summaryrefslogtreecommitdiffstats
path: root/src/libknot/xdp/xdp.c
blob: 8286884f5e0cdc4027dca829625f623da640beea (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
/*  Copyright (C) 2023 CZ.NIC, z.s.p.o. <knot-dns@labs.nic.cz>

    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 3 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <https://www.gnu.org/licenses/>.
 */

#include <assert.h>
#include <errno.h>
#include <netinet/in.h>
#include <linux/if_ether.h>
#include <linux/udp.h>
#include <stddef.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <unistd.h>

#include "libknot/attribute.h"
#include "libknot/endian.h"
#include "libknot/errcode.h"
#include "libknot/xdp/bpf-consts.h"
#include "libknot/xdp/bpf-user.h"
#include "libknot/xdp/eth.h"
#include "libknot/xdp/msg_init.h"
#include "libknot/xdp/protocols.h"
#include "libknot/xdp/xdp.h"
#include "contrib/macros.h"
#include "contrib/net.h"

#define FRAME_SIZE		2048

#define FRAME_COUNT_TX		2048
#define FRAME_COUNT_RX		2048
#define FRAME_COUNT		(FRAME_COUNT_TX + FRAME_COUNT_RX)

#define RING_LEN_TX		FRAME_COUNT_TX
#define RING_LEN_CQ		FRAME_COUNT_TX
#define RING_LEN_RX		FRAME_COUNT_RX
/* It's recommended that the FQ ring size >= HW RX ring size + AF_XDP RX ring size. */
#define RING_LEN_FQ		(2 * FRAME_COUNT_RX)

#define ALLOC_RETRY_NUM		15
#define ALLOC_RETRY_DELAY	20 // In nanoseconds.

/* With recent compilers we statically check #defines for settings that
 * get refused by AF_XDP drivers (in current versions, at least). */
#if (__STDC_VERSION__ >= 201112L)
#define IS_POWER_OF_2(n) (((n) & (n - 1)) == 0)
_Static_assert((FRAME_SIZE == 4096 || FRAME_SIZE == 2048)
	&& IS_POWER_OF_2(RING_LEN_TX) && IS_POWER_OF_2(RING_LEN_RX)
	&& IS_POWER_OF_2(RING_LEN_CQ) && IS_POWER_OF_2(RING_LEN_FQ)
	&& FRAME_COUNT_TX <= (1 << 16) /* see tx_free_indices */
	, "Incorrect #define combination for AF_XDP.");
#endif

struct umem_frame {
	uint8_t bytes[FRAME_SIZE];
};

static int configure_xsk_umem(struct kxsk_umem **out_umem, bool extra_frames)
{
	/* Allocate memory and call driver to create the UMEM. */
	struct kxsk_umem *umem = calloc(1,
		offsetof(struct kxsk_umem, tx_free_indices)
		+ sizeof(umem->tx_free_indices[0]) * FRAME_COUNT_TX);
	if (umem == NULL) {
		return KNOT_ENOMEM;
	}

	size_t frame_count = FRAME_COUNT + (extra_frames ? FRAME_COUNT_RX : 0);

	int ret = posix_memalign((void **)&umem->frames, getpagesize(),
	                         FRAME_SIZE * frame_count);
	if (ret != 0) {
		free(umem);
		return KNOT_ENOMEM;
	}

	const struct xsk_umem_config config = {
		.fill_size = RING_LEN_FQ,
		.comp_size = RING_LEN_CQ,
		.frame_size = FRAME_SIZE,
		.frame_headroom = KNOT_XDP_PKT_ALIGNMENT,
	};

	ret = xsk_umem__create(&umem->umem, umem->frames, FRAME_SIZE * frame_count,
	                       &umem->fq, &umem->cq, &config);
	if (ret != KNOT_EOK) {
		free(umem->frames);
		free(umem);
		return ret;
	}
	*out_umem = umem;

	/* Designate the starting chunk of buffers for TX, and put them onto the stack. */
	umem->tx_free_count = FRAME_COUNT_TX;
	for (uint32_t i = 0; i < FRAME_COUNT_TX; ++i) {
		umem->tx_free_indices[i] = i;
	}

	/* Designate the rest of buffers for RX, and pass them to the driver. */
	uint32_t idx = 0;
	ret = xsk_ring_prod__reserve(&umem->fq, frame_count - FRAME_COUNT_TX, &idx);
	if (ret != frame_count - FRAME_COUNT_TX) {
		assert(0);
		return KNOT_ERROR;
	}
	assert(idx == 0);
	for (uint32_t i = FRAME_COUNT_TX; i < frame_count; ++i) {
		*xsk_ring_prod__fill_addr(&umem->fq, idx++) = i * FRAME_SIZE;
	}
	xsk_ring_prod__submit(&umem->fq, frame_count - FRAME_COUNT_TX);

	return KNOT_EOK;
}

static void deconfigure_xsk_umem(struct kxsk_umem *umem)
{
	(void)xsk_umem__delete(umem->umem);
	free(umem->frames);
	free(umem);
}

static int configure_xsk_socket(struct kxsk_umem *umem,
                                const struct kxsk_iface *iface,
                                knot_xdp_socket_t **out_sock,
                                const knot_xdp_config_t *config)
{
	knot_xdp_socket_t *xsk_info = calloc(1, sizeof(*xsk_info));
	if (xsk_info == NULL) {
		return KNOT_ENOMEM;
	}
	xsk_info->iface = iface;
	xsk_info->umem = umem;

	uint16_t bind_flags = 0;
	if (config != NULL && config->force_copy) {
		bind_flags |= XDP_COPY;
	}

	const struct xsk_socket_config sock_conf = {
		.tx_size = RING_LEN_TX,
		.rx_size = RING_LEN_RX,
		.libbpf_flags = XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD,
		.bind_flags = bind_flags,
	};

	int ret = xsk_socket__create(&xsk_info->xsk, iface->if_name,
	                             iface->if_queue, umem->umem,
	                             &xsk_info->rx, &xsk_info->tx, &sock_conf);
	if (ret != 0) {
		free(xsk_info);
		return ret;
	}

	*out_sock = xsk_info;
	return KNOT_EOK;
}

_public_
int knot_xdp_init(knot_xdp_socket_t **socket, const char *if_name, int if_queue,
                  knot_xdp_filter_flag_t flags, uint16_t udp_port, uint16_t quic_port,
                  knot_xdp_load_bpf_t load_bpf, const knot_xdp_config_t *xdp_config)
{
	if (socket == NULL || if_name == NULL ||
	    (udp_port == quic_port && (flags & KNOT_XDP_FILTER_UDP) && (flags & KNOT_XDP_FILTER_QUIC)) ||
	    (flags & (KNOT_XDP_FILTER_UDP | KNOT_XDP_FILTER_TCP | KNOT_XDP_FILTER_QUIC)) == 0) {
		return KNOT_EINVAL;
	}

	struct kxsk_iface *iface;
	const bool generic_xdp = (xdp_config != NULL && xdp_config->force_generic);
	int ret = kxsk_iface_new(if_name, if_queue, load_bpf, generic_xdp, &iface);
	if (ret != KNOT_EOK) {
		return ret;
	}

	/* Initialize shared packet_buffer for umem usage. */
	struct kxsk_umem *umem = NULL;
	ret = configure_xsk_umem(&umem, xdp_config->extra_frames);
	if (ret != KNOT_EOK) {
		kxsk_iface_free(iface);
		return ret;
	}

	ret = configure_xsk_socket(umem, iface, socket, xdp_config);
	if (ret != KNOT_EOK) {
		deconfigure_xsk_umem(umem);
		kxsk_iface_free(iface);
		return ret;
	}

	(*socket)->frame_limit = FRAME_SIZE;
	ret = knot_eth_mtu(if_name);
	if (ret > 0) {
		(*socket)->frame_limit = MIN((unsigned)ret, (*socket)->frame_limit);
	}

	if (flags & KNOT_XDP_FILTER_ROUTE) {
		ret = knot_eth_vlans(&(*socket)->vlan_map, &(*socket)->vlan_map_max);
		if (ret != KNOT_EOK) {
			xsk_socket__delete((*socket)->xsk);
			deconfigure_xsk_umem(umem);
			kxsk_iface_free(iface);
			free(*socket);
			*socket = NULL;
			return ret;
		}
	}

	ret = kxsk_socket_start(iface, flags, udp_port, quic_port, (*socket)->xsk);
	if (ret != KNOT_EOK) {
		free((*socket)->vlan_map);
		xsk_socket__delete((*socket)->xsk);
		deconfigure_xsk_umem(umem);
		kxsk_iface_free(iface);
		free(*socket);
		*socket = NULL;
		return ret;
	}

	return ret;
}

_public_
void knot_xdp_deinit(knot_xdp_socket_t *socket)
{
	if (socket == NULL) {
		return;
	}
	if (unlikely(socket->send_mock != NULL)) {
		free(socket);
		return;
	}

	kxsk_socket_stop(socket->iface);
	xsk_socket__delete(socket->xsk);
	deconfigure_xsk_umem(socket->umem);

	kxsk_iface_free((struct kxsk_iface *)/*const-cast*/socket->iface);
	free(socket->vlan_map);
	free(socket);
}

_public_
int knot_xdp_socket_fd(knot_xdp_socket_t *socket)
{
	if (socket == NULL) {
		return 0;
	}

	return xsk_socket__fd(socket->xsk);
}

static void tx_free_relative(struct kxsk_umem *umem, uint64_t addr_relative)
{
	/* The address may not point to *start* of buffer, but `/` solves that. */
	uint64_t index = addr_relative / FRAME_SIZE;
	assert(index < FRAME_COUNT);
	umem->tx_free_indices[umem->tx_free_count++] = index;
}

_public_
void knot_xdp_send_prepare(knot_xdp_socket_t *socket)
{
	if (socket == NULL || unlikely(socket->send_mock != NULL)) {
		return;
	}

	struct kxsk_umem *const umem = socket->umem;
	struct xsk_ring_cons *const cq = &umem->cq;

	uint32_t idx = 0;
	const uint32_t completed = xsk_ring_cons__peek(cq, UINT32_MAX, &idx);
	if (completed == 0) {
		return;
	}
	assert(umem->tx_free_count + completed <= FRAME_COUNT_TX);

	for (uint32_t i = 0; i < completed; ++i) {
		uint64_t addr_relative = *xsk_ring_cons__comp_addr(cq, idx++);
		tx_free_relative(umem, addr_relative);
	}

	xsk_ring_cons__release(cq, completed);
}

static struct umem_frame *alloc_tx_frame(knot_xdp_socket_t *socket)
{
	if (unlikely(socket->send_mock != NULL)) {
		return malloc(sizeof(struct umem_frame));
	}

	const struct timespec delay = { .tv_nsec = ALLOC_RETRY_DELAY };
	struct kxsk_umem *umem = socket->umem;

	for (int i = 0; unlikely(umem->tx_free_count == 0); i++) {
		if (i == ALLOC_RETRY_NUM) {
			return NULL;
		}
		nanosleep(&delay, NULL);
		knot_xdp_send_prepare(socket);
	}

	uint32_t index = umem->tx_free_indices[--umem->tx_free_count];
	return umem->frames + index;
}

static void prepare_payload(knot_xdp_msg_t *msg, void *uframe)
{
	size_t hdr_len = prot_write_hdrs_len(msg);
	msg->payload.iov_base = uframe + hdr_len + KNOT_XDP_PKT_ALIGNMENT;
	msg->payload.iov_len = FRAME_SIZE - hdr_len - KNOT_XDP_PKT_ALIGNMENT;
}

_public_
int knot_xdp_send_alloc(knot_xdp_socket_t *socket, knot_xdp_msg_flag_t flags,
                        knot_xdp_msg_t *out)
{
	if (socket == NULL || out == NULL) {
		return KNOT_EINVAL;
	}

	struct umem_frame *uframe = alloc_tx_frame(socket);
	if (uframe == NULL) {
		return KNOT_ENOMEM;
	}

	msg_init(out, flags);
	prepare_payload(out, uframe);

	return KNOT_EOK;
}

_public_
int knot_xdp_reply_alloc(knot_xdp_socket_t *socket, const knot_xdp_msg_t *query,
                         knot_xdp_msg_t *out)
{
	if (socket == NULL || query == NULL || out == NULL) {
		return KNOT_EINVAL;
	}

	struct umem_frame *uframe = alloc_tx_frame(socket);
	if (uframe == NULL) {
		return KNOT_ENOMEM;
	}

	msg_init_reply(out, query);
	prepare_payload(out, uframe);

	return KNOT_EOK;
}

static void free_unsent(knot_xdp_socket_t *socket, const knot_xdp_msg_t *msg)
{
	if (unlikely(socket->send_mock != NULL)) {
		free(msg->payload.iov_base - prot_write_hdrs_len(msg) - KNOT_XDP_PKT_ALIGNMENT);
		return;
	}
	uint64_t addr_relative = (uint8_t *)msg->payload.iov_base
	                         - socket->umem->frames->bytes;
	tx_free_relative(socket->umem, addr_relative);
}

_public_
int knot_xdp_send(knot_xdp_socket_t *socket, const knot_xdp_msg_t msgs[],
                  uint32_t count, uint32_t *sent)
{
	if (socket == NULL || msgs == NULL || sent == NULL) {
		return KNOT_EINVAL;
	}
	if (unlikely(socket->send_mock != NULL)) {
		int ret = socket->send_mock(socket, msgs, count, sent);
		for (uint32_t i = 0; i < count; ++i) {
			free_unsent(socket, &msgs[i]);
		}
		return ret;
	}

	/* Now we want to do something close to
	 *   xsk_ring_prod__reserve(&socket->tx, count, *idx)
	 * but we don't know in advance if we utilize *whole* `count`,
	 * and the API doesn't allow "cancelling reservations".
	 * Therefore we handle `socket->tx.cached_prod` by hand.
	 */
	if (xsk_prod_nb_free(&socket->tx, count) < count) {
		/* This situation was sometimes observed in the emulated XDP mode. */
		for (uint32_t i = 0; i < count; ++i) {
			free_unsent(socket, &msgs[i]);
		}
		return KNOT_ENOBUFS;
	}
	uint32_t idx = socket->tx.cached_prod;

	for (uint32_t i = 0; i < count; ++i) {
		const knot_xdp_msg_t *msg = &msgs[i];

		if (empty_msg(msg)) {
			free_unsent(socket, msg);
		} else {
			size_t hdr_len = prot_write_hdrs_len(msg);
			size_t tot_len = hdr_len + msg->payload.iov_len;
			uint8_t *msg_beg = msg->payload.iov_base - hdr_len;
			uint16_t mss = MIN(socket->frame_limit - hdr_len, KNOT_TCP_MSS);
			prot_write_eth(msg_beg, msg, msg_beg + tot_len, mss);

			*xsk_ring_prod__tx_desc(&socket->tx, idx++) = (struct xdp_desc) {
				.addr = msg_beg - socket->umem->frames->bytes,
				.len = tot_len,
			};
		}
	}

	*sent = idx - socket->tx.cached_prod;
	assert(*sent <= count);
	socket->tx.cached_prod = idx;
	xsk_ring_prod__submit(&socket->tx, *sent);
	socket->kernel_needs_wakeup = true;

	return KNOT_EOK;
}

_public_
void knot_xdp_send_free(knot_xdp_socket_t *socket, const knot_xdp_msg_t msgs[],
                        uint32_t count)
{
	for (uint32_t i = 0; i < count; i++) {
		free_unsent(socket, &msgs[i]);
	}
}

_public_
int knot_xdp_send_finish(knot_xdp_socket_t *socket)
{
	if (socket == NULL) {
		return KNOT_EINVAL;
	}

	/* Trigger sending queued packets. */
	if (!socket->kernel_needs_wakeup) {
		return KNOT_EOK;
	}

	int ret = sendto(xsk_socket__fd(socket->xsk), NULL, 0, MSG_DONTWAIT, NULL, 0);
	const bool is_ok = (ret >= 0);
	// List of "safe" errors taken from
	// https://github.com/torvalds/linux/blame/master/samples/bpf/xdpsock_user.c
	const bool is_again = !is_ok && (errno == ENOBUFS || errno == EAGAIN
	                                || errno == EBUSY || errno == ENETDOWN);
	// Some of the !is_ok cases are a little unclear - what to do about the syscall,
	// including how caller of _sendmsg_finish() should react.
	if (is_ok || !is_again) {
		socket->kernel_needs_wakeup = false;
	}
	if (is_again) {
		return KNOT_EAGAIN;
	} else if (is_ok) {
		return KNOT_EOK;
	} else {
		return -errno;
	}
	/* This syscall might be avoided with a newer kernel feature (>= 5.4):
	   https://www.kernel.org/doc/html/latest/networking/af_xdp.html#xdp-use-need-wakeup-bind-flag
	   Unfortunately it's not easy to continue supporting older kernels
	   when using this feature on newer ones.
	 */
}

_public_
int knot_xdp_recv(knot_xdp_socket_t *socket, knot_xdp_msg_t msgs[],
                  uint32_t max_count, uint32_t *count, size_t *wire_size)
{
	if (socket == NULL || msgs == NULL || count == NULL) {
		return KNOT_EINVAL;
	}

	uint32_t idx = 0;
	const uint32_t available = xsk_ring_cons__peek(&socket->rx, max_count, &idx);
	if (available == 0) {
		*count = 0;
		return KNOT_EOK;
	}
	assert(available <= max_count);

	for (uint32_t i = 0; i < available; ++i) {
		knot_xdp_msg_t *msg = &msgs[i];
		const struct xdp_desc *desc = xsk_ring_cons__rx_desc(&socket->rx, idx++);
		uint8_t *uframe_p = (uint8_t *)socket->umem->frames + desc->addr;

		void *payl_end;
		void *payl_start = prot_read_eth(uframe_p, msg, &payl_end,
		                                 socket->vlan_map, socket->vlan_map_max);

		msg->payload.iov_base = payl_start;
		msg->payload.iov_len = payl_end - payl_start;
		msg->mss = MIN(msg->mss, FRAME_SIZE - (payl_start - (void *)uframe_p));

		if (wire_size != NULL) {
			(*wire_size) += desc->len;
		}
	}

	xsk_ring_cons__release(&socket->rx, available);
	*count = available;

	return KNOT_EOK;
}

static uint8_t *msg_uframe_ptr(const knot_xdp_msg_t *msg)
{
	return NULL + ((msg->payload.iov_base - NULL) & ~(FRAME_SIZE - 1));
}

_public_
void knot_xdp_recv_finish(knot_xdp_socket_t *socket, const knot_xdp_msg_t msgs[],
                          uint32_t count)
{
	if (socket == NULL || msgs == NULL) {
		return;
	}

	const struct timespec delay = { .tv_nsec = ALLOC_RETRY_DELAY };

	struct kxsk_umem *const umem = socket->umem;
	struct xsk_ring_prod *const fq = &umem->fq;

	uint32_t idx = 0;
	uint32_t reserved = xsk_ring_prod__reserve(fq, count, &idx);
	for (int i = 0; unlikely(reserved < count); i++) {
		if (i == ALLOC_RETRY_NUM) {
			return;
		}
		nanosleep(&delay, NULL);
		reserved = xsk_ring_prod__reserve(fq, count, &idx);
	}

	for (uint32_t i = 0; i < reserved; ++i) {
		uint8_t *uframe_p = msg_uframe_ptr(&msgs[i]);
		uint64_t offset = uframe_p - umem->frames->bytes;
		*xsk_ring_prod__fill_addr(fq, idx++) = offset;
	}

	xsk_ring_prod__submit(fq, reserved);
}

_public_
void knot_xdp_socket_info(const knot_xdp_socket_t *socket, FILE *file)
{
	if (socket == NULL || file == NULL) {
		return;
	}

	// The number of busy frames
	#define RING_BUSY(ring) \
		((*(ring)->producer - *(ring)->consumer) & (ring)->mask)

	#define RING_PRINFO(name, ring) \
		fprintf(file, "Ring %s: size %4d, busy %4d (prod %4d, cons %4d)\n", \
		        name, (unsigned)(ring)->size, \
		        (unsigned)RING_BUSY((ring)), \
		        (unsigned)*(ring)->producer, (unsigned)*(ring)->consumer)

	const int rx_busyf = RING_BUSY(&socket->umem->fq) + RING_BUSY(&socket->rx);
	fprintf(file, "\nLOST RX frames: %4d", (int)(FRAME_COUNT_RX - rx_busyf));

	const int tx_busyf = RING_BUSY(&socket->umem->cq) + RING_BUSY(&socket->tx);
	const int tx_freef = socket->umem->tx_free_count;
	fprintf(file, "\nLOST TX frames: %4d\n", (int)(FRAME_COUNT_TX - tx_busyf - tx_freef));

	RING_PRINFO("FQ", &socket->umem->fq);
	RING_PRINFO("RX", &socket->rx);
	RING_PRINFO("TX", &socket->tx);
	RING_PRINFO("CQ", &socket->umem->cq);
	fprintf(file, "TX free frames: %4d\n", tx_freef);
}