Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * INET An implementation of the TCP/IP protocol suite for the LINUX |
| 3 | * operating system. INET is implemented using the BSD Socket |
| 4 | * interface as the means of communication with the user level. |
| 5 | * |
| 6 | * The IP fragmentation functionality. |
| 7 | * |
| 8 | * Version: $Id: ip_fragment.c,v 1.59 2002/01/12 07:54:56 davem Exp $ |
| 9 | * |
| 10 | * Authors: Fred N. van Kempen <waltje@uWalt.NL.Mugnet.ORG> |
| 11 | * Alan Cox <Alan.Cox@linux.org> |
| 12 | * |
| 13 | * Fixes: |
| 14 | * Alan Cox : Split from ip.c , see ip_input.c for history. |
| 15 | * David S. Miller : Begin massive cleanup... |
| 16 | * Andi Kleen : Add sysctls. |
| 17 | * xxxx : Overlapfrag bug. |
| 18 | * Ultima : ip_expire() kernel panic. |
| 19 | * Bill Hawes : Frag accounting and evictor fixes. |
| 20 | * John McDonald : 0 length frag bug. |
| 21 | * Alexey Kuznetsov: SMP races, threading, cleanup. |
| 22 | * Patrick McHardy : LRU queue of frag heads for evictor. |
| 23 | */ |
| 24 | |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 25 | #include <linux/compiler.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include <linux/module.h> |
| 27 | #include <linux/types.h> |
| 28 | #include <linux/mm.h> |
| 29 | #include <linux/jiffies.h> |
| 30 | #include <linux/skbuff.h> |
| 31 | #include <linux/list.h> |
| 32 | #include <linux/ip.h> |
| 33 | #include <linux/icmp.h> |
| 34 | #include <linux/netdevice.h> |
| 35 | #include <linux/jhash.h> |
| 36 | #include <linux/random.h> |
| 37 | #include <net/sock.h> |
| 38 | #include <net/ip.h> |
| 39 | #include <net/icmp.h> |
| 40 | #include <net/checksum.h> |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 41 | #include <net/inetpeer.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | #include <linux/tcp.h> |
| 43 | #include <linux/udp.h> |
| 44 | #include <linux/inet.h> |
| 45 | #include <linux/netfilter_ipv4.h> |
| 46 | |
| 47 | /* NOTE. Logic of IP defragmentation is parallel to corresponding IPv6 |
| 48 | * code now. If you change something here, _PLEASE_ update ipv6/reassembly.c |
| 49 | * as well. Or notify me, at least. --ANK |
| 50 | */ |
| 51 | |
| 52 | /* Fragment cache limits. We will commit 256K at one time. Should we |
| 53 | * cross that limit we will prune down to 192K. This should cope with |
| 54 | * even the most extreme cases without allowing an attacker to measurably |
| 55 | * harm machine performance. |
| 56 | */ |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 57 | int sysctl_ipfrag_high_thresh __read_mostly = 256*1024; |
| 58 | int sysctl_ipfrag_low_thresh __read_mostly = 192*1024; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 60 | int sysctl_ipfrag_max_dist __read_mostly = 64; |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 61 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 62 | /* Important NOTE! Fragment queue must be destroyed before MSL expires. |
| 63 | * RFC791 is wrong proposing to prolongate timer each fragment arrival by TTL. |
| 64 | */ |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 65 | int sysctl_ipfrag_time __read_mostly = IP_FRAG_TIME; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 66 | |
| 67 | struct ipfrag_skb_cb |
| 68 | { |
| 69 | struct inet_skb_parm h; |
| 70 | int offset; |
| 71 | }; |
| 72 | |
| 73 | #define FRAG_CB(skb) ((struct ipfrag_skb_cb*)((skb)->cb)) |
| 74 | |
| 75 | /* Describe an entry in the "incomplete datagrams" queue. */ |
| 76 | struct ipq { |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 77 | struct hlist_node list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | struct list_head lru_list; /* lru list member */ |
| 79 | u32 user; |
Al Viro | 1827777 | 2006-09-26 22:19:02 -0700 | [diff] [blame] | 80 | __be32 saddr; |
| 81 | __be32 daddr; |
| 82 | __be16 id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | u8 protocol; |
| 84 | u8 last_in; |
| 85 | #define COMPLETE 4 |
| 86 | #define FIRST_IN 2 |
| 87 | #define LAST_IN 1 |
| 88 | |
| 89 | struct sk_buff *fragments; /* linked list of received fragments */ |
| 90 | int len; /* total length of original datagram */ |
| 91 | int meat; |
| 92 | spinlock_t lock; |
| 93 | atomic_t refcnt; |
| 94 | struct timer_list timer; /* when will this queue expire? */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | struct timeval stamp; |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 96 | int iif; |
| 97 | unsigned int rid; |
| 98 | struct inet_peer *peer; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | }; |
| 100 | |
| 101 | /* Hash table. */ |
| 102 | |
| 103 | #define IPQ_HASHSZ 64 |
| 104 | |
| 105 | /* Per-bucket lock is easy to add now. */ |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 106 | static struct hlist_head ipq_hash[IPQ_HASHSZ]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 107 | static DEFINE_RWLOCK(ipfrag_lock); |
| 108 | static u32 ipfrag_hash_rnd; |
| 109 | static LIST_HEAD(ipq_lru_list); |
| 110 | int ip_frag_nqueues = 0; |
| 111 | |
| 112 | static __inline__ void __ipq_unlink(struct ipq *qp) |
| 113 | { |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 114 | hlist_del(&qp->list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | list_del(&qp->lru_list); |
| 116 | ip_frag_nqueues--; |
| 117 | } |
| 118 | |
| 119 | static __inline__ void ipq_unlink(struct ipq *ipq) |
| 120 | { |
| 121 | write_lock(&ipfrag_lock); |
| 122 | __ipq_unlink(ipq); |
| 123 | write_unlock(&ipfrag_lock); |
| 124 | } |
| 125 | |
Al Viro | 1827777 | 2006-09-26 22:19:02 -0700 | [diff] [blame] | 126 | static unsigned int ipqhashfn(__be16 id, __be32 saddr, __be32 daddr, u8 prot) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | { |
Al Viro | 1827777 | 2006-09-26 22:19:02 -0700 | [diff] [blame] | 128 | return jhash_3words((__force u32)id << 16 | prot, |
| 129 | (__force u32)saddr, (__force u32)daddr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 130 | ipfrag_hash_rnd) & (IPQ_HASHSZ - 1); |
| 131 | } |
| 132 | |
| 133 | static struct timer_list ipfrag_secret_timer; |
Brian Haley | ab32ea5 | 2006-09-22 14:15:41 -0700 | [diff] [blame] | 134 | int sysctl_ipfrag_secret_interval __read_mostly = 10 * 60 * HZ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | |
| 136 | static void ipfrag_secret_rebuild(unsigned long dummy) |
| 137 | { |
| 138 | unsigned long now = jiffies; |
| 139 | int i; |
| 140 | |
| 141 | write_lock(&ipfrag_lock); |
| 142 | get_random_bytes(&ipfrag_hash_rnd, sizeof(u32)); |
| 143 | for (i = 0; i < IPQ_HASHSZ; i++) { |
| 144 | struct ipq *q; |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 145 | struct hlist_node *p, *n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 147 | hlist_for_each_entry_safe(q, p, n, &ipq_hash[i], list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | unsigned int hval = ipqhashfn(q->id, q->saddr, |
| 149 | q->daddr, q->protocol); |
| 150 | |
| 151 | if (hval != i) { |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 152 | hlist_del(&q->list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | |
| 154 | /* Relink to new hash chain. */ |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 155 | hlist_add_head(&q->list, &ipq_hash[hval]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 156 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | } |
| 158 | } |
| 159 | write_unlock(&ipfrag_lock); |
| 160 | |
| 161 | mod_timer(&ipfrag_secret_timer, now + sysctl_ipfrag_secret_interval); |
| 162 | } |
| 163 | |
| 164 | atomic_t ip_frag_mem = ATOMIC_INIT(0); /* Memory used for fragments */ |
| 165 | |
| 166 | /* Memory Tracking Functions. */ |
| 167 | static __inline__ void frag_kfree_skb(struct sk_buff *skb, int *work) |
| 168 | { |
| 169 | if (work) |
| 170 | *work -= skb->truesize; |
| 171 | atomic_sub(skb->truesize, &ip_frag_mem); |
| 172 | kfree_skb(skb); |
| 173 | } |
| 174 | |
| 175 | static __inline__ void frag_free_queue(struct ipq *qp, int *work) |
| 176 | { |
| 177 | if (work) |
| 178 | *work -= sizeof(struct ipq); |
| 179 | atomic_sub(sizeof(struct ipq), &ip_frag_mem); |
| 180 | kfree(qp); |
| 181 | } |
| 182 | |
| 183 | static __inline__ struct ipq *frag_alloc_queue(void) |
| 184 | { |
| 185 | struct ipq *qp = kmalloc(sizeof(struct ipq), GFP_ATOMIC); |
| 186 | |
| 187 | if(!qp) |
| 188 | return NULL; |
| 189 | atomic_add(sizeof(struct ipq), &ip_frag_mem); |
| 190 | return qp; |
| 191 | } |
| 192 | |
| 193 | |
| 194 | /* Destruction primitives. */ |
| 195 | |
| 196 | /* Complete destruction of ipq. */ |
| 197 | static void ip_frag_destroy(struct ipq *qp, int *work) |
| 198 | { |
| 199 | struct sk_buff *fp; |
| 200 | |
| 201 | BUG_TRAP(qp->last_in&COMPLETE); |
| 202 | BUG_TRAP(del_timer(&qp->timer) == 0); |
| 203 | |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 204 | if (qp->peer) |
| 205 | inet_putpeer(qp->peer); |
| 206 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | /* Release all fragment data. */ |
| 208 | fp = qp->fragments; |
| 209 | while (fp) { |
| 210 | struct sk_buff *xp = fp->next; |
| 211 | |
| 212 | frag_kfree_skb(fp, work); |
| 213 | fp = xp; |
| 214 | } |
| 215 | |
| 216 | /* Finally, release the queue descriptor itself. */ |
| 217 | frag_free_queue(qp, work); |
| 218 | } |
| 219 | |
| 220 | static __inline__ void ipq_put(struct ipq *ipq, int *work) |
| 221 | { |
| 222 | if (atomic_dec_and_test(&ipq->refcnt)) |
| 223 | ip_frag_destroy(ipq, work); |
| 224 | } |
| 225 | |
| 226 | /* Kill ipq entry. It is not destroyed immediately, |
| 227 | * because caller (and someone more) holds reference count. |
| 228 | */ |
| 229 | static void ipq_kill(struct ipq *ipq) |
| 230 | { |
| 231 | if (del_timer(&ipq->timer)) |
| 232 | atomic_dec(&ipq->refcnt); |
| 233 | |
| 234 | if (!(ipq->last_in & COMPLETE)) { |
| 235 | ipq_unlink(ipq); |
| 236 | atomic_dec(&ipq->refcnt); |
| 237 | ipq->last_in |= COMPLETE; |
| 238 | } |
| 239 | } |
| 240 | |
| 241 | /* Memory limiting on fragments. Evictor trashes the oldest |
| 242 | * fragment queue until we are back under the threshold. |
| 243 | */ |
| 244 | static void ip_evictor(void) |
| 245 | { |
| 246 | struct ipq *qp; |
| 247 | struct list_head *tmp; |
| 248 | int work; |
| 249 | |
| 250 | work = atomic_read(&ip_frag_mem) - sysctl_ipfrag_low_thresh; |
| 251 | if (work <= 0) |
| 252 | return; |
| 253 | |
| 254 | while (work > 0) { |
| 255 | read_lock(&ipfrag_lock); |
| 256 | if (list_empty(&ipq_lru_list)) { |
| 257 | read_unlock(&ipfrag_lock); |
| 258 | return; |
| 259 | } |
| 260 | tmp = ipq_lru_list.next; |
| 261 | qp = list_entry(tmp, struct ipq, lru_list); |
| 262 | atomic_inc(&qp->refcnt); |
| 263 | read_unlock(&ipfrag_lock); |
| 264 | |
| 265 | spin_lock(&qp->lock); |
| 266 | if (!(qp->last_in&COMPLETE)) |
| 267 | ipq_kill(qp); |
| 268 | spin_unlock(&qp->lock); |
| 269 | |
| 270 | ipq_put(qp, &work); |
| 271 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | /* |
| 276 | * Oops, a fragment queue timed out. Kill it and send an ICMP reply. |
| 277 | */ |
| 278 | static void ip_expire(unsigned long arg) |
| 279 | { |
| 280 | struct ipq *qp = (struct ipq *) arg; |
| 281 | |
| 282 | spin_lock(&qp->lock); |
| 283 | |
| 284 | if (qp->last_in & COMPLETE) |
| 285 | goto out; |
| 286 | |
| 287 | ipq_kill(qp); |
| 288 | |
| 289 | IP_INC_STATS_BH(IPSTATS_MIB_REASMTIMEOUT); |
| 290 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); |
| 291 | |
| 292 | if ((qp->last_in&FIRST_IN) && qp->fragments != NULL) { |
| 293 | struct sk_buff *head = qp->fragments; |
| 294 | /* Send an ICMP "Fragment Reassembly Timeout" message. */ |
| 295 | if ((head->dev = dev_get_by_index(qp->iif)) != NULL) { |
| 296 | icmp_send(head, ICMP_TIME_EXCEEDED, ICMP_EXC_FRAGTIME, 0); |
| 297 | dev_put(head->dev); |
| 298 | } |
| 299 | } |
| 300 | out: |
| 301 | spin_unlock(&qp->lock); |
| 302 | ipq_put(qp, NULL); |
| 303 | } |
| 304 | |
| 305 | /* Creation primitives. */ |
| 306 | |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 307 | static struct ipq *ip_frag_intern(struct ipq *qp_in) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | { |
| 309 | struct ipq *qp; |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 310 | #ifdef CONFIG_SMP |
| 311 | struct hlist_node *n; |
| 312 | #endif |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 313 | unsigned int hash; |
| 314 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | write_lock(&ipfrag_lock); |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 316 | hash = ipqhashfn(qp_in->id, qp_in->saddr, qp_in->daddr, |
| 317 | qp_in->protocol); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | #ifdef CONFIG_SMP |
| 319 | /* With SMP race we have to recheck hash table, because |
| 320 | * such entry could be created on other cpu, while we |
| 321 | * promoted read lock to write lock. |
| 322 | */ |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 323 | hlist_for_each_entry(qp, n, &ipq_hash[hash], list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 324 | if(qp->id == qp_in->id && |
| 325 | qp->saddr == qp_in->saddr && |
| 326 | qp->daddr == qp_in->daddr && |
| 327 | qp->protocol == qp_in->protocol && |
| 328 | qp->user == qp_in->user) { |
| 329 | atomic_inc(&qp->refcnt); |
| 330 | write_unlock(&ipfrag_lock); |
| 331 | qp_in->last_in |= COMPLETE; |
| 332 | ipq_put(qp_in, NULL); |
| 333 | return qp; |
| 334 | } |
| 335 | } |
| 336 | #endif |
| 337 | qp = qp_in; |
| 338 | |
| 339 | if (!mod_timer(&qp->timer, jiffies + sysctl_ipfrag_time)) |
| 340 | atomic_inc(&qp->refcnt); |
| 341 | |
| 342 | atomic_inc(&qp->refcnt); |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 343 | hlist_add_head(&qp->list, &ipq_hash[hash]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | INIT_LIST_HEAD(&qp->lru_list); |
| 345 | list_add_tail(&qp->lru_list, &ipq_lru_list); |
| 346 | ip_frag_nqueues++; |
| 347 | write_unlock(&ipfrag_lock); |
| 348 | return qp; |
| 349 | } |
| 350 | |
| 351 | /* Add an entry to the 'ipq' queue for a newly received IP datagram. */ |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 352 | static struct ipq *ip_frag_create(struct iphdr *iph, u32 user) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 353 | { |
| 354 | struct ipq *qp; |
| 355 | |
| 356 | if ((qp = frag_alloc_queue()) == NULL) |
| 357 | goto out_nomem; |
| 358 | |
| 359 | qp->protocol = iph->protocol; |
| 360 | qp->last_in = 0; |
| 361 | qp->id = iph->id; |
| 362 | qp->saddr = iph->saddr; |
| 363 | qp->daddr = iph->daddr; |
| 364 | qp->user = user; |
| 365 | qp->len = 0; |
| 366 | qp->meat = 0; |
| 367 | qp->fragments = NULL; |
| 368 | qp->iif = 0; |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 369 | qp->peer = sysctl_ipfrag_max_dist ? inet_getpeer(iph->saddr, 1) : NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | |
| 371 | /* Initialize a timer for this entry. */ |
| 372 | init_timer(&qp->timer); |
| 373 | qp->timer.data = (unsigned long) qp; /* pointer to queue */ |
| 374 | qp->timer.function = ip_expire; /* expire function */ |
| 375 | spin_lock_init(&qp->lock); |
| 376 | atomic_set(&qp->refcnt, 1); |
| 377 | |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 378 | return ip_frag_intern(qp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | |
| 380 | out_nomem: |
Patrick McHardy | 64ce207 | 2005-08-09 20:50:53 -0700 | [diff] [blame] | 381 | LIMIT_NETDEBUG(KERN_ERR "ip_frag_create: no memory left !\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | return NULL; |
| 383 | } |
| 384 | |
| 385 | /* Find the correct entry in the "incomplete datagrams" queue for |
| 386 | * this IP datagram, and create new one, if nothing is found. |
| 387 | */ |
| 388 | static inline struct ipq *ip_find(struct iphdr *iph, u32 user) |
| 389 | { |
Alexey Dobriyan | 76ab608 | 2006-01-06 13:24:29 -0800 | [diff] [blame] | 390 | __be16 id = iph->id; |
Al Viro | 1827777 | 2006-09-26 22:19:02 -0700 | [diff] [blame] | 391 | __be32 saddr = iph->saddr; |
| 392 | __be32 daddr = iph->daddr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | __u8 protocol = iph->protocol; |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 394 | unsigned int hash; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | struct ipq *qp; |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 396 | struct hlist_node *n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | |
| 398 | read_lock(&ipfrag_lock); |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 399 | hash = ipqhashfn(id, saddr, daddr, protocol); |
Yasuyuki Kozakai | e7c8a41 | 2005-11-16 12:55:37 -0800 | [diff] [blame] | 400 | hlist_for_each_entry(qp, n, &ipq_hash[hash], list) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 401 | if(qp->id == id && |
| 402 | qp->saddr == saddr && |
| 403 | qp->daddr == daddr && |
| 404 | qp->protocol == protocol && |
| 405 | qp->user == user) { |
| 406 | atomic_inc(&qp->refcnt); |
| 407 | read_unlock(&ipfrag_lock); |
| 408 | return qp; |
| 409 | } |
| 410 | } |
| 411 | read_unlock(&ipfrag_lock); |
| 412 | |
David S. Miller | 55c0022 | 2006-04-09 22:43:55 -0700 | [diff] [blame] | 413 | return ip_frag_create(iph, user); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | } |
| 415 | |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 416 | /* Is the fragment too far ahead to be part of ipq? */ |
| 417 | static inline int ip_frag_too_far(struct ipq *qp) |
| 418 | { |
| 419 | struct inet_peer *peer = qp->peer; |
| 420 | unsigned int max = sysctl_ipfrag_max_dist; |
| 421 | unsigned int start, end; |
| 422 | |
| 423 | int rc; |
| 424 | |
| 425 | if (!peer || !max) |
| 426 | return 0; |
| 427 | |
| 428 | start = qp->rid; |
| 429 | end = atomic_inc_return(&peer->rid); |
| 430 | qp->rid = end; |
| 431 | |
| 432 | rc = qp->fragments && (end - start) > max; |
| 433 | |
| 434 | if (rc) { |
| 435 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); |
| 436 | } |
| 437 | |
| 438 | return rc; |
| 439 | } |
| 440 | |
| 441 | static int ip_frag_reinit(struct ipq *qp) |
| 442 | { |
| 443 | struct sk_buff *fp; |
| 444 | |
| 445 | if (!mod_timer(&qp->timer, jiffies + sysctl_ipfrag_time)) { |
| 446 | atomic_inc(&qp->refcnt); |
| 447 | return -ETIMEDOUT; |
| 448 | } |
| 449 | |
| 450 | fp = qp->fragments; |
| 451 | do { |
| 452 | struct sk_buff *xp = fp->next; |
| 453 | frag_kfree_skb(fp, NULL); |
| 454 | fp = xp; |
| 455 | } while (fp); |
| 456 | |
| 457 | qp->last_in = 0; |
| 458 | qp->len = 0; |
| 459 | qp->meat = 0; |
| 460 | qp->fragments = NULL; |
| 461 | qp->iif = 0; |
| 462 | |
| 463 | return 0; |
| 464 | } |
| 465 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | /* Add new segment to existing queue. */ |
| 467 | static void ip_frag_queue(struct ipq *qp, struct sk_buff *skb) |
| 468 | { |
| 469 | struct sk_buff *prev, *next; |
| 470 | int flags, offset; |
| 471 | int ihl, end; |
| 472 | |
| 473 | if (qp->last_in & COMPLETE) |
| 474 | goto err; |
| 475 | |
Herbert Xu | 89cee8b | 2005-12-13 23:14:27 -0800 | [diff] [blame] | 476 | if (!(IPCB(skb)->flags & IPSKB_FRAG_COMPLETE) && |
| 477 | unlikely(ip_frag_too_far(qp)) && unlikely(ip_frag_reinit(qp))) { |
| 478 | ipq_kill(qp); |
| 479 | goto err; |
| 480 | } |
| 481 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | offset = ntohs(skb->nh.iph->frag_off); |
| 483 | flags = offset & ~IP_OFFSET; |
| 484 | offset &= IP_OFFSET; |
| 485 | offset <<= 3; /* offset is in 8-byte chunks */ |
| 486 | ihl = skb->nh.iph->ihl * 4; |
| 487 | |
| 488 | /* Determine the position of this fragment. */ |
| 489 | end = offset + skb->len - ihl; |
| 490 | |
| 491 | /* Is this the final fragment? */ |
| 492 | if ((flags & IP_MF) == 0) { |
| 493 | /* If we already have some bits beyond end |
| 494 | * or have different end, the segment is corrrupted. |
| 495 | */ |
| 496 | if (end < qp->len || |
| 497 | ((qp->last_in & LAST_IN) && end != qp->len)) |
| 498 | goto err; |
| 499 | qp->last_in |= LAST_IN; |
| 500 | qp->len = end; |
| 501 | } else { |
| 502 | if (end&7) { |
| 503 | end &= ~7; |
| 504 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) |
| 505 | skb->ip_summed = CHECKSUM_NONE; |
| 506 | } |
| 507 | if (end > qp->len) { |
| 508 | /* Some bits beyond end -> corruption. */ |
| 509 | if (qp->last_in & LAST_IN) |
| 510 | goto err; |
| 511 | qp->len = end; |
| 512 | } |
| 513 | } |
| 514 | if (end == offset) |
| 515 | goto err; |
| 516 | |
| 517 | if (pskb_pull(skb, ihl) == NULL) |
| 518 | goto err; |
Stephen Hemminger | 48bc41a | 2005-09-06 15:51:48 -0700 | [diff] [blame] | 519 | if (pskb_trim_rcsum(skb, end-offset)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | goto err; |
| 521 | |
| 522 | /* Find out which fragments are in front and at the back of us |
| 523 | * in the chain of fragments so far. We must know where to put |
| 524 | * this fragment, right? |
| 525 | */ |
| 526 | prev = NULL; |
| 527 | for(next = qp->fragments; next != NULL; next = next->next) { |
| 528 | if (FRAG_CB(next)->offset >= offset) |
| 529 | break; /* bingo! */ |
| 530 | prev = next; |
| 531 | } |
| 532 | |
| 533 | /* We found where to put this one. Check for overlap with |
| 534 | * preceding fragment, and, if needed, align things so that |
| 535 | * any overlaps are eliminated. |
| 536 | */ |
| 537 | if (prev) { |
| 538 | int i = (FRAG_CB(prev)->offset + prev->len) - offset; |
| 539 | |
| 540 | if (i > 0) { |
| 541 | offset += i; |
| 542 | if (end <= offset) |
| 543 | goto err; |
| 544 | if (!pskb_pull(skb, i)) |
| 545 | goto err; |
| 546 | if (skb->ip_summed != CHECKSUM_UNNECESSARY) |
| 547 | skb->ip_summed = CHECKSUM_NONE; |
| 548 | } |
| 549 | } |
| 550 | |
| 551 | while (next && FRAG_CB(next)->offset < end) { |
| 552 | int i = end - FRAG_CB(next)->offset; /* overlap is 'i' bytes */ |
| 553 | |
| 554 | if (i < next->len) { |
| 555 | /* Eat head of the next overlapped fragment |
| 556 | * and leave the loop. The next ones cannot overlap. |
| 557 | */ |
| 558 | if (!pskb_pull(next, i)) |
| 559 | goto err; |
| 560 | FRAG_CB(next)->offset += i; |
| 561 | qp->meat -= i; |
| 562 | if (next->ip_summed != CHECKSUM_UNNECESSARY) |
| 563 | next->ip_summed = CHECKSUM_NONE; |
| 564 | break; |
| 565 | } else { |
| 566 | struct sk_buff *free_it = next; |
| 567 | |
| 568 | /* Old fragmnet is completely overridden with |
| 569 | * new one drop it. |
| 570 | */ |
| 571 | next = next->next; |
| 572 | |
| 573 | if (prev) |
| 574 | prev->next = next; |
| 575 | else |
| 576 | qp->fragments = next; |
| 577 | |
| 578 | qp->meat -= free_it->len; |
| 579 | frag_kfree_skb(free_it, NULL); |
| 580 | } |
| 581 | } |
| 582 | |
| 583 | FRAG_CB(skb)->offset = offset; |
| 584 | |
| 585 | /* Insert this fragment in the chain of fragments. */ |
| 586 | skb->next = next; |
| 587 | if (prev) |
| 588 | prev->next = skb; |
| 589 | else |
| 590 | qp->fragments = skb; |
| 591 | |
| 592 | if (skb->dev) |
| 593 | qp->iif = skb->dev->ifindex; |
| 594 | skb->dev = NULL; |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 595 | skb_get_timestamp(skb, &qp->stamp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | qp->meat += skb->len; |
| 597 | atomic_add(skb->truesize, &ip_frag_mem); |
| 598 | if (offset == 0) |
| 599 | qp->last_in |= FIRST_IN; |
| 600 | |
| 601 | write_lock(&ipfrag_lock); |
| 602 | list_move_tail(&qp->lru_list, &ipq_lru_list); |
| 603 | write_unlock(&ipfrag_lock); |
| 604 | |
| 605 | return; |
| 606 | |
| 607 | err: |
| 608 | kfree_skb(skb); |
| 609 | } |
| 610 | |
| 611 | |
| 612 | /* Build a new IP datagram from all its fragments. */ |
| 613 | |
| 614 | static struct sk_buff *ip_frag_reasm(struct ipq *qp, struct net_device *dev) |
| 615 | { |
| 616 | struct iphdr *iph; |
| 617 | struct sk_buff *fp, *head = qp->fragments; |
| 618 | int len; |
| 619 | int ihlen; |
| 620 | |
| 621 | ipq_kill(qp); |
| 622 | |
| 623 | BUG_TRAP(head != NULL); |
| 624 | BUG_TRAP(FRAG_CB(head)->offset == 0); |
| 625 | |
| 626 | /* Allocate a new buffer for the datagram. */ |
| 627 | ihlen = head->nh.iph->ihl*4; |
| 628 | len = ihlen + qp->len; |
| 629 | |
| 630 | if(len > 65535) |
| 631 | goto out_oversize; |
| 632 | |
| 633 | /* Head of list must not be cloned. */ |
| 634 | if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) |
| 635 | goto out_nomem; |
| 636 | |
| 637 | /* If the first fragment is fragmented itself, we split |
| 638 | * it to two chunks: the first with data and paged part |
| 639 | * and the second, holding only fragments. */ |
| 640 | if (skb_shinfo(head)->frag_list) { |
| 641 | struct sk_buff *clone; |
| 642 | int i, plen = 0; |
| 643 | |
| 644 | if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) |
| 645 | goto out_nomem; |
| 646 | clone->next = head->next; |
| 647 | head->next = clone; |
| 648 | skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list; |
| 649 | skb_shinfo(head)->frag_list = NULL; |
| 650 | for (i=0; i<skb_shinfo(head)->nr_frags; i++) |
| 651 | plen += skb_shinfo(head)->frags[i].size; |
| 652 | clone->len = clone->data_len = head->data_len - plen; |
| 653 | head->data_len -= clone->len; |
| 654 | head->len -= clone->len; |
| 655 | clone->csum = 0; |
| 656 | clone->ip_summed = head->ip_summed; |
| 657 | atomic_add(clone->truesize, &ip_frag_mem); |
| 658 | } |
| 659 | |
| 660 | skb_shinfo(head)->frag_list = head->next; |
| 661 | skb_push(head, head->data - head->nh.raw); |
| 662 | atomic_sub(head->truesize, &ip_frag_mem); |
| 663 | |
| 664 | for (fp=head->next; fp; fp = fp->next) { |
| 665 | head->data_len += fp->len; |
| 666 | head->len += fp->len; |
| 667 | if (head->ip_summed != fp->ip_summed) |
| 668 | head->ip_summed = CHECKSUM_NONE; |
Patrick McHardy | 84fa793 | 2006-08-29 16:44:56 -0700 | [diff] [blame] | 669 | else if (head->ip_summed == CHECKSUM_COMPLETE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 670 | head->csum = csum_add(head->csum, fp->csum); |
| 671 | head->truesize += fp->truesize; |
| 672 | atomic_sub(fp->truesize, &ip_frag_mem); |
| 673 | } |
| 674 | |
| 675 | head->next = NULL; |
| 676 | head->dev = dev; |
Patrick McHardy | a61bbcf | 2005-08-14 17:24:31 -0700 | [diff] [blame] | 677 | skb_set_timestamp(head, &qp->stamp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | |
| 679 | iph = head->nh.iph; |
| 680 | iph->frag_off = 0; |
| 681 | iph->tot_len = htons(len); |
| 682 | IP_INC_STATS_BH(IPSTATS_MIB_REASMOKS); |
| 683 | qp->fragments = NULL; |
| 684 | return head; |
| 685 | |
| 686 | out_nomem: |
Patrick McHardy | 64ce207 | 2005-08-09 20:50:53 -0700 | [diff] [blame] | 687 | LIMIT_NETDEBUG(KERN_ERR "IP: queue_glue: no memory for gluing " |
| 688 | "queue %p\n", qp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | goto out_fail; |
| 690 | out_oversize: |
| 691 | if (net_ratelimit()) |
| 692 | printk(KERN_INFO |
| 693 | "Oversized IP packet from %d.%d.%d.%d.\n", |
| 694 | NIPQUAD(qp->saddr)); |
| 695 | out_fail: |
| 696 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); |
| 697 | return NULL; |
| 698 | } |
| 699 | |
| 700 | /* Process an incoming IP datagram fragment. */ |
| 701 | struct sk_buff *ip_defrag(struct sk_buff *skb, u32 user) |
| 702 | { |
| 703 | struct iphdr *iph = skb->nh.iph; |
| 704 | struct ipq *qp; |
| 705 | struct net_device *dev; |
| 706 | |
| 707 | IP_INC_STATS_BH(IPSTATS_MIB_REASMREQDS); |
| 708 | |
| 709 | /* Start by cleaning up the memory. */ |
| 710 | if (atomic_read(&ip_frag_mem) > sysctl_ipfrag_high_thresh) |
| 711 | ip_evictor(); |
| 712 | |
| 713 | dev = skb->dev; |
| 714 | |
| 715 | /* Lookup (or create) queue header */ |
| 716 | if ((qp = ip_find(iph, user)) != NULL) { |
| 717 | struct sk_buff *ret = NULL; |
| 718 | |
| 719 | spin_lock(&qp->lock); |
| 720 | |
| 721 | ip_frag_queue(qp, skb); |
| 722 | |
| 723 | if (qp->last_in == (FIRST_IN|LAST_IN) && |
| 724 | qp->meat == qp->len) |
| 725 | ret = ip_frag_reasm(qp, dev); |
| 726 | |
| 727 | spin_unlock(&qp->lock); |
| 728 | ipq_put(qp, NULL); |
| 729 | return ret; |
| 730 | } |
| 731 | |
| 732 | IP_INC_STATS_BH(IPSTATS_MIB_REASMFAILS); |
| 733 | kfree_skb(skb); |
| 734 | return NULL; |
| 735 | } |
| 736 | |
| 737 | void ipfrag_init(void) |
| 738 | { |
| 739 | ipfrag_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^ |
| 740 | (jiffies ^ (jiffies >> 6))); |
| 741 | |
| 742 | init_timer(&ipfrag_secret_timer); |
| 743 | ipfrag_secret_timer.function = ipfrag_secret_rebuild; |
| 744 | ipfrag_secret_timer.expires = jiffies + sysctl_ipfrag_secret_interval; |
| 745 | add_timer(&ipfrag_secret_timer); |
| 746 | } |
| 747 | |
| 748 | EXPORT_SYMBOL(ip_defrag); |