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_byteorder.h>
44 #include <rte_mbuf_ptype.h>
51 * Packet Offload Features Flags. It also carry packet type information.
52 * Critical resources. Both rx/tx shared these bits. Be cautious on any change
54 * - RX flags start at bit position zero, and get added to the left of previous
56 * - The most-significant 3 bits are reserved for generic mbuf flags
57 * - TX flags therefore start at bit position 60 (i.e. 63-3), and new flags get
58 * added to the right of the previously defined flags i.e. they should count
59 * downwards, not upwards.
61 * Keep these flags synchronized with rte_get_rx_ol_flag_name() and
62 * rte_get_tx_ol_flag_name().
66 * The RX packet is a 802.1q VLAN packet, and the tci has been
67 * saved in in mbuf->vlan_tci.
68 * If the flag PKT_RX_VLAN_STRIPPED is also present, the VLAN
69 * header has been stripped from mbuf data, else it is still
72 #define PKT_RX_VLAN (1ULL << 0)
74 #define PKT_RX_RSS_HASH (1ULL << 1) /**< RX packet with RSS hash result. */
75 #define PKT_RX_FDIR (1ULL << 2) /**< RX packet with FDIR match indicate. */
79 * Checking this flag alone is deprecated: check the 2 bits of
80 * PKT_RX_L4_CKSUM_MASK.
81 * This flag was set when the L4 checksum of a packet was detected as
82 * wrong by the hardware.
84 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3)
88 * Checking this flag alone is deprecated: check the 2 bits of
89 * PKT_RX_IP_CKSUM_MASK.
90 * This flag was set when the IP checksum of a packet was detected as
91 * wrong by the hardware.
93 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4)
95 #define PKT_RX_EIP_CKSUM_BAD (1ULL << 5) /**< External IP header checksum error. */
98 * A vlan has been stripped by the hardware and its tci is saved in
99 * mbuf->vlan_tci. This can only happen if vlan stripping is enabled
100 * in the RX configuration of the PMD.
101 * When PKT_RX_VLAN_STRIPPED is set, PKT_RX_VLAN must also be set.
103 #define PKT_RX_VLAN_STRIPPED (1ULL << 6)
106 * Mask of bits used to determine the status of RX IP checksum.
107 * - PKT_RX_IP_CKSUM_UNKNOWN: no information about the RX IP checksum
108 * - PKT_RX_IP_CKSUM_BAD: the IP checksum in the packet is wrong
109 * - PKT_RX_IP_CKSUM_GOOD: the IP checksum in the packet is valid
110 * - PKT_RX_IP_CKSUM_NONE: the IP checksum is not correct in the packet
111 * data, but the integrity of the IP header is verified.
113 #define PKT_RX_IP_CKSUM_MASK ((1ULL << 4) | (1ULL << 7))
115 #define PKT_RX_IP_CKSUM_UNKNOWN 0
116 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4)
117 #define PKT_RX_IP_CKSUM_GOOD (1ULL << 7)
118 #define PKT_RX_IP_CKSUM_NONE ((1ULL << 4) | (1ULL << 7))
121 * Mask of bits used to determine the status of RX L4 checksum.
122 * - PKT_RX_L4_CKSUM_UNKNOWN: no information about the RX L4 checksum
123 * - PKT_RX_L4_CKSUM_BAD: the L4 checksum in the packet is wrong
124 * - PKT_RX_L4_CKSUM_GOOD: the L4 checksum in the packet is valid
125 * - PKT_RX_L4_CKSUM_NONE: the L4 checksum is not correct in the packet
126 * data, but the integrity of the L4 data is verified.
128 #define PKT_RX_L4_CKSUM_MASK ((1ULL << 3) | (1ULL << 8))
130 #define PKT_RX_L4_CKSUM_UNKNOWN 0
131 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3)
132 #define PKT_RX_L4_CKSUM_GOOD (1ULL << 8)
133 #define PKT_RX_L4_CKSUM_NONE ((1ULL << 3) | (1ULL << 8))
135 #define PKT_RX_IEEE1588_PTP (1ULL << 9) /**< RX IEEE1588 L2 Ethernet PT Packet. */
136 #define PKT_RX_IEEE1588_TMST (1ULL << 10) /**< RX IEEE1588 L2/L4 timestamped packet.*/
137 #define PKT_RX_FDIR_ID (1ULL << 13) /**< FD id reported if FDIR match. */
138 #define PKT_RX_FDIR_FLX (1ULL << 14) /**< Flexible bytes reported if FDIR match. */
141 * The 2 vlans have been stripped by the hardware and their tci are
142 * saved in mbuf->vlan_tci (inner) and mbuf->vlan_tci_outer (outer).
143 * This can only happen if vlan stripping is enabled in the RX
144 * configuration of the PMD.
145 * When PKT_RX_QINQ_STRIPPED is set, the flags (PKT_RX_VLAN |
146 * PKT_RX_VLAN_STRIPPED | PKT_RX_QINQ) must also be set.
148 #define PKT_RX_QINQ_STRIPPED (1ULL << 15)
151 * When packets are coalesced by a hardware or virtual driver, this flag
152 * can be set in the RX mbuf, meaning that the m->tso_segsz field is
153 * valid and is set to the segment size of original packets.
155 #define PKT_RX_LRO (1ULL << 16)
158 * Indicate that the timestamp field in the mbuf is valid.
160 #define PKT_RX_TIMESTAMP (1ULL << 17)
163 * Indicate that security offload processing was applied on the RX packet.
165 #define PKT_RX_SEC_OFFLOAD (1ULL << 18)
168 * Indicate that security offload processing failed on the RX packet.
170 #define PKT_RX_SEC_OFFLOAD_FAILED (1ULL << 19)
173 * The RX packet is a double VLAN, and the outer tci has been
174 * saved in in mbuf->vlan_tci_outer. If PKT_RX_QINQ set, PKT_RX_VLAN
175 * also should be set and inner tci should be saved to mbuf->vlan_tci.
176 * If the flag PKT_RX_QINQ_STRIPPED is also present, both VLANs
177 * headers have been stripped from mbuf data, else they are still
180 #define PKT_RX_QINQ (1ULL << 20)
183 * Mask of bits used to determine the status of outer RX L4 checksum.
184 * - PKT_RX_OUTER_L4_CKSUM_UNKNOWN: no info about the outer RX L4 checksum
185 * - PKT_RX_OUTER_L4_CKSUM_BAD: the outer L4 checksum in the packet is wrong
186 * - PKT_RX_OUTER_L4_CKSUM_GOOD: the outer L4 checksum in the packet is valid
187 * - PKT_RX_OUTER_L4_CKSUM_INVALID: invalid outer L4 checksum state.
189 * The detection of PKT_RX_OUTER_L4_CKSUM_GOOD shall be based on the given
190 * HW capability, At minimum, the PMD should support
191 * PKT_RX_OUTER_L4_CKSUM_UNKNOWN and PKT_RX_OUTER_L4_CKSUM_BAD states
192 * if the DEV_RX_OFFLOAD_OUTER_UDP_CKSUM offload is available.
194 #define PKT_RX_OUTER_L4_CKSUM_MASK ((1ULL << 21) | (1ULL << 22))
196 #define PKT_RX_OUTER_L4_CKSUM_UNKNOWN 0
197 #define PKT_RX_OUTER_L4_CKSUM_BAD (1ULL << 21)
198 #define PKT_RX_OUTER_L4_CKSUM_GOOD (1ULL << 22)
199 #define PKT_RX_OUTER_L4_CKSUM_INVALID ((1ULL << 21) | (1ULL << 22))
201 /* add new RX flags here */
203 /* add new TX flags here */
206 * Indicate that the metadata field in the mbuf is in use.
208 #define PKT_TX_METADATA (1ULL << 40)
211 * Outer UDP checksum offload flag. This flag is used for enabling
212 * outer UDP checksum in PMD. To use outer UDP checksum, the user needs to
213 * 1) Enable the following in mbuff,
214 * a) Fill outer_l2_len and outer_l3_len in mbuf.
215 * b) Set the PKT_TX_OUTER_UDP_CKSUM flag.
216 * c) Set the PKT_TX_OUTER_IPV4 or PKT_TX_OUTER_IPV6 flag.
217 * 2) Configure DEV_TX_OFFLOAD_OUTER_UDP_CKSUM offload flag.
219 #define PKT_TX_OUTER_UDP_CKSUM (1ULL << 41)
222 * UDP Fragmentation Offload flag. This flag is used for enabling UDP
223 * fragmentation in SW or in HW. When use UFO, mbuf->tso_segsz is used
224 * to store the MSS of UDP fragments.
226 #define PKT_TX_UDP_SEG (1ULL << 42)
229 * Request security offload processing on the TX packet.
231 #define PKT_TX_SEC_OFFLOAD (1ULL << 43)
234 * Offload the MACsec. This flag must be set by the application to enable
235 * this offload feature for a packet to be transmitted.
237 #define PKT_TX_MACSEC (1ULL << 44)
240 * Bits 45:48 used for the tunnel type.
241 * The tunnel type must be specified for TSO or checksum on the inner part
243 * These flags can be used with PKT_TX_TCP_SEG for TSO, or PKT_TX_xxx_CKSUM.
244 * The mbuf fields for inner and outer header lengths are required:
245 * outer_l2_len, outer_l3_len, l2_len, l3_len, l4_len and tso_segsz for TSO.
247 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
248 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
249 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
250 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
251 /** TX packet with MPLS-in-UDP RFC 7510 header. */
252 #define PKT_TX_TUNNEL_MPLSINUDP (0x5ULL << 45)
253 #define PKT_TX_TUNNEL_VXLAN_GPE (0x6ULL << 45)
255 * Generic IP encapsulated tunnel type, used for TSO and checksum offload.
256 * It can be used for tunnels which are not standards or listed above.
257 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_GRE
258 * or PKT_TX_TUNNEL_IPIP if possible.
259 * The ethdev must be configured with DEV_TX_OFFLOAD_IP_TNL_TSO.
260 * Outer and inner checksums are done according to the existing flags like
262 * Specific tunnel headers that contain payload length, sequence id
263 * or checksum are not expected to be updated.
265 #define PKT_TX_TUNNEL_IP (0xDULL << 45)
267 * Generic UDP encapsulated tunnel type, used for TSO and checksum offload.
268 * UDP tunnel type implies outer IP layer.
269 * It can be used for tunnels which are not standards or listed above.
270 * It is preferred to use specific tunnel flags like PKT_TX_TUNNEL_VXLAN
272 * The ethdev must be configured with DEV_TX_OFFLOAD_UDP_TNL_TSO.
273 * Outer and inner checksums are done according to the existing flags like
275 * Specific tunnel headers that contain payload length, sequence id
276 * or checksum are not expected to be updated.
278 #define PKT_TX_TUNNEL_UDP (0xEULL << 45)
279 /* add new TX TUNNEL type here */
280 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
283 * Double VLAN insertion (QinQ) request to driver, driver may offload the
284 * insertion based on device capability.
285 * mbuf 'vlan_tci' & 'vlan_tci_outer' must be valid when this flag is set.
287 #define PKT_TX_QINQ (1ULL << 49)
288 /* this old name is deprecated */
289 #define PKT_TX_QINQ_PKT PKT_TX_QINQ
292 * TCP segmentation offload. To enable this offload feature for a
293 * packet to be transmitted on hardware supporting TSO:
294 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
296 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
297 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag
298 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
300 #define PKT_TX_TCP_SEG (1ULL << 50)
302 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
305 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
306 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
307 * L4 checksum offload, the user needs to:
308 * - fill l2_len and l3_len in mbuf
309 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
310 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
312 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
313 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
314 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
315 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
316 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
319 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
320 * also be set by the application, although a PMD will only check
322 * - fill the mbuf offload information: l2_len, l3_len
324 #define PKT_TX_IP_CKSUM (1ULL << 54)
327 * Packet is IPv4. This flag must be set when using any offload feature
328 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
329 * packet. If the packet is a tunneled packet, this flag is related to
332 #define PKT_TX_IPV4 (1ULL << 55)
335 * Packet is IPv6. This flag must be set when using an offload feature
336 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
337 * packet. If the packet is a tunneled packet, this flag is related to
340 #define PKT_TX_IPV6 (1ULL << 56)
343 * VLAN tag insertion request to driver, driver may offload the insertion
344 * based on the device capability.
345 * mbuf 'vlan_tci' field must be valid when this flag is set.
347 #define PKT_TX_VLAN (1ULL << 57)
348 /* this old name is deprecated */
349 #define PKT_TX_VLAN_PKT PKT_TX_VLAN
352 * Offload the IP checksum of an external header in the hardware. The
353 * flag PKT_TX_OUTER_IPV4 should also be set by the application, although
354 * a PMD will only check PKT_TX_OUTER_IP_CKSUM.
355 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
357 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
360 * Packet outer header is IPv4. This flag must be set when using any
361 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
362 * outer header of the tunneled packet is an IPv4 packet.
364 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
367 * Packet outer header is IPv6. This flag must be set when using any
368 * outer offload feature (L4 checksum) to tell the NIC that the outer
369 * header of the tunneled packet is an IPv6 packet.
371 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
374 * Bitmask of all supported packet Tx offload features flags,
375 * which can be set for packet.
377 #define PKT_TX_OFFLOAD_MASK ( \
378 PKT_TX_OUTER_IPV6 | \
379 PKT_TX_OUTER_IPV4 | \
380 PKT_TX_OUTER_IP_CKSUM | \
386 PKT_TX_IEEE1588_TMST | \
389 PKT_TX_TUNNEL_MASK | \
391 PKT_TX_SEC_OFFLOAD | \
393 PKT_TX_OUTER_UDP_CKSUM | \
397 * Mbuf having an external buffer attached. shinfo in mbuf must be filled.
399 #define EXT_ATTACHED_MBUF (1ULL << 61)
401 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
403 /** Alignment constraint of mbuf private area. */
404 #define RTE_MBUF_PRIV_ALIGN 8
407 * Get the name of a RX offload flag
410 * The mask describing the flag.
412 * The name of this flag, or NULL if it's not a valid RX flag.
414 const char *rte_get_rx_ol_flag_name(uint64_t mask);
417 * Dump the list of RX offload flags in a buffer
420 * The mask describing the RX flags.
424 * The length of the buffer.
426 * 0 on success, (-1) on error.
428 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
431 * Get the name of a TX offload flag
434 * The mask describing the flag. Usually only one bit must be set.
435 * Several bits can be given if they belong to the same mask.
436 * Ex: PKT_TX_L4_MASK.
438 * The name of this flag, or NULL if it's not a valid TX flag.
440 const char *rte_get_tx_ol_flag_name(uint64_t mask);
443 * Dump the list of TX offload flags in a buffer
446 * The mask describing the TX flags.
450 * The length of the buffer.
452 * 0 on success, (-1) on error.
454 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
457 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
458 * splitting it into multiple segments.
459 * So, for mbufs that planned to be involved into RX/TX, the recommended
460 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
462 #define RTE_MBUF_DEFAULT_DATAROOM 2048
463 #define RTE_MBUF_DEFAULT_BUF_SIZE \
464 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
466 /* define a set of marker types that can be used to refer to set points in the
469 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
471 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
473 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
474 * with a single assignment */
476 struct rte_mbuf_sched {
477 uint32_t queue_id; /**< Queue ID. */
478 uint8_t traffic_class;
479 /**< Traffic class ID. Traffic class 0
480 * is the highest priority traffic class.
483 /**< Color. @see enum rte_color.*/
484 uint16_t reserved; /**< Reserved. */
485 }; /**< Hierarchical scheduler */
488 * enum for the tx_offload bit-fields lenghts and offsets.
489 * defines the layout of rte_mbuf tx_offload field.
492 RTE_MBUF_L2_LEN_BITS = 7,
493 RTE_MBUF_L3_LEN_BITS = 9,
494 RTE_MBUF_L4_LEN_BITS = 8,
495 RTE_MBUF_TSO_SEGSZ_BITS = 16,
496 RTE_MBUF_OUTL3_LEN_BITS = 9,
497 RTE_MBUF_OUTL2_LEN_BITS = 7,
498 RTE_MBUF_TXOFLD_UNUSED_BITS = sizeof(uint64_t) * CHAR_BIT -
499 RTE_MBUF_L2_LEN_BITS -
500 RTE_MBUF_L3_LEN_BITS -
501 RTE_MBUF_L4_LEN_BITS -
502 RTE_MBUF_TSO_SEGSZ_BITS -
503 RTE_MBUF_OUTL3_LEN_BITS -
504 RTE_MBUF_OUTL2_LEN_BITS,
505 #if RTE_BYTE_ORDER == RTE_BIG_ENDIAN
506 RTE_MBUF_L2_LEN_OFS =
507 sizeof(uint64_t) * CHAR_BIT - RTE_MBUF_L2_LEN_BITS
508 RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS - RTE_MBUF_L3_LEN_BITS,
509 RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS - RTE_MBUF_L4_LEN_BITS,
510 RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS - RTE_MBUF_TSO_SEGSZ_BITS,
511 RTE_MBUF_OUTL3_LEN_OFS =
512 RTE_MBUF_TSO_SEGSZ_OFS - RTE_MBUF_OUTL3_LEN_BITS,
513 RTE_MBUF_OUTL2_LEN_OFS =
514 RTE_MBUF_OUTL3_LEN_OFS - RTE_MBUF_OUTL2_LEN_BITS,
515 RTE_MBUF_TXOFLD_UNUSED_OFS =
516 RTE_MBUF_OUTL2_LEN_OFS - RTE_MBUF_TXOFLD_UNUSED_BITS,
518 RTE_MBUF_L2_LEN_OFS = 0,
519 RTE_MBUF_L3_LEN_OFS = RTE_MBUF_L2_LEN_OFS + RTE_MBUF_L2_LEN_BITS,
520 RTE_MBUF_L4_LEN_OFS = RTE_MBUF_L3_LEN_OFS + RTE_MBUF_L3_LEN_BITS,
521 RTE_MBUF_TSO_SEGSZ_OFS = RTE_MBUF_L4_LEN_OFS + RTE_MBUF_L4_LEN_BITS,
522 RTE_MBUF_OUTL3_LEN_OFS =
523 RTE_MBUF_TSO_SEGSZ_OFS + RTE_MBUF_TSO_SEGSZ_BITS,
524 RTE_MBUF_OUTL2_LEN_OFS =
525 RTE_MBUF_OUTL3_LEN_OFS + RTE_MBUF_OUTL3_LEN_BITS,
526 RTE_MBUF_TXOFLD_UNUSED_OFS =
527 RTE_MBUF_OUTL2_LEN_OFS + RTE_MBUF_OUTL2_LEN_BITS,
532 * The generic rte_mbuf, containing a packet mbuf.
537 void *buf_addr; /**< Virtual address of segment buffer. */
539 * Physical address of segment buffer.
540 * Force alignment to 8-bytes, so as to ensure we have the exact
541 * same mbuf cacheline0 layout for 32-bit and 64-bit. This makes
542 * working on vector drivers easier.
547 rte_iova_t buf_physaddr; /**< deprecated */
548 } __rte_aligned(sizeof(rte_iova_t));
550 /* next 8 bytes are initialised on RX descriptor rearm */
555 * Reference counter. Its size should at least equal to the size
556 * of port field (16 bits), to support zero-copy broadcast.
557 * It should only be accessed using the following functions:
558 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
559 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
560 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
565 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
566 uint16_t refcnt; /**< Non-atomically accessed refcnt */
568 uint16_t nb_segs; /**< Number of segments. */
570 /** Input port (16 bits to support more than 256 virtual ports).
571 * The event eth Tx adapter uses this field to specify the output port.
575 uint64_t ol_flags; /**< Offload features. */
577 /* remaining bytes are set on RX when pulling packet from descriptor */
578 MARKER rx_descriptor_fields1;
581 * The packet type, which is the combination of outer/inner L2, L3, L4
582 * and tunnel types. The packet_type is about data really present in the
583 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
584 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
585 * vlan is stripped from the data.
589 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
591 uint32_t l2_type:4; /**< (Outer) L2 type. */
592 uint32_t l3_type:4; /**< (Outer) L3 type. */
593 uint32_t l4_type:4; /**< (Outer) L4 type. */
594 uint32_t tun_type:4; /**< Tunnel type. */
597 uint8_t inner_esp_next_proto;
598 /**< ESP next protocol type, valid if
599 * RTE_PTYPE_TUNNEL_ESP tunnel type is set
604 uint8_t inner_l2_type:4;
605 /**< Inner L2 type. */
606 uint8_t inner_l3_type:4;
607 /**< Inner L3 type. */
610 uint32_t inner_l4_type:4; /**< Inner L4 type. */
614 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
615 uint16_t data_len; /**< Amount of data in segment buffer. */
616 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN is set. */
622 uint32_t rss; /**< RSS hash result if RSS enabled */
630 /**< Second 4 flexible bytes */
633 /**< First 4 flexible bytes or FD ID, dependent
634 * on PKT_RX_FDIR_* flag in ol_flags.
636 } fdir; /**< Filter identifier if FDIR enabled */
637 struct rte_mbuf_sched sched;
638 /**< Hierarchical scheduler : 8 bytes */
643 /**< The event eth Tx adapter uses this field
644 * to store Tx queue id.
645 * @see rte_event_eth_tx_adapter_txq_set()
647 } txadapter; /**< Eventdev ethdev Tx adapter */
648 /**< User defined tags. See rte_distributor_process() */
650 } hash; /**< hash information */
653 * Application specific metadata value
654 * for egress flow rule match.
655 * Valid if PKT_TX_METADATA is set.
656 * Located here to allow conjunct use
657 * with hash.sched.hi.
659 uint32_t tx_metadata;
664 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ is set. */
665 uint16_t vlan_tci_outer;
667 uint16_t buf_len; /**< Length of segment buffer. */
669 /** Valid if PKT_RX_TIMESTAMP is set. The unit and time reference
670 * are not normalized but are always the same for a given port.
674 /* second cache line - fields only used in slow path or on TX */
675 MARKER cacheline1 __rte_cache_min_aligned;
679 void *userdata; /**< Can be used for external metadata */
680 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
683 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
684 struct rte_mbuf *next; /**< Next segment of scattered packet. */
686 /* fields to support TX offloads */
689 uint64_t tx_offload; /**< combined for easy fetch */
692 uint64_t l2_len:RTE_MBUF_L2_LEN_BITS;
693 /**< L2 (MAC) Header Length for non-tunneling pkt.
694 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
696 uint64_t l3_len:RTE_MBUF_L3_LEN_BITS;
697 /**< L3 (IP) Header Length. */
698 uint64_t l4_len:RTE_MBUF_L4_LEN_BITS;
699 /**< L4 (TCP/UDP) Header Length. */
700 uint64_t tso_segsz:RTE_MBUF_TSO_SEGSZ_BITS;
701 /**< TCP TSO segment size */
703 /* fields for TX offloading of tunnels */
704 uint64_t outer_l3_len:RTE_MBUF_OUTL3_LEN_BITS;
705 /**< Outer L3 (IP) Hdr Length. */
706 uint64_t outer_l2_len:RTE_MBUF_OUTL2_LEN_BITS;
707 /**< Outer L2 (MAC) Hdr Length. */
709 /* uint64_t unused:RTE_MBUF_TXOFLD_UNUSED_BITS; */
713 /** Size of the application private data. In case of an indirect
714 * mbuf, it stores the direct mbuf private data size. */
717 /** Timesync flags for use with IEEE1588. */
720 /** Sequence number. See also rte_reorder_insert(). */
723 /** Shared data for external buffer attached to mbuf. See
724 * rte_pktmbuf_attach_extbuf().
726 struct rte_mbuf_ext_shared_info *shinfo;
728 } __rte_cache_aligned;
731 * Function typedef of callback to free externally attached buffer.
733 typedef void (*rte_mbuf_extbuf_free_callback_t)(void *addr, void *opaque);
736 * Shared data at the end of an external buffer.
738 struct rte_mbuf_ext_shared_info {
739 rte_mbuf_extbuf_free_callback_t free_cb; /**< Free callback function */
740 void *fcb_opaque; /**< Free callback argument */
741 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
744 /**< Maximum number of nb_segs allowed. */
745 #define RTE_MBUF_MAX_NB_SEGS UINT16_MAX
748 * Prefetch the first part of the mbuf
750 * The first 64 bytes of the mbuf corresponds to fields that are used early
751 * in the receive path. If the cache line of the architecture is higher than
752 * 64B, the second part will also be prefetched.
755 * The pointer to the mbuf.
758 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
760 rte_prefetch0(&m->cacheline0);
764 * Prefetch the second part of the mbuf
766 * The next 64 bytes of the mbuf corresponds to fields that are used in the
767 * transmit path. If the cache line of the architecture is higher than 64B,
768 * this function does nothing as it is expected that the full mbuf is
772 * The pointer to the mbuf.
775 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
777 #if RTE_CACHE_LINE_SIZE == 64
778 rte_prefetch0(&m->cacheline1);
785 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
788 * Return the IO address of the beginning of the mbuf data
791 * The pointer to the mbuf.
793 * The IO address of the beginning of the mbuf data
795 static inline rte_iova_t
796 rte_mbuf_data_iova(const struct rte_mbuf *mb)
798 return mb->buf_iova + mb->data_off;
802 static inline phys_addr_t
803 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
805 return rte_mbuf_data_iova(mb);
809 * Return the default IO address of the beginning of the mbuf data
811 * This function is used by drivers in their receive function, as it
812 * returns the location where data should be written by the NIC, taking
813 * the default headroom in account.
816 * The pointer to the mbuf.
818 * The IO address of the beginning of the mbuf data
820 static inline rte_iova_t
821 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
823 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
827 static inline phys_addr_t
828 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
830 return rte_mbuf_data_iova_default(mb);
834 * Return the mbuf owning the data buffer address of an indirect mbuf.
837 * The pointer to the indirect mbuf.
839 * The address of the direct mbuf corresponding to buffer_addr.
841 static inline struct rte_mbuf *
842 rte_mbuf_from_indirect(struct rte_mbuf *mi)
844 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
848 * Return address of buffer embedded in the given mbuf.
850 * The return value shall be same as mb->buf_addr if the mbuf is already
851 * initialized and direct. However, this API is useful if mempool of the
852 * mbuf is already known because it doesn't need to access mbuf contents in
853 * order to get the mempool pointer.
856 * @b EXPERIMENTAL: This API may change without prior notice.
857 * This will be used by rte_mbuf_to_baddr() which has redundant code once
858 * experimental tag is removed.
861 * The pointer to the mbuf.
863 * The pointer to the mempool of the mbuf.
865 * The pointer of the mbuf buffer.
867 static inline char * __rte_experimental
868 rte_mbuf_buf_addr(struct rte_mbuf *mb, struct rte_mempool *mp)
870 return (char *)mb + sizeof(*mb) + rte_pktmbuf_priv_size(mp);
874 * Return the default address of the beginning of the mbuf data.
877 * @b EXPERIMENTAL: This API may change without prior notice.
880 * The pointer to the mbuf.
882 * The pointer of the beginning of the mbuf data.
884 static inline char * __rte_experimental
885 rte_mbuf_data_addr_default(struct rte_mbuf *mb)
887 return rte_mbuf_buf_addr(mb, mb->pool) + RTE_PKTMBUF_HEADROOM;
891 * Return address of buffer embedded in the given mbuf.
893 * @note: Accessing mempool pointer of a mbuf is expensive because the
894 * pointer is stored in the 2nd cache line of mbuf. If mempool is known, it
895 * is better not to reference the mempool pointer in mbuf but calling
896 * rte_mbuf_buf_addr() would be more efficient.
899 * The pointer to the mbuf.
901 * The address of the data buffer owned by the mbuf.
904 rte_mbuf_to_baddr(struct rte_mbuf *md)
906 #ifdef ALLOW_EXPERIMENTAL_API
907 return rte_mbuf_buf_addr(md, md->pool);
910 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
916 * Return the starting address of the private data area embedded in
919 * Note that no check is made to ensure that a private data area
920 * actually exists in the supplied mbuf.
923 * The pointer to the mbuf.
925 * The starting address of the private data area of the given mbuf.
927 static inline void * __rte_experimental
928 rte_mbuf_to_priv(struct rte_mbuf *m)
930 return RTE_PTR_ADD(m, sizeof(struct rte_mbuf));
934 * Returns TRUE if given mbuf is cloned by mbuf indirection, or FALSE
937 * If a mbuf has its data in another mbuf and references it by mbuf
938 * indirection, this mbuf can be defined as a cloned mbuf.
940 #define RTE_MBUF_CLONED(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
943 * Returns TRUE if given mbuf has an external buffer, or FALSE otherwise.
945 * External buffer is a user-provided anonymous buffer.
947 #define RTE_MBUF_HAS_EXTBUF(mb) ((mb)->ol_flags & EXT_ATTACHED_MBUF)
950 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
952 * If a mbuf embeds its own data after the rte_mbuf structure, this mbuf
953 * can be defined as a direct mbuf.
955 #define RTE_MBUF_DIRECT(mb) \
956 (!((mb)->ol_flags & (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF)))
959 * Private data in case of pktmbuf pool.
961 * A structure that contains some pktmbuf_pool-specific data that are
962 * appended after the mempool structure (in private data).
964 struct rte_pktmbuf_pool_private {
965 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
966 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
969 #ifdef RTE_LIBRTE_MBUF_DEBUG
971 /** check mbuf type in debug mode */
972 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
974 #else /* RTE_LIBRTE_MBUF_DEBUG */
976 /** check mbuf type in debug mode */
977 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
979 #endif /* RTE_LIBRTE_MBUF_DEBUG */
981 #ifdef RTE_MBUF_REFCNT_ATOMIC
984 * Reads the value of an mbuf's refcnt.
988 * Reference count number.
990 static inline uint16_t
991 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
993 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
997 * Sets an mbuf's refcnt to a defined value.
1004 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
1006 rte_atomic16_set(&m->refcnt_atomic, (int16_t)new_value);
1010 static inline uint16_t
1011 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
1013 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
1017 * Adds given value to an mbuf's refcnt and returns its new value.
1021 * Value to add/subtract
1025 static inline uint16_t
1026 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
1029 * The atomic_add is an expensive operation, so we don't want to
1030 * call it in the case where we know we are the unique holder of
1031 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
1032 * operation has to be used because concurrent accesses on the
1033 * reference counter can occur.
1035 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1037 rte_mbuf_refcnt_set(m, (uint16_t)value);
1038 return (uint16_t)value;
1041 return __rte_mbuf_refcnt_update(m, value);
1044 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
1047 static inline uint16_t
1048 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
1050 m->refcnt = (uint16_t)(m->refcnt + value);
1055 * Adds given value to an mbuf's refcnt and returns its new value.
1057 static inline uint16_t
1058 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
1060 return __rte_mbuf_refcnt_update(m, value);
1064 * Reads the value of an mbuf's refcnt.
1066 static inline uint16_t
1067 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
1073 * Sets an mbuf's refcnt to the defined value.
1076 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
1078 m->refcnt = new_value;
1081 #endif /* RTE_MBUF_REFCNT_ATOMIC */
1084 * Reads the refcnt of an external buffer.
1087 * Shared data of the external buffer.
1089 * Reference count number.
1091 static inline uint16_t
1092 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
1094 return (uint16_t)(rte_atomic16_read(&shinfo->refcnt_atomic));
1098 * Set refcnt of an external buffer.
1101 * Shared data of the external buffer.
1106 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
1109 rte_atomic16_set(&shinfo->refcnt_atomic, (int16_t)new_value);
1113 * Add given value to refcnt of an external buffer and return its new
1117 * Shared data of the external buffer.
1119 * Value to add/subtract
1123 static inline uint16_t
1124 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
1127 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
1129 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
1130 return (uint16_t)value;
1133 return (uint16_t)rte_atomic16_add_return(&shinfo->refcnt_atomic, value);
1136 /** Mbuf prefetch */
1137 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
1144 * Sanity checks on an mbuf.
1146 * Check the consistency of the given mbuf. The function will cause a
1147 * panic if corruption is detected.
1150 * The mbuf to be checked.
1152 * True if the mbuf is a packet header, false if it is a sub-segment
1153 * of a packet (in this case, some fields like nb_segs are not checked)
1156 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
1159 * Sanity checks on a mbuf.
1161 * Almost like rte_mbuf_sanity_check(), but this function gives the reason
1162 * if corruption is detected rather than panic.
1165 * The mbuf to be checked.
1167 * True if the mbuf is a packet header, false if it is a sub-segment
1168 * of a packet (in this case, some fields like nb_segs are not checked)
1170 * A reference to a string pointer where to store the reason why a mbuf is
1171 * considered invalid.
1173 * - 0 if no issue has been found, reason is left untouched.
1174 * - -1 if a problem is detected, reason then points to a string describing
1175 * the reason why the mbuf is deemed invalid.
1178 int rte_mbuf_check(const struct rte_mbuf *m, int is_header,
1179 const char **reason);
1181 #define MBUF_RAW_ALLOC_CHECK(m) do { \
1182 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
1183 RTE_ASSERT((m)->next == NULL); \
1184 RTE_ASSERT((m)->nb_segs == 1); \
1185 __rte_mbuf_sanity_check(m, 0); \
1189 * Allocate an uninitialized mbuf from mempool *mp*.
1191 * This function can be used by PMDs (especially in RX functions) to
1192 * allocate an uninitialized mbuf. The driver is responsible of
1193 * initializing all the required fields. See rte_pktmbuf_reset().
1194 * For standard needs, prefer rte_pktmbuf_alloc().
1196 * The caller can expect that the following fields of the mbuf structure
1197 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
1198 * next=NULL, pool, priv_size. The other fields must be initialized
1202 * The mempool from which mbuf is allocated.
1204 * - The pointer to the new mbuf on success.
1205 * - NULL if allocation failed.
1207 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
1211 if (rte_mempool_get(mp, (void **)&m) < 0)
1213 MBUF_RAW_ALLOC_CHECK(m);
1218 * Put mbuf back into its original mempool.
1220 * The caller must ensure that the mbuf is direct and properly
1221 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
1222 * rte_pktmbuf_prefree_seg().
1224 * This function should be used with care, when optimization is
1225 * required. For standard needs, prefer rte_pktmbuf_free() or
1226 * rte_pktmbuf_free_seg().
1229 * The mbuf to be freed.
1231 static __rte_always_inline void
1232 rte_mbuf_raw_free(struct rte_mbuf *m)
1234 RTE_ASSERT(RTE_MBUF_DIRECT(m));
1235 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
1236 RTE_ASSERT(m->next == NULL);
1237 RTE_ASSERT(m->nb_segs == 1);
1238 __rte_mbuf_sanity_check(m, 0);
1239 rte_mempool_put(m->pool, m);
1243 * The packet mbuf constructor.
1245 * This function initializes some fields in the mbuf structure that are
1246 * not modified by the user once created (origin pool, buffer start
1247 * address, and so on). This function is given as a callback function to
1248 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
1251 * The mempool from which mbufs originate.
1253 * A pointer that can be used by the user to retrieve useful information
1254 * for mbuf initialization. This pointer is the opaque argument passed to
1255 * rte_mempool_obj_iter() or rte_mempool_create().
1257 * The mbuf to initialize.
1259 * The index of the mbuf in the pool table.
1261 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
1262 void *m, unsigned i);
1266 * A packet mbuf pool constructor.
1268 * This function initializes the mempool private data in the case of a
1269 * pktmbuf pool. This private data is needed by the driver. The
1270 * function must be called on the mempool before it is used, or it
1271 * can be given as a callback function to rte_mempool_create() at
1272 * pool creation. It can be extended by the user, for example, to
1273 * provide another packet size.
1276 * The mempool from which mbufs originate.
1278 * A pointer that can be used by the user to retrieve useful information
1279 * for mbuf initialization. This pointer is the opaque argument passed to
1280 * rte_mempool_create().
1282 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
1285 * Create a mbuf pool.
1287 * This function creates and initializes a packet mbuf pool. It is
1288 * a wrapper to rte_mempool functions.
1291 * The name of the mbuf pool.
1293 * The number of elements in the mbuf pool. The optimum size (in terms
1294 * of memory usage) for a mempool is when n is a power of two minus one:
1297 * Size of the per-core object cache. See rte_mempool_create() for
1300 * Size of application private are between the rte_mbuf structure
1301 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1302 * @param data_room_size
1303 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1305 * The socket identifier where the memory should be allocated. The
1306 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1309 * The pointer to the new allocated mempool, on success. NULL on error
1310 * with rte_errno set appropriately. Possible rte_errno values include:
1311 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1312 * - E_RTE_SECONDARY - function was called from a secondary process instance
1313 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1314 * - ENOSPC - the maximum number of memzones has already been allocated
1315 * - EEXIST - a memzone with the same name already exists
1316 * - ENOMEM - no appropriate memory area found in which to create memzone
1318 struct rte_mempool *
1319 rte_pktmbuf_pool_create(const char *name, unsigned n,
1320 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
1324 * Create a mbuf pool with a given mempool ops name
1326 * This function creates and initializes a packet mbuf pool. It is
1327 * a wrapper to rte_mempool functions.
1330 * The name of the mbuf pool.
1332 * The number of elements in the mbuf pool. The optimum size (in terms
1333 * of memory usage) for a mempool is when n is a power of two minus one:
1336 * Size of the per-core object cache. See rte_mempool_create() for
1339 * Size of application private are between the rte_mbuf structure
1340 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
1341 * @param data_room_size
1342 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
1344 * The socket identifier where the memory should be allocated. The
1345 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
1348 * The mempool ops name to be used for this mempool instead of
1349 * default mempool. The value can be *NULL* to use default mempool.
1351 * The pointer to the new allocated mempool, on success. NULL on error
1352 * with rte_errno set appropriately. Possible rte_errno values include:
1353 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
1354 * - E_RTE_SECONDARY - function was called from a secondary process instance
1355 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
1356 * - ENOSPC - the maximum number of memzones has already been allocated
1357 * - EEXIST - a memzone with the same name already exists
1358 * - ENOMEM - no appropriate memory area found in which to create memzone
1360 struct rte_mempool *
1361 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
1362 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
1363 int socket_id, const char *ops_name);
1366 * Get the data room size of mbufs stored in a pktmbuf_pool
1368 * The data room size is the amount of data that can be stored in a
1369 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
1372 * The packet mbuf pool.
1374 * The data room size of mbufs stored in this mempool.
1376 static inline uint16_t
1377 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
1379 struct rte_pktmbuf_pool_private *mbp_priv;
1381 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1382 return mbp_priv->mbuf_data_room_size;
1386 * Get the application private size of mbufs stored in a pktmbuf_pool
1388 * The private size of mbuf is a zone located between the rte_mbuf
1389 * structure and the data buffer where an application can store data
1390 * associated to a packet.
1393 * The packet mbuf pool.
1395 * The private size of mbufs stored in this mempool.
1397 static inline uint16_t
1398 rte_pktmbuf_priv_size(struct rte_mempool *mp)
1400 struct rte_pktmbuf_pool_private *mbp_priv;
1402 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
1403 return mbp_priv->mbuf_priv_size;
1407 * Reset the data_off field of a packet mbuf to its default value.
1409 * The given mbuf must have only one segment, which should be empty.
1412 * The packet mbuf's data_off field has to be reset.
1414 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
1416 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
1417 (uint16_t)m->buf_len);
1421 * Reset the fields of a packet mbuf to their default values.
1423 * The given mbuf must have only one segment.
1426 * The packet mbuf to be resetted.
1428 #define MBUF_INVALID_PORT UINT16_MAX
1430 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
1436 m->vlan_tci_outer = 0;
1438 m->port = MBUF_INVALID_PORT;
1442 rte_pktmbuf_reset_headroom(m);
1445 __rte_mbuf_sanity_check(m, 1);
1449 * Allocate a new mbuf from a mempool.
1451 * This new mbuf contains one segment, which has a length of 0. The pointer
1452 * to data is initialized to have some bytes of headroom in the buffer
1453 * (if buffer size allows).
1456 * The mempool from which the mbuf is allocated.
1458 * - The pointer to the new mbuf on success.
1459 * - NULL if allocation failed.
1461 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
1464 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
1465 rte_pktmbuf_reset(m);
1470 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1474 * The mempool from which mbufs are allocated.
1476 * Array of pointers to mbufs
1481 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
1483 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1484 struct rte_mbuf **mbufs, unsigned count)
1489 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1493 /* To understand duff's device on loop unwinding optimization, see
1494 * https://en.wikipedia.org/wiki/Duff's_device.
1495 * Here while() loop is used rather than do() while{} to avoid extra
1496 * check if count is zero.
1498 switch (count % 4) {
1500 while (idx != count) {
1501 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1502 rte_pktmbuf_reset(mbufs[idx]);
1506 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1507 rte_pktmbuf_reset(mbufs[idx]);
1511 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1512 rte_pktmbuf_reset(mbufs[idx]);
1516 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
1517 rte_pktmbuf_reset(mbufs[idx]);
1526 * Initialize shared data at the end of an external buffer before attaching
1527 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
1528 * initialization but a helper function to simply spare a few bytes at the
1529 * end of the buffer for shared data. If shared data is allocated
1530 * separately, this should not be called but application has to properly
1531 * initialize the shared data according to its need.
1533 * Free callback and its argument is saved and the refcnt is set to 1.
1536 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
1537 * after this initialization. This shall be used for
1538 * ``rte_pktmbuf_attach_extbuf()``
1541 * The pointer to the external buffer.
1542 * @param [in,out] buf_len
1543 * The pointer to length of the external buffer. Input value must be
1544 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
1545 * padding for alignment. If not enough, this function will return NULL.
1546 * Adjusted buffer length will be returned through this pointer.
1548 * Free callback function to call when the external buffer needs to be
1551 * Argument for the free callback function.
1554 * A pointer to the initialized shared data on success, return NULL
1557 static inline struct rte_mbuf_ext_shared_info *
1558 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
1559 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
1561 struct rte_mbuf_ext_shared_info *shinfo;
1562 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
1565 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
1567 if (addr <= buf_addr)
1570 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
1571 shinfo->free_cb = free_cb;
1572 shinfo->fcb_opaque = fcb_opaque;
1573 rte_mbuf_ext_refcnt_set(shinfo, 1);
1575 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
1580 * Attach an external buffer to a mbuf.
1582 * User-managed anonymous buffer can be attached to an mbuf. When attaching
1583 * it, corresponding free callback function and its argument should be
1584 * provided via shinfo. This callback function will be called once all the
1585 * mbufs are detached from the buffer (refcnt becomes zero).
1587 * The headroom for the attaching mbuf will be set to zero and this can be
1588 * properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1589 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1591 * More mbufs can be attached to the same external buffer by
1592 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1595 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1596 * ``rte_pktmbuf_detach()``.
1598 * Memory for shared data must be provided and user must initialize all of
1599 * the content properly, escpecially free callback and refcnt. The pointer
1600 * of shared data will be stored in m->shinfo.
1601 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1602 * bytes at the end of buffer for the shared data, store free callback and
1603 * its argument and set the refcnt to 1. The following is an example:
1605 * struct rte_mbuf_ext_shared_info *shinfo =
1606 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1607 * free_cb, fcb_arg);
1608 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1609 * rte_pktmbuf_reset_headroom(m);
1610 * rte_pktmbuf_adj(m, data_len);
1612 * Attaching an external buffer is quite similar to mbuf indirection in
1613 * replacing buffer addresses and length of a mbuf, but a few differences:
1614 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1615 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1616 * In such cases, the buffer area of a direct mbuf must be read-only. But
1617 * external buffer has its own refcnt and it starts from 1. Unless
1618 * multiple mbufs are attached to a mbuf having an external buffer, the
1619 * external buffer is writable.
1620 * - There's no need to allocate buffer from a mempool. Any buffer can be
1621 * attached with appropriate free callback and its IO address.
1622 * - Smaller metadata is required to maintain shared data such as refcnt.
1625 * The pointer to the mbuf.
1627 * The pointer to the external buffer.
1629 * IO address of the external buffer.
1631 * The size of the external buffer.
1633 * User-provided memory for shared data of the external buffer.
1636 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1637 rte_iova_t buf_iova, uint16_t buf_len,
1638 struct rte_mbuf_ext_shared_info *shinfo)
1640 /* mbuf should not be read-only */
1641 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1642 RTE_ASSERT(shinfo->free_cb != NULL);
1644 m->buf_addr = buf_addr;
1645 m->buf_iova = buf_iova;
1646 m->buf_len = buf_len;
1651 m->ol_flags |= EXT_ATTACHED_MBUF;
1656 * Detach the external buffer attached to a mbuf, same as
1657 * ``rte_pktmbuf_detach()``
1660 * The mbuf having external buffer.
1662 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1665 * Attach packet mbuf to another packet mbuf.
1667 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1668 * an external buffer, the mbuf being attached will be attached to the
1669 * external buffer instead of mbuf indirection.
1671 * Otherwise, the mbuf will be indirectly attached. After attachment we
1672 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1673 * 'direct'. The direct mbuf's reference counter is incremented.
1675 * Right now, not supported:
1676 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1677 * - mbuf we trying to attach (mi) is used by someone else
1678 * e.g. it's reference counter is greater then 1.
1681 * The indirect packet mbuf.
1683 * The packet mbuf we're attaching to.
1685 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1687 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1688 rte_mbuf_refcnt_read(mi) == 1);
1690 if (RTE_MBUF_HAS_EXTBUF(m)) {
1691 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1692 mi->ol_flags = m->ol_flags;
1693 mi->shinfo = m->shinfo;
1695 /* if m is not direct, get the mbuf that embeds the data */
1696 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1697 mi->priv_size = m->priv_size;
1698 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1701 mi->buf_iova = m->buf_iova;
1702 mi->buf_addr = m->buf_addr;
1703 mi->buf_len = m->buf_len;
1705 mi->data_off = m->data_off;
1706 mi->data_len = m->data_len;
1708 mi->vlan_tci = m->vlan_tci;
1709 mi->vlan_tci_outer = m->vlan_tci_outer;
1710 mi->tx_offload = m->tx_offload;
1714 mi->pkt_len = mi->data_len;
1716 mi->packet_type = m->packet_type;
1717 mi->timestamp = m->timestamp;
1719 __rte_mbuf_sanity_check(mi, 1);
1720 __rte_mbuf_sanity_check(m, 0);
1724 * @internal used by rte_pktmbuf_detach().
1726 * Decrement the reference counter of the external buffer. When the
1727 * reference counter becomes 0, the buffer is freed by pre-registered
1731 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1733 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1734 RTE_ASSERT(m->shinfo != NULL);
1736 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1737 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1741 * @internal used by rte_pktmbuf_detach().
1743 * Decrement the direct mbuf's reference counter. When the reference
1744 * counter becomes 0, the direct mbuf is freed.
1747 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1749 struct rte_mbuf *md;
1751 RTE_ASSERT(RTE_MBUF_CLONED(m));
1753 md = rte_mbuf_from_indirect(m);
1755 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1758 rte_mbuf_refcnt_set(md, 1);
1759 rte_mbuf_raw_free(md);
1764 * Detach a packet mbuf from external buffer or direct buffer.
1766 * - decrement refcnt and free the external/direct buffer if refcnt
1768 * - restore original mbuf address and length values.
1769 * - reset pktmbuf data and data_len to their default values.
1771 * All other fields of the given packet mbuf will be left intact.
1774 * The indirect attached packet mbuf.
1776 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1778 struct rte_mempool *mp = m->pool;
1779 uint32_t mbuf_size, buf_len;
1782 if (RTE_MBUF_HAS_EXTBUF(m))
1783 __rte_pktmbuf_free_extbuf(m);
1785 __rte_pktmbuf_free_direct(m);
1787 priv_size = rte_pktmbuf_priv_size(mp);
1788 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1789 buf_len = rte_pktmbuf_data_room_size(mp);
1791 m->priv_size = priv_size;
1792 m->buf_addr = (char *)m + mbuf_size;
1793 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1794 m->buf_len = (uint16_t)buf_len;
1795 rte_pktmbuf_reset_headroom(m);
1801 * Decrease reference counter and unlink a mbuf segment
1803 * This function does the same than a free, except that it does not
1804 * return the segment to its pool.
1805 * It decreases the reference counter, and if it reaches 0, it is
1806 * detached from its parent for an indirect mbuf.
1809 * The mbuf to be unlinked
1811 * - (m) if it is the last reference. It can be recycled or freed.
1812 * - (NULL) if the mbuf still has remaining references on it.
1814 static __rte_always_inline struct rte_mbuf *
1815 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1817 __rte_mbuf_sanity_check(m, 0);
1819 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1821 if (!RTE_MBUF_DIRECT(m))
1822 rte_pktmbuf_detach(m);
1824 if (m->next != NULL) {
1831 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1833 if (!RTE_MBUF_DIRECT(m))
1834 rte_pktmbuf_detach(m);
1836 if (m->next != NULL) {
1840 rte_mbuf_refcnt_set(m, 1);
1848 * Free a segment of a packet mbuf into its original mempool.
1850 * Free an mbuf, without parsing other segments in case of chained
1854 * The packet mbuf segment to be freed.
1856 static __rte_always_inline void
1857 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1859 m = rte_pktmbuf_prefree_seg(m);
1860 if (likely(m != NULL))
1861 rte_mbuf_raw_free(m);
1865 * Free a packet mbuf back into its original mempool.
1867 * Free an mbuf, and all its segments in case of chained buffers. Each
1868 * segment is added back into its original mempool.
1871 * The packet mbuf to be freed. If NULL, the function does nothing.
1873 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1875 struct rte_mbuf *m_next;
1878 __rte_mbuf_sanity_check(m, 1);
1882 rte_pktmbuf_free_seg(m);
1888 * Creates a "clone" of the given packet mbuf.
1890 * Walks through all segments of the given packet mbuf, and for each of them:
1891 * - Creates a new packet mbuf from the given pool.
1892 * - Attaches newly created mbuf to the segment.
1893 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1894 * from the original packet mbuf.
1897 * The packet mbuf to be cloned.
1899 * The mempool from which the "clone" mbufs are allocated.
1901 * - The pointer to the new "clone" mbuf on success.
1902 * - NULL if allocation fails.
1904 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1905 struct rte_mempool *mp)
1907 struct rte_mbuf *mc, *mi, **prev;
1911 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1916 pktlen = md->pkt_len;
1921 rte_pktmbuf_attach(mi, md);
1924 } while ((md = md->next) != NULL &&
1925 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1929 mc->pkt_len = pktlen;
1931 /* Allocation of new indirect segment failed */
1932 if (unlikely (mi == NULL)) {
1933 rte_pktmbuf_free(mc);
1937 __rte_mbuf_sanity_check(mc, 1);
1942 * Adds given value to the refcnt of all packet mbuf segments.
1944 * Walks through all segments of given packet mbuf and for each of them
1945 * invokes rte_mbuf_refcnt_update().
1948 * The packet mbuf whose refcnt to be updated.
1950 * The value to add to the mbuf's segments refcnt.
1952 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1954 __rte_mbuf_sanity_check(m, 1);
1957 rte_mbuf_refcnt_update(m, v);
1958 } while ((m = m->next) != NULL);
1962 * Get the headroom in a packet mbuf.
1967 * The length of the headroom.
1969 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1971 __rte_mbuf_sanity_check(m, 0);
1976 * Get the tailroom of a packet mbuf.
1981 * The length of the tailroom.
1983 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1985 __rte_mbuf_sanity_check(m, 0);
1986 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1991 * Get the last segment of the packet.
1996 * The last segment of the given mbuf.
1998 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
2000 __rte_mbuf_sanity_check(m, 1);
2001 while (m->next != NULL)
2007 * A macro that points to an offset into the data in the mbuf.
2009 * The returned pointer is cast to type t. Before using this
2010 * function, the user must ensure that the first segment is large
2011 * enough to accommodate its data.
2016 * The offset into the mbuf data.
2018 * The type to cast the result into.
2020 #define rte_pktmbuf_mtod_offset(m, t, o) \
2021 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
2024 * A macro that points to the start of the data in the mbuf.
2026 * The returned pointer is cast to type t. Before using this
2027 * function, the user must ensure that the first segment is large
2028 * enough to accommodate its data.
2033 * The type to cast the result into.
2035 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
2038 * A macro that returns the IO address that points to an offset of the
2039 * start of the data in the mbuf
2044 * The offset into the data to calculate address from.
2046 #define rte_pktmbuf_iova_offset(m, o) \
2047 (rte_iova_t)((m)->buf_iova + (m)->data_off + (o))
2050 #define rte_pktmbuf_mtophys_offset(m, o) \
2051 rte_pktmbuf_iova_offset(m, o)
2054 * A macro that returns the IO address that points to the start of the
2060 #define rte_pktmbuf_iova(m) rte_pktmbuf_iova_offset(m, 0)
2063 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
2066 * A macro that returns the length of the packet.
2068 * The value can be read or assigned.
2073 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
2076 * A macro that returns the length of the segment.
2078 * The value can be read or assigned.
2083 #define rte_pktmbuf_data_len(m) ((m)->data_len)
2086 * Prepend len bytes to an mbuf data area.
2088 * Returns a pointer to the new
2089 * data start address. If there is not enough headroom in the first
2090 * segment, the function will return NULL, without modifying the mbuf.
2095 * The amount of data to prepend (in bytes).
2097 * A pointer to the start of the newly prepended data, or
2098 * NULL if there is not enough headroom space in the first segment
2100 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
2103 __rte_mbuf_sanity_check(m, 1);
2105 if (unlikely(len > rte_pktmbuf_headroom(m)))
2108 /* NB: elaborating the subtraction like this instead of using
2109 * -= allows us to ensure the result type is uint16_t
2110 * avoiding compiler warnings on gcc 8.1 at least */
2111 m->data_off = (uint16_t)(m->data_off - len);
2112 m->data_len = (uint16_t)(m->data_len + len);
2113 m->pkt_len = (m->pkt_len + len);
2115 return (char *)m->buf_addr + m->data_off;
2119 * Append len bytes to an mbuf.
2121 * Append len bytes to an mbuf and return a pointer to the start address
2122 * of the added data. If there is not enough tailroom in the last
2123 * segment, the function will return NULL, without modifying the mbuf.
2128 * The amount of data to append (in bytes).
2130 * A pointer to the start of the newly appended data, or
2131 * NULL if there is not enough tailroom space in the last segment
2133 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
2136 struct rte_mbuf *m_last;
2138 __rte_mbuf_sanity_check(m, 1);
2140 m_last = rte_pktmbuf_lastseg(m);
2141 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
2144 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
2145 m_last->data_len = (uint16_t)(m_last->data_len + len);
2146 m->pkt_len = (m->pkt_len + len);
2147 return (char*) tail;
2151 * Remove len bytes at the beginning of an mbuf.
2153 * Returns a pointer to the start address of the new data area. If the
2154 * length is greater than the length of the first segment, then the
2155 * function will fail and return NULL, without modifying the mbuf.
2160 * The amount of data to remove (in bytes).
2162 * A pointer to the new start of the data.
2164 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
2166 __rte_mbuf_sanity_check(m, 1);
2168 if (unlikely(len > m->data_len))
2171 /* NB: elaborating the addition like this instead of using
2172 * += allows us to ensure the result type is uint16_t
2173 * avoiding compiler warnings on gcc 8.1 at least */
2174 m->data_len = (uint16_t)(m->data_len - len);
2175 m->data_off = (uint16_t)(m->data_off + len);
2176 m->pkt_len = (m->pkt_len - len);
2177 return (char *)m->buf_addr + m->data_off;
2181 * Remove len bytes of data at the end of the mbuf.
2183 * If the length is greater than the length of the last segment, the
2184 * function will fail and return -1 without modifying the mbuf.
2189 * The amount of data to remove (in bytes).
2194 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
2196 struct rte_mbuf *m_last;
2198 __rte_mbuf_sanity_check(m, 1);
2200 m_last = rte_pktmbuf_lastseg(m);
2201 if (unlikely(len > m_last->data_len))
2204 m_last->data_len = (uint16_t)(m_last->data_len - len);
2205 m->pkt_len = (m->pkt_len - len);
2210 * Test if mbuf data is contiguous.
2215 * - 1, if all data is contiguous (one segment).
2216 * - 0, if there is several segments.
2218 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
2220 __rte_mbuf_sanity_check(m, 1);
2221 return !!(m->nb_segs == 1);
2225 * @internal used by rte_pktmbuf_read().
2227 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
2228 uint32_t len, void *buf);
2231 * Read len data bytes in a mbuf at specified offset.
2233 * If the data is contiguous, return the pointer in the mbuf data, else
2234 * copy the data in the buffer provided by the user and return its
2238 * The pointer to the mbuf.
2240 * The offset of the data in the mbuf.
2242 * The amount of bytes to read.
2244 * The buffer where data is copied if it is not contiguous in mbuf
2245 * data. Its length should be at least equal to the len parameter.
2247 * The pointer to the data, either in the mbuf if it is contiguous,
2248 * or in the user buffer. If mbuf is too small, NULL is returned.
2250 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
2251 uint32_t off, uint32_t len, void *buf)
2253 if (likely(off + len <= rte_pktmbuf_data_len(m)))
2254 return rte_pktmbuf_mtod_offset(m, char *, off);
2256 return __rte_pktmbuf_read(m, off, len, buf);
2260 * Chain an mbuf to another, thereby creating a segmented packet.
2262 * Note: The implementation will do a linear walk over the segments to find
2263 * the tail entry. For cases when there are many segments, it's better to
2264 * chain the entries manually.
2267 * The head of the mbuf chain (the first packet)
2269 * The mbuf to put last in the chain
2273 * - -EOVERFLOW, if the chain segment limit exceeded
2275 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
2277 struct rte_mbuf *cur_tail;
2279 /* Check for number-of-segments-overflow */
2280 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
2283 /* Chain 'tail' onto the old tail */
2284 cur_tail = rte_pktmbuf_lastseg(head);
2285 cur_tail->next = tail;
2287 /* accumulate number of segments and total length.
2288 * NB: elaborating the addition like this instead of using
2289 * -= allows us to ensure the result type is uint16_t
2290 * avoiding compiler warnings on gcc 8.1 at least */
2291 head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
2292 head->pkt_len += tail->pkt_len;
2294 /* pkt_len is only set in the head */
2295 tail->pkt_len = tail->data_len;
2302 * @b EXPERIMENTAL: This API may change without prior notice.
2304 * For given input values generate raw tx_offload value.
2305 * Note that it is caller responsibility to make sure that input parameters
2306 * don't exceed maximum bit-field values.
2316 * outer_l3_len value.
2318 * outer_l2_len value.
2322 * raw tx_offload value.
2324 static __rte_always_inline uint64_t
2325 rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
2326 uint64_t ol3, uint64_t ol2, uint64_t unused)
2328 return il2 << RTE_MBUF_L2_LEN_OFS |
2329 il3 << RTE_MBUF_L3_LEN_OFS |
2330 il4 << RTE_MBUF_L4_LEN_OFS |
2331 tso << RTE_MBUF_TSO_SEGSZ_OFS |
2332 ol3 << RTE_MBUF_OUTL3_LEN_OFS |
2333 ol2 << RTE_MBUF_OUTL2_LEN_OFS |
2334 unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
2338 * Validate general requirements for Tx offload in mbuf.
2340 * This function checks correctness and completeness of Tx offload settings.
2343 * The packet mbuf to be validated.
2345 * 0 if packet is valid
2348 rte_validate_tx_offload(const struct rte_mbuf *m)
2350 uint64_t ol_flags = m->ol_flags;
2352 /* Does packet set any of available offloads? */
2353 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
2356 /* IP checksum can be counted only for IPv4 packet */
2357 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
2360 /* IP type not set when required */
2361 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
2362 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
2365 /* Check requirements for TSO packet */
2366 if (ol_flags & PKT_TX_TCP_SEG)
2367 if ((m->tso_segsz == 0) ||
2368 ((ol_flags & PKT_TX_IPV4) &&
2369 !(ol_flags & PKT_TX_IP_CKSUM)))
2372 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
2373 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
2374 !(ol_flags & PKT_TX_OUTER_IPV4))
2381 * Linearize data in mbuf.
2383 * This function moves the mbuf data in the first segment if there is enough
2384 * tailroom. The subsequent segments are unchained and freed.
2393 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
2395 size_t seg_len, copy_len;
2397 struct rte_mbuf *m_next;
2400 if (rte_pktmbuf_is_contiguous(mbuf))
2403 /* Extend first segment to the total packet length */
2404 copy_len = rte_pktmbuf_pkt_len(mbuf) - rte_pktmbuf_data_len(mbuf);
2406 if (unlikely(copy_len > rte_pktmbuf_tailroom(mbuf)))
2409 buffer = rte_pktmbuf_mtod_offset(mbuf, char *, mbuf->data_len);
2410 mbuf->data_len = (uint16_t)(mbuf->pkt_len);
2412 /* Append data from next segments to the first one */
2417 seg_len = rte_pktmbuf_data_len(m);
2418 rte_memcpy(buffer, rte_pktmbuf_mtod(m, char *), seg_len);
2421 rte_pktmbuf_free_seg(m);
2432 * Dump an mbuf structure to a file.
2434 * Dump all fields for the given packet mbuf and all its associated
2435 * segments (in the case of a chained buffer).
2438 * A pointer to a file for output
2442 * If dump_len != 0, also dump the "dump_len" first data bytes of
2445 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
2448 * Get the value of mbuf sched queue_id field.
2450 static inline uint32_t
2451 rte_mbuf_sched_queue_get(const struct rte_mbuf *m)
2453 return m->hash.sched.queue_id;
2457 * Get the value of mbuf sched traffic_class field.
2459 static inline uint8_t
2460 rte_mbuf_sched_traffic_class_get(const struct rte_mbuf *m)
2462 return m->hash.sched.traffic_class;
2466 * Get the value of mbuf sched color field.
2468 static inline uint8_t
2469 rte_mbuf_sched_color_get(const struct rte_mbuf *m)
2471 return m->hash.sched.color;
2475 * Get the values of mbuf sched queue_id, traffic_class and color.
2480 * Returns the queue id
2481 * @param traffic_class
2482 * Returns the traffic class id
2484 * Returns the colour id
2487 rte_mbuf_sched_get(const struct rte_mbuf *m, uint32_t *queue_id,
2488 uint8_t *traffic_class,
2491 struct rte_mbuf_sched sched = m->hash.sched;
2493 *queue_id = sched.queue_id;
2494 *traffic_class = sched.traffic_class;
2495 *color = sched.color;
2499 * Set the mbuf sched queue_id to the defined value.
2502 rte_mbuf_sched_queue_set(struct rte_mbuf *m, uint32_t queue_id)
2504 m->hash.sched.queue_id = queue_id;
2508 * Set the mbuf sched traffic_class id to the defined value.
2511 rte_mbuf_sched_traffic_class_set(struct rte_mbuf *m, uint8_t traffic_class)
2513 m->hash.sched.traffic_class = traffic_class;
2517 * Set the mbuf sched color id to the defined value.
2520 rte_mbuf_sched_color_set(struct rte_mbuf *m, uint8_t color)
2522 m->hash.sched.color = color;
2526 * Set the mbuf sched queue_id, traffic_class and color.
2531 * Queue id value to be set
2532 * @param traffic_class
2533 * Traffic class id value to be set
2535 * Color id to be set
2538 rte_mbuf_sched_set(struct rte_mbuf *m, uint32_t queue_id,
2539 uint8_t traffic_class,
2542 m->hash.sched = (struct rte_mbuf_sched){
2543 .queue_id = queue_id,
2544 .traffic_class = traffic_class,
2554 #endif /* _RTE_MBUF_H_ */