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         /* 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 };
  17
  18 /**
  19  * fragment queue flags
  20  *
  21  * @INET_FRAG_FIRST_IN: first fragment has arrived
  22  * @INET_FRAG_LAST_IN: final fragment has arrived
  23  * @INET_FRAG_COMPLETE: frag queue has been processed and is due for destruction
  24  * @INET_FRAG_EVICTED: frag queue is being evicted
  25  */
  26 enum {
  27         INET_FRAG_FIRST_IN      = BIT(0),
  28         INET_FRAG_LAST_IN       = BIT(1),
  29         INET_FRAG_COMPLETE      = BIT(2),
  30         INET_FRAG_EVICTED       = BIT(3)
  31 };
  32
  33 /**
  34  * struct inet_frag_queue - fragment queue
  35  *
  36  * @lock: spinlock protecting the queue
  37  * @timer: queue expiration timer
  38  * @list: hash bucket list
  39  * @refcnt: reference count of the queue
  40  * @fragments: received fragments head
  41  * @fragments_tail: received fragments tail
  42  * @stamp: timestamp of the last received fragment
  43  * @len: total length of the original datagram
  44  * @meat: length of received fragments so far
  45  * @flags: fragment queue flags
  46  * @max_size: (ipv4 only) maximum received fragment size with IP_DF set
  47  * @net: namespace that this frag belongs to
  48  */
  49 struct 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 };
  63
  64 #define INETFRAGS_HASHSZ        1024
  65
  66 /* averaged:
  67  * max_depth = default ipfrag_high_thresh / INETFRAGS_HASHSZ /
  68  *             rounded up (SKB_TRUELEN(0) + sizeof(struct ipq or
  69  *             struct frag_queue))
  70  */
  71 #define INETFRAGS_MAXDEPTH      128
  72
  73 struct inet_frag_bucket {
  74         struct hlist_head       chain;
  75         spinlock_t              chain_lock;
  76 };
  77
  78 struct inet_frags {
  79         struct inet_frag_bucket hash[INETFRAGS_HASHSZ];
  80
  81         struct work_struct      frags_work;
  82         unsigned int next_bucket;
  83         unsigned long last_rebuild_jiffies;
  84         bool rebuild;
  85
  86         /* The first call to hashfn is responsible to initialize
  87          * rnd. This is best done with net_get_random_once.
  88          *
  89          * rnd_seqlock is used to let hash insertion detect
  90          * when it needs to re-lookup the hash chain to use.
  91          */
  92         u32                     rnd;
  93         seqlock_t               rnd_seqlock;
  94         int                     qsize;
  95
  96         unsigned int            (*hashfn)(const struct inet_frag_queue *);
  97         bool                    (*match)(const struct inet_frag_queue *q,
  98                                          const void *arg);
  99         void                    (*constructor)(struct inet_frag_queue *q,
 100                                                const void *arg);
 101         void                    (*destructor)(struct inet_frag_queue *);
 102         void                    (*skb_free)(struct sk_buff *);
 103         void                    (*frag_expire)(unsigned long data);
 104         struct kmem_cache       *frags_cachep;
 105         const char              *frags_cache_name;
 106 };
 107
 108 int inet_frags_init(struct inet_frags *);
 109 void inet_frags_fini(struct inet_frags *);
 110
 111 void inet_frags_init_net(struct netns_frags *nf);
 112 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f);
 113
 114 void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f);
 115 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f);
 116 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
 117                 struct inet_frags *f, void *key, unsigned int hash);
 118
 119 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
 120                                    const char *prefix);
 121
 122 static inline void inet_frag_put(struct inet_frag_queue *q, struct inet_frags *f)
 123 {
 124         if (atomic_dec_and_test(&q->refcnt))
 125                 inet_frag_destroy(q, f);
 126 }
 127
 128 /* Memory Tracking Functions. */
 129
 130 /* The default percpu_counter batch size is not big enough to scale to
 131  * fragmentation mem acct sizes.
 132  * The mem size of a 64K fragment is approx:
 133  *  (44 fragments * 2944 truesize) + frag_queue struct(200) = 129736 bytes
 134  */
 135 static unsigned int frag_percpu_counter_batch = 130000;
 136
 137 static inline int frag_mem_limit(struct netns_frags *nf)
 138 {
 139         return percpu_counter_read(&nf->mem);
 140 }
 141
 142 static inline void sub_frag_mem_limit(struct inet_frag_queue *q, int i)
 143 {
 144         __percpu_counter_add(&q->net->mem, -i, frag_percpu_counter_batch);
 145 }
 146
 147 static inline void add_frag_mem_limit(struct inet_frag_queue *q, int i)
 148 {
 149         __percpu_counter_add(&q->net->mem, i, frag_percpu_counter_batch);
 150 }
 151
 152 static inline void init_frag_mem_limit(struct netns_frags *nf)
 153 {
 154         percpu_counter_init(&nf->mem, 0, GFP_KERNEL);
 155 }
 156
 157 static inline unsigned int sum_frag_mem_limit(struct netns_frags *nf)
 158 {
 159         unsigned int res;
 160
 161         local_bh_disable();
 162         res = percpu_counter_sum_positive(&nf->mem);
 163         local_bh_enable();
 164
 165         return res;
 166 }
 167
 168 /* RFC 3168 support :
 169  * We want to check ECN values of all fragments, do detect invalid combinations.
 170  * In ipq->ecn, we store the OR value of each ip4_frag_ecn() fragment value.
 171  */
 172 #define IPFRAG_ECN_NOT_ECT      0x01 /* one frag had ECN_NOT_ECT */
 173 #define IPFRAG_ECN_ECT_1        0x02 /* one frag had ECN_ECT_1 */
 174 #define IPFRAG_ECN_ECT_0        0x04 /* one frag had ECN_ECT_0 */
 175 #define IPFRAG_ECN_CE           0x08 /* one frag had ECN_CE */
 176
 177 extern const u8 ip_frag_ecn_table[16];
 178
 179 #endif