xref: /linux/include/net/inet_frag.h (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 #ifndef __NET_FRAG_H__
2 #define __NET_FRAG_H__
3 
4 #include <linux/percpu_counter.h>
5 
6 struct netns_frags {
7 	/* The percpu_counter "mem" need to be cacheline aligned.
8 	 *  mem.count must not share cacheline with other writers
9 	 */
10 	struct percpu_counter   mem ____cacheline_aligned_in_smp;
11 
12 	/* sysctls */
13 	int			timeout;
14 	int			high_thresh;
15 	int			low_thresh;
16 };
17 
18 /**
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
24  */
25 enum {
26 	INET_FRAG_FIRST_IN	= BIT(0),
27 	INET_FRAG_LAST_IN	= BIT(1),
28 	INET_FRAG_COMPLETE	= BIT(2),
29 };
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
44  * @max_size: maximum received fragment size
45  * @net: namespace that this frag belongs to
46  * @list_evictor: list of queues to forcefully evict (e.g. due to low memory)
47  */
48 struct inet_frag_queue {
49 	spinlock_t		lock;
50 	struct timer_list	timer;
51 	struct hlist_node	list;
52 	atomic_t		refcnt;
53 	struct sk_buff		*fragments;
54 	struct sk_buff		*fragments_tail;
55 	ktime_t			stamp;
56 	int			len;
57 	int			meat;
58 	__u8			flags;
59 	u16			max_size;
60 	struct netns_frags	*net;
61 	struct hlist_node	list_evictor;
62 };
63 
64 #define INETFRAGS_HASHSZ	1024
65 
66 /* 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  */
71 #define INETFRAGS_MAXDEPTH	128
72 
73 struct inet_frag_bucket {
74 	struct hlist_head	chain;
75 	spinlock_t		chain_lock;
76 };
77 
78 struct inet_frags {
79 	struct inet_frag_bucket	hash[INETFRAGS_HASHSZ];
80 
81 	struct work_struct	frags_work;
82 	unsigned int next_bucket;
83 	unsigned long last_rebuild_jiffies;
84 	bool rebuild;
85 
86 	/* The first call to hashfn is responsible to initialize
87 	 * rnd. This is best done with net_get_random_once.
88 	 *
89 	 * rnd_seqlock is used to let hash insertion detect
90 	 * when it needs to re-lookup the hash chain to use.
91 	 */
92 	u32			rnd;
93 	seqlock_t		rnd_seqlock;
94 	int			qsize;
95 
96 	unsigned int		(*hashfn)(const struct inet_frag_queue *);
97 	bool			(*match)(const struct inet_frag_queue *q,
98 					 const void *arg);
99 	void			(*constructor)(struct inet_frag_queue *q,
100 					       const void *arg);
101 	void			(*destructor)(struct inet_frag_queue *);
102 	void			(*skb_free)(struct sk_buff *);
103 	void			(*frag_expire)(unsigned long data);
104 	struct kmem_cache	*frags_cachep;
105 	const char		*frags_cache_name;
106 };
107 
108 int inet_frags_init(struct inet_frags *);
109 void inet_frags_fini(struct inet_frags *);
110 
111 void inet_frags_init_net(struct netns_frags *nf);
112 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
113 
114 void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);
115 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f);
116 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
117 		struct inet_frags *f, void *key, unsigned int hash);
118 
119 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
120 				   const char *prefix);
121 
122 static inline void inet_frag_put(struct inet_frag_queue *q, struct inet_frags *f)
123 {
124 	if (atomic_dec_and_test(&q->refcnt))
125 		inet_frag_destroy(q, f);
126 }
127 
128 static inline bool inet_frag_evicting(struct inet_frag_queue *q)
129 {
130 	return !hlist_unhashed(&q->list_evictor);
131 }
132 
133 /* Memory Tracking Functions. */
134 
135 /* 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  */
140 static unsigned int frag_percpu_counter_batch = 130000;
141 
142 static inline int frag_mem_limit(struct netns_frags *nf)
143 {
144 	return percpu_counter_read(&nf->mem);
145 }
146 
147 static inline void sub_frag_mem_limit(struct netns_frags *nf, int i)
148 {
149 	__percpu_counter_add(&nf->mem, -i, frag_percpu_counter_batch);
150 }
151 
152 static inline void add_frag_mem_limit(struct netns_frags *nf, int i)
153 {
154 	__percpu_counter_add(&nf->mem, i, frag_percpu_counter_batch);
155 }
156 
157 static inline void init_frag_mem_limit(struct netns_frags *nf)
158 {
159 	percpu_counter_init(&nf->mem, 0, GFP_KERNEL);
160 }
161 
162 static inline unsigned int sum_frag_mem_limit(struct netns_frags *nf)
163 {
164 	unsigned int res;
165 
166 	local_bh_disable();
167 	res = percpu_counter_sum_positive(&nf->mem);
168 	local_bh_enable();
169 
170 	return res;
171 }
172 
173 /* 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 
182 extern const u8 ip_frag_ecn_table[16];
183 
184 #endif
185