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 static inline phys_addr_t
156 rte_mbuf_data_dma_addr(const struct rte_mbuf *mb)
158 return rte_mbuf_data_iova(mb);
162 * Return the default IO address of the beginning of the mbuf data
164 * This function is used by drivers in their receive function, as it
165 * returns the location where data should be written by the NIC, taking
166 * the default headroom in account.
169 * The pointer to the mbuf.
171 * The IO address of the beginning of the mbuf data
173 static inline rte_iova_t
174 rte_mbuf_data_iova_default(const struct rte_mbuf *mb)
176 return mb->buf_iova + RTE_PKTMBUF_HEADROOM;
180 static inline phys_addr_t
181 rte_mbuf_data_dma_addr_default(const struct rte_mbuf *mb)
183 return rte_mbuf_data_iova_default(mb);
187 * Return the mbuf owning the data buffer address of an indirect mbuf.
190 * The pointer to the indirect mbuf.
192 * The address of the direct mbuf corresponding to buffer_addr.
194 static inline struct rte_mbuf *
195 rte_mbuf_from_indirect(struct rte_mbuf *mi)
197 return (struct rte_mbuf *)RTE_PTR_SUB(mi->buf_addr, sizeof(*mi) + mi->priv_size);
201 * Return address of buffer embedded in the given mbuf.
203 * The return value shall be same as mb->buf_addr if the mbuf is already
204 * initialized and direct. However, this API is useful if mempool of the
205 * mbuf is already known because it doesn't need to access mbuf contents in
206 * order to get the mempool pointer.
209 * @b EXPERIMENTAL: This API may change without prior notice.
210 * This will be used by rte_mbuf_to_baddr() which has redundant code once
211 * experimental tag is removed.
214 * The pointer to the mbuf.
216 * The pointer to the mempool of the mbuf.
218 * The pointer of the mbuf buffer.
222 rte_mbuf_buf_addr(struct rte_mbuf *mb, struct rte_mempool *mp)
224 return (char *)mb + sizeof(*mb) + rte_pktmbuf_priv_size(mp);
228 * Return the default address of the beginning of the mbuf data.
231 * @b EXPERIMENTAL: This API may change without prior notice.
234 * The pointer to the mbuf.
236 * The pointer of the beginning of the mbuf data.
240 rte_mbuf_data_addr_default(__rte_unused struct rte_mbuf *mb)
242 /* gcc complains about calling this experimental function even
243 * when not using it. Hide it with ALLOW_EXPERIMENTAL_API.
245 #ifdef ALLOW_EXPERIMENTAL_API
246 return rte_mbuf_buf_addr(mb, mb->pool) + RTE_PKTMBUF_HEADROOM;
253 * Return address of buffer embedded in the given mbuf.
255 * @note: Accessing mempool pointer of a mbuf is expensive because the
256 * pointer is stored in the 2nd cache line of mbuf. If mempool is known, it
257 * is better not to reference the mempool pointer in mbuf but calling
258 * rte_mbuf_buf_addr() would be more efficient.
261 * The pointer to the mbuf.
263 * The address of the data buffer owned by the mbuf.
266 rte_mbuf_to_baddr(struct rte_mbuf *md)
268 #ifdef ALLOW_EXPERIMENTAL_API
269 return rte_mbuf_buf_addr(md, md->pool);
272 buffer_addr = (char *)md + sizeof(*md) + rte_pktmbuf_priv_size(md->pool);
278 * Return the starting address of the private data area embedded in
281 * Note that no check is made to ensure that a private data area
282 * actually exists in the supplied mbuf.
285 * The pointer to the mbuf.
287 * The starting address of the private data area of the given mbuf.
291 rte_mbuf_to_priv(struct rte_mbuf *m)
293 return RTE_PTR_ADD(m, sizeof(struct rte_mbuf));
297 * Private data in case of pktmbuf pool.
299 * A structure that contains some pktmbuf_pool-specific data that are
300 * appended after the mempool structure (in private data).
302 struct rte_pktmbuf_pool_private {
303 uint16_t mbuf_data_room_size; /**< Size of data space in each mbuf. */
304 uint16_t mbuf_priv_size; /**< Size of private area in each mbuf. */
305 uint32_t flags; /**< reserved for future use. */
309 * Return the flags from private data in an mempool structure.
312 * A pointer to the mempool structure.
314 * The flags from the private data structure.
316 static inline uint32_t
317 rte_pktmbuf_priv_flags(struct rte_mempool *mp)
319 struct rte_pktmbuf_pool_private *mbp_priv;
321 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
322 return mbp_priv->flags;
326 * When set, pktmbuf mempool will hold only mbufs with pinned external
327 * buffer. The external buffer will be attached to the mbuf at the
328 * memory pool creation and will never be detached by the mbuf free calls.
329 * mbuf should not contain any room for data after the mbuf structure.
331 #define RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF (1 << 0)
334 * Returns non zero if given mbuf has a pinned external buffer, or zero
335 * otherwise. The pinned external buffer is allocated at pool creation
336 * time and should not be freed on mbuf freeing.
338 * External buffer is a user-provided anonymous buffer.
340 #define RTE_MBUF_HAS_PINNED_EXTBUF(mb) \
341 (rte_pktmbuf_priv_flags(mb->pool) & RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF)
343 #ifdef RTE_LIBRTE_MBUF_DEBUG
345 /** check mbuf type in debug mode */
346 #define __rte_mbuf_sanity_check(m, is_h) rte_mbuf_sanity_check(m, is_h)
348 #else /* RTE_LIBRTE_MBUF_DEBUG */
350 /** check mbuf type in debug mode */
351 #define __rte_mbuf_sanity_check(m, is_h) do { } while (0)
353 #endif /* RTE_LIBRTE_MBUF_DEBUG */
355 #ifdef RTE_MBUF_REFCNT_ATOMIC
358 * Reads the value of an mbuf's refcnt.
362 * Reference count number.
364 static inline uint16_t
365 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
367 return __atomic_load_n(&m->refcnt, __ATOMIC_RELAXED);
371 * Sets an mbuf's refcnt to a defined value.
378 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
380 __atomic_store_n(&m->refcnt, new_value, __ATOMIC_RELAXED);
384 static inline uint16_t
385 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
387 return __atomic_add_fetch(&m->refcnt, (uint16_t)value,
392 * Adds given value to an mbuf's refcnt and returns its new value.
396 * Value to add/subtract
400 static inline uint16_t
401 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
404 * The atomic_add is an expensive operation, so we don't want to
405 * call it in the case where we know we are the unique holder of
406 * this mbuf (i.e. ref_cnt == 1). Otherwise, an atomic
407 * operation has to be used because concurrent accesses on the
408 * reference counter can occur.
410 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
412 rte_mbuf_refcnt_set(m, (uint16_t)value);
413 return (uint16_t)value;
416 return __rte_mbuf_refcnt_update(m, value);
419 #else /* ! RTE_MBUF_REFCNT_ATOMIC */
422 static inline uint16_t
423 __rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
425 m->refcnt = (uint16_t)(m->refcnt + value);
430 * Adds given value to an mbuf's refcnt and returns its new value.
432 static inline uint16_t
433 rte_mbuf_refcnt_update(struct rte_mbuf *m, int16_t value)
435 return __rte_mbuf_refcnt_update(m, value);
439 * Reads the value of an mbuf's refcnt.
441 static inline uint16_t
442 rte_mbuf_refcnt_read(const struct rte_mbuf *m)
448 * Sets an mbuf's refcnt to the defined value.
451 rte_mbuf_refcnt_set(struct rte_mbuf *m, uint16_t new_value)
453 m->refcnt = new_value;
456 #endif /* RTE_MBUF_REFCNT_ATOMIC */
459 * Reads the refcnt of an external buffer.
462 * Shared data of the external buffer.
464 * Reference count number.
466 static inline uint16_t
467 rte_mbuf_ext_refcnt_read(const struct rte_mbuf_ext_shared_info *shinfo)
469 return __atomic_load_n(&shinfo->refcnt, __ATOMIC_RELAXED);
473 * Set refcnt of an external buffer.
476 * Shared data of the external buffer.
481 rte_mbuf_ext_refcnt_set(struct rte_mbuf_ext_shared_info *shinfo,
484 __atomic_store_n(&shinfo->refcnt, new_value, __ATOMIC_RELAXED);
488 * Add given value to refcnt of an external buffer and return its new
492 * Shared data of the external buffer.
494 * Value to add/subtract
498 static inline uint16_t
499 rte_mbuf_ext_refcnt_update(struct rte_mbuf_ext_shared_info *shinfo,
502 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1)) {
504 rte_mbuf_ext_refcnt_set(shinfo, (uint16_t)value);
505 return (uint16_t)value;
508 return __atomic_add_fetch(&shinfo->refcnt, (uint16_t)value,
513 #define RTE_MBUF_PREFETCH_TO_FREE(m) do { \
520 * Sanity checks on an mbuf.
522 * Check the consistency of the given mbuf. The function will cause a
523 * panic if corruption is detected.
526 * The mbuf to be checked.
528 * True if the mbuf is a packet header, false if it is a sub-segment
529 * of a packet (in this case, some fields like nb_segs are not checked)
532 rte_mbuf_sanity_check(const struct rte_mbuf *m, int is_header);
535 * Sanity checks on a mbuf.
537 * Almost like rte_mbuf_sanity_check(), but this function gives the reason
538 * if corruption is detected rather than panic.
541 * The mbuf to be checked.
543 * True if the mbuf is a packet header, false if it is a sub-segment
544 * of a packet (in this case, some fields like nb_segs are not checked)
546 * A reference to a string pointer where to store the reason why a mbuf is
547 * considered invalid.
549 * - 0 if no issue has been found, reason is left untouched.
550 * - -1 if a problem is detected, reason then points to a string describing
551 * the reason why the mbuf is deemed invalid.
554 int rte_mbuf_check(const struct rte_mbuf *m, int is_header,
555 const char **reason);
557 #define MBUF_RAW_ALLOC_CHECK(m) do { \
558 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1); \
559 RTE_ASSERT((m)->next == NULL); \
560 RTE_ASSERT((m)->nb_segs == 1); \
561 __rte_mbuf_sanity_check(m, 0); \
565 * Allocate an uninitialized mbuf from mempool *mp*.
567 * This function can be used by PMDs (especially in RX functions) to
568 * allocate an uninitialized mbuf. The driver is responsible of
569 * initializing all the required fields. See rte_pktmbuf_reset().
570 * For standard needs, prefer rte_pktmbuf_alloc().
572 * The caller can expect that the following fields of the mbuf structure
573 * are initialized: buf_addr, buf_iova, buf_len, refcnt=1, nb_segs=1,
574 * next=NULL, pool, priv_size. The other fields must be initialized
578 * The mempool from which mbuf is allocated.
580 * - The pointer to the new mbuf on success.
581 * - NULL if allocation failed.
583 static inline struct rte_mbuf *rte_mbuf_raw_alloc(struct rte_mempool *mp)
587 if (rte_mempool_get(mp, (void **)&m) < 0)
589 MBUF_RAW_ALLOC_CHECK(m);
594 * Put mbuf back into its original mempool.
596 * The caller must ensure that the mbuf is direct and properly
597 * reinitialized (refcnt=1, next=NULL, nb_segs=1), as done by
598 * rte_pktmbuf_prefree_seg().
600 * This function should be used with care, when optimization is
601 * required. For standard needs, prefer rte_pktmbuf_free() or
602 * rte_pktmbuf_free_seg().
605 * The mbuf to be freed.
607 static __rte_always_inline void
608 rte_mbuf_raw_free(struct rte_mbuf *m)
610 RTE_ASSERT(!RTE_MBUF_CLONED(m) &&
611 (!RTE_MBUF_HAS_EXTBUF(m) || RTE_MBUF_HAS_PINNED_EXTBUF(m)));
612 RTE_ASSERT(rte_mbuf_refcnt_read(m) == 1);
613 RTE_ASSERT(m->next == NULL);
614 RTE_ASSERT(m->nb_segs == 1);
615 __rte_mbuf_sanity_check(m, 0);
616 rte_mempool_put(m->pool, m);
620 * The packet mbuf constructor.
622 * This function initializes some fields in the mbuf structure that are
623 * not modified by the user once created (origin pool, buffer start
624 * address, and so on). This function is given as a callback function to
625 * rte_mempool_obj_iter() or rte_mempool_create() at pool creation time.
628 * The mempool from which mbufs originate.
630 * A pointer that can be used by the user to retrieve useful information
631 * for mbuf initialization. This pointer is the opaque argument passed to
632 * rte_mempool_obj_iter() or rte_mempool_create().
634 * The mbuf to initialize.
636 * The index of the mbuf in the pool table.
638 void rte_pktmbuf_init(struct rte_mempool *mp, void *opaque_arg,
639 void *m, unsigned i);
642 * A packet mbuf pool constructor.
644 * This function initializes the mempool private data in the case of a
645 * pktmbuf pool. This private data is needed by the driver. The
646 * function must be called on the mempool before it is used, or it
647 * can be given as a callback function to rte_mempool_create() at
648 * pool creation. It can be extended by the user, for example, to
649 * provide another packet size.
652 * The mempool from which mbufs originate.
654 * A pointer that can be used by the user to retrieve useful information
655 * for mbuf initialization. This pointer is the opaque argument passed to
656 * rte_mempool_create().
658 void rte_pktmbuf_pool_init(struct rte_mempool *mp, void *opaque_arg);
661 * Create a mbuf pool.
663 * This function creates and initializes a packet mbuf pool. It is
664 * a wrapper to rte_mempool functions.
667 * The name of the mbuf pool.
669 * The number of elements in the mbuf pool. The optimum size (in terms
670 * of memory usage) for a mempool is when n is a power of two minus one:
673 * Size of the per-core object cache. See rte_mempool_create() for
676 * Size of application private are between the rte_mbuf structure
677 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
678 * @param data_room_size
679 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
681 * The socket identifier where the memory should be allocated. The
682 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
685 * The pointer to the new allocated mempool, on success. NULL on error
686 * with rte_errno set appropriately. Possible rte_errno values include:
687 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
688 * - E_RTE_SECONDARY - function was called from a secondary process instance
689 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
690 * - ENOSPC - the maximum number of memzones has already been allocated
691 * - EEXIST - a memzone with the same name already exists
692 * - ENOMEM - no appropriate memory area found in which to create memzone
695 rte_pktmbuf_pool_create(const char *name, unsigned n,
696 unsigned cache_size, uint16_t priv_size, uint16_t data_room_size,
700 * Create a mbuf pool with a given mempool ops name
702 * This function creates and initializes a packet mbuf pool. It is
703 * a wrapper to rte_mempool functions.
706 * The name of the mbuf pool.
708 * The number of elements in the mbuf pool. The optimum size (in terms
709 * of memory usage) for a mempool is when n is a power of two minus one:
712 * Size of the per-core object cache. See rte_mempool_create() for
715 * Size of application private are between the rte_mbuf structure
716 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
717 * @param data_room_size
718 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
720 * The socket identifier where the memory should be allocated. The
721 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
724 * The mempool ops name to be used for this mempool instead of
725 * default mempool. The value can be *NULL* to use default mempool.
727 * The pointer to the new allocated mempool, on success. NULL on error
728 * with rte_errno set appropriately. Possible rte_errno values include:
729 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
730 * - E_RTE_SECONDARY - function was called from a secondary process instance
731 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
732 * - ENOSPC - the maximum number of memzones has already been allocated
733 * - EEXIST - a memzone with the same name already exists
734 * - ENOMEM - no appropriate memory area found in which to create memzone
737 rte_pktmbuf_pool_create_by_ops(const char *name, unsigned int n,
738 unsigned int cache_size, uint16_t priv_size, uint16_t data_room_size,
739 int socket_id, const char *ops_name);
741 /** A structure that describes the pinned external buffer segment. */
742 struct rte_pktmbuf_extmem {
743 void *buf_ptr; /**< The virtual address of data buffer. */
744 rte_iova_t buf_iova; /**< The IO address of the data buffer. */
745 size_t buf_len; /**< External buffer length in bytes. */
746 uint16_t elt_size; /**< mbuf element size in bytes. */
750 * Create a mbuf pool with external pinned data buffers.
752 * This function creates and initializes a packet mbuf pool that contains
753 * only mbufs with external buffer. It is a wrapper to rte_mempool functions.
756 * The name of the mbuf pool.
758 * The number of elements in the mbuf pool. The optimum size (in terms
759 * of memory usage) for a mempool is when n is a power of two minus one:
762 * Size of the per-core object cache. See rte_mempool_create() for
765 * Size of application private are between the rte_mbuf structure
766 * and the data buffer. This value must be aligned to RTE_MBUF_PRIV_ALIGN.
767 * @param data_room_size
768 * Size of data buffer in each mbuf, including RTE_PKTMBUF_HEADROOM.
770 * The socket identifier where the memory should be allocated. The
771 * value can be *SOCKET_ID_ANY* if there is no NUMA constraint for the
774 * Pointer to the array of structures describing the external memory
775 * for data buffers. It is caller responsibility to register this memory
776 * with rte_extmem_register() (if needed), map this memory to appropriate
777 * physical device, etc.
779 * Number of elements in the ext_mem array.
781 * The pointer to the new allocated mempool, on success. NULL on error
782 * with rte_errno set appropriately. Possible rte_errno values include:
783 * - E_RTE_NO_CONFIG - function could not get pointer to rte_config structure
784 * - E_RTE_SECONDARY - function was called from a secondary process instance
785 * - EINVAL - cache size provided is too large, or priv_size is not aligned.
786 * - ENOSPC - the maximum number of memzones has already been allocated
787 * - EEXIST - a memzone with the same name already exists
788 * - ENOMEM - no appropriate memory area found in which to create memzone
792 rte_pktmbuf_pool_create_extbuf(const char *name, unsigned int n,
793 unsigned int cache_size, uint16_t priv_size,
794 uint16_t data_room_size, int socket_id,
795 const struct rte_pktmbuf_extmem *ext_mem,
796 unsigned int ext_num);
799 * Get the data room size of mbufs stored in a pktmbuf_pool
801 * The data room size is the amount of data that can be stored in a
802 * mbuf including the headroom (RTE_PKTMBUF_HEADROOM).
805 * The packet mbuf pool.
807 * The data room size of mbufs stored in this mempool.
809 static inline uint16_t
810 rte_pktmbuf_data_room_size(struct rte_mempool *mp)
812 struct rte_pktmbuf_pool_private *mbp_priv;
814 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
815 return mbp_priv->mbuf_data_room_size;
819 * Get the application private size of mbufs stored in a pktmbuf_pool
821 * The private size of mbuf is a zone located between the rte_mbuf
822 * structure and the data buffer where an application can store data
823 * associated to a packet.
826 * The packet mbuf pool.
828 * The private size of mbufs stored in this mempool.
830 static inline uint16_t
831 rte_pktmbuf_priv_size(struct rte_mempool *mp)
833 struct rte_pktmbuf_pool_private *mbp_priv;
835 mbp_priv = (struct rte_pktmbuf_pool_private *)rte_mempool_get_priv(mp);
836 return mbp_priv->mbuf_priv_size;
840 * Reset the data_off field of a packet mbuf to its default value.
842 * The given mbuf must have only one segment, which should be empty.
845 * The packet mbuf's data_off field has to be reset.
847 static inline void rte_pktmbuf_reset_headroom(struct rte_mbuf *m)
849 m->data_off = (uint16_t)RTE_MIN((uint16_t)RTE_PKTMBUF_HEADROOM,
850 (uint16_t)m->buf_len);
854 * Reset the fields of a packet mbuf to their default values.
856 * The given mbuf must have only one segment.
859 * The packet mbuf to be reset.
861 #define MBUF_INVALID_PORT UINT16_MAX
863 static inline void rte_pktmbuf_reset(struct rte_mbuf *m)
869 m->vlan_tci_outer = 0;
871 m->port = MBUF_INVALID_PORT;
873 m->ol_flags &= EXT_ATTACHED_MBUF;
875 rte_pktmbuf_reset_headroom(m);
878 __rte_mbuf_sanity_check(m, 1);
882 * Allocate a new mbuf from a mempool.
884 * This new mbuf contains one segment, which has a length of 0. The pointer
885 * to data is initialized to have some bytes of headroom in the buffer
886 * (if buffer size allows).
889 * The mempool from which the mbuf is allocated.
891 * - The pointer to the new mbuf on success.
892 * - NULL if allocation failed.
894 static inline struct rte_mbuf *rte_pktmbuf_alloc(struct rte_mempool *mp)
897 if ((m = rte_mbuf_raw_alloc(mp)) != NULL)
898 rte_pktmbuf_reset(m);
903 * Allocate a bulk of mbufs, initialize refcnt and reset the fields to default
907 * The mempool from which mbufs are allocated.
909 * Array of pointers to mbufs
914 * - -ENOENT: Not enough entries in the mempool; no mbufs are retrieved.
916 static inline int rte_pktmbuf_alloc_bulk(struct rte_mempool *pool,
917 struct rte_mbuf **mbufs, unsigned count)
922 rc = rte_mempool_get_bulk(pool, (void **)mbufs, count);
926 /* To understand duff's device on loop unwinding optimization, see
927 * https://en.wikipedia.org/wiki/Duff's_device.
928 * Here while() loop is used rather than do() while{} to avoid extra
929 * check if count is zero.
933 while (idx != count) {
934 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
935 rte_pktmbuf_reset(mbufs[idx]);
939 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
940 rte_pktmbuf_reset(mbufs[idx]);
944 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
945 rte_pktmbuf_reset(mbufs[idx]);
949 MBUF_RAW_ALLOC_CHECK(mbufs[idx]);
950 rte_pktmbuf_reset(mbufs[idx]);
959 * Initialize shared data at the end of an external buffer before attaching
960 * to a mbuf by ``rte_pktmbuf_attach_extbuf()``. This is not a mandatory
961 * initialization but a helper function to simply spare a few bytes at the
962 * end of the buffer for shared data. If shared data is allocated
963 * separately, this should not be called but application has to properly
964 * initialize the shared data according to its need.
966 * Free callback and its argument is saved and the refcnt is set to 1.
969 * The value of buf_len will be reduced to RTE_PTR_DIFF(shinfo, buf_addr)
970 * after this initialization. This shall be used for
971 * ``rte_pktmbuf_attach_extbuf()``
974 * The pointer to the external buffer.
975 * @param [in,out] buf_len
976 * The pointer to length of the external buffer. Input value must be
977 * larger than the size of ``struct rte_mbuf_ext_shared_info`` and
978 * padding for alignment. If not enough, this function will return NULL.
979 * Adjusted buffer length will be returned through this pointer.
981 * Free callback function to call when the external buffer needs to be
984 * Argument for the free callback function.
987 * A pointer to the initialized shared data on success, return NULL
990 static inline struct rte_mbuf_ext_shared_info *
991 rte_pktmbuf_ext_shinfo_init_helper(void *buf_addr, uint16_t *buf_len,
992 rte_mbuf_extbuf_free_callback_t free_cb, void *fcb_opaque)
994 struct rte_mbuf_ext_shared_info *shinfo;
995 void *buf_end = RTE_PTR_ADD(buf_addr, *buf_len);
998 addr = RTE_PTR_ALIGN_FLOOR(RTE_PTR_SUB(buf_end, sizeof(*shinfo)),
1000 if (addr <= buf_addr)
1003 shinfo = (struct rte_mbuf_ext_shared_info *)addr;
1004 shinfo->free_cb = free_cb;
1005 shinfo->fcb_opaque = fcb_opaque;
1006 rte_mbuf_ext_refcnt_set(shinfo, 1);
1008 *buf_len = (uint16_t)RTE_PTR_DIFF(shinfo, buf_addr);
1013 * Attach an external buffer to a mbuf.
1015 * User-managed anonymous buffer can be attached to an mbuf. When attaching
1016 * it, corresponding free callback function and its argument should be
1017 * provided via shinfo. This callback function will be called once all the
1018 * mbufs are detached from the buffer (refcnt becomes zero).
1020 * The headroom length of the attaching mbuf will be set to zero and this
1021 * can be properly adjusted after attachment. For example, ``rte_pktmbuf_adj()``
1022 * or ``rte_pktmbuf_reset_headroom()`` might be used.
1024 * Similarly, the packet length is initialized to 0. If the buffer contains
1025 * data, the user has to adjust ``data_len`` and the ``pkt_len`` field of
1026 * the mbuf accordingly.
1028 * More mbufs can be attached to the same external buffer by
1029 * ``rte_pktmbuf_attach()`` once the external buffer has been attached by
1032 * Detachment can be done by either ``rte_pktmbuf_detach_extbuf()`` or
1033 * ``rte_pktmbuf_detach()``.
1035 * Memory for shared data must be provided and user must initialize all of
1036 * the content properly, especially free callback and refcnt. The pointer
1037 * of shared data will be stored in m->shinfo.
1038 * ``rte_pktmbuf_ext_shinfo_init_helper`` can help to simply spare a few
1039 * bytes at the end of buffer for the shared data, store free callback and
1040 * its argument and set the refcnt to 1. The following is an example:
1042 * struct rte_mbuf_ext_shared_info *shinfo =
1043 * rte_pktmbuf_ext_shinfo_init_helper(buf_addr, &buf_len,
1044 * free_cb, fcb_arg);
1045 * rte_pktmbuf_attach_extbuf(m, buf_addr, buf_iova, buf_len, shinfo);
1046 * rte_pktmbuf_reset_headroom(m);
1047 * rte_pktmbuf_adj(m, data_len);
1049 * Attaching an external buffer is quite similar to mbuf indirection in
1050 * replacing buffer addresses and length of a mbuf, but a few differences:
1051 * - When an indirect mbuf is attached, refcnt of the direct mbuf would be
1052 * 2 as long as the direct mbuf itself isn't freed after the attachment.
1053 * In such cases, the buffer area of a direct mbuf must be read-only. But
1054 * external buffer has its own refcnt and it starts from 1. Unless
1055 * multiple mbufs are attached to a mbuf having an external buffer, the
1056 * external buffer is writable.
1057 * - There's no need to allocate buffer from a mempool. Any buffer can be
1058 * attached with appropriate free callback and its IO address.
1059 * - Smaller metadata is required to maintain shared data such as refcnt.
1062 * The pointer to the mbuf.
1064 * The pointer to the external buffer.
1066 * IO address of the external buffer.
1068 * The size of the external buffer.
1070 * User-provided memory for shared data of the external buffer.
1073 rte_pktmbuf_attach_extbuf(struct rte_mbuf *m, void *buf_addr,
1074 rte_iova_t buf_iova, uint16_t buf_len,
1075 struct rte_mbuf_ext_shared_info *shinfo)
1077 /* mbuf should not be read-only */
1078 RTE_ASSERT(RTE_MBUF_DIRECT(m) && rte_mbuf_refcnt_read(m) == 1);
1079 RTE_ASSERT(shinfo->free_cb != NULL);
1081 m->buf_addr = buf_addr;
1082 m->buf_iova = buf_iova;
1083 m->buf_len = buf_len;
1088 m->ol_flags |= EXT_ATTACHED_MBUF;
1093 * Detach the external buffer attached to a mbuf, same as
1094 * ``rte_pktmbuf_detach()``
1097 * The mbuf having external buffer.
1099 #define rte_pktmbuf_detach_extbuf(m) rte_pktmbuf_detach(m)
1102 * Copy dynamic fields from msrc to mdst.
1105 * The destination mbuf.
1110 rte_mbuf_dynfield_copy(struct rte_mbuf *mdst, const struct rte_mbuf *msrc)
1112 memcpy(&mdst->dynfield1, msrc->dynfield1, sizeof(mdst->dynfield1));
1117 __rte_pktmbuf_copy_hdr(struct rte_mbuf *mdst, const struct rte_mbuf *msrc)
1119 mdst->port = msrc->port;
1120 mdst->vlan_tci = msrc->vlan_tci;
1121 mdst->vlan_tci_outer = msrc->vlan_tci_outer;
1122 mdst->tx_offload = msrc->tx_offload;
1123 mdst->hash = msrc->hash;
1124 mdst->packet_type = msrc->packet_type;
1125 mdst->timestamp = msrc->timestamp;
1126 rte_mbuf_dynfield_copy(mdst, msrc);
1130 * Attach packet mbuf to another packet mbuf.
1132 * If the mbuf we are attaching to isn't a direct buffer and is attached to
1133 * an external buffer, the mbuf being attached will be attached to the
1134 * external buffer instead of mbuf indirection.
1136 * Otherwise, the mbuf will be indirectly attached. After attachment we
1137 * refer the mbuf we attached as 'indirect', while mbuf we attached to as
1138 * 'direct'. The direct mbuf's reference counter is incremented.
1140 * Right now, not supported:
1141 * - attachment for already indirect mbuf (e.g. - mi has to be direct).
1142 * - mbuf we trying to attach (mi) is used by someone else
1143 * e.g. it's reference counter is greater then 1.
1146 * The indirect packet mbuf.
1148 * The packet mbuf we're attaching to.
1150 static inline void rte_pktmbuf_attach(struct rte_mbuf *mi, struct rte_mbuf *m)
1152 RTE_ASSERT(RTE_MBUF_DIRECT(mi) &&
1153 rte_mbuf_refcnt_read(mi) == 1);
1155 if (RTE_MBUF_HAS_EXTBUF(m)) {
1156 rte_mbuf_ext_refcnt_update(m->shinfo, 1);
1157 mi->ol_flags = m->ol_flags;
1158 mi->shinfo = m->shinfo;
1160 /* if m is not direct, get the mbuf that embeds the data */
1161 rte_mbuf_refcnt_update(rte_mbuf_from_indirect(m), 1);
1162 mi->priv_size = m->priv_size;
1163 mi->ol_flags = m->ol_flags | IND_ATTACHED_MBUF;
1166 __rte_pktmbuf_copy_hdr(mi, m);
1168 mi->data_off = m->data_off;
1169 mi->data_len = m->data_len;
1170 mi->buf_iova = m->buf_iova;
1171 mi->buf_addr = m->buf_addr;
1172 mi->buf_len = m->buf_len;
1175 mi->pkt_len = mi->data_len;
1178 __rte_mbuf_sanity_check(mi, 1);
1179 __rte_mbuf_sanity_check(m, 0);
1183 * @internal used by rte_pktmbuf_detach().
1185 * Decrement the reference counter of the external buffer. When the
1186 * reference counter becomes 0, the buffer is freed by pre-registered
1190 __rte_pktmbuf_free_extbuf(struct rte_mbuf *m)
1192 RTE_ASSERT(RTE_MBUF_HAS_EXTBUF(m));
1193 RTE_ASSERT(m->shinfo != NULL);
1195 if (rte_mbuf_ext_refcnt_update(m->shinfo, -1) == 0)
1196 m->shinfo->free_cb(m->buf_addr, m->shinfo->fcb_opaque);
1200 * @internal used by rte_pktmbuf_detach().
1202 * Decrement the direct mbuf's reference counter. When the reference
1203 * counter becomes 0, the direct mbuf is freed.
1206 __rte_pktmbuf_free_direct(struct rte_mbuf *m)
1208 struct rte_mbuf *md;
1210 RTE_ASSERT(RTE_MBUF_CLONED(m));
1212 md = rte_mbuf_from_indirect(m);
1214 if (rte_mbuf_refcnt_update(md, -1) == 0) {
1217 rte_mbuf_refcnt_set(md, 1);
1218 rte_mbuf_raw_free(md);
1223 * Detach a packet mbuf from external buffer or direct buffer.
1225 * - decrement refcnt and free the external/direct buffer if refcnt
1227 * - restore original mbuf address and length values.
1228 * - reset pktmbuf data and data_len to their default values.
1230 * All other fields of the given packet mbuf will be left intact.
1232 * If the packet mbuf was allocated from the pool with pinned
1233 * external buffers the rte_pktmbuf_detach does nothing with the
1234 * mbuf of this kind, because the pinned buffers are not supposed
1238 * The indirect attached packet mbuf.
1240 static inline void rte_pktmbuf_detach(struct rte_mbuf *m)
1242 struct rte_mempool *mp = m->pool;
1243 uint32_t mbuf_size, buf_len;
1246 if (RTE_MBUF_HAS_EXTBUF(m)) {
1248 * The mbuf has the external attached buffer,
1249 * we should check the type of the memory pool where
1250 * the mbuf was allocated from to detect the pinned
1253 uint32_t flags = rte_pktmbuf_priv_flags(mp);
1255 if (flags & RTE_PKTMBUF_POOL_F_PINNED_EXT_BUF) {
1257 * The pinned external buffer should not be
1258 * detached from its backing mbuf, just exit.
1262 __rte_pktmbuf_free_extbuf(m);
1264 __rte_pktmbuf_free_direct(m);
1266 priv_size = rte_pktmbuf_priv_size(mp);
1267 mbuf_size = (uint32_t)(sizeof(struct rte_mbuf) + priv_size);
1268 buf_len = rte_pktmbuf_data_room_size(mp);
1270 m->priv_size = priv_size;
1271 m->buf_addr = (char *)m + mbuf_size;
1272 m->buf_iova = rte_mempool_virt2iova(m) + mbuf_size;
1273 m->buf_len = (uint16_t)buf_len;
1274 rte_pktmbuf_reset_headroom(m);
1280 * @internal Handle the packet mbufs with attached pinned external buffer
1281 * on the mbuf freeing:
1283 * - return zero if reference counter in shinfo is one. It means there is
1284 * no more reference to this pinned buffer and mbuf can be returned to
1287 * - otherwise (if reference counter is not one), decrement reference
1288 * counter and return non-zero value to prevent freeing the backing mbuf.
1290 * Returns non zero if mbuf should not be freed.
1292 static inline int __rte_pktmbuf_pinned_extbuf_decref(struct rte_mbuf *m)
1294 struct rte_mbuf_ext_shared_info *shinfo;
1296 /* Clear flags, mbuf is being freed. */
1297 m->ol_flags = EXT_ATTACHED_MBUF;
1300 /* Optimize for performance - do not dec/reinit */
1301 if (likely(rte_mbuf_ext_refcnt_read(shinfo) == 1))
1305 * Direct usage of add primitive to avoid
1306 * duplication of comparing with one.
1308 if (likely(__atomic_add_fetch(&shinfo->refcnt, (uint16_t)-1,
1312 /* Reinitialize counter before mbuf freeing. */
1313 rte_mbuf_ext_refcnt_set(shinfo, 1);
1318 * Decrease reference counter and unlink a mbuf segment
1320 * This function does the same than a free, except that it does not
1321 * return the segment to its pool.
1322 * It decreases the reference counter, and if it reaches 0, it is
1323 * detached from its parent for an indirect mbuf.
1326 * The mbuf to be unlinked
1328 * - (m) if it is the last reference. It can be recycled or freed.
1329 * - (NULL) if the mbuf still has remaining references on it.
1331 static __rte_always_inline struct rte_mbuf *
1332 rte_pktmbuf_prefree_seg(struct rte_mbuf *m)
1334 __rte_mbuf_sanity_check(m, 0);
1336 if (likely(rte_mbuf_refcnt_read(m) == 1)) {
1338 if (!RTE_MBUF_DIRECT(m)) {
1339 rte_pktmbuf_detach(m);
1340 if (RTE_MBUF_HAS_EXTBUF(m) &&
1341 RTE_MBUF_HAS_PINNED_EXTBUF(m) &&
1342 __rte_pktmbuf_pinned_extbuf_decref(m))
1346 if (m->next != NULL) {
1353 } else if (__rte_mbuf_refcnt_update(m, -1) == 0) {
1355 if (!RTE_MBUF_DIRECT(m)) {
1356 rte_pktmbuf_detach(m);
1357 if (RTE_MBUF_HAS_EXTBUF(m) &&
1358 RTE_MBUF_HAS_PINNED_EXTBUF(m) &&
1359 __rte_pktmbuf_pinned_extbuf_decref(m))
1363 if (m->next != NULL) {
1367 rte_mbuf_refcnt_set(m, 1);
1375 * Free a segment of a packet mbuf into its original mempool.
1377 * Free an mbuf, without parsing other segments in case of chained
1381 * The packet mbuf segment to be freed.
1383 static __rte_always_inline void
1384 rte_pktmbuf_free_seg(struct rte_mbuf *m)
1386 m = rte_pktmbuf_prefree_seg(m);
1387 if (likely(m != NULL))
1388 rte_mbuf_raw_free(m);
1392 * Free a packet mbuf back into its original mempool.
1394 * Free an mbuf, and all its segments in case of chained buffers. Each
1395 * segment is added back into its original mempool.
1398 * The packet mbuf to be freed. If NULL, the function does nothing.
1400 static inline void rte_pktmbuf_free(struct rte_mbuf *m)
1402 struct rte_mbuf *m_next;
1405 __rte_mbuf_sanity_check(m, 1);
1409 rte_pktmbuf_free_seg(m);
1415 * Free a bulk of packet mbufs back into their original mempools.
1417 * Free a bulk of mbufs, and all their segments in case of chained buffers.
1418 * Each segment is added back into its original mempool.
1421 * Array of pointers to packet mbufs.
1422 * The array may contain NULL pointers.
1427 void rte_pktmbuf_free_bulk(struct rte_mbuf **mbufs, unsigned int count);
1430 * Create a "clone" of the given packet mbuf.
1432 * Walks through all segments of the given packet mbuf, and for each of them:
1433 * - Creates a new packet mbuf from the given pool.
1434 * - Attaches newly created mbuf to the segment.
1435 * Then updates pkt_len and nb_segs of the "clone" packet mbuf to match values
1436 * from the original packet mbuf.
1439 * The packet mbuf to be cloned.
1441 * The mempool from which the "clone" mbufs are allocated.
1443 * - The pointer to the new "clone" mbuf on success.
1444 * - NULL if allocation fails.
1447 rte_pktmbuf_clone(struct rte_mbuf *md, struct rte_mempool *mp);
1450 * Create a full copy of a given packet mbuf.
1452 * Copies all the data from a given packet mbuf to a newly allocated
1453 * set of mbufs. The private data are is not copied.
1456 * The packet mbuf to be copiedd.
1458 * The mempool from which the "clone" mbufs are allocated.
1460 * The number of bytes to skip before copying.
1461 * If the mbuf does not have that many bytes, it is an error
1462 * and NULL is returned.
1464 * The upper limit on bytes to copy. Passing UINT32_MAX
1465 * means all data (after offset).
1467 * - The pointer to the new "clone" mbuf on success.
1468 * - NULL if allocation fails.
1472 rte_pktmbuf_copy(const struct rte_mbuf *m, struct rte_mempool *mp,
1473 uint32_t offset, uint32_t length);
1476 * Adds given value to the refcnt of all packet mbuf segments.
1478 * Walks through all segments of given packet mbuf and for each of them
1479 * invokes rte_mbuf_refcnt_update().
1482 * The packet mbuf whose refcnt to be updated.
1484 * The value to add to the mbuf's segments refcnt.
1486 static inline void rte_pktmbuf_refcnt_update(struct rte_mbuf *m, int16_t v)
1488 __rte_mbuf_sanity_check(m, 1);
1491 rte_mbuf_refcnt_update(m, v);
1492 } while ((m = m->next) != NULL);
1496 * Get the headroom in a packet mbuf.
1501 * The length of the headroom.
1503 static inline uint16_t rte_pktmbuf_headroom(const struct rte_mbuf *m)
1505 __rte_mbuf_sanity_check(m, 0);
1510 * Get the tailroom of a packet mbuf.
1515 * The length of the tailroom.
1517 static inline uint16_t rte_pktmbuf_tailroom(const struct rte_mbuf *m)
1519 __rte_mbuf_sanity_check(m, 0);
1520 return (uint16_t)(m->buf_len - rte_pktmbuf_headroom(m) -
1525 * Get the last segment of the packet.
1530 * The last segment of the given mbuf.
1532 static inline struct rte_mbuf *rte_pktmbuf_lastseg(struct rte_mbuf *m)
1534 __rte_mbuf_sanity_check(m, 1);
1535 while (m->next != NULL)
1541 #define rte_pktmbuf_mtophys_offset(m, o) \
1542 rte_pktmbuf_iova_offset(m, o)
1545 #define rte_pktmbuf_mtophys(m) rte_pktmbuf_iova(m)
1548 * A macro that returns the length of the packet.
1550 * The value can be read or assigned.
1555 #define rte_pktmbuf_pkt_len(m) ((m)->pkt_len)
1558 * A macro that returns the length of the segment.
1560 * The value can be read or assigned.
1565 #define rte_pktmbuf_data_len(m) ((m)->data_len)
1568 * Prepend len bytes to an mbuf data area.
1570 * Returns a pointer to the new
1571 * data start address. If there is not enough headroom in the first
1572 * segment, the function will return NULL, without modifying the mbuf.
1577 * The amount of data to prepend (in bytes).
1579 * A pointer to the start of the newly prepended data, or
1580 * NULL if there is not enough headroom space in the first segment
1582 static inline char *rte_pktmbuf_prepend(struct rte_mbuf *m,
1585 __rte_mbuf_sanity_check(m, 1);
1587 if (unlikely(len > rte_pktmbuf_headroom(m)))
1590 /* NB: elaborating the subtraction like this instead of using
1591 * -= allows us to ensure the result type is uint16_t
1592 * avoiding compiler warnings on gcc 8.1 at least */
1593 m->data_off = (uint16_t)(m->data_off - len);
1594 m->data_len = (uint16_t)(m->data_len + len);
1595 m->pkt_len = (m->pkt_len + len);
1597 return (char *)m->buf_addr + m->data_off;
1601 * Append len bytes to an mbuf.
1603 * Append len bytes to an mbuf and return a pointer to the start address
1604 * of the added data. If there is not enough tailroom in the last
1605 * segment, the function will return NULL, without modifying the mbuf.
1610 * The amount of data to append (in bytes).
1612 * A pointer to the start of the newly appended data, or
1613 * NULL if there is not enough tailroom space in the last segment
1615 static inline char *rte_pktmbuf_append(struct rte_mbuf *m, uint16_t len)
1618 struct rte_mbuf *m_last;
1620 __rte_mbuf_sanity_check(m, 1);
1622 m_last = rte_pktmbuf_lastseg(m);
1623 if (unlikely(len > rte_pktmbuf_tailroom(m_last)))
1626 tail = (char *)m_last->buf_addr + m_last->data_off + m_last->data_len;
1627 m_last->data_len = (uint16_t)(m_last->data_len + len);
1628 m->pkt_len = (m->pkt_len + len);
1629 return (char*) tail;
1633 * Remove len bytes at the beginning of an mbuf.
1635 * Returns a pointer to the start address of the new data area. If the
1636 * length is greater than the length of the first segment, then the
1637 * function will fail and return NULL, without modifying the mbuf.
1642 * The amount of data to remove (in bytes).
1644 * A pointer to the new start of the data.
1646 static inline char *rte_pktmbuf_adj(struct rte_mbuf *m, uint16_t len)
1648 __rte_mbuf_sanity_check(m, 1);
1650 if (unlikely(len > m->data_len))
1653 /* NB: elaborating the addition like this instead of using
1654 * += allows us to ensure the result type is uint16_t
1655 * avoiding compiler warnings on gcc 8.1 at least */
1656 m->data_len = (uint16_t)(m->data_len - len);
1657 m->data_off = (uint16_t)(m->data_off + len);
1658 m->pkt_len = (m->pkt_len - len);
1659 return (char *)m->buf_addr + m->data_off;
1663 * Remove len bytes of data at the end of the mbuf.
1665 * If the length is greater than the length of the last segment, the
1666 * function will fail and return -1 without modifying the mbuf.
1671 * The amount of data to remove (in bytes).
1676 static inline int rte_pktmbuf_trim(struct rte_mbuf *m, uint16_t len)
1678 struct rte_mbuf *m_last;
1680 __rte_mbuf_sanity_check(m, 1);
1682 m_last = rte_pktmbuf_lastseg(m);
1683 if (unlikely(len > m_last->data_len))
1686 m_last->data_len = (uint16_t)(m_last->data_len - len);
1687 m->pkt_len = (m->pkt_len - len);
1692 * Test if mbuf data is contiguous.
1697 * - 1, if all data is contiguous (one segment).
1698 * - 0, if there is several segments.
1700 static inline int rte_pktmbuf_is_contiguous(const struct rte_mbuf *m)
1702 __rte_mbuf_sanity_check(m, 1);
1703 return m->nb_segs == 1;
1707 * @internal used by rte_pktmbuf_read().
1709 const void *__rte_pktmbuf_read(const struct rte_mbuf *m, uint32_t off,
1710 uint32_t len, void *buf);
1713 * Read len data bytes in a mbuf at specified offset.
1715 * If the data is contiguous, return the pointer in the mbuf data, else
1716 * copy the data in the buffer provided by the user and return its
1720 * The pointer to the mbuf.
1722 * The offset of the data in the mbuf.
1724 * The amount of bytes to read.
1726 * The buffer where data is copied if it is not contiguous in mbuf
1727 * data. Its length should be at least equal to the len parameter.
1729 * The pointer to the data, either in the mbuf if it is contiguous,
1730 * or in the user buffer. If mbuf is too small, NULL is returned.
1732 static inline const void *rte_pktmbuf_read(const struct rte_mbuf *m,
1733 uint32_t off, uint32_t len, void *buf)
1735 if (likely(off + len <= rte_pktmbuf_data_len(m)))
1736 return rte_pktmbuf_mtod_offset(m, char *, off);
1738 return __rte_pktmbuf_read(m, off, len, buf);
1742 * Chain an mbuf to another, thereby creating a segmented packet.
1744 * Note: The implementation will do a linear walk over the segments to find
1745 * the tail entry. For cases when there are many segments, it's better to
1746 * chain the entries manually.
1749 * The head of the mbuf chain (the first packet)
1751 * The mbuf to put last in the chain
1755 * - -EOVERFLOW, if the chain segment limit exceeded
1757 static inline int rte_pktmbuf_chain(struct rte_mbuf *head, struct rte_mbuf *tail)
1759 struct rte_mbuf *cur_tail;
1761 /* Check for number-of-segments-overflow */
1762 if (head->nb_segs + tail->nb_segs > RTE_MBUF_MAX_NB_SEGS)
1765 /* Chain 'tail' onto the old tail */
1766 cur_tail = rte_pktmbuf_lastseg(head);
1767 cur_tail->next = tail;
1769 /* accumulate number of segments and total length.
1770 * NB: elaborating the addition like this instead of using
1771 * -= allows us to ensure the result type is uint16_t
1772 * avoiding compiler warnings on gcc 8.1 at least */
1773 head->nb_segs = (uint16_t)(head->nb_segs + tail->nb_segs);
1774 head->pkt_len += tail->pkt_len;
1776 /* pkt_len is only set in the head */
1777 tail->pkt_len = tail->data_len;
1784 * @b EXPERIMENTAL: This API may change without prior notice.
1786 * For given input values generate raw tx_offload value.
1787 * Note that it is caller responsibility to make sure that input parameters
1788 * don't exceed maximum bit-field values.
1798 * outer_l3_len value.
1800 * outer_l2_len value.
1804 * raw tx_offload value.
1806 static __rte_always_inline uint64_t
1807 rte_mbuf_tx_offload(uint64_t il2, uint64_t il3, uint64_t il4, uint64_t tso,
1808 uint64_t ol3, uint64_t ol2, uint64_t unused)
1810 return il2 << RTE_MBUF_L2_LEN_OFS |
1811 il3 << RTE_MBUF_L3_LEN_OFS |
1812 il4 << RTE_MBUF_L4_LEN_OFS |
1813 tso << RTE_MBUF_TSO_SEGSZ_OFS |
1814 ol3 << RTE_MBUF_OUTL3_LEN_OFS |
1815 ol2 << RTE_MBUF_OUTL2_LEN_OFS |
1816 unused << RTE_MBUF_TXOFLD_UNUSED_OFS;
1820 * Validate general requirements for Tx offload in mbuf.
1822 * This function checks correctness and completeness of Tx offload settings.
1825 * The packet mbuf to be validated.
1827 * 0 if packet is valid
1830 rte_validate_tx_offload(const struct rte_mbuf *m)
1832 uint64_t ol_flags = m->ol_flags;
1834 /* Does packet set any of available offloads? */
1835 if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
1838 /* IP checksum can be counted only for IPv4 packet */
1839 if ((ol_flags & PKT_TX_IP_CKSUM) && (ol_flags & PKT_TX_IPV6))
1842 /* IP type not set when required */
1843 if (ol_flags & (PKT_TX_L4_MASK | PKT_TX_TCP_SEG))
1844 if (!(ol_flags & (PKT_TX_IPV4 | PKT_TX_IPV6)))
1847 /* Check requirements for TSO packet */
1848 if (ol_flags & PKT_TX_TCP_SEG)
1849 if ((m->tso_segsz == 0) ||
1850 ((ol_flags & PKT_TX_IPV4) &&
1851 !(ol_flags & PKT_TX_IP_CKSUM)))
1854 /* PKT_TX_OUTER_IP_CKSUM set for non outer IPv4 packet. */
1855 if ((ol_flags & PKT_TX_OUTER_IP_CKSUM) &&
1856 !(ol_flags & PKT_TX_OUTER_IPV4))
1863 * @internal used by rte_pktmbuf_linearize().
1865 int __rte_pktmbuf_linearize(struct rte_mbuf *mbuf);
1868 * Linearize data in mbuf.
1870 * This function moves the mbuf data in the first segment if there is enough
1871 * tailroom. The subsequent segments are unchained and freed.
1880 rte_pktmbuf_linearize(struct rte_mbuf *mbuf)
1882 if (rte_pktmbuf_is_contiguous(mbuf))
1884 return __rte_pktmbuf_linearize(mbuf);
1888 * Dump an mbuf structure to a file.
1890 * Dump all fields for the given packet mbuf and all its associated
1891 * segments (in the case of a chained buffer).
1894 * A pointer to a file for output
1898 * If dump_len != 0, also dump the "dump_len" first data bytes of
1901 void rte_pktmbuf_dump(FILE *f, const struct rte_mbuf *m, unsigned dump_len);
1904 * Get the value of mbuf sched queue_id field.
1906 static inline uint32_t
1907 rte_mbuf_sched_queue_get(const struct rte_mbuf *m)
1909 return m->hash.sched.queue_id;
1913 * Get the value of mbuf sched traffic_class field.
1915 static inline uint8_t
1916 rte_mbuf_sched_traffic_class_get(const struct rte_mbuf *m)
1918 return m->hash.sched.traffic_class;
1922 * Get the value of mbuf sched color field.
1924 static inline uint8_t
1925 rte_mbuf_sched_color_get(const struct rte_mbuf *m)
1927 return m->hash.sched.color;
1931 * Get the values of mbuf sched queue_id, traffic_class and color.
1936 * Returns the queue id
1937 * @param traffic_class
1938 * Returns the traffic class id
1940 * Returns the colour id
1943 rte_mbuf_sched_get(const struct rte_mbuf *m, uint32_t *queue_id,
1944 uint8_t *traffic_class,
1947 struct rte_mbuf_sched sched = m->hash.sched;
1949 *queue_id = sched.queue_id;
1950 *traffic_class = sched.traffic_class;
1951 *color = sched.color;
1955 * Set the mbuf sched queue_id to the defined value.
1958 rte_mbuf_sched_queue_set(struct rte_mbuf *m, uint32_t queue_id)
1960 m->hash.sched.queue_id = queue_id;
1964 * Set the mbuf sched traffic_class id to the defined value.
1967 rte_mbuf_sched_traffic_class_set(struct rte_mbuf *m, uint8_t traffic_class)
1969 m->hash.sched.traffic_class = traffic_class;
1973 * Set the mbuf sched color id to the defined value.
1976 rte_mbuf_sched_color_set(struct rte_mbuf *m, uint8_t color)
1978 m->hash.sched.color = color;
1982 * Set the mbuf sched queue_id, traffic_class and color.
1987 * Queue id value to be set
1988 * @param traffic_class
1989 * Traffic class id value to be set
1991 * Color id to be set
1994 rte_mbuf_sched_set(struct rte_mbuf *m, uint32_t queue_id,
1995 uint8_t traffic_class,
1998 m->hash.sched = (struct rte_mbuf_sched){
1999 .queue_id = queue_id,
2000 .traffic_class = traffic_class,
2010 #endif /* _RTE_MBUF_H_ */