xref: /linux/include/net/inet_frag.h (revision 352d3ef47efb6f36d44f645387f7746a6fcb4035)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_FRAG_H__
3 #define __NET_FRAG_H__
4 
5 #include <linux/rhashtable-types.h>
6 #include <linux/completion.h>
7 #include <linux/in6.h>
8 #include <linux/rbtree_types.h>
9 #include <linux/refcount.h>
10 #include <net/dropreason-core.h>
11 
12 /* Per netns frag queues directory */
13 struct fqdir {
14 	/* sysctls */
15 	long			high_thresh;
16 	int			timeout;
17 	int			max_dist;
18 	struct inet_frags	*f;
19 	struct net		*net;
20 	bool			dead;
21 
22 	struct rhashtable       rhashtable ____cacheline_aligned_in_smp;
23 
24 	/* Keep atomic mem on separate cachelines in structs that include it */
25 	atomic_long_t		mem ____cacheline_aligned_in_smp;
26 	struct work_struct	destroy_work;
27 	struct llist_node	free_list;
28 };
29 
30 /**
31  * fragment queue flags
32  *
33  * @INET_FRAG_FIRST_IN: first fragment has arrived
34  * @INET_FRAG_LAST_IN: final fragment has arrived
35  * @INET_FRAG_COMPLETE: frag queue has been processed and is due for destruction
36  * @INET_FRAG_HASH_DEAD: inet_frag_kill() has not removed fq from rhashtable
37  * @INET_FRAG_DROP: if skbs must be dropped (instead of being consumed)
38  */
39 enum {
40 	INET_FRAG_FIRST_IN	= BIT(0),
41 	INET_FRAG_LAST_IN	= BIT(1),
42 	INET_FRAG_COMPLETE	= BIT(2),
43 	INET_FRAG_HASH_DEAD	= BIT(3),
44 	INET_FRAG_DROP		= BIT(4),
45 };
46 
47 struct frag_v4_compare_key {
48 	__be32		saddr;
49 	__be32		daddr;
50 	u32		user;
51 	u32		vif;
52 	__be16		id;
53 	u16		protocol;
54 };
55 
56 struct frag_v6_compare_key {
57 	struct in6_addr	saddr;
58 	struct in6_addr	daddr;
59 	u32		user;
60 	__be32		id;
61 	u32		iif;
62 };
63 
64 /**
65  * struct inet_frag_queue - fragment queue
66  *
67  * @node: rhash node
68  * @key: keys identifying this frag.
69  * @timer: queue expiration timer
70  * @lock: spinlock protecting this frag
71  * @refcnt: reference count of the queue
72  * @rb_fragments: received fragments rb-tree root
73  * @fragments_tail: received fragments tail
74  * @last_run_head: the head of the last "run". see ip_fragment.c
75  * @stamp: timestamp of the last received fragment
76  * @len: total length of the original datagram
77  * @meat: length of received fragments so far
78  * @mono_delivery_time: stamp has a mono delivery time (EDT)
79  * @flags: fragment queue flags
80  * @max_size: maximum received fragment size
81  * @fqdir: pointer to struct fqdir
82  * @rcu: rcu head for freeing deferall
83  */
84 struct inet_frag_queue {
85 	struct rhash_head	node;
86 	union {
87 		struct frag_v4_compare_key v4;
88 		struct frag_v6_compare_key v6;
89 	} key;
90 	struct timer_list	timer;
91 	spinlock_t		lock;
92 	refcount_t		refcnt;
93 	struct rb_root		rb_fragments;
94 	struct sk_buff		*fragments_tail;
95 	struct sk_buff		*last_run_head;
96 	ktime_t			stamp;
97 	int			len;
98 	int			meat;
99 	u8			mono_delivery_time;
100 	__u8			flags;
101 	u16			max_size;
102 	struct fqdir		*fqdir;
103 	struct rcu_head		rcu;
104 };
105 
106 struct inet_frags {
107 	unsigned int		qsize;
108 
109 	void			(*constructor)(struct inet_frag_queue *q,
110 					       const void *arg);
111 	void			(*destructor)(struct inet_frag_queue *);
112 	void			(*frag_expire)(struct timer_list *t);
113 	struct kmem_cache	*frags_cachep;
114 	const char		*frags_cache_name;
115 	struct rhashtable_params rhash_params;
116 	refcount_t		refcnt;
117 	struct completion	completion;
118 };
119 
120 int inet_frags_init(struct inet_frags *);
121 void inet_frags_fini(struct inet_frags *);
122 
123 int fqdir_init(struct fqdir **fqdirp, struct inet_frags *f, struct net *net);
124 
125 static inline void fqdir_pre_exit(struct fqdir *fqdir)
126 {
127 	/* Prevent creation of new frags.
128 	 * Pairs with READ_ONCE() in inet_frag_find().
129 	 */
130 	WRITE_ONCE(fqdir->high_thresh, 0);
131 
132 	/* Pairs with READ_ONCE() in inet_frag_kill(), ip_expire()
133 	 * and ip6frag_expire_frag_queue().
134 	 */
135 	WRITE_ONCE(fqdir->dead, true);
136 }
137 void fqdir_exit(struct fqdir *fqdir);
138 
139 void inet_frag_kill(struct inet_frag_queue *q);
140 void inet_frag_destroy(struct inet_frag_queue *q);
141 struct inet_frag_queue *inet_frag_find(struct fqdir *fqdir, void *key);
142 
143 /* Free all skbs in the queue; return the sum of their truesizes. */
144 unsigned int inet_frag_rbtree_purge(struct rb_root *root,
145 				    enum skb_drop_reason reason);
146 
147 static inline void inet_frag_put(struct inet_frag_queue *q)
148 {
149 	if (refcount_dec_and_test(&q->refcnt))
150 		inet_frag_destroy(q);
151 }
152 
153 /* Memory Tracking Functions. */
154 
155 static inline long frag_mem_limit(const struct fqdir *fqdir)
156 {
157 	return atomic_long_read(&fqdir->mem);
158 }
159 
160 static inline void sub_frag_mem_limit(struct fqdir *fqdir, long val)
161 {
162 	atomic_long_sub(val, &fqdir->mem);
163 }
164 
165 static inline void add_frag_mem_limit(struct fqdir *fqdir, long val)
166 {
167 	atomic_long_add(val, &fqdir->mem);
168 }
169 
170 /* RFC 3168 support :
171  * We want to check ECN values of all fragments, do detect invalid combinations.
172  * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
173  */
174 #define	IPFRAG_ECN_NOT_ECT	0x01 /* one frag had ECN_NOT_ECT */
175 #define	IPFRAG_ECN_ECT_1	0x02 /* one frag had ECN_ECT_1 */
176 #define	IPFRAG_ECN_ECT_0	0x04 /* one frag had ECN_ECT_0 */
177 #define	IPFRAG_ECN_CE		0x08 /* one frag had ECN_CE */
178 
179 extern const u8 ip_frag_ecn_table[16];
180 
181 /* Return values of inet_frag_queue_insert() */
182 #define IPFRAG_OK	0
183 #define IPFRAG_DUP	1
184 #define IPFRAG_OVERLAP	2
185 int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb,
186 			   int offset, int end);
187 void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb,
188 			      struct sk_buff *parent);
189 void inet_frag_reasm_finish(struct inet_frag_queue *q, struct sk_buff *head,
190 			    void *reasm_data, bool try_coalesce);
191 struct sk_buff *inet_frag_pull_head(struct inet_frag_queue *q);
192 
193 #endif
194