net/ipv4/inet_fragment.c

   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/sock.h>
  25 #include <net/inet_frag.h>
  26 #include <net/inet_ecn.h>
  27
  28 /* Given the OR values of all fragments, apply RFC 3168 5.3 requirements
  29  * Value : 0xff if frame should be dropped.
  30  *         0 or INET_ECN_CE value, to be ORed in to final iph->tos field
  31  */
  32 const u8 ip_frag_ecn_table[16] = {
  33         /* at least one fragment had CE, and others ECT_0 or ECT_1 */
  34         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0]                      = INET_ECN_CE,
  35         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1]                      = INET_ECN_CE,
  36         [IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1]   = INET_ECN_CE,
  37
  38         /* invalid combinations : drop frame */
  39         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE] = 0xff,
  40         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0] = 0xff,
  41         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_1] = 0xff,
  42         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
  43         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0] = 0xff,
  44         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_1] = 0xff,
  45         [IPFRAG_ECN_NOT_ECT | IPFRAG_ECN_CE | IPFRAG_ECN_ECT_0 | IPFRAG_ECN_ECT_1] = 0xff,
  46 };
  47 EXPORT_SYMBOL(ip_frag_ecn_table);
  48
  49 static void inet_frag_secret_rebuild(unsigned long dummy)
  50 {
  51         struct inet_frags *f = (struct inet_frags *)dummy;
  52         unsigned long now = jiffies;
  53         int i;
  54
  55         /* Per bucket lock NOT needed here, due to write lock protection */
  56         write_lock(&f->lock);
  57
  58         get_random_bytes(&f->rnd, sizeof(u32));
  59         for (i = 0; i < INETFRAGS_HASHSZ; i++) {
  60                 struct inet_frag_bucket *hb;
  61                 struct inet_frag_queue *q;
  62                 struct hlist_node *n;
  63
  64                 hb = &f->hash[i];
  65                 hlist_for_each_entry_safe(q, n, &hb->chain, list) {
  66                         unsigned int hval = f->hashfn(q);
  67
  68                         if (hval != i) {
  69                                 struct inet_frag_bucket *hb_dest;
  70
  71                                 hlist_del(&q->list);
  72
  73                                 /* Relink to new hash chain. */
  74                                 hb_dest = &f->hash[hval];
  75                                 hlist_add_head(&q->list, &hb_dest->chain);
  76                         }
  77                 }
  78         }
  79         write_unlock(&f->lock);
  80
  81         mod_timer(&f->secret_timer, now + f->secret_interval);
  82 }
  83
  84 void inet_frags_init(struct inet_frags *f)
  85 {
  86         int i;
  87
  88         for (i = 0; i < INETFRAGS_HASHSZ; i++) {
  89                 struct inet_frag_bucket *hb = &f->hash[i];
  90
  91                 spin_lock_init(&hb->chain_lock);
  92                 INIT_HLIST_HEAD(&hb->chain);
  93         }
  94         rwlock_init(&f->lock);
  95
  96         setup_timer(&f->secret_timer, inet_frag_secret_rebuild,
  97                         (unsigned long)f);
  98         f->secret_timer.expires = jiffies + f->secret_interval;
  99         add_timer(&f->secret_timer);
 100 }
 101 EXPORT_SYMBOL(inet_frags_init);
 102
 103 void inet_frags_init_net(struct netns_frags *nf)
 104 {
 105         nf->nqueues = 0;
 106         init_frag_mem_limit(nf);
 107         INIT_LIST_HEAD(&nf->lru_list);
 108         spin_lock_init(&nf->lru_lock);
 109 }
 110 EXPORT_SYMBOL(inet_frags_init_net);
 111
 112 void inet_frags_fini(struct inet_frags *f)
 113 {
 114         del_timer(&f->secret_timer);
 115 }
 116 EXPORT_SYMBOL(inet_frags_fini);
 117
 118 void inet_frags_exit_net(struct netns_frags *nf, struct inet_frags *f)
 119 {
 120         nf->low_thresh = 0;
 121
 122         local_bh_disable();
 123         inet_frag_evictor(nf, f, true);
 124         local_bh_enable();
 125
 126         percpu_counter_destroy(&nf->mem);
 127 }
 128 EXPORT_SYMBOL(inet_frags_exit_net);
 129
 130 static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
 131 {
 132         struct inet_frag_bucket *hb;
 133         unsigned int hash;
 134
 135         read_lock(&f->lock);
 136         hash = f->hashfn(fq);
 137         hb = &f->hash[hash];
 138
 139         spin_lock(&hb->chain_lock);
 140         hlist_del(&fq->list);
 141         spin_unlock(&hb->chain_lock);
 142
 143         read_unlock(&f->lock);
 144         inet_frag_lru_del(fq);
 145 }
 146
 147 void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
 148 {
 149         if (del_timer(&fq->timer))
 150                 atomic_dec(&fq->refcnt);
 151
 152         if (!(fq->last_in & INET_FRAG_COMPLETE)) {
 153                 fq_unlink(fq, f);
 154                 atomic_dec(&fq->refcnt);
 155                 fq->last_in |= INET_FRAG_COMPLETE;
 156         }
 157 }
 158 EXPORT_SYMBOL(inet_frag_kill);
 159
 160 static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
 161                 struct sk_buff *skb)
 162 {
 163         if (f->skb_free)
 164                 f->skb_free(skb);
 165         kfree_skb(skb);
 166 }
 167
 168 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
 169                                         int *work)
 170 {
 171         struct sk_buff *fp;
 172         struct netns_frags *nf;
 173         unsigned int sum, sum_truesize = 0;
 174
 175         WARN_ON(!(q->last_in & INET_FRAG_COMPLETE));
 176         WARN_ON(del_timer(&q->timer) != 0);
 177
 178         /* Release all fragment data. */
 179         fp = q->fragments;
 180         nf = q->net;
 181         while (fp) {
 182                 struct sk_buff *xp = fp->next;
 183
 184                 sum_truesize += fp->truesize;
 185                 frag_kfree_skb(nf, f, fp);
 186                 fp = xp;
 187         }
 188         sum = sum_truesize + f->qsize;
 189         if (work)
 190                 *work -= sum;
 191         sub_frag_mem_limit(q, sum);
 192
 193         if (f->destructor)
 194                 f->destructor(q);
 195         kfree(q);
 196
 197 }
 198 EXPORT_SYMBOL(inet_frag_destroy);
 199
 200 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force)
 201 {
 202         struct inet_frag_queue *q;
 203         int work, evicted = 0;
 204
 205         if (!force) {
 206                 if (frag_mem_limit(nf) <= nf->high_thresh)
 207                         return 0;
 208         }
 209
 210         work = frag_mem_limit(nf) - nf->low_thresh;
 211         while (work > 0) {
 212                 spin_lock(&nf->lru_lock);
 213
 214                 if (list_empty(&nf->lru_list)) {
 215                         spin_unlock(&nf->lru_lock);
 216                         break;
 217                 }
 218
 219                 q = list_first_entry(&nf->lru_list,
 220                                 struct inet_frag_queue, lru_list);
 221                 atomic_inc(&q->refcnt);
 222                 /* Remove q from list to avoid several CPUs grabbing it */
 223                 list_del_init(&q->lru_list);
 224
 225                 spin_unlock(&nf->lru_lock);
 226
 227                 spin_lock(&q->lock);
 228                 if (!(q->last_in & INET_FRAG_COMPLETE))
 229                         inet_frag_kill(q, f);
 230                 spin_unlock(&q->lock);
 231
 232                 if (atomic_dec_and_test(&q->refcnt))
 233                         inet_frag_destroy(q, f, &work);
 234                 evicted++;
 235         }
 236
 237         return evicted;
 238 }
 239 EXPORT_SYMBOL(inet_frag_evictor);
 240
 241 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
 242                 struct inet_frag_queue *qp_in, struct inet_frags *f,
 243                 void *arg)
 244 {
 245         struct inet_frag_bucket *hb;
 246         struct inet_frag_queue *qp;
 247         unsigned int hash;
 248
 249         read_lock(&f->lock); /* Protects against hash rebuild */
 250         /*
 251          * While we stayed w/o the lock other CPU could update
 252          * the rnd seed, so we need to re-calculate the hash
 253          * chain. Fortunatelly the qp_in can be used to get one.
 254          */
 255         hash = f->hashfn(qp_in);
 256         hb = &f->hash[hash];
 257         spin_lock(&hb->chain_lock);
 258
 259 #ifdef CONFIG_SMP
 260         /* With SMP race we have to recheck hash table, because
 261          * such entry could be created on other cpu, while we
 262          * released the hash bucket lock.
 263          */
 264         hlist_for_each_entry(qp, &hb->chain, list) {
 265                 if (qp->net == nf && f->match(qp, arg)) {
 266                         atomic_inc(&qp->refcnt);
 267                         spin_unlock(&hb->chain_lock);
 268                         read_unlock(&f->lock);
 269                         qp_in->last_in |= INET_FRAG_COMPLETE;
 270                         inet_frag_put(qp_in, f);
 271                         return qp;
 272                 }
 273         }
 274 #endif
 275         qp = qp_in;
 276         if (!mod_timer(&qp->timer, jiffies + nf->timeout))
 277                 atomic_inc(&qp->refcnt);
 278
 279         atomic_inc(&qp->refcnt);
 280         hlist_add_head(&qp->list, &hb->chain);
 281         spin_unlock(&hb->chain_lock);
 282         read_unlock(&f->lock);
 283         inet_frag_lru_add(nf, qp);
 284         return qp;
 285 }
 286
 287 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
 288                 struct inet_frags *f, void *arg)
 289 {
 290         struct inet_frag_queue *q;
 291
 292         q = kzalloc(f->qsize, GFP_ATOMIC);
 293         if (q == NULL)
 294                 return NULL;
 295
 296         q->net = nf;
 297         f->constructor(q, arg);
 298         add_frag_mem_limit(q, f->qsize);
 299
 300         setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
 301         spin_lock_init(&q->lock);
 302         atomic_set(&q->refcnt, 1);
 303         INIT_LIST_HEAD(&q->lru_list);
 304
 305         return q;
 306 }
 307
 308 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
 309                 struct inet_frags *f, void *arg)
 310 {
 311         struct inet_frag_queue *q;
 312
 313         q = inet_frag_alloc(nf, f, arg);
 314         if (q == NULL)
 315                 return NULL;
 316
 317         return inet_frag_intern(nf, q, f, arg);
 318 }
 319
 320 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
 321                 struct inet_frags *f, void *key, unsigned int hash)
 322         __releases(&f->lock)
 323 {
 324         struct inet_frag_bucket *hb;
 325         struct inet_frag_queue *q;
 326         int depth = 0;
 327
 328         hb = &f->hash[hash];
 329
 330         spin_lock(&hb->chain_lock);
 331         hlist_for_each_entry(q, &hb->chain, list) {
 332                 if (q->net == nf && f->match(q, key)) {
 333                         atomic_inc(&q->refcnt);
 334                         spin_unlock(&hb->chain_lock);
 335                         read_unlock(&f->lock);
 336                         return q;
 337                 }
 338                 depth++;
 339         }
 340         spin_unlock(&hb->chain_lock);
 341         read_unlock(&f->lock);
 342
 343         if (depth <= INETFRAGS_MAXDEPTH)
 344                 return inet_frag_create(nf, f, key);
 345         else
 346                 return ERR_PTR(-ENOBUFS);
 347 }
 348 EXPORT_SYMBOL(inet_frag_find);
 349
 350 void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
 351                                    const char *prefix)
 352 {
 353         static const char msg[] = "inet_frag_find: Fragment hash bucket"
 354                 " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
 355                 ". Dropping fragment.\n";
 356
 357         if (PTR_ERR(q) == -ENOBUFS)
 358                 LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg);
 359 }
 360 EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);