1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation.
3 * Copyright 2014 6WIND S.A.
13 * The mbuf library provides the ability to create and destroy buffers
14 * that may be used by the RTE application to store message
15 * buffers. The message buffers are stored in a mempool, using the
16 * RTE mempool library.
18 * The preferred way to create a mbuf pool is to use
19 * rte_pktmbuf_pool_create(). However, in some situations, an
20 * application may want to have more control (ex: populate the pool with
21 * specific memory), in this case it is possible to use functions from
22 * rte_mempool. See how rte_pktmbuf_pool_create() is implemented for
25 * This library provides an API to allocate/free packet mbufs, which are
26 * used to carry network packets.
28 * To understand the concepts of packet buffers or mbufs, you
29 * should read "TCP/IP Illustrated, Volume 2: The Implementation,
30 * Addison-Wesley, 1995, ISBN 0-201-63354-X from Richard Stevens"
31 * http://www.kohala.com/start/tcpipiv2.html
35 #include <rte_compat.h>
36 #include <rte_common.h>
37 #include <rte_config.h>
38 #include <rte_mempool.h>
39 #include <rte_memory.h>
40 #include <rte_atomic.h>
41 #include <rte_prefetch.h>
42 #include <rte_branch_prediction.h>
43 #include <rte_mbuf_ptype.h>
50 * Packet Offload Features Flags. It also carry packet type information.
51 * Critical resources. Both rx/tx shared these bits. Be cautious on any change
53 * - RX flags start at bit position zero, and get added to the left of previous
55 * - The most-significant 3 bits are reserved for generic mbuf flags
56 * - TX flags therefore start at bit position 60 (i.e. 63-3), and new flags get
57 * added to the right of the previously defined flags i.e. they should count
58 * downwards, not upwards.
60 * Keep these flags synchronized with rte_get_rx_ol_flag_name() and
61 * rte_get_tx_ol_flag_name().
65 * The RX packet is a 802.1q VLAN packet, and the tci has been
66 * saved in in mbuf->vlan_tci.
67 * If the flag PKT_RX_VLAN_STRIPPED is also present, the VLAN
68 * header has been stripped from mbuf data, else it is still
71 #define PKT_RX_VLAN (1ULL << 0)
73 #define PKT_RX_RSS_HASH (1ULL << 1) /**< RX packet with RSS hash result. */
74 #define PKT_RX_FDIR (1ULL << 2) /**< RX packet with FDIR match indicate. */
78 * Checking this flag alone is deprecated: check the 2 bits of
79 * PKT_RX_L4_CKSUM_MASK.
80 * This flag was set when the L4 checksum of a packet was detected as
81 * wrong by the hardware.
83 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3)
87 * Checking this flag alone is deprecated: check the 2 bits of
88 * PKT_RX_IP_CKSUM_MASK.
89 * This flag was set when the IP checksum of a packet was detected as
90 * wrong by the hardware.
92 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4)
94 #define PKT_RX_EIP_CKSUM_BAD (1ULL << 5) /**< External IP header checksum error. */
97 * A vlan has been stripped by the hardware and its tci is saved in
98 * mbuf->vlan_tci. This can only happen if vlan stripping is enabled
99 * in the RX configuration of the PMD.
100 * When PKT_RX_VLAN_STRIPPED is set, PKT_RX_VLAN must also be set.
102 #define PKT_RX_VLAN_STRIPPED (1ULL << 6)
105 * Mask of bits used to determine the status of RX IP checksum.
106 * - PKT_RX_IP_CKSUM_UNKNOWN: no information about the RX IP checksum
107 * - PKT_RX_IP_CKSUM_BAD: the IP checksum in the packet is wrong
108 * - PKT_RX_IP_CKSUM_GOOD: the IP checksum in the packet is valid
109 * - PKT_RX_IP_CKSUM_NONE: the IP checksum is not correct in the packet
110 * data, but the integrity of the IP header is verified.
112 #define PKT_RX_IP_CKSUM_MASK ((1ULL << 4) | (1ULL << 7))
114 #define PKT_RX_IP_CKSUM_UNKNOWN 0
115 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4)
116 #define PKT_RX_IP_CKSUM_GOOD (1ULL << 7)
117 #define PKT_RX_IP_CKSUM_NONE ((1ULL << 4) | (1ULL << 7))
120 * Mask of bits used to determine the status of RX L4 checksum.
121 * - PKT_RX_L4_CKSUM_UNKNOWN: no information about the RX L4 checksum
122 * - PKT_RX_L4_CKSUM_BAD: the L4 checksum in the packet is wrong
123 * - PKT_RX_L4_CKSUM_GOOD: the L4 checksum in the packet is valid
124 * - PKT_RX_L4_CKSUM_NONE: the L4 checksum is not correct in the packet
125 * data, but the integrity of the L4 data is verified.
127 #define PKT_RX_L4_CKSUM_MASK ((1ULL << 3) | (1ULL << 8))
129 #define PKT_RX_L4_CKSUM_UNKNOWN 0
130 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3)
131 #define PKT_RX_L4_CKSUM_GOOD (1ULL << 8)
132 #define PKT_RX_L4_CKSUM_NONE ((1ULL << 3) | (1ULL << 8))
134 #define PKT_RX_IEEE1588_PTP (1ULL << 9) /**< RX IEEE1588 L2 Ethernet PT Packet. */
135 #define PKT_RX_IEEE1588_TMST (1ULL << 10) /**< RX IEEE1588 L2/L4 timestamped packet.*/
136 #define PKT_RX_FDIR_ID (1ULL << 13) /**< FD id reported if FDIR match. */
137 #define PKT_RX_FDIR_FLX (1ULL << 14) /**< Flexible bytes reported if FDIR match. */
140 * The 2 vlans have been stripped by the hardware and their tci are
141 * saved in mbuf->vlan_tci (inner) and mbuf->vlan_tci_outer (outer).
142 * This can only happen if vlan stripping is enabled in the RX
143 * configuration of the PMD. If this flag is set,
144 * When PKT_RX_QINQ_STRIPPED is set, the flags (PKT_RX_VLAN |
145 * PKT_RX_VLAN_STRIPPED | PKT_RX_QINQ) must also be set.
147 #define PKT_RX_QINQ_STRIPPED (1ULL << 15)
150 * When packets are coalesced by a hardware or virtual driver, this flag
151 * can be set in the RX mbuf, meaning that the m->tso_segsz field is
152 * valid and is set to the segment size of original packets.
154 #define PKT_RX_LRO (1ULL << 16)
157 * Indicate that the timestamp field in the mbuf is valid.
159 #define PKT_RX_TIMESTAMP (1ULL << 17)
162 * Indicate that security offload processing was applied on the RX packet.
164 #define PKT_RX_SEC_OFFLOAD (1ULL << 18)
167 * Indicate that security offload processing failed on the RX packet.
169 #define PKT_RX_SEC_OFFLOAD_FAILED (1ULL << 19)
172 * The RX packet is a double VLAN, and the outer tci has been
173 * saved in in mbuf->vlan_tci_outer.
174 * If the flag PKT_RX_QINQ_STRIPPED is also present, both VLANs
175 * headers have been stripped from mbuf data, else they are still
178 #define PKT_RX_QINQ (1ULL << 20)
180 /* add new RX flags here */
182 /* add new TX flags here */
185 * UDP Fragmentation Offload flag. This flag is used for enabling UDP
186 * fragmentation in SW or in HW. When use UFO, mbuf->tso_segsz is used
187 * to store the MSS of UDP fragments.
189 #define PKT_TX_UDP_SEG (1ULL << 42)
192 * Request security offload processing on the TX packet.
194 #define PKT_TX_SEC_OFFLOAD (1ULL << 43)
197 * Offload the MACsec. This flag must be set by the application to enable
198 * this offload feature for a packet to be transmitted.
200 #define PKT_TX_MACSEC (1ULL << 44)
203 * Bits 45:48 used for the tunnel type.
204 * The tunnel type must be specified for TSO or checksum on the inner part
206 * These flags can be used with PKT_TX_TCP_SEG for TSO, or PKT_TX_xxx_CKSUM.
207 * The mbuf fields for inner and outer header lengths are required:
208 * outer_l2_len, outer_l3_len, l2_len, l3_len, l4_len and tso_segsz for TSO.
210 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
211 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
212 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
213 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
214 /** TX packet with MPLS-in-UDP RFC 7510 header. */
215 #define PKT_TX_TUNNEL_MPLSINUDP (0x5ULL << 45)
216 #define PKT_TX_TUNNEL_VXLAN_GPE (0x6ULL << 45)
218 * Generic IP encapsulated tunnel type, used for TSO and checksum offload.
219 * It can be used for tunnels which are not standards or listed above.
220 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_GRE
221 * or PKT_TX_TUNNEL_IPIP if possible.
222 * The ethdev must be configured with DEV_TX_OFFLOAD_IP_TNL_TSO.
223 * Outer and inner checksums are done according to the existing flags like
225 * Specific tunnel headers that contain payload length, sequence id
226 * or checksum are not expected to be updated.
228 #define PKT_TX_TUNNEL_IP (0xDULL << 45)
230 * Generic UDP encapsulated tunnel type, used for TSO and checksum offload.
231 * UDP tunnel type implies outer IP layer.
232 * It can be used for tunnels which are not standards or listed above.
233 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_VXLAN
235 * The ethdev must be configured with DEV_TX_OFFLOAD_UDP_TNL_TSO.
236 * Outer and inner checksums are done according to the existing flags like
238 * Specific tunnel headers that contain payload length, sequence id
239 * or checksum are not expected to be updated.
241 #define PKT_TX_TUNNEL_UDP (0xEULL << 45)
242 /* add new TX TUNNEL type here */
243 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
246 * Second VLAN insertion (QinQ) flag.
248 #define PKT_TX_QINQ (1ULL << 49) /**< TX packet with double VLAN inserted. */
249 /* this old name is deprecated */
250 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
253 * TCP segmentation offload. To enable this offload feature for a
254 * packet to be transmitted on hardware supporting TSO:
255 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
257 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
258 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag
259 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
261 #define PKT_TX_TCP_SEG (1ULL << 50)
263 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
266 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
267 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
268 * L4 checksum offload, the user needs to:
269 * - fill l2_len and l3_len in mbuf
270 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
271 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
273 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
274 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
275 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
276 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
277 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
280 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
281 * also be set by the application, although a PMD will only check
283 * - fill the mbuf offload information: l2_len, l3_len
285 #define PKT_TX_IP_CKSUM (1ULL << 54)
288 * Packet is IPv4. This flag must be set when using any offload feature
289 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
290 * packet. If the packet is a tunneled packet, this flag is related to
293 #define PKT_TX_IPV4 (1ULL << 55)
296 * Packet is IPv6. This flag must be set when using an offload feature
297 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
298 * packet. If the packet is a tunneled packet, this flag is related to
301 #define PKT_TX_IPV6 (1ULL << 56)
304 * TX packet is a 802.1q VLAN packet.
306 #define PKT_TX_VLAN (1ULL << 57)
307 /* this old name is deprecated */
308 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
311 * Offload the IP checksum of an external header in the hardware. The
312 * flag PKT_TX_OUTER_IPV4 should also be set by the application, although
313 * a PMD will only check PKT_TX_OUTER_IP_CKSUM.
314 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
316 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
319 * Packet outer header is IPv4. This flag must be set when using any
320 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
321 * outer header of the tunneled packet is an IPv4 packet.
323 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
326 * Packet outer header is IPv6. This flag must be set when using any
327 * outer offload feature (L4 checksum) to tell the NIC that the outer
328 * header of the tunneled packet is an IPv6 packet.
330 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
333 * Bitmask of all supported packet Tx offload features flags,
334 * which can be set for packet.
336 #define PKT_TX_OFFLOAD_MASK ( \
337 PKT_TX_OUTER_IPV6 | \
338 PKT_TX_OUTER_IPV4 | \
339 PKT_TX_OUTER_IP_CKSUM | \
345 PKT_TX_IEEE1588_TMST | \
348 PKT_TX_TUNNEL_MASK | \
350 PKT_TX_SEC_OFFLOAD | \
354 * Mbuf having an external buffer attached. shinfo in mbuf must be filled.
356 #define EXT_ATTACHED_MBUF (1ULL << 61)
358 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
360 /** Alignment constraint of mbuf private area. */
361 #define RTE_MBUF_PRIV_ALIGN 8
364 * Get the name of a RX offload flag
367 * The mask describing the flag.
369 * The name of this flag, or NULL if it's not a valid RX flag.
371 const char *rte_get_rx_ol_flag_name(uint64_t mask);
374 * Dump the list of RX offload flags in a buffer
377 * The mask describing the RX flags.
381 * The length of the buffer.
383 * 0 on success, (-1) on error.
385 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
388 * Get the name of a TX offload flag
391 * The mask describing the flag. Usually only one bit must be set.
392 * Several bits can be given if they belong to the same mask.
393 * Ex: PKT_TX_L4_MASK.
395 * The name of this flag, or NULL if it's not a valid TX flag.
397 const char *rte_get_tx_ol_flag_name(uint64_t mask);
400 * Dump the list of TX offload flags in a buffer
403 * The mask describing the TX flags.
407 * The length of the buffer.
409 * 0 on success, (-1) on error.
411 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
414 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
415 * splitting it into multiple segments.
416 * So, for mbufs that planned to be involved into RX/TX, the recommended
417 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
419 #define RTE_MBUF_DEFAULT_DATAROOM 2048
420 #define RTE_MBUF_DEFAULT_BUF_SIZE \
421 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
423 /* define a set of marker types that can be used to refer to set points in the
426 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
428 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
430 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
431 * with a single assignment */
434 * The generic rte_mbuf, containing a packet mbuf.
439 void *buf_addr; /**< Virtual address of segment buffer. */
441 * Physical address of segment buffer.
442 * Force alignment to 8-bytes, so as to ensure we have the exact
443 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
444 * working on vector drivers easier.
449 rte_iova_t buf_physaddr; /**< deprecated */
450 } __rte_aligned(sizeof(rte_iova_t));
452 /* next 8 bytes are initialised on RX descriptor rearm */
457 * Reference counter. Its size should at least equal to the size
458 * of port field (16 bits), to support zero-copy broadcast.
459 * It should only be accessed using the following functions:
460 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
461 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
462 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
467 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
468 uint16_t refcnt; /**< Non-atomically accessed refcnt */
470 uint16_t nb_segs; /**< Number of segments. */
472 /** Input port (16 bits to support more than 256 virtual ports).
473 * The event eth Tx adapter uses this field to specify the output port.
477 uint64_t ol_flags; /**< Offload features. */
479 /* remaining bytes are set on RX when pulling packet from descriptor */
480 MARKER rx_descriptor_fields1;
483 * The packet type, which is the combination of outer/inner L2, L3, L4
484 * and tunnel types. The packet_type is about data really present in the
485 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
486 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
487 * vlan is stripped from the data.
491 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
493 uint32_t l2_type:4; /**< (Outer) L2 type. */
494 uint32_t l3_type:4; /**< (Outer) L3 type. */
495 uint32_t l4_type:4; /**< (Outer) L4 type. */
496 uint32_t tun_type:4; /**< Tunnel type. */
499 uint8_t inner_esp_next_proto;
500 /**< ESP next protocol type, valid if
501 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
506 uint8_t inner_l2_type:4;
507 /**< Inner L2 type. */
508 uint8_t inner_l3_type:4;
509 /**< Inner L3 type. */
512 uint32_t inner_l4_type:4; /**< Inner L4 type. */
516 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
517 uint16_t data_len; /**< Amount of data in segment buffer. */
518 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
522 uint32_t rss; /**< RSS hash result if RSS enabled */
531 /**< Second 4 flexible bytes */
534 /**< First 4 flexible bytes or FD ID, dependent on
535 PKT_RX_FDIR_* flag in ol_flags. */
536 } fdir; /**< Filter identifier if FDIR enabled */
540 /**< The event eth Tx adapter uses this field to store
541 * Tx queue id. @see rte_event_eth_tx_adapter_txq_set()
543 } sched; /**< Hierarchical scheduler */
544 uint32_t usr; /**< User defined tags. See rte_distributor_process() */
545 } hash; /**< hash information */
547 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
548 uint16_t vlan_tci_outer;
550 uint16_t buf_len; /**< Length of segment buffer. */
552 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
553 * are not normalized but are always the same for a given port.
557 /* second cache line - fields only used in slow path or on TX */
558 MARKER cacheline1 __rte_cache_min_aligned;
562 void *userdata; /**< Can be used for external metadata */
563 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
566 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
567 struct rte_mbuf *next; /**< Next segment of scattered packet. */
569 /* fields to support TX offloads */
572 uint64_t tx_offload; /**< combined for easy fetch */
576 /**< L2 (MAC) Header Length for non-tunneling pkt.
577 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
579 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
580 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
581 uint64_t tso_segsz:16; /**< TCP TSO segment size */
583 /* fields for TX offloading of tunnels */
584 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
585 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
587 /* uint64_t unused:8; */
591 /** Size of the application private data. In case of an indirect
592 * mbuf, it stores the direct mbuf private data size. */
595 /** Timesync flags for use with IEEE1588. */
598 /** Sequence number. See also rte_reorder_insert(). */
601 /** Shared data for external buffer attached to mbuf. See
602 * rte_pktmbuf_attach_extbuf().
604 struct rte_mbuf_ext_shared_info *shinfo;
606 } __rte_cache_aligned;
609 * Function typedef of callback to free externally attached buffer.
611 typedef void (*rte_mbuf_extbuf_free_callback_t)(void *addr, void *opaque);
614 * Shared data at the end of an external buffer.
616 struct rte_mbuf_ext_shared_info {
617 rte_mbuf_extbuf_free_callback_t free_cb; /**< Free callback function */
618 void *fcb_opaque; /**< Free callback argument */
619 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
622 /**< Maximum number of nb_segs allowed. */
623 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
626 * Prefetch the first part of the mbuf
628 * The first 64 bytes of the mbuf corresponds to fields that are used early
629 * in the receive path. If the cache line of the architecture is higher than
630 * 64B, the second part will also be prefetched.
633 * The pointer to the mbuf.
636 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
638 rte_prefetch0(&m->cacheline0);
642 * Prefetch the second part of the mbuf
644 * The next 64 bytes of the mbuf corresponds to fields that are used in the
645 * transmit path. If the cache line of the architecture is higher than 64B,
646 * this function does nothing as it is expected that the full mbuf is
650 * The pointer to the mbuf.
653 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
655 #if RTE_CACHE_LINE_SIZE == 64
656 rte_prefetch0(&m->cacheline1);
663 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
666 * Return the IO address of the beginning of the mbuf data
669 * The pointer to the mbuf.
671 * The IO address of the beginning of the mbuf data
673 static inline rte_iova_t
674 rte_mbuf_data_iova(const struct rte_mbuf *mb)
676 return mb->buf_iova + mb->data_off;
680 static inline phys_addr_t
681 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
683 return rte_mbuf_data_iova(mb);
687 * Return the default IO address of the beginning of the mbuf data
689 * This function is used by drivers in their receive function, as it
690 * returns the location where data should be written by the NIC, taking
691 * the default headroom in account.
694 * The pointer to the mbuf.
696 * The IO address of the beginning of the mbuf data
698 static inline rte_iova_t
699 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
701 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
705 static inline phys_addr_t
706 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
708 return rte_mbuf_data_iova_default(mb);
712 * Return the mbuf owning the data buffer address of an indirect mbuf.
715 * The pointer to the indirect mbuf.
717 * The address of the direct mbuf corresponding to buffer_addr.
719 static inline struct rte_mbuf *
720 rte_mbuf_from_indirect(struct rte_mbuf *mi)
722 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
726 * Return the buffer address embedded in the given mbuf.
729 * The pointer to the mbuf.
731 * The address of the data buffer owned by the mbuf.
734 rte_mbuf_to_baddr(struct rte_mbuf *md)
737 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
742 * Return the starting address of the private data area embedded in
745 * Note that no check is made to ensure that a private data area
746 * actually exists in the supplied mbuf.
749 * The pointer to the mbuf.
751 * The starting address of the private data area of the given mbuf.
753 static inline void * __rte_experimental
754 rte_mbuf_to_priv(struct rte_mbuf *m)
756 return RTE_PTR_ADD(m, sizeof(struct rte_mbuf));
760 * Returns TRUE if given mbuf is cloned by mbuf indirection, or FALSE
763 * If a mbuf has its data in another mbuf and references it by mbuf
764 * indirection, this mbuf can be defined as a cloned mbuf.
766 #define RTE_MBUF_CLONED(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
770 * Use RTE_MBUF_CLONED().
772 #define RTE_MBUF_INDIRECT(mb) RTE_MBUF_CLONED(mb)
775 * Returns TRUE if given mbuf has an external buffer, or FALSE otherwise.
777 * External buffer is a user-provided anonymous buffer.
779 #define RTE_MBUF_HAS_EXTBUF(mb) ((mb)->ol_flags & EXT_ATTACHED_MBUF)
782 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
784 * If a mbuf embeds its own data after the rte_mbuf structure, this mbuf
785 * can be defined as a direct mbuf.
787 #define RTE_MBUF_DIRECT(mb) \
788 (!((mb)->ol_flags & (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF)))
791 * Private data in case of pktmbuf pool.
793 * A structure that contains some pktmbuf_pool-specific data that are
794 * appended after the mempool structure (in private data).
796 struct rte_pktmbuf_pool_private {
797 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
798 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
801 #ifdef RTE_LIBRTE_MBUF_DEBUG
803 /** check mbuf type in debug mode */
804 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
806 #else /* RTE_LIBRTE_MBUF_DEBUG */
808 /** check mbuf type in debug mode */
809 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
811 #endif /* RTE_LIBRTE_MBUF_DEBUG */
813 #ifdef RTE_MBUF_REFCNT_ATOMIC
816 * Reads the value of an mbuf's refcnt.
820 * Reference count number.
822 static inline uint16_t
823 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
825 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
829 * Sets an mbuf's refcnt to a defined value.
836 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
838 rte_atomic16_set(&m->refcnt_atomic, (int16_t)new_value);
842 static inline uint16_t
843 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
845 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
849 * Adds given value to an mbuf's refcnt and returns its new value.
853 * Value to add/subtract
857 static inline uint16_t
858 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
861 * The atomic_add is an expensive operation, so we don't want to
862 * call it in the case where we know we are the uniq holder of
863 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
864 * operation has to be used because concurrent accesses on the
865 * reference counter can occur.
867 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
869 rte_mbuf_refcnt_set(m, (uint16_t)value);
870 return (uint16_t)value;
873 return __rte_mbuf_refcnt_update(m, value);
876 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
879 static inline uint16_t
880 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
882 m->refcnt = (uint16_t)(m->refcnt + value);
887 * Adds given value to an mbuf's refcnt and returns its new value.
889 static inline uint16_t
890 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
892 return __rte_mbuf_refcnt_update(m, value);
896 * Reads the value of an mbuf's refcnt.
898 static inline uint16_t
899 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
905 * Sets an mbuf's refcnt to the defined value.
908 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
910 m->refcnt = new_value;
913 #endif /* RTE_MBUF_REFCNT_ATOMIC */
916 * Reads the refcnt of an external buffer.
919 * Shared data of the external buffer.
921 * Reference count number.
923 static inline uint16_t
924 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
926 return (uint16_t)(rte_atomic16_read(&shinfo->refcnt_atomic));
930 * Set refcnt of an external buffer.
933 * Shared data of the external buffer.
938 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
941 rte_atomic16_set(&shinfo->refcnt_atomic, (int16_t)new_value);
945 * Add given value to refcnt of an external buffer and return its new
949 * Shared data of the external buffer.
951 * Value to add/subtract
955 static inline uint16_t
956 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
959 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
961 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
962 return (uint16_t)value;
965 return (uint16_t)rte_atomic16_add_return(&shinfo->refcnt_atomic, value);
969 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
976 * Sanity checks on an mbuf.
978 * Check the consistency of the given mbuf. The function will cause a
979 * panic if corruption is detected.
982 * The mbuf to be checked.
984 * True if the mbuf is a packet header, false if it is a sub-segment
985 * of a packet (in this case, some fields like nb_segs are not checked)
988 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
990 #define MBUF_RAW_ALLOC_CHECK(m) do { \
991 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
992 RTE_ASSERT((m)->next == NULL); \
993 RTE_ASSERT((m)->nb_segs == 1); \
994 __rte_mbuf_sanity_check(m, 0); \
998 * Allocate an uninitialized mbuf from mempool *mp*.
1000 * This function can be used by PMDs (especially in RX functions) to
1001 * allocate an uninitialized mbuf. The driver is responsible of
1002 * initializing all the required fields. See rte_pktmbuf_reset().
1003 * For standard needs, prefer rte_pktmbuf_alloc().
1005 * The caller can expect that the following fields of the mbuf structure
1006 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
1007 * next=NULL, pool, priv_size. The other fields must be initialized
1011 * The mempool from which mbuf is allocated.
1013 * - The pointer to the new mbuf on success.
1014 * - NULL if allocation failed.
1016 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
1020 if (rte_mempool_get(mp, (void **)&m) < 0)
1022 MBUF_RAW_ALLOC_CHECK(m);
1027 * Put mbuf back into its original mempool.
1029 * The caller must ensure that the mbuf is direct and properly
1030 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
1031 * rte_pktmbuf_prefree_seg().
1033 * This function should be used with care, when optimization is
1034 * required. For standard needs, prefer rte_pktmbuf_free() or
1035 * rte_pktmbuf_free_seg().
1038 * The mbuf to be freed.
1040 static __rte_always_inline void
1041 rte_mbuf_raw_free(struct rte_mbuf *m)
1043 RTE_ASSERT(RTE_MBUF_DIRECT(m));
1044 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
1045 RTE_ASSERT(m->next == NULL);
1046 RTE_ASSERT(m->nb_segs == 1);
1047 __rte_mbuf_sanity_check(m, 0);
1048 rte_mempool_put(m->pool, m);
1052 * The packet mbuf constructor.
1054 * This function initializes some fields in the mbuf structure that are
1055 * not modified by the user once created (origin pool, buffer start
1056 * address, and so on). This function is given as a callback function to
1057 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
1060 * The mempool from which mbufs originate.
1062 * A pointer that can be used by the user to retrieve useful information
1063 * for mbuf initialization. This pointer is the opaque argument passed to
1064 * rte_mempool_obj_iter() or rte_mempool_create().
1066 * The mbuf to initialize.
1068 * The index of the mbuf in the pool table.
1070 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
1071 void *m, unsigned i);
1075 * A packet mbuf pool constructor.
1077 * This function initializes the mempool private data in the case of a
1078 * pktmbuf pool. This private data is needed by the driver. The
1079 * function must be called on the mempool before it is used, or it
1080 * can be given as a callback function to rte_mempool_create() at
1081 * pool creation. It can be extended by the user, for example, to
1082 * provide another packet size.
1085 * The mempool from which mbufs originate.
1087 * A pointer that can be used by the user to retrieve useful information
1088 * for mbuf initialization. This pointer is the opaque argument passed to
1089 * rte_mempool_create().
1091 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
1094 * Create a mbuf pool.
1096 * This function creates and initializes a packet mbuf pool. It is
1097 * a wrapper to rte_mempool functions.
1100 * The name of the mbuf pool.
1102 * The number of elements in the mbuf pool. The optimum size (in terms
1103 * of memory usage) for a mempool is when n is a power of two minus one:
1106 * Size of the per-core object cache. See rte_mempool_create() for
1109 * Size of application private are between the rte_mbuf structure
1110 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1111 * @param data_room_size
1112 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1114 * The socket identifier where the memory should be allocated. The
1115 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1118 * The pointer to the new allocated mempool, on success. NULL on error
1119 * with rte_errno set appropriately. Possible rte_errno values include:
1120 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1121 * - E_RTE_SECONDARY - function was called from a secondary process instance
1122 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1123 * - ENOSPC - the maximum number of memzones has already been allocated
1124 * - EEXIST - a memzone with the same name already exists
1125 * - ENOMEM - no appropriate memory area found in which to create memzone
1127 struct rte_mempool *
1128 rte_pktmbuf_pool_create(const char *name, unsigned n,
1129 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
1133 * Create a mbuf pool with a given mempool ops name
1135 * This function creates and initializes a packet mbuf pool. It is
1136 * a wrapper to rte_mempool functions.
1139 * The name of the mbuf pool.
1141 * The number of elements in the mbuf pool. The optimum size (in terms
1142 * of memory usage) for a mempool is when n is a power of two minus one:
1145 * Size of the per-core object cache. See rte_mempool_create() for
1148 * Size of application private are between the rte_mbuf structure
1149 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1150 * @param data_room_size
1151 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1153 * The socket identifier where the memory should be allocated. The
1154 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1157 * The mempool ops name to be used for this mempool instead of
1158 * default mempool. The value can be *NULL* to use default mempool.
1160 * The pointer to the new allocated mempool, on success. NULL on error
1161 * with rte_errno set appropriately. Possible rte_errno values include:
1162 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1163 * - E_RTE_SECONDARY - function was called from a secondary process instance
1164 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1165 * - ENOSPC - the maximum number of memzones has already been allocated
1166 * - EEXIST - a memzone with the same name already exists
1167 * - ENOMEM - no appropriate memory area found in which to create memzone
1169 struct rte_mempool *
1170 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1171 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1172 int socket_id, const char *ops_name);
1175 * Get the data room size of mbufs stored in a pktmbuf_pool
1177 * The data room size is the amount of data that can be stored in a
1178 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1181 * The packet mbuf pool.
1183 * The data room size of mbufs stored in this mempool.
1185 static inline uint16_t
1186 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1188 struct rte_pktmbuf_pool_private *mbp_priv;
1190 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1191 return mbp_priv->mbuf_data_room_size;
1195 * Get the application private size of mbufs stored in a pktmbuf_pool
1197 * The private size of mbuf is a zone located between the rte_mbuf
1198 * structure and the data buffer where an application can store data
1199 * associated to a packet.
1202 * The packet mbuf pool.
1204 * The private size of mbufs stored in this mempool.
1206 static inline uint16_t
1207 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1209 struct rte_pktmbuf_pool_private *mbp_priv;
1211 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1212 return mbp_priv->mbuf_priv_size;
1216 * Reset the data_off field of a packet mbuf to its default value.
1218 * The given mbuf must have only one segment, which should be empty.
1221 * The packet mbuf's data_off field has to be reset.
1223 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1225 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
1226 (uint16_t)m->buf_len);
1230 * Reset the fields of a packet mbuf to their default values.
1232 * The given mbuf must have only one segment.
1235 * The packet mbuf to be resetted.
1237 #define MBUF_INVALID_PORT UINT16_MAX
1239 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1245 m->vlan_tci_outer = 0;
1247 m->port = MBUF_INVALID_PORT;
1251 rte_pktmbuf_reset_headroom(m);
1254 __rte_mbuf_sanity_check(m, 1);
1258 * Allocate a new mbuf from a mempool.
1260 * This new mbuf contains one segment, which has a length of 0. The pointer
1261 * to data is initialized to have some bytes of headroom in the buffer
1262 * (if buffer size allows).
1265 * The mempool from which the mbuf is allocated.
1267 * - The pointer to the new mbuf on success.
1268 * - NULL if allocation failed.
1270 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1273 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1274 rte_pktmbuf_reset(m);
1279 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1283 * The mempool from which mbufs are allocated.
1285 * Array of pointers to mbufs
1290 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1292 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1293 struct rte_mbuf **mbufs, unsigned count)
1298 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1302 /* To understand duff's device on loop unwinding optimization, see
1303 * https://en.wikipedia.org/wiki/Duff's_device.
1304 * Here while() loop is used rather than do() while{} to avoid extra
1305 * check if count is zero.
1307 switch (count % 4) {
1309 while (idx != count) {
1310 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1311 rte_pktmbuf_reset(mbufs[idx]);
1315 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1316 rte_pktmbuf_reset(mbufs[idx]);
1320 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1321 rte_pktmbuf_reset(mbufs[idx]);
1325 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1326 rte_pktmbuf_reset(mbufs[idx]);
1335 * Initialize shared data at the end of an external buffer before attaching
1336 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
1337 * initialization but a helper function to simply spare a few bytes at the
1338 * end of the buffer for shared data. If shared data is allocated
1339 * separately, this should not be called but application has to properly
1340 * initialize the shared data according to its need.
1342 * Free callback and its argument is saved and the refcnt is set to 1.
1345 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
1346 * after this initialization. This shall be used for
1347 * ``rte_pktmbuf_attach_extbuf()``
1350 * The pointer to the external buffer.
1351 * @param [in,out] buf_len
1352 * The pointer to length of the external buffer. Input value must be
1353 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
1354 * padding for alignment. If not enough, this function will return NULL.
1355 * Adjusted buffer length will be returned through this pointer.
1357 * Free callback function to call when the external buffer needs to be
1360 * Argument for the free callback function.
1363 * A pointer to the initialized shared data on success, return NULL
1366 static inline struct rte_mbuf_ext_shared_info *
1367 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
1368 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
1370 struct rte_mbuf_ext_shared_info *shinfo;
1371 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
1374 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
1376 if (addr <= buf_addr)
1379 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
1380 shinfo->free_cb = free_cb;
1381 shinfo->fcb_opaque = fcb_opaque;
1382 rte_mbuf_ext_refcnt_set(shinfo, 1);
1384 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
1389 * Attach an external buffer to a mbuf.
1391 * User-managed anonymous buffer can be attached to an mbuf. When attaching
1392 * it, corresponding free callback function and its argument should be
1393 * provided via shinfo. This callback function will be called once all the
1394 * mbufs are detached from the buffer (refcnt becomes zero).
1396 * The headroom for the attaching mbuf will be set to zero and this can be
1397 * properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1398 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1400 * More mbufs can be attached to the same external buffer by
1401 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1404 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1405 * ``rte_pktmbuf_detach()``.
1407 * Memory for shared data must be provided and user must initialize all of
1408 * the content properly, escpecially free callback and refcnt. The pointer
1409 * of shared data will be stored in m->shinfo.
1410 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1411 * bytes at the end of buffer for the shared data, store free callback and
1412 * its argument and set the refcnt to 1. The following is an example:
1414 * struct rte_mbuf_ext_shared_info *shinfo =
1415 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1416 * free_cb, fcb_arg);
1417 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1418 * rte_pktmbuf_reset_headroom(m);
1419 * rte_pktmbuf_adj(m, data_len);
1421 * Attaching an external buffer is quite similar to mbuf indirection in
1422 * replacing buffer addresses and length of a mbuf, but a few differences:
1423 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1424 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1425 * In such cases, the buffer area of a direct mbuf must be read-only. But
1426 * external buffer has its own refcnt and it starts from 1. Unless
1427 * multiple mbufs are attached to a mbuf having an external buffer, the
1428 * external buffer is writable.
1429 * - There's no need to allocate buffer from a mempool. Any buffer can be
1430 * attached with appropriate free callback and its IO address.
1431 * - Smaller metadata is required to maintain shared data such as refcnt.
1434 * @b EXPERIMENTAL: This API may change without prior notice.
1435 * Once external buffer is enabled by allowing experimental API,
1436 * ``RTE_MBUF_DIRECT()`` and ``RTE_MBUF_INDIRECT()`` are no longer
1437 * exclusive. A mbuf can be considered direct if it is neither indirect nor
1438 * having external buffer.
1441 * The pointer to the mbuf.
1443 * The pointer to the external buffer.
1445 * IO address of the external buffer.
1447 * The size of the external buffer.
1449 * User-provided memory for shared data of the external buffer.
1451 static inline void __rte_experimental
1452 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1453 rte_iova_t buf_iova, uint16_t buf_len,
1454 struct rte_mbuf_ext_shared_info *shinfo)
1456 /* mbuf should not be read-only */
1457 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1458 RTE_ASSERT(shinfo->free_cb != NULL);
1460 m->buf_addr = buf_addr;
1461 m->buf_iova = buf_iova;
1462 m->buf_len = buf_len;
1467 m->ol_flags |= EXT_ATTACHED_MBUF;
1472 * Detach the external buffer attached to a mbuf, same as
1473 * ``rte_pktmbuf_detach()``
1476 * The mbuf having external buffer.
1478 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1481 * Attach packet mbuf to another packet mbuf.
1483 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1484 * an external buffer, the mbuf being attached will be attached to the
1485 * external buffer instead of mbuf indirection.
1487 * Otherwise, the mbuf will be indirectly attached. After attachment we
1488 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1489 * 'direct'. The direct mbuf's reference counter is incremented.
1491 * Right now, not supported:
1492 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1493 * - mbuf we trying to attach (mi) is used by someone else
1494 * e.g. it's reference counter is greater then 1.
1497 * The indirect packet mbuf.
1499 * The packet mbuf we're attaching to.
1501 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1503 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1504 rte_mbuf_refcnt_read(mi) == 1);
1506 if (RTE_MBUF_HAS_EXTBUF(m)) {
1507 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1508 mi->ol_flags = m->ol_flags;
1509 mi->shinfo = m->shinfo;
1511 /* if m is not direct, get the mbuf that embeds the data */
1512 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1513 mi->priv_size = m->priv_size;
1514 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1517 mi->buf_iova = m->buf_iova;
1518 mi->buf_addr = m->buf_addr;
1519 mi->buf_len = m->buf_len;
1521 mi->data_off = m->data_off;
1522 mi->data_len = m->data_len;
1524 mi->vlan_tci = m->vlan_tci;
1525 mi->vlan_tci_outer = m->vlan_tci_outer;
1526 mi->tx_offload = m->tx_offload;
1530 mi->pkt_len = mi->data_len;
1532 mi->packet_type = m->packet_type;
1533 mi->timestamp = m->timestamp;
1535 __rte_mbuf_sanity_check(mi, 1);
1536 __rte_mbuf_sanity_check(m, 0);
1540 * @internal used by rte_pktmbuf_detach().
1542 * Decrement the reference counter of the external buffer. When the
1543 * reference counter becomes 0, the buffer is freed by pre-registered
1547 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1549 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1550 RTE_ASSERT(m->shinfo != NULL);
1552 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1553 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1557 * @internal used by rte_pktmbuf_detach().
1559 * Decrement the direct mbuf's reference counter. When the reference
1560 * counter becomes 0, the direct mbuf is freed.
1563 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1565 struct rte_mbuf *md;
1567 RTE_ASSERT(RTE_MBUF_INDIRECT(m));
1569 md = rte_mbuf_from_indirect(m);
1571 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1574 rte_mbuf_refcnt_set(md, 1);
1575 rte_mbuf_raw_free(md);
1580 * Detach a packet mbuf from external buffer or direct buffer.
1582 * - decrement refcnt and free the external/direct buffer if refcnt
1584 * - restore original mbuf address and length values.
1585 * - reset pktmbuf data and data_len to their default values.
1587 * All other fields of the given packet mbuf will be left intact.
1590 * The indirect attached packet mbuf.
1592 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1594 struct rte_mempool *mp = m->pool;
1595 uint32_t mbuf_size, buf_len;
1598 if (RTE_MBUF_HAS_EXTBUF(m))
1599 __rte_pktmbuf_free_extbuf(m);
1601 __rte_pktmbuf_free_direct(m);
1603 priv_size = rte_pktmbuf_priv_size(mp);
1604 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1605 buf_len = rte_pktmbuf_data_room_size(mp);
1607 m->priv_size = priv_size;
1608 m->buf_addr = (char *)m + mbuf_size;
1609 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1610 m->buf_len = (uint16_t)buf_len;
1611 rte_pktmbuf_reset_headroom(m);
1617 * Decrease reference counter and unlink a mbuf segment
1619 * This function does the same than a free, except that it does not
1620 * return the segment to its pool.
1621 * It decreases the reference counter, and if it reaches 0, it is
1622 * detached from its parent for an indirect mbuf.
1625 * The mbuf to be unlinked
1627 * - (m) if it is the last reference. It can be recycled or freed.
1628 * - (NULL) if the mbuf still has remaining references on it.
1630 static __rte_always_inline struct rte_mbuf *
1631 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1633 __rte_mbuf_sanity_check(m, 0);
1635 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1637 if (!RTE_MBUF_DIRECT(m))
1638 rte_pktmbuf_detach(m);
1640 if (m->next != NULL) {
1647 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1649 if (!RTE_MBUF_DIRECT(m))
1650 rte_pktmbuf_detach(m);
1652 if (m->next != NULL) {
1656 rte_mbuf_refcnt_set(m, 1);
1664 * Free a segment of a packet mbuf into its original mempool.
1666 * Free an mbuf, without parsing other segments in case of chained
1670 * The packet mbuf segment to be freed.
1672 static __rte_always_inline void
1673 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1675 m = rte_pktmbuf_prefree_seg(m);
1676 if (likely(m != NULL))
1677 rte_mbuf_raw_free(m);
1681 * Free a packet mbuf back into its original mempool.
1683 * Free an mbuf, and all its segments in case of chained buffers. Each
1684 * segment is added back into its original mempool.
1687 * The packet mbuf to be freed. If NULL, the function does nothing.
1689 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1691 struct rte_mbuf *m_next;
1694 __rte_mbuf_sanity_check(m, 1);
1698 rte_pktmbuf_free_seg(m);
1704 * Creates a "clone" of the given packet mbuf.
1706 * Walks through all segments of the given packet mbuf, and for each of them:
1707 * - Creates a new packet mbuf from the given pool.
1708 * - Attaches newly created mbuf to the segment.
1709 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1710 * from the original packet mbuf.
1713 * The packet mbuf to be cloned.
1715 * The mempool from which the "clone" mbufs are allocated.
1717 * - The pointer to the new "clone" mbuf on success.
1718 * - NULL if allocation fails.
1720 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1721 struct rte_mempool *mp)
1723 struct rte_mbuf *mc, *mi, **prev;
1727 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1732 pktlen = md->pkt_len;
1737 rte_pktmbuf_attach(mi, md);
1740 } while ((md = md->next) != NULL &&
1741 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1745 mc->pkt_len = pktlen;
1747 /* Allocation of new indirect segment failed */
1748 if (unlikely (mi == NULL)) {
1749 rte_pktmbuf_free(mc);
1753 __rte_mbuf_sanity_check(mc, 1);
1758 * Adds given value to the refcnt of all packet mbuf segments.
1760 * Walks through all segments of given packet mbuf and for each of them
1761 * invokes rte_mbuf_refcnt_update().
1764 * The packet mbuf whose refcnt to be updated.
1766 * The value to add to the mbuf's segments refcnt.
1768 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1770 __rte_mbuf_sanity_check(m, 1);
1773 rte_mbuf_refcnt_update(m, v);
1774 } while ((m = m->next) != NULL);
1778 * Get the headroom in a packet mbuf.
1783 * The length of the headroom.
1785 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1787 __rte_mbuf_sanity_check(m, 0);
1792 * Get the tailroom of a packet mbuf.
1797 * The length of the tailroom.
1799 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1801 __rte_mbuf_sanity_check(m, 0);
1802 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1807 * Get the last segment of the packet.
1812 * The last segment of the given mbuf.
1814 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1816 __rte_mbuf_sanity_check(m, 1);
1817 while (m->next != NULL)
1823 * A macro that points to an offset into the data in the mbuf.
1825 * The returned pointer is cast to type t. Before using this
1826 * function, the user must ensure that the first segment is large
1827 * enough to accommodate its data.
1832 * The offset into the mbuf data.
1834 * The type to cast the result into.
1836 #define rte_pktmbuf_mtod_offset(m, t, o) \
1837 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1840 * A macro that points to the start of the data in the mbuf.
1842 * The returned pointer is cast to type t. Before using this
1843 * function, the user must ensure that the first segment is large
1844 * enough to accommodate its data.
1849 * The type to cast the result into.
1851 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1854 * A macro that returns the IO address that points to an offset of the
1855 * start of the data in the mbuf
1860 * The offset into the data to calculate address from.
1862 #define rte_pktmbuf_iova_offset(m, o) \
1863 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
1866 #define rte_pktmbuf_mtophys_offset(m, o) \
1867 rte_pktmbuf_iova_offset(m, o)
1870 * A macro that returns the IO address that points to the start of the
1876 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
1879 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1882 * A macro that returns the length of the packet.
1884 * The value can be read or assigned.
1889 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1892 * A macro that returns the length of the segment.
1894 * The value can be read or assigned.
1899 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1902 * Prepend len bytes to an mbuf data area.
1904 * Returns a pointer to the new
1905 * data start address. If there is not enough headroom in the first
1906 * segment, the function will return NULL, without modifying the mbuf.
1911 * The amount of data to prepend (in bytes).
1913 * A pointer to the start of the newly prepended data, or
1914 * NULL if there is not enough headroom space in the first segment
1916 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1919 __rte_mbuf_sanity_check(m, 1);
1921 if (unlikely(len > rte_pktmbuf_headroom(m)))
1924 /* NB: elaborating the subtraction like this instead of using
1925 * -= allows us to ensure the result type is uint16_t
1926 * avoiding compiler warnings on gcc 8.1 at least */
1927 m->data_off = (uint16_t)(m->data_off - len);
1928 m->data_len = (uint16_t)(m->data_len + len);
1929 m->pkt_len = (m->pkt_len + len);
1931 return (char *)m->buf_addr + m->data_off;
1935 * Append len bytes to an mbuf.
1937 * Append len bytes to an mbuf and return a pointer to the start address
1938 * of the added data. If there is not enough tailroom in the last
1939 * segment, the function will return NULL, without modifying the mbuf.
1944 * The amount of data to append (in bytes).
1946 * A pointer to the start of the newly appended data, or
1947 * NULL if there is not enough tailroom space in the last segment
1949 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1952 struct rte_mbuf *m_last;
1954 __rte_mbuf_sanity_check(m, 1);
1956 m_last = rte_pktmbuf_lastseg(m);
1957 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1960 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1961 m_last->data_len = (uint16_t)(m_last->data_len + len);
1962 m->pkt_len = (m->pkt_len + len);
1963 return (char*) tail;
1967 * Remove len bytes at the beginning of an mbuf.
1969 * Returns a pointer to the start address of the new data area. If the
1970 * length is greater than the length of the first segment, then the
1971 * function will fail and return NULL, without modifying the mbuf.
1976 * The amount of data to remove (in bytes).
1978 * A pointer to the new start of the data.
1980 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1982 __rte_mbuf_sanity_check(m, 1);
1984 if (unlikely(len > m->data_len))
1987 /* NB: elaborating the addition like this instead of using
1988 * += allows us to ensure the result type is uint16_t
1989 * avoiding compiler warnings on gcc 8.1 at least */
1990 m->data_len = (uint16_t)(m->data_len - len);
1991 m->data_off = (uint16_t)(m->data_off + len);
1992 m->pkt_len = (m->pkt_len - len);
1993 return (char *)m->buf_addr + m->data_off;
1997 * Remove len bytes of data at the end of the mbuf.
1999 * If the length is greater than the length of the last segment, the
2000 * function will fail and return -1 without modifying the mbuf.
2005 * The amount of data to remove (in bytes).
2010 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
2012 struct rte_mbuf *m_last;
2014 __rte_mbuf_sanity_check(m, 1);
2016 m_last = rte_pktmbuf_lastseg(m);
2017 if (unlikely(len > m_last->data_len))
2020 m_last->data_len = (uint16_t)(m_last->data_len - len);
2021 m->pkt_len = (m->pkt_len - len);
2026 * Test if mbuf data is contiguous.
2031 * - 1, if all data is contiguous (one segment).
2032 * - 0, if there is several segments.
2034 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
2036 __rte_mbuf_sanity_check(m, 1);
2037 return !!(m->nb_segs == 1);
2041 * @internal used by rte_pktmbuf_read().
2043 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
2044 uint32_t len, void *buf);
2047 * Read len data bytes in a mbuf at specified offset.
2049 * If the data is contiguous, return the pointer in the mbuf data, else
2050 * copy the data in the buffer provided by the user and return its
2054 * The pointer to the mbuf.
2056 * The offset of the data in the mbuf.
2058 * The amount of bytes to read.
2060 * The buffer where data is copied if it is not contiguous in mbuf
2061 * data. Its length should be at least equal to the len parameter.
2063 * The pointer to the data, either in the mbuf if it is contiguous,
2064 * or in the user buffer. If mbuf is too small, NULL is returned.
2066 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
2067 uint32_t off, uint32_t len, void *buf)
2069 if (likely(off + len <= rte_pktmbuf_data_len(m)))
2070 return rte_pktmbuf_mtod_offset(m, char *, off);
2072 return __rte_pktmbuf_read(m, off, len, buf);
2076 * Chain an mbuf to another, thereby creating a segmented packet.
2078 * Note: The implementation will do a linear walk over the segments to find
2079 * the tail entry. For cases when there are many segments, it's better to
2080 * chain the entries manually.
2083 * The head of the mbuf chain (the first packet)
2085 * The mbuf to put last in the chain
2089 * - -EOVERFLOW, if the chain segment limit exceeded
2091 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
2093 struct rte_mbuf *cur_tail;
2095 /* Check for number-of-segments-overflow */
2096 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
2099 /* Chain 'tail' onto the old tail */
2100 cur_tail = rte_pktmbuf_lastseg(head);
2101 cur_tail->next = tail;
2103 /* accumulate number of segments and total length.
2104 * NB: elaborating the addition like this instead of using
2105 * -= allows us to ensure the result type is uint16_t
2106 * avoiding compiler warnings on gcc 8.1 at least */
2107 head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
2108 head->pkt_len += tail->pkt_len;
2110 /* pkt_len is only set in the head */
2111 tail->pkt_len = tail->data_len;
2117 * Validate general requirements for Tx offload in mbuf.
2119 * This function checks correctness and completeness of Tx offload settings.
2122 * The packet mbuf to be validated.
2124 * 0 if packet is valid
2127 rte_validate_tx_offload(const struct rte_mbuf *m)
2129 uint64_t ol_flags = m->ol_flags;
2130 uint64_t inner_l3_offset = m->l2_len;
2132 /* Does packet set any of available offloads? */
2133 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
2136 if (ol_flags & PKT_TX_OUTER_IP_CKSUM)
2137 /* NB: elaborating the addition like this instead of using
2138 * += gives the result uint64_t type instead of int,
2139 * avoiding compiler warnings on gcc 8.1 at least */
2140 inner_l3_offset = inner_l3_offset + m->outer_l2_len +
2143 /* Headers are fragmented */
2144 if (rte_pktmbuf_data_len(m) < inner_l3_offset + m->l3_len + m->l4_len)
2147 /* IP checksum can be counted only for IPv4 packet */
2148 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
2151 /* IP type not set when required */
2152 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
2153 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
2156 /* Check requirements for TSO packet */
2157 if (ol_flags & PKT_TX_TCP_SEG)
2158 if ((m->tso_segsz == 0) ||
2159 ((ol_flags & PKT_TX_IPV4) &&
2160 !(ol_flags & PKT_TX_IP_CKSUM)))
2163 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
2164 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
2165 !(ol_flags & PKT_TX_OUTER_IPV4))
2172 * Linearize data in mbuf.
2174 * This function moves the mbuf data in the first segment if there is enough
2175 * tailroom. The subsequent segments are unchained and freed.
2184 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
2186 size_t seg_len, copy_len;
2188 struct rte_mbuf *m_next;
2191 if (rte_pktmbuf_is_contiguous(mbuf))
2194 /* Extend first segment to the total packet length */
2195 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
2197 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
2200 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
2201 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
2203 /* Append data from next segments to the first one */
2208 seg_len = rte_pktmbuf_data_len(m);
2209 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
2212 rte_pktmbuf_free_seg(m);
2223 * Dump an mbuf structure to a file.
2225 * Dump all fields for the given packet mbuf and all its associated
2226 * segments (in the case of a chained buffer).
2229 * A pointer to a file for output
2233 * If dump_len != 0, also dump the "dump_len" first data bytes of
2236 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
2242 #endif /* _RTE_MBUF_H_ */