4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 * Copyright 2014 6WIND S.A.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * * Neither the name of Intel Corporation nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42 * The mbuf library provides the ability to create and destroy buffers
43 * that may be used by the RTE application to store message
44 * buffers. The message buffers are stored in a mempool, using the
45 * RTE mempool library.
47 * This library provide an API to allocate/free packet mbufs, which are
48 * used to carry network packets.
50 * To understand the concepts of packet buffers or mbufs, you
51 * should read "TCP/IP Illustrated, Volume 2: The Implementation,
52 * Addison-Wesley, 1995, ISBN 0-201-63354-X from Richard Stevens"
53 * http://www.kohala.com/start/tcpipiv2.html
57 #include <rte_common.h>
58 #include <rte_mempool.h>
59 #include <rte_memory.h>
60 #include <rte_atomic.h>
61 #include <rte_prefetch.h>
62 #include <rte_branch_prediction.h>
63 #include <rte_mbuf_ptype.h>
70 * Packet Offload Features Flags. It also carry packet type information.
71 * Critical resources. Both rx/tx shared these bits. Be cautious on any change
73 * - RX flags start at bit position zero, and get added to the left of previous
75 * - The most-significant 3 bits are reserved for generic mbuf flags
76 * - TX flags therefore start at bit position 60 (i.e. 63-3), and new flags get
77 * added to the right of the previously defined flags i.e. they should count
78 * downwards, not upwards.
80 * Keep these flags synchronized with rte_get_rx_ol_flag_name() and
81 * rte_get_tx_ol_flag_name().
85 * RX packet is a 802.1q VLAN packet. This flag was set by PMDs when
86 * the packet is recognized as a VLAN, but the behavior between PMDs
87 * was not the same. This flag is kept for some time to avoid breaking
88 * applications and should be replaced by PKT_RX_VLAN_STRIPPED.
90 #define PKT_RX_VLAN_PKT (1ULL << 0)
92 #define PKT_RX_RSS_HASH (1ULL << 1) /**< RX packet with RSS hash result. */
93 #define PKT_RX_FDIR (1ULL << 2) /**< RX packet with FDIR match indicate. */
94 #define PKT_RX_L4_CKSUM_BAD (1ULL << 3) /**< L4 cksum of RX pkt. is not OK. */
95 #define PKT_RX_IP_CKSUM_BAD (1ULL << 4) /**< IP cksum of RX pkt. is not OK. */
96 #define PKT_RX_EIP_CKSUM_BAD (1ULL << 5) /**< External IP header checksum error. */
99 * A vlan has been stripped by the hardware and its tci is saved in
100 * mbuf->vlan_tci. This can only happen if vlan stripping is enabled
101 * in the RX configuration of the PMD.
103 #define PKT_RX_VLAN_STRIPPED (1ULL << 6)
105 /* hole, some bits can be reused here */
107 #define PKT_RX_IEEE1588_PTP (1ULL << 9) /**< RX IEEE1588 L2 Ethernet PT Packet. */
108 #define PKT_RX_IEEE1588_TMST (1ULL << 10) /**< RX IEEE1588 L2/L4 timestamped packet.*/
109 #define PKT_RX_FDIR_ID (1ULL << 13) /**< FD id reported if FDIR match. */
110 #define PKT_RX_FDIR_FLX (1ULL << 14) /**< Flexible bytes reported if FDIR match. */
113 * The 2 vlans have been stripped by the hardware and their tci are
114 * saved in mbuf->vlan_tci (inner) and mbuf->vlan_tci_outer (outer).
115 * This can only happen if vlan stripping is enabled in the RX
116 * configuration of the PMD. If this flag is set, PKT_RX_VLAN_STRIPPED
119 #define PKT_RX_QINQ_STRIPPED (1ULL << 15)
123 * RX packet with double VLAN stripped.
124 * This flag is replaced by PKT_RX_QINQ_STRIPPED.
126 #define PKT_RX_QINQ_PKT PKT_RX_QINQ_STRIPPED
128 /* add new RX flags here */
130 /* add new TX flags here */
133 * Bits 45:48 used for the tunnel type.
134 * When doing Tx offload like TSO or checksum, the HW needs to configure the
135 * tunnel type into the HW descriptors.
137 #define PKT_TX_TUNNEL_VXLAN (0x1ULL << 45)
138 #define PKT_TX_TUNNEL_GRE (0x2ULL << 45)
139 #define PKT_TX_TUNNEL_IPIP (0x3ULL << 45)
140 #define PKT_TX_TUNNEL_GENEVE (0x4ULL << 45)
141 /* add new TX TUNNEL type here */
142 #define PKT_TX_TUNNEL_MASK (0xFULL << 45)
145 * Second VLAN insertion (QinQ) flag.
147 #define PKT_TX_QINQ_PKT (1ULL << 49) /**< TX packet with double VLAN inserted. */
150 * TCP segmentation offload. To enable this offload feature for a
151 * packet to be transmitted on hardware supporting TSO:
152 * - set the PKT_TX_TCP_SEG flag in mbuf->ol_flags (this flag implies
154 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
155 * - if it's IPv4, set the PKT_TX_IP_CKSUM flag and write the IP checksum
157 * - fill the mbuf offload information: l2_len, l3_len, l4_len, tso_segsz
158 * - calculate the pseudo header checksum without taking ip_len in account,
159 * and set it in the TCP header. Refer to rte_ipv4_phdr_cksum() and
160 * rte_ipv6_phdr_cksum() that can be used as helpers.
162 #define PKT_TX_TCP_SEG (1ULL << 50)
164 #define PKT_TX_IEEE1588_TMST (1ULL << 51) /**< TX IEEE1588 packet to timestamp. */
167 * Bits 52+53 used for L4 packet type with checksum enabled: 00: Reserved,
168 * 01: TCP checksum, 10: SCTP checksum, 11: UDP checksum. To use hardware
169 * L4 checksum offload, the user needs to:
170 * - fill l2_len and l3_len in mbuf
171 * - set the flags PKT_TX_TCP_CKSUM, PKT_TX_SCTP_CKSUM or PKT_TX_UDP_CKSUM
172 * - set the flag PKT_TX_IPV4 or PKT_TX_IPV6
173 * - calculate the pseudo header checksum and set it in the L4 header (only
174 * for TCP or UDP). See rte_ipv4_phdr_cksum() and rte_ipv6_phdr_cksum().
175 * For SCTP, set the crc field to 0.
177 #define PKT_TX_L4_NO_CKSUM (0ULL << 52) /**< Disable L4 cksum of TX pkt. */
178 #define PKT_TX_TCP_CKSUM (1ULL << 52) /**< TCP cksum of TX pkt. computed by NIC. */
179 #define PKT_TX_SCTP_CKSUM (2ULL << 52) /**< SCTP cksum of TX pkt. computed by NIC. */
180 #define PKT_TX_UDP_CKSUM (3ULL << 52) /**< UDP cksum of TX pkt. computed by NIC. */
181 #define PKT_TX_L4_MASK (3ULL << 52) /**< Mask for L4 cksum offload request. */
184 * Offload the IP checksum in the hardware. The flag PKT_TX_IPV4 should
185 * also be set by the application, although a PMD will only check
187 * - set the IP checksum field in the packet to 0
188 * - fill the mbuf offload information: l2_len, l3_len
190 #define PKT_TX_IP_CKSUM (1ULL << 54)
193 * Packet is IPv4. This flag must be set when using any offload feature
194 * (TSO, L3 or L4 checksum) to tell the NIC that the packet is an IPv4
195 * packet. If the packet is a tunneled packet, this flag is related to
198 #define PKT_TX_IPV4 (1ULL << 55)
201 * Packet is IPv6. This flag must be set when using an offload feature
202 * (TSO or L4 checksum) to tell the NIC that the packet is an IPv6
203 * packet. If the packet is a tunneled packet, this flag is related to
206 #define PKT_TX_IPV6 (1ULL << 56)
208 #define PKT_TX_VLAN_PKT (1ULL << 57) /**< TX packet is a 802.1q VLAN packet. */
211 * Offload the IP checksum of an external header in the hardware. The
212 * flag PKT_TX_OUTER_IPV4 should also be set by the application, alto ugh
213 * a PMD will only check PKT_TX_IP_CKSUM. The IP checksum field in the
214 * packet must be set to 0.
215 * - set the outer IP checksum field in the packet to 0
216 * - fill the mbuf offload information: outer_l2_len, outer_l3_len
218 #define PKT_TX_OUTER_IP_CKSUM (1ULL << 58)
221 * Packet outer header is IPv4. This flag must be set when using any
222 * outer offload feature (L3 or L4 checksum) to tell the NIC that the
223 * outer header of the tunneled packet is an IPv4 packet.
225 #define PKT_TX_OUTER_IPV4 (1ULL << 59)
228 * Packet outer header is IPv6. This flag must be set when using any
229 * outer offload feature (L4 checksum) to tell the NIC that the outer
230 * header of the tunneled packet is an IPv6 packet.
232 #define PKT_TX_OUTER_IPV6 (1ULL << 60)
234 #define __RESERVED (1ULL << 61) /**< reserved for future mbuf use */
236 #define IND_ATTACHED_MBUF (1ULL << 62) /**< Indirect attached mbuf */
238 /* Use final bit of flags to indicate a control mbuf */
239 #define CTRL_MBUF_FLAG (1ULL << 63) /**< Mbuf contains control data */
241 /** Alignment constraint of mbuf private area. */
242 #define RTE_MBUF_PRIV_ALIGN 8
245 * Get the name of a RX offload flag
248 * The mask describing the flag.
250 * The name of this flag, or NULL if it's not a valid RX flag.
252 const char *rte_get_rx_ol_flag_name(uint64_t mask);
255 * Dump the list of RX offload flags in a buffer
258 * The mask describing the RX flags.
262 * The length of the buffer.
264 * 0 on success, (-1) on error.
266 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
269 * Get the name of a TX offload flag
272 * The mask describing the flag. Usually only one bit must be set.
273 * Several bits can be given if they belong to the same mask.
274 * Ex: PKT_TX_L4_MASK.
276 * The name of this flag, or NULL if it's not a valid TX flag.
278 const char *rte_get_tx_ol_flag_name(uint64_t mask);
281 * Dump the list of TX offload flags in a buffer
284 * The mask describing the TX flags.
288 * The length of the buffer.
290 * 0 on success, (-1) on error.
292 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
295 * Some NICs need at least 2KB buffer to RX standard Ethernet frame without
296 * splitting it into multiple segments.
297 * So, for mbufs that planned to be involved into RX/TX, the recommended
298 * minimal buffer length is 2KB + RTE_PKTMBUF_HEADROOM.
300 #define RTE_MBUF_DEFAULT_DATAROOM 2048
301 #define RTE_MBUF_DEFAULT_BUF_SIZE \
302 (RTE_MBUF_DEFAULT_DATAROOM + RTE_PKTMBUF_HEADROOM)
304 /* define a set of marker types that can be used to refer to set points in the
307 typedef void *MARKER[0]; /**< generic marker for a point in a structure */
309 typedef uint8_t MARKER8[0]; /**< generic marker with 1B alignment */
311 typedef uint64_t MARKER64[0]; /**< marker that allows us to overwrite 8 bytes
312 * with a single assignment */
315 * The generic rte_mbuf, containing a packet mbuf.
320 void *buf_addr; /**< Virtual address of segment buffer. */
321 phys_addr_t buf_physaddr; /**< Physical address of segment buffer. */
323 uint16_t buf_len; /**< Length of segment buffer. */
325 /* next 6 bytes are initialised on RX descriptor rearm */
330 * 16-bit Reference counter.
331 * It should only be accessed using the following functions:
332 * rte_mbuf_refcnt_update(), rte_mbuf_refcnt_read(), and
333 * rte_mbuf_refcnt_set(). The functionality of these functions (atomic,
334 * or non-atomic) is controlled by the CONFIG_RTE_MBUF_REFCNT_ATOMIC
339 rte_atomic16_t refcnt_atomic; /**< Atomically accessed refcnt */
340 uint16_t refcnt; /**< Non-atomically accessed refcnt */
342 uint8_t nb_segs; /**< Number of segments. */
343 uint8_t port; /**< Input port. */
345 uint64_t ol_flags; /**< Offload features. */
347 /* remaining bytes are set on RX when pulling packet from descriptor */
348 MARKER rx_descriptor_fields1;
351 * The packet type, which is the combination of outer/inner L2, L3, L4
352 * and tunnel types. The packet_type is about data really present in the
353 * mbuf. Example: if vlan stripping is enabled, a received vlan packet
354 * would have RTE_PTYPE_L2_ETHER and not RTE_PTYPE_L2_VLAN because the
355 * vlan is stripped from the data.
359 uint32_t packet_type; /**< L2/L3/L4 and tunnel information. */
361 uint32_t l2_type:4; /**< (Outer) L2 type. */
362 uint32_t l3_type:4; /**< (Outer) L3 type. */
363 uint32_t l4_type:4; /**< (Outer) L4 type. */
364 uint32_t tun_type:4; /**< Tunnel type. */
365 uint32_t inner_l2_type:4; /**< Inner L2 type. */
366 uint32_t inner_l3_type:4; /**< Inner L3 type. */
367 uint32_t inner_l4_type:4; /**< Inner L4 type. */
371 uint32_t pkt_len; /**< Total pkt len: sum of all segments. */
372 uint16_t data_len; /**< Amount of data in segment buffer. */
373 /** VLAN TCI (CPU order), valid if PKT_RX_VLAN_STRIPPED is set. */
377 uint32_t rss; /**< RSS hash result if RSS enabled */
386 /**< Second 4 flexible bytes */
389 /**< First 4 flexible bytes or FD ID, dependent on
390 PKT_RX_FDIR_* flag in ol_flags. */
391 } fdir; /**< Filter identifier if FDIR enabled */
395 } sched; /**< Hierarchical scheduler */
396 uint32_t usr; /**< User defined tags. See rte_distributor_process() */
397 } hash; /**< hash information */
399 uint32_t seqn; /**< Sequence number. See also rte_reorder_insert() */
401 /** Outer VLAN TCI (CPU order), valid if PKT_RX_QINQ_STRIPPED is set. */
402 uint16_t vlan_tci_outer;
404 /* second cache line - fields only used in slow path or on TX */
405 MARKER cacheline1 __rte_cache_min_aligned;
409 void *userdata; /**< Can be used for external metadata */
410 uint64_t udata64; /**< Allow 8-byte userdata on 32-bit */
413 struct rte_mempool *pool; /**< Pool from which mbuf was allocated. */
414 struct rte_mbuf *next; /**< Next segment of scattered packet. */
416 /* fields to support TX offloads */
419 uint64_t tx_offload; /**< combined for easy fetch */
423 /**< L2 (MAC) Header Length for non-tunneling pkt.
424 * Outer_L4_len + ... + Inner_L2_len for tunneling pkt.
426 uint64_t l3_len:9; /**< L3 (IP) Header Length. */
427 uint64_t l4_len:8; /**< L4 (TCP/UDP) Header Length. */
428 uint64_t tso_segsz:16; /**< TCP TSO segment size */
430 /* fields for TX offloading of tunnels */
431 uint64_t outer_l3_len:9; /**< Outer L3 (IP) Hdr Length. */
432 uint64_t outer_l2_len:7; /**< Outer L2 (MAC) Hdr Length. */
434 /* uint64_t unused:8; */
438 /** Size of the application private data. In case of an indirect
439 * mbuf, it stores the direct mbuf private data size. */
442 /** Timesync flags for use with IEEE1588. */
444 } __rte_cache_aligned;
447 * Prefetch the first part of the mbuf
449 * The first 64 bytes of the mbuf corresponds to fields that are used early
450 * in the receive path. If the cache line of the architecture is higher than
451 * 64B, the second part will also be prefetched.
454 * The pointer to the mbuf.
457 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
459 rte_prefetch0(&m->cacheline0);
463 * Prefetch the second part of the mbuf
465 * The next 64 bytes of the mbuf corresponds to fields that are used in the
466 * transmit path. If the cache line of the architecture is higher than 64B,
467 * this function does nothing as it is expected that the full mbuf is
471 * The pointer to the mbuf.
474 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
476 #if RTE_CACHE_LINE_SIZE == 64
477 rte_prefetch0(&m->cacheline1);
484 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
487 * Return the DMA address of the beginning of the mbuf data
490 * The pointer to the mbuf.
492 * The physical address of the beginning of the mbuf data
494 static inline phys_addr_t
495 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
497 return mb->buf_physaddr + mb->data_off;
501 * Return the default DMA address of the beginning of the mbuf data
503 * This function is used by drivers in their receive function, as it
504 * returns the location where data should be written by the NIC, taking
505 * the default headroom in account.
508 * The pointer to the mbuf.
510 * The physical address of the beginning of the mbuf data
512 static inline phys_addr_t
513 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
515 return mb->buf_physaddr + RTE_PKTMBUF_HEADROOM;
519 * Return the mbuf owning the data buffer address of an indirect mbuf.
522 * The pointer to the indirect mbuf.
524 * The address of the direct mbuf corresponding to buffer_addr.
526 static inline struct rte_mbuf *
527 rte_mbuf_from_indirect(struct rte_mbuf *mi)
529 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
533 * Return the buffer address embedded in the given mbuf.
536 * The pointer to the mbuf.
538 * The address of the data buffer owned by the mbuf.
541 rte_mbuf_to_baddr(struct rte_mbuf *md)
544 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
549 * Returns TRUE if given mbuf is indirect, or FALSE otherwise.
551 #define RTE_MBUF_INDIRECT(mb) ((mb)->ol_flags & IND_ATTACHED_MBUF)
554 * Returns TRUE if given mbuf is direct, or FALSE otherwise.
556 #define RTE_MBUF_DIRECT(mb) (!RTE_MBUF_INDIRECT(mb))
559 * Private data in case of pktmbuf pool.
561 * A structure that contains some pktmbuf_pool-specific data that are
562 * appended after the mempool structure (in private data).
564 struct rte_pktmbuf_pool_private {
565 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
566 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
569 #ifdef RTE_LIBRTE_MBUF_DEBUG
571 /** check mbuf type in debug mode */
572 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
574 #else /* RTE_LIBRTE_MBUF_DEBUG */
576 /** check mbuf type in debug mode */
577 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
579 #endif /* RTE_LIBRTE_MBUF_DEBUG */
581 #ifdef RTE_MBUF_REFCNT_ATOMIC
584 * Reads the value of an mbuf's refcnt.
588 * Reference count number.
590 static inline uint16_t
591 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
593 return (uint16_t)(rte_atomic16_read(&m->refcnt_atomic));
597 * Sets an mbuf's refcnt to a defined value.
604 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
606 rte_atomic16_set(&m->refcnt_atomic, new_value);
610 * Adds given value to an mbuf's refcnt and returns its new value.
614 * Value to add/subtract
618 static inline uint16_t
619 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
622 * The atomic_add is an expensive operation, so we don't want to
623 * call it in the case where we know we are the uniq holder of
624 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
625 * operation has to be used because concurrent accesses on the
626 * reference counter can occur.
628 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
629 rte_mbuf_refcnt_set(m, 1 + value);
633 return (uint16_t)(rte_atomic16_add_return(&m->refcnt_atomic, value));
636 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
639 * Adds given value to an mbuf's refcnt and returns its new value.
641 static inline uint16_t
642 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
644 m->refcnt = (uint16_t)(m->refcnt + value);
649 * Reads the value of an mbuf's refcnt.
651 static inline uint16_t
652 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
658 * Sets an mbuf's refcnt to the defined value.
661 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
663 m->refcnt = new_value;
666 #endif /* RTE_MBUF_REFCNT_ATOMIC */
669 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
676 * Sanity checks on an mbuf.
678 * Check the consistency of the given mbuf. The function will cause a
679 * panic if corruption is detected.
682 * The mbuf to be checked.
684 * True if the mbuf is a packet header, false if it is a sub-segment
685 * of a packet (in this case, some fields like nb_segs are not checked)
688 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
691 * Allocate an unitialized mbuf from mempool *mp*.
693 * This function can be used by PMDs (especially in RX functions) to
694 * allocate an unitialized mbuf. The driver is responsible of
695 * initializing all the required fields. See rte_pktmbuf_reset().
696 * For standard needs, prefer rte_pktmbuf_alloc().
699 * The mempool from which mbuf is allocated.
701 * - The pointer to the new mbuf on success.
702 * - NULL if allocation failed.
704 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
709 if (rte_mempool_get(mp, &mb) < 0)
711 m = (struct rte_mbuf *)mb;
712 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 0);
713 rte_mbuf_refcnt_set(m, 1);
714 __rte_mbuf_sanity_check(m, 0);
720 * @internal Put mbuf back into its original mempool.
721 * The use of that function is reserved for RTE internal needs.
722 * Please use rte_pktmbuf_free().
725 * The mbuf to be freed.
727 static inline void __attribute__((always_inline))
728 __rte_mbuf_raw_free(struct rte_mbuf *m)
730 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 0);
731 rte_mempool_put(m->pool, m);
734 /* Operations on ctrl mbuf */
737 * The control mbuf constructor.
739 * This function initializes some fields in an mbuf structure that are
740 * not modified by the user once created (mbuf type, origin pool, buffer
741 * start address, and so on). This function is given as a callback function
742 * to rte_mempool_create() at pool creation time.
745 * The mempool from which the mbuf is allocated.
747 * A pointer that can be used by the user to retrieve useful information
748 * for mbuf initialization. This pointer comes from the ``init_arg``
749 * parameter of rte_mempool_create().
751 * The mbuf to initialize.
753 * The index of the mbuf in the pool table.
755 void rte_ctrlmbuf_init(struct rte_mempool *mp, void *opaque_arg,
756 void *m, unsigned i);
759 * Allocate a new mbuf (type is ctrl) from mempool *mp*.
761 * This new mbuf is initialized with data pointing to the beginning of
762 * buffer, and with a length of zero.
765 * The mempool from which the mbuf is allocated.
767 * - The pointer to the new mbuf on success.
768 * - NULL if allocation failed.
770 #define rte_ctrlmbuf_alloc(mp) rte_pktmbuf_alloc(mp)
773 * Free a control mbuf back into its original mempool.
776 * The control mbuf to be freed.
778 #define rte_ctrlmbuf_free(m) rte_pktmbuf_free(m)
781 * A macro that returns the pointer to the carried data.
783 * The value that can be read or assigned.
788 #define rte_ctrlmbuf_data(m) ((char *)((m)->buf_addr) + (m)->data_off)
791 * A macro that returns the length of the carried data.
793 * The value that can be read or assigned.
798 #define rte_ctrlmbuf_len(m) rte_pktmbuf_data_len(m)
801 * Tests if an mbuf is a control mbuf
804 * The mbuf to be tested
806 * - True (1) if the mbuf is a control mbuf
807 * - False(0) otherwise
810 rte_is_ctrlmbuf(struct rte_mbuf *m)
812 return !!(m->ol_flags & CTRL_MBUF_FLAG);
815 /* Operations on pkt mbuf */
818 * The packet mbuf constructor.
820 * This function initializes some fields in the mbuf structure that are
821 * not modified by the user once created (origin pool, buffer start
822 * address, and so on). This function is given as a callback function to
823 * rte_mempool_create() at pool creation time.
826 * The mempool from which mbufs originate.
828 * A pointer that can be used by the user to retrieve useful information
829 * for mbuf initialization. This pointer comes from the ``init_arg``
830 * parameter of rte_mempool_create().
832 * The mbuf to initialize.
834 * The index of the mbuf in the pool table.
836 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
837 void *m, unsigned i);
841 * A packet mbuf pool constructor.
843 * This function initializes the mempool private data in the case of a
844 * pktmbuf pool. This private data is needed by the driver. The
845 * function is given as a callback function to rte_mempool_create() at
846 * pool creation. It can be extended by the user, for example, to
847 * provide another packet size.
850 * The mempool from which mbufs originate.
852 * A pointer that can be used by the user to retrieve useful information
853 * for mbuf initialization. This pointer comes from the ``init_arg``
854 * parameter of rte_mempool_create().
856 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
859 * Create a mbuf pool.
861 * This function creates and initializes a packet mbuf pool. It is
862 * a wrapper to rte_mempool_create() with the proper packet constructor
863 * and mempool constructor.
866 * The name of the mbuf pool.
868 * The number of elements in the mbuf pool. The optimum size (in terms
869 * of memory usage) for a mempool is when n is a power of two minus one:
872 * Size of the per-core object cache. See rte_mempool_create() for
875 * Size of application private are between the rte_mbuf structure
876 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
877 * @param data_room_size
878 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
880 * The socket identifier where the memory should be allocated. The
881 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
884 * The pointer to the new allocated mempool, on success. NULL on error
885 * with rte_errno set appropriately. Possible rte_errno values include:
886 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
887 * - E_RTE_SECONDARY - function was called from a secondary process instance
888 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
889 * - ENOSPC - the maximum number of memzones has already been allocated
890 * - EEXIST - a memzone with the same name already exists
891 * - ENOMEM - no appropriate memory area found in which to create memzone
894 rte_pktmbuf_pool_create(const char *name, unsigned n,
895 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
899 * Get the data room size of mbufs stored in a pktmbuf_pool
901 * The data room size is the amount of data that can be stored in a
902 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
905 * The packet mbuf pool.
907 * The data room size of mbufs stored in this mempool.
909 static inline uint16_t
910 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
912 struct rte_pktmbuf_pool_private *mbp_priv;
914 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
915 return mbp_priv->mbuf_data_room_size;
919 * Get the application private size of mbufs stored in a pktmbuf_pool
921 * The private size of mbuf is a zone located between the rte_mbuf
922 * structure and the data buffer where an application can store data
923 * associated to a packet.
926 * The packet mbuf pool.
928 * The private size of mbufs stored in this mempool.
930 static inline uint16_t
931 rte_pktmbuf_priv_size(struct rte_mempool *mp)
933 struct rte_pktmbuf_pool_private *mbp_priv;
935 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
936 return mbp_priv->mbuf_priv_size;
940 * Reset the data_off field of a packet mbuf to its default value.
942 * The given mbuf must have only one segment, which should be empty.
945 * The packet mbuf's data_off field has to be reset.
947 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
949 m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, (uint16_t)m->buf_len);
953 * Reset the fields of a packet mbuf to their default values.
955 * The given mbuf must have only one segment.
958 * The packet mbuf to be resetted.
960 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
966 m->vlan_tci_outer = 0;
972 rte_pktmbuf_reset_headroom(m);
975 __rte_mbuf_sanity_check(m, 1);
979 * Allocate a new mbuf from a mempool.
981 * This new mbuf contains one segment, which has a length of 0. The pointer
982 * to data is initialized to have some bytes of headroom in the buffer
983 * (if buffer size allows).
986 * The mempool from which the mbuf is allocated.
988 * - The pointer to the new mbuf on success.
989 * - NULL if allocation failed.
991 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
994 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
995 rte_pktmbuf_reset(m);
1000 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
1004 * The mempool from which mbufs are allocated.
1006 * Array of pointers to mbufs
1012 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
1013 struct rte_mbuf **mbufs, unsigned count)
1018 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
1022 /* To understand duff's device on loop unwinding optimization, see
1023 * https://en.wikipedia.org/wiki/Duff's_device.
1024 * Here while() loop is used rather than do() while{} to avoid extra
1025 * check if count is zero.
1027 switch (count % 4) {
1029 while (idx != count) {
1030 RTE_ASSERT(rte_mbuf_refcnt_read(mbufs[idx]) == 0);
1031 rte_mbuf_refcnt_set(mbufs[idx], 1);
1032 rte_pktmbuf_reset(mbufs[idx]);
1035 RTE_ASSERT(rte_mbuf_refcnt_read(mbufs[idx]) == 0);
1036 rte_mbuf_refcnt_set(mbufs[idx], 1);
1037 rte_pktmbuf_reset(mbufs[idx]);
1040 RTE_ASSERT(rte_mbuf_refcnt_read(mbufs[idx]) == 0);
1041 rte_mbuf_refcnt_set(mbufs[idx], 1);
1042 rte_pktmbuf_reset(mbufs[idx]);
1045 RTE_ASSERT(rte_mbuf_refcnt_read(mbufs[idx]) == 0);
1046 rte_mbuf_refcnt_set(mbufs[idx], 1);
1047 rte_pktmbuf_reset(mbufs[idx]);
1055 * Attach packet mbuf to another packet mbuf.
1057 * After attachment we refer the mbuf we attached as 'indirect',
1058 * while mbuf we attached to as 'direct'.
1059 * The direct mbuf's reference counter is incremented.
1061 * Right now, not supported:
1062 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1063 * - mbuf we trying to attach (mi) is used by someone else
1064 * e.g. it's reference counter is greater then 1.
1067 * The indirect packet mbuf.
1069 * The packet mbuf we're attaching to.
1071 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1073 struct rte_mbuf *md;
1075 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1076 rte_mbuf_refcnt_read(mi) == 1);
1078 /* if m is not direct, get the mbuf that embeds the data */
1079 if (RTE_MBUF_DIRECT(m))
1082 md = rte_mbuf_from_indirect(m);
1084 rte_mbuf_refcnt_update(md, 1);
1085 mi->priv_size = m->priv_size;
1086 mi->buf_physaddr = m->buf_physaddr;
1087 mi->buf_addr = m->buf_addr;
1088 mi->buf_len = m->buf_len;
1091 mi->data_off = m->data_off;
1092 mi->data_len = m->data_len;
1094 mi->vlan_tci = m->vlan_tci;
1095 mi->vlan_tci_outer = m->vlan_tci_outer;
1096 mi->tx_offload = m->tx_offload;
1100 mi->pkt_len = mi->data_len;
1102 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1103 mi->packet_type = m->packet_type;
1105 __rte_mbuf_sanity_check(mi, 1);
1106 __rte_mbuf_sanity_check(m, 0);
1110 * Detach an indirect packet mbuf.
1112 * - restore original mbuf address and length values.
1113 * - reset pktmbuf data and data_len to their default values.
1114 * - decrement the direct mbuf's reference counter. When the
1115 * reference counter becomes 0, the direct mbuf is freed.
1117 * All other fields of the given packet mbuf will be left intact.
1120 * The indirect attached packet mbuf.
1122 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1124 struct rte_mbuf *md = rte_mbuf_from_indirect(m);
1125 struct rte_mempool *mp = m->pool;
1126 uint32_t mbuf_size, buf_len, priv_size;
1128 priv_size = rte_pktmbuf_priv_size(mp);
1129 mbuf_size = sizeof(struct rte_mbuf) + priv_size;
1130 buf_len = rte_pktmbuf_data_room_size(mp);
1132 m->priv_size = priv_size;
1133 m->buf_addr = (char *)m + mbuf_size;
1134 m->buf_physaddr = rte_mempool_virt2phy(mp, m) + mbuf_size;
1135 m->buf_len = (uint16_t)buf_len;
1136 rte_pktmbuf_reset_headroom(m);
1140 if (rte_mbuf_refcnt_update(md, -1) == 0)
1141 __rte_mbuf_raw_free(md);
1144 static inline struct rte_mbuf* __attribute__((always_inline))
1145 __rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1147 __rte_mbuf_sanity_check(m, 0);
1149 if (likely(rte_mbuf_refcnt_update(m, -1) == 0)) {
1150 /* if this is an indirect mbuf, it is detached. */
1151 if (RTE_MBUF_INDIRECT(m))
1152 rte_pktmbuf_detach(m);
1159 * Free a segment of a packet mbuf into its original mempool.
1161 * Free an mbuf, without parsing other segments in case of chained
1165 * The packet mbuf segment to be freed.
1167 static inline void __attribute__((always_inline))
1168 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1170 if (likely(NULL != (m = __rte_pktmbuf_prefree_seg(m)))) {
1172 __rte_mbuf_raw_free(m);
1177 * Free a packet mbuf back into its original mempool.
1179 * Free an mbuf, and all its segments in case of chained buffers. Each
1180 * segment is added back into its original mempool.
1183 * The packet mbuf to be freed.
1185 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1187 struct rte_mbuf *m_next;
1189 __rte_mbuf_sanity_check(m, 1);
1193 rte_pktmbuf_free_seg(m);
1199 * Creates a "clone" of the given packet mbuf.
1201 * Walks through all segments of the given packet mbuf, and for each of them:
1202 * - Creates a new packet mbuf from the given pool.
1203 * - Attaches newly created mbuf to the segment.
1204 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1205 * from the original packet mbuf.
1208 * The packet mbuf to be cloned.
1210 * The mempool from which the "clone" mbufs are allocated.
1212 * - The pointer to the new "clone" mbuf on success.
1213 * - NULL if allocation fails.
1215 static inline struct rte_mbuf *rte_pktmbuf_clone(struct rte_mbuf *md,
1216 struct rte_mempool *mp)
1218 struct rte_mbuf *mc, *mi, **prev;
1222 if (unlikely ((mc = rte_pktmbuf_alloc(mp)) == NULL))
1227 pktlen = md->pkt_len;
1232 rte_pktmbuf_attach(mi, md);
1235 } while ((md = md->next) != NULL &&
1236 (mi = rte_pktmbuf_alloc(mp)) != NULL);
1240 mc->pkt_len = pktlen;
1242 /* Allocation of new indirect segment failed */
1243 if (unlikely (mi == NULL)) {
1244 rte_pktmbuf_free(mc);
1248 __rte_mbuf_sanity_check(mc, 1);
1253 * Adds given value to the refcnt of all packet mbuf segments.
1255 * Walks through all segments of given packet mbuf and for each of them
1256 * invokes rte_mbuf_refcnt_update().
1259 * The packet mbuf whose refcnt to be updated.
1261 * The value to add to the mbuf's segments refcnt.
1263 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1265 __rte_mbuf_sanity_check(m, 1);
1268 rte_mbuf_refcnt_update(m, v);
1269 } while ((m = m->next) != NULL);
1273 * Get the headroom in a packet mbuf.
1278 * The length of the headroom.
1280 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1282 __rte_mbuf_sanity_check(m, 1);
1287 * Get the tailroom of a packet mbuf.
1292 * The length of the tailroom.
1294 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1296 __rte_mbuf_sanity_check(m, 1);
1297 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1302 * Get the last segment of the packet.
1307 * The last segment of the given mbuf.
1309 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1311 struct rte_mbuf *m2 = (struct rte_mbuf *)m;
1313 __rte_mbuf_sanity_check(m, 1);
1314 while (m2->next != NULL)
1320 * A macro that points to an offset into the data in the mbuf.
1322 * The returned pointer is cast to type t. Before using this
1323 * function, the user must ensure that the first segment is large
1324 * enough to accommodate its data.
1329 * The offset into the mbuf data.
1331 * The type to cast the result into.
1333 #define rte_pktmbuf_mtod_offset(m, t, o) \
1334 ((t)((char *)(m)->buf_addr + (m)->data_off + (o)))
1337 * A macro that points to the start of the data in the mbuf.
1339 * The returned pointer is cast to type t. Before using this
1340 * function, the user must ensure that the first segment is large
1341 * enough to accommodate its data.
1346 * The type to cast the result into.
1348 #define rte_pktmbuf_mtod(m, t) rte_pktmbuf_mtod_offset(m, t, 0)
1351 * A macro that returns the physical address that points to an offset of the
1352 * start of the data in the mbuf
1357 * The offset into the data to calculate address from.
1359 #define rte_pktmbuf_mtophys_offset(m, o) \
1360 (phys_addr_t)((m)->buf_physaddr + (m)->data_off + (o))
1363 * A macro that returns the physical address that points to the start of the
1369 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_mtophys_offset(m, 0)
1372 * A macro that returns the length of the packet.
1374 * The value can be read or assigned.
1379 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1382 * A macro that returns the length of the segment.
1384 * The value can be read or assigned.
1389 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1392 * Prepend len bytes to an mbuf data area.
1394 * Returns a pointer to the new
1395 * data start address. If there is not enough headroom in the first
1396 * segment, the function will return NULL, without modifying the mbuf.
1401 * The amount of data to prepend (in bytes).
1403 * A pointer to the start of the newly prepended data, or
1404 * NULL if there is not enough headroom space in the first segment
1406 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1409 __rte_mbuf_sanity_check(m, 1);
1411 if (unlikely(len > rte_pktmbuf_headroom(m)))
1415 m->data_len = (uint16_t)(m->data_len + len);
1416 m->pkt_len = (m->pkt_len + len);
1418 return (char *)m->buf_addr + m->data_off;
1422 * Append len bytes to an mbuf.
1424 * Append len bytes to an mbuf and return a pointer to the start address
1425 * of the added data. If there is not enough tailroom in the last
1426 * segment, the function will return NULL, without modifying the mbuf.
1431 * The amount of data to append (in bytes).
1433 * A pointer to the start of the newly appended data, or
1434 * NULL if there is not enough tailroom space in the last segment
1436 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1439 struct rte_mbuf *m_last;
1441 __rte_mbuf_sanity_check(m, 1);
1443 m_last = rte_pktmbuf_lastseg(m);
1444 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1447 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1448 m_last->data_len = (uint16_t)(m_last->data_len + len);
1449 m->pkt_len = (m->pkt_len + len);
1450 return (char*) tail;
1454 * Remove len bytes at the beginning of an mbuf.
1456 * Returns a pointer to the start address of the new data area. If the
1457 * length is greater than the length of the first segment, then the
1458 * function will fail and return NULL, without modifying the mbuf.
1463 * The amount of data to remove (in bytes).
1465 * A pointer to the new start of the data.
1467 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1469 __rte_mbuf_sanity_check(m, 1);
1471 if (unlikely(len > m->data_len))
1474 m->data_len = (uint16_t)(m->data_len - len);
1476 m->pkt_len = (m->pkt_len - len);
1477 return (char *)m->buf_addr + m->data_off;
1481 * Remove len bytes of data at the end of the mbuf.
1483 * If the length is greater than the length of the last segment, the
1484 * function will fail and return -1 without modifying the mbuf.
1489 * The amount of data to remove (in bytes).
1494 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
1496 struct rte_mbuf *m_last;
1498 __rte_mbuf_sanity_check(m, 1);
1500 m_last = rte_pktmbuf_lastseg(m);
1501 if (unlikely(len > m_last->data_len))
1504 m_last->data_len = (uint16_t)(m_last->data_len - len);
1505 m->pkt_len = (m->pkt_len - len);
1510 * Test if mbuf data is contiguous.
1515 * - 1, if all data is contiguous (one segment).
1516 * - 0, if there is several segments.
1518 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
1520 __rte_mbuf_sanity_check(m, 1);
1521 return !!(m->nb_segs == 1);
1525 * @internal used by rte_pktmbuf_read().
1527 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
1528 uint32_t len, void *buf);
1531 * Read len data bytes in a mbuf at specified offset.
1533 * If the data is contiguous, return the pointer in the mbuf data, else
1534 * copy the data in the buffer provided by the user and return its
1538 * The pointer to the mbuf.
1540 * The offset of the data in the mbuf.
1542 * The amount of bytes to read.
1544 * The buffer where data is copied if it is not contigous in mbuf
1545 * data. Its length should be at least equal to the len parameter.
1547 * The pointer to the data, either in the mbuf if it is contiguous,
1548 * or in the user buffer. If mbuf is too small, NULL is returned.
1550 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
1551 uint32_t off, uint32_t len, void *buf)
1553 if (likely(off + len <= rte_pktmbuf_data_len(m)))
1554 return rte_pktmbuf_mtod_offset(m, char *, off);
1556 return __rte_pktmbuf_read(m, off, len, buf);
1560 * Chain an mbuf to another, thereby creating a segmented packet.
1562 * Note: The implementation will do a linear walk over the segments to find
1563 * the tail entry. For cases when there are many segments, it's better to
1564 * chain the entries manually.
1567 * The head of the mbuf chain (the first packet)
1569 * The mbuf to put last in the chain
1573 * - -EOVERFLOW, if the chain is full (256 entries)
1575 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
1577 struct rte_mbuf *cur_tail;
1579 /* Check for number-of-segments-overflow */
1580 if (head->nb_segs + tail->nb_segs >= 1 << (sizeof(head->nb_segs) * 8))
1583 /* Chain 'tail' onto the old tail */
1584 cur_tail = rte_pktmbuf_lastseg(head);
1585 cur_tail->next = tail;
1587 /* accumulate number of segments and total length. */
1588 head->nb_segs = (uint8_t)(head->nb_segs + tail->nb_segs);
1589 head->pkt_len += tail->pkt_len;
1591 /* pkt_len is only set in the head */
1592 tail->pkt_len = tail->data_len;
1598 * Dump an mbuf structure to the console.
1600 * Dump all fields for the given packet mbuf and all its associated
1601 * segments (in the case of a chained buffer).
1604 * A pointer to a file for output
1608 * If dump_len != 0, also dump the "dump_len" first data bytes of
1611 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
1617 #endif /* _RTE_MBUF_H_ */