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.
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. If PKT_RX_QINQ set, PKT_RX_VLAN
174 * also should be set and inner tci should be saved to mbuf->vlan_tci.
175 * If the flag PKT_RX_QINQ_STRIPPED is also present, both VLANs
176 * headers have been stripped from mbuf data, else they are still
179 #define PKT_RX_QINQ (1ULL << 20)
182 * Mask of bits used to determine the status of outer RX L4 checksum.
183 * - PKT_RX_OUTER_L4_CKSUM_UNKNOWN: no info about the outer RX L4 checksum
184 * - PKT_RX_OUTER_L4_CKSUM_BAD: the outer L4 checksum in the packet is wrong
185 * - PKT_RX_OUTER_L4_CKSUM_GOOD: the outer L4 checksum in the packet is valid
186 * - PKT_RX_OUTER_L4_CKSUM_INVALID: invalid outer L4 checksum state.
188 * The detection of PKT_RX_OUTER_L4_CKSUM_GOOD shall be based on the given
189 * HW capability, At minimum, the PMD should support
190 * PKT_RX_OUTER_L4_CKSUM_UNKNOWN and PKT_RX_OUTER_L4_CKSUM_BAD states
191 * if the DEV_RX_OFFLOAD_OUTER_UDP_CKSUM offload is available.
193 #define PKT_RX_OUTER_L4_CKSUM_MASK ((1ULL << 21) | (1ULL << 22))
195 #define PKT_RX_OUTER_L4_CKSUM_UNKNOWN 0
196 #define PKT_RX_OUTER_L4_CKSUM_BAD (1ULL << 21)
197 #define PKT_RX_OUTER_L4_CKSUM_GOOD (1ULL << 22)
198 #define PKT_RX_OUTER_L4_CKSUM_INVALID ((1ULL << 21) | (1ULL << 22))
200 /* add new RX flags here */
202 /* add new TX flags here */
205 * Outer UDP checksum offload flag. This flag is used for enabling
206 * outer UDP checksum in PMD. To use outer UDP checksum, the user needs to
207 * 1) Enable the following in mbuff,
208 * a) Fill outer_l2_len and outer_l3_len in mbuf.
209 * b) Set the PKT_TX_OUTER_UDP_CKSUM flag.
210 * c) Set the PKT_TX_OUTER_IPV4 or PKT_TX_OUTER_IPV6 flag.
211 * 2) Configure DEV_TX_OFFLOAD_OUTER_UDP_CKSUM offload flag.
213 #define PKT_TX_OUTER_UDP_CKSUM (1ULL << 41)
216 * UDP Fragmentation Offload flag. This flag is used for enabling UDP
217 * fragmentation in SW or in HW. When use UFO, mbuf->tso_segsz is used
218 * to store the MSS of UDP fragments.
220 #define PKT_TX_UDP_SEG (1ULL << 42)
223 * Request security offload processing on the TX packet.
225 #define PKT_TX_SEC_OFFLOAD (1ULL << 43)
228 * Offload the MACsec. This flag must be set by the application to enable
229 * this offload feature for a packet to be transmitted.
231 #define PKT_TX_MACSEC (1ULL << 44)
234 * Bits 45:48 used for the tunnel type.
235 * The tunnel type must be specified for TSO or checksum on the inner part
237 * These flags can be used with PKT_TX_TCP_SEG for TSO, or PKT_TX_xxx_CKSUM.
238 * The mbuf fields for inner and outer header lengths are required:
239 * outer_l2_len, outer_l3_len, l2_len, l3_len, l4_len and tso_segsz for TSO.
241 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
242 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
243 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
244 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
245 /** TX packet with MPLS-in-UDP RFC 7510 header. */
246 #define PKT_TX_TUNNEL_MPLSINUDP (0x5ULL << 45)
247 #define PKT_TX_TUNNEL_VXLAN_GPE (0x6ULL << 45)
249 * Generic IP encapsulated tunnel type, used for TSO and checksum offload.
250 * It can be used for tunnels which are not standards or listed above.
251 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_GRE
252 * or PKT_TX_TUNNEL_IPIP if possible.
253 * The ethdev must be configured with DEV_TX_OFFLOAD_IP_TNL_TSO.
254 * Outer and inner checksums are done according to the existing flags like
256 * Specific tunnel headers that contain payload length, sequence id
257 * or checksum are not expected to be updated.
259 #define PKT_TX_TUNNEL_IP (0xDULL << 45)
261 * Generic UDP encapsulated tunnel type, used for TSO and checksum offload.
262 * UDP tunnel type implies outer IP layer.
263 * It can be used for tunnels which are not standards or listed above.
264 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_VXLAN
266 * The ethdev must be configured with DEV_TX_OFFLOAD_UDP_TNL_TSO.
267 * Outer and inner checksums are done according to the existing flags like
269 * Specific tunnel headers that contain payload length, sequence id
270 * or checksum are not expected to be updated.
272 #define PKT_TX_TUNNEL_UDP (0xEULL << 45)
273 /* add new TX TUNNEL type here */
274 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
277 * Second VLAN insertion (QinQ) flag.
279 #define PKT_TX_QINQ (1ULL << 49) /**< TX packet with double VLAN inserted. */
280 /* this old name is deprecated */
281 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
284 * TCP segmentation offload. To enable this offload feature for a
285 * packet to be transmitted on hardware supporting TSO:
286 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
288 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
289 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag
290 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
292 #define PKT_TX_TCP_SEG (1ULL << 50)
294 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
297 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
298 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
299 * L4 checksum offload, the user needs to:
300 * - fill l2_len and l3_len in mbuf
301 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
302 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
304 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
305 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
306 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
307 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
308 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
311 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
312 * also be set by the application, although a PMD will only check
314 * - fill the mbuf offload information: l2_len, l3_len
316 #define PKT_TX_IP_CKSUM (1ULL << 54)
319 * Packet is IPv4. This flag must be set when using any offload feature
320 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
321 * packet. If the packet is a tunneled packet, this flag is related to
324 #define PKT_TX_IPV4 (1ULL << 55)
327 * Packet is IPv6. This flag must be set when using an offload feature
328 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
329 * packet. If the packet is a tunneled packet, this flag is related to
332 #define PKT_TX_IPV6 (1ULL << 56)
335 * TX packet is a 802.1q VLAN packet.
337 #define PKT_TX_VLAN (1ULL << 57)
338 /* this old name is deprecated */
339 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
342 * Offload the IP checksum of an external header in the hardware. The
343 * flag PKT_TX_OUTER_IPV4 should also be set by the application, although
344 * a PMD will only check PKT_TX_OUTER_IP_CKSUM.
345 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
347 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
350 * Packet outer header is IPv4. This flag must be set when using any
351 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
352 * outer header of the tunneled packet is an IPv4 packet.
354 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
357 * Packet outer header is IPv6. This flag must be set when using any
358 * outer offload feature (L4 checksum) to tell the NIC that the outer
359 * header of the tunneled packet is an IPv6 packet.
361 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
364 * Bitmask of all supported packet Tx offload features flags,
365 * which can be set for packet.
367 #define PKT_TX_OFFLOAD_MASK ( \
368 PKT_TX_OUTER_IPV6 | \
369 PKT_TX_OUTER_IPV4 | \
370 PKT_TX_OUTER_IP_CKSUM | \
376 PKT_TX_IEEE1588_TMST | \
379 PKT_TX_TUNNEL_MASK | \
381 PKT_TX_SEC_OFFLOAD | \
383 PKT_TX_OUTER_UDP_CKSUM)
386 * Mbuf having an external buffer attached. shinfo in mbuf must be filled.
388 #define EXT_ATTACHED_MBUF (1ULL << 61)
390 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
392 /** Alignment constraint of mbuf private area. */
393 #define RTE_MBUF_PRIV_ALIGN 8
396 * Get the name of a RX offload flag
399 * The mask describing the flag.
401 * The name of this flag, or NULL if it's not a valid RX flag.
403 const char *rte_get_rx_ol_flag_name(uint64_t mask);
406 * Dump the list of RX offload flags in a buffer
409 * The mask describing the RX flags.
413 * The length of the buffer.
415 * 0 on success, (-1) on error.
417 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
420 * Get the name of a TX offload flag
423 * The mask describing the flag. Usually only one bit must be set.
424 * Several bits can be given if they belong to the same mask.
425 * Ex: PKT_TX_L4_MASK.
427 * The name of this flag, or NULL if it's not a valid TX flag.
429 const char *rte_get_tx_ol_flag_name(uint64_t mask);
432 * Dump the list of TX offload flags in a buffer
435 * The mask describing the TX flags.
439 * The length of the buffer.
441 * 0 on success, (-1) on error.
443 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
446 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
447 * splitting it into multiple segments.
448 * So, for mbufs that planned to be involved into RX/TX, the recommended
449 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
451 #define RTE_MBUF_DEFAULT_DATAROOM 2048
452 #define RTE_MBUF_DEFAULT_BUF_SIZE \
453 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
455 /* define a set of marker types that can be used to refer to set points in the
458 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
460 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
462 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
463 * with a single assignment */
466 * The generic rte_mbuf, containing a packet mbuf.
471 void *buf_addr; /**< Virtual address of segment buffer. */
473 * Physical address of segment buffer.
474 * Force alignment to 8-bytes, so as to ensure we have the exact
475 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
476 * working on vector drivers easier.
481 rte_iova_t buf_physaddr; /**< deprecated */
482 } __rte_aligned(sizeof(rte_iova_t));
484 /* next 8 bytes are initialised on RX descriptor rearm */
489 * Reference counter. Its size should at least equal to the size
490 * of port field (16 bits), to support zero-copy broadcast.
491 * It should only be accessed using the following functions:
492 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
493 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
494 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
499 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
500 uint16_t refcnt; /**< Non-atomically accessed refcnt */
502 uint16_t nb_segs; /**< Number of segments. */
504 /** Input port (16 bits to support more than 256 virtual ports).
505 * The event eth Tx adapter uses this field to specify the output port.
509 uint64_t ol_flags; /**< Offload features. */
511 /* remaining bytes are set on RX when pulling packet from descriptor */
512 MARKER rx_descriptor_fields1;
515 * The packet type, which is the combination of outer/inner L2, L3, L4
516 * and tunnel types. The packet_type is about data really present in the
517 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
518 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
519 * vlan is stripped from the data.
523 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
525 uint32_t l2_type:4; /**< (Outer) L2 type. */
526 uint32_t l3_type:4; /**< (Outer) L3 type. */
527 uint32_t l4_type:4; /**< (Outer) L4 type. */
528 uint32_t tun_type:4; /**< Tunnel type. */
531 uint8_t inner_esp_next_proto;
532 /**< ESP next protocol type, valid if
533 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
538 uint8_t inner_l2_type:4;
539 /**< Inner L2 type. */
540 uint8_t inner_l3_type:4;
541 /**< Inner L3 type. */
544 uint32_t inner_l4_type:4; /**< Inner L4 type. */
548 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
549 uint16_t data_len; /**< Amount of data in segment buffer. */
550 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
554 uint32_t rss; /**< RSS hash result if RSS enabled */
563 /**< Second 4 flexible bytes */
566 /**< First 4 flexible bytes or FD ID, dependent on
567 PKT_RX_FDIR_* flag in ol_flags. */
568 } fdir; /**< Filter identifier if FDIR enabled */
572 /**< The event eth Tx adapter uses this field to store
573 * Tx queue id. @see rte_event_eth_tx_adapter_txq_set()
575 } sched; /**< Hierarchical scheduler */
576 uint32_t usr; /**< User defined tags. See rte_distributor_process() */
577 } hash; /**< hash information */
579 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
580 uint16_t vlan_tci_outer;
582 uint16_t buf_len; /**< Length of segment buffer. */
584 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
585 * are not normalized but are always the same for a given port.
589 /* second cache line - fields only used in slow path or on TX */
590 MARKER cacheline1 __rte_cache_min_aligned;
594 void *userdata; /**< Can be used for external metadata */
595 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
598 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
599 struct rte_mbuf *next; /**< Next segment of scattered packet. */
601 /* fields to support TX offloads */
604 uint64_t tx_offload; /**< combined for easy fetch */
608 /**< L2 (MAC) Header Length for non-tunneling pkt.
609 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
611 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
612 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
613 uint64_t tso_segsz:16; /**< TCP TSO segment size */
615 /* fields for TX offloading of tunnels */
616 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
617 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
619 /* uint64_t unused:8; */
623 /** Size of the application private data. In case of an indirect
624 * mbuf, it stores the direct mbuf private data size. */
627 /** Timesync flags for use with IEEE1588. */
630 /** Sequence number. See also rte_reorder_insert(). */
633 /** Shared data for external buffer attached to mbuf. See
634 * rte_pktmbuf_attach_extbuf().
636 struct rte_mbuf_ext_shared_info *shinfo;
638 } __rte_cache_aligned;
641 * Function typedef of callback to free externally attached buffer.
643 typedef void (*rte_mbuf_extbuf_free_callback_t)(void *addr, void *opaque);
646 * Shared data at the end of an external buffer.
648 struct rte_mbuf_ext_shared_info {
649 rte_mbuf_extbuf_free_callback_t free_cb; /**< Free callback function */
650 void *fcb_opaque; /**< Free callback argument */
651 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
654 /**< Maximum number of nb_segs allowed. */
655 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
658 * Prefetch the first part of the mbuf
660 * The first 64 bytes of the mbuf corresponds to fields that are used early
661 * in the receive path. If the cache line of the architecture is higher than
662 * 64B, the second part will also be prefetched.
665 * The pointer to the mbuf.
668 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
670 rte_prefetch0(&m->cacheline0);
674 * Prefetch the second part of the mbuf
676 * The next 64 bytes of the mbuf corresponds to fields that are used in the
677 * transmit path. If the cache line of the architecture is higher than 64B,
678 * this function does nothing as it is expected that the full mbuf is
682 * The pointer to the mbuf.
685 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
687 #if RTE_CACHE_LINE_SIZE == 64
688 rte_prefetch0(&m->cacheline1);
695 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
698 * Return the IO address of the beginning of the mbuf data
701 * The pointer to the mbuf.
703 * The IO address of the beginning of the mbuf data
705 static inline rte_iova_t
706 rte_mbuf_data_iova(const struct rte_mbuf *mb)
708 return mb->buf_iova + mb->data_off;
712 static inline phys_addr_t
713 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
715 return rte_mbuf_data_iova(mb);
719 * Return the default IO address of the beginning of the mbuf data
721 * This function is used by drivers in their receive function, as it
722 * returns the location where data should be written by the NIC, taking
723 * the default headroom in account.
726 * The pointer to the mbuf.
728 * The IO address of the beginning of the mbuf data
730 static inline rte_iova_t
731 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
733 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
737 static inline phys_addr_t
738 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
740 return rte_mbuf_data_iova_default(mb);
744 * Return the mbuf owning the data buffer address of an indirect mbuf.
747 * The pointer to the indirect mbuf.
749 * The address of the direct mbuf corresponding to buffer_addr.
751 static inline struct rte_mbuf *
752 rte_mbuf_from_indirect(struct rte_mbuf *mi)
754 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
758 * Return the buffer address embedded in the given mbuf.
761 * The pointer to the mbuf.
763 * The address of the data buffer owned by the mbuf.
766 rte_mbuf_to_baddr(struct rte_mbuf *md)
769 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
774 * Return the starting address of the private data area embedded in
777 * Note that no check is made to ensure that a private data area
778 * actually exists in the supplied mbuf.
781 * The pointer to the mbuf.
783 * The starting address of the private data area of the given mbuf.
785 static inline void * __rte_experimental
786 rte_mbuf_to_priv(struct rte_mbuf *m)
788 return RTE_PTR_ADD(m, sizeof(struct rte_mbuf));
792 * Returns TRUE if given mbuf is cloned by mbuf indirection, or FALSE
795 * If a mbuf has its data in another mbuf and references it by mbuf
796 * indirection, this mbuf can be defined as a cloned mbuf.
798 #define RTE_MBUF_CLONED(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
802 * Use RTE_MBUF_CLONED().
804 #define RTE_MBUF_INDIRECT(mb) RTE_MBUF_CLONED(mb)
807 * Returns TRUE if given mbuf has an external buffer, or FALSE otherwise.
809 * External buffer is a user-provided anonymous buffer.
811 #define RTE_MBUF_HAS_EXTBUF(mb) ((mb)->ol_flags & EXT_ATTACHED_MBUF)
814 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
816 * If a mbuf embeds its own data after the rte_mbuf structure, this mbuf
817 * can be defined as a direct mbuf.
819 #define RTE_MBUF_DIRECT(mb) \
820 (!((mb)->ol_flags & (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF)))
823 * Private data in case of pktmbuf pool.
825 * A structure that contains some pktmbuf_pool-specific data that are
826 * appended after the mempool structure (in private data).
828 struct rte_pktmbuf_pool_private {
829 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
830 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
833 #ifdef RTE_LIBRTE_MBUF_DEBUG
835 /** check mbuf type in debug mode */
836 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
838 #else /* RTE_LIBRTE_MBUF_DEBUG */
840 /** check mbuf type in debug mode */
841 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
843 #endif /* RTE_LIBRTE_MBUF_DEBUG */
845 #ifdef RTE_MBUF_REFCNT_ATOMIC
848 * Reads the value of an mbuf's refcnt.
852 * Reference count number.
854 static inline uint16_t
855 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
857 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
861 * Sets an mbuf's refcnt to a defined value.
868 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
870 rte_atomic16_set(&m->refcnt_atomic, (int16_t)new_value);
874 static inline uint16_t
875 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
877 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
881 * Adds given value to an mbuf's refcnt and returns its new value.
885 * Value to add/subtract
889 static inline uint16_t
890 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
893 * The atomic_add is an expensive operation, so we don't want to
894 * call it in the case where we know we are the uniq holder of
895 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
896 * operation has to be used because concurrent accesses on the
897 * reference counter can occur.
899 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
901 rte_mbuf_refcnt_set(m, (uint16_t)value);
902 return (uint16_t)value;
905 return __rte_mbuf_refcnt_update(m, value);
908 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
911 static inline uint16_t
912 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
914 m->refcnt = (uint16_t)(m->refcnt + value);
919 * Adds given value to an mbuf's refcnt and returns its new value.
921 static inline uint16_t
922 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
924 return __rte_mbuf_refcnt_update(m, value);
928 * Reads the value of an mbuf's refcnt.
930 static inline uint16_t
931 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
937 * Sets an mbuf's refcnt to the defined value.
940 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
942 m->refcnt = new_value;
945 #endif /* RTE_MBUF_REFCNT_ATOMIC */
948 * Reads the refcnt of an external buffer.
951 * Shared data of the external buffer.
953 * Reference count number.
955 static inline uint16_t
956 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
958 return (uint16_t)(rte_atomic16_read(&shinfo->refcnt_atomic));
962 * Set refcnt of an external buffer.
965 * Shared data of the external buffer.
970 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
973 rte_atomic16_set(&shinfo->refcnt_atomic, (int16_t)new_value);
977 * Add given value to refcnt of an external buffer and return its new
981 * Shared data of the external buffer.
983 * Value to add/subtract
987 static inline uint16_t
988 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
991 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
993 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
994 return (uint16_t)value;
997 return (uint16_t)rte_atomic16_add_return(&shinfo->refcnt_atomic, value);
1000 /** Mbuf prefetch */
1001 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
1008 * Sanity checks on an mbuf.
1010 * Check the consistency of the given mbuf. The function will cause a
1011 * panic if corruption is detected.
1014 * The mbuf to be checked.
1016 * True if the mbuf is a packet header, false if it is a sub-segment
1017 * of a packet (in this case, some fields like nb_segs are not checked)
1020 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
1022 #define MBUF_RAW_ALLOC_CHECK(m) do { \
1023 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
1024 RTE_ASSERT((m)->next == NULL); \
1025 RTE_ASSERT((m)->nb_segs == 1); \
1026 __rte_mbuf_sanity_check(m, 0); \
1030 * Allocate an uninitialized mbuf from mempool *mp*.
1032 * This function can be used by PMDs (especially in RX functions) to
1033 * allocate an uninitialized mbuf. The driver is responsible of
1034 * initializing all the required fields. See rte_pktmbuf_reset().
1035 * For standard needs, prefer rte_pktmbuf_alloc().
1037 * The caller can expect that the following fields of the mbuf structure
1038 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
1039 * next=NULL, pool, priv_size. The other fields must be initialized
1043 * The mempool from which mbuf is allocated.
1045 * - The pointer to the new mbuf on success.
1046 * - NULL if allocation failed.
1048 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
1052 if (rte_mempool_get(mp, (void **)&m) < 0)
1054 MBUF_RAW_ALLOC_CHECK(m);
1059 * Put mbuf back into its original mempool.
1061 * The caller must ensure that the mbuf is direct and properly
1062 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
1063 * rte_pktmbuf_prefree_seg().
1065 * This function should be used with care, when optimization is
1066 * required. For standard needs, prefer rte_pktmbuf_free() or
1067 * rte_pktmbuf_free_seg().
1070 * The mbuf to be freed.
1072 static __rte_always_inline void
1073 rte_mbuf_raw_free(struct rte_mbuf *m)
1075 RTE_ASSERT(RTE_MBUF_DIRECT(m));
1076 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
1077 RTE_ASSERT(m->next == NULL);
1078 RTE_ASSERT(m->nb_segs == 1);
1079 __rte_mbuf_sanity_check(m, 0);
1080 rte_mempool_put(m->pool, m);
1084 * The packet mbuf constructor.
1086 * This function initializes some fields in the mbuf structure that are
1087 * not modified by the user once created (origin pool, buffer start
1088 * address, and so on). This function is given as a callback function to
1089 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
1092 * The mempool from which mbufs originate.
1094 * A pointer that can be used by the user to retrieve useful information
1095 * for mbuf initialization. This pointer is the opaque argument passed to
1096 * rte_mempool_obj_iter() or rte_mempool_create().
1098 * The mbuf to initialize.
1100 * The index of the mbuf in the pool table.
1102 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
1103 void *m, unsigned i);
1107 * A packet mbuf pool constructor.
1109 * This function initializes the mempool private data in the case of a
1110 * pktmbuf pool. This private data is needed by the driver. The
1111 * function must be called on the mempool before it is used, or it
1112 * can be given as a callback function to rte_mempool_create() at
1113 * pool creation. It can be extended by the user, for example, to
1114 * provide another packet size.
1117 * The mempool from which mbufs originate.
1119 * A pointer that can be used by the user to retrieve useful information
1120 * for mbuf initialization. This pointer is the opaque argument passed to
1121 * rte_mempool_create().
1123 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
1126 * Create a mbuf pool.
1128 * This function creates and initializes a packet mbuf pool. It is
1129 * a wrapper to rte_mempool functions.
1132 * The name of the mbuf pool.
1134 * The number of elements in the mbuf pool. The optimum size (in terms
1135 * of memory usage) for a mempool is when n is a power of two minus one:
1138 * Size of the per-core object cache. See rte_mempool_create() for
1141 * Size of application private are between the rte_mbuf structure
1142 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1143 * @param data_room_size
1144 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1146 * The socket identifier where the memory should be allocated. The
1147 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1150 * The pointer to the new allocated mempool, on success. NULL on error
1151 * with rte_errno set appropriately. Possible rte_errno values include:
1152 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1153 * - E_RTE_SECONDARY - function was called from a secondary process instance
1154 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1155 * - ENOSPC - the maximum number of memzones has already been allocated
1156 * - EEXIST - a memzone with the same name already exists
1157 * - ENOMEM - no appropriate memory area found in which to create memzone
1159 struct rte_mempool *
1160 rte_pktmbuf_pool_create(const char *name, unsigned n,
1161 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
1165 * Create a mbuf pool with a given mempool ops name
1167 * This function creates and initializes a packet mbuf pool. It is
1168 * a wrapper to rte_mempool functions.
1171 * The name of the mbuf pool.
1173 * The number of elements in the mbuf pool. The optimum size (in terms
1174 * of memory usage) for a mempool is when n is a power of two minus one:
1177 * Size of the per-core object cache. See rte_mempool_create() for
1180 * Size of application private are between the rte_mbuf structure
1181 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1182 * @param data_room_size
1183 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1185 * The socket identifier where the memory should be allocated. The
1186 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1189 * The mempool ops name to be used for this mempool instead of
1190 * default mempool. The value can be *NULL* to use default mempool.
1192 * The pointer to the new allocated mempool, on success. NULL on error
1193 * with rte_errno set appropriately. Possible rte_errno values include:
1194 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1195 * - E_RTE_SECONDARY - function was called from a secondary process instance
1196 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1197 * - ENOSPC - the maximum number of memzones has already been allocated
1198 * - EEXIST - a memzone with the same name already exists
1199 * - ENOMEM - no appropriate memory area found in which to create memzone
1201 struct rte_mempool *
1202 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1203 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1204 int socket_id, const char *ops_name);
1207 * Get the data room size of mbufs stored in a pktmbuf_pool
1209 * The data room size is the amount of data that can be stored in a
1210 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1213 * The packet mbuf pool.
1215 * The data room size of mbufs stored in this mempool.
1217 static inline uint16_t
1218 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1220 struct rte_pktmbuf_pool_private *mbp_priv;
1222 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1223 return mbp_priv->mbuf_data_room_size;
1227 * Get the application private size of mbufs stored in a pktmbuf_pool
1229 * The private size of mbuf is a zone located between the rte_mbuf
1230 * structure and the data buffer where an application can store data
1231 * associated to a packet.
1234 * The packet mbuf pool.
1236 * The private size of mbufs stored in this mempool.
1238 static inline uint16_t
1239 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1241 struct rte_pktmbuf_pool_private *mbp_priv;
1243 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1244 return mbp_priv->mbuf_priv_size;
1248 * Reset the data_off field of a packet mbuf to its default value.
1250 * The given mbuf must have only one segment, which should be empty.
1253 * The packet mbuf's data_off field has to be reset.
1255 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1257 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
1258 (uint16_t)m->buf_len);
1262 * Reset the fields of a packet mbuf to their default values.
1264 * The given mbuf must have only one segment.
1267 * The packet mbuf to be resetted.
1269 #define MBUF_INVALID_PORT UINT16_MAX
1271 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1277 m->vlan_tci_outer = 0;
1279 m->port = MBUF_INVALID_PORT;
1283 rte_pktmbuf_reset_headroom(m);
1286 __rte_mbuf_sanity_check(m, 1);
1290 * Allocate a new mbuf from a mempool.
1292 * This new mbuf contains one segment, which has a length of 0. The pointer
1293 * to data is initialized to have some bytes of headroom in the buffer
1294 * (if buffer size allows).
1297 * The mempool from which the mbuf is allocated.
1299 * - The pointer to the new mbuf on success.
1300 * - NULL if allocation failed.
1302 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1305 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1306 rte_pktmbuf_reset(m);
1311 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1315 * The mempool from which mbufs are allocated.
1317 * Array of pointers to mbufs
1322 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1324 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1325 struct rte_mbuf **mbufs, unsigned count)
1330 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1334 /* To understand duff's device on loop unwinding optimization, see
1335 * https://en.wikipedia.org/wiki/Duff's_device.
1336 * Here while() loop is used rather than do() while{} to avoid extra
1337 * check if count is zero.
1339 switch (count % 4) {
1341 while (idx != count) {
1342 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1343 rte_pktmbuf_reset(mbufs[idx]);
1347 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1348 rte_pktmbuf_reset(mbufs[idx]);
1352 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1353 rte_pktmbuf_reset(mbufs[idx]);
1357 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1358 rte_pktmbuf_reset(mbufs[idx]);
1367 * Initialize shared data at the end of an external buffer before attaching
1368 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
1369 * initialization but a helper function to simply spare a few bytes at the
1370 * end of the buffer for shared data. If shared data is allocated
1371 * separately, this should not be called but application has to properly
1372 * initialize the shared data according to its need.
1374 * Free callback and its argument is saved and the refcnt is set to 1.
1377 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
1378 * after this initialization. This shall be used for
1379 * ``rte_pktmbuf_attach_extbuf()``
1382 * The pointer to the external buffer.
1383 * @param [in,out] buf_len
1384 * The pointer to length of the external buffer. Input value must be
1385 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
1386 * padding for alignment. If not enough, this function will return NULL.
1387 * Adjusted buffer length will be returned through this pointer.
1389 * Free callback function to call when the external buffer needs to be
1392 * Argument for the free callback function.
1395 * A pointer to the initialized shared data on success, return NULL
1398 static inline struct rte_mbuf_ext_shared_info *
1399 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
1400 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
1402 struct rte_mbuf_ext_shared_info *shinfo;
1403 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
1406 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
1408 if (addr <= buf_addr)
1411 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
1412 shinfo->free_cb = free_cb;
1413 shinfo->fcb_opaque = fcb_opaque;
1414 rte_mbuf_ext_refcnt_set(shinfo, 1);
1416 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
1421 * Attach an external buffer to a mbuf.
1423 * User-managed anonymous buffer can be attached to an mbuf. When attaching
1424 * it, corresponding free callback function and its argument should be
1425 * provided via shinfo. This callback function will be called once all the
1426 * mbufs are detached from the buffer (refcnt becomes zero).
1428 * The headroom for the attaching mbuf will be set to zero and this can be
1429 * properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1430 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1432 * More mbufs can be attached to the same external buffer by
1433 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1436 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1437 * ``rte_pktmbuf_detach()``.
1439 * Memory for shared data must be provided and user must initialize all of
1440 * the content properly, escpecially free callback and refcnt. The pointer
1441 * of shared data will be stored in m->shinfo.
1442 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1443 * bytes at the end of buffer for the shared data, store free callback and
1444 * its argument and set the refcnt to 1. The following is an example:
1446 * struct rte_mbuf_ext_shared_info *shinfo =
1447 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1448 * free_cb, fcb_arg);
1449 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1450 * rte_pktmbuf_reset_headroom(m);
1451 * rte_pktmbuf_adj(m, data_len);
1453 * Attaching an external buffer is quite similar to mbuf indirection in
1454 * replacing buffer addresses and length of a mbuf, but a few differences:
1455 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1456 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1457 * In such cases, the buffer area of a direct mbuf must be read-only. But
1458 * external buffer has its own refcnt and it starts from 1. Unless
1459 * multiple mbufs are attached to a mbuf having an external buffer, the
1460 * external buffer is writable.
1461 * - There's no need to allocate buffer from a mempool. Any buffer can be
1462 * attached with appropriate free callback and its IO address.
1463 * - Smaller metadata is required to maintain shared data such as refcnt.
1466 * @b EXPERIMENTAL: This API may change without prior notice.
1467 * Once external buffer is enabled by allowing experimental API,
1468 * ``RTE_MBUF_DIRECT()`` and ``RTE_MBUF_INDIRECT()`` are no longer
1469 * exclusive. A mbuf can be considered direct if it is neither indirect nor
1470 * having external buffer.
1473 * The pointer to the mbuf.
1475 * The pointer to the external buffer.
1477 * IO address of the external buffer.
1479 * The size of the external buffer.
1481 * User-provided memory for shared data of the external buffer.
1483 static inline void __rte_experimental
1484 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1485 rte_iova_t buf_iova, uint16_t buf_len,
1486 struct rte_mbuf_ext_shared_info *shinfo)
1488 /* mbuf should not be read-only */
1489 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1490 RTE_ASSERT(shinfo->free_cb != NULL);
1492 m->buf_addr = buf_addr;
1493 m->buf_iova = buf_iova;
1494 m->buf_len = buf_len;
1499 m->ol_flags |= EXT_ATTACHED_MBUF;
1504 * Detach the external buffer attached to a mbuf, same as
1505 * ``rte_pktmbuf_detach()``
1508 * The mbuf having external buffer.
1510 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1513 * Attach packet mbuf to another packet mbuf.
1515 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1516 * an external buffer, the mbuf being attached will be attached to the
1517 * external buffer instead of mbuf indirection.
1519 * Otherwise, the mbuf will be indirectly attached. After attachment we
1520 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1521 * 'direct'. The direct mbuf's reference counter is incremented.
1523 * Right now, not supported:
1524 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1525 * - mbuf we trying to attach (mi) is used by someone else
1526 * e.g. it's reference counter is greater then 1.
1529 * The indirect packet mbuf.
1531 * The packet mbuf we're attaching to.
1533 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1535 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1536 rte_mbuf_refcnt_read(mi) == 1);
1538 if (RTE_MBUF_HAS_EXTBUF(m)) {
1539 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1540 mi->ol_flags = m->ol_flags;
1541 mi->shinfo = m->shinfo;
1543 /* if m is not direct, get the mbuf that embeds the data */
1544 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1545 mi->priv_size = m->priv_size;
1546 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1549 mi->buf_iova = m->buf_iova;
1550 mi->buf_addr = m->buf_addr;
1551 mi->buf_len = m->buf_len;
1553 mi->data_off = m->data_off;
1554 mi->data_len = m->data_len;
1556 mi->vlan_tci = m->vlan_tci;
1557 mi->vlan_tci_outer = m->vlan_tci_outer;
1558 mi->tx_offload = m->tx_offload;
1562 mi->pkt_len = mi->data_len;
1564 mi->packet_type = m->packet_type;
1565 mi->timestamp = m->timestamp;
1567 __rte_mbuf_sanity_check(mi, 1);
1568 __rte_mbuf_sanity_check(m, 0);
1572 * @internal used by rte_pktmbuf_detach().
1574 * Decrement the reference counter of the external buffer. When the
1575 * reference counter becomes 0, the buffer is freed by pre-registered
1579 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1581 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1582 RTE_ASSERT(m->shinfo != NULL);
1584 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1585 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1589 * @internal used by rte_pktmbuf_detach().
1591 * Decrement the direct mbuf's reference counter. When the reference
1592 * counter becomes 0, the direct mbuf is freed.
1595 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1597 struct rte_mbuf *md;
1599 RTE_ASSERT(RTE_MBUF_INDIRECT(m));
1601 md = rte_mbuf_from_indirect(m);
1603 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1606 rte_mbuf_refcnt_set(md, 1);
1607 rte_mbuf_raw_free(md);
1612 * Detach a packet mbuf from external buffer or direct buffer.
1614 * - decrement refcnt and free the external/direct buffer if refcnt
1616 * - restore original mbuf address and length values.
1617 * - reset pktmbuf data and data_len to their default values.
1619 * All other fields of the given packet mbuf will be left intact.
1622 * The indirect attached packet mbuf.
1624 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1626 struct rte_mempool *mp = m->pool;
1627 uint32_t mbuf_size, buf_len;
1630 if (RTE_MBUF_HAS_EXTBUF(m))
1631 __rte_pktmbuf_free_extbuf(m);
1633 __rte_pktmbuf_free_direct(m);
1635 priv_size = rte_pktmbuf_priv_size(mp);
1636 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1637 buf_len = rte_pktmbuf_data_room_size(mp);
1639 m->priv_size = priv_size;
1640 m->buf_addr = (char *)m + mbuf_size;
1641 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1642 m->buf_len = (uint16_t)buf_len;
1643 rte_pktmbuf_reset_headroom(m);
1649 * Decrease reference counter and unlink a mbuf segment
1651 * This function does the same than a free, except that it does not
1652 * return the segment to its pool.
1653 * It decreases the reference counter, and if it reaches 0, it is
1654 * detached from its parent for an indirect mbuf.
1657 * The mbuf to be unlinked
1659 * - (m) if it is the last reference. It can be recycled or freed.
1660 * - (NULL) if the mbuf still has remaining references on it.
1662 static __rte_always_inline struct rte_mbuf *
1663 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1665 __rte_mbuf_sanity_check(m, 0);
1667 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1669 if (!RTE_MBUF_DIRECT(m))
1670 rte_pktmbuf_detach(m);
1672 if (m->next != NULL) {
1679 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1681 if (!RTE_MBUF_DIRECT(m))
1682 rte_pktmbuf_detach(m);
1684 if (m->next != NULL) {
1688 rte_mbuf_refcnt_set(m, 1);
1696 * Free a segment of a packet mbuf into its original mempool.
1698 * Free an mbuf, without parsing other segments in case of chained
1702 * The packet mbuf segment to be freed.
1704 static __rte_always_inline void
1705 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1707 m = rte_pktmbuf_prefree_seg(m);
1708 if (likely(m != NULL))
1709 rte_mbuf_raw_free(m);
1713 * Free a packet mbuf back into its original mempool.
1715 * Free an mbuf, and all its segments in case of chained buffers. Each
1716 * segment is added back into its original mempool.
1719 * The packet mbuf to be freed. If NULL, the function does nothing.
1721 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1723 struct rte_mbuf *m_next;
1726 __rte_mbuf_sanity_check(m, 1);
1730 rte_pktmbuf_free_seg(m);
1736 * Creates a "clone" of the given packet mbuf.
1738 * Walks through all segments of the given packet mbuf, and for each of them:
1739 * - Creates a new packet mbuf from the given pool.
1740 * - Attaches newly created mbuf to the segment.
1741 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1742 * from the original packet mbuf.
1745 * The packet mbuf to be cloned.
1747 * The mempool from which the "clone" mbufs are allocated.
1749 * - The pointer to the new "clone" mbuf on success.
1750 * - NULL if allocation fails.
1752 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1753 struct rte_mempool *mp)
1755 struct rte_mbuf *mc, *mi, **prev;
1759 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1764 pktlen = md->pkt_len;
1769 rte_pktmbuf_attach(mi, md);
1772 } while ((md = md->next) != NULL &&
1773 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1777 mc->pkt_len = pktlen;
1779 /* Allocation of new indirect segment failed */
1780 if (unlikely (mi == NULL)) {
1781 rte_pktmbuf_free(mc);
1785 __rte_mbuf_sanity_check(mc, 1);
1790 * Adds given value to the refcnt of all packet mbuf segments.
1792 * Walks through all segments of given packet mbuf and for each of them
1793 * invokes rte_mbuf_refcnt_update().
1796 * The packet mbuf whose refcnt to be updated.
1798 * The value to add to the mbuf's segments refcnt.
1800 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1802 __rte_mbuf_sanity_check(m, 1);
1805 rte_mbuf_refcnt_update(m, v);
1806 } while ((m = m->next) != NULL);
1810 * Get the headroom in a packet mbuf.
1815 * The length of the headroom.
1817 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1819 __rte_mbuf_sanity_check(m, 0);
1824 * Get the tailroom of a packet mbuf.
1829 * The length of the tailroom.
1831 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1833 __rte_mbuf_sanity_check(m, 0);
1834 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1839 * Get the last segment of the packet.
1844 * The last segment of the given mbuf.
1846 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1848 __rte_mbuf_sanity_check(m, 1);
1849 while (m->next != NULL)
1855 * A macro that points to an offset into the data in the mbuf.
1857 * The returned pointer is cast to type t. Before using this
1858 * function, the user must ensure that the first segment is large
1859 * enough to accommodate its data.
1864 * The offset into the mbuf data.
1866 * The type to cast the result into.
1868 #define rte_pktmbuf_mtod_offset(m, t, o) \
1869 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1872 * A macro that points to the start of the data in the mbuf.
1874 * The returned pointer is cast to type t. Before using this
1875 * function, the user must ensure that the first segment is large
1876 * enough to accommodate its data.
1881 * The type to cast the result into.
1883 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1886 * A macro that returns the IO address that points to an offset of the
1887 * start of the data in the mbuf
1892 * The offset into the data to calculate address from.
1894 #define rte_pktmbuf_iova_offset(m, o) \
1895 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
1898 #define rte_pktmbuf_mtophys_offset(m, o) \
1899 rte_pktmbuf_iova_offset(m, o)
1902 * A macro that returns the IO address that points to the start of the
1908 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
1911 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1914 * A macro that returns the length of the packet.
1916 * The value can be read or assigned.
1921 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1924 * A macro that returns the length of the segment.
1926 * The value can be read or assigned.
1931 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1934 * Prepend len bytes to an mbuf data area.
1936 * Returns a pointer to the new
1937 * data start address. If there is not enough headroom in the first
1938 * segment, the function will return NULL, without modifying the mbuf.
1943 * The amount of data to prepend (in bytes).
1945 * A pointer to the start of the newly prepended data, or
1946 * NULL if there is not enough headroom space in the first segment
1948 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1951 __rte_mbuf_sanity_check(m, 1);
1953 if (unlikely(len > rte_pktmbuf_headroom(m)))
1956 /* NB: elaborating the subtraction like this instead of using
1957 * -= allows us to ensure the result type is uint16_t
1958 * avoiding compiler warnings on gcc 8.1 at least */
1959 m->data_off = (uint16_t)(m->data_off - len);
1960 m->data_len = (uint16_t)(m->data_len + len);
1961 m->pkt_len = (m->pkt_len + len);
1963 return (char *)m->buf_addr + m->data_off;
1967 * Append len bytes to an mbuf.
1969 * Append len bytes to an mbuf and return a pointer to the start address
1970 * of the added data. If there is not enough tailroom in the last
1971 * segment, the function will return NULL, without modifying the mbuf.
1976 * The amount of data to append (in bytes).
1978 * A pointer to the start of the newly appended data, or
1979 * NULL if there is not enough tailroom space in the last segment
1981 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1984 struct rte_mbuf *m_last;
1986 __rte_mbuf_sanity_check(m, 1);
1988 m_last = rte_pktmbuf_lastseg(m);
1989 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1992 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1993 m_last->data_len = (uint16_t)(m_last->data_len + len);
1994 m->pkt_len = (m->pkt_len + len);
1995 return (char*) tail;
1999 * Remove len bytes at the beginning of an mbuf.
2001 * Returns a pointer to the start address of the new data area. If the
2002 * length is greater than the length of the first segment, then the
2003 * function will fail and return NULL, without modifying the mbuf.
2008 * The amount of data to remove (in bytes).
2010 * A pointer to the new start of the data.
2012 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
2014 __rte_mbuf_sanity_check(m, 1);
2016 if (unlikely(len > m->data_len))
2019 /* NB: elaborating the addition like this instead of using
2020 * += allows us to ensure the result type is uint16_t
2021 * avoiding compiler warnings on gcc 8.1 at least */
2022 m->data_len = (uint16_t)(m->data_len - len);
2023 m->data_off = (uint16_t)(m->data_off + len);
2024 m->pkt_len = (m->pkt_len - len);
2025 return (char *)m->buf_addr + m->data_off;
2029 * Remove len bytes of data at the end of the mbuf.
2031 * If the length is greater than the length of the last segment, the
2032 * function will fail and return -1 without modifying the mbuf.
2037 * The amount of data to remove (in bytes).
2042 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
2044 struct rte_mbuf *m_last;
2046 __rte_mbuf_sanity_check(m, 1);
2048 m_last = rte_pktmbuf_lastseg(m);
2049 if (unlikely(len > m_last->data_len))
2052 m_last->data_len = (uint16_t)(m_last->data_len - len);
2053 m->pkt_len = (m->pkt_len - len);
2058 * Test if mbuf data is contiguous.
2063 * - 1, if all data is contiguous (one segment).
2064 * - 0, if there is several segments.
2066 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
2068 __rte_mbuf_sanity_check(m, 1);
2069 return !!(m->nb_segs == 1);
2073 * @internal used by rte_pktmbuf_read().
2075 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
2076 uint32_t len, void *buf);
2079 * Read len data bytes in a mbuf at specified offset.
2081 * If the data is contiguous, return the pointer in the mbuf data, else
2082 * copy the data in the buffer provided by the user and return its
2086 * The pointer to the mbuf.
2088 * The offset of the data in the mbuf.
2090 * The amount of bytes to read.
2092 * The buffer where data is copied if it is not contiguous in mbuf
2093 * data. Its length should be at least equal to the len parameter.
2095 * The pointer to the data, either in the mbuf if it is contiguous,
2096 * or in the user buffer. If mbuf is too small, NULL is returned.
2098 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
2099 uint32_t off, uint32_t len, void *buf)
2101 if (likely(off + len <= rte_pktmbuf_data_len(m)))
2102 return rte_pktmbuf_mtod_offset(m, char *, off);
2104 return __rte_pktmbuf_read(m, off, len, buf);
2108 * Chain an mbuf to another, thereby creating a segmented packet.
2110 * Note: The implementation will do a linear walk over the segments to find
2111 * the tail entry. For cases when there are many segments, it's better to
2112 * chain the entries manually.
2115 * The head of the mbuf chain (the first packet)
2117 * The mbuf to put last in the chain
2121 * - -EOVERFLOW, if the chain segment limit exceeded
2123 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
2125 struct rte_mbuf *cur_tail;
2127 /* Check for number-of-segments-overflow */
2128 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
2131 /* Chain 'tail' onto the old tail */
2132 cur_tail = rte_pktmbuf_lastseg(head);
2133 cur_tail->next = tail;
2135 /* accumulate number of segments and total length.
2136 * NB: elaborating the addition like this instead of using
2137 * -= allows us to ensure the result type is uint16_t
2138 * avoiding compiler warnings on gcc 8.1 at least */
2139 head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
2140 head->pkt_len += tail->pkt_len;
2142 /* pkt_len is only set in the head */
2143 tail->pkt_len = tail->data_len;
2149 * Validate general requirements for Tx offload in mbuf.
2151 * This function checks correctness and completeness of Tx offload settings.
2154 * The packet mbuf to be validated.
2156 * 0 if packet is valid
2159 rte_validate_tx_offload(const struct rte_mbuf *m)
2161 uint64_t ol_flags = m->ol_flags;
2162 uint64_t inner_l3_offset = m->l2_len;
2164 /* Does packet set any of available offloads? */
2165 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
2168 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
2169 /* NB: elaborating the addition like this instead of using
2170 * += gives the result uint64_t type instead of int,
2171 * avoiding compiler warnings on gcc 8.1 at least */
2172 inner_l3_offset = inner_l3_offset + m->outer_l2_len +
2175 /* Headers are fragmented */
2176 if (rte_pktmbuf_data_len(m) < inner_l3_offset + m->l3_len + m->l4_len)
2179 /* IP checksum can be counted only for IPv4 packet */
2180 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
2183 /* IP type not set when required */
2184 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
2185 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
2188 /* Check requirements for TSO packet */
2189 if (ol_flags & PKT_TX_TCP_SEG)
2190 if ((m->tso_segsz == 0) ||
2191 ((ol_flags & PKT_TX_IPV4) &&
2192 !(ol_flags & PKT_TX_IP_CKSUM)))
2195 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
2196 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
2197 !(ol_flags & PKT_TX_OUTER_IPV4))
2204 * Linearize data in mbuf.
2206 * This function moves the mbuf data in the first segment if there is enough
2207 * tailroom. The subsequent segments are unchained and freed.
2216 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
2218 size_t seg_len, copy_len;
2220 struct rte_mbuf *m_next;
2223 if (rte_pktmbuf_is_contiguous(mbuf))
2226 /* Extend first segment to the total packet length */
2227 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
2229 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
2232 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
2233 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
2235 /* Append data from next segments to the first one */
2240 seg_len = rte_pktmbuf_data_len(m);
2241 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
2244 rte_pktmbuf_free_seg(m);
2255 * Dump an mbuf structure to a file.
2257 * Dump all fields for the given packet mbuf and all its associated
2258 * segments (in the case of a chained buffer).
2261 * A pointer to a file for output
2265 * If dump_len != 0, also dump the "dump_len" first data bytes of
2268 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
2274 #endif /* _RTE_MBUF_H_ */