1 /* 2 * inet fragments management 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 * 9 * Authors: Pavel Emelyanov <xemul@openvz.org> 10 * Started as consolidation of ipv4/ip_fragment.c, 11 * ipv6/reassembly. and ipv6 nf conntrack reassembly 12 */ 13 14 #include <linux/list.h> 15 #include <linux/spinlock.h> 16 #include <linux/module.h> 17 #include <linux/timer.h> 18 #include <linux/mm.h> 19 #include <linux/random.h> 20 #include <linux/skbuff.h> 21 #include <linux/rtnetlink.h> 22 #include <linux/slab.h> 23 24 #include <net/inet_frag.h> 25 26 static void inet_frag_secret_rebuild(unsigned long dummy) 27 { 28 struct inet_frags *f = (struct inet_frags *)dummy; 29 unsigned long now = jiffies; 30 int i; 31 32 write_lock(&f->lock); 33 get_random_bytes(&f->rnd, sizeof(u32)); 34 for (i = 0; i < INETFRAGS_HASHSZ; i++) { 35 struct inet_frag_queue *q; 36 struct hlist_node *p, *n; 37 38 hlist_for_each_entry_safe(q, p, n, &f->hash[i], list) { 39 unsigned int hval = f->hashfn(q); 40 41 if (hval != i) { 42 hlist_del(&q->list); 43 44 /* Relink to new hash chain. */ 45 hlist_add_head(&q->list, &f->hash[hval]); 46 } 47 } 48 } 49 write_unlock(&f->lock); 50 51 mod_timer(&f->secret_timer, now + f->secret_interval); 52 } 53 54 void inet_frags_init(struct inet_frags *f) 55 { 56 int i; 57 58 for (i = 0; i < INETFRAGS_HASHSZ; i++) 59 INIT_HLIST_HEAD(&f->hash[i]); 60 61 rwlock_init(&f->lock); 62 63 f->rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ 64 (jiffies ^ (jiffies >> 6))); 65 66 setup_timer(&f->secret_timer, inet_frag_secret_rebuild, 67 (unsigned long)f); 68 f->secret_timer.expires = jiffies + f->secret_interval; 69 add_timer(&f->secret_timer); 70 } 71 EXPORT_SYMBOL(inet_frags_init); 72 73 void inet_frags_init_net(struct netns_frags *nf) 74 { 75 nf->nqueues = 0; 76 atomic_set(&nf->mem, 0); 77 INIT_LIST_HEAD(&nf->lru_list); 78 } 79 EXPORT_SYMBOL(inet_frags_init_net); 80 81 void inet_frags_fini(struct inet_frags *f) 82 { 83 del_timer(&f->secret_timer); 84 } 85 EXPORT_SYMBOL(inet_frags_fini); 86 87 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) 88 { 89 nf->low_thresh = 0; 90 91 local_bh_disable(); 92 inet_frag_evictor(nf, f); 93 local_bh_enable(); 94 } 95 EXPORT_SYMBOL(inet_frags_exit_net); 96 97 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) 98 { 99 write_lock(&f->lock); 100 hlist_del(&fq->list); 101 list_del(&fq->lru_list); 102 fq->net->nqueues--; 103 write_unlock(&f->lock); 104 } 105 106 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) 107 { 108 if (del_timer(&fq->timer)) 109 atomic_dec(&fq->refcnt); 110 111 if (!(fq->last_in & INET_FRAG_COMPLETE)) { 112 fq_unlink(fq, f); 113 atomic_dec(&fq->refcnt); 114 fq->last_in |= INET_FRAG_COMPLETE; 115 } 116 } 117 118 EXPORT_SYMBOL(inet_frag_kill); 119 120 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, 121 struct sk_buff *skb, int *work) 122 { 123 if (work) 124 *work -= skb->truesize; 125 126 atomic_sub(skb->truesize, &nf->mem); 127 if (f->skb_free) 128 f->skb_free(skb); 129 kfree_skb(skb); 130 } 131 132 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, 133 int *work) 134 { 135 struct sk_buff *fp; 136 struct netns_frags *nf; 137 138 WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); 139 WARN_ON(del_timer(&q->timer) != 0); 140 141 /* Release all fragment data. */ 142 fp = q->fragments; 143 nf = q->net; 144 while (fp) { 145 struct sk_buff *xp = fp->next; 146 147 frag_kfree_skb(nf, f, fp, work); 148 fp = xp; 149 } 150 151 if (work) 152 *work -= f->qsize; 153 atomic_sub(f->qsize, &nf->mem); 154 155 if (f->destructor) 156 f->destructor(q); 157 kfree(q); 158 159 } 160 EXPORT_SYMBOL(inet_frag_destroy); 161 162 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f) 163 { 164 struct inet_frag_queue *q; 165 int work, evicted = 0; 166 167 work = atomic_read(&nf->mem) - nf->low_thresh; 168 while (work > 0) { 169 read_lock(&f->lock); 170 if (list_empty(&nf->lru_list)) { 171 read_unlock(&f->lock); 172 break; 173 } 174 175 q = list_first_entry(&nf->lru_list, 176 struct inet_frag_queue, lru_list); 177 atomic_inc(&q->refcnt); 178 read_unlock(&f->lock); 179 180 spin_lock(&q->lock); 181 if (!(q->last_in & INET_FRAG_COMPLETE)) 182 inet_frag_kill(q, f); 183 spin_unlock(&q->lock); 184 185 if (atomic_dec_and_test(&q->refcnt)) 186 inet_frag_destroy(q, f, &work); 187 evicted++; 188 } 189 190 return evicted; 191 } 192 EXPORT_SYMBOL(inet_frag_evictor); 193 194 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, 195 struct inet_frag_queue *qp_in, struct inet_frags *f, 196 void *arg) 197 { 198 struct inet_frag_queue *qp; 199 #ifdef CONFIG_SMP 200 struct hlist_node *n; 201 #endif 202 unsigned int hash; 203 204 write_lock(&f->lock); 205 /* 206 * While we stayed w/o the lock other CPU could update 207 * the rnd seed, so we need to re-calculate the hash 208 * chain. Fortunatelly the qp_in can be used to get one. 209 */ 210 hash = f->hashfn(qp_in); 211 #ifdef CONFIG_SMP 212 /* With SMP race we have to recheck hash table, because 213 * such entry could be created on other cpu, while we 214 * promoted read lock to write lock. 215 */ 216 hlist_for_each_entry(qp, n, &f->hash[hash], list) { 217 if (qp->net == nf && f->match(qp, arg)) { 218 atomic_inc(&qp->refcnt); 219 write_unlock(&f->lock); 220 qp_in->last_in |= INET_FRAG_COMPLETE; 221 inet_frag_put(qp_in, f); 222 return qp; 223 } 224 } 225 #endif 226 qp = qp_in; 227 if (!mod_timer(&qp->timer, jiffies + nf->timeout)) 228 atomic_inc(&qp->refcnt); 229 230 atomic_inc(&qp->refcnt); 231 hlist_add_head(&qp->list, &f->hash[hash]); 232 list_add_tail(&qp->lru_list, &nf->lru_list); 233 nf->nqueues++; 234 write_unlock(&f->lock); 235 return qp; 236 } 237 238 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, 239 struct inet_frags *f, void *arg) 240 { 241 struct inet_frag_queue *q; 242 243 q = kzalloc(f->qsize, GFP_ATOMIC); 244 if (q == NULL) 245 return NULL; 246 247 f->constructor(q, arg); 248 atomic_add(f->qsize, &nf->mem); 249 setup_timer(&q->timer, f->frag_expire, (unsigned long)q); 250 spin_lock_init(&q->lock); 251 atomic_set(&q->refcnt, 1); 252 q->net = nf; 253 254 return q; 255 } 256 257 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, 258 struct inet_frags *f, void *arg) 259 { 260 struct inet_frag_queue *q; 261 262 q = inet_frag_alloc(nf, f, arg); 263 if (q == NULL) 264 return NULL; 265 266 return inet_frag_intern(nf, q, f, arg); 267 } 268 269 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, 270 struct inet_frags *f, void *key, unsigned int hash) 271 __releases(&f->lock) 272 { 273 struct inet_frag_queue *q; 274 struct hlist_node *n; 275 276 hlist_for_each_entry(q, n, &f->hash[hash], list) { 277 if (q->net == nf && f->match(q, key)) { 278 atomic_inc(&q->refcnt); 279 read_unlock(&f->lock); 280 return q; 281 } 282 } 283 read_unlock(&f->lock); 284 285 return inet_frag_create(nf, f, key); 286 } 287 EXPORT_SYMBOL(inet_frag_find); 288