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 init_frag_mem_limit(nf); 77 INIT_LIST_HEAD(&nf->lru_list); 78 spin_lock_init(&nf->lru_lock); 79 } 80 EXPORT_SYMBOL(inet_frags_init_net); 81 82 void inet_frags_fini(struct inet_frags *f) 83 { 84 del_timer(&f->secret_timer); 85 } 86 EXPORT_SYMBOL(inet_frags_fini); 87 88 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f) 89 { 90 nf->low_thresh = 0; 91 92 local_bh_disable(); 93 inet_frag_evictor(nf, f, true); 94 local_bh_enable(); 95 96 percpu_counter_destroy(&nf->mem); 97 } 98 EXPORT_SYMBOL(inet_frags_exit_net); 99 100 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f) 101 { 102 write_lock(&f->lock); 103 hlist_del(&fq->list); 104 fq->net->nqueues--; 105 write_unlock(&f->lock); 106 inet_frag_lru_del(fq); 107 } 108 109 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f) 110 { 111 if (del_timer(&fq->timer)) 112 atomic_dec(&fq->refcnt); 113 114 if (!(fq->last_in & INET_FRAG_COMPLETE)) { 115 fq_unlink(fq, f); 116 atomic_dec(&fq->refcnt); 117 fq->last_in |= INET_FRAG_COMPLETE; 118 } 119 } 120 EXPORT_SYMBOL(inet_frag_kill); 121 122 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, 123 struct sk_buff *skb) 124 { 125 if (f->skb_free) 126 f->skb_free(skb); 127 kfree_skb(skb); 128 } 129 130 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, 131 int *work) 132 { 133 struct sk_buff *fp; 134 struct netns_frags *nf; 135 unsigned int sum, sum_truesize = 0; 136 137 WARN_ON(!(q->last_in & INET_FRAG_COMPLETE)); 138 WARN_ON(del_timer(&q->timer) != 0); 139 140 /* Release all fragment data. */ 141 fp = q->fragments; 142 nf = q->net; 143 while (fp) { 144 struct sk_buff *xp = fp->next; 145 146 sum_truesize += fp->truesize; 147 frag_kfree_skb(nf, f, fp); 148 fp = xp; 149 } 150 sum = sum_truesize + f->qsize; 151 if (work) 152 *work -= sum; 153 sub_frag_mem_limit(q, sum); 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, bool force) 163 { 164 struct inet_frag_queue *q; 165 int work, evicted = 0; 166 167 if (!force) { 168 if (frag_mem_limit(nf) <= nf->high_thresh) 169 return 0; 170 } 171 172 work = frag_mem_limit(nf) - nf->low_thresh; 173 while (work > 0) { 174 spin_lock(&nf->lru_lock); 175 176 if (list_empty(&nf->lru_list)) { 177 spin_unlock(&nf->lru_lock); 178 break; 179 } 180 181 q = list_first_entry(&nf->lru_list, 182 struct inet_frag_queue, lru_list); 183 atomic_inc(&q->refcnt); 184 spin_unlock(&nf->lru_lock); 185 186 spin_lock(&q->lock); 187 if (!(q->last_in & INET_FRAG_COMPLETE)) 188 inet_frag_kill(q, f); 189 spin_unlock(&q->lock); 190 191 if (atomic_dec_and_test(&q->refcnt)) 192 inet_frag_destroy(q, f, &work); 193 evicted++; 194 } 195 196 return evicted; 197 } 198 EXPORT_SYMBOL(inet_frag_evictor); 199 200 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, 201 struct inet_frag_queue *qp_in, struct inet_frags *f, 202 void *arg) 203 { 204 struct inet_frag_queue *qp; 205 #ifdef CONFIG_SMP 206 struct hlist_node *n; 207 #endif 208 unsigned int hash; 209 210 write_lock(&f->lock); 211 /* 212 * While we stayed w/o the lock other CPU could update 213 * the rnd seed, so we need to re-calculate the hash 214 * chain. Fortunatelly the qp_in can be used to get one. 215 */ 216 hash = f->hashfn(qp_in); 217 #ifdef CONFIG_SMP 218 /* With SMP race we have to recheck hash table, because 219 * such entry could be created on other cpu, while we 220 * promoted read lock to write lock. 221 */ 222 hlist_for_each_entry(qp, n, &f->hash[hash], list) { 223 if (qp->net == nf && f->match(qp, arg)) { 224 atomic_inc(&qp->refcnt); 225 write_unlock(&f->lock); 226 qp_in->last_in |= INET_FRAG_COMPLETE; 227 inet_frag_put(qp_in, f); 228 return qp; 229 } 230 } 231 #endif 232 qp = qp_in; 233 if (!mod_timer(&qp->timer, jiffies + nf->timeout)) 234 atomic_inc(&qp->refcnt); 235 236 atomic_inc(&qp->refcnt); 237 hlist_add_head(&qp->list, &f->hash[hash]); 238 nf->nqueues++; 239 write_unlock(&f->lock); 240 inet_frag_lru_add(nf, qp); 241 return qp; 242 } 243 244 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, 245 struct inet_frags *f, void *arg) 246 { 247 struct inet_frag_queue *q; 248 249 q = kzalloc(f->qsize, GFP_ATOMIC); 250 if (q == NULL) 251 return NULL; 252 253 q->net = nf; 254 f->constructor(q, arg); 255 add_frag_mem_limit(q, f->qsize); 256 257 setup_timer(&q->timer, f->frag_expire, (unsigned long)q); 258 spin_lock_init(&q->lock); 259 atomic_set(&q->refcnt, 1); 260 261 return q; 262 } 263 264 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, 265 struct inet_frags *f, void *arg) 266 { 267 struct inet_frag_queue *q; 268 269 q = inet_frag_alloc(nf, f, arg); 270 if (q == NULL) 271 return NULL; 272 273 return inet_frag_intern(nf, q, f, arg); 274 } 275 276 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, 277 struct inet_frags *f, void *key, unsigned int hash) 278 __releases(&f->lock) 279 { 280 struct inet_frag_queue *q; 281 struct hlist_node *n; 282 283 hlist_for_each_entry(q, n, &f->hash[hash], list) { 284 if (q->net == nf && f->match(q, key)) { 285 atomic_inc(&q->refcnt); 286 read_unlock(&f->lock); 287 return q; 288 } 289 } 290 read_unlock(&f->lock); 291 292 return inet_frag_create(nf, f, key); 293 } 294 EXPORT_SYMBOL(inet_frag_find); 295