xref: /xnu-8020.140.41/bsd/kern/kern_debug.c (revision 27b03b360a988dfd3dfdf34262bb0042026747cc)
1 /*
2  * Copyright (c) 2020 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 
29 #if CONFIG_DEBUG_SYSCALL_REJECTION
30 
31 #include <kern/bits.h>
32 #include <kern/exc_guard.h>
33 #include <kern/exception.h>
34 #include <kern/kalloc.h>
35 #include <kern/simple_lock.h>
36 #include <kern/startup.h>
37 #include <kern/syscall_sw.h>
38 
39 #include <pexpert/pexpert.h>
40 
41 #include <sys/syscall.h>
42 #include <sys/sysctl.h>
43 #include <sys/sysent.h>
44 #include <sys/systm.h>
45 #include <sys/types.h>
46 #include <sys/user.h>
47 
48 #include <sys/kern_debug.h>
49 
50 #define SYSCALL_REJECTION_MODE_IGNORE   0
51 #define SYSCALL_REJECTION_MODE_GUARD    1
52 #define SYSCALL_REJECTION_MODE_CRASH    2
53 
54 int debug_syscall_rejection_mode = 0;
55 SYSCTL_INT(_kern, OID_AUTO, debug_syscall_rejection_mode, CTLFLAG_RW | CTLFLAG_LOCKED,
56     &debug_syscall_rejection_mode, 0, "0: ignore, 1: non-fatal, 2: crash");
57 
58 static size_t const predefined_masks = 2; // 0: null mask (all 0), 1: all mask (all 1)
59 
60 /*
61  * The number of masks is derived from the mask selector data type:
62  *
63  * A selector is just made of an index into syscall_rejection_masks,
64  * with the exception of the highest bit, which indicates whether the
65  * mask is to be added as an "allow" mask or a "deny" mask.
66  * Additionally, predefined masks don't actually have storage and are
67  * handled specially, so syscall_rejection_masks starts with the first
68  * non-predefined mask (and is sized appropriately).
69  */
70 static size_t const syscall_rejection_mask_count =
71     (1 << (8 * sizeof(syscall_rejection_selector_t) - predefined_masks)) - 1;
72 static syscall_rejection_mask_t syscall_rejection_masks[syscall_rejection_mask_count];
73 
74 #define SR_MASK_SIZE (BITMAP_SIZE(mach_trap_count + nsysent))
75 
76 static LCK_GRP_DECLARE(syscall_rejection_lck_grp, "syscall rejection lock");
77 static LCK_MTX_DECLARE(syscall_rejection_mtx, &syscall_rejection_lck_grp);
78 
79 extern kern_return_t task_violated_guard(mach_exception_code_t, mach_exception_subcode_t, void *);
80 
81 bool
debug_syscall_rejection_handle(int syscall_mach_trap_number)82 debug_syscall_rejection_handle(int syscall_mach_trap_number)
83 {
84 	bool fatal = false;
85 
86 	switch (debug_syscall_rejection_mode) {
87 	case SYSCALL_REJECTION_MODE_CRASH:
88 		fatal = true;
89 		OS_FALLTHROUGH;
90 	case SYSCALL_REJECTION_MODE_GUARD: {
91 		mach_exception_code_t code = 0;
92 		EXC_GUARD_ENCODE_TYPE(code, GUARD_TYPE_REJECTED_SC);
93 		EXC_GUARD_ENCODE_FLAVOR(code, 0);
94 		EXC_GUARD_ENCODE_TARGET(code, syscall_mach_trap_number < 0);
95 		mach_exception_subcode_t subcode =
96 		    syscall_mach_trap_number < 0 ? -syscall_mach_trap_number : syscall_mach_trap_number;
97 
98 		if (!fatal) {
99 			task_violated_guard(code, subcode, NULL);
100 		} else {
101 			thread_guard_violation(current_thread(), code, subcode, fatal);
102 		}
103 		break;
104 	};
105 	default:
106 		/* ignore */
107 		;
108 	}
109 	return fatal;
110 }
111 
112 void
rejected_syscall_guard_ast(thread_t __unused t,mach_exception_data_type_t code,mach_exception_data_type_t subcode)113 rejected_syscall_guard_ast(
114 	thread_t __unused t,
115 	mach_exception_data_type_t code,
116 	mach_exception_data_type_t subcode)
117 {
118 	task_exception_notify(EXC_GUARD, code, subcode);
119 	proc_t p = current_proc();
120 	psignal(p, SIGSYS);
121 }
122 
123 
124 static void
_syscall_rejection_apply_mask(syscall_rejection_mask_t dest,const syscall_rejection_mask_t src,bool apply_as_allow)125 _syscall_rejection_apply_mask(syscall_rejection_mask_t dest, const syscall_rejection_mask_t src, bool apply_as_allow)
126 {
127 	assert(dest != NULL);
128 	assert(src != NULL);
129 
130 	if (apply_as_allow) {
131 		bitmap_or(dest, dest, src, mach_trap_count + nsysent);
132 	} else {
133 		bitmap_and_not(dest, dest, src, mach_trap_count + nsysent);
134 	}
135 }
136 
137 /*
138  * The masks to apply are passed to the kernel as packed selectors,
139  * which are just however many of the selector data type fit into one
140  * (or more) fields of the natural word size (i.e. a register). This
141  * avoids copying from user space.
142  *
143  * More specifically, at the time of this writing, a selector is 1
144  * byte wide, and there is only one uint64_t argument
145  * (args->packed_selectors), so up to 8 selectors can be specified,
146  * which are then stuffed into the 64 bits of the argument. If less
147  * than 8 masks are requested to be applied, the remaining selectors
148  * will just be left as 0, which naturally resolves as the "empty" or
149  * "NULL" mask that changes nothing.
150  *
151  * The libsyscall wrapper provides a more convenient interface where
152  * an array (up to 8 elements long) and its length are passed in,
153  * which the wrapper then packs into packed_selectors of the actual
154  * system call.
155  */
156 
157 int
debug_syscall_reject(struct proc * p __unused,struct debug_syscall_reject_args * args,int * retval)158 debug_syscall_reject(struct proc *p __unused, struct debug_syscall_reject_args *args, int *retval)
159 {
160 	int error = 0;
161 
162 	*retval = 0;
163 
164 	if (debug_syscall_rejection_mode == SYSCALL_REJECTION_MODE_IGNORE) {
165 		return 0;
166 	}
167 
168 	uthread_t ut = current_uthread();
169 
170 	bitmap_t mask[SR_MASK_SIZE / sizeof(bitmap_t)];
171 	// syscall rejection masks are always reset to "deny all"
172 	memset(mask, 0, SR_MASK_SIZE);
173 
174 	lck_mtx_lock(&syscall_rejection_mtx);
175 
176 	for (int i = 0; i < sizeof(args->packed_selectors) / sizeof(syscall_rejection_selector_t); i++) {
177 		syscall_rejection_selector_t selector = ((syscall_rejection_selector_t const *)&(args->packed_selectors))[i];
178 		bool const is_allow_mask = selector & SYSCALL_REJECTION_IS_ALLOW_MASK;
179 		int const mask_index = selector & SYSCALL_REJECTION_INDEX_MASK;
180 
181 		if (mask_index == SYSCALL_REJECTION_NULL) {
182 			// mask 0 is always empty (nothing to apply)
183 			continue;
184 		}
185 
186 		if (mask_index == SYSCALL_REJECTION_ALL) {
187 			// mask 1 is always full (overrides everything)
188 			memset(mask, is_allow_mask ? 0xff : 0x00, SR_MASK_SIZE);
189 			continue;
190 		}
191 
192 		syscall_rejection_mask_t mask_to_apply = syscall_rejection_masks[mask_index - predefined_masks];
193 
194 		if (mask_to_apply == NULL) {
195 			error = ENOENT;
196 			goto out_locked;
197 		}
198 
199 		_syscall_rejection_apply_mask(mask, mask_to_apply, is_allow_mask);
200 	}
201 
202 	if (ut->syscall_rejection_mask == NULL) {
203 		/* Not RT-safe, but only necessary once. */
204 
205 		ut->syscall_rejection_mask = kalloc_data(SR_MASK_SIZE, Z_WAITOK);
206 
207 		if (ut->syscall_rejection_mask == NULL) {
208 			error = ENOMEM;
209 			goto out_locked;
210 		}
211 	}
212 
213 	memcpy(ut->syscall_rejection_mask, mask, SR_MASK_SIZE);
214 
215 out_locked:
216 	lck_mtx_unlock(&syscall_rejection_mtx);
217 
218 	return error;
219 }
220 
221 static bool
_syscall_rejection_add(syscall_rejection_mask_t dst,char const * name)222 _syscall_rejection_add(syscall_rejection_mask_t dst, char const *name)
223 {
224 	/*
225 	 * Yes, this function is O(n+m), making the whole act of setting a
226 	 * mask O(l*(n+m)), but defining masks is done rarely enough (and
227 	 * i, n and m small enough) for this to not matter.
228 	 */
229 
230 	for (int i = 0; i < mach_trap_count; i++) {
231 		if (strcmp(mach_syscall_name_table[i], name) == 0) {
232 			bitmap_set(dst, i);
233 			return true;
234 		}
235 	}
236 
237 	extern char const *syscallnames[];
238 
239 	for (int i = 0; i < nsysent; i++) {
240 		if (strcmp(syscallnames[i], name) == 0) {
241 			bitmap_set(dst, i + mach_trap_count);
242 			return true;
243 		}
244 	}
245 
246 	printf("%s: trying to add non-existing syscall/mach trap '%s'\n", __func__, name);
247 	return false;
248 }
249 
250 /* Pretty much arbitrary, we just don't want userspace to pass
251  * unreasonably large buffers to parse. */
252 static size_t const max_input_size = 16 * PAGE_MAX_SIZE;
253 
254 static int
_sysctl_debug_syscall_rejection_masks(struct sysctl_oid __unused * oidp,void * __unused arg1,int __unused arg2,struct sysctl_req * req)255 _sysctl_debug_syscall_rejection_masks(struct sysctl_oid __unused *oidp, void * __unused arg1, int __unused arg2,
256     struct sysctl_req *req)
257 {
258 	if (req->newptr == 0) {
259 		return 0;
260 	}
261 
262 	if (req->newlen > max_input_size) {
263 		return E2BIG;
264 	}
265 
266 	size_t const len = req->newlen;
267 	char *buf = kalloc_data(len + 1, Z_WAITOK);
268 
269 	if (buf == NULL) {
270 		return ENOMEM;
271 	}
272 
273 	/*
274 	 * sysctl_io_string always copies out the given buffer as the
275 	 * "old" value if requested.  We could construct a text
276 	 * representation of existing masks, but this is not particularly
277 	 * interesting, so we just return the dummy string "<masks>".
278 	 */
279 	strlcpy(buf, "<masks>", len + 1);
280 	int changed = 0;
281 	int error = sysctl_io_string(req, buf, len + 1, 0, &changed);
282 
283 	if (error != 0 || !changed) {
284 		goto out;
285 	}
286 
287 	char const *p = buf;
288 
289 	int id = 0;
290 	int l = 0;
291 	int n = sscanf(p, "%i: %n", &id, &l);
292 
293 	if (n != 1 || id < predefined_masks || id > syscall_rejection_mask_count + predefined_masks) {
294 		printf("%s: invalid mask id %i (or conversion failed)\n", __FUNCTION__, id);
295 		error = EINVAL;
296 		goto out;
297 	}
298 
299 	p += l;
300 
301 	syscall_rejection_mask_t new_mask = kalloc_data(SR_MASK_SIZE,
302 	    Z_WAITOK | Z_ZERO);
303 	if (new_mask == NULL) {
304 		printf("%s: allocating new mask for id %i failed\n", __FUNCTION__, id);
305 		error = ENOMEM;
306 		goto out;
307 	}
308 
309 	size_t const max_name_len = 128;
310 	char name[max_name_len];
311 
312 	error = 0;
313 
314 	while (p < buf + len && *p != 0) {
315 		name[0] = 0;
316 		n = sscanf(p, "%127s %n", name, &l);
317 		if (n != 1 || name[0] == 0) {
318 			error = EINVAL;
319 			kfree_data(new_mask, SR_MASK_SIZE);
320 			goto out;
321 		}
322 
323 		if (!_syscall_rejection_add(new_mask, name)) {
324 			error = ENOENT;
325 			kfree_data(new_mask, SR_MASK_SIZE);
326 			goto out;
327 		}
328 
329 		p += l;
330 	}
331 
332 
333 	syscall_rejection_mask_t to_free = NULL;
334 
335 	lck_mtx_lock(&syscall_rejection_mtx);
336 
337 	syscall_rejection_mask_t *target_mask = &syscall_rejection_masks[id - predefined_masks];
338 
339 	to_free = *target_mask;
340 	*target_mask = new_mask;
341 
342 	lck_mtx_unlock(&syscall_rejection_mtx);
343 
344 	kfree_data(to_free, SR_MASK_SIZE);
345 out:
346 
347 	kfree_data(buf, len);
348 	return error;
349 }
350 
351 SYSCTL_PROC(_kern, OID_AUTO, syscall_rejection_masks, CTLTYPE_STRING | CTLFLAG_WR | CTLFLAG_MASKED | CTLFLAG_LOCKED,
352     0, 0, _sysctl_debug_syscall_rejection_masks, "A", "system call rejection masks");
353 
354 #else /* CONFIG_DEBUG_SYSCALL_REJECTION */
355 
356 #include <sys/kern_debug.h>
357 
358 int
debug_syscall_reject(struct proc * p __unused,struct debug_syscall_reject_args * __unused args,int __unused * ret)359 debug_syscall_reject(struct proc *p __unused, struct debug_syscall_reject_args * __unused args, int __unused *ret)
360 {
361 	/* not supported. */
362 	return ENOTSUP;
363 }
364 
365 #endif /* CONFIG_DEBUG_SYSCALL_REJECTION */
366