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_prefetch.h>
41 #include <rte_branch_prediction.h>
42 #include <rte_byteorder.h>
43 #include <rte_mbuf_ptype.h>
44 #include <rte_mbuf_core.h>
51 * Get the name of a RX offload flag
54 * The mask describing the flag.
56 * The name of this flag, or NULL if it's not a valid RX flag.
58 const char *rte_get_rx_ol_flag_name(uint64_t mask);
61 * Dump the list of RX offload flags in a buffer
64 * The mask describing the RX flags.
68 * The length of the buffer.
70 * 0 on success, (-1) on error.
72 int rte_get_rx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
75 * Get the name of a TX offload flag
78 * The mask describing the flag. Usually only one bit must be set.
79 * Several bits can be given if they belong to the same mask.
82 * The name of this flag, or NULL if it's not a valid TX flag.
84 const char *rte_get_tx_ol_flag_name(uint64_t mask);
87 * Dump the list of TX offload flags in a buffer
90 * The mask describing the TX flags.
94 * The length of the buffer.
96 * 0 on success, (-1) on error.
98 int rte_get_tx_ol_flag_list(uint64_t mask, char *buf, size_t buflen);
101 * Prefetch the first part of the mbuf
103 * The first 64 bytes of the mbuf corresponds to fields that are used early
104 * in the receive path. If the cache line of the architecture is higher than
105 * 64B, the second part will also be prefetched.
108 * The pointer to the mbuf.
111 rte_mbuf_prefetch_part1(struct rte_mbuf *m)
113 rte_prefetch0(&m->cacheline0);
117 * Prefetch the second part of the mbuf
119 * The next 64 bytes of the mbuf corresponds to fields that are used in the
120 * transmit path. If the cache line of the architecture is higher than 64B,
121 * this function does nothing as it is expected that the full mbuf is
125 * The pointer to the mbuf.
128 rte_mbuf_prefetch_part2(struct rte_mbuf *m)
130 #if RTE_CACHE_LINE_SIZE == 64
131 rte_prefetch0(&m->cacheline1);
138 static inline uint16_t rte_pktmbuf_priv_size(struct rte_mempool *mp);
141 * Return the IO address of the beginning of the mbuf data
144 * The pointer to the mbuf.
146 * The IO address of the beginning of the mbuf data
148 static inline rte_iova_t
149 rte_mbuf_data_iova(const struct rte_mbuf *mb)
151 return mb->buf_iova + mb->data_off;
155 * Return the default IO address of the beginning of the mbuf data
157 * This function is used by drivers in their receive function, as it
158 * returns the location where data should be written by the NIC, taking
159 * the default headroom in account.
162 * The pointer to the mbuf.
164 * The IO address of the beginning of the mbuf data
166 static inline rte_iova_t
167 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
169 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
173 * Return the mbuf owning the data buffer address of an indirect mbuf.
176 * The pointer to the indirect mbuf.
178 * The address of the direct mbuf corresponding to buffer_addr.
180 static inline struct rte_mbuf *
181 rte_mbuf_from_indirect(struct rte_mbuf *mi)
183 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
187 * Return address of buffer embedded in the given mbuf.
189 * The return value shall be same as mb->buf_addr if the mbuf is already
190 * initialized and direct. However, this API is useful if mempool of the
191 * mbuf is already known because it doesn't need to access mbuf contents in
192 * order to get the mempool pointer.
195 * The pointer to the mbuf.
197 * The pointer to the mempool of the mbuf.
199 * The pointer of the mbuf buffer.
202 rte_mbuf_buf_addr(struct rte_mbuf *mb, struct rte_mempool *mp)
204 return (char *)mb + sizeof(*mb) + rte_pktmbuf_priv_size(mp);
208 * Return the default address of the beginning of the mbuf data.
211 * The pointer to the mbuf.
213 * The pointer of the beginning of the mbuf data.
216 rte_mbuf_data_addr_default(struct rte_mbuf *mb)
218 return rte_mbuf_buf_addr(mb, mb->pool) + RTE_PKTMBUF_HEADROOM;
222 * Return address of buffer embedded in the given mbuf.
224 * @note: Accessing mempool pointer of a mbuf is expensive because the
225 * pointer is stored in the 2nd cache line of mbuf. If mempool is known, it
226 * is better not to reference the mempool pointer in mbuf but calling
227 * rte_mbuf_buf_addr() would be more efficient.
230 * The pointer to the mbuf.
232 * The address of the data buffer owned by the mbuf.
235 rte_mbuf_to_baddr(struct rte_mbuf *md)
237 return rte_mbuf_buf_addr(md, md->pool);
241 * Return the starting address of the private data area embedded in
244 * Note that no check is made to ensure that a private data area
245 * actually exists in the supplied mbuf.
248 * The pointer to the mbuf.
250 * The starting address of the private data area of the given mbuf.
253 rte_mbuf_to_priv(struct rte_mbuf *m)
255 return RTE_PTR_ADD(m, sizeof(struct rte_mbuf));
259 * Private data in case of pktmbuf pool.
261 * A structure that contains some pktmbuf_pool-specific data that are
262 * appended after the mempool structure (in private data).
264 struct rte_pktmbuf_pool_private {
265 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
266 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
267 uint32_t flags; /**< reserved for future use. */
271 * Return the flags from private data in an mempool structure.
274 * A pointer to the mempool structure.
276 * The flags from the private data structure.
278 static inline uint32_t
279 rte_pktmbuf_priv_flags(struct rte_mempool *mp)
281 struct rte_pktmbuf_pool_private *mbp_priv;
283 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
284 return mbp_priv->flags;
288 * When set, pktmbuf mempool will hold only mbufs with pinned external
289 * buffer. The external buffer will be attached to the mbuf at the
290 * memory pool creation and will never be detached by the mbuf free calls.
291 * mbuf should not contain any room for data after the mbuf structure.
293 #define RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF (1 << 0)
296 * Returns non zero if given mbuf has a pinned external buffer, or zero
297 * otherwise. The pinned external buffer is allocated at pool creation
298 * time and should not be freed on mbuf freeing.
300 * External buffer is a user-provided anonymous buffer.
302 #define RTE_MBUF_HAS_PINNED_EXTBUF(mb) \
303 (rte_pktmbuf_priv_flags(mb->pool) & RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF)
305 #ifdef RTE_LIBRTE_MBUF_DEBUG
307 /** check mbuf type in debug mode */
308 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
310 #else /* RTE_LIBRTE_MBUF_DEBUG */
312 /** check mbuf type in debug mode */
313 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
315 #endif /* RTE_LIBRTE_MBUF_DEBUG */
317 #ifdef RTE_MBUF_REFCNT_ATOMIC
320 * Reads the value of an mbuf's refcnt.
324 * Reference count number.
326 static inline uint16_t
327 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
329 return __atomic_load_n(&m->refcnt, __ATOMIC_RELAXED);
333 * Sets an mbuf's refcnt to a defined value.
340 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
342 __atomic_store_n(&m->refcnt, new_value, __ATOMIC_RELAXED);
346 static inline uint16_t
347 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
349 return __atomic_add_fetch(&m->refcnt, (uint16_t)value,
354 * Adds given value to an mbuf's refcnt and returns its new value.
358 * Value to add/subtract
362 static inline uint16_t
363 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
366 * The atomic_add is an expensive operation, so we don't want to
367 * call it in the case where we know we are the unique holder of
368 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
369 * operation has to be used because concurrent accesses on the
370 * reference counter can occur.
372 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
374 rte_mbuf_refcnt_set(m, (uint16_t)value);
375 return (uint16_t)value;
378 return __rte_mbuf_refcnt_update(m, value);
381 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
384 static inline uint16_t
385 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
387 m->refcnt = (uint16_t)(m->refcnt + value);
392 * Adds given value to an mbuf's refcnt and returns its new value.
394 static inline uint16_t
395 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
397 return __rte_mbuf_refcnt_update(m, value);
401 * Reads the value of an mbuf's refcnt.
403 static inline uint16_t
404 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
410 * Sets an mbuf's refcnt to the defined value.
413 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
415 m->refcnt = new_value;
418 #endif /* RTE_MBUF_REFCNT_ATOMIC */
421 * Reads the refcnt of an external buffer.
424 * Shared data of the external buffer.
426 * Reference count number.
428 static inline uint16_t
429 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
431 return __atomic_load_n(&shinfo->refcnt, __ATOMIC_RELAXED);
435 * Set refcnt of an external buffer.
438 * Shared data of the external buffer.
443 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
446 __atomic_store_n(&shinfo->refcnt, new_value, __ATOMIC_RELAXED);
450 * Add given value to refcnt of an external buffer and return its new
454 * Shared data of the external buffer.
456 * Value to add/subtract
460 static inline uint16_t
461 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
464 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
466 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
467 return (uint16_t)value;
470 return __atomic_add_fetch(&shinfo->refcnt, (uint16_t)value,
475 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
482 * Sanity checks on an mbuf.
484 * Check the consistency of the given mbuf. The function will cause a
485 * panic if corruption is detected.
488 * The mbuf to be checked.
490 * True if the mbuf is a packet header, false if it is a sub-segment
491 * of a packet (in this case, some fields like nb_segs are not checked)
494 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
497 * Sanity checks on a mbuf.
499 * Almost like rte_mbuf_sanity_check(), but this function gives the reason
500 * if corruption is detected rather than panic.
503 * The mbuf to be checked.
505 * True if the mbuf is a packet header, false if it is a sub-segment
506 * of a packet (in this case, some fields like nb_segs are not checked)
508 * A reference to a string pointer where to store the reason why a mbuf is
509 * considered invalid.
511 * - 0 if no issue has been found, reason is left untouched.
512 * - -1 if a problem is detected, reason then points to a string describing
513 * the reason why the mbuf is deemed invalid.
516 int rte_mbuf_check(const struct rte_mbuf *m, int is_header,
517 const char **reason);
520 * Sanity checks on a reinitialized mbuf in debug mode.
522 * Check the consistency of the given reinitialized mbuf.
523 * The function will cause a panic if corruption is detected.
525 * Check that the mbuf is properly reinitialized (refcnt=1, next=NULL,
526 * nb_segs=1), as done by rte_pktmbuf_prefree_seg().
529 * The mbuf to be checked.
531 static __rte_always_inline void
532 __rte_mbuf_raw_sanity_check(__rte_unused const struct rte_mbuf *m)
534 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
535 RTE_ASSERT(m->next == NULL);
536 RTE_ASSERT(m->nb_segs == 1);
537 __rte_mbuf_sanity_check(m, 0);
540 /** For backwards compatibility. */
541 #define MBUF_RAW_ALLOC_CHECK(m) __rte_mbuf_raw_sanity_check(m)
544 * Allocate an uninitialized mbuf from mempool *mp*.
546 * This function can be used by PMDs (especially in RX functions) to
547 * allocate an uninitialized mbuf. The driver is responsible of
548 * initializing all the required fields. See rte_pktmbuf_reset().
549 * For standard needs, prefer rte_pktmbuf_alloc().
551 * The caller can expect that the following fields of the mbuf structure
552 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
553 * next=NULL, pool, priv_size. The other fields must be initialized
557 * The mempool from which mbuf is allocated.
559 * - The pointer to the new mbuf on success.
560 * - NULL if allocation failed.
562 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
566 if (rte_mempool_get(mp, (void **)&m) < 0)
568 __rte_mbuf_raw_sanity_check(m);
573 * Put mbuf back into its original mempool.
575 * The caller must ensure that the mbuf is direct and properly
576 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
577 * rte_pktmbuf_prefree_seg().
579 * This function should be used with care, when optimization is
580 * required. For standard needs, prefer rte_pktmbuf_free() or
581 * rte_pktmbuf_free_seg().
584 * The mbuf to be freed.
586 static __rte_always_inline void
587 rte_mbuf_raw_free(struct rte_mbuf *m)
589 RTE_ASSERT(!RTE_MBUF_CLONED(m) &&
590 (!RTE_MBUF_HAS_EXTBUF(m) || RTE_MBUF_HAS_PINNED_EXTBUF(m)));
591 __rte_mbuf_raw_sanity_check(m);
592 rte_mempool_put(m->pool, m);
596 * The packet mbuf constructor.
598 * This function initializes some fields in the mbuf structure that are
599 * not modified by the user once created (origin pool, buffer start
600 * address, and so on). This function is given as a callback function to
601 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
603 * This function expects that the mempool private area was previously
604 * initialized with rte_pktmbuf_pool_init().
607 * The mempool from which mbufs originate.
609 * A pointer that can be used by the user to retrieve useful information
610 * for mbuf initialization. This pointer is the opaque argument passed to
611 * rte_mempool_obj_iter() or rte_mempool_create().
613 * The mbuf to initialize.
615 * The index of the mbuf in the pool table.
617 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
618 void *m, unsigned i);
621 * A packet mbuf pool constructor.
623 * This function initializes the mempool private data in the case of a
624 * pktmbuf pool. This private data is needed by the driver. The
625 * function must be called on the mempool before it is used, or it
626 * can be given as a callback function to rte_mempool_create() at
627 * pool creation. It can be extended by the user, for example, to
628 * provide another packet size.
630 * The mempool private area size must be at least equal to
631 * sizeof(struct rte_pktmbuf_pool_private).
634 * The mempool from which mbufs originate.
636 * A pointer that can be used by the user to retrieve useful information
637 * for mbuf initialization. This pointer is the opaque argument passed to
638 * rte_mempool_create().
640 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
643 * Create a mbuf pool.
645 * This function creates and initializes a packet mbuf pool. It is
646 * a wrapper to rte_mempool functions.
649 * The name of the mbuf pool.
651 * The number of elements in the mbuf pool. The optimum size (in terms
652 * of memory usage) for a mempool is when n is a power of two minus one:
655 * Size of the per-core object cache. See rte_mempool_create() for
658 * Size of application private are between the rte_mbuf structure
659 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
660 * @param data_room_size
661 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
663 * The socket identifier where the memory should be allocated. The
664 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
667 * The pointer to the new allocated mempool, on success. NULL on error
668 * with rte_errno set appropriately. Possible rte_errno values include:
669 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
670 * - E_RTE_SECONDARY - function was called from a secondary process instance
671 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
672 * - ENOSPC - the maximum number of memzones has already been allocated
673 * - EEXIST - a memzone with the same name already exists
674 * - ENOMEM - no appropriate memory area found in which to create memzone
677 rte_pktmbuf_pool_create(const char *name, unsigned n,
678 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
682 * Create a mbuf pool with a given mempool ops name
684 * This function creates and initializes a packet mbuf pool. It is
685 * a wrapper to rte_mempool functions.
688 * The name of the mbuf pool.
690 * The number of elements in the mbuf pool. The optimum size (in terms
691 * of memory usage) for a mempool is when n is a power of two minus one:
694 * Size of the per-core object cache. See rte_mempool_create() for
697 * Size of application private are between the rte_mbuf structure
698 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
699 * @param data_room_size
700 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
702 * The socket identifier where the memory should be allocated. The
703 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
706 * The mempool ops name to be used for this mempool instead of
707 * default mempool. The value can be *NULL* to use default mempool.
709 * The pointer to the new allocated mempool, on success. NULL on error
710 * with rte_errno set appropriately. Possible rte_errno values include:
711 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
712 * - E_RTE_SECONDARY - function was called from a secondary process instance
713 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
714 * - ENOSPC - the maximum number of memzones has already been allocated
715 * - EEXIST - a memzone with the same name already exists
716 * - ENOMEM - no appropriate memory area found in which to create memzone
719 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
720 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
721 int socket_id, const char *ops_name);
723 /** A structure that describes the pinned external buffer segment. */
724 struct rte_pktmbuf_extmem {
725 void *buf_ptr; /**< The virtual address of data buffer. */
726 rte_iova_t buf_iova; /**< The IO address of the data buffer. */
727 size_t buf_len; /**< External buffer length in bytes. */
728 uint16_t elt_size; /**< mbuf element size in bytes. */
732 * Create a mbuf pool with external pinned data buffers.
734 * This function creates and initializes a packet mbuf pool that contains
735 * only mbufs with external buffer. It is a wrapper to rte_mempool functions.
738 * The name of the mbuf pool.
740 * The number of elements in the mbuf pool. The optimum size (in terms
741 * of memory usage) for a mempool is when n is a power of two minus one:
744 * Size of the per-core object cache. See rte_mempool_create() for
747 * Size of application private are between the rte_mbuf structure
748 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
749 * @param data_room_size
750 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
752 * The socket identifier where the memory should be allocated. The
753 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
756 * Pointer to the array of structures describing the external memory
757 * for data buffers. It is caller responsibility to register this memory
758 * with rte_extmem_register() (if needed), map this memory to appropriate
759 * physical device, etc.
761 * Number of elements in the ext_mem array.
763 * The pointer to the new allocated mempool, on success. NULL on error
764 * with rte_errno set appropriately. Possible rte_errno values include:
765 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
766 * - E_RTE_SECONDARY - function was called from a secondary process instance
767 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
768 * - ENOSPC - the maximum number of memzones has already been allocated
769 * - EEXIST - a memzone with the same name already exists
770 * - ENOMEM - no appropriate memory area found in which to create memzone
774 rte_pktmbuf_pool_create_extbuf(const char *name, unsigned int n,
775 unsigned int cache_size, uint16_t priv_size,
776 uint16_t data_room_size, int socket_id,
777 const struct rte_pktmbuf_extmem *ext_mem,
778 unsigned int ext_num);
781 * Get the data room size of mbufs stored in a pktmbuf_pool
783 * The data room size is the amount of data that can be stored in a
784 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
787 * The packet mbuf pool.
789 * The data room size of mbufs stored in this mempool.
791 static inline uint16_t
792 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
794 struct rte_pktmbuf_pool_private *mbp_priv;
796 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
797 return mbp_priv->mbuf_data_room_size;
801 * Get the application private size of mbufs stored in a pktmbuf_pool
803 * The private size of mbuf is a zone located between the rte_mbuf
804 * structure and the data buffer where an application can store data
805 * associated to a packet.
808 * The packet mbuf pool.
810 * The private size of mbufs stored in this mempool.
812 static inline uint16_t
813 rte_pktmbuf_priv_size(struct rte_mempool *mp)
815 struct rte_pktmbuf_pool_private *mbp_priv;
817 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
818 return mbp_priv->mbuf_priv_size;
822 * Reset the data_off field of a packet mbuf to its default value.
824 * The given mbuf must have only one segment, which should be empty.
827 * The packet mbuf's data_off field has to be reset.
829 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
831 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
832 (uint16_t)m->buf_len);
836 * Reset the fields of a packet mbuf to their default values.
838 * The given mbuf must have only one segment.
841 * The packet mbuf to be reset.
843 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
849 m->vlan_tci_outer = 0;
851 m->port = RTE_MBUF_PORT_INVALID;
853 m->ol_flags &= EXT_ATTACHED_MBUF;
855 rte_pktmbuf_reset_headroom(m);
858 __rte_mbuf_sanity_check(m, 1);
862 * Allocate a new mbuf from a mempool.
864 * This new mbuf contains one segment, which has a length of 0. The pointer
865 * to data is initialized to have some bytes of headroom in the buffer
866 * (if buffer size allows).
869 * The mempool from which the mbuf is allocated.
871 * - The pointer to the new mbuf on success.
872 * - NULL if allocation failed.
874 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
877 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
878 rte_pktmbuf_reset(m);
883 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
887 * The mempool from which mbufs are allocated.
889 * Array of pointers to mbufs
894 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
896 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
897 struct rte_mbuf **mbufs, unsigned count)
902 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
906 /* To understand duff's device on loop unwinding optimization, see
907 * https://en.wikipedia.org/wiki/Duff's_device.
908 * Here while() loop is used rather than do() while{} to avoid extra
909 * check if count is zero.
913 while (idx != count) {
914 __rte_mbuf_raw_sanity_check(mbufs[idx]);
915 rte_pktmbuf_reset(mbufs[idx]);
919 __rte_mbuf_raw_sanity_check(mbufs[idx]);
920 rte_pktmbuf_reset(mbufs[idx]);
924 __rte_mbuf_raw_sanity_check(mbufs[idx]);
925 rte_pktmbuf_reset(mbufs[idx]);
929 __rte_mbuf_raw_sanity_check(mbufs[idx]);
930 rte_pktmbuf_reset(mbufs[idx]);
939 * Initialize shared data at the end of an external buffer before attaching
940 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
941 * initialization but a helper function to simply spare a few bytes at the
942 * end of the buffer for shared data. If shared data is allocated
943 * separately, this should not be called but application has to properly
944 * initialize the shared data according to its need.
946 * Free callback and its argument is saved and the refcnt is set to 1.
949 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
950 * after this initialization. This shall be used for
951 * ``rte_pktmbuf_attach_extbuf()``
954 * The pointer to the external buffer.
955 * @param [in,out] buf_len
956 * The pointer to length of the external buffer. Input value must be
957 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
958 * padding for alignment. If not enough, this function will return NULL.
959 * Adjusted buffer length will be returned through this pointer.
961 * Free callback function to call when the external buffer needs to be
964 * Argument for the free callback function.
967 * A pointer to the initialized shared data on success, return NULL
970 static inline struct rte_mbuf_ext_shared_info *
971 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
972 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
974 struct rte_mbuf_ext_shared_info *shinfo;
975 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
978 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
980 if (addr <= buf_addr)
983 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
984 shinfo->free_cb = free_cb;
985 shinfo->fcb_opaque = fcb_opaque;
986 rte_mbuf_ext_refcnt_set(shinfo, 1);
988 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
993 * Attach an external buffer to a mbuf.
995 * User-managed anonymous buffer can be attached to an mbuf. When attaching
996 * it, corresponding free callback function and its argument should be
997 * provided via shinfo. This callback function will be called once all the
998 * mbufs are detached from the buffer (refcnt becomes zero).
1000 * The headroom length of the attaching mbuf will be set to zero and this
1001 * can be properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1002 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1004 * Similarly, the packet length is initialized to 0. If the buffer contains
1005 * data, the user has to adjust ``data_len`` and the ``pkt_len`` field of
1006 * the mbuf accordingly.
1008 * More mbufs can be attached to the same external buffer by
1009 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1012 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1013 * ``rte_pktmbuf_detach()``.
1015 * Memory for shared data must be provided and user must initialize all of
1016 * the content properly, especially free callback and refcnt. The pointer
1017 * of shared data will be stored in m->shinfo.
1018 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1019 * bytes at the end of buffer for the shared data, store free callback and
1020 * its argument and set the refcnt to 1. The following is an example:
1022 * struct rte_mbuf_ext_shared_info *shinfo =
1023 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1024 * free_cb, fcb_arg);
1025 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1026 * rte_pktmbuf_reset_headroom(m);
1027 * rte_pktmbuf_adj(m, data_len);
1029 * Attaching an external buffer is quite similar to mbuf indirection in
1030 * replacing buffer addresses and length of a mbuf, but a few differences:
1031 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1032 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1033 * In such cases, the buffer area of a direct mbuf must be read-only. But
1034 * external buffer has its own refcnt and it starts from 1. Unless
1035 * multiple mbufs are attached to a mbuf having an external buffer, the
1036 * external buffer is writable.
1037 * - There's no need to allocate buffer from a mempool. Any buffer can be
1038 * attached with appropriate free callback and its IO address.
1039 * - Smaller metadata is required to maintain shared data such as refcnt.
1042 * The pointer to the mbuf.
1044 * The pointer to the external buffer.
1046 * IO address of the external buffer.
1048 * The size of the external buffer.
1050 * User-provided memory for shared data of the external buffer.
1053 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1054 rte_iova_t buf_iova, uint16_t buf_len,
1055 struct rte_mbuf_ext_shared_info *shinfo)
1057 /* mbuf should not be read-only */
1058 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1059 RTE_ASSERT(shinfo->free_cb != NULL);
1061 m->buf_addr = buf_addr;
1062 m->buf_iova = buf_iova;
1063 m->buf_len = buf_len;
1068 m->ol_flags |= EXT_ATTACHED_MBUF;
1073 * Detach the external buffer attached to a mbuf, same as
1074 * ``rte_pktmbuf_detach()``
1077 * The mbuf having external buffer.
1079 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1082 * Copy dynamic fields from msrc to mdst.
1085 * The destination mbuf.
1090 rte_mbuf_dynfield_copy(struct rte_mbuf *mdst, const struct rte_mbuf *msrc)
1092 memcpy(&mdst->dynfield1, msrc->dynfield1, sizeof(mdst->dynfield1));
1097 __rte_pktmbuf_copy_hdr(struct rte_mbuf *mdst, const struct rte_mbuf *msrc)
1099 mdst->port = msrc->port;
1100 mdst->vlan_tci = msrc->vlan_tci;
1101 mdst->vlan_tci_outer = msrc->vlan_tci_outer;
1102 mdst->tx_offload = msrc->tx_offload;
1103 mdst->hash = msrc->hash;
1104 mdst->packet_type = msrc->packet_type;
1105 rte_mbuf_dynfield_copy(mdst, msrc);
1109 * Attach packet mbuf to another packet mbuf.
1111 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1112 * an external buffer, the mbuf being attached will be attached to the
1113 * external buffer instead of mbuf indirection.
1115 * Otherwise, the mbuf will be indirectly attached. After attachment we
1116 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1117 * 'direct'. The direct mbuf's reference counter is incremented.
1119 * Right now, not supported:
1120 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1121 * - mbuf we trying to attach (mi) is used by someone else
1122 * e.g. it's reference counter is greater then 1.
1125 * The indirect packet mbuf.
1127 * The packet mbuf we're attaching to.
1129 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1131 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1132 rte_mbuf_refcnt_read(mi) == 1);
1134 if (RTE_MBUF_HAS_EXTBUF(m)) {
1135 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1136 mi->ol_flags = m->ol_flags;
1137 mi->shinfo = m->shinfo;
1139 /* if m is not direct, get the mbuf that embeds the data */
1140 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1141 mi->priv_size = m->priv_size;
1142 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1145 __rte_pktmbuf_copy_hdr(mi, m);
1147 mi->data_off = m->data_off;
1148 mi->data_len = m->data_len;
1149 mi->buf_iova = m->buf_iova;
1150 mi->buf_addr = m->buf_addr;
1151 mi->buf_len = m->buf_len;
1154 mi->pkt_len = mi->data_len;
1157 __rte_mbuf_sanity_check(mi, 1);
1158 __rte_mbuf_sanity_check(m, 0);
1162 * @internal used by rte_pktmbuf_detach().
1164 * Decrement the reference counter of the external buffer. When the
1165 * reference counter becomes 0, the buffer is freed by pre-registered
1169 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1171 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1172 RTE_ASSERT(m->shinfo != NULL);
1174 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1175 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1179 * @internal used by rte_pktmbuf_detach().
1181 * Decrement the direct mbuf's reference counter. When the reference
1182 * counter becomes 0, the direct mbuf is freed.
1185 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1187 struct rte_mbuf *md;
1189 RTE_ASSERT(RTE_MBUF_CLONED(m));
1191 md = rte_mbuf_from_indirect(m);
1193 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1196 rte_mbuf_refcnt_set(md, 1);
1197 rte_mbuf_raw_free(md);
1202 * Detach a packet mbuf from external buffer or direct buffer.
1204 * - decrement refcnt and free the external/direct buffer if refcnt
1206 * - restore original mbuf address and length values.
1207 * - reset pktmbuf data and data_len to their default values.
1209 * All other fields of the given packet mbuf will be left intact.
1211 * If the packet mbuf was allocated from the pool with pinned
1212 * external buffers the rte_pktmbuf_detach does nothing with the
1213 * mbuf of this kind, because the pinned buffers are not supposed
1217 * The indirect attached packet mbuf.
1219 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1221 struct rte_mempool *mp = m->pool;
1222 uint32_t mbuf_size, buf_len;
1225 if (RTE_MBUF_HAS_EXTBUF(m)) {
1227 * The mbuf has the external attached buffer,
1228 * we should check the type of the memory pool where
1229 * the mbuf was allocated from to detect the pinned
1232 uint32_t flags = rte_pktmbuf_priv_flags(mp);
1234 if (flags & RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF) {
1236 * The pinned external buffer should not be
1237 * detached from its backing mbuf, just exit.
1241 __rte_pktmbuf_free_extbuf(m);
1243 __rte_pktmbuf_free_direct(m);
1245 priv_size = rte_pktmbuf_priv_size(mp);
1246 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1247 buf_len = rte_pktmbuf_data_room_size(mp);
1249 m->priv_size = priv_size;
1250 m->buf_addr = (char *)m + mbuf_size;
1251 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1252 m->buf_len = (uint16_t)buf_len;
1253 rte_pktmbuf_reset_headroom(m);
1259 * @internal Handle the packet mbufs with attached pinned external buffer
1260 * on the mbuf freeing:
1262 * - return zero if reference counter in shinfo is one. It means there is
1263 * no more reference to this pinned buffer and mbuf can be returned to
1266 * - otherwise (if reference counter is not one), decrement reference
1267 * counter and return non-zero value to prevent freeing the backing mbuf.
1269 * Returns non zero if mbuf should not be freed.
1271 static inline int __rte_pktmbuf_pinned_extbuf_decref(struct rte_mbuf *m)
1273 struct rte_mbuf_ext_shared_info *shinfo;
1275 /* Clear flags, mbuf is being freed. */
1276 m->ol_flags = EXT_ATTACHED_MBUF;
1279 /* Optimize for performance - do not dec/reinit */
1280 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1))
1284 * Direct usage of add primitive to avoid
1285 * duplication of comparing with one.
1287 if (likely(__atomic_add_fetch(&shinfo->refcnt, (uint16_t)-1,
1291 /* Reinitialize counter before mbuf freeing. */
1292 rte_mbuf_ext_refcnt_set(shinfo, 1);
1297 * Decrease reference counter and unlink a mbuf segment
1299 * This function does the same than a free, except that it does not
1300 * return the segment to its pool.
1301 * It decreases the reference counter, and if it reaches 0, it is
1302 * detached from its parent for an indirect mbuf.
1305 * The mbuf to be unlinked
1307 * - (m) if it is the last reference. It can be recycled or freed.
1308 * - (NULL) if the mbuf still has remaining references on it.
1310 static __rte_always_inline struct rte_mbuf *
1311 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1313 __rte_mbuf_sanity_check(m, 0);
1315 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1317 if (!RTE_MBUF_DIRECT(m)) {
1318 rte_pktmbuf_detach(m);
1319 if (RTE_MBUF_HAS_EXTBUF(m) &&
1320 RTE_MBUF_HAS_PINNED_EXTBUF(m) &&
1321 __rte_pktmbuf_pinned_extbuf_decref(m))
1325 if (m->next != NULL) {
1332 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1334 if (!RTE_MBUF_DIRECT(m)) {
1335 rte_pktmbuf_detach(m);
1336 if (RTE_MBUF_HAS_EXTBUF(m) &&
1337 RTE_MBUF_HAS_PINNED_EXTBUF(m) &&
1338 __rte_pktmbuf_pinned_extbuf_decref(m))
1342 if (m->next != NULL) {
1346 rte_mbuf_refcnt_set(m, 1);
1354 * Free a segment of a packet mbuf into its original mempool.
1356 * Free an mbuf, without parsing other segments in case of chained
1360 * The packet mbuf segment to be freed.
1362 static __rte_always_inline void
1363 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1365 m = rte_pktmbuf_prefree_seg(m);
1366 if (likely(m != NULL))
1367 rte_mbuf_raw_free(m);
1371 * Free a packet mbuf back into its original mempool.
1373 * Free an mbuf, and all its segments in case of chained buffers. Each
1374 * segment is added back into its original mempool.
1377 * The packet mbuf to be freed. If NULL, the function does nothing.
1379 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1381 struct rte_mbuf *m_next;
1384 __rte_mbuf_sanity_check(m, 1);
1388 rte_pktmbuf_free_seg(m);
1394 * Free a bulk of packet mbufs back into their original mempools.
1396 * Free a bulk of mbufs, and all their segments in case of chained buffers.
1397 * Each segment is added back into its original mempool.
1400 * Array of pointers to packet mbufs.
1401 * The array may contain NULL pointers.
1406 void rte_pktmbuf_free_bulk(struct rte_mbuf **mbufs, unsigned int count);
1409 * Create a "clone" of the given packet mbuf.
1411 * Walks through all segments of the given packet mbuf, and for each of them:
1412 * - Creates a new packet mbuf from the given pool.
1413 * - Attaches newly created mbuf to the segment.
1414 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1415 * from the original packet mbuf.
1418 * The packet mbuf to be cloned.
1420 * The mempool from which the "clone" mbufs are allocated.
1422 * - The pointer to the new "clone" mbuf on success.
1423 * - NULL if allocation fails.
1426 rte_pktmbuf_clone(struct rte_mbuf *md, struct rte_mempool *mp);
1429 * Create a full copy of a given packet mbuf.
1431 * Copies all the data from a given packet mbuf to a newly allocated
1432 * set of mbufs. The private data are is not copied.
1435 * The packet mbuf to be copiedd.
1437 * The mempool from which the "clone" mbufs are allocated.
1439 * The number of bytes to skip before copying.
1440 * If the mbuf does not have that many bytes, it is an error
1441 * and NULL is returned.
1443 * The upper limit on bytes to copy. Passing UINT32_MAX
1444 * means all data (after offset).
1446 * - The pointer to the new "clone" mbuf on success.
1447 * - NULL if allocation fails.
1451 rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
1452 uint32_t offset, uint32_t length);
1455 * Adds given value to the refcnt of all packet mbuf segments.
1457 * Walks through all segments of given packet mbuf and for each of them
1458 * invokes rte_mbuf_refcnt_update().
1461 * The packet mbuf whose refcnt to be updated.
1463 * The value to add to the mbuf's segments refcnt.
1465 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1467 __rte_mbuf_sanity_check(m, 1);
1470 rte_mbuf_refcnt_update(m, v);
1471 } while ((m = m->next) != NULL);
1475 * Get the headroom in a packet mbuf.
1480 * The length of the headroom.
1482 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1484 __rte_mbuf_sanity_check(m, 0);
1489 * Get the tailroom of a packet mbuf.
1494 * The length of the tailroom.
1496 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1498 __rte_mbuf_sanity_check(m, 0);
1499 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1504 * Get the last segment of the packet.
1509 * The last segment of the given mbuf.
1511 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1513 __rte_mbuf_sanity_check(m, 1);
1514 while (m->next != NULL)
1520 * A macro that returns the length of the packet.
1522 * The value can be read or assigned.
1527 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1530 * A macro that returns the length of the segment.
1532 * The value can be read or assigned.
1537 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1540 * Prepend len bytes to an mbuf data area.
1542 * Returns a pointer to the new
1543 * data start address. If there is not enough headroom in the first
1544 * segment, the function will return NULL, without modifying the mbuf.
1549 * The amount of data to prepend (in bytes).
1551 * A pointer to the start of the newly prepended data, or
1552 * NULL if there is not enough headroom space in the first segment
1554 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1557 __rte_mbuf_sanity_check(m, 1);
1559 if (unlikely(len > rte_pktmbuf_headroom(m)))
1562 /* NB: elaborating the subtraction like this instead of using
1563 * -= allows us to ensure the result type is uint16_t
1564 * avoiding compiler warnings on gcc 8.1 at least */
1565 m->data_off = (uint16_t)(m->data_off - len);
1566 m->data_len = (uint16_t)(m->data_len + len);
1567 m->pkt_len = (m->pkt_len + len);
1569 return (char *)m->buf_addr + m->data_off;
1573 * Append len bytes to an mbuf.
1575 * Append len bytes to an mbuf and return a pointer to the start address
1576 * of the added data. If there is not enough tailroom in the last
1577 * segment, the function will return NULL, without modifying the mbuf.
1582 * The amount of data to append (in bytes).
1584 * A pointer to the start of the newly appended data, or
1585 * NULL if there is not enough tailroom space in the last segment
1587 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1590 struct rte_mbuf *m_last;
1592 __rte_mbuf_sanity_check(m, 1);
1594 m_last = rte_pktmbuf_lastseg(m);
1595 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1598 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1599 m_last->data_len = (uint16_t)(m_last->data_len + len);
1600 m->pkt_len = (m->pkt_len + len);
1601 return (char*) tail;
1605 * Remove len bytes at the beginning of an mbuf.
1607 * Returns a pointer to the start address of the new data area. If the
1608 * length is greater than the length of the first segment, then the
1609 * function will fail and return NULL, without modifying the mbuf.
1614 * The amount of data to remove (in bytes).
1616 * A pointer to the new start of the data.
1618 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1620 __rte_mbuf_sanity_check(m, 1);
1622 if (unlikely(len > m->data_len))
1625 /* NB: elaborating the addition like this instead of using
1626 * += allows us to ensure the result type is uint16_t
1627 * avoiding compiler warnings on gcc 8.1 at least */
1628 m->data_len = (uint16_t)(m->data_len - len);
1629 m->data_off = (uint16_t)(m->data_off + len);
1630 m->pkt_len = (m->pkt_len - len);
1631 return (char *)m->buf_addr + m->data_off;
1635 * Remove len bytes of data at the end of the mbuf.
1637 * If the length is greater than the length of the last segment, the
1638 * function will fail and return -1 without modifying the mbuf.
1643 * The amount of data to remove (in bytes).
1648 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
1650 struct rte_mbuf *m_last;
1652 __rte_mbuf_sanity_check(m, 1);
1654 m_last = rte_pktmbuf_lastseg(m);
1655 if (unlikely(len > m_last->data_len))
1658 m_last->data_len = (uint16_t)(m_last->data_len - len);
1659 m->pkt_len = (m->pkt_len - len);
1664 * Test if mbuf data is contiguous.
1669 * - 1, if all data is contiguous (one segment).
1670 * - 0, if there is several segments.
1672 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
1674 __rte_mbuf_sanity_check(m, 1);
1675 return m->nb_segs == 1;
1679 * @internal used by rte_pktmbuf_read().
1681 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
1682 uint32_t len, void *buf);
1685 * Read len data bytes in a mbuf at specified offset.
1687 * If the data is contiguous, return the pointer in the mbuf data, else
1688 * copy the data in the buffer provided by the user and return its
1692 * The pointer to the mbuf.
1694 * The offset of the data in the mbuf.
1696 * The amount of bytes to read.
1698 * The buffer where data is copied if it is not contiguous in mbuf
1699 * data. Its length should be at least equal to the len parameter.
1701 * The pointer to the data, either in the mbuf if it is contiguous,
1702 * or in the user buffer. If mbuf is too small, NULL is returned.
1704 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
1705 uint32_t off, uint32_t len, void *buf)
1707 if (likely(off + len <= rte_pktmbuf_data_len(m)))
1708 return rte_pktmbuf_mtod_offset(m, char *, off);
1710 return __rte_pktmbuf_read(m, off, len, buf);
1714 * Chain an mbuf to another, thereby creating a segmented packet.
1716 * Note: The implementation will do a linear walk over the segments to find
1717 * the tail entry. For cases when there are many segments, it's better to
1718 * chain the entries manually.
1721 * The head of the mbuf chain (the first packet)
1723 * The mbuf to put last in the chain
1727 * - -EOVERFLOW, if the chain segment limit exceeded
1729 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
1731 struct rte_mbuf *cur_tail;
1733 /* Check for number-of-segments-overflow */
1734 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
1737 /* Chain 'tail' onto the old tail */
1738 cur_tail = rte_pktmbuf_lastseg(head);
1739 cur_tail->next = tail;
1741 /* accumulate number of segments and total length.
1742 * NB: elaborating the addition like this instead of using
1743 * -= allows us to ensure the result type is uint16_t
1744 * avoiding compiler warnings on gcc 8.1 at least */
1745 head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
1746 head->pkt_len += tail->pkt_len;
1748 /* pkt_len is only set in the head */
1749 tail->pkt_len = tail->data_len;
1756 * @b EXPERIMENTAL: This API may change without prior notice.
1758 * For given input values generate raw tx_offload value.
1759 * Note that it is caller responsibility to make sure that input parameters
1760 * don't exceed maximum bit-field values.
1770 * outer_l3_len value.
1772 * outer_l2_len value.
1776 * raw tx_offload value.
1778 static __rte_always_inline uint64_t
1779 rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
1780 uint64_t ol3, uint64_t ol2, uint64_t unused)
1782 return il2 << RTE_MBUF_L2_LEN_OFS |
1783 il3 << RTE_MBUF_L3_LEN_OFS |
1784 il4 << RTE_MBUF_L4_LEN_OFS |
1785 tso << RTE_MBUF_TSO_SEGSZ_OFS |
1786 ol3 << RTE_MBUF_OUTL3_LEN_OFS |
1787 ol2 << RTE_MBUF_OUTL2_LEN_OFS |
1788 unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
1792 * Validate general requirements for Tx offload in mbuf.
1794 * This function checks correctness and completeness of Tx offload settings.
1797 * The packet mbuf to be validated.
1799 * 0 if packet is valid
1802 rte_validate_tx_offload(const struct rte_mbuf *m)
1804 uint64_t ol_flags = m->ol_flags;
1806 /* Does packet set any of available offloads? */
1807 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
1810 /* IP checksum can be counted only for IPv4 packet */
1811 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
1814 /* IP type not set when required */
1815 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
1816 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
1819 /* Check requirements for TSO packet */
1820 if (ol_flags & PKT_TX_TCP_SEG)
1821 if ((m->tso_segsz == 0) ||
1822 ((ol_flags & PKT_TX_IPV4) &&
1823 !(ol_flags & PKT_TX_IP_CKSUM)))
1826 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
1827 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
1828 !(ol_flags & PKT_TX_OUTER_IPV4))
1835 * @internal used by rte_pktmbuf_linearize().
1837 int __rte_pktmbuf_linearize(struct rte_mbuf *mbuf);
1840 * Linearize data in mbuf.
1842 * This function moves the mbuf data in the first segment if there is enough
1843 * tailroom. The subsequent segments are unchained and freed.
1852 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
1854 if (rte_pktmbuf_is_contiguous(mbuf))
1856 return __rte_pktmbuf_linearize(mbuf);
1860 * Dump an mbuf structure to a file.
1862 * Dump all fields for the given packet mbuf and all its associated
1863 * segments (in the case of a chained buffer).
1866 * A pointer to a file for output
1870 * If dump_len != 0, also dump the "dump_len" first data bytes of
1873 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
1876 * Get the value of mbuf sched queue_id field.
1878 static inline uint32_t
1879 rte_mbuf_sched_queue_get(const struct rte_mbuf *m)
1881 return m->hash.sched.queue_id;
1885 * Get the value of mbuf sched traffic_class field.
1887 static inline uint8_t
1888 rte_mbuf_sched_traffic_class_get(const struct rte_mbuf *m)
1890 return m->hash.sched.traffic_class;
1894 * Get the value of mbuf sched color field.
1896 static inline uint8_t
1897 rte_mbuf_sched_color_get(const struct rte_mbuf *m)
1899 return m->hash.sched.color;
1903 * Get the values of mbuf sched queue_id, traffic_class and color.
1908 * Returns the queue id
1909 * @param traffic_class
1910 * Returns the traffic class id
1912 * Returns the colour id
1915 rte_mbuf_sched_get(const struct rte_mbuf *m, uint32_t *queue_id,
1916 uint8_t *traffic_class,
1919 struct rte_mbuf_sched sched = m->hash.sched;
1921 *queue_id = sched.queue_id;
1922 *traffic_class = sched.traffic_class;
1923 *color = sched.color;
1927 * Set the mbuf sched queue_id to the defined value.
1930 rte_mbuf_sched_queue_set(struct rte_mbuf *m, uint32_t queue_id)
1932 m->hash.sched.queue_id = queue_id;
1936 * Set the mbuf sched traffic_class id to the defined value.
1939 rte_mbuf_sched_traffic_class_set(struct rte_mbuf *m, uint8_t traffic_class)
1941 m->hash.sched.traffic_class = traffic_class;
1945 * Set the mbuf sched color id to the defined value.
1948 rte_mbuf_sched_color_set(struct rte_mbuf *m, uint8_t color)
1950 m->hash.sched.color = color;
1954 * Set the mbuf sched queue_id, traffic_class and color.
1959 * Queue id value to be set
1960 * @param traffic_class
1961 * Traffic class id value to be set
1963 * Color id to be set
1966 rte_mbuf_sched_set(struct rte_mbuf *m, uint32_t queue_id,
1967 uint8_t traffic_class,
1970 m->hash.sched = (struct rte_mbuf_sched){
1971 .queue_id = queue_id,
1972 .traffic_class = traffic_class,
1982 #endif /* _RTE_MBUF_H_ */