FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_exec.c
1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 1993, David Greenman
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/asan.h>
41 #include <sys/capsicum.h>
42 #include <sys/eventhandler.h>
43 #include <sys/exec.h>
44 #include <sys/fcntl.h>
45 #include <sys/filedesc.h>
46 #include <sys/imgact.h>
47 #include <sys/imgact_elf.h>
48 #include <sys/kernel.h>
49 #include <sys/lock.h>
50 #include <sys/malloc.h>
51 #include <sys/mman.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/ptrace.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/shm.h>
64 #include <sys/signalvar.h>
65 #include <sys/smp.h>
66 #include <sys/stat.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/timers.h>
72 #include <sys/umtx.h>
73 #include <sys/vnode.h>
74 #include <sys/wait.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/pmap.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_pager.h>
88
89 #ifdef HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92
93 #include <machine/reg.h>
94
95 #include <security/audit/audit.h>
96 #include <security/mac/mac_framework.h>
97
98 #ifdef KDTRACE_HOOKS
99 #include <sys/dtrace_bsd.h>
100 dtrace_execexit_func_t dtrace_fasttrap_exec;
101 #endif
102
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
107
108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
109
110 int coredump_pack_fileinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
112 &coredump_pack_fileinfo, 0,
113 "Enable file path packing in 'procstat -f' coredump notes");
114
115 int coredump_pack_vmmapinfo = 1;
116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
117 &coredump_pack_vmmapinfo, 0,
118 "Enable file path packing in 'procstat -v' coredump notes");
119
120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
123 static int do_execve(struct thread *td, struct image_args *args,
124 struct mac *mac_p, struct vmspace *oldvmspace);
125
126 /* XXX This should be vm_size_t. */
127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
128 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
129 "Location of process' ps_strings structure");
130
131 /* XXX This should be vm_size_t. */
132 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
133 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
134 "Top of process stack");
135
136 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
137 NULL, 0, sysctl_kern_stackprot, "I",
138 "Stack memory permissions");
139
140 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
141 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
142 &ps_arg_cache_limit, 0,
143 "Process' command line characters cache limit");
144
145 static int disallow_high_osrel;
146 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
147 &disallow_high_osrel, 0,
148 "Disallow execution of binaries built for higher version of the world");
149
150 static int map_at_zero = 0;
151 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
152 "Permit processes to map an object at virtual address 0.");
153
154 int core_dump_can_intr = 1;
155 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
156 &core_dump_can_intr, 0,
157 "Core dumping interruptible with SIGKILL");
158
159 static int
160 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
161 {
162 struct proc *p;
163 int error;
164
165 p = curproc;
166 #ifdef SCTL_MASK32
167 if (req->flags & SCTL_MASK32) {
168 unsigned int val;
169 val = (unsigned int)p->p_sysent->sv_psstrings;
170 error = SYSCTL_OUT(req, &val, sizeof(val));
171 } else
172 #endif
173 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
174 sizeof(p->p_sysent->sv_psstrings));
175 return error;
176 }
177
178 static int
179 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
180 {
181 struct proc *p;
182 int error;
183
184 p = curproc;
185 #ifdef SCTL_MASK32
186 if (req->flags & SCTL_MASK32) {
187 unsigned int val;
188 val = (unsigned int)p->p_sysent->sv_usrstack;
189 error = SYSCTL_OUT(req, &val, sizeof(val));
190 } else
191 #endif
192 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
193 sizeof(p->p_sysent->sv_usrstack));
194 return error;
195 }
196
197 static int
198 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
199 {
200 struct proc *p;
201
202 p = curproc;
203 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
204 sizeof(p->p_sysent->sv_stackprot)));
205 }
206
207 /*
208 * Each of the items is a pointer to a `const struct execsw', hence the
209 * double pointer here.
210 */
211 static const struct execsw **execsw;
212
213 #ifndef _SYS_SYSPROTO_H_
214 struct execve_args {
215 char *fname;
216 char **argv;
217 char **envv;
218 };
219 #endif
220
221 int
222 sys_execve(struct thread *td, struct execve_args *uap)
223 {
224 struct image_args args;
225 struct vmspace *oldvmspace;
226 int error;
227
228 error = pre_execve(td, &oldvmspace);
229 if (error != 0)
230 return (error);
231 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
232 uap->argv, uap->envv);
233 if (error == 0)
234 error = kern_execve(td, &args, NULL, oldvmspace);
235 post_execve(td, error, oldvmspace);
236 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
237 return (error);
238 }
239
240 #ifndef _SYS_SYSPROTO_H_
241 struct fexecve_args {
242 int fd;
243 char **argv;
244 char **envv;
245 };
246 #endif
247 int
248 sys_fexecve(struct thread *td, struct fexecve_args *uap)
249 {
250 struct image_args args;
251 struct vmspace *oldvmspace;
252 int error;
253
254 error = pre_execve(td, &oldvmspace);
255 if (error != 0)
256 return (error);
257 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
258 uap->argv, uap->envv);
259 if (error == 0) {
260 args.fd = uap->fd;
261 error = kern_execve(td, &args, NULL, oldvmspace);
262 }
263 post_execve(td, error, oldvmspace);
264 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
265 return (error);
266 }
267
268 #ifndef _SYS_SYSPROTO_H_
269 struct __mac_execve_args {
270 char *fname;
271 char **argv;
272 char **envv;
273 struct mac *mac_p;
274 };
275 #endif
276
277 int
278 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
279 {
280 #ifdef MAC
281 struct image_args args;
282 struct vmspace *oldvmspace;
283 int error;
284
285 error = pre_execve(td, &oldvmspace);
286 if (error != 0)
287 return (error);
288 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
289 uap->argv, uap->envv);
290 if (error == 0)
291 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
292 post_execve(td, error, oldvmspace);
293 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
294 return (error);
295 #else
296 return (ENOSYS);
297 #endif
298 }
299
300 int
301 pre_execve(struct thread *td, struct vmspace **oldvmspace)
302 {
303 struct proc *p;
304 int error;
305
306 KASSERT(td == curthread, ("non-current thread %p", td));
307 error = 0;
308 p = td->td_proc;
309 if ((p->p_flag & P_HADTHREADS) != 0) {
310 PROC_LOCK(p);
311 if (thread_single(p, SINGLE_BOUNDARY) != 0)
312 error = ERESTART;
313 PROC_UNLOCK(p);
314 }
315 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
316 ("nested execve"));
317 *oldvmspace = p->p_vmspace;
318 return (error);
319 }
320
321 void
322 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
323 {
324 struct proc *p;
325
326 KASSERT(td == curthread, ("non-current thread %p", td));
327 p = td->td_proc;
328 if ((p->p_flag & P_HADTHREADS) != 0) {
329 PROC_LOCK(p);
330 /*
331 * If success, we upgrade to SINGLE_EXIT state to
332 * force other threads to suicide.
333 */
334 if (error == EJUSTRETURN)
335 thread_single(p, SINGLE_EXIT);
336 else
337 thread_single_end(p, SINGLE_BOUNDARY);
338 PROC_UNLOCK(p);
339 }
340 exec_cleanup(td, oldvmspace);
341 }
342
343 /*
344 * kern_execve() has the astonishing property of not always returning to
345 * the caller. If sufficiently bad things happen during the call to
346 * do_execve(), it can end up calling exit1(); as a result, callers must
347 * avoid doing anything which they might need to undo (e.g., allocating
348 * memory).
349 */
350 int
351 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
352 struct vmspace *oldvmspace)
353 {
354
355 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
356 exec_args_get_begin_envv(args) - args->begin_argv);
357 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
358 args->endp - exec_args_get_begin_envv(args));
359 return (do_execve(td, args, mac_p, oldvmspace));
360 }
361
362 static void
363 execve_nosetid(struct image_params *imgp)
364 {
365 imgp->credential_setid = false;
366 if (imgp->newcred != NULL) {
367 crfree(imgp->newcred);
368 imgp->newcred = NULL;
369 }
370 }
371
372 /*
373 * In-kernel implementation of execve(). All arguments are assumed to be
374 * userspace pointers from the passed thread.
375 */
376 static int
377 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
378 struct vmspace *oldvmspace)
379 {
380 struct proc *p = td->td_proc;
381 struct nameidata nd;
382 struct ucred *oldcred;
383 struct uidinfo *euip = NULL;
384 uintptr_t stack_base;
385 struct image_params image_params, *imgp;
386 struct vattr attr;
387 int (*img_first)(struct image_params *);
388 struct pargs *oldargs = NULL, *newargs = NULL;
389 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
390 #ifdef KTRACE
391 struct ktr_io_params *kiop;
392 #endif
393 struct vnode *oldtextvp, *newtextvp;
394 struct vnode *oldtextdvp, *newtextdvp;
395 char *oldbinname, *newbinname;
396 bool credential_changing;
397 #ifdef MAC
398 struct label *interpvplabel = NULL;
399 bool will_transition;
400 #endif
401 #ifdef HWPMC_HOOKS
402 struct pmckern_procexec pe;
403 #endif
404 int error, i, orig_osrel;
405 uint32_t orig_fctl0;
406 size_t freepath_size;
407 static const char fexecv_proc_title[] = "(fexecv)";
408
409 imgp = &image_params;
410 oldtextvp = oldtextdvp = NULL;
411 newtextvp = newtextdvp = NULL;
412 newbinname = oldbinname = NULL;
413 #ifdef KTRACE
414 kiop = NULL;
415 #endif
416
417 /*
418 * Lock the process and set the P_INEXEC flag to indicate that
419 * it should be left alone until we're done here. This is
420 * necessary to avoid race conditions - e.g. in ptrace() -
421 * that might allow a local user to illicitly obtain elevated
422 * privileges.
423 */
424 PROC_LOCK(p);
425 KASSERT((p->p_flag & P_INEXEC) == 0,
426 ("%s(): process already has P_INEXEC flag", __func__));
427 p->p_flag |= P_INEXEC;
428 PROC_UNLOCK(p);
429
430 /*
431 * Initialize part of the common data
432 */
433 bzero(imgp, sizeof(*imgp));
434 imgp->proc = p;
435 imgp->attr = &attr;
436 imgp->args = args;
437 oldcred = p->p_ucred;
438 orig_osrel = p->p_osrel;
439 orig_fctl0 = p->p_fctl0;
440
441 #ifdef MAC
442 error = mac_execve_enter(imgp, mac_p);
443 if (error)
444 goto exec_fail;
445 #endif
446
447 SDT_PROBE1(proc, , , exec, args->fname);
448
449 interpret:
450 if (args->fname != NULL) {
451 #ifdef CAPABILITY_MODE
452 /*
453 * While capability mode can't reach this point via direct
454 * path arguments to execve(), we also don't allow
455 * interpreters to be used in capability mode (for now).
456 * Catch indirect lookups and return a permissions error.
457 */
458 if (IN_CAPABILITY_MODE(td)) {
459 error = ECAPMODE;
460 goto exec_fail;
461 }
462 #endif
463
464 /*
465 * Translate the file name. namei() returns a vnode
466 * pointer in ni_vp among other things.
467 */
468 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
469 SAVENAME | AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
470 args->fname, td);
471
472 error = namei(&nd);
473 if (error)
474 goto exec_fail;
475
476 newtextvp = nd.ni_vp;
477 newtextdvp = nd.ni_dvp;
478 nd.ni_dvp = NULL;
479 newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
480 M_WAITOK);
481 memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
482 newbinname[nd.ni_cnd.cn_namelen] = '\0';
483 imgp->vp = newtextvp;
484
485 /*
486 * Do the best to calculate the full path to the image file.
487 */
488 if (args->fname[0] == '/') {
489 imgp->execpath = args->fname;
490 } else {
491 VOP_UNLOCK(imgp->vp);
492 freepath_size = MAXPATHLEN;
493 if (vn_fullpath_hardlink(newtextvp, newtextdvp,
494 newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
495 &imgp->freepath, &freepath_size) != 0)
496 imgp->execpath = args->fname;
497 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
498 }
499 } else {
500 AUDIT_ARG_FD(args->fd);
501
502 /*
503 * If the descriptors was not opened with O_PATH, then
504 * we require that it was opened with O_EXEC or
505 * O_RDONLY. In either case, exec_check_permissions()
506 * below checks _current_ file access mode regardless
507 * of the permissions additionally checked at the
508 * open(2).
509 */
510 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
511 &newtextvp);
512 if (error != 0)
513 goto exec_fail;
514
515 if (vn_fullpath(newtextvp, &imgp->execpath,
516 &imgp->freepath) != 0)
517 imgp->execpath = args->fname;
518 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
519 AUDIT_ARG_VNODE1(newtextvp);
520 imgp->vp = newtextvp;
521 }
522
523 /*
524 * Check file permissions. Also 'opens' file and sets its vnode to
525 * text mode.
526 */
527 error = exec_check_permissions(imgp);
528 if (error)
529 goto exec_fail_dealloc;
530
531 imgp->object = imgp->vp->v_object;
532 if (imgp->object != NULL)
533 vm_object_reference(imgp->object);
534
535 error = exec_map_first_page(imgp);
536 if (error)
537 goto exec_fail_dealloc;
538
539 imgp->proc->p_osrel = 0;
540 imgp->proc->p_fctl0 = 0;
541
542 /*
543 * Implement image setuid/setgid.
544 *
545 * Determine new credentials before attempting image activators
546 * so that it can be used by process_exec handlers to determine
547 * credential/setid changes.
548 *
549 * Don't honor setuid/setgid if the filesystem prohibits it or if
550 * the process is being traced.
551 *
552 * We disable setuid/setgid/etc in capability mode on the basis
553 * that most setugid applications are not written with that
554 * environment in mind, and will therefore almost certainly operate
555 * incorrectly. In principle there's no reason that setugid
556 * applications might not be useful in capability mode, so we may want
557 * to reconsider this conservative design choice in the future.
558 *
559 * XXXMAC: For the time being, use NOSUID to also prohibit
560 * transitions on the file system.
561 */
562 credential_changing = false;
563 credential_changing |= (attr.va_mode & S_ISUID) &&
564 oldcred->cr_uid != attr.va_uid;
565 credential_changing |= (attr.va_mode & S_ISGID) &&
566 oldcred->cr_gid != attr.va_gid;
567 #ifdef MAC
568 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
569 interpvplabel, imgp) != 0;
570 credential_changing |= will_transition;
571 #endif
572
573 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
574 if (credential_changing)
575 imgp->proc->p_pdeathsig = 0;
576
577 if (credential_changing &&
578 #ifdef CAPABILITY_MODE
579 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
580 #endif
581 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
582 (p->p_flag & P_TRACED) == 0) {
583 imgp->credential_setid = true;
584 VOP_UNLOCK(imgp->vp);
585 imgp->newcred = crdup(oldcred);
586 if (attr.va_mode & S_ISUID) {
587 euip = uifind(attr.va_uid);
588 change_euid(imgp->newcred, euip);
589 }
590 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
591 if (attr.va_mode & S_ISGID)
592 change_egid(imgp->newcred, attr.va_gid);
593 /*
594 * Implement correct POSIX saved-id behavior.
595 *
596 * XXXMAC: Note that the current logic will save the
597 * uid and gid if a MAC domain transition occurs, even
598 * though maybe it shouldn't.
599 */
600 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
601 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
602 } else {
603 /*
604 * Implement correct POSIX saved-id behavior.
605 *
606 * XXX: It's not clear that the existing behavior is
607 * POSIX-compliant. A number of sources indicate that the
608 * saved uid/gid should only be updated if the new ruid is
609 * not equal to the old ruid, or the new euid is not equal
610 * to the old euid and the new euid is not equal to the old
611 * ruid. The FreeBSD code always updates the saved uid/gid.
612 * Also, this code uses the new (replaced) euid and egid as
613 * the source, which may or may not be the right ones to use.
614 */
615 if (oldcred->cr_svuid != oldcred->cr_uid ||
616 oldcred->cr_svgid != oldcred->cr_gid) {
617 VOP_UNLOCK(imgp->vp);
618 imgp->newcred = crdup(oldcred);
619 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
620 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
621 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
622 }
623 }
624 /* The new credentials are installed into the process later. */
625
626 /*
627 * If the current process has a special image activator it
628 * wants to try first, call it. For example, emulating shell
629 * scripts differently.
630 */
631 error = -1;
632 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
633 error = img_first(imgp);
634
635 /*
636 * Loop through the list of image activators, calling each one.
637 * An activator returns -1 if there is no match, 0 on success,
638 * and an error otherwise.
639 */
640 for (i = 0; error == -1 && execsw[i]; ++i) {
641 if (execsw[i]->ex_imgact == NULL ||
642 execsw[i]->ex_imgact == img_first) {
643 continue;
644 }
645 error = (*execsw[i]->ex_imgact)(imgp);
646 }
647
648 if (error) {
649 if (error == -1)
650 error = ENOEXEC;
651 goto exec_fail_dealloc;
652 }
653
654 /*
655 * Special interpreter operation, cleanup and loop up to try to
656 * activate the interpreter.
657 */
658 if (imgp->interpreted) {
659 exec_unmap_first_page(imgp);
660 /*
661 * The text reference needs to be removed for scripts.
662 * There is a short period before we determine that
663 * something is a script where text reference is active.
664 * The vnode lock is held over this entire period
665 * so nothing should illegitimately be blocked.
666 */
667 MPASS(imgp->textset);
668 VOP_UNSET_TEXT_CHECKED(newtextvp);
669 imgp->textset = false;
670 /* free name buffer and old vnode */
671 #ifdef MAC
672 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
673 #endif
674 if (imgp->opened) {
675 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
676 imgp->opened = false;
677 }
678 vput(newtextvp);
679 imgp->vp = newtextvp = NULL;
680 if (args->fname != NULL) {
681 if (newtextdvp != NULL) {
682 vrele(newtextdvp);
683 newtextdvp = NULL;
684 }
685 NDFREE(&nd, NDF_ONLY_PNBUF);
686 free(newbinname, M_PARGS);
687 newbinname = NULL;
688 }
689 vm_object_deallocate(imgp->object);
690 imgp->object = NULL;
691 execve_nosetid(imgp);
692 imgp->execpath = NULL;
693 free(imgp->freepath, M_TEMP);
694 imgp->freepath = NULL;
695 /* set new name to that of the interpreter */
696 args->fname = imgp->interpreter_name;
697 goto interpret;
698 }
699
700 /*
701 * NB: We unlock the vnode here because it is believed that none
702 * of the sv_copyout_strings/sv_fixup operations require the vnode.
703 */
704 VOP_UNLOCK(imgp->vp);
705
706 if (disallow_high_osrel &&
707 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
708 error = ENOEXEC;
709 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
710 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
711 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
712 goto exec_fail_dealloc;
713 }
714
715 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
716 if (SV_PROC_FLAG(p, SV_CAPSICUM))
717 sys_cap_enter(td, NULL);
718
719 /*
720 * Copy out strings (args and env) and initialize stack base.
721 */
722 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
723 if (error != 0) {
724 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
725 goto exec_fail_dealloc;
726 }
727
728 /*
729 * Stack setup.
730 */
731 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
732 if (error != 0) {
733 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
734 goto exec_fail_dealloc;
735 }
736
737 if (args->fdp != NULL) {
738 /* Install a brand new file descriptor table. */
739 fdinstall_remapped(td, args->fdp);
740 args->fdp = NULL;
741 } else {
742 /*
743 * Keep on using the existing file descriptor table. For
744 * security and other reasons, the file descriptor table
745 * cannot be shared after an exec.
746 */
747 fdunshare(td);
748 pdunshare(td);
749 /* close files on exec */
750 fdcloseexec(td);
751 }
752
753 /*
754 * Malloc things before we need locks.
755 */
756 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
757 /* Cache arguments if they fit inside our allowance */
758 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
759 newargs = pargs_alloc(i);
760 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
761 }
762
763 /*
764 * For security and other reasons, signal handlers cannot
765 * be shared after an exec. The new process gets a copy of the old
766 * handlers. In execsigs(), the new process will have its signals
767 * reset.
768 */
769 if (sigacts_shared(p->p_sigacts)) {
770 oldsigacts = p->p_sigacts;
771 newsigacts = sigacts_alloc();
772 sigacts_copy(newsigacts, oldsigacts);
773 }
774
775 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
776
777 PROC_LOCK(p);
778 if (oldsigacts)
779 p->p_sigacts = newsigacts;
780 /* Stop profiling */
781 stopprofclock(p);
782
783 /* reset caught signals */
784 execsigs(p);
785
786 /* name this process - nameiexec(p, ndp) */
787 bzero(p->p_comm, sizeof(p->p_comm));
788 if (args->fname)
789 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
790 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
791 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
792 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
793 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
794 #ifdef KTR
795 sched_clear_tdname(td);
796 #endif
797
798 /*
799 * mark as execed, wakeup the process that vforked (if any) and tell
800 * it that it now has its own resources back
801 */
802 p->p_flag |= P_EXEC;
803 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
804 p->p_flag2 &= ~P2_NOTRACE;
805 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
806 p->p_flag2 &= ~P2_STKGAP_DISABLE;
807 if (p->p_flag & P_PPWAIT) {
808 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
809 cv_broadcast(&p->p_pwait);
810 /* STOPs are no longer ignored, arrange for AST */
811 signotify(td);
812 }
813
814 if ((imgp->sysent->sv_setid_allowed != NULL &&
815 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
816 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
817 execve_nosetid(imgp);
818
819 /*
820 * Implement image setuid/setgid installation.
821 */
822 if (imgp->credential_setid) {
823 /*
824 * Turn off syscall tracing for set-id programs, except for
825 * root. Record any set-id flags first to make sure that
826 * we do not regain any tracing during a possible block.
827 */
828 setsugid(p);
829 #ifdef KTRACE
830 kiop = ktrprocexec(p);
831 #endif
832 /*
833 * Close any file descriptors 0..2 that reference procfs,
834 * then make sure file descriptors 0..2 are in use.
835 *
836 * Both fdsetugidsafety() and fdcheckstd() may call functions
837 * taking sleepable locks, so temporarily drop our locks.
838 */
839 PROC_UNLOCK(p);
840 VOP_UNLOCK(imgp->vp);
841 fdsetugidsafety(td);
842 error = fdcheckstd(td);
843 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
844 if (error != 0)
845 goto exec_fail_dealloc;
846 PROC_LOCK(p);
847 #ifdef MAC
848 if (will_transition) {
849 mac_vnode_execve_transition(oldcred, imgp->newcred,
850 imgp->vp, interpvplabel, imgp);
851 }
852 #endif
853 } else {
854 if (oldcred->cr_uid == oldcred->cr_ruid &&
855 oldcred->cr_gid == oldcred->cr_rgid)
856 p->p_flag &= ~P_SUGID;
857 }
858 /*
859 * Set the new credentials.
860 */
861 if (imgp->newcred != NULL) {
862 proc_set_cred(p, imgp->newcred);
863 crfree(oldcred);
864 oldcred = NULL;
865 }
866
867 /*
868 * Store the vp for use in kern.proc.pathname. This vnode was
869 * referenced by namei() or by fexecve variant of fname handling.
870 */
871 oldtextvp = p->p_textvp;
872 p->p_textvp = newtextvp;
873 oldtextdvp = p->p_textdvp;
874 p->p_textdvp = newtextdvp;
875 newtextdvp = NULL;
876 oldbinname = p->p_binname;
877 p->p_binname = newbinname;
878 newbinname = NULL;
879
880 #ifdef KDTRACE_HOOKS
881 /*
882 * Tell the DTrace fasttrap provider about the exec if it
883 * has declared an interest.
884 */
885 if (dtrace_fasttrap_exec)
886 dtrace_fasttrap_exec(p);
887 #endif
888
889 /*
890 * Notify others that we exec'd, and clear the P_INEXEC flag
891 * as we're now a bona fide freshly-execed process.
892 */
893 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
894 p->p_flag &= ~P_INEXEC;
895
896 /* clear "fork but no exec" flag, as we _are_ execing */
897 p->p_acflag &= ~AFORK;
898
899 /*
900 * Free any previous argument cache and replace it with
901 * the new argument cache, if any.
902 */
903 oldargs = p->p_args;
904 p->p_args = newargs;
905 newargs = NULL;
906
907 PROC_UNLOCK(p);
908
909 #ifdef HWPMC_HOOKS
910 /*
911 * Check if system-wide sampling is in effect or if the
912 * current process is using PMCs. If so, do exec() time
913 * processing. This processing needs to happen AFTER the
914 * P_INEXEC flag is cleared.
915 */
916 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
917 VOP_UNLOCK(imgp->vp);
918 pe.pm_credentialschanged = credential_changing;
919 pe.pm_entryaddr = imgp->entry_addr;
920
921 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
922 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
923 }
924 #endif
925
926 /* Set values passed into the program in registers. */
927 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
928
929 VOP_MMAPPED(imgp->vp);
930
931 SDT_PROBE1(proc, , , exec__success, args->fname);
932
933 exec_fail_dealloc:
934 if (error != 0) {
935 p->p_osrel = orig_osrel;
936 p->p_fctl0 = orig_fctl0;
937 }
938
939 if (imgp->firstpage != NULL)
940 exec_unmap_first_page(imgp);
941
942 if (imgp->vp != NULL) {
943 if (imgp->opened)
944 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
945 if (imgp->textset)
946 VOP_UNSET_TEXT_CHECKED(imgp->vp);
947 if (error != 0)
948 vput(imgp->vp);
949 else
950 VOP_UNLOCK(imgp->vp);
951 if (args->fname != NULL)
952 NDFREE(&nd, NDF_ONLY_PNBUF);
953 if (newtextdvp != NULL)
954 vrele(newtextdvp);
955 free(newbinname, M_PARGS);
956 }
957
958 if (imgp->object != NULL)
959 vm_object_deallocate(imgp->object);
960
961 free(imgp->freepath, M_TEMP);
962
963 if (error == 0) {
964 if (p->p_ptevents & PTRACE_EXEC) {
965 PROC_LOCK(p);
966 if (p->p_ptevents & PTRACE_EXEC)
967 td->td_dbgflags |= TDB_EXEC;
968 PROC_UNLOCK(p);
969 }
970 } else {
971 exec_fail:
972 /* we're done here, clear P_INEXEC */
973 PROC_LOCK(p);
974 p->p_flag &= ~P_INEXEC;
975 PROC_UNLOCK(p);
976
977 SDT_PROBE1(proc, , , exec__failure, error);
978 }
979
980 if (imgp->newcred != NULL && oldcred != NULL)
981 crfree(imgp->newcred);
982
983 #ifdef MAC
984 mac_execve_exit(imgp);
985 mac_execve_interpreter_exit(interpvplabel);
986 #endif
987 exec_free_args(args);
988
989 /*
990 * Handle deferred decrement of ref counts.
991 */
992 if (oldtextvp != NULL)
993 vrele(oldtextvp);
994 if (oldtextdvp != NULL)
995 vrele(oldtextdvp);
996 free(oldbinname, M_PARGS);
997 #ifdef KTRACE
998 ktr_io_params_free(kiop);
999 #endif
1000 pargs_drop(oldargs);
1001 pargs_drop(newargs);
1002 if (oldsigacts != NULL)
1003 sigacts_free(oldsigacts);
1004 if (euip != NULL)
1005 uifree(euip);
1006
1007 if (error && imgp->vmspace_destroyed) {
1008 /* sorry, no more process anymore. exit gracefully */
1009 exec_cleanup(td, oldvmspace);
1010 exit1(td, 0, SIGABRT);
1011 /* NOT REACHED */
1012 }
1013
1014 #ifdef KTRACE
1015 if (error == 0)
1016 ktrprocctor(p);
1017 #endif
1018
1019 /*
1020 * We don't want cpu_set_syscall_retval() to overwrite any of
1021 * the register values put in place by exec_setregs().
1022 * Implementations of cpu_set_syscall_retval() will leave
1023 * registers unmodified when returning EJUSTRETURN.
1024 */
1025 return (error == 0 ? EJUSTRETURN : error);
1026 }
1027
1028 void
1029 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1030 {
1031 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1032 KASSERT(td->td_proc->p_vmspace != oldvmspace,
1033 ("oldvmspace still used"));
1034 vmspace_free(oldvmspace);
1035 td->td_pflags &= ~TDP_EXECVMSPC;
1036 }
1037 }
1038
1039 int
1040 exec_map_first_page(struct image_params *imgp)
1041 {
1042 vm_object_t object;
1043 vm_page_t m;
1044 int error;
1045
1046 if (imgp->firstpage != NULL)
1047 exec_unmap_first_page(imgp);
1048
1049 object = imgp->vp->v_object;
1050 if (object == NULL)
1051 return (EACCES);
1052 #if VM_NRESERVLEVEL > 0
1053 if ((object->flags & OBJ_COLORED) == 0) {
1054 VM_OBJECT_WLOCK(object);
1055 vm_object_color(object, 0);
1056 VM_OBJECT_WUNLOCK(object);
1057 }
1058 #endif
1059 error = vm_page_grab_valid_unlocked(&m, object, 0,
1060 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1061 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1062
1063 if (error != VM_PAGER_OK)
1064 return (EIO);
1065 imgp->firstpage = sf_buf_alloc(m, 0);
1066 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1067
1068 return (0);
1069 }
1070
1071 void
1072 exec_unmap_first_page(struct image_params *imgp)
1073 {
1074 vm_page_t m;
1075
1076 if (imgp->firstpage != NULL) {
1077 m = sf_buf_page(imgp->firstpage);
1078 sf_buf_free(imgp->firstpage);
1079 imgp->firstpage = NULL;
1080 vm_page_unwire(m, PQ_ACTIVE);
1081 }
1082 }
1083
1084 void
1085 exec_onexec_old(struct thread *td)
1086 {
1087 sigfastblock_clear(td);
1088 umtx_exec(td->td_proc);
1089 }
1090
1091 /*
1092 * This is an optimization which removes the unmanaged shared page
1093 * mapping. In combination with pmap_remove_pages(), which cleans all
1094 * managed mappings in the process' vmspace pmap, no work will be left
1095 * for pmap_remove(min, max).
1096 */
1097 void
1098 exec_free_abi_mappings(struct proc *p)
1099 {
1100 struct vmspace *vmspace;
1101 struct sysentvec *sv;
1102
1103 vmspace = p->p_vmspace;
1104 if (refcount_load(&vmspace->vm_refcnt) != 1)
1105 return;
1106
1107 sv = p->p_sysent;
1108 if (sv->sv_shared_page_obj == NULL)
1109 return;
1110
1111 pmap_remove(vmspace_pmap(vmspace), sv->sv_shared_page_base,
1112 sv->sv_shared_page_base + sv->sv_shared_page_len);
1113 }
1114
1115 /*
1116 * Destroy old address space, and allocate a new stack.
1117 * The new stack is only sgrowsiz large because it is grown
1118 * automatically on a page fault.
1119 */
1120 int
1121 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1122 {
1123 int error;
1124 struct proc *p = imgp->proc;
1125 struct vmspace *vmspace = p->p_vmspace;
1126 struct thread *td = curthread;
1127 vm_object_t obj;
1128 struct rlimit rlim_stack;
1129 vm_offset_t sv_minuser, stack_addr;
1130 vm_map_t map;
1131 vm_prot_t stack_prot;
1132 u_long ssiz;
1133
1134 imgp->vmspace_destroyed = true;
1135 imgp->sysent = sv;
1136
1137 if (p->p_sysent->sv_onexec_old != NULL)
1138 p->p_sysent->sv_onexec_old(td);
1139 itimers_exec(p);
1140 if (sv->sv_onexec != NULL)
1141 sv->sv_onexec(p, imgp);
1142
1143 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1144
1145 /*
1146 * Blow away entire process VM, if address space not shared,
1147 * otherwise, create a new VM space so that other threads are
1148 * not disrupted
1149 */
1150 map = &vmspace->vm_map;
1151 if (map_at_zero)
1152 sv_minuser = sv->sv_minuser;
1153 else
1154 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1155 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1156 vm_map_min(map) == sv_minuser &&
1157 vm_map_max(map) == sv->sv_maxuser &&
1158 cpu_exec_vmspace_reuse(p, map)) {
1159 exec_free_abi_mappings(p);
1160 shmexit(vmspace);
1161 pmap_remove_pages(vmspace_pmap(vmspace));
1162 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1163 /*
1164 * An exec terminates mlockall(MCL_FUTURE).
1165 * ASLR and W^X states must be re-evaluated.
1166 */
1167 vm_map_lock(map);
1168 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1169 MAP_ASLR_IGNSTART | MAP_WXORX);
1170 vm_map_unlock(map);
1171 } else {
1172 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1173 if (error)
1174 return (error);
1175 vmspace = p->p_vmspace;
1176 map = &vmspace->vm_map;
1177 }
1178 map->flags |= imgp->map_flags;
1179
1180 /* Map a shared page */
1181 obj = sv->sv_shared_page_obj;
1182 if (obj != NULL) {
1183 vm_object_reference(obj);
1184 error = vm_map_fixed(map, obj, 0,
1185 sv->sv_shared_page_base, sv->sv_shared_page_len,
1186 VM_PROT_READ | VM_PROT_EXECUTE,
1187 VM_PROT_READ | VM_PROT_EXECUTE,
1188 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1189 if (error != KERN_SUCCESS) {
1190 vm_object_deallocate(obj);
1191 return (vm_mmap_to_errno(error));
1192 }
1193 }
1194
1195 /* Allocate a new stack */
1196 if (imgp->stack_sz != 0) {
1197 ssiz = trunc_page(imgp->stack_sz);
1198 PROC_LOCK(p);
1199 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1200 PROC_UNLOCK(p);
1201 if (ssiz > rlim_stack.rlim_max)
1202 ssiz = rlim_stack.rlim_max;
1203 if (ssiz > rlim_stack.rlim_cur) {
1204 rlim_stack.rlim_cur = ssiz;
1205 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1206 }
1207 } else if (sv->sv_maxssiz != NULL) {
1208 ssiz = *sv->sv_maxssiz;
1209 } else {
1210 ssiz = maxssiz;
1211 }
1212 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1213 if (ssiz < imgp->eff_stack_sz)
1214 imgp->eff_stack_sz = ssiz;
1215 stack_addr = sv->sv_usrstack - ssiz;
1216 stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1217 imgp->stack_prot : sv->sv_stackprot;
1218 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1219 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1220 if (error != KERN_SUCCESS) {
1221 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1222 "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1223 stack_prot, error, vm_mmap_to_errno(error));
1224 return (vm_mmap_to_errno(error));
1225 }
1226
1227 /*
1228 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1229 * are still used to enforce the stack rlimit on the process stack.
1230 */
1231 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1232 vmspace->vm_maxsaddr = (char *)stack_addr;
1233
1234 return (0);
1235 }
1236
1237 /*
1238 * Copy out argument and environment strings from the old process address
1239 * space into the temporary string buffer.
1240 */
1241 int
1242 exec_copyin_args(struct image_args *args, const char *fname,
1243 enum uio_seg segflg, char **argv, char **envv)
1244 {
1245 u_long arg, env;
1246 int error;
1247
1248 bzero(args, sizeof(*args));
1249 if (argv == NULL)
1250 return (EFAULT);
1251
1252 /*
1253 * Allocate demand-paged memory for the file name, argument, and
1254 * environment strings.
1255 */
1256 error = exec_alloc_args(args);
1257 if (error != 0)
1258 return (error);
1259
1260 /*
1261 * Copy the file name.
1262 */
1263 error = exec_args_add_fname(args, fname, segflg);
1264 if (error != 0)
1265 goto err_exit;
1266
1267 /*
1268 * extract arguments first
1269 */
1270 for (;;) {
1271 error = fueword(argv++, &arg);
1272 if (error == -1) {
1273 error = EFAULT;
1274 goto err_exit;
1275 }
1276 if (arg == 0)
1277 break;
1278 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1279 UIO_USERSPACE);
1280 if (error != 0)
1281 goto err_exit;
1282 }
1283
1284 /*
1285 * extract environment strings
1286 */
1287 if (envv) {
1288 for (;;) {
1289 error = fueword(envv++, &env);
1290 if (error == -1) {
1291 error = EFAULT;
1292 goto err_exit;
1293 }
1294 if (env == 0)
1295 break;
1296 error = exec_args_add_env(args,
1297 (char *)(uintptr_t)env, UIO_USERSPACE);
1298 if (error != 0)
1299 goto err_exit;
1300 }
1301 }
1302
1303 return (0);
1304
1305 err_exit:
1306 exec_free_args(args);
1307 return (error);
1308 }
1309
1310 int
1311 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1312 const void *data, size_t datalen, const int *fds, size_t fdslen)
1313 {
1314 struct filedesc *ofdp;
1315 const char *p;
1316 int *kfds;
1317 int error;
1318
1319 memset(args, '\0', sizeof(*args));
1320 ofdp = td->td_proc->p_fd;
1321 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1322 return (E2BIG);
1323 error = exec_alloc_args(args);
1324 if (error != 0)
1325 return (error);
1326
1327 args->begin_argv = args->buf;
1328 args->stringspace = ARG_MAX;
1329
1330 if (datalen > 0) {
1331 /*
1332 * Argument buffer has been provided. Copy it into the
1333 * kernel as a single string and add a terminating null
1334 * byte.
1335 */
1336 error = copyin(data, args->begin_argv, datalen);
1337 if (error != 0)
1338 goto err_exit;
1339 args->begin_argv[datalen] = '\0';
1340 args->endp = args->begin_argv + datalen + 1;
1341 args->stringspace -= datalen + 1;
1342
1343 /*
1344 * Traditional argument counting. Count the number of
1345 * null bytes.
1346 */
1347 for (p = args->begin_argv; p < args->endp; ++p)
1348 if (*p == '\0')
1349 ++args->argc;
1350 } else {
1351 /* No argument buffer provided. */
1352 args->endp = args->begin_argv;
1353 }
1354
1355 /* Create new file descriptor table. */
1356 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1357 error = copyin(fds, kfds, fdslen * sizeof(int));
1358 if (error != 0) {
1359 free(kfds, M_TEMP);
1360 goto err_exit;
1361 }
1362 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1363 free(kfds, M_TEMP);
1364 if (error != 0)
1365 goto err_exit;
1366
1367 return (0);
1368 err_exit:
1369 exec_free_args(args);
1370 return (error);
1371 }
1372
1373 struct exec_args_kva {
1374 vm_offset_t addr;
1375 u_int gen;
1376 SLIST_ENTRY(exec_args_kva) next;
1377 };
1378
1379 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1380
1381 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1382 static struct mtx exec_args_kva_mtx;
1383 static u_int exec_args_gen;
1384
1385 static void
1386 exec_prealloc_args_kva(void *arg __unused)
1387 {
1388 struct exec_args_kva *argkva;
1389 u_int i;
1390
1391 SLIST_INIT(&exec_args_kva_freelist);
1392 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1393 for (i = 0; i < exec_map_entries; i++) {
1394 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1395 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1396 argkva->gen = exec_args_gen;
1397 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1398 }
1399 }
1400 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1401
1402 static vm_offset_t
1403 exec_alloc_args_kva(void **cookie)
1404 {
1405 struct exec_args_kva *argkva;
1406
1407 argkva = (void *)atomic_readandclear_ptr(
1408 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1409 if (argkva == NULL) {
1410 mtx_lock(&exec_args_kva_mtx);
1411 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1412 (void)mtx_sleep(&exec_args_kva_freelist,
1413 &exec_args_kva_mtx, 0, "execkva", 0);
1414 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1415 mtx_unlock(&exec_args_kva_mtx);
1416 }
1417 kasan_mark((void *)argkva->addr, exec_map_entry_size,
1418 exec_map_entry_size, 0);
1419 *(struct exec_args_kva **)cookie = argkva;
1420 return (argkva->addr);
1421 }
1422
1423 static void
1424 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1425 {
1426 vm_offset_t base;
1427
1428 base = argkva->addr;
1429 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1430 KASAN_EXEC_ARGS_FREED);
1431 if (argkva->gen != gen) {
1432 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1433 MADV_FREE);
1434 argkva->gen = gen;
1435 }
1436 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1437 (uintptr_t)NULL, (uintptr_t)argkva)) {
1438 mtx_lock(&exec_args_kva_mtx);
1439 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1440 wakeup_one(&exec_args_kva_freelist);
1441 mtx_unlock(&exec_args_kva_mtx);
1442 }
1443 }
1444
1445 static void
1446 exec_free_args_kva(void *cookie)
1447 {
1448
1449 exec_release_args_kva(cookie, exec_args_gen);
1450 }
1451
1452 static void
1453 exec_args_kva_lowmem(void *arg __unused)
1454 {
1455 SLIST_HEAD(, exec_args_kva) head;
1456 struct exec_args_kva *argkva;
1457 u_int gen;
1458 int i;
1459
1460 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1461
1462 /*
1463 * Force an madvise of each KVA range. Any currently allocated ranges
1464 * will have MADV_FREE applied once they are freed.
1465 */
1466 SLIST_INIT(&head);
1467 mtx_lock(&exec_args_kva_mtx);
1468 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1469 mtx_unlock(&exec_args_kva_mtx);
1470 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1471 SLIST_REMOVE_HEAD(&head, next);
1472 exec_release_args_kva(argkva, gen);
1473 }
1474
1475 CPU_FOREACH(i) {
1476 argkva = (void *)atomic_readandclear_ptr(
1477 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1478 if (argkva != NULL)
1479 exec_release_args_kva(argkva, gen);
1480 }
1481 }
1482 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1483 EVENTHANDLER_PRI_ANY);
1484
1485 /*
1486 * Allocate temporary demand-paged, zero-filled memory for the file name,
1487 * argument, and environment strings.
1488 */
1489 int
1490 exec_alloc_args(struct image_args *args)
1491 {
1492
1493 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1494 return (0);
1495 }
1496
1497 void
1498 exec_free_args(struct image_args *args)
1499 {
1500
1501 if (args->buf != NULL) {
1502 exec_free_args_kva(args->bufkva);
1503 args->buf = NULL;
1504 }
1505 if (args->fname_buf != NULL) {
1506 free(args->fname_buf, M_TEMP);
1507 args->fname_buf = NULL;
1508 }
1509 if (args->fdp != NULL)
1510 fdescfree_remapped(args->fdp);
1511 }
1512
1513 /*
1514 * A set to functions to fill struct image args.
1515 *
1516 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1517 * fname) before the other functions. All exec_args_add_arg() calls must
1518 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1519 * may be called any time after exec_args_add_fname().
1520 *
1521 * exec_args_add_fname() - install path to be executed
1522 * exec_args_add_arg() - append an argument string
1523 * exec_args_add_env() - append an env string
1524 * exec_args_adjust_args() - adjust location of the argument list to
1525 * allow new arguments to be prepended
1526 */
1527 int
1528 exec_args_add_fname(struct image_args *args, const char *fname,
1529 enum uio_seg segflg)
1530 {
1531 int error;
1532 size_t length;
1533
1534 KASSERT(args->fname == NULL, ("fname already appended"));
1535 KASSERT(args->endp == NULL, ("already appending to args"));
1536
1537 if (fname != NULL) {
1538 args->fname = args->buf;
1539 error = segflg == UIO_SYSSPACE ?
1540 copystr(fname, args->fname, PATH_MAX, &length) :
1541 copyinstr(fname, args->fname, PATH_MAX, &length);
1542 if (error != 0)
1543 return (error == ENAMETOOLONG ? E2BIG : error);
1544 } else
1545 length = 0;
1546
1547 /* Set up for _arg_*()/_env_*() */
1548 args->endp = args->buf + length;
1549 /* begin_argv must be set and kept updated */
1550 args->begin_argv = args->endp;
1551 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1552 ("too little space remaining for arguments %zu < %zu",
1553 exec_map_entry_size - length, (size_t)ARG_MAX));
1554 args->stringspace = ARG_MAX;
1555
1556 return (0);
1557 }
1558
1559 static int
1560 exec_args_add_str(struct image_args *args, const char *str,
1561 enum uio_seg segflg, int *countp)
1562 {
1563 int error;
1564 size_t length;
1565
1566 KASSERT(args->endp != NULL, ("endp not initialized"));
1567 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1568
1569 error = (segflg == UIO_SYSSPACE) ?
1570 copystr(str, args->endp, args->stringspace, &length) :
1571 copyinstr(str, args->endp, args->stringspace, &length);
1572 if (error != 0)
1573 return (error == ENAMETOOLONG ? E2BIG : error);
1574 args->stringspace -= length;
1575 args->endp += length;
1576 (*countp)++;
1577
1578 return (0);
1579 }
1580
1581 int
1582 exec_args_add_arg(struct image_args *args, const char *argp,
1583 enum uio_seg segflg)
1584 {
1585
1586 KASSERT(args->envc == 0, ("appending args after env"));
1587
1588 return (exec_args_add_str(args, argp, segflg, &args->argc));
1589 }
1590
1591 int
1592 exec_args_add_env(struct image_args *args, const char *envp,
1593 enum uio_seg segflg)
1594 {
1595
1596 if (args->envc == 0)
1597 args->begin_envv = args->endp;
1598
1599 return (exec_args_add_str(args, envp, segflg, &args->envc));
1600 }
1601
1602 int
1603 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1604 {
1605 ssize_t offset;
1606
1607 KASSERT(args->endp != NULL, ("endp not initialized"));
1608 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1609
1610 offset = extend - consume;
1611 if (args->stringspace < offset)
1612 return (E2BIG);
1613 memmove(args->begin_argv + extend, args->begin_argv + consume,
1614 args->endp - args->begin_argv + consume);
1615 if (args->envc > 0)
1616 args->begin_envv += offset;
1617 args->endp += offset;
1618 args->stringspace -= offset;
1619 return (0);
1620 }
1621
1622 char *
1623 exec_args_get_begin_envv(struct image_args *args)
1624 {
1625
1626 KASSERT(args->endp != NULL, ("endp not initialized"));
1627
1628 if (args->envc > 0)
1629 return (args->begin_envv);
1630 return (args->endp);
1631 }
1632
1633 void
1634 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1635 {
1636 if (imgp->sysent->sv_stackgap == NULL ||
1637 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1638 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1639 (imgp->map_flags & MAP_ASLR) == 0)
1640 return;
1641 imgp->sysent->sv_stackgap(imgp, dp);
1642 }
1643
1644 /*
1645 * Copy strings out to the new process address space, constructing new arg
1646 * and env vector tables. Return a pointer to the base so that it can be used
1647 * as the initial stack pointer.
1648 */
1649 int
1650 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1651 {
1652 int argc, envc;
1653 char **vectp;
1654 char *stringp;
1655 uintptr_t destp, ustringp;
1656 struct ps_strings *arginfo;
1657 struct proc *p;
1658 size_t execpath_len;
1659 int error, szsigcode, szps;
1660 char canary[sizeof(long) * 8];
1661
1662 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1663 /*
1664 * Calculate string base and vector table pointers.
1665 * Also deal with signal trampoline code for this exec type.
1666 */
1667 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1668 execpath_len = strlen(imgp->execpath) + 1;
1669 else
1670 execpath_len = 0;
1671 p = imgp->proc;
1672 szsigcode = 0;
1673 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1674 imgp->ps_strings = arginfo;
1675 if (p->p_sysent->sv_sigcode_base == 0) {
1676 if (p->p_sysent->sv_szsigcode != NULL)
1677 szsigcode = *(p->p_sysent->sv_szsigcode);
1678 }
1679 destp = (uintptr_t)arginfo;
1680
1681 /*
1682 * install sigcode
1683 */
1684 if (szsigcode != 0) {
1685 destp -= szsigcode;
1686 destp = rounddown2(destp, sizeof(void *));
1687 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1688 szsigcode);
1689 if (error != 0)
1690 return (error);
1691 }
1692
1693 /*
1694 * Copy the image path for the rtld.
1695 */
1696 if (execpath_len != 0) {
1697 destp -= execpath_len;
1698 destp = rounddown2(destp, sizeof(void *));
1699 imgp->execpathp = (void *)destp;
1700 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1701 if (error != 0)
1702 return (error);
1703 }
1704
1705 /*
1706 * Prepare the canary for SSP.
1707 */
1708 arc4rand(canary, sizeof(canary), 0);
1709 destp -= sizeof(canary);
1710 imgp->canary = (void *)destp;
1711 error = copyout(canary, imgp->canary, sizeof(canary));
1712 if (error != 0)
1713 return (error);
1714 imgp->canarylen = sizeof(canary);
1715
1716 /*
1717 * Prepare the pagesizes array.
1718 */
1719 destp -= szps;
1720 destp = rounddown2(destp, sizeof(void *));
1721 imgp->pagesizes = (void *)destp;
1722 error = copyout(pagesizes, imgp->pagesizes, szps);
1723 if (error != 0)
1724 return (error);
1725 imgp->pagesizeslen = szps;
1726
1727 /*
1728 * Allocate room for the argument and environment strings.
1729 */
1730 destp -= ARG_MAX - imgp->args->stringspace;
1731 destp = rounddown2(destp, sizeof(void *));
1732 ustringp = destp;
1733
1734 exec_stackgap(imgp, &destp);
1735
1736 if (imgp->auxargs) {
1737 /*
1738 * Allocate room on the stack for the ELF auxargs
1739 * array. It has up to AT_COUNT entries.
1740 */
1741 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1742 destp = rounddown2(destp, sizeof(void *));
1743 }
1744
1745 vectp = (char **)destp;
1746
1747 /*
1748 * Allocate room for the argv[] and env vectors including the
1749 * terminating NULL pointers.
1750 */
1751 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1752
1753 /*
1754 * vectp also becomes our initial stack base
1755 */
1756 *stack_base = (uintptr_t)vectp;
1757
1758 stringp = imgp->args->begin_argv;
1759 argc = imgp->args->argc;
1760 envc = imgp->args->envc;
1761
1762 /*
1763 * Copy out strings - arguments and environment.
1764 */
1765 error = copyout(stringp, (void *)ustringp,
1766 ARG_MAX - imgp->args->stringspace);
1767 if (error != 0)
1768 return (error);
1769
1770 /*
1771 * Fill in "ps_strings" struct for ps, w, etc.
1772 */
1773 imgp->argv = vectp;
1774 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1775 suword32(&arginfo->ps_nargvstr, argc) != 0)
1776 return (EFAULT);
1777
1778 /*
1779 * Fill in argument portion of vector table.
1780 */
1781 for (; argc > 0; --argc) {
1782 if (suword(vectp++, ustringp) != 0)
1783 return (EFAULT);
1784 while (*stringp++ != 0)
1785 ustringp++;
1786 ustringp++;
1787 }
1788
1789 /* a null vector table pointer separates the argp's from the envp's */
1790 if (suword(vectp++, 0) != 0)
1791 return (EFAULT);
1792
1793 imgp->envv = vectp;
1794 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1795 suword32(&arginfo->ps_nenvstr, envc) != 0)
1796 return (EFAULT);
1797
1798 /*
1799 * Fill in environment portion of vector table.
1800 */
1801 for (; envc > 0; --envc) {
1802 if (suword(vectp++, ustringp) != 0)
1803 return (EFAULT);
1804 while (*stringp++ != 0)
1805 ustringp++;
1806 ustringp++;
1807 }
1808
1809 /* end of vector table is a null pointer */
1810 if (suword(vectp, 0) != 0)
1811 return (EFAULT);
1812
1813 if (imgp->auxargs) {
1814 vectp++;
1815 error = imgp->sysent->sv_copyout_auxargs(imgp,
1816 (uintptr_t)vectp);
1817 if (error != 0)
1818 return (error);
1819 }
1820
1821 return (0);
1822 }
1823
1824 /*
1825 * Check permissions of file to execute.
1826 * Called with imgp->vp locked.
1827 * Return 0 for success or error code on failure.
1828 */
1829 int
1830 exec_check_permissions(struct image_params *imgp)
1831 {
1832 struct vnode *vp = imgp->vp;
1833 struct vattr *attr = imgp->attr;
1834 struct thread *td;
1835 int error;
1836
1837 td = curthread;
1838
1839 /* Get file attributes */
1840 error = VOP_GETATTR(vp, attr, td->td_ucred);
1841 if (error)
1842 return (error);
1843
1844 #ifdef MAC
1845 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1846 if (error)
1847 return (error);
1848 #endif
1849
1850 /*
1851 * 1) Check if file execution is disabled for the filesystem that
1852 * this file resides on.
1853 * 2) Ensure that at least one execute bit is on. Otherwise, a
1854 * privileged user will always succeed, and we don't want this
1855 * to happen unless the file really is executable.
1856 * 3) Ensure that the file is a regular file.
1857 */
1858 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1859 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1860 (attr->va_type != VREG))
1861 return (EACCES);
1862
1863 /*
1864 * Zero length files can't be exec'd
1865 */
1866 if (attr->va_size == 0)
1867 return (ENOEXEC);
1868
1869 /*
1870 * Check for execute permission to file based on current credentials.
1871 */
1872 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1873 if (error)
1874 return (error);
1875
1876 /*
1877 * Check number of open-for-writes on the file and deny execution
1878 * if there are any.
1879 *
1880 * Add a text reference now so no one can write to the
1881 * executable while we're activating it.
1882 *
1883 * Remember if this was set before and unset it in case this is not
1884 * actually an executable image.
1885 */
1886 error = VOP_SET_TEXT(vp);
1887 if (error != 0)
1888 return (error);
1889 imgp->textset = true;
1890
1891 /*
1892 * Call filesystem specific open routine (which does nothing in the
1893 * general case).
1894 */
1895 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1896 if (error == 0)
1897 imgp->opened = true;
1898 return (error);
1899 }
1900
1901 /*
1902 * Exec handler registration
1903 */
1904 int
1905 exec_register(const struct execsw *execsw_arg)
1906 {
1907 const struct execsw **es, **xs, **newexecsw;
1908 u_int count = 2; /* New slot and trailing NULL */
1909
1910 if (execsw)
1911 for (es = execsw; *es; es++)
1912 count++;
1913 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1914 xs = newexecsw;
1915 if (execsw)
1916 for (es = execsw; *es; es++)
1917 *xs++ = *es;
1918 *xs++ = execsw_arg;
1919 *xs = NULL;
1920 if (execsw)
1921 free(execsw, M_TEMP);
1922 execsw = newexecsw;
1923 return (0);
1924 }
1925
1926 int
1927 exec_unregister(const struct execsw *execsw_arg)
1928 {
1929 const struct execsw **es, **xs, **newexecsw;
1930 int count = 1;
1931
1932 if (execsw == NULL)
1933 panic("unregister with no handlers left?\n");
1934
1935 for (es = execsw; *es; es++) {
1936 if (*es == execsw_arg)
1937 break;
1938 }
1939 if (*es == NULL)
1940 return (ENOENT);
1941 for (es = execsw; *es; es++)
1942 if (*es != execsw_arg)
1943 count++;
1944 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1945 xs = newexecsw;
1946 for (es = execsw; *es; es++)
1947 if (*es != execsw_arg)
1948 *xs++ = *es;
1949 *xs = NULL;
1950 if (execsw)
1951 free(execsw, M_TEMP);
1952 execsw = newexecsw;
1953 return (0);
1954 }
Cache object: 77ed20ffb124e83f2c5cb0534277a48e
|