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 * Indicate that the metadata field in the mbuf is in use.
207 #define PKT_TX_METADATA (1ULL << 40)
210 * Outer UDP checksum offload flag. This flag is used for enabling
211 * outer UDP checksum in PMD. To use outer UDP checksum, the user needs to
212 * 1) Enable the following in mbuff,
213 * a) Fill outer_l2_len and outer_l3_len in mbuf.
214 * b) Set the PKT_TX_OUTER_UDP_CKSUM flag.
215 * c) Set the PKT_TX_OUTER_IPV4 or PKT_TX_OUTER_IPV6 flag.
216 * 2) Configure DEV_TX_OFFLOAD_OUTER_UDP_CKSUM offload flag.
218 #define PKT_TX_OUTER_UDP_CKSUM (1ULL << 41)
221 * UDP Fragmentation Offload flag. This flag is used for enabling UDP
222 * fragmentation in SW or in HW. When use UFO, mbuf->tso_segsz is used
223 * to store the MSS of UDP fragments.
225 #define PKT_TX_UDP_SEG (1ULL << 42)
228 * Request security offload processing on the TX packet.
230 #define PKT_TX_SEC_OFFLOAD (1ULL << 43)
233 * Offload the MACsec. This flag must be set by the application to enable
234 * this offload feature for a packet to be transmitted.
236 #define PKT_TX_MACSEC (1ULL << 44)
239 * Bits 45:48 used for the tunnel type.
240 * The tunnel type must be specified for TSO or checksum on the inner part
242 * These flags can be used with PKT_TX_TCP_SEG for TSO, or PKT_TX_xxx_CKSUM.
243 * The mbuf fields for inner and outer header lengths are required:
244 * outer_l2_len, outer_l3_len, l2_len, l3_len, l4_len and tso_segsz for TSO.
246 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
247 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
248 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
249 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
250 /** TX packet with MPLS-in-UDP RFC 7510 header. */
251 #define PKT_TX_TUNNEL_MPLSINUDP (0x5ULL << 45)
252 #define PKT_TX_TUNNEL_VXLAN_GPE (0x6ULL << 45)
254 * Generic IP encapsulated tunnel type, used for TSO and checksum offload.
255 * It can be used for tunnels which are not standards or listed above.
256 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_GRE
257 * or PKT_TX_TUNNEL_IPIP if possible.
258 * The ethdev must be configured with DEV_TX_OFFLOAD_IP_TNL_TSO.
259 * Outer and inner checksums are done according to the existing flags like
261 * Specific tunnel headers that contain payload length, sequence id
262 * or checksum are not expected to be updated.
264 #define PKT_TX_TUNNEL_IP (0xDULL << 45)
266 * Generic UDP encapsulated tunnel type, used for TSO and checksum offload.
267 * UDP tunnel type implies outer IP layer.
268 * It can be used for tunnels which are not standards or listed above.
269 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_VXLAN
271 * The ethdev must be configured with DEV_TX_OFFLOAD_UDP_TNL_TSO.
272 * Outer and inner checksums are done according to the existing flags like
274 * Specific tunnel headers that contain payload length, sequence id
275 * or checksum are not expected to be updated.
277 #define PKT_TX_TUNNEL_UDP (0xEULL << 45)
278 /* add new TX TUNNEL type here */
279 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
282 * Second VLAN insertion (QinQ) flag.
284 #define PKT_TX_QINQ (1ULL << 49) /**< TX packet with double VLAN inserted. */
285 /* this old name is deprecated */
286 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
289 * TCP segmentation offload. To enable this offload feature for a
290 * packet to be transmitted on hardware supporting TSO:
291 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
293 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
294 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag
295 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
297 #define PKT_TX_TCP_SEG (1ULL << 50)
299 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
302 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
303 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
304 * L4 checksum offload, the user needs to:
305 * - fill l2_len and l3_len in mbuf
306 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
307 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
309 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
310 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
311 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
312 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
313 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
316 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
317 * also be set by the application, although a PMD will only check
319 * - fill the mbuf offload information: l2_len, l3_len
321 #define PKT_TX_IP_CKSUM (1ULL << 54)
324 * Packet is IPv4. This flag must be set when using any offload feature
325 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
326 * packet. If the packet is a tunneled packet, this flag is related to
329 #define PKT_TX_IPV4 (1ULL << 55)
332 * Packet is IPv6. This flag must be set when using an offload feature
333 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
334 * packet. If the packet is a tunneled packet, this flag is related to
337 #define PKT_TX_IPV6 (1ULL << 56)
340 * TX packet is a 802.1q VLAN packet.
342 #define PKT_TX_VLAN (1ULL << 57)
343 /* this old name is deprecated */
344 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
347 * Offload the IP checksum of an external header in the hardware. The
348 * flag PKT_TX_OUTER_IPV4 should also be set by the application, although
349 * a PMD will only check PKT_TX_OUTER_IP_CKSUM.
350 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
352 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
355 * Packet outer header is IPv4. This flag must be set when using any
356 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
357 * outer header of the tunneled packet is an IPv4 packet.
359 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
362 * Packet outer header is IPv6. This flag must be set when using any
363 * outer offload feature (L4 checksum) to tell the NIC that the outer
364 * header of the tunneled packet is an IPv6 packet.
366 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
369 * Bitmask of all supported packet Tx offload features flags,
370 * which can be set for packet.
372 #define PKT_TX_OFFLOAD_MASK ( \
373 PKT_TX_OUTER_IPV6 | \
374 PKT_TX_OUTER_IPV4 | \
375 PKT_TX_OUTER_IP_CKSUM | \
381 PKT_TX_IEEE1588_TMST | \
384 PKT_TX_TUNNEL_MASK | \
386 PKT_TX_SEC_OFFLOAD | \
388 PKT_TX_OUTER_UDP_CKSUM | \
392 * Mbuf having an external buffer attached. shinfo in mbuf must be filled.
394 #define EXT_ATTACHED_MBUF (1ULL << 61)
396 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
398 /** Alignment constraint of mbuf private area. */
399 #define RTE_MBUF_PRIV_ALIGN 8
402 * Get the name of a RX offload flag
405 * The mask describing the flag.
407 * The name of this flag, or NULL if it's not a valid RX flag.
409 const char *rte_get_rx_ol_flag_name(uint64_t mask);
412 * Dump the list of RX offload flags in a buffer
415 * The mask describing the RX flags.
419 * The length of the buffer.
421 * 0 on success, (-1) on error.
423 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
426 * Get the name of a TX offload flag
429 * The mask describing the flag. Usually only one bit must be set.
430 * Several bits can be given if they belong to the same mask.
431 * Ex: PKT_TX_L4_MASK.
433 * The name of this flag, or NULL if it's not a valid TX flag.
435 const char *rte_get_tx_ol_flag_name(uint64_t mask);
438 * Dump the list of TX offload flags in a buffer
441 * The mask describing the TX flags.
445 * The length of the buffer.
447 * 0 on success, (-1) on error.
449 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
452 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
453 * splitting it into multiple segments.
454 * So, for mbufs that planned to be involved into RX/TX, the recommended
455 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
457 #define RTE_MBUF_DEFAULT_DATAROOM 2048
458 #define RTE_MBUF_DEFAULT_BUF_SIZE \
459 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
461 /* define a set of marker types that can be used to refer to set points in the
464 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
466 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
468 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
469 * with a single assignment */
472 * The generic rte_mbuf, containing a packet mbuf.
477 void *buf_addr; /**< Virtual address of segment buffer. */
479 * Physical address of segment buffer.
480 * Force alignment to 8-bytes, so as to ensure we have the exact
481 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
482 * working on vector drivers easier.
487 rte_iova_t buf_physaddr; /**< deprecated */
488 } __rte_aligned(sizeof(rte_iova_t));
490 /* next 8 bytes are initialised on RX descriptor rearm */
495 * Reference counter. Its size should at least equal to the size
496 * of port field (16 bits), to support zero-copy broadcast.
497 * It should only be accessed using the following functions:
498 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
499 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
500 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
505 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
506 uint16_t refcnt; /**< Non-atomically accessed refcnt */
508 uint16_t nb_segs; /**< Number of segments. */
510 /** Input port (16 bits to support more than 256 virtual ports).
511 * The event eth Tx adapter uses this field to specify the output port.
515 uint64_t ol_flags; /**< Offload features. */
517 /* remaining bytes are set on RX when pulling packet from descriptor */
518 MARKER rx_descriptor_fields1;
521 * The packet type, which is the combination of outer/inner L2, L3, L4
522 * and tunnel types. The packet_type is about data really present in the
523 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
524 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
525 * vlan is stripped from the data.
529 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
531 uint32_t l2_type:4; /**< (Outer) L2 type. */
532 uint32_t l3_type:4; /**< (Outer) L3 type. */
533 uint32_t l4_type:4; /**< (Outer) L4 type. */
534 uint32_t tun_type:4; /**< Tunnel type. */
537 uint8_t inner_esp_next_proto;
538 /**< ESP next protocol type, valid if
539 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
544 uint8_t inner_l2_type:4;
545 /**< Inner L2 type. */
546 uint8_t inner_l3_type:4;
547 /**< Inner L3 type. */
550 uint32_t inner_l4_type:4; /**< Inner L4 type. */
554 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
555 uint16_t data_len; /**< Amount of data in segment buffer. */
556 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
562 uint32_t rss; /**< RSS hash result if RSS enabled */
570 /**< Second 4 flexible bytes */
573 /**< First 4 flexible bytes or FD ID, dependent
574 * on PKT_RX_FDIR_* flag in ol_flags.
576 } fdir; /**< Filter identifier if FDIR enabled */
580 /**< The event eth Tx adapter uses this field
581 * to store Tx queue id.
582 * @see rte_event_eth_tx_adapter_txq_set()
584 } sched; /**< Hierarchical scheduler */
585 /**< User defined tags. See rte_distributor_process() */
587 } hash; /**< hash information */
590 * Application specific metadata value
591 * for egress flow rule match.
592 * Valid if PKT_TX_METADATA is set.
593 * Located here to allow conjunct use
594 * with hash.sched.hi.
596 uint32_t tx_metadata;
601 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
602 uint16_t vlan_tci_outer;
604 uint16_t buf_len; /**< Length of segment buffer. */
606 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
607 * are not normalized but are always the same for a given port.
611 /* second cache line - fields only used in slow path or on TX */
612 MARKER cacheline1 __rte_cache_min_aligned;
616 void *userdata; /**< Can be used for external metadata */
617 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
620 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
621 struct rte_mbuf *next; /**< Next segment of scattered packet. */
623 /* fields to support TX offloads */
626 uint64_t tx_offload; /**< combined for easy fetch */
630 /**< L2 (MAC) Header Length for non-tunneling pkt.
631 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
633 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
634 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
635 uint64_t tso_segsz:16; /**< TCP TSO segment size */
637 /* fields for TX offloading of tunnels */
638 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
639 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
641 /* uint64_t unused:8; */
645 /** Size of the application private data. In case of an indirect
646 * mbuf, it stores the direct mbuf private data size. */
649 /** Timesync flags for use with IEEE1588. */
652 /** Sequence number. See also rte_reorder_insert(). */
655 /** Shared data for external buffer attached to mbuf. See
656 * rte_pktmbuf_attach_extbuf().
658 struct rte_mbuf_ext_shared_info *shinfo;
660 } __rte_cache_aligned;
663 * Function typedef of callback to free externally attached buffer.
665 typedef void (*rte_mbuf_extbuf_free_callback_t)(void *addr, void *opaque);
668 * Shared data at the end of an external buffer.
670 struct rte_mbuf_ext_shared_info {
671 rte_mbuf_extbuf_free_callback_t free_cb; /**< Free callback function */
672 void *fcb_opaque; /**< Free callback argument */
673 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
676 /**< Maximum number of nb_segs allowed. */
677 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
680 * Prefetch the first part of the mbuf
682 * The first 64 bytes of the mbuf corresponds to fields that are used early
683 * in the receive path. If the cache line of the architecture is higher than
684 * 64B, the second part will also be prefetched.
687 * The pointer to the mbuf.
690 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
692 rte_prefetch0(&m->cacheline0);
696 * Prefetch the second part of the mbuf
698 * The next 64 bytes of the mbuf corresponds to fields that are used in the
699 * transmit path. If the cache line of the architecture is higher than 64B,
700 * this function does nothing as it is expected that the full mbuf is
704 * The pointer to the mbuf.
707 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
709 #if RTE_CACHE_LINE_SIZE == 64
710 rte_prefetch0(&m->cacheline1);
717 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
720 * Return the IO address of the beginning of the mbuf data
723 * The pointer to the mbuf.
725 * The IO address of the beginning of the mbuf data
727 static inline rte_iova_t
728 rte_mbuf_data_iova(const struct rte_mbuf *mb)
730 return mb->buf_iova + mb->data_off;
734 static inline phys_addr_t
735 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
737 return rte_mbuf_data_iova(mb);
741 * Return the default IO address of the beginning of the mbuf data
743 * This function is used by drivers in their receive function, as it
744 * returns the location where data should be written by the NIC, taking
745 * the default headroom in account.
748 * The pointer to the mbuf.
750 * The IO address of the beginning of the mbuf data
752 static inline rte_iova_t
753 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
755 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
759 static inline phys_addr_t
760 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
762 return rte_mbuf_data_iova_default(mb);
766 * Return the mbuf owning the data buffer address of an indirect mbuf.
769 * The pointer to the indirect mbuf.
771 * The address of the direct mbuf corresponding to buffer_addr.
773 static inline struct rte_mbuf *
774 rte_mbuf_from_indirect(struct rte_mbuf *mi)
776 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
780 * Return the buffer address embedded in the given mbuf.
783 * The pointer to the mbuf.
785 * The address of the data buffer owned by the mbuf.
788 rte_mbuf_to_baddr(struct rte_mbuf *md)
791 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
796 * Return the starting address of the private data area embedded in
799 * Note that no check is made to ensure that a private data area
800 * actually exists in the supplied mbuf.
803 * The pointer to the mbuf.
805 * The starting address of the private data area of the given mbuf.
807 static inline void * __rte_experimental
808 rte_mbuf_to_priv(struct rte_mbuf *m)
810 return RTE_PTR_ADD(m, sizeof(struct rte_mbuf));
814 * Returns TRUE if given mbuf is cloned by mbuf indirection, or FALSE
817 * If a mbuf has its data in another mbuf and references it by mbuf
818 * indirection, this mbuf can be defined as a cloned mbuf.
820 #define RTE_MBUF_CLONED(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
824 * Use RTE_MBUF_CLONED().
826 #define RTE_MBUF_INDIRECT(mb) RTE_MBUF_CLONED(mb)
829 * Returns TRUE if given mbuf has an external buffer, or FALSE otherwise.
831 * External buffer is a user-provided anonymous buffer.
833 #define RTE_MBUF_HAS_EXTBUF(mb) ((mb)->ol_flags & EXT_ATTACHED_MBUF)
836 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
838 * If a mbuf embeds its own data after the rte_mbuf structure, this mbuf
839 * can be defined as a direct mbuf.
841 #define RTE_MBUF_DIRECT(mb) \
842 (!((mb)->ol_flags & (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF)))
845 * Private data in case of pktmbuf pool.
847 * A structure that contains some pktmbuf_pool-specific data that are
848 * appended after the mempool structure (in private data).
850 struct rte_pktmbuf_pool_private {
851 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
852 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
855 #ifdef RTE_LIBRTE_MBUF_DEBUG
857 /** check mbuf type in debug mode */
858 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
860 #else /* RTE_LIBRTE_MBUF_DEBUG */
862 /** check mbuf type in debug mode */
863 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
865 #endif /* RTE_LIBRTE_MBUF_DEBUG */
867 #ifdef RTE_MBUF_REFCNT_ATOMIC
870 * Reads the value of an mbuf's refcnt.
874 * Reference count number.
876 static inline uint16_t
877 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
879 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
883 * Sets an mbuf's refcnt to a defined value.
890 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
892 rte_atomic16_set(&m->refcnt_atomic, (int16_t)new_value);
896 static inline uint16_t
897 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
899 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
903 * Adds given value to an mbuf's refcnt and returns its new value.
907 * Value to add/subtract
911 static inline uint16_t
912 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
915 * The atomic_add is an expensive operation, so we don't want to
916 * call it in the case where we know we are the uniq holder of
917 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
918 * operation has to be used because concurrent accesses on the
919 * reference counter can occur.
921 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
923 rte_mbuf_refcnt_set(m, (uint16_t)value);
924 return (uint16_t)value;
927 return __rte_mbuf_refcnt_update(m, value);
930 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
933 static inline uint16_t
934 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
936 m->refcnt = (uint16_t)(m->refcnt + value);
941 * Adds given value to an mbuf's refcnt and returns its new value.
943 static inline uint16_t
944 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
946 return __rte_mbuf_refcnt_update(m, value);
950 * Reads the value of an mbuf's refcnt.
952 static inline uint16_t
953 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
959 * Sets an mbuf's refcnt to the defined value.
962 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
964 m->refcnt = new_value;
967 #endif /* RTE_MBUF_REFCNT_ATOMIC */
970 * Reads the refcnt of an external buffer.
973 * Shared data of the external buffer.
975 * Reference count number.
977 static inline uint16_t
978 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
980 return (uint16_t)(rte_atomic16_read(&shinfo->refcnt_atomic));
984 * Set refcnt of an external buffer.
987 * Shared data of the external buffer.
992 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
995 rte_atomic16_set(&shinfo->refcnt_atomic, (int16_t)new_value);
999 * Add given value to refcnt of an external buffer and return its new
1003 * Shared data of the external buffer.
1005 * Value to add/subtract
1009 static inline uint16_t
1010 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
1013 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
1015 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
1016 return (uint16_t)value;
1019 return (uint16_t)rte_atomic16_add_return(&shinfo->refcnt_atomic, value);
1022 /** Mbuf prefetch */
1023 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
1030 * Sanity checks on an mbuf.
1032 * Check the consistency of the given mbuf. The function will cause a
1033 * panic if corruption is detected.
1036 * The mbuf to be checked.
1038 * True if the mbuf is a packet header, false if it is a sub-segment
1039 * of a packet (in this case, some fields like nb_segs are not checked)
1042 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
1044 #define MBUF_RAW_ALLOC_CHECK(m) do { \
1045 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
1046 RTE_ASSERT((m)->next == NULL); \
1047 RTE_ASSERT((m)->nb_segs == 1); \
1048 __rte_mbuf_sanity_check(m, 0); \
1052 * Allocate an uninitialized mbuf from mempool *mp*.
1054 * This function can be used by PMDs (especially in RX functions) to
1055 * allocate an uninitialized mbuf. The driver is responsible of
1056 * initializing all the required fields. See rte_pktmbuf_reset().
1057 * For standard needs, prefer rte_pktmbuf_alloc().
1059 * The caller can expect that the following fields of the mbuf structure
1060 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
1061 * next=NULL, pool, priv_size. The other fields must be initialized
1065 * The mempool from which mbuf is allocated.
1067 * - The pointer to the new mbuf on success.
1068 * - NULL if allocation failed.
1070 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
1074 if (rte_mempool_get(mp, (void **)&m) < 0)
1076 MBUF_RAW_ALLOC_CHECK(m);
1081 * Put mbuf back into its original mempool.
1083 * The caller must ensure that the mbuf is direct and properly
1084 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
1085 * rte_pktmbuf_prefree_seg().
1087 * This function should be used with care, when optimization is
1088 * required. For standard needs, prefer rte_pktmbuf_free() or
1089 * rte_pktmbuf_free_seg().
1092 * The mbuf to be freed.
1094 static __rte_always_inline void
1095 rte_mbuf_raw_free(struct rte_mbuf *m)
1097 RTE_ASSERT(RTE_MBUF_DIRECT(m));
1098 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
1099 RTE_ASSERT(m->next == NULL);
1100 RTE_ASSERT(m->nb_segs == 1);
1101 __rte_mbuf_sanity_check(m, 0);
1102 rte_mempool_put(m->pool, m);
1106 * The packet mbuf constructor.
1108 * This function initializes some fields in the mbuf structure that are
1109 * not modified by the user once created (origin pool, buffer start
1110 * address, and so on). This function is given as a callback function to
1111 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
1114 * The mempool from which mbufs originate.
1116 * A pointer that can be used by the user to retrieve useful information
1117 * for mbuf initialization. This pointer is the opaque argument passed to
1118 * rte_mempool_obj_iter() or rte_mempool_create().
1120 * The mbuf to initialize.
1122 * The index of the mbuf in the pool table.
1124 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
1125 void *m, unsigned i);
1129 * A packet mbuf pool constructor.
1131 * This function initializes the mempool private data in the case of a
1132 * pktmbuf pool. This private data is needed by the driver. The
1133 * function must be called on the mempool before it is used, or it
1134 * can be given as a callback function to rte_mempool_create() at
1135 * pool creation. It can be extended by the user, for example, to
1136 * provide another packet size.
1139 * The mempool from which mbufs originate.
1141 * A pointer that can be used by the user to retrieve useful information
1142 * for mbuf initialization. This pointer is the opaque argument passed to
1143 * rte_mempool_create().
1145 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
1148 * Create a mbuf pool.
1150 * This function creates and initializes a packet mbuf pool. It is
1151 * a wrapper to rte_mempool functions.
1154 * The name of the mbuf pool.
1156 * The number of elements in the mbuf pool. The optimum size (in terms
1157 * of memory usage) for a mempool is when n is a power of two minus one:
1160 * Size of the per-core object cache. See rte_mempool_create() for
1163 * Size of application private are between the rte_mbuf structure
1164 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1165 * @param data_room_size
1166 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1168 * The socket identifier where the memory should be allocated. The
1169 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1172 * The pointer to the new allocated mempool, on success. NULL on error
1173 * with rte_errno set appropriately. Possible rte_errno values include:
1174 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1175 * - E_RTE_SECONDARY - function was called from a secondary process instance
1176 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1177 * - ENOSPC - the maximum number of memzones has already been allocated
1178 * - EEXIST - a memzone with the same name already exists
1179 * - ENOMEM - no appropriate memory area found in which to create memzone
1181 struct rte_mempool *
1182 rte_pktmbuf_pool_create(const char *name, unsigned n,
1183 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
1187 * Create a mbuf pool with a given mempool ops name
1189 * This function creates and initializes a packet mbuf pool. It is
1190 * a wrapper to rte_mempool functions.
1193 * The name of the mbuf pool.
1195 * The number of elements in the mbuf pool. The optimum size (in terms
1196 * of memory usage) for a mempool is when n is a power of two minus one:
1199 * Size of the per-core object cache. See rte_mempool_create() for
1202 * Size of application private are between the rte_mbuf structure
1203 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1204 * @param data_room_size
1205 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1207 * The socket identifier where the memory should be allocated. The
1208 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1211 * The mempool ops name to be used for this mempool instead of
1212 * default mempool. The value can be *NULL* to use default mempool.
1214 * The pointer to the new allocated mempool, on success. NULL on error
1215 * with rte_errno set appropriately. Possible rte_errno values include:
1216 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1217 * - E_RTE_SECONDARY - function was called from a secondary process instance
1218 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1219 * - ENOSPC - the maximum number of memzones has already been allocated
1220 * - EEXIST - a memzone with the same name already exists
1221 * - ENOMEM - no appropriate memory area found in which to create memzone
1223 struct rte_mempool *
1224 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1225 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1226 int socket_id, const char *ops_name);
1229 * Get the data room size of mbufs stored in a pktmbuf_pool
1231 * The data room size is the amount of data that can be stored in a
1232 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1235 * The packet mbuf pool.
1237 * The data room size of mbufs stored in this mempool.
1239 static inline uint16_t
1240 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1242 struct rte_pktmbuf_pool_private *mbp_priv;
1244 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1245 return mbp_priv->mbuf_data_room_size;
1249 * Get the application private size of mbufs stored in a pktmbuf_pool
1251 * The private size of mbuf is a zone located between the rte_mbuf
1252 * structure and the data buffer where an application can store data
1253 * associated to a packet.
1256 * The packet mbuf pool.
1258 * The private size of mbufs stored in this mempool.
1260 static inline uint16_t
1261 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1263 struct rte_pktmbuf_pool_private *mbp_priv;
1265 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1266 return mbp_priv->mbuf_priv_size;
1270 * Reset the data_off field of a packet mbuf to its default value.
1272 * The given mbuf must have only one segment, which should be empty.
1275 * The packet mbuf's data_off field has to be reset.
1277 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1279 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
1280 (uint16_t)m->buf_len);
1284 * Reset the fields of a packet mbuf to their default values.
1286 * The given mbuf must have only one segment.
1289 * The packet mbuf to be resetted.
1291 #define MBUF_INVALID_PORT UINT16_MAX
1293 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1299 m->vlan_tci_outer = 0;
1301 m->port = MBUF_INVALID_PORT;
1305 rte_pktmbuf_reset_headroom(m);
1308 __rte_mbuf_sanity_check(m, 1);
1312 * Allocate a new mbuf from a mempool.
1314 * This new mbuf contains one segment, which has a length of 0. The pointer
1315 * to data is initialized to have some bytes of headroom in the buffer
1316 * (if buffer size allows).
1319 * The mempool from which the mbuf is allocated.
1321 * - The pointer to the new mbuf on success.
1322 * - NULL if allocation failed.
1324 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1327 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1328 rte_pktmbuf_reset(m);
1333 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1337 * The mempool from which mbufs are allocated.
1339 * Array of pointers to mbufs
1344 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1346 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1347 struct rte_mbuf **mbufs, unsigned count)
1352 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1356 /* To understand duff's device on loop unwinding optimization, see
1357 * https://en.wikipedia.org/wiki/Duff's_device.
1358 * Here while() loop is used rather than do() while{} to avoid extra
1359 * check if count is zero.
1361 switch (count % 4) {
1363 while (idx != count) {
1364 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1365 rte_pktmbuf_reset(mbufs[idx]);
1369 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1370 rte_pktmbuf_reset(mbufs[idx]);
1374 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1375 rte_pktmbuf_reset(mbufs[idx]);
1379 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1380 rte_pktmbuf_reset(mbufs[idx]);
1389 * Initialize shared data at the end of an external buffer before attaching
1390 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
1391 * initialization but a helper function to simply spare a few bytes at the
1392 * end of the buffer for shared data. If shared data is allocated
1393 * separately, this should not be called but application has to properly
1394 * initialize the shared data according to its need.
1396 * Free callback and its argument is saved and the refcnt is set to 1.
1399 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
1400 * after this initialization. This shall be used for
1401 * ``rte_pktmbuf_attach_extbuf()``
1404 * The pointer to the external buffer.
1405 * @param [in,out] buf_len
1406 * The pointer to length of the external buffer. Input value must be
1407 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
1408 * padding for alignment. If not enough, this function will return NULL.
1409 * Adjusted buffer length will be returned through this pointer.
1411 * Free callback function to call when the external buffer needs to be
1414 * Argument for the free callback function.
1417 * A pointer to the initialized shared data on success, return NULL
1420 static inline struct rte_mbuf_ext_shared_info *
1421 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
1422 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
1424 struct rte_mbuf_ext_shared_info *shinfo;
1425 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
1428 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
1430 if (addr <= buf_addr)
1433 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
1434 shinfo->free_cb = free_cb;
1435 shinfo->fcb_opaque = fcb_opaque;
1436 rte_mbuf_ext_refcnt_set(shinfo, 1);
1438 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
1443 * Attach an external buffer to a mbuf.
1445 * User-managed anonymous buffer can be attached to an mbuf. When attaching
1446 * it, corresponding free callback function and its argument should be
1447 * provided via shinfo. This callback function will be called once all the
1448 * mbufs are detached from the buffer (refcnt becomes zero).
1450 * The headroom for the attaching mbuf will be set to zero and this can be
1451 * properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1452 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1454 * More mbufs can be attached to the same external buffer by
1455 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1458 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1459 * ``rte_pktmbuf_detach()``.
1461 * Memory for shared data must be provided and user must initialize all of
1462 * the content properly, escpecially free callback and refcnt. The pointer
1463 * of shared data will be stored in m->shinfo.
1464 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1465 * bytes at the end of buffer for the shared data, store free callback and
1466 * its argument and set the refcnt to 1. The following is an example:
1468 * struct rte_mbuf_ext_shared_info *shinfo =
1469 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1470 * free_cb, fcb_arg);
1471 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1472 * rte_pktmbuf_reset_headroom(m);
1473 * rte_pktmbuf_adj(m, data_len);
1475 * Attaching an external buffer is quite similar to mbuf indirection in
1476 * replacing buffer addresses and length of a mbuf, but a few differences:
1477 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1478 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1479 * In such cases, the buffer area of a direct mbuf must be read-only. But
1480 * external buffer has its own refcnt and it starts from 1. Unless
1481 * multiple mbufs are attached to a mbuf having an external buffer, the
1482 * external buffer is writable.
1483 * - There's no need to allocate buffer from a mempool. Any buffer can be
1484 * attached with appropriate free callback and its IO address.
1485 * - Smaller metadata is required to maintain shared data such as refcnt.
1488 * @b EXPERIMENTAL: This API may change without prior notice.
1489 * Once external buffer is enabled by allowing experimental API,
1490 * ``RTE_MBUF_DIRECT()`` and ``RTE_MBUF_INDIRECT()`` are no longer
1491 * exclusive. A mbuf can be considered direct if it is neither indirect nor
1492 * having external buffer.
1495 * The pointer to the mbuf.
1497 * The pointer to the external buffer.
1499 * IO address of the external buffer.
1501 * The size of the external buffer.
1503 * User-provided memory for shared data of the external buffer.
1505 static inline void __rte_experimental
1506 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1507 rte_iova_t buf_iova, uint16_t buf_len,
1508 struct rte_mbuf_ext_shared_info *shinfo)
1510 /* mbuf should not be read-only */
1511 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1512 RTE_ASSERT(shinfo->free_cb != NULL);
1514 m->buf_addr = buf_addr;
1515 m->buf_iova = buf_iova;
1516 m->buf_len = buf_len;
1521 m->ol_flags |= EXT_ATTACHED_MBUF;
1526 * Detach the external buffer attached to a mbuf, same as
1527 * ``rte_pktmbuf_detach()``
1530 * The mbuf having external buffer.
1532 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1535 * Attach packet mbuf to another packet mbuf.
1537 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1538 * an external buffer, the mbuf being attached will be attached to the
1539 * external buffer instead of mbuf indirection.
1541 * Otherwise, the mbuf will be indirectly attached. After attachment we
1542 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1543 * 'direct'. The direct mbuf's reference counter is incremented.
1545 * Right now, not supported:
1546 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1547 * - mbuf we trying to attach (mi) is used by someone else
1548 * e.g. it's reference counter is greater then 1.
1551 * The indirect packet mbuf.
1553 * The packet mbuf we're attaching to.
1555 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1557 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1558 rte_mbuf_refcnt_read(mi) == 1);
1560 if (RTE_MBUF_HAS_EXTBUF(m)) {
1561 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1562 mi->ol_flags = m->ol_flags;
1563 mi->shinfo = m->shinfo;
1565 /* if m is not direct, get the mbuf that embeds the data */
1566 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1567 mi->priv_size = m->priv_size;
1568 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1571 mi->buf_iova = m->buf_iova;
1572 mi->buf_addr = m->buf_addr;
1573 mi->buf_len = m->buf_len;
1575 mi->data_off = m->data_off;
1576 mi->data_len = m->data_len;
1578 mi->vlan_tci = m->vlan_tci;
1579 mi->vlan_tci_outer = m->vlan_tci_outer;
1580 mi->tx_offload = m->tx_offload;
1584 mi->pkt_len = mi->data_len;
1586 mi->packet_type = m->packet_type;
1587 mi->timestamp = m->timestamp;
1589 __rte_mbuf_sanity_check(mi, 1);
1590 __rte_mbuf_sanity_check(m, 0);
1594 * @internal used by rte_pktmbuf_detach().
1596 * Decrement the reference counter of the external buffer. When the
1597 * reference counter becomes 0, the buffer is freed by pre-registered
1601 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1603 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1604 RTE_ASSERT(m->shinfo != NULL);
1606 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1607 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1611 * @internal used by rte_pktmbuf_detach().
1613 * Decrement the direct mbuf's reference counter. When the reference
1614 * counter becomes 0, the direct mbuf is freed.
1617 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1619 struct rte_mbuf *md;
1621 RTE_ASSERT(RTE_MBUF_INDIRECT(m));
1623 md = rte_mbuf_from_indirect(m);
1625 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1628 rte_mbuf_refcnt_set(md, 1);
1629 rte_mbuf_raw_free(md);
1634 * Detach a packet mbuf from external buffer or direct buffer.
1636 * - decrement refcnt and free the external/direct buffer if refcnt
1638 * - restore original mbuf address and length values.
1639 * - reset pktmbuf data and data_len to their default values.
1641 * All other fields of the given packet mbuf will be left intact.
1644 * The indirect attached packet mbuf.
1646 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1648 struct rte_mempool *mp = m->pool;
1649 uint32_t mbuf_size, buf_len;
1652 if (RTE_MBUF_HAS_EXTBUF(m))
1653 __rte_pktmbuf_free_extbuf(m);
1655 __rte_pktmbuf_free_direct(m);
1657 priv_size = rte_pktmbuf_priv_size(mp);
1658 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1659 buf_len = rte_pktmbuf_data_room_size(mp);
1661 m->priv_size = priv_size;
1662 m->buf_addr = (char *)m + mbuf_size;
1663 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1664 m->buf_len = (uint16_t)buf_len;
1665 rte_pktmbuf_reset_headroom(m);
1671 * Decrease reference counter and unlink a mbuf segment
1673 * This function does the same than a free, except that it does not
1674 * return the segment to its pool.
1675 * It decreases the reference counter, and if it reaches 0, it is
1676 * detached from its parent for an indirect mbuf.
1679 * The mbuf to be unlinked
1681 * - (m) if it is the last reference. It can be recycled or freed.
1682 * - (NULL) if the mbuf still has remaining references on it.
1684 static __rte_always_inline struct rte_mbuf *
1685 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1687 __rte_mbuf_sanity_check(m, 0);
1689 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1691 if (!RTE_MBUF_DIRECT(m))
1692 rte_pktmbuf_detach(m);
1694 if (m->next != NULL) {
1701 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1703 if (!RTE_MBUF_DIRECT(m))
1704 rte_pktmbuf_detach(m);
1706 if (m->next != NULL) {
1710 rte_mbuf_refcnt_set(m, 1);
1718 * Free a segment of a packet mbuf into its original mempool.
1720 * Free an mbuf, without parsing other segments in case of chained
1724 * The packet mbuf segment to be freed.
1726 static __rte_always_inline void
1727 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1729 m = rte_pktmbuf_prefree_seg(m);
1730 if (likely(m != NULL))
1731 rte_mbuf_raw_free(m);
1735 * Free a packet mbuf back into its original mempool.
1737 * Free an mbuf, and all its segments in case of chained buffers. Each
1738 * segment is added back into its original mempool.
1741 * The packet mbuf to be freed. If NULL, the function does nothing.
1743 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1745 struct rte_mbuf *m_next;
1748 __rte_mbuf_sanity_check(m, 1);
1752 rte_pktmbuf_free_seg(m);
1758 * Creates a "clone" of the given packet mbuf.
1760 * Walks through all segments of the given packet mbuf, and for each of them:
1761 * - Creates a new packet mbuf from the given pool.
1762 * - Attaches newly created mbuf to the segment.
1763 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1764 * from the original packet mbuf.
1767 * The packet mbuf to be cloned.
1769 * The mempool from which the "clone" mbufs are allocated.
1771 * - The pointer to the new "clone" mbuf on success.
1772 * - NULL if allocation fails.
1774 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1775 struct rte_mempool *mp)
1777 struct rte_mbuf *mc, *mi, **prev;
1781 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1786 pktlen = md->pkt_len;
1791 rte_pktmbuf_attach(mi, md);
1794 } while ((md = md->next) != NULL &&
1795 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1799 mc->pkt_len = pktlen;
1801 /* Allocation of new indirect segment failed */
1802 if (unlikely (mi == NULL)) {
1803 rte_pktmbuf_free(mc);
1807 __rte_mbuf_sanity_check(mc, 1);
1812 * Adds given value to the refcnt of all packet mbuf segments.
1814 * Walks through all segments of given packet mbuf and for each of them
1815 * invokes rte_mbuf_refcnt_update().
1818 * The packet mbuf whose refcnt to be updated.
1820 * The value to add to the mbuf's segments refcnt.
1822 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1824 __rte_mbuf_sanity_check(m, 1);
1827 rte_mbuf_refcnt_update(m, v);
1828 } while ((m = m->next) != NULL);
1832 * Get the headroom in a packet mbuf.
1837 * The length of the headroom.
1839 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1841 __rte_mbuf_sanity_check(m, 0);
1846 * Get the tailroom of a packet mbuf.
1851 * The length of the tailroom.
1853 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1855 __rte_mbuf_sanity_check(m, 0);
1856 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1861 * Get the last segment of the packet.
1866 * The last segment of the given mbuf.
1868 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1870 __rte_mbuf_sanity_check(m, 1);
1871 while (m->next != NULL)
1877 * A macro that points to an offset into the data in the mbuf.
1879 * The returned pointer is cast to type t. Before using this
1880 * function, the user must ensure that the first segment is large
1881 * enough to accommodate its data.
1886 * The offset into the mbuf data.
1888 * The type to cast the result into.
1890 #define rte_pktmbuf_mtod_offset(m, t, o) \
1891 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1894 * A macro that points to the start of the data in the mbuf.
1896 * The returned pointer is cast to type t. Before using this
1897 * function, the user must ensure that the first segment is large
1898 * enough to accommodate its data.
1903 * The type to cast the result into.
1905 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1908 * A macro that returns the IO address that points to an offset of the
1909 * start of the data in the mbuf
1914 * The offset into the data to calculate address from.
1916 #define rte_pktmbuf_iova_offset(m, o) \
1917 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
1920 #define rte_pktmbuf_mtophys_offset(m, o) \
1921 rte_pktmbuf_iova_offset(m, o)
1924 * A macro that returns the IO address that points to the start of the
1930 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
1933 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1936 * A macro that returns the length of the packet.
1938 * The value can be read or assigned.
1943 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1946 * A macro that returns the length of the segment.
1948 * The value can be read or assigned.
1953 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1956 * Prepend len bytes to an mbuf data area.
1958 * Returns a pointer to the new
1959 * data start address. If there is not enough headroom in the first
1960 * segment, the function will return NULL, without modifying the mbuf.
1965 * The amount of data to prepend (in bytes).
1967 * A pointer to the start of the newly prepended data, or
1968 * NULL if there is not enough headroom space in the first segment
1970 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1973 __rte_mbuf_sanity_check(m, 1);
1975 if (unlikely(len > rte_pktmbuf_headroom(m)))
1978 /* NB: elaborating the subtraction like this instead of using
1979 * -= allows us to ensure the result type is uint16_t
1980 * avoiding compiler warnings on gcc 8.1 at least */
1981 m->data_off = (uint16_t)(m->data_off - len);
1982 m->data_len = (uint16_t)(m->data_len + len);
1983 m->pkt_len = (m->pkt_len + len);
1985 return (char *)m->buf_addr + m->data_off;
1989 * Append len bytes to an mbuf.
1991 * Append len bytes to an mbuf and return a pointer to the start address
1992 * of the added data. If there is not enough tailroom in the last
1993 * segment, the function will return NULL, without modifying the mbuf.
1998 * The amount of data to append (in bytes).
2000 * A pointer to the start of the newly appended data, or
2001 * NULL if there is not enough tailroom space in the last segment
2003 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
2006 struct rte_mbuf *m_last;
2008 __rte_mbuf_sanity_check(m, 1);
2010 m_last = rte_pktmbuf_lastseg(m);
2011 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
2014 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
2015 m_last->data_len = (uint16_t)(m_last->data_len + len);
2016 m->pkt_len = (m->pkt_len + len);
2017 return (char*) tail;
2021 * Remove len bytes at the beginning of an mbuf.
2023 * Returns a pointer to the start address of the new data area. If the
2024 * length is greater than the length of the first segment, then the
2025 * function will fail and return NULL, without modifying the mbuf.
2030 * The amount of data to remove (in bytes).
2032 * A pointer to the new start of the data.
2034 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
2036 __rte_mbuf_sanity_check(m, 1);
2038 if (unlikely(len > m->data_len))
2041 /* NB: elaborating the addition like this instead of using
2042 * += allows us to ensure the result type is uint16_t
2043 * avoiding compiler warnings on gcc 8.1 at least */
2044 m->data_len = (uint16_t)(m->data_len - len);
2045 m->data_off = (uint16_t)(m->data_off + len);
2046 m->pkt_len = (m->pkt_len - len);
2047 return (char *)m->buf_addr + m->data_off;
2051 * Remove len bytes of data at the end of the mbuf.
2053 * If the length is greater than the length of the last segment, the
2054 * function will fail and return -1 without modifying the mbuf.
2059 * The amount of data to remove (in bytes).
2064 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
2066 struct rte_mbuf *m_last;
2068 __rte_mbuf_sanity_check(m, 1);
2070 m_last = rte_pktmbuf_lastseg(m);
2071 if (unlikely(len > m_last->data_len))
2074 m_last->data_len = (uint16_t)(m_last->data_len - len);
2075 m->pkt_len = (m->pkt_len - len);
2080 * Test if mbuf data is contiguous.
2085 * - 1, if all data is contiguous (one segment).
2086 * - 0, if there is several segments.
2088 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
2090 __rte_mbuf_sanity_check(m, 1);
2091 return !!(m->nb_segs == 1);
2095 * @internal used by rte_pktmbuf_read().
2097 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
2098 uint32_t len, void *buf);
2101 * Read len data bytes in a mbuf at specified offset.
2103 * If the data is contiguous, return the pointer in the mbuf data, else
2104 * copy the data in the buffer provided by the user and return its
2108 * The pointer to the mbuf.
2110 * The offset of the data in the mbuf.
2112 * The amount of bytes to read.
2114 * The buffer where data is copied if it is not contiguous in mbuf
2115 * data. Its length should be at least equal to the len parameter.
2117 * The pointer to the data, either in the mbuf if it is contiguous,
2118 * or in the user buffer. If mbuf is too small, NULL is returned.
2120 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
2121 uint32_t off, uint32_t len, void *buf)
2123 if (likely(off + len <= rte_pktmbuf_data_len(m)))
2124 return rte_pktmbuf_mtod_offset(m, char *, off);
2126 return __rte_pktmbuf_read(m, off, len, buf);
2130 * Chain an mbuf to another, thereby creating a segmented packet.
2132 * Note: The implementation will do a linear walk over the segments to find
2133 * the tail entry. For cases when there are many segments, it's better to
2134 * chain the entries manually.
2137 * The head of the mbuf chain (the first packet)
2139 * The mbuf to put last in the chain
2143 * - -EOVERFLOW, if the chain segment limit exceeded
2145 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
2147 struct rte_mbuf *cur_tail;
2149 /* Check for number-of-segments-overflow */
2150 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
2153 /* Chain 'tail' onto the old tail */
2154 cur_tail = rte_pktmbuf_lastseg(head);
2155 cur_tail->next = tail;
2157 /* accumulate number of segments and total length.
2158 * NB: elaborating the addition like this instead of using
2159 * -= allows us to ensure the result type is uint16_t
2160 * avoiding compiler warnings on gcc 8.1 at least */
2161 head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
2162 head->pkt_len += tail->pkt_len;
2164 /* pkt_len is only set in the head */
2165 tail->pkt_len = tail->data_len;
2171 * Validate general requirements for Tx offload in mbuf.
2173 * This function checks correctness and completeness of Tx offload settings.
2176 * The packet mbuf to be validated.
2178 * 0 if packet is valid
2181 rte_validate_tx_offload(const struct rte_mbuf *m)
2183 uint64_t ol_flags = m->ol_flags;
2184 uint64_t inner_l3_offset = m->l2_len;
2186 /* Does packet set any of available offloads? */
2187 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
2190 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
2191 /* NB: elaborating the addition like this instead of using
2192 * += gives the result uint64_t type instead of int,
2193 * avoiding compiler warnings on gcc 8.1 at least */
2194 inner_l3_offset = inner_l3_offset + m->outer_l2_len +
2197 /* Headers are fragmented */
2198 if (rte_pktmbuf_data_len(m) < inner_l3_offset + m->l3_len + m->l4_len)
2201 /* IP checksum can be counted only for IPv4 packet */
2202 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
2205 /* IP type not set when required */
2206 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
2207 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
2210 /* Check requirements for TSO packet */
2211 if (ol_flags & PKT_TX_TCP_SEG)
2212 if ((m->tso_segsz == 0) ||
2213 ((ol_flags & PKT_TX_IPV4) &&
2214 !(ol_flags & PKT_TX_IP_CKSUM)))
2217 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
2218 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
2219 !(ol_flags & PKT_TX_OUTER_IPV4))
2226 * Linearize data in mbuf.
2228 * This function moves the mbuf data in the first segment if there is enough
2229 * tailroom. The subsequent segments are unchained and freed.
2238 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
2240 size_t seg_len, copy_len;
2242 struct rte_mbuf *m_next;
2245 if (rte_pktmbuf_is_contiguous(mbuf))
2248 /* Extend first segment to the total packet length */
2249 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
2251 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
2254 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
2255 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
2257 /* Append data from next segments to the first one */
2262 seg_len = rte_pktmbuf_data_len(m);
2263 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
2266 rte_pktmbuf_free_seg(m);
2277 * Dump an mbuf structure to a file.
2279 * Dump all fields for the given packet mbuf and all its associated
2280 * segments (in the case of a chained buffer).
2283 * A pointer to a file for output
2287 * If dump_len != 0, also dump the "dump_len" first data bytes of
2290 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
2296 #endif /* _RTE_MBUF_H_ */