2 * Copyright (c) 2007-2013 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
21 #include <linux/uaccess.h>
22 #include <linux/netdevice.h>
23 #include <linux/etherdevice.h>
24 #include <linux/if_ether.h>
25 #include <linux/if_vlan.h>
26 #include <net/llc_pdu.h>
27 #include <linux/kernel.h>
28 #include <linux/hash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
33 #include <linux/rcupdate.h>
34 #include <linux/if_arp.h>
36 #include <linux/ipv6.h>
37 #include <linux/sctp.h>
38 #include <linux/tcp.h>
39 #include <linux/udp.h>
40 #include <linux/icmp.h>
41 #include <linux/icmpv6.h>
42 #include <linux/rculist.h>
45 #include <net/ndisc.h>
49 #define TBL_MIN_BUCKETS 1024
50 #define REHASH_INTERVAL (10 * 60 * HZ)
52 static struct kmem_cache
*flow_cache
;
53 struct kmem_cache
*flow_stats_cache __read_mostly
;
55 static u16
range_n_bytes(const struct sw_flow_key_range
*range
)
57 return range
->end
- range
->start
;
60 void ovs_flow_mask_key(struct sw_flow_key
*dst
, const struct sw_flow_key
*src
,
61 const struct sw_flow_mask
*mask
)
63 const long *m
= (const long *)((const u8
*)&mask
->key
+
65 const long *s
= (const long *)((const u8
*)src
+
67 long *d
= (long *)((u8
*)dst
+ mask
->range
.start
);
70 /* The memory outside of the 'mask->range' are not set since
71 * further operations on 'dst' only uses contents within
74 for (i
= 0; i
< range_n_bytes(&mask
->range
); i
+= sizeof(long))
78 struct sw_flow
*ovs_flow_alloc(void)
81 struct flow_stats
*stats
;
84 flow
= kmem_cache_alloc(flow_cache
, GFP_KERNEL
);
86 return ERR_PTR(-ENOMEM
);
90 flow
->stats_last_writer
= NUMA_NO_NODE
;
92 /* Initialize the default stat node. */
93 stats
= kmem_cache_alloc_node(flow_stats_cache
,
94 GFP_KERNEL
| __GFP_ZERO
, 0);
98 spin_lock_init(&stats
->lock
);
100 RCU_INIT_POINTER(flow
->stats
[0], stats
);
104 RCU_INIT_POINTER(flow
->stats
[node
], NULL
);
108 kmem_cache_free(flow_cache
, flow
);
109 return ERR_PTR(-ENOMEM
);
112 int ovs_flow_tbl_count(struct flow_table
*table
)
117 static struct flex_array
*alloc_buckets(unsigned int n_buckets
)
119 struct flex_array
*buckets
;
122 buckets
= flex_array_alloc(sizeof(struct hlist_head
),
123 n_buckets
, GFP_KERNEL
);
127 err
= flex_array_prealloc(buckets
, 0, n_buckets
, GFP_KERNEL
);
129 flex_array_free(buckets
);
133 for (i
= 0; i
< n_buckets
; i
++)
134 INIT_HLIST_HEAD((struct hlist_head
*)
135 flex_array_get(buckets
, i
));
140 static void flow_free(struct sw_flow
*flow
)
144 kfree((struct sw_flow_actions __force
*)flow
->sf_acts
);
146 if (flow
->stats
[node
])
147 kmem_cache_free(flow_stats_cache
,
148 (struct flow_stats __force
*)flow
->stats
[node
]);
149 kmem_cache_free(flow_cache
, flow
);
152 static void rcu_free_flow_callback(struct rcu_head
*rcu
)
154 struct sw_flow
*flow
= container_of(rcu
, struct sw_flow
, rcu
);
159 static void rcu_free_sw_flow_mask_cb(struct rcu_head
*rcu
)
161 struct sw_flow_mask
*mask
= container_of(rcu
, struct sw_flow_mask
, rcu
);
166 void ovs_flow_free(struct sw_flow
*flow
, bool deferred
)
172 call_rcu(&flow
->rcu
, rcu_free_flow_callback
);
177 static void free_buckets(struct flex_array
*buckets
)
179 flex_array_free(buckets
);
183 static void __table_instance_destroy(struct table_instance
*ti
)
185 free_buckets(ti
->buckets
);
189 static struct table_instance
*table_instance_alloc(int new_size
)
191 struct table_instance
*ti
= kmalloc(sizeof(*ti
), GFP_KERNEL
);
196 ti
->buckets
= alloc_buckets(new_size
);
202 ti
->n_buckets
= new_size
;
204 ti
->keep_flows
= false;
205 get_random_bytes(&ti
->hash_seed
, sizeof(u32
));
210 int ovs_flow_tbl_init(struct flow_table
*table
)
212 struct table_instance
*ti
;
214 ti
= table_instance_alloc(TBL_MIN_BUCKETS
);
219 rcu_assign_pointer(table
->ti
, ti
);
220 INIT_LIST_HEAD(&table
->mask_list
);
221 table
->last_rehash
= jiffies
;
226 static void flow_tbl_destroy_rcu_cb(struct rcu_head
*rcu
)
228 struct table_instance
*ti
= container_of(rcu
, struct table_instance
, rcu
);
230 __table_instance_destroy(ti
);
233 static void table_instance_destroy(struct table_instance
*ti
, bool deferred
)
243 for (i
= 0; i
< ti
->n_buckets
; i
++) {
244 struct sw_flow
*flow
;
245 struct hlist_head
*head
= flex_array_get(ti
->buckets
, i
);
246 struct hlist_node
*n
;
247 int ver
= ti
->node_ver
;
249 hlist_for_each_entry_safe(flow
, n
, head
, hash_node
[ver
]) {
250 hlist_del_rcu(&flow
->hash_node
[ver
]);
251 ovs_flow_free(flow
, deferred
);
257 call_rcu(&ti
->rcu
, flow_tbl_destroy_rcu_cb
);
259 __table_instance_destroy(ti
);
262 void ovs_flow_tbl_destroy(struct flow_table
*table
, bool deferred
)
264 struct table_instance
*ti
= ovsl_dereference(table
->ti
);
266 table_instance_destroy(ti
, deferred
);
269 struct sw_flow
*ovs_flow_tbl_dump_next(struct table_instance
*ti
,
270 u32
*bucket
, u32
*last
)
272 struct sw_flow
*flow
;
273 struct hlist_head
*head
;
278 while (*bucket
< ti
->n_buckets
) {
280 head
= flex_array_get(ti
->buckets
, *bucket
);
281 hlist_for_each_entry_rcu(flow
, head
, hash_node
[ver
]) {
296 static struct hlist_head
*find_bucket(struct table_instance
*ti
, u32 hash
)
298 hash
= jhash_1word(hash
, ti
->hash_seed
);
299 return flex_array_get(ti
->buckets
,
300 (hash
& (ti
->n_buckets
- 1)));
303 static void table_instance_insert(struct table_instance
*ti
, struct sw_flow
*flow
)
305 struct hlist_head
*head
;
307 head
= find_bucket(ti
, flow
->hash
);
308 hlist_add_head_rcu(&flow
->hash_node
[ti
->node_ver
], head
);
311 static void flow_table_copy_flows(struct table_instance
*old
,
312 struct table_instance
*new)
317 old_ver
= old
->node_ver
;
318 new->node_ver
= !old_ver
;
320 /* Insert in new table. */
321 for (i
= 0; i
< old
->n_buckets
; i
++) {
322 struct sw_flow
*flow
;
323 struct hlist_head
*head
;
325 head
= flex_array_get(old
->buckets
, i
);
327 hlist_for_each_entry(flow
, head
, hash_node
[old_ver
])
328 table_instance_insert(new, flow
);
331 old
->keep_flows
= true;
334 static struct table_instance
*table_instance_rehash(struct table_instance
*ti
,
337 struct table_instance
*new_ti
;
339 new_ti
= table_instance_alloc(n_buckets
);
343 flow_table_copy_flows(ti
, new_ti
);
348 int ovs_flow_tbl_flush(struct flow_table
*flow_table
)
350 struct table_instance
*old_ti
;
351 struct table_instance
*new_ti
;
353 old_ti
= ovsl_dereference(flow_table
->ti
);
354 new_ti
= table_instance_alloc(TBL_MIN_BUCKETS
);
358 rcu_assign_pointer(flow_table
->ti
, new_ti
);
359 flow_table
->last_rehash
= jiffies
;
360 flow_table
->count
= 0;
362 table_instance_destroy(old_ti
, true);
366 static u32
flow_hash(const struct sw_flow_key
*key
, int key_start
,
369 const u32
*hash_key
= (const u32
*)((const u8
*)key
+ key_start
);
370 int hash_u32s
= (key_end
- key_start
) >> 2;
372 /* Make sure number of hash bytes are multiple of u32. */
373 BUILD_BUG_ON(sizeof(long) % sizeof(u32
));
375 return arch_fast_hash2(hash_key
, hash_u32s
, 0);
378 static int flow_key_start(const struct sw_flow_key
*key
)
380 if (key
->tun_key
.ipv4_dst
)
383 return rounddown(offsetof(struct sw_flow_key
, phy
),
387 static bool cmp_key(const struct sw_flow_key
*key1
,
388 const struct sw_flow_key
*key2
,
389 int key_start
, int key_end
)
391 const long *cp1
= (const long *)((const u8
*)key1
+ key_start
);
392 const long *cp2
= (const long *)((const u8
*)key2
+ key_start
);
396 for (i
= key_start
; i
< key_end
; i
+= sizeof(long))
397 diffs
|= *cp1
++ ^ *cp2
++;
402 static bool flow_cmp_masked_key(const struct sw_flow
*flow
,
403 const struct sw_flow_key
*key
,
404 int key_start
, int key_end
)
406 return cmp_key(&flow
->key
, key
, key_start
, key_end
);
409 bool ovs_flow_cmp_unmasked_key(const struct sw_flow
*flow
,
410 struct sw_flow_match
*match
)
412 struct sw_flow_key
*key
= match
->key
;
413 int key_start
= flow_key_start(key
);
414 int key_end
= match
->range
.end
;
416 return cmp_key(&flow
->unmasked_key
, key
, key_start
, key_end
);
419 static struct sw_flow
*masked_flow_lookup(struct table_instance
*ti
,
420 const struct sw_flow_key
*unmasked
,
421 struct sw_flow_mask
*mask
)
423 struct sw_flow
*flow
;
424 struct hlist_head
*head
;
425 int key_start
= mask
->range
.start
;
426 int key_end
= mask
->range
.end
;
428 struct sw_flow_key masked_key
;
430 ovs_flow_mask_key(&masked_key
, unmasked
, mask
);
431 hash
= flow_hash(&masked_key
, key_start
, key_end
);
432 head
= find_bucket(ti
, hash
);
433 hlist_for_each_entry_rcu(flow
, head
, hash_node
[ti
->node_ver
]) {
434 if (flow
->mask
== mask
&& flow
->hash
== hash
&&
435 flow_cmp_masked_key(flow
, &masked_key
,
442 struct sw_flow
*ovs_flow_tbl_lookup_stats(struct flow_table
*tbl
,
443 const struct sw_flow_key
*key
,
446 struct table_instance
*ti
= rcu_dereference_ovsl(tbl
->ti
);
447 struct sw_flow_mask
*mask
;
448 struct sw_flow
*flow
;
451 list_for_each_entry_rcu(mask
, &tbl
->mask_list
, list
) {
453 flow
= masked_flow_lookup(ti
, key
, mask
);
454 if (flow
) /* Found */
460 struct sw_flow
*ovs_flow_tbl_lookup(struct flow_table
*tbl
,
461 const struct sw_flow_key
*key
)
463 u32 __always_unused n_mask_hit
;
465 return ovs_flow_tbl_lookup_stats(tbl
, key
, &n_mask_hit
);
468 int ovs_flow_tbl_num_masks(const struct flow_table
*table
)
470 struct sw_flow_mask
*mask
;
473 list_for_each_entry(mask
, &table
->mask_list
, list
)
479 static struct table_instance
*table_instance_expand(struct table_instance
*ti
)
481 return table_instance_rehash(ti
, ti
->n_buckets
* 2);
484 /* Remove 'mask' from the mask list, if it is not needed any more. */
485 static void flow_mask_remove(struct flow_table
*tbl
, struct sw_flow_mask
*mask
)
488 /* ovs-lock is required to protect mask-refcount and
492 BUG_ON(!mask
->ref_count
);
495 if (!mask
->ref_count
) {
496 list_del_rcu(&mask
->list
);
497 call_rcu(&mask
->rcu
, rcu_free_sw_flow_mask_cb
);
502 /* Must be called with OVS mutex held. */
503 void ovs_flow_tbl_remove(struct flow_table
*table
, struct sw_flow
*flow
)
505 struct table_instance
*ti
= ovsl_dereference(table
->ti
);
507 BUG_ON(table
->count
== 0);
508 hlist_del_rcu(&flow
->hash_node
[ti
->node_ver
]);
511 /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
512 * accessible as long as the RCU read lock is held. */
513 flow_mask_remove(table
, flow
->mask
);
516 static struct sw_flow_mask
*mask_alloc(void)
518 struct sw_flow_mask
*mask
;
520 mask
= kmalloc(sizeof(*mask
), GFP_KERNEL
);
527 static bool mask_equal(const struct sw_flow_mask
*a
,
528 const struct sw_flow_mask
*b
)
530 const u8
*a_
= (const u8
*)&a
->key
+ a
->range
.start
;
531 const u8
*b_
= (const u8
*)&b
->key
+ b
->range
.start
;
533 return (a
->range
.end
== b
->range
.end
)
534 && (a
->range
.start
== b
->range
.start
)
535 && (memcmp(a_
, b_
, range_n_bytes(&a
->range
)) == 0);
538 static struct sw_flow_mask
*flow_mask_find(const struct flow_table
*tbl
,
539 const struct sw_flow_mask
*mask
)
541 struct list_head
*ml
;
543 list_for_each(ml
, &tbl
->mask_list
) {
544 struct sw_flow_mask
*m
;
545 m
= container_of(ml
, struct sw_flow_mask
, list
);
546 if (mask_equal(mask
, m
))
553 /* Add 'mask' into the mask list, if it is not already there. */
554 static int flow_mask_insert(struct flow_table
*tbl
, struct sw_flow
*flow
,
555 struct sw_flow_mask
*new)
557 struct sw_flow_mask
*mask
;
558 mask
= flow_mask_find(tbl
, new);
560 /* Allocate a new mask if none exsits. */
564 mask
->key
= new->key
;
565 mask
->range
= new->range
;
566 list_add_rcu(&mask
->list
, &tbl
->mask_list
);
568 BUG_ON(!mask
->ref_count
);
576 /* Must be called with OVS mutex held. */
577 int ovs_flow_tbl_insert(struct flow_table
*table
, struct sw_flow
*flow
,
578 struct sw_flow_mask
*mask
)
580 struct table_instance
*new_ti
= NULL
;
581 struct table_instance
*ti
;
584 err
= flow_mask_insert(table
, flow
, mask
);
588 flow
->hash
= flow_hash(&flow
->key
, flow
->mask
->range
.start
,
589 flow
->mask
->range
.end
);
590 ti
= ovsl_dereference(table
->ti
);
591 table_instance_insert(ti
, flow
);
594 /* Expand table, if necessary, to make room. */
595 if (table
->count
> ti
->n_buckets
)
596 new_ti
= table_instance_expand(ti
);
597 else if (time_after(jiffies
, table
->last_rehash
+ REHASH_INTERVAL
))
598 new_ti
= table_instance_rehash(ti
, ti
->n_buckets
);
601 rcu_assign_pointer(table
->ti
, new_ti
);
602 table_instance_destroy(ti
, true);
603 table
->last_rehash
= jiffies
;
608 /* Initializes the flow module.
609 * Returns zero if successful or a negative error code. */
610 int ovs_flow_init(void)
612 BUILD_BUG_ON(__alignof__(struct sw_flow_key
) % __alignof__(long));
613 BUILD_BUG_ON(sizeof(struct sw_flow_key
) % sizeof(long));
615 flow_cache
= kmem_cache_create("sw_flow", sizeof(struct sw_flow
)
616 + (num_possible_nodes()
617 * sizeof(struct flow_stats
*)),
619 if (flow_cache
== NULL
)
623 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats
),
624 0, SLAB_HWCACHE_ALIGN
, NULL
);
625 if (flow_stats_cache
== NULL
) {
626 kmem_cache_destroy(flow_cache
);
634 /* Uninitializes the flow module. */
635 void ovs_flow_exit(void)
637 kmem_cache_destroy(flow_stats_cache
);
638 kmem_cache_destroy(flow_cache
);