linux/net/ipv4/inet_fragment.c
<<
>>
Prefs
   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 */
  32const 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};
  47EXPORT_SYMBOL(ip_frag_ecn_table);
  48
  49static 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
  84void 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}
 101EXPORT_SYMBOL(inet_frags_init);
 102
 103void 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}
 110EXPORT_SYMBOL(inet_frags_init_net);
 111
 112void inet_frags_fini(struct inet_frags *f)
 113{
 114        del_timer(&f->secret_timer);
 115}
 116EXPORT_SYMBOL(inet_frags_fini);
 117
 118void 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}
 126EXPORT_SYMBOL(inet_frags_exit_net);
 127
 128static inline void fq_unlink(struct inet_frag_queue *fq, struct inet_frags *f)
 129{
 130        struct inet_frag_bucket *hb;
 131        unsigned int hash;
 132
 133        read_lock(&f->lock);
 134        hash = f->hashfn(fq);
 135        hb = &f->hash[hash];
 136
 137        spin_lock(&hb->chain_lock);
 138        hlist_del(&fq->list);
 139        spin_unlock(&hb->chain_lock);
 140
 141        read_unlock(&f->lock);
 142        inet_frag_lru_del(fq);
 143}
 144
 145void inet_frag_kill(struct inet_frag_queue *fq, struct inet_frags *f)
 146{
 147        if (del_timer(&fq->timer))
 148                atomic_dec(&fq->refcnt);
 149
 150        if (!(fq->last_in & INET_FRAG_COMPLETE)) {
 151                fq_unlink(fq, f);
 152                atomic_dec(&fq->refcnt);
 153                fq->last_in |= INET_FRAG_COMPLETE;
 154        }
 155}
 156EXPORT_SYMBOL(inet_frag_kill);
 157
 158static inline void frag_kfree_skb(struct netns_frags *nf, struct inet_frags *f,
 159                struct sk_buff *skb)
 160{
 161        if (f->skb_free)
 162                f->skb_free(skb);
 163        kfree_skb(skb);
 164}
 165
 166void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
 167                                        int *work)
 168{
 169        struct sk_buff *fp;
 170        struct netns_frags *nf;
 171        unsigned int sum, sum_truesize = 0;
 172
 173        WARN_ON(!(q->last_in & INET_FRAG_COMPLETE));
 174        WARN_ON(del_timer(&q->timer) != 0);
 175
 176        /* Release all fragment data. */
 177        fp = q->fragments;
 178        nf = q->net;
 179        if (fp) {
 180                do {
 181                        struct sk_buff *xp = fp->next;
 182
 183                        sum_truesize += fp->truesize;
 184                        frag_kfree_skb(nf, f, fp);
 185                        fp = xp;
 186                } while (fp);
 187        } else {
 188                sum_truesize = inet_frag_rbtree_purge(&q->rb_fragments);
 189        }
 190        sum = sum_truesize + f->qsize;
 191        if (work)
 192                *work -= sum;
 193        sub_frag_mem_limit(q, sum);
 194
 195        if (f->destructor)
 196                f->destructor(q);
 197        kfree(q);
 198
 199}
 200EXPORT_SYMBOL(inet_frag_destroy);
 201
 202int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force)
 203{
 204        struct inet_frag_queue *q;
 205        int work, evicted = 0;
 206
 207        if (!force) {
 208                if (frag_mem_limit(nf) <= nf->high_thresh)
 209                        return 0;
 210        }
 211
 212        work = frag_mem_limit(nf) - nf->low_thresh;
 213        while (work > 0 || force) {
 214                spin_lock(&nf->lru_lock);
 215
 216                if (list_empty(&nf->lru_list)) {
 217                        spin_unlock(&nf->lru_lock);
 218                        break;
 219                }
 220
 221                q = list_first_entry(&nf->lru_list,
 222                                struct inet_frag_queue, lru_list);
 223                atomic_inc(&q->refcnt);
 224                /* Remove q from list to avoid several CPUs grabbing it */
 225                list_del_init(&q->lru_list);
 226
 227                spin_unlock(&nf->lru_lock);
 228
 229                spin_lock(&q->lock);
 230                if (!(q->last_in & INET_FRAG_COMPLETE))
 231                        inet_frag_kill(q, f);
 232                spin_unlock(&q->lock);
 233
 234                if (atomic_dec_and_test(&q->refcnt))
 235                        inet_frag_destroy(q, f, &work);
 236                evicted++;
 237        }
 238
 239        return evicted;
 240}
 241EXPORT_SYMBOL(inet_frag_evictor);
 242
 243static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
 244                struct inet_frag_queue *qp_in, struct inet_frags *f,
 245                void *arg)
 246{
 247        struct inet_frag_bucket *hb;
 248        struct inet_frag_queue *qp;
 249        unsigned int hash;
 250
 251        read_lock(&f->lock); /* Protects against hash rebuild */
 252        /*
 253         * While we stayed w/o the lock other CPU could update
 254         * the rnd seed, so we need to re-calculate the hash
 255         * chain. Fortunatelly the qp_in can be used to get one.
 256         */
 257        hash = f->hashfn(qp_in);
 258        hb = &f->hash[hash];
 259        spin_lock(&hb->chain_lock);
 260
 261#ifdef CONFIG_SMP
 262        /* With SMP race we have to recheck hash table, because
 263         * such entry could be created on other cpu, while we
 264         * released the hash bucket lock.
 265         */
 266        hlist_for_each_entry(qp, &hb->chain, list) {
 267                if (qp->net == nf && f->match(qp, arg)) {
 268                        atomic_inc(&qp->refcnt);
 269                        spin_unlock(&hb->chain_lock);
 270                        read_unlock(&f->lock);
 271                        qp_in->last_in |= INET_FRAG_COMPLETE;
 272                        inet_frag_put(qp_in, f);
 273                        return qp;
 274                }
 275        }
 276#endif
 277        qp = qp_in;
 278        if (!mod_timer(&qp->timer, jiffies + nf->timeout))
 279                atomic_inc(&qp->refcnt);
 280
 281        atomic_inc(&qp->refcnt);
 282        hlist_add_head(&qp->list, &hb->chain);
 283        inet_frag_lru_add(nf, qp);
 284        spin_unlock(&hb->chain_lock);
 285        read_unlock(&f->lock);
 286
 287        return qp;
 288}
 289
 290static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
 291                struct inet_frags *f, void *arg)
 292{
 293        struct inet_frag_queue *q;
 294
 295        q = kzalloc(f->qsize, GFP_ATOMIC);
 296        if (q == NULL)
 297                return NULL;
 298
 299        q->net = nf;
 300        f->constructor(q, arg);
 301        add_frag_mem_limit(q, f->qsize);
 302
 303        setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
 304        spin_lock_init(&q->lock);
 305        atomic_set(&q->refcnt, 1);
 306        INIT_LIST_HEAD(&q->lru_list);
 307
 308        return q;
 309}
 310
 311static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
 312                struct inet_frags *f, void *arg)
 313{
 314        struct inet_frag_queue *q;
 315
 316        q = inet_frag_alloc(nf, f, arg);
 317        if (q == NULL)
 318                return NULL;
 319
 320        return inet_frag_intern(nf, q, f, arg);
 321}
 322
 323struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
 324                struct inet_frags *f, void *key, unsigned int hash)
 325        __releases(&f->lock)
 326{
 327        struct inet_frag_bucket *hb;
 328        struct inet_frag_queue *q;
 329        int depth = 0;
 330
 331        hb = &f->hash[hash];
 332
 333        spin_lock(&hb->chain_lock);
 334        hlist_for_each_entry(q, &hb->chain, list) {
 335                if (q->net == nf && f->match(q, key)) {
 336                        atomic_inc(&q->refcnt);
 337                        spin_unlock(&hb->chain_lock);
 338                        read_unlock(&f->lock);
 339                        return q;
 340                }
 341                depth++;
 342        }
 343        spin_unlock(&hb->chain_lock);
 344        read_unlock(&f->lock);
 345
 346        if (depth <= INETFRAGS_MAXDEPTH)
 347                return inet_frag_create(nf, f, key);
 348        else
 349                return ERR_PTR(-ENOBUFS);
 350}
 351EXPORT_SYMBOL(inet_frag_find);
 352
 353void inet_frag_maybe_warn_overflow(struct inet_frag_queue *q,
 354                                   const char *prefix)
 355{
 356        static const char msg[] = "inet_frag_find: Fragment hash bucket"
 357                " list length grew over limit " __stringify(INETFRAGS_MAXDEPTH)
 358                ". Dropping fragment.\n";
 359
 360        if (PTR_ERR(q) == -ENOBUFS)
 361                LIMIT_NETDEBUG(KERN_WARNING "%s%s", prefix, msg);
 362}
 363EXPORT_SYMBOL(inet_frag_maybe_warn_overflow);
 364