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, true); 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 EXPORT_SYMBOL(inet_frag_kill); 118 119 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f, 120 struct sk_buff *skb, int *work) 121 { 122 if (work) 123 *work -= skb->truesize; 124 125 atomic_sub(skb->truesize, &nf->mem); 126 if (f->skb_free) 127 f->skb_free(skb); 128 kfree_skb(skb); 129 } 130 131 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f, 132 int *work) 133 { 134 struct sk_buff *fp; 135 struct netns_frags *nf; 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 frag_kfree_skb(nf, f, fp, work); 147 fp = xp; 148 } 149 150 if (work) 151 *work -= f->qsize; 152 atomic_sub(f->qsize, &nf->mem); 153 154 if (f->destructor) 155 f->destructor(q); 156 kfree(q); 157 158 } 159 EXPORT_SYMBOL(inet_frag_destroy); 160 161 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force) 162 { 163 struct inet_frag_queue *q; 164 int work, evicted = 0; 165 166 if (!force) { 167 if (atomic_read(&nf->mem) <= nf->high_thresh) 168 return 0; 169 } 170 171 work = atomic_read(&nf->mem) - nf->low_thresh; 172 while (work > 0) { 173 read_lock(&f->lock); 174 if (list_empty(&nf->lru_list)) { 175 read_unlock(&f->lock); 176 break; 177 } 178 179 q = list_first_entry(&nf->lru_list, 180 struct inet_frag_queue, lru_list); 181 atomic_inc(&q->refcnt); 182 read_unlock(&f->lock); 183 184 spin_lock(&q->lock); 185 if (!(q->last_in & INET_FRAG_COMPLETE)) 186 inet_frag_kill(q, f); 187 spin_unlock(&q->lock); 188 189 if (atomic_dec_and_test(&q->refcnt)) 190 inet_frag_destroy(q, f, &work); 191 evicted++; 192 } 193 194 return evicted; 195 } 196 EXPORT_SYMBOL(inet_frag_evictor); 197 198 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf, 199 struct inet_frag_queue *qp_in, struct inet_frags *f, 200 void *arg) 201 { 202 struct inet_frag_queue *qp; 203 #ifdef CONFIG_SMP 204 struct hlist_node *n; 205 #endif 206 unsigned int hash; 207 208 write_lock(&f->lock); 209 /* 210 * While we stayed w/o the lock other CPU could update 211 * the rnd seed, so we need to re-calculate the hash 212 * chain. Fortunatelly the qp_in can be used to get one. 213 */ 214 hash = f->hashfn(qp_in); 215 #ifdef CONFIG_SMP 216 /* With SMP race we have to recheck hash table, because 217 * such entry could be created on other cpu, while we 218 * promoted read lock to write lock. 219 */ 220 hlist_for_each_entry(qp, n, &f->hash[hash], list) { 221 if (qp->net == nf && f->match(qp, arg)) { 222 atomic_inc(&qp->refcnt); 223 write_unlock(&f->lock); 224 qp_in->last_in |= INET_FRAG_COMPLETE; 225 inet_frag_put(qp_in, f); 226 return qp; 227 } 228 } 229 #endif 230 qp = qp_in; 231 if (!mod_timer(&qp->timer, jiffies + nf->timeout)) 232 atomic_inc(&qp->refcnt); 233 234 atomic_inc(&qp->refcnt); 235 hlist_add_head(&qp->list, &f->hash[hash]); 236 list_add_tail(&qp->lru_list, &nf->lru_list); 237 nf->nqueues++; 238 write_unlock(&f->lock); 239 return qp; 240 } 241 242 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf, 243 struct inet_frags *f, void *arg) 244 { 245 struct inet_frag_queue *q; 246 247 q = kzalloc(f->qsize, GFP_ATOMIC); 248 if (q == NULL) 249 return NULL; 250 251 q->net = nf; 252 f->constructor(q, arg); 253 atomic_add(f->qsize, &nf->mem); 254 setup_timer(&q->timer, f->frag_expire, (unsigned long)q); 255 spin_lock_init(&q->lock); 256 atomic_set(&q->refcnt, 1); 257 258 return q; 259 } 260 261 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf, 262 struct inet_frags *f, void *arg) 263 { 264 struct inet_frag_queue *q; 265 266 q = inet_frag_alloc(nf, f, arg); 267 if (q == NULL) 268 return NULL; 269 270 return inet_frag_intern(nf, q, f, arg); 271 } 272 273 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf, 274 struct inet_frags *f, void *key, unsigned int hash) 275 __releases(&f->lock) 276 { 277 struct inet_frag_queue *q; 278 struct hlist_node *n; 279 280 hlist_for_each_entry(q, n, &f->hash[hash], list) { 281 if (q->net == nf && f->match(q, key)) { 282 atomic_inc(&q->refcnt); 283 read_unlock(&f->lock); 284 return q; 285 } 286 } 287 read_unlock(&f->lock); 288 289 return inet_frag_create(nf, f, key); 290 } 291 EXPORT_SYMBOL(inet_frag_find); 292