Loading...
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
8/* Per netns frag queues directory */
9struct fqdir {
10 /* sysctls */
11 long high_thresh;
12 long low_thresh;
13 int timeout;
14 int max_dist;
15 struct inet_frags *f;
16 struct net *net;
17 bool dead;
18
19 struct rhashtable rhashtable ____cacheline_aligned_in_smp;
20
21 /* Keep atomic mem on separate cachelines in structs that include it */
22 atomic_long_t mem ____cacheline_aligned_in_smp;
23 struct work_struct destroy_work;
24};
25
26/**
27 * fragment queue flags
28 *
29 * @INET_FRAG_FIRST_IN: first fragment has arrived
30 * @INET_FRAG_LAST_IN: final fragment has arrived
31 * @INET_FRAG_COMPLETE: frag queue has been processed and is due for destruction
32 * @INET_FRAG_HASH_DEAD: inet_frag_kill() has not removed fq from rhashtable
33 */
34enum {
35 INET_FRAG_FIRST_IN = BIT(0),
36 INET_FRAG_LAST_IN = BIT(1),
37 INET_FRAG_COMPLETE = BIT(2),
38 INET_FRAG_HASH_DEAD = BIT(3),
39};
40
41struct frag_v4_compare_key {
42 __be32 saddr;
43 __be32 daddr;
44 u32 user;
45 u32 vif;
46 __be16 id;
47 u16 protocol;
48};
49
50struct frag_v6_compare_key {
51 struct in6_addr saddr;
52 struct in6_addr daddr;
53 u32 user;
54 __be32 id;
55 u32 iif;
56};
57
58/**
59 * struct inet_frag_queue - fragment queue
60 *
61 * @node: rhash node
62 * @key: keys identifying this frag.
63 * @timer: queue expiration timer
64 * @lock: spinlock protecting this frag
65 * @refcnt: reference count of the queue
66 * @rb_fragments: received fragments rb-tree root
67 * @fragments_tail: received fragments tail
68 * @last_run_head: the head of the last "run". see ip_fragment.c
69 * @stamp: timestamp of the last received fragment
70 * @len: total length of the original datagram
71 * @meat: length of received fragments so far
72 * @flags: fragment queue flags
73 * @max_size: maximum received fragment size
74 * @fqdir: pointer to struct fqdir
75 * @rcu: rcu head for freeing deferall
76 */
77struct inet_frag_queue {
78 struct rhash_head node;
79 union {
80 struct frag_v4_compare_key v4;
81 struct frag_v6_compare_key v6;
82 } key;
83 struct timer_list timer;
84 spinlock_t lock;
85 refcount_t refcnt;
86 struct rb_root rb_fragments;
87 struct sk_buff *fragments_tail;
88 struct sk_buff *last_run_head;
89 ktime_t stamp;
90 int len;
91 int meat;
92 __u8 flags;
93 u16 max_size;
94 struct fqdir *fqdir;
95 struct rcu_head rcu;
96};
97
98struct inet_frags {
99 unsigned int qsize;
100
101 void (*constructor)(struct inet_frag_queue *q,
102 const void *arg);
103 void (*destructor)(struct inet_frag_queue *);
104 void (*frag_expire)(struct timer_list *t);
105 struct kmem_cache *frags_cachep;
106 const char *frags_cache_name;
107 struct rhashtable_params rhash_params;
108 refcount_t refcnt;
109 struct completion completion;
110};
111
112int inet_frags_init(struct inet_frags *);
113void inet_frags_fini(struct inet_frags *);
114
115int fqdir_init(struct fqdir **fqdirp, struct inet_frags *f, struct net *net);
116
117static inline void fqdir_pre_exit(struct fqdir *fqdir)
118{
119 fqdir->high_thresh = 0; /* prevent creation of new frags */
120 fqdir->dead = true;
121}
122void fqdir_exit(struct fqdir *fqdir);
123
124void inet_frag_kill(struct inet_frag_queue *q);
125void inet_frag_destroy(struct inet_frag_queue *q);
126struct inet_frag_queue *inet_frag_find(struct fqdir *fqdir, void *key);
127
128/* Free all skbs in the queue; return the sum of their truesizes. */
129unsigned int inet_frag_rbtree_purge(struct rb_root *root);
130
131static inline void inet_frag_put(struct inet_frag_queue *q)
132{
133 if (refcount_dec_and_test(&q->refcnt))
134 inet_frag_destroy(q);
135}
136
137/* Memory Tracking Functions. */
138
139static inline long frag_mem_limit(const struct fqdir *fqdir)
140{
141 return atomic_long_read(&fqdir->mem);
142}
143
144static inline void sub_frag_mem_limit(struct fqdir *fqdir, long val)
145{
146 atomic_long_sub(val, &fqdir->mem);
147}
148
149static inline void add_frag_mem_limit(struct fqdir *fqdir, long val)
150{
151 atomic_long_add(val, &fqdir->mem);
152}
153
154/* RFC 3168 support :
155 * We want to check ECN values of all fragments, do detect invalid combinations.
156 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
157 */
158#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
159#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
160#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
161#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
162
163extern const u8 ip_frag_ecn_table[16];
164
165/* Return values of inet_frag_queue_insert() */
166#define IPFRAG_OK 0
167#define IPFRAG_DUP 1
168#define IPFRAG_OVERLAP 2
169int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb,
170 int offset, int end);
171void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb,
172 struct sk_buff *parent);
173void inet_frag_reasm_finish(struct inet_frag_queue *q, struct sk_buff *head,
174 void *reasm_data, bool try_coalesce);
175struct sk_buff *inet_frag_pull_head(struct inet_frag_queue *q);
176
177#endif
1#ifndef __NET_FRAG_H__
2#define __NET_FRAG_H__
3
4#include <linux/percpu_counter.h>
5
6struct 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 int max_dist;
17};
18
19/**
20 * fragment queue flags
21 *
22 * @INET_FRAG_FIRST_IN: first fragment has arrived
23 * @INET_FRAG_LAST_IN: final fragment has arrived
24 * @INET_FRAG_COMPLETE: frag queue has been processed and is due for destruction
25 */
26enum {
27 INET_FRAG_FIRST_IN = BIT(0),
28 INET_FRAG_LAST_IN = BIT(1),
29 INET_FRAG_COMPLETE = BIT(2),
30};
31
32/**
33 * struct inet_frag_queue - fragment queue
34 *
35 * @lock: spinlock protecting the queue
36 * @timer: queue expiration timer
37 * @list: hash bucket list
38 * @refcnt: reference count of the queue
39 * @fragments: received fragments head
40 * @fragments_tail: received fragments tail
41 * @stamp: timestamp of the last received fragment
42 * @len: total length of the original datagram
43 * @meat: length of received fragments so far
44 * @flags: fragment queue flags
45 * @max_size: maximum received fragment size
46 * @net: namespace that this frag belongs to
47 * @list_evictor: list of queues to forcefully evict (e.g. due to low memory)
48 */
49struct inet_frag_queue {
50 spinlock_t lock;
51 struct timer_list timer;
52 struct hlist_node list;
53 atomic_t refcnt;
54 struct sk_buff *fragments;
55 struct sk_buff *fragments_tail;
56 ktime_t stamp;
57 int len;
58 int meat;
59 __u8 flags;
60 u16 max_size;
61 struct netns_frags *net;
62 struct hlist_node list_evictor;
63};
64
65#define INETFRAGS_HASHSZ 1024
66
67/* averaged:
68 * max_depth = default ipfrag_high_thresh / INETFRAGS_HASHSZ /
69 * rounded up (SKB_TRUELEN(0) + sizeof(struct ipq or
70 * struct frag_queue))
71 */
72#define INETFRAGS_MAXDEPTH 128
73
74struct inet_frag_bucket {
75 struct hlist_head chain;
76 spinlock_t chain_lock;
77};
78
79struct inet_frags {
80 struct inet_frag_bucket hash[INETFRAGS_HASHSZ];
81
82 struct work_struct frags_work;
83 unsigned int next_bucket;
84 unsigned long last_rebuild_jiffies;
85 bool rebuild;
86
87 /* The first call to hashfn is responsible to initialize
88 * rnd. This is best done with net_get_random_once.
89 *
90 * rnd_seqlock is used to let hash insertion detect
91 * when it needs to re-lookup the hash chain to use.
92 */
93 u32 rnd;
94 seqlock_t rnd_seqlock;
95 int qsize;
96
97 unsigned int (*hashfn)(const struct inet_frag_queue *);
98 bool (*match)(const struct inet_frag_queue *q,
99 const void *arg);
100 void (*constructor)(struct inet_frag_queue *q,
101 const void *arg);
102 void (*destructor)(struct inet_frag_queue *);
103 void (*frag_expire)(unsigned long data);
104 struct kmem_cache *frags_cachep;
105 const char *frags_cache_name;
106};
107
108int inet_frags_init(struct inet_frags *);
109void inet_frags_fini(struct inet_frags *);
110
111static inline int inet_frags_init_net(struct netns_frags *nf)
112{
113 return percpu_counter_init(&nf->mem, 0, GFP_KERNEL);
114}
115static inline void inet_frags_uninit_net(struct netns_frags *nf)
116{
117 percpu_counter_destroy(&nf->mem);
118}
119
120void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
121
122void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);
123void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f);
124struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
125 struct inet_frags *f, void *key, unsigned int hash);
126
127void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
128 const char *prefix);
129
130static inline void inet_frag_put(struct inet_frag_queue *q, struct inet_frags *f)
131{
132 if (atomic_dec_and_test(&q->refcnt))
133 inet_frag_destroy(q, f);
134}
135
136static inline bool inet_frag_evicting(struct inet_frag_queue *q)
137{
138 return !hlist_unhashed(&q->list_evictor);
139}
140
141/* Memory Tracking Functions. */
142
143/* The default percpu_counter batch size is not big enough to scale to
144 * fragmentation mem acct sizes.
145 * The mem size of a 64K fragment is approx:
146 * (44 fragments * 2944 truesize) + frag_queue struct(200) = 129736 bytes
147 */
148static unsigned int frag_percpu_counter_batch = 130000;
149
150static inline int frag_mem_limit(struct netns_frags *nf)
151{
152 return percpu_counter_read(&nf->mem);
153}
154
155static inline void sub_frag_mem_limit(struct netns_frags *nf, int i)
156{
157 __percpu_counter_add(&nf->mem, -i, frag_percpu_counter_batch);
158}
159
160static inline void add_frag_mem_limit(struct netns_frags *nf, int i)
161{
162 __percpu_counter_add(&nf->mem, i, frag_percpu_counter_batch);
163}
164
165static inline unsigned int sum_frag_mem_limit(struct netns_frags *nf)
166{
167 unsigned int res;
168
169 local_bh_disable();
170 res = percpu_counter_sum_positive(&nf->mem);
171 local_bh_enable();
172
173 return res;
174}
175
176/* RFC 3168 support :
177 * We want to check ECN values of all fragments, do detect invalid combinations.
178 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
179 */
180#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
181#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
182#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
183#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
184
185extern const u8 ip_frag_ecn_table[16];
186
187#endif