include/net/inet_frag.h

   1 #ifndef __NET_FRAG_H__
   2 #define __NET_FRAG_H__
   3
   4 #include <linux/percpu_counter.h>
   5
   6 struct netns_frags {
   7         struct list_head        lru_list;
   8         spinlock_t              lru_lock;
   9
  10         /* The percpu_counter "mem" need to be cacheline aligned.
  11          *  mem.count must not share cacheline with other writers
  12          */
  13         struct percpu_counter   mem ____cacheline_aligned_in_smp;
  14
  15         /* sysctls */
  16         int                     timeout;
  17         int                     high_thresh;
  18         int                     low_thresh;
  19 };
  20
  21 struct inet_frag_queue {
  22         spinlock_t              lock;
  23         struct timer_list       timer;      /* when will this queue expire? */
  24         struct list_head        lru_list;   /* lru list member */
  25         struct hlist_node       list;
  26         atomic_t                refcnt;
  27         struct sk_buff          *fragments; /* list of received fragments */
  28         struct sk_buff          *fragments_tail;
  29         ktime_t                 stamp;
  30         int                     len;        /* total length of orig datagram */
  31         int                     meat;
  32         __u8                    last_in;    /* first/last segment arrived? */
  33
  34 #define INET_FRAG_EVICTED       8
  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
  53 struct inet_frag_bucket {
  54         struct hlist_head       chain;
  55         spinlock_t              chain_lock;
  56 };
  57
  58 struct 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         struct work_struct      frags_work;
  69         unsigned int next_bucket;
  70
  71         /* The first call to hashfn is responsible to initialize
  72          * rnd. This is best done with net_get_random_once.
  73          */
  74         u32                     rnd;
  75         int                     qsize;
  76
  77         unsigned int            (*hashfn)(const struct inet_frag_queue *);
  78         bool                    (*match)(const struct inet_frag_queue *q,
  79                                          const void *arg);
  80         void                    (*constructor)(struct inet_frag_queue *q,
  81                                                const void *arg);
  82         void                    (*destructor)(struct inet_frag_queue *);
  83         void                    (*skb_free)(struct sk_buff *);
  84         void                    (*frag_expire)(unsigned long data);
  85 };
  86
  87 void inet_frags_init(struct inet_frags *);
  88 void inet_frags_fini(struct inet_frags *);
  89
  90 void inet_frags_init_net(struct netns_frags *nf);
  91 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
  92
  93 void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);
  94 void inet_frag_destroy(struct inet_frag_queue *q,
  95                                 struct inet_frags *f, int *work);
  96 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
  97                 struct inet_frags *f, void *key, unsigned int hash)
  98         __releases(&f->lock);
  99 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
 100                                    const char *prefix);
 101
 102 static inline void inet_frag_put(struct inet_frag_queue *q, struct inet_frags *f)
 103 {
 104         if (atomic_dec_and_test(&q->refcnt))
 105                 inet_frag_destroy(q, f, NULL);
 106 }
 107
 108 /* Memory Tracking Functions. */
 109
 110 /* The default percpu_counter batch size is not big enough to scale to
 111  * fragmentation mem acct sizes.
 112  * The mem size of a 64K fragment is approx:
 113  *  (44 fragments * 2944 truesize) + frag_queue struct(200) = 129736 bytes
 114  */
 115 static unsigned int frag_percpu_counter_batch = 130000;
 116
 117 static inline int frag_mem_limit(struct netns_frags *nf)
 118 {
 119         return percpu_counter_read(&nf->mem);
 120 }
 121
 122 static inline void sub_frag_mem_limit(struct inet_frag_queue *q, int i)
 123 {
 124         __percpu_counter_add(&q->net->mem, -i, frag_percpu_counter_batch);
 125 }
 126
 127 static inline void add_frag_mem_limit(struct inet_frag_queue *q, int i)
 128 {
 129         __percpu_counter_add(&q->net->mem, i, frag_percpu_counter_batch);
 130 }
 131
 132 static inline void init_frag_mem_limit(struct netns_frags *nf)
 133 {
 134         percpu_counter_init(&nf->mem, 0);
 135 }
 136
 137 static inline unsigned int sum_frag_mem_limit(struct netns_frags *nf)
 138 {
 139         unsigned int res;
 140
 141         local_bh_disable();
 142         res = percpu_counter_sum_positive(&nf->mem);
 143         local_bh_enable();
 144
 145         return res;
 146 }
 147
 148 static inline void inet_frag_lru_move(struct inet_frag_queue *q)
 149 {
 150         spin_lock(&q->net->lru_lock);
 151         if (!list_empty(&q->lru_list))
 152                 list_move_tail(&q->lru_list, &q->net->lru_list);
 153         spin_unlock(&q->net->lru_lock);
 154 }
 155
 156 static inline void inet_frag_lru_del(struct inet_frag_queue *q)
 157 {
 158         spin_lock(&q->net->lru_lock);
 159         list_del_init(&q->lru_list);
 160         spin_unlock(&q->net->lru_lock);
 161 }
 162
 163 static inline void inet_frag_lru_add(struct netns_frags *nf,
 164                                      struct inet_frag_queue *q)
 165 {
 166         spin_lock(&nf->lru_lock);
 167         list_add_tail(&q->lru_list, &nf->lru_list);
 168         spin_unlock(&nf->lru_lock);
 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
 180 extern const u8 ip_frag_ecn_table[16];
 181
 182 #endif