blob: 53eead2da74324b47ef8c772663f9bcb4fd1d0e1 [file] [log] [blame]
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -07001#ifndef __NET_FRAG_H__
2#define __NET_FRAG_H__
3
Jesper Dangaard Brouer6d7b8572013-01-28 23:45:33 +00004#include <linux/percpu_counter.h>
5
Pavel Emelyanovac18e752008-01-22 06:02:14 -08006struct netns_frags {
Jesper Dangaard Brouer6d7b8572013-01-28 23:45:33 +00007 /* The percpu_counter "mem" need to be cacheline aligned.
8 * mem.count must not share cacheline with other writers
Jesper Dangaard Brouercd39a782013-01-28 23:44:14 +00009 */
Jesper Dangaard Brouer6d7b8572013-01-28 23:45:33 +000010 struct percpu_counter mem ____cacheline_aligned_in_smp;
11
Pavel Emelyanovb2fd5322008-01-22 06:09:37 -080012 /* sysctls */
13 int timeout;
Pavel Emelyanove31e0bdc72008-01-22 06:10:13 -080014 int high_thresh;
15 int low_thresh;
Pavel Emelyanovac18e752008-01-22 06:02:14 -080016};
17
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020018/**
19 * fragment queue flags
20 *
21 * @INET_FRAG_FIRST_IN: first fragment has arrived
22 * @INET_FRAG_LAST_IN: final fragment has arrived
23 * @INET_FRAG_COMPLETE: frag queue has been processed and is due for destruction
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020024 */
25enum {
26 INET_FRAG_FIRST_IN = BIT(0),
27 INET_FRAG_LAST_IN = BIT(1),
28 INET_FRAG_COMPLETE = BIT(2),
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020029};
30
31/**
32 * struct inet_frag_queue - fragment queue
33 *
34 * @lock: spinlock protecting the queue
35 * @timer: queue expiration timer
36 * @list: hash bucket list
37 * @refcnt: reference count of the queue
38 * @fragments: received fragments head
39 * @fragments_tail: received fragments tail
40 * @stamp: timestamp of the last received fragment
41 * @len: total length of the original datagram
42 * @meat: length of received fragments so far
43 * @flags: fragment queue flags
Florian Westphald6b915e2015-05-22 16:32:51 +020044 * @max_size: maximum received fragment size
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020045 * @net: namespace that this frag belongs to
Florian Westphald1fe1942015-07-23 12:05:37 +020046 * @list_evictor: list of queues to forcefully evict (e.g. due to low memory)
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020047 */
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070048struct inet_frag_queue {
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070049 spinlock_t lock;
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020050 struct timer_list timer;
Jesper Dangaard Brouer6e34a8b2013-01-28 23:44:49 +000051 struct hlist_node list;
52 atomic_t refcnt;
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020053 struct sk_buff *fragments;
Changli Gaod6bebca2010-06-29 04:39:37 +000054 struct sk_buff *fragments_tail;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070055 ktime_t stamp;
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020056 int len;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070057 int meat;
Nikolay Aleksandrov1ab19342014-08-01 12:29:45 +020058 __u8 flags;
Patrick McHardy5f2d04f2012-08-26 19:13:55 +020059 u16 max_size;
Jesper Dangaard Brouer6e34a8b2013-01-28 23:44:49 +000060 struct netns_frags *net;
Florian Westphald1fe1942015-07-23 12:05:37 +020061 struct hlist_node list_evictor;
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -070062};
63
Jesper Dangaard Brouera4c40092013-04-25 09:52:25 +000064#define INETFRAGS_HASHSZ 1024
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070065
Hannes Frederic Sowa5a3da1f2013-03-15 11:32:30 +000066/* averaged:
67 * max_depth = default ipfrag_high_thresh / INETFRAGS_HASHSZ /
68 * rounded up (SKB_TRUELEN(0) + sizeof(struct ipq or
69 * struct frag_queue))
70 */
Florian Westphalb13d3cb2014-07-24 16:50:32 +020071#define INETFRAGS_MAXDEPTH 128
Hannes Frederic Sowa5a3da1f2013-03-15 11:32:30 +000072
Jesper Dangaard Brouer19952cc42013-04-03 23:38:16 +000073struct inet_frag_bucket {
74 struct hlist_head chain;
75 spinlock_t chain_lock;
76};
77
Pavel Emelyanov7eb95152007-10-15 02:31:52 -070078struct inet_frags {
Jesper Dangaard Brouer19952cc42013-04-03 23:38:16 +000079 struct inet_frag_bucket hash[INETFRAGS_HASHSZ];
Hannes Frederic Sowa7088ad72013-10-23 11:06:57 +020080
Florian Westphalb13d3cb2014-07-24 16:50:32 +020081 struct work_struct frags_work;
82 unsigned int next_bucket;
Florian Westphale3a57d12014-07-24 16:50:35 +020083 unsigned long last_rebuild_jiffies;
84 bool rebuild;
Florian Westphalb13d3cb2014-07-24 16:50:32 +020085
Hannes Frederic Sowa7088ad72013-10-23 11:06:57 +020086 /* The first call to hashfn is responsible to initialize
87 * rnd. This is best done with net_get_random_once.
Florian Westphalab1c7242014-07-24 16:50:36 +020088 *
89 * rnd_seqlock is used to let hash insertion detect
90 * when it needs to re-lookup the hash chain to use.
Hannes Frederic Sowa7088ad72013-10-23 11:06:57 +020091 */
Jesper Dangaard Brouer5f8e1e82013-01-28 23:44:37 +000092 u32 rnd;
Florian Westphalab1c7242014-07-24 16:50:36 +020093 seqlock_t rnd_seqlock;
Jesper Dangaard Brouer5f8e1e82013-01-28 23:44:37 +000094 int qsize;
Pavel Emelyanov321a3a92007-10-15 02:38:08 -070095
Florian Westphal36c77782014-07-24 16:50:29 +020096 unsigned int (*hashfn)(const struct inet_frag_queue *);
97 bool (*match)(const struct inet_frag_queue *q,
98 const void *arg);
Pavel Emelyanovc6fda282007-10-17 19:46:47 -070099 void (*constructor)(struct inet_frag_queue *q,
Florian Westphal36c77782014-07-24 16:50:29 +0200100 const void *arg);
Pavel Emelyanov1e4b8282007-10-15 02:39:14 -0700101 void (*destructor)(struct inet_frag_queue *);
102 void (*skb_free)(struct sk_buff *);
Pavel Emelyanove521db92007-10-17 19:45:23 -0700103 void (*frag_expire)(unsigned long data);
Nikolay Aleksandrovd4ad4d22014-08-01 12:29:48 +0200104 struct kmem_cache *frags_cachep;
105 const char *frags_cache_name;
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700106};
107
Nikolay Aleksandrovd4ad4d22014-08-01 12:29:48 +0200108int inet_frags_init(struct inet_frags *);
Pavel Emelyanov7eb95152007-10-15 02:31:52 -0700109void inet_frags_fini(struct inet_frags *);
110
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800111void inet_frags_init_net(struct netns_frags *nf);
Pavel Emelyanov81566e82008-01-22 06:12:39 -0800112void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
Pavel Emelyanove5a2bb82008-01-22 06:06:23 -0800113
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700114void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);
Florian Westphal3fd588e2014-07-24 16:50:34 +0200115void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f);
Pavel Emelyanovac18e752008-01-22 06:02:14 -0800116struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
Florian Westphalab1c7242014-07-24 16:50:36 +0200117 struct inet_frags *f, void *key, unsigned int hash);
118
Hannes Frederic Sowa5a3da1f2013-03-15 11:32:30 +0000119void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
120 const char *prefix);
Pavel Emelyanov277e6502007-10-15 02:37:18 -0700121
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700122static inline void inet_frag_put(struct inet_frag_queue *q, struct inet_frags *f)
123{
124 if (atomic_dec_and_test(&q->refcnt))
Florian Westphal3fd588e2014-07-24 16:50:34 +0200125 inet_frag_destroy(q, f);
Pavel Emelyanov762cc402007-10-15 02:41:56 -0700126}
127
Nikolay Aleksandrovcaaecdd2015-07-23 12:05:40 +0200128static inline bool inet_frag_evicting(struct inet_frag_queue *q)
129{
130 return !hlist_unhashed(&q->list_evictor);
131}
132
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000133/* Memory Tracking Functions. */
134
Jesper Dangaard Brouer6d7b8572013-01-28 23:45:33 +0000135/* The default percpu_counter batch size is not big enough to scale to
136 * fragmentation mem acct sizes.
137 * The mem size of a 64K fragment is approx:
138 * (44 fragments * 2944 truesize) + frag_queue struct(200) = 129736 bytes
139 */
140static unsigned int frag_percpu_counter_batch = 130000;
141
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000142static inline int frag_mem_limit(struct netns_frags *nf)
143{
Jesper Dangaard Brouer6d7b8572013-01-28 23:45:33 +0000144 return percpu_counter_read(&nf->mem);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000145}
146
Florian Westphal0e60d242015-07-23 12:05:38 +0200147static inline void sub_frag_mem_limit(struct netns_frags *nf, int i)
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000148{
Florian Westphal0e60d242015-07-23 12:05:38 +0200149 __percpu_counter_add(&nf->mem, -i, frag_percpu_counter_batch);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000150}
151
Florian Westphal0e60d242015-07-23 12:05:38 +0200152static inline void add_frag_mem_limit(struct netns_frags *nf, int i)
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000153{
Florian Westphal0e60d242015-07-23 12:05:38 +0200154 __percpu_counter_add(&nf->mem, i, frag_percpu_counter_batch);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000155}
156
157static inline void init_frag_mem_limit(struct netns_frags *nf)
158{
Tejun Heo908c7f12014-09-08 09:51:29 +0900159 percpu_counter_init(&nf->mem, 0, GFP_KERNEL);
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000160}
161
Florian Westphal36c77782014-07-24 16:50:29 +0200162static inline unsigned int sum_frag_mem_limit(struct netns_frags *nf)
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000163{
Florian Westphal36c77782014-07-24 16:50:29 +0200164 unsigned int res;
Eric Dumazet4cfb0482013-02-22 07:43:35 +0000165
166 local_bh_disable();
167 res = percpu_counter_sum_positive(&nf->mem);
168 local_bh_enable();
169
170 return res;
Jesper Dangaard Brouerd4336732013-01-28 23:45:12 +0000171}
172
Hannes Frederic Sowabe991972013-03-22 08:24:37 +0000173/* RFC 3168 support :
174 * We want to check ECN values of all fragments, do detect invalid combinations.
175 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
176 */
177#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
178#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
179#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
180#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
181
182extern const u8 ip_frag_ecn_table[16];
183
Pavel Emelyanov5ab11c92007-10-15 02:24:19 -0700184#endif