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/jhash.h>
29 #include <linux/jiffies.h>
30 #include <linux/llc.h>
31 #include <linux/module.h>
33 #include <linux/rcupdate.h>
34 #include <linux/cpumask.h>
35 #include <linux/if_arp.h>
37 #include <linux/ipv6.h>
38 #include <linux/sctp.h>
39 #include <linux/tcp.h>
40 #include <linux/udp.h>
41 #include <linux/icmp.h>
42 #include <linux/icmpv6.h>
43 #include <linux/rculist.h>
46 #include <net/ndisc.h>
48 #include "flow_netlink.h"
50 #define TBL_MIN_BUCKETS 1024
51 #define MASK_ARRAY_SIZE_MIN 16
52 #define REHASH_INTERVAL (10 * 60 * HZ)
54 #define MC_HASH_SHIFT 8
55 #define MC_HASH_ENTRIES (1u << MC_HASH_SHIFT)
56 #define MC_HASH_SEGS ((sizeof(uint32_t) * 8) / MC_HASH_SHIFT)
58 static struct kmem_cache
*flow_cache
;
59 struct kmem_cache
*flow_stats_cache __read_mostly
;
61 static u16
range_n_bytes(const struct sw_flow_key_range
*range
)
63 return range
->end
- range
->start
;
66 void ovs_flow_mask_key(struct sw_flow_key
*dst
, const struct sw_flow_key
*src
,
67 bool full
, const struct sw_flow_mask
*mask
)
69 int start
= full
? 0 : mask
->range
.start
;
70 int len
= full
? sizeof *dst
: range_n_bytes(&mask
->range
);
71 const long *m
= (const long *)((const u8
*)&mask
->key
+ start
);
72 const long *s
= (const long *)((const u8
*)src
+ start
);
73 long *d
= (long *)((u8
*)dst
+ start
);
76 /* If 'full' is true then all of 'dst' is fully initialized. Otherwise,
77 * if 'full' is false the memory outside of the 'mask->range' is left
78 * uninitialized. This can be used as an optimization when further
79 * operations on 'dst' only use contents within 'mask->range'.
81 for (i
= 0; i
< len
; i
+= sizeof(long))
85 struct sw_flow
*ovs_flow_alloc(void)
88 struct flow_stats
*stats
;
90 flow
= kmem_cache_zalloc(flow_cache
, GFP_KERNEL
);
92 return ERR_PTR(-ENOMEM
);
94 flow
->stats_last_writer
= -1;
96 /* Initialize the default stat node. */
97 stats
= kmem_cache_alloc_node(flow_stats_cache
,
98 GFP_KERNEL
| __GFP_ZERO
,
99 node_online(0) ? 0 : NUMA_NO_NODE
);
103 spin_lock_init(&stats
->lock
);
105 RCU_INIT_POINTER(flow
->stats
[0], stats
);
107 cpumask_set_cpu(0, &flow
->cpu_used_mask
);
111 kmem_cache_free(flow_cache
, flow
);
112 return ERR_PTR(-ENOMEM
);
115 int ovs_flow_tbl_count(const struct flow_table
*table
)
120 static struct flex_array
*alloc_buckets(unsigned int n_buckets
)
122 struct flex_array
*buckets
;
125 buckets
= flex_array_alloc(sizeof(struct hlist_head
),
126 n_buckets
, GFP_KERNEL
);
130 err
= flex_array_prealloc(buckets
, 0, n_buckets
, GFP_KERNEL
);
132 flex_array_free(buckets
);
136 for (i
= 0; i
< n_buckets
; i
++)
137 INIT_HLIST_HEAD((struct hlist_head
*)
138 flex_array_get(buckets
, i
));
143 static void flow_free(struct sw_flow
*flow
)
147 if (ovs_identifier_is_key(&flow
->id
))
148 kfree(flow
->id
.unmasked_key
);
150 ovs_nla_free_flow_actions((struct sw_flow_actions __force
*)flow
->sf_acts
);
151 /* We open code this to make sure cpu 0 is always considered */
152 for (cpu
= 0; cpu
< nr_cpu_ids
; cpu
= cpumask_next(cpu
, &flow
->cpu_used_mask
))
153 if (flow
->stats
[cpu
])
154 kmem_cache_free(flow_stats_cache
,
155 rcu_dereference_raw(flow
->stats
[cpu
]));
156 kmem_cache_free(flow_cache
, flow
);
159 static void rcu_free_flow_callback(struct rcu_head
*rcu
)
161 struct sw_flow
*flow
= container_of(rcu
, struct sw_flow
, 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 static void mask_array_rcu_cb(struct rcu_head
*rcu
)
212 struct mask_array
*ma
= container_of(rcu
, struct mask_array
, rcu
);
217 static struct mask_array
*tbl_mask_array_alloc(int size
)
219 struct mask_array
*new;
221 size
= max(MASK_ARRAY_SIZE_MIN
, size
);
222 new = kzalloc(sizeof(struct mask_array
) +
223 sizeof(struct sw_flow_mask
*) * size
, GFP_KERNEL
);
233 static int tbl_mask_array_realloc(struct flow_table
*tbl
, int size
)
235 struct mask_array
*old
;
236 struct mask_array
*new;
238 new = tbl_mask_array_alloc(size
);
242 old
= ovsl_dereference(tbl
->mask_array
);
246 for (i
= 0; i
< old
->max
; i
++) {
247 if (ovsl_dereference(old
->masks
[i
]))
248 new->masks
[count
++] = old
->masks
[i
];
253 rcu_assign_pointer(tbl
->mask_array
, new);
256 call_rcu(&old
->rcu
, mask_array_rcu_cb
);
261 int ovs_flow_tbl_init(struct flow_table
*table
)
263 struct table_instance
*ti
, *ufid_ti
;
264 struct mask_array
*ma
;
266 table
->mask_cache
= __alloc_percpu(sizeof(struct mask_cache_entry
) *
267 MC_HASH_ENTRIES
, __alignof__(struct mask_cache_entry
));
268 if (!table
->mask_cache
)
271 ma
= tbl_mask_array_alloc(MASK_ARRAY_SIZE_MIN
);
273 goto free_mask_cache
;
275 ti
= table_instance_alloc(TBL_MIN_BUCKETS
);
277 goto free_mask_array
;
279 ufid_ti
= table_instance_alloc(TBL_MIN_BUCKETS
);
283 rcu_assign_pointer(table
->ti
, ti
);
284 rcu_assign_pointer(table
->ufid_ti
, ufid_ti
);
285 rcu_assign_pointer(table
->mask_array
, ma
);
286 table
->last_rehash
= jiffies
;
288 table
->ufid_count
= 0;
292 __table_instance_destroy(ti
);
296 free_percpu(table
->mask_cache
);
300 static void flow_tbl_destroy_rcu_cb(struct rcu_head
*rcu
)
302 struct table_instance
*ti
= container_of(rcu
, struct table_instance
, rcu
);
304 __table_instance_destroy(ti
);
307 static void table_instance_destroy(struct table_instance
*ti
,
308 struct table_instance
*ufid_ti
,
320 for (i
= 0; i
< ti
->n_buckets
; i
++) {
321 struct sw_flow
*flow
;
322 struct hlist_head
*head
= flex_array_get(ti
->buckets
, i
);
323 struct hlist_node
*n
;
324 int ver
= ti
->node_ver
;
325 int ufid_ver
= ufid_ti
->node_ver
;
327 hlist_for_each_entry_safe(flow
, n
, head
, flow_table
.node
[ver
]) {
328 hlist_del_rcu(&flow
->flow_table
.node
[ver
]);
329 if (ovs_identifier_is_ufid(&flow
->id
))
330 hlist_del_rcu(&flow
->ufid_table
.node
[ufid_ver
]);
331 ovs_flow_free(flow
, deferred
);
337 call_rcu(&ti
->rcu
, flow_tbl_destroy_rcu_cb
);
338 call_rcu(&ufid_ti
->rcu
, flow_tbl_destroy_rcu_cb
);
340 __table_instance_destroy(ti
);
341 __table_instance_destroy(ufid_ti
);
345 /* No need for locking this function is called from RCU callback or
348 void ovs_flow_tbl_destroy(struct flow_table
*table
)
350 struct table_instance
*ti
= rcu_dereference_raw(table
->ti
);
351 struct table_instance
*ufid_ti
= rcu_dereference_raw(table
->ufid_ti
);
353 free_percpu(table
->mask_cache
);
354 kfree(rcu_dereference_raw(table
->mask_array
));
355 table_instance_destroy(ti
, ufid_ti
, false);
358 struct sw_flow
*ovs_flow_tbl_dump_next(struct table_instance
*ti
,
359 u32
*bucket
, u32
*last
)
361 struct sw_flow
*flow
;
362 struct hlist_head
*head
;
367 while (*bucket
< ti
->n_buckets
) {
369 head
= flex_array_get(ti
->buckets
, *bucket
);
370 hlist_for_each_entry_rcu(flow
, head
, flow_table
.node
[ver
]) {
385 static struct hlist_head
*find_bucket(struct table_instance
*ti
, u32 hash
)
387 hash
= jhash_1word(hash
, ti
->hash_seed
);
388 return flex_array_get(ti
->buckets
,
389 (hash
& (ti
->n_buckets
- 1)));
392 static void table_instance_insert(struct table_instance
*ti
,
393 struct sw_flow
*flow
)
395 struct hlist_head
*head
;
397 head
= find_bucket(ti
, flow
->flow_table
.hash
);
398 hlist_add_head_rcu(&flow
->flow_table
.node
[ti
->node_ver
], head
);
401 static void ufid_table_instance_insert(struct table_instance
*ti
,
402 struct sw_flow
*flow
)
404 struct hlist_head
*head
;
406 head
= find_bucket(ti
, flow
->ufid_table
.hash
);
407 hlist_add_head_rcu(&flow
->ufid_table
.node
[ti
->node_ver
], head
);
410 static void flow_table_copy_flows(struct table_instance
*old
,
411 struct table_instance
*new, bool ufid
)
416 old_ver
= old
->node_ver
;
417 new->node_ver
= !old_ver
;
419 /* Insert in new table. */
420 for (i
= 0; i
< old
->n_buckets
; i
++) {
421 struct sw_flow
*flow
;
422 struct hlist_head
*head
;
424 head
= flex_array_get(old
->buckets
, i
);
427 hlist_for_each_entry(flow
, head
,
428 ufid_table
.node
[old_ver
])
429 ufid_table_instance_insert(new, flow
);
431 hlist_for_each_entry(flow
, head
,
432 flow_table
.node
[old_ver
])
433 table_instance_insert(new, flow
);
436 old
->keep_flows
= true;
439 static struct table_instance
*table_instance_rehash(struct table_instance
*ti
,
440 int n_buckets
, bool ufid
)
442 struct table_instance
*new_ti
;
444 new_ti
= table_instance_alloc(n_buckets
);
448 flow_table_copy_flows(ti
, new_ti
, ufid
);
453 int ovs_flow_tbl_flush(struct flow_table
*flow_table
)
455 struct table_instance
*old_ti
, *new_ti
;
456 struct table_instance
*old_ufid_ti
, *new_ufid_ti
;
458 new_ti
= table_instance_alloc(TBL_MIN_BUCKETS
);
461 new_ufid_ti
= table_instance_alloc(TBL_MIN_BUCKETS
);
465 old_ti
= ovsl_dereference(flow_table
->ti
);
466 old_ufid_ti
= ovsl_dereference(flow_table
->ufid_ti
);
468 rcu_assign_pointer(flow_table
->ti
, new_ti
);
469 rcu_assign_pointer(flow_table
->ufid_ti
, new_ufid_ti
);
470 flow_table
->last_rehash
= jiffies
;
471 flow_table
->count
= 0;
472 flow_table
->ufid_count
= 0;
474 table_instance_destroy(old_ti
, old_ufid_ti
, true);
478 __table_instance_destroy(new_ti
);
482 static u32
flow_hash(const struct sw_flow_key
*key
,
483 const struct sw_flow_key_range
*range
)
485 int key_start
= range
->start
;
486 int key_end
= range
->end
;
487 const u32
*hash_key
= (const u32
*)((const u8
*)key
+ key_start
);
488 int hash_u32s
= (key_end
- key_start
) >> 2;
490 /* Make sure number of hash bytes are multiple of u32. */
491 BUILD_BUG_ON(sizeof(long) % sizeof(u32
));
493 return jhash2(hash_key
, hash_u32s
, 0);
496 static int flow_key_start(const struct sw_flow_key
*key
)
501 return rounddown(offsetof(struct sw_flow_key
, phy
),
505 static bool cmp_key(const struct sw_flow_key
*key1
,
506 const struct sw_flow_key
*key2
,
507 int key_start
, int key_end
)
509 const long *cp1
= (const long *)((const u8
*)key1
+ key_start
);
510 const long *cp2
= (const long *)((const u8
*)key2
+ key_start
);
514 for (i
= key_start
; i
< key_end
; i
+= sizeof(long))
515 diffs
|= *cp1
++ ^ *cp2
++;
520 static bool flow_cmp_masked_key(const struct sw_flow
*flow
,
521 const struct sw_flow_key
*key
,
522 const struct sw_flow_key_range
*range
)
524 return cmp_key(&flow
->key
, key
, range
->start
, range
->end
);
527 static bool ovs_flow_cmp_unmasked_key(const struct sw_flow
*flow
,
528 const struct sw_flow_match
*match
)
530 struct sw_flow_key
*key
= match
->key
;
531 int key_start
= flow_key_start(key
);
532 int key_end
= match
->range
.end
;
534 BUG_ON(ovs_identifier_is_ufid(&flow
->id
));
535 return cmp_key(flow
->id
.unmasked_key
, key
, key_start
, key_end
);
538 static struct sw_flow
*masked_flow_lookup(struct table_instance
*ti
,
539 const struct sw_flow_key
*unmasked
,
540 const struct sw_flow_mask
*mask
,
543 struct sw_flow
*flow
;
544 struct hlist_head
*head
;
546 struct sw_flow_key masked_key
;
548 ovs_flow_mask_key(&masked_key
, unmasked
, false, mask
);
549 hash
= flow_hash(&masked_key
, &mask
->range
);
550 head
= find_bucket(ti
, hash
);
552 hlist_for_each_entry_rcu(flow
, head
, flow_table
.node
[ti
->node_ver
]) {
553 if (flow
->mask
== mask
&& flow
->flow_table
.hash
== hash
&&
554 flow_cmp_masked_key(flow
, &masked_key
, &mask
->range
))
560 /* Flow lookup does full lookup on flow table. It starts with
561 * mask from index passed in *index.
563 static struct sw_flow
*flow_lookup(struct flow_table
*tbl
,
564 struct table_instance
*ti
,
565 const struct mask_array
*ma
,
566 const struct sw_flow_key
*key
,
570 struct sw_flow_mask
*mask
;
571 struct sw_flow
*flow
;
574 if (*index
< ma
->max
) {
575 mask
= rcu_dereference_ovsl(ma
->masks
[*index
]);
577 flow
= masked_flow_lookup(ti
, key
, mask
, n_mask_hit
);
583 for (i
= 0; i
< ma
->max
; i
++) {
588 mask
= rcu_dereference_ovsl(ma
->masks
[i
]);
592 flow
= masked_flow_lookup(ti
, key
, mask
, n_mask_hit
);
593 if (flow
) { /* Found */
603 * mask_cache maps flow to probable mask. This cache is not tightly
604 * coupled cache, It means updates to mask list can result in inconsistent
605 * cache entry in mask cache.
606 * This is per cpu cache and is divided in MC_HASH_SEGS segments.
607 * In case of a hash collision the entry is hashed in next segment.
609 struct sw_flow
*ovs_flow_tbl_lookup_stats(struct flow_table
*tbl
,
610 const struct sw_flow_key
*key
,
614 struct mask_array
*ma
= rcu_dereference(tbl
->mask_array
);
615 struct table_instance
*ti
= rcu_dereference(tbl
->ti
);
616 struct mask_cache_entry
*entries
, *ce
;
617 struct sw_flow
*flow
;
622 if (unlikely(!skb_hash
)) {
625 return flow_lookup(tbl
, ti
, ma
, key
, n_mask_hit
, &mask_index
);
628 /* Pre and post recirulation flows usually have the same skb_hash
629 * value. To avoid hash collisions, rehash the 'skb_hash' with
632 skb_hash
= jhash_1word(skb_hash
, key
->recirc_id
);
636 entries
= this_cpu_ptr(tbl
->mask_cache
);
638 /* Find the cache entry 'ce' to operate on. */
639 for (seg
= 0; seg
< MC_HASH_SEGS
; seg
++) {
640 int index
= hash
& (MC_HASH_ENTRIES
- 1);
641 struct mask_cache_entry
*e
;
644 if (e
->skb_hash
== skb_hash
) {
645 flow
= flow_lookup(tbl
, ti
, ma
, key
, n_mask_hit
,
652 if (!ce
|| e
->skb_hash
< ce
->skb_hash
)
653 ce
= e
; /* A better replacement cache candidate. */
655 hash
>>= MC_HASH_SHIFT
;
658 /* Cache miss, do full lookup. */
659 flow
= flow_lookup(tbl
, ti
, ma
, key
, n_mask_hit
, &ce
->mask_index
);
661 ce
->skb_hash
= skb_hash
;
666 struct sw_flow
*ovs_flow_tbl_lookup(struct flow_table
*tbl
,
667 const struct sw_flow_key
*key
)
669 struct table_instance
*ti
= rcu_dereference_ovsl(tbl
->ti
);
670 struct mask_array
*ma
= rcu_dereference_ovsl(tbl
->mask_array
);
671 u32 __always_unused n_mask_hit
;
674 return flow_lookup(tbl
, ti
, ma
, key
, &n_mask_hit
, &index
);
677 struct sw_flow
*ovs_flow_tbl_lookup_exact(struct flow_table
*tbl
,
678 const struct sw_flow_match
*match
)
680 struct mask_array
*ma
= ovsl_dereference(tbl
->mask_array
);
683 /* Always called under ovs-mutex. */
684 for (i
= 0; i
< ma
->max
; i
++) {
685 struct table_instance
*ti
= ovsl_dereference(tbl
->ti
);
686 u32 __always_unused n_mask_hit
;
687 struct sw_flow_mask
*mask
;
688 struct sw_flow
*flow
;
690 mask
= ovsl_dereference(ma
->masks
[i
]);
693 flow
= masked_flow_lookup(ti
, match
->key
, mask
, &n_mask_hit
);
694 if (flow
&& ovs_identifier_is_key(&flow
->id
) &&
695 ovs_flow_cmp_unmasked_key(flow
, match
))
701 static u32
ufid_hash(const struct sw_flow_id
*sfid
)
703 return jhash(sfid
->ufid
, sfid
->ufid_len
, 0);
706 static bool ovs_flow_cmp_ufid(const struct sw_flow
*flow
,
707 const struct sw_flow_id
*sfid
)
709 if (flow
->id
.ufid_len
!= sfid
->ufid_len
)
712 return !memcmp(flow
->id
.ufid
, sfid
->ufid
, sfid
->ufid_len
);
715 bool ovs_flow_cmp(const struct sw_flow
*flow
, const struct sw_flow_match
*match
)
717 if (ovs_identifier_is_ufid(&flow
->id
))
718 return flow_cmp_masked_key(flow
, match
->key
, &match
->range
);
720 return ovs_flow_cmp_unmasked_key(flow
, match
);
723 struct sw_flow
*ovs_flow_tbl_lookup_ufid(struct flow_table
*tbl
,
724 const struct sw_flow_id
*ufid
)
726 struct table_instance
*ti
= rcu_dereference_ovsl(tbl
->ufid_ti
);
727 struct sw_flow
*flow
;
728 struct hlist_head
*head
;
731 hash
= ufid_hash(ufid
);
732 head
= find_bucket(ti
, hash
);
733 hlist_for_each_entry_rcu(flow
, head
, ufid_table
.node
[ti
->node_ver
]) {
734 if (flow
->ufid_table
.hash
== hash
&&
735 ovs_flow_cmp_ufid(flow
, ufid
))
741 int ovs_flow_tbl_num_masks(const struct flow_table
*table
)
743 struct mask_array
*ma
;
745 ma
= rcu_dereference_ovsl(table
->mask_array
);
749 static struct table_instance
*table_instance_expand(struct table_instance
*ti
,
752 return table_instance_rehash(ti
, ti
->n_buckets
* 2, ufid
);
755 static void tbl_mask_array_delete_mask(struct mask_array
*ma
,
756 struct sw_flow_mask
*mask
)
760 /* Remove the deleted mask pointers from the array */
761 for (i
= 0; i
< ma
->max
; i
++) {
762 if (mask
== ovsl_dereference(ma
->masks
[i
])) {
763 RCU_INIT_POINTER(ma
->masks
[i
], NULL
);
765 kfree_rcu(mask
, rcu
);
772 /* Remove 'mask' from the mask list, if it is not needed any more. */
773 static void flow_mask_remove(struct flow_table
*tbl
, struct sw_flow_mask
*mask
)
776 /* ovs-lock is required to protect mask-refcount and
780 BUG_ON(!mask
->ref_count
);
783 if (!mask
->ref_count
) {
784 struct mask_array
*ma
;
786 ma
= ovsl_dereference(tbl
->mask_array
);
787 tbl_mask_array_delete_mask(ma
, mask
);
789 /* Shrink the mask array if necessary. */
790 if (ma
->max
>= (MASK_ARRAY_SIZE_MIN
* 2) &&
791 ma
->count
<= (ma
->max
/ 3))
792 tbl_mask_array_realloc(tbl
, ma
->max
/ 2);
798 /* Must be called with OVS mutex held. */
799 void ovs_flow_tbl_remove(struct flow_table
*table
, struct sw_flow
*flow
)
801 struct table_instance
*ti
= ovsl_dereference(table
->ti
);
802 struct table_instance
*ufid_ti
= ovsl_dereference(table
->ufid_ti
);
804 BUG_ON(table
->count
== 0);
805 hlist_del_rcu(&flow
->flow_table
.node
[ti
->node_ver
]);
807 if (ovs_identifier_is_ufid(&flow
->id
)) {
808 hlist_del_rcu(&flow
->ufid_table
.node
[ufid_ti
->node_ver
]);
812 /* RCU delete the mask. 'flow->mask' is not NULLed, as it should be
813 * accessible as long as the RCU read lock is held.
815 flow_mask_remove(table
, flow
->mask
);
818 static struct sw_flow_mask
*mask_alloc(void)
820 struct sw_flow_mask
*mask
;
822 mask
= kmalloc(sizeof(*mask
), GFP_KERNEL
);
829 static bool mask_equal(const struct sw_flow_mask
*a
,
830 const struct sw_flow_mask
*b
)
832 const u8
*a_
= (const u8
*)&a
->key
+ a
->range
.start
;
833 const u8
*b_
= (const u8
*)&b
->key
+ b
->range
.start
;
835 return (a
->range
.end
== b
->range
.end
)
836 && (a
->range
.start
== b
->range
.start
)
837 && (memcmp(a_
, b_
, range_n_bytes(&a
->range
)) == 0);
840 static struct sw_flow_mask
*flow_mask_find(const struct flow_table
*tbl
,
841 const struct sw_flow_mask
*mask
)
843 struct mask_array
*ma
;
846 ma
= ovsl_dereference(tbl
->mask_array
);
847 for (i
= 0; i
< ma
->max
; i
++) {
848 struct sw_flow_mask
*t
;
850 t
= ovsl_dereference(ma
->masks
[i
]);
851 if (t
&& mask_equal(mask
, t
))
858 /* Add 'mask' into the mask list, if it is not already there. */
859 static int flow_mask_insert(struct flow_table
*tbl
, struct sw_flow
*flow
,
860 const struct sw_flow_mask
*new)
862 struct sw_flow_mask
*mask
;
864 mask
= flow_mask_find(tbl
, new);
866 struct mask_array
*ma
;
869 /* Allocate a new mask if none exsits. */
874 mask
->key
= new->key
;
875 mask
->range
= new->range
;
877 /* Add mask to mask-list. */
878 ma
= ovsl_dereference(tbl
->mask_array
);
879 if (ma
->count
>= ma
->max
) {
882 err
= tbl_mask_array_realloc(tbl
, ma
->max
+
883 MASK_ARRAY_SIZE_MIN
);
888 ma
= ovsl_dereference(tbl
->mask_array
);
891 for (i
= 0; i
< ma
->max
; i
++) {
892 struct sw_flow_mask
*t
;
894 t
= ovsl_dereference(ma
->masks
[i
]);
896 rcu_assign_pointer(ma
->masks
[i
], mask
);
903 BUG_ON(!mask
->ref_count
);
911 /* Must be called with OVS mutex held. */
912 static void flow_key_insert(struct flow_table
*table
, struct sw_flow
*flow
)
914 struct table_instance
*new_ti
= NULL
;
915 struct table_instance
*ti
;
917 flow
->flow_table
.hash
= flow_hash(&flow
->key
, &flow
->mask
->range
);
918 ti
= ovsl_dereference(table
->ti
);
919 table_instance_insert(ti
, flow
);
922 /* Expand table, if necessary, to make room. */
923 if (table
->count
> ti
->n_buckets
)
924 new_ti
= table_instance_expand(ti
, false);
925 else if (time_after(jiffies
, table
->last_rehash
+ REHASH_INTERVAL
))
926 new_ti
= table_instance_rehash(ti
, ti
->n_buckets
, false);
929 rcu_assign_pointer(table
->ti
, new_ti
);
930 call_rcu(&ti
->rcu
, flow_tbl_destroy_rcu_cb
);
931 table
->last_rehash
= jiffies
;
935 /* Must be called with OVS mutex held. */
936 static void flow_ufid_insert(struct flow_table
*table
, struct sw_flow
*flow
)
938 struct table_instance
*ti
;
940 flow
->ufid_table
.hash
= ufid_hash(&flow
->id
);
941 ti
= ovsl_dereference(table
->ufid_ti
);
942 ufid_table_instance_insert(ti
, flow
);
945 /* Expand table, if necessary, to make room. */
946 if (table
->ufid_count
> ti
->n_buckets
) {
947 struct table_instance
*new_ti
;
949 new_ti
= table_instance_expand(ti
, true);
951 rcu_assign_pointer(table
->ufid_ti
, new_ti
);
952 call_rcu(&ti
->rcu
, flow_tbl_destroy_rcu_cb
);
957 /* Must be called with OVS mutex held. */
958 int ovs_flow_tbl_insert(struct flow_table
*table
, struct sw_flow
*flow
,
959 const struct sw_flow_mask
*mask
)
963 err
= flow_mask_insert(table
, flow
, mask
);
966 flow_key_insert(table
, flow
);
967 if (ovs_identifier_is_ufid(&flow
->id
))
968 flow_ufid_insert(table
, flow
);
973 /* Initializes the flow module.
974 * Returns zero if successful or a negative error code.
976 int ovs_flow_init(void)
978 BUILD_BUG_ON(__alignof__(struct sw_flow_key
) % __alignof__(long));
979 BUILD_BUG_ON(sizeof(struct sw_flow_key
) % sizeof(long));
981 flow_cache
= kmem_cache_create("sw_flow", sizeof(struct sw_flow
)
983 * sizeof(struct flow_stats
*)),
985 if (flow_cache
== NULL
)
989 = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats
),
990 0, SLAB_HWCACHE_ALIGN
, NULL
);
991 if (flow_stats_cache
== NULL
) {
992 kmem_cache_destroy(flow_cache
);
1000 /* Uninitializes the flow module. */
1001 void ovs_flow_exit(void)
1003 kmem_cache_destroy(flow_stats_cache
);
1004 kmem_cache_destroy(flow_cache
);