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 * When doing Tx offload like TSO or checksum, the HW needs to configure the
205 * tunnel type into the HW descriptors.
207 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
208 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
209 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
210 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
211 /**< TX packet with MPLS-in-UDP RFC 7510 header. */
212 #define PKT_TX_TUNNEL_MPLSINUDP (0x5ULL << 45)
213 /* add new TX TUNNEL type here */
214 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
217 * Second VLAN insertion (QinQ) flag.
219 #define PKT_TX_QINQ (1ULL << 49) /**< TX packet with double VLAN inserted. */
220 /* this old name is deprecated */
221 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
224 * TCP segmentation offload. To enable this offload feature for a
225 * packet to be transmitted on hardware supporting TSO:
226 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
228 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
229 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag and write the IP checksum
231 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
232 * - calculate the pseudo header checksum without taking ip_len in account,
233 * and set it in the TCP header. Refer to rte_ipv4_phdr_cksum() and
234 * rte_ipv6_phdr_cksum() that can be used as helpers.
236 #define PKT_TX_TCP_SEG (1ULL << 50)
238 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
241 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
242 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
243 * L4 checksum offload, the user needs to:
244 * - fill l2_len and l3_len in mbuf
245 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
246 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
247 * - calculate the pseudo header checksum and set it in the L4 header (only
248 * for TCP or UDP). See rte_ipv4_phdr_cksum() and rte_ipv6_phdr_cksum().
249 * For SCTP, set the crc field to 0.
251 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
252 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
253 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
254 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
255 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
258 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
259 * also be set by the application, although a PMD will only check
261 * - set the IP checksum field in the packet to 0
262 * - fill the mbuf offload information: l2_len, l3_len
264 #define PKT_TX_IP_CKSUM (1ULL << 54)
267 * Packet is IPv4. This flag must be set when using any offload feature
268 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
269 * packet. If the packet is a tunneled packet, this flag is related to
272 #define PKT_TX_IPV4 (1ULL << 55)
275 * Packet is IPv6. This flag must be set when using an offload feature
276 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
277 * packet. If the packet is a tunneled packet, this flag is related to
280 #define PKT_TX_IPV6 (1ULL << 56)
283 * TX packet is a 802.1q VLAN packet.
285 #define PKT_TX_VLAN (1ULL << 57)
286 /* this old name is deprecated */
287 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
290 * Offload the IP checksum of an external header in the hardware. The
291 * flag PKT_TX_OUTER_IPV4 should also be set by the application, alto ugh
292 * a PMD will only check PKT_TX_IP_CKSUM. The IP checksum field in the
293 * packet must be set to 0.
294 * - set the outer IP checksum field in the packet to 0
295 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
297 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
300 * Packet outer header is IPv4. This flag must be set when using any
301 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
302 * outer header of the tunneled packet is an IPv4 packet.
304 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
307 * Packet outer header is IPv6. This flag must be set when using any
308 * outer offload feature (L4 checksum) to tell the NIC that the outer
309 * header of the tunneled packet is an IPv6 packet.
311 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
314 * Bitmask of all supported packet Tx offload features flags,
315 * which can be set for packet.
317 #define PKT_TX_OFFLOAD_MASK ( \
320 PKT_TX_OUTER_IP_CKSUM | \
322 PKT_TX_IEEE1588_TMST | \
325 PKT_TX_TUNNEL_MASK | \
329 #define __RESERVED (1ULL << 61) /**< reserved for future mbuf use */
331 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
333 /** Alignment constraint of mbuf private area. */
334 #define RTE_MBUF_PRIV_ALIGN 8
337 * Get the name of a RX offload flag
340 * The mask describing the flag.
342 * The name of this flag, or NULL if it's not a valid RX flag.
344 const char *rte_get_rx_ol_flag_name(uint64_t mask);
347 * Dump the list of RX offload flags in a buffer
350 * The mask describing the RX flags.
354 * The length of the buffer.
356 * 0 on success, (-1) on error.
358 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
361 * Get the name of a TX offload flag
364 * The mask describing the flag. Usually only one bit must be set.
365 * Several bits can be given if they belong to the same mask.
366 * Ex: PKT_TX_L4_MASK.
368 * The name of this flag, or NULL if it's not a valid TX flag.
370 const char *rte_get_tx_ol_flag_name(uint64_t mask);
373 * Dump the list of TX offload flags in a buffer
376 * The mask describing the TX flags.
380 * The length of the buffer.
382 * 0 on success, (-1) on error.
384 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
387 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
388 * splitting it into multiple segments.
389 * So, for mbufs that planned to be involved into RX/TX, the recommended
390 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
392 #define RTE_MBUF_DEFAULT_DATAROOM 2048
393 #define RTE_MBUF_DEFAULT_BUF_SIZE \
394 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
396 /* define a set of marker types that can be used to refer to set points in the
399 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
401 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
403 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
404 * with a single assignment */
407 * The generic rte_mbuf, containing a packet mbuf.
412 void *buf_addr; /**< Virtual address of segment buffer. */
414 * Physical address of segment buffer.
415 * Force alignment to 8-bytes, so as to ensure we have the exact
416 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
417 * working on vector drivers easier.
422 rte_iova_t buf_physaddr; /**< deprecated */
423 } __rte_aligned(sizeof(rte_iova_t));
425 /* next 8 bytes are initialised on RX descriptor rearm */
430 * Reference counter. Its size should at least equal to the size
431 * of port field (16 bits), to support zero-copy broadcast.
432 * It should only be accessed using the following functions:
433 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
434 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
435 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
440 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
441 uint16_t refcnt; /**< Non-atomically accessed refcnt */
443 uint16_t nb_segs; /**< Number of segments. */
445 /** Input port (16 bits to support more than 256 virtual ports). */
448 uint64_t ol_flags; /**< Offload features. */
450 /* remaining bytes are set on RX when pulling packet from descriptor */
451 MARKER rx_descriptor_fields1;
454 * The packet type, which is the combination of outer/inner L2, L3, L4
455 * and tunnel types. The packet_type is about data really present in the
456 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
457 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
458 * vlan is stripped from the data.
462 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
464 uint32_t l2_type:4; /**< (Outer) L2 type. */
465 uint32_t l3_type:4; /**< (Outer) L3 type. */
466 uint32_t l4_type:4; /**< (Outer) L4 type. */
467 uint32_t tun_type:4; /**< Tunnel type. */
470 uint8_t inner_esp_next_proto;
471 /**< ESP next protocol type, valid if
472 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
477 uint8_t inner_l2_type:4;
478 /**< Inner L2 type. */
479 uint8_t inner_l3_type:4;
480 /**< Inner L3 type. */
483 uint32_t inner_l4_type:4; /**< Inner L4 type. */
487 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
488 uint16_t data_len; /**< Amount of data in segment buffer. */
489 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
493 uint32_t rss; /**< RSS hash result if RSS enabled */
502 /**< Second 4 flexible bytes */
505 /**< First 4 flexible bytes or FD ID, dependent on
506 PKT_RX_FDIR_* flag in ol_flags. */
507 } fdir; /**< Filter identifier if FDIR enabled */
511 } sched; /**< Hierarchical scheduler */
512 uint32_t usr; /**< User defined tags. See rte_distributor_process() */
513 } hash; /**< hash information */
515 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
516 uint16_t vlan_tci_outer;
518 uint16_t buf_len; /**< Length of segment buffer. */
520 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
521 * are not normalized but are always the same for a given port.
525 /* second cache line - fields only used in slow path or on TX */
526 MARKER cacheline1 __rte_cache_min_aligned;
530 void *userdata; /**< Can be used for external metadata */
531 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
534 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
535 struct rte_mbuf *next; /**< Next segment of scattered packet. */
537 /* fields to support TX offloads */
540 uint64_t tx_offload; /**< combined for easy fetch */
544 /**< L2 (MAC) Header Length for non-tunneling pkt.
545 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
547 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
548 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
549 uint64_t tso_segsz:16; /**< TCP TSO segment size */
551 /* fields for TX offloading of tunnels */
552 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
553 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
555 /* uint64_t unused:8; */
559 /** Size of the application private data. In case of an indirect
560 * mbuf, it stores the direct mbuf private data size. */
563 /** Timesync flags for use with IEEE1588. */
566 /** Sequence number. See also rte_reorder_insert(). */
569 } __rte_cache_aligned;
571 /**< Maximum number of nb_segs allowed. */
572 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
575 * Prefetch the first part of the mbuf
577 * The first 64 bytes of the mbuf corresponds to fields that are used early
578 * in the receive path. If the cache line of the architecture is higher than
579 * 64B, the second part will also be prefetched.
582 * The pointer to the mbuf.
585 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
587 rte_prefetch0(&m->cacheline0);
591 * Prefetch the second part of the mbuf
593 * The next 64 bytes of the mbuf corresponds to fields that are used in the
594 * transmit path. If the cache line of the architecture is higher than 64B,
595 * this function does nothing as it is expected that the full mbuf is
599 * The pointer to the mbuf.
602 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
604 #if RTE_CACHE_LINE_SIZE == 64
605 rte_prefetch0(&m->cacheline1);
612 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
615 * Return the IO address of the beginning of the mbuf data
618 * The pointer to the mbuf.
620 * The IO address of the beginning of the mbuf data
622 static inline rte_iova_t
623 rte_mbuf_data_iova(const struct rte_mbuf *mb)
625 return mb->buf_iova + mb->data_off;
629 static inline phys_addr_t
630 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
632 return rte_mbuf_data_iova(mb);
636 * Return the default IO address of the beginning of the mbuf data
638 * This function is used by drivers in their receive function, as it
639 * returns the location where data should be written by the NIC, taking
640 * the default headroom in account.
643 * The pointer to the mbuf.
645 * The IO address of the beginning of the mbuf data
647 static inline rte_iova_t
648 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
650 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
654 static inline phys_addr_t
655 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
657 return rte_mbuf_data_iova_default(mb);
661 * Return the mbuf owning the data buffer address of an indirect mbuf.
664 * The pointer to the indirect mbuf.
666 * The address of the direct mbuf corresponding to buffer_addr.
668 static inline struct rte_mbuf *
669 rte_mbuf_from_indirect(struct rte_mbuf *mi)
671 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
675 * Return the buffer address embedded in the given mbuf.
678 * The pointer to the mbuf.
680 * The address of the data buffer owned by the mbuf.
683 rte_mbuf_to_baddr(struct rte_mbuf *md)
686 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
691 * Returns TRUE if given mbuf is indirect, or FALSE otherwise.
693 #define RTE_MBUF_INDIRECT(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
696 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
698 #define RTE_MBUF_DIRECT(mb) (!RTE_MBUF_INDIRECT(mb))
701 * Private data in case of pktmbuf pool.
703 * A structure that contains some pktmbuf_pool-specific data that are
704 * appended after the mempool structure (in private data).
706 struct rte_pktmbuf_pool_private {
707 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
708 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
711 #ifdef RTE_LIBRTE_MBUF_DEBUG
713 /** check mbuf type in debug mode */
714 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
716 #else /* RTE_LIBRTE_MBUF_DEBUG */
718 /** check mbuf type in debug mode */
719 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
721 #endif /* RTE_LIBRTE_MBUF_DEBUG */
723 #ifdef RTE_MBUF_REFCNT_ATOMIC
726 * Reads the value of an mbuf's refcnt.
730 * Reference count number.
732 static inline uint16_t
733 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
735 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
739 * Sets an mbuf's refcnt to a defined value.
746 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
748 rte_atomic16_set(&m->refcnt_atomic, new_value);
752 static inline uint16_t
753 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
755 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
759 * Adds given value to an mbuf's refcnt and returns its new value.
763 * Value to add/subtract
767 static inline uint16_t
768 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
771 * The atomic_add is an expensive operation, so we don't want to
772 * call it in the case where we know we are the uniq holder of
773 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
774 * operation has to be used because concurrent accesses on the
775 * reference counter can occur.
777 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
778 rte_mbuf_refcnt_set(m, 1 + value);
782 return __rte_mbuf_refcnt_update(m, value);
785 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
788 static inline uint16_t
789 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
791 m->refcnt = (uint16_t)(m->refcnt + value);
796 * Adds given value to an mbuf's refcnt and returns its new value.
798 static inline uint16_t
799 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
801 return __rte_mbuf_refcnt_update(m, value);
805 * Reads the value of an mbuf's refcnt.
807 static inline uint16_t
808 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
814 * Sets an mbuf's refcnt to the defined value.
817 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
819 m->refcnt = new_value;
822 #endif /* RTE_MBUF_REFCNT_ATOMIC */
825 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
832 * Sanity checks on an mbuf.
834 * Check the consistency of the given mbuf. The function will cause a
835 * panic if corruption is detected.
838 * The mbuf to be checked.
840 * True if the mbuf is a packet header, false if it is a sub-segment
841 * of a packet (in this case, some fields like nb_segs are not checked)
844 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
846 #define MBUF_RAW_ALLOC_CHECK(m) do { \
847 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
848 RTE_ASSERT((m)->next == NULL); \
849 RTE_ASSERT((m)->nb_segs == 1); \
850 __rte_mbuf_sanity_check(m, 0); \
854 * Allocate an uninitialized mbuf from mempool *mp*.
856 * This function can be used by PMDs (especially in RX functions) to
857 * allocate an uninitialized mbuf. The driver is responsible of
858 * initializing all the required fields. See rte_pktmbuf_reset().
859 * For standard needs, prefer rte_pktmbuf_alloc().
861 * The caller can expect that the following fields of the mbuf structure
862 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
863 * next=NULL, pool, priv_size. The other fields must be initialized
867 * The mempool from which mbuf is allocated.
869 * - The pointer to the new mbuf on success.
870 * - NULL if allocation failed.
872 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
876 if (rte_mempool_get(mp, (void **)&m) < 0)
878 MBUF_RAW_ALLOC_CHECK(m);
883 * Put mbuf back into its original mempool.
885 * The caller must ensure that the mbuf is direct and properly
886 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
887 * rte_pktmbuf_prefree_seg().
889 * This function should be used with care, when optimization is
890 * required. For standard needs, prefer rte_pktmbuf_free() or
891 * rte_pktmbuf_free_seg().
894 * The mbuf to be freed.
896 static __rte_always_inline void
897 rte_mbuf_raw_free(struct rte_mbuf *m)
899 RTE_ASSERT(RTE_MBUF_DIRECT(m));
900 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
901 RTE_ASSERT(m->next == NULL);
902 RTE_ASSERT(m->nb_segs == 1);
903 __rte_mbuf_sanity_check(m, 0);
904 rte_mempool_put(m->pool, m);
907 /* compat with older versions */
910 __rte_mbuf_raw_free(struct rte_mbuf *m)
912 rte_mbuf_raw_free(m);
916 * The packet mbuf constructor.
918 * This function initializes some fields in the mbuf structure that are
919 * not modified by the user once created (origin pool, buffer start
920 * address, and so on). This function is given as a callback function to
921 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
924 * The mempool from which mbufs originate.
926 * A pointer that can be used by the user to retrieve useful information
927 * for mbuf initialization. This pointer is the opaque argument passed to
928 * rte_mempool_obj_iter() or rte_mempool_create().
930 * The mbuf to initialize.
932 * The index of the mbuf in the pool table.
934 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
935 void *m, unsigned i);
939 * A packet mbuf pool constructor.
941 * This function initializes the mempool private data in the case of a
942 * pktmbuf pool. This private data is needed by the driver. The
943 * function must be called on the mempool before it is used, or it
944 * can be given as a callback function to rte_mempool_create() at
945 * pool creation. It can be extended by the user, for example, to
946 * provide another packet size.
949 * The mempool from which mbufs originate.
951 * A pointer that can be used by the user to retrieve useful information
952 * for mbuf initialization. This pointer is the opaque argument passed to
953 * rte_mempool_create().
955 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
958 * Create a mbuf pool.
960 * This function creates and initializes a packet mbuf pool. It is
961 * a wrapper to rte_mempool functions.
964 * The name of the mbuf pool.
966 * The number of elements in the mbuf pool. The optimum size (in terms
967 * of memory usage) for a mempool is when n is a power of two minus one:
970 * Size of the per-core object cache. See rte_mempool_create() for
973 * Size of application private are between the rte_mbuf structure
974 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
975 * @param data_room_size
976 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
978 * The socket identifier where the memory should be allocated. The
979 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
982 * The pointer to the new allocated mempool, on success. NULL on error
983 * with rte_errno set appropriately. Possible rte_errno values include:
984 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
985 * - E_RTE_SECONDARY - function was called from a secondary process instance
986 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
987 * - ENOSPC - the maximum number of memzones has already been allocated
988 * - EEXIST - a memzone with the same name already exists
989 * - ENOMEM - no appropriate memory area found in which to create memzone
992 rte_pktmbuf_pool_create(const char *name, unsigned n,
993 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
997 * Create a mbuf pool with a given mempool ops name
999 * This function creates and initializes a packet mbuf pool. It is
1000 * a wrapper to rte_mempool functions.
1003 * The name of the mbuf pool.
1005 * The number of elements in the mbuf pool. The optimum size (in terms
1006 * of memory usage) for a mempool is when n is a power of two minus one:
1009 * Size of the per-core object cache. See rte_mempool_create() for
1012 * Size of application private are between the rte_mbuf structure
1013 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1014 * @param data_room_size
1015 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1017 * The socket identifier where the memory should be allocated. The
1018 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1021 * The mempool ops name to be used for this mempool instead of
1022 * default mempool. The value can be *NULL* to use default mempool.
1024 * The pointer to the new allocated mempool, on success. NULL on error
1025 * with rte_errno set appropriately. Possible rte_errno values include:
1026 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1027 * - E_RTE_SECONDARY - function was called from a secondary process instance
1028 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1029 * - ENOSPC - the maximum number of memzones has already been allocated
1030 * - EEXIST - a memzone with the same name already exists
1031 * - ENOMEM - no appropriate memory area found in which to create memzone
1033 struct rte_mempool * __rte_experimental
1034 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1035 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1036 int socket_id, const char *ops_name);
1039 * Get the data room size of mbufs stored in a pktmbuf_pool
1041 * The data room size is the amount of data that can be stored in a
1042 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1045 * The packet mbuf pool.
1047 * The data room size of mbufs stored in this mempool.
1049 static inline uint16_t
1050 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1052 struct rte_pktmbuf_pool_private *mbp_priv;
1054 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1055 return mbp_priv->mbuf_data_room_size;
1059 * Get the application private size of mbufs stored in a pktmbuf_pool
1061 * The private size of mbuf is a zone located between the rte_mbuf
1062 * structure and the data buffer where an application can store data
1063 * associated to a packet.
1066 * The packet mbuf pool.
1068 * The private size of mbufs stored in this mempool.
1070 static inline uint16_t
1071 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1073 struct rte_pktmbuf_pool_private *mbp_priv;
1075 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1076 return mbp_priv->mbuf_priv_size;
1080 * Reset the data_off field of a packet mbuf to its default value.
1082 * The given mbuf must have only one segment, which should be empty.
1085 * The packet mbuf's data_off field has to be reset.
1087 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1089 m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, (uint16_t)m->buf_len);
1093 * Reset the fields of a packet mbuf to their default values.
1095 * The given mbuf must have only one segment.
1098 * The packet mbuf to be resetted.
1100 #define MBUF_INVALID_PORT UINT16_MAX
1102 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1108 m->vlan_tci_outer = 0;
1110 m->port = MBUF_INVALID_PORT;
1114 rte_pktmbuf_reset_headroom(m);
1117 __rte_mbuf_sanity_check(m, 1);
1121 * Allocate a new mbuf from a mempool.
1123 * This new mbuf contains one segment, which has a length of 0. The pointer
1124 * to data is initialized to have some bytes of headroom in the buffer
1125 * (if buffer size allows).
1128 * The mempool from which the mbuf is allocated.
1130 * - The pointer to the new mbuf on success.
1131 * - NULL if allocation failed.
1133 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1136 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1137 rte_pktmbuf_reset(m);
1142 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1146 * The mempool from which mbufs are allocated.
1148 * Array of pointers to mbufs
1153 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1155 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1156 struct rte_mbuf **mbufs, unsigned count)
1161 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1165 /* To understand duff's device on loop unwinding optimization, see
1166 * https://en.wikipedia.org/wiki/Duff's_device.
1167 * Here while() loop is used rather than do() while{} to avoid extra
1168 * check if count is zero.
1170 switch (count % 4) {
1172 while (idx != count) {
1173 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1174 rte_pktmbuf_reset(mbufs[idx]);
1178 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1179 rte_pktmbuf_reset(mbufs[idx]);
1183 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1184 rte_pktmbuf_reset(mbufs[idx]);
1188 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1189 rte_pktmbuf_reset(mbufs[idx]);
1198 * Attach packet mbuf to another packet mbuf.
1200 * After attachment we refer the mbuf we attached as 'indirect',
1201 * while mbuf we attached to as 'direct'.
1202 * The direct mbuf's reference counter is incremented.
1204 * Right now, not supported:
1205 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1206 * - mbuf we trying to attach (mi) is used by someone else
1207 * e.g. it's reference counter is greater then 1.
1210 * The indirect packet mbuf.
1212 * The packet mbuf we're attaching to.
1214 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1216 struct rte_mbuf *md;
1218 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1219 rte_mbuf_refcnt_read(mi) == 1);
1221 /* if m is not direct, get the mbuf that embeds the data */
1222 if (RTE_MBUF_DIRECT(m))
1225 md = rte_mbuf_from_indirect(m);
1227 rte_mbuf_refcnt_update(md, 1);
1228 mi->priv_size = m->priv_size;
1229 mi->buf_iova = m->buf_iova;
1230 mi->buf_addr = m->buf_addr;
1231 mi->buf_len = m->buf_len;
1233 mi->data_off = m->data_off;
1234 mi->data_len = m->data_len;
1236 mi->vlan_tci = m->vlan_tci;
1237 mi->vlan_tci_outer = m->vlan_tci_outer;
1238 mi->tx_offload = m->tx_offload;
1242 mi->pkt_len = mi->data_len;
1244 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1245 mi->packet_type = m->packet_type;
1246 mi->timestamp = m->timestamp;
1248 __rte_mbuf_sanity_check(mi, 1);
1249 __rte_mbuf_sanity_check(m, 0);
1253 * Detach an indirect packet mbuf.
1255 * - restore original mbuf address and length values.
1256 * - reset pktmbuf data and data_len to their default values.
1257 * - decrement the direct mbuf's reference counter. When the
1258 * reference counter becomes 0, the direct mbuf is freed.
1260 * All other fields of the given packet mbuf will be left intact.
1263 * The indirect attached packet mbuf.
1265 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1267 struct rte_mbuf *md = rte_mbuf_from_indirect(m);
1268 struct rte_mempool *mp = m->pool;
1269 uint32_t mbuf_size, buf_len, priv_size;
1271 priv_size = rte_pktmbuf_priv_size(mp);
1272 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
1273 buf_len = rte_pktmbuf_data_room_size(mp);
1275 m->priv_size = priv_size;
1276 m->buf_addr = (char *)m + mbuf_size;
1277 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1278 m->buf_len = (uint16_t)buf_len;
1279 rte_pktmbuf_reset_headroom(m);
1283 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1286 rte_mbuf_refcnt_set(md, 1);
1287 rte_mbuf_raw_free(md);
1292 * Decrease reference counter and unlink a mbuf segment
1294 * This function does the same than a free, except that it does not
1295 * return the segment to its pool.
1296 * It decreases the reference counter, and if it reaches 0, it is
1297 * detached from its parent for an indirect mbuf.
1300 * The mbuf to be unlinked
1302 * - (m) if it is the last reference. It can be recycled or freed.
1303 * - (NULL) if the mbuf still has remaining references on it.
1305 static __rte_always_inline struct rte_mbuf *
1306 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1308 __rte_mbuf_sanity_check(m, 0);
1310 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1312 if (RTE_MBUF_INDIRECT(m))
1313 rte_pktmbuf_detach(m);
1315 if (m->next != NULL) {
1322 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1324 if (RTE_MBUF_INDIRECT(m))
1325 rte_pktmbuf_detach(m);
1327 if (m->next != NULL) {
1331 rte_mbuf_refcnt_set(m, 1);
1338 /* deprecated, replaced by rte_pktmbuf_prefree_seg() */
1340 static inline struct rte_mbuf *
1341 __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1343 return rte_pktmbuf_prefree_seg(m);
1347 * Free a segment of a packet mbuf into its original mempool.
1349 * Free an mbuf, without parsing other segments in case of chained
1353 * The packet mbuf segment to be freed.
1355 static __rte_always_inline void
1356 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1358 m = rte_pktmbuf_prefree_seg(m);
1359 if (likely(m != NULL))
1360 rte_mbuf_raw_free(m);
1364 * Free a packet mbuf back into its original mempool.
1366 * Free an mbuf, and all its segments in case of chained buffers. Each
1367 * segment is added back into its original mempool.
1370 * The packet mbuf to be freed. If NULL, the function does nothing.
1372 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1374 struct rte_mbuf *m_next;
1377 __rte_mbuf_sanity_check(m, 1);
1381 rte_pktmbuf_free_seg(m);
1387 * Creates a "clone" of the given packet mbuf.
1389 * Walks through all segments of the given packet mbuf, and for each of them:
1390 * - Creates a new packet mbuf from the given pool.
1391 * - Attaches newly created mbuf to the segment.
1392 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1393 * from the original packet mbuf.
1396 * The packet mbuf to be cloned.
1398 * The mempool from which the "clone" mbufs are allocated.
1400 * - The pointer to the new "clone" mbuf on success.
1401 * - NULL if allocation fails.
1403 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1404 struct rte_mempool *mp)
1406 struct rte_mbuf *mc, *mi, **prev;
1410 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1415 pktlen = md->pkt_len;
1420 rte_pktmbuf_attach(mi, md);
1423 } while ((md = md->next) != NULL &&
1424 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1428 mc->pkt_len = pktlen;
1430 /* Allocation of new indirect segment failed */
1431 if (unlikely (mi == NULL)) {
1432 rte_pktmbuf_free(mc);
1436 __rte_mbuf_sanity_check(mc, 1);
1441 * Adds given value to the refcnt of all packet mbuf segments.
1443 * Walks through all segments of given packet mbuf and for each of them
1444 * invokes rte_mbuf_refcnt_update().
1447 * The packet mbuf whose refcnt to be updated.
1449 * The value to add to the mbuf's segments refcnt.
1451 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1453 __rte_mbuf_sanity_check(m, 1);
1456 rte_mbuf_refcnt_update(m, v);
1457 } while ((m = m->next) != NULL);
1461 * Get the headroom in a packet mbuf.
1466 * The length of the headroom.
1468 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1470 __rte_mbuf_sanity_check(m, 0);
1475 * Get the tailroom of a packet mbuf.
1480 * The length of the tailroom.
1482 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1484 __rte_mbuf_sanity_check(m, 0);
1485 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1490 * Get the last segment of the packet.
1495 * The last segment of the given mbuf.
1497 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1499 __rte_mbuf_sanity_check(m, 1);
1500 while (m->next != NULL)
1506 * A macro that points to an offset into the data in the mbuf.
1508 * The returned pointer is cast to type t. Before using this
1509 * function, the user must ensure that the first segment is large
1510 * enough to accommodate its data.
1515 * The offset into the mbuf data.
1517 * The type to cast the result into.
1519 #define rte_pktmbuf_mtod_offset(m, t, o) \
1520 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1523 * A macro that points to the start of the data in the mbuf.
1525 * The returned pointer is cast to type t. Before using this
1526 * function, the user must ensure that the first segment is large
1527 * enough to accommodate its data.
1532 * The type to cast the result into.
1534 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1537 * A macro that returns the IO address that points to an offset of the
1538 * start of the data in the mbuf
1543 * The offset into the data to calculate address from.
1545 #define rte_pktmbuf_iova_offset(m, o) \
1546 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
1549 #define rte_pktmbuf_mtophys_offset(m, o) \
1550 rte_pktmbuf_iova_offset(m, o)
1553 * A macro that returns the IO address that points to the start of the
1559 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
1562 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1565 * A macro that returns the length of the packet.
1567 * The value can be read or assigned.
1572 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1575 * A macro that returns the length of the segment.
1577 * The value can be read or assigned.
1582 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1585 * Prepend len bytes to an mbuf data area.
1587 * Returns a pointer to the new
1588 * data start address. If there is not enough headroom in the first
1589 * segment, the function will return NULL, without modifying the mbuf.
1594 * The amount of data to prepend (in bytes).
1596 * A pointer to the start of the newly prepended data, or
1597 * NULL if there is not enough headroom space in the first segment
1599 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1602 __rte_mbuf_sanity_check(m, 1);
1604 if (unlikely(len > rte_pktmbuf_headroom(m)))
1608 m->data_len = (uint16_t)(m->data_len + len);
1609 m->pkt_len = (m->pkt_len + len);
1611 return (char *)m->buf_addr + m->data_off;
1615 * Append len bytes to an mbuf.
1617 * Append len bytes to an mbuf and return a pointer to the start address
1618 * of the added data. If there is not enough tailroom in the last
1619 * segment, the function will return NULL, without modifying the mbuf.
1624 * The amount of data to append (in bytes).
1626 * A pointer to the start of the newly appended data, or
1627 * NULL if there is not enough tailroom space in the last segment
1629 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1632 struct rte_mbuf *m_last;
1634 __rte_mbuf_sanity_check(m, 1);
1636 m_last = rte_pktmbuf_lastseg(m);
1637 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1640 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1641 m_last->data_len = (uint16_t)(m_last->data_len + len);
1642 m->pkt_len = (m->pkt_len + len);
1643 return (char*) tail;
1647 * Remove len bytes at the beginning of an mbuf.
1649 * Returns a pointer to the start address of the new data area. If the
1650 * length is greater than the length of the first segment, then the
1651 * function will fail and return NULL, without modifying the mbuf.
1656 * The amount of data to remove (in bytes).
1658 * A pointer to the new start of the data.
1660 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1662 __rte_mbuf_sanity_check(m, 1);
1664 if (unlikely(len > m->data_len))
1667 m->data_len = (uint16_t)(m->data_len - len);
1669 m->pkt_len = (m->pkt_len - len);
1670 return (char *)m->buf_addr + m->data_off;
1674 * Remove len bytes of data at the end of the mbuf.
1676 * If the length is greater than the length of the last segment, the
1677 * function will fail and return -1 without modifying the mbuf.
1682 * The amount of data to remove (in bytes).
1687 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
1689 struct rte_mbuf *m_last;
1691 __rte_mbuf_sanity_check(m, 1);
1693 m_last = rte_pktmbuf_lastseg(m);
1694 if (unlikely(len > m_last->data_len))
1697 m_last->data_len = (uint16_t)(m_last->data_len - len);
1698 m->pkt_len = (m->pkt_len - len);
1703 * Test if mbuf data is contiguous.
1708 * - 1, if all data is contiguous (one segment).
1709 * - 0, if there is several segments.
1711 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
1713 __rte_mbuf_sanity_check(m, 1);
1714 return !!(m->nb_segs == 1);
1718 * @internal used by rte_pktmbuf_read().
1720 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
1721 uint32_t len, void *buf);
1724 * Read len data bytes in a mbuf at specified offset.
1726 * If the data is contiguous, return the pointer in the mbuf data, else
1727 * copy the data in the buffer provided by the user and return its
1731 * The pointer to the mbuf.
1733 * The offset of the data in the mbuf.
1735 * The amount of bytes to read.
1737 * The buffer where data is copied if it is not contiguous in mbuf
1738 * data. Its length should be at least equal to the len parameter.
1740 * The pointer to the data, either in the mbuf if it is contiguous,
1741 * or in the user buffer. If mbuf is too small, NULL is returned.
1743 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
1744 uint32_t off, uint32_t len, void *buf)
1746 if (likely(off + len <= rte_pktmbuf_data_len(m)))
1747 return rte_pktmbuf_mtod_offset(m, char *, off);
1749 return __rte_pktmbuf_read(m, off, len, buf);
1753 * Chain an mbuf to another, thereby creating a segmented packet.
1755 * Note: The implementation will do a linear walk over the segments to find
1756 * the tail entry. For cases when there are many segments, it's better to
1757 * chain the entries manually.
1760 * The head of the mbuf chain (the first packet)
1762 * The mbuf to put last in the chain
1766 * - -EOVERFLOW, if the chain segment limit exceeded
1768 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
1770 struct rte_mbuf *cur_tail;
1772 /* Check for number-of-segments-overflow */
1773 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
1776 /* Chain 'tail' onto the old tail */
1777 cur_tail = rte_pktmbuf_lastseg(head);
1778 cur_tail->next = tail;
1780 /* accumulate number of segments and total length. */
1781 head->nb_segs += tail->nb_segs;
1782 head->pkt_len += tail->pkt_len;
1784 /* pkt_len is only set in the head */
1785 tail->pkt_len = tail->data_len;
1791 * Validate general requirements for Tx offload in mbuf.
1793 * This function checks correctness and completeness of Tx offload settings.
1796 * The packet mbuf to be validated.
1798 * 0 if packet is valid
1801 rte_validate_tx_offload(const struct rte_mbuf *m)
1803 uint64_t ol_flags = m->ol_flags;
1804 uint64_t inner_l3_offset = m->l2_len;
1806 /* Does packet set any of available offloads? */
1807 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
1810 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
1811 inner_l3_offset += m->outer_l2_len + m->outer_l3_len;
1813 /* Headers are fragmented */
1814 if (rte_pktmbuf_data_len(m) < inner_l3_offset + m->l3_len + m->l4_len)
1817 /* IP checksum can be counted only for IPv4 packet */
1818 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
1821 /* IP type not set when required */
1822 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
1823 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
1826 /* Check requirements for TSO packet */
1827 if (ol_flags & PKT_TX_TCP_SEG)
1828 if ((m->tso_segsz == 0) ||
1829 ((ol_flags & PKT_TX_IPV4) &&
1830 !(ol_flags & PKT_TX_IP_CKSUM)))
1833 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
1834 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
1835 !(ol_flags & PKT_TX_OUTER_IPV4))
1842 * Linearize data in mbuf.
1844 * This function moves the mbuf data in the first segment if there is enough
1845 * tailroom. The subsequent segments are unchained and freed.
1854 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
1856 int seg_len, copy_len;
1858 struct rte_mbuf *m_next;
1861 if (rte_pktmbuf_is_contiguous(mbuf))
1864 /* Extend first segment to the total packet length */
1865 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
1867 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
1870 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
1871 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
1873 /* Append data from next segments to the first one */
1878 seg_len = rte_pktmbuf_data_len(m);
1879 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
1882 rte_pktmbuf_free_seg(m);
1893 * Dump an mbuf structure to a file.
1895 * Dump all fields for the given packet mbuf and all its associated
1896 * segments (in the case of a chained buffer).
1899 * A pointer to a file for output
1903 * If dump_len != 0, also dump the "dump_len" first data bytes of
1906 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
1912 #endif /* _RTE_MBUF_H_ */