Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
avc.c
Go to the documentation of this file.
1 /*
2  * Implementation of the kernel access vector cache (AVC).
3  *
4  * Authors: Stephen Smalley, <[email protected]>
5  * James Morris <[email protected]>
6  *
7  * Update: KaiGai, Kohei <[email protected]>
8  * Replaced the avc_lock spinlock by RCU.
9  *
10  * Copyright (C) 2003 Red Hat, Inc., James Morris <[email protected]>
11  *
12  * This program is free software; you can redistribute it and/or modify
13  * it under the terms of the GNU General Public License version 2,
14  * as published by the Free Software Foundation.
15  */
16 #include <linux/types.h>
17 #include <linux/stddef.h>
18 #include <linux/kernel.h>
19 #include <linux/slab.h>
20 #include <linux/fs.h>
21 #include <linux/dcache.h>
22 #include <linux/init.h>
23 #include <linux/skbuff.h>
24 #include <linux/percpu.h>
25 #include <net/sock.h>
26 #include <linux/un.h>
27 #include <net/af_unix.h>
28 #include <linux/ip.h>
29 #include <linux/audit.h>
30 #include <linux/ipv6.h>
31 #include <net/ipv6.h>
32 #include "avc.h"
33 #include "avc_ss.h"
34 #include "classmap.h"
35 
36 #define AVC_CACHE_SLOTS 512
37 #define AVC_DEF_CACHE_THRESHOLD 512
38 #define AVC_CACHE_RECLAIM 16
39 
40 #ifdef CONFIG_SECURITY_SELINUX_AVC_STATS
41 #define avc_cache_stats_incr(field) this_cpu_inc(avc_cache_stats.field)
42 #else
43 #define avc_cache_stats_incr(field) do {} while (0)
44 #endif
45 
46 struct avc_entry {
50  struct av_decision avd;
51 };
52 
53 struct avc_node {
54  struct avc_entry ae;
55  struct hlist_node list; /* anchored in avc_cache->slots[i] */
56  struct rcu_head rhead;
57 };
58 
59 struct avc_cache {
60  struct hlist_head slots[AVC_CACHE_SLOTS]; /* head for avc_node->list */
61  spinlock_t slots_lock[AVC_CACHE_SLOTS]; /* lock for writes */
62  atomic_t lru_hint; /* LRU hint for reclaim scan */
64  u32 latest_notif; /* latest revocation notification */
65 };
66 
71 };
72 
73 /* Exported via selinufs */
75 
76 #ifdef CONFIG_SECURITY_SELINUX_AVC_STATS
78 #endif
79 
80 static struct avc_cache avc_cache;
81 static struct avc_callback_node *avc_callbacks;
82 static struct kmem_cache *avc_node_cachep;
83 
84 static inline int avc_hash(u32 ssid, u32 tsid, u16 tclass)
85 {
86  return (ssid ^ (tsid<<2) ^ (tclass<<4)) & (AVC_CACHE_SLOTS - 1);
87 }
88 
94 static void avc_dump_av(struct audit_buffer *ab, u16 tclass, u32 av)
95 {
96  const char **perms;
97  int i, perm;
98 
99  if (av == 0) {
100  audit_log_format(ab, " null");
101  return;
102  }
103 
104  perms = secclass_map[tclass-1].perms;
105 
106  audit_log_format(ab, " {");
107  i = 0;
108  perm = 1;
109  while (i < (sizeof(av) * 8)) {
110  if ((perm & av) && perms[i]) {
111  audit_log_format(ab, " %s", perms[i]);
112  av &= ~perm;
113  }
114  i++;
115  perm <<= 1;
116  }
117 
118  if (av)
119  audit_log_format(ab, " 0x%x", av);
120 
121  audit_log_format(ab, " }");
122 }
123 
130 static void avc_dump_query(struct audit_buffer *ab, u32 ssid, u32 tsid, u16 tclass)
131 {
132  int rc;
133  char *scontext;
134  u32 scontext_len;
135 
136  rc = security_sid_to_context(ssid, &scontext, &scontext_len);
137  if (rc)
138  audit_log_format(ab, "ssid=%d", ssid);
139  else {
140  audit_log_format(ab, "scontext=%s", scontext);
141  kfree(scontext);
142  }
143 
144  rc = security_sid_to_context(tsid, &scontext, &scontext_len);
145  if (rc)
146  audit_log_format(ab, " tsid=%d", tsid);
147  else {
148  audit_log_format(ab, " tcontext=%s", scontext);
149  kfree(scontext);
150  }
151 
152  BUG_ON(tclass >= ARRAY_SIZE(secclass_map));
153  audit_log_format(ab, " tclass=%s", secclass_map[tclass-1].name);
154 }
155 
161 void __init avc_init(void)
162 {
163  int i;
164 
165  for (i = 0; i < AVC_CACHE_SLOTS; i++) {
168  }
171 
172  avc_node_cachep = kmem_cache_create("avc_node", sizeof(struct avc_node),
173  0, SLAB_PANIC, NULL);
174 
175  audit_log(current->audit_context, GFP_KERNEL, AUDIT_KERNEL, "AVC INITIALIZED\n");
176 }
177 
179 {
180  int i, chain_len, max_chain_len, slots_used;
181  struct avc_node *node;
182  struct hlist_head *head;
183 
184  rcu_read_lock();
185 
186  slots_used = 0;
187  max_chain_len = 0;
188  for (i = 0; i < AVC_CACHE_SLOTS; i++) {
189  head = &avc_cache.slots[i];
190  if (!hlist_empty(head)) {
191  struct hlist_node *next;
192 
193  slots_used++;
194  chain_len = 0;
195  hlist_for_each_entry_rcu(node, next, head, list)
196  chain_len++;
197  if (chain_len > max_chain_len)
198  max_chain_len = chain_len;
199  }
200  }
201 
202  rcu_read_unlock();
203 
204  return scnprintf(page, PAGE_SIZE, "entries: %d\nbuckets used: %d/%d\n"
205  "longest chain: %d\n",
207  slots_used, AVC_CACHE_SLOTS, max_chain_len);
208 }
209 
210 static void avc_node_free(struct rcu_head *rhead)
211 {
212  struct avc_node *node = container_of(rhead, struct avc_node, rhead);
213  kmem_cache_free(avc_node_cachep, node);
214  avc_cache_stats_incr(frees);
215 }
216 
217 static void avc_node_delete(struct avc_node *node)
218 {
219  hlist_del_rcu(&node->list);
220  call_rcu(&node->rhead, avc_node_free);
222 }
223 
224 static void avc_node_kill(struct avc_node *node)
225 {
226  kmem_cache_free(avc_node_cachep, node);
227  avc_cache_stats_incr(frees);
229 }
230 
231 static void avc_node_replace(struct avc_node *new, struct avc_node *old)
232 {
233  hlist_replace_rcu(&old->list, &new->list);
234  call_rcu(&old->rhead, avc_node_free);
236 }
237 
238 static inline int avc_reclaim_node(void)
239 {
240  struct avc_node *node;
241  int hvalue, try, ecx;
242  unsigned long flags;
243  struct hlist_head *head;
244  struct hlist_node *next;
245  spinlock_t *lock;
246 
247  for (try = 0, ecx = 0; try < AVC_CACHE_SLOTS; try++) {
249  head = &avc_cache.slots[hvalue];
250  lock = &avc_cache.slots_lock[hvalue];
251 
252  if (!spin_trylock_irqsave(lock, flags))
253  continue;
254 
255  rcu_read_lock();
256  hlist_for_each_entry(node, next, head, list) {
257  avc_node_delete(node);
258  avc_cache_stats_incr(reclaims);
259  ecx++;
260  if (ecx >= AVC_CACHE_RECLAIM) {
261  rcu_read_unlock();
262  spin_unlock_irqrestore(lock, flags);
263  goto out;
264  }
265  }
266  rcu_read_unlock();
267  spin_unlock_irqrestore(lock, flags);
268  }
269 out:
270  return ecx;
271 }
272 
273 static struct avc_node *avc_alloc_node(void)
274 {
275  struct avc_node *node;
276 
277  node = kmem_cache_zalloc(avc_node_cachep, GFP_ATOMIC|__GFP_NOMEMALLOC);
278  if (!node)
279  goto out;
280 
281  INIT_HLIST_NODE(&node->list);
282  avc_cache_stats_incr(allocations);
283 
285  avc_reclaim_node();
286 
287 out:
288  return node;
289 }
290 
291 static void avc_node_populate(struct avc_node *node, u32 ssid, u32 tsid, u16 tclass, struct av_decision *avd)
292 {
293  node->ae.ssid = ssid;
294  node->ae.tsid = tsid;
295  node->ae.tclass = tclass;
296  memcpy(&node->ae.avd, avd, sizeof(node->ae.avd));
297 }
298 
299 static inline struct avc_node *avc_search_node(u32 ssid, u32 tsid, u16 tclass)
300 {
301  struct avc_node *node, *ret = NULL;
302  int hvalue;
303  struct hlist_head *head;
304  struct hlist_node *next;
305 
306  hvalue = avc_hash(ssid, tsid, tclass);
307  head = &avc_cache.slots[hvalue];
308  hlist_for_each_entry_rcu(node, next, head, list) {
309  if (ssid == node->ae.ssid &&
310  tclass == node->ae.tclass &&
311  tsid == node->ae.tsid) {
312  ret = node;
313  break;
314  }
315  }
316 
317  return ret;
318 }
319 
332 static struct avc_node *avc_lookup(u32 ssid, u32 tsid, u16 tclass)
333 {
334  struct avc_node *node;
335 
336  avc_cache_stats_incr(lookups);
337  node = avc_search_node(ssid, tsid, tclass);
338 
339  if (node)
340  return node;
341 
342  avc_cache_stats_incr(misses);
343  return NULL;
344 }
345 
346 static int avc_latest_notif_update(int seqno, int is_insert)
347 {
348  int ret = 0;
349  static DEFINE_SPINLOCK(notif_lock);
350  unsigned long flag;
351 
352  spin_lock_irqsave(&notif_lock, flag);
353  if (is_insert) {
354  if (seqno < avc_cache.latest_notif) {
355  printk(KERN_WARNING "SELinux: avc: seqno %d < latest_notif %d\n",
356  seqno, avc_cache.latest_notif);
357  ret = -EAGAIN;
358  }
359  } else {
360  if (seqno > avc_cache.latest_notif)
362  }
363  spin_unlock_irqrestore(&notif_lock, flag);
364 
365  return ret;
366 }
367 
385 static struct avc_node *avc_insert(u32 ssid, u32 tsid, u16 tclass, struct av_decision *avd)
386 {
387  struct avc_node *pos, *node = NULL;
388  int hvalue;
389  unsigned long flag;
390 
391  if (avc_latest_notif_update(avd->seqno, 1))
392  goto out;
393 
394  node = avc_alloc_node();
395  if (node) {
396  struct hlist_head *head;
397  struct hlist_node *next;
398  spinlock_t *lock;
399 
400  hvalue = avc_hash(ssid, tsid, tclass);
401  avc_node_populate(node, ssid, tsid, tclass, avd);
402 
403  head = &avc_cache.slots[hvalue];
404  lock = &avc_cache.slots_lock[hvalue];
405 
406  spin_lock_irqsave(lock, flag);
407  hlist_for_each_entry(pos, next, head, list) {
408  if (pos->ae.ssid == ssid &&
409  pos->ae.tsid == tsid &&
410  pos->ae.tclass == tclass) {
411  avc_node_replace(node, pos);
412  goto found;
413  }
414  }
415  hlist_add_head_rcu(&node->list, head);
416 found:
417  spin_unlock_irqrestore(lock, flag);
418  }
419 out:
420  return node;
421 }
422 
429 static void avc_audit_pre_callback(struct audit_buffer *ab, void *a)
430 {
431  struct common_audit_data *ad = a;
432  audit_log_format(ab, "avc: %s ",
433  ad->selinux_audit_data->denied ? "denied" : "granted");
434  avc_dump_av(ab, ad->selinux_audit_data->tclass,
435  ad->selinux_audit_data->audited);
436  audit_log_format(ab, " for ");
437 }
438 
445 static void avc_audit_post_callback(struct audit_buffer *ab, void *a)
446 {
447  struct common_audit_data *ad = a;
448  audit_log_format(ab, " ");
449  avc_dump_query(ab, ad->selinux_audit_data->ssid,
450  ad->selinux_audit_data->tsid,
451  ad->selinux_audit_data->tclass);
452 }
453 
454 /* This is the slow part of avc audit with big stack footprint */
455 noinline int slow_avc_audit(u32 ssid, u32 tsid, u16 tclass,
456  u32 requested, u32 audited, u32 denied,
457  struct common_audit_data *a,
458  unsigned flags)
459 {
460  struct common_audit_data stack_data;
461  struct selinux_audit_data sad;
462 
463  if (!a) {
464  a = &stack_data;
466  }
467 
468  /*
469  * When in a RCU walk do the audit on the RCU retry. This is because
470  * the collection of the dname in an inode audit message is not RCU
471  * safe. Note this may drop some audits when the situation changes
472  * during retry. However this is logically just as if the operation
473  * happened a little later.
474  */
475  if ((a->type == LSM_AUDIT_DATA_INODE) &&
476  (flags & MAY_NOT_BLOCK))
477  return -ECHILD;
478 
479  sad.tclass = tclass;
480  sad.requested = requested;
481  sad.ssid = ssid;
482  sad.tsid = tsid;
483  sad.audited = audited;
484  sad.denied = denied;
485 
486  a->selinux_audit_data = &sad;
487 
488  common_lsm_audit(a, avc_audit_pre_callback, avc_audit_post_callback);
489  return 0;
490 }
491 
502 {
503  struct avc_callback_node *c;
504  int rc = 0;
505 
506  c = kmalloc(sizeof(*c), GFP_KERNEL);
507  if (!c) {
508  rc = -ENOMEM;
509  goto out;
510  }
511 
512  c->callback = callback;
513  c->events = events;
514  c->next = avc_callbacks;
515  avc_callbacks = c;
516 out:
517  return rc;
518 }
519 
520 static inline int avc_sidcmp(u32 x, u32 y)
521 {
522  return (x == y || x == SECSID_WILD || y == SECSID_WILD);
523 }
524 
537 static int avc_update_node(u32 event, u32 perms, u32 ssid, u32 tsid, u16 tclass,
538  u32 seqno)
539 {
540  int hvalue, rc = 0;
541  unsigned long flag;
542  struct avc_node *pos, *node, *orig = NULL;
543  struct hlist_head *head;
544  struct hlist_node *next;
545  spinlock_t *lock;
546 
547  node = avc_alloc_node();
548  if (!node) {
549  rc = -ENOMEM;
550  goto out;
551  }
552 
553  /* Lock the target slot */
554  hvalue = avc_hash(ssid, tsid, tclass);
555 
556  head = &avc_cache.slots[hvalue];
557  lock = &avc_cache.slots_lock[hvalue];
558 
559  spin_lock_irqsave(lock, flag);
560 
561  hlist_for_each_entry(pos, next, head, list) {
562  if (ssid == pos->ae.ssid &&
563  tsid == pos->ae.tsid &&
564  tclass == pos->ae.tclass &&
565  seqno == pos->ae.avd.seqno){
566  orig = pos;
567  break;
568  }
569  }
570 
571  if (!orig) {
572  rc = -ENOENT;
573  avc_node_kill(node);
574  goto out_unlock;
575  }
576 
577  /*
578  * Copy and replace original node.
579  */
580 
581  avc_node_populate(node, ssid, tsid, tclass, &orig->ae.avd);
582 
583  switch (event) {
584  case AVC_CALLBACK_GRANT:
585  node->ae.avd.allowed |= perms;
586  break;
588  case AVC_CALLBACK_REVOKE:
589  node->ae.avd.allowed &= ~perms;
590  break;
592  node->ae.avd.auditallow |= perms;
593  break;
595  node->ae.avd.auditallow &= ~perms;
596  break;
598  node->ae.avd.auditdeny |= perms;
599  break;
601  node->ae.avd.auditdeny &= ~perms;
602  break;
603  }
604  avc_node_replace(node, orig);
605 out_unlock:
606  spin_unlock_irqrestore(lock, flag);
607 out:
608  return rc;
609 }
610 
614 static void avc_flush(void)
615 {
616  struct hlist_head *head;
617  struct hlist_node *next;
618  struct avc_node *node;
619  spinlock_t *lock;
620  unsigned long flag;
621  int i;
622 
623  for (i = 0; i < AVC_CACHE_SLOTS; i++) {
624  head = &avc_cache.slots[i];
625  lock = &avc_cache.slots_lock[i];
626 
627  spin_lock_irqsave(lock, flag);
628  /*
629  * With preemptable RCU, the outer spinlock does not
630  * prevent RCU grace periods from ending.
631  */
632  rcu_read_lock();
633  hlist_for_each_entry(node, next, head, list)
634  avc_node_delete(node);
635  rcu_read_unlock();
636  spin_unlock_irqrestore(lock, flag);
637  }
638 }
639 
644 int avc_ss_reset(u32 seqno)
645 {
646  struct avc_callback_node *c;
647  int rc = 0, tmprc;
648 
649  avc_flush();
650 
651  for (c = avc_callbacks; c; c = c->next) {
652  if (c->events & AVC_CALLBACK_RESET) {
653  tmprc = c->callback(AVC_CALLBACK_RESET);
654  /* save the first error encountered for the return
655  value and continue processing the callbacks */
656  if (!rc)
657  rc = tmprc;
658  }
659  }
660 
661  avc_latest_notif_update(seqno, 0);
662  return rc;
663 }
664 
665 /*
666  * Slow-path helper function for avc_has_perm_noaudit,
667  * when the avc_node lookup fails. We get called with
668  * the RCU read lock held, and need to return with it
669  * still held, but drop if for the security compute.
670  *
671  * Don't inline this, since it's the slow-path and just
672  * results in a bigger stack frame.
673  */
674 static noinline struct avc_node *avc_compute_av(u32 ssid, u32 tsid,
675  u16 tclass, struct av_decision *avd)
676 {
677  rcu_read_unlock();
678  security_compute_av(ssid, tsid, tclass, avd);
679  rcu_read_lock();
680  return avc_insert(ssid, tsid, tclass, avd);
681 }
682 
683 static noinline int avc_denied(u32 ssid, u32 tsid,
684  u16 tclass, u32 requested,
685  unsigned flags,
686  struct av_decision *avd)
687 {
688  if (flags & AVC_STRICT)
689  return -EACCES;
690 
692  return -EACCES;
693 
694  avc_update_node(AVC_CALLBACK_GRANT, requested, ssid,
695  tsid, tclass, avd->seqno);
696  return 0;
697 }
698 
699 
720 inline int avc_has_perm_noaudit(u32 ssid, u32 tsid,
721  u16 tclass, u32 requested,
722  unsigned flags,
723  struct av_decision *avd)
724 {
725  struct avc_node *node;
726  int rc = 0;
727  u32 denied;
728 
729  BUG_ON(!requested);
730 
731  rcu_read_lock();
732 
733  node = avc_lookup(ssid, tsid, tclass);
734  if (unlikely(!node)) {
735  node = avc_compute_av(ssid, tsid, tclass, avd);
736  } else {
737  memcpy(avd, &node->ae.avd, sizeof(*avd));
738  avd = &node->ae.avd;
739  }
740 
741  denied = requested & ~(avd->allowed);
742  if (unlikely(denied))
743  rc = avc_denied(ssid, tsid, tclass, requested, flags, avd);
744 
745  rcu_read_unlock();
746  return rc;
747 }
748 
766 int avc_has_perm_flags(u32 ssid, u32 tsid, u16 tclass,
767  u32 requested, struct common_audit_data *auditdata,
768  unsigned flags)
769 {
770  struct av_decision avd;
771  int rc, rc2;
772 
773  rc = avc_has_perm_noaudit(ssid, tsid, tclass, requested, 0, &avd);
774 
775  rc2 = avc_audit(ssid, tsid, tclass, requested, &avd, rc, auditdata,
776  flags);
777  if (rc2)
778  return rc2;
779  return rc;
780 }
781 
783 {
784  return avc_cache.latest_notif;
785 }
786 
787 void avc_disable(void)
788 {
789  /*
790  * If you are looking at this because you have realized that we are
791  * not destroying the avc_node_cachep it might be easy to fix, but
792  * I don't know the memory barrier semantics well enough to know. It's
793  * possible that some other task dereferenced security_ops when
794  * it still pointed to selinux operations. If that is the case it's
795  * possible that it is about to use the avc and is about to need the
796  * avc_node_cachep. I know I could wrap the security.c security_ops call
797  * in an rcu_lock, but seriously, it's not worth it. Instead I just flush
798  * the cache and get that memory back.
799  */
800  if (avc_node_cachep) {
801  avc_flush();
802  /* kmem_cache_destroy(avc_node_cachep); */
803  }
804 }