1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
5 #ifndef _RTE_IP_FRAG_H_
6 #define _RTE_IP_FRAG_H_
10 * RTE IP Fragmentation and Reassembly
12 * Implementation of IP packet fragmentation and reassembly.
22 #include <rte_config.h>
23 #include <rte_malloc.h>
24 #include <rte_memory.h>
26 #include <rte_byteorder.h>
31 IP_LAST_FRAG_IDX, /**< index of last fragment */
32 IP_FIRST_FRAG_IDX, /**< index of first fragment */
33 IP_MIN_FRAG_NUM, /**< minimum number of fragments */
34 IP_MAX_FRAG_NUM = RTE_LIBRTE_IP_FRAG_MAX_FRAG,
35 /**< maximum number of fragments per packet */
38 /** @internal fragmented mbuf */
40 uint16_t ofs; /**< offset into the packet */
41 uint16_t len; /**< length of fragment */
42 struct rte_mbuf *mb; /**< fragment mbuf */
45 /** @internal <src addr, dst_addr, id> to uniquely identify fragmented datagram. */
48 /**< src and dst address, only first 8 bytes used for IPv4 */
51 uint64_t id_key_len; /**< combined for easy fetch */
54 uint32_t id; /**< packet id */
55 uint32_t key_len; /**< src/dst key length */
61 * @internal Fragmented packet to reassemble.
62 * First two entries in the frags[] array are for the last and first fragments.
65 TAILQ_ENTRY(ip_frag_pkt) lru; /**< LRU list */
66 struct ip_frag_key key; /**< fragmentation key */
67 uint64_t start; /**< creation timestamp */
68 uint32_t total_size; /**< expected reassembled size */
69 uint32_t frag_size; /**< size of fragments received */
70 uint32_t last_idx; /**< index of next entry to fill */
71 struct ip_frag frags[IP_MAX_FRAG_NUM]; /**< fragments */
72 } __rte_cache_aligned;
74 #define IP_FRAG_DEATH_ROW_LEN 32 /**< death row size (in packets) */
76 /* death row size in mbufs */
77 #define IP_FRAG_DEATH_ROW_MBUF_LEN (IP_FRAG_DEATH_ROW_LEN * (IP_MAX_FRAG_NUM + 1))
79 /** mbuf death row (packets to be freed) */
80 struct rte_ip_frag_death_row {
81 uint32_t cnt; /**< number of mbufs currently on death row */
82 struct rte_mbuf *row[IP_FRAG_DEATH_ROW_MBUF_LEN];
83 /**< mbufs to be freed */
86 TAILQ_HEAD(ip_pkt_list, ip_frag_pkt); /**< @internal fragments tailq */
88 /** fragmentation table statistics */
89 struct ip_frag_tbl_stat {
90 uint64_t find_num; /**< total # of find/insert attempts. */
91 uint64_t add_num; /**< # of add ops. */
92 uint64_t del_num; /**< # of del ops. */
93 uint64_t reuse_num; /**< # of reuse (del/add) ops. */
94 uint64_t fail_total; /**< total # of add failures. */
95 uint64_t fail_nospace; /**< # of 'no space' add failures. */
96 } __rte_cache_aligned;
98 /** fragmentation table */
99 struct rte_ip_frag_tbl {
100 uint64_t max_cycles; /**< ttl for table entries. */
101 uint32_t entry_mask; /**< hash value mask. */
102 uint32_t max_entries; /**< max entries allowed. */
103 uint32_t use_entries; /**< entries in use. */
104 uint32_t bucket_entries; /**< hash associativity. */
105 uint32_t nb_entries; /**< total size of the table. */
106 uint32_t nb_buckets; /**< num of associativity lines. */
107 struct ip_frag_pkt *last; /**< last used entry. */
108 struct ip_pkt_list lru; /**< LRU list for table entries. */
109 struct ip_frag_tbl_stat stat; /**< statistics counters. */
110 __extension__ struct ip_frag_pkt pkt[0]; /**< hash table. */
113 /** IPv6 fragment extension header */
114 #define RTE_IPV6_EHDR_MF_SHIFT 0
115 #define RTE_IPV6_EHDR_MF_MASK 1
116 #define RTE_IPV6_EHDR_FO_SHIFT 3
117 #define RTE_IPV6_EHDR_FO_MASK (~((1 << RTE_IPV6_EHDR_FO_SHIFT) - 1))
118 #define RTE_IPV6_EHDR_FO_ALIGN (1 << RTE_IPV6_EHDR_FO_SHIFT)
120 #define RTE_IPV6_FRAG_USED_MASK \
121 (RTE_IPV6_EHDR_MF_MASK | RTE_IPV6_EHDR_FO_MASK)
123 #define RTE_IPV6_GET_MF(x) ((x) & RTE_IPV6_EHDR_MF_MASK)
124 #define RTE_IPV6_GET_FO(x) ((x) >> RTE_IPV6_EHDR_FO_SHIFT)
126 #define RTE_IPV6_SET_FRAG_DATA(fo, mf) \
127 (((fo) & RTE_IPV6_EHDR_FO_MASK) | ((mf) & RTE_IPV6_EHDR_MF_MASK))
129 struct ipv6_extension_fragment {
130 uint8_t next_header; /**< Next header type */
131 uint8_t reserved; /**< Reserved */
132 uint16_t frag_data; /**< All fragmentation data */
133 uint32_t id; /**< Packet ID */
134 } __attribute__((__packed__));
139 * Create a new IP fragmentation table.
142 * Number of buckets in the hash table.
143 * @param bucket_entries
144 * Number of entries per bucket (e.g. hash associativity).
145 * Should be power of two.
147 * Maximum number of entries that could be stored in the table.
148 * The value should be less or equal then bucket_num * bucket_entries.
150 * Maximum TTL in cycles for each fragmented packet.
152 * The *socket_id* argument is the socket identifier in the case of
153 * NUMA. The value can be *SOCKET_ID_ANY* if there is no NUMA constraints.
155 * The pointer to the new allocated fragmentation table, on success. NULL on error.
157 struct rte_ip_frag_tbl * rte_ip_frag_table_create(uint32_t bucket_num,
158 uint32_t bucket_entries, uint32_t max_entries,
159 uint64_t max_cycles, int socket_id);
162 * Free allocated IP fragmentation table.
165 * Fragmentation table to free.
168 rte_ip_frag_table_destroy(struct rte_ip_frag_tbl *tbl);
171 * This function implements the fragmentation of IPv6 packets.
176 * Array storing the output fragments.
178 * Number of fragments.
180 * Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv6
181 * datagrams. This value includes the size of the IPv6 header.
183 * MBUF pool used for allocating direct buffers for the output fragments.
184 * @param pool_indirect
185 * MBUF pool used for allocating indirect buffers for the output fragments.
187 * Upon successful completion - number of output fragments placed
188 * in the pkts_out array.
189 * Otherwise - (-1) * errno.
192 rte_ipv6_fragment_packet(struct rte_mbuf *pkt_in,
193 struct rte_mbuf **pkts_out,
194 uint16_t nb_pkts_out,
196 struct rte_mempool *pool_direct,
197 struct rte_mempool *pool_indirect);
200 * This function implements reassembly of fragmented IPv6 packets.
201 * Incoming mbuf should have its l2_len/l3_len fields setup correctly.
204 * Table where to lookup/add the fragmented packet.
206 * Death row to free buffers to
208 * Incoming mbuf with IPv6 fragment.
210 * Fragment arrival timestamp.
212 * Pointer to the IPv6 header.
214 * Pointer to the IPv6 fragment extension header.
216 * Pointer to mbuf for reassembled packet, or NULL if:
217 * - an error occurred.
218 * - not all fragments of the packet are collected yet.
220 struct rte_mbuf *rte_ipv6_frag_reassemble_packet(struct rte_ip_frag_tbl *tbl,
221 struct rte_ip_frag_death_row *dr,
222 struct rte_mbuf *mb, uint64_t tms, struct ipv6_hdr *ip_hdr,
223 struct ipv6_extension_fragment *frag_hdr);
226 * Return a pointer to the packet's fragment header, if found.
227 * It only looks at the extension header that's right after the fixed IPv6
228 * header, and doesn't follow the whole chain of extension headers.
231 * Pointer to the IPv6 header.
233 * Pointer to the IPv6 fragment extension header, or NULL if it's not
236 static inline struct ipv6_extension_fragment *
237 rte_ipv6_frag_get_ipv6_fragment_header(struct ipv6_hdr *hdr)
239 if (hdr->proto == IPPROTO_FRAGMENT) {
240 return (struct ipv6_extension_fragment *) ++hdr;
247 * IPv4 fragmentation.
249 * This function implements the fragmentation of IPv4 packets.
254 * Array storing the output fragments.
256 * Number of fragments.
258 * Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv4
259 * datagrams. This value includes the size of the IPv4 header.
261 * MBUF pool used for allocating direct buffers for the output fragments.
262 * @param pool_indirect
263 * MBUF pool used for allocating indirect buffers for the output fragments.
265 * Upon successful completion - number of output fragments placed
266 * in the pkts_out array.
267 * Otherwise - (-1) * errno.
269 int32_t rte_ipv4_fragment_packet(struct rte_mbuf *pkt_in,
270 struct rte_mbuf **pkts_out,
271 uint16_t nb_pkts_out, uint16_t mtu_size,
272 struct rte_mempool *pool_direct,
273 struct rte_mempool *pool_indirect);
276 * This function implements reassembly of fragmented IPv4 packets.
277 * Incoming mbufs should have its l2_len/l3_len fields setup correclty.
280 * Table where to lookup/add the fragmented packet.
282 * Death row to free buffers to
284 * Incoming mbuf with IPv4 fragment.
286 * Fragment arrival timestamp.
288 * Pointer to the IPV4 header inside the fragment.
290 * Pointer to mbuf for reassembled packet, or NULL if:
291 * - an error occurred.
292 * - not all fragments of the packet are collected yet.
294 struct rte_mbuf * rte_ipv4_frag_reassemble_packet(struct rte_ip_frag_tbl *tbl,
295 struct rte_ip_frag_death_row *dr,
296 struct rte_mbuf *mb, uint64_t tms, struct ipv4_hdr *ip_hdr);
299 * Check if the IPv4 packet is fragmented
302 * IPv4 header of the packet
304 * 1 if fragmented, 0 if not fragmented
307 rte_ipv4_frag_pkt_is_fragmented(const struct ipv4_hdr * hdr) {
308 uint16_t flag_offset, ip_flag, ip_ofs;
310 flag_offset = rte_be_to_cpu_16(hdr->fragment_offset);
311 ip_ofs = (uint16_t)(flag_offset & IPV4_HDR_OFFSET_MASK);
312 ip_flag = (uint16_t)(flag_offset & IPV4_HDR_MF_FLAG);
314 return ip_flag != 0 || ip_ofs != 0;
318 * Free mbufs on a given death row.
321 * Death row to free mbufs in.
323 * How many buffers to prefetch before freeing.
325 void rte_ip_frag_free_death_row(struct rte_ip_frag_death_row *dr,
330 * Dump fragmentation table statistics to file.
333 * File to dump statistics to
335 * Fragmentation table to dump statistics from
338 rte_ip_frag_table_statistics_dump(FILE * f, const struct rte_ip_frag_tbl *tbl);
341 * Delete expired fragments
344 * Table to delete expired fragments from
346 * Death row to free buffers to
350 void __rte_experimental
351 rte_frag_table_del_expired_entries(struct rte_ip_frag_tbl *tbl,
352 struct rte_ip_frag_death_row *dr, uint64_t tms);
358 #endif /* _RTE_IP_FRAG_H_ */