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/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 }
  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)
 121 {
 122         if (f->skb_free)
 123                 f->skb_free(skb);
 124         kfree_skb(skb);
 125 }
 126
 127 void inet_frag_destroy(struct inet_frag_queue *q, struct inet_frags *f,
 128                                         int *work)
 129 {
 130         struct sk_buff *fp;
 131         struct netns_frags *nf;
 132         unsigned int sum, sum_truesize = 0;
 133
 134         WARN_ON(!(q->last_in & INET_FRAG_COMPLETE));
 135         WARN_ON(del_timer(&q->timer) != 0);
 136
 137         /* Release all fragment data. */
 138         fp = q->fragments;
 139         nf = q->net;
 140         while (fp) {
 141                 struct sk_buff *xp = fp->next;
 142
 143                 sum_truesize += fp->truesize;
 144                 frag_kfree_skb(nf, f, fp);
 145                 fp = xp;
 146         }
 147         sum = sum_truesize + f->qsize;
 148         if (work)
 149                 *work -= sum;
 150         sub_frag_mem_limit(q, sum);
 151
 152         if (f->destructor)
 153                 f->destructor(q);
 154         kfree(q);
 155
 156 }
 157 EXPORT_SYMBOL(inet_frag_destroy);
 158
 159 int inet_frag_evictor(struct netns_frags *nf, struct inet_frags *f, bool force)
 160 {
 161         struct inet_frag_queue *q;
 162         int work, evicted = 0;
 163
 164         if (!force) {
 165                 if (frag_mem_limit(nf) <= nf->high_thresh)
 166                         return 0;
 167         }
 168
 169         work = frag_mem_limit(nf) - nf->low_thresh;
 170         while (work > 0) {
 171                 read_lock(&f->lock);
 172                 if (list_empty(&nf->lru_list)) {
 173                         read_unlock(&f->lock);
 174                         break;
 175                 }
 176
 177                 q = list_first_entry(&nf->lru_list,
 178                                 struct inet_frag_queue, lru_list);
 179                 atomic_inc(&q->refcnt);
 180                 read_unlock(&f->lock);
 181
 182                 spin_lock(&q->lock);
 183                 if (!(q->last_in & INET_FRAG_COMPLETE))
 184                         inet_frag_kill(q, f);
 185                 spin_unlock(&q->lock);
 186
 187                 if (atomic_dec_and_test(&q->refcnt))
 188                         inet_frag_destroy(q, f, &work);
 189                 evicted++;
 190         }
 191
 192         return evicted;
 193 }
 194 EXPORT_SYMBOL(inet_frag_evictor);
 195
 196 static struct inet_frag_queue *inet_frag_intern(struct netns_frags *nf,
 197                 struct inet_frag_queue *qp_in, struct inet_frags *f,
 198                 void *arg)
 199 {
 200         struct inet_frag_queue *qp;
 201 #ifdef CONFIG_SMP
 202         struct hlist_node *n;
 203 #endif
 204         unsigned int hash;
 205
 206         write_lock(&f->lock);
 207         /*
 208          * While we stayed w/o the lock other CPU could update
 209          * the rnd seed, so we need to re-calculate the hash
 210          * chain. Fortunatelly the qp_in can be used to get one.
 211          */
 212         hash = f->hashfn(qp_in);
 213 #ifdef CONFIG_SMP
 214         /* With SMP race we have to recheck hash table, because
 215          * such entry could be created on other cpu, while we
 216          * promoted read lock to write lock.
 217          */
 218         hlist_for_each_entry(qp, n, &f->hash[hash], list) {
 219                 if (qp->net == nf && f->match(qp, arg)) {
 220                         atomic_inc(&qp->refcnt);
 221                         write_unlock(&f->lock);
 222                         qp_in->last_in |= INET_FRAG_COMPLETE;
 223                         inet_frag_put(qp_in, f);
 224                         return qp;
 225                 }
 226         }
 227 #endif
 228         qp = qp_in;
 229         if (!mod_timer(&qp->timer, jiffies + nf->timeout))
 230                 atomic_inc(&qp->refcnt);
 231
 232         atomic_inc(&qp->refcnt);
 233         hlist_add_head(&qp->list, &f->hash[hash]);
 234         list_add_tail(&qp->lru_list, &nf->lru_list);
 235         nf->nqueues++;
 236         write_unlock(&f->lock);
 237         return qp;
 238 }
 239
 240 static struct inet_frag_queue *inet_frag_alloc(struct netns_frags *nf,
 241                 struct inet_frags *f, void *arg)
 242 {
 243         struct inet_frag_queue *q;
 244
 245         q = kzalloc(f->qsize, GFP_ATOMIC);
 246         if (q == NULL)
 247                 return NULL;
 248
 249         q->net = nf;
 250         f->constructor(q, arg);
 251         add_frag_mem_limit(q, f->qsize);
 252
 253         setup_timer(&q->timer, f->frag_expire, (unsigned long)q);
 254         spin_lock_init(&q->lock);
 255         atomic_set(&q->refcnt, 1);
 256
 257         return q;
 258 }
 259
 260 static struct inet_frag_queue *inet_frag_create(struct netns_frags *nf,
 261                 struct inet_frags *f, void *arg)
 262 {
 263         struct inet_frag_queue *q;
 264
 265         q = inet_frag_alloc(nf, f, arg);
 266         if (q == NULL)
 267                 return NULL;
 268
 269         return inet_frag_intern(nf, q, f, arg);
 270 }
 271
 272 struct inet_frag_queue *inet_frag_find(struct netns_frags *nf,
 273                 struct inet_frags *f, void *key, unsigned int hash)
 274         __releases(&f->lock)
 275 {
 276         struct inet_frag_queue *q;
 277         struct hlist_node *n;
 278
 279         hlist_for_each_entry(q, n, &f->hash[hash], list) {
 280                 if (q->net == nf && f->match(q, key)) {
 281                         atomic_inc(&q->refcnt);
 282                         read_unlock(&f->lock);
 283                         return q;
 284                 }
 285         }
 286         read_unlock(&f->lock);
 287
 288         return inet_frag_create(nf, f, key);
 289 }
 290 EXPORT_SYMBOL(inet_frag_find);