xref: /linux/security/selinux/ss/sidtab.c (revision acbf6de674ef7b1b5870b25e7b3c695bf84273d0)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Implementation of the SID table type.
4  *
5  * Original author: Stephen Smalley, <stephen.smalley.work@gmail.com>
6  * Author: Ondrej Mosnacek, <omosnacek@gmail.com>
7  *
8  * Copyright (C) 2018 Red Hat, Inc.
9  */
10 #include <linux/errno.h>
11 #include <linux/kernel.h>
12 #include <linux/list.h>
13 #include <linux/rcupdate.h>
14 #include <linux/slab.h>
15 #include <linux/sched.h>
16 #include <linux/spinlock.h>
17 #include <asm/barrier.h>
18 #include "flask.h"
19 #include "security.h"
20 #include "sidtab.h"
21 #include "services.h"
22 
23 struct sidtab_str_cache {
24 	struct rcu_head rcu_member;
25 	struct list_head lru_member;
26 	struct sidtab_entry *parent;
27 	u32 len;
28 	char str[];
29 };
30 
31 #define index_to_sid(index) ((index) + SECINITSID_NUM + 1)
32 #define sid_to_index(sid) ((sid) - (SECINITSID_NUM + 1))
33 
34 int sidtab_init(struct sidtab *s)
35 {
36 	u32 i;
37 
38 	memset(s->roots, 0, sizeof(s->roots));
39 
40 	for (i = 0; i < SECINITSID_NUM; i++)
41 		s->isids[i].set = 0;
42 
43 	s->frozen = false;
44 	s->count = 0;
45 	s->convert = NULL;
46 	hash_init(s->context_to_sid);
47 
48 	spin_lock_init(&s->lock);
49 
50 #if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
51 	s->cache_free_slots = CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE;
52 	INIT_LIST_HEAD(&s->cache_lru_list);
53 	spin_lock_init(&s->cache_lock);
54 #endif
55 
56 	return 0;
57 }
58 
59 static u32 context_to_sid(struct sidtab *s, struct context *context, u32 hash)
60 {
61 	struct sidtab_entry *entry;
62 	u32 sid = 0;
63 
64 	rcu_read_lock();
65 	hash_for_each_possible_rcu(s->context_to_sid, entry, list, hash) {
66 		if (entry->hash != hash)
67 			continue;
68 		if (context_cmp(&entry->context, context)) {
69 			sid = entry->sid;
70 			break;
71 		}
72 	}
73 	rcu_read_unlock();
74 	return sid;
75 }
76 
77 int sidtab_set_initial(struct sidtab *s, u32 sid, struct context *context)
78 {
79 	struct sidtab_isid_entry *isid;
80 	u32 hash;
81 	int rc;
82 
83 	if (sid == 0 || sid > SECINITSID_NUM)
84 		return -EINVAL;
85 
86 	isid = &s->isids[sid - 1];
87 
88 	rc = context_cpy(&isid->entry.context, context);
89 	if (rc)
90 		return rc;
91 
92 #if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
93 	isid->entry.cache = NULL;
94 #endif
95 	isid->set = 1;
96 
97 	hash = context_compute_hash(context);
98 
99 	/*
100 	 * Multiple initial sids may map to the same context. Check that this
101 	 * context is not already represented in the context_to_sid hashtable
102 	 * to avoid duplicate entries and long linked lists upon hash
103 	 * collision.
104 	 */
105 	if (!context_to_sid(s, context, hash)) {
106 		isid->entry.sid = sid;
107 		isid->entry.hash = hash;
108 		hash_add(s->context_to_sid, &isid->entry.list, hash);
109 	}
110 
111 	return 0;
112 }
113 
114 int sidtab_hash_stats(struct sidtab *sidtab, char *page)
115 {
116 	int i;
117 	int chain_len = 0;
118 	int slots_used = 0;
119 	int entries = 0;
120 	int max_chain_len = 0;
121 	int cur_bucket = 0;
122 	struct sidtab_entry *entry;
123 
124 	rcu_read_lock();
125 	hash_for_each_rcu(sidtab->context_to_sid, i, entry, list) {
126 		entries++;
127 		if (i == cur_bucket) {
128 			chain_len++;
129 			if (chain_len == 1)
130 				slots_used++;
131 		} else {
132 			cur_bucket = i;
133 			if (chain_len > max_chain_len)
134 				max_chain_len = chain_len;
135 			chain_len = 0;
136 		}
137 	}
138 	rcu_read_unlock();
139 
140 	if (chain_len > max_chain_len)
141 		max_chain_len = chain_len;
142 
143 	return scnprintf(page, PAGE_SIZE, "entries: %d\nbuckets used: %d/%d\n"
144 			 "longest chain: %d\n", entries,
145 			 slots_used, SIDTAB_HASH_BUCKETS, max_chain_len);
146 }
147 
148 static u32 sidtab_level_from_count(u32 count)
149 {
150 	u32 capacity = SIDTAB_LEAF_ENTRIES;
151 	u32 level = 0;
152 
153 	while (count > capacity) {
154 		capacity <<= SIDTAB_INNER_SHIFT;
155 		++level;
156 	}
157 	return level;
158 }
159 
160 static int sidtab_alloc_roots(struct sidtab *s, u32 level)
161 {
162 	u32 l;
163 
164 	if (!s->roots[0].ptr_leaf) {
165 		s->roots[0].ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
166 					       GFP_ATOMIC);
167 		if (!s->roots[0].ptr_leaf)
168 			return -ENOMEM;
169 	}
170 	for (l = 1; l <= level; ++l)
171 		if (!s->roots[l].ptr_inner) {
172 			s->roots[l].ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
173 							GFP_ATOMIC);
174 			if (!s->roots[l].ptr_inner)
175 				return -ENOMEM;
176 			s->roots[l].ptr_inner->entries[0] = s->roots[l - 1];
177 		}
178 	return 0;
179 }
180 
181 static struct sidtab_entry *sidtab_do_lookup(struct sidtab *s, u32 index,
182 					     int alloc)
183 {
184 	union sidtab_entry_inner *entry;
185 	u32 level, capacity_shift, leaf_index = index / SIDTAB_LEAF_ENTRIES;
186 
187 	/* find the level of the subtree we need */
188 	level = sidtab_level_from_count(index + 1);
189 	capacity_shift = level * SIDTAB_INNER_SHIFT;
190 
191 	/* allocate roots if needed */
192 	if (alloc && sidtab_alloc_roots(s, level) != 0)
193 		return NULL;
194 
195 	/* lookup inside the subtree */
196 	entry = &s->roots[level];
197 	while (level != 0) {
198 		capacity_shift -= SIDTAB_INNER_SHIFT;
199 		--level;
200 
201 		entry = &entry->ptr_inner->entries[leaf_index >> capacity_shift];
202 		leaf_index &= ((u32)1 << capacity_shift) - 1;
203 
204 		if (!entry->ptr_inner) {
205 			if (alloc)
206 				entry->ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
207 							   GFP_ATOMIC);
208 			if (!entry->ptr_inner)
209 				return NULL;
210 		}
211 	}
212 	if (!entry->ptr_leaf) {
213 		if (alloc)
214 			entry->ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
215 						  GFP_ATOMIC);
216 		if (!entry->ptr_leaf)
217 			return NULL;
218 	}
219 	return &entry->ptr_leaf->entries[index % SIDTAB_LEAF_ENTRIES];
220 }
221 
222 static struct sidtab_entry *sidtab_lookup(struct sidtab *s, u32 index)
223 {
224 	/* read entries only after reading count */
225 	u32 count = smp_load_acquire(&s->count);
226 
227 	if (index >= count)
228 		return NULL;
229 
230 	return sidtab_do_lookup(s, index, 0);
231 }
232 
233 static struct sidtab_entry *sidtab_lookup_initial(struct sidtab *s, u32 sid)
234 {
235 	return s->isids[sid - 1].set ? &s->isids[sid - 1].entry : NULL;
236 }
237 
238 static struct sidtab_entry *sidtab_search_core(struct sidtab *s, u32 sid,
239 					       int force)
240 {
241 	if (sid != 0) {
242 		struct sidtab_entry *entry;
243 
244 		if (sid > SECINITSID_NUM)
245 			entry = sidtab_lookup(s, sid_to_index(sid));
246 		else
247 			entry = sidtab_lookup_initial(s, sid);
248 		if (entry && (!entry->context.len || force))
249 			return entry;
250 	}
251 
252 	return sidtab_lookup_initial(s, SECINITSID_UNLABELED);
253 }
254 
255 struct sidtab_entry *sidtab_search_entry(struct sidtab *s, u32 sid)
256 {
257 	return sidtab_search_core(s, sid, 0);
258 }
259 
260 struct sidtab_entry *sidtab_search_entry_force(struct sidtab *s, u32 sid)
261 {
262 	return sidtab_search_core(s, sid, 1);
263 }
264 
265 int sidtab_context_to_sid(struct sidtab *s, struct context *context,
266 			  u32 *sid)
267 {
268 	unsigned long flags;
269 	u32 count, hash = context_compute_hash(context);
270 	struct sidtab_convert_params *convert;
271 	struct sidtab_entry *dst, *dst_convert;
272 	int rc;
273 
274 	*sid = context_to_sid(s, context, hash);
275 	if (*sid)
276 		return 0;
277 
278 	/* lock-free search failed: lock, re-search, and insert if not found */
279 	spin_lock_irqsave(&s->lock, flags);
280 
281 	rc = 0;
282 	*sid = context_to_sid(s, context, hash);
283 	if (*sid)
284 		goto out_unlock;
285 
286 	if (unlikely(s->frozen)) {
287 		/*
288 		 * This sidtab is now frozen - tell the caller to abort and
289 		 * get the new one.
290 		 */
291 		rc = -ESTALE;
292 		goto out_unlock;
293 	}
294 
295 	count = s->count;
296 
297 	/* bail out if we already reached max entries */
298 	rc = -EOVERFLOW;
299 	if (count >= SIDTAB_MAX)
300 		goto out_unlock;
301 
302 	/* insert context into new entry */
303 	rc = -ENOMEM;
304 	dst = sidtab_do_lookup(s, count, 1);
305 	if (!dst)
306 		goto out_unlock;
307 
308 	dst->sid = index_to_sid(count);
309 	dst->hash = hash;
310 
311 	rc = context_cpy(&dst->context, context);
312 	if (rc)
313 		goto out_unlock;
314 
315 	/*
316 	 * if we are building a new sidtab, we need to convert the context
317 	 * and insert it there as well
318 	 */
319 	convert = s->convert;
320 	if (convert) {
321 		struct sidtab *target = convert->target;
322 
323 		rc = -ENOMEM;
324 		dst_convert = sidtab_do_lookup(target, count, 1);
325 		if (!dst_convert) {
326 			context_destroy(&dst->context);
327 			goto out_unlock;
328 		}
329 
330 		rc = services_convert_context(convert->args,
331 					      context, &dst_convert->context,
332 					      GFP_ATOMIC);
333 		if (rc) {
334 			context_destroy(&dst->context);
335 			goto out_unlock;
336 		}
337 		dst_convert->sid = index_to_sid(count);
338 		dst_convert->hash = context_compute_hash(&dst_convert->context);
339 		target->count = count + 1;
340 
341 		hash_add_rcu(target->context_to_sid,
342 			     &dst_convert->list, dst_convert->hash);
343 	}
344 
345 	if (context->len)
346 		pr_info("SELinux:  Context %s is not valid (left unmapped).\n",
347 			context->str);
348 
349 	*sid = index_to_sid(count);
350 
351 	/* write entries before updating count */
352 	smp_store_release(&s->count, count + 1);
353 	hash_add_rcu(s->context_to_sid, &dst->list, dst->hash);
354 
355 	rc = 0;
356 out_unlock:
357 	spin_unlock_irqrestore(&s->lock, flags);
358 	return rc;
359 }
360 
361 static void sidtab_convert_hashtable(struct sidtab *s, u32 count)
362 {
363 	struct sidtab_entry *entry;
364 	u32 i;
365 
366 	for (i = 0; i < count; i++) {
367 		entry = sidtab_do_lookup(s, i, 0);
368 		entry->sid = index_to_sid(i);
369 		entry->hash = context_compute_hash(&entry->context);
370 
371 		hash_add_rcu(s->context_to_sid, &entry->list, entry->hash);
372 	}
373 }
374 
375 static int sidtab_convert_tree(union sidtab_entry_inner *edst,
376 			       union sidtab_entry_inner *esrc,
377 			       u32 *pos, u32 count, u32 level,
378 			       struct sidtab_convert_params *convert)
379 {
380 	int rc;
381 	u32 i;
382 
383 	if (level != 0) {
384 		if (!edst->ptr_inner) {
385 			edst->ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
386 						  GFP_KERNEL);
387 			if (!edst->ptr_inner)
388 				return -ENOMEM;
389 		}
390 		i = 0;
391 		while (i < SIDTAB_INNER_ENTRIES && *pos < count) {
392 			rc = sidtab_convert_tree(&edst->ptr_inner->entries[i],
393 						 &esrc->ptr_inner->entries[i],
394 						 pos, count, level - 1,
395 						 convert);
396 			if (rc)
397 				return rc;
398 			i++;
399 		}
400 	} else {
401 		if (!edst->ptr_leaf) {
402 			edst->ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
403 						 GFP_KERNEL);
404 			if (!edst->ptr_leaf)
405 				return -ENOMEM;
406 		}
407 		i = 0;
408 		while (i < SIDTAB_LEAF_ENTRIES && *pos < count) {
409 			rc = services_convert_context(convert->args,
410 					&esrc->ptr_leaf->entries[i].context,
411 					&edst->ptr_leaf->entries[i].context,
412 					GFP_KERNEL);
413 			if (rc)
414 				return rc;
415 			(*pos)++;
416 			i++;
417 		}
418 		cond_resched();
419 	}
420 	return 0;
421 }
422 
423 int sidtab_convert(struct sidtab *s, struct sidtab_convert_params *params)
424 {
425 	unsigned long flags;
426 	u32 count, level, pos;
427 	int rc;
428 
429 	spin_lock_irqsave(&s->lock, flags);
430 
431 	/* concurrent policy loads are not allowed */
432 	if (s->convert) {
433 		spin_unlock_irqrestore(&s->lock, flags);
434 		return -EBUSY;
435 	}
436 
437 	count = s->count;
438 	level = sidtab_level_from_count(count);
439 
440 	/* allocate last leaf in the new sidtab (to avoid race with
441 	 * live convert)
442 	 */
443 	rc = sidtab_do_lookup(params->target, count - 1, 1) ? 0 : -ENOMEM;
444 	if (rc) {
445 		spin_unlock_irqrestore(&s->lock, flags);
446 		return rc;
447 	}
448 
449 	/* set count in case no new entries are added during conversion */
450 	params->target->count = count;
451 
452 	/* enable live convert of new entries */
453 	s->convert = params;
454 
455 	/* we can safely convert the tree outside the lock */
456 	spin_unlock_irqrestore(&s->lock, flags);
457 
458 	pr_info("SELinux:  Converting %u SID table entries...\n", count);
459 
460 	/* convert all entries not covered by live convert */
461 	pos = 0;
462 	rc = sidtab_convert_tree(&params->target->roots[level],
463 				 &s->roots[level], &pos, count, level, params);
464 	if (rc) {
465 		/* we need to keep the old table - disable live convert */
466 		spin_lock_irqsave(&s->lock, flags);
467 		s->convert = NULL;
468 		spin_unlock_irqrestore(&s->lock, flags);
469 		return rc;
470 	}
471 	/*
472 	 * The hashtable can also be modified in sidtab_context_to_sid()
473 	 * so we must re-acquire the lock here.
474 	 */
475 	spin_lock_irqsave(&s->lock, flags);
476 	sidtab_convert_hashtable(params->target, count);
477 	spin_unlock_irqrestore(&s->lock, flags);
478 
479 	return 0;
480 }
481 
482 void sidtab_cancel_convert(struct sidtab *s)
483 {
484 	unsigned long flags;
485 
486 	/* cancelling policy load - disable live convert of sidtab */
487 	spin_lock_irqsave(&s->lock, flags);
488 	s->convert = NULL;
489 	spin_unlock_irqrestore(&s->lock, flags);
490 }
491 
492 void sidtab_freeze_begin(struct sidtab *s, unsigned long *flags) __acquires(&s->lock)
493 {
494 	spin_lock_irqsave(&s->lock, *flags);
495 	s->frozen = true;
496 	s->convert = NULL;
497 }
498 void sidtab_freeze_end(struct sidtab *s, unsigned long *flags) __releases(&s->lock)
499 {
500 	spin_unlock_irqrestore(&s->lock, *flags);
501 }
502 
503 static void sidtab_destroy_entry(struct sidtab_entry *entry)
504 {
505 	context_destroy(&entry->context);
506 #if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
507 	kfree(rcu_dereference_raw(entry->cache));
508 #endif
509 }
510 
511 static void sidtab_destroy_tree(union sidtab_entry_inner entry, u32 level)
512 {
513 	u32 i;
514 
515 	if (level != 0) {
516 		struct sidtab_node_inner *node = entry.ptr_inner;
517 
518 		if (!node)
519 			return;
520 
521 		for (i = 0; i < SIDTAB_INNER_ENTRIES; i++)
522 			sidtab_destroy_tree(node->entries[i], level - 1);
523 		kfree(node);
524 	} else {
525 		struct sidtab_node_leaf *node = entry.ptr_leaf;
526 
527 		if (!node)
528 			return;
529 
530 		for (i = 0; i < SIDTAB_LEAF_ENTRIES; i++)
531 			sidtab_destroy_entry(&node->entries[i]);
532 		kfree(node);
533 	}
534 }
535 
536 void sidtab_destroy(struct sidtab *s)
537 {
538 	u32 i, level;
539 
540 	for (i = 0; i < SECINITSID_NUM; i++)
541 		if (s->isids[i].set)
542 			sidtab_destroy_entry(&s->isids[i].entry);
543 
544 	level = SIDTAB_MAX_LEVEL;
545 	while (level && !s->roots[level].ptr_inner)
546 		--level;
547 
548 	sidtab_destroy_tree(s->roots[level], level);
549 	/*
550 	 * The context_to_sid hashtable's objects are all shared
551 	 * with the isids array and context tree, and so don't need
552 	 * to be cleaned up here.
553 	 */
554 }
555 
556 #if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
557 
558 void sidtab_sid2str_put(struct sidtab *s, struct sidtab_entry *entry,
559 			const char *str, u32 str_len)
560 {
561 	struct sidtab_str_cache *cache, *victim = NULL;
562 	unsigned long flags;
563 
564 	/* do not cache invalid contexts */
565 	if (entry->context.len)
566 		return;
567 
568 	spin_lock_irqsave(&s->cache_lock, flags);
569 
570 	cache = rcu_dereference_protected(entry->cache,
571 					  lockdep_is_held(&s->cache_lock));
572 	if (cache) {
573 		/* entry in cache - just bump to the head of LRU list */
574 		list_move(&cache->lru_member, &s->cache_lru_list);
575 		goto out_unlock;
576 	}
577 
578 	cache = kmalloc(struct_size(cache, str, str_len), GFP_ATOMIC);
579 	if (!cache)
580 		goto out_unlock;
581 
582 	if (s->cache_free_slots == 0) {
583 		/* pop a cache entry from the tail and free it */
584 		victim = container_of(s->cache_lru_list.prev,
585 				      struct sidtab_str_cache, lru_member);
586 		list_del(&victim->lru_member);
587 		rcu_assign_pointer(victim->parent->cache, NULL);
588 	} else {
589 		s->cache_free_slots--;
590 	}
591 	cache->parent = entry;
592 	cache->len = str_len;
593 	memcpy(cache->str, str, str_len);
594 	list_add(&cache->lru_member, &s->cache_lru_list);
595 
596 	rcu_assign_pointer(entry->cache, cache);
597 
598 out_unlock:
599 	spin_unlock_irqrestore(&s->cache_lock, flags);
600 	kfree_rcu(victim, rcu_member);
601 }
602 
603 int sidtab_sid2str_get(struct sidtab *s, struct sidtab_entry *entry,
604 		       char **out, u32 *out_len)
605 {
606 	struct sidtab_str_cache *cache;
607 	int rc = 0;
608 
609 	if (entry->context.len)
610 		return -ENOENT; /* do not cache invalid contexts */
611 
612 	rcu_read_lock();
613 
614 	cache = rcu_dereference(entry->cache);
615 	if (!cache) {
616 		rc = -ENOENT;
617 	} else {
618 		*out_len = cache->len;
619 		if (out) {
620 			*out = kmemdup(cache->str, cache->len, GFP_ATOMIC);
621 			if (!*out)
622 				rc = -ENOMEM;
623 		}
624 	}
625 
626 	rcu_read_unlock();
627 
628 	if (!rc && out)
629 		sidtab_sid2str_put(s, entry, *out, *out_len);
630 	return rc;
631 }
632 
633 #endif /* CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0 */
634