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 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 */ 26 enum { 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 */ 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 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 74 struct inet_frag_bucket { 75 struct hlist_head chain; 76 spinlock_t chain_lock; 77 }; 78 79 struct 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 }; 107 108 int inet_frags_init(struct inet_frags *); 109 void inet_frags_fini(struct inet_frags *); 110 111 static inline int inet_frags_init_net(struct netns_frags *nf) 112 { 113 return percpu_counter_init(&nf->mem, 0, GFP_KERNEL); 114 } 115 static inline void inet_frags_uninit_net(struct netns_frags *nf) 116 { 117 percpu_counter_destroy(&nf->mem); 118 } 119 120 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f); 121 122 void inet_frag_kill(struct inet_frag_queue *q, struct inet_frags *f); 123 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f); 124 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, 125 struct inet_frags *f, void *key, unsigned int hash); 126 127 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q, 128 const char *prefix); 129 130 static 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 } 135 136 static 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 */ 148 static unsigned int frag_percpu_counter_batch = 130000; 149 150 static inline int frag_mem_limit(struct netns_frags *nf) 151 { 152 return percpu_counter_read(&nf->mem); 153 } 154 155 static 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 } 159 160 static 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 } 164 165 static 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 */ 184 185 extern const u8 ip_frag_ecn_table[16]; 186 187 #endif 188