xref: /dpdk/lib/ip_frag/rte_ip_frag.h (revision daa02b5cddbb8e11b31d41e2bf7bb1ae64dcae2f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation
3  */
4 
5 #ifndef _RTE_IP_FRAG_H_
6 #define _RTE_IP_FRAG_H_
7 
8 /**
9  * @file
10  * RTE IP Fragmentation and Reassembly
11  *
12  * Implementation of IP packet fragmentation and reassembly.
13  */
14 
15 #ifdef __cplusplus
16 extern "C" {
17 #endif
18 
19 #include <stdint.h>
20 #include <stdio.h>
21 
22 #include <rte_config.h>
23 #include <rte_malloc.h>
24 #include <rte_memory.h>
25 #include <rte_ip.h>
26 #include <rte_byteorder.h>
27 
28 struct rte_mbuf;
29 
30 enum {
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 */
36 };
37 
38 /** @internal fragmented mbuf */
39 struct ip_frag {
40 	uint16_t ofs;          /**< offset into the packet */
41 	uint16_t len;          /**< length of fragment */
42 	struct rte_mbuf *mb;   /**< fragment mbuf */
43 };
44 
45 /** @internal <src addr, dst_addr, id> to uniquely identify fragmented datagram. */
46 struct ip_frag_key {
47 	uint64_t src_dst[4];
48 	/**< src and dst address, only first 8 bytes used for IPv4 */
49 	RTE_STD_C11
50 	union {
51 		uint64_t id_key_len; /**< combined for easy fetch */
52 		__extension__
53 		struct {
54 			uint32_t id;       /**< packet id */
55 			uint32_t key_len;  /**< src/dst key length */
56 		};
57 	};
58 };
59 
60 /**
61  * @internal Fragmented packet to reassemble.
62  * First two entries in the frags[] array are for the last and first fragments.
63  */
64 struct ip_frag_pkt {
65 	RTE_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;
73 
74 #define IP_FRAG_DEATH_ROW_LEN 32 /**< death row size (in packets) */
75 
76 /* death row size in mbufs */
77 #define IP_FRAG_DEATH_ROW_MBUF_LEN (IP_FRAG_DEATH_ROW_LEN * (IP_MAX_FRAG_NUM + 1))
78 
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 */
84 };
85 
86 RTE_TAILQ_HEAD(ip_pkt_list, ip_frag_pkt); /**< @internal fragments tailq */
87 
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;
97 
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. */
111 };
112 
113 /* struct ipv6_extension_fragment moved to librte_net/rte_ip.h and renamed. */
114 #define ipv6_extension_fragment	rte_ipv6_fragment_ext
115 
116 /**
117  * Create a new IP fragmentation table.
118  *
119  * @param bucket_num
120  *   Number of buckets in the hash table.
121  * @param bucket_entries
122  *   Number of entries per bucket (e.g. hash associativity).
123  *   Should be power of two.
124  * @param max_entries
125  *   Maximum number of entries that could be stored in the table.
126  *   The value should be less or equal then bucket_num * bucket_entries.
127  * @param max_cycles
128  *   Maximum TTL in cycles for each fragmented packet.
129  * @param socket_id
130  *   The *socket_id* argument is the socket identifier in the case of
131  *   NUMA. The value can be *SOCKET_ID_ANY* if there is no NUMA constraints.
132  * @return
133  *   The pointer to the new allocated fragmentation table, on success. NULL on error.
134  */
135 struct rte_ip_frag_tbl * rte_ip_frag_table_create(uint32_t bucket_num,
136 		uint32_t bucket_entries,  uint32_t max_entries,
137 		uint64_t max_cycles, int socket_id);
138 
139 /**
140  * Free allocated IP fragmentation table.
141  *
142  * @param tbl
143  *   Fragmentation table to free.
144  */
145 void
146 rte_ip_frag_table_destroy(struct rte_ip_frag_tbl *tbl);
147 
148 /**
149  * This function implements the fragmentation of IPv6 packets.
150  *
151  * @param pkt_in
152  *   The input packet.
153  * @param pkts_out
154  *   Array storing the output fragments.
155  * @param nb_pkts_out
156  *   Number of fragments.
157  * @param mtu_size
158  *   Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv6
159  *   datagrams. This value includes the size of the IPv6 header.
160  * @param pool_direct
161  *   MBUF pool used for allocating direct buffers for the output fragments.
162  * @param pool_indirect
163  *   MBUF pool used for allocating indirect buffers for the output fragments.
164  * @return
165  *   Upon successful completion - number of output fragments placed
166  *   in the pkts_out array.
167  *   Otherwise - (-1) * errno.
168  */
169 int32_t
170 rte_ipv6_fragment_packet(struct rte_mbuf *pkt_in,
171 		struct rte_mbuf **pkts_out,
172 		uint16_t nb_pkts_out,
173 		uint16_t mtu_size,
174 		struct rte_mempool *pool_direct,
175 		struct rte_mempool *pool_indirect);
176 
177 /**
178  * This function implements reassembly of fragmented IPv6 packets.
179  * Incoming mbuf should have its l2_len/l3_len fields setup correctly.
180  *
181  * @param tbl
182  *   Table where to lookup/add the fragmented packet.
183  * @param dr
184  *   Death row to free buffers to
185  * @param mb
186  *   Incoming mbuf with IPv6 fragment.
187  * @param tms
188  *   Fragment arrival timestamp.
189  * @param ip_hdr
190  *   Pointer to the IPv6 header.
191  * @param frag_hdr
192  *   Pointer to the IPv6 fragment extension header.
193  * @return
194  *   Pointer to mbuf for reassembled packet, or NULL if:
195  *   - an error occurred.
196  *   - not all fragments of the packet are collected yet.
197  */
198 struct rte_mbuf *rte_ipv6_frag_reassemble_packet(struct rte_ip_frag_tbl *tbl,
199 		struct rte_ip_frag_death_row *dr,
200 		struct rte_mbuf *mb, uint64_t tms, struct rte_ipv6_hdr *ip_hdr,
201 		struct ipv6_extension_fragment *frag_hdr);
202 
203 /**
204  * Return a pointer to the packet's fragment header, if found.
205  * It only looks at the extension header that's right after the fixed IPv6
206  * header, and doesn't follow the whole chain of extension headers.
207  *
208  * @param hdr
209  *   Pointer to the IPv6 header.
210  * @return
211  *   Pointer to the IPv6 fragment extension header, or NULL if it's not
212  *   present.
213  */
214 static inline struct ipv6_extension_fragment *
215 rte_ipv6_frag_get_ipv6_fragment_header(struct rte_ipv6_hdr *hdr)
216 {
217 	if (hdr->proto == IPPROTO_FRAGMENT) {
218 		return (struct ipv6_extension_fragment *) ++hdr;
219 	}
220 	else
221 		return NULL;
222 }
223 
224 /**
225  * IPv4 fragmentation.
226  *
227  * This function implements the fragmentation of IPv4 packets.
228  *
229  * @param pkt_in
230  *   The input packet.
231  * @param pkts_out
232  *   Array storing the output fragments.
233  * @param nb_pkts_out
234  *   Number of fragments.
235  * @param mtu_size
236  *   Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv4
237  *   datagrams. This value includes the size of the IPv4 header.
238  * @param pool_direct
239  *   MBUF pool used for allocating direct buffers for the output fragments.
240  * @param pool_indirect
241  *   MBUF pool used for allocating indirect buffers for the output fragments.
242  * @return
243  *   Upon successful completion - number of output fragments placed
244  *   in the pkts_out array.
245  *   Otherwise - (-1) * errno.
246  */
247 int32_t rte_ipv4_fragment_packet(struct rte_mbuf *pkt_in,
248 			struct rte_mbuf **pkts_out,
249 			uint16_t nb_pkts_out, uint16_t mtu_size,
250 			struct rte_mempool *pool_direct,
251 			struct rte_mempool *pool_indirect);
252 
253 /**
254  * This function implements reassembly of fragmented IPv4 packets.
255  * Incoming mbufs should have its l2_len/l3_len fields setup correctly.
256  *
257  * @param tbl
258  *   Table where to lookup/add the fragmented packet.
259  * @param dr
260  *   Death row to free buffers to
261  * @param mb
262  *   Incoming mbuf with IPv4 fragment.
263  * @param tms
264  *   Fragment arrival timestamp.
265  * @param ip_hdr
266  *   Pointer to the IPV4 header inside the fragment.
267  * @return
268  *   Pointer to mbuf for reassembled packet, or NULL if:
269  *   - an error occurred.
270  *   - not all fragments of the packet are collected yet.
271  */
272 struct rte_mbuf * rte_ipv4_frag_reassemble_packet(struct rte_ip_frag_tbl *tbl,
273 		struct rte_ip_frag_death_row *dr,
274 		struct rte_mbuf *mb, uint64_t tms, struct rte_ipv4_hdr *ip_hdr);
275 
276 /**
277  * Check if the IPv4 packet is fragmented
278  *
279  * @param hdr
280  *   IPv4 header of the packet
281  * @return
282  *   1 if fragmented, 0 if not fragmented
283  */
284 static inline int
285 rte_ipv4_frag_pkt_is_fragmented(const struct rte_ipv4_hdr *hdr)
286 {
287 	uint16_t flag_offset, ip_flag, ip_ofs;
288 
289 	flag_offset = rte_be_to_cpu_16(hdr->fragment_offset);
290 	ip_ofs = (uint16_t)(flag_offset & RTE_IPV4_HDR_OFFSET_MASK);
291 	ip_flag = (uint16_t)(flag_offset & RTE_IPV4_HDR_MF_FLAG);
292 
293 	return ip_flag != 0 || ip_ofs  != 0;
294 }
295 
296 /**
297  * Free mbufs on a given death row.
298  *
299  * @param dr
300  *   Death row to free mbufs in.
301  * @param prefetch
302  *   How many buffers to prefetch before freeing.
303  */
304 void rte_ip_frag_free_death_row(struct rte_ip_frag_death_row *dr,
305 		uint32_t prefetch);
306 
307 
308 /**
309  * Dump fragmentation table statistics to file.
310  *
311  * @param f
312  *   File to dump statistics to
313  * @param tbl
314  *   Fragmentation table to dump statistics from
315  */
316 void
317 rte_ip_frag_table_statistics_dump(FILE * f, const struct rte_ip_frag_tbl *tbl);
318 
319 /**
320  * Delete expired fragments
321  *
322  * @param tbl
323  *   Table to delete expired fragments from
324  * @param dr
325  *   Death row to free buffers to
326  * @param tms
327  *   Current timestamp
328  */
329 __rte_experimental
330 void
331 rte_frag_table_del_expired_entries(struct rte_ip_frag_tbl *tbl,
332 	struct rte_ip_frag_death_row *dr, uint64_t tms);
333 
334 #ifdef __cplusplus
335 }
336 #endif
337 
338 #endif /* _RTE_IP_FRAG_H_ */
339