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 /* 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