PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
uavc.c
Go to the documentation of this file.
1 /* -------------------------------------------------------------------------
2  *
3  * contrib/sepgsql/uavc.c
4  *
5  * Implementation of userspace access vector cache; that enables to cache
6  * access control decisions recently used, and reduce number of kernel
7  * invocations to avoid unnecessary performance hit.
8  *
9  * Copyright (c) 2011-2017, PostgreSQL Global Development Group
10  *
11  * -------------------------------------------------------------------------
12  */
13 #include "postgres.h"
14 
15 #include "access/hash.h"
16 #include "catalog/pg_proc.h"
17 #include "commands/seclabel.h"
18 #include "storage/ipc.h"
19 #include "utils/guc.h"
20 #include "utils/memutils.h"
21 
22 #include "sepgsql.h"
23 
24 /*
25  * avc_cache
26  *
27  * It enables to cache access control decision (and behavior on execution of
28  * trusted procedure, db_procedure class only) for a particular pair of
29  * security labels and object class in userspace.
30  */
31 typedef struct
32 {
33  uint32 hash; /* hash value of this cache entry */
34  char *scontext; /* security context of the subject */
35  char *tcontext; /* security context of the target */
36  uint16 tclass; /* object class of the target */
37 
38  uint32 allowed; /* permissions to be allowed */
39  uint32 auditallow; /* permissions to be audited on allowed */
40  uint32 auditdeny; /* permissions to be audited on denied */
41 
42  bool permissive; /* true, if permissive rule */
43  bool hot_cache; /* true, if recently referenced */
45  /* true, if tcontext is valid */
46  char *ncontext; /* temporary scontext on execution of trusted
47  * procedure, or NULL elsewhere */
48 } avc_cache;
49 
50 /*
51  * Declaration of static variables
52  */
53 #define AVC_NUM_SLOTS 512
54 #define AVC_NUM_RECLAIM 16
55 #define AVC_DEF_THRESHOLD 384
56 
58 static List *avc_slots[AVC_NUM_SLOTS]; /* avc's hash buckets */
59 static int avc_num_caches; /* number of caches currently used */
60 static int avc_lru_hint; /* index of the buckets to be reclaimed next */
61 static int avc_threshold; /* threshold to launch cache-reclaiming */
62 static char *avc_unlabeled; /* system 'unlabeled' label */
63 
64 /*
65  * Hash function
66  */
67 static uint32
68 sepgsql_avc_hash(const char *scontext, const char *tcontext, uint16 tclass)
69 {
70  return hash_any((const unsigned char *) scontext, strlen(scontext))
71  ^ hash_any((const unsigned char *) tcontext, strlen(tcontext))
72  ^ tclass;
73 }
74 
75 /*
76  * Reset all the avc caches
77  */
78 static void
80 {
81  MemoryContextReset(avc_mem_cxt);
82 
83  memset(avc_slots, 0, sizeof(List *) * AVC_NUM_SLOTS);
84  avc_num_caches = 0;
85  avc_lru_hint = 0;
87 }
88 
89 /*
90  * Reclaim caches recently unreferenced
91  */
92 static void
94 {
95  ListCell *cell;
96  ListCell *next;
97  ListCell *prev;
98  int index;
99 
101  {
102  index = avc_lru_hint;
103 
104  prev = NULL;
105  for (cell = list_head(avc_slots[index]); cell; cell = next)
106  {
107  avc_cache *cache = lfirst(cell);
108 
109  next = lnext(cell);
110  if (!cache->hot_cache)
111  {
112  avc_slots[index]
113  = list_delete_cell(avc_slots[index], cell, prev);
114 
115  pfree(cache->scontext);
116  pfree(cache->tcontext);
117  if (cache->ncontext)
118  pfree(cache->ncontext);
119  pfree(cache);
120 
121  avc_num_caches--;
122  }
123  else
124  {
125  cache->hot_cache = false;
126  prev = cell;
127  }
128  }
130  }
131 }
132 
133 /* -------------------------------------------------------------------------
134  *
135  * sepgsql_avc_check_valid
136  *
137  * This function checks whether the cached entries are still valid. If
138  * the security policy has been reloaded (or any other events that requires
139  * resetting userspace caches has occurred) since the last reference to
140  * the access vector cache, we must flush the cache.
141  *
142  * Access control decisions must be atomic, but multiple system calls may
143  * be required to make a decision; thus, when referencing the access vector
144  * cache, we must loop until we complete without an intervening cache flush
145  * event. In practice, looping even once should be very rare. Callers should
146  * do something like this:
147  *
148  * sepgsql_avc_check_valid();
149  * do {
150  * :
151  * <reference to uavc>
152  * :
153  * } while (!sepgsql_avc_check_valid())
154  *
155  * -------------------------------------------------------------------------
156  */
157 static bool
159 {
160  if (selinux_status_updated() > 0)
161  {
163 
164  return false;
165  }
166  return true;
167 }
168 
169 /*
170  * sepgsql_avc_unlabeled
171  *
172  * Returns an alternative label to be applied when no label or an invalid
173  * label would otherwise be assigned.
174  */
175 static char *
177 {
178  if (!avc_unlabeled)
179  {
180  security_context_t unlabeled;
181 
182  if (security_get_initial_context_raw("unlabeled", &unlabeled) < 0)
183  ereport(ERROR,
184  (errcode(ERRCODE_INTERNAL_ERROR),
185  errmsg("SELinux: failed to get initial security label: %m")));
186  PG_TRY();
187  {
188  avc_unlabeled = MemoryContextStrdup(avc_mem_cxt, unlabeled);
189  }
190  PG_CATCH();
191  {
192  freecon(unlabeled);
193  PG_RE_THROW();
194  }
195  PG_END_TRY();
196 
197  freecon(unlabeled);
198  }
199  return avc_unlabeled;
200 }
201 
202 /*
203  * sepgsql_avc_compute
204  *
205  * A fallback path, when cache mishit. It asks SELinux its access control
206  * decision for the supplied pair of security context and object class.
207  */
208 static avc_cache *
209 sepgsql_avc_compute(const char *scontext, const char *tcontext, uint16 tclass)
210 {
211  char *ucontext = NULL;
212  char *ncontext = NULL;
213  MemoryContext oldctx;
214  avc_cache *cache;
215  uint32 hash;
216  int index;
217  struct av_decision avd;
218 
219  hash = sepgsql_avc_hash(scontext, tcontext, tclass);
220  index = hash % AVC_NUM_SLOTS;
221 
222  /*
223  * Validation check of the supplied security context. Because it always
224  * invoke system-call, frequent check should be avoided. Unless security
225  * policy is reloaded, validation status shall be kept, so we also cache
226  * whether the supplied security context was valid, or not.
227  */
228  if (security_check_context_raw((security_context_t) tcontext) != 0)
229  ucontext = sepgsql_avc_unlabeled();
230 
231  /*
232  * Ask SELinux its access control decision
233  */
234  if (!ucontext)
235  sepgsql_compute_avd(scontext, tcontext, tclass, &avd);
236  else
237  sepgsql_compute_avd(scontext, ucontext, tclass, &avd);
238 
239  /*
240  * It also caches a security label to be switched when a client labeled as
241  * 'scontext' executes a procedure labeled as 'tcontext', not only access
242  * control decision on the procedure. The security label to be switched
243  * shall be computed uniquely on a pair of 'scontext' and 'tcontext',
244  * thus, it is reasonable to cache the new label on avc, and enables to
245  * reduce unnecessary system calls. It shall be referenced at
246  * sepgsql_needs_fmgr_hook to check whether the supplied function is a
247  * trusted procedure, or not.
248  */
249  if (tclass == SEPG_CLASS_DB_PROCEDURE)
250  {
251  if (!ucontext)
252  ncontext = sepgsql_compute_create(scontext, tcontext,
254  else
255  ncontext = sepgsql_compute_create(scontext, ucontext,
257  if (strcmp(scontext, ncontext) == 0)
258  {
259  pfree(ncontext);
260  ncontext = NULL;
261  }
262  }
263 
264  /*
265  * Set up an avc_cache object
266  */
267  oldctx = MemoryContextSwitchTo(avc_mem_cxt);
268 
269  cache = palloc0(sizeof(avc_cache));
270 
271  cache->hash = hash;
272  cache->scontext = pstrdup(scontext);
273  cache->tcontext = pstrdup(tcontext);
274  cache->tclass = tclass;
275 
276  cache->allowed = avd.allowed;
277  cache->auditallow = avd.auditallow;
278  cache->auditdeny = avd.auditdeny;
279  cache->hot_cache = true;
280  if (avd.flags & SELINUX_AVD_FLAGS_PERMISSIVE)
281  cache->permissive = true;
282  if (!ucontext)
283  cache->tcontext_is_valid = true;
284  if (ncontext)
285  cache->ncontext = pstrdup(ncontext);
286 
287  avc_num_caches++;
288 
291 
292  avc_slots[index] = lcons(cache, avc_slots[index]);
293 
294  MemoryContextSwitchTo(oldctx);
295 
296  return cache;
297 }
298 
299 /*
300  * sepgsql_avc_lookup
301  *
302  * Look up a cache entry that matches the supplied security contexts and
303  * object class. If not found, create a new cache entry.
304  */
305 static avc_cache *
306 sepgsql_avc_lookup(const char *scontext, const char *tcontext, uint16 tclass)
307 {
308  avc_cache *cache;
309  ListCell *cell;
310  uint32 hash;
311  int index;
312 
313  hash = sepgsql_avc_hash(scontext, tcontext, tclass);
314  index = hash % AVC_NUM_SLOTS;
315 
316  foreach(cell, avc_slots[index])
317  {
318  cache = lfirst(cell);
319 
320  if (cache->hash == hash &&
321  cache->tclass == tclass &&
322  strcmp(cache->tcontext, tcontext) == 0 &&
323  strcmp(cache->scontext, scontext) == 0)
324  {
325  cache->hot_cache = true;
326  return cache;
327  }
328  }
329  /* not found, so insert a new cache */
330  return sepgsql_avc_compute(scontext, tcontext, tclass);
331 }
332 
333 /*
334  * sepgsql_avc_check_perms(_label)
335  *
336  * It returns 'true', if the security policy suggested to allow the required
337  * permissions. Otherwise, it returns 'false' or raises an error according
338  * to the 'abort_on_violation' argument.
339  * The 'tobject' and 'tclass' identify the target object being referenced,
340  * and 'required' is a bitmask of permissions (SEPG_*__*) defined for each
341  * object classes.
342  * The 'audit_name' is the object name (optional). If SEPGSQL_AVC_NOAUDIT
343  * was supplied, it means to skip all the audit messages.
344  */
345 bool
346 sepgsql_avc_check_perms_label(const char *tcontext,
347  uint16 tclass, uint32 required,
348  const char *audit_name,
349  bool abort_on_violation)
350 {
351  char *scontext = sepgsql_get_client_label();
352  avc_cache *cache;
353  uint32 denied;
354  uint32 audited;
355  bool result;
356 
358  do
359  {
360  result = true;
361 
362  /*
363  * If the target object is unlabeled, we perform the check using the
364  * label supplied by sepgsql_avc_unlabeled().
365  */
366  if (tcontext)
367  cache = sepgsql_avc_lookup(scontext, tcontext, tclass);
368  else
369  cache = sepgsql_avc_lookup(scontext,
370  sepgsql_avc_unlabeled(), tclass);
371 
372  denied = required & ~cache->allowed;
373 
374  /*
375  * Compute permissions to be audited
376  */
378  audited = (denied ? (denied & ~0) : (required & ~0));
379  else
380  audited = denied ? (denied & cache->auditdeny)
381  : (required & cache->auditallow);
382 
383  if (denied)
384  {
385  /*
386  * In permissive mode or permissive domain, violated permissions
387  * shall be audited to the log files at once, and then implicitly
388  * allowed to avoid a flood of access denied logs, because the
389  * purpose of permissive mode/domain is to collect a violation log
390  * that will make it possible to fix up the security policy.
391  */
392  if (!sepgsql_getenforce() || cache->permissive)
393  cache->allowed |= required;
394  else
395  result = false;
396  }
397  } while (!sepgsql_avc_check_valid());
398 
399  /*
400  * In the case when we have something auditable actions here,
401  * sepgsql_audit_log shall be called with text representation of security
402  * labels for both of subject and object. It records this access
403  * violation, so DBA will be able to find out unexpected security problems
404  * later.
405  */
406  if (audited != 0 &&
407  audit_name != SEPGSQL_AVC_NOAUDIT &&
409  {
410  sepgsql_audit_log(denied != 0,
411  cache->scontext,
412  cache->tcontext_is_valid ?
413  cache->tcontext : sepgsql_avc_unlabeled(),
414  cache->tclass,
415  audited,
416  audit_name);
417  }
418 
419  if (abort_on_violation && !result)
420  ereport(ERROR,
421  (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
422  errmsg("SELinux: security policy violation")));
423 
424  return result;
425 }
426 
427 bool
429  uint16 tclass, uint32 required,
430  const char *audit_name,
431  bool abort_on_violation)
432 {
433  char *tcontext = GetSecurityLabel(tobject, SEPGSQL_LABEL_TAG);
434  bool rc;
435 
436  rc = sepgsql_avc_check_perms_label(tcontext,
437  tclass, required,
438  audit_name, abort_on_violation);
439  if (tcontext)
440  pfree(tcontext);
441 
442  return rc;
443 }
444 
445 /*
446  * sepgsql_avc_trusted_proc
447  *
448  * If the supplied function OID is configured as a trusted procedure, this
449  * function will return a security label to be used during the execution of
450  * that function. Otherwise, it returns NULL.
451  */
452 char *
454 {
455  char *scontext = sepgsql_get_client_label();
456  char *tcontext;
457  ObjectAddress tobject;
458  avc_cache *cache;
459 
460  tobject.classId = ProcedureRelationId;
461  tobject.objectId = functionId;
462  tobject.objectSubId = 0;
463  tcontext = GetSecurityLabel(&tobject, SEPGSQL_LABEL_TAG);
464 
466  do
467  {
468  if (tcontext)
469  cache = sepgsql_avc_lookup(scontext, tcontext,
471  else
472  cache = sepgsql_avc_lookup(scontext, sepgsql_avc_unlabeled(),
474  } while (!sepgsql_avc_check_valid());
475 
476  return cache->ncontext;
477 }
478 
479 /*
480  * sepgsql_avc_exit
481  *
482  * Clean up userspace AVC on process exit.
483  */
484 static void
486 {
487  selinux_status_close();
488 }
489 
490 /*
491  * sepgsql_avc_init
492  *
493  * Initialize the userspace AVC. This should be called from _PG_init.
494  */
495 void
497 {
498  int rc;
499 
500  /*
501  * All the avc stuff shall be allocated in avc_mem_cxt
502  */
504  "userspace access vector cache",
506  memset(avc_slots, 0, sizeof(avc_slots));
507  avc_num_caches = 0;
508  avc_lru_hint = 0;
510 
511  /*
512  * SELinux allows to mmap(2) its kernel status page in read-only mode to
513  * inform userspace applications its status updating (such as policy
514  * reloading) without system-call invocations. This feature is only
515  * supported in Linux-2.6.38 or later, however, libselinux provides a
516  * fallback mode to know its status using netlink sockets.
517  */
518  rc = selinux_status_open(1);
519  if (rc < 0)
520  ereport(ERROR,
521  (errcode(ERRCODE_INTERNAL_ERROR),
522  errmsg("SELinux: could not open selinux status : %m")));
523  else if (rc > 0)
524  ereport(LOG,
525  (errmsg("SELinux: kernel status page uses fallback mode")));
526 
527  /* Arrange to close selinux status page on process exit. */
529 }
#define SEPGSQL_MODE_INTERNAL
Definition: sepgsql.h:30
static char * avc_unlabeled
Definition: uavc.c:62
static int32 next
Definition: blutils.c:210
bool permissive
Definition: uavc.c:42
char * ncontext
Definition: uavc.c:46
#define AVC_DEF_THRESHOLD
Definition: uavc.c:55
static uint32 sepgsql_avc_hash(const char *scontext, const char *tcontext, uint16 tclass)
Definition: uavc.c:68
void on_proc_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:292
static void sepgsql_avc_reset(void)
Definition: uavc.c:79
#define ProcedureRelationId
Definition: pg_proc.h:33
char * pstrdup(const char *in)
Definition: mcxt.c:1077
void sepgsql_compute_avd(const char *scontext, const char *tcontext, uint16 tclass, struct av_decision *avd)
Definition: selinux.c:732
Definition: uavc.c:31
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
int errcode(int sqlerrcode)
Definition: elog.c:575
bool sepgsql_getenforce(void)
Definition: selinux.c:648
static char * sepgsql_avc_unlabeled(void)
Definition: uavc.c:176
uint32 hash
Definition: uavc.c:33
#define AVC_NUM_RECLAIM
Definition: uavc.c:54
return result
Definition: formatting.c:1633
void MemoryContextReset(MemoryContext context)
Definition: mcxt.c:135
uint32 allowed
Definition: uavc.c:38
uint32 auditallow
Definition: uavc.c:39
static avc_cache * sepgsql_avc_lookup(const char *scontext, const char *tcontext, uint16 tclass)
Definition: uavc.c:306
#define LOG
Definition: elog.h:26
unsigned int Oid
Definition: postgres_ext.h:31
static avc_cache * sepgsql_avc_compute(const char *scontext, const char *tcontext, uint16 tclass)
Definition: uavc.c:209
#define SEPGSQL_AVC_NOAUDIT
Definition: sepgsql.h:255
uint16 tclass
Definition: uavc.c:36
Definition: type.h:89
#define SEPG_CLASS_PROCESS
Definition: sepgsql.h:36
int sepgsql_get_mode(void)
Definition: selinux.c:622
unsigned short uint16
Definition: c.h:267
void pfree(void *pointer)
Definition: mcxt.c:950
bool hot_cache
Definition: uavc.c:43
bool tcontext_is_valid
Definition: uavc.c:44
static MemoryContext avc_mem_cxt
Definition: uavc.c:57
#define ERROR
Definition: elog.h:43
void sepgsql_audit_log(bool denied, const char *scontext, const char *tcontext, uint16 tclass, uint32 audited, const char *audit_name)
Definition: selinux.c:677
char * GetSecurityLabel(const ObjectAddress *object, const char *provider)
Definition: seclabel.c:195
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:165
char * tcontext
Definition: uavc.c:35
bool sepgsql_get_debug_audit(void)
Definition: hooks.c:76
char * scontext
Definition: uavc.c:34
#define SEPG_CLASS_DB_PROCEDURE
Definition: sepgsql.h:48
void sepgsql_avc_init(void)
Definition: uavc.c:496
static void sepgsql_avc_exit(int code, Datum arg)
Definition: uavc.c:485
static ListCell * list_head(const List *l)
Definition: pg_list.h:77
#define SEPGSQL_LABEL_TAG
Definition: sepgsql.h:23
unsigned int uint32
Definition: c.h:268
static int avc_lru_hint
Definition: uavc.c:60
char * sepgsql_avc_trusted_proc(Oid functionId)
Definition: uavc.c:453
#define lnext(lc)
Definition: pg_list.h:105
#define ereport(elevel, rest)
Definition: elog.h:122
MemoryContext TopMemoryContext
Definition: mcxt.c:43
static int avc_num_caches
Definition: uavc.c:59
#define AVC_NUM_SLOTS
Definition: uavc.c:53
List * list_delete_cell(List *list, ListCell *cell, ListCell *prev)
Definition: list.c:528
MemoryContext AllocSetContextCreate(MemoryContext parent, const char *name, Size minContextSize, Size initBlockSize, Size maxBlockSize)
Definition: aset.c:322
void * palloc0(Size size)
Definition: mcxt.c:878
uintptr_t Datum
Definition: postgres.h:372
List * lcons(void *datum, List *list)
Definition: list.c:259
#define PG_CATCH()
Definition: elog.h:293
bool sepgsql_avc_check_perms(const ObjectAddress *tobject, uint16 tclass, uint32 required, const char *audit_name, bool abort_on_violation)
Definition: uavc.c:428
#define NULL
Definition: c.h:229
#define lfirst(lc)
Definition: pg_list.h:106
Datum hash_any(register const unsigned char *k, register int keylen)
Definition: hashfunc.c:307
static List * avc_slots[AVC_NUM_SLOTS]
Definition: uavc.c:58
static void sepgsql_avc_reclaim(void)
Definition: uavc.c:93
#define PG_RE_THROW()
Definition: elog.h:314
char * sepgsql_get_client_label(void)
Definition: label.c:91
static bool sepgsql_avc_check_valid(void)
Definition: uavc.c:158
uint32 auditdeny
Definition: uavc.c:40
char * sepgsql_compute_create(const char *scontext, const char *tcontext, uint16 tclass, const char *objname)
Definition: selinux.c:837
int errmsg(const char *fmt,...)
Definition: elog.c:797
char * MemoryContextStrdup(MemoryContext context, const char *string)
Definition: mcxt.c:1064
void * arg
#define PG_TRY()
Definition: elog.h:284
Definition: pg_list.h:45
static unsigned hash(unsigned *uv, int n)
Definition: rege_dfa.c:541
bool sepgsql_avc_check_perms_label(const char *tcontext, uint16 tclass, uint32 required, const char *audit_name, bool abort_on_violation)
Definition: uavc.c:346
#define PG_END_TRY()
Definition: elog.h:300
static int avc_threshold
Definition: uavc.c:61