1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation.
3 * Copyright 2014 6WIND S.A.
13 * The mbuf library provides the ability to create and destroy buffers
14 * that may be used by the RTE application to store message
15 * buffers. The message buffers are stored in a mempool, using the
16 * RTE mempool library.
18 * The preferred way to create a mbuf pool is to use
19 * rte_pktmbuf_pool_create(). However, in some situations, an
20 * application may want to have more control (ex: populate the pool with
21 * specific memory), in this case it is possible to use functions from
22 * rte_mempool. See how rte_pktmbuf_pool_create() is implemented for
25 * This library provides an API to allocate/free packet mbufs, which are
26 * used to carry network packets.
28 * To understand the concepts of packet buffers or mbufs, you
29 * should read "TCP/IP Illustrated, Volume 2: The Implementation,
30 * Addison-Wesley, 1995, ISBN 0-201-63354-X from Richard Stevens"
31 * http://www.kohala.com/start/tcpipiv2.html
35 #include <rte_compat.h>
36 #include <rte_common.h>
37 #include <rte_config.h>
38 #include <rte_mempool.h>
39 #include <rte_memory.h>
40 #include <rte_atomic.h>
41 #include <rte_prefetch.h>
42 #include <rte_branch_prediction.h>
43 #include <rte_mbuf_ptype.h>
50 * Packet Offload Features Flags. It also carry packet type information.
51 * Critical resources. Both rx/tx shared these bits. Be cautious on any change
53 * - RX flags start at bit position zero, and get added to the left of previous
55 * - The most-significant 3 bits are reserved for generic mbuf flags
56 * - TX flags therefore start at bit position 60 (i.e. 63-3), and new flags get
57 * added to the right of the previously defined flags i.e. they should count
58 * downwards, not upwards.
60 * Keep these flags synchronized with rte_get_rx_ol_flag_name() and
61 * rte_get_tx_ol_flag_name().
65 * The RX packet is a 802.1q VLAN packet, and the tci has been
66 * saved in in mbuf->vlan_tci.
67 * If the flag PKT_RX_VLAN_STRIPPED is also present, the VLAN
68 * header has been stripped from mbuf data, else it is still
71 #define PKT_RX_VLAN (1ULL << 0)
73 #define PKT_RX_RSS_HASH (1ULL << 1) /**< RX packet with RSS hash result. */
74 #define PKT_RX_FDIR (1ULL << 2) /**< RX packet with FDIR match indicate. */
78 * Checking this flag alone is deprecated: check the 2 bits of
79 * PKT_RX_L4_CKSUM_MASK.
80 * This flag was set when the L4 checksum of a packet was detected as
81 * wrong by the hardware.
83 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3)
87 * Checking this flag alone is deprecated: check the 2 bits of
88 * PKT_RX_IP_CKSUM_MASK.
89 * This flag was set when the IP checksum of a packet was detected as
90 * wrong by the hardware.
92 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4)
94 #define PKT_RX_EIP_CKSUM_BAD (1ULL << 5) /**< External IP header checksum error. */
97 * A vlan has been stripped by the hardware and its tci is saved in
98 * mbuf->vlan_tci. This can only happen if vlan stripping is enabled
99 * in the RX configuration of the PMD.
100 * When PKT_RX_VLAN_STRIPPED is set, PKT_RX_VLAN must also be set.
102 #define PKT_RX_VLAN_STRIPPED (1ULL << 6)
105 * Mask of bits used to determine the status of RX IP checksum.
106 * - PKT_RX_IP_CKSUM_UNKNOWN: no information about the RX IP checksum
107 * - PKT_RX_IP_CKSUM_BAD: the IP checksum in the packet is wrong
108 * - PKT_RX_IP_CKSUM_GOOD: the IP checksum in the packet is valid
109 * - PKT_RX_IP_CKSUM_NONE: the IP checksum is not correct in the packet
110 * data, but the integrity of the IP header is verified.
112 #define PKT_RX_IP_CKSUM_MASK ((1ULL << 4) | (1ULL << 7))
114 #define PKT_RX_IP_CKSUM_UNKNOWN 0
115 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4)
116 #define PKT_RX_IP_CKSUM_GOOD (1ULL << 7)
117 #define PKT_RX_IP_CKSUM_NONE ((1ULL << 4) | (1ULL << 7))
120 * Mask of bits used to determine the status of RX L4 checksum.
121 * - PKT_RX_L4_CKSUM_UNKNOWN: no information about the RX L4 checksum
122 * - PKT_RX_L4_CKSUM_BAD: the L4 checksum in the packet is wrong
123 * - PKT_RX_L4_CKSUM_GOOD: the L4 checksum in the packet is valid
124 * - PKT_RX_L4_CKSUM_NONE: the L4 checksum is not correct in the packet
125 * data, but the integrity of the L4 data is verified.
127 #define PKT_RX_L4_CKSUM_MASK ((1ULL << 3) | (1ULL << 8))
129 #define PKT_RX_L4_CKSUM_UNKNOWN 0
130 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3)
131 #define PKT_RX_L4_CKSUM_GOOD (1ULL << 8)
132 #define PKT_RX_L4_CKSUM_NONE ((1ULL << 3) | (1ULL << 8))
134 #define PKT_RX_IEEE1588_PTP (1ULL << 9) /**< RX IEEE1588 L2 Ethernet PT Packet. */
135 #define PKT_RX_IEEE1588_TMST (1ULL << 10) /**< RX IEEE1588 L2/L4 timestamped packet.*/
136 #define PKT_RX_FDIR_ID (1ULL << 13) /**< FD id reported if FDIR match. */
137 #define PKT_RX_FDIR_FLX (1ULL << 14) /**< Flexible bytes reported if FDIR match. */
140 * The 2 vlans have been stripped by the hardware and their tci are
141 * saved in mbuf->vlan_tci (inner) and mbuf->vlan_tci_outer (outer).
142 * This can only happen if vlan stripping is enabled in the RX
143 * configuration of the PMD. If this flag is set,
144 * When PKT_RX_QINQ_STRIPPED is set, the flags (PKT_RX_VLAN |
145 * PKT_RX_VLAN_STRIPPED | PKT_RX_QINQ) must also be set.
147 #define PKT_RX_QINQ_STRIPPED (1ULL << 15)
150 * When packets are coalesced by a hardware or virtual driver, this flag
151 * can be set in the RX mbuf, meaning that the m->tso_segsz field is
152 * valid and is set to the segment size of original packets.
154 #define PKT_RX_LRO (1ULL << 16)
157 * Indicate that the timestamp field in the mbuf is valid.
159 #define PKT_RX_TIMESTAMP (1ULL << 17)
162 * Indicate that security offload processing was applied on the RX packet.
164 #define PKT_RX_SEC_OFFLOAD (1ULL << 18)
167 * Indicate that security offload processing failed on the RX packet.
169 #define PKT_RX_SEC_OFFLOAD_FAILED (1ULL << 19)
172 * The RX packet is a double VLAN, and the outer tci has been
173 * saved in in mbuf->vlan_tci_outer.
174 * If the flag PKT_RX_QINQ_STRIPPED is also present, both VLANs
175 * headers have been stripped from mbuf data, else they are still
178 #define PKT_RX_QINQ (1ULL << 20)
180 /* add new RX flags here */
182 /* add new TX flags here */
185 * UDP Fragmentation Offload flag. This flag is used for enabling UDP
186 * fragmentation in SW or in HW. When use UFO, mbuf->tso_segsz is used
187 * to store the MSS of UDP fragments.
189 #define PKT_TX_UDP_SEG (1ULL << 42)
192 * Request security offload processing on the TX packet.
194 #define PKT_TX_SEC_OFFLOAD (1ULL << 43)
197 * Offload the MACsec. This flag must be set by the application to enable
198 * this offload feature for a packet to be transmitted.
200 #define PKT_TX_MACSEC (1ULL << 44)
203 * Bits 45:48 used for the tunnel type.
204 * The tunnel type must be specified for TSO or checksum on the inner part
206 * These flags can be used with PKT_TX_TCP_SEG for TSO, or PKT_TX_xxx_CKSUM.
207 * The mbuf fields for inner and outer header lengths are required:
208 * outer_l2_len, outer_l3_len, l2_len, l3_len, l4_len and tso_segsz for TSO.
210 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
211 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
212 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
213 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
214 /** TX packet with MPLS-in-UDP RFC 7510 header. */
215 #define PKT_TX_TUNNEL_MPLSINUDP (0x5ULL << 45)
216 /* add new TX TUNNEL type here */
217 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
220 * Second VLAN insertion (QinQ) flag.
222 #define PKT_TX_QINQ (1ULL << 49) /**< TX packet with double VLAN inserted. */
223 /* this old name is deprecated */
224 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
227 * TCP segmentation offload. To enable this offload feature for a
228 * packet to be transmitted on hardware supporting TSO:
229 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
231 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
232 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag
233 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
235 #define PKT_TX_TCP_SEG (1ULL << 50)
237 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
240 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
241 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
242 * L4 checksum offload, the user needs to:
243 * - fill l2_len and l3_len in mbuf
244 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
245 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
247 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
248 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
249 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
250 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
251 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
254 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
255 * also be set by the application, although a PMD will only check
257 * - fill the mbuf offload information: l2_len, l3_len
259 #define PKT_TX_IP_CKSUM (1ULL << 54)
262 * Packet is IPv4. This flag must be set when using any offload feature
263 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
264 * packet. If the packet is a tunneled packet, this flag is related to
267 #define PKT_TX_IPV4 (1ULL << 55)
270 * Packet is IPv6. This flag must be set when using an offload feature
271 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
272 * packet. If the packet is a tunneled packet, this flag is related to
275 #define PKT_TX_IPV6 (1ULL << 56)
278 * TX packet is a 802.1q VLAN packet.
280 #define PKT_TX_VLAN (1ULL << 57)
281 /* this old name is deprecated */
282 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
285 * Offload the IP checksum of an external header in the hardware. The
286 * flag PKT_TX_OUTER_IPV4 should also be set by the application, although
287 * a PMD will only check PKT_TX_OUTER_IP_CKSUM.
288 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
290 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
293 * Packet outer header is IPv4. This flag must be set when using any
294 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
295 * outer header of the tunneled packet is an IPv4 packet.
297 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
300 * Packet outer header is IPv6. This flag must be set when using any
301 * outer offload feature (L4 checksum) to tell the NIC that the outer
302 * header of the tunneled packet is an IPv6 packet.
304 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
307 * Bitmask of all supported packet Tx offload features flags,
308 * which can be set for packet.
310 #define PKT_TX_OFFLOAD_MASK ( \
313 PKT_TX_OUTER_IP_CKSUM | \
315 PKT_TX_IEEE1588_TMST | \
318 PKT_TX_TUNNEL_MASK | \
322 #define __RESERVED (1ULL << 61) /**< reserved for future mbuf use */
324 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
326 /** Alignment constraint of mbuf private area. */
327 #define RTE_MBUF_PRIV_ALIGN 8
330 * Get the name of a RX offload flag
333 * The mask describing the flag.
335 * The name of this flag, or NULL if it's not a valid RX flag.
337 const char *rte_get_rx_ol_flag_name(uint64_t mask);
340 * Dump the list of RX offload flags in a buffer
343 * The mask describing the RX flags.
347 * The length of the buffer.
349 * 0 on success, (-1) on error.
351 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
354 * Get the name of a TX offload flag
357 * The mask describing the flag. Usually only one bit must be set.
358 * Several bits can be given if they belong to the same mask.
359 * Ex: PKT_TX_L4_MASK.
361 * The name of this flag, or NULL if it's not a valid TX flag.
363 const char *rte_get_tx_ol_flag_name(uint64_t mask);
366 * Dump the list of TX offload flags in a buffer
369 * The mask describing the TX flags.
373 * The length of the buffer.
375 * 0 on success, (-1) on error.
377 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
380 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
381 * splitting it into multiple segments.
382 * So, for mbufs that planned to be involved into RX/TX, the recommended
383 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
385 #define RTE_MBUF_DEFAULT_DATAROOM 2048
386 #define RTE_MBUF_DEFAULT_BUF_SIZE \
387 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
389 /* define a set of marker types that can be used to refer to set points in the
392 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
394 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
396 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
397 * with a single assignment */
400 * The generic rte_mbuf, containing a packet mbuf.
405 void *buf_addr; /**< Virtual address of segment buffer. */
407 * Physical address of segment buffer.
408 * Force alignment to 8-bytes, so as to ensure we have the exact
409 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
410 * working on vector drivers easier.
415 rte_iova_t buf_physaddr; /**< deprecated */
416 } __rte_aligned(sizeof(rte_iova_t));
418 /* next 8 bytes are initialised on RX descriptor rearm */
423 * Reference counter. Its size should at least equal to the size
424 * of port field (16 bits), to support zero-copy broadcast.
425 * It should only be accessed using the following functions:
426 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
427 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
428 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
433 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
434 uint16_t refcnt; /**< Non-atomically accessed refcnt */
436 uint16_t nb_segs; /**< Number of segments. */
438 /** Input port (16 bits to support more than 256 virtual ports). */
441 uint64_t ol_flags; /**< Offload features. */
443 /* remaining bytes are set on RX when pulling packet from descriptor */
444 MARKER rx_descriptor_fields1;
447 * The packet type, which is the combination of outer/inner L2, L3, L4
448 * and tunnel types. The packet_type is about data really present in the
449 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
450 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
451 * vlan is stripped from the data.
455 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
457 uint32_t l2_type:4; /**< (Outer) L2 type. */
458 uint32_t l3_type:4; /**< (Outer) L3 type. */
459 uint32_t l4_type:4; /**< (Outer) L4 type. */
460 uint32_t tun_type:4; /**< Tunnel type. */
463 uint8_t inner_esp_next_proto;
464 /**< ESP next protocol type, valid if
465 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
470 uint8_t inner_l2_type:4;
471 /**< Inner L2 type. */
472 uint8_t inner_l3_type:4;
473 /**< Inner L3 type. */
476 uint32_t inner_l4_type:4; /**< Inner L4 type. */
480 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
481 uint16_t data_len; /**< Amount of data in segment buffer. */
482 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
486 uint32_t rss; /**< RSS hash result if RSS enabled */
495 /**< Second 4 flexible bytes */
498 /**< First 4 flexible bytes or FD ID, dependent on
499 PKT_RX_FDIR_* flag in ol_flags. */
500 } fdir; /**< Filter identifier if FDIR enabled */
504 } sched; /**< Hierarchical scheduler */
505 uint32_t usr; /**< User defined tags. See rte_distributor_process() */
506 } hash; /**< hash information */
508 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
509 uint16_t vlan_tci_outer;
511 uint16_t buf_len; /**< Length of segment buffer. */
513 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
514 * are not normalized but are always the same for a given port.
518 /* second cache line - fields only used in slow path or on TX */
519 MARKER cacheline1 __rte_cache_min_aligned;
523 void *userdata; /**< Can be used for external metadata */
524 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
527 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
528 struct rte_mbuf *next; /**< Next segment of scattered packet. */
530 /* fields to support TX offloads */
533 uint64_t tx_offload; /**< combined for easy fetch */
537 /**< L2 (MAC) Header Length for non-tunneling pkt.
538 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
540 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
541 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
542 uint64_t tso_segsz:16; /**< TCP TSO segment size */
544 /* fields for TX offloading of tunnels */
545 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
546 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
548 /* uint64_t unused:8; */
552 /** Size of the application private data. In case of an indirect
553 * mbuf, it stores the direct mbuf private data size. */
556 /** Timesync flags for use with IEEE1588. */
559 /** Sequence number. See also rte_reorder_insert(). */
562 } __rte_cache_aligned;
564 /**< Maximum number of nb_segs allowed. */
565 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
568 * Prefetch the first part of the mbuf
570 * The first 64 bytes of the mbuf corresponds to fields that are used early
571 * in the receive path. If the cache line of the architecture is higher than
572 * 64B, the second part will also be prefetched.
575 * The pointer to the mbuf.
578 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
580 rte_prefetch0(&m->cacheline0);
584 * Prefetch the second part of the mbuf
586 * The next 64 bytes of the mbuf corresponds to fields that are used in the
587 * transmit path. If the cache line of the architecture is higher than 64B,
588 * this function does nothing as it is expected that the full mbuf is
592 * The pointer to the mbuf.
595 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
597 #if RTE_CACHE_LINE_SIZE == 64
598 rte_prefetch0(&m->cacheline1);
605 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
608 * Return the IO address of the beginning of the mbuf data
611 * The pointer to the mbuf.
613 * The IO address of the beginning of the mbuf data
615 static inline rte_iova_t
616 rte_mbuf_data_iova(const struct rte_mbuf *mb)
618 return mb->buf_iova + mb->data_off;
622 static inline phys_addr_t
623 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
625 return rte_mbuf_data_iova(mb);
629 * Return the default IO address of the beginning of the mbuf data
631 * This function is used by drivers in their receive function, as it
632 * returns the location where data should be written by the NIC, taking
633 * the default headroom in account.
636 * The pointer to the mbuf.
638 * The IO address of the beginning of the mbuf data
640 static inline rte_iova_t
641 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
643 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
647 static inline phys_addr_t
648 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
650 return rte_mbuf_data_iova_default(mb);
654 * Return the mbuf owning the data buffer address of an indirect mbuf.
657 * The pointer to the indirect mbuf.
659 * The address of the direct mbuf corresponding to buffer_addr.
661 static inline struct rte_mbuf *
662 rte_mbuf_from_indirect(struct rte_mbuf *mi)
664 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
668 * Return the buffer address embedded in the given mbuf.
671 * The pointer to the mbuf.
673 * The address of the data buffer owned by the mbuf.
676 rte_mbuf_to_baddr(struct rte_mbuf *md)
679 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
684 * Returns TRUE if given mbuf is indirect, or FALSE otherwise.
686 #define RTE_MBUF_INDIRECT(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
689 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
691 #define RTE_MBUF_DIRECT(mb) (!RTE_MBUF_INDIRECT(mb))
694 * Private data in case of pktmbuf pool.
696 * A structure that contains some pktmbuf_pool-specific data that are
697 * appended after the mempool structure (in private data).
699 struct rte_pktmbuf_pool_private {
700 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
701 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
704 #ifdef RTE_LIBRTE_MBUF_DEBUG
706 /** check mbuf type in debug mode */
707 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
709 #else /* RTE_LIBRTE_MBUF_DEBUG */
711 /** check mbuf type in debug mode */
712 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
714 #endif /* RTE_LIBRTE_MBUF_DEBUG */
716 #ifdef RTE_MBUF_REFCNT_ATOMIC
719 * Reads the value of an mbuf's refcnt.
723 * Reference count number.
725 static inline uint16_t
726 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
728 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
732 * Sets an mbuf's refcnt to a defined value.
739 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
741 rte_atomic16_set(&m->refcnt_atomic, new_value);
745 static inline uint16_t
746 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
748 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
752 * Adds given value to an mbuf's refcnt and returns its new value.
756 * Value to add/subtract
760 static inline uint16_t
761 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
764 * The atomic_add is an expensive operation, so we don't want to
765 * call it in the case where we know we are the uniq holder of
766 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
767 * operation has to be used because concurrent accesses on the
768 * reference counter can occur.
770 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
771 rte_mbuf_refcnt_set(m, 1 + value);
775 return __rte_mbuf_refcnt_update(m, value);
778 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
781 static inline uint16_t
782 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
784 m->refcnt = (uint16_t)(m->refcnt + value);
789 * Adds given value to an mbuf's refcnt and returns its new value.
791 static inline uint16_t
792 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
794 return __rte_mbuf_refcnt_update(m, value);
798 * Reads the value of an mbuf's refcnt.
800 static inline uint16_t
801 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
807 * Sets an mbuf's refcnt to the defined value.
810 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
812 m->refcnt = new_value;
815 #endif /* RTE_MBUF_REFCNT_ATOMIC */
818 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
825 * Sanity checks on an mbuf.
827 * Check the consistency of the given mbuf. The function will cause a
828 * panic if corruption is detected.
831 * The mbuf to be checked.
833 * True if the mbuf is a packet header, false if it is a sub-segment
834 * of a packet (in this case, some fields like nb_segs are not checked)
837 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
839 #define MBUF_RAW_ALLOC_CHECK(m) do { \
840 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
841 RTE_ASSERT((m)->next == NULL); \
842 RTE_ASSERT((m)->nb_segs == 1); \
843 __rte_mbuf_sanity_check(m, 0); \
847 * Allocate an uninitialized mbuf from mempool *mp*.
849 * This function can be used by PMDs (especially in RX functions) to
850 * allocate an uninitialized mbuf. The driver is responsible of
851 * initializing all the required fields. See rte_pktmbuf_reset().
852 * For standard needs, prefer rte_pktmbuf_alloc().
854 * The caller can expect that the following fields of the mbuf structure
855 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
856 * next=NULL, pool, priv_size. The other fields must be initialized
860 * The mempool from which mbuf is allocated.
862 * - The pointer to the new mbuf on success.
863 * - NULL if allocation failed.
865 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
869 if (rte_mempool_get(mp, (void **)&m) < 0)
871 MBUF_RAW_ALLOC_CHECK(m);
876 * Put mbuf back into its original mempool.
878 * The caller must ensure that the mbuf is direct and properly
879 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
880 * rte_pktmbuf_prefree_seg().
882 * This function should be used with care, when optimization is
883 * required. For standard needs, prefer rte_pktmbuf_free() or
884 * rte_pktmbuf_free_seg().
887 * The mbuf to be freed.
889 static __rte_always_inline void
890 rte_mbuf_raw_free(struct rte_mbuf *m)
892 RTE_ASSERT(RTE_MBUF_DIRECT(m));
893 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
894 RTE_ASSERT(m->next == NULL);
895 RTE_ASSERT(m->nb_segs == 1);
896 __rte_mbuf_sanity_check(m, 0);
897 rte_mempool_put(m->pool, m);
900 /* compat with older versions */
903 __rte_mbuf_raw_free(struct rte_mbuf *m)
905 rte_mbuf_raw_free(m);
909 * The packet mbuf constructor.
911 * This function initializes some fields in the mbuf structure that are
912 * not modified by the user once created (origin pool, buffer start
913 * address, and so on). This function is given as a callback function to
914 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
917 * The mempool from which mbufs originate.
919 * A pointer that can be used by the user to retrieve useful information
920 * for mbuf initialization. This pointer is the opaque argument passed to
921 * rte_mempool_obj_iter() or rte_mempool_create().
923 * The mbuf to initialize.
925 * The index of the mbuf in the pool table.
927 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
928 void *m, unsigned i);
932 * A packet mbuf pool constructor.
934 * This function initializes the mempool private data in the case of a
935 * pktmbuf pool. This private data is needed by the driver. The
936 * function must be called on the mempool before it is used, or it
937 * can be given as a callback function to rte_mempool_create() at
938 * pool creation. It can be extended by the user, for example, to
939 * provide another packet size.
942 * The mempool from which mbufs originate.
944 * A pointer that can be used by the user to retrieve useful information
945 * for mbuf initialization. This pointer is the opaque argument passed to
946 * rte_mempool_create().
948 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
951 * Create a mbuf pool.
953 * This function creates and initializes a packet mbuf pool. It is
954 * a wrapper to rte_mempool functions.
957 * The name of the mbuf pool.
959 * The number of elements in the mbuf pool. The optimum size (in terms
960 * of memory usage) for a mempool is when n is a power of two minus one:
963 * Size of the per-core object cache. See rte_mempool_create() for
966 * Size of application private are between the rte_mbuf structure
967 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
968 * @param data_room_size
969 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
971 * The socket identifier where the memory should be allocated. The
972 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
975 * The pointer to the new allocated mempool, on success. NULL on error
976 * with rte_errno set appropriately. Possible rte_errno values include:
977 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
978 * - E_RTE_SECONDARY - function was called from a secondary process instance
979 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
980 * - ENOSPC - the maximum number of memzones has already been allocated
981 * - EEXIST - a memzone with the same name already exists
982 * - ENOMEM - no appropriate memory area found in which to create memzone
985 rte_pktmbuf_pool_create(const char *name, unsigned n,
986 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
990 * Create a mbuf pool with a given mempool ops name
992 * This function creates and initializes a packet mbuf pool. It is
993 * a wrapper to rte_mempool functions.
996 * The name of the mbuf pool.
998 * The number of elements in the mbuf pool. The optimum size (in terms
999 * of memory usage) for a mempool is when n is a power of two minus one:
1002 * Size of the per-core object cache. See rte_mempool_create() for
1005 * Size of application private are between the rte_mbuf structure
1006 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1007 * @param data_room_size
1008 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1010 * The socket identifier where the memory should be allocated. The
1011 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1014 * The mempool ops name to be used for this mempool instead of
1015 * default mempool. The value can be *NULL* to use default mempool.
1017 * The pointer to the new allocated mempool, on success. NULL on error
1018 * with rte_errno set appropriately. Possible rte_errno values include:
1019 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1020 * - E_RTE_SECONDARY - function was called from a secondary process instance
1021 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1022 * - ENOSPC - the maximum number of memzones has already been allocated
1023 * - EEXIST - a memzone with the same name already exists
1024 * - ENOMEM - no appropriate memory area found in which to create memzone
1026 struct rte_mempool * __rte_experimental
1027 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1028 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1029 int socket_id, const char *ops_name);
1032 * Get the data room size of mbufs stored in a pktmbuf_pool
1034 * The data room size is the amount of data that can be stored in a
1035 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1038 * The packet mbuf pool.
1040 * The data room size of mbufs stored in this mempool.
1042 static inline uint16_t
1043 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1045 struct rte_pktmbuf_pool_private *mbp_priv;
1047 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1048 return mbp_priv->mbuf_data_room_size;
1052 * Get the application private size of mbufs stored in a pktmbuf_pool
1054 * The private size of mbuf is a zone located between the rte_mbuf
1055 * structure and the data buffer where an application can store data
1056 * associated to a packet.
1059 * The packet mbuf pool.
1061 * The private size of mbufs stored in this mempool.
1063 static inline uint16_t
1064 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1066 struct rte_pktmbuf_pool_private *mbp_priv;
1068 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1069 return mbp_priv->mbuf_priv_size;
1073 * Reset the data_off field of a packet mbuf to its default value.
1075 * The given mbuf must have only one segment, which should be empty.
1078 * The packet mbuf's data_off field has to be reset.
1080 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1082 m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, (uint16_t)m->buf_len);
1086 * Reset the fields of a packet mbuf to their default values.
1088 * The given mbuf must have only one segment.
1091 * The packet mbuf to be resetted.
1093 #define MBUF_INVALID_PORT UINT16_MAX
1095 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1101 m->vlan_tci_outer = 0;
1103 m->port = MBUF_INVALID_PORT;
1107 rte_pktmbuf_reset_headroom(m);
1110 __rte_mbuf_sanity_check(m, 1);
1114 * Allocate a new mbuf from a mempool.
1116 * This new mbuf contains one segment, which has a length of 0. The pointer
1117 * to data is initialized to have some bytes of headroom in the buffer
1118 * (if buffer size allows).
1121 * The mempool from which the mbuf is allocated.
1123 * - The pointer to the new mbuf on success.
1124 * - NULL if allocation failed.
1126 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1129 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1130 rte_pktmbuf_reset(m);
1135 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1139 * The mempool from which mbufs are allocated.
1141 * Array of pointers to mbufs
1146 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1148 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1149 struct rte_mbuf **mbufs, unsigned count)
1154 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1158 /* To understand duff's device on loop unwinding optimization, see
1159 * https://en.wikipedia.org/wiki/Duff's_device.
1160 * Here while() loop is used rather than do() while{} to avoid extra
1161 * check if count is zero.
1163 switch (count % 4) {
1165 while (idx != count) {
1166 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1167 rte_pktmbuf_reset(mbufs[idx]);
1171 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1172 rte_pktmbuf_reset(mbufs[idx]);
1176 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1177 rte_pktmbuf_reset(mbufs[idx]);
1181 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1182 rte_pktmbuf_reset(mbufs[idx]);
1191 * Attach packet mbuf to another packet mbuf.
1193 * After attachment we refer the mbuf we attached as 'indirect',
1194 * while mbuf we attached to as 'direct'.
1195 * The direct mbuf's reference counter is incremented.
1197 * Right now, not supported:
1198 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1199 * - mbuf we trying to attach (mi) is used by someone else
1200 * e.g. it's reference counter is greater then 1.
1203 * The indirect packet mbuf.
1205 * The packet mbuf we're attaching to.
1207 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1209 struct rte_mbuf *md;
1211 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1212 rte_mbuf_refcnt_read(mi) == 1);
1214 /* if m is not direct, get the mbuf that embeds the data */
1215 if (RTE_MBUF_DIRECT(m))
1218 md = rte_mbuf_from_indirect(m);
1220 rte_mbuf_refcnt_update(md, 1);
1221 mi->priv_size = m->priv_size;
1222 mi->buf_iova = m->buf_iova;
1223 mi->buf_addr = m->buf_addr;
1224 mi->buf_len = m->buf_len;
1226 mi->data_off = m->data_off;
1227 mi->data_len = m->data_len;
1229 mi->vlan_tci = m->vlan_tci;
1230 mi->vlan_tci_outer = m->vlan_tci_outer;
1231 mi->tx_offload = m->tx_offload;
1235 mi->pkt_len = mi->data_len;
1237 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1238 mi->packet_type = m->packet_type;
1239 mi->timestamp = m->timestamp;
1241 __rte_mbuf_sanity_check(mi, 1);
1242 __rte_mbuf_sanity_check(m, 0);
1246 * Detach an indirect packet mbuf.
1248 * - restore original mbuf address and length values.
1249 * - reset pktmbuf data and data_len to their default values.
1250 * - decrement the direct mbuf's reference counter. When the
1251 * reference counter becomes 0, the direct mbuf is freed.
1253 * All other fields of the given packet mbuf will be left intact.
1256 * The indirect attached packet mbuf.
1258 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1260 struct rte_mbuf *md = rte_mbuf_from_indirect(m);
1261 struct rte_mempool *mp = m->pool;
1262 uint32_t mbuf_size, buf_len, priv_size;
1264 priv_size = rte_pktmbuf_priv_size(mp);
1265 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
1266 buf_len = rte_pktmbuf_data_room_size(mp);
1268 m->priv_size = priv_size;
1269 m->buf_addr = (char *)m + mbuf_size;
1270 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1271 m->buf_len = (uint16_t)buf_len;
1272 rte_pktmbuf_reset_headroom(m);
1276 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1279 rte_mbuf_refcnt_set(md, 1);
1280 rte_mbuf_raw_free(md);
1285 * Decrease reference counter and unlink a mbuf segment
1287 * This function does the same than a free, except that it does not
1288 * return the segment to its pool.
1289 * It decreases the reference counter, and if it reaches 0, it is
1290 * detached from its parent for an indirect mbuf.
1293 * The mbuf to be unlinked
1295 * - (m) if it is the last reference. It can be recycled or freed.
1296 * - (NULL) if the mbuf still has remaining references on it.
1298 static __rte_always_inline struct rte_mbuf *
1299 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1301 __rte_mbuf_sanity_check(m, 0);
1303 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1305 if (RTE_MBUF_INDIRECT(m))
1306 rte_pktmbuf_detach(m);
1308 if (m->next != NULL) {
1315 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1317 if (RTE_MBUF_INDIRECT(m))
1318 rte_pktmbuf_detach(m);
1320 if (m->next != NULL) {
1324 rte_mbuf_refcnt_set(m, 1);
1331 /* deprecated, replaced by rte_pktmbuf_prefree_seg() */
1333 static inline struct rte_mbuf *
1334 __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1336 return rte_pktmbuf_prefree_seg(m);
1340 * Free a segment of a packet mbuf into its original mempool.
1342 * Free an mbuf, without parsing other segments in case of chained
1346 * The packet mbuf segment to be freed.
1348 static __rte_always_inline void
1349 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1351 m = rte_pktmbuf_prefree_seg(m);
1352 if (likely(m != NULL))
1353 rte_mbuf_raw_free(m);
1357 * Free a packet mbuf back into its original mempool.
1359 * Free an mbuf, and all its segments in case of chained buffers. Each
1360 * segment is added back into its original mempool.
1363 * The packet mbuf to be freed. If NULL, the function does nothing.
1365 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1367 struct rte_mbuf *m_next;
1370 __rte_mbuf_sanity_check(m, 1);
1374 rte_pktmbuf_free_seg(m);
1380 * Creates a "clone" of the given packet mbuf.
1382 * Walks through all segments of the given packet mbuf, and for each of them:
1383 * - Creates a new packet mbuf from the given pool.
1384 * - Attaches newly created mbuf to the segment.
1385 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1386 * from the original packet mbuf.
1389 * The packet mbuf to be cloned.
1391 * The mempool from which the "clone" mbufs are allocated.
1393 * - The pointer to the new "clone" mbuf on success.
1394 * - NULL if allocation fails.
1396 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1397 struct rte_mempool *mp)
1399 struct rte_mbuf *mc, *mi, **prev;
1403 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1408 pktlen = md->pkt_len;
1413 rte_pktmbuf_attach(mi, md);
1416 } while ((md = md->next) != NULL &&
1417 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1421 mc->pkt_len = pktlen;
1423 /* Allocation of new indirect segment failed */
1424 if (unlikely (mi == NULL)) {
1425 rte_pktmbuf_free(mc);
1429 __rte_mbuf_sanity_check(mc, 1);
1434 * Adds given value to the refcnt of all packet mbuf segments.
1436 * Walks through all segments of given packet mbuf and for each of them
1437 * invokes rte_mbuf_refcnt_update().
1440 * The packet mbuf whose refcnt to be updated.
1442 * The value to add to the mbuf's segments refcnt.
1444 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1446 __rte_mbuf_sanity_check(m, 1);
1449 rte_mbuf_refcnt_update(m, v);
1450 } while ((m = m->next) != NULL);
1454 * Get the headroom in a packet mbuf.
1459 * The length of the headroom.
1461 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1463 __rte_mbuf_sanity_check(m, 0);
1468 * Get the tailroom of a packet mbuf.
1473 * The length of the tailroom.
1475 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1477 __rte_mbuf_sanity_check(m, 0);
1478 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1483 * Get the last segment of the packet.
1488 * The last segment of the given mbuf.
1490 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1492 __rte_mbuf_sanity_check(m, 1);
1493 while (m->next != NULL)
1499 * A macro that points to an offset into the data in the mbuf.
1501 * The returned pointer is cast to type t. Before using this
1502 * function, the user must ensure that the first segment is large
1503 * enough to accommodate its data.
1508 * The offset into the mbuf data.
1510 * The type to cast the result into.
1512 #define rte_pktmbuf_mtod_offset(m, t, o) \
1513 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1516 * A macro that points to the start of the data in the mbuf.
1518 * The returned pointer is cast to type t. Before using this
1519 * function, the user must ensure that the first segment is large
1520 * enough to accommodate its data.
1525 * The type to cast the result into.
1527 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1530 * A macro that returns the IO address that points to an offset of the
1531 * start of the data in the mbuf
1536 * The offset into the data to calculate address from.
1538 #define rte_pktmbuf_iova_offset(m, o) \
1539 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
1542 #define rte_pktmbuf_mtophys_offset(m, o) \
1543 rte_pktmbuf_iova_offset(m, o)
1546 * A macro that returns the IO address that points to the start of the
1552 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
1555 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1558 * A macro that returns the length of the packet.
1560 * The value can be read or assigned.
1565 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1568 * A macro that returns the length of the segment.
1570 * The value can be read or assigned.
1575 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1578 * Prepend len bytes to an mbuf data area.
1580 * Returns a pointer to the new
1581 * data start address. If there is not enough headroom in the first
1582 * segment, the function will return NULL, without modifying the mbuf.
1587 * The amount of data to prepend (in bytes).
1589 * A pointer to the start of the newly prepended data, or
1590 * NULL if there is not enough headroom space in the first segment
1592 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1595 __rte_mbuf_sanity_check(m, 1);
1597 if (unlikely(len > rte_pktmbuf_headroom(m)))
1601 m->data_len = (uint16_t)(m->data_len + len);
1602 m->pkt_len = (m->pkt_len + len);
1604 return (char *)m->buf_addr + m->data_off;
1608 * Append len bytes to an mbuf.
1610 * Append len bytes to an mbuf and return a pointer to the start address
1611 * of the added data. If there is not enough tailroom in the last
1612 * segment, the function will return NULL, without modifying the mbuf.
1617 * The amount of data to append (in bytes).
1619 * A pointer to the start of the newly appended data, or
1620 * NULL if there is not enough tailroom space in the last segment
1622 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1625 struct rte_mbuf *m_last;
1627 __rte_mbuf_sanity_check(m, 1);
1629 m_last = rte_pktmbuf_lastseg(m);
1630 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1633 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1634 m_last->data_len = (uint16_t)(m_last->data_len + len);
1635 m->pkt_len = (m->pkt_len + len);
1636 return (char*) tail;
1640 * Remove len bytes at the beginning of an mbuf.
1642 * Returns a pointer to the start address of the new data area. If the
1643 * length is greater than the length of the first segment, then the
1644 * function will fail and return NULL, without modifying the mbuf.
1649 * The amount of data to remove (in bytes).
1651 * A pointer to the new start of the data.
1653 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1655 __rte_mbuf_sanity_check(m, 1);
1657 if (unlikely(len > m->data_len))
1660 m->data_len = (uint16_t)(m->data_len - len);
1662 m->pkt_len = (m->pkt_len - len);
1663 return (char *)m->buf_addr + m->data_off;
1667 * Remove len bytes of data at the end of the mbuf.
1669 * If the length is greater than the length of the last segment, the
1670 * function will fail and return -1 without modifying the mbuf.
1675 * The amount of data to remove (in bytes).
1680 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
1682 struct rte_mbuf *m_last;
1684 __rte_mbuf_sanity_check(m, 1);
1686 m_last = rte_pktmbuf_lastseg(m);
1687 if (unlikely(len > m_last->data_len))
1690 m_last->data_len = (uint16_t)(m_last->data_len - len);
1691 m->pkt_len = (m->pkt_len - len);
1696 * Test if mbuf data is contiguous.
1701 * - 1, if all data is contiguous (one segment).
1702 * - 0, if there is several segments.
1704 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
1706 __rte_mbuf_sanity_check(m, 1);
1707 return !!(m->nb_segs == 1);
1711 * @internal used by rte_pktmbuf_read().
1713 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
1714 uint32_t len, void *buf);
1717 * Read len data bytes in a mbuf at specified offset.
1719 * If the data is contiguous, return the pointer in the mbuf data, else
1720 * copy the data in the buffer provided by the user and return its
1724 * The pointer to the mbuf.
1726 * The offset of the data in the mbuf.
1728 * The amount of bytes to read.
1730 * The buffer where data is copied if it is not contiguous in mbuf
1731 * data. Its length should be at least equal to the len parameter.
1733 * The pointer to the data, either in the mbuf if it is contiguous,
1734 * or in the user buffer. If mbuf is too small, NULL is returned.
1736 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
1737 uint32_t off, uint32_t len, void *buf)
1739 if (likely(off + len <= rte_pktmbuf_data_len(m)))
1740 return rte_pktmbuf_mtod_offset(m, char *, off);
1742 return __rte_pktmbuf_read(m, off, len, buf);
1746 * Chain an mbuf to another, thereby creating a segmented packet.
1748 * Note: The implementation will do a linear walk over the segments to find
1749 * the tail entry. For cases when there are many segments, it's better to
1750 * chain the entries manually.
1753 * The head of the mbuf chain (the first packet)
1755 * The mbuf to put last in the chain
1759 * - -EOVERFLOW, if the chain segment limit exceeded
1761 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
1763 struct rte_mbuf *cur_tail;
1765 /* Check for number-of-segments-overflow */
1766 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
1769 /* Chain 'tail' onto the old tail */
1770 cur_tail = rte_pktmbuf_lastseg(head);
1771 cur_tail->next = tail;
1773 /* accumulate number of segments and total length. */
1774 head->nb_segs += tail->nb_segs;
1775 head->pkt_len += tail->pkt_len;
1777 /* pkt_len is only set in the head */
1778 tail->pkt_len = tail->data_len;
1784 * Validate general requirements for Tx offload in mbuf.
1786 * This function checks correctness and completeness of Tx offload settings.
1789 * The packet mbuf to be validated.
1791 * 0 if packet is valid
1794 rte_validate_tx_offload(const struct rte_mbuf *m)
1796 uint64_t ol_flags = m->ol_flags;
1797 uint64_t inner_l3_offset = m->l2_len;
1799 /* Does packet set any of available offloads? */
1800 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
1803 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
1804 inner_l3_offset += m->outer_l2_len + m->outer_l3_len;
1806 /* Headers are fragmented */
1807 if (rte_pktmbuf_data_len(m) < inner_l3_offset + m->l3_len + m->l4_len)
1810 /* IP checksum can be counted only for IPv4 packet */
1811 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
1814 /* IP type not set when required */
1815 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
1816 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
1819 /* Check requirements for TSO packet */
1820 if (ol_flags & PKT_TX_TCP_SEG)
1821 if ((m->tso_segsz == 0) ||
1822 ((ol_flags & PKT_TX_IPV4) &&
1823 !(ol_flags & PKT_TX_IP_CKSUM)))
1826 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
1827 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
1828 !(ol_flags & PKT_TX_OUTER_IPV4))
1835 * Linearize data in mbuf.
1837 * This function moves the mbuf data in the first segment if there is enough
1838 * tailroom. The subsequent segments are unchained and freed.
1847 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
1849 int seg_len, copy_len;
1851 struct rte_mbuf *m_next;
1854 if (rte_pktmbuf_is_contiguous(mbuf))
1857 /* Extend first segment to the total packet length */
1858 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
1860 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
1863 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
1864 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
1866 /* Append data from next segments to the first one */
1871 seg_len = rte_pktmbuf_data_len(m);
1872 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
1875 rte_pktmbuf_free_seg(m);
1886 * Dump an mbuf structure to a file.
1888 * Dump all fields for the given packet mbuf and all its associated
1889 * segments (in the case of a chained buffer).
1892 * A pointer to a file for output
1896 * If dump_len != 0, also dump the "dump_len" first data bytes of
1899 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
1905 #endif /* _RTE_MBUF_H_ */