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 #define PKT_TX_TUNNEL_VXLAN_GPE (0x6ULL << 45)
218 * Generic IP encapsulated tunnel type, used for TSO and checksum offload.
219 * It can be used for tunnels which are not standards or listed above.
220 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_GRE
221 * or PKT_TX_TUNNEL_IPIP if possible.
222 * The ethdev must be configured with DEV_TX_OFFLOAD_IP_TNL_TSO.
223 * Outer and inner checksums are done according to the existing flags like
225 * Specific tunnel headers that contain payload length, sequence id
226 * or checksum are not expected to be updated.
228 #define PKT_TX_TUNNEL_IP (0xDULL << 45)
230 * Generic UDP encapsulated tunnel type, used for TSO and checksum offload.
231 * UDP tunnel type implies outer IP layer.
232 * It can be used for tunnels which are not standards or listed above.
233 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_VXLAN
235 * The ethdev must be configured with DEV_TX_OFFLOAD_UDP_TNL_TSO.
236 * Outer and inner checksums are done according to the existing flags like
238 * Specific tunnel headers that contain payload length, sequence id
239 * or checksum are not expected to be updated.
241 #define PKT_TX_TUNNEL_UDP (0xEULL << 45)
242 /* add new TX TUNNEL type here */
243 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
246 * Second VLAN insertion (QinQ) flag.
248 #define PKT_TX_QINQ (1ULL << 49) /**< TX packet with double VLAN inserted. */
249 /* this old name is deprecated */
250 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
253 * TCP segmentation offload. To enable this offload feature for a
254 * packet to be transmitted on hardware supporting TSO:
255 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
257 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
258 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag
259 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
261 #define PKT_TX_TCP_SEG (1ULL << 50)
263 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
266 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
267 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
268 * L4 checksum offload, the user needs to:
269 * - fill l2_len and l3_len in mbuf
270 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
271 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
273 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
274 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
275 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
276 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
277 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
280 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
281 * also be set by the application, although a PMD will only check
283 * - fill the mbuf offload information: l2_len, l3_len
285 #define PKT_TX_IP_CKSUM (1ULL << 54)
288 * Packet is IPv4. This flag must be set when using any offload feature
289 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
290 * packet. If the packet is a tunneled packet, this flag is related to
293 #define PKT_TX_IPV4 (1ULL << 55)
296 * Packet is IPv6. This flag must be set when using an offload feature
297 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
298 * packet. If the packet is a tunneled packet, this flag is related to
301 #define PKT_TX_IPV6 (1ULL << 56)
304 * TX packet is a 802.1q VLAN packet.
306 #define PKT_TX_VLAN (1ULL << 57)
307 /* this old name is deprecated */
308 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
311 * Offload the IP checksum of an external header in the hardware. The
312 * flag PKT_TX_OUTER_IPV4 should also be set by the application, although
313 * a PMD will only check PKT_TX_OUTER_IP_CKSUM.
314 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
316 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
319 * Packet outer header is IPv4. This flag must be set when using any
320 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
321 * outer header of the tunneled packet is an IPv4 packet.
323 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
326 * Packet outer header is IPv6. This flag must be set when using any
327 * outer offload feature (L4 checksum) to tell the NIC that the outer
328 * header of the tunneled packet is an IPv6 packet.
330 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
333 * Bitmask of all supported packet Tx offload features flags,
334 * which can be set for packet.
336 #define PKT_TX_OFFLOAD_MASK ( \
339 PKT_TX_OUTER_IP_CKSUM | \
341 PKT_TX_IEEE1588_TMST | \
344 PKT_TX_TUNNEL_MASK | \
349 * Mbuf having an external buffer attached. shinfo in mbuf must be filled.
351 #define EXT_ATTACHED_MBUF (1ULL << 61)
353 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
355 /** Alignment constraint of mbuf private area. */
356 #define RTE_MBUF_PRIV_ALIGN 8
359 * Get the name of a RX offload flag
362 * The mask describing the flag.
364 * The name of this flag, or NULL if it's not a valid RX flag.
366 const char *rte_get_rx_ol_flag_name(uint64_t mask);
369 * Dump the list of RX offload flags in a buffer
372 * The mask describing the RX flags.
376 * The length of the buffer.
378 * 0 on success, (-1) on error.
380 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
383 * Get the name of a TX offload flag
386 * The mask describing the flag. Usually only one bit must be set.
387 * Several bits can be given if they belong to the same mask.
388 * Ex: PKT_TX_L4_MASK.
390 * The name of this flag, or NULL if it's not a valid TX flag.
392 const char *rte_get_tx_ol_flag_name(uint64_t mask);
395 * Dump the list of TX offload flags in a buffer
398 * The mask describing the TX flags.
402 * The length of the buffer.
404 * 0 on success, (-1) on error.
406 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
409 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
410 * splitting it into multiple segments.
411 * So, for mbufs that planned to be involved into RX/TX, the recommended
412 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
414 #define RTE_MBUF_DEFAULT_DATAROOM 2048
415 #define RTE_MBUF_DEFAULT_BUF_SIZE \
416 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
418 /* define a set of marker types that can be used to refer to set points in the
421 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
423 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
425 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
426 * with a single assignment */
429 * The generic rte_mbuf, containing a packet mbuf.
434 void *buf_addr; /**< Virtual address of segment buffer. */
436 * Physical address of segment buffer.
437 * Force alignment to 8-bytes, so as to ensure we have the exact
438 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
439 * working on vector drivers easier.
444 rte_iova_t buf_physaddr; /**< deprecated */
445 } __rte_aligned(sizeof(rte_iova_t));
447 /* next 8 bytes are initialised on RX descriptor rearm */
452 * Reference counter. Its size should at least equal to the size
453 * of port field (16 bits), to support zero-copy broadcast.
454 * It should only be accessed using the following functions:
455 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
456 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
457 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
462 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
463 uint16_t refcnt; /**< Non-atomically accessed refcnt */
465 uint16_t nb_segs; /**< Number of segments. */
467 /** Input port (16 bits to support more than 256 virtual ports). */
470 uint64_t ol_flags; /**< Offload features. */
472 /* remaining bytes are set on RX when pulling packet from descriptor */
473 MARKER rx_descriptor_fields1;
476 * The packet type, which is the combination of outer/inner L2, L3, L4
477 * and tunnel types. The packet_type is about data really present in the
478 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
479 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
480 * vlan is stripped from the data.
484 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
486 uint32_t l2_type:4; /**< (Outer) L2 type. */
487 uint32_t l3_type:4; /**< (Outer) L3 type. */
488 uint32_t l4_type:4; /**< (Outer) L4 type. */
489 uint32_t tun_type:4; /**< Tunnel type. */
492 uint8_t inner_esp_next_proto;
493 /**< ESP next protocol type, valid if
494 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
499 uint8_t inner_l2_type:4;
500 /**< Inner L2 type. */
501 uint8_t inner_l3_type:4;
502 /**< Inner L3 type. */
505 uint32_t inner_l4_type:4; /**< Inner L4 type. */
509 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
510 uint16_t data_len; /**< Amount of data in segment buffer. */
511 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
515 uint32_t rss; /**< RSS hash result if RSS enabled */
524 /**< Second 4 flexible bytes */
527 /**< First 4 flexible bytes or FD ID, dependent on
528 PKT_RX_FDIR_* flag in ol_flags. */
529 } fdir; /**< Filter identifier if FDIR enabled */
533 } sched; /**< Hierarchical scheduler */
534 uint32_t usr; /**< User defined tags. See rte_distributor_process() */
535 } hash; /**< hash information */
537 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
538 uint16_t vlan_tci_outer;
540 uint16_t buf_len; /**< Length of segment buffer. */
542 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
543 * are not normalized but are always the same for a given port.
547 /* second cache line - fields only used in slow path or on TX */
548 MARKER cacheline1 __rte_cache_min_aligned;
552 void *userdata; /**< Can be used for external metadata */
553 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
556 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
557 struct rte_mbuf *next; /**< Next segment of scattered packet. */
559 /* fields to support TX offloads */
562 uint64_t tx_offload; /**< combined for easy fetch */
566 /**< L2 (MAC) Header Length for non-tunneling pkt.
567 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
569 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
570 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
571 uint64_t tso_segsz:16; /**< TCP TSO segment size */
573 /* fields for TX offloading of tunnels */
574 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
575 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
577 /* uint64_t unused:8; */
581 /** Size of the application private data. In case of an indirect
582 * mbuf, it stores the direct mbuf private data size. */
585 /** Timesync flags for use with IEEE1588. */
588 /** Sequence number. See also rte_reorder_insert(). */
591 /** Shared data for external buffer attached to mbuf. See
592 * rte_pktmbuf_attach_extbuf().
594 struct rte_mbuf_ext_shared_info *shinfo;
596 } __rte_cache_aligned;
599 * Function typedef of callback to free externally attached buffer.
601 typedef void (*rte_mbuf_extbuf_free_callback_t)(void *addr, void *opaque);
604 * Shared data at the end of an external buffer.
606 struct rte_mbuf_ext_shared_info {
607 rte_mbuf_extbuf_free_callback_t free_cb; /**< Free callback function */
608 void *fcb_opaque; /**< Free callback argument */
609 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
612 /**< Maximum number of nb_segs allowed. */
613 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
616 * Prefetch the first part of the mbuf
618 * The first 64 bytes of the mbuf corresponds to fields that are used early
619 * in the receive path. If the cache line of the architecture is higher than
620 * 64B, the second part will also be prefetched.
623 * The pointer to the mbuf.
626 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
628 rte_prefetch0(&m->cacheline0);
632 * Prefetch the second part of the mbuf
634 * The next 64 bytes of the mbuf corresponds to fields that are used in the
635 * transmit path. If the cache line of the architecture is higher than 64B,
636 * this function does nothing as it is expected that the full mbuf is
640 * The pointer to the mbuf.
643 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
645 #if RTE_CACHE_LINE_SIZE == 64
646 rte_prefetch0(&m->cacheline1);
653 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
656 * Return the IO address of the beginning of the mbuf data
659 * The pointer to the mbuf.
661 * The IO address of the beginning of the mbuf data
663 static inline rte_iova_t
664 rte_mbuf_data_iova(const struct rte_mbuf *mb)
666 return mb->buf_iova + mb->data_off;
670 static inline phys_addr_t
671 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
673 return rte_mbuf_data_iova(mb);
677 * Return the default IO address of the beginning of the mbuf data
679 * This function is used by drivers in their receive function, as it
680 * returns the location where data should be written by the NIC, taking
681 * the default headroom in account.
684 * The pointer to the mbuf.
686 * The IO address of the beginning of the mbuf data
688 static inline rte_iova_t
689 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
691 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
695 static inline phys_addr_t
696 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
698 return rte_mbuf_data_iova_default(mb);
702 * Return the mbuf owning the data buffer address of an indirect mbuf.
705 * The pointer to the indirect mbuf.
707 * The address of the direct mbuf corresponding to buffer_addr.
709 static inline struct rte_mbuf *
710 rte_mbuf_from_indirect(struct rte_mbuf *mi)
712 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
716 * Return the buffer address embedded in the given mbuf.
719 * The pointer to the mbuf.
721 * The address of the data buffer owned by the mbuf.
724 rte_mbuf_to_baddr(struct rte_mbuf *md)
727 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
732 * Returns TRUE if given mbuf is cloned by mbuf indirection, or FALSE
735 * If a mbuf has its data in another mbuf and references it by mbuf
736 * indirection, this mbuf can be defined as a cloned mbuf.
738 #define RTE_MBUF_CLONED(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
741 * Returns TRUE if given mbuf is indirect, or FALSE otherwise.
743 #define RTE_MBUF_INDIRECT(mb) RTE_MBUF_CLONED(mb)
746 * Returns TRUE if given mbuf has an external buffer, or FALSE otherwise.
748 * External buffer is a user-provided anonymous buffer.
750 #define RTE_MBUF_HAS_EXTBUF(mb) ((mb)->ol_flags & EXT_ATTACHED_MBUF)
753 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
755 * If a mbuf embeds its own data after the rte_mbuf structure, this mbuf
756 * can be defined as a direct mbuf.
758 #define RTE_MBUF_DIRECT(mb) \
759 (!((mb)->ol_flags & (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF)))
762 * Private data in case of pktmbuf pool.
764 * A structure that contains some pktmbuf_pool-specific data that are
765 * appended after the mempool structure (in private data).
767 struct rte_pktmbuf_pool_private {
768 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
769 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
772 #ifdef RTE_LIBRTE_MBUF_DEBUG
774 /** check mbuf type in debug mode */
775 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
777 #else /* RTE_LIBRTE_MBUF_DEBUG */
779 /** check mbuf type in debug mode */
780 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
782 #endif /* RTE_LIBRTE_MBUF_DEBUG */
784 #ifdef RTE_MBUF_REFCNT_ATOMIC
787 * Reads the value of an mbuf's refcnt.
791 * Reference count number.
793 static inline uint16_t
794 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
796 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
800 * Sets an mbuf's refcnt to a defined value.
807 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
809 rte_atomic16_set(&m->refcnt_atomic, (int16_t)new_value);
813 static inline uint16_t
814 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
816 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
820 * Adds given value to an mbuf's refcnt and returns its new value.
824 * Value to add/subtract
828 static inline uint16_t
829 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
832 * The atomic_add is an expensive operation, so we don't want to
833 * call it in the case where we know we are the uniq holder of
834 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
835 * operation has to be used because concurrent accesses on the
836 * reference counter can occur.
838 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
840 rte_mbuf_refcnt_set(m, (uint16_t)value);
841 return (uint16_t)value;
844 return __rte_mbuf_refcnt_update(m, value);
847 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
850 static inline uint16_t
851 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
853 m->refcnt = (uint16_t)(m->refcnt + value);
858 * Adds given value to an mbuf's refcnt and returns its new value.
860 static inline uint16_t
861 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
863 return __rte_mbuf_refcnt_update(m, value);
867 * Reads the value of an mbuf's refcnt.
869 static inline uint16_t
870 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
876 * Sets an mbuf's refcnt to the defined value.
879 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
881 m->refcnt = new_value;
884 #endif /* RTE_MBUF_REFCNT_ATOMIC */
887 * Reads the refcnt of an external buffer.
890 * Shared data of the external buffer.
892 * Reference count number.
894 static inline uint16_t
895 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
897 return (uint16_t)(rte_atomic16_read(&shinfo->refcnt_atomic));
901 * Set refcnt of an external buffer.
904 * Shared data of the external buffer.
909 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
912 rte_atomic16_set(&shinfo->refcnt_atomic, (int16_t)new_value);
916 * Add given value to refcnt of an external buffer and return its new
920 * Shared data of the external buffer.
922 * Value to add/subtract
926 static inline uint16_t
927 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
930 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
932 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
933 return (uint16_t)value;
936 return (uint16_t)rte_atomic16_add_return(&shinfo->refcnt_atomic, value);
940 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
947 * Sanity checks on an mbuf.
949 * Check the consistency of the given mbuf. The function will cause a
950 * panic if corruption is detected.
953 * The mbuf to be checked.
955 * True if the mbuf is a packet header, false if it is a sub-segment
956 * of a packet (in this case, some fields like nb_segs are not checked)
959 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
961 #define MBUF_RAW_ALLOC_CHECK(m) do { \
962 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
963 RTE_ASSERT((m)->next == NULL); \
964 RTE_ASSERT((m)->nb_segs == 1); \
965 __rte_mbuf_sanity_check(m, 0); \
969 * Allocate an uninitialized mbuf from mempool *mp*.
971 * This function can be used by PMDs (especially in RX functions) to
972 * allocate an uninitialized mbuf. The driver is responsible of
973 * initializing all the required fields. See rte_pktmbuf_reset().
974 * For standard needs, prefer rte_pktmbuf_alloc().
976 * The caller can expect that the following fields of the mbuf structure
977 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
978 * next=NULL, pool, priv_size. The other fields must be initialized
982 * The mempool from which mbuf is allocated.
984 * - The pointer to the new mbuf on success.
985 * - NULL if allocation failed.
987 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
991 if (rte_mempool_get(mp, (void **)&m) < 0)
993 MBUF_RAW_ALLOC_CHECK(m);
998 * Put mbuf back into its original mempool.
1000 * The caller must ensure that the mbuf is direct and properly
1001 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
1002 * rte_pktmbuf_prefree_seg().
1004 * This function should be used with care, when optimization is
1005 * required. For standard needs, prefer rte_pktmbuf_free() or
1006 * rte_pktmbuf_free_seg().
1009 * The mbuf to be freed.
1011 static __rte_always_inline void
1012 rte_mbuf_raw_free(struct rte_mbuf *m)
1014 RTE_ASSERT(RTE_MBUF_DIRECT(m));
1015 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
1016 RTE_ASSERT(m->next == NULL);
1017 RTE_ASSERT(m->nb_segs == 1);
1018 __rte_mbuf_sanity_check(m, 0);
1019 rte_mempool_put(m->pool, m);
1022 /* compat with older versions */
1025 __rte_mbuf_raw_free(struct rte_mbuf *m)
1027 rte_mbuf_raw_free(m);
1031 * The packet mbuf constructor.
1033 * This function initializes some fields in the mbuf structure that are
1034 * not modified by the user once created (origin pool, buffer start
1035 * address, and so on). This function is given as a callback function to
1036 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
1039 * The mempool from which mbufs originate.
1041 * A pointer that can be used by the user to retrieve useful information
1042 * for mbuf initialization. This pointer is the opaque argument passed to
1043 * rte_mempool_obj_iter() or rte_mempool_create().
1045 * The mbuf to initialize.
1047 * The index of the mbuf in the pool table.
1049 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
1050 void *m, unsigned i);
1054 * A packet mbuf pool constructor.
1056 * This function initializes the mempool private data in the case of a
1057 * pktmbuf pool. This private data is needed by the driver. The
1058 * function must be called on the mempool before it is used, or it
1059 * can be given as a callback function to rte_mempool_create() at
1060 * pool creation. It can be extended by the user, for example, to
1061 * provide another packet size.
1064 * The mempool from which mbufs originate.
1066 * A pointer that can be used by the user to retrieve useful information
1067 * for mbuf initialization. This pointer is the opaque argument passed to
1068 * rte_mempool_create().
1070 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
1073 * Create a mbuf pool.
1075 * This function creates and initializes a packet mbuf pool. It is
1076 * a wrapper to rte_mempool functions.
1079 * The name of the mbuf pool.
1081 * The number of elements in the mbuf pool. The optimum size (in terms
1082 * of memory usage) for a mempool is when n is a power of two minus one:
1085 * Size of the per-core object cache. See rte_mempool_create() for
1088 * Size of application private are between the rte_mbuf structure
1089 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1090 * @param data_room_size
1091 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1093 * The socket identifier where the memory should be allocated. The
1094 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1097 * The pointer to the new allocated mempool, on success. NULL on error
1098 * with rte_errno set appropriately. Possible rte_errno values include:
1099 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1100 * - E_RTE_SECONDARY - function was called from a secondary process instance
1101 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1102 * - ENOSPC - the maximum number of memzones has already been allocated
1103 * - EEXIST - a memzone with the same name already exists
1104 * - ENOMEM - no appropriate memory area found in which to create memzone
1106 struct rte_mempool *
1107 rte_pktmbuf_pool_create(const char *name, unsigned n,
1108 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
1112 * Create a mbuf pool with a given mempool ops name
1114 * This function creates and initializes a packet mbuf pool. It is
1115 * a wrapper to rte_mempool functions.
1118 * The name of the mbuf pool.
1120 * The number of elements in the mbuf pool. The optimum size (in terms
1121 * of memory usage) for a mempool is when n is a power of two minus one:
1124 * Size of the per-core object cache. See rte_mempool_create() for
1127 * Size of application private are between the rte_mbuf structure
1128 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1129 * @param data_room_size
1130 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1132 * The socket identifier where the memory should be allocated. The
1133 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1136 * The mempool ops name to be used for this mempool instead of
1137 * default mempool. The value can be *NULL* to use default mempool.
1139 * The pointer to the new allocated mempool, on success. NULL on error
1140 * with rte_errno set appropriately. Possible rte_errno values include:
1141 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1142 * - E_RTE_SECONDARY - function was called from a secondary process instance
1143 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1144 * - ENOSPC - the maximum number of memzones has already been allocated
1145 * - EEXIST - a memzone with the same name already exists
1146 * - ENOMEM - no appropriate memory area found in which to create memzone
1148 struct rte_mempool * __rte_experimental
1149 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1150 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1151 int socket_id, const char *ops_name);
1154 * Get the data room size of mbufs stored in a pktmbuf_pool
1156 * The data room size is the amount of data that can be stored in a
1157 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1160 * The packet mbuf pool.
1162 * The data room size of mbufs stored in this mempool.
1164 static inline uint16_t
1165 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1167 struct rte_pktmbuf_pool_private *mbp_priv;
1169 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1170 return mbp_priv->mbuf_data_room_size;
1174 * Get the application private size of mbufs stored in a pktmbuf_pool
1176 * The private size of mbuf is a zone located between the rte_mbuf
1177 * structure and the data buffer where an application can store data
1178 * associated to a packet.
1181 * The packet mbuf pool.
1183 * The private size of mbufs stored in this mempool.
1185 static inline uint16_t
1186 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1188 struct rte_pktmbuf_pool_private *mbp_priv;
1190 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1191 return mbp_priv->mbuf_priv_size;
1195 * Reset the data_off field of a packet mbuf to its default value.
1197 * The given mbuf must have only one segment, which should be empty.
1200 * The packet mbuf's data_off field has to be reset.
1202 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1204 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
1205 (uint16_t)m->buf_len);
1209 * Reset the fields of a packet mbuf to their default values.
1211 * The given mbuf must have only one segment.
1214 * The packet mbuf to be resetted.
1216 #define MBUF_INVALID_PORT UINT16_MAX
1218 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1224 m->vlan_tci_outer = 0;
1226 m->port = MBUF_INVALID_PORT;
1230 rte_pktmbuf_reset_headroom(m);
1233 __rte_mbuf_sanity_check(m, 1);
1237 * Allocate a new mbuf from a mempool.
1239 * This new mbuf contains one segment, which has a length of 0. The pointer
1240 * to data is initialized to have some bytes of headroom in the buffer
1241 * (if buffer size allows).
1244 * The mempool from which the mbuf is allocated.
1246 * - The pointer to the new mbuf on success.
1247 * - NULL if allocation failed.
1249 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1252 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1253 rte_pktmbuf_reset(m);
1258 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1262 * The mempool from which mbufs are allocated.
1264 * Array of pointers to mbufs
1269 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1271 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1272 struct rte_mbuf **mbufs, unsigned count)
1277 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1281 /* To understand duff's device on loop unwinding optimization, see
1282 * https://en.wikipedia.org/wiki/Duff's_device.
1283 * Here while() loop is used rather than do() while{} to avoid extra
1284 * check if count is zero.
1286 switch (count % 4) {
1288 while (idx != count) {
1289 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1290 rte_pktmbuf_reset(mbufs[idx]);
1294 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1295 rte_pktmbuf_reset(mbufs[idx]);
1299 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1300 rte_pktmbuf_reset(mbufs[idx]);
1304 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1305 rte_pktmbuf_reset(mbufs[idx]);
1314 * Initialize shared data at the end of an external buffer before attaching
1315 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
1316 * initialization but a helper function to simply spare a few bytes at the
1317 * end of the buffer for shared data. If shared data is allocated
1318 * separately, this should not be called but application has to properly
1319 * initialize the shared data according to its need.
1321 * Free callback and its argument is saved and the refcnt is set to 1.
1324 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
1325 * after this initialization. This shall be used for
1326 * ``rte_pktmbuf_attach_extbuf()``
1329 * The pointer to the external buffer.
1330 * @param [in,out] buf_len
1331 * The pointer to length of the external buffer. Input value must be
1332 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
1333 * padding for alignment. If not enough, this function will return NULL.
1334 * Adjusted buffer length will be returned through this pointer.
1336 * Free callback function to call when the external buffer needs to be
1339 * Argument for the free callback function.
1342 * A pointer to the initialized shared data on success, return NULL
1345 static inline struct rte_mbuf_ext_shared_info *
1346 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
1347 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
1349 struct rte_mbuf_ext_shared_info *shinfo;
1350 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
1353 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
1355 if (addr <= buf_addr)
1358 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
1359 shinfo->free_cb = free_cb;
1360 shinfo->fcb_opaque = fcb_opaque;
1361 rte_mbuf_ext_refcnt_set(shinfo, 1);
1363 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
1368 * Attach an external buffer to a mbuf.
1370 * User-managed anonymous buffer can be attached to an mbuf. When attaching
1371 * it, corresponding free callback function and its argument should be
1372 * provided via shinfo. This callback function will be called once all the
1373 * mbufs are detached from the buffer (refcnt becomes zero).
1375 * The headroom for the attaching mbuf will be set to zero and this can be
1376 * properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1377 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1379 * More mbufs can be attached to the same external buffer by
1380 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1383 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1384 * ``rte_pktmbuf_detach()``.
1386 * Memory for shared data must be provided and user must initialize all of
1387 * the content properly, escpecially free callback and refcnt. The pointer
1388 * of shared data will be stored in m->shinfo.
1389 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1390 * bytes at the end of buffer for the shared data, store free callback and
1391 * its argument and set the refcnt to 1. The following is an example:
1393 * struct rte_mbuf_ext_shared_info *shinfo =
1394 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1395 * free_cb, fcb_arg);
1396 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1397 * rte_pktmbuf_reset_headroom(m);
1398 * rte_pktmbuf_adj(m, data_len);
1400 * Attaching an external buffer is quite similar to mbuf indirection in
1401 * replacing buffer addresses and length of a mbuf, but a few differences:
1402 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1403 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1404 * In such cases, the buffer area of a direct mbuf must be read-only. But
1405 * external buffer has its own refcnt and it starts from 1. Unless
1406 * multiple mbufs are attached to a mbuf having an external buffer, the
1407 * external buffer is writable.
1408 * - There's no need to allocate buffer from a mempool. Any buffer can be
1409 * attached with appropriate free callback and its IO address.
1410 * - Smaller metadata is required to maintain shared data such as refcnt.
1413 * @b EXPERIMENTAL: This API may change without prior notice.
1414 * Once external buffer is enabled by allowing experimental API,
1415 * ``RTE_MBUF_DIRECT()`` and ``RTE_MBUF_INDIRECT()`` are no longer
1416 * exclusive. A mbuf can be considered direct if it is neither indirect nor
1417 * having external buffer.
1420 * The pointer to the mbuf.
1422 * The pointer to the external buffer.
1424 * IO address of the external buffer.
1426 * The size of the external buffer.
1428 * User-provided memory for shared data of the external buffer.
1430 static inline void __rte_experimental
1431 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1432 rte_iova_t buf_iova, uint16_t buf_len,
1433 struct rte_mbuf_ext_shared_info *shinfo)
1435 /* mbuf should not be read-only */
1436 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1437 RTE_ASSERT(shinfo->free_cb != NULL);
1439 m->buf_addr = buf_addr;
1440 m->buf_iova = buf_iova;
1441 m->buf_len = buf_len;
1446 m->ol_flags |= EXT_ATTACHED_MBUF;
1451 * Detach the external buffer attached to a mbuf, same as
1452 * ``rte_pktmbuf_detach()``
1455 * The mbuf having external buffer.
1457 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1460 * Attach packet mbuf to another packet mbuf.
1462 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1463 * an external buffer, the mbuf being attached will be attached to the
1464 * external buffer instead of mbuf indirection.
1466 * Otherwise, the mbuf will be indirectly attached. After attachment we
1467 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1468 * 'direct'. The direct mbuf's reference counter is incremented.
1470 * Right now, not supported:
1471 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1472 * - mbuf we trying to attach (mi) is used by someone else
1473 * e.g. it's reference counter is greater then 1.
1476 * The indirect packet mbuf.
1478 * The packet mbuf we're attaching to.
1480 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1482 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1483 rte_mbuf_refcnt_read(mi) == 1);
1485 if (RTE_MBUF_HAS_EXTBUF(m)) {
1486 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1487 mi->ol_flags = m->ol_flags;
1488 mi->shinfo = m->shinfo;
1490 /* if m is not direct, get the mbuf that embeds the data */
1491 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1492 mi->priv_size = m->priv_size;
1493 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1496 mi->buf_iova = m->buf_iova;
1497 mi->buf_addr = m->buf_addr;
1498 mi->buf_len = m->buf_len;
1500 mi->data_off = m->data_off;
1501 mi->data_len = m->data_len;
1503 mi->vlan_tci = m->vlan_tci;
1504 mi->vlan_tci_outer = m->vlan_tci_outer;
1505 mi->tx_offload = m->tx_offload;
1509 mi->pkt_len = mi->data_len;
1511 mi->packet_type = m->packet_type;
1512 mi->timestamp = m->timestamp;
1514 __rte_mbuf_sanity_check(mi, 1);
1515 __rte_mbuf_sanity_check(m, 0);
1519 * @internal used by rte_pktmbuf_detach().
1521 * Decrement the reference counter of the external buffer. When the
1522 * reference counter becomes 0, the buffer is freed by pre-registered
1526 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1528 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1529 RTE_ASSERT(m->shinfo != NULL);
1531 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1532 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1536 * @internal used by rte_pktmbuf_detach().
1538 * Decrement the direct mbuf's reference counter. When the reference
1539 * counter becomes 0, the direct mbuf is freed.
1542 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1544 struct rte_mbuf *md;
1546 RTE_ASSERT(RTE_MBUF_INDIRECT(m));
1548 md = rte_mbuf_from_indirect(m);
1550 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1553 rte_mbuf_refcnt_set(md, 1);
1554 rte_mbuf_raw_free(md);
1559 * Detach a packet mbuf from external buffer or direct buffer.
1561 * - decrement refcnt and free the external/direct buffer if refcnt
1563 * - restore original mbuf address and length values.
1564 * - reset pktmbuf data and data_len to their default values.
1566 * All other fields of the given packet mbuf will be left intact.
1569 * The indirect attached packet mbuf.
1571 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1573 struct rte_mempool *mp = m->pool;
1574 uint32_t mbuf_size, buf_len;
1577 if (RTE_MBUF_HAS_EXTBUF(m))
1578 __rte_pktmbuf_free_extbuf(m);
1580 __rte_pktmbuf_free_direct(m);
1582 priv_size = rte_pktmbuf_priv_size(mp);
1583 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1584 buf_len = rte_pktmbuf_data_room_size(mp);
1586 m->priv_size = priv_size;
1587 m->buf_addr = (char *)m + mbuf_size;
1588 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1589 m->buf_len = (uint16_t)buf_len;
1590 rte_pktmbuf_reset_headroom(m);
1596 * Decrease reference counter and unlink a mbuf segment
1598 * This function does the same than a free, except that it does not
1599 * return the segment to its pool.
1600 * It decreases the reference counter, and if it reaches 0, it is
1601 * detached from its parent for an indirect mbuf.
1604 * The mbuf to be unlinked
1606 * - (m) if it is the last reference. It can be recycled or freed.
1607 * - (NULL) if the mbuf still has remaining references on it.
1609 static __rte_always_inline struct rte_mbuf *
1610 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1612 __rte_mbuf_sanity_check(m, 0);
1614 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1616 if (!RTE_MBUF_DIRECT(m))
1617 rte_pktmbuf_detach(m);
1619 if (m->next != NULL) {
1626 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1628 if (!RTE_MBUF_DIRECT(m))
1629 rte_pktmbuf_detach(m);
1631 if (m->next != NULL) {
1635 rte_mbuf_refcnt_set(m, 1);
1642 /* deprecated, replaced by rte_pktmbuf_prefree_seg() */
1644 static inline struct rte_mbuf *
1645 __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1647 return rte_pktmbuf_prefree_seg(m);
1651 * Free a segment of a packet mbuf into its original mempool.
1653 * Free an mbuf, without parsing other segments in case of chained
1657 * The packet mbuf segment to be freed.
1659 static __rte_always_inline void
1660 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1662 m = rte_pktmbuf_prefree_seg(m);
1663 if (likely(m != NULL))
1664 rte_mbuf_raw_free(m);
1668 * Free a packet mbuf back into its original mempool.
1670 * Free an mbuf, and all its segments in case of chained buffers. Each
1671 * segment is added back into its original mempool.
1674 * The packet mbuf to be freed. If NULL, the function does nothing.
1676 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1678 struct rte_mbuf *m_next;
1681 __rte_mbuf_sanity_check(m, 1);
1685 rte_pktmbuf_free_seg(m);
1691 * Creates a "clone" of the given packet mbuf.
1693 * Walks through all segments of the given packet mbuf, and for each of them:
1694 * - Creates a new packet mbuf from the given pool.
1695 * - Attaches newly created mbuf to the segment.
1696 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1697 * from the original packet mbuf.
1700 * The packet mbuf to be cloned.
1702 * The mempool from which the "clone" mbufs are allocated.
1704 * - The pointer to the new "clone" mbuf on success.
1705 * - NULL if allocation fails.
1707 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1708 struct rte_mempool *mp)
1710 struct rte_mbuf *mc, *mi, **prev;
1714 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1719 pktlen = md->pkt_len;
1724 rte_pktmbuf_attach(mi, md);
1727 } while ((md = md->next) != NULL &&
1728 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1732 mc->pkt_len = pktlen;
1734 /* Allocation of new indirect segment failed */
1735 if (unlikely (mi == NULL)) {
1736 rte_pktmbuf_free(mc);
1740 __rte_mbuf_sanity_check(mc, 1);
1745 * Adds given value to the refcnt of all packet mbuf segments.
1747 * Walks through all segments of given packet mbuf and for each of them
1748 * invokes rte_mbuf_refcnt_update().
1751 * The packet mbuf whose refcnt to be updated.
1753 * The value to add to the mbuf's segments refcnt.
1755 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1757 __rte_mbuf_sanity_check(m, 1);
1760 rte_mbuf_refcnt_update(m, v);
1761 } while ((m = m->next) != NULL);
1765 * Get the headroom in a packet mbuf.
1770 * The length of the headroom.
1772 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1774 __rte_mbuf_sanity_check(m, 0);
1779 * Get the tailroom of a packet mbuf.
1784 * The length of the tailroom.
1786 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1788 __rte_mbuf_sanity_check(m, 0);
1789 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1794 * Get the last segment of the packet.
1799 * The last segment of the given mbuf.
1801 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1803 __rte_mbuf_sanity_check(m, 1);
1804 while (m->next != NULL)
1810 * A macro that points to an offset into the data in the mbuf.
1812 * The returned pointer is cast to type t. Before using this
1813 * function, the user must ensure that the first segment is large
1814 * enough to accommodate its data.
1819 * The offset into the mbuf data.
1821 * The type to cast the result into.
1823 #define rte_pktmbuf_mtod_offset(m, t, o) \
1824 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1827 * A macro that points to the start of the data in the mbuf.
1829 * The returned pointer is cast to type t. Before using this
1830 * function, the user must ensure that the first segment is large
1831 * enough to accommodate its data.
1836 * The type to cast the result into.
1838 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1841 * A macro that returns the IO address that points to an offset of the
1842 * start of the data in the mbuf
1847 * The offset into the data to calculate address from.
1849 #define rte_pktmbuf_iova_offset(m, o) \
1850 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
1853 #define rte_pktmbuf_mtophys_offset(m, o) \
1854 rte_pktmbuf_iova_offset(m, o)
1857 * A macro that returns the IO address that points to the start of the
1863 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
1866 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1869 * A macro that returns the length of the packet.
1871 * The value can be read or assigned.
1876 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1879 * A macro that returns the length of the segment.
1881 * The value can be read or assigned.
1886 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1889 * Prepend len bytes to an mbuf data area.
1891 * Returns a pointer to the new
1892 * data start address. If there is not enough headroom in the first
1893 * segment, the function will return NULL, without modifying the mbuf.
1898 * The amount of data to prepend (in bytes).
1900 * A pointer to the start of the newly prepended data, or
1901 * NULL if there is not enough headroom space in the first segment
1903 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1906 __rte_mbuf_sanity_check(m, 1);
1908 if (unlikely(len > rte_pktmbuf_headroom(m)))
1912 m->data_len = (uint16_t)(m->data_len + len);
1913 m->pkt_len = (m->pkt_len + len);
1915 return (char *)m->buf_addr + m->data_off;
1919 * Append len bytes to an mbuf.
1921 * Append len bytes to an mbuf and return a pointer to the start address
1922 * of the added data. If there is not enough tailroom in the last
1923 * segment, the function will return NULL, without modifying the mbuf.
1928 * The amount of data to append (in bytes).
1930 * A pointer to the start of the newly appended data, or
1931 * NULL if there is not enough tailroom space in the last segment
1933 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1936 struct rte_mbuf *m_last;
1938 __rte_mbuf_sanity_check(m, 1);
1940 m_last = rte_pktmbuf_lastseg(m);
1941 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1944 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1945 m_last->data_len = (uint16_t)(m_last->data_len + len);
1946 m->pkt_len = (m->pkt_len + len);
1947 return (char*) tail;
1951 * Remove len bytes at the beginning of an mbuf.
1953 * Returns a pointer to the start address of the new data area. If the
1954 * length is greater than the length of the first segment, then the
1955 * function will fail and return NULL, without modifying the mbuf.
1960 * The amount of data to remove (in bytes).
1962 * A pointer to the new start of the data.
1964 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1966 __rte_mbuf_sanity_check(m, 1);
1968 if (unlikely(len > m->data_len))
1971 m->data_len = (uint16_t)(m->data_len - len);
1973 m->pkt_len = (m->pkt_len - len);
1974 return (char *)m->buf_addr + m->data_off;
1978 * Remove len bytes of data at the end of the mbuf.
1980 * If the length is greater than the length of the last segment, the
1981 * function will fail and return -1 without modifying the mbuf.
1986 * The amount of data to remove (in bytes).
1991 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
1993 struct rte_mbuf *m_last;
1995 __rte_mbuf_sanity_check(m, 1);
1997 m_last = rte_pktmbuf_lastseg(m);
1998 if (unlikely(len > m_last->data_len))
2001 m_last->data_len = (uint16_t)(m_last->data_len - len);
2002 m->pkt_len = (m->pkt_len - len);
2007 * Test if mbuf data is contiguous.
2012 * - 1, if all data is contiguous (one segment).
2013 * - 0, if there is several segments.
2015 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
2017 __rte_mbuf_sanity_check(m, 1);
2018 return !!(m->nb_segs == 1);
2022 * @internal used by rte_pktmbuf_read().
2024 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
2025 uint32_t len, void *buf);
2028 * Read len data bytes in a mbuf at specified offset.
2030 * If the data is contiguous, return the pointer in the mbuf data, else
2031 * copy the data in the buffer provided by the user and return its
2035 * The pointer to the mbuf.
2037 * The offset of the data in the mbuf.
2039 * The amount of bytes to read.
2041 * The buffer where data is copied if it is not contiguous in mbuf
2042 * data. Its length should be at least equal to the len parameter.
2044 * The pointer to the data, either in the mbuf if it is contiguous,
2045 * or in the user buffer. If mbuf is too small, NULL is returned.
2047 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
2048 uint32_t off, uint32_t len, void *buf)
2050 if (likely(off + len <= rte_pktmbuf_data_len(m)))
2051 return rte_pktmbuf_mtod_offset(m, char *, off);
2053 return __rte_pktmbuf_read(m, off, len, buf);
2057 * Chain an mbuf to another, thereby creating a segmented packet.
2059 * Note: The implementation will do a linear walk over the segments to find
2060 * the tail entry. For cases when there are many segments, it's better to
2061 * chain the entries manually.
2064 * The head of the mbuf chain (the first packet)
2066 * The mbuf to put last in the chain
2070 * - -EOVERFLOW, if the chain segment limit exceeded
2072 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
2074 struct rte_mbuf *cur_tail;
2076 /* Check for number-of-segments-overflow */
2077 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
2080 /* Chain 'tail' onto the old tail */
2081 cur_tail = rte_pktmbuf_lastseg(head);
2082 cur_tail->next = tail;
2084 /* accumulate number of segments and total length. */
2085 head->nb_segs += tail->nb_segs;
2086 head->pkt_len += tail->pkt_len;
2088 /* pkt_len is only set in the head */
2089 tail->pkt_len = tail->data_len;
2095 * Validate general requirements for Tx offload in mbuf.
2097 * This function checks correctness and completeness of Tx offload settings.
2100 * The packet mbuf to be validated.
2102 * 0 if packet is valid
2105 rte_validate_tx_offload(const struct rte_mbuf *m)
2107 uint64_t ol_flags = m->ol_flags;
2108 uint64_t inner_l3_offset = m->l2_len;
2110 /* Does packet set any of available offloads? */
2111 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
2114 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
2115 /* NB: elaborating the addition like this instead of using
2116 * += gives the result uint64_t type instead of int,
2117 * avoiding compiler warnings on gcc 8.1 at least */
2118 inner_l3_offset = inner_l3_offset + m->outer_l2_len +
2121 /* Headers are fragmented */
2122 if (rte_pktmbuf_data_len(m) < inner_l3_offset + m->l3_len + m->l4_len)
2125 /* IP checksum can be counted only for IPv4 packet */
2126 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
2129 /* IP type not set when required */
2130 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
2131 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
2134 /* Check requirements for TSO packet */
2135 if (ol_flags & PKT_TX_TCP_SEG)
2136 if ((m->tso_segsz == 0) ||
2137 ((ol_flags & PKT_TX_IPV4) &&
2138 !(ol_flags & PKT_TX_IP_CKSUM)))
2141 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
2142 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
2143 !(ol_flags & PKT_TX_OUTER_IPV4))
2150 * Linearize data in mbuf.
2152 * This function moves the mbuf data in the first segment if there is enough
2153 * tailroom. The subsequent segments are unchained and freed.
2162 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
2164 size_t seg_len, copy_len;
2166 struct rte_mbuf *m_next;
2169 if (rte_pktmbuf_is_contiguous(mbuf))
2172 /* Extend first segment to the total packet length */
2173 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
2175 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
2178 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
2179 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
2181 /* Append data from next segments to the first one */
2186 seg_len = rte_pktmbuf_data_len(m);
2187 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
2190 rte_pktmbuf_free_seg(m);
2201 * Dump an mbuf structure to a file.
2203 * Dump all fields for the given packet mbuf and all its associated
2204 * segments (in the case of a chained buffer).
2207 * A pointer to a file for output
2211 * If dump_len != 0, also dump the "dump_len" first data bytes of
2214 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
2220 #endif /* _RTE_MBUF_H_ */