npf_tableset.c revision 1.33 1 1.1 rmind /*-
2 1.33 rmind * Copyright (c) 2009-2019 The NetBSD Foundation, Inc.
3 1.1 rmind * All rights reserved.
4 1.1 rmind *
5 1.1 rmind * This material is based upon work partially supported by The
6 1.1 rmind * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
7 1.1 rmind *
8 1.1 rmind * Redistribution and use in source and binary forms, with or without
9 1.1 rmind * modification, are permitted provided that the following conditions
10 1.1 rmind * are met:
11 1.1 rmind * 1. Redistributions of source code must retain the above copyright
12 1.1 rmind * notice, this list of conditions and the following disclaimer.
13 1.1 rmind * 2. Redistributions in binary form must reproduce the above copyright
14 1.1 rmind * notice, this list of conditions and the following disclaimer in the
15 1.1 rmind * documentation and/or other materials provided with the distribution.
16 1.1 rmind *
17 1.1 rmind * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
18 1.1 rmind * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
19 1.1 rmind * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
20 1.1 rmind * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
21 1.1 rmind * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 1.1 rmind * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 1.1 rmind * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 1.1 rmind * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 1.1 rmind * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 1.1 rmind * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 1.1 rmind * POSSIBILITY OF SUCH DAMAGE.
28 1.1 rmind */
29 1.1 rmind
30 1.1 rmind /*
31 1.4 rmind * NPF tableset module.
32 1.1 rmind *
33 1.15 rmind * Notes
34 1.15 rmind *
35 1.15 rmind * The tableset is an array of tables. After the creation, the array
36 1.15 rmind * is immutable. The caller is responsible to synchronise the access
37 1.29 rmind * to the tableset.
38 1.33 rmind *
39 1.33 rmind * Warning (not applicable for the userspace npfkern):
40 1.33 rmind *
41 1.33 rmind * The thmap_put()/thmap_del() are not called from the interrupt
42 1.33 rmind * context and are protected by a mutex(9), therefore they do not
43 1.33 rmind * SPL wrappers -- see the comment at the top of the npf_conndb.c
44 1.33 rmind * source file.
45 1.1 rmind */
46 1.1 rmind
47 1.25 christos #ifdef _KERNEL
48 1.1 rmind #include <sys/cdefs.h>
49 1.33 rmind __KERNEL_RCSID(0, "$NetBSD: npf_tableset.c,v 1.33 2019/07/23 00:52:01 rmind Exp $");
50 1.1 rmind
51 1.1 rmind #include <sys/param.h>
52 1.10 rmind #include <sys/types.h>
53 1.1 rmind
54 1.1 rmind #include <sys/atomic.h>
55 1.21 rmind #include <sys/cdbr.h>
56 1.1 rmind #include <sys/kmem.h>
57 1.1 rmind #include <sys/pool.h>
58 1.1 rmind #include <sys/queue.h>
59 1.28 rmind #include <sys/mutex.h>
60 1.29 rmind #include <sys/thmap.h>
61 1.1 rmind
62 1.25 christos #include "lpm.h"
63 1.25 christos #endif
64 1.25 christos
65 1.1 rmind #include "npf_impl.h"
66 1.1 rmind
67 1.15 rmind typedef struct npf_tblent {
68 1.24 christos LIST_ENTRY(npf_tblent) te_listent;
69 1.24 christos uint16_t te_preflen;
70 1.24 christos uint16_t te_alen;
71 1.13 rmind npf_addr_t te_addr;
72 1.15 rmind } npf_tblent_t;
73 1.1 rmind
74 1.29 rmind #define NPF_ADDRLEN2IDX(alen) ((alen) >> 4)
75 1.29 rmind #define NPF_ADDR_SLOTS (2)
76 1.1 rmind
77 1.1 rmind struct npf_table {
78 1.19 rmind /*
79 1.29 rmind * The storage type can be: a) hashmap b) LPM c) cdb.
80 1.19 rmind * There are separate trees for IPv4 and IPv6.
81 1.19 rmind */
82 1.21 rmind union {
83 1.21 rmind struct {
84 1.29 rmind thmap_t * t_map;
85 1.29 rmind LIST_HEAD(, npf_tblent) t_gc;
86 1.21 rmind };
87 1.29 rmind lpm_t * t_lpm;
88 1.21 rmind struct {
89 1.21 rmind void * t_blob;
90 1.21 rmind size_t t_bsize;
91 1.21 rmind struct cdbr * t_cdb;
92 1.21 rmind };
93 1.29 rmind struct {
94 1.29 rmind npf_tblent_t ** t_elements[NPF_ADDR_SLOTS];
95 1.29 rmind unsigned t_allocated[NPF_ADDR_SLOTS];
96 1.29 rmind unsigned t_used[NPF_ADDR_SLOTS];
97 1.29 rmind };
98 1.21 rmind } /* C11 */;
99 1.29 rmind LIST_HEAD(, npf_tblent) t_list;
100 1.29 rmind unsigned t_nitems;
101 1.19 rmind
102 1.19 rmind /*
103 1.19 rmind * Table ID, type and lock. The ID may change during the
104 1.19 rmind * config reload, it is protected by the npf_config_lock.
105 1.19 rmind */
106 1.19 rmind int t_type;
107 1.29 rmind unsigned t_id;
108 1.28 rmind kmutex_t t_lock;
109 1.19 rmind
110 1.29 rmind /* Reference count and table name. */
111 1.29 rmind unsigned t_refcnt;
112 1.19 rmind char t_name[NPF_TABLE_MAXNAMELEN];
113 1.19 rmind };
114 1.19 rmind
115 1.19 rmind struct npf_tableset {
116 1.29 rmind unsigned ts_nitems;
117 1.19 rmind npf_table_t * ts_map[];
118 1.1 rmind };
119 1.1 rmind
120 1.19 rmind #define NPF_TABLESET_SIZE(n) \
121 1.19 rmind (offsetof(npf_tableset_t, ts_map[n]) * sizeof(npf_table_t *))
122 1.19 rmind
123 1.29 rmind #define NPF_IFADDR_STEP 4
124 1.13 rmind
125 1.13 rmind static pool_cache_t tblent_cache __read_mostly;
126 1.1 rmind
127 1.1 rmind /*
128 1.1 rmind * npf_table_sysinit: initialise tableset structures.
129 1.1 rmind */
130 1.4 rmind void
131 1.1 rmind npf_tableset_sysinit(void)
132 1.1 rmind {
133 1.29 rmind tblent_cache = pool_cache_init(sizeof(npf_tblent_t), 0,
134 1.14 rmind 0, 0, "npftblpl", NULL, IPL_NONE, NULL, NULL, NULL);
135 1.1 rmind }
136 1.1 rmind
137 1.1 rmind void
138 1.1 rmind npf_tableset_sysfini(void)
139 1.1 rmind {
140 1.1 rmind pool_cache_destroy(tblent_cache);
141 1.1 rmind }
142 1.1 rmind
143 1.1 rmind npf_tableset_t *
144 1.19 rmind npf_tableset_create(u_int nitems)
145 1.1 rmind {
146 1.19 rmind npf_tableset_t *ts = kmem_zalloc(NPF_TABLESET_SIZE(nitems), KM_SLEEP);
147 1.19 rmind ts->ts_nitems = nitems;
148 1.19 rmind return ts;
149 1.1 rmind }
150 1.1 rmind
151 1.1 rmind void
152 1.19 rmind npf_tableset_destroy(npf_tableset_t *ts)
153 1.1 rmind {
154 1.1 rmind /*
155 1.19 rmind * Destroy all tables (no references should be held, since the
156 1.19 rmind * ruleset should be destroyed before).
157 1.1 rmind */
158 1.19 rmind for (u_int tid = 0; tid < ts->ts_nitems; tid++) {
159 1.19 rmind npf_table_t *t = ts->ts_map[tid];
160 1.19 rmind
161 1.17 rmind if (t && atomic_dec_uint_nv(&t->t_refcnt) == 0) {
162 1.1 rmind npf_table_destroy(t);
163 1.1 rmind }
164 1.1 rmind }
165 1.19 rmind kmem_free(ts, NPF_TABLESET_SIZE(ts->ts_nitems));
166 1.1 rmind }
167 1.1 rmind
168 1.1 rmind /*
169 1.1 rmind * npf_tableset_insert: insert the table into the specified tableset.
170 1.1 rmind *
171 1.13 rmind * => Returns 0 on success. Fails and returns error if ID is already used.
172 1.1 rmind */
173 1.1 rmind int
174 1.19 rmind npf_tableset_insert(npf_tableset_t *ts, npf_table_t *t)
175 1.1 rmind {
176 1.1 rmind const u_int tid = t->t_id;
177 1.1 rmind int error;
178 1.1 rmind
179 1.19 rmind KASSERT((u_int)tid < ts->ts_nitems);
180 1.1 rmind
181 1.19 rmind if (ts->ts_map[tid] == NULL) {
182 1.17 rmind atomic_inc_uint(&t->t_refcnt);
183 1.19 rmind ts->ts_map[tid] = t;
184 1.1 rmind error = 0;
185 1.1 rmind } else {
186 1.1 rmind error = EEXIST;
187 1.1 rmind }
188 1.1 rmind return error;
189 1.1 rmind }
190 1.1 rmind
191 1.26 rmind npf_table_t *
192 1.26 rmind npf_tableset_swap(npf_tableset_t *ts, npf_table_t *newt)
193 1.26 rmind {
194 1.26 rmind const u_int tid = newt->t_id;
195 1.26 rmind npf_table_t *oldt = ts->ts_map[tid];
196 1.26 rmind
197 1.26 rmind KASSERT(tid < ts->ts_nitems);
198 1.26 rmind KASSERT(oldt->t_id == newt->t_id);
199 1.26 rmind
200 1.26 rmind newt->t_refcnt = oldt->t_refcnt;
201 1.26 rmind oldt->t_refcnt = 0;
202 1.26 rmind
203 1.26 rmind return atomic_swap_ptr(&ts->ts_map[tid], newt);
204 1.26 rmind }
205 1.26 rmind
206 1.1 rmind /*
207 1.19 rmind * npf_tableset_getbyname: look for a table in the set given the name.
208 1.19 rmind */
209 1.19 rmind npf_table_t *
210 1.19 rmind npf_tableset_getbyname(npf_tableset_t *ts, const char *name)
211 1.19 rmind {
212 1.19 rmind npf_table_t *t;
213 1.19 rmind
214 1.19 rmind for (u_int tid = 0; tid < ts->ts_nitems; tid++) {
215 1.19 rmind if ((t = ts->ts_map[tid]) == NULL)
216 1.19 rmind continue;
217 1.19 rmind if (strcmp(name, t->t_name) == 0)
218 1.19 rmind return t;
219 1.19 rmind }
220 1.19 rmind return NULL;
221 1.19 rmind }
222 1.19 rmind
223 1.19 rmind npf_table_t *
224 1.19 rmind npf_tableset_getbyid(npf_tableset_t *ts, u_int tid)
225 1.19 rmind {
226 1.19 rmind if (__predict_true(tid < ts->ts_nitems)) {
227 1.19 rmind return ts->ts_map[tid];
228 1.19 rmind }
229 1.19 rmind return NULL;
230 1.19 rmind }
231 1.19 rmind
232 1.19 rmind /*
233 1.15 rmind * npf_tableset_reload: iterate all tables and if the new table is of the
234 1.15 rmind * same type and has no items, then we preserve the old one and its entries.
235 1.15 rmind *
236 1.15 rmind * => The caller is responsible for providing synchronisation.
237 1.15 rmind */
238 1.15 rmind void
239 1.25 christos npf_tableset_reload(npf_t *npf, npf_tableset_t *nts, npf_tableset_t *ots)
240 1.15 rmind {
241 1.19 rmind for (u_int tid = 0; tid < nts->ts_nitems; tid++) {
242 1.19 rmind npf_table_t *t, *ot;
243 1.19 rmind
244 1.19 rmind if ((t = nts->ts_map[tid]) == NULL) {
245 1.19 rmind continue;
246 1.19 rmind }
247 1.15 rmind
248 1.19 rmind /* If our table has entries, just load it. */
249 1.19 rmind if (t->t_nitems) {
250 1.15 rmind continue;
251 1.15 rmind }
252 1.19 rmind
253 1.19 rmind /* Look for a currently existing table with such name. */
254 1.19 rmind ot = npf_tableset_getbyname(ots, t->t_name);
255 1.19 rmind if (ot == NULL) {
256 1.19 rmind /* Not found: we have a new table. */
257 1.19 rmind continue;
258 1.19 rmind }
259 1.19 rmind
260 1.19 rmind /* Found. Did the type change? */
261 1.19 rmind if (t->t_type != ot->t_type) {
262 1.19 rmind /* Yes, load the new. */
263 1.15 rmind continue;
264 1.15 rmind }
265 1.17 rmind
266 1.17 rmind /*
267 1.19 rmind * Preserve the current table. Acquire a reference since
268 1.19 rmind * we are keeping it in the old table set. Update its ID.
269 1.17 rmind */
270 1.17 rmind atomic_inc_uint(&ot->t_refcnt);
271 1.19 rmind nts->ts_map[tid] = ot;
272 1.19 rmind
273 1.25 christos KASSERT(npf_config_locked_p(npf));
274 1.19 rmind ot->t_id = tid;
275 1.17 rmind
276 1.21 rmind /* Destroy the new table (we hold the only reference). */
277 1.17 rmind t->t_refcnt--;
278 1.15 rmind npf_table_destroy(t);
279 1.15 rmind }
280 1.15 rmind }
281 1.15 rmind
282 1.22 rmind int
283 1.28 rmind npf_tableset_export(npf_t *npf, const npf_tableset_t *ts, nvlist_t *npf_dict)
284 1.20 rmind {
285 1.20 rmind const npf_table_t *t;
286 1.20 rmind
287 1.25 christos KASSERT(npf_config_locked_p(npf));
288 1.20 rmind
289 1.20 rmind for (u_int tid = 0; tid < ts->ts_nitems; tid++) {
290 1.28 rmind nvlist_t *table;
291 1.28 rmind
292 1.20 rmind if ((t = ts->ts_map[tid]) == NULL) {
293 1.20 rmind continue;
294 1.20 rmind }
295 1.28 rmind table = nvlist_create(0);
296 1.28 rmind nvlist_add_string(table, "name", t->t_name);
297 1.28 rmind nvlist_add_number(table, "type", t->t_type);
298 1.28 rmind nvlist_add_number(table, "id", tid);
299 1.20 rmind
300 1.28 rmind nvlist_append_nvlist_array(npf_dict, "tables", table);
301 1.28 rmind nvlist_destroy(table);
302 1.20 rmind }
303 1.22 rmind return 0;
304 1.20 rmind }
305 1.20 rmind
306 1.15 rmind /*
307 1.13 rmind * Few helper routines.
308 1.1 rmind */
309 1.1 rmind
310 1.29 rmind static void
311 1.29 rmind table_ipset_flush(npf_table_t *t)
312 1.1 rmind {
313 1.13 rmind npf_tblent_t *ent;
314 1.1 rmind
315 1.29 rmind while ((ent = LIST_FIRST(&t->t_list)) != NULL) {
316 1.29 rmind thmap_del(t->t_map, &ent->te_addr, ent->te_alen);
317 1.29 rmind LIST_REMOVE(ent, te_listent);
318 1.29 rmind pool_cache_put(tblent_cache, ent);
319 1.13 rmind }
320 1.29 rmind t->t_nitems = 0;
321 1.1 rmind }
322 1.1 rmind
323 1.13 rmind static void
324 1.29 rmind table_tree_flush(npf_table_t *t)
325 1.18 rmind {
326 1.29 rmind npf_tblent_t *ent;
327 1.18 rmind
328 1.29 rmind while ((ent = LIST_FIRST(&t->t_list)) != NULL) {
329 1.29 rmind LIST_REMOVE(ent, te_listent);
330 1.29 rmind pool_cache_put(tblent_cache, ent);
331 1.18 rmind }
332 1.29 rmind lpm_clear(t->t_lpm, NULL, NULL);
333 1.29 rmind t->t_nitems = 0;
334 1.18 rmind }
335 1.18 rmind
336 1.18 rmind static void
337 1.29 rmind table_ifaddr_flush(npf_table_t *t)
338 1.1 rmind {
339 1.13 rmind npf_tblent_t *ent;
340 1.1 rmind
341 1.29 rmind for (unsigned i = 0; i < NPF_ADDR_SLOTS; i++) {
342 1.29 rmind size_t len;
343 1.29 rmind
344 1.29 rmind if (!t->t_allocated[i]) {
345 1.29 rmind KASSERT(t->t_elements[i] == NULL);
346 1.29 rmind continue;
347 1.29 rmind }
348 1.29 rmind len = t->t_allocated[i] * sizeof(npf_tblent_t *);
349 1.29 rmind kmem_free(t->t_elements[i], len);
350 1.29 rmind t->t_elements[i] = NULL;
351 1.29 rmind t->t_allocated[i] = 0;
352 1.29 rmind t->t_used[i] = 0;
353 1.29 rmind }
354 1.24 christos while ((ent = LIST_FIRST(&t->t_list)) != NULL) {
355 1.24 christos LIST_REMOVE(ent, te_listent);
356 1.13 rmind pool_cache_put(tblent_cache, ent);
357 1.13 rmind }
358 1.29 rmind t->t_nitems = 0;
359 1.1 rmind }
360 1.1 rmind
361 1.1 rmind /*
362 1.1 rmind * npf_table_create: create table with a specified ID.
363 1.1 rmind */
364 1.1 rmind npf_table_t *
365 1.21 rmind npf_table_create(const char *name, u_int tid, int type,
366 1.28 rmind const void *blob, size_t size)
367 1.1 rmind {
368 1.1 rmind npf_table_t *t;
369 1.1 rmind
370 1.25 christos t = kmem_zalloc(sizeof(npf_table_t), KM_SLEEP);
371 1.19 rmind strlcpy(t->t_name, name, NPF_TABLE_MAXNAMELEN);
372 1.1 rmind
373 1.1 rmind switch (type) {
374 1.29 rmind case NPF_TABLE_LPM:
375 1.30 christos t->t_lpm = lpm_create(KM_NOSLEEP);
376 1.29 rmind if (t->t_lpm == NULL) {
377 1.24 christos goto out;
378 1.25 christos }
379 1.24 christos LIST_INIT(&t->t_list);
380 1.1 rmind break;
381 1.29 rmind case NPF_TABLE_IPSET:
382 1.29 rmind t->t_map = thmap_create(0, NULL, THMAP_NOCOPY);
383 1.29 rmind if (t->t_map == NULL) {
384 1.24 christos goto out;
385 1.25 christos }
386 1.1 rmind break;
387 1.29 rmind case NPF_TABLE_CONST:
388 1.28 rmind t->t_blob = kmem_alloc(size, KM_SLEEP);
389 1.28 rmind if (t->t_blob == NULL) {
390 1.28 rmind goto out;
391 1.28 rmind }
392 1.28 rmind memcpy(t->t_blob, blob, size);
393 1.21 rmind t->t_bsize = size;
394 1.28 rmind
395 1.28 rmind t->t_cdb = cdbr_open_mem(t->t_blob, size,
396 1.28 rmind CDBR_DEFAULT, NULL, NULL);
397 1.21 rmind if (t->t_cdb == NULL) {
398 1.28 rmind kmem_free(t->t_blob, t->t_bsize);
399 1.24 christos goto out;
400 1.21 rmind }
401 1.21 rmind t->t_nitems = cdbr_entries(t->t_cdb);
402 1.21 rmind break;
403 1.29 rmind case NPF_TABLE_IFADDR:
404 1.29 rmind break;
405 1.1 rmind default:
406 1.1 rmind KASSERT(false);
407 1.1 rmind }
408 1.30 christos mutex_init(&t->t_lock, MUTEX_DEFAULT, IPL_NET);
409 1.1 rmind t->t_type = type;
410 1.1 rmind t->t_id = tid;
411 1.1 rmind return t;
412 1.24 christos out:
413 1.25 christos kmem_free(t, sizeof(npf_table_t));
414 1.24 christos return NULL;
415 1.1 rmind }
416 1.1 rmind
417 1.1 rmind /*
418 1.1 rmind * npf_table_destroy: free all table entries and table itself.
419 1.1 rmind */
420 1.1 rmind void
421 1.1 rmind npf_table_destroy(npf_table_t *t)
422 1.1 rmind {
423 1.17 rmind KASSERT(t->t_refcnt == 0);
424 1.1 rmind
425 1.1 rmind switch (t->t_type) {
426 1.29 rmind case NPF_TABLE_IPSET:
427 1.29 rmind table_ipset_flush(t);
428 1.29 rmind npf_table_gc(NULL, t);
429 1.29 rmind thmap_destroy(t->t_map);
430 1.1 rmind break;
431 1.29 rmind case NPF_TABLE_LPM:
432 1.24 christos table_tree_flush(t);
433 1.24 christos lpm_destroy(t->t_lpm);
434 1.1 rmind break;
435 1.29 rmind case NPF_TABLE_CONST:
436 1.21 rmind cdbr_close(t->t_cdb);
437 1.28 rmind kmem_free(t->t_blob, t->t_bsize);
438 1.21 rmind break;
439 1.29 rmind case NPF_TABLE_IFADDR:
440 1.29 rmind table_ifaddr_flush(t);
441 1.29 rmind break;
442 1.1 rmind default:
443 1.1 rmind KASSERT(false);
444 1.1 rmind }
445 1.28 rmind mutex_destroy(&t->t_lock);
446 1.25 christos kmem_free(t, sizeof(npf_table_t));
447 1.1 rmind }
448 1.1 rmind
449 1.26 rmind u_int
450 1.26 rmind npf_table_getid(npf_table_t *t)
451 1.26 rmind {
452 1.26 rmind return t->t_id;
453 1.26 rmind }
454 1.26 rmind
455 1.1 rmind /*
456 1.19 rmind * npf_table_check: validate the name, ID and type.
457 1.13 rmind */
458 1.1 rmind int
459 1.28 rmind npf_table_check(npf_tableset_t *ts, const char *name, uint64_t tid, uint64_t type)
460 1.1 rmind {
461 1.28 rmind if (tid >= ts->ts_nitems) {
462 1.1 rmind return EINVAL;
463 1.1 rmind }
464 1.19 rmind if (ts->ts_map[tid] != NULL) {
465 1.1 rmind return EEXIST;
466 1.1 rmind }
467 1.21 rmind switch (type) {
468 1.29 rmind case NPF_TABLE_LPM:
469 1.29 rmind case NPF_TABLE_IPSET:
470 1.29 rmind case NPF_TABLE_CONST:
471 1.29 rmind case NPF_TABLE_IFADDR:
472 1.21 rmind break;
473 1.21 rmind default:
474 1.1 rmind return EINVAL;
475 1.1 rmind }
476 1.19 rmind if (strlen(name) >= NPF_TABLE_MAXNAMELEN) {
477 1.19 rmind return ENAMETOOLONG;
478 1.19 rmind }
479 1.19 rmind if (npf_tableset_getbyname(ts, name)) {
480 1.20 rmind return EEXIST;
481 1.19 rmind }
482 1.1 rmind return 0;
483 1.1 rmind }
484 1.1 rmind
485 1.13 rmind static int
486 1.29 rmind table_ifaddr_insert(npf_table_t *t, const int alen, npf_tblent_t *ent)
487 1.29 rmind {
488 1.29 rmind const unsigned aidx = NPF_ADDRLEN2IDX(alen);
489 1.29 rmind const unsigned allocated = t->t_allocated[aidx];
490 1.29 rmind const unsigned used = t->t_used[aidx];
491 1.13 rmind
492 1.13 rmind /*
493 1.29 rmind * No need to check for duplicates.
494 1.13 rmind */
495 1.29 rmind if (allocated <= used) {
496 1.29 rmind npf_tblent_t **old_elements = t->t_elements[aidx];
497 1.29 rmind npf_tblent_t **elements;
498 1.29 rmind size_t toalloc, newsize;
499 1.29 rmind
500 1.29 rmind toalloc = roundup2(allocated + 1, NPF_IFADDR_STEP);
501 1.29 rmind newsize = toalloc * sizeof(npf_tblent_t *);
502 1.29 rmind
503 1.31 christos elements = kmem_zalloc(newsize, KM_NOSLEEP);
504 1.32 christos if (elements == NULL) {
505 1.32 christos return ENOMEM;
506 1.32 christos }
507 1.29 rmind for (unsigned i = 0; i < used; i++) {
508 1.29 rmind elements[i] = old_elements[i];
509 1.29 rmind }
510 1.29 rmind if (allocated) {
511 1.29 rmind const size_t len = allocated * sizeof(npf_tblent_t *);
512 1.29 rmind KASSERT(old_elements != NULL);
513 1.29 rmind kmem_free(old_elements, len);
514 1.29 rmind }
515 1.29 rmind t->t_elements[aidx] = elements;
516 1.29 rmind t->t_allocated[aidx] = toalloc;
517 1.13 rmind }
518 1.29 rmind t->t_elements[aidx][used] = ent;
519 1.29 rmind t->t_used[aidx]++;
520 1.32 christos return 0;
521 1.13 rmind }
522 1.13 rmind
523 1.1 rmind /*
524 1.13 rmind * npf_table_insert: add an IP CIDR entry into the table.
525 1.1 rmind */
526 1.1 rmind int
527 1.19 rmind npf_table_insert(npf_table_t *t, const int alen,
528 1.6 zoltan const npf_addr_t *addr, const npf_netmask_t mask)
529 1.1 rmind {
530 1.13 rmind npf_tblent_t *ent;
531 1.13 rmind int error;
532 1.1 rmind
533 1.33 rmind error = npf_netmask_check(alen, mask);
534 1.13 rmind if (error) {
535 1.13 rmind return error;
536 1.8 rmind }
537 1.12 rmind ent = pool_cache_get(tblent_cache, PR_WAITOK);
538 1.13 rmind memcpy(&ent->te_addr, addr, alen);
539 1.13 rmind ent->te_alen = alen;
540 1.29 rmind ent->te_preflen = 0;
541 1.1 rmind
542 1.13 rmind /*
543 1.13 rmind * Insert the entry. Return an error on duplicate.
544 1.13 rmind */
545 1.28 rmind mutex_enter(&t->t_lock);
546 1.1 rmind switch (t->t_type) {
547 1.29 rmind case NPF_TABLE_IPSET:
548 1.13 rmind /*
549 1.29 rmind * Hashmap supports only IPs.
550 1.29 rmind *
551 1.29 rmind * Note: the key must be already persistent, since we
552 1.29 rmind * use THMAP_NOCOPY.
553 1.13 rmind */
554 1.13 rmind if (mask != NPF_NO_NETMASK) {
555 1.13 rmind error = EINVAL;
556 1.13 rmind break;
557 1.1 rmind }
558 1.29 rmind if (thmap_put(t->t_map, &ent->te_addr, alen, ent) == ent) {
559 1.29 rmind LIST_INSERT_HEAD(&t->t_list, ent, te_listent);
560 1.15 rmind t->t_nitems++;
561 1.1 rmind } else {
562 1.1 rmind error = EEXIST;
563 1.1 rmind }
564 1.1 rmind break;
565 1.29 rmind case NPF_TABLE_LPM: {
566 1.24 christos const unsigned preflen =
567 1.24 christos (mask == NPF_NO_NETMASK) ? (alen * 8) : mask;
568 1.29 rmind ent->te_preflen = preflen;
569 1.29 rmind
570 1.24 christos if (lpm_lookup(t->t_lpm, addr, alen) == NULL &&
571 1.24 christos lpm_insert(t->t_lpm, addr, alen, preflen, ent) == 0) {
572 1.24 christos LIST_INSERT_HEAD(&t->t_list, ent, te_listent);
573 1.15 rmind t->t_nitems++;
574 1.15 rmind error = 0;
575 1.13 rmind } else {
576 1.15 rmind error = EEXIST;
577 1.1 rmind }
578 1.1 rmind break;
579 1.13 rmind }
580 1.29 rmind case NPF_TABLE_CONST:
581 1.21 rmind error = EINVAL;
582 1.21 rmind break;
583 1.29 rmind case NPF_TABLE_IFADDR:
584 1.32 christos if ((error = table_ifaddr_insert(t, alen, ent)) != 0) {
585 1.32 christos break;
586 1.32 christos }
587 1.29 rmind LIST_INSERT_HEAD(&t->t_list, ent, te_listent);
588 1.29 rmind t->t_nitems++;
589 1.29 rmind break;
590 1.1 rmind default:
591 1.1 rmind KASSERT(false);
592 1.1 rmind }
593 1.28 rmind mutex_exit(&t->t_lock);
594 1.1 rmind
595 1.8 rmind if (error) {
596 1.12 rmind pool_cache_put(tblent_cache, ent);
597 1.1 rmind }
598 1.1 rmind return error;
599 1.1 rmind }
600 1.1 rmind
601 1.1 rmind /*
602 1.13 rmind * npf_table_remove: remove the IP CIDR entry from the table.
603 1.1 rmind */
604 1.1 rmind int
605 1.19 rmind npf_table_remove(npf_table_t *t, const int alen,
606 1.6 zoltan const npf_addr_t *addr, const npf_netmask_t mask)
607 1.1 rmind {
608 1.21 rmind npf_tblent_t *ent = NULL;
609 1.29 rmind int error;
610 1.1 rmind
611 1.33 rmind error = npf_netmask_check(alen, mask);
612 1.13 rmind if (error) {
613 1.13 rmind return error;
614 1.8 rmind }
615 1.15 rmind
616 1.28 rmind mutex_enter(&t->t_lock);
617 1.13 rmind switch (t->t_type) {
618 1.29 rmind case NPF_TABLE_IPSET:
619 1.29 rmind ent = thmap_del(t->t_map, addr, alen);
620 1.12 rmind if (__predict_true(ent != NULL)) {
621 1.24 christos LIST_REMOVE(ent, te_listent);
622 1.29 rmind LIST_INSERT_HEAD(&t->t_gc, ent, te_listent);
623 1.29 rmind ent = NULL; // to be G/C'ed
624 1.15 rmind t->t_nitems--;
625 1.29 rmind } else {
626 1.29 rmind error = ENOENT;
627 1.1 rmind }
628 1.1 rmind break;
629 1.29 rmind case NPF_TABLE_LPM:
630 1.24 christos ent = lpm_lookup(t->t_lpm, addr, alen);
631 1.12 rmind if (__predict_true(ent != NULL)) {
632 1.24 christos LIST_REMOVE(ent, te_listent);
633 1.24 christos lpm_remove(t->t_lpm, &ent->te_addr,
634 1.24 christos ent->te_alen, ent->te_preflen);
635 1.15 rmind t->t_nitems--;
636 1.29 rmind } else {
637 1.29 rmind error = ENOENT;
638 1.1 rmind }
639 1.1 rmind break;
640 1.29 rmind case NPF_TABLE_CONST:
641 1.29 rmind case NPF_TABLE_IFADDR:
642 1.21 rmind error = EINVAL;
643 1.21 rmind break;
644 1.1 rmind default:
645 1.1 rmind KASSERT(false);
646 1.13 rmind ent = NULL;
647 1.1 rmind }
648 1.28 rmind mutex_exit(&t->t_lock);
649 1.1 rmind
650 1.21 rmind if (ent) {
651 1.21 rmind pool_cache_put(tblent_cache, ent);
652 1.1 rmind }
653 1.21 rmind return error;
654 1.1 rmind }
655 1.1 rmind
656 1.1 rmind /*
657 1.13 rmind * npf_table_lookup: find the table according to ID, lookup and match
658 1.13 rmind * the contents with the specified IP address.
659 1.1 rmind */
660 1.1 rmind int
661 1.19 rmind npf_table_lookup(npf_table_t *t, const int alen, const npf_addr_t *addr)
662 1.1 rmind {
663 1.21 rmind const void *data;
664 1.21 rmind size_t dlen;
665 1.21 rmind bool found;
666 1.29 rmind int error;
667 1.1 rmind
668 1.33 rmind error = npf_netmask_check(alen, NPF_NO_NETMASK);
669 1.29 rmind if (error) {
670 1.29 rmind return error;
671 1.13 rmind }
672 1.13 rmind
673 1.1 rmind switch (t->t_type) {
674 1.29 rmind case NPF_TABLE_IPSET:
675 1.29 rmind found = thmap_get(t->t_map, addr, alen) != NULL;
676 1.1 rmind break;
677 1.29 rmind case NPF_TABLE_LPM:
678 1.28 rmind mutex_enter(&t->t_lock);
679 1.24 christos found = lpm_lookup(t->t_lpm, addr, alen) != NULL;
680 1.28 rmind mutex_exit(&t->t_lock);
681 1.21 rmind break;
682 1.29 rmind case NPF_TABLE_CONST:
683 1.21 rmind if (cdbr_find(t->t_cdb, addr, alen, &data, &dlen) == 0) {
684 1.29 rmind found = dlen == (unsigned)alen &&
685 1.25 christos memcmp(addr, data, dlen) == 0;
686 1.21 rmind } else {
687 1.21 rmind found = false;
688 1.21 rmind }
689 1.1 rmind break;
690 1.29 rmind case NPF_TABLE_IFADDR: {
691 1.29 rmind const unsigned aidx = NPF_ADDRLEN2IDX(alen);
692 1.29 rmind
693 1.29 rmind found = false;
694 1.29 rmind for (unsigned i = 0; i < t->t_used[aidx]; i++) {
695 1.29 rmind const npf_tblent_t *elm = t->t_elements[aidx][i];
696 1.29 rmind
697 1.29 rmind KASSERT(elm->te_alen == alen);
698 1.29 rmind
699 1.29 rmind if (memcmp(&elm->te_addr, addr, alen) == 0) {
700 1.29 rmind found = true;
701 1.29 rmind break;
702 1.29 rmind }
703 1.29 rmind }
704 1.29 rmind break;
705 1.29 rmind }
706 1.1 rmind default:
707 1.1 rmind KASSERT(false);
708 1.21 rmind found = false;
709 1.1 rmind }
710 1.1 rmind
711 1.21 rmind return found ? 0 : ENOENT;
712 1.1 rmind }
713 1.15 rmind
714 1.29 rmind npf_addr_t *
715 1.29 rmind npf_table_getsome(npf_table_t *t, const int alen, unsigned idx)
716 1.29 rmind {
717 1.29 rmind const unsigned aidx = NPF_ADDRLEN2IDX(alen);
718 1.29 rmind npf_tblent_t *elm;
719 1.29 rmind unsigned nitems;
720 1.29 rmind
721 1.29 rmind KASSERT(t->t_type == NPF_TABLE_IFADDR);
722 1.29 rmind KASSERT(aidx < NPF_ADDR_SLOTS);
723 1.29 rmind
724 1.29 rmind nitems = t->t_used[aidx];
725 1.29 rmind if (nitems == 0) {
726 1.29 rmind return NULL;
727 1.29 rmind }
728 1.29 rmind
729 1.29 rmind /*
730 1.29 rmind * No need to acquire the lock, since the table is immutable.
731 1.29 rmind */
732 1.29 rmind elm = t->t_elements[aidx][idx % nitems];
733 1.29 rmind return &elm->te_addr;
734 1.29 rmind }
735 1.29 rmind
736 1.15 rmind static int
737 1.21 rmind table_ent_copyout(const npf_addr_t *addr, const int alen, npf_netmask_t mask,
738 1.15 rmind void *ubuf, size_t len, size_t *off)
739 1.15 rmind {
740 1.15 rmind void *ubufp = (uint8_t *)ubuf + *off;
741 1.15 rmind npf_ioctl_ent_t uent;
742 1.15 rmind
743 1.15 rmind if ((*off += sizeof(npf_ioctl_ent_t)) > len) {
744 1.15 rmind return ENOMEM;
745 1.15 rmind }
746 1.21 rmind uent.alen = alen;
747 1.21 rmind memcpy(&uent.addr, addr, sizeof(npf_addr_t));
748 1.15 rmind uent.mask = mask;
749 1.15 rmind
750 1.15 rmind return copyout(&uent, ubufp, sizeof(npf_ioctl_ent_t));
751 1.15 rmind }
752 1.15 rmind
753 1.15 rmind static int
754 1.29 rmind table_generic_list(const npf_table_t *t, void *ubuf, size_t len)
755 1.15 rmind {
756 1.24 christos npf_tblent_t *ent;
757 1.24 christos size_t off = 0;
758 1.15 rmind int error = 0;
759 1.15 rmind
760 1.24 christos LIST_FOREACH(ent, &t->t_list, te_listent) {
761 1.24 christos error = table_ent_copyout(&ent->te_addr,
762 1.29 rmind ent->te_alen, ent->te_preflen, ubuf, len, &off);
763 1.21 rmind if (error)
764 1.21 rmind break;
765 1.21 rmind }
766 1.21 rmind return error;
767 1.21 rmind }
768 1.21 rmind
769 1.21 rmind static int
770 1.21 rmind table_cdb_list(npf_table_t *t, void *ubuf, size_t len)
771 1.21 rmind {
772 1.21 rmind size_t off = 0, dlen;
773 1.21 rmind const void *data;
774 1.21 rmind int error = 0;
775 1.21 rmind
776 1.21 rmind for (size_t i = 0; i < t->t_nitems; i++) {
777 1.21 rmind if (cdbr_get(t->t_cdb, i, &data, &dlen) != 0) {
778 1.21 rmind return EINVAL;
779 1.21 rmind }
780 1.21 rmind error = table_ent_copyout(data, dlen, 0, ubuf, len, &off);
781 1.15 rmind if (error)
782 1.15 rmind break;
783 1.15 rmind }
784 1.15 rmind return error;
785 1.15 rmind }
786 1.15 rmind
787 1.15 rmind /*
788 1.15 rmind * npf_table_list: copy a list of all table entries into a userspace buffer.
789 1.15 rmind */
790 1.15 rmind int
791 1.19 rmind npf_table_list(npf_table_t *t, void *ubuf, size_t len)
792 1.15 rmind {
793 1.15 rmind int error = 0;
794 1.15 rmind
795 1.28 rmind mutex_enter(&t->t_lock);
796 1.15 rmind switch (t->t_type) {
797 1.29 rmind case NPF_TABLE_IPSET:
798 1.29 rmind error = table_generic_list(t, ubuf, len);
799 1.15 rmind break;
800 1.29 rmind case NPF_TABLE_LPM:
801 1.29 rmind error = table_generic_list(t, ubuf, len);
802 1.16 rmind break;
803 1.29 rmind case NPF_TABLE_CONST:
804 1.21 rmind error = table_cdb_list(t, ubuf, len);
805 1.21 rmind break;
806 1.29 rmind case NPF_TABLE_IFADDR:
807 1.29 rmind error = table_generic_list(t, ubuf, len);
808 1.29 rmind break;
809 1.15 rmind default:
810 1.15 rmind KASSERT(false);
811 1.15 rmind }
812 1.28 rmind mutex_exit(&t->t_lock);
813 1.15 rmind
814 1.15 rmind return error;
815 1.15 rmind }
816 1.18 rmind
817 1.18 rmind /*
818 1.18 rmind * npf_table_flush: remove all table entries.
819 1.18 rmind */
820 1.18 rmind int
821 1.19 rmind npf_table_flush(npf_table_t *t)
822 1.18 rmind {
823 1.21 rmind int error = 0;
824 1.21 rmind
825 1.28 rmind mutex_enter(&t->t_lock);
826 1.18 rmind switch (t->t_type) {
827 1.29 rmind case NPF_TABLE_IPSET:
828 1.29 rmind table_ipset_flush(t);
829 1.18 rmind break;
830 1.29 rmind case NPF_TABLE_LPM:
831 1.24 christos table_tree_flush(t);
832 1.18 rmind break;
833 1.29 rmind case NPF_TABLE_CONST:
834 1.29 rmind case NPF_TABLE_IFADDR:
835 1.21 rmind error = EINVAL;
836 1.21 rmind break;
837 1.18 rmind default:
838 1.18 rmind KASSERT(false);
839 1.18 rmind }
840 1.28 rmind mutex_exit(&t->t_lock);
841 1.21 rmind return error;
842 1.18 rmind }
843 1.29 rmind
844 1.29 rmind void
845 1.29 rmind npf_table_gc(npf_t *npf, npf_table_t *t)
846 1.29 rmind {
847 1.29 rmind npf_tblent_t *ent;
848 1.29 rmind void *ref;
849 1.29 rmind
850 1.29 rmind if (t->t_type != NPF_TABLE_IPSET || LIST_EMPTY(&t->t_gc)) {
851 1.29 rmind return;
852 1.29 rmind }
853 1.29 rmind
854 1.29 rmind ref = thmap_stage_gc(t->t_map);
855 1.29 rmind if (npf) {
856 1.29 rmind npf_config_locked_p(npf);
857 1.29 rmind npf_config_sync(npf);
858 1.29 rmind }
859 1.29 rmind thmap_gc(t->t_map, ref);
860 1.29 rmind
861 1.29 rmind while ((ent = LIST_FIRST(&t->t_gc)) != NULL) {
862 1.29 rmind LIST_REMOVE(ent, te_listent);
863 1.29 rmind pool_cache_put(tblent_cache, ent);
864 1.29 rmind }
865 1.29 rmind }
866