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_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
33 #include "opt_thrworkq.h"
34 #include "opt_vm.h"
35
36 #include <sys/param.h>
37 #include <sys/capsicum.h>
38 #include <sys/systm.h>
39 #include <sys/eventhandler.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/sysproto.h>
43 #include <sys/signalvar.h>
44 #include <sys/kernel.h>
45 #include <sys/mount.h>
46 #include <sys/filedesc.h>
47 #include <sys/fcntl.h>
48 #include <sys/acct.h>
49 #include <sys/exec.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
52 #include <sys/wait.h>
53 #include <sys/malloc.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/pioctl.h>
57 #include <sys/namei.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
65 #include <sys/shm.h>
66 #include <sys/sysctl.h>
67 #ifdef THRWORKQ
68 #include <sys/thrworkq.h>
69 #endif
70 #include <sys/vnode.h>
71 #include <sys/stat.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_pager.h>
85
86 #ifdef HWPMC_HOOKS
87 #include <sys/pmckern.h>
88 #endif
89
90 #include <machine/reg.h>
91
92 #include <security/audit/audit.h>
93 #include <security/mac/mac_framework.h>
94
95 #ifdef KDTRACE_HOOKS
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t dtrace_fasttrap_exec;
98 #endif
99
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
104
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106
107 int coredump_pack_fileinfo = 1;
108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
109 &coredump_pack_fileinfo, 0,
110 "Enable file path packing in 'procstat -f' coredump notes");
111
112 int coredump_pack_vmmapinfo = 1;
113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
114 &coredump_pack_vmmapinfo, 0,
115 "Enable file path packing in 'procstat -v' coredump notes");
116
117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
120 static int do_execve(struct thread *td, struct image_args *args,
121 struct mac *mac_p);
122
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
125 NULL, 0, sysctl_kern_ps_strings, "LU", "");
126
127 /* XXX This should be vm_size_t. */
128 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
129 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
130
131 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
132 NULL, 0, sysctl_kern_stackprot, "I", "");
133
134 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
135 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
136 &ps_arg_cache_limit, 0, "");
137
138 static int disallow_high_osrel;
139 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
140 &disallow_high_osrel, 0,
141 "Disallow execution of binaries built for higher version of the world");
142
143 static int map_at_zero = 0;
144 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
145 "Permit processes to map an object at virtual address 0.");
146
147 static int
sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)148 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
149 {
150 struct proc *p;
151 int error;
152
153 p = curproc;
154 #ifdef SCTL_MASK32
155 if (req->flags & SCTL_MASK32) {
156 unsigned int val;
157 val = (unsigned int)p->p_sysent->sv_psstrings;
158 error = SYSCTL_OUT(req, &val, sizeof(val));
159 } else
160 #endif
161 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
162 sizeof(p->p_sysent->sv_psstrings));
163 return error;
164 }
165
166 static int
sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)167 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
168 {
169 struct proc *p;
170 int error;
171
172 p = curproc;
173 #ifdef SCTL_MASK32
174 if (req->flags & SCTL_MASK32) {
175 unsigned int val;
176 val = (unsigned int)p->p_sysent->sv_usrstack;
177 error = SYSCTL_OUT(req, &val, sizeof(val));
178 } else
179 #endif
180 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
181 sizeof(p->p_sysent->sv_usrstack));
182 return error;
183 }
184
185 static int
sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)186 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
187 {
188 struct proc *p;
189
190 p = curproc;
191 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
192 sizeof(p->p_sysent->sv_stackprot)));
193 }
194
195 /*
196 * Each of the items is a pointer to a `const struct execsw', hence the
197 * double pointer here.
198 */
199 static const struct execsw **execsw;
200
201 #ifndef _SYS_SYSPROTO_H_
202 struct execve_args {
203 char *fname;
204 char **argv;
205 char **envv;
206 };
207 #endif
208
209 int
sys_execve(struct thread * td,struct execve_args * uap)210 sys_execve(struct thread *td, struct execve_args *uap)
211 {
212 struct image_args args;
213 struct vmspace *oldvmspace;
214 int error;
215
216 error = pre_execve(td, &oldvmspace);
217
218 if (error != 0)
219 return (error);
220 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
221 uap->argv, uap->envv);
222 if (error == 0)
223 error = kern_execve(td, &args, NULL);
224 post_execve(td, error, oldvmspace);
225 return (error);
226 }
227
228 #ifndef _SYS_SYSPROTO_H_
229 struct fexecve_args {
230 int fd;
231 char **argv;
232 char **envv;
233 }
234 #endif
235 int
sys_fexecve(struct thread * td,struct fexecve_args * uap)236 sys_fexecve(struct thread *td, struct fexecve_args *uap)
237 {
238 struct image_args args;
239 struct vmspace *oldvmspace;
240 int error;
241
242 error = pre_execve(td, &oldvmspace);
243 if (error != 0)
244 return (error);
245 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
246 uap->argv, uap->envv);
247 if (error == 0) {
248 args.fd = uap->fd;
249 error = kern_execve(td, &args, NULL);
250 }
251 post_execve(td, error, oldvmspace);
252 return (error);
253 }
254
255 #ifndef _SYS_SYSPROTO_H_
256 struct __mac_execve_args {
257 char *fname;
258 char **argv;
259 char **envv;
260 struct mac *mac_p;
261 };
262 #endif
263
264 int
sys___mac_execve(struct thread * td,struct __mac_execve_args * uap)265 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
266 {
267 #ifdef MAC
268 struct image_args args;
269 struct vmspace *oldvmspace;
270 int error;
271
272 error = pre_execve(td, &oldvmspace);
273 if (error != 0)
274 return (error);
275 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
276 uap->argv, uap->envv);
277 if (error == 0)
278 error = kern_execve(td, &args, uap->mac_p);
279 post_execve(td, error, oldvmspace);
280 return (error);
281 #else
282 return (ENOSYS);
283 #endif
284 }
285
286 int
pre_execve(struct thread * td,struct vmspace ** oldvmspace)287 pre_execve(struct thread *td, struct vmspace **oldvmspace)
288 {
289 struct proc *p;
290 int error;
291
292 KASSERT(td == curthread, ("non-current thread %p", td));
293 error = 0;
294 p = td->td_proc;
295 if ((p->p_flag & P_HADTHREADS) != 0) {
296 PROC_LOCK(p);
297 if (thread_single(p, SINGLE_BOUNDARY) != 0)
298 error = ERESTART;
299 PROC_UNLOCK(p);
300 }
301 #ifdef THRWORKQ
302 if (error == 0)
303 thrworkq_exit(p);
304 #endif
305 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
306 ("nested execve"));
307 *oldvmspace = p->p_vmspace;
308 return (error);
309 }
310
311 void
post_execve(struct thread * td,int error,struct vmspace * oldvmspace)312 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
313 {
314 struct proc *p;
315
316 KASSERT(td == curthread, ("non-current thread %p", td));
317 p = td->td_proc;
318 if ((p->p_flag & P_HADTHREADS) != 0) {
319 PROC_LOCK(p);
320 /*
321 * If success, we upgrade to SINGLE_EXIT state to
322 * force other threads to suicide.
323 */
324 if (error == 0)
325 thread_single(p, SINGLE_EXIT);
326 else
327 thread_single_end(p, SINGLE_BOUNDARY);
328 PROC_UNLOCK(p);
329 }
330 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
331 KASSERT(p->p_vmspace != oldvmspace,
332 ("oldvmspace still used"));
333 vmspace_free(oldvmspace);
334 td->td_pflags &= ~TDP_EXECVMSPC;
335 }
336 }
337
338 /*
339 * XXX: kern_execve has the astonishing property of not always returning to
340 * the caller. If sufficiently bad things happen during the call to
341 * do_execve(), it can end up calling exit1(); as a result, callers must
342 * avoid doing anything which they might need to undo (e.g., allocating
343 * memory).
344 */
345 int
kern_execve(struct thread * td,struct image_args * args,struct mac * mac_p)346 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
347 {
348
349 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
350 args->begin_envv - args->begin_argv);
351 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
352 args->endp - args->begin_envv);
353 return (do_execve(td, args, mac_p));
354 }
355
356 /*
357 * In-kernel implementation of execve(). All arguments are assumed to be
358 * userspace pointers from the passed thread.
359 */
360 static int
do_execve(td,args,mac_p)361 do_execve(td, args, mac_p)
362 struct thread *td;
363 struct image_args *args;
364 struct mac *mac_p;
365 {
366 struct proc *p = td->td_proc;
367 struct nameidata nd;
368 struct ucred *newcred = NULL, *oldcred;
369 struct uidinfo *euip = NULL;
370 register_t *stack_base;
371 int error, i;
372 struct image_params image_params, *imgp;
373 struct vattr attr;
374 int (*img_first)(struct image_params *);
375 struct pargs *oldargs = NULL, *newargs = NULL;
376 struct sigacts *oldsigacts, *newsigacts;
377 #ifdef KTRACE
378 struct vnode *tracevp = NULL;
379 struct ucred *tracecred = NULL;
380 #endif
381 struct vnode *oldtextvp = NULL, *newtextvp;
382 cap_rights_t rights;
383 int credential_changing;
384 int textset;
385 #ifdef MAC
386 struct label *interpvplabel = NULL;
387 int will_transition;
388 #endif
389 #ifdef HWPMC_HOOKS
390 struct pmckern_procexec pe;
391 #endif
392 static const char fexecv_proc_title[] = "(fexecv)";
393
394 imgp = &image_params;
395
396 /*
397 * Lock the process and set the P_INEXEC flag to indicate that
398 * it should be left alone until we're done here. This is
399 * necessary to avoid race conditions - e.g. in ptrace() -
400 * that might allow a local user to illicitly obtain elevated
401 * privileges.
402 */
403 PROC_LOCK(p);
404 KASSERT((p->p_flag & P_INEXEC) == 0,
405 ("%s(): process already has P_INEXEC flag", __func__));
406 p->p_flag |= P_INEXEC;
407 PROC_UNLOCK(p);
408
409 /*
410 * Initialize part of the common data
411 */
412 bzero(imgp, sizeof(*imgp));
413 imgp->proc = p;
414 imgp->attr = &attr;
415 imgp->args = args;
416
417 #ifdef MAC
418 error = mac_execve_enter(imgp, mac_p);
419 if (error)
420 goto exec_fail;
421 #endif
422
423 /*
424 * Translate the file name. namei() returns a vnode pointer
425 * in ni_vp amoung other things.
426 *
427 * XXXAUDIT: It would be desirable to also audit the name of the
428 * interpreter if this is an interpreted binary.
429 */
430 if (args->fname != NULL) {
431 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
432 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
433 }
434
435 SDT_PROBE1(proc, , , exec, args->fname);
436
437 interpret:
438 if (args->fname != NULL) {
439 #ifdef CAPABILITY_MODE
440 /*
441 * While capability mode can't reach this point via direct
442 * path arguments to execve(), we also don't allow
443 * interpreters to be used in capability mode (for now).
444 * Catch indirect lookups and return a permissions error.
445 */
446 if (IN_CAPABILITY_MODE(td)) {
447 error = ECAPMODE;
448 goto exec_fail;
449 }
450 #endif
451 error = namei(&nd);
452 if (error)
453 goto exec_fail;
454
455 newtextvp = nd.ni_vp;
456 imgp->vp = newtextvp;
457 } else {
458 AUDIT_ARG_FD(args->fd);
459 /*
460 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
461 */
462 error = fgetvp_exec(td, args->fd,
463 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
464 if (error)
465 goto exec_fail;
466 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
467 AUDIT_ARG_VNODE1(newtextvp);
468 imgp->vp = newtextvp;
469 }
470
471 /*
472 * Check file permissions (also 'opens' file)
473 */
474 error = exec_check_permissions(imgp);
475 if (error)
476 goto exec_fail_dealloc;
477
478 imgp->object = imgp->vp->v_object;
479 if (imgp->object != NULL)
480 vm_object_reference(imgp->object);
481
482 /*
483 * Set VV_TEXT now so no one can write to the executable while we're
484 * activating it.
485 *
486 * Remember if this was set before and unset it in case this is not
487 * actually an executable image.
488 */
489 textset = VOP_IS_TEXT(imgp->vp);
490 VOP_SET_TEXT(imgp->vp);
491
492 error = exec_map_first_page(imgp);
493 if (error)
494 goto exec_fail_dealloc;
495
496 imgp->proc->p_osrel = 0;
497 /*
498 * If the current process has a special image activator it
499 * wants to try first, call it. For example, emulating shell
500 * scripts differently.
501 */
502 error = -1;
503 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
504 error = img_first(imgp);
505
506 /*
507 * Loop through the list of image activators, calling each one.
508 * An activator returns -1 if there is no match, 0 on success,
509 * and an error otherwise.
510 */
511 for (i = 0; error == -1 && execsw[i]; ++i) {
512 if (execsw[i]->ex_imgact == NULL ||
513 execsw[i]->ex_imgact == img_first) {
514 continue;
515 }
516 error = (*execsw[i]->ex_imgact)(imgp);
517 }
518
519 if (error) {
520 if (error == -1) {
521 if (textset == 0)
522 VOP_UNSET_TEXT(imgp->vp);
523 error = ENOEXEC;
524 }
525 goto exec_fail_dealloc;
526 }
527
528 /*
529 * Special interpreter operation, cleanup and loop up to try to
530 * activate the interpreter.
531 */
532 if (imgp->interpreted) {
533 exec_unmap_first_page(imgp);
534 /*
535 * VV_TEXT needs to be unset for scripts. There is a short
536 * period before we determine that something is a script where
537 * VV_TEXT will be set. The vnode lock is held over this
538 * entire period so nothing should illegitimately be blocked.
539 */
540 VOP_UNSET_TEXT(imgp->vp);
541 /* free name buffer and old vnode */
542 if (args->fname != NULL)
543 NDFREE(&nd, NDF_ONLY_PNBUF);
544 #ifdef MAC
545 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
546 #endif
547 if (imgp->opened) {
548 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
549 imgp->opened = 0;
550 }
551 vput(newtextvp);
552 vm_object_deallocate(imgp->object);
553 imgp->object = NULL;
554 /* set new name to that of the interpreter */
555 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
556 UIO_SYSSPACE, imgp->interpreter_name, td);
557 args->fname = imgp->interpreter_name;
558 goto interpret;
559 }
560
561 /*
562 * NB: We unlock the vnode here because it is believed that none
563 * of the sv_copyout_strings/sv_fixup operations require the vnode.
564 */
565 VOP_UNLOCK(imgp->vp, 0);
566
567 /*
568 * Do the best to calculate the full path to the image file.
569 */
570 if (imgp->auxargs != NULL &&
571 ((args->fname != NULL && args->fname[0] == '/') ||
572 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
573 imgp->execpath = args->fname;
574
575 if (disallow_high_osrel &&
576 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
577 error = ENOEXEC;
578 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
579 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
580 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
581 goto exec_fail_dealloc;
582 }
583
584 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
585 if (SV_PROC_FLAG(p, SV_CAPSICUM))
586 sys_cap_enter(td, NULL);
587
588 /*
589 * Copy out strings (args and env) and initialize stack base
590 */
591 if (p->p_sysent->sv_copyout_strings)
592 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
593 else
594 stack_base = exec_copyout_strings(imgp);
595
596 /*
597 * If custom stack fixup routine present for this process
598 * let it do the stack setup.
599 * Else stuff argument count as first item on stack
600 */
601 if (p->p_sysent->sv_fixup != NULL)
602 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
603 else
604 suword(--stack_base, imgp->args->argc);
605
606 if (args->fdp != NULL) {
607 /* Install a brand new file descriptor table. */
608 fdinstall_remapped(td, args->fdp);
609 args->fdp = NULL;
610 } else {
611 /*
612 * Keep on using the existing file descriptor table. For
613 * security and other reasons, the file descriptor table
614 * cannot be shared after an exec.
615 */
616 fdunshare(td);
617 /* close files on exec */
618 fdcloseexec(td);
619 }
620
621 /*
622 * Malloc things before we need locks.
623 */
624 i = imgp->args->begin_envv - imgp->args->begin_argv;
625 /* Cache arguments if they fit inside our allowance */
626 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
627 newargs = pargs_alloc(i);
628 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
629 }
630
631 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
632
633 /*
634 * For security and other reasons, signal handlers cannot
635 * be shared after an exec. The new process gets a copy of the old
636 * handlers. In execsigs(), the new process will have its signals
637 * reset.
638 */
639 if (sigacts_shared(p->p_sigacts)) {
640 oldsigacts = p->p_sigacts;
641 newsigacts = sigacts_alloc();
642 sigacts_copy(newsigacts, oldsigacts);
643 } else {
644 oldsigacts = NULL;
645 newsigacts = NULL; /* satisfy gcc */
646 }
647
648 PROC_LOCK(p);
649 if (oldsigacts)
650 p->p_sigacts = newsigacts;
651 oldcred = p->p_ucred;
652 /* Stop profiling */
653 stopprofclock(p);
654
655 /* reset caught signals */
656 execsigs(p);
657
658 /* name this process - nameiexec(p, ndp) */
659 bzero(p->p_comm, sizeof(p->p_comm));
660 if (args->fname)
661 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
662 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
663 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
664 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
665 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
666 #ifdef KTR
667 sched_clear_tdname(td);
668 #endif
669
670 /*
671 * mark as execed, wakeup the process that vforked (if any) and tell
672 * it that it now has its own resources back
673 */
674 p->p_flag |= P_EXEC;
675 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
676 p->p_flag2 &= ~P2_NOTRACE;
677 if (p->p_flag & P_PPWAIT) {
678 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
679 cv_broadcast(&p->p_pwait);
680 }
681
682 /*
683 * Implement image setuid/setgid.
684 *
685 * Don't honor setuid/setgid if the filesystem prohibits it or if
686 * the process is being traced.
687 *
688 * We disable setuid/setgid/etc in compatibility mode on the basis
689 * that most setugid applications are not written with that
690 * environment in mind, and will therefore almost certainly operate
691 * incorrectly. In principle there's no reason that setugid
692 * applications might not be useful in capability mode, so we may want
693 * to reconsider this conservative design choice in the future.
694 *
695 * XXXMAC: For the time being, use NOSUID to also prohibit
696 * transitions on the file system.
697 */
698 credential_changing = 0;
699 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
700 attr.va_uid;
701 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
702 attr.va_gid;
703 #ifdef MAC
704 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
705 interpvplabel, imgp);
706 credential_changing |= will_transition;
707 #endif
708
709 if (credential_changing &&
710 #ifdef CAPABILITY_MODE
711 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
712 #endif
713 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
714 (p->p_flag & P_TRACED) == 0) {
715 /*
716 * Turn off syscall tracing for set-id programs, except for
717 * root. Record any set-id flags first to make sure that
718 * we do not regain any tracing during a possible block.
719 */
720 setsugid(p);
721
722 #ifdef KTRACE
723 if (p->p_tracecred != NULL &&
724 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
725 ktrprocexec(p, &tracecred, &tracevp);
726 #endif
727 /*
728 * Close any file descriptors 0..2 that reference procfs,
729 * then make sure file descriptors 0..2 are in use.
730 *
731 * Both fdsetugidsafety() and fdcheckstd() may call functions
732 * taking sleepable locks, so temporarily drop our locks.
733 */
734 PROC_UNLOCK(p);
735 VOP_UNLOCK(imgp->vp, 0);
736 fdsetugidsafety(td);
737 error = fdcheckstd(td);
738 if (error != 0)
739 goto done1;
740 newcred = crdup(oldcred);
741 euip = uifind(attr.va_uid);
742 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
743 PROC_LOCK(p);
744 /*
745 * Set the new credentials.
746 */
747 if (attr.va_mode & S_ISUID)
748 change_euid(newcred, euip);
749 if (attr.va_mode & S_ISGID)
750 change_egid(newcred, attr.va_gid);
751 #ifdef MAC
752 if (will_transition) {
753 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
754 interpvplabel, imgp);
755 }
756 #endif
757 /*
758 * Implement correct POSIX saved-id behavior.
759 *
760 * XXXMAC: Note that the current logic will save the
761 * uid and gid if a MAC domain transition occurs, even
762 * though maybe it shouldn't.
763 */
764 change_svuid(newcred, newcred->cr_uid);
765 change_svgid(newcred, newcred->cr_gid);
766 proc_set_cred(p, newcred);
767 } else {
768 if (oldcred->cr_uid == oldcred->cr_ruid &&
769 oldcred->cr_gid == oldcred->cr_rgid)
770 p->p_flag &= ~P_SUGID;
771 /*
772 * Implement correct POSIX saved-id behavior.
773 *
774 * XXX: It's not clear that the existing behavior is
775 * POSIX-compliant. A number of sources indicate that the
776 * saved uid/gid should only be updated if the new ruid is
777 * not equal to the old ruid, or the new euid is not equal
778 * to the old euid and the new euid is not equal to the old
779 * ruid. The FreeBSD code always updates the saved uid/gid.
780 * Also, this code uses the new (replaced) euid and egid as
781 * the source, which may or may not be the right ones to use.
782 */
783 if (oldcred->cr_svuid != oldcred->cr_uid ||
784 oldcred->cr_svgid != oldcred->cr_gid) {
785 PROC_UNLOCK(p);
786 VOP_UNLOCK(imgp->vp, 0);
787 newcred = crdup(oldcred);
788 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
789 PROC_LOCK(p);
790 change_svuid(newcred, newcred->cr_uid);
791 change_svgid(newcred, newcred->cr_gid);
792 proc_set_cred(p, newcred);
793 }
794 }
795
796 /*
797 * Store the vp for use in procfs. This vnode was referenced by namei
798 * or fgetvp_exec.
799 */
800 oldtextvp = p->p_textvp;
801 p->p_textvp = newtextvp;
802
803 #ifdef KDTRACE_HOOKS
804 /*
805 * Tell the DTrace fasttrap provider about the exec if it
806 * has declared an interest.
807 */
808 if (dtrace_fasttrap_exec)
809 dtrace_fasttrap_exec(p);
810 #endif
811
812 /*
813 * Notify others that we exec'd, and clear the P_INEXEC flag
814 * as we're now a bona fide freshly-execed process.
815 */
816 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
817 p->p_flag &= ~P_INEXEC;
818
819 /* clear "fork but no exec" flag, as we _are_ execing */
820 p->p_acflag &= ~AFORK;
821
822 /*
823 * Free any previous argument cache and replace it with
824 * the new argument cache, if any.
825 */
826 oldargs = p->p_args;
827 p->p_args = newargs;
828 newargs = NULL;
829
830 #ifdef HWPMC_HOOKS
831 /*
832 * Check if system-wide sampling is in effect or if the
833 * current process is using PMCs. If so, do exec() time
834 * processing. This processing needs to happen AFTER the
835 * P_INEXEC flag is cleared.
836 *
837 * The proc lock needs to be released before taking the PMC
838 * SX.
839 */
840 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
841 PROC_UNLOCK(p);
842 VOP_UNLOCK(imgp->vp, 0);
843 pe.pm_credentialschanged = credential_changing;
844 pe.pm_entryaddr = imgp->entry_addr;
845
846 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
847 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
848 } else
849 PROC_UNLOCK(p);
850 #else /* !HWPMC_HOOKS */
851 PROC_UNLOCK(p);
852 #endif
853
854 /* Set values passed into the program in registers. */
855 if (p->p_sysent->sv_setregs)
856 (*p->p_sysent->sv_setregs)(td, imgp,
857 (u_long)(uintptr_t)stack_base);
858 else
859 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
860
861 vfs_mark_atime(imgp->vp, td->td_ucred);
862
863 SDT_PROBE1(proc, , , exec__success, args->fname);
864
865 VOP_UNLOCK(imgp->vp, 0);
866 done1:
867 /*
868 * Free any resources malloc'd earlier that we didn't use.
869 */
870 if (euip != NULL)
871 uifree(euip);
872 if (newcred != NULL)
873 crfree(oldcred);
874
875 /*
876 * Handle deferred decrement of ref counts.
877 */
878 if (oldtextvp != NULL)
879 vrele(oldtextvp);
880 #ifdef KTRACE
881 if (tracevp != NULL)
882 vrele(tracevp);
883 if (tracecred != NULL)
884 crfree(tracecred);
885 #endif
886 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
887 pargs_drop(oldargs);
888 pargs_drop(newargs);
889 if (oldsigacts != NULL)
890 sigacts_free(oldsigacts);
891
892 exec_fail_dealloc:
893
894 /*
895 * free various allocated resources
896 */
897 if (imgp->firstpage != NULL)
898 exec_unmap_first_page(imgp);
899
900 if (imgp->vp != NULL) {
901 if (args->fname)
902 NDFREE(&nd, NDF_ONLY_PNBUF);
903 if (imgp->opened)
904 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
905 if (error != 0)
906 vput(imgp->vp);
907 else
908 VOP_UNLOCK(imgp->vp, 0);
909 }
910
911 if (imgp->object != NULL)
912 vm_object_deallocate(imgp->object);
913
914 free(imgp->freepath, M_TEMP);
915
916 if (error == 0) {
917 PROC_LOCK(p);
918 td->td_dbgflags |= TDB_EXEC;
919 PROC_UNLOCK(p);
920
921 /*
922 * Stop the process here if its stop event mask has
923 * the S_EXEC bit set.
924 */
925 STOPEVENT(p, S_EXEC, 0);
926 goto done2;
927 }
928
929 exec_fail:
930 /* we're done here, clear P_INEXEC */
931 PROC_LOCK(p);
932 p->p_flag &= ~P_INEXEC;
933 PROC_UNLOCK(p);
934
935 SDT_PROBE1(proc, , , exec__failure, error);
936
937 done2:
938 #ifdef MAC
939 mac_execve_exit(imgp);
940 mac_execve_interpreter_exit(interpvplabel);
941 #endif
942 exec_free_args(args);
943
944 if (error && imgp->vmspace_destroyed) {
945 /* sorry, no more process anymore. exit gracefully */
946 exit1(td, 0, SIGABRT);
947 /* NOT REACHED */
948 }
949
950 #ifdef KTRACE
951 if (error == 0)
952 ktrprocctor(p);
953 #endif
954
955 return (error);
956 }
957
958 int
exec_map_first_page(imgp)959 exec_map_first_page(imgp)
960 struct image_params *imgp;
961 {
962 int rv, i, after, initial_pagein;
963 vm_page_t ma[VM_INITIAL_PAGEIN];
964 vm_object_t object;
965
966 if (imgp->firstpage != NULL)
967 exec_unmap_first_page(imgp);
968
969 object = imgp->vp->v_object;
970 if (object == NULL)
971 return (EACCES);
972 VM_OBJECT_WLOCK(object);
973 #if VM_NRESERVLEVEL > 0
974 vm_object_color(object, 0);
975 #endif
976 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
977 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
978 if (!vm_pager_has_page(object, 0, NULL, &after)) {
979 vm_page_lock(ma[0]);
980 vm_page_free(ma[0]);
981 vm_page_unlock(ma[0]);
982 vm_page_xunbusy(ma[0]);
983 VM_OBJECT_WUNLOCK(object);
984 return (EIO);
985 }
986 initial_pagein = min(after, VM_INITIAL_PAGEIN);
987 KASSERT(initial_pagein <= object->size,
988 ("%s: initial_pagein %d object->size %ju",
989 __func__, initial_pagein, (uintmax_t )object->size));
990 for (i = 1; i < initial_pagein; i++) {
991 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
992 if (ma[i]->valid)
993 break;
994 if (vm_page_tryxbusy(ma[i]))
995 break;
996 } else {
997 ma[i] = vm_page_alloc(object, i, VM_ALLOC_NORMAL);
998 if (ma[i] == NULL)
999 break;
1000 }
1001 }
1002 initial_pagein = i;
1003 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1004 if (rv != VM_PAGER_OK) {
1005 for (i = 0; i < initial_pagein; i++) {
1006 vm_page_lock(ma[i]);
1007 vm_page_free(ma[i]);
1008 vm_page_unlock(ma[i]);
1009 vm_page_xunbusy(ma[i]);
1010 }
1011 VM_OBJECT_WUNLOCK(object);
1012 return (EIO);
1013 }
1014 for (i = 1; i < initial_pagein; i++)
1015 vm_page_readahead_finish(ma[i]);
1016 }
1017 vm_page_xunbusy(ma[0]);
1018 vm_page_lock(ma[0]);
1019 vm_page_hold(ma[0]);
1020 vm_page_activate(ma[0]);
1021 vm_page_unlock(ma[0]);
1022 VM_OBJECT_WUNLOCK(object);
1023
1024 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1025 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1026
1027 return (0);
1028 }
1029
1030 void
exec_unmap_first_page(imgp)1031 exec_unmap_first_page(imgp)
1032 struct image_params *imgp;
1033 {
1034 vm_page_t m;
1035
1036 if (imgp->firstpage != NULL) {
1037 m = sf_buf_page(imgp->firstpage);
1038 sf_buf_free(imgp->firstpage);
1039 imgp->firstpage = NULL;
1040 vm_page_lock(m);
1041 vm_page_unhold(m);
1042 vm_page_unlock(m);
1043 }
1044 }
1045
1046 /*
1047 * Destroy old address space, and allocate a new stack
1048 * The new stack is only SGROWSIZ large because it is grown
1049 * automatically in trap.c.
1050 */
1051 int
exec_new_vmspace(imgp,sv)1052 exec_new_vmspace(imgp, sv)
1053 struct image_params *imgp;
1054 struct sysentvec *sv;
1055 {
1056 int error;
1057 struct proc *p = imgp->proc;
1058 struct vmspace *vmspace = p->p_vmspace;
1059 vm_object_t obj;
1060 struct rlimit rlim_stack;
1061 vm_offset_t sv_minuser, stack_addr;
1062 vm_map_t map;
1063 u_long ssiz;
1064
1065 imgp->vmspace_destroyed = 1;
1066 imgp->sysent = sv;
1067
1068 /* May be called with Giant held */
1069 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1070
1071 /*
1072 * Blow away entire process VM, if address space not shared,
1073 * otherwise, create a new VM space so that other threads are
1074 * not disrupted
1075 */
1076 map = &vmspace->vm_map;
1077 if (map_at_zero)
1078 sv_minuser = sv->sv_minuser;
1079 else
1080 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1081 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1082 vm_map_max(map) == sv->sv_maxuser) {
1083 shmexit(vmspace);
1084 pmap_remove_pages(vmspace_pmap(vmspace));
1085 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1086 } else {
1087 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1088 if (error)
1089 return (error);
1090 vmspace = p->p_vmspace;
1091 map = &vmspace->vm_map;
1092 }
1093
1094 /* Map a shared page */
1095 obj = sv->sv_shared_page_obj;
1096 if (obj != NULL) {
1097 vm_object_reference(obj);
1098 error = vm_map_fixed(map, obj, 0,
1099 sv->sv_shared_page_base, sv->sv_shared_page_len,
1100 VM_PROT_READ | VM_PROT_EXECUTE,
1101 VM_PROT_READ | VM_PROT_EXECUTE,
1102 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1103 if (error) {
1104 vm_object_deallocate(obj);
1105 return (error);
1106 }
1107 }
1108
1109 /* Allocate a new stack */
1110 if (imgp->stack_sz != 0) {
1111 ssiz = trunc_page(imgp->stack_sz);
1112 PROC_LOCK(p);
1113 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1114 PROC_UNLOCK(p);
1115 if (ssiz > rlim_stack.rlim_max)
1116 ssiz = rlim_stack.rlim_max;
1117 if (ssiz > rlim_stack.rlim_cur) {
1118 rlim_stack.rlim_cur = ssiz;
1119 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1120 }
1121 } else if (sv->sv_maxssiz != NULL) {
1122 ssiz = *sv->sv_maxssiz;
1123 } else {
1124 ssiz = maxssiz;
1125 }
1126 stack_addr = sv->sv_usrstack - ssiz;
1127 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1128 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1129 sv->sv_stackprot,
1130 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1131 if (error)
1132 return (error);
1133
1134 /*
1135 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1136 * are still used to enforce the stack rlimit on the process stack.
1137 */
1138 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1139 vmspace->vm_maxsaddr = (char *)stack_addr;
1140
1141 return (0);
1142 }
1143
1144 /*
1145 * Copy out argument and environment strings from the old process address
1146 * space into the temporary string buffer.
1147 */
1148 int
exec_copyin_args(struct image_args * args,char * fname,enum uio_seg segflg,char ** argv,char ** envv)1149 exec_copyin_args(struct image_args *args, char *fname,
1150 enum uio_seg segflg, char **argv, char **envv)
1151 {
1152 u_long argp, envp;
1153 int error;
1154 size_t length;
1155
1156 bzero(args, sizeof(*args));
1157 if (argv == NULL)
1158 return (EFAULT);
1159
1160 /*
1161 * Allocate demand-paged memory for the file name, argument, and
1162 * environment strings.
1163 */
1164 error = exec_alloc_args(args);
1165 if (error != 0)
1166 return (error);
1167
1168 /*
1169 * Copy the file name.
1170 */
1171 if (fname != NULL) {
1172 args->fname = args->buf;
1173 error = (segflg == UIO_SYSSPACE) ?
1174 copystr(fname, args->fname, PATH_MAX, &length) :
1175 copyinstr(fname, args->fname, PATH_MAX, &length);
1176 if (error != 0)
1177 goto err_exit;
1178 } else
1179 length = 0;
1180
1181 args->begin_argv = args->buf + length;
1182 args->endp = args->begin_argv;
1183 args->stringspace = ARG_MAX;
1184
1185 /*
1186 * extract arguments first
1187 */
1188 for (;;) {
1189 error = fueword(argv++, &argp);
1190 if (error == -1) {
1191 error = EFAULT;
1192 goto err_exit;
1193 }
1194 if (argp == 0)
1195 break;
1196 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1197 args->stringspace, &length);
1198 if (error != 0) {
1199 if (error == ENAMETOOLONG)
1200 error = E2BIG;
1201 goto err_exit;
1202 }
1203 args->stringspace -= length;
1204 args->endp += length;
1205 args->argc++;
1206 }
1207
1208 args->begin_envv = args->endp;
1209
1210 /*
1211 * extract environment strings
1212 */
1213 if (envv) {
1214 for (;;) {
1215 error = fueword(envv++, &envp);
1216 if (error == -1) {
1217 error = EFAULT;
1218 goto err_exit;
1219 }
1220 if (envp == 0)
1221 break;
1222 error = copyinstr((void *)(uintptr_t)envp,
1223 args->endp, args->stringspace, &length);
1224 if (error != 0) {
1225 if (error == ENAMETOOLONG)
1226 error = E2BIG;
1227 goto err_exit;
1228 }
1229 args->stringspace -= length;
1230 args->endp += length;
1231 args->envc++;
1232 }
1233 }
1234
1235 return (0);
1236
1237 err_exit:
1238 exec_free_args(args);
1239 return (error);
1240 }
1241
1242 int
exec_copyin_data_fds(struct thread * td,struct image_args * args,const void * data,size_t datalen,const int * fds,size_t fdslen)1243 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1244 const void *data, size_t datalen, const int *fds, size_t fdslen)
1245 {
1246 struct filedesc *ofdp;
1247 const char *p;
1248 int *kfds;
1249 int error;
1250
1251 memset(args, '\0', sizeof(*args));
1252 ofdp = td->td_proc->p_fd;
1253 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1254 return (E2BIG);
1255 error = exec_alloc_args(args);
1256 if (error != 0)
1257 return (error);
1258
1259 args->begin_argv = args->buf;
1260 args->stringspace = ARG_MAX;
1261
1262 if (datalen > 0) {
1263 /*
1264 * Argument buffer has been provided. Copy it into the
1265 * kernel as a single string and add a terminating null
1266 * byte.
1267 */
1268 error = copyin(data, args->begin_argv, datalen);
1269 if (error != 0)
1270 goto err_exit;
1271 args->begin_argv[datalen] = '\0';
1272 args->endp = args->begin_argv + datalen + 1;
1273 args->stringspace -= datalen + 1;
1274
1275 /*
1276 * Traditional argument counting. Count the number of
1277 * null bytes.
1278 */
1279 for (p = args->begin_argv; p < args->endp; ++p)
1280 if (*p == '\0')
1281 ++args->argc;
1282 } else {
1283 /* No argument buffer provided. */
1284 args->endp = args->begin_argv;
1285 }
1286 /* There are no environment variables. */
1287 args->begin_envv = args->endp;
1288
1289 /* Create new file descriptor table. */
1290 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1291 error = copyin(fds, kfds, fdslen * sizeof(int));
1292 if (error != 0) {
1293 free(kfds, M_TEMP);
1294 goto err_exit;
1295 }
1296 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1297 free(kfds, M_TEMP);
1298 if (error != 0)
1299 goto err_exit;
1300
1301 return (0);
1302 err_exit:
1303 exec_free_args(args);
1304 return (error);
1305 }
1306
1307 /*
1308 * Allocate temporary demand-paged, zero-filled memory for the file name,
1309 * argument, and environment strings. Returns zero if the allocation succeeds
1310 * and ENOMEM otherwise.
1311 */
1312 int
exec_alloc_args(struct image_args * args)1313 exec_alloc_args(struct image_args *args)
1314 {
1315
1316 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1317 return (args->buf != NULL ? 0 : ENOMEM);
1318 }
1319
1320 void
exec_free_args(struct image_args * args)1321 exec_free_args(struct image_args *args)
1322 {
1323
1324 if (args->buf != NULL) {
1325 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1326 PATH_MAX + ARG_MAX);
1327 args->buf = NULL;
1328 }
1329 if (args->fname_buf != NULL) {
1330 free(args->fname_buf, M_TEMP);
1331 args->fname_buf = NULL;
1332 }
1333 if (args->fdp != NULL)
1334 fdescfree_remapped(args->fdp);
1335 }
1336
1337 /*
1338 * Copy strings out to the new process address space, constructing new arg
1339 * and env vector tables. Return a pointer to the base so that it can be used
1340 * as the initial stack pointer.
1341 */
1342 register_t *
exec_copyout_strings(imgp)1343 exec_copyout_strings(imgp)
1344 struct image_params *imgp;
1345 {
1346 int argc, envc;
1347 char **vectp;
1348 char *stringp;
1349 uintptr_t destp;
1350 register_t *stack_base;
1351 struct ps_strings *arginfo;
1352 struct proc *p;
1353 size_t execpath_len;
1354 int szsigcode, szps;
1355 char canary[sizeof(long) * 8];
1356
1357 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1358 /*
1359 * Calculate string base and vector table pointers.
1360 * Also deal with signal trampoline code for this exec type.
1361 */
1362 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1363 execpath_len = strlen(imgp->execpath) + 1;
1364 else
1365 execpath_len = 0;
1366 p = imgp->proc;
1367 szsigcode = 0;
1368 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1369 if (p->p_sysent->sv_sigcode_base == 0) {
1370 if (p->p_sysent->sv_szsigcode != NULL)
1371 szsigcode = *(p->p_sysent->sv_szsigcode);
1372 }
1373 destp = (uintptr_t)arginfo;
1374
1375 /*
1376 * install sigcode
1377 */
1378 if (szsigcode != 0) {
1379 destp -= szsigcode;
1380 destp = rounddown2(destp, sizeof(void *));
1381 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1382 }
1383
1384 /*
1385 * Copy the image path for the rtld.
1386 */
1387 if (execpath_len != 0) {
1388 destp -= execpath_len;
1389 imgp->execpathp = destp;
1390 copyout(imgp->execpath, (void *)destp, execpath_len);
1391 }
1392
1393 /*
1394 * Prepare the canary for SSP.
1395 */
1396 arc4rand(canary, sizeof(canary), 0);
1397 destp -= sizeof(canary);
1398 imgp->canary = destp;
1399 copyout(canary, (void *)destp, sizeof(canary));
1400 imgp->canarylen = sizeof(canary);
1401
1402 /*
1403 * Prepare the pagesizes array.
1404 */
1405 destp -= szps;
1406 destp = rounddown2(destp, sizeof(void *));
1407 imgp->pagesizes = destp;
1408 copyout(pagesizes, (void *)destp, szps);
1409 imgp->pagesizeslen = szps;
1410
1411 destp -= ARG_MAX - imgp->args->stringspace;
1412 destp = rounddown2(destp, sizeof(void *));
1413
1414 /*
1415 * If we have a valid auxargs ptr, prepare some room
1416 * on the stack.
1417 */
1418 if (imgp->auxargs) {
1419 /*
1420 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1421 * lower compatibility.
1422 */
1423 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1424 (AT_COUNT * 2);
1425 /*
1426 * The '+ 2' is for the null pointers at the end of each of
1427 * the arg and env vector sets,and imgp->auxarg_size is room
1428 * for argument of Runtime loader.
1429 */
1430 vectp = (char **)(destp - (imgp->args->argc +
1431 imgp->args->envc + 2 + imgp->auxarg_size)
1432 * sizeof(char *));
1433 } else {
1434 /*
1435 * The '+ 2' is for the null pointers at the end of each of
1436 * the arg and env vector sets
1437 */
1438 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1439 + 2) * sizeof(char *));
1440 }
1441
1442 /*
1443 * vectp also becomes our initial stack base
1444 */
1445 stack_base = (register_t *)vectp;
1446
1447 stringp = imgp->args->begin_argv;
1448 argc = imgp->args->argc;
1449 envc = imgp->args->envc;
1450
1451 /*
1452 * Copy out strings - arguments and environment.
1453 */
1454 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1455
1456 /*
1457 * Fill in "ps_strings" struct for ps, w, etc.
1458 */
1459 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1460 suword32(&arginfo->ps_nargvstr, argc);
1461
1462 /*
1463 * Fill in argument portion of vector table.
1464 */
1465 for (; argc > 0; --argc) {
1466 suword(vectp++, (long)(intptr_t)destp);
1467 while (*stringp++ != 0)
1468 destp++;
1469 destp++;
1470 }
1471
1472 /* a null vector table pointer separates the argp's from the envp's */
1473 suword(vectp++, 0);
1474
1475 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1476 suword32(&arginfo->ps_nenvstr, envc);
1477
1478 /*
1479 * Fill in environment portion of vector table.
1480 */
1481 for (; envc > 0; --envc) {
1482 suword(vectp++, (long)(intptr_t)destp);
1483 while (*stringp++ != 0)
1484 destp++;
1485 destp++;
1486 }
1487
1488 /* end of vector table is a null pointer */
1489 suword(vectp, 0);
1490
1491 return (stack_base);
1492 }
1493
1494 /*
1495 * Check permissions of file to execute.
1496 * Called with imgp->vp locked.
1497 * Return 0 for success or error code on failure.
1498 */
1499 int
exec_check_permissions(imgp)1500 exec_check_permissions(imgp)
1501 struct image_params *imgp;
1502 {
1503 struct vnode *vp = imgp->vp;
1504 struct vattr *attr = imgp->attr;
1505 struct thread *td;
1506 int error, writecount;
1507
1508 td = curthread;
1509
1510 /* Get file attributes */
1511 error = VOP_GETATTR(vp, attr, td->td_ucred);
1512 if (error)
1513 return (error);
1514
1515 #ifdef MAC
1516 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1517 if (error)
1518 return (error);
1519 #endif
1520
1521 /*
1522 * 1) Check if file execution is disabled for the filesystem that
1523 * this file resides on.
1524 * 2) Ensure that at least one execute bit is on. Otherwise, a
1525 * privileged user will always succeed, and we don't want this
1526 * to happen unless the file really is executable.
1527 * 3) Ensure that the file is a regular file.
1528 */
1529 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1530 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1531 (attr->va_type != VREG))
1532 return (EACCES);
1533
1534 /*
1535 * Zero length files can't be exec'd
1536 */
1537 if (attr->va_size == 0)
1538 return (ENOEXEC);
1539
1540 /*
1541 * Check for execute permission to file based on current credentials.
1542 */
1543 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1544 if (error)
1545 return (error);
1546
1547 /*
1548 * Check number of open-for-writes on the file and deny execution
1549 * if there are any.
1550 */
1551 error = VOP_GET_WRITECOUNT(vp, &writecount);
1552 if (error != 0)
1553 return (error);
1554 if (writecount != 0)
1555 return (ETXTBSY);
1556
1557 /*
1558 * Call filesystem specific open routine (which does nothing in the
1559 * general case).
1560 */
1561 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1562 if (error == 0)
1563 imgp->opened = 1;
1564 return (error);
1565 }
1566
1567 /*
1568 * Exec handler registration
1569 */
1570 int
exec_register(execsw_arg)1571 exec_register(execsw_arg)
1572 const struct execsw *execsw_arg;
1573 {
1574 const struct execsw **es, **xs, **newexecsw;
1575 int count = 2; /* New slot and trailing NULL */
1576
1577 if (execsw)
1578 for (es = execsw; *es; es++)
1579 count++;
1580 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1581 if (newexecsw == NULL)
1582 return (ENOMEM);
1583 xs = newexecsw;
1584 if (execsw)
1585 for (es = execsw; *es; es++)
1586 *xs++ = *es;
1587 *xs++ = execsw_arg;
1588 *xs = NULL;
1589 if (execsw)
1590 free(execsw, M_TEMP);
1591 execsw = newexecsw;
1592 return (0);
1593 }
1594
1595 int
exec_unregister(execsw_arg)1596 exec_unregister(execsw_arg)
1597 const struct execsw *execsw_arg;
1598 {
1599 const struct execsw **es, **xs, **newexecsw;
1600 int count = 1;
1601
1602 if (execsw == NULL)
1603 panic("unregister with no handlers left?\n");
1604
1605 for (es = execsw; *es; es++) {
1606 if (*es == execsw_arg)
1607 break;
1608 }
1609 if (*es == NULL)
1610 return (ENOENT);
1611 for (es = execsw; *es; es++)
1612 if (*es != execsw_arg)
1613 count++;
1614 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1615 if (newexecsw == NULL)
1616 return (ENOMEM);
1617 xs = newexecsw;
1618 for (es = execsw; *es; es++)
1619 if (*es != execsw_arg)
1620 *xs++ = *es;
1621 *xs = NULL;
1622 if (execsw)
1623 free(execsw, M_TEMP);
1624 execsw = newexecsw;
1625 return (0);
1626 }
1627