 ma[0] = vm_page_lookup(object, 0);
849 if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
850 (ma[0]->valid == 0)) {
851 if (ma[0]) {
852 vm_page_lock_queues();
853 pmap_remove_all(ma[0]);
854 vm_page_free(ma[0]);
855 vm_page_unlock_queues();
856 }
857 VM_OBJECT_UNLOCK(object);
858 return (EIO);
859 }
860 }
861 vm_page_lock_queues();
862 vm_page_hold(ma[0]);
863 vm_page_wakeup(ma[0]);
864 vm_page_unlock_queues();
865 VM_OBJECT_UNLOCK(object);
866
867 imgp->firstpage = sf_buf_alloc(ma[0], 0);
868 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
869
870 return (0);
871 }
872
873 void
874 exec_unmap_first_page(imgp)
875 struct image_params *imgp;
876 {
877 vm_page_t m;
878
879 if (imgp->firstpage != NULL) {
880 m = sf_buf_page(imgp->firstpage);
881 sf_buf_free(imgp->firstpage);
882 imgp->firstpage = NULL;
883 vm_page_lock_queues();
884 vm_page_unhold(m);
885 vm_page_unlock_queues();
886 }
887 }
888
889 /*
890 * Destroy old address space, and allocate a new stack
891 * The new stack is only SGROWSIZ large because it is grown
892 * automatically in trap.c.
893 */
894 int
895 exec_new_vmspace(imgp, sv)
896 struct image_params *imgp;
897 struct sysentvec *sv;
898 {
899 int error;
900 struct proc *p = imgp->proc;
901 struct vmspace *vmspace = p->p_vmspace;
902 vm_offset_t stack_addr;
903 vm_map_t map;
904
905 imgp->vmspace_destroyed = 1;
906
907 /* Called with Giant held, do not depend on it! */
908 EVENTHANDLER_INVOKE(process_exec, p);
909
910 /*
911 * Here is as good a place as any to do any resource limit cleanups.
912 * This is needed if a 64 bit binary exec's a 32 bit binary - the
913 * data size limit may need to be changed to a value that makes
914 * sense for the 32 bit binary.
915 */
916 if (sv->sv_fixlimits != NULL)
917 sv->sv_fixlimits(imgp);
918
919 /*
920 * Blow away entire process VM, if address space not shared,
921 * otherwise, create a new VM space so that other threads are
922 * not disrupted
923 */
924 map = &vmspace->vm_map;
925 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
926 vm_map_max(map) == sv->sv_maxuser) {
927 shmexit(vmspace);
928 pmap_remove_pages(vmspace_pmap(vmspace), vm_map_min(map),
929 vm_map_max(map));
930 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
931 } else {
932 vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
933 vmspace = p->p_vmspace;
934 map = &vmspace->vm_map;
935 }
936
937 /* Allocate a new stack */
938 stack_addr = sv->sv_usrstack - maxssiz;
939 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
940 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
941 if (error)
942 return (error);
943
944 #ifdef __ia64__
945 /* Allocate a new register stack */
946 stack_addr = IA64_BACKINGSTORE;
947 error = vm_map_stack(map, stack_addr, (vm_size_t)maxssiz,
948 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
949 if (error)
950 return (error);
951 #endif
952
953 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
954 * VM_STACK case, but they are still used to monitor the size of the
955 * process stack so we can check the stack rlimit.
956 */
957 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
958 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - maxssiz;
959
960 return (0);
961 }
962
963 /*
964 * Copy out argument and environment strings from the old process
965 * address space into the temporary string buffer.
966 */
967 int
968 exec_copyin_args(struct image_args *args, char *fname,
969 enum uio_seg segflg, char **argv, char **envv)
970 {
971 char *argp, *envp;
972 int error;
973 size_t length;
974
975 error = 0;
976
977 bzero(args, sizeof(*args));
978 if (argv == NULL)
979 return (EFAULT);
980 /*
981 * Allocate temporary demand zeroed space for argument and
982 * environment strings:
983 *
984 * o ARG_MAX for argument and environment;
985 * o MAXSHELLCMDLEN for the name of interpreters.
986 */
987 args->buf = (char *) kmem_alloc_wait(exec_map,
988 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
989 if (args->buf == NULL)
990 return (ENOMEM);
991 args->begin_argv = args->buf;
992 args->endp = args->begin_argv;
993 args->stringspace = ARG_MAX;
994
995 args->fname = args->buf + ARG_MAX;
996
997 /*
998 * Copy the file name.
999 */
1000 error = (segflg == UIO_SYSSPACE) ?
1001 copystr(fname, args->fname, PATH_MAX, &length) :
1002 copyinstr(fname, args->fname, PATH_MAX, &length);
1003 if (error != 0)
1004 return (error);
1005
1006 /*
1007 * extract arguments first
1008 */
1009 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1010 if (argp == (caddr_t) -1)
1011 return (EFAULT);
1012 if ((error = copyinstr(argp, args->endp,
1013 args->stringspace, &length))) {
1014 if (error == ENAMETOOLONG)
1015 return (E2BIG);
1016 return (error);
1017 }
1018 args->stringspace -= length;
1019 args->endp += length;
1020 args->argc++;
1021 }
1022
1023 args->begin_envv = args->endp;
1024
1025 /*
1026 * extract environment strings
1027 */
1028 if (envv) {
1029 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1030 if (envp == (caddr_t)-1)
1031 return (EFAULT);
1032 if ((error = copyinstr(envp, args->endp,
1033 args->stringspace, &length))) {
1034 if (error == ENAMETOOLONG)
1035 return (E2BIG);
1036 return (error);
1037 }
1038 args->stringspace -= length;
1039 args->endp += length;
1040 args->envc++;
1041 }
1042 }
1043
1044 return (0);
1045 }
1046
1047 void
1048 exec_free_args(struct image_args *args)
1049 {
1050
1051 if (args->buf) {
1052 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1053 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1054 args->buf = NULL;
1055 }
1056 }
1057
1058 /*
1059 * Copy strings out to the new process address space, constructing
1060 * new arg and env vector tables. Return a pointer to the base
1061 * so that it can be used as the initial stack pointer.
1062 */
1063 register_t *
1064 exec_copyout_strings(imgp)
1065 struct image_params *imgp;
1066 {
1067 int argc, envc;
1068 char **vectp;
1069 char *stringp, *destp;
1070 register_t *stack_base;
1071 struct ps_strings *arginfo;
1072 struct proc *p;
1073 int szsigcode;
1074
1075 /*
1076 * Calculate string base and vector table pointers.
1077 * Also deal with signal trampoline code for this exec type.
1078 */
1079 p = imgp->proc;
1080 szsigcode = 0;
1081 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1082 if (p->p_sysent->sv_szsigcode != NULL)
1083 szsigcode = *(p->p_sysent->sv_szsigcode);
1084 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1085 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1086
1087 /*
1088 * install sigcode
1089 */
1090 if (szsigcode)
1091 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1092 szsigcode), szsigcode);
1093
1094 /*
1095 * If we have a valid auxargs ptr, prepare some room
1096 * on the stack.
1097 */
1098 if (imgp->auxargs) {
1099 /*
1100 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1101 * lower compatibility.
1102 */
1103 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1104 (AT_COUNT * 2);
1105 /*
1106 * The '+ 2' is for the null pointers at the end of each of
1107 * the arg and env vector sets,and imgp->auxarg_size is room
1108 * for argument of Runtime loader.
1109 */
1110 vectp = (char **)(destp - (imgp->args->argc +
1111 imgp->args->envc + 2 + imgp->auxarg_size) *
1112 sizeof(char *));
1113
1114 } else {
1115 /*
1116 * The '+ 2' is for the null pointers at the end of each of
1117 * the arg and env vector sets
1118 */
1119 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1120 sizeof(char *));
1121 }
1122
1123 /*
1124 * vectp also becomes our initial stack base
1125 */
1126 stack_base = (register_t *)vectp;
1127
1128 stringp = imgp->args->begin_argv;
1129 argc = imgp->args->argc;
1130 envc = imgp->args->envc;
1131
1132 /*
1133 * Copy out strings - arguments and environment.
1134 */
1135 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1136
1137 /*
1138 * Fill in "ps_strings" struct for ps, w, etc.
1139 */
1140 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1141 suword(&arginfo->ps_nargvstr, argc);
1142
1143 /*
1144 * Fill in argument portion of vector table.
1145 */
1146 for (; argc > 0; --argc) {
1147 suword(vectp++, (long)(intptr_t)destp);
1148 while (*stringp++ != 0)
1149 destp++;
1150 destp++;
1151 }
1152
1153 /* a null vector table pointer separates the argp's from the envp's */
1154 suword(vectp++, 0);
1155
1156 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1157 suword(&arginfo->ps_nenvstr, envc);
1158
1159 /*
1160 * Fill in environment portion of vector table.
1161 */
1162 for (; envc > 0; --envc) {
1163 suword(vectp++, (long)(intptr_t)destp);
1164 while (*stringp++ != 0)
1165 destp++;
1166 destp++;
1167 }
1168
1169 /* end of vector table is a null pointer */
1170 suword(vectp, 0);
1171
1172 return (stack_base);
1173 }
1174
1175 /*
1176 * Check permissions of file to execute.
1177 * Called with imgp->vp locked.
1178 * Return 0 for success or error code on failure.
1179 */
1180 int
1181 exec_check_permissions(imgp)
1182 struct image_params *imgp;
1183 {
1184 struct vnode *vp = imgp->vp;
1185 struct vattr *attr = imgp->attr;
1186 struct thread *td;
1187 int error;
1188
1189 td = curthread; /* XXXKSE */
1190
1191 /* Get file attributes */
1192 error = VOP_GETATTR(vp, attr, td->td_ucred, td);
1193 if (error)
1194 return (error);
1195
1196 #ifdef MAC
1197 error = mac_check_vnode_exec(td->td_ucred, imgp->vp, imgp);
1198 if (error)
1199 return (error);
1200 #endif
1201
1202 /*
1203 * 1) Check if file execution is disabled for the filesystem that this
1204 * file resides on.
1205 * 2) Insure that at least one execute bit is on - otherwise root
1206 * will always succeed, and we don't want to happen unless the
1207 * file really is executable.
1208 * 3) Insure that the file is a regular file.
1209 */
1210 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1211 ((attr->va_mode & 0111) == 0) ||
1212 (attr->va_type != VREG))
1213 return (EACCES);
1214
1215 /*
1216 * Zero length files can't be exec'd
1217 */
1218 if (attr->va_size == 0)
1219 return (ENOEXEC);
1220
1221 /*
1222 * Check for execute permission to file based on current credentials.
1223 */
1224 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1225 if (error)
1226 return (error);
1227
1228 /*
1229 * Check number of open-for-writes on the file and deny execution
1230 * if there are any.
1231 */
1232 if (vp->v_writecount)
1233 return (ETXTBSY);
1234
1235 /*
1236 * Call filesystem specific open routine (which does nothing in the
1237 * general case).
1238 */
1239 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, -1);
1240 return (error);
1241 }
1242
1243 /*
1244 * Exec handler registration
1245 */
1246 int
1247 exec_register(execsw_arg)
1248 const struct execsw *execsw_arg;
1249 {
1250 const struct execsw **es, **xs, **newexecsw;
1251 int count = 2; /* New slot and trailing NULL */
1252
1253 if (execsw)
1254 for (es = execsw; *es; es++)
1255 count++;
1256 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1257 if (newexecsw == NULL)
1258 return (ENOMEM);
1259 xs = newexecsw;
1260 if (execsw)
1261 for (es = execsw; *es; es++)
1262 *xs++ = *es;
1263 *xs++ = execsw_arg;
1264 *xs = NULL;
1265 if (execsw)
1266 free(execsw, M_TEMP);
1267 execsw = newexecsw;
1268 return (0);
1269 }
1270
1271 int
1272 exec_unregister(execsw_arg)
1273 const struct execsw *execsw_arg;
1274 {
1275 const struct execsw **es, **xs, **newexecsw;
1276 int count = 1;
1277
1278 if (execsw == NULL)
1279 panic("unregister with no handlers left?\n");
1280
1281 for (es = execsw; *es; es++) {
1282 if (*es == execsw_arg)
1283 break;
1284 }
1285 if (*es == NULL)
1286 return (ENOENT);
1287 for (es = execsw; *es; es++)
1288 if (*es != execsw_arg)
1289 count++;
1290 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1291 if (newexecsw == NULL)
1292 return (ENOMEM);
1293 xs = newexecsw;
1294 for (es = execsw; *es; es++)
1295 if (*es != execsw_arg)
1296 *xs++ = *es;
1297 *xs = NULL;
1298 if (execsw)
1299 free(execsw, M_TEMP);
1300 execsw = newexecsw;
1301 return (0);
1302 }
Cache object: 60dc10850ba3a3888d1b9431e8a61ce8
|