xref: /dpdk/doc/guides/prog_guide/cryptodev_lib.rst (revision 8711af290f353f727989684de2e75c7c41d2779a)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright(c) 2016-2020 Intel Corporation.
3
4Cryptography Device Library
5===========================
6
7The cryptodev library provides a Crypto device framework for management and
8provisioning of hardware and software Crypto poll mode drivers, defining generic
9APIs which support a number of different Crypto operations. The framework
10currently only supports cipher, authentication, chained cipher/authentication
11and AEAD symmetric and asymmetric Crypto operations.
12
13The usages in security protocols are discussed in the :doc:`../howto/security` guide.
14
15Design Principles
16-----------------
17
18The cryptodev library follows the same basic principles as those used in DPDK's
19Ethernet Device framework. The Crypto framework provides a generic Crypto device
20framework which supports both physical (hardware) and virtual (software) Crypto
21devices as well as a generic Crypto API which allows Crypto devices to be
22managed and configured and supports Crypto operations to be provisioned on
23Crypto poll mode driver.
24
25
26Device Management
27-----------------
28
29Device Creation
30~~~~~~~~~~~~~~~
31
32Physical Crypto devices are discovered during the PCI probe/enumeration of the
33EAL function which is executed at DPDK initialization, based on
34their PCI device identifier, each unique PCI BDF (bus/bridge, device,
35function). Specific physical Crypto devices, like other physical devices in DPDK
36can be listed using the EAL command line options.
37
38Virtual devices can be created by two mechanisms, either using the EAL command
39line options or from within the application using an EAL API directly.
40
41From the command line using the --vdev EAL option
42
43.. code-block:: console
44
45   --vdev  'crypto_aesni_mb0,max_nb_queue_pairs=2,socket_id=0'
46
47.. Note::
48
49   * If DPDK application requires multiple software crypto PMD devices then required
50     number of ``--vdev`` with appropriate libraries are to be added.
51
52   * An Application with crypto PMD instances sharing the same library requires unique ID.
53
54   Example: ``--vdev  'crypto_aesni_mb0' --vdev  'crypto_aesni_mb1'``
55
56Or using the rte_vdev_init API within the application code.
57
58.. code-block:: c
59
60   rte_vdev_init("crypto_aesni_mb",
61                     "max_nb_queue_pairs=2,socket_id=0")
62
63All virtual Crypto devices support the following initialization parameters:
64
65* ``max_nb_queue_pairs`` - maximum number of queue pairs supported by the device.
66* ``socket_id`` - socket on which to allocate the device resources on.
67
68
69Device Identification
70~~~~~~~~~~~~~~~~~~~~~
71
72Each device, whether virtual or physical is uniquely designated by two
73identifiers:
74
75- A unique device index used to designate the Crypto device in all functions
76  exported by the cryptodev API.
77
78- A device name used to designate the Crypto device in console messages, for
79  administration or debugging purposes. For ease of use, the port name includes
80  the port index.
81
82
83Device Configuration
84~~~~~~~~~~~~~~~~~~~~
85
86The configuration of each Crypto device includes the following operations:
87
88- Allocation of resources, including hardware resources if a physical device.
89- Resetting the device into a well-known default state.
90- Initialization of statistics counters.
91
92The rte_cryptodev_configure API is used to configure a Crypto device.
93
94.. code-block:: c
95
96   int rte_cryptodev_configure(uint8_t dev_id,
97                               struct rte_cryptodev_config *config)
98
99The ``rte_cryptodev_config`` structure is used to pass the configuration
100parameters for socket selection and number of queue pairs.
101
102.. literalinclude:: ../../../lib/cryptodev/rte_cryptodev.h
103   :language: c
104   :start-after: Structure rte_cryptodev_config 8<
105   :end-before: >8 End of structure rte_cryptodev_config.
106
107
108Configuration of Queue Pairs
109~~~~~~~~~~~~~~~~~~~~~~~~~~~~
110
111Each Crypto devices queue pair is individually configured through the
112``rte_cryptodev_queue_pair_setup`` API.
113Each queue pairs resources may be allocated on a specified socket.
114
115.. code-block:: c
116
117    int rte_cryptodev_queue_pair_setup(uint8_t dev_id, uint16_t queue_pair_id,
118                const struct rte_cryptodev_qp_conf *qp_conf,
119                int socket_id)
120
121
122.. literalinclude:: ../../../lib/cryptodev/rte_cryptodev.h
123   :language: c
124   :start-after: Structure rte_cryptodev_qp_conf 8<
125   :end-before: >8 End of structure rte_cryptodev_qp_conf.
126
127
128The field ``mp_session`` is used for creating temporary session to process
129the crypto operations in the session-less mode.
130They can be the same other different mempools. Please note not all Cryptodev
131PMDs supports session-less mode.
132
133
134Logical Cores, Memory and Queues Pair Relationships
135~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
136
137The Crypto device Library as the Poll Mode Driver library support NUMA for when
138a processor’s logical cores and interfaces utilize its local memory. Therefore
139Crypto operations, and in the case of symmetric Crypto operations, the session
140and the mbuf being operated on, should be allocated from memory pools created
141in the local memory. The buffers should, if possible, remain on the local
142processor to obtain the best performance results and buffer descriptors should
143be populated with mbufs allocated from a mempool allocated from local memory.
144
145The run-to-completion model also performs better, especially in the case of
146virtual Crypto devices, if the Crypto operation and session and data buffer is
147in local memory instead of a remote processor's memory. This is also true for
148the pipe-line model provided all logical cores used are located on the same
149processor.
150
151Multiple logical cores should never share the same queue pair for enqueuing
152operations or dequeuing operations on the same Crypto device since this would
153require global locks and hinder performance. It is however possible to use a
154different logical core to dequeue an operation on a queue pair from the logical
155core which it was enqueued on. This means that a crypto burst enqueue/dequeue
156APIs are a logical place to transition from one logical core to another in a
157packet processing pipeline.
158
159
160Device Features and Capabilities
161---------------------------------
162
163Crypto devices define their functionality through two mechanisms, global device
164features and algorithm capabilities. Global devices features identify device
165wide level features which are applicable to the whole device such as
166the device having hardware acceleration or supporting symmetric and/or asymmetric
167Crypto operations.
168
169The capabilities mechanism defines the individual algorithms/functions which
170the device supports, such as a specific symmetric Crypto cipher,
171authentication operation or Authenticated Encryption with Associated Data
172(AEAD) operation.
173
174
175Device Features
176~~~~~~~~~~~~~~~
177
178Currently the following Crypto device features are defined:
179
180* Symmetric Crypto operations
181* Asymmetric Crypto operations
182* Chaining of symmetric Crypto operations
183* SSE accelerated SIMD vector operations
184* AVX accelerated SIMD vector operations
185* AVX2 accelerated SIMD vector operations
186* AESNI accelerated instructions
187* Hardware off-load processing
188
189
190Device Operation Capabilities
191~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
192
193Crypto capabilities which identify particular algorithm which the Crypto PMD
194supports are  defined by the operation type, the operation transform, the
195transform identifier and then the particulars of the transform. For the full
196scope of the Crypto capability see the definition of the structure in the
197*DPDK API Reference*.
198
199.. code-block:: c
200
201   struct rte_cryptodev_capabilities;
202
203Each Crypto poll mode driver defines its own private array of capabilities
204for the operations it supports. Below is an example of the capabilities for a
205PMD which supports the authentication algorithm SHA1_HMAC and the cipher
206algorithm AES_CBC.
207
208.. code-block:: c
209
210    static const struct rte_cryptodev_capabilities pmd_capabilities[] = {
211        {    /* SHA1 HMAC */
212            .op = RTE_CRYPTO_OP_TYPE_SYMMETRIC,
213            .sym = {
214                .xform_type = RTE_CRYPTO_SYM_XFORM_AUTH,
215                .auth = {
216                    .algo = RTE_CRYPTO_AUTH_SHA1_HMAC,
217                    .block_size = 64,
218                    .key_size = {
219                        .min = 64,
220                        .max = 64,
221                        .increment = 0
222                    },
223                    .digest_size = {
224                        .min = 12,
225                        .max = 12,
226                        .increment = 0
227                    },
228                    .aad_size = { 0 },
229                    .iv_size = { 0 }
230                }
231            }
232        },
233        {    /* AES CBC */
234            .op = RTE_CRYPTO_OP_TYPE_SYMMETRIC,
235            .sym = {
236                .xform_type = RTE_CRYPTO_SYM_XFORM_CIPHER,
237                .cipher = {
238                    .algo = RTE_CRYPTO_CIPHER_AES_CBC,
239                    .block_size = 16,
240                    .key_size = {
241                        .min = 16,
242                        .max = 32,
243                        .increment = 8
244                    },
245                    .iv_size = {
246                        .min = 16,
247                        .max = 16,
248                        .increment = 0
249                    }
250                }
251            }
252        }
253    }
254
255
256Capabilities Discovery
257~~~~~~~~~~~~~~~~~~~~~~
258
259Discovering the features and capabilities of a Crypto device poll mode driver
260is achieved through the ``rte_cryptodev_info_get`` function.
261
262.. code-block:: c
263
264   void rte_cryptodev_info_get(uint8_t dev_id,
265                               struct rte_cryptodev_info *dev_info);
266
267This allows the user to query a specific Crypto PMD and get all the device
268features and capabilities. The ``rte_cryptodev_info`` structure contains all the
269relevant information for the device.
270
271.. literalinclude:: ../../../lib/cryptodev/rte_cryptodev.h
272   :language: c
273   :start-after: Structure rte_cryptodev_info 8<
274   :end-before: >8 End of structure rte_cryptodev_info.
275
276
277Operation Processing
278--------------------
279
280Scheduling of Crypto operations on DPDK's application data path is
281performed using a burst oriented asynchronous API set. A queue pair on a Crypto
282device accepts a burst of Crypto operations using enqueue burst API. On physical
283Crypto devices the enqueue burst API will place the operations to be processed
284on the devices hardware input queue, for virtual devices the processing of the
285Crypto operations is usually completed during the enqueue call to the Crypto
286device. The dequeue burst API will retrieve any processed operations available
287from the queue pair on the Crypto device, from physical devices this is usually
288directly from the devices processed queue, and for virtual device's from a
289``rte_ring`` where processed operations are placed after being processed on the
290enqueue call.
291
292
293Private data
294~~~~~~~~~~~~
295For session-based operations, the set and get API provides a mechanism for an
296application to store and retrieve the private user data information stored along
297with the crypto session.
298
299For example, suppose an application is submitting a crypto operation with a session
300associated and wants to indicate private user data information which is required to be
301used after completion of the crypto operation. In this case, the application can use
302the set API to set the user data and retrieve it using get API.
303
304.. code-block:: c
305
306	int rte_cryptodev_sym_session_set_user_data(
307		struct rte_cryptodev_sym_session *sess,	void *data, uint16_t size);
308
309	void * rte_cryptodev_sym_session_get_user_data(
310		struct rte_cryptodev_sym_session *sess);
311
312Please note the ``size`` passed to set API cannot be bigger than the predefined
313``user_data_sz`` when creating the session header mempool, otherwise the
314function will return error. Also when ``user_data_sz`` was defined as ``0`` when
315creating the session header mempool, the get API will always return ``NULL``.
316
317For session-less mode, the private user data information can be placed along with the
318``struct rte_crypto_op``. The ``rte_crypto_op::private_data_offset`` indicates the
319start of private data information. The offset is counted from the start of the
320rte_crypto_op including other crypto information such as the IVs (since there can
321be an IV also for authentication).
322
323User callback APIs
324~~~~~~~~~~~~~~~~~~
325The add APIs configures a user callback function to be called for each burst of crypto
326ops received/sent on a given crypto device queue pair. The return value is a pointer
327that can be used later to remove the callback using remove API. Application is expected
328to register a callback function of type ``rte_cryptodev_callback_fn``. Multiple callback
329functions can be added for a given queue pair. API does not restrict on maximum number of
330callbacks.
331
332Callbacks registered by application would not survive ``rte_cryptodev_configure`` as it
333reinitializes the callback list. It is user responsibility to remove all installed
334callbacks before calling ``rte_cryptodev_configure`` to avoid possible memory leakage.
335
336So, the application is expected to add user callback after ``rte_cryptodev_configure``.
337The callbacks can also be added at the runtime. These callbacks get executed when
338``rte_cryptodev_enqueue_burst``/``rte_cryptodev_dequeue_burst`` is called.
339
340.. code-block:: c
341
342	struct rte_cryptodev_cb *
343		rte_cryptodev_add_enq_callback(uint8_t dev_id, uint16_t qp_id,
344					       rte_cryptodev_callback_fn cb_fn,
345					       void *cb_arg);
346
347	struct rte_cryptodev_cb *
348		rte_cryptodev_add_deq_callback(uint8_t dev_id, uint16_t qp_id,
349					       rte_cryptodev_callback_fn cb_fn,
350					       void *cb_arg);
351
352	uint16_t (* rte_cryptodev_callback_fn)(uint16_t dev_id, uint16_t qp_id,
353					       struct rte_crypto_op **ops,
354					       uint16_t nb_ops, void *user_param);
355
356The remove API removes a callback function added by
357``rte_cryptodev_add_enq_callback``/``rte_cryptodev_add_deq_callback``.
358
359.. code-block:: c
360
361	int rte_cryptodev_remove_enq_callback(uint8_t dev_id, uint16_t qp_id,
362					      struct rte_cryptodev_cb *cb);
363
364	int rte_cryptodev_remove_deq_callback(uint8_t dev_id, uint16_t qp_id,
365					      struct rte_cryptodev_cb *cb);
366
367
368Enqueue / Dequeue Burst APIs
369~~~~~~~~~~~~~~~~~~~~~~~~~~~~
370
371The burst enqueue API uses a Crypto device identifier and a queue pair
372identifier to specify the Crypto device queue pair to schedule the processing on.
373The ``nb_ops`` parameter is the number of operations to process which are
374supplied in the ``ops`` array of ``rte_crypto_op`` structures.
375The enqueue function returns the number of operations it actually enqueued for
376processing, a return value equal to ``nb_ops`` means that all packets have been
377enqueued.
378
379.. code-block:: c
380
381   uint16_t rte_cryptodev_enqueue_burst(uint8_t dev_id, uint16_t qp_id,
382                                        struct rte_crypto_op **ops, uint16_t nb_ops)
383
384The dequeue API uses the same format as the enqueue API of processed but
385the ``nb_ops`` and ``ops`` parameters are now used to specify the max processed
386operations the user wishes to retrieve and the location in which to store them.
387The API call returns the actual number of processed operations returned, this
388can never be larger than ``nb_ops``.
389
390.. code-block:: c
391
392   uint16_t rte_cryptodev_dequeue_burst(uint8_t dev_id, uint16_t qp_id,
393                                        struct rte_crypto_op **ops, uint16_t nb_ops)
394
395
396Operation Representation
397~~~~~~~~~~~~~~~~~~~~~~~~
398
399An Crypto operation is represented by an rte_crypto_op structure, which is a
400generic metadata container for all necessary information required for the
401Crypto operation to be processed on a particular Crypto device poll mode driver.
402
403.. figure:: img/crypto_op.*
404
405The operation structure includes the operation type, the operation status
406and the session type (session-based/less), a reference to the operation
407specific data, which can vary in size and content depending on the operation
408being provisioned. It also contains the source mempool for the operation,
409if it allocated from a mempool.
410
411If Crypto operations are allocated from a Crypto operation mempool, see next
412section, there is also the ability to allocate private memory with the
413operation for applications purposes.
414
415Application software is responsible for specifying all the operation specific
416fields in the ``rte_crypto_op`` structure which are then used by the Crypto PMD
417to process the requested operation.
418
419
420Operation Management and Allocation
421~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
422
423The cryptodev library provides an API set for managing Crypto operations which
424utilize the Mempool Library to allocate operation buffers. Therefore, it ensures
425that the crypto operation is interleaved optimally across the channels and
426ranks for optimal processing.
427A ``rte_crypto_op`` contains a field indicating the pool that it originated from.
428When calling ``rte_crypto_op_free(op)``, the operation returns to its original pool.
429
430.. code-block:: c
431
432   extern struct rte_mempool *
433   rte_crypto_op_pool_create(const char *name, enum rte_crypto_op_type type,
434                             unsigned nb_elts, unsigned cache_size, uint16_t priv_size,
435                             int socket_id);
436
437During pool creation ``rte_crypto_op_init()`` is called as a constructor to
438initialize each Crypto operation which subsequently calls
439``__rte_crypto_op_reset()`` to configure any operation type specific fields based
440on the type parameter.
441
442
443``rte_crypto_op_alloc()`` and ``rte_crypto_op_bulk_alloc()`` are used to allocate
444Crypto operations of a specific type from a given Crypto operation mempool.
445``__rte_crypto_op_reset()`` is called on each operation before being returned to
446allocate to a user so the operation is always in a good known state before use
447by the application.
448
449.. code-block:: c
450
451   struct rte_crypto_op *rte_crypto_op_alloc(struct rte_mempool *mempool,
452                                             enum rte_crypto_op_type type)
453
454   unsigned rte_crypto_op_bulk_alloc(struct rte_mempool *mempool,
455                                     enum rte_crypto_op_type type,
456                                     struct rte_crypto_op **ops, uint16_t nb_ops)
457
458``rte_crypto_op_free()`` is called by the application to return an operation to
459its allocating pool.
460
461.. code-block:: c
462
463   void rte_crypto_op_free(struct rte_crypto_op *op)
464
465
466Symmetric Cryptography Support
467------------------------------
468
469The cryptodev library currently provides support for the following symmetric
470Crypto operations; cipher, authentication, including chaining of these
471operations, as well as also supporting AEAD operations.
472
473
474Session and Session Management
475~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
476
477Sessions are used in symmetric cryptographic processing to store the immutable
478data defined in a cryptographic transform which is used in the operation
479processing of a packet flow. Sessions are used to manage information such as
480expand cipher keys and HMAC IPADs and OPADs, which need to be calculated for a
481particular Crypto operation, but are immutable on a packet to packet basis for
482a flow. Crypto sessions cache this immutable data in a optimal way for the
483underlying PMD and this allows further acceleration of the offload of
484Crypto workloads.
485
486The Crypto device framework provides APIs to create session mempool and allocate
487and initialize sessions for crypto devices, where sessions are mempool objects.
488The application has to use ``rte_cryptodev_sym_session_pool_create()`` to
489create the session mempool header and the private data with the size specified
490by the user through the ``elt_size`` parameter in the function.
491The session private data is for the driver to initialize and access
492during crypto operations, hence the ``elt_size`` should be big enough
493for all drivers that will share this mempool.
494To obtain the proper session private data size of a crypto device,
495the user can call ``rte_cryptodev_sym_get_private_session_size()`` function.
496In case of heterogeneous crypto devices which will share the same session mempool,
497the maximum session private data size of them should be passed.
498
499Once the session mempools have been created, ``rte_cryptodev_sym_session_create()``
500is used to allocate and initialize the session from the given mempool.
501The created session can ONLY be used by the crypto devices sharing the same driver ID
502as the device ID passed into the function as the parameter.
503In addition, a symmetric transform chain is used to specify the operation and its parameters.
504See the section below for details on transforms.
505
506When a session is no longer used, user must call ``rte_cryptodev_sym_session_free()``
507to uninitialize the session data and return the session
508back to the mempool it belongs.
509
510
511Transforms and Transform Chaining
512~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
513
514Symmetric Crypto transforms (``rte_crypto_sym_xform``) are the mechanism used
515to specify the details of the Crypto operation. For chaining of symmetric
516operations such as cipher encrypt and authentication generate, the next pointer
517allows transform to be chained together. Crypto devices which support chaining
518must publish the chaining of symmetric Crypto operations feature flag. Allocation of the
519xform structure is in the application domain. To allow future API extensions in a
520backwardly compatible manner, e.g. addition of a new parameter, the application should
521zero the full xform struct before populating it.
522
523Currently there are three transforms types cipher, authentication and AEAD.
524Also it is important to note that the order in which the
525transforms are passed indicates the order of the chaining.
526
527.. literalinclude:: ../../../lib/cryptodev/rte_crypto_sym.h
528   :language: c
529   :start-after: Structure rte_crypto_sym_xform 8<
530   :end-before: >8 End of structure rte_crypto_sym_xform.
531
532The API does not place a limit on the number of transforms that can be chained
533together but this will be limited by the underlying Crypto device poll mode
534driver which is processing the operation.
535
536.. figure:: img/crypto_xform_chain.*
537
538
539Symmetric Operations
540~~~~~~~~~~~~~~~~~~~~
541
542The symmetric Crypto operation structure contains all the mutable data relating
543to performing symmetric cryptographic processing on a referenced mbuf data
544buffer. It is used for either cipher, authentication, AEAD and chained
545operations.
546
547As a minimum the symmetric operation must have a source data buffer (``m_src``),
548a valid session (or transform chain if in session-less mode) and the minimum
549authentication/ cipher/ AEAD parameters required depending on the type of operation
550specified in the session or the transform
551chain.
552
553.. literalinclude:: ../../../lib/cryptodev/rte_crypto_sym.h
554   :language: c
555   :start-after: Structure rte_crypto_sym_op 8<
556   :end-before: >8 End of structure rte_crypto_sym_op.
557
558
559Synchronous mode
560----------------
561
562Some cryptodevs support synchronous mode alongside with a standard asynchronous
563mode. In that case operations are performed directly when calling
564``rte_cryptodev_sym_cpu_crypto_process`` method instead of enqueuing and
565dequeuing an operation before. This mode of operation allows cryptodevs which
566utilize CPU cryptographic acceleration to have significant performance boost
567comparing to standard asynchronous approach. Cryptodevs supporting synchronous
568mode have ``RTE_CRYPTODEV_FF_SYM_CPU_CRYPTO`` feature flag set.
569
570To perform a synchronous operation a call to
571``rte_cryptodev_sym_cpu_crypto_process`` has to be made with vectorized
572operation descriptor (``struct rte_crypto_sym_vec``) containing:
573
574- ``num`` - number of operations to perform,
575- pointer to an array of size ``num`` containing a scatter-gather list
576  descriptors of performed operations (``struct rte_crypto_sgl``). Each instance
577  of ``struct rte_crypto_sgl`` consists of a number of segments and a pointer to
578  an array of segment descriptors ``struct rte_crypto_vec``;
579- pointers to arrays of size ``num`` containing IV, AAD and digest information
580  in the ``cpu_crypto`` sub-structure,
581- pointer to an array of size ``num`` where status information will be stored
582  for each operation.
583
584Function returns a number of successfully completed operations and sets
585appropriate status number for each operation in the status array provided as
586a call argument. Status different than zero must be treated as error.
587
588For more details, e.g. how to convert an mbuf to an SGL, please refer to an
589example usage in the IPsec library implementation.
590
591Cryptodev Raw Data-path APIs
592~~~~~~~~~~~~~~~~~~~~~~~~~~~~
593
594The Crypto Raw data-path APIs are a set of APIs designed to enable external
595libraries/applications to leverage the cryptographic processing provided by
596DPDK crypto PMDs through the cryptodev API but in a manner that is not
597dependent on native DPDK data structures (eg. rte_mbuf, rte_crypto_op, ... etc)
598in their data-path implementation.
599
600The raw data-path APIs have the following advantages:
601
602- External data structure friendly design. The new APIs uses the operation
603  descriptor ``struct rte_crypto_sym_vec`` that supports raw data pointer and
604  IOVA addresses as input. Moreover, the APIs does not require the user to
605  allocate the descriptor from mempool, nor requiring mbufs to describe input
606  data's virtual and IOVA addresses. All these features made the translation
607  from user's own data structure into the descriptor easier and more efficient.
608
609- Flexible enqueue and dequeue operation. The raw data-path APIs gives the
610  user more control to the enqueue and dequeue operations, including the
611  capability of precious enqueue/dequeue count, abandoning enqueue or dequeue
612  at any time, and operation status translation and set on the fly.
613
614Cryptodev PMDs which support the raw data-path APIs will have
615``RTE_CRYPTODEV_FF_SYM_RAW_DP`` feature flag presented. To use this feature,
616the user shall create a local ``struct rte_crypto_raw_dp_ctx`` buffer and
617extend to at least the length returned by ``rte_cryptodev_get_raw_dp_ctx_size``
618function call. The created buffer is then initialized using
619``rte_cryptodev_configure_raw_dp_ctx`` function with the ``is_update``
620parameter as 0. The library and the crypto device driver will then set the
621buffer and attach either the cryptodev sym session, the rte_security session,
622or the cryptodev xform for session-less operation into the ctx buffer, and
623set the corresponding enqueue and dequeue function handlers based on the
624algorithm information stored in the session or xform. When the ``is_update``
625parameter passed into ``rte_cryptodev_configure_raw_dp_ctx`` is 1, the driver
626will not initialize the buffer but only update the session or xform and
627the function handlers accordingly.
628
629After the ``struct rte_crypto_raw_dp_ctx`` buffer is initialized, it is now
630ready for enqueue and dequeue operation. There are two different enqueue
631functions: ``rte_cryptodev_raw_enqueue`` to enqueue single raw data
632operation, and ``rte_cryptodev_raw_enqueue_burst`` to enqueue a descriptor
633with multiple operations. In case of the application uses similar approach to
634``struct rte_crypto_sym_vec`` to manage its data burst but with different
635data structure, using the ``rte_cryptodev_raw_enqueue_burst`` function may be
636less efficient as this is a situation where the application has to loop over
637all crypto operations to assemble the ``struct rte_crypto_sym_vec`` descriptor
638from its own data structure, and then the driver will loop over them again to
639translate every operation in the descriptor to the driver's specific queue data.
640The ``rte_cryptodev_raw_enqueue`` should be used to save one loop for each data
641burst instead.
642
643The ``rte_cryptodev_raw_enqueue`` and ``rte_cryptodev_raw_enqueue_burst``
644functions will return or set the enqueue status. ``rte_cryptodev_raw_enqueue``
645will return the status directly, ``rte_cryptodev_raw_enqueue_burst`` will
646return the number of operations enqueued or stored (explained as follows) and
647set the ``enqueue_status`` buffer provided by the user. The possible
648enqueue status values are:
649
650- ``1``: the operation(s) is/are enqueued successfully.
651- ``0``: the operation(s) is/are cached successfully in the crypto device queue
652  but is not actually enqueued. The user shall call
653  ``rte_cryptodev_raw_enqueue_done`` function after the expected operations
654  are stored. The crypto device will then start enqueuing all of them at
655  once.
656- The negative integer: error occurred during enqueue.
657
658Calling ``rte_cryptodev_configure_raw_dp_ctx`` with the parameter ``is_update``
659set as 0 twice without the enqueue function returning or setting enqueue status
660to 1 or ``rte_cryptodev_raw_enqueue_done`` function being called in between will
661invalidate any operation stored in the device queue but not enqueued. This
662feature is useful when the user wants to abandon partially enqueued operations
663for a failed enqueue burst operation and try enqueuing in a whole later.
664
665Similar as enqueue, there are two dequeue functions:
666``rte_cryptodev_raw_dequeue`` for dequeuing single operation, and
667``rte_cryptodev_raw_dequeue_burst`` for dequeuing a burst of operations (e.g.
668all operations in a ``struct rte_crypto_sym_vec`` descriptor). The
669``rte_cryptodev_raw_dequeue_burst`` function allows the user to provide callback
670functions to retrieve dequeue count from the enqueued user data and write the
671expected status value to the user data on the fly. The dequeue functions also
672set the dequeue status:
673
674- ``1``: the operation(s) is/are dequeued successfully.
675- ``0``: the operation(s) is/are completed but is not actually dequeued (hence
676  still kept in the device queue). The user shall call the
677  ``rte_cryptodev_raw_dequeue_done`` function after the expected number of
678  operations (e.g. all operations in a descriptor) are dequeued. The crypto
679  device driver will then free them from the queue at once.
680- The negative integer: error occurred during dequeue.
681
682Calling ``rte_cryptodev_configure_raw_dp_ctx`` with the parameter ``is_update``
683set as 0 twice without the dequeue functions execution changed dequeue_status
684to 1 or ``rte_cryptodev_raw_dequeue_done`` function being called in between will
685revert the crypto device queue's dequeue effort to the moment when the
686``struct rte_crypto_raw_dp_ctx`` buffer is initialized. This feature is useful
687when the user wants to abandon partially dequeued data and try dequeuing again
688later in a whole.
689
690There are a few limitations to the raw data path APIs:
691
692* Only support in-place operations.
693* APIs are NOT thread-safe.
694* CANNOT mix the raw data-path API's enqueue with rte_cryptodev_enqueue_burst,
695  or vice versa.
696
697See *DPDK API Reference* for details on each API definitions.
698
699Sample code
700-----------
701
702There are various sample applications that show how to use the cryptodev library,
703such as the L2fwd with Crypto sample application (L2fwd-crypto) and
704the IPsec Security Gateway application (ipsec-secgw).
705
706While these applications demonstrate how an application can be created to perform
707generic crypto operation, the required complexity hides the basic steps of
708how to use the cryptodev APIs.
709
710The following sample code shows the basic steps to encrypt several buffers
711with AES-CBC (although performing other crypto operations is similar),
712using one of the crypto PMDs available in DPDK.
713
714.. code-block:: c
715
716    /*
717     * Simple example to encrypt several buffers with AES-CBC using
718     * the Cryptodev APIs.
719     */
720
721    #define MAX_SESSIONS         1024
722    #define NUM_MBUFS            1024
723    #define POOL_CACHE_SIZE      128
724    #define BURST_SIZE           32
725    #define BUFFER_SIZE          1024
726    #define AES_CBC_IV_LENGTH    16
727    #define AES_CBC_KEY_LENGTH   16
728    #define IV_OFFSET            (sizeof(struct rte_crypto_op) + \
729                                 sizeof(struct rte_crypto_sym_op))
730
731    struct rte_mempool *mbuf_pool, *crypto_op_pool;
732    struct rte_mempool *session_pool, *session_priv_pool;
733    unsigned int session_size;
734    int ret;
735
736    /* Initialize EAL. */
737    ret = rte_eal_init(argc, argv);
738    if (ret < 0)
739        rte_exit(EXIT_FAILURE, "Invalid EAL arguments\n");
740
741    uint8_t socket_id = rte_socket_id();
742
743    /* Create the mbuf pool. */
744    mbuf_pool = rte_pktmbuf_pool_create("mbuf_pool",
745                                    NUM_MBUFS,
746                                    POOL_CACHE_SIZE,
747                                    0,
748                                    RTE_MBUF_DEFAULT_BUF_SIZE,
749                                    socket_id);
750    if (mbuf_pool == NULL)
751        rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
752
753    /*
754     * The IV is always placed after the crypto operation,
755     * so some private data is required to be reserved.
756     */
757    unsigned int crypto_op_private_data = AES_CBC_IV_LENGTH;
758
759    /* Create crypto operation pool. */
760    crypto_op_pool = rte_crypto_op_pool_create("crypto_op_pool",
761                                            RTE_CRYPTO_OP_TYPE_SYMMETRIC,
762                                            NUM_MBUFS,
763                                            POOL_CACHE_SIZE,
764                                            crypto_op_private_data,
765                                            socket_id);
766    if (crypto_op_pool == NULL)
767        rte_exit(EXIT_FAILURE, "Cannot create crypto op pool\n");
768
769    /* Create the virtual crypto device. */
770    char args[128];
771    const char *crypto_name = "crypto_aesni_mb0";
772    snprintf(args, sizeof(args), "socket_id=%d", socket_id);
773    ret = rte_vdev_init(crypto_name, args);
774    if (ret != 0)
775        rte_exit(EXIT_FAILURE, "Cannot create virtual device");
776
777    uint8_t cdev_id = rte_cryptodev_get_dev_id(crypto_name);
778
779    /* Get private session data size. */
780    session_size = rte_cryptodev_sym_get_private_session_size(cdev_id);
781
782    #ifdef USE_TWO_MEMPOOLS
783    /* Create session mempool for the session header. */
784    session_pool = rte_cryptodev_sym_session_pool_create("session_pool",
785                                    MAX_SESSIONS,
786                                    0,
787                                    POOL_CACHE_SIZE,
788                                    0,
789                                    socket_id);
790
791    /*
792     * Create session private data mempool for the
793     * private session data for the crypto device.
794     */
795    session_priv_pool = rte_mempool_create("session_pool",
796                                    MAX_SESSIONS,
797                                    session_size,
798                                    POOL_CACHE_SIZE,
799                                    0, NULL, NULL, NULL,
800                                    NULL, socket_id,
801                                    0);
802
803    #else
804    /* Use of the same mempool for session header and private data */
805	session_pool = rte_cryptodev_sym_session_pool_create("session_pool",
806                                    MAX_SESSIONS * 2,
807                                    session_size,
808                                    POOL_CACHE_SIZE,
809                                    0,
810                                    socket_id);
811
812	session_priv_pool = session_pool;
813
814    #endif
815
816    /* Configure the crypto device. */
817    struct rte_cryptodev_config conf = {
818        .nb_queue_pairs = 1,
819        .socket_id = socket_id
820    };
821
822    struct rte_cryptodev_qp_conf qp_conf = {
823        .nb_descriptors = 2048,
824        .mp_session = session_pool,
825        .mp_session_private = session_priv_pool
826    };
827
828    if (rte_cryptodev_configure(cdev_id, &conf) < 0)
829        rte_exit(EXIT_FAILURE, "Failed to configure cryptodev %u", cdev_id);
830
831    if (rte_cryptodev_queue_pair_setup(cdev_id, 0, &qp_conf, socket_id) < 0)
832        rte_exit(EXIT_FAILURE, "Failed to setup queue pair\n");
833
834    if (rte_cryptodev_start(cdev_id) < 0)
835        rte_exit(EXIT_FAILURE, "Failed to start device\n");
836
837    /* Create the crypto transform. */
838    uint8_t cipher_key[16] = {0};
839    struct rte_crypto_sym_xform cipher_xform = {
840        .next = NULL,
841        .type = RTE_CRYPTO_SYM_XFORM_CIPHER,
842        .cipher = {
843            .op = RTE_CRYPTO_CIPHER_OP_ENCRYPT,
844            .algo = RTE_CRYPTO_CIPHER_AES_CBC,
845            .key = {
846                .data = cipher_key,
847                .length = AES_CBC_KEY_LENGTH
848            },
849            .iv = {
850                .offset = IV_OFFSET,
851                .length = AES_CBC_IV_LENGTH
852            }
853        }
854    };
855
856    /* Create crypto session and initialize it for the crypto device. */
857    struct rte_cryptodev_sym_session *session;
858    session = rte_cryptodev_sym_session_create(cdev_id, &cipher_xform,
859                    session_pool);
860    if (session == NULL)
861        rte_exit(EXIT_FAILURE, "Session could not be created\n");
862
863    /* Get a burst of crypto operations. */
864    struct rte_crypto_op *crypto_ops[BURST_SIZE];
865    if (rte_crypto_op_bulk_alloc(crypto_op_pool,
866                            RTE_CRYPTO_OP_TYPE_SYMMETRIC,
867                            crypto_ops, BURST_SIZE) == 0)
868        rte_exit(EXIT_FAILURE, "Not enough crypto operations available\n");
869
870    /* Get a burst of mbufs. */
871    struct rte_mbuf *mbufs[BURST_SIZE];
872    if (rte_pktmbuf_alloc_bulk(mbuf_pool, mbufs, BURST_SIZE) < 0)
873        rte_exit(EXIT_FAILURE, "Not enough mbufs available");
874
875    /* Initialize the mbufs and append them to the crypto operations. */
876    unsigned int i;
877    for (i = 0; i < BURST_SIZE; i++) {
878        if (rte_pktmbuf_append(mbufs[i], BUFFER_SIZE) == NULL)
879            rte_exit(EXIT_FAILURE, "Not enough room in the mbuf\n");
880        crypto_ops[i]->sym->m_src = mbufs[i];
881    }
882
883    /* Set up the crypto operations. */
884    for (i = 0; i < BURST_SIZE; i++) {
885        struct rte_crypto_op *op = crypto_ops[i];
886        /* Modify bytes of the IV at the end of the crypto operation */
887        uint8_t *iv_ptr = rte_crypto_op_ctod_offset(op, uint8_t *,
888                                                IV_OFFSET);
889
890        generate_random_bytes(iv_ptr, AES_CBC_IV_LENGTH);
891
892        op->sym->cipher.data.offset = 0;
893        op->sym->cipher.data.length = BUFFER_SIZE;
894
895        /* Attach the crypto session to the operation */
896        rte_crypto_op_attach_sym_session(op, session);
897    }
898
899    /* Enqueue the crypto operations in the crypto device. */
900    uint16_t num_enqueued_ops = rte_cryptodev_enqueue_burst(cdev_id, 0,
901                                            crypto_ops, BURST_SIZE);
902
903    /*
904     * Dequeue the crypto operations until all the operations
905     * are processed in the crypto device.
906     */
907    uint16_t num_dequeued_ops, total_num_dequeued_ops = 0;
908    do {
909        struct rte_crypto_op *dequeued_ops[BURST_SIZE];
910        num_dequeued_ops = rte_cryptodev_dequeue_burst(cdev_id, 0,
911                                        dequeued_ops, BURST_SIZE);
912        total_num_dequeued_ops += num_dequeued_ops;
913
914        /* Check if operation was processed successfully */
915        for (i = 0; i < num_dequeued_ops; i++) {
916            if (dequeued_ops[i]->status != RTE_CRYPTO_OP_STATUS_SUCCESS)
917                rte_exit(EXIT_FAILURE,
918                        "Some operations were not processed correctly");
919        }
920
921        rte_mempool_put_bulk(crypto_op_pool, (void **)dequeued_ops,
922                                            num_dequeued_ops);
923    } while (total_num_dequeued_ops < num_enqueued_ops);
924
925Asymmetric Cryptography
926-----------------------
927
928The cryptodev library currently provides support for the following asymmetric
929Crypto operations; RSA, Modular exponentiation and inversion, Diffie-Hellman and
930Elliptic Curve Diffie-Hellman public and/or private key generation and shared
931secret compute, DSA and EdDSA signature generation and verification.
932
933Session and Session Management
934~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
935
936Sessions are used in asymmetric cryptographic processing to store the immutable
937data defined in asymmetric cryptographic transform which is further used in the
938operation processing. Sessions typically stores information, such as, public
939and private key information or domain params or prime modulus data i.e. immutable
940across data sets. Crypto sessions cache this immutable data in a optimal way for the
941underlying PMD and this allows further acceleration of the offload of Crypto workloads.
942
943Like symmetric, the Crypto device framework provides APIs to allocate and initialize
944asymmetric sessions for crypto devices, where sessions are mempool objects.
945It is the application's responsibility to create and manage the session mempools.
946Application using both symmetric and asymmetric sessions should allocate and maintain
947different sessions pools for each type.
948
949An application can use ``rte_cryptodev_asym_session_pool_create()`` to create a mempool
950with a specified number of elements. The element size will allow for the session header,
951and the max private session size.
952The max private session size is chosen based on available crypto devices,
953the biggest private session size is used. This means any of those devices can be used,
954and the mempool element will have available space for its private session data.
955
956Once the session mempools have been created, ``rte_cryptodev_asym_session_create()``
957is used to allocate and initialize an asymmetric session from the given mempool.
958An asymmetric transform chain is used to specify the operation and its parameters.
959See the section below for details on transforms.
960
961When a session is no longer used, user must call ``rte_cryptodev_asym_session_clear()``
962for each of the crypto devices that are using the session, to free all driver
963private asymmetric session data. Once this is done, session should be freed using
964``rte_cryptodev_asym_session_free()`` which returns them to their mempool.
965
966Asymmetric Sessionless Support
967~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
968
969Asymmetric crypto framework supports session-less operations as well.
970
971Fields that should be set by user are:
972
973Member xform of struct rte_crypto_asym_op should point to the user created rte_crypto_asym_xform.
974Note that rte_crypto_asym_xform should be immutable for the lifetime of associated crypto_op.
975
976Member sess_type of rte_crypto_op should also be set to RTE_CRYPTO_OP_SESSIONLESS.
977
978Transforms and Transform Chaining
979~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
980
981Asymmetric Crypto transforms (``rte_crypto_asym_xform``) are the mechanism used
982to specify the details of the asymmetric Crypto operation. Next pointer within
983xform allows transform to be chained together. Also it is important to note that
984the order in which the transforms are passed indicates the order of the chaining. Allocation
985of the xform structure is in the application domain. To allow future API extensions in a
986backwardly compatible manner, e.g. addition of a new parameter, the application should
987zero the full xform struct before populating it.
988
989Not all asymmetric crypto xforms are supported for chaining. Currently supported
990asymmetric crypto chaining is Diffie-Hellman private key generation followed by
991public generation. Also, currently API does not support chaining of symmetric and
992asymmetric crypto xforms.
993
994Each xform defines specific asymmetric crypto algo. Currently supported are:
995* RSA
996* Modular operations (Exponentiation and Inverse)
997* Diffie-Hellman
998* DSA
999* Elliptic Curve Diffie-Hellman
1000* None - special case where PMD may support a passthrough mode. More for diagnostic purpose
1001
1002See *DPDK API Reference* for details on each rte_crypto_xxx_xform struct
1003
1004Asymmetric Operations
1005~~~~~~~~~~~~~~~~~~~~~
1006
1007The asymmetric Crypto operation structure contains all the mutable data relating
1008to asymmetric cryptographic processing on an input data buffer. It uses either
1009RSA, Modular, Diffie-Hellman or DSA operations depending upon session it is attached
1010to.
1011
1012Every operation must carry a valid session handle which further carries information
1013on xform or xform-chain to be performed on op. Every xform type defines its own set
1014of operational params in their respective rte_crypto_xxx_op_param struct. Depending
1015on xform information within session, PMD picks up and process respective op_param
1016struct.
1017Unlike symmetric, asymmetric operations do not use mbufs for input/output.
1018They operate on data buffer of type ``rte_crypto_param``.
1019
1020See *DPDK API Reference* for details on each rte_crypto_xxx_op_param struct
1021
1022Private user data
1023~~~~~~~~~~~~~~~~~
1024
1025Similar to symmetric above, asymmetric also has a set and get API that provides a
1026mechanism for an application to store and retrieve the private user data information
1027stored along with the crypto session.
1028
1029.. code-block:: c
1030
1031	int rte_cryptodev_asym_session_set_user_data(void *sess,
1032		void *data, uint16_t size);
1033
1034	void * rte_cryptodev_asym_session_get_user_data(void *sess);
1035
1036Please note the ``size`` passed to set API cannot be bigger than the predefined
1037``user_data_sz`` when creating the session mempool, otherwise the function will
1038return an error. Also when ``user_data_sz`` was defined as ``0`` when
1039creating the session mempool, the get API will always return ``NULL``.
1040
1041Asymmetric crypto Sample code
1042-----------------------------
1043
1044There's a unit test application test_cryptodev_asym.c inside unit test framework that
1045show how to setup and process asymmetric operations using cryptodev library.
1046
1047The following code samples are taken from the test application mentioned above,
1048and show basic steps to compute modular exponentiation using an openssl PMD
1049available in DPDK (performing other crypto operations is similar except change
1050to respective op and xform setup).
1051
1052.. note::
1053   The following code snippets are taken from multiple functions, so variable
1054   names may differ slightly between sections.
1055
1056Configure the virtual device, queue pairs, crypto op pool and session mempool.
1057
1058.. literalinclude:: ../../../app/test/test_cryptodev_asym.c
1059   :language: c
1060   :start-after: Device, op pool and session configuration for asymmetric crypto. 8<
1061   :end-before: >8 End of device, op pool and session configuration for asymmetric crypto section.
1062   :dedent: 1
1063
1064Create MODEX data vectors.
1065
1066.. literalinclude:: ../../../app/test/test_cryptodev_mod_test_vectors.h
1067   :language: c
1068   :start-after: MODEX data. 8<
1069   :end-before: >8 End of MODEX data.
1070
1071Setup crypto xform to do modular exponentiation using data vectors.
1072
1073.. literalinclude:: ../../../app/test/test_cryptodev_mod_test_vectors.h
1074   :language: c
1075   :start-after: MODEX vector. 8<
1076   :end-before: >8 End of MODEX vector.
1077
1078Generate crypto op, create and attach a session, then process packets.
1079
1080.. literalinclude:: ../../../app/test/test_cryptodev_asym.c
1081   :language: c
1082   :start-after: Create op, create session, and process packets. 8<
1083   :end-before: >8 End of create op, create session, and process packets section.
1084   :dedent: 1
1085
1086.. note::
1087   The ``rte_cryptodev_asym_session`` struct is hidden from the application.
1088   The ``sess`` pointer used above is a void pointer.
1089
1090
1091Asymmetric Crypto Device API
1092~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1093
1094The cryptodev Library API is described in the
1095`DPDK API Reference <https://doc.dpdk.org/api/>`_
1096
1097
1098Device Statistics
1099-----------------
1100
1101The Cryptodev library has support for displaying Crypto device information
1102through the Telemetry interface. Telemetry commands that can be used
1103are shown below.
1104
1105#. Get the list of available Crypto devices by ID::
1106
1107     --> /cryptodev/list
1108     {"/cryptodev/list": [0, 1, 2, 3]}
1109
1110#. Get general information from a Crypto device::
1111
1112     --> /cryptodev/info,0
1113     {"/cryptodev/info": {"device_name": "0000:1c:01.0_qat_sym",
1114     "max_nb_queue_pairs": 2}}
1115
1116#. Get the statistics for a particular Crypto device::
1117
1118     --> /cryptodev/stats,0
1119     {"/cryptodev/stats": {"enqueued_count": 0, "dequeued_count": 0,
1120     "enqueue_err_count": 0, "dequeue_err_count": 0}}
1121
1122#. Get the capabilities of a particular Crypto device::
1123
1124     --> /cryptodev/caps,0
1125     {"/cryptodev/caps": {"crypto_caps": [<array of serialized bytes of
1126     capabilities>], "crypto_caps_n": <number of capabilities>}}
1127
1128For more information on how to use the Telemetry interface, see
1129the :doc:`../howto/telemetry`.
1130