Linux Audio

Check our new training course

Yocto / OpenEmbedded training

Feb 10-13, 2025
Register
Loading...
v4.6
 
  1/*
  2 * Implementation of the SID table type.
  3 *
  4 * Author : Stephen Smalley, <sds@epoch.ncsc.mil>
 
 
 
  5 */
 
  6#include <linux/kernel.h>
 
 
  7#include <linux/slab.h>
 
  8#include <linux/spinlock.h>
  9#include <linux/errno.h>
 10#include "flask.h"
 11#include "security.h"
 12#include "sidtab.h"
 13
 14#define SIDTAB_HASH(sid) \
 15(sid & SIDTAB_HASH_MASK)
 
 
 
 
 
 
 
 
 16
 17int sidtab_init(struct sidtab *s)
 18{
 19	int i;
 
 
 
 
 
 
 
 
 
 20
 21	s->htable = kmalloc(sizeof(*(s->htable)) * SIDTAB_SIZE, GFP_ATOMIC);
 22	if (!s->htable)
 23		return -ENOMEM;
 24	for (i = 0; i < SIDTAB_SIZE; i++)
 25		s->htable[i] = NULL;
 26	s->nel = 0;
 27	s->next_sid = 1;
 28	s->shutdown = 0;
 29	spin_lock_init(&s->lock);
 
 
 
 
 
 
 
 30	return 0;
 31}
 32
 33int sidtab_insert(struct sidtab *s, u32 sid, struct context *context)
 34{
 35	int hvalue, rc = 0;
 36	struct sidtab_node *prev, *cur, *newnode;
 37
 38	if (!s) {
 39		rc = -ENOMEM;
 40		goto out;
 
 
 
 
 
 41	}
 
 
 
 42
 43	hvalue = SIDTAB_HASH(sid);
 44	prev = NULL;
 45	cur = s->htable[hvalue];
 46	while (cur && sid > cur->sid) {
 47		prev = cur;
 48		cur = cur->next;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 49	}
 50
 51	if (cur && sid == cur->sid) {
 52		rc = -EEXIST;
 53		goto out;
 54	}
 55
 56	newnode = kmalloc(sizeof(*newnode), GFP_ATOMIC);
 57	if (newnode == NULL) {
 58		rc = -ENOMEM;
 59		goto out;
 60	}
 61	newnode->sid = sid;
 62	if (context_cpy(&newnode->context, context)) {
 63		kfree(newnode);
 64		rc = -ENOMEM;
 65		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 66	}
 
 67
 68	if (prev) {
 69		newnode->next = prev->next;
 70		wmb();
 71		prev->next = newnode;
 72	} else {
 73		newnode->next = s->htable[hvalue];
 74		wmb();
 75		s->htable[hvalue] = newnode;
 
 
 
 
 
 
 
 
 76	}
 
 
 77
 78	s->nel++;
 79	if (sid >= s->next_sid)
 80		s->next_sid = sid + 1;
 81out:
 82	return rc;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 83}
 84
 85static struct context *sidtab_search_core(struct sidtab *s, u32 sid, int force)
 
 86{
 87	int hvalue;
 88	struct sidtab_node *cur;
 89
 90	if (!s)
 
 
 
 
 
 91		return NULL;
 92
 93	hvalue = SIDTAB_HASH(sid);
 94	cur = s->htable[hvalue];
 95	while (cur && sid > cur->sid)
 96		cur = cur->next;
 97
 98	if (force && cur && sid == cur->sid && cur->context.len)
 99		return &cur->context;
100
101	if (cur == NULL || sid != cur->sid || cur->context.len) {
102		/* Remap invalid SIDs to the unlabeled SID. */
103		sid = SECINITSID_UNLABELED;
104		hvalue = SIDTAB_HASH(sid);
105		cur = s->htable[hvalue];
106		while (cur && sid > cur->sid)
107			cur = cur->next;
108		if (!cur || sid != cur->sid)
 
 
 
 
 
 
109			return NULL;
110	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
111
112	return &cur->context;
113}
114
115struct context *sidtab_search(struct sidtab *s, u32 sid)
116{
117	return sidtab_search_core(s, sid, 0);
118}
119
120struct context *sidtab_search_force(struct sidtab *s, u32 sid)
121{
122	return sidtab_search_core(s, sid, 1);
123}
124
125int sidtab_map(struct sidtab *s,
126	       int (*apply) (u32 sid,
127			     struct context *context,
128			     void *args),
129	       void *args)
130{
131	int i, rc = 0;
132	struct sidtab_node *cur;
133
134	if (!s)
135		goto out;
136
137	for (i = 0; i < SIDTAB_SIZE; i++) {
138		cur = s->htable[i];
139		while (cur) {
140			rc = apply(cur->sid, &cur->context, args);
141			if (rc)
142				goto out;
143			cur = cur->next;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
144		}
 
 
 
 
 
 
145	}
146out:
 
 
 
 
 
 
 
 
 
 
 
 
 
147	return rc;
148}
149
150static void sidtab_update_cache(struct sidtab *s, struct sidtab_node *n, int loc)
151{
152	BUG_ON(loc >= SIDTAB_CACHE_LEN);
 
153
154	while (loc > 0) {
155		s->cache[loc] = s->cache[loc - 1];
156		loc--;
 
 
 
157	}
158	s->cache[0] = n;
159}
160
161static inline u32 sidtab_search_context(struct sidtab *s,
162						  struct context *context)
 
 
163{
164	int i;
165	struct sidtab_node *cur;
166
167	for (i = 0; i < SIDTAB_SIZE; i++) {
168		cur = s->htable[i];
169		while (cur) {
170			if (context_cmp(&cur->context, context)) {
171				sidtab_update_cache(s, cur, SIDTAB_CACHE_LEN - 1);
172				return cur->sid;
173			}
174			cur = cur->next;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
175		}
 
 
 
 
 
 
 
 
 
 
 
176	}
177	return 0;
178}
179
180static inline u32 sidtab_search_cache(struct sidtab *s, struct context *context)
181{
182	int i;
183	struct sidtab_node *node;
 
184
185	for (i = 0; i < SIDTAB_CACHE_LEN; i++) {
186		node = s->cache[i];
187		if (unlikely(!node))
188			return 0;
189		if (context_cmp(&node->context, context)) {
190			sidtab_update_cache(s, node, i);
191			return node->sid;
192		}
193	}
194	return 0;
195}
196
197int sidtab_context_to_sid(struct sidtab *s,
198			  struct context *context,
199			  u32 *out_sid)
200{
201	u32 sid;
202	int ret = 0;
203	unsigned long flags;
204
205	*out_sid = SECSID_NULL;
 
 
 
 
 
 
 
 
 
 
206
207	sid  = sidtab_search_cache(s, context);
208	if (!sid)
209		sid = sidtab_search_context(s, context);
210	if (!sid) {
 
 
 
 
 
 
 
 
 
 
211		spin_lock_irqsave(&s->lock, flags);
212		/* Rescan now that we hold the lock. */
213		sid = sidtab_search_context(s, context);
214		if (sid)
215			goto unlock_out;
216		/* No SID exists for the context.  Allocate a new one. */
217		if (s->next_sid == UINT_MAX || s->shutdown) {
218			ret = -ENOMEM;
219			goto unlock_out;
220		}
221		sid = s->next_sid++;
222		if (context->len)
223			printk(KERN_INFO
224		       "SELinux:  Context %s is not valid (left unmapped).\n",
225			       context->str);
226		ret = sidtab_insert(s, sid, context);
227		if (ret)
228			s->next_sid--;
229unlock_out:
230		spin_unlock_irqrestore(&s->lock, flags);
 
231	}
 
 
 
 
 
 
 
232
233	if (ret)
234		return ret;
235
236	*out_sid = sid;
237	return 0;
238}
239
240void sidtab_hash_eval(struct sidtab *h, char *tag)
241{
242	int i, chain_len, slots_used, max_chain_len;
243	struct sidtab_node *cur;
 
 
 
244
245	slots_used = 0;
246	max_chain_len = 0;
247	for (i = 0; i < SIDTAB_SIZE; i++) {
248		cur = h->htable[i];
249		if (cur) {
250			slots_used++;
251			chain_len = 0;
252			while (cur) {
253				chain_len++;
254				cur = cur->next;
255			}
256
257			if (chain_len > max_chain_len)
258				max_chain_len = chain_len;
259		}
260	}
 
261
262	printk(KERN_DEBUG "%s:  %d entries and %d/%d buckets used, longest "
263	       "chain length %d\n", tag, h->nel, slots_used, SIDTAB_SIZE,
264	       max_chain_len);
 
 
 
 
 
 
 
 
 
 
265}
266
267void sidtab_destroy(struct sidtab *s)
268{
269	int i;
270	struct sidtab_node *cur, *temp;
271
272	if (!s)
273		return;
 
274
275	for (i = 0; i < SIDTAB_SIZE; i++) {
276		cur = s->htable[i];
277		while (cur) {
278			temp = cur;
279			cur = cur->next;
280			context_destroy(&temp->context);
281			kfree(temp);
282		}
283		s->htable[i] = NULL;
284	}
285	kfree(s->htable);
286	s->htable = NULL;
287	s->nel = 0;
288	s->next_sid = 1;
289}
290
291void sidtab_set(struct sidtab *dst, struct sidtab *src)
 
 
 
292{
 
293	unsigned long flags;
294	int i;
295
296	spin_lock_irqsave(&src->lock, flags);
297	dst->htable = src->htable;
298	dst->nel = src->nel;
299	dst->next_sid = src->next_sid;
300	dst->shutdown = 0;
301	for (i = 0; i < SIDTAB_CACHE_LEN; i++)
302		dst->cache[i] = NULL;
303	spin_unlock_irqrestore(&src->lock, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
304}
305
306void sidtab_shutdown(struct sidtab *s)
 
307{
308	unsigned long flags;
 
309
310	spin_lock_irqsave(&s->lock, flags);
311	s->shutdown = 1;
312	spin_unlock_irqrestore(&s->lock, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
313}
v5.9
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Implementation of the SID table type.
  4 *
  5 * Original author: Stephen Smalley, <sds@tycho.nsa.gov>
  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
 22struct sidtab_str_cache {
 23	struct rcu_head rcu_member;
 24	struct list_head lru_member;
 25	struct sidtab_entry *parent;
 26	u32 len;
 27	char str[];
 28};
 29
 30#define index_to_sid(index) (index + SECINITSID_NUM + 1)
 31#define sid_to_index(sid) (sid - (SECINITSID_NUM + 1))
 32
 33int sidtab_init(struct sidtab *s)
 34{
 35	u32 i;
 36
 37	memset(s->roots, 0, sizeof(s->roots));
 38
 39	for (i = 0; i < SECINITSID_NUM; i++)
 40		s->isids[i].set = 0;
 41
 42	s->count = 0;
 43	s->convert = NULL;
 44	hash_init(s->context_to_sid);
 45
 
 
 
 
 
 
 
 
 46	spin_lock_init(&s->lock);
 47
 48#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
 49	s->cache_free_slots = CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE;
 50	INIT_LIST_HEAD(&s->cache_lru_list);
 51	spin_lock_init(&s->cache_lock);
 52#endif
 53
 54	return 0;
 55}
 56
 57static u32 context_to_sid(struct sidtab *s, struct context *context, u32 hash)
 58{
 59	struct sidtab_entry *entry;
 60	u32 sid = 0;
 61
 62	rcu_read_lock();
 63	hash_for_each_possible_rcu(s->context_to_sid, entry, list, hash) {
 64		if (entry->hash != hash)
 65			continue;
 66		if (context_cmp(&entry->context, context)) {
 67			sid = entry->sid;
 68			break;
 69		}
 70	}
 71	rcu_read_unlock();
 72	return sid;
 73}
 74
 75int sidtab_set_initial(struct sidtab *s, u32 sid, struct context *context)
 76{
 77	struct sidtab_isid_entry *isid;
 78	u32 hash;
 79	int rc;
 80
 81	if (sid == 0 || sid > SECINITSID_NUM)
 82		return -EINVAL;
 83
 84	isid = &s->isids[sid - 1];
 85
 86	rc = context_cpy(&isid->entry.context, context);
 87	if (rc)
 88		return rc;
 89
 90#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
 91	isid->entry.cache = NULL;
 92#endif
 93	isid->set = 1;
 94
 95	hash = context_compute_hash(context);
 96
 97	/*
 98	 * Multiple initial sids may map to the same context. Check that this
 99	 * context is not already represented in the context_to_sid hashtable
100	 * to avoid duplicate entries and long linked lists upon hash
101	 * collision.
102	 */
103	if (!context_to_sid(s, context, hash)) {
104		isid->entry.sid = sid;
105		isid->entry.hash = hash;
106		hash_add(s->context_to_sid, &isid->entry.list, hash);
107	}
108
109	return 0;
110}
 
 
111
112int sidtab_hash_stats(struct sidtab *sidtab, char *page)
113{
114	int i;
115	int chain_len = 0;
116	int slots_used = 0;
117	int entries = 0;
118	int max_chain_len = 0;
119	int cur_bucket = 0;
120	struct sidtab_entry *entry;
121
122	rcu_read_lock();
123	hash_for_each_rcu(sidtab->context_to_sid, i, entry, list) {
124		entries++;
125		if (i == cur_bucket) {
126			chain_len++;
127			if (chain_len == 1)
128				slots_used++;
129		} else {
130			cur_bucket = i;
131			if (chain_len > max_chain_len)
132				max_chain_len = chain_len;
133			chain_len = 0;
134		}
135	}
136	rcu_read_unlock();
137
138	if (chain_len > max_chain_len)
139		max_chain_len = chain_len;
140
141	return scnprintf(page, PAGE_SIZE, "entries: %d\nbuckets used: %d/%d\n"
142			 "longest chain: %d\n", entries,
143			 slots_used, SIDTAB_HASH_BUCKETS, max_chain_len);
144}
145
146static u32 sidtab_level_from_count(u32 count)
147{
148	u32 capacity = SIDTAB_LEAF_ENTRIES;
149	u32 level = 0;
150
151	while (count > capacity) {
152		capacity <<= SIDTAB_INNER_SHIFT;
153		++level;
154	}
155	return level;
156}
157
158static int sidtab_alloc_roots(struct sidtab *s, u32 level)
159{
160	u32 l;
161
162	if (!s->roots[0].ptr_leaf) {
163		s->roots[0].ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
164					       GFP_ATOMIC);
165		if (!s->roots[0].ptr_leaf)
166			return -ENOMEM;
167	}
168	for (l = 1; l <= level; ++l)
169		if (!s->roots[l].ptr_inner) {
170			s->roots[l].ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
171							GFP_ATOMIC);
172			if (!s->roots[l].ptr_inner)
173				return -ENOMEM;
174			s->roots[l].ptr_inner->entries[0] = s->roots[l - 1];
175		}
176	return 0;
177}
178
179static struct sidtab_entry *sidtab_do_lookup(struct sidtab *s, u32 index,
180					     int alloc)
181{
182	union sidtab_entry_inner *entry;
183	u32 level, capacity_shift, leaf_index = index / SIDTAB_LEAF_ENTRIES;
184
185	/* find the level of the subtree we need */
186	level = sidtab_level_from_count(index + 1);
187	capacity_shift = level * SIDTAB_INNER_SHIFT;
188
189	/* allocate roots if needed */
190	if (alloc && sidtab_alloc_roots(s, level) != 0)
191		return NULL;
192
193	/* lookup inside the subtree */
194	entry = &s->roots[level];
195	while (level != 0) {
196		capacity_shift -= SIDTAB_INNER_SHIFT;
197		--level;
198
199		entry = &entry->ptr_inner->entries[leaf_index >> capacity_shift];
200		leaf_index &= ((u32)1 << capacity_shift) - 1;
201
202		if (!entry->ptr_inner) {
203			if (alloc)
204				entry->ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
205							   GFP_ATOMIC);
206			if (!entry->ptr_inner)
207				return NULL;
208		}
209	}
210	if (!entry->ptr_leaf) {
211		if (alloc)
212			entry->ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
213						  GFP_ATOMIC);
214		if (!entry->ptr_leaf)
215			return NULL;
216	}
217	return &entry->ptr_leaf->entries[index % SIDTAB_LEAF_ENTRIES];
218}
219
220static struct sidtab_entry *sidtab_lookup(struct sidtab *s, u32 index)
221{
222	/* read entries only after reading count */
223	u32 count = smp_load_acquire(&s->count);
224
225	if (index >= count)
226		return NULL;
227
228	return sidtab_do_lookup(s, index, 0);
229}
230
231static struct sidtab_entry *sidtab_lookup_initial(struct sidtab *s, u32 sid)
232{
233	return s->isids[sid - 1].set ? &s->isids[sid - 1].entry : NULL;
234}
235
236static struct sidtab_entry *sidtab_search_core(struct sidtab *s, u32 sid,
237					       int force)
238{
239	if (sid != 0) {
240		struct sidtab_entry *entry;
241
242		if (sid > SECINITSID_NUM)
243			entry = sidtab_lookup(s, sid_to_index(sid));
244		else
245			entry = sidtab_lookup_initial(s, sid);
246		if (entry && (!entry->context.len || force))
247			return entry;
248	}
249
250	return sidtab_lookup_initial(s, SECINITSID_UNLABELED);
251}
252
253struct sidtab_entry *sidtab_search_entry(struct sidtab *s, u32 sid)
254{
255	return sidtab_search_core(s, sid, 0);
256}
257
258struct sidtab_entry *sidtab_search_entry_force(struct sidtab *s, u32 sid)
259{
260	return sidtab_search_core(s, sid, 1);
261}
262
263int sidtab_context_to_sid(struct sidtab *s, struct context *context,
264			  u32 *sid)
265{
266	unsigned long flags;
267	u32 count, hash = context_compute_hash(context);
268	struct sidtab_convert_params *convert;
269	struct sidtab_entry *dst, *dst_convert;
270	int rc;
271
272	*sid = context_to_sid(s, context, hash);
273	if (*sid)
274		return 0;
275
276	/* lock-free search failed: lock, re-search, and insert if not found */
277	spin_lock_irqsave(&s->lock, flags);
278
279	rc = 0;
280	*sid = context_to_sid(s, context, hash);
281	if (*sid)
282		goto out_unlock;
283
284	count = s->count;
285	convert = s->convert;
286
287	/* bail out if we already reached max entries */
288	rc = -EOVERFLOW;
289	if (count >= SIDTAB_MAX)
290		goto out_unlock;
291
292	/* insert context into new entry */
293	rc = -ENOMEM;
294	dst = sidtab_do_lookup(s, count, 1);
295	if (!dst)
296		goto out_unlock;
297
298	dst->sid = index_to_sid(count);
299	dst->hash = hash;
300
301	rc = context_cpy(&dst->context, context);
302	if (rc)
303		goto out_unlock;
304
305	/*
306	 * if we are building a new sidtab, we need to convert the context
307	 * and insert it there as well
308	 */
309	if (convert) {
310		rc = -ENOMEM;
311		dst_convert = sidtab_do_lookup(convert->target, count, 1);
312		if (!dst_convert) {
313			context_destroy(&dst->context);
314			goto out_unlock;
315		}
316
317		rc = convert->func(context, &dst_convert->context,
318				   convert->args);
319		if (rc) {
320			context_destroy(&dst->context);
321			goto out_unlock;
322		}
323		dst_convert->sid = index_to_sid(count);
324		dst_convert->hash = context_compute_hash(&dst_convert->context);
325		convert->target->count = count + 1;
326
327		hash_add_rcu(convert->target->context_to_sid,
328			     &dst_convert->list, dst_convert->hash);
329	}
330
331	if (context->len)
332		pr_info("SELinux:  Context %s is not valid (left unmapped).\n",
333			context->str);
334
335	*sid = index_to_sid(count);
336
337	/* write entries before updating count */
338	smp_store_release(&s->count, count + 1);
339	hash_add_rcu(s->context_to_sid, &dst->list, dst->hash);
340
341	rc = 0;
342out_unlock:
343	spin_unlock_irqrestore(&s->lock, flags);
344	return rc;
345}
346
347static void sidtab_convert_hashtable(struct sidtab *s, u32 count)
348{
349	struct sidtab_entry *entry;
350	u32 i;
351
352	for (i = 0; i < count; i++) {
353		entry = sidtab_do_lookup(s, i, 0);
354		entry->sid = index_to_sid(i);
355		entry->hash = context_compute_hash(&entry->context);
356
357		hash_add_rcu(s->context_to_sid, &entry->list, entry->hash);
358	}
 
359}
360
361static int sidtab_convert_tree(union sidtab_entry_inner *edst,
362			       union sidtab_entry_inner *esrc,
363			       u32 *pos, u32 count, u32 level,
364			       struct sidtab_convert_params *convert)
365{
366	int rc;
367	u32 i;
368
369	if (level != 0) {
370		if (!edst->ptr_inner) {
371			edst->ptr_inner = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
372						  GFP_KERNEL);
373			if (!edst->ptr_inner)
374				return -ENOMEM;
375		}
376		i = 0;
377		while (i < SIDTAB_INNER_ENTRIES && *pos < count) {
378			rc = sidtab_convert_tree(&edst->ptr_inner->entries[i],
379						 &esrc->ptr_inner->entries[i],
380						 pos, count, level - 1,
381						 convert);
382			if (rc)
383				return rc;
384			i++;
385		}
386	} else {
387		if (!edst->ptr_leaf) {
388			edst->ptr_leaf = kzalloc(SIDTAB_NODE_ALLOC_SIZE,
389						 GFP_KERNEL);
390			if (!edst->ptr_leaf)
391				return -ENOMEM;
392		}
393		i = 0;
394		while (i < SIDTAB_LEAF_ENTRIES && *pos < count) {
395			rc = convert->func(&esrc->ptr_leaf->entries[i].context,
396					   &edst->ptr_leaf->entries[i].context,
397					   convert->args);
398			if (rc)
399				return rc;
400			(*pos)++;
401			i++;
402		}
403		cond_resched();
404	}
405	return 0;
406}
407
408int sidtab_convert(struct sidtab *s, struct sidtab_convert_params *params)
409{
410	unsigned long flags;
411	u32 count, level, pos;
412	int rc;
413
414	spin_lock_irqsave(&s->lock, flags);
415
416	/* concurrent policy loads are not allowed */
417	if (s->convert) {
418		spin_unlock_irqrestore(&s->lock, flags);
419		return -EBUSY;
 
 
420	}
 
 
421
422	count = s->count;
423	level = sidtab_level_from_count(count);
 
 
 
 
 
424
425	/* allocate last leaf in the new sidtab (to avoid race with
426	 * live convert)
427	 */
428	rc = sidtab_do_lookup(params->target, count - 1, 1) ? 0 : -ENOMEM;
429	if (rc) {
430		spin_unlock_irqrestore(&s->lock, flags);
431		return rc;
432	}
433
434	/* set count in case no new entries are added during conversion */
435	params->target->count = count;
436
437	/* enable live convert of new entries */
438	s->convert = params;
439
440	/* we can safely convert the tree outside the lock */
441	spin_unlock_irqrestore(&s->lock, flags);
442
443	pr_info("SELinux:  Converting %u SID table entries...\n", count);
444
445	/* convert all entries not covered by live convert */
446	pos = 0;
447	rc = sidtab_convert_tree(&params->target->roots[level],
448				 &s->roots[level], &pos, count, level, params);
449	if (rc) {
450		/* we need to keep the old table - disable live convert */
451		spin_lock_irqsave(&s->lock, flags);
452		s->convert = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
453		spin_unlock_irqrestore(&s->lock, flags);
454		return rc;
455	}
456	/*
457	 * The hashtable can also be modified in sidtab_context_to_sid()
458	 * so we must re-acquire the lock here.
459	 */
460	spin_lock_irqsave(&s->lock, flags);
461	sidtab_convert_hashtable(params->target, count);
462	spin_unlock_irqrestore(&s->lock, flags);
463
 
 
 
 
464	return 0;
465}
466
467static void sidtab_destroy_entry(struct sidtab_entry *entry)
468{
469	context_destroy(&entry->context);
470#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
471	kfree(rcu_dereference_raw(entry->cache));
472#endif
473}
474
475static void sidtab_destroy_tree(union sidtab_entry_inner entry, u32 level)
476{
477	u32 i;
 
 
 
 
 
 
 
 
478
479	if (level != 0) {
480		struct sidtab_node_inner *node = entry.ptr_inner;
481
482		if (!node)
483			return;
484
485		for (i = 0; i < SIDTAB_INNER_ENTRIES; i++)
486			sidtab_destroy_tree(node->entries[i], level - 1);
487		kfree(node);
488	} else {
489		struct sidtab_node_leaf *node = entry.ptr_leaf;
490
491		if (!node)
492			return;
493
494		for (i = 0; i < SIDTAB_LEAF_ENTRIES; i++)
495			sidtab_destroy_entry(&node->entries[i]);
496		kfree(node);
497	}
498}
499
500void sidtab_destroy(struct sidtab *s)
501{
502	u32 i, level;
 
503
504	for (i = 0; i < SECINITSID_NUM; i++)
505		if (s->isids[i].set)
506			sidtab_destroy_entry(&s->isids[i].entry);
507
508	level = SIDTAB_MAX_LEVEL;
509	while (level && !s->roots[level].ptr_inner)
510		--level;
511
512	sidtab_destroy_tree(s->roots[level], level);
513	/*
514	 * The context_to_sid hashtable's objects are all shared
515	 * with the isids array and context tree, and so don't need
516	 * to be cleaned up here.
517	 */
 
 
 
 
518}
519
520#if CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0
521
522void sidtab_sid2str_put(struct sidtab *s, struct sidtab_entry *entry,
523			const char *str, u32 str_len)
524{
525	struct sidtab_str_cache *cache, *victim = NULL;
526	unsigned long flags;
 
527
528	/* do not cache invalid contexts */
529	if (entry->context.len)
530		return;
531
532	spin_lock_irqsave(&s->cache_lock, flags);
533
534	cache = rcu_dereference_protected(entry->cache,
535					  lockdep_is_held(&s->cache_lock));
536	if (cache) {
537		/* entry in cache - just bump to the head of LRU list */
538		list_move(&cache->lru_member, &s->cache_lru_list);
539		goto out_unlock;
540	}
541
542	cache = kmalloc(sizeof(struct sidtab_str_cache) + str_len, GFP_ATOMIC);
543	if (!cache)
544		goto out_unlock;
545
546	if (s->cache_free_slots == 0) {
547		/* pop a cache entry from the tail and free it */
548		victim = container_of(s->cache_lru_list.prev,
549				      struct sidtab_str_cache, lru_member);
550		list_del(&victim->lru_member);
551		rcu_assign_pointer(victim->parent->cache, NULL);
552	} else {
553		s->cache_free_slots--;
554	}
555	cache->parent = entry;
556	cache->len = str_len;
557	memcpy(cache->str, str, str_len);
558	list_add(&cache->lru_member, &s->cache_lru_list);
559
560	rcu_assign_pointer(entry->cache, cache);
561
562out_unlock:
563	spin_unlock_irqrestore(&s->cache_lock, flags);
564	kfree_rcu(victim, rcu_member);
565}
566
567int sidtab_sid2str_get(struct sidtab *s, struct sidtab_entry *entry,
568		       char **out, u32 *out_len)
569{
570	struct sidtab_str_cache *cache;
571	int rc = 0;
572
573	if (entry->context.len)
574		return -ENOENT; /* do not cache invalid contexts */
575
576	rcu_read_lock();
577
578	cache = rcu_dereference(entry->cache);
579	if (!cache) {
580		rc = -ENOENT;
581	} else {
582		*out_len = cache->len;
583		if (out) {
584			*out = kmemdup(cache->str, cache->len, GFP_ATOMIC);
585			if (!*out)
586				rc = -ENOMEM;
587		}
588	}
589
590	rcu_read_unlock();
591
592	if (!rc && out)
593		sidtab_sid2str_put(s, entry, *out, *out_len);
594	return rc;
595}
596
597#endif /* CONFIG_SECURITY_SELINUX_SID2STR_CACHE_SIZE > 0 */