Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * INET     An implementation of the TCP/IP protocol suite for the LINUX
0004  *      operating system.  INET is implemented using the  BSD Socket
0005  *      interface as the means of communication with the user level.
0006  *
0007  *      The IP fragmentation functionality.
0008  *
0009  * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG>
0010  *      Alan Cox <alan@lxorguk.ukuu.org.uk>
0011  *
0012  * Fixes:
0013  *      Alan Cox    :   Split from ip.c , see ip_input.c for history.
0014  *      David S. Miller :   Begin massive cleanup...
0015  *      Andi Kleen  :   Add sysctls.
0016  *      xxxx        :   Overlapfrag bug.
0017  *      Ultima          :       ip_expire() kernel panic.
0018  *      Bill Hawes  :   Frag accounting and evictor fixes.
0019  *      John McDonald   :   0 length frag bug.
0020  *      Alexey Kuznetsov:   SMP races, threading, cleanup.
0021  *      Patrick McHardy :   LRU queue of frag heads for evictor.
0022  */
0023 
0024 #define pr_fmt(fmt) "IPv4: " fmt
0025 
0026 #include <linux/compiler.h>
0027 #include <linux/module.h>
0028 #include <linux/types.h>
0029 #include <linux/mm.h>
0030 #include <linux/jiffies.h>
0031 #include <linux/skbuff.h>
0032 #include <linux/list.h>
0033 #include <linux/ip.h>
0034 #include <linux/icmp.h>
0035 #include <linux/netdevice.h>
0036 #include <linux/jhash.h>
0037 #include <linux/random.h>
0038 #include <linux/slab.h>
0039 #include <net/route.h>
0040 #include <net/dst.h>
0041 #include <net/sock.h>
0042 #include <net/ip.h>
0043 #include <net/icmp.h>
0044 #include <net/checksum.h>
0045 #include <net/inetpeer.h>
0046 #include <net/inet_frag.h>
0047 #include <linux/tcp.h>
0048 #include <linux/udp.h>
0049 #include <linux/inet.h>
0050 #include <linux/netfilter_ipv4.h>
0051 #include <net/inet_ecn.h>
0052 #include <net/l3mdev.h>
0053 
0054 /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6
0055  * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c
0056  * as well. Or notify me, at least. --ANK
0057  */
0058 static const char ip_frag_cache_name[] = "ip4-frags";
0059 
0060 /* Describe an entry in the "incomplete datagrams" queue. */
0061 struct ipq {
0062     struct inet_frag_queue q;
0063 
0064     u8      ecn; /* RFC3168 support */
0065     u16     max_df_size; /* largest frag with DF set seen */
0066     int             iif;
0067     unsigned int    rid;
0068     struct inet_peer *peer;
0069 };
0070 
0071 static u8 ip4_frag_ecn(u8 tos)
0072 {
0073     return 1 << (tos & INET_ECN_MASK);
0074 }
0075 
0076 static struct inet_frags ip4_frags;
0077 
0078 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
0079              struct sk_buff *prev_tail, struct net_device *dev);
0080 
0081 
0082 static void ip4_frag_init(struct inet_frag_queue *q, const void *a)
0083 {
0084     struct ipq *qp = container_of(q, struct ipq, q);
0085     struct net *net = q->fqdir->net;
0086 
0087     const struct frag_v4_compare_key *key = a;
0088 
0089     q->key.v4 = *key;
0090     qp->ecn = 0;
0091     qp->peer = q->fqdir->max_dist ?
0092         inet_getpeer_v4(net->ipv4.peers, key->saddr, key->vif, 1) :
0093         NULL;
0094 }
0095 
0096 static void ip4_frag_free(struct inet_frag_queue *q)
0097 {
0098     struct ipq *qp;
0099 
0100     qp = container_of(q, struct ipq, q);
0101     if (qp->peer)
0102         inet_putpeer(qp->peer);
0103 }
0104 
0105 
0106 /* Destruction primitives. */
0107 
0108 static void ipq_put(struct ipq *ipq)
0109 {
0110     inet_frag_put(&ipq->q);
0111 }
0112 
0113 /* Kill ipq entry. It is not destroyed immediately,
0114  * because caller (and someone more) holds reference count.
0115  */
0116 static void ipq_kill(struct ipq *ipq)
0117 {
0118     inet_frag_kill(&ipq->q);
0119 }
0120 
0121 static bool frag_expire_skip_icmp(u32 user)
0122 {
0123     return user == IP_DEFRAG_AF_PACKET ||
0124            ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_IN,
0125                      __IP_DEFRAG_CONNTRACK_IN_END) ||
0126            ip_defrag_user_in_between(user, IP_DEFRAG_CONNTRACK_BRIDGE_IN,
0127                      __IP_DEFRAG_CONNTRACK_BRIDGE_IN);
0128 }
0129 
0130 /*
0131  * Oops, a fragment queue timed out.  Kill it and send an ICMP reply.
0132  */
0133 static void ip_expire(struct timer_list *t)
0134 {
0135     struct inet_frag_queue *frag = from_timer(frag, t, timer);
0136     const struct iphdr *iph;
0137     struct sk_buff *head = NULL;
0138     struct net *net;
0139     struct ipq *qp;
0140     int err;
0141 
0142     qp = container_of(frag, struct ipq, q);
0143     net = qp->q.fqdir->net;
0144 
0145     rcu_read_lock();
0146 
0147     /* Paired with WRITE_ONCE() in fqdir_pre_exit(). */
0148     if (READ_ONCE(qp->q.fqdir->dead))
0149         goto out_rcu_unlock;
0150 
0151     spin_lock(&qp->q.lock);
0152 
0153     if (qp->q.flags & INET_FRAG_COMPLETE)
0154         goto out;
0155 
0156     ipq_kill(qp);
0157     __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
0158     __IP_INC_STATS(net, IPSTATS_MIB_REASMTIMEOUT);
0159 
0160     if (!(qp->q.flags & INET_FRAG_FIRST_IN))
0161         goto out;
0162 
0163     /* sk_buff::dev and sk_buff::rbnode are unionized. So we
0164      * pull the head out of the tree in order to be able to
0165      * deal with head->dev.
0166      */
0167     head = inet_frag_pull_head(&qp->q);
0168     if (!head)
0169         goto out;
0170     head->dev = dev_get_by_index_rcu(net, qp->iif);
0171     if (!head->dev)
0172         goto out;
0173 
0174 
0175     /* skb has no dst, perform route lookup again */
0176     iph = ip_hdr(head);
0177     err = ip_route_input_noref(head, iph->daddr, iph->saddr,
0178                        iph->tos, head->dev);
0179     if (err)
0180         goto out;
0181 
0182     /* Only an end host needs to send an ICMP
0183      * "Fragment Reassembly Timeout" message, per RFC792.
0184      */
0185     if (frag_expire_skip_icmp(qp->q.key.v4.user) &&
0186         (skb_rtable(head)->rt_type != RTN_LOCAL))
0187         goto out;
0188 
0189     spin_unlock(&qp->q.lock);
0190     icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0);
0191     goto out_rcu_unlock;
0192 
0193 out:
0194     spin_unlock(&qp->q.lock);
0195 out_rcu_unlock:
0196     rcu_read_unlock();
0197     kfree_skb(head);
0198     ipq_put(qp);
0199 }
0200 
0201 /* Find the correct entry in the "incomplete datagrams" queue for
0202  * this IP datagram, and create new one, if nothing is found.
0203  */
0204 static struct ipq *ip_find(struct net *net, struct iphdr *iph,
0205                u32 user, int vif)
0206 {
0207     struct frag_v4_compare_key key = {
0208         .saddr = iph->saddr,
0209         .daddr = iph->daddr,
0210         .user = user,
0211         .vif = vif,
0212         .id = iph->id,
0213         .protocol = iph->protocol,
0214     };
0215     struct inet_frag_queue *q;
0216 
0217     q = inet_frag_find(net->ipv4.fqdir, &key);
0218     if (!q)
0219         return NULL;
0220 
0221     return container_of(q, struct ipq, q);
0222 }
0223 
0224 /* Is the fragment too far ahead to be part of ipq? */
0225 static int ip_frag_too_far(struct ipq *qp)
0226 {
0227     struct inet_peer *peer = qp->peer;
0228     unsigned int max = qp->q.fqdir->max_dist;
0229     unsigned int start, end;
0230 
0231     int rc;
0232 
0233     if (!peer || !max)
0234         return 0;
0235 
0236     start = qp->rid;
0237     end = atomic_inc_return(&peer->rid);
0238     qp->rid = end;
0239 
0240     rc = qp->q.fragments_tail && (end - start) > max;
0241 
0242     if (rc)
0243         __IP_INC_STATS(qp->q.fqdir->net, IPSTATS_MIB_REASMFAILS);
0244 
0245     return rc;
0246 }
0247 
0248 static int ip_frag_reinit(struct ipq *qp)
0249 {
0250     unsigned int sum_truesize = 0;
0251 
0252     if (!mod_timer(&qp->q.timer, jiffies + qp->q.fqdir->timeout)) {
0253         refcount_inc(&qp->q.refcnt);
0254         return -ETIMEDOUT;
0255     }
0256 
0257     sum_truesize = inet_frag_rbtree_purge(&qp->q.rb_fragments);
0258     sub_frag_mem_limit(qp->q.fqdir, sum_truesize);
0259 
0260     qp->q.flags = 0;
0261     qp->q.len = 0;
0262     qp->q.meat = 0;
0263     qp->q.rb_fragments = RB_ROOT;
0264     qp->q.fragments_tail = NULL;
0265     qp->q.last_run_head = NULL;
0266     qp->iif = 0;
0267     qp->ecn = 0;
0268 
0269     return 0;
0270 }
0271 
0272 /* Add new segment to existing queue. */
0273 static int ip_frag_queue(struct ipq *qp, struct sk_buff *skb)
0274 {
0275     struct net *net = qp->q.fqdir->net;
0276     int ihl, end, flags, offset;
0277     struct sk_buff *prev_tail;
0278     struct net_device *dev;
0279     unsigned int fragsize;
0280     int err = -ENOENT;
0281     u8 ecn;
0282 
0283     if (qp->q.flags & INET_FRAG_COMPLETE)
0284         goto err;
0285 
0286     if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) &&
0287         unlikely(ip_frag_too_far(qp)) &&
0288         unlikely(err = ip_frag_reinit(qp))) {
0289         ipq_kill(qp);
0290         goto err;
0291     }
0292 
0293     ecn = ip4_frag_ecn(ip_hdr(skb)->tos);
0294     offset = ntohs(ip_hdr(skb)->frag_off);
0295     flags = offset & ~IP_OFFSET;
0296     offset &= IP_OFFSET;
0297     offset <<= 3;       /* offset is in 8-byte chunks */
0298     ihl = ip_hdrlen(skb);
0299 
0300     /* Determine the position of this fragment. */
0301     end = offset + skb->len - skb_network_offset(skb) - ihl;
0302     err = -EINVAL;
0303 
0304     /* Is this the final fragment? */
0305     if ((flags & IP_MF) == 0) {
0306         /* If we already have some bits beyond end
0307          * or have different end, the segment is corrupted.
0308          */
0309         if (end < qp->q.len ||
0310             ((qp->q.flags & INET_FRAG_LAST_IN) && end != qp->q.len))
0311             goto discard_qp;
0312         qp->q.flags |= INET_FRAG_LAST_IN;
0313         qp->q.len = end;
0314     } else {
0315         if (end&7) {
0316             end &= ~7;
0317             if (skb->ip_summed != CHECKSUM_UNNECESSARY)
0318                 skb->ip_summed = CHECKSUM_NONE;
0319         }
0320         if (end > qp->q.len) {
0321             /* Some bits beyond end -> corruption. */
0322             if (qp->q.flags & INET_FRAG_LAST_IN)
0323                 goto discard_qp;
0324             qp->q.len = end;
0325         }
0326     }
0327     if (end == offset)
0328         goto discard_qp;
0329 
0330     err = -ENOMEM;
0331     if (!pskb_pull(skb, skb_network_offset(skb) + ihl))
0332         goto discard_qp;
0333 
0334     err = pskb_trim_rcsum(skb, end - offset);
0335     if (err)
0336         goto discard_qp;
0337 
0338     /* Note : skb->rbnode and skb->dev share the same location. */
0339     dev = skb->dev;
0340     /* Makes sure compiler wont do silly aliasing games */
0341     barrier();
0342 
0343     prev_tail = qp->q.fragments_tail;
0344     err = inet_frag_queue_insert(&qp->q, skb, offset, end);
0345     if (err)
0346         goto insert_error;
0347 
0348     if (dev)
0349         qp->iif = dev->ifindex;
0350 
0351     qp->q.stamp = skb->tstamp;
0352     qp->q.mono_delivery_time = skb->mono_delivery_time;
0353     qp->q.meat += skb->len;
0354     qp->ecn |= ecn;
0355     add_frag_mem_limit(qp->q.fqdir, skb->truesize);
0356     if (offset == 0)
0357         qp->q.flags |= INET_FRAG_FIRST_IN;
0358 
0359     fragsize = skb->len + ihl;
0360 
0361     if (fragsize > qp->q.max_size)
0362         qp->q.max_size = fragsize;
0363 
0364     if (ip_hdr(skb)->frag_off & htons(IP_DF) &&
0365         fragsize > qp->max_df_size)
0366         qp->max_df_size = fragsize;
0367 
0368     if (qp->q.flags == (INET_FRAG_FIRST_IN | INET_FRAG_LAST_IN) &&
0369         qp->q.meat == qp->q.len) {
0370         unsigned long orefdst = skb->_skb_refdst;
0371 
0372         skb->_skb_refdst = 0UL;
0373         err = ip_frag_reasm(qp, skb, prev_tail, dev);
0374         skb->_skb_refdst = orefdst;
0375         if (err)
0376             inet_frag_kill(&qp->q);
0377         return err;
0378     }
0379 
0380     skb_dst_drop(skb);
0381     return -EINPROGRESS;
0382 
0383 insert_error:
0384     if (err == IPFRAG_DUP) {
0385         kfree_skb(skb);
0386         return -EINVAL;
0387     }
0388     err = -EINVAL;
0389     __IP_INC_STATS(net, IPSTATS_MIB_REASM_OVERLAPS);
0390 discard_qp:
0391     inet_frag_kill(&qp->q);
0392     __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
0393 err:
0394     kfree_skb(skb);
0395     return err;
0396 }
0397 
0398 static bool ip_frag_coalesce_ok(const struct ipq *qp)
0399 {
0400     return qp->q.key.v4.user == IP_DEFRAG_LOCAL_DELIVER;
0401 }
0402 
0403 /* Build a new IP datagram from all its fragments. */
0404 static int ip_frag_reasm(struct ipq *qp, struct sk_buff *skb,
0405              struct sk_buff *prev_tail, struct net_device *dev)
0406 {
0407     struct net *net = qp->q.fqdir->net;
0408     struct iphdr *iph;
0409     void *reasm_data;
0410     int len, err;
0411     u8 ecn;
0412 
0413     ipq_kill(qp);
0414 
0415     ecn = ip_frag_ecn_table[qp->ecn];
0416     if (unlikely(ecn == 0xff)) {
0417         err = -EINVAL;
0418         goto out_fail;
0419     }
0420 
0421     /* Make the one we just received the head. */
0422     reasm_data = inet_frag_reasm_prepare(&qp->q, skb, prev_tail);
0423     if (!reasm_data)
0424         goto out_nomem;
0425 
0426     len = ip_hdrlen(skb) + qp->q.len;
0427     err = -E2BIG;
0428     if (len > 65535)
0429         goto out_oversize;
0430 
0431     inet_frag_reasm_finish(&qp->q, skb, reasm_data,
0432                    ip_frag_coalesce_ok(qp));
0433 
0434     skb->dev = dev;
0435     IPCB(skb)->frag_max_size = max(qp->max_df_size, qp->q.max_size);
0436 
0437     iph = ip_hdr(skb);
0438     iph->tot_len = htons(len);
0439     iph->tos |= ecn;
0440 
0441     /* When we set IP_DF on a refragmented skb we must also force a
0442      * call to ip_fragment to avoid forwarding a DF-skb of size s while
0443      * original sender only sent fragments of size f (where f < s).
0444      *
0445      * We only set DF/IPSKB_FRAG_PMTU if such DF fragment was the largest
0446      * frag seen to avoid sending tiny DF-fragments in case skb was built
0447      * from one very small df-fragment and one large non-df frag.
0448      */
0449     if (qp->max_df_size == qp->q.max_size) {
0450         IPCB(skb)->flags |= IPSKB_FRAG_PMTU;
0451         iph->frag_off = htons(IP_DF);
0452     } else {
0453         iph->frag_off = 0;
0454     }
0455 
0456     ip_send_check(iph);
0457 
0458     __IP_INC_STATS(net, IPSTATS_MIB_REASMOKS);
0459     qp->q.rb_fragments = RB_ROOT;
0460     qp->q.fragments_tail = NULL;
0461     qp->q.last_run_head = NULL;
0462     return 0;
0463 
0464 out_nomem:
0465     net_dbg_ratelimited("queue_glue: no memory for gluing queue %p\n", qp);
0466     err = -ENOMEM;
0467     goto out_fail;
0468 out_oversize:
0469     net_info_ratelimited("Oversized IP packet from %pI4\n", &qp->q.key.v4.saddr);
0470 out_fail:
0471     __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
0472     return err;
0473 }
0474 
0475 /* Process an incoming IP datagram fragment. */
0476 int ip_defrag(struct net *net, struct sk_buff *skb, u32 user)
0477 {
0478     struct net_device *dev = skb->dev ? : skb_dst(skb)->dev;
0479     int vif = l3mdev_master_ifindex_rcu(dev);
0480     struct ipq *qp;
0481 
0482     __IP_INC_STATS(net, IPSTATS_MIB_REASMREQDS);
0483     skb_orphan(skb);
0484 
0485     /* Lookup (or create) queue header */
0486     qp = ip_find(net, ip_hdr(skb), user, vif);
0487     if (qp) {
0488         int ret;
0489 
0490         spin_lock(&qp->q.lock);
0491 
0492         ret = ip_frag_queue(qp, skb);
0493 
0494         spin_unlock(&qp->q.lock);
0495         ipq_put(qp);
0496         return ret;
0497     }
0498 
0499     __IP_INC_STATS(net, IPSTATS_MIB_REASMFAILS);
0500     kfree_skb(skb);
0501     return -ENOMEM;
0502 }
0503 EXPORT_SYMBOL(ip_defrag);
0504 
0505 struct sk_buff *ip_check_defrag(struct net *net, struct sk_buff *skb, u32 user)
0506 {
0507     struct iphdr iph;
0508     int netoff;
0509     u32 len;
0510 
0511     if (skb->protocol != htons(ETH_P_IP))
0512         return skb;
0513 
0514     netoff = skb_network_offset(skb);
0515 
0516     if (skb_copy_bits(skb, netoff, &iph, sizeof(iph)) < 0)
0517         return skb;
0518 
0519     if (iph.ihl < 5 || iph.version != 4)
0520         return skb;
0521 
0522     len = ntohs(iph.tot_len);
0523     if (skb->len < netoff + len || len < (iph.ihl * 4))
0524         return skb;
0525 
0526     if (ip_is_fragment(&iph)) {
0527         skb = skb_share_check(skb, GFP_ATOMIC);
0528         if (skb) {
0529             if (!pskb_may_pull(skb, netoff + iph.ihl * 4)) {
0530                 kfree_skb(skb);
0531                 return NULL;
0532             }
0533             if (pskb_trim_rcsum(skb, netoff + len)) {
0534                 kfree_skb(skb);
0535                 return NULL;
0536             }
0537             memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
0538             if (ip_defrag(net, skb, user))
0539                 return NULL;
0540             skb_clear_hash(skb);
0541         }
0542     }
0543     return skb;
0544 }
0545 EXPORT_SYMBOL(ip_check_defrag);
0546 
0547 #ifdef CONFIG_SYSCTL
0548 static int dist_min;
0549 
0550 static struct ctl_table ip4_frags_ns_ctl_table[] = {
0551     {
0552         .procname   = "ipfrag_high_thresh",
0553         .maxlen     = sizeof(unsigned long),
0554         .mode       = 0644,
0555         .proc_handler   = proc_doulongvec_minmax,
0556     },
0557     {
0558         .procname   = "ipfrag_low_thresh",
0559         .maxlen     = sizeof(unsigned long),
0560         .mode       = 0644,
0561         .proc_handler   = proc_doulongvec_minmax,
0562     },
0563     {
0564         .procname   = "ipfrag_time",
0565         .maxlen     = sizeof(int),
0566         .mode       = 0644,
0567         .proc_handler   = proc_dointvec_jiffies,
0568     },
0569     {
0570         .procname   = "ipfrag_max_dist",
0571         .maxlen     = sizeof(int),
0572         .mode       = 0644,
0573         .proc_handler   = proc_dointvec_minmax,
0574         .extra1     = &dist_min,
0575     },
0576     { }
0577 };
0578 
0579 /* secret interval has been deprecated */
0580 static int ip4_frags_secret_interval_unused;
0581 static struct ctl_table ip4_frags_ctl_table[] = {
0582     {
0583         .procname   = "ipfrag_secret_interval",
0584         .data       = &ip4_frags_secret_interval_unused,
0585         .maxlen     = sizeof(int),
0586         .mode       = 0644,
0587         .proc_handler   = proc_dointvec_jiffies,
0588     },
0589     { }
0590 };
0591 
0592 static int __net_init ip4_frags_ns_ctl_register(struct net *net)
0593 {
0594     struct ctl_table *table;
0595     struct ctl_table_header *hdr;
0596 
0597     table = ip4_frags_ns_ctl_table;
0598     if (!net_eq(net, &init_net)) {
0599         table = kmemdup(table, sizeof(ip4_frags_ns_ctl_table), GFP_KERNEL);
0600         if (!table)
0601             goto err_alloc;
0602 
0603     }
0604     table[0].data   = &net->ipv4.fqdir->high_thresh;
0605     table[0].extra1 = &net->ipv4.fqdir->low_thresh;
0606     table[1].data   = &net->ipv4.fqdir->low_thresh;
0607     table[1].extra2 = &net->ipv4.fqdir->high_thresh;
0608     table[2].data   = &net->ipv4.fqdir->timeout;
0609     table[3].data   = &net->ipv4.fqdir->max_dist;
0610 
0611     hdr = register_net_sysctl(net, "net/ipv4", table);
0612     if (!hdr)
0613         goto err_reg;
0614 
0615     net->ipv4.frags_hdr = hdr;
0616     return 0;
0617 
0618 err_reg:
0619     if (!net_eq(net, &init_net))
0620         kfree(table);
0621 err_alloc:
0622     return -ENOMEM;
0623 }
0624 
0625 static void __net_exit ip4_frags_ns_ctl_unregister(struct net *net)
0626 {
0627     struct ctl_table *table;
0628 
0629     table = net->ipv4.frags_hdr->ctl_table_arg;
0630     unregister_net_sysctl_table(net->ipv4.frags_hdr);
0631     kfree(table);
0632 }
0633 
0634 static void __init ip4_frags_ctl_register(void)
0635 {
0636     register_net_sysctl(&init_net, "net/ipv4", ip4_frags_ctl_table);
0637 }
0638 #else
0639 static int ip4_frags_ns_ctl_register(struct net *net)
0640 {
0641     return 0;
0642 }
0643 
0644 static void ip4_frags_ns_ctl_unregister(struct net *net)
0645 {
0646 }
0647 
0648 static void __init ip4_frags_ctl_register(void)
0649 {
0650 }
0651 #endif
0652 
0653 static int __net_init ipv4_frags_init_net(struct net *net)
0654 {
0655     int res;
0656 
0657     res = fqdir_init(&net->ipv4.fqdir, &ip4_frags, net);
0658     if (res < 0)
0659         return res;
0660     /* Fragment cache limits.
0661      *
0662      * The fragment memory accounting code, (tries to) account for
0663      * the real memory usage, by measuring both the size of frag
0664      * queue struct (inet_frag_queue (ipv4:ipq/ipv6:frag_queue))
0665      * and the SKB's truesize.
0666      *
0667      * A 64K fragment consumes 129736 bytes (44*2944)+200
0668      * (1500 truesize == 2944, sizeof(struct ipq) == 200)
0669      *
0670      * We will commit 4MB at one time. Should we cross that limit
0671      * we will prune down to 3MB, making room for approx 8 big 64K
0672      * fragments 8x128k.
0673      */
0674     net->ipv4.fqdir->high_thresh = 4 * 1024 * 1024;
0675     net->ipv4.fqdir->low_thresh  = 3 * 1024 * 1024;
0676     /*
0677      * Important NOTE! Fragment queue must be destroyed before MSL expires.
0678      * RFC791 is wrong proposing to prolongate timer each fragment arrival
0679      * by TTL.
0680      */
0681     net->ipv4.fqdir->timeout = IP_FRAG_TIME;
0682 
0683     net->ipv4.fqdir->max_dist = 64;
0684 
0685     res = ip4_frags_ns_ctl_register(net);
0686     if (res < 0)
0687         fqdir_exit(net->ipv4.fqdir);
0688     return res;
0689 }
0690 
0691 static void __net_exit ipv4_frags_pre_exit_net(struct net *net)
0692 {
0693     fqdir_pre_exit(net->ipv4.fqdir);
0694 }
0695 
0696 static void __net_exit ipv4_frags_exit_net(struct net *net)
0697 {
0698     ip4_frags_ns_ctl_unregister(net);
0699     fqdir_exit(net->ipv4.fqdir);
0700 }
0701 
0702 static struct pernet_operations ip4_frags_ops = {
0703     .init       = ipv4_frags_init_net,
0704     .pre_exit   = ipv4_frags_pre_exit_net,
0705     .exit       = ipv4_frags_exit_net,
0706 };
0707 
0708 
0709 static u32 ip4_key_hashfn(const void *data, u32 len, u32 seed)
0710 {
0711     return jhash2(data,
0712               sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
0713 }
0714 
0715 static u32 ip4_obj_hashfn(const void *data, u32 len, u32 seed)
0716 {
0717     const struct inet_frag_queue *fq = data;
0718 
0719     return jhash2((const u32 *)&fq->key.v4,
0720               sizeof(struct frag_v4_compare_key) / sizeof(u32), seed);
0721 }
0722 
0723 static int ip4_obj_cmpfn(struct rhashtable_compare_arg *arg, const void *ptr)
0724 {
0725     const struct frag_v4_compare_key *key = arg->key;
0726     const struct inet_frag_queue *fq = ptr;
0727 
0728     return !!memcmp(&fq->key, key, sizeof(*key));
0729 }
0730 
0731 static const struct rhashtable_params ip4_rhash_params = {
0732     .head_offset        = offsetof(struct inet_frag_queue, node),
0733     .key_offset     = offsetof(struct inet_frag_queue, key),
0734     .key_len        = sizeof(struct frag_v4_compare_key),
0735     .hashfn         = ip4_key_hashfn,
0736     .obj_hashfn     = ip4_obj_hashfn,
0737     .obj_cmpfn      = ip4_obj_cmpfn,
0738     .automatic_shrinking    = true,
0739 };
0740 
0741 void __init ipfrag_init(void)
0742 {
0743     ip4_frags.constructor = ip4_frag_init;
0744     ip4_frags.destructor = ip4_frag_free;
0745     ip4_frags.qsize = sizeof(struct ipq);
0746     ip4_frags.frag_expire = ip_expire;
0747     ip4_frags.frags_cache_name = ip_frag_cache_name;
0748     ip4_frags.rhash_params = ip4_rhash_params;
0749     if (inet_frags_init(&ip4_frags))
0750         panic("IP: failed to allocate ip4_frags cache\n");
0751     ip4_frags_ctl_register();
0752     register_pernet_subsys(&ip4_frags_ops);
0753 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!