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#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 */
13struct fqdir {
14 /* sysctls */
15 long high_thresh;
16 long low_thresh;
17 int timeout;
18 int max_dist;
19 struct inet_frags *f;
20 struct net *net;
21 bool dead;
22
23 struct rhashtable rhashtable ____cacheline_aligned_in_smp;
24
25 /* Keep atomic mem on separate cachelines in structs that include it */
26 atomic_long_t mem ____cacheline_aligned_in_smp;
27 struct work_struct destroy_work;
28 struct llist_node free_list;
29};
30
31/**
32 * enum: fragment queue flags
33 *
34 * @INET_FRAG_FIRST_IN: first fragment has arrived
35 * @INET_FRAG_LAST_IN: final fragment has arrived
36 * @INET_FRAG_COMPLETE: frag queue has been processed and is due for destruction
37 * @INET_FRAG_HASH_DEAD: inet_frag_kill() has not removed fq from rhashtable
38 * @INET_FRAG_DROP: if skbs must be dropped (instead of being consumed)
39 */
40enum {
41 INET_FRAG_FIRST_IN = BIT(0),
42 INET_FRAG_LAST_IN = BIT(1),
43 INET_FRAG_COMPLETE = BIT(2),
44 INET_FRAG_HASH_DEAD = BIT(3),
45 INET_FRAG_DROP = BIT(4),
46};
47
48struct frag_v4_compare_key {
49 __be32 saddr;
50 __be32 daddr;
51 u32 user;
52 u32 vif;
53 __be16 id;
54 u16 protocol;
55};
56
57struct frag_v6_compare_key {
58 struct in6_addr saddr;
59 struct in6_addr daddr;
60 u32 user;
61 __be32 id;
62 u32 iif;
63};
64
65/**
66 * struct inet_frag_queue - fragment queue
67 *
68 * @node: rhash node
69 * @key: keys identifying this frag.
70 * @timer: queue expiration timer
71 * @lock: spinlock protecting this frag
72 * @refcnt: reference count of the queue
73 * @rb_fragments: received fragments rb-tree root
74 * @fragments_tail: received fragments tail
75 * @last_run_head: the head of the last "run". see ip_fragment.c
76 * @stamp: timestamp of the last received fragment
77 * @len: total length of the original datagram
78 * @meat: length of received fragments so far
79 * @tstamp_type: stamp has a mono delivery time (EDT)
80 * @flags: fragment queue flags
81 * @max_size: maximum received fragment size
82 * @fqdir: pointer to struct fqdir
83 * @rcu: rcu head for freeing deferall
84 */
85struct inet_frag_queue {
86 struct rhash_head node;
87 union {
88 struct frag_v4_compare_key v4;
89 struct frag_v6_compare_key v6;
90 } key;
91 struct timer_list timer;
92 spinlock_t lock;
93 refcount_t refcnt;
94 struct rb_root rb_fragments;
95 struct sk_buff *fragments_tail;
96 struct sk_buff *last_run_head;
97 ktime_t stamp;
98 int len;
99 int meat;
100 u8 tstamp_type;
101 __u8 flags;
102 u16 max_size;
103 struct fqdir *fqdir;
104 struct rcu_head rcu;
105};
106
107struct inet_frags {
108 unsigned int qsize;
109
110 void (*constructor)(struct inet_frag_queue *q,
111 const void *arg);
112 void (*destructor)(struct inet_frag_queue *);
113 void (*frag_expire)(struct timer_list *t);
114 struct kmem_cache *frags_cachep;
115 const char *frags_cache_name;
116 struct rhashtable_params rhash_params;
117 refcount_t refcnt;
118 struct completion completion;
119};
120
121int inet_frags_init(struct inet_frags *);
122void inet_frags_fini(struct inet_frags *);
123
124int fqdir_init(struct fqdir **fqdirp, struct inet_frags *f, struct net *net);
125
126static inline void fqdir_pre_exit(struct fqdir *fqdir)
127{
128 /* Prevent creation of new frags.
129 * Pairs with READ_ONCE() in inet_frag_find().
130 */
131 WRITE_ONCE(fqdir->high_thresh, 0);
132
133 /* Pairs with READ_ONCE() in inet_frag_kill(), ip_expire()
134 * and ip6frag_expire_frag_queue().
135 */
136 WRITE_ONCE(fqdir->dead, true);
137}
138void fqdir_exit(struct fqdir *fqdir);
139
140void inet_frag_kill(struct inet_frag_queue *q);
141void inet_frag_destroy(struct inet_frag_queue *q);
142struct inet_frag_queue *inet_frag_find(struct fqdir *fqdir, void *key);
143
144/* Free all skbs in the queue; return the sum of their truesizes. */
145unsigned int inet_frag_rbtree_purge(struct rb_root *root,
146 enum skb_drop_reason reason);
147
148static inline void inet_frag_put(struct inet_frag_queue *q)
149{
150 if (refcount_dec_and_test(&q->refcnt))
151 inet_frag_destroy(q);
152}
153
154/* Memory Tracking Functions. */
155
156static inline long frag_mem_limit(const struct fqdir *fqdir)
157{
158 return atomic_long_read(&fqdir->mem);
159}
160
161static inline void sub_frag_mem_limit(struct fqdir *fqdir, long val)
162{
163 atomic_long_sub(val, &fqdir->mem);
164}
165
166static inline void add_frag_mem_limit(struct fqdir *fqdir, long val)
167{
168 atomic_long_add(val, &fqdir->mem);
169}
170
171/* RFC 3168 support :
172 * We want to check ECN values of all fragments, do detect invalid combinations.
173 * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
174 */
175#define IPFRAG_ECN_NOT_ECT 0x01 /* one frag had ECN_NOT_ECT */
176#define IPFRAG_ECN_ECT_1 0x02 /* one frag had ECN_ECT_1 */
177#define IPFRAG_ECN_ECT_0 0x04 /* one frag had ECN_ECT_0 */
178#define IPFRAG_ECN_CE 0x08 /* one frag had ECN_CE */
179
180extern const u8 ip_frag_ecn_table[16];
181
182/* Return values of inet_frag_queue_insert() */
183#define IPFRAG_OK 0
184#define IPFRAG_DUP 1
185#define IPFRAG_OVERLAP 2
186int inet_frag_queue_insert(struct inet_frag_queue *q, struct sk_buff *skb,
187 int offset, int end);
188void *inet_frag_reasm_prepare(struct inet_frag_queue *q, struct sk_buff *skb,
189 struct sk_buff *parent);
190void inet_frag_reasm_finish(struct inet_frag_queue *q, struct sk_buff *head,
191 void *reasm_data, bool try_coalesce);
192struct sk_buff *inet_frag_pull_head(struct inet_frag_queue *q);
193
194#endif
1#ifndef __NET_FRAG_H__
2#define __NET_FRAG_H__
3
4#include <linux/percpu_counter.h>
5
6struct netns_frags {
7 int nqueues;
8 struct list_head lru_list;
9 spinlock_t lru_lock;
10
11 /* The percpu_counter "mem" need to be cacheline aligned.
12 * mem.count must not share cacheline with other writers
13 */
14 struct percpu_counter mem ____cacheline_aligned_in_smp;
15
16 /* sysctls */
17 int timeout;
18 int high_thresh;
19 int low_thresh;
20};
21
22struct inet_frag_queue {
23 spinlock_t lock;
24 struct timer_list timer; /* when will this queue expire? */
25 struct list_head lru_list; /* lru list member */
26 struct hlist_node list;
27 atomic_t refcnt;
28 struct sk_buff *fragments; /* list of received fragments */
29 struct sk_buff *fragments_tail;
30 ktime_t stamp;
31 int len; /* total length of orig datagram */
32 int meat;
33 __u8 last_in; /* first/last segment arrived? */
34
35#define INET_FRAG_COMPLETE 4
36#define INET_FRAG_FIRST_IN 2
37#define INET_FRAG_LAST_IN 1
38
39 u16 max_size;
40
41 struct netns_frags *net;
42};
43
44#define INETFRAGS_HASHSZ 1024
45
46/* averaged:
47 * max_depth = default ipfrag_high_thresh / INETFRAGS_HASHSZ /
48 * rounded up (SKB_TRUELEN(0) + sizeof(struct ipq or
49 * struct frag_queue))
50 */
51#define INETFRAGS_MAXDEPTH 128
52
53struct inet_frag_bucket {
54 struct hlist_head chain;
55 spinlock_t chain_lock;
56};
57
58struct inet_frags {
59 struct inet_frag_bucket hash[INETFRAGS_HASHSZ];
60 /* This rwlock is a global lock (seperate per IPv4, IPv6 and
61 * netfilter). Important to keep this on a seperate cacheline.
62 * Its primarily a rebuild protection rwlock.
63 */
64 rwlock_t lock ____cacheline_aligned_in_smp;
65 int secret_interval;
66 struct timer_list secret_timer;
67
68 /* The first call to hashfn is responsible to initialize
69 * rnd. This is best done with net_get_random_once.
70 */
71 u32 rnd;
72 int qsize;
73
74 unsigned int (*hashfn)(struct inet_frag_queue *);
75 bool (*match)(struct inet_frag_queue *q, void *arg);
76 void (*constructor)(struct inet_frag_queue *q,
77 void *arg);
78 void (*destructor)(struct inet_frag_queue *);
79 void (*skb_free)(struct sk_buff *);
80 void (*frag_expire)(unsigned long data);
81};
82
83void inet_frags_init(struct inet_frags *);
84void inet_frags_fini(struct inet_frags *);
85
86void inet_frags_init_net(struct netns_frags *nf);
87void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
88
89void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);
90void inet_frag_destroy(struct inet_frag_queue *q,
91 struct inet_frags *f, int *work);
92int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force);
93struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
94 struct inet_frags *f, void *key, unsigned int hash)
95 __releases(&f->lock);
96void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
97 const char *prefix);
98
99static inline void inet_frag_put(struct inet_frag_queue *q, struct inet_frags *f)
100{
101 if (atomic_dec_and_test(&q->refcnt))
102 inet_frag_destroy(q, f, NULL);
103}
104
105/* Memory Tracking Functions. */
106
107/* The default percpu_counter batch size is not big enough to scale to
108 * fragmentation mem acct sizes.
109 * The mem size of a 64K fragment is approx:
110 * (44 fragments * 2944 truesize) + frag_queue struct(200) = 129736 bytes
111 */
112static unsigned int frag_percpu_counter_batch = 130000;
113
114static inline int frag_mem_limit(struct netns_frags *nf)
115{
116 return percpu_counter_read(&nf->mem);
117}
118
119static inline void sub_frag_mem_limit(struct inet_frag_queue *q, int i)
120{
121 __percpu_counter_add(&q->net->mem, -i, frag_percpu_counter_batch);
122}
123
124static inline void add_frag_mem_limit(struct inet_frag_queue *q, int i)
125{
126 __percpu_counter_add(&q->net->mem, i, frag_percpu_counter_batch);
127}
128
129static inline void init_frag_mem_limit(struct netns_frags *nf)
130{
131 percpu_counter_init(&nf->mem, 0);
132}
133
134static inline int sum_frag_mem_limit(struct netns_frags *nf)
135{
136 int res;
137
138 local_bh_disable();
139 res = percpu_counter_sum_positive(&nf->mem);
140 local_bh_enable();
141
142 return res;
143}
144
145static inline void inet_frag_lru_move(struct inet_frag_queue *q)
146{
147 spin_lock(&q->net->lru_lock);
148 if (!list_empty(&q->lru_list))
149 list_move_tail(&q->lru_list, &q->net->lru_list);
150 spin_unlock(&q->net->lru_lock);
151}
152
153static inline void inet_frag_lru_del(struct inet_frag_queue *q)
154{
155 spin_lock(&q->net->lru_lock);
156 list_del_init(&q->lru_list);
157 q->net->nqueues--;
158 spin_unlock(&q->net->lru_lock);
159}
160
161static inline void inet_frag_lru_add(struct netns_frags *nf,
162 struct inet_frag_queue *q)
163{
164 spin_lock(&nf->lru_lock);
165 list_add_tail(&q->lru_list, &nf->lru_list);
166 q->net->nqueues++;
167 spin_unlock(&nf->lru_lock);
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
179extern const u8 ip_frag_ecn_table[16];
180
181#endif