]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/kern_exec.c
projects / apple / xnu.git / blob
? search:


eb333d4b1a33df93d2b5515180b3f1e60f7155db
[apple/xnu.git] / bsd / kern / kern_exec.c
1 /*
2 * Copyright (c) 2000-2019 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
29 /*
30 * Mach Operating System
31 * Copyright (c) 1987 Carnegie-Mellon University
32 * All rights reserved. The CMU software License Agreement specifies
33 * the terms and conditions for use and redistribution.
34 */
35
36 /*-
37 * Copyright (c) 1982, 1986, 1991, 1993
38 * The Regents of the University of California. All rights reserved.
39 * (c) UNIX System Laboratories, Inc.
40 * All or some portions of this file are derived from material licensed
41 * to the University of California by American Telephone and Telegraph
42 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
43 * the permission of UNIX System Laboratories, Inc.
44 *
45 * Redistribution and use in source and binary forms, with or without
46 * modification, are permitted provided that the following conditions
47 * are met:
48 * 1. Redistributions of source code must retain the above copyright
49 * notice, this list of conditions and the following disclaimer.
50 * 2. Redistributions in binary form must reproduce the above copyright
51 * notice, this list of conditions and the following disclaimer in the
52 * documentation and/or other materials provided with the distribution.
53 * 3. All advertising materials mentioning features or use of this software
54 * must display the following acknowledgement:
55 * This product includes software developed by the University of
56 * California, Berkeley and its contributors.
57 * 4. Neither the name of the University nor the names of its contributors
58 * may be used to endorse or promote products derived from this software
59 * without specific prior written permission.
60 *
61 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
62 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
63 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
64 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
65 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
66 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
67 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
68 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
69 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
70 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
71 * SUCH DAMAGE.
72 *
73 * from: @(#)kern_exec.c 8.1 (Berkeley) 6/10/93
74 */
75 /*
76 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
77 * support for mandatory and extensible security protections. This notice
78 * is included in support of clause 2.2 (b) of the Apple Public License,
79 * Version 2.0.
80 */
81 #include <machine/reg.h>
82 #include <machine/cpu_capabilities.h>
83
84 #include <sys/param.h>
85 #include <sys/systm.h>
86 #include <sys/filedesc.h>
87 #include <sys/kernel.h>
88 #include <sys/proc_internal.h>
89 #include <sys/kauth.h>
90 #include <sys/user.h>
91 #include <sys/socketvar.h>
92 #include <sys/malloc.h>
93 #include <sys/namei.h>
94 #include <sys/mount_internal.h>
95 #include <sys/vnode_internal.h>
96 #include <sys/file_internal.h>
97 #include <sys/stat.h>
98 #include <sys/uio_internal.h>
99 #include <sys/acct.h>
100 #include <sys/exec.h>
101 #include <sys/kdebug.h>
102 #include <sys/signal.h>
103 #include <sys/aio_kern.h>
104 #include <sys/sysproto.h>
105 #include <sys/persona.h>
106 #include <sys/reason.h>
107 #if SYSV_SHM
108 #include <sys/shm_internal.h> /* shmexec() */
109 #endif
110 #include <sys/ubc_internal.h> /* ubc_map() */
111 #include <sys/spawn.h>
112 #include <sys/spawn_internal.h>
113 #include <sys/process_policy.h>
114 #include <sys/codesign.h>
115 #include <sys/random.h>
116 #include <crypto/sha1.h>
117
118 #include <libkern/libkern.h>
119
120 #include <security/audit/audit.h>
121
122 #include <ipc/ipc_types.h>
123
124 #include <mach/mach_param.h>
125 #include <mach/mach_types.h>
126 #include <mach/port.h>
127 #include <mach/task.h>
128 #include <mach/task_access.h>
129 #include <mach/thread_act.h>
130 #include <mach/vm_map.h>
131 #include <mach/mach_vm.h>
132 #include <mach/vm_param.h>
133
134 #include <kern/sched_prim.h> /* thread_wakeup() */
135 #include <kern/affinity.h>
136 #include <kern/assert.h>
137 #include <kern/task.h>
138 #include <kern/coalition.h>
139 #include <kern/policy_internal.h>
140 #include <kern/kalloc.h>
141
142 #include <os/log.h>
143
144 #if CONFIG_MACF
145 #include <security/mac_framework.h>
146 #include <security/mac_mach_internal.h>
147 #endif
148
149 #if CONFIG_AUDIT
150 #include <bsm/audit_kevents.h>
151 #endif
152
153 #if CONFIG_ARCADE
154 #include <kern/arcade.h>
155 #endif
156
157 #include <vm/vm_map.h>
158 #include <vm/vm_kern.h>
159 #include <vm/vm_protos.h>
160 #include <vm/vm_kern.h>
161 #include <vm/vm_fault.h>
162 #include <vm/vm_pageout.h>
163
164 #include <kdp/kdp_dyld.h>
165
166 #include <machine/machine_routines.h>
167 #include <machine/pal_routines.h>
168
169 #include <pexpert/pexpert.h>
170
171 #if CONFIG_MEMORYSTATUS
172 #include <sys/kern_memorystatus.h>
173 #endif
174
175 #include <IOKit/IOBSD.h>
176
177 extern boolean_t vm_darkwake_mode;
178
179 extern int bootarg_execfailurereports; /* bsd_init.c */
180
181 #if CONFIG_DTRACE
182 /* Do not include dtrace.h, it redefines kmem_[alloc/free] */
183 extern void dtrace_proc_exec(proc_t);
184 extern void (*dtrace_proc_waitfor_exec_ptr)(proc_t);
185
186 /*
187 * Since dtrace_proc_waitfor_exec_ptr can be added/removed in dtrace_subr.c,
188 * we will store its value before actually calling it.
189 */
190 static void (*dtrace_proc_waitfor_hook)(proc_t) = NULL;
191
192 #include <sys/dtrace_ptss.h>
193 #endif
194
195 /* support for child creation in exec after vfork */
196 thread_t fork_create_child(task_t parent_task,
197 coalition_t *parent_coalition,
198 proc_t child_proc,
199 int inherit_memory,
200 int is_64bit_addr,
201 int is_64bit_data,
202 int in_exec);
203 void vfork_exit(proc_t p, int rv);
204 extern void proc_apply_task_networkbg_internal(proc_t, thread_t);
205 extern void task_set_did_exec_flag(task_t task);
206 extern void task_clear_exec_copy_flag(task_t task);
207 proc_t proc_exec_switch_task(proc_t p, task_t old_task, task_t new_task, thread_t new_thread);
208 boolean_t task_is_active(task_t);
209 boolean_t thread_is_active(thread_t thread);
210 void thread_copy_resource_info(thread_t dst_thread, thread_t src_thread);
211 void *ipc_importance_exec_switch_task(task_t old_task, task_t new_task);
212 extern void ipc_importance_release(void *elem);
213 extern boolean_t task_has_watchports(task_t task);
214
215 /*
216 * Mach things for which prototypes are unavailable from Mach headers
217 */
218 #define IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND 0x1
219 void ipc_task_reset(
220 task_t task);
221 void ipc_thread_reset(
222 thread_t thread);
223 kern_return_t ipc_object_copyin(
224 ipc_space_t space,
225 mach_port_name_t name,
226 mach_msg_type_name_t msgt_name,
227 ipc_object_t *objectp,
228 mach_port_context_t context,
229 mach_msg_guard_flags_t *guard_flags,
230 uint32_t kmsg_flags);
231 void ipc_port_release_send(ipc_port_t);
232
233 #if DEVELOPMENT || DEBUG
234 void task_importance_update_owner_info(task_t);
235 #endif
236
237 extern struct savearea *get_user_regs(thread_t);
238
239 __attribute__((noinline)) int __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(mach_port_t task_access_port, int32_t new_pid);
240
241 #include <kern/thread.h>
242 #include <kern/task.h>
243 #include <kern/ast.h>
244 #include <kern/mach_loader.h>
245 #include <kern/mach_fat.h>
246 #include <mach-o/fat.h>
247 #include <mach-o/loader.h>
248 #include <machine/vmparam.h>
249 #include <sys/imgact.h>
250
251 #include <sys/sdt.h>
252
253
254 /*
255 * EAI_ITERLIMIT The maximum number of times to iterate an image
256 * activator in exec_activate_image() before treating
257 * it as malformed/corrupt.
258 */
259 #define EAI_ITERLIMIT 3
260
261 /*
262 * For #! interpreter parsing
263 */
264 #define IS_WHITESPACE(ch) ((ch == ' ') || (ch == '\t'))
265 #define IS_EOL(ch) ((ch == '#') || (ch == '\n'))
266
267 extern vm_map_t bsd_pageable_map;
268 extern const struct fileops vnops;
269 extern int nextpidversion;
270
271 #define USER_ADDR_ALIGN(addr, val) \
272 ( ( (user_addr_t)(addr) + (val) - 1) \
273 & ~((val) - 1) )
274
275 /* Platform Code Exec Logging */
276 static int platform_exec_logging = 0;
277
278 SYSCTL_DECL(_security_mac);
279
280 SYSCTL_INT(_security_mac, OID_AUTO, platform_exec_logging, CTLFLAG_RW, &platform_exec_logging, 0,
281 "log cdhashes for all platform binary executions");
282
283 static os_log_t peLog = OS_LOG_DEFAULT;
284
285 struct exec_port_actions {
286 uint32_t portwatch_count;
287 uint32_t registered_count;
288 ipc_port_t *portwatch_array;
289 ipc_port_t *registered_array;
290 };
291
292 struct image_params; /* Forward */
293 static int exec_activate_image(struct image_params *imgp);
294 static int exec_copyout_strings(struct image_params *imgp, user_addr_t *stackp);
295 static int load_return_to_errno(load_return_t lrtn);
296 static int execargs_alloc(struct image_params *imgp);
297 static int execargs_free(struct image_params *imgp);
298 static int exec_check_permissions(struct image_params *imgp);
299 static int exec_extract_strings(struct image_params *imgp);
300 static int exec_add_apple_strings(struct image_params *imgp, const load_result_t *load_result);
301 static int exec_handle_sugid(struct image_params *imgp);
302 static int sugid_scripts = 0;
303 SYSCTL_INT(_kern, OID_AUTO, sugid_scripts, CTLFLAG_RW | CTLFLAG_LOCKED, &sugid_scripts, 0, "");
304 static kern_return_t create_unix_stack(vm_map_t map, load_result_t* load_result, proc_t p);
305 static int copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size);
306 static void exec_resettextvp(proc_t, struct image_params *);
307 static int check_for_signature(proc_t, struct image_params *);
308 static void exec_prefault_data(proc_t, struct image_params *, load_result_t *);
309 static errno_t exec_handle_port_actions(struct image_params *imgp,
310 struct exec_port_actions *port_actions);
311 static errno_t exec_handle_spawnattr_policy(proc_t p, thread_t thread, int psa_apptype, uint64_t psa_qos_clamp,
312 uint64_t psa_darwin_role, struct exec_port_actions *port_actions);
313 static void exec_port_actions_destroy(struct exec_port_actions *port_actions);
314
315 /*
316 * exec_add_user_string
317 *
318 * Add the requested string to the string space area.
319 *
320 * Parameters; struct image_params * image parameter block
321 * user_addr_t string to add to strings area
322 * int segment from which string comes
323 * boolean_t TRUE if string contributes to NCARGS
324 *
325 * Returns: 0 Success
326 * !0 Failure errno from copyinstr()
327 *
328 * Implicit returns:
329 * (imgp->ip_strendp) updated location of next add, if any
330 * (imgp->ip_strspace) updated byte count of space remaining
331 * (imgp->ip_argspace) updated byte count of space in NCARGS
332 */
333 static int
334 exec_add_user_string(struct image_params *imgp, user_addr_t str, int seg, boolean_t is_ncargs)
335 {
336 int error = 0;
337
338 do {
339 size_t len = 0;
340 int space;
341
342 if (is_ncargs) {
343 space = imgp->ip_argspace; /* by definition smaller than ip_strspace */
344 } else {
345 space = imgp->ip_strspace;
346 }
347
348 if (space <= 0) {
349 error = E2BIG;
350 break;
351 }
352
353 if (!UIO_SEG_IS_USER_SPACE(seg)) {
354 char *kstr = CAST_DOWN(char *, str); /* SAFE */
355 error = copystr(kstr, imgp->ip_strendp, space, &len);
356 } else {
357 error = copyinstr(str, imgp->ip_strendp, space, &len);
358 }
359
360 imgp->ip_strendp += len;
361 imgp->ip_strspace -= len;
362 if (is_ncargs) {
363 imgp->ip_argspace -= len;
364 }
365 } while (error == ENAMETOOLONG);
366
367 return error;
368 }
369
370 /*
371 * dyld is now passed the executable path as a getenv-like variable
372 * in the same fashion as the stack_guard and malloc_entropy keys.
373 */
374 #define EXECUTABLE_KEY "executable_path="
375
376 /*
377 * exec_save_path
378 *
379 * To support new app package launching for Mac OS X, the dyld needs the
380 * first argument to execve() stored on the user stack.
381 *
382 * Save the executable path name at the bottom of the strings area and set
383 * the argument vector pointer to the location following that to indicate
384 * the start of the argument and environment tuples, setting the remaining
385 * string space count to the size of the string area minus the path length.
386 *
387 * Parameters; struct image_params * image parameter block
388 * char * path used to invoke program
389 * int segment from which path comes
390 *
391 * Returns: int 0 Success
392 * EFAULT Bad address
393 * copy[in]str:EFAULT Bad address
394 * copy[in]str:ENAMETOOLONG Filename too long
395 *
396 * Implicit returns:
397 * (imgp->ip_strings) saved path
398 * (imgp->ip_strspace) space remaining in ip_strings
399 * (imgp->ip_strendp) start of remaining copy area
400 * (imgp->ip_argspace) space remaining of NCARGS
401 * (imgp->ip_applec) Initial applev[0]
402 *
403 * Note: We have to do this before the initial namei() since in the
404 * path contains symbolic links, namei() will overwrite the
405 * original path buffer contents. If the last symbolic link
406 * resolved was a relative pathname, we would lose the original
407 * "path", which could be an absolute pathname. This might be
408 * unacceptable for dyld.
409 */
410 static int
411 exec_save_path(struct image_params *imgp, user_addr_t path, int seg, const char **excpath)
412 {
413 int error;
414 size_t len;
415 char *kpath;
416
417 // imgp->ip_strings can come out of a cache, so we need to obliterate the
418 // old path.
419 memset(imgp->ip_strings, '\0', strlen(EXECUTABLE_KEY) + MAXPATHLEN);
420
421 len = MIN(MAXPATHLEN, imgp->ip_strspace);
422
423 switch (seg) {
424 case UIO_USERSPACE32:
425 case UIO_USERSPACE64: /* Same for copyin()... */
426 error = copyinstr(path, imgp->ip_strings + strlen(EXECUTABLE_KEY), len, &len);
427 break;
428 case UIO_SYSSPACE:
429 kpath = CAST_DOWN(char *, path); /* SAFE */
430 error = copystr(kpath, imgp->ip_strings + strlen(EXECUTABLE_KEY), len, &len);
431 break;
432 default:
433 error = EFAULT;
434 break;
435 }
436
437 if (!error) {
438 bcopy(EXECUTABLE_KEY, imgp->ip_strings, strlen(EXECUTABLE_KEY));
439 len += strlen(EXECUTABLE_KEY);
440
441 imgp->ip_strendp += len;
442 imgp->ip_strspace -= len;
443
444 if (excpath) {
445 *excpath = imgp->ip_strings + strlen(EXECUTABLE_KEY);
446 }
447 }
448
449 return error;
450 }
451
452 /*
453 * exec_reset_save_path
454 *
455 * If we detect a shell script, we need to reset the string area
456 * state so that the interpreter can be saved onto the stack.
457 *
458 * Parameters; struct image_params * image parameter block
459 *
460 * Returns: int 0 Success
461 *
462 * Implicit returns:
463 * (imgp->ip_strings) saved path
464 * (imgp->ip_strspace) space remaining in ip_strings
465 * (imgp->ip_strendp) start of remaining copy area
466 * (imgp->ip_argspace) space remaining of NCARGS
467 *
468 */
469 static int
470 exec_reset_save_path(struct image_params *imgp)
471 {
472 imgp->ip_strendp = imgp->ip_strings;
473 imgp->ip_argspace = NCARGS;
474 imgp->ip_strspace = (NCARGS + PAGE_SIZE);
475
476 return 0;
477 }
478
479 /*
480 * exec_shell_imgact
481 *
482 * Image activator for interpreter scripts. If the image begins with
483 * the characters "#!", then it is an interpreter script. Verify the
484 * length of the script line indicating the interpreter is not in
485 * excess of the maximum allowed size. If this is the case, then
486 * break out the arguments, if any, which are separated by white
487 * space, and copy them into the argument save area as if they were
488 * provided on the command line before all other arguments. The line
489 * ends when we encounter a comment character ('#') or newline.
490 *
491 * Parameters; struct image_params * image parameter block
492 *
493 * Returns: -1 not an interpreter (keep looking)
494 * -3 Success: interpreter: relookup
495 * >0 Failure: interpreter: error number
496 *
497 * A return value other than -1 indicates subsequent image activators should
498 * not be given the opportunity to attempt to activate the image.
499 */
500 static int
501 exec_shell_imgact(struct image_params *imgp)
502 {
503 char *vdata = imgp->ip_vdata;
504 char *ihp;
505 char *line_startp, *line_endp;
506 char *interp;
507
508 /*
509 * Make sure it's a shell script. If we've already redirected
510 * from an interpreted file once, don't do it again.
511 */
512 if (vdata[0] != '#' ||
513 vdata[1] != '!' ||
514 (imgp->ip_flags & IMGPF_INTERPRET) != 0) {
515 return -1;
516 }
517
518 if (imgp->ip_origcputype != 0) {
519 /* Fat header previously matched, don't allow shell script inside */
520 return -1;
521 }
522
523 imgp->ip_flags |= IMGPF_INTERPRET;
524 imgp->ip_interp_sugid_fd = -1;
525 imgp->ip_interp_buffer[0] = '\0';
526
527 /* Check to see if SUGID scripts are permitted. If they aren't then
528 * clear the SUGID bits.
529 * imgp->ip_vattr is known to be valid.
530 */
531 if (sugid_scripts == 0) {
532 imgp->ip_origvattr->va_mode &= ~(VSUID | VSGID);
533 }
534
535 /* Try to find the first non-whitespace character */
536 for (ihp = &vdata[2]; ihp < &vdata[IMG_SHSIZE]; ihp++) {
537 if (IS_EOL(*ihp)) {
538 /* Did not find interpreter, "#!\n" */
539 return ENOEXEC;
540 } else if (IS_WHITESPACE(*ihp)) {
541 /* Whitespace, like "#! /bin/sh\n", keep going. */
542 } else {
543 /* Found start of interpreter */
544 break;
545 }
546 }
547
548 if (ihp == &vdata[IMG_SHSIZE]) {
549 /* All whitespace, like "#! " */
550 return ENOEXEC;
551 }
552
553 line_startp = ihp;
554
555 /* Try to find the end of the interpreter+args string */
556 for (; ihp < &vdata[IMG_SHSIZE]; ihp++) {
557 if (IS_EOL(*ihp)) {
558 /* Got it */
559 break;
560 } else {
561 /* Still part of interpreter or args */
562 }
563 }
564
565 if (ihp == &vdata[IMG_SHSIZE]) {
566 /* A long line, like "#! blah blah blah" without end */
567 return ENOEXEC;
568 }
569
570 /* Backtrack until we find the last non-whitespace */
571 while (IS_EOL(*ihp) || IS_WHITESPACE(*ihp)) {
572 ihp--;
573 }
574
575 /* The character after the last non-whitespace is our logical end of line */
576 line_endp = ihp + 1;
577
578 /*
579 * Now we have pointers to the usable part of:
580 *
581 * "#! /usr/bin/int first second third \n"
582 * ^ line_startp ^ line_endp
583 */
584
585 /* copy the interpreter name */
586 interp = imgp->ip_interp_buffer;
587 for (ihp = line_startp; (ihp < line_endp) && !IS_WHITESPACE(*ihp); ihp++) {
588 *interp++ = *ihp;
589 }
590 *interp = '\0';
591
592 exec_reset_save_path(imgp);
593 exec_save_path(imgp, CAST_USER_ADDR_T(imgp->ip_interp_buffer),
594 UIO_SYSSPACE, NULL);
595
596 /* Copy the entire interpreter + args for later processing into argv[] */
597 interp = imgp->ip_interp_buffer;
598 for (ihp = line_startp; (ihp < line_endp); ihp++) {
599 *interp++ = *ihp;
600 }
601 *interp = '\0';
602
603 #if !SECURE_KERNEL
604 /*
605 * If we have an SUID or SGID script, create a file descriptor
606 * from the vnode and pass /dev/fd/%d instead of the actual
607 * path name so that the script does not get opened twice
608 */
609 if (imgp->ip_origvattr->va_mode & (VSUID | VSGID)) {
610 proc_t p;
611 struct fileproc *fp;
612 int fd;
613 int error;
614
615 p = vfs_context_proc(imgp->ip_vfs_context);
616 error = falloc(p, &fp, &fd, imgp->ip_vfs_context);
617 if (error) {
618 return error;
619 }
620
621 fp->f_fglob->fg_flag = FREAD;
622 fp->f_fglob->fg_ops = &vnops;
623 fp->f_fglob->fg_data = (caddr_t)imgp->ip_vp;
624
625 proc_fdlock(p);
626 procfdtbl_releasefd(p, fd, NULL);
627 fp_drop(p, fd, fp, 1);
628 proc_fdunlock(p);
629 vnode_ref(imgp->ip_vp);
630
631 imgp->ip_interp_sugid_fd = fd;
632 }
633 #endif
634
635 return -3;
636 }
637
638
639
640 /*
641 * exec_fat_imgact
642 *
643 * Image activator for fat 1.0 binaries. If the binary is fat, then we
644 * need to select an image from it internally, and make that the image
645 * we are going to attempt to execute. At present, this consists of
646 * reloading the first page for the image with a first page from the
647 * offset location indicated by the fat header.
648 *
649 * Parameters; struct image_params * image parameter block
650 *
651 * Returns: -1 not a fat binary (keep looking)
652 * -2 Success: encapsulated binary: reread
653 * >0 Failure: error number
654 *
655 * Important: This image activator is byte order neutral.
656 *
657 * Note: A return value other than -1 indicates subsequent image
658 * activators should not be given the opportunity to attempt
659 * to activate the image.
660 *
661 * If we find an encapsulated binary, we make no assertions
662 * about its validity; instead, we leave that up to a rescan
663 * for an activator to claim it, and, if it is claimed by one,
664 * that activator is responsible for determining validity.
665 */
666 static int
667 exec_fat_imgact(struct image_params *imgp)
668 {
669 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
670 kauth_cred_t cred = kauth_cred_proc_ref(p);
671 struct fat_header *fat_header = (struct fat_header *)imgp->ip_vdata;
672 struct _posix_spawnattr *psa = NULL;
673 struct fat_arch fat_arch;
674 int resid, error;
675 load_return_t lret;
676
677 if (imgp->ip_origcputype != 0) {
678 /* Fat header previously matched, don't allow another fat file inside */
679 error = -1; /* not claimed */
680 goto bad;
681 }
682
683 /* Make sure it's a fat binary */
684 if (OSSwapBigToHostInt32(fat_header->magic) != FAT_MAGIC) {
685 error = -1; /* not claimed */
686 goto bad;
687 }
688
689 /* imgp->ip_vdata has PAGE_SIZE, zerofilled if the file is smaller */
690 lret = fatfile_validate_fatarches((vm_offset_t)fat_header, PAGE_SIZE);
691 if (lret != LOAD_SUCCESS) {
692 error = load_return_to_errno(lret);
693 goto bad;
694 }
695
696 /* If posix_spawn binprefs exist, respect those prefs. */
697 psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
698 if (psa != NULL && psa->psa_binprefs[0] != 0) {
699 uint32_t pr = 0;
700
701 /* Check each preference listed against all arches in header */
702 for (pr = 0; pr < NBINPREFS; pr++) {
703 cpu_type_t pref = psa->psa_binprefs[pr];
704 if (pref == 0) {
705 /* No suitable arch in the pref list */
706 error = EBADARCH;
707 goto bad;
708 }
709
710 if (pref == CPU_TYPE_ANY) {
711 /* Fall through to regular grading */
712 goto regular_grading;
713 }
714
715 lret = fatfile_getbestarch_for_cputype(pref,
716 (vm_offset_t)fat_header,
717 PAGE_SIZE,
718 imgp,
719 &fat_arch);
720 if (lret == LOAD_SUCCESS) {
721 goto use_arch;
722 }
723 }
724
725 /* Requested binary preference was not honored */
726 error = EBADEXEC;
727 goto bad;
728 }
729
730 regular_grading:
731 /* Look up our preferred architecture in the fat file. */
732 lret = fatfile_getbestarch((vm_offset_t)fat_header,
733 PAGE_SIZE,
734 imgp,
735 &fat_arch);
736 if (lret != LOAD_SUCCESS) {
737 error = load_return_to_errno(lret);
738 goto bad;
739 }
740
741 use_arch:
742 /* Read the Mach-O header out of fat_arch */
743 error = vn_rdwr(UIO_READ, imgp->ip_vp, imgp->ip_vdata,
744 PAGE_SIZE, fat_arch.offset,
745 UIO_SYSSPACE, (IO_UNIT | IO_NODELOCKED),
746 cred, &resid, p);
747 if (error) {
748 goto bad;
749 }
750
751 if (resid) {
752 memset(imgp->ip_vdata + (PAGE_SIZE - resid), 0x0, resid);
753 }
754
755 /* Success. Indicate we have identified an encapsulated binary */
756 error = -2;
757 imgp->ip_arch_offset = (user_size_t)fat_arch.offset;
758 imgp->ip_arch_size = (user_size_t)fat_arch.size;
759 imgp->ip_origcputype = fat_arch.cputype;
760 imgp->ip_origcpusubtype = fat_arch.cpusubtype;
761
762 bad:
763 kauth_cred_unref(&cred);
764 return error;
765 }
766
767 static int
768 activate_exec_state(task_t task, proc_t p, thread_t thread, load_result_t *result)
769 {
770 int ret;
771
772 task_set_dyld_info(task, MACH_VM_MIN_ADDRESS, 0);
773 task_set_64bit(task, result->is_64bit_addr, result->is_64bit_data);
774 if (result->is_64bit_addr) {
775 OSBitOrAtomic(P_LP64, &p->p_flag);
776 } else {
777 OSBitAndAtomic(~((uint32_t)P_LP64), &p->p_flag);
778 }
779 task_set_mach_header_address(task, result->mach_header);
780
781 ret = thread_state_initialize(thread);
782 if (ret != KERN_SUCCESS) {
783 return ret;
784 }
785
786 if (result->threadstate) {
787 uint32_t *ts = result->threadstate;
788 uint32_t total_size = result->threadstate_sz;
789
790 while (total_size > 0) {
791 uint32_t flavor = *ts++;
792 uint32_t size = *ts++;
793
794 ret = thread_setstatus(thread, flavor, (thread_state_t)ts, size);
795 if (ret) {
796 return ret;
797 }
798 ts += size;
799 total_size -= (size + 2) * sizeof(uint32_t);
800 }
801 }
802
803 thread_setentrypoint(thread, result->entry_point);
804
805 return KERN_SUCCESS;
806 }
807
808
809 /*
810 * Set p->p_comm and p->p_name to the name passed to exec
811 */
812 static void
813 set_proc_name(struct image_params *imgp, proc_t p)
814 {
815 int p_name_len = sizeof(p->p_name) - 1;
816
817 if (imgp->ip_ndp->ni_cnd.cn_namelen > p_name_len) {
818 imgp->ip_ndp->ni_cnd.cn_namelen = p_name_len;
819 }
820
821 bcopy((caddr_t)imgp->ip_ndp->ni_cnd.cn_nameptr, (caddr_t)p->p_name,
822 (unsigned)imgp->ip_ndp->ni_cnd.cn_namelen);
823 p->p_name[imgp->ip_ndp->ni_cnd.cn_namelen] = '\0';
824
825 if (imgp->ip_ndp->ni_cnd.cn_namelen > MAXCOMLEN) {
826 imgp->ip_ndp->ni_cnd.cn_namelen = MAXCOMLEN;
827 }
828
829 bcopy((caddr_t)imgp->ip_ndp->ni_cnd.cn_nameptr, (caddr_t)p->p_comm,
830 (unsigned)imgp->ip_ndp->ni_cnd.cn_namelen);
831 p->p_comm[imgp->ip_ndp->ni_cnd.cn_namelen] = '\0';
832 }
833
834 /*
835 * exec_mach_imgact
836 *
837 * Image activator for mach-o 1.0 binaries.
838 *
839 * Parameters; struct image_params * image parameter block
840 *
841 * Returns: -1 not a fat binary (keep looking)
842 * -2 Success: encapsulated binary: reread
843 * >0 Failure: error number
844 * EBADARCH Mach-o binary, but with an unrecognized
845 * architecture
846 * ENOMEM No memory for child process after -
847 * can only happen after vfork()
848 *
849 * Important: This image activator is NOT byte order neutral.
850 *
851 * Note: A return value other than -1 indicates subsequent image
852 * activators should not be given the opportunity to attempt
853 * to activate the image.
854 *
855 * TODO: More gracefully handle failures after vfork
856 */
857 static int
858 exec_mach_imgact(struct image_params *imgp)
859 {
860 struct mach_header *mach_header = (struct mach_header *)imgp->ip_vdata;
861 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
862 int error = 0;
863 task_t task;
864 task_t new_task = NULL; /* protected by vfexec */
865 thread_t thread;
866 struct uthread *uthread;
867 vm_map_t old_map = VM_MAP_NULL;
868 vm_map_t map = VM_MAP_NULL;
869 load_return_t lret;
870 load_result_t load_result = {};
871 struct _posix_spawnattr *psa = NULL;
872 int spawn = (imgp->ip_flags & IMGPF_SPAWN);
873 int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC);
874 int exec = (imgp->ip_flags & IMGPF_EXEC);
875 os_reason_t exec_failure_reason = OS_REASON_NULL;
876
877 /*
878 * make sure it's a Mach-O 1.0 or Mach-O 2.0 binary; the difference
879 * is a reserved field on the end, so for the most part, we can
880 * treat them as if they were identical. Reverse-endian Mach-O
881 * binaries are recognized but not compatible.
882 */
883 if ((mach_header->magic == MH_CIGAM) ||
884 (mach_header->magic == MH_CIGAM_64)) {
885 error = EBADARCH;
886 goto bad;
887 }
888
889 if ((mach_header->magic != MH_MAGIC) &&
890 (mach_header->magic != MH_MAGIC_64)) {
891 error = -1;
892 goto bad;
893 }
894
895 if (mach_header->filetype != MH_EXECUTE) {
896 error = -1;
897 goto bad;
898 }
899
900 if (imgp->ip_origcputype != 0) {
901 /* Fat header previously had an idea about this thin file */
902 if (imgp->ip_origcputype != mach_header->cputype ||
903 imgp->ip_origcpusubtype != mach_header->cpusubtype) {
904 error = EBADARCH;
905 goto bad;
906 }
907 } else {
908 imgp->ip_origcputype = mach_header->cputype;
909 imgp->ip_origcpusubtype = mach_header->cpusubtype;
910 }
911
912 task = current_task();
913 thread = current_thread();
914 uthread = get_bsdthread_info(thread);
915
916 if ((mach_header->cputype & CPU_ARCH_ABI64) == CPU_ARCH_ABI64) {
917 imgp->ip_flags |= IMGPF_IS_64BIT_ADDR | IMGPF_IS_64BIT_DATA;
918 }
919
920 /* If posix_spawn binprefs exist, respect those prefs. */
921 psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
922 if (psa != NULL && psa->psa_binprefs[0] != 0) {
923 int pr = 0;
924 for (pr = 0; pr < NBINPREFS; pr++) {
925 cpu_type_t pref = psa->psa_binprefs[pr];
926 if (pref == 0) {
927 /* No suitable arch in the pref list */
928 error = EBADARCH;
929 goto bad;
930 }
931
932 if (pref == CPU_TYPE_ANY) {
933 /* Jump to regular grading */
934 goto grade;
935 }
936
937 if (pref == imgp->ip_origcputype) {
938 /* We have a match! */
939 goto grade;
940 }
941 }
942 error = EBADARCH;
943 goto bad;
944 }
945 grade:
946 if (!grade_binary(imgp->ip_origcputype, imgp->ip_origcpusubtype & ~CPU_SUBTYPE_MASK, TRUE)) {
947 error = EBADARCH;
948 goto bad;
949 }
950
951 if (validate_potential_simulator_binary(imgp->ip_origcputype, imgp,
952 imgp->ip_arch_offset, imgp->ip_arch_size) != LOAD_SUCCESS) {
953 #if __x86_64__
954 const char *excpath;
955 error = exec_save_path(imgp, imgp->ip_user_fname, imgp->ip_seg, &excpath);
956 os_log_error(OS_LOG_DEFAULT, "Unsupported 32-bit executable: \"%s\"", (error) ? imgp->ip_vp->v_name : excpath);
957 #endif
958 error = EBADARCH;
959 goto bad;
960 }
961
962 #if defined(HAS_APPLE_PAC)
963 assert(mach_header->cputype == CPU_TYPE_ARM64
964 );
965
966 if (((mach_header->cputype == CPU_TYPE_ARM64 &&
967 (mach_header->cpusubtype & ~CPU_SUBTYPE_MASK) == CPU_SUBTYPE_ARM64E)
968 ) && (CPU_SUBTYPE_ARM64_PTR_AUTH_VERSION(mach_header->cpusubtype) == 0)) {
969 imgp->ip_flags &= ~IMGPF_NOJOP;
970 } else {
971 imgp->ip_flags |= IMGPF_NOJOP;
972 }
973 #endif
974
975 /* Copy in arguments/environment from the old process */
976 error = exec_extract_strings(imgp);
977 if (error) {
978 goto bad;
979 }
980
981 AUDIT_ARG(argv, imgp->ip_startargv, imgp->ip_argc,
982 imgp->ip_endargv - imgp->ip_startargv);
983 AUDIT_ARG(envv, imgp->ip_endargv, imgp->ip_envc,
984 imgp->ip_endenvv - imgp->ip_endargv);
985
986 /*
987 * We are being called to activate an image subsequent to a vfork()
988 * operation; in this case, we know that our task, thread, and
989 * uthread are actually those of our parent, and our proc, which we
990 * obtained indirectly from the image_params vfs_context_t, is the
991 * new child process.
992 */
993 if (vfexec) {
994 imgp->ip_new_thread = fork_create_child(task,
995 NULL,
996 p,
997 FALSE,
998 (imgp->ip_flags & IMGPF_IS_64BIT_ADDR),
999 (imgp->ip_flags & IMGPF_IS_64BIT_DATA),
1000 FALSE);
1001 /* task and thread ref returned, will be released in __mac_execve */
1002 if (imgp->ip_new_thread == NULL) {
1003 error = ENOMEM;
1004 goto bad;
1005 }
1006 }
1007
1008
1009 /* reset local idea of thread, uthread, task */
1010 thread = imgp->ip_new_thread;
1011 uthread = get_bsdthread_info(thread);
1012 task = new_task = get_threadtask(thread);
1013
1014 /*
1015 * Load the Mach-O file.
1016 *
1017 * NOTE: An error after this point indicates we have potentially
1018 * destroyed or overwritten some process state while attempting an
1019 * execve() following a vfork(), which is an unrecoverable condition.
1020 * We send the new process an immediate SIGKILL to avoid it executing
1021 * any instructions in the mutated address space. For true spawns,
1022 * this is not the case, and "too late" is still not too late to
1023 * return an error code to the parent process.
1024 */
1025
1026 /*
1027 * Actually load the image file we previously decided to load.
1028 */
1029 lret = load_machfile(imgp, mach_header, thread, &map, &load_result);
1030 if (lret != LOAD_SUCCESS) {
1031 error = load_return_to_errno(lret);
1032
1033 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1034 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_BAD_MACHO, 0, 0);
1035 if (lret == LOAD_BADMACHO_UPX) {
1036 set_proc_name(imgp, p);
1037 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_UPX);
1038 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1039 } else {
1040 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_BAD_MACHO);
1041
1042 if (bootarg_execfailurereports) {
1043 set_proc_name(imgp, p);
1044 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1045 }
1046 }
1047
1048 exec_failure_reason->osr_flags |= OS_REASON_FLAG_CONSISTENT_FAILURE;
1049
1050 goto badtoolate;
1051 }
1052
1053 proc_lock(p);
1054 p->p_cputype = imgp->ip_origcputype;
1055 p->p_cpusubtype = imgp->ip_origcpusubtype;
1056 p->p_platform = load_result.ip_platform;
1057 p->p_sdk = load_result.lr_sdk;
1058 proc_unlock(p);
1059
1060 vm_map_set_user_wire_limit(map, p->p_rlimit[RLIMIT_MEMLOCK].rlim_cur);
1061
1062 /*
1063 * Set code-signing flags if this binary is signed, or if parent has
1064 * requested them on exec.
1065 */
1066 if (load_result.csflags & CS_VALID) {
1067 imgp->ip_csflags |= load_result.csflags &
1068 (CS_VALID | CS_SIGNED | CS_DEV_CODE |
1069 CS_HARD | CS_KILL | CS_RESTRICT | CS_ENFORCEMENT | CS_REQUIRE_LV |
1070 CS_FORCED_LV | CS_ENTITLEMENTS_VALIDATED | CS_DYLD_PLATFORM | CS_RUNTIME |
1071 CS_ENTITLEMENT_FLAGS |
1072 CS_EXEC_SET_HARD | CS_EXEC_SET_KILL | CS_EXEC_SET_ENFORCEMENT);
1073 } else {
1074 imgp->ip_csflags &= ~CS_VALID;
1075 }
1076
1077 if (p->p_csflags & CS_EXEC_SET_HARD) {
1078 imgp->ip_csflags |= CS_HARD;
1079 }
1080 if (p->p_csflags & CS_EXEC_SET_KILL) {
1081 imgp->ip_csflags |= CS_KILL;
1082 }
1083 if (p->p_csflags & CS_EXEC_SET_ENFORCEMENT) {
1084 imgp->ip_csflags |= CS_ENFORCEMENT;
1085 }
1086 if (p->p_csflags & CS_EXEC_INHERIT_SIP) {
1087 if (p->p_csflags & CS_INSTALLER) {
1088 imgp->ip_csflags |= CS_INSTALLER;
1089 }
1090 if (p->p_csflags & CS_DATAVAULT_CONTROLLER) {
1091 imgp->ip_csflags |= CS_DATAVAULT_CONTROLLER;
1092 }
1093 if (p->p_csflags & CS_NVRAM_UNRESTRICTED) {
1094 imgp->ip_csflags |= CS_NVRAM_UNRESTRICTED;
1095 }
1096 }
1097
1098 /*
1099 * Set up the system reserved areas in the new address space.
1100 */
1101 int cpu_subtype;
1102 cpu_subtype = 0; /* all cpu_subtypes use the same shared region */
1103 #if defined(HAS_APPLE_PAC)
1104 if (cpu_type() == CPU_TYPE_ARM64 &&
1105 (p->p_cpusubtype & ~CPU_SUBTYPE_MASK) == CPU_SUBTYPE_ARM64E) {
1106 assertf(p->p_cputype == CPU_TYPE_ARM64,
1107 "p %p cpu_type() 0x%x p->p_cputype 0x%x p->p_cpusubtype 0x%x",
1108 p, cpu_type(), p->p_cputype, p->p_cpusubtype);
1109 /*
1110 * arm64e uses pointer authentication, so request a separate
1111 * shared region for this CPU subtype.
1112 */
1113 cpu_subtype = p->p_cpusubtype & ~CPU_SUBTYPE_MASK;
1114 }
1115 #endif /* HAS_APPLE_PAC */
1116 vm_map_exec(map, task, load_result.is_64bit_addr, (void *)p->p_fd->fd_rdir, cpu_type(), cpu_subtype);
1117
1118 /*
1119 * Close file descriptors which specify close-on-exec.
1120 */
1121 fdexec(p, psa != NULL ? psa->psa_flags : 0, exec);
1122
1123 /*
1124 * deal with set[ug]id.
1125 */
1126 error = exec_handle_sugid(imgp);
1127 if (error) {
1128 vm_map_deallocate(map);
1129
1130 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1131 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_SUGID_FAILURE, 0, 0);
1132
1133 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_SUGID_FAILURE);
1134 if (bootarg_execfailurereports) {
1135 set_proc_name(imgp, p);
1136 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1137 }
1138
1139 goto badtoolate;
1140 }
1141
1142 /*
1143 * Commit to new map.
1144 *
1145 * Swap the new map for the old for target task, which consumes
1146 * our new map reference but each leaves us responsible for the
1147 * old_map reference. That lets us get off the pmap associated
1148 * with it, and then we can release it.
1149 *
1150 * The map needs to be set on the target task which is different
1151 * than current task, thus swap_task_map is used instead of
1152 * vm_map_switch.
1153 */
1154 old_map = swap_task_map(task, thread, map);
1155 vm_map_deallocate(old_map);
1156 old_map = NULL;
1157
1158 lret = activate_exec_state(task, p, thread, &load_result);
1159 if (lret != KERN_SUCCESS) {
1160 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1161 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_ACTV_THREADSTATE, 0, 0);
1162
1163 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_ACTV_THREADSTATE);
1164 if (bootarg_execfailurereports) {
1165 set_proc_name(imgp, p);
1166 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1167 }
1168
1169 goto badtoolate;
1170 }
1171
1172 /*
1173 * deal with voucher on exec-calling thread.
1174 */
1175 if (imgp->ip_new_thread == NULL) {
1176 thread_set_mach_voucher(current_thread(), IPC_VOUCHER_NULL);
1177 }
1178
1179 /* Make sure we won't interrupt ourself signalling a partial process */
1180 if (!vfexec && !spawn && (p->p_lflag & P_LTRACED)) {
1181 psignal(p, SIGTRAP);
1182 }
1183
1184 if (load_result.unixproc &&
1185 create_unix_stack(get_task_map(task),
1186 &load_result,
1187 p) != KERN_SUCCESS) {
1188 error = load_return_to_errno(LOAD_NOSPACE);
1189
1190 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1191 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_STACK_ALLOC, 0, 0);
1192
1193 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_STACK_ALLOC);
1194 if (bootarg_execfailurereports) {
1195 set_proc_name(imgp, p);
1196 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1197 }
1198
1199 goto badtoolate;
1200 }
1201
1202 error = exec_add_apple_strings(imgp, &load_result);
1203 if (error) {
1204 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1205 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_APPLE_STRING_INIT, 0, 0);
1206
1207 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_APPLE_STRING_INIT);
1208 if (bootarg_execfailurereports) {
1209 set_proc_name(imgp, p);
1210 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1211 }
1212 goto badtoolate;
1213 }
1214
1215 /* Switch to target task's map to copy out strings */
1216 old_map = vm_map_switch(get_task_map(task));
1217
1218 if (load_result.unixproc) {
1219 user_addr_t ap;
1220
1221 /*
1222 * Copy the strings area out into the new process address
1223 * space.
1224 */
1225 ap = p->user_stack;
1226 error = exec_copyout_strings(imgp, &ap);
1227 if (error) {
1228 vm_map_switch(old_map);
1229
1230 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1231 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_STRINGS, 0, 0);
1232
1233 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_STRINGS);
1234 if (bootarg_execfailurereports) {
1235 set_proc_name(imgp, p);
1236 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1237 }
1238 goto badtoolate;
1239 }
1240 /* Set the stack */
1241 thread_setuserstack(thread, ap);
1242 }
1243
1244 if (load_result.dynlinker) {
1245 uint64_t ap;
1246 int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT_ADDR) ? 8 : 4;
1247
1248 /* Adjust the stack */
1249 ap = thread_adjuserstack(thread, -new_ptr_size);
1250 error = copyoutptr(load_result.mach_header, ap, new_ptr_size);
1251
1252 if (error) {
1253 vm_map_switch(old_map);
1254
1255 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
1256 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_DYNLINKER, 0, 0);
1257
1258 exec_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_COPYOUT_DYNLINKER);
1259 if (bootarg_execfailurereports) {
1260 set_proc_name(imgp, p);
1261 exec_failure_reason->osr_flags |= OS_REASON_FLAG_GENERATE_CRASH_REPORT;
1262 }
1263 goto badtoolate;
1264 }
1265 task_set_dyld_info(task, load_result.all_image_info_addr,
1266 load_result.all_image_info_size);
1267 }
1268
1269 /* Avoid immediate VM faults back into kernel */
1270 exec_prefault_data(p, imgp, &load_result);
1271
1272 vm_map_switch(old_map);
1273
1274 /*
1275 * Reset signal state.
1276 */
1277 execsigs(p, thread);
1278
1279 /*
1280 * need to cancel async IO requests that can be cancelled and wait for those
1281 * already active. MAY BLOCK!
1282 */
1283 _aio_exec( p );
1284
1285 #if SYSV_SHM
1286 /* FIXME: Till vmspace inherit is fixed: */
1287 if (!vfexec && p->vm_shm) {
1288 shmexec(p);
1289 }
1290 #endif
1291 #if SYSV_SEM
1292 /* Clean up the semaphores */
1293 semexit(p);
1294 #endif
1295
1296 /*
1297 * Remember file name for accounting.
1298 */
1299 p->p_acflag &= ~AFORK;
1300
1301 set_proc_name(imgp, p);
1302
1303 #if CONFIG_SECLUDED_MEMORY
1304 if (secluded_for_apps &&
1305 load_result.platform_binary) {
1306 if (strncmp(p->p_name,
1307 "Camera",
1308 sizeof(p->p_name)) == 0) {
1309 task_set_could_use_secluded_mem(task, TRUE);
1310 } else {
1311 task_set_could_use_secluded_mem(task, FALSE);
1312 }
1313 if (strncmp(p->p_name,
1314 "mediaserverd",
1315 sizeof(p->p_name)) == 0) {
1316 task_set_could_also_use_secluded_mem(task, TRUE);
1317 }
1318 }
1319 #endif /* CONFIG_SECLUDED_MEMORY */
1320
1321 #if __arm64__
1322 if (load_result.legacy_footprint) {
1323 task_set_legacy_footprint(task);
1324 }
1325 #endif /* __arm64__ */
1326
1327 pal_dbg_set_task_name(task);
1328
1329 /*
1330 * The load result will have already been munged by AMFI to include the
1331 * platform binary flag if boot-args dictated it (AMFI will mark anything
1332 * that doesn't go through the upcall path as a platform binary if its
1333 * enforcement is disabled).
1334 */
1335 if (load_result.platform_binary) {
1336 if (cs_debug) {
1337 printf("setting platform binary on task: pid = %d\n", p->p_pid);
1338 }
1339
1340 /*
1341 * We must use 'task' here because the proc's task has not yet been
1342 * switched to the new one.
1343 */
1344 task_set_platform_binary(task, TRUE);
1345 } else {
1346 if (cs_debug) {
1347 printf("clearing platform binary on task: pid = %d\n", p->p_pid);
1348 }
1349
1350 task_set_platform_binary(task, FALSE);
1351 }
1352
1353 #if DEVELOPMENT || DEBUG
1354 /*
1355 * Update the pid an proc name for importance base if any
1356 */
1357 task_importance_update_owner_info(task);
1358 #endif
1359
1360 memcpy(&p->p_uuid[0], &load_result.uuid[0], sizeof(p->p_uuid));
1361
1362 #if CONFIG_DTRACE
1363 dtrace_proc_exec(p);
1364 #endif
1365
1366 if (kdebug_enable) {
1367 long args[4] = {};
1368
1369 uintptr_t fsid = 0, fileid = 0;
1370 if (imgp->ip_vattr) {
1371 uint64_t fsid64 = vnode_get_va_fsid(imgp->ip_vattr);
1372 fsid = fsid64;
1373 fileid = imgp->ip_vattr->va_fileid;
1374 // check for (unexpected) overflow and trace zero in that case
1375 if (fsid != fsid64 || fileid != imgp->ip_vattr->va_fileid) {
1376 fsid = fileid = 0;
1377 }
1378 }
1379 KERNEL_DEBUG_CONSTANT_IST1(TRACE_DATA_EXEC, p->p_pid, fsid, fileid, 0,
1380 (uintptr_t)thread_tid(thread));
1381
1382 /*
1383 * Collect the pathname for tracing
1384 */
1385 kdbg_trace_string(p, &args[0], &args[1], &args[2], &args[3]);
1386 KERNEL_DEBUG_CONSTANT_IST1(TRACE_STRING_EXEC, args[0], args[1],
1387 args[2], args[3], (uintptr_t)thread_tid(thread));
1388 }
1389
1390 /*
1391 * If posix_spawned with the START_SUSPENDED flag, stop the
1392 * process before it runs.
1393 */
1394 if (imgp->ip_px_sa != NULL) {
1395 psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
1396 if (psa->psa_flags & POSIX_SPAWN_START_SUSPENDED) {
1397 proc_lock(p);
1398 p->p_stat = SSTOP;
1399 proc_unlock(p);
1400 (void) task_suspend_internal(task);
1401 }
1402 }
1403
1404 /*
1405 * mark as execed, wakeup the process that vforked (if any) and tell
1406 * it that it now has its own resources back
1407 */
1408 OSBitOrAtomic(P_EXEC, &p->p_flag);
1409 proc_resetregister(p);
1410 if (p->p_pptr && (p->p_lflag & P_LPPWAIT)) {
1411 proc_lock(p);
1412 p->p_lflag &= ~P_LPPWAIT;
1413 proc_unlock(p);
1414 wakeup((caddr_t)p->p_pptr);
1415 }
1416
1417 /*
1418 * Pay for our earlier safety; deliver the delayed signals from
1419 * the incomplete vfexec process now that it's complete.
1420 */
1421 if (vfexec && (p->p_lflag & P_LTRACED)) {
1422 psignal_vfork(p, new_task, thread, SIGTRAP);
1423 }
1424
1425 goto done;
1426
1427 badtoolate:
1428 /* Don't allow child process to execute any instructions */
1429 if (!spawn) {
1430 if (vfexec) {
1431 assert(exec_failure_reason != OS_REASON_NULL);
1432 psignal_vfork_with_reason(p, new_task, thread, SIGKILL, exec_failure_reason);
1433 exec_failure_reason = OS_REASON_NULL;
1434 } else {
1435 assert(exec_failure_reason != OS_REASON_NULL);
1436 psignal_with_reason(p, SIGKILL, exec_failure_reason);
1437 exec_failure_reason = OS_REASON_NULL;
1438
1439 if (exec) {
1440 /* Terminate the exec copy task */
1441 task_terminate_internal(task);
1442 }
1443 }
1444
1445 /* We can't stop this system call at this point, so just pretend we succeeded */
1446 error = 0;
1447 } else {
1448 os_reason_free(exec_failure_reason);
1449 exec_failure_reason = OS_REASON_NULL;
1450 }
1451
1452 done:
1453 if (load_result.threadstate) {
1454 kfree(load_result.threadstate, load_result.threadstate_sz);
1455 load_result.threadstate = NULL;
1456 }
1457
1458 bad:
1459 /* If we hit this, we likely would have leaked an exit reason */
1460 assert(exec_failure_reason == OS_REASON_NULL);
1461 return error;
1462 }
1463
1464
1465
1466
1467 /*
1468 * Our image activator table; this is the table of the image types we are
1469 * capable of loading. We list them in order of preference to ensure the
1470 * fastest image load speed.
1471 *
1472 * XXX hardcoded, for now; should use linker sets
1473 */
1474 struct execsw {
1475 int(*const ex_imgact)(struct image_params *);
1476 const char *ex_name;
1477 }const execsw[] = {
1478 { exec_mach_imgact, "Mach-o Binary" },
1479 { exec_fat_imgact, "Fat Binary" },
1480 { exec_shell_imgact, "Interpreter Script" },
1481 { NULL, NULL}
1482 };
1483
1484
1485 /*
1486 * exec_activate_image
1487 *
1488 * Description: Iterate through the available image activators, and activate
1489 * the image associated with the imgp structure. We start with
1490 * the activator for Mach-o binaries followed by that for Fat binaries
1491 * for Interpreter scripts.
1492 *
1493 * Parameters: struct image_params * Image parameter block
1494 *
1495 * Returns: 0 Success
1496 * EBADEXEC The executable is corrupt/unknown
1497 * execargs_alloc:EINVAL Invalid argument
1498 * execargs_alloc:EACCES Permission denied
1499 * execargs_alloc:EINTR Interrupted function
1500 * execargs_alloc:ENOMEM Not enough space
1501 * exec_save_path:EFAULT Bad address
1502 * exec_save_path:ENAMETOOLONG Filename too long
1503 * exec_check_permissions:EACCES Permission denied
1504 * exec_check_permissions:ENOEXEC Executable file format error
1505 * exec_check_permissions:ETXTBSY Text file busy [misuse of error code]
1506 * exec_check_permissions:???
1507 * namei:???
1508 * vn_rdwr:??? [anything vn_rdwr can return]
1509 * <ex_imgact>:??? [anything an imgact can return]
1510 * EDEADLK Process is being terminated
1511 */
1512 static int
1513 exec_activate_image(struct image_params *imgp)
1514 {
1515 struct nameidata *ndp = NULL;
1516 const char *excpath;
1517 int error;
1518 int resid;
1519 int once = 1; /* save SGUID-ness for interpreted files */
1520 int i;
1521 int itercount = 0;
1522 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
1523
1524 error = execargs_alloc(imgp);
1525 if (error) {
1526 goto bad_notrans;
1527 }
1528
1529 error = exec_save_path(imgp, imgp->ip_user_fname, imgp->ip_seg, &excpath);
1530 if (error) {
1531 goto bad_notrans;
1532 }
1533
1534 /* Use excpath, which contains the copyin-ed exec path */
1535 DTRACE_PROC1(exec, uintptr_t, excpath);
1536
1537 MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
1538 if (ndp == NULL) {
1539 error = ENOMEM;
1540 goto bad_notrans;
1541 }
1542
1543 NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF | AUDITVNPATH1,
1544 UIO_SYSSPACE, CAST_USER_ADDR_T(excpath), imgp->ip_vfs_context);
1545
1546 again:
1547 error = namei(ndp);
1548 if (error) {
1549 goto bad_notrans;
1550 }
1551 imgp->ip_ndp = ndp; /* successful namei(); call nameidone() later */
1552 imgp->ip_vp = ndp->ni_vp; /* if set, need to vnode_put() at some point */
1553
1554 /*
1555 * Before we start the transition from binary A to binary B, make
1556 * sure another thread hasn't started exiting the process. We grab
1557 * the proc lock to check p_lflag initially, and the transition
1558 * mechanism ensures that the value doesn't change after we release
1559 * the lock.
1560 */
1561 proc_lock(p);
1562 if (p->p_lflag & P_LEXIT) {
1563 error = EDEADLK;
1564 proc_unlock(p);
1565 goto bad_notrans;
1566 }
1567 error = proc_transstart(p, 1, 0);
1568 proc_unlock(p);
1569 if (error) {
1570 goto bad_notrans;
1571 }
1572
1573 error = exec_check_permissions(imgp);
1574 if (error) {
1575 goto bad;
1576 }
1577
1578 /* Copy; avoid invocation of an interpreter overwriting the original */
1579 if (once) {
1580 once = 0;
1581 *imgp->ip_origvattr = *imgp->ip_vattr;
1582 }
1583
1584 error = vn_rdwr(UIO_READ, imgp->ip_vp, imgp->ip_vdata, PAGE_SIZE, 0,
1585 UIO_SYSSPACE, IO_NODELOCKED,
1586 vfs_context_ucred(imgp->ip_vfs_context),
1587 &resid, vfs_context_proc(imgp->ip_vfs_context));
1588 if (error) {
1589 goto bad;
1590 }
1591
1592 if (resid) {
1593 memset(imgp->ip_vdata + (PAGE_SIZE - resid), 0x0, resid);
1594 }
1595
1596 encapsulated_binary:
1597 /* Limit the number of iterations we will attempt on each binary */
1598 if (++itercount > EAI_ITERLIMIT) {
1599 error = EBADEXEC;
1600 goto bad;
1601 }
1602 error = -1;
1603 for (i = 0; error == -1 && execsw[i].ex_imgact != NULL; i++) {
1604 error = (*execsw[i].ex_imgact)(imgp);
1605
1606 switch (error) {
1607 /* case -1: not claimed: continue */
1608 case -2: /* Encapsulated binary, imgp->ip_XXX set for next iteration */
1609 goto encapsulated_binary;
1610
1611 case -3: /* Interpreter */
1612 #if CONFIG_MACF
1613 /*
1614 * Copy the script label for later use. Note that
1615 * the label can be different when the script is
1616 * actually read by the interpreter.
1617 */
1618 if (imgp->ip_scriptlabelp) {
1619 mac_vnode_label_free(imgp->ip_scriptlabelp);
1620 }
1621 imgp->ip_scriptlabelp = mac_vnode_label_alloc();
1622 if (imgp->ip_scriptlabelp == NULL) {
1623 error = ENOMEM;
1624 break;
1625 }
1626 mac_vnode_label_copy(imgp->ip_vp->v_label,
1627 imgp->ip_scriptlabelp);
1628
1629 /*
1630 * Take a ref of the script vnode for later use.
1631 */
1632 if (imgp->ip_scriptvp) {
1633 vnode_put(imgp->ip_scriptvp);
1634 imgp->ip_scriptvp = NULLVP;
1635 }
1636 if (vnode_getwithref(imgp->ip_vp) == 0) {
1637 imgp->ip_scriptvp = imgp->ip_vp;
1638 }
1639 #endif
1640
1641 nameidone(ndp);
1642
1643 vnode_put(imgp->ip_vp);
1644 imgp->ip_vp = NULL; /* already put */
1645 imgp->ip_ndp = NULL; /* already nameidone */
1646
1647 /* Use excpath, which exec_shell_imgact reset to the interpreter */
1648 NDINIT(ndp, LOOKUP, OP_LOOKUP, FOLLOW | LOCKLEAF,
1649 UIO_SYSSPACE, CAST_USER_ADDR_T(excpath), imgp->ip_vfs_context);
1650
1651 proc_transend(p, 0);
1652 goto again;
1653
1654 default:
1655 break;
1656 }
1657 }
1658
1659 if (error == 0) {
1660 if (imgp->ip_flags & IMGPF_INTERPRET && ndp->ni_vp) {
1661 AUDIT_ARG(vnpath, ndp->ni_vp, ARG_VNODE2);
1662 }
1663
1664 /*
1665 * Call out to allow 3rd party notification of exec.
1666 * Ignore result of kauth_authorize_fileop call.
1667 */
1668 if (kauth_authorize_fileop_has_listeners()) {
1669 kauth_authorize_fileop(vfs_context_ucred(imgp->ip_vfs_context),
1670 KAUTH_FILEOP_EXEC,
1671 (uintptr_t)ndp->ni_vp, 0);
1672 }
1673 }
1674 bad:
1675 proc_transend(p, 0);
1676
1677 bad_notrans:
1678 if (imgp->ip_strings) {
1679 execargs_free(imgp);
1680 }
1681 if (imgp->ip_ndp) {
1682 nameidone(imgp->ip_ndp);
1683 }
1684 if (ndp) {
1685 FREE(ndp, M_TEMP);
1686 }
1687
1688 return error;
1689 }
1690
1691 /*
1692 * exec_validate_spawnattr_policy
1693 *
1694 * Description: Validates the entitlements required to set the apptype.
1695 *
1696 * Parameters: int psa_apptype posix spawn attribute apptype
1697 *
1698 * Returns: 0 Success
1699 * EPERM Failure
1700 */
1701 static errno_t
1702 exec_validate_spawnattr_policy(int psa_apptype)
1703 {
1704 if ((psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) != 0) {
1705 int proctype = psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK;
1706 if (proctype == POSIX_SPAWN_PROC_TYPE_DRIVER) {
1707 if (!IOTaskHasEntitlement(current_task(), POSIX_SPAWN_ENTITLEMENT_DRIVER)) {
1708 return EPERM;
1709 }
1710 }
1711 }
1712
1713 return 0;
1714 }
1715
1716 /*
1717 * exec_handle_spawnattr_policy
1718 *
1719 * Description: Decode and apply the posix_spawn apptype, qos clamp, and watchport ports to the task.
1720 *
1721 * Parameters: proc_t p process to apply attributes to
1722 * int psa_apptype posix spawn attribute apptype
1723 *
1724 * Returns: 0 Success
1725 */
1726 static errno_t
1727 exec_handle_spawnattr_policy(proc_t p, thread_t thread, int psa_apptype, uint64_t psa_qos_clamp,
1728 uint64_t psa_darwin_role, struct exec_port_actions *port_actions)
1729 {
1730 int apptype = TASK_APPTYPE_NONE;
1731 int qos_clamp = THREAD_QOS_UNSPECIFIED;
1732 int role = TASK_UNSPECIFIED;
1733
1734 if ((psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) != 0) {
1735 int proctype = psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK;
1736
1737 switch (proctype) {
1738 case POSIX_SPAWN_PROC_TYPE_DAEMON_INTERACTIVE:
1739 apptype = TASK_APPTYPE_DAEMON_INTERACTIVE;
1740 break;
1741 case POSIX_SPAWN_PROC_TYPE_DAEMON_STANDARD:
1742 apptype = TASK_APPTYPE_DAEMON_STANDARD;
1743 break;
1744 case POSIX_SPAWN_PROC_TYPE_DAEMON_ADAPTIVE:
1745 apptype = TASK_APPTYPE_DAEMON_ADAPTIVE;
1746 break;
1747 case POSIX_SPAWN_PROC_TYPE_DAEMON_BACKGROUND:
1748 apptype = TASK_APPTYPE_DAEMON_BACKGROUND;
1749 break;
1750 case POSIX_SPAWN_PROC_TYPE_APP_DEFAULT:
1751 apptype = TASK_APPTYPE_APP_DEFAULT;
1752 break;
1753 #if !CONFIG_EMBEDDED
1754 case POSIX_SPAWN_PROC_TYPE_APP_TAL:
1755 apptype = TASK_APPTYPE_APP_TAL;
1756 break;
1757 #endif /* !CONFIG_EMBEDDED */
1758 case POSIX_SPAWN_PROC_TYPE_DRIVER:
1759 apptype = TASK_APPTYPE_DRIVER;
1760 break;
1761 default:
1762 apptype = TASK_APPTYPE_NONE;
1763 /* TODO: Should an invalid value here fail the spawn? */
1764 break;
1765 }
1766 }
1767
1768 if (psa_qos_clamp != POSIX_SPAWN_PROC_CLAMP_NONE) {
1769 switch (psa_qos_clamp) {
1770 case POSIX_SPAWN_PROC_CLAMP_UTILITY:
1771 qos_clamp = THREAD_QOS_UTILITY;
1772 break;
1773 case POSIX_SPAWN_PROC_CLAMP_BACKGROUND:
1774 qos_clamp = THREAD_QOS_BACKGROUND;
1775 break;
1776 case POSIX_SPAWN_PROC_CLAMP_MAINTENANCE:
1777 qos_clamp = THREAD_QOS_MAINTENANCE;
1778 break;
1779 default:
1780 qos_clamp = THREAD_QOS_UNSPECIFIED;
1781 /* TODO: Should an invalid value here fail the spawn? */
1782 break;
1783 }
1784 }
1785
1786 if (psa_darwin_role != PRIO_DARWIN_ROLE_DEFAULT) {
1787 proc_darwin_role_to_task_role(psa_darwin_role, &role);
1788 }
1789
1790 if (apptype != TASK_APPTYPE_NONE ||
1791 qos_clamp != THREAD_QOS_UNSPECIFIED ||
1792 role != TASK_UNSPECIFIED ||
1793 port_actions->portwatch_count) {
1794 proc_set_task_spawnpolicy(p->task, thread, apptype, qos_clamp, role,
1795 port_actions->portwatch_array, port_actions->portwatch_count);
1796 }
1797
1798 if (port_actions->registered_count) {
1799 if (mach_ports_register(p->task, port_actions->registered_array,
1800 port_actions->registered_count)) {
1801 return EINVAL;
1802 }
1803 /* mach_ports_register() consumed the array */
1804 port_actions->registered_array = NULL;
1805 port_actions->registered_count = 0;
1806 }
1807
1808 return 0;
1809 }
1810
1811 static void
1812 exec_port_actions_destroy(struct exec_port_actions *port_actions)
1813 {
1814 if (port_actions->portwatch_array) {
1815 for (uint32_t i = 0; i < port_actions->portwatch_count; i++) {
1816 ipc_port_t port = NULL;
1817 if ((port = port_actions->portwatch_array[i]) != NULL) {
1818 ipc_port_release_send(port);
1819 }
1820 }
1821 kfree(port_actions->portwatch_array,
1822 port_actions->portwatch_count * sizeof(ipc_port_t *));
1823 }
1824
1825 if (port_actions->registered_array) {
1826 for (uint32_t i = 0; i < port_actions->registered_count; i++) {
1827 ipc_port_t port = NULL;
1828 if ((port = port_actions->registered_array[i]) != NULL) {
1829 ipc_port_release_send(port);
1830 }
1831 }
1832 kfree(port_actions->registered_array,
1833 port_actions->registered_count * sizeof(ipc_port_t *));
1834 }
1835 }
1836
1837 /*
1838 * exec_handle_port_actions
1839 *
1840 * Description: Go through the _posix_port_actions_t contents,
1841 * calling task_set_special_port, task_set_exception_ports
1842 * and/or audit_session_spawnjoin for the current task.
1843 *
1844 * Parameters: struct image_params * Image parameter block
1845 *
1846 * Returns: 0 Success
1847 * EINVAL Failure
1848 * ENOTSUP Illegal posix_spawn attr flag was set
1849 */
1850 static errno_t
1851 exec_handle_port_actions(struct image_params *imgp,
1852 struct exec_port_actions *actions)
1853 {
1854 _posix_spawn_port_actions_t pacts = imgp->ip_px_spa;
1855 #if CONFIG_AUDIT
1856 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
1857 #endif
1858 _ps_port_action_t *act = NULL;
1859 task_t task = get_threadtask(imgp->ip_new_thread);
1860 ipc_port_t port = NULL;
1861 errno_t ret = 0;
1862 int i, portwatch_i = 0, registered_i = 0;
1863 kern_return_t kr;
1864 boolean_t task_has_watchport_boost = task_has_watchports(current_task());
1865 boolean_t in_exec = (imgp->ip_flags & IMGPF_EXEC);
1866
1867 for (i = 0; i < pacts->pspa_count; i++) {
1868 act = &pacts->pspa_actions[i];
1869
1870 switch (act->port_type) {
1871 case PSPA_SPECIAL:
1872 case PSPA_EXCEPTION:
1873 #if CONFIG_AUDIT
1874 case PSPA_AU_SESSION:
1875 #endif
1876 break;
1877 case PSPA_IMP_WATCHPORTS:
1878 if (++actions->portwatch_count > TASK_MAX_WATCHPORT_COUNT) {
1879 ret = EINVAL;
1880 goto done;
1881 }
1882 break;
1883 case PSPA_REGISTERED_PORTS:
1884 if (++actions->registered_count > TASK_PORT_REGISTER_MAX) {
1885 ret = EINVAL;
1886 goto done;
1887 }
1888 break;
1889 default:
1890 ret = EINVAL;
1891 goto done;
1892 }
1893 }
1894
1895 if (actions->portwatch_count) {
1896 if (in_exec && task_has_watchport_boost) {
1897 ret = EINVAL;
1898 goto done;
1899 }
1900 actions->portwatch_array =
1901 kalloc(sizeof(ipc_port_t *) * actions->portwatch_count);
1902 if (actions->portwatch_array == NULL) {
1903 ret = ENOMEM;
1904 goto done;
1905 }
1906 bzero(actions->portwatch_array,
1907 sizeof(ipc_port_t *) * actions->portwatch_count);
1908 }
1909
1910 if (actions->registered_count) {
1911 actions->registered_array =
1912 kalloc(sizeof(ipc_port_t *) * actions->registered_count);
1913 if (actions->registered_array == NULL) {
1914 ret = ENOMEM;
1915 goto done;
1916 }
1917 bzero(actions->registered_array,
1918 sizeof(ipc_port_t *) * actions->registered_count);
1919 }
1920
1921 for (i = 0; i < pacts->pspa_count; i++) {
1922 act = &pacts->pspa_actions[i];
1923
1924 if (MACH_PORT_VALID(act->new_port)) {
1925 kr = ipc_object_copyin(get_task_ipcspace(current_task()),
1926 act->new_port, MACH_MSG_TYPE_COPY_SEND,
1927 (ipc_object_t *) &port, 0, NULL, IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND);
1928
1929 if (kr != KERN_SUCCESS) {
1930 ret = EINVAL;
1931 goto done;
1932 }
1933 } else {
1934 /* it's NULL or DEAD */
1935 port = CAST_MACH_NAME_TO_PORT(act->new_port);
1936 }
1937
1938 switch (act->port_type) {
1939 case PSPA_SPECIAL:
1940 kr = task_set_special_port(task, act->which, port);
1941
1942 if (kr != KERN_SUCCESS) {
1943 ret = EINVAL;
1944 }
1945 break;
1946
1947 case PSPA_EXCEPTION:
1948 kr = task_set_exception_ports(task, act->mask, port,
1949 act->behavior, act->flavor);
1950 if (kr != KERN_SUCCESS) {
1951 ret = EINVAL;
1952 }
1953 break;
1954 #if CONFIG_AUDIT
1955 case PSPA_AU_SESSION:
1956 ret = audit_session_spawnjoin(p, task, port);
1957 if (ret) {
1958 /* audit_session_spawnjoin() has already dropped the reference in case of error. */
1959 goto done;
1960 }
1961
1962 break;
1963 #endif
1964 case PSPA_IMP_WATCHPORTS:
1965 if (actions->portwatch_array) {
1966 /* hold on to this till end of spawn */
1967 actions->portwatch_array[portwatch_i++] = port;
1968 } else {
1969 ipc_port_release_send(port);
1970 }
1971 break;
1972 case PSPA_REGISTERED_PORTS:
1973 /* hold on to this till end of spawn */
1974 actions->registered_array[registered_i++] = port;
1975 break;
1976 default:
1977 ret = EINVAL;
1978 break;
1979 }
1980
1981 if (ret) {
1982 /* action failed, so release port resources */
1983 ipc_port_release_send(port);
1984 break;
1985 }
1986 }
1987
1988 done:
1989 if (0 != ret) {
1990 DTRACE_PROC1(spawn__port__failure, mach_port_name_t, act->new_port);
1991 }
1992 return ret;
1993 }
1994
1995 /*
1996 * exec_handle_file_actions
1997 *
1998 * Description: Go through the _posix_file_actions_t contents applying the
1999 * open, close, and dup2 operations to the open file table for
2000 * the current process.
2001 *
2002 * Parameters: struct image_params * Image parameter block
2003 *
2004 * Returns: 0 Success
2005 * ???
2006 *
2007 * Note: Actions are applied in the order specified, with the credential
2008 * of the parent process. This is done to permit the parent
2009 * process to utilize POSIX_SPAWN_RESETIDS to drop privilege in
2010 * the child following operations the child may in fact not be
2011 * normally permitted to perform.
2012 */
2013 static int
2014 exec_handle_file_actions(struct image_params *imgp, short psa_flags)
2015 {
2016 int error = 0;
2017 int action;
2018 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
2019 _posix_spawn_file_actions_t px_sfap = imgp->ip_px_sfa;
2020 int ival[2]; /* dummy retval for system calls) */
2021 #if CONFIG_AUDIT
2022 struct uthread *uthread = get_bsdthread_info(current_thread());
2023 #endif
2024
2025 for (action = 0; action < px_sfap->psfa_act_count; action++) {
2026 _psfa_action_t *psfa = &px_sfap->psfa_act_acts[action];
2027
2028 switch (psfa->psfaa_type) {
2029 case PSFA_OPEN: {
2030 /*
2031 * Open is different, in that it requires the use of
2032 * a path argument, which is normally copied in from
2033 * user space; because of this, we have to support an
2034 * open from kernel space that passes an address space
2035 * context of UIO_SYSSPACE, and casts the address
2036 * argument to a user_addr_t.
2037 */
2038 char *bufp = NULL;
2039 struct vnode_attr *vap;
2040 struct nameidata *ndp;
2041 int mode = psfa->psfaa_openargs.psfao_mode;
2042 struct dup2_args dup2a;
2043 struct close_nocancel_args ca;
2044 int origfd;
2045
2046 MALLOC(bufp, char *, sizeof(*vap) + sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
2047 if (bufp == NULL) {
2048 error = ENOMEM;
2049 break;
2050 }
2051
2052 vap = (struct vnode_attr *) bufp;
2053 ndp = (struct nameidata *) (bufp + sizeof(*vap));
2054
2055 VATTR_INIT(vap);
2056 /* Mask off all but regular access permissions */
2057 mode = ((mode & ~p->p_fd->fd_cmask) & ALLPERMS) & ~S_ISTXT;
2058 VATTR_SET(vap, va_mode, mode & ACCESSPERMS);
2059
2060 AUDIT_SUBCALL_ENTER(OPEN, p, uthread);
2061
2062 NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW | AUDITVNPATH1, UIO_SYSSPACE,
2063 CAST_USER_ADDR_T(psfa->psfaa_openargs.psfao_path),
2064 imgp->ip_vfs_context);
2065
2066 error = open1(imgp->ip_vfs_context,
2067 ndp,
2068 psfa->psfaa_openargs.psfao_oflag,
2069 vap,
2070 fileproc_alloc_init, NULL,
2071 ival);
2072
2073 FREE(bufp, M_TEMP);
2074
2075 AUDIT_SUBCALL_EXIT(uthread, error);
2076
2077 /*
2078 * If there's an error, or we get the right fd by
2079 * accident, then drop out here. This is easier than
2080 * reworking all the open code to preallocate fd
2081 * slots, and internally taking one as an argument.
2082 */
2083 if (error || ival[0] == psfa->psfaa_filedes) {
2084 break;
2085 }
2086
2087 origfd = ival[0];
2088 /*
2089 * If we didn't fall out from an error, we ended up
2090 * with the wrong fd; so now we've got to try to dup2
2091 * it to the right one.
2092 */
2093 dup2a.from = origfd;
2094 dup2a.to = psfa->psfaa_filedes;
2095
2096 /*
2097 * The dup2() system call implementation sets
2098 * ival to newfd in the success case, but we
2099 * can ignore that, since if we didn't get the
2100 * fd we wanted, the error will stop us.
2101 */
2102 AUDIT_SUBCALL_ENTER(DUP2, p, uthread);
2103 error = dup2(p, &dup2a, ival);
2104 AUDIT_SUBCALL_EXIT(uthread, error);
2105 if (error) {
2106 break;
2107 }
2108
2109 /*
2110 * Finally, close the original fd.
2111 */
2112 ca.fd = origfd;
2113
2114 AUDIT_SUBCALL_ENTER(CLOSE, p, uthread);
2115 error = close_nocancel(p, &ca, ival);
2116 AUDIT_SUBCALL_EXIT(uthread, error);
2117 }
2118 break;
2119
2120 case PSFA_DUP2: {
2121 struct dup2_args dup2a;
2122
2123 dup2a.from = psfa->psfaa_filedes;
2124 dup2a.to = psfa->psfaa_dup2args.psfad_newfiledes;
2125
2126 /*
2127 * The dup2() system call implementation sets
2128 * ival to newfd in the success case, but we
2129 * can ignore that, since if we didn't get the
2130 * fd we wanted, the error will stop us.
2131 */
2132 AUDIT_SUBCALL_ENTER(DUP2, p, uthread);
2133 error = dup2(p, &dup2a, ival);
2134 AUDIT_SUBCALL_EXIT(uthread, error);
2135 }
2136 break;
2137
2138 case PSFA_FILEPORT_DUP2: {
2139 ipc_port_t port;
2140 kern_return_t kr;
2141 struct dup2_args dup2a;
2142 struct close_nocancel_args ca;
2143
2144 if (!MACH_PORT_VALID(psfa->psfaa_fileport)) {
2145 error = EINVAL;
2146 break;
2147 }
2148
2149 kr = ipc_object_copyin(get_task_ipcspace(current_task()),
2150 psfa->psfaa_fileport, MACH_MSG_TYPE_COPY_SEND,
2151 (ipc_object_t *) &port, 0, NULL, IPC_KMSG_FLAGS_ALLOW_IMMOVABLE_SEND);
2152
2153 if (kr != KERN_SUCCESS) {
2154 error = EINVAL;
2155 break;
2156 }
2157
2158 error = fileport_makefd_internal(p, port, 0, ival);
2159
2160 if (IPC_PORT_NULL != port) {
2161 ipc_port_release_send(port);
2162 }
2163
2164 if (error || ival[0] == psfa->psfaa_dup2args.psfad_newfiledes) {
2165 break;
2166 }
2167
2168 dup2a.from = ca.fd = ival[0];
2169 dup2a.to = psfa->psfaa_dup2args.psfad_newfiledes;
2170 AUDIT_SUBCALL_ENTER(DUP2, p, uthread);
2171 error = dup2(p, &dup2a, ival);
2172 AUDIT_SUBCALL_EXIT(uthread, error);
2173 if (error) {
2174 break;
2175 }
2176
2177 AUDIT_SUBCALL_ENTER(CLOSE, p, uthread);
2178 error = close_nocancel(p, &ca, ival);
2179 AUDIT_SUBCALL_EXIT(uthread, error);
2180 }
2181 break;
2182
2183 case PSFA_CLOSE: {
2184 struct close_nocancel_args ca;
2185
2186 ca.fd = psfa->psfaa_filedes;
2187
2188 AUDIT_SUBCALL_ENTER(CLOSE, p, uthread);
2189 error = close_nocancel(p, &ca, ival);
2190 AUDIT_SUBCALL_EXIT(uthread, error);
2191 }
2192 break;
2193
2194 case PSFA_INHERIT: {
2195 struct fcntl_nocancel_args fcntla;
2196
2197 /*
2198 * Check to see if the descriptor exists, and
2199 * ensure it's -not- marked as close-on-exec.
2200 *
2201 * Attempting to "inherit" a guarded fd will
2202 * result in a error.
2203 */
2204 fcntla.fd = psfa->psfaa_filedes;
2205 fcntla.cmd = F_GETFD;
2206 if ((error = fcntl_nocancel(p, &fcntla, ival)) != 0) {
2207 break;
2208 }
2209
2210 if ((ival[0] & FD_CLOEXEC) == FD_CLOEXEC) {
2211 fcntla.fd = psfa->psfaa_filedes;
2212 fcntla.cmd = F_SETFD;
2213 fcntla.arg = ival[0] & ~FD_CLOEXEC;
2214 error = fcntl_nocancel(p, &fcntla, ival);
2215 }
2216 }
2217 break;
2218
2219 case PSFA_CHDIR: {
2220 /*
2221 * Chdir is different, in that it requires the use of
2222 * a path argument, which is normally copied in from
2223 * user space; because of this, we have to support a
2224 * chdir from kernel space that passes an address space
2225 * context of UIO_SYSSPACE, and casts the address
2226 * argument to a user_addr_t.
2227 */
2228 struct nameidata nd;
2229
2230 AUDIT_SUBCALL_ENTER(CHDIR, p, uthread);
2231 NDINIT(&nd, LOOKUP, OP_CHDIR, FOLLOW | AUDITVNPATH1, UIO_SYSSPACE,
2232 CAST_USER_ADDR_T(psfa->psfaa_chdirargs.psfac_path),
2233 imgp->ip_vfs_context);
2234
2235 error = chdir_internal(p, imgp->ip_vfs_context, &nd, 0);
2236 AUDIT_SUBCALL_EXIT(uthread, error);
2237 }
2238 break;
2239
2240 case PSFA_FCHDIR: {
2241 struct fchdir_args fchdira;
2242
2243 fchdira.fd = psfa->psfaa_filedes;
2244
2245 AUDIT_SUBCALL_ENTER(FCHDIR, p, uthread);
2246 error = fchdir(p, &fchdira, ival);
2247 AUDIT_SUBCALL_EXIT(uthread, error);
2248 }
2249 break;
2250
2251 default:
2252 error = EINVAL;
2253 break;
2254 }
2255
2256 /* All file actions failures are considered fatal, per POSIX */
2257
2258 if (error) {
2259 if (PSFA_OPEN == psfa->psfaa_type) {
2260 DTRACE_PROC1(spawn__open__failure, uintptr_t,
2261 psfa->psfaa_openargs.psfao_path);
2262 } else {
2263 DTRACE_PROC1(spawn__fd__failure, int, psfa->psfaa_filedes);
2264 }
2265 break;
2266 }
2267 }
2268
2269 if (error != 0 || (psa_flags & POSIX_SPAWN_CLOEXEC_DEFAULT) == 0) {
2270 return error;
2271 }
2272
2273 /*
2274 * If POSIX_SPAWN_CLOEXEC_DEFAULT is set, behave (during
2275 * this spawn only) as if "close on exec" is the default
2276 * disposition of all pre-existing file descriptors. In this case,
2277 * the list of file descriptors mentioned in the file actions
2278 * are the only ones that can be inherited, so mark them now.
2279 *
2280 * The actual closing part comes later, in fdexec().
2281 */
2282 proc_fdlock(p);
2283 for (action = 0; action < px_sfap->psfa_act_count; action++) {
2284 _psfa_action_t *psfa = &px_sfap->psfa_act_acts[action];
2285 int fd = psfa->psfaa_filedes;
2286
2287 switch (psfa->psfaa_type) {
2288 case PSFA_DUP2:
2289 case PSFA_FILEPORT_DUP2:
2290 fd = psfa->psfaa_dup2args.psfad_newfiledes;
2291 /*FALLTHROUGH*/
2292 case PSFA_OPEN:
2293 case PSFA_INHERIT:
2294 *fdflags(p, fd) |= UF_INHERIT;
2295 break;
2296
2297 case PSFA_CLOSE:
2298 case PSFA_CHDIR:
2299 case PSFA_FCHDIR:
2300 /*
2301 * Although PSFA_FCHDIR does have a file descriptor, it is not
2302 * *creating* one, thus we do not automatically mark it for
2303 * inheritance under POSIX_SPAWN_CLOEXEC_DEFAULT. A client that
2304 * wishes it to be inherited should use the PSFA_INHERIT action
2305 * explicitly.
2306 */
2307 break;
2308 }
2309 }
2310 proc_fdunlock(p);
2311
2312 return 0;
2313 }
2314
2315 #if CONFIG_MACF
2316 /*
2317 * exec_spawnattr_getmacpolicyinfo
2318 */
2319 void *
2320 exec_spawnattr_getmacpolicyinfo(const void *macextensions, const char *policyname, size_t *lenp)
2321 {
2322 const struct _posix_spawn_mac_policy_extensions *psmx = macextensions;
2323 int i;
2324
2325 if (psmx == NULL) {
2326 return NULL;
2327 }
2328
2329 for (i = 0; i < psmx->psmx_count; i++) {
2330 const _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[i];
2331 if (strncmp(extension->policyname, policyname, sizeof(extension->policyname)) == 0) {
2332 if (lenp != NULL) {
2333 *lenp = extension->datalen;
2334 }
2335 return extension->datap;
2336 }
2337 }
2338
2339 if (lenp != NULL) {
2340 *lenp = 0;
2341 }
2342 return NULL;
2343 }
2344
2345 static int
2346 spawn_copyin_macpolicyinfo(const struct user__posix_spawn_args_desc *px_args, _posix_spawn_mac_policy_extensions_t *psmxp)
2347 {
2348 _posix_spawn_mac_policy_extensions_t psmx = NULL;
2349 int error = 0;
2350 int copycnt = 0;
2351 int i = 0;
2352
2353 *psmxp = NULL;
2354
2355 if (px_args->mac_extensions_size < PS_MAC_EXTENSIONS_SIZE(1) ||
2356 px_args->mac_extensions_size > PAGE_SIZE) {
2357 error = EINVAL;
2358 goto bad;
2359 }
2360
2361 MALLOC(psmx, _posix_spawn_mac_policy_extensions_t, px_args->mac_extensions_size, M_TEMP, M_WAITOK);
2362 if ((error = copyin(px_args->mac_extensions, psmx, px_args->mac_extensions_size)) != 0) {
2363 goto bad;
2364 }
2365
2366 size_t extsize = PS_MAC_EXTENSIONS_SIZE(psmx->psmx_count);
2367 if (extsize == 0 || extsize > px_args->mac_extensions_size) {
2368 error = EINVAL;
2369 goto bad;
2370 }
2371
2372 for (i = 0; i < psmx->psmx_count; i++) {
2373 _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[i];
2374 if (extension->datalen == 0 || extension->datalen > PAGE_SIZE) {
2375 error = EINVAL;
2376 goto bad;
2377 }
2378 }
2379
2380 for (copycnt = 0; copycnt < psmx->psmx_count; copycnt++) {
2381 _ps_mac_policy_extension_t *extension = &psmx->psmx_extensions[copycnt];
2382 void *data = NULL;
2383
2384 MALLOC(data, void *, extension->datalen, M_TEMP, M_WAITOK);
2385 if ((error = copyin(extension->data, data, extension->datalen)) != 0) {
2386 FREE(data, M_TEMP);
2387 goto bad;
2388 }
2389 extension->datap = data;
2390 }
2391
2392 *psmxp = psmx;
2393 return 0;
2394
2395 bad:
2396 if (psmx != NULL) {
2397 for (i = 0; i < copycnt; i++) {
2398 FREE(psmx->psmx_extensions[i].datap, M_TEMP);
2399 }
2400 FREE(psmx, M_TEMP);
2401 }
2402 return error;
2403 }
2404
2405 static void
2406 spawn_free_macpolicyinfo(_posix_spawn_mac_policy_extensions_t psmx)
2407 {
2408 int i;
2409
2410 if (psmx == NULL) {
2411 return;
2412 }
2413 for (i = 0; i < psmx->psmx_count; i++) {
2414 FREE(psmx->psmx_extensions[i].datap, M_TEMP);
2415 }
2416 FREE(psmx, M_TEMP);
2417 }
2418 #endif /* CONFIG_MACF */
2419
2420 #if CONFIG_COALITIONS
2421 static inline void
2422 spawn_coalitions_release_all(coalition_t coal[COALITION_NUM_TYPES])
2423 {
2424 for (int c = 0; c < COALITION_NUM_TYPES; c++) {
2425 if (coal[c]) {
2426 coalition_remove_active(coal[c]);
2427 coalition_release(coal[c]);
2428 }
2429 }
2430 }
2431 #endif
2432
2433 #if CONFIG_PERSONAS
2434 static int
2435 spawn_validate_persona(struct _posix_spawn_persona_info *px_persona)
2436 {
2437 int error = 0;
2438 struct persona *persona = NULL;
2439 int verify = px_persona->pspi_flags & POSIX_SPAWN_PERSONA_FLAGS_VERIFY;
2440
2441 if (!IOTaskHasEntitlement(current_task(), PERSONA_MGMT_ENTITLEMENT)) {
2442 return EPERM;
2443 }
2444
2445 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
2446 if (px_persona->pspi_ngroups > NGROUPS_MAX) {
2447 return EINVAL;
2448 }
2449 }
2450
2451 persona = persona_lookup(px_persona->pspi_id);
2452 if (!persona) {
2453 error = ESRCH;
2454 goto out;
2455 }
2456
2457 if (verify) {
2458 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_UID) {
2459 if (px_persona->pspi_uid != persona_get_uid(persona)) {
2460 error = EINVAL;
2461 goto out;
2462 }
2463 }
2464 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GID) {
2465 if (px_persona->pspi_gid != persona_get_gid(persona)) {
2466 error = EINVAL;
2467 goto out;
2468 }
2469 }
2470 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
2471 unsigned ngroups = 0;
2472 gid_t groups[NGROUPS_MAX];
2473
2474 if (persona_get_groups(persona, &ngroups, groups,
2475 px_persona->pspi_ngroups) != 0) {
2476 error = EINVAL;
2477 goto out;
2478 }
2479 if (ngroups != px_persona->pspi_ngroups) {
2480 error = EINVAL;
2481 goto out;
2482 }
2483 while (ngroups--) {
2484 if (px_persona->pspi_groups[ngroups] != groups[ngroups]) {
2485 error = EINVAL;
2486 goto out;
2487 }
2488 }
2489 if (px_persona->pspi_gmuid != persona_get_gmuid(persona)) {
2490 error = EINVAL;
2491 goto out;
2492 }
2493 }
2494 }
2495
2496 out:
2497 if (persona) {
2498 persona_put(persona);
2499 }
2500
2501 return error;
2502 }
2503
2504 static int
2505 spawn_persona_adopt(proc_t p, struct _posix_spawn_persona_info *px_persona)
2506 {
2507 int ret;
2508 kauth_cred_t cred;
2509 struct persona *persona = NULL;
2510 int override = !!(px_persona->pspi_flags & POSIX_SPAWN_PERSONA_FLAGS_OVERRIDE);
2511
2512 if (!override) {
2513 return persona_proc_adopt_id(p, px_persona->pspi_id, NULL);
2514 }
2515
2516 /*
2517 * we want to spawn into the given persona, but we want to override
2518 * the kauth with a different UID/GID combo
2519 */
2520 persona = persona_lookup(px_persona->pspi_id);
2521 if (!persona) {
2522 return ESRCH;
2523 }
2524
2525 cred = persona_get_cred(persona);
2526 if (!cred) {
2527 ret = EINVAL;
2528 goto out;
2529 }
2530
2531 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_UID) {
2532 cred = kauth_cred_setresuid(cred,
2533 px_persona->pspi_uid,
2534 px_persona->pspi_uid,
2535 px_persona->pspi_uid,
2536 KAUTH_UID_NONE);
2537 }
2538
2539 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GID) {
2540 cred = kauth_cred_setresgid(cred,
2541 px_persona->pspi_gid,
2542 px_persona->pspi_gid,
2543 px_persona->pspi_gid);
2544 }
2545
2546 if (px_persona->pspi_flags & POSIX_SPAWN_PERSONA_GROUPS) {
2547 cred = kauth_cred_setgroups(cred,
2548 px_persona->pspi_groups,
2549 px_persona->pspi_ngroups,
2550 px_persona->pspi_gmuid);
2551 }
2552
2553 ret = persona_proc_adopt(p, persona, cred);
2554
2555 out:
2556 persona_put(persona);
2557 return ret;
2558 }
2559 #endif
2560
2561 #if __arm64__
2562 extern int legacy_footprint_entitlement_mode;
2563 static inline void
2564 proc_legacy_footprint_entitled(proc_t p, task_t task, const char *caller)
2565 {
2566 #pragma unused(p, caller)
2567 boolean_t legacy_footprint_entitled;
2568
2569 switch (legacy_footprint_entitlement_mode) {
2570 case LEGACY_FOOTPRINT_ENTITLEMENT_IGNORE:
2571 /* the entitlement is ignored */
2572 break;
2573 case LEGACY_FOOTPRINT_ENTITLEMENT_IOS11_ACCT:
2574 /* the entitlement grants iOS11 legacy accounting */
2575 legacy_footprint_entitled = IOTaskHasEntitlement(task,
2576 "com.apple.private.memory.legacy_footprint");
2577 if (legacy_footprint_entitled) {
2578 task_set_legacy_footprint(task);
2579 }
2580 break;
2581 case LEGACY_FOOTPRINT_ENTITLEMENT_LIMIT_INCREASE:
2582 /* the entitlement grants a footprint limit increase */
2583 legacy_footprint_entitled = IOTaskHasEntitlement(task,
2584 "com.apple.private.memory.legacy_footprint");
2585 if (legacy_footprint_entitled) {
2586 task_set_extra_footprint_limit(task);
2587 }
2588 break;
2589 default:
2590 break;
2591 }
2592 }
2593
2594 static inline void
2595 proc_ios13extended_footprint_entitled(proc_t p, task_t task, const char *caller)
2596 {
2597 #pragma unused(p, caller)
2598 boolean_t ios13extended_footprint_entitled;
2599
2600 /* the entitlement grants a footprint limit increase */
2601 ios13extended_footprint_entitled = IOTaskHasEntitlement(task,
2602 "com.apple.developer.memory.ios13extended_footprint");
2603 if (ios13extended_footprint_entitled) {
2604 task_set_ios13extended_footprint_limit(task);
2605 }
2606 }
2607 #endif /* __arm64__ */
2608
2609 /*
2610 * Apply a modification on the proc's kauth cred until it converges.
2611 *
2612 * `update` consumes its argument to return a new kauth cred.
2613 */
2614 static void
2615 apply_kauth_cred_update(proc_t p,
2616 kauth_cred_t (^update)(kauth_cred_t orig_cred))
2617 {
2618 kauth_cred_t my_cred, my_new_cred;
2619
2620 my_cred = kauth_cred_proc_ref(p);
2621 for (;;) {
2622 my_new_cred = update(my_cred);
2623 if (my_cred == my_new_cred) {
2624 kauth_cred_unref(&my_new_cred);
2625 break;
2626 }
2627
2628 /* try update cred on proc */
2629 proc_ucred_lock(p);
2630
2631 if (p->p_ucred == my_cred) {
2632 /* base pointer didn't change, donate our ref */
2633 p->p_ucred = my_new_cred;
2634 PROC_UPDATE_CREDS_ONPROC(p);
2635 proc_ucred_unlock(p);
2636
2637 /* drop p->p_ucred reference */
2638 kauth_cred_unref(&my_cred);
2639 break;
2640 }
2641
2642 /* base pointer changed, retry */
2643 my_cred = p->p_ucred;
2644 kauth_cred_ref(my_cred);
2645 proc_ucred_unlock(p);
2646
2647 kauth_cred_unref(&my_new_cred);
2648 }
2649 }
2650
2651 static int
2652 spawn_posix_cred_adopt(proc_t p,
2653 struct _posix_spawn_posix_cred_info *px_pcred_info)
2654 {
2655 int error = 0;
2656
2657 if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_GID) {
2658 struct setgid_args args = {
2659 .gid = px_pcred_info->pspci_gid,
2660 };
2661 error = setgid(p, &args, NULL);
2662 if (error) {
2663 return error;
2664 }
2665 }
2666
2667 if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_GROUPS) {
2668 error = setgroups_internal(p,
2669 px_pcred_info->pspci_ngroups,
2670 px_pcred_info->pspci_groups,
2671 px_pcred_info->pspci_gmuid);
2672 if (error) {
2673 return error;
2674 }
2675 }
2676
2677 if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_UID) {
2678 struct setuid_args args = {
2679 .uid = px_pcred_info->pspci_uid,
2680 };
2681 error = setuid(p, &args, NULL);
2682 if (error) {
2683 return error;
2684 }
2685 }
2686 return 0;
2687 }
2688
2689 /*
2690 * posix_spawn
2691 *
2692 * Parameters: uap->pid Pointer to pid return area
2693 * uap->fname File name to exec
2694 * uap->argp Argument list
2695 * uap->envp Environment list
2696 *
2697 * Returns: 0 Success
2698 * EINVAL Invalid argument
2699 * ENOTSUP Not supported
2700 * ENOEXEC Executable file format error
2701 * exec_activate_image:EINVAL Invalid argument
2702 * exec_activate_image:EACCES Permission denied
2703 * exec_activate_image:EINTR Interrupted function
2704 * exec_activate_image:ENOMEM Not enough space
2705 * exec_activate_image:EFAULT Bad address
2706 * exec_activate_image:ENAMETOOLONG Filename too long
2707 * exec_activate_image:ENOEXEC Executable file format error
2708 * exec_activate_image:ETXTBSY Text file busy [misuse of error code]
2709 * exec_activate_image:EAUTH Image decryption failed
2710 * exec_activate_image:EBADEXEC The executable is corrupt/unknown
2711 * exec_activate_image:???
2712 * mac_execve_enter:???
2713 *
2714 * TODO: Expect to need __mac_posix_spawn() at some point...
2715 * Handle posix_spawnattr_t
2716 * Handle posix_spawn_file_actions_t
2717 */
2718 int
2719 posix_spawn(proc_t ap, struct posix_spawn_args *uap, int32_t *retval)
2720 {
2721 proc_t p = ap; /* quiet bogus GCC vfork() warning */
2722 user_addr_t pid = uap->pid;
2723 int ival[2]; /* dummy retval for setpgid() */
2724 char *bufp = NULL;
2725 struct image_params *imgp;
2726 struct vnode_attr *vap;
2727 struct vnode_attr *origvap;
2728 struct uthread *uthread = 0; /* compiler complains if not set to 0*/
2729 int error, sig;
2730 int is_64 = IS_64BIT_PROCESS(p);
2731 struct vfs_context context;
2732 struct user__posix_spawn_args_desc px_args;
2733 struct _posix_spawnattr px_sa;
2734 _posix_spawn_file_actions_t px_sfap = NULL;
2735 _posix_spawn_port_actions_t px_spap = NULL;
2736 struct __kern_sigaction vec;
2737 boolean_t spawn_no_exec = FALSE;
2738 boolean_t proc_transit_set = TRUE;
2739 boolean_t exec_done = FALSE;
2740 struct exec_port_actions port_actions = { };
2741 vm_size_t px_sa_offset = offsetof(struct _posix_spawnattr, psa_ports);
2742 task_t old_task = current_task();
2743 task_t new_task = NULL;
2744 boolean_t should_release_proc_ref = FALSE;
2745 void *inherit = NULL;
2746 #if CONFIG_PERSONAS
2747 struct _posix_spawn_persona_info *px_persona = NULL;
2748 #endif
2749 struct _posix_spawn_posix_cred_info *px_pcred_info = NULL;
2750
2751 /*
2752 * Allocate a big chunk for locals instead of using stack since these
2753 * structures are pretty big.
2754 */
2755 MALLOC(bufp, char *, (sizeof(*imgp) + sizeof(*vap) + sizeof(*origvap)), M_TEMP, M_WAITOK | M_ZERO);
2756 imgp = (struct image_params *) bufp;
2757 if (bufp == NULL) {
2758 error = ENOMEM;
2759 goto bad;
2760 }
2761 vap = (struct vnode_attr *) (bufp + sizeof(*imgp));
2762 origvap = (struct vnode_attr *) (bufp + sizeof(*imgp) + sizeof(*vap));
2763
2764 /* Initialize the common data in the image_params structure */
2765 imgp->ip_user_fname = uap->path;
2766 imgp->ip_user_argv = uap->argv;
2767 imgp->ip_user_envv = uap->envp;
2768 imgp->ip_vattr = vap;
2769 imgp->ip_origvattr = origvap;
2770 imgp->ip_vfs_context = &context;
2771 imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT_ADDR : IMGPF_NONE);
2772 imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
2773 imgp->ip_mac_return = 0;
2774 imgp->ip_px_persona = NULL;
2775 imgp->ip_px_pcred_info = NULL;
2776 imgp->ip_cs_error = OS_REASON_NULL;
2777 imgp->ip_simulator_binary = IMGPF_SB_DEFAULT;
2778
2779 if (uap->adesc != USER_ADDR_NULL) {
2780 if (is_64) {
2781 error = copyin(uap->adesc, &px_args, sizeof(px_args));
2782 } else {
2783 struct user32__posix_spawn_args_desc px_args32;
2784
2785 error = copyin(uap->adesc, &px_args32, sizeof(px_args32));
2786
2787 /*
2788 * Convert arguments descriptor from external 32 bit
2789 * representation to internal 64 bit representation
2790 */
2791 px_args.attr_size = px_args32.attr_size;
2792 px_args.attrp = CAST_USER_ADDR_T(px_args32.attrp);
2793 px_args.file_actions_size = px_args32.file_actions_size;
2794 px_args.file_actions = CAST_USER_ADDR_T(px_args32.file_actions);
2795 px_args.port_actions_size = px_args32.port_actions_size;
2796 px_args.port_actions = CAST_USER_ADDR_T(px_args32.port_actions);
2797 px_args.mac_extensions_size = px_args32.mac_extensions_size;
2798 px_args.mac_extensions = CAST_USER_ADDR_T(px_args32.mac_extensions);
2799 px_args.coal_info_size = px_args32.coal_info_size;
2800 px_args.coal_info = CAST_USER_ADDR_T(px_args32.coal_info);
2801 px_args.persona_info_size = px_args32.persona_info_size;
2802 px_args.persona_info = CAST_USER_ADDR_T(px_args32.persona_info);
2803 px_args.posix_cred_info_size = px_args32.posix_cred_info_size;
2804 px_args.posix_cred_info = CAST_USER_ADDR_T(px_args32.posix_cred_info);
2805 }
2806 if (error) {
2807 goto bad;
2808 }
2809
2810 if (px_args.attr_size != 0) {
2811 /*
2812 * We are not copying the port_actions pointer,
2813 * because we already have it from px_args.
2814 * This is a bit fragile: <rdar://problem/16427422>
2815 */
2816
2817 if ((error = copyin(px_args.attrp, &px_sa, px_sa_offset)) != 0) {
2818 goto bad;
2819 }
2820
2821 bzero((void *)((unsigned long) &px_sa + px_sa_offset), sizeof(px_sa) - px_sa_offset );
2822
2823 imgp->ip_px_sa = &px_sa;
2824 }
2825 if (px_args.file_actions_size != 0) {
2826 /* Limit file_actions to allowed number of open files */
2827 int maxfa = (p->p_limit ? p->p_rlimit[RLIMIT_NOFILE].rlim_cur : NOFILE);
2828 size_t maxfa_size = PSF_ACTIONS_SIZE(maxfa);
2829 if (px_args.file_actions_size < PSF_ACTIONS_SIZE(1) ||
2830 maxfa_size == 0 || px_args.file_actions_size > maxfa_size) {
2831 error = EINVAL;
2832 goto bad;
2833 }
2834 MALLOC(px_sfap, _posix_spawn_file_actions_t, px_args.file_actions_size, M_TEMP, M_WAITOK);
2835 if (px_sfap == NULL) {
2836 error = ENOMEM;
2837 goto bad;
2838 }
2839 imgp->ip_px_sfa = px_sfap;
2840
2841 if ((error = copyin(px_args.file_actions, px_sfap,
2842 px_args.file_actions_size)) != 0) {
2843 goto bad;
2844 }
2845
2846 /* Verify that the action count matches the struct size */
2847 size_t psfsize = PSF_ACTIONS_SIZE(px_sfap->psfa_act_count);
2848 if (psfsize == 0 || psfsize != px_args.file_actions_size) {
2849 error = EINVAL;
2850 goto bad;
2851 }
2852 }
2853 if (px_args.port_actions_size != 0) {
2854 /* Limit port_actions to one page of data */
2855 if (px_args.port_actions_size < PS_PORT_ACTIONS_SIZE(1) ||
2856 px_args.port_actions_size > PAGE_SIZE) {
2857 error = EINVAL;
2858 goto bad;
2859 }
2860
2861 MALLOC(px_spap, _posix_spawn_port_actions_t,
2862 px_args.port_actions_size, M_TEMP, M_WAITOK);
2863 if (px_spap == NULL) {
2864 error = ENOMEM;
2865 goto bad;
2866 }
2867 imgp->ip_px_spa = px_spap;
2868
2869 if ((error = copyin(px_args.port_actions, px_spap,
2870 px_args.port_actions_size)) != 0) {
2871 goto bad;
2872 }
2873
2874 /* Verify that the action count matches the struct size */
2875 size_t pasize = PS_PORT_ACTIONS_SIZE(px_spap->pspa_count);
2876 if (pasize == 0 || pasize != px_args.port_actions_size) {
2877 error = EINVAL;
2878 goto bad;
2879 }
2880 }
2881 #if CONFIG_PERSONAS
2882 /* copy in the persona info */
2883 if (px_args.persona_info_size != 0 && px_args.persona_info != 0) {
2884 /* for now, we need the exact same struct in user space */
2885 if (px_args.persona_info_size != sizeof(*px_persona)) {
2886 error = ERANGE;
2887 goto bad;
2888 }
2889
2890 MALLOC(px_persona, struct _posix_spawn_persona_info *, px_args.persona_info_size, M_TEMP, M_WAITOK | M_ZERO);
2891 if (px_persona == NULL) {
2892 error = ENOMEM;
2893 goto bad;
2894 }
2895 imgp->ip_px_persona = px_persona;
2896
2897 if ((error = copyin(px_args.persona_info, px_persona,
2898 px_args.persona_info_size)) != 0) {
2899 goto bad;
2900 }
2901 if ((error = spawn_validate_persona(px_persona)) != 0) {
2902 goto bad;
2903 }
2904 }
2905 #endif
2906 /* copy in the posix cred info */
2907 if (px_args.posix_cred_info_size != 0 && px_args.posix_cred_info != 0) {
2908 /* for now, we need the exact same struct in user space */
2909 if (px_args.posix_cred_info_size != sizeof(*px_pcred_info)) {
2910 error = ERANGE;
2911 goto bad;
2912 }
2913
2914 if (!kauth_cred_issuser(kauth_cred_get())) {
2915 error = EPERM;
2916 goto bad;
2917 }
2918
2919 MALLOC(px_pcred_info, struct _posix_spawn_posix_cred_info *,
2920 px_args.posix_cred_info_size, M_TEMP, M_WAITOK | M_ZERO);
2921 if (px_pcred_info == NULL) {
2922 error = ENOMEM;
2923 goto bad;
2924 }
2925 imgp->ip_px_pcred_info = px_pcred_info;
2926
2927 if ((error = copyin(px_args.posix_cred_info, px_pcred_info,
2928 px_args.posix_cred_info_size)) != 0) {
2929 goto bad;
2930 }
2931
2932 if (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_GROUPS) {
2933 if (px_pcred_info->pspci_ngroups > NGROUPS_MAX) {
2934 error = EINVAL;
2935 goto bad;
2936 }
2937 }
2938 }
2939 #if CONFIG_MACF
2940 if (px_args.mac_extensions_size != 0) {
2941 if ((error = spawn_copyin_macpolicyinfo(&px_args, (_posix_spawn_mac_policy_extensions_t *)&imgp->ip_px_smpx)) != 0) {
2942 goto bad;
2943 }
2944 }
2945 #endif /* CONFIG_MACF */
2946 }
2947
2948 /* set uthread to parent */
2949 uthread = get_bsdthread_info(current_thread());
2950
2951 /*
2952 * <rdar://6640530>; this does not result in a behaviour change
2953 * relative to Leopard, so there should not be any existing code
2954 * which depends on it.
2955 */
2956 if (uthread->uu_flag & UT_VFORK) {
2957 error = EINVAL;
2958 goto bad;
2959 }
2960
2961 if (imgp->ip_px_sa != NULL) {
2962 struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
2963 if ((error = exec_validate_spawnattr_policy(psa->psa_apptype)) != 0) {
2964 goto bad;
2965 }
2966 }
2967
2968 /*
2969 * If we don't have the extension flag that turns "posix_spawn()"
2970 * into "execve() with options", then we will be creating a new
2971 * process which does not inherit memory from the parent process,
2972 * which is one of the most expensive things about using fork()
2973 * and execve().
2974 */
2975 if (imgp->ip_px_sa == NULL || !(px_sa.psa_flags & POSIX_SPAWN_SETEXEC)) {
2976 /* Set the new task's coalition, if it is requested. */
2977 coalition_t coal[COALITION_NUM_TYPES] = { COALITION_NULL };
2978 #if CONFIG_COALITIONS
2979 int i, ncoals;
2980 kern_return_t kr = KERN_SUCCESS;
2981 struct _posix_spawn_coalition_info coal_info;
2982 int coal_role[COALITION_NUM_TYPES];
2983
2984 if (imgp->ip_px_sa == NULL || !px_args.coal_info) {
2985 goto do_fork1;
2986 }
2987
2988 memset(&coal_info, 0, sizeof(coal_info));
2989
2990 if (px_args.coal_info_size > sizeof(coal_info)) {
2991 px_args.coal_info_size = sizeof(coal_info);
2992 }
2993 error = copyin(px_args.coal_info,
2994 &coal_info, px_args.coal_info_size);
2995 if (error != 0) {
2996 goto bad;
2997 }
2998
2999 ncoals = 0;
3000 for (i = 0; i < COALITION_NUM_TYPES; i++) {
3001 uint64_t cid = coal_info.psci_info[i].psci_id;
3002 if (cid != 0) {
3003 /*
3004 * don't allow tasks which are not in a
3005 * privileged coalition to spawn processes
3006 * into coalitions other than their own
3007 */
3008 if (!task_is_in_privileged_coalition(p->task, i)) {
3009 coal_dbg("ERROR: %d not in privilegd "
3010 "coalition of type %d",
3011 p->p_pid, i);
3012 spawn_coalitions_release_all(coal);
3013 error = EPERM;
3014 goto bad;
3015 }
3016
3017 coal_dbg("searching for coalition id:%llu", cid);
3018 /*
3019 * take a reference and activation on the
3020 * coalition to guard against free-while-spawn
3021 * races
3022 */
3023 coal[i] = coalition_find_and_activate_by_id(cid);
3024 if (coal[i] == COALITION_NULL) {
3025 coal_dbg("could not find coalition id:%llu "
3026 "(perhaps it has been terminated or reaped)", cid);
3027 /*
3028 * release any other coalition's we
3029 * may have a reference to
3030 */
3031 spawn_coalitions_release_all(coal);
3032 error = ESRCH;
3033 goto bad;
3034 }
3035 if (coalition_type(coal[i]) != i) {
3036 coal_dbg("coalition with id:%lld is not of type:%d"
3037 " (it's type:%d)", cid, i, coalition_type(coal[i]));
3038 error = ESRCH;
3039 goto bad;
3040 }
3041 coal_role[i] = coal_info.psci_info[i].psci_role;
3042 ncoals++;
3043 }
3044 }
3045 if (ncoals < COALITION_NUM_TYPES) {
3046 /*
3047 * If the user is attempting to spawn into a subset of
3048 * the known coalition types, then make sure they have
3049 * _at_least_ specified a resource coalition. If not,
3050 * the following fork1() call will implicitly force an
3051 * inheritance from 'p' and won't actually spawn the
3052 * new task into the coalitions the user specified.
3053 * (also the call to coalitions_set_roles will panic)
3054 */
3055 if (coal[COALITION_TYPE_RESOURCE] == COALITION_NULL) {
3056 spawn_coalitions_release_all(coal);
3057 error = EINVAL;
3058 goto bad;
3059 }
3060 }
3061 do_fork1:
3062 #endif /* CONFIG_COALITIONS */
3063
3064 /*
3065 * note that this will implicitly inherit the
3066 * caller's persona (if it exists)
3067 */
3068 error = fork1(p, &imgp->ip_new_thread, PROC_CREATE_SPAWN, coal);
3069 /* returns a thread and task reference */
3070
3071 if (error == 0) {
3072 new_task = get_threadtask(imgp->ip_new_thread);
3073 }
3074 #if CONFIG_COALITIONS
3075 /* set the roles of this task within each given coalition */
3076 if (error == 0) {
3077 kr = coalitions_set_roles(coal, new_task, coal_role);
3078 if (kr != KERN_SUCCESS) {
3079 error = EINVAL;
3080 }
3081 if (kdebug_debugid_enabled(MACHDBG_CODE(DBG_MACH_COALITION,
3082 MACH_COALITION_ADOPT))) {
3083 for (i = 0; i < COALITION_NUM_TYPES; i++) {
3084 if (coal[i] != COALITION_NULL) {
3085 /*
3086 * On 32-bit targets, uniqueid
3087 * will get truncated to 32 bits
3088 */
3089 KDBG_RELEASE(MACHDBG_CODE(
3090 DBG_MACH_COALITION,
3091 MACH_COALITION_ADOPT),
3092 coalition_id(coal[i]),
3093 get_task_uniqueid(new_task));
3094 }
3095 }
3096 }
3097 }
3098
3099 /* drop our references and activations - fork1() now holds them */
3100 spawn_coalitions_release_all(coal);
3101 #endif /* CONFIG_COALITIONS */
3102 if (error != 0) {
3103 goto bad;
3104 }
3105 imgp->ip_flags |= IMGPF_SPAWN; /* spawn w/o exec */
3106 spawn_no_exec = TRUE; /* used in later tests */
3107 } else {
3108 /*
3109 * For execve case, create a new task and thread
3110 * which points to current_proc. The current_proc will point
3111 * to the new task after image activation and proc ref drain.
3112 *
3113 * proc (current_proc) <----- old_task (current_task)
3114 * ^ | ^
3115 * | | |
3116 * | ----------------------------------
3117 * |
3118 * --------- new_task (task marked as TF_EXEC_COPY)
3119 *
3120 * After image activation, the proc will point to the new task
3121 * and would look like following.
3122 *
3123 * proc (current_proc) <----- old_task (current_task, marked as TPF_DID_EXEC)
3124 * ^ |
3125 * | |
3126 * | ----------> new_task
3127 * | |
3128 * -----------------
3129 *
3130 * During exec any transition from new_task -> proc is fine, but don't allow
3131 * transition from proc->task, since it will modify old_task.
3132 */
3133 imgp->ip_new_thread = fork_create_child(old_task,
3134 NULL,
3135 p,
3136 FALSE,
3137 p->p_flag & P_LP64,
3138 task_get_64bit_data(old_task),
3139 TRUE);
3140 /* task and thread ref returned by fork_create_child */
3141 if (imgp->ip_new_thread == NULL) {
3142 error = ENOMEM;
3143 goto bad;
3144 }
3145
3146 new_task = get_threadtask(imgp->ip_new_thread);
3147 imgp->ip_flags |= IMGPF_EXEC;
3148 }
3149
3150 if (spawn_no_exec) {
3151 p = (proc_t)get_bsdthreadtask_info(imgp->ip_new_thread);
3152
3153 /*
3154 * We had to wait until this point before firing the
3155 * proc:::create probe, otherwise p would not point to the
3156 * child process.
3157 */
3158 DTRACE_PROC1(create, proc_t, p);
3159 }
3160 assert(p != NULL);
3161
3162 context.vc_thread = imgp->ip_new_thread;
3163 context.vc_ucred = p->p_ucred; /* XXX must NOT be kauth_cred_get() */
3164
3165 /*
3166 * Post fdcopy(), pre exec_handle_sugid() - this is where we want
3167 * to handle the file_actions. Since vfork() also ends up setting
3168 * us into the parent process group, and saved off the signal flags,
3169 * this is also where we want to handle the spawn flags.
3170 */
3171
3172 /* Has spawn file actions? */
3173 if (imgp->ip_px_sfa != NULL) {
3174 /*
3175 * The POSIX_SPAWN_CLOEXEC_DEFAULT flag
3176 * is handled in exec_handle_file_actions().
3177 */
3178 #if CONFIG_AUDIT
3179 /*
3180 * The file actions auditing can overwrite the upath of
3181 * AUE_POSIX_SPAWN audit record. Save the audit record.
3182 */
3183 struct kaudit_record *save_uu_ar = uthread->uu_ar;
3184 uthread->uu_ar = NULL;
3185 #endif
3186 error = exec_handle_file_actions(imgp,
3187 imgp->ip_px_sa != NULL ? px_sa.psa_flags : 0);
3188 #if CONFIG_AUDIT
3189 /* Restore the AUE_POSIX_SPAWN audit record. */
3190 uthread->uu_ar = save_uu_ar;
3191 #endif
3192 if (error != 0) {
3193 goto bad;
3194 }
3195 }
3196
3197 /* Has spawn port actions? */
3198 if (imgp->ip_px_spa != NULL) {
3199 #if CONFIG_AUDIT
3200 /*
3201 * Do the same for the port actions as we did for the file
3202 * actions. Save the AUE_POSIX_SPAWN audit record.
3203 */
3204 struct kaudit_record *save_uu_ar = uthread->uu_ar;
3205 uthread->uu_ar = NULL;
3206 #endif
3207 error = exec_handle_port_actions(imgp, &port_actions);
3208 #if CONFIG_AUDIT
3209 /* Restore the AUE_POSIX_SPAWN audit record. */
3210 uthread->uu_ar = save_uu_ar;
3211 #endif
3212 if (error != 0) {
3213 goto bad;
3214 }
3215 }
3216
3217 /* Has spawn attr? */
3218 if (imgp->ip_px_sa != NULL) {
3219 /*
3220 * Reset UID/GID to parent's RUID/RGID; This works only
3221 * because the operation occurs *after* the vfork() and
3222 * before the call to exec_handle_sugid() by the image
3223 * activator called from exec_activate_image(). POSIX
3224 * requires that any setuid/setgid bits on the process
3225 * image will take precedence over the spawn attributes
3226 * (re)setting them.
3227 *
3228 * Modifications to p_ucred must be guarded using the
3229 * proc's ucred lock. This prevents others from accessing
3230 * a garbage credential.
3231 */
3232 if (px_sa.psa_flags & POSIX_SPAWN_RESETIDS) {
3233 apply_kauth_cred_update(p, ^kauth_cred_t (kauth_cred_t my_cred){
3234 return kauth_cred_setuidgid(my_cred,
3235 kauth_cred_getruid(my_cred),
3236 kauth_cred_getrgid(my_cred));
3237 });
3238 }
3239
3240 if (imgp->ip_px_pcred_info) {
3241 if (!spawn_no_exec) {
3242 error = ENOTSUP;
3243 goto bad;
3244 }
3245
3246 error = spawn_posix_cred_adopt(p, imgp->ip_px_pcred_info);
3247 if (error != 0) {
3248 goto bad;
3249 }
3250 }
3251
3252 #if CONFIG_PERSONAS
3253 if (imgp->ip_px_persona != NULL) {
3254 if (!spawn_no_exec) {
3255 error = ENOTSUP;
3256 goto bad;
3257 }
3258
3259 /*
3260 * If we were asked to spawn a process into a new persona,
3261 * do the credential switch now (which may override the UID/GID
3262 * inherit done just above). It's important to do this switch
3263 * before image activation both for reasons stated above, and
3264 * to ensure that the new persona has access to the image/file
3265 * being executed.
3266 */
3267 error = spawn_persona_adopt(p, imgp->ip_px_persona);
3268 if (error != 0) {
3269 goto bad;
3270 }
3271 }
3272 #endif /* CONFIG_PERSONAS */
3273 #if !SECURE_KERNEL
3274 /*
3275 * Disable ASLR for the spawned process.
3276 *
3277 * But only do so if we are not embedded + RELEASE.
3278 * While embedded allows for a boot-arg (-disable_aslr)
3279 * to deal with this (which itself is only honored on
3280 * DEVELOPMENT or DEBUG builds of xnu), it is often
3281 * useful or necessary to disable ASLR on a per-process
3282 * basis for unit testing and debugging.
3283 */
3284 if (px_sa.psa_flags & _POSIX_SPAWN_DISABLE_ASLR) {
3285 OSBitOrAtomic(P_DISABLE_ASLR, &p->p_flag);
3286 }
3287 #endif /* !SECURE_KERNEL */
3288
3289 /* Randomize high bits of ASLR slide */
3290 if (px_sa.psa_flags & _POSIX_SPAWN_HIGH_BITS_ASLR) {
3291 imgp->ip_flags |= IMGPF_HIGH_BITS_ASLR;
3292 }
3293
3294 #if !SECURE_KERNEL
3295 /*
3296 * Forcibly disallow execution from data pages for the spawned process
3297 * even if it would otherwise be permitted by the architecture default.
3298 */
3299 if (px_sa.psa_flags & _POSIX_SPAWN_ALLOW_DATA_EXEC) {
3300 imgp->ip_flags |= IMGPF_ALLOW_DATA_EXEC;
3301 }
3302 #endif /* !SECURE_KERNEL */
3303
3304 if ((px_sa.psa_apptype & POSIX_SPAWN_PROC_TYPE_MASK) ==
3305 POSIX_SPAWN_PROC_TYPE_DRIVER) {
3306 imgp->ip_flags |= IMGPF_DRIVER;
3307 }
3308 }
3309
3310 /*
3311 * Disable ASLR during image activation. This occurs either if the
3312 * _POSIX_SPAWN_DISABLE_ASLR attribute was found above or if
3313 * P_DISABLE_ASLR was inherited from the parent process.
3314 */
3315 if (p->p_flag & P_DISABLE_ASLR) {
3316 imgp->ip_flags |= IMGPF_DISABLE_ASLR;
3317 }
3318
3319 /*
3320 * Clear transition flag so we won't hang if exec_activate_image() causes
3321 * an automount (and launchd does a proc sysctl to service it).
3322 *
3323 * <rdar://problem/6848672>, <rdar://problem/5959568>.
3324 */
3325 if (spawn_no_exec) {
3326 proc_transend(p, 0);
3327 proc_transit_set = 0;
3328 }
3329
3330 #if MAC_SPAWN /* XXX */
3331 if (uap->mac_p != USER_ADDR_NULL) {
3332 error = mac_execve_enter(uap->mac_p, imgp);
3333 if (error) {
3334 goto bad;
3335 }
3336 }
3337 #endif
3338
3339 /*
3340 * Activate the image
3341 */
3342 error = exec_activate_image(imgp);
3343 #if defined(HAS_APPLE_PAC)
3344 ml_task_set_disable_user_jop(new_task, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
3345 ml_thread_set_disable_user_jop(imgp->ip_new_thread, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
3346 #endif
3347
3348 if (error == 0 && !spawn_no_exec) {
3349 p = proc_exec_switch_task(p, old_task, new_task, imgp->ip_new_thread);
3350 /* proc ref returned */
3351 should_release_proc_ref = TRUE;
3352
3353 /*
3354 * Need to transfer pending watch port boosts to the new task while still making
3355 * sure that the old task remains in the importance linkage. Create an importance
3356 * linkage from old task to new task, then switch the task importance base
3357 * of old task and new task. After the switch the port watch boost will be
3358 * boosting the new task and new task will be donating importance to old task.
3359 */
3360 inherit = ipc_importance_exec_switch_task(old_task, new_task);
3361 }
3362
3363 if (error == 0) {
3364 /* process completed the exec */
3365 exec_done = TRUE;
3366 } else if (error == -1) {
3367 /* Image not claimed by any activator? */
3368 error = ENOEXEC;
3369 }
3370
3371 if (!error && imgp->ip_px_sa != NULL) {
3372 thread_t child_thread = imgp->ip_new_thread;
3373 uthread_t child_uthread = get_bsdthread_info(child_thread);
3374
3375 /*
3376 * Because of POSIX_SPAWN_SETEXEC, we need to handle this after image
3377 * activation, else when image activation fails (before the point of no
3378 * return) would leave the parent process in a modified state.
3379 */
3380 if (px_sa.psa_flags & POSIX_SPAWN_SETPGROUP) {
3381 struct setpgid_args spga;
3382 spga.pid = p->p_pid;
3383 spga.pgid = px_sa.psa_pgroup;
3384 /*
3385 * Effectively, call setpgid() system call; works
3386 * because there are no pointer arguments.
3387 */
3388 if ((error = setpgid(p, &spga, ival)) != 0) {
3389 goto bad;
3390 }
3391 }
3392
3393 if (px_sa.psa_flags & POSIX_SPAWN_SETSID) {
3394 error = setsid_internal(p);
3395 if (error != 0) {
3396 goto bad;
3397 }
3398 }
3399
3400 /*
3401 * If we have a spawn attr, and it contains signal related flags,
3402 * the we need to process them in the "context" of the new child
3403 * process, so we have to process it following image activation,
3404 * prior to making the thread runnable in user space. This is
3405 * necessitated by some signal information being per-thread rather
3406 * than per-process, and we don't have the new allocation in hand
3407 * until after the image is activated.
3408 */
3409
3410 /*
3411 * Mask a list of signals, instead of them being unmasked, if
3412 * they were unmasked in the parent; note that some signals
3413 * are not maskable.
3414 */
3415 if (px_sa.psa_flags & POSIX_SPAWN_SETSIGMASK) {
3416 child_uthread->uu_sigmask = (px_sa.psa_sigmask & ~sigcantmask);
3417 }
3418 /*
3419 * Default a list of signals instead of ignoring them, if
3420 * they were ignored in the parent. Note that we pass
3421 * spawn_no_exec to setsigvec() to indicate that we called
3422 * fork1() and therefore do not need to call proc_signalstart()
3423 * internally.
3424 */
3425 if (px_sa.psa_flags & POSIX_SPAWN_SETSIGDEF) {
3426 vec.sa_handler = SIG_DFL;
3427 vec.sa_tramp = 0;
3428 vec.sa_mask = 0;
3429 vec.sa_flags = 0;
3430 for (sig = 1; sig < NSIG; sig++) {
3431 if (px_sa.psa_sigdefault & (1 << (sig - 1))) {
3432 error = setsigvec(p, child_thread, sig, &vec, spawn_no_exec);
3433 }
3434 }
3435 }
3436
3437 /*
3438 * Activate the CPU usage monitor, if requested. This is done via a task-wide, per-thread CPU
3439 * usage limit, which will generate a resource exceeded exception if any one thread exceeds the
3440 * limit.
3441 *
3442 * Userland gives us interval in seconds, and the kernel SPI expects nanoseconds.
3443 */
3444 if (px_sa.psa_cpumonitor_percent != 0) {
3445 /*
3446 * Always treat a CPU monitor activation coming from spawn as entitled. Requiring
3447 * an entitlement to configure the monitor a certain way seems silly, since
3448 * whomever is turning it on could just as easily choose not to do so.
3449 */
3450 error = proc_set_task_ruse_cpu(p->task,
3451 TASK_POLICY_RESOURCE_ATTRIBUTE_NOTIFY_EXC,
3452 px_sa.psa_cpumonitor_percent,
3453 px_sa.psa_cpumonitor_interval * NSEC_PER_SEC,
3454 0, TRUE);
3455 }
3456
3457
3458 if (px_pcred_info &&
3459 (px_pcred_info->pspci_flags & POSIX_SPAWN_POSIX_CRED_LOGIN)) {
3460 /*
3461 * setlogin() must happen after setsid()
3462 */
3463 setlogin_internal(p, px_pcred_info->pspci_login);
3464 }
3465 }
3466
3467 bad:
3468
3469 if (error == 0) {
3470 /* reset delay idle sleep status if set */
3471 #if !CONFIG_EMBEDDED
3472 if ((p->p_flag & P_DELAYIDLESLEEP) == P_DELAYIDLESLEEP) {
3473 OSBitAndAtomic(~((uint32_t)P_DELAYIDLESLEEP), &p->p_flag);
3474 }
3475 #endif /* !CONFIG_EMBEDDED */
3476 /* upon successful spawn, re/set the proc control state */
3477 if (imgp->ip_px_sa != NULL) {
3478 switch (px_sa.psa_pcontrol) {
3479 case POSIX_SPAWN_PCONTROL_THROTTLE:
3480 p->p_pcaction = P_PCTHROTTLE;
3481 break;
3482 case POSIX_SPAWN_PCONTROL_SUSPEND:
3483 p->p_pcaction = P_PCSUSP;
3484 break;
3485 case POSIX_SPAWN_PCONTROL_KILL:
3486 p->p_pcaction = P_PCKILL;
3487 break;
3488 case POSIX_SPAWN_PCONTROL_NONE:
3489 default:
3490 p->p_pcaction = 0;
3491 break;
3492 }
3493 ;
3494 }
3495 exec_resettextvp(p, imgp);
3496
3497 #if CONFIG_MEMORYSTATUS
3498 /* Set jetsam priority for DriverKit processes */
3499 if (px_sa.psa_apptype == POSIX_SPAWN_PROC_TYPE_DRIVER) {
3500 px_sa.psa_priority = JETSAM_PRIORITY_DRIVER_APPLE;
3501 }
3502
3503 /* Has jetsam attributes? */
3504 if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_SET)) {
3505 /*
3506 * With 2-level high-water-mark support, POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND is no
3507 * longer relevant, as background limits are described via the inactive limit slots.
3508 *
3509 * That said, however, if the POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND is passed in,
3510 * we attempt to mimic previous behavior by forcing the BG limit data into the
3511 * inactive/non-fatal mode and force the active slots to hold system_wide/fatal mode.
3512 */
3513
3514 if (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_HIWATER_BACKGROUND) {
3515 memorystatus_update(p, px_sa.psa_priority, 0, FALSE, /* assertion priority */
3516 (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
3517 TRUE,
3518 -1, TRUE,
3519 px_sa.psa_memlimit_inactive, FALSE);
3520 } else {
3521 memorystatus_update(p, px_sa.psa_priority, 0, FALSE, /* assertion priority */
3522 (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_USE_EFFECTIVE_PRIORITY),
3523 TRUE,
3524 px_sa.psa_memlimit_active,
3525 (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_ACTIVE_FATAL),
3526 px_sa.psa_memlimit_inactive,
3527 (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_MEMLIMIT_INACTIVE_FATAL));
3528 }
3529 }
3530
3531 /* Has jetsam relaunch behavior? */
3532 if (imgp->ip_px_sa != NULL && (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_MASK)) {
3533 /*
3534 * Launchd has passed in data indicating the behavior of this process in response to jetsam.
3535 * This data would be used by the jetsam subsystem to determine the position and protection
3536 * offered to this process on dirty -> clean transitions.
3537 */
3538 int relaunch_flags = P_MEMSTAT_RELAUNCH_UNKNOWN;
3539 switch (px_sa.psa_jetsam_flags & POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_MASK) {
3540 case POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_LOW:
3541 relaunch_flags = P_MEMSTAT_RELAUNCH_LOW;
3542 break;
3543 case POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_MED:
3544 relaunch_flags = P_MEMSTAT_RELAUNCH_MED;
3545 break;
3546 case POSIX_SPAWN_JETSAM_RELAUNCH_BEHAVIOR_HIGH:
3547 relaunch_flags = P_MEMSTAT_RELAUNCH_HIGH;
3548 break;
3549 default:
3550 break;
3551 }
3552 memorystatus_relaunch_flags_update(p, relaunch_flags);
3553 }
3554
3555 #endif /* CONFIG_MEMORYSTATUS */
3556 if (imgp->ip_px_sa != NULL && px_sa.psa_thread_limit > 0) {
3557 task_set_thread_limit(new_task, (uint16_t)px_sa.psa_thread_limit);
3558 }
3559 }
3560
3561 /*
3562 * If we successfully called fork1(), we always need to do this;
3563 * we identify this case by noting the IMGPF_SPAWN flag. This is
3564 * because we come back from that call with signals blocked in the
3565 * child, and we have to unblock them, but we want to wait until
3566 * after we've performed any spawn actions. This has to happen
3567 * before check_for_signature(), which uses psignal.
3568 */
3569 if (spawn_no_exec) {
3570 if (proc_transit_set) {
3571 proc_transend(p, 0);
3572 }
3573
3574 /*
3575 * Drop the signal lock on the child which was taken on our
3576 * behalf by forkproc()/cloneproc() to prevent signals being
3577 * received by the child in a partially constructed state.
3578 */
3579 proc_signalend(p, 0);
3580
3581 /* flag the 'fork' has occurred */
3582 proc_knote(p->p_pptr, NOTE_FORK | p->p_pid);
3583 }
3584
3585 /* flag exec has occurred, notify only if it has not failed due to FP Key error */
3586 if (!error && ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0)) {
3587 proc_knote(p, NOTE_EXEC);
3588 }
3589
3590
3591 if (error == 0) {
3592 /*
3593 * We need to initialize the bank context behind the protection of
3594 * the proc_trans lock to prevent a race with exit. We can't do this during
3595 * exec_activate_image because task_bank_init checks entitlements that
3596 * aren't loaded until subsequent calls (including exec_resettextvp).
3597 */
3598 error = proc_transstart(p, 0, 0);
3599
3600 if (error == 0) {
3601 task_bank_init(new_task);
3602 proc_transend(p, 0);
3603 }
3604
3605 #if __arm64__
3606 proc_legacy_footprint_entitled(p, new_task, __FUNCTION__);
3607 proc_ios13extended_footprint_entitled(p, new_task, __FUNCTION__);
3608 #endif /* __arm64__ */
3609 }
3610
3611 /* Inherit task role from old task to new task for exec */
3612 if (error == 0 && !spawn_no_exec) {
3613 proc_inherit_task_role(new_task, old_task);
3614 }
3615
3616 #if CONFIG_ARCADE
3617 if (error == 0) {
3618 /*
3619 * Check to see if we need to trigger an arcade upcall AST now
3620 * that the vnode has been reset on the task.
3621 */
3622 arcade_prepare(new_task, imgp->ip_new_thread);
3623 }
3624 #endif /* CONFIG_ARCADE */
3625
3626 /* Clear the initial wait on the thread before handling spawn policy */
3627 if (imgp && imgp->ip_new_thread) {
3628 task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_INITIAL_WAIT);
3629 }
3630
3631 /*
3632 * Apply the spawnattr policy, apptype (which primes the task for importance donation),
3633 * and bind any portwatch ports to the new task.
3634 * This must be done after the exec so that the child's thread is ready,
3635 * and after the in transit state has been released, because priority is
3636 * dropped here so we need to be prepared for a potentially long preemption interval
3637 *
3638 * TODO: Consider splitting this up into separate phases
3639 */
3640 if (error == 0 && imgp->ip_px_sa != NULL) {
3641 struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
3642
3643 error = exec_handle_spawnattr_policy(p, imgp->ip_new_thread, psa->psa_apptype, psa->psa_qos_clamp,
3644 psa->psa_darwin_role, &port_actions);
3645 }
3646
3647 /* Transfer the turnstile watchport boost to new task if in exec */
3648 if (error == 0 && !spawn_no_exec) {
3649 task_transfer_turnstile_watchports(old_task, new_task, imgp->ip_new_thread);
3650 }
3651
3652 /*
3653 * Apply the requested maximum address.
3654 */
3655 if (error == 0 && imgp->ip_px_sa != NULL) {
3656 struct _posix_spawnattr *psa = (struct _posix_spawnattr *) imgp->ip_px_sa;
3657
3658 if (psa->psa_max_addr) {
3659 vm_map_set_max_addr(get_task_map(new_task), psa->psa_max_addr);
3660 }
3661 }
3662
3663 if (error == 0) {
3664 /* Apply the main thread qos */
3665 thread_t main_thread = imgp->ip_new_thread;
3666 task_set_main_thread_qos(new_task, main_thread);
3667
3668 #if CONFIG_MACF
3669 /*
3670 * Processes with the MAP_JIT entitlement are permitted to have
3671 * a jumbo-size map.
3672 */
3673 if (mac_proc_check_map_anon(p, 0, 0, 0, MAP_JIT, NULL) == 0) {
3674 vm_map_set_jumbo(get_task_map(new_task));
3675 vm_map_set_jit_entitled(get_task_map(new_task));
3676 }
3677 #endif /* CONFIG_MACF */
3678 }
3679
3680 /*
3681 * Release any ports we kept around for binding to the new task
3682 * We need to release the rights even if the posix_spawn has failed.
3683 */
3684 if (imgp->ip_px_spa != NULL) {
3685 exec_port_actions_destroy(&port_actions);
3686 }
3687
3688 /*
3689 * We have to delay operations which might throw a signal until after
3690 * the signals have been unblocked; however, we want that to happen
3691 * after exec_resettextvp() so that the textvp is correct when they
3692 * fire.
3693 */
3694 if (error == 0) {
3695 error = check_for_signature(p, imgp);
3696
3697 /*
3698 * Pay for our earlier safety; deliver the delayed signals from
3699 * the incomplete spawn process now that it's complete.
3700 */
3701 if (imgp != NULL && spawn_no_exec && (p->p_lflag & P_LTRACED)) {
3702 psignal_vfork(p, p->task, imgp->ip_new_thread, SIGTRAP);
3703 }
3704
3705 if (error == 0 && !spawn_no_exec) {
3706 KDBG(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXEC),
3707 p->p_pid);
3708 }
3709 }
3710
3711
3712 if (imgp != NULL) {
3713 if (imgp->ip_vp) {
3714 vnode_put(imgp->ip_vp);
3715 }
3716 if (imgp->ip_scriptvp) {
3717 vnode_put(imgp->ip_scriptvp);
3718 }
3719 if (imgp->ip_strings) {
3720 execargs_free(imgp);
3721 }
3722 if (imgp->ip_px_sfa != NULL) {
3723 FREE(imgp->ip_px_sfa, M_TEMP);
3724 }
3725 if (imgp->ip_px_spa != NULL) {
3726 FREE(imgp->ip_px_spa, M_TEMP);
3727 }
3728 #if CONFIG_PERSONAS
3729 if (imgp->ip_px_persona != NULL) {
3730 FREE(imgp->ip_px_persona, M_TEMP);
3731 }
3732 #endif
3733 if (imgp->ip_px_pcred_info != NULL) {
3734 FREE(imgp->ip_px_pcred_info, M_TEMP);
3735 }
3736 #if CONFIG_MACF
3737 if (imgp->ip_px_smpx != NULL) {
3738 spawn_free_macpolicyinfo(imgp->ip_px_smpx);
3739 }
3740 if (imgp->ip_execlabelp) {
3741 mac_cred_label_free(imgp->ip_execlabelp);
3742 }
3743 if (imgp->ip_scriptlabelp) {
3744 mac_vnode_label_free(imgp->ip_scriptlabelp);
3745 }
3746 if (imgp->ip_cs_error != OS_REASON_NULL) {
3747 os_reason_free(imgp->ip_cs_error);
3748 imgp->ip_cs_error = OS_REASON_NULL;
3749 }
3750 #endif
3751 }
3752
3753 #if CONFIG_DTRACE
3754 if (spawn_no_exec) {
3755 /*
3756 * In the original DTrace reference implementation,
3757 * posix_spawn() was a libc routine that just
3758 * did vfork(2) then exec(2). Thus the proc::: probes
3759 * are very fork/exec oriented. The details of this
3760 * in-kernel implementation of posix_spawn() is different
3761 * (while producing the same process-observable effects)
3762 * particularly w.r.t. errors, and which thread/process
3763 * is constructing what on behalf of whom.
3764 */
3765 if (error) {
3766 DTRACE_PROC1(spawn__failure, int, error);
3767 } else {
3768 DTRACE_PROC(spawn__success);
3769 /*
3770 * Some DTrace scripts, e.g. newproc.d in
3771 * /usr/bin, rely on the the 'exec-success'
3772 * probe being fired in the child after the
3773 * new process image has been constructed
3774 * in order to determine the associated pid.
3775 *
3776 * So, even though the parent built the image
3777 * here, for compatibility, mark the new thread
3778 * so 'exec-success' fires on it as it leaves
3779 * the kernel.
3780 */
3781 dtrace_thread_didexec(imgp->ip_new_thread);
3782 }
3783 } else {
3784 if (error) {
3785 DTRACE_PROC1(exec__failure, int, error);
3786 } else {
3787 dtrace_thread_didexec(imgp->ip_new_thread);
3788 }
3789 }
3790
3791 if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
3792 (*dtrace_proc_waitfor_hook)(p);
3793 }
3794 #endif
3795
3796 #if CONFIG_AUDIT
3797 if (!error && AUDIT_ENABLED() && p) {
3798 /* Add the CDHash of the new process to the audit record */
3799 uint8_t *cdhash = cs_get_cdhash(p);
3800 if (cdhash) {
3801 AUDIT_ARG(data, cdhash, sizeof(uint8_t), CS_CDHASH_LEN);
3802 }
3803 }
3804 #endif
3805
3806 /*
3807 * clear bsd_info from old task if it did exec.
3808 */
3809 if (task_did_exec(old_task)) {
3810 set_bsdtask_info(old_task, NULL);
3811 }
3812
3813 /* clear bsd_info from new task and terminate it if exec failed */
3814 if (new_task != NULL && task_is_exec_copy(new_task)) {
3815 set_bsdtask_info(new_task, NULL);
3816 task_terminate_internal(new_task);
3817 }
3818
3819 /* Return to both the parent and the child? */
3820 if (imgp != NULL && spawn_no_exec) {
3821 /*
3822 * If the parent wants the pid, copy it out
3823 */
3824 if (pid != USER_ADDR_NULL) {
3825 _Static_assert(sizeof(p->p_pid) == 4, "posix_spawn() assumes a 32-bit pid_t");
3826 bool aligned = (pid & 3) == 0;
3827 if (aligned) {
3828 (void)copyout_atomic32(p->p_pid, pid);
3829 } else {
3830 (void)suword(pid, p->p_pid);
3831 }
3832 }
3833 retval[0] = error;
3834
3835 /*
3836 * If we had an error, perform an internal reap ; this is
3837 * entirely safe, as we have a real process backing us.
3838 */
3839 if (error) {
3840 proc_list_lock();
3841 p->p_listflag |= P_LIST_DEADPARENT;
3842 proc_list_unlock();
3843 proc_lock(p);
3844 /* make sure no one else has killed it off... */
3845 if (p->p_stat != SZOMB && p->exit_thread == NULL) {
3846 p->exit_thread = current_thread();
3847 proc_unlock(p);
3848 exit1(p, 1, (int *)NULL);
3849 } else {
3850 /* someone is doing it for us; just skip it */
3851 proc_unlock(p);
3852 }
3853 }
3854 }
3855
3856 /*
3857 * Do not terminate the current task, if proc_exec_switch_task did not
3858 * switch the tasks, terminating the current task without the switch would
3859 * result in loosing the SIGKILL status.
3860 */
3861 if (task_did_exec(old_task)) {
3862 /* Terminate the current task, since exec will start in new task */
3863 task_terminate_internal(old_task);
3864 }
3865
3866 /* Release the thread ref returned by fork_create_child/fork1 */
3867 if (imgp != NULL && imgp->ip_new_thread) {
3868 /* wake up the new thread */
3869 task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_FINAL_WAIT);
3870 thread_deallocate(imgp->ip_new_thread);
3871 imgp->ip_new_thread = NULL;
3872 }
3873
3874 /* Release the ref returned by fork_create_child/fork1 */
3875 if (new_task) {
3876 task_deallocate(new_task);
3877 new_task = NULL;
3878 }
3879
3880 if (should_release_proc_ref) {
3881 proc_rele(p);
3882 }
3883
3884 if (bufp != NULL) {
3885 FREE(bufp, M_TEMP);
3886 }
3887
3888 if (inherit != NULL) {
3889 ipc_importance_release(inherit);
3890 }
3891
3892 return error;
3893 }
3894
3895 /*
3896 * proc_exec_switch_task
3897 *
3898 * Parameters: p proc
3899 * old_task task before exec
3900 * new_task task after exec
3901 * new_thread thread in new task
3902 *
3903 * Returns: proc.
3904 *
3905 * Note: The function will switch the task pointer of proc
3906 * from old task to new task. The switch needs to happen
3907 * after draining all proc refs and inside a proc translock.
3908 * In the case of failure to switch the task, which might happen
3909 * if the process received a SIGKILL or jetsam killed it, it will make
3910 * sure that the new tasks terminates. User proc ref returned
3911 * to caller.
3912 *
3913 * This function is called after point of no return, in the case
3914 * failure to switch, it will terminate the new task and swallow the
3915 * error and let the terminated process complete exec and die.
3916 */
3917 proc_t
3918 proc_exec_switch_task(proc_t p, task_t old_task, task_t new_task, thread_t new_thread)
3919 {
3920 int error = 0;
3921 boolean_t task_active;
3922 boolean_t proc_active;
3923 boolean_t thread_active;
3924 thread_t old_thread = current_thread();
3925
3926 /*
3927 * Switch the task pointer of proc to new task.
3928 * Before switching the task, wait for proc_refdrain.
3929 * After the switch happens, the proc can disappear,
3930 * take a ref before it disappears. Waiting for
3931 * proc_refdrain in exec will block all other threads
3932 * trying to take a proc ref, boost the current thread
3933 * to avoid priority inversion.
3934 */
3935 thread_set_exec_promotion(old_thread);
3936 p = proc_refdrain_with_refwait(p, TRUE);
3937 /* extra proc ref returned to the caller */
3938
3939 assert(get_threadtask(new_thread) == new_task);
3940 task_active = task_is_active(new_task);
3941
3942 /* Take the proc_translock to change the task ptr */
3943 proc_lock(p);
3944 proc_active = !(p->p_lflag & P_LEXIT);
3945
3946 /* Check if the current thread is not aborted due to SIGKILL */
3947 thread_active = thread_is_active(old_thread);
3948
3949 /*
3950 * Do not switch the task if the new task or proc is already terminated
3951 * as a result of error in exec past point of no return
3952 */
3953 if (proc_active && task_active && thread_active) {
3954 error = proc_transstart(p, 1, 0);
3955 if (error == 0) {
3956 uthread_t new_uthread = get_bsdthread_info(new_thread);
3957 uthread_t old_uthread = get_bsdthread_info(current_thread());
3958
3959 /*
3960 * bsd_info of old_task will get cleared in execve and posix_spawn
3961 * after firing exec-success/error dtrace probe.
3962 */
3963 p->task = new_task;
3964
3965 /* Clear dispatchqueue and workloop ast offset */
3966 p->p_dispatchqueue_offset = 0;
3967 p->p_dispatchqueue_serialno_offset = 0;
3968 p->p_dispatchqueue_label_offset = 0;
3969 p->p_return_to_kernel_offset = 0;
3970
3971 /* Copy the signal state, dtrace state and set bsd ast on new thread */
3972 act_set_astbsd(new_thread);
3973 new_uthread->uu_siglist = old_uthread->uu_siglist;
3974 new_uthread->uu_sigwait = old_uthread->uu_sigwait;
3975 new_uthread->uu_sigmask = old_uthread->uu_sigmask;
3976 new_uthread->uu_oldmask = old_uthread->uu_oldmask;
3977 new_uthread->uu_vforkmask = old_uthread->uu_vforkmask;
3978 new_uthread->uu_exit_reason = old_uthread->uu_exit_reason;
3979 #if CONFIG_DTRACE
3980 new_uthread->t_dtrace_sig = old_uthread->t_dtrace_sig;
3981 new_uthread->t_dtrace_stop = old_uthread->t_dtrace_stop;
3982 new_uthread->t_dtrace_resumepid = old_uthread->t_dtrace_resumepid;
3983 assert(new_uthread->t_dtrace_scratch == NULL);
3984 new_uthread->t_dtrace_scratch = old_uthread->t_dtrace_scratch;
3985
3986 old_uthread->t_dtrace_sig = 0;
3987 old_uthread->t_dtrace_stop = 0;
3988 old_uthread->t_dtrace_resumepid = 0;
3989 old_uthread->t_dtrace_scratch = NULL;
3990 #endif
3991 /* Copy the resource accounting info */
3992 thread_copy_resource_info(new_thread, current_thread());
3993
3994 /* Clear the exit reason and signal state on old thread */
3995 old_uthread->uu_exit_reason = NULL;
3996 old_uthread->uu_siglist = 0;
3997
3998 /* Add the new uthread to proc uthlist and remove the old one */
3999 TAILQ_INSERT_TAIL(&p->p_uthlist, new_uthread, uu_list);
4000 TAILQ_REMOVE(&p->p_uthlist, old_uthread, uu_list);
4001
4002 task_set_did_exec_flag(old_task);
4003 task_clear_exec_copy_flag(new_task);
4004
4005 task_copy_fields_for_exec(new_task, old_task);
4006
4007 proc_transend(p, 1);
4008 }
4009 }
4010
4011 proc_unlock(p);
4012 proc_refwake(p);
4013 thread_clear_exec_promotion(old_thread);
4014
4015 if (error != 0 || !task_active || !proc_active || !thread_active) {
4016 task_terminate_internal(new_task);
4017 }
4018
4019 return p;
4020 }
4021
4022 /*
4023 * execve
4024 *
4025 * Parameters: uap->fname File name to exec
4026 * uap->argp Argument list
4027 * uap->envp Environment list
4028 *
4029 * Returns: 0 Success
4030 * __mac_execve:EINVAL Invalid argument
4031 * __mac_execve:ENOTSUP Invalid argument
4032 * __mac_execve:EACCES Permission denied
4033 * __mac_execve:EINTR Interrupted function
4034 * __mac_execve:ENOMEM Not enough space
4035 * __mac_execve:EFAULT Bad address
4036 * __mac_execve:ENAMETOOLONG Filename too long
4037 * __mac_execve:ENOEXEC Executable file format error
4038 * __mac_execve:ETXTBSY Text file busy [misuse of error code]
4039 * __mac_execve:???
4040 *
4041 * TODO: Dynamic linker header address on stack is copied via suword()
4042 */
4043 /* ARGSUSED */
4044 int
4045 execve(proc_t p, struct execve_args *uap, int32_t *retval)
4046 {
4047 struct __mac_execve_args muap;
4048 int err;
4049
4050 memoryshot(VM_EXECVE, DBG_FUNC_NONE);
4051
4052 muap.fname = uap->fname;
4053 muap.argp = uap->argp;
4054 muap.envp = uap->envp;
4055 muap.mac_p = USER_ADDR_NULL;
4056 err = __mac_execve(p, &muap, retval);
4057
4058 return err;
4059 }
4060
4061 /*
4062 * __mac_execve
4063 *
4064 * Parameters: uap->fname File name to exec
4065 * uap->argp Argument list
4066 * uap->envp Environment list
4067 * uap->mac_p MAC label supplied by caller
4068 *
4069 * Returns: 0 Success
4070 * EINVAL Invalid argument
4071 * ENOTSUP Not supported
4072 * ENOEXEC Executable file format error
4073 * exec_activate_image:EINVAL Invalid argument
4074 * exec_activate_image:EACCES Permission denied
4075 * exec_activate_image:EINTR Interrupted function
4076 * exec_activate_image:ENOMEM Not enough space
4077 * exec_activate_image:EFAULT Bad address
4078 * exec_activate_image:ENAMETOOLONG Filename too long
4079 * exec_activate_image:ENOEXEC Executable file format error
4080 * exec_activate_image:ETXTBSY Text file busy [misuse of error code]
4081 * exec_activate_image:EBADEXEC The executable is corrupt/unknown
4082 * exec_activate_image:???
4083 * mac_execve_enter:???
4084 *
4085 * TODO: Dynamic linker header address on stack is copied via suword()
4086 */
4087 int
4088 __mac_execve(proc_t p, struct __mac_execve_args *uap, int32_t *retval)
4089 {
4090 char *bufp = NULL;
4091 struct image_params *imgp;
4092 struct vnode_attr *vap;
4093 struct vnode_attr *origvap;
4094 int error;
4095 int is_64 = IS_64BIT_PROCESS(p);
4096 struct vfs_context context;
4097 struct uthread *uthread;
4098 task_t old_task = current_task();
4099 task_t new_task = NULL;
4100 boolean_t should_release_proc_ref = FALSE;
4101 boolean_t exec_done = FALSE;
4102 boolean_t in_vfexec = FALSE;
4103 void *inherit = NULL;
4104
4105 context.vc_thread = current_thread();
4106 context.vc_ucred = kauth_cred_proc_ref(p); /* XXX must NOT be kauth_cred_get() */
4107
4108 /* Allocate a big chunk for locals instead of using stack since these
4109 * structures a pretty big.
4110 */
4111 MALLOC(bufp, char *, (sizeof(*imgp) + sizeof(*vap) + sizeof(*origvap)), M_TEMP, M_WAITOK | M_ZERO);
4112 imgp = (struct image_params *) bufp;
4113 if (bufp == NULL) {
4114 error = ENOMEM;
4115 goto exit_with_error;
4116 }
4117 vap = (struct vnode_attr *) (bufp + sizeof(*imgp));
4118 origvap = (struct vnode_attr *) (bufp + sizeof(*imgp) + sizeof(*vap));
4119
4120 /* Initialize the common data in the image_params structure */
4121 imgp->ip_user_fname = uap->fname;
4122 imgp->ip_user_argv = uap->argp;
4123 imgp->ip_user_envv = uap->envp;
4124 imgp->ip_vattr = vap;
4125 imgp->ip_origvattr = origvap;
4126 imgp->ip_vfs_context = &context;
4127 imgp->ip_flags = (is_64 ? IMGPF_WAS_64BIT_ADDR : IMGPF_NONE) | ((p->p_flag & P_DISABLE_ASLR) ? IMGPF_DISABLE_ASLR : IMGPF_NONE);
4128 imgp->ip_seg = (is_64 ? UIO_USERSPACE64 : UIO_USERSPACE32);
4129 imgp->ip_mac_return = 0;
4130 imgp->ip_cs_error = OS_REASON_NULL;
4131 imgp->ip_simulator_binary = IMGPF_SB_DEFAULT;
4132
4133 #if CONFIG_MACF
4134 if (uap->mac_p != USER_ADDR_NULL) {
4135 error = mac_execve_enter(uap->mac_p, imgp);
4136 if (error) {
4137 kauth_cred_unref(&context.vc_ucred);
4138 goto exit_with_error;
4139 }
4140 }
4141 #endif
4142 uthread = get_bsdthread_info(current_thread());
4143 if (uthread->uu_flag & UT_VFORK) {
4144 imgp->ip_flags |= IMGPF_VFORK_EXEC;
4145 in_vfexec = TRUE;
4146 } else {
4147 imgp->ip_flags |= IMGPF_EXEC;
4148
4149 /*
4150 * For execve case, create a new task and thread
4151 * which points to current_proc. The current_proc will point
4152 * to the new task after image activation and proc ref drain.
4153 *
4154 * proc (current_proc) <----- old_task (current_task)
4155 * ^ | ^
4156 * | | |
4157 * | ----------------------------------
4158 * |
4159 * --------- new_task (task marked as TF_EXEC_COPY)
4160 *
4161 * After image activation, the proc will point to the new task
4162 * and would look like following.
4163 *
4164 * proc (current_proc) <----- old_task (current_task, marked as TPF_DID_EXEC)
4165 * ^ |
4166 * | |
4167 * | ----------> new_task
4168 * | |
4169 * -----------------
4170 *
4171 * During exec any transition from new_task -> proc is fine, but don't allow
4172 * transition from proc->task, since it will modify old_task.
4173 */
4174 imgp->ip_new_thread = fork_create_child(old_task,
4175 NULL,
4176 p,
4177 FALSE,
4178 p->p_flag & P_LP64,
4179 task_get_64bit_data(old_task),
4180 TRUE);
4181 /* task and thread ref returned by fork_create_child */
4182 if (imgp->ip_new_thread == NULL) {
4183 error = ENOMEM;
4184 goto exit_with_error;
4185 }
4186
4187 new_task = get_threadtask(imgp->ip_new_thread);
4188 context.vc_thread = imgp->ip_new_thread;
4189 }
4190
4191 error = exec_activate_image(imgp);
4192 /* thread and task ref returned for vfexec case */
4193
4194 if (imgp->ip_new_thread != NULL) {
4195 /*
4196 * task reference might be returned by exec_activate_image
4197 * for vfexec.
4198 */
4199 new_task = get_threadtask(imgp->ip_new_thread);
4200 #if defined(HAS_APPLE_PAC)
4201 ml_task_set_disable_user_jop(new_task, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
4202 ml_thread_set_disable_user_jop(imgp->ip_new_thread, imgp->ip_flags & IMGPF_NOJOP ? TRUE : FALSE);
4203 #endif
4204 }
4205
4206 if (!error && !in_vfexec) {
4207 p = proc_exec_switch_task(p, old_task, new_task, imgp->ip_new_thread);
4208 /* proc ref returned */
4209 should_release_proc_ref = TRUE;
4210
4211 /*
4212 * Need to transfer pending watch port boosts to the new task while still making
4213 * sure that the old task remains in the importance linkage. Create an importance
4214 * linkage from old task to new task, then switch the task importance base
4215 * of old task and new task. After the switch the port watch boost will be
4216 * boosting the new task and new task will be donating importance to old task.
4217 */
4218 inherit = ipc_importance_exec_switch_task(old_task, new_task);
4219 }
4220
4221 kauth_cred_unref(&context.vc_ucred);
4222
4223 /* Image not claimed by any activator? */
4224 if (error == -1) {
4225 error = ENOEXEC;
4226 }
4227
4228 if (!error) {
4229 exec_done = TRUE;
4230 assert(imgp->ip_new_thread != NULL);
4231
4232 exec_resettextvp(p, imgp);
4233 error = check_for_signature(p, imgp);
4234 }
4235
4236 /* flag exec has occurred, notify only if it has not failed due to FP Key error */
4237 if (exec_done && ((p->p_lflag & P_LTERM_DECRYPTFAIL) == 0)) {
4238 proc_knote(p, NOTE_EXEC);
4239 }
4240
4241 if (imgp->ip_vp != NULLVP) {
4242 vnode_put(imgp->ip_vp);
4243 }
4244 if (imgp->ip_scriptvp != NULLVP) {
4245 vnode_put(imgp->ip_scriptvp);
4246 }
4247 if (imgp->ip_strings) {
4248 execargs_free(imgp);
4249 }
4250 #if CONFIG_MACF
4251 if (imgp->ip_execlabelp) {
4252 mac_cred_label_free(imgp->ip_execlabelp);
4253 }
4254 if (imgp->ip_scriptlabelp) {
4255 mac_vnode_label_free(imgp->ip_scriptlabelp);
4256 }
4257 #endif
4258 if (imgp->ip_cs_error != OS_REASON_NULL) {
4259 os_reason_free(imgp->ip_cs_error);
4260 imgp->ip_cs_error = OS_REASON_NULL;
4261 }
4262
4263 if (!error) {
4264 /*
4265 * We need to initialize the bank context behind the protection of
4266 * the proc_trans lock to prevent a race with exit. We can't do this during
4267 * exec_activate_image because task_bank_init checks entitlements that
4268 * aren't loaded until subsequent calls (including exec_resettextvp).
4269 */
4270 error = proc_transstart(p, 0, 0);
4271 }
4272
4273 if (!error) {
4274 task_bank_init(new_task);
4275 proc_transend(p, 0);
4276
4277 #if __arm64__
4278 proc_legacy_footprint_entitled(p, new_task, __FUNCTION__);
4279 proc_ios13extended_footprint_entitled(p, new_task, __FUNCTION__);
4280 #endif /* __arm64__ */
4281
4282 /* Sever any extant thread affinity */
4283 thread_affinity_exec(current_thread());
4284
4285 /* Inherit task role from old task to new task for exec */
4286 if (!in_vfexec) {
4287 proc_inherit_task_role(new_task, old_task);
4288 }
4289
4290 thread_t main_thread = imgp->ip_new_thread;
4291
4292 task_set_main_thread_qos(new_task, main_thread);
4293
4294 #if CONFIG_ARCADE
4295 /*
4296 * Check to see if we need to trigger an arcade upcall AST now
4297 * that the vnode has been reset on the task.
4298 */
4299 arcade_prepare(new_task, imgp->ip_new_thread);
4300 #endif /* CONFIG_ARCADE */
4301
4302 #if CONFIG_MACF
4303 /*
4304 * Processes with the MAP_JIT entitlement are permitted to have
4305 * a jumbo-size map.
4306 */
4307 if (mac_proc_check_map_anon(p, 0, 0, 0, MAP_JIT, NULL) == 0) {
4308 vm_map_set_jumbo(get_task_map(new_task));
4309 vm_map_set_jit_entitled(get_task_map(new_task));
4310 }
4311 #endif /* CONFIG_MACF */
4312
4313 if (vm_darkwake_mode == TRUE) {
4314 /*
4315 * This process is being launched when the system
4316 * is in darkwake. So mark it specially. This will
4317 * cause all its pages to be entered in the background Q.
4318 */
4319 task_set_darkwake_mode(new_task, vm_darkwake_mode);
4320 }
4321
4322 #if CONFIG_DTRACE
4323 dtrace_thread_didexec(imgp->ip_new_thread);
4324
4325 if ((dtrace_proc_waitfor_hook = dtrace_proc_waitfor_exec_ptr) != NULL) {
4326 (*dtrace_proc_waitfor_hook)(p);
4327 }
4328 #endif
4329
4330 #if CONFIG_AUDIT
4331 if (!error && AUDIT_ENABLED() && p) {
4332 /* Add the CDHash of the new process to the audit record */
4333 uint8_t *cdhash = cs_get_cdhash(p);
4334 if (cdhash) {
4335 AUDIT_ARG(data, cdhash, sizeof(uint8_t), CS_CDHASH_LEN);
4336 }
4337 }
4338 #endif
4339
4340 if (in_vfexec) {
4341 vfork_return(p, retval, p->p_pid);
4342 }
4343 } else {
4344 DTRACE_PROC1(exec__failure, int, error);
4345 }
4346
4347 exit_with_error:
4348
4349 /*
4350 * clear bsd_info from old task if it did exec.
4351 */
4352 if (task_did_exec(old_task)) {
4353 set_bsdtask_info(old_task, NULL);
4354 }
4355
4356 /* clear bsd_info from new task and terminate it if exec failed */
4357 if (new_task != NULL && task_is_exec_copy(new_task)) {
4358 set_bsdtask_info(new_task, NULL);
4359 task_terminate_internal(new_task);
4360 }
4361
4362 if (imgp != NULL) {
4363 /* Clear the initial wait on the thread transferring watchports */
4364 if (imgp->ip_new_thread) {
4365 task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_INITIAL_WAIT);
4366 }
4367
4368 /* Transfer the watchport boost to new task */
4369 if (!error && !in_vfexec) {
4370 task_transfer_turnstile_watchports(old_task,
4371 new_task, imgp->ip_new_thread);
4372 }
4373 /*
4374 * Do not terminate the current task, if proc_exec_switch_task did not
4375 * switch the tasks, terminating the current task without the switch would
4376 * result in loosing the SIGKILL status.
4377 */
4378 if (task_did_exec(old_task)) {
4379 /* Terminate the current task, since exec will start in new task */
4380 task_terminate_internal(old_task);
4381 }
4382
4383 /* Release the thread ref returned by fork_create_child */
4384 if (imgp->ip_new_thread) {
4385 /* wake up the new exec thread */
4386 task_clear_return_wait(get_threadtask(imgp->ip_new_thread), TCRW_CLEAR_FINAL_WAIT);
4387 thread_deallocate(imgp->ip_new_thread);
4388 imgp->ip_new_thread = NULL;
4389 }
4390 }
4391
4392 /* Release the ref returned by fork_create_child */
4393 if (new_task) {
4394 task_deallocate(new_task);
4395 new_task = NULL;
4396 }
4397
4398 if (should_release_proc_ref) {
4399 proc_rele(p);
4400 }
4401
4402 if (bufp != NULL) {
4403 FREE(bufp, M_TEMP);
4404 }
4405
4406 if (inherit != NULL) {
4407 ipc_importance_release(inherit);
4408 }
4409
4410 return error;
4411 }
4412
4413
4414 /*
4415 * copyinptr
4416 *
4417 * Description: Copy a pointer in from user space to a user_addr_t in kernel
4418 * space, based on 32/64 bitness of the user space
4419 *
4420 * Parameters: froma User space address
4421 * toptr Address of kernel space user_addr_t
4422 * ptr_size 4/8, based on 'froma' address space
4423 *
4424 * Returns: 0 Success
4425 * EFAULT Bad 'froma'
4426 *
4427 * Implicit returns:
4428 * *ptr_size Modified
4429 */
4430 static int
4431 copyinptr(user_addr_t froma, user_addr_t *toptr, int ptr_size)
4432 {
4433 int error;
4434
4435 if (ptr_size == 4) {
4436 /* 64 bit value containing 32 bit address */
4437 unsigned int i = 0;
4438
4439 error = copyin(froma, &i, 4);
4440 *toptr = CAST_USER_ADDR_T(i); /* SAFE */
4441 } else {
4442 error = copyin(froma, toptr, 8);
4443 }
4444 return error;
4445 }
4446
4447
4448 /*
4449 * copyoutptr
4450 *
4451 * Description: Copy a pointer out from a user_addr_t in kernel space to
4452 * user space, based on 32/64 bitness of the user space
4453 *
4454 * Parameters: ua User space address to copy to
4455 * ptr Address of kernel space user_addr_t
4456 * ptr_size 4/8, based on 'ua' address space
4457 *
4458 * Returns: 0 Success
4459 * EFAULT Bad 'ua'
4460 *
4461 */
4462 static int
4463 copyoutptr(user_addr_t ua, user_addr_t ptr, int ptr_size)
4464 {
4465 int error;
4466
4467 if (ptr_size == 4) {
4468 /* 64 bit value containing 32 bit address */
4469 unsigned int i = CAST_DOWN_EXPLICIT(unsigned int, ua); /* SAFE */
4470
4471 error = copyout(&i, ptr, 4);
4472 } else {
4473 error = copyout(&ua, ptr, 8);
4474 }
4475 return error;
4476 }
4477
4478
4479 /*
4480 * exec_copyout_strings
4481 *
4482 * Copy out the strings segment to user space. The strings segment is put
4483 * on a preinitialized stack frame.
4484 *
4485 * Parameters: struct image_params * the image parameter block
4486 * int * a pointer to the stack offset variable
4487 *
4488 * Returns: 0 Success
4489 * !0 Faiure: errno
4490 *
4491 * Implicit returns:
4492 * (*stackp) The stack offset, modified
4493 *
4494 * Note: The strings segment layout is backward, from the beginning
4495 * of the top of the stack to consume the minimal amount of
4496 * space possible; the returned stack pointer points to the
4497 * end of the area consumed (stacks grow downward).
4498 *
4499 * argc is an int; arg[i] are pointers; env[i] are pointers;
4500 * the 0's are (void *)NULL's
4501 *
4502 * The stack frame layout is:
4503 *
4504 * +-------------+ <- p->user_stack
4505 * | 16b |
4506 * +-------------+
4507 * | STRING AREA |
4508 * | : |
4509 * | : |
4510 * | : |
4511 * +- -- -- -- --+
4512 * | PATH AREA |
4513 * +-------------+
4514 * | 0 |
4515 * +-------------+
4516 * | applev[n] |
4517 * +-------------+
4518 * :
4519 * :
4520 * +-------------+
4521 * | applev[1] |
4522 * +-------------+
4523 * | exec_path / |
4524 * | applev[0] |
4525 * +-------------+
4526 * | 0 |
4527 * +-------------+
4528 * | env[n] |
4529 * +-------------+
4530 * :
4531 * :
4532 * +-------------+
4533 * | env[0] |
4534 * +-------------+
4535 * | 0 |
4536 * +-------------+
4537 * | arg[argc-1] |
4538 * +-------------+
4539 * :
4540 * :
4541 * +-------------+
4542 * | arg[0] |
4543 * +-------------+
4544 * | argc |
4545 * sp-> +-------------+
4546 *
4547 * Although technically a part of the STRING AREA, we treat the PATH AREA as
4548 * a separate entity. This allows us to align the beginning of the PATH AREA
4549 * to a pointer boundary so that the exec_path, env[i], and argv[i] pointers
4550 * which preceed it on the stack are properly aligned.
4551 */
4552
4553 static int
4554 exec_copyout_strings(struct image_params *imgp, user_addr_t *stackp)
4555 {
4556 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
4557 int ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT_ADDR) ? 8 : 4;
4558 int ptr_area_size;
4559 void *ptr_buffer_start, *ptr_buffer;
4560 int string_size;
4561
4562 user_addr_t string_area; /* *argv[], *env[] */
4563 user_addr_t ptr_area; /* argv[], env[], applev[] */
4564 user_addr_t argc_area; /* argc */
4565 user_addr_t stack;
4566 int error;
4567
4568 unsigned i;
4569 struct copyout_desc {
4570 char *start_string;
4571 int count;
4572 #if CONFIG_DTRACE
4573 user_addr_t *dtrace_cookie;
4574 #endif
4575 boolean_t null_term;
4576 } descriptors[] = {
4577 {
4578 .start_string = imgp->ip_startargv,
4579 .count = imgp->ip_argc,
4580 #if CONFIG_DTRACE
4581 .dtrace_cookie = &p->p_dtrace_argv,
4582 #endif
4583 .null_term = TRUE
4584 },
4585 {
4586 .start_string = imgp->ip_endargv,
4587 .count = imgp->ip_envc,
4588 #if CONFIG_DTRACE
4589 .dtrace_cookie = &p->p_dtrace_envp,
4590 #endif
4591 .null_term = TRUE
4592 },
4593 {
4594 .start_string = imgp->ip_strings,
4595 .count = 1,
4596 #if CONFIG_DTRACE
4597 .dtrace_cookie = NULL,
4598 #endif
4599 .null_term = FALSE
4600 },
4601 {
4602 .start_string = imgp->ip_endenvv,
4603 .count = imgp->ip_applec - 1, /* exec_path handled above */
4604 #if CONFIG_DTRACE
4605 .dtrace_cookie = NULL,
4606 #endif
4607 .null_term = TRUE
4608 }
4609 };
4610
4611 stack = *stackp;
4612
4613 /*
4614 * All previous contributors to the string area
4615 * should have aligned their sub-area
4616 */
4617 if (imgp->ip_strspace % ptr_size != 0) {
4618 error = EINVAL;
4619 goto bad;
4620 }
4621
4622 /* Grow the stack down for the strings we've been building up */
4623 string_size = imgp->ip_strendp - imgp->ip_strings;
4624 stack -= string_size;
4625 string_area = stack;
4626
4627 /*
4628 * Need room for one pointer for each string, plus
4629 * one for the NULLs terminating the argv, envv, and apple areas.
4630 */
4631 ptr_area_size = (imgp->ip_argc + imgp->ip_envc + imgp->ip_applec + 3) * ptr_size;
4632 stack -= ptr_area_size;
4633 ptr_area = stack;
4634
4635 /* We'll construct all the pointer arrays in our string buffer,
4636 * which we already know is aligned properly, and ip_argspace
4637 * was used to verify we have enough space.
4638 */
4639 ptr_buffer_start = ptr_buffer = (void *)imgp->ip_strendp;
4640
4641 /*
4642 * Need room for pointer-aligned argc slot.
4643 */
4644 stack -= ptr_size;
4645 argc_area = stack;
4646
4647 /*
4648 * Record the size of the arguments area so that sysctl_procargs()
4649 * can return the argument area without having to parse the arguments.
4650 */
4651 proc_lock(p);
4652 p->p_argc = imgp->ip_argc;
4653 p->p_argslen = (int)(*stackp - string_area);
4654 proc_unlock(p);
4655
4656 /* Return the initial stack address: the location of argc */
4657 *stackp = stack;
4658
4659 /*
4660 * Copy out the entire strings area.
4661 */
4662 error = copyout(imgp->ip_strings, string_area,
4663 string_size);
4664 if (error) {
4665 goto bad;
4666 }
4667
4668 for (i = 0; i < sizeof(descriptors) / sizeof(descriptors[0]); i++) {
4669 char *cur_string = descriptors[i].start_string;
4670 int j;
4671
4672 #if CONFIG_DTRACE
4673 if (descriptors[i].dtrace_cookie) {
4674 proc_lock(p);
4675 *descriptors[i].dtrace_cookie = ptr_area + ((uintptr_t)ptr_buffer - (uintptr_t)ptr_buffer_start); /* dtrace convenience */
4676 proc_unlock(p);
4677 }
4678 #endif /* CONFIG_DTRACE */
4679
4680 /*
4681 * For each segment (argv, envv, applev), copy as many pointers as requested
4682 * to our pointer buffer.
4683 */
4684 for (j = 0; j < descriptors[i].count; j++) {
4685 user_addr_t cur_address = string_area + (cur_string - imgp->ip_strings);
4686
4687 /* Copy out the pointer to the current string. Alignment has been verified */
4688 if (ptr_size == 8) {
4689 *(uint64_t *)ptr_buffer = (uint64_t)cur_address;
4690 } else {
4691 *(uint32_t *)ptr_buffer = (uint32_t)cur_address;
4692 }
4693
4694 ptr_buffer = (void *)((uintptr_t)ptr_buffer + ptr_size);
4695 cur_string += strlen(cur_string) + 1; /* Only a NUL between strings in the same area */
4696 }
4697
4698 if (descriptors[i].null_term) {
4699 if (ptr_size == 8) {
4700 *(uint64_t *)ptr_buffer = 0ULL;
4701 } else {
4702 *(uint32_t *)ptr_buffer = 0;
4703 }
4704
4705 ptr_buffer = (void *)((uintptr_t)ptr_buffer + ptr_size);
4706 }
4707 }
4708
4709 /*
4710 * Copy out all our pointer arrays in bulk.
4711 */
4712 error = copyout(ptr_buffer_start, ptr_area,
4713 ptr_area_size);
4714 if (error) {
4715 goto bad;
4716 }
4717
4718 /* argc (int32, stored in a ptr_size area) */
4719 error = copyoutptr((user_addr_t)imgp->ip_argc, argc_area, ptr_size);
4720 if (error) {
4721 goto bad;
4722 }
4723
4724 bad:
4725 return error;
4726 }
4727
4728
4729 /*
4730 * exec_extract_strings
4731 *
4732 * Copy arguments and environment from user space into work area; we may
4733 * have already copied some early arguments into the work area, and if
4734 * so, any arguments opied in are appended to those already there.
4735 * This function is the primary manipulator of ip_argspace, since
4736 * these are the arguments the client of execve(2) knows about. After
4737 * each argv[]/envv[] string is copied, we charge the string length
4738 * and argv[]/envv[] pointer slot to ip_argspace, so that we can
4739 * full preflight the arg list size.
4740 *
4741 * Parameters: struct image_params * the image parameter block
4742 *
4743 * Returns: 0 Success
4744 * !0 Failure: errno
4745 *
4746 * Implicit returns;
4747 * (imgp->ip_argc) Count of arguments, updated
4748 * (imgp->ip_envc) Count of environment strings, updated
4749 * (imgp->ip_argspace) Count of remaining of NCARGS
4750 * (imgp->ip_interp_buffer) Interpreter and args (mutated in place)
4751 *
4752 *
4753 * Note: The argument and environment vectors are user space pointers
4754 * to arrays of user space pointers.
4755 */
4756 static int
4757 exec_extract_strings(struct image_params *imgp)
4758 {
4759 int error = 0;
4760 int ptr_size = (imgp->ip_flags & IMGPF_WAS_64BIT_ADDR) ? 8 : 4;
4761 int new_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT_ADDR) ? 8 : 4;
4762 user_addr_t argv = imgp->ip_user_argv;
4763 user_addr_t envv = imgp->ip_user_envv;
4764
4765 /*
4766 * Adjust space reserved for the path name by however much padding it
4767 * needs. Doing this here since we didn't know if this would be a 32-
4768 * or 64-bit process back in exec_save_path.
4769 */
4770 while (imgp->ip_strspace % new_ptr_size != 0) {
4771 *imgp->ip_strendp++ = '\0';
4772 imgp->ip_strspace--;
4773 /* imgp->ip_argspace--; not counted towards exec args total */
4774 }
4775
4776 /*
4777 * From now on, we start attributing string space to ip_argspace
4778 */
4779 imgp->ip_startargv = imgp->ip_strendp;
4780 imgp->ip_argc = 0;
4781
4782 if ((imgp->ip_flags & IMGPF_INTERPRET) != 0) {
4783 user_addr_t arg;
4784 char *argstart, *ch;
4785
4786 /* First, the arguments in the "#!" string are tokenized and extracted. */
4787 argstart = imgp->ip_interp_buffer;
4788 while (argstart) {
4789 ch = argstart;
4790 while (*ch && !IS_WHITESPACE(*ch)) {
4791 ch++;
4792 }
4793
4794 if (*ch == '\0') {
4795 /* last argument, no need to NUL-terminate */
4796 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(argstart), UIO_SYSSPACE, TRUE);
4797 argstart = NULL;
4798 } else {
4799 /* NUL-terminate */
4800 *ch = '\0';
4801 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(argstart), UIO_SYSSPACE, TRUE);
4802
4803 /*
4804 * Find the next string. We know spaces at the end of the string have already
4805 * been stripped.
4806 */
4807 argstart = ch + 1;
4808 while (IS_WHITESPACE(*argstart)) {
4809 argstart++;
4810 }
4811 }
4812
4813 /* Error-check, regardless of whether this is the last interpreter arg or not */
4814 if (error) {
4815 goto bad;
4816 }
4817 if (imgp->ip_argspace < new_ptr_size) {
4818 error = E2BIG;
4819 goto bad;
4820 }
4821 imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */
4822 imgp->ip_argc++;
4823 }
4824
4825 if (argv != 0LL) {
4826 /*
4827 * If we are running an interpreter, replace the av[0] that was
4828 * passed to execve() with the path name that was
4829 * passed to execve() for interpreters which do not use the PATH
4830 * to locate their script arguments.
4831 */
4832 error = copyinptr(argv, &arg, ptr_size);
4833 if (error) {
4834 goto bad;
4835 }
4836 if (arg != 0LL) {
4837 argv += ptr_size; /* consume without using */
4838 }
4839 }
4840
4841 if (imgp->ip_interp_sugid_fd != -1) {
4842 char temp[19]; /* "/dev/fd/" + 10 digits + NUL */
4843 snprintf(temp, sizeof(temp), "/dev/fd/%d", imgp->ip_interp_sugid_fd);
4844 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(temp), UIO_SYSSPACE, TRUE);
4845 } else {
4846 error = exec_add_user_string(imgp, imgp->ip_user_fname, imgp->ip_seg, TRUE);
4847 }
4848
4849 if (error) {
4850 goto bad;
4851 }
4852 if (imgp->ip_argspace < new_ptr_size) {
4853 error = E2BIG;
4854 goto bad;
4855 }
4856 imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */
4857 imgp->ip_argc++;
4858 }
4859
4860 while (argv != 0LL) {
4861 user_addr_t arg;
4862
4863 error = copyinptr(argv, &arg, ptr_size);
4864 if (error) {
4865 goto bad;
4866 }
4867
4868 if (arg == 0LL) {
4869 break;
4870 }
4871
4872 argv += ptr_size;
4873
4874 /*
4875 * av[n...] = arg[n]
4876 */
4877 error = exec_add_user_string(imgp, arg, imgp->ip_seg, TRUE);
4878 if (error) {
4879 goto bad;
4880 }
4881 if (imgp->ip_argspace < new_ptr_size) {
4882 error = E2BIG;
4883 goto bad;
4884 }
4885 imgp->ip_argspace -= new_ptr_size; /* to hold argv[] entry */
4886 imgp->ip_argc++;
4887 }
4888
4889 /* Save space for argv[] NULL terminator */
4890 if (imgp->ip_argspace < new_ptr_size) {
4891 error = E2BIG;
4892 goto bad;
4893 }
4894 imgp->ip_argspace -= new_ptr_size;
4895
4896 /* Note where the args ends and env begins. */
4897 imgp->ip_endargv = imgp->ip_strendp;
4898 imgp->ip_envc = 0;
4899
4900 /* Now, get the environment */
4901 while (envv != 0LL) {
4902 user_addr_t env;
4903
4904 error = copyinptr(envv, &env, ptr_size);
4905 if (error) {
4906 goto bad;
4907 }
4908
4909 envv += ptr_size;
4910 if (env == 0LL) {
4911 break;
4912 }
4913 /*
4914 * av[n...] = env[n]
4915 */
4916 error = exec_add_user_string(imgp, env, imgp->ip_seg, TRUE);
4917 if (error) {
4918 goto bad;
4919 }
4920 if (imgp->ip_argspace < new_ptr_size) {
4921 error = E2BIG;
4922 goto bad;
4923 }
4924 imgp->ip_argspace -= new_ptr_size; /* to hold envv[] entry */
4925 imgp->ip_envc++;
4926 }
4927
4928 /* Save space for envv[] NULL terminator */
4929 if (imgp->ip_argspace < new_ptr_size) {
4930 error = E2BIG;
4931 goto bad;
4932 }
4933 imgp->ip_argspace -= new_ptr_size;
4934
4935 /* Align the tail of the combined argv+envv area */
4936 while (imgp->ip_strspace % new_ptr_size != 0) {
4937 if (imgp->ip_argspace < 1) {
4938 error = E2BIG;
4939 goto bad;
4940 }
4941 *imgp->ip_strendp++ = '\0';
4942 imgp->ip_strspace--;
4943 imgp->ip_argspace--;
4944 }
4945
4946 /* Note where the envv ends and applev begins. */
4947 imgp->ip_endenvv = imgp->ip_strendp;
4948
4949 /*
4950 * From now on, we are no longer charging argument
4951 * space to ip_argspace.
4952 */
4953
4954 bad:
4955 return error;
4956 }
4957
4958 /*
4959 * Libc has an 8-element array set up for stack guard values. It only fills
4960 * in one of those entries, and both gcc and llvm seem to use only a single
4961 * 8-byte guard. Until somebody needs more than an 8-byte guard value, don't
4962 * do the work to construct them.
4963 */
4964 #define GUARD_VALUES 1
4965 #define GUARD_KEY "stack_guard="
4966
4967 /*
4968 * System malloc needs some entropy when it is initialized.
4969 */
4970 #define ENTROPY_VALUES 2
4971 #define ENTROPY_KEY "malloc_entropy="
4972
4973 /*
4974 * libplatform needs a random pointer-obfuscation value when it is initialized.
4975 */
4976 #define PTR_MUNGE_VALUES 1
4977 #define PTR_MUNGE_KEY "ptr_munge="
4978
4979 /*
4980 * System malloc engages nanozone for UIAPP.
4981 */
4982 #define NANO_ENGAGE_KEY "MallocNanoZone=1"
4983
4984 #define PFZ_KEY "pfz="
4985 extern user32_addr_t commpage_text32_location;
4986 extern user64_addr_t commpage_text64_location;
4987
4988 #define MAIN_STACK_VALUES 4
4989 #define MAIN_STACK_KEY "main_stack="
4990
4991 #define FSID_KEY "executable_file="
4992 #define DYLD_FSID_KEY "dyld_file="
4993 #define CDHASH_KEY "executable_cdhash="
4994 #define DYLD_FLAGS_KEY "dyld_flags="
4995
4996 #define FSID_MAX_STRING "0x1234567890abcdef,0x1234567890abcdef"
4997
4998 #define HEX_STR_LEN 18 // 64-bit hex value "0x0123456701234567"
4999
5000 static int
5001 exec_add_entropy_key(struct image_params *imgp,
5002 const char *key,
5003 int values,
5004 boolean_t embedNUL)
5005 {
5006 const int limit = 8;
5007 uint64_t entropy[limit];
5008 char str[strlen(key) + (HEX_STR_LEN + 1) * limit + 1];
5009 if (values > limit) {
5010 values = limit;
5011 }
5012
5013 read_random(entropy, sizeof(entropy[0]) * values);
5014
5015 if (embedNUL) {
5016 entropy[0] &= ~(0xffull << 8);
5017 }
5018
5019 int len = snprintf(str, sizeof(str), "%s0x%llx", key, entropy[0]);
5020 int remaining = sizeof(str) - len;
5021 for (int i = 1; i < values && remaining > 0; ++i) {
5022 int start = sizeof(str) - remaining;
5023 len = snprintf(&str[start], remaining, ",0x%llx", entropy[i]);
5024 remaining -= len;
5025 }
5026
5027 return exec_add_user_string(imgp, CAST_USER_ADDR_T(str), UIO_SYSSPACE, FALSE);
5028 }
5029
5030 /*
5031 * Build up the contents of the apple[] string vector
5032 */
5033 #if (DEVELOPMENT || DEBUG)
5034 uint64_t dyld_flags = 0;
5035 #endif
5036
5037 static int
5038 exec_add_apple_strings(struct image_params *imgp,
5039 const load_result_t *load_result)
5040 {
5041 int error;
5042 int img_ptr_size = (imgp->ip_flags & IMGPF_IS_64BIT_ADDR) ? 8 : 4;
5043
5044 /* exec_save_path stored the first string */
5045 imgp->ip_applec = 1;
5046
5047 /* adding the pfz string */
5048 {
5049 char pfz_string[strlen(PFZ_KEY) + HEX_STR_LEN + 1];
5050
5051 if (img_ptr_size == 8) {
5052 snprintf(pfz_string, sizeof(pfz_string), PFZ_KEY "0x%llx", commpage_text64_location);
5053 } else {
5054 snprintf(pfz_string, sizeof(pfz_string), PFZ_KEY "0x%x", commpage_text32_location);
5055 }
5056 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(pfz_string), UIO_SYSSPACE, FALSE);
5057 if (error) {
5058 goto bad;
5059 }
5060 imgp->ip_applec++;
5061 }
5062
5063 /* adding the NANO_ENGAGE_KEY key */
5064 if (imgp->ip_px_sa) {
5065 int proc_flags = (((struct _posix_spawnattr *) imgp->ip_px_sa)->psa_flags);
5066
5067 if ((proc_flags & _POSIX_SPAWN_NANO_ALLOCATOR) == _POSIX_SPAWN_NANO_ALLOCATOR) {
5068 const char *nano_string = NANO_ENGAGE_KEY;
5069 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(nano_string), UIO_SYSSPACE, FALSE);
5070 if (error) {
5071 goto bad;
5072 }
5073 imgp->ip_applec++;
5074 }
5075 }
5076
5077 /*
5078 * Supply libc with a collection of random values to use when
5079 * implementing -fstack-protector.
5080 *
5081 * (The first random string always contains an embedded NUL so that
5082 * __stack_chk_guard also protects against C string vulnerabilities)
5083 */
5084 error = exec_add_entropy_key(imgp, GUARD_KEY, GUARD_VALUES, TRUE);
5085 if (error) {
5086 goto bad;
5087 }
5088 imgp->ip_applec++;
5089
5090 /*
5091 * Supply libc with entropy for system malloc.
5092 */
5093 error = exec_add_entropy_key(imgp, ENTROPY_KEY, ENTROPY_VALUES, FALSE);
5094 if (error) {
5095 goto bad;
5096 }
5097 imgp->ip_applec++;
5098
5099 /*
5100 * Supply libpthread & libplatform with a random value to use for pointer
5101 * obfuscation.
5102 */
5103 error = exec_add_entropy_key(imgp, PTR_MUNGE_KEY, PTR_MUNGE_VALUES, FALSE);
5104 if (error) {
5105 goto bad;
5106 }
5107 imgp->ip_applec++;
5108
5109 /*
5110 * Add MAIN_STACK_KEY: Supplies the address and size of the main thread's
5111 * stack if it was allocated by the kernel.
5112 *
5113 * The guard page is not included in this stack size as libpthread
5114 * expects to add it back in after receiving this value.
5115 */
5116 if (load_result->unixproc) {
5117 char stack_string[strlen(MAIN_STACK_KEY) + (HEX_STR_LEN + 1) * MAIN_STACK_VALUES + 1];
5118 snprintf(stack_string, sizeof(stack_string),
5119 MAIN_STACK_KEY "0x%llx,0x%llx,0x%llx,0x%llx",
5120 (uint64_t)load_result->user_stack,
5121 (uint64_t)load_result->user_stack_size,
5122 (uint64_t)load_result->user_stack_alloc,
5123 (uint64_t)load_result->user_stack_alloc_size);
5124 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(stack_string), UIO_SYSSPACE, FALSE);
5125 if (error) {
5126 goto bad;
5127 }
5128 imgp->ip_applec++;
5129 }
5130
5131 if (imgp->ip_vattr) {
5132 uint64_t fsid = vnode_get_va_fsid(imgp->ip_vattr);
5133 uint64_t fsobjid = imgp->ip_vattr->va_fileid;
5134
5135 char fsid_string[strlen(FSID_KEY) + strlen(FSID_MAX_STRING) + 1];
5136 snprintf(fsid_string, sizeof(fsid_string),
5137 FSID_KEY "0x%llx,0x%llx", fsid, fsobjid);
5138 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(fsid_string), UIO_SYSSPACE, FALSE);
5139 if (error) {
5140 goto bad;
5141 }
5142 imgp->ip_applec++;
5143 }
5144
5145 if (imgp->ip_dyld_fsid || imgp->ip_dyld_fsobjid) {
5146 char fsid_string[strlen(DYLD_FSID_KEY) + strlen(FSID_MAX_STRING) + 1];
5147 snprintf(fsid_string, sizeof(fsid_string),
5148 DYLD_FSID_KEY "0x%llx,0x%llx", imgp->ip_dyld_fsid, imgp->ip_dyld_fsobjid);
5149 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(fsid_string), UIO_SYSSPACE, FALSE);
5150 if (error) {
5151 goto bad;
5152 }
5153 imgp->ip_applec++;
5154 }
5155
5156 uint8_t cdhash[SHA1_RESULTLEN];
5157 int cdhash_errror = ubc_cs_getcdhash(imgp->ip_vp, imgp->ip_arch_offset, cdhash);
5158 if (cdhash_errror == 0) {
5159 char hash_string[strlen(CDHASH_KEY) + 2 * SHA1_RESULTLEN + 1];
5160 strncpy(hash_string, CDHASH_KEY, sizeof(hash_string));
5161 char *p = hash_string + sizeof(CDHASH_KEY) - 1;
5162 for (int i = 0; i < SHA1_RESULTLEN; i++) {
5163 snprintf(p, 3, "%02x", (int) cdhash[i]);
5164 p += 2;
5165 }
5166 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(hash_string), UIO_SYSSPACE, FALSE);
5167 if (error) {
5168 goto bad;
5169 }
5170 imgp->ip_applec++;
5171 }
5172 #if (DEVELOPMENT || DEBUG)
5173 if (dyld_flags) {
5174 char dyld_flags_string[strlen(DYLD_FLAGS_KEY) + HEX_STR_LEN + 1];
5175 snprintf(dyld_flags_string, sizeof(dyld_flags_string), DYLD_FLAGS_KEY "0x%llx", dyld_flags);
5176 error = exec_add_user_string(imgp, CAST_USER_ADDR_T(dyld_flags_string), UIO_SYSSPACE, FALSE);
5177 if (error) {
5178 goto bad;
5179 }
5180 imgp->ip_applec++;
5181 }
5182 #endif
5183
5184 /* Align the tail of the combined applev area */
5185 while (imgp->ip_strspace % img_ptr_size != 0) {
5186 *imgp->ip_strendp++ = '\0';
5187 imgp->ip_strspace--;
5188 }
5189
5190 bad:
5191 return error;
5192 }
5193
5194 #define unix_stack_size(p) (p->p_rlimit[RLIMIT_STACK].rlim_cur)
5195
5196 /*
5197 * exec_check_permissions
5198 *
5199 * Description: Verify that the file that is being attempted to be executed
5200 * is in fact allowed to be executed based on it POSIX file
5201 * permissions and other access control criteria
5202 *
5203 * Parameters: struct image_params * the image parameter block
5204 *
5205 * Returns: 0 Success
5206 * EACCES Permission denied
5207 * ENOEXEC Executable file format error
5208 * ETXTBSY Text file busy [misuse of error code]
5209 * vnode_getattr:???
5210 * vnode_authorize:???
5211 */
5212 static int
5213 exec_check_permissions(struct image_params *imgp)
5214 {
5215 struct vnode *vp = imgp->ip_vp;
5216 struct vnode_attr *vap = imgp->ip_vattr;
5217 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
5218 int error;
5219 kauth_action_t action;
5220
5221 /* Only allow execution of regular files */
5222 if (!vnode_isreg(vp)) {
5223 return EACCES;
5224 }
5225
5226 /* Get the file attributes that we will be using here and elsewhere */
5227 VATTR_INIT(vap);
5228 VATTR_WANTED(vap, va_uid);
5229 VATTR_WANTED(vap, va_gid);
5230 VATTR_WANTED(vap, va_mode);
5231 VATTR_WANTED(vap, va_fsid);
5232 VATTR_WANTED(vap, va_fsid64);
5233 VATTR_WANTED(vap, va_fileid);
5234 VATTR_WANTED(vap, va_data_size);
5235 if ((error = vnode_getattr(vp, vap, imgp->ip_vfs_context)) != 0) {
5236 return error;
5237 }
5238
5239 /*
5240 * Ensure that at least one execute bit is on - otherwise root
5241 * will always succeed, and we don't want to happen unless the
5242 * file really is executable.
5243 */
5244 if (!vfs_authopaque(vnode_mount(vp)) && ((vap->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0)) {
5245 return EACCES;
5246 }
5247
5248 /* Disallow zero length files */
5249 if (vap->va_data_size == 0) {
5250 return ENOEXEC;
5251 }
5252
5253 imgp->ip_arch_offset = (user_size_t)0;
5254 imgp->ip_arch_size = vap->va_data_size;
5255
5256 /* Disable setuid-ness for traced programs or if MNT_NOSUID */
5257 if ((vp->v_mount->mnt_flag & MNT_NOSUID) || (p->p_lflag & P_LTRACED)) {
5258 vap->va_mode &= ~(VSUID | VSGID);
5259 }
5260
5261 /*
5262 * Disable _POSIX_SPAWN_ALLOW_DATA_EXEC and _POSIX_SPAWN_DISABLE_ASLR
5263 * flags for setuid/setgid binaries.
5264 */
5265 if (vap->va_mode & (VSUID | VSGID)) {
5266 imgp->ip_flags &= ~(IMGPF_ALLOW_DATA_EXEC | IMGPF_DISABLE_ASLR);
5267 }
5268
5269 #if CONFIG_MACF
5270 error = mac_vnode_check_exec(imgp->ip_vfs_context, vp, imgp);
5271 if (error) {
5272 return error;
5273 }
5274 #endif
5275
5276 /* Check for execute permission */
5277 action = KAUTH_VNODE_EXECUTE;
5278 /* Traced images must also be readable */
5279 if (p->p_lflag & P_LTRACED) {
5280 action |= KAUTH_VNODE_READ_DATA;
5281 }
5282 if ((error = vnode_authorize(vp, NULL, action, imgp->ip_vfs_context)) != 0) {
5283 return error;
5284 }
5285
5286 #if 0
5287 /* Don't let it run if anyone had it open for writing */
5288 vnode_lock(vp);
5289 if (vp->v_writecount) {
5290 panic("going to return ETXTBSY %x", vp);
5291 vnode_unlock(vp);
5292 return ETXTBSY;
5293 }
5294 vnode_unlock(vp);
5295 #endif
5296
5297
5298 /* XXX May want to indicate to underlying FS that vnode is open */
5299
5300 return error;
5301 }
5302
5303
5304 /*
5305 * exec_handle_sugid
5306 *
5307 * Initially clear the P_SUGID in the process flags; if an SUGID process is
5308 * exec'ing a non-SUGID image, then this is the point of no return.
5309 *
5310 * If the image being activated is SUGID, then replace the credential with a
5311 * copy, disable tracing (unless the tracing process is root), reset the
5312 * mach task port to revoke it, set the P_SUGID bit,
5313 *
5314 * If the saved user and group ID will be changing, then make sure it happens
5315 * to a new credential, rather than a shared one.
5316 *
5317 * Set the security token (this is probably obsolete, given that the token
5318 * should not technically be separate from the credential itself).
5319 *
5320 * Parameters: struct image_params * the image parameter block
5321 *
5322 * Returns: void No failure indication
5323 *
5324 * Implicit returns:
5325 * <process credential> Potentially modified/replaced
5326 * <task port> Potentially revoked
5327 * <process flags> P_SUGID bit potentially modified
5328 * <security token> Potentially modified
5329 */
5330 static int
5331 exec_handle_sugid(struct image_params *imgp)
5332 {
5333 proc_t p = vfs_context_proc(imgp->ip_vfs_context);
5334 kauth_cred_t cred = vfs_context_ucred(imgp->ip_vfs_context);
5335 int i;
5336 int leave_sugid_clear = 0;
5337 int mac_reset_ipc = 0;
5338 int error = 0;
5339 task_t task = NULL;
5340 #if CONFIG_MACF
5341 int mac_transition, disjoint_cred = 0;
5342 int label_update_return = 0;
5343
5344 /*
5345 * Determine whether a call to update the MAC label will result in the
5346 * credential changing.
5347 *
5348 * Note: MAC policies which do not actually end up modifying
5349 * the label subsequently are strongly encouraged to
5350 * return 0 for this check, since a non-zero answer will
5351 * slow down the exec fast path for normal binaries.
5352 */
5353 mac_transition = mac_cred_check_label_update_execve(
5354 imgp->ip_vfs_context,
5355 imgp->ip_vp,
5356 imgp->ip_arch_offset,
5357 imgp->ip_scriptvp,
5358 imgp->ip_scriptlabelp,
5359 imgp->ip_execlabelp,
5360 p,
5361 imgp->ip_px_smpx);
5362 #endif
5363
5364 OSBitAndAtomic(~((uint32_t)P_SUGID), &p->p_flag);
5365
5366 /*
5367 * Order of the following is important; group checks must go last,
5368 * as we use the success of the 'ismember' check combined with the
5369 * failure of the explicit match to indicate that we will be setting
5370 * the egid of the process even though the new process did not
5371 * require VSUID/VSGID bits in order for it to set the new group as
5372 * its egid.
5373 *
5374 * Note: Technically, by this we are implying a call to
5375 * setegid() in the new process, rather than implying
5376 * it used its VSGID bit to set the effective group,
5377 * even though there is no code in that process to make
5378 * such a call.
5379 */
5380 if (((imgp->ip_origvattr->va_mode & VSUID) != 0 &&
5381 kauth_cred_getuid(cred) != imgp->ip_origvattr->va_uid) ||
5382 ((imgp->ip_origvattr->va_mode & VSGID) != 0 &&
5383 ((kauth_cred_ismember_gid(cred, imgp->ip_origvattr->va_gid, &leave_sugid_clear) || !leave_sugid_clear) ||
5384 (kauth_cred_getgid(cred) != imgp->ip_origvattr->va_gid)))) {
5385 #if CONFIG_MACF
5386 /* label for MAC transition and neither VSUID nor VSGID */
5387 handle_mac_transition:
5388 #endif
5389
5390 #if !SECURE_KERNEL
5391 /*
5392 * Replace the credential with a copy of itself if euid or
5393 * egid change.
5394 *
5395 * Note: setuid binaries will automatically opt out of
5396 * group resolver participation as a side effect
5397 * of this operation. This is an intentional
5398 * part of the security model, which requires a
5399 * participating credential be established by
5400 * escalating privilege, setting up all other
5401 * aspects of the credential including whether
5402 * or not to participate in external group
5403 * membership resolution, then dropping their
5404 * effective privilege to that of the desired
5405 * final credential state.
5406 *
5407 * Modifications to p_ucred must be guarded using the
5408 * proc's ucred lock. This prevents others from accessing
5409 * a garbage credential.
5410 */
5411 if (imgp->ip_origvattr->va_mode & VSUID) {
5412 apply_kauth_cred_update(p, ^kauth_cred_t (kauth_cred_t my_cred) {
5413 return kauth_cred_setresuid(my_cred,
5414 KAUTH_UID_NONE,
5415 imgp->ip_origvattr->va_uid,
5416 imgp->ip_origvattr->va_uid,
5417 KAUTH_UID_NONE);
5418 });
5419 }
5420
5421 if (imgp->ip_origvattr->va_mode & VSGID) {
5422 apply_kauth_cred_update(p, ^kauth_cred_t (kauth_cred_t my_cred) {
5423 return kauth_cred_setresgid(my_cred,
5424 KAUTH_GID_NONE,
5425 imgp->ip_origvattr->va_gid,
5426 imgp->ip_origvattr->va_gid);
5427 });
5428 }
5429 #endif /* !SECURE_KERNEL */
5430
5431 #if CONFIG_MACF
5432 /*
5433 * If a policy has indicated that it will transition the label,
5434 * before making the call into the MAC policies, get a new
5435 * duplicate credential, so they can modify it without
5436 * modifying any others sharing it.
5437 */
5438 if (mac_transition) {
5439 /*
5440 * This hook may generate upcalls that require
5441 * importance donation from the kernel.
5442 * (23925818)
5443 */
5444 thread_t thread = current_thread();
5445 thread_enable_send_importance(thread, TRUE);
5446 kauth_proc_label_update_execve(p,
5447 imgp->ip_vfs_context,
5448 imgp->ip_vp,
5449 imgp->ip_arch_offset,
5450 imgp->ip_scriptvp,
5451 imgp->ip_scriptlabelp,
5452 imgp->ip_execlabelp,
5453 &imgp->ip_csflags,
5454 imgp->ip_px_smpx,
5455 &disjoint_cred, /* will be non zero if disjoint */
5456 &label_update_return);
5457 thread_enable_send_importance(thread, FALSE);
5458
5459 if (disjoint_cred) {
5460 /*
5461 * If updating the MAC label resulted in a
5462 * disjoint credential, flag that we need to
5463 * set the P_SUGID bit. This protects
5464 * against debuggers being attached by an
5465 * insufficiently privileged process onto the
5466 * result of a transition to a more privileged
5467 * credential.
5468 */
5469 leave_sugid_clear = 0;
5470 }
5471
5472 imgp->ip_mac_return = label_update_return;
5473 }
5474
5475 mac_reset_ipc = mac_proc_check_inherit_ipc_ports(p, p->p_textvp, p->p_textoff, imgp->ip_vp, imgp->ip_arch_offset, imgp->ip_scriptvp);
5476
5477 #endif /* CONFIG_MACF */
5478
5479 /*
5480 * If 'leave_sugid_clear' is non-zero, then we passed the
5481 * VSUID and MACF checks, and successfully determined that
5482 * the previous cred was a member of the VSGID group, but
5483 * that it was not the default at the time of the execve,
5484 * and that the post-labelling credential was not disjoint.
5485 * So we don't set the P_SUGID or reset mach ports and fds
5486 * on the basis of simply running this code.
5487 */
5488 if (mac_reset_ipc || !leave_sugid_clear) {
5489 /*
5490 * Have mach reset the task and thread ports.
5491 * We don't want anyone who had the ports before
5492 * a setuid exec to be able to access/control the
5493 * task/thread after.
5494 */
5495 ipc_task_reset((imgp->ip_new_thread != NULL) ?
5496 get_threadtask(imgp->ip_new_thread) : p->task);
5497 ipc_thread_reset((imgp->ip_new_thread != NULL) ?
5498 imgp->ip_new_thread : current_thread());
5499 }
5500
5501 if (!leave_sugid_clear) {
5502 /*
5503 * Flag the process as setuid.
5504 */
5505 OSBitOrAtomic(P_SUGID, &p->p_flag);
5506
5507 /*
5508 * Radar 2261856; setuid security hole fix
5509 * XXX For setuid processes, attempt to ensure that
5510 * stdin, stdout, and stderr are already allocated.
5511 * We do not want userland to accidentally allocate
5512 * descriptors in this range which has implied meaning
5513 * to libc.
5514 */
5515 for (i = 0; i < 3; i++) {
5516 if (p->p_fd->fd_ofiles[i] != NULL) {
5517 continue;
5518 }
5519
5520 /*
5521 * Do the kernel equivalent of
5522 *
5523 * if i == 0
5524 * (void) open("/dev/null", O_RDONLY);
5525 * else
5526 * (void) open("/dev/null", O_WRONLY);
5527 */
5528
5529 struct fileproc *fp;
5530 int indx;
5531 int flag;
5532 struct nameidata *ndp = NULL;
5533
5534 if (i == 0) {
5535 flag = FREAD;
5536 } else {
5537 flag = FWRITE;
5538 }
5539
5540 if ((error = falloc(p,
5541 &fp, &indx, imgp->ip_vfs_context)) != 0) {
5542 continue;
5543 }
5544
5545 MALLOC(ndp, struct nameidata *, sizeof(*ndp), M_TEMP, M_WAITOK | M_ZERO);
5546 if (ndp == NULL) {
5547 fp_free(p, indx, fp);
5548 error = ENOMEM;
5549 break;
5550 }
5551
5552 NDINIT(ndp, LOOKUP, OP_OPEN, FOLLOW, UIO_SYSSPACE,
5553 CAST_USER_ADDR_T("/dev/null"),
5554 imgp->ip_vfs_context);
5555
5556 if ((error = vn_open(ndp, flag, 0)) != 0) {
5557 fp_free(p, indx, fp);
5558 FREE(ndp, M_TEMP);
5559 break;
5560 }
5561
5562 struct fileglob *fg = fp->f_fglob;
5563
5564 fg->fg_flag = flag;
5565 fg->fg_ops = &vnops;
5566 fg->fg_data = ndp->ni_vp;
5567
5568 vnode_put(ndp->ni_vp);
5569
5570 proc_fdlock(p);
5571 procfdtbl_releasefd(p, indx, NULL);
5572 fp_drop(p, indx, fp, 1);
5573 proc_fdunlock(p);
5574
5575 FREE(ndp, M_TEMP);
5576 }
5577 }
5578 }
5579 #if CONFIG_MACF
5580 else {
5581 /*
5582 * We are here because we were told that the MAC label will
5583 * be transitioned, and the binary is not VSUID or VSGID; to
5584 * deal with this case, we could either duplicate a lot of
5585 * code, or we can indicate we want to default the P_SUGID
5586 * bit clear and jump back up.
5587 */
5588 if (mac_transition) {
5589 leave_sugid_clear = 1;
5590 goto handle_mac_transition;
5591 }
5592 }
5593
5594 #endif /* CONFIG_MACF */
5595
5596 /*
5597 * Implement the semantic where the effective user and group become
5598 * the saved user and group in exec'ed programs.
5599 *
5600 * Modifications to p_ucred must be guarded using the
5601 * proc's ucred lock. This prevents others from accessing
5602 * a garbage credential.
5603 */
5604 apply_kauth_cred_update(p, ^kauth_cred_t (kauth_cred_t my_cred) {
5605 return kauth_cred_setsvuidgid(my_cred,
5606 kauth_cred_getuid(my_cred),
5607 kauth_cred_getgid(my_cred));
5608 });
5609
5610 /* Update the process' identity version and set the security token */
5611 p->p_idversion = OSIncrementAtomic(&nextpidversion);
5612
5613 if (imgp->ip_new_thread != NULL) {
5614 task = get_threadtask(imgp->ip_new_thread);
5615 } else {
5616 task = p->task;
5617 }
5618 set_security_token_task_internal(p, task);
5619
5620 return error;
5621 }
5622
5623
5624 /*
5625 * create_unix_stack
5626 *
5627 * Description: Set the user stack address for the process to the provided
5628 * address. If a custom stack was not set as a result of the
5629 * load process (i.e. as specified by the image file for the
5630 * executable), then allocate the stack in the provided map and
5631 * set up appropriate guard pages for enforcing administrative
5632 * limits on stack growth, if they end up being needed.
5633 *
5634 * Parameters: p Process to set stack on
5635 * load_result Information from mach-o load commands
5636 * map Address map in which to allocate the new stack
5637 *
5638 * Returns: KERN_SUCCESS Stack successfully created
5639 * !KERN_SUCCESS Mach failure code
5640 */
5641 static kern_return_t
5642 create_unix_stack(vm_map_t map, load_result_t* load_result,
5643 proc_t p)
5644 {
5645 mach_vm_size_t size, prot_size;
5646 mach_vm_offset_t addr, prot_addr;
5647 kern_return_t kr;
5648
5649 mach_vm_address_t user_stack = load_result->user_stack;
5650
5651 proc_lock(p);
5652 p->user_stack = user_stack;
5653 if (load_result->custom_stack) {
5654 p->p_lflag |= P_LCUSTOM_STACK;
5655 }
5656 proc_unlock(p);
5657
5658 if (load_result->user_stack_alloc_size > 0) {
5659 /*
5660 * Allocate enough space for the maximum stack size we
5661 * will ever authorize and an extra page to act as
5662 * a guard page for stack overflows. For default stacks,
5663 * vm_initial_limit_stack takes care of the extra guard page.
5664 * Otherwise we must allocate it ourselves.
5665 */
5666 if (mach_vm_round_page_overflow(load_result->user_stack_alloc_size, &size)) {
5667 return KERN_INVALID_ARGUMENT;
5668 }
5669 addr = mach_vm_trunc_page(load_result->user_stack - size);
5670 kr = mach_vm_allocate_kernel(map, &addr, size,
5671 VM_FLAGS_FIXED, VM_MEMORY_STACK);
5672 if (kr != KERN_SUCCESS) {
5673 // Can't allocate at default location, try anywhere
5674 addr = 0;
5675 kr = mach_vm_allocate_kernel(map, &addr, size,
5676 VM_FLAGS_ANYWHERE, VM_MEMORY_STACK);
5677 if (kr != KERN_SUCCESS) {
5678 return kr;
5679 }
5680
5681 user_stack = addr + size;
5682 load_result->user_stack = user_stack;
5683
5684 proc_lock(p);
5685 p->user_stack = user_stack;
5686 proc_unlock(p);
5687 }
5688
5689 load_result->user_stack_alloc = addr;
5690
5691 /*
5692 * And prevent access to what's above the current stack
5693 * size limit for this process.
5694 */
5695 if (load_result->user_stack_size == 0) {
5696 proc_list_lock();
5697 load_result->user_stack_size = unix_stack_size(p);
5698 proc_list_unlock();
5699 prot_size = mach_vm_trunc_page(size - load_result->user_stack_size);
5700 } else {
5701 prot_size = PAGE_SIZE;
5702 }
5703
5704 prot_addr = addr;
5705 kr = mach_vm_protect(map,
5706 prot_addr,
5707 prot_size,
5708 FALSE,
5709 VM_PROT_NONE);
5710 if (kr != KERN_SUCCESS) {
5711 (void)mach_vm_deallocate(map, addr, size);
5712 return kr;
5713 }
5714 }
5715
5716 return KERN_SUCCESS;
5717 }
5718
5719 #include <sys/reboot.h>
5720
5721 /*
5722 * load_init_program_at_path
5723 *
5724 * Description: Load the "init" program; in most cases, this will be "launchd"
5725 *
5726 * Parameters: p Process to call execve() to create
5727 * the "init" program
5728 * scratch_addr Page in p, scratch space
5729 * path NULL terminated path
5730 *
5731 * Returns: KERN_SUCCESS Success
5732 * !KERN_SUCCESS See execve/mac_execve for error codes
5733 *
5734 * Notes: The process that is passed in is the first manufactured
5735 * process on the system, and gets here via bsd_ast() firing
5736 * for the first time. This is done to ensure that bsd_init()
5737 * has run to completion.
5738 *
5739 * The address map of the first manufactured process matches the
5740 * word width of the kernel. Once the self-exec completes, the
5741 * initproc might be different.
5742 */
5743 static int
5744 load_init_program_at_path(proc_t p, user_addr_t scratch_addr, const char* path)
5745 {
5746 int retval[2];
5747 int error;
5748 struct execve_args init_exec_args;
5749 user_addr_t argv0 = USER_ADDR_NULL, argv1 = USER_ADDR_NULL;
5750
5751 /*
5752 * Validate inputs and pre-conditions
5753 */
5754 assert(p);
5755 assert(scratch_addr);
5756 assert(path);
5757
5758 /*
5759 * Copy out program name.
5760 */
5761 size_t path_length = strlen(path) + 1;
5762 argv0 = scratch_addr;
5763 error = copyout(path, argv0, path_length);
5764 if (error) {
5765 return error;
5766 }
5767
5768 scratch_addr = USER_ADDR_ALIGN(scratch_addr + path_length, sizeof(user_addr_t));
5769
5770 /*
5771 * Put out first (and only) argument, similarly.
5772 * Assumes everything fits in a page as allocated above.
5773 */
5774 if (boothowto & RB_SINGLE) {
5775 const char *init_args = "-s";
5776 size_t init_args_length = strlen(init_args) + 1;
5777
5778 argv1 = scratch_addr;
5779 error = copyout(init_args, argv1, init_args_length);
5780 if (error) {
5781 return error;
5782 }
5783
5784 scratch_addr = USER_ADDR_ALIGN(scratch_addr + init_args_length, sizeof(user_addr_t));
5785 }
5786
5787 if (proc_is64bit(p)) {
5788 user64_addr_t argv64bit[3] = {};
5789
5790 argv64bit[0] = argv0;
5791 argv64bit[1] = argv1;
5792 argv64bit[2] = USER_ADDR_NULL;
5793
5794 error = copyout(argv64bit, scratch_addr, sizeof(argv64bit));
5795 if (error) {
5796 return error;
5797 }
5798 } else {
5799 user32_addr_t argv32bit[3] = {};
5800
5801 argv32bit[0] = (user32_addr_t)argv0;
5802 argv32bit[1] = (user32_addr_t)argv1;
5803 argv32bit[2] = USER_ADDR_NULL;
5804
5805 error = copyout(argv32bit, scratch_addr, sizeof(argv32bit));
5806 if (error) {
5807 return error;
5808 }
5809 }
5810
5811 /*
5812 * Set up argument block for fake call to execve.
5813 */
5814 init_exec_args.fname = argv0;
5815 init_exec_args.argp = scratch_addr;
5816 init_exec_args.envp = USER_ADDR_NULL;
5817
5818 /*
5819 * So that init task is set with uid,gid 0 token
5820 */
5821 set_security_token(p);
5822
5823 return execve(p, &init_exec_args, retval);
5824 }
5825
5826 static const char * init_programs[] = {
5827 #if DEBUG
5828 "/usr/local/sbin/launchd.debug",
5829 #endif
5830 #if DEVELOPMENT || DEBUG
5831 "/usr/local/sbin/launchd.development",
5832 #endif
5833 "/sbin/launchd",
5834 };
5835
5836 /*
5837 * load_init_program
5838 *
5839 * Description: Load the "init" program; in most cases, this will be "launchd"
5840 *
5841 * Parameters: p Process to call execve() to create
5842 * the "init" program
5843 *
5844 * Returns: (void)
5845 *
5846 * Notes: The process that is passed in is the first manufactured
5847 * process on the system, and gets here via bsd_ast() firing
5848 * for the first time. This is done to ensure that bsd_init()
5849 * has run to completion.
5850 *
5851 * In DEBUG & DEVELOPMENT builds, the launchdsuffix boot-arg
5852 * may be used to select a specific launchd executable. As with
5853 * the kcsuffix boot-arg, setting launchdsuffix to "" or "release"
5854 * will force /sbin/launchd to be selected.
5855 *
5856 * Search order by build:
5857 *
5858 * DEBUG DEVELOPMENT RELEASE PATH
5859 * ----------------------------------------------------------------------------------
5860 * 1 1 NA /usr/local/sbin/launchd.$LAUNCHDSUFFIX
5861 * 2 NA NA /usr/local/sbin/launchd.debug
5862 * 3 2 NA /usr/local/sbin/launchd.development
5863 * 4 3 1 /sbin/launchd
5864 */
5865 void
5866 load_init_program(proc_t p)
5867 {
5868 uint32_t i;
5869 int error;
5870 vm_map_t map = current_map();
5871 mach_vm_offset_t scratch_addr = 0;
5872 mach_vm_size_t map_page_size = vm_map_page_size(map);
5873
5874 (void) mach_vm_allocate_kernel(map, &scratch_addr, map_page_size, VM_FLAGS_ANYWHERE, VM_KERN_MEMORY_NONE);
5875 #if CONFIG_MEMORYSTATUS
5876 (void) memorystatus_init_at_boot_snapshot();
5877 #endif /* CONFIG_MEMORYSTATUS */
5878
5879 #if DEBUG || DEVELOPMENT
5880 /* Check for boot-arg suffix first */
5881 char launchd_suffix[64];
5882 if (PE_parse_boot_argn("launchdsuffix", launchd_suffix, sizeof(launchd_suffix))) {
5883 char launchd_path[128];
5884 boolean_t is_release_suffix = ((launchd_suffix[0] == 0) ||
5885 (strcmp(launchd_suffix, "release") == 0));
5886
5887 if (is_release_suffix) {
5888 printf("load_init_program: attempting to load /sbin/launchd\n");
5889 error = load_init_program_at_path(p, (user_addr_t)scratch_addr, "/sbin/launchd");
5890 if (!error) {
5891 return;
5892 }
5893
5894 panic("Process 1 exec of launchd.release failed, errno %d", error);
5895 } else {
5896 strlcpy(launchd_path, "/usr/local/sbin/launchd.", sizeof(launchd_path));
5897 strlcat(launchd_path, launchd_suffix, sizeof(launchd_path));
5898
5899 printf("load_init_program: attempting to load %s\n", launchd_path);
5900 error = load_init_program_at_path(p, (user_addr_t)scratch_addr, launchd_path);
5901 if (!error) {
5902 return;
5903 } else {
5904 printf("load_init_program: failed loading %s: errno %d\n", launchd_path, error);
5905 }
5906 }
5907 }
5908 #endif
5909
5910 error = ENOENT;
5911 for (i = 0; i < sizeof(init_programs) / sizeof(init_programs[0]); i++) {
5912 printf("load_init_program: attempting to load %s\n", init_programs[i]);
5913 error = load_init_program_at_path(p, (user_addr_t)scratch_addr, init_programs[i]);
5914 if (!error) {
5915 return;
5916 } else {
5917 printf("load_init_program: failed loading %s: errno %d\n", init_programs[i], error);
5918 }
5919 }
5920
5921 panic("Process 1 exec of %s failed, errno %d", ((i == 0) ? "<null>" : init_programs[i - 1]), error);
5922 }
5923
5924 /*
5925 * load_return_to_errno
5926 *
5927 * Description: Convert a load_return_t (Mach error) to an errno (BSD error)
5928 *
5929 * Parameters: lrtn Mach error number
5930 *
5931 * Returns: (int) BSD error number
5932 * 0 Success
5933 * EBADARCH Bad architecture
5934 * EBADMACHO Bad Mach object file
5935 * ESHLIBVERS Bad shared library version
5936 * ENOMEM Out of memory/resource shortage
5937 * EACCES Access denied
5938 * ENOENT Entry not found (usually "file does
5939 * does not exist")
5940 * EIO An I/O error occurred
5941 * EBADEXEC The executable is corrupt/unknown
5942 */
5943 static int
5944 load_return_to_errno(load_return_t lrtn)
5945 {
5946 switch (lrtn) {
5947 case LOAD_SUCCESS:
5948 return 0;
5949 case LOAD_BADARCH:
5950 return EBADARCH;
5951 case LOAD_BADMACHO:
5952 case LOAD_BADMACHO_UPX:
5953 return EBADMACHO;
5954 case LOAD_SHLIB:
5955 return ESHLIBVERS;
5956 case LOAD_NOSPACE:
5957 case LOAD_RESOURCE:
5958 return ENOMEM;
5959 case LOAD_PROTECT:
5960 return EACCES;
5961 case LOAD_ENOENT:
5962 return ENOENT;
5963 case LOAD_IOERROR:
5964 return EIO;
5965 case LOAD_DECRYPTFAIL:
5966 return EAUTH;
5967 case LOAD_FAILURE:
5968 default:
5969 return EBADEXEC;
5970 }
5971 }
5972
5973 #include <mach/mach_types.h>
5974 #include <mach/vm_prot.h>
5975 #include <mach/semaphore.h>
5976 #include <mach/sync_policy.h>
5977 #include <kern/clock.h>
5978 #include <mach/kern_return.h>
5979
5980 /*
5981 * execargs_alloc
5982 *
5983 * Description: Allocate the block of memory used by the execve arguments.
5984 * At the same time, we allocate a page so that we can read in
5985 * the first page of the image.
5986 *
5987 * Parameters: struct image_params * the image parameter block
5988 *
5989 * Returns: 0 Success
5990 * EINVAL Invalid argument
5991 * EACCES Permission denied
5992 * EINTR Interrupted function
5993 * ENOMEM Not enough space
5994 *
5995 * Notes: This is a temporary allocation into the kernel address space
5996 * to enable us to copy arguments in from user space. This is
5997 * necessitated by not mapping the process calling execve() into
5998 * the kernel address space during the execve() system call.
5999 *
6000 * We assemble the argument and environment, etc., into this
6001 * region before copying it as a single block into the child
6002 * process address space (at the top or bottom of the stack,
6003 * depending on which way the stack grows; see the function
6004 * exec_copyout_strings() for details).
6005 *
6006 * This ends up with a second (possibly unnecessary) copy compared
6007 * with assembing the data directly into the child address space,
6008 * instead, but since we cannot be guaranteed that the parent has
6009 * not modified its environment, we can't really know that it's
6010 * really a block there as well.
6011 */
6012
6013
6014 static int execargs_waiters = 0;
6015 lck_mtx_t *execargs_cache_lock;
6016
6017 static void
6018 execargs_lock_lock(void)
6019 {
6020 lck_mtx_lock_spin(execargs_cache_lock);
6021 }
6022
6023 static void
6024 execargs_lock_unlock(void)
6025 {
6026 lck_mtx_unlock(execargs_cache_lock);
6027 }
6028
6029 static wait_result_t
6030 execargs_lock_sleep(void)
6031 {
6032 return lck_mtx_sleep(execargs_cache_lock, LCK_SLEEP_DEFAULT, &execargs_free_count, THREAD_INTERRUPTIBLE);
6033 }
6034
6035 static kern_return_t
6036 execargs_purgeable_allocate(char **execarg_address)
6037 {
6038 kern_return_t kr = vm_allocate_kernel(bsd_pageable_map, (vm_offset_t *)execarg_address, BSD_PAGEABLE_SIZE_PER_EXEC, VM_FLAGS_ANYWHERE | VM_FLAGS_PURGABLE, VM_KERN_MEMORY_NONE);
6039 assert(kr == KERN_SUCCESS);
6040 return kr;
6041 }
6042
6043 static kern_return_t
6044 execargs_purgeable_reference(void *execarg_address)
6045 {
6046 int state = VM_PURGABLE_NONVOLATILE;
6047 kern_return_t kr = vm_purgable_control(bsd_pageable_map, (vm_offset_t) execarg_address, VM_PURGABLE_SET_STATE, &state);
6048
6049 assert(kr == KERN_SUCCESS);
6050 return kr;
6051 }
6052
6053 static kern_return_t
6054 execargs_purgeable_volatilize(void *execarg_address)
6055 {
6056 int state = VM_PURGABLE_VOLATILE | VM_PURGABLE_ORDERING_OBSOLETE;
6057 kern_return_t kr;
6058 kr = vm_purgable_control(bsd_pageable_map, (vm_offset_t) execarg_address, VM_PURGABLE_SET_STATE, &state);
6059
6060 assert(kr == KERN_SUCCESS);
6061
6062 return kr;
6063 }
6064
6065 static void
6066 execargs_wakeup_waiters(void)
6067 {
6068 thread_wakeup(&execargs_free_count);
6069 }
6070
6071 static int
6072 execargs_alloc(struct image_params *imgp)
6073 {
6074 kern_return_t kret;
6075 wait_result_t res;
6076 int i, cache_index = -1;
6077
6078 execargs_lock_lock();
6079
6080 while (execargs_free_count == 0) {
6081 execargs_waiters++;
6082 res = execargs_lock_sleep();
6083 execargs_waiters--;
6084 if (res != THREAD_AWAKENED) {
6085 execargs_lock_unlock();
6086 return EINTR;
6087 }
6088 }
6089
6090 execargs_free_count--;
6091
6092 for (i = 0; i < execargs_cache_size; i++) {
6093 vm_offset_t element = execargs_cache[i];
6094 if (element) {
6095 cache_index = i;
6096 imgp->ip_strings = (char *)(execargs_cache[i]);
6097 execargs_cache[i] = 0;
6098 break;
6099 }
6100 }
6101
6102 assert(execargs_free_count >= 0);
6103
6104 execargs_lock_unlock();
6105
6106 if (cache_index == -1) {
6107 kret = execargs_purgeable_allocate(&imgp->ip_strings);
6108 } else {
6109 kret = execargs_purgeable_reference(imgp->ip_strings);
6110 }
6111
6112 assert(kret == KERN_SUCCESS);
6113 if (kret != KERN_SUCCESS) {
6114 return ENOMEM;
6115 }
6116
6117 /* last page used to read in file headers */
6118 imgp->ip_vdata = imgp->ip_strings + (NCARGS + PAGE_SIZE);
6119 imgp->ip_strendp = imgp->ip_strings;
6120 imgp->ip_argspace = NCARGS;
6121 imgp->ip_strspace = (NCARGS + PAGE_SIZE);
6122
6123 return 0;
6124 }
6125
6126 /*
6127 * execargs_free
6128 *
6129 * Description: Free the block of memory used by the execve arguments and the
6130 * first page of the executable by a previous call to the function
6131 * execargs_alloc().
6132 *
6133 * Parameters: struct image_params * the image parameter block
6134 *
6135 * Returns: 0 Success
6136 * EINVAL Invalid argument
6137 * EINTR Oeration interrupted
6138 */
6139 static int
6140 execargs_free(struct image_params *imgp)
6141 {
6142 kern_return_t kret;
6143 int i;
6144 boolean_t needs_wakeup = FALSE;
6145
6146 kret = execargs_purgeable_volatilize(imgp->ip_strings);
6147
6148 execargs_lock_lock();
6149 execargs_free_count++;
6150
6151 for (i = 0; i < execargs_cache_size; i++) {
6152 vm_offset_t element = execargs_cache[i];
6153 if (element == 0) {
6154 execargs_cache[i] = (vm_offset_t) imgp->ip_strings;
6155 imgp->ip_strings = NULL;
6156 break;
6157 }
6158 }
6159
6160 assert(imgp->ip_strings == NULL);
6161
6162 if (execargs_waiters > 0) {
6163 needs_wakeup = TRUE;
6164 }
6165
6166 execargs_lock_unlock();
6167
6168 if (needs_wakeup == TRUE) {
6169 execargs_wakeup_waiters();
6170 }
6171
6172 return kret == KERN_SUCCESS ? 0 : EINVAL;
6173 }
6174
6175 static void
6176 exec_resettextvp(proc_t p, struct image_params *imgp)
6177 {
6178 vnode_t vp;
6179 off_t offset;
6180 vnode_t tvp = p->p_textvp;
6181 int ret;
6182
6183 vp = imgp->ip_vp;
6184 offset = imgp->ip_arch_offset;
6185
6186 if (vp == NULLVP) {
6187 panic("exec_resettextvp: expected valid vp");
6188 }
6189
6190 ret = vnode_ref(vp);
6191 proc_lock(p);
6192 if (ret == 0) {
6193 p->p_textvp = vp;
6194 p->p_textoff = offset;
6195 } else {
6196 p->p_textvp = NULLVP; /* this is paranoia */
6197 p->p_textoff = 0;
6198 }
6199 proc_unlock(p);
6200
6201 if (tvp != NULLVP) {
6202 if (vnode_getwithref(tvp) == 0) {
6203 vnode_rele(tvp);
6204 vnode_put(tvp);
6205 }
6206 }
6207 }
6208
6209 // Includes the 0-byte (therefore "SIZE" instead of "LEN").
6210 static const size_t CS_CDHASH_STRING_SIZE = CS_CDHASH_LEN * 2 + 1;
6211
6212 static void
6213 cdhash_to_string(char str[CS_CDHASH_STRING_SIZE], uint8_t const * const cdhash)
6214 {
6215 static char const nibble[] = "0123456789abcdef";
6216
6217 /* Apparently still the safest way to get a hex representation
6218 * of binary data.
6219 * xnu's printf routines have %*D/%20D in theory, but "not really", see:
6220 * <rdar://problem/33328859> confusion around %*D/%nD in printf
6221 */
6222 for (int i = 0; i < CS_CDHASH_LEN; ++i) {
6223 str[i * 2] = nibble[(cdhash[i] & 0xf0) >> 4];
6224 str[i * 2 + 1] = nibble[cdhash[i] & 0x0f];
6225 }
6226 str[CS_CDHASH_STRING_SIZE - 1] = 0;
6227 }
6228
6229 /*
6230 * __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__
6231 *
6232 * Description: Waits for the userspace daemon to respond to the request
6233 * we made. Function declared non inline to be visible in
6234 * stackshots and spindumps as well as debugging.
6235 */
6236 __attribute__((noinline)) int
6237 __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(mach_port_t task_access_port, int32_t new_pid)
6238 {
6239 return find_code_signature(task_access_port, new_pid);
6240 }
6241
6242 static int
6243 check_for_signature(proc_t p, struct image_params *imgp)
6244 {
6245 mach_port_t port = IPC_PORT_NULL;
6246 kern_return_t kr = KERN_FAILURE;
6247 int error = EACCES;
6248 boolean_t unexpected_failure = FALSE;
6249 struct cs_blob *csb;
6250 boolean_t require_success = FALSE;
6251 int spawn = (imgp->ip_flags & IMGPF_SPAWN);
6252 int vfexec = (imgp->ip_flags & IMGPF_VFORK_EXEC);
6253 os_reason_t signature_failure_reason = OS_REASON_NULL;
6254
6255 /*
6256 * Override inherited code signing flags with the
6257 * ones for the process that is being successfully
6258 * loaded
6259 */
6260 proc_lock(p);
6261 p->p_csflags = imgp->ip_csflags;
6262 proc_unlock(p);
6263
6264 /* Set the switch_protect flag on the map */
6265 if (p->p_csflags & (CS_HARD | CS_KILL)) {
6266 vm_map_switch_protect(get_task_map(p->task), TRUE);
6267 }
6268
6269 /*
6270 * image activation may be failed due to policy
6271 * which is unexpected but security framework does not
6272 * approve of exec, kill and return immediately.
6273 */
6274 if (imgp->ip_mac_return != 0) {
6275 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
6276 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_SECURITY_POLICY, 0, 0);
6277 signature_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_SECURITY_POLICY);
6278 error = imgp->ip_mac_return;
6279 unexpected_failure = TRUE;
6280 goto done;
6281 }
6282
6283 if (imgp->ip_cs_error != OS_REASON_NULL) {
6284 signature_failure_reason = imgp->ip_cs_error;
6285 imgp->ip_cs_error = OS_REASON_NULL;
6286 error = EACCES;
6287 goto done;
6288 }
6289
6290 /* If the code signature came through the image activation path, we skip the
6291 * taskgated / externally attached path. */
6292 if (imgp->ip_csflags & CS_SIGNED) {
6293 error = 0;
6294 goto done;
6295 }
6296
6297 /* The rest of the code is for signatures that either already have been externally
6298 * attached (likely, but not necessarily by a previous run through the taskgated
6299 * path), or that will now be attached by taskgated. */
6300
6301 kr = task_get_task_access_port(p->task, &port);
6302 if (KERN_SUCCESS != kr || !IPC_PORT_VALID(port)) {
6303 error = 0;
6304 if (require_success) {
6305 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
6306 p->p_pid, OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASK_ACCESS_PORT, 0, 0);
6307 signature_failure_reason = os_reason_create(OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASK_ACCESS_PORT);
6308 error = EACCES;
6309 }
6310 goto done;
6311 }
6312
6313 /*
6314 * taskgated returns KERN_SUCCESS if it has completed its work
6315 * and the exec should continue, KERN_FAILURE if the exec should
6316 * fail, or it may error out with different error code in an
6317 * event of mig failure (e.g. process was signalled during the
6318 * rpc call, taskgated died, mig server died etc.).
6319 */
6320
6321 kr = __EXEC_WAITING_ON_TASKGATED_CODE_SIGNATURE_UPCALL__(port, p->p_pid);
6322 switch (kr) {
6323 case KERN_SUCCESS:
6324 error = 0;
6325 break;
6326 case KERN_FAILURE:
6327 error = EACCES;
6328
6329 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
6330 p->p_pid, OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASKGATED_INVALID_SIG, 0, 0);
6331 signature_failure_reason = os_reason_create(OS_REASON_CODESIGNING, CODESIGNING_EXIT_REASON_TASKGATED_INVALID_SIG);
6332 goto done;
6333 default:
6334 error = EACCES;
6335
6336 KERNEL_DEBUG_CONSTANT(BSDDBG_CODE(DBG_BSD_PROC, BSD_PROC_EXITREASON_CREATE) | DBG_FUNC_NONE,
6337 p->p_pid, OS_REASON_EXEC, EXEC_EXIT_REASON_TASKGATED_OTHER, 0, 0);
6338 signature_failure_reason = os_reason_create(OS_REASON_EXEC, EXEC_EXIT_REASON_TASKGATED_OTHER);
6339 unexpected_failure = TRUE;
6340 goto done;
6341 }
6342
6343 /* Only do this if exec_resettextvp() did not fail */
6344 if (p->p_textvp != NULLVP) {
6345 csb = ubc_cs_blob_get(p->p_textvp, -1, p->p_textoff);
6346
6347 if (csb != NULL) {
6348 /* As the enforcement we can do here is very limited, we only allow things that
6349 * are the only reason why this code path still exists:
6350 * Adhoc signed non-platform binaries without special cs_flags and without any
6351 * entitlements (unrestricted ones still pass AMFI). */
6352 if (
6353 /* Revalidate the blob if necessary through bumped generation count. */
6354 (ubc_cs_generation_check(p->p_textvp) == 0 ||
6355 ubc_cs_blob_revalidate(p->p_textvp, csb, imgp, 0) == 0) &&
6356 /* Only CS_ADHOC, no CS_KILL, CS_HARD etc. */
6357 (csb->csb_flags & CS_ALLOWED_MACHO) == CS_ADHOC &&
6358 /* If it has a CMS blob, it's not adhoc. The CS_ADHOC flag can lie. */
6359 csblob_find_blob_bytes((const uint8_t *)csb->csb_mem_kaddr, csb->csb_mem_size,
6360 CSSLOT_SIGNATURESLOT,
6361 CSMAGIC_BLOBWRAPPER) == NULL &&
6362 /* It could still be in a trust cache (unlikely with CS_ADHOC), or a magic path. */
6363 csb->csb_platform_binary == 0 &&
6364 /* No entitlements, not even unrestricted ones. */
6365 csb->csb_entitlements_blob == NULL) {
6366 proc_lock(p);
6367 p->p_csflags |= CS_SIGNED | CS_VALID;
6368 proc_unlock(p);
6369 } else {
6370 uint8_t cdhash[CS_CDHASH_LEN];
6371 char cdhash_string[CS_CDHASH_STRING_SIZE];
6372 proc_getcdhash(p, cdhash);
6373 cdhash_to_string(cdhash_string, cdhash);
6374 printf("ignoring detached code signature on '%s' with cdhash '%s' "
6375 "because it is invalid, or not a simple adhoc signature.\n",
6376 p->p_name, cdhash_string);
6377 }
6378 }
6379 }
6380
6381 done:
6382 if (0 == error) {
6383 /* The process's code signature related properties are
6384 * fully set up, so this is an opportune moment to log
6385 * platform binary execution, if desired. */
6386 if (platform_exec_logging != 0 && csproc_get_platform_binary(p)) {
6387 uint8_t cdhash[CS_CDHASH_LEN];
6388 char cdhash_string[CS_CDHASH_STRING_SIZE];
6389 proc_getcdhash(p, cdhash);
6390 cdhash_to_string(cdhash_string, cdhash);
6391
6392 os_log(peLog, "CS Platform Exec Logging: Executing platform signed binary "
6393 "'%s' with cdhash %s\n", p->p_name, cdhash_string);
6394 }
6395 } else {
6396 if (!unexpected_failure) {
6397 p->p_csflags |= CS_KILLED;
6398 }
6399 /* make very sure execution fails */
6400 if (vfexec || spawn) {
6401 assert(signature_failure_reason != OS_REASON_NULL);
6402 psignal_vfork_with_reason(p, p->task, imgp->ip_new_thread,
6403 SIGKILL, signature_failure_reason);
6404 signature_failure_reason = OS_REASON_NULL;
6405 error = 0;
6406 } else {
6407 assert(signature_failure_reason != OS_REASON_NULL);
6408 psignal_with_reason(p, SIGKILL, signature_failure_reason);
6409 signature_failure_reason = OS_REASON_NULL;
6410 }
6411 }
6412
6413 if (port != IPC_PORT_NULL) {
6414 ipc_port_release_send(port);
6415 }
6416
6417 /* If we hit this, we likely would have leaked an exit reason */
6418 assert(signature_failure_reason == OS_REASON_NULL);
6419 return error;
6420 }
6421
6422 /*
6423 * Typically as soon as we start executing this process, the
6424 * first instruction will trigger a VM fault to bring the text
6425 * pages (as executable) into the address space, followed soon
6426 * thereafter by dyld data structures (for dynamic executable).
6427 * To optimize this, as well as improve support for hardware
6428 * debuggers that can only access resident pages present
6429 * in the process' page tables, we prefault some pages if
6430 * possible. Errors are non-fatal.
6431 */
6432 static void
6433 exec_prefault_data(proc_t p __unused, struct image_params *imgp, load_result_t *load_result)
6434 {
6435 int ret;
6436 size_t expected_all_image_infos_size;
6437
6438 /*
6439 * Prefault executable or dyld entry point.
6440 */
6441 vm_fault(current_map(),
6442 vm_map_trunc_page(load_result->entry_point,
6443 vm_map_page_mask(current_map())),
6444 VM_PROT_READ | VM_PROT_EXECUTE,
6445 FALSE, VM_KERN_MEMORY_NONE,
6446 THREAD_UNINT, NULL, 0);
6447
6448 if (imgp->ip_flags & IMGPF_IS_64BIT_ADDR) {
6449 expected_all_image_infos_size = sizeof(struct user64_dyld_all_image_infos);
6450 } else {
6451 expected_all_image_infos_size = sizeof(struct user32_dyld_all_image_infos);
6452 }
6453
6454 /* Decode dyld anchor structure from <mach-o/dyld_images.h> */
6455 if (load_result->dynlinker &&
6456 load_result->all_image_info_addr &&
6457 load_result->all_image_info_size >= expected_all_image_infos_size) {
6458 union {
6459 struct user64_dyld_all_image_infos infos64;
6460 struct user32_dyld_all_image_infos infos32;
6461 } all_image_infos;
6462
6463 /*
6464 * Pre-fault to avoid copyin() going through the trap handler
6465 * and recovery path.
6466 */
6467 vm_fault(current_map(),
6468 vm_map_trunc_page(load_result->all_image_info_addr,
6469 vm_map_page_mask(current_map())),
6470 VM_PROT_READ | VM_PROT_WRITE,
6471 FALSE, VM_KERN_MEMORY_NONE,
6472 THREAD_UNINT, NULL, 0);
6473 if ((load_result->all_image_info_addr & PAGE_MASK) + expected_all_image_infos_size > PAGE_SIZE) {
6474 /* all_image_infos straddles a page */
6475 vm_fault(current_map(),
6476 vm_map_trunc_page(load_result->all_image_info_addr + expected_all_image_infos_size - 1,
6477 vm_map_page_mask(current_map())),
6478 VM_PROT_READ | VM_PROT_WRITE,
6479 FALSE, VM_KERN_MEMORY_NONE,
6480 THREAD_UNINT, NULL, 0);
6481 }
6482
6483 ret = copyin(load_result->all_image_info_addr,
6484 &all_image_infos,
6485 expected_all_image_infos_size);
6486 if (ret == 0 && all_image_infos.infos32.version >= DYLD_ALL_IMAGE_INFOS_ADDRESS_MINIMUM_VERSION) {
6487 user_addr_t notification_address;
6488 user_addr_t dyld_image_address;
6489 user_addr_t dyld_version_address;
6490 user_addr_t dyld_all_image_infos_address;
6491 user_addr_t dyld_slide_amount;
6492
6493 if (imgp->ip_flags & IMGPF_IS_64BIT_ADDR) {
6494 notification_address = all_image_infos.infos64.notification;
6495 dyld_image_address = all_image_infos.infos64.dyldImageLoadAddress;
6496 dyld_version_address = all_image_infos.infos64.dyldVersion;
6497 dyld_all_image_infos_address = all_image_infos.infos64.dyldAllImageInfosAddress;
6498 } else {
6499 notification_address = all_image_infos.infos32.notification;
6500 dyld_image_address = all_image_infos.infos32.dyldImageLoadAddress;
6501 dyld_version_address = all_image_infos.infos32.dyldVersion;
6502 dyld_all_image_infos_address = all_image_infos.infos32.dyldAllImageInfosAddress;
6503 }
6504
6505 /*
6506 * dyld statically sets up the all_image_infos in its Mach-O
6507 * binary at static link time, with pointers relative to its default
6508 * load address. Since ASLR might slide dyld before its first
6509 * instruction is executed, "dyld_slide_amount" tells us how far
6510 * dyld was loaded compared to its default expected load address.
6511 * All other pointers into dyld's image should be adjusted by this
6512 * amount. At some point later, dyld will fix up pointers to take
6513 * into account the slide, at which point the all_image_infos_address
6514 * field in the structure will match the runtime load address, and
6515 * "dyld_slide_amount" will be 0, if we were to consult it again.
6516 */
6517
6518 dyld_slide_amount = load_result->all_image_info_addr - dyld_all_image_infos_address;
6519
6520 #if 0
6521 kprintf("exec_prefault: 0x%016llx 0x%08x 0x%016llx 0x%016llx 0x%016llx 0x%016llx\n",
6522 (uint64_t)load_result->all_image_info_addr,
6523 all_image_infos.infos32.version,
6524 (uint64_t)notification_address,
6525 (uint64_t)dyld_image_address,
6526 (uint64_t)dyld_version_address,
6527 (uint64_t)dyld_all_image_infos_address);
6528 #endif
6529
6530 vm_fault(current_map(),
6531 vm_map_trunc_page(notification_address + dyld_slide_amount,
6532 vm_map_page_mask(current_map())),
6533 VM_PROT_READ | VM_PROT_EXECUTE,
6534 FALSE, VM_KERN_MEMORY_NONE,
6535 THREAD_UNINT, NULL, 0);
6536 vm_fault(current_map(),
6537 vm_map_trunc_page(dyld_image_address + dyld_slide_amount,
6538 vm_map_page_mask(current_map())),
6539 VM_PROT_READ | VM_PROT_EXECUTE,
6540 FALSE, VM_KERN_MEMORY_NONE,
6541 THREAD_UNINT, NULL, 0);
6542 vm_fault(current_map(),
6543 vm_map_trunc_page(dyld_version_address + dyld_slide_amount,
6544 vm_map_page_mask(current_map())),
6545 VM_PROT_READ,
6546 FALSE, VM_KERN_MEMORY_NONE,
6547 THREAD_UNINT, NULL, 0);
6548 vm_fault(current_map(),
6549 vm_map_trunc_page(dyld_all_image_infos_address + dyld_slide_amount,
6550 vm_map_page_mask(current_map())),
6551 VM_PROT_READ | VM_PROT_WRITE,
6552 FALSE, VM_KERN_MEMORY_NONE,
6553 THREAD_UNINT, NULL, 0);
6554 }
6555 }
6556 }
mirror of XNU