xref: /linux/tools/lib/bpf/ringbuf.c (revision 23c996fc2bc1978a02c64eddb90b4ab5d309c8df)
1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 /*
3  * Ring buffer operations.
4  *
5  * Copyright (C) 2020 Facebook, Inc.
6  */
7 #ifndef _GNU_SOURCE
8 #define _GNU_SOURCE
9 #endif
10 #include <stdlib.h>
11 #include <stdio.h>
12 #include <errno.h>
13 #include <unistd.h>
14 #include <linux/err.h>
15 #include <linux/bpf.h>
16 #include <asm/barrier.h>
17 #include <sys/mman.h>
18 #include <sys/epoll.h>
19 #include <time.h>
20 
21 #include "libbpf.h"
22 #include "libbpf_internal.h"
23 #include "bpf.h"
24 
25 struct ring {
26 	ring_buffer_sample_fn sample_cb;
27 	void *ctx;
28 	void *data;
29 	unsigned long *consumer_pos;
30 	unsigned long *producer_pos;
31 	unsigned long mask;
32 	int map_fd;
33 };
34 
35 struct ring_buffer {
36 	struct epoll_event *events;
37 	struct ring **rings;
38 	size_t page_size;
39 	int epoll_fd;
40 	int ring_cnt;
41 };
42 
43 struct user_ring_buffer {
44 	struct epoll_event event;
45 	unsigned long *consumer_pos;
46 	unsigned long *producer_pos;
47 	void *data;
48 	unsigned long mask;
49 	size_t page_size;
50 	int map_fd;
51 	int epoll_fd;
52 };
53 
54 /* 8-byte ring buffer header structure */
55 struct ringbuf_hdr {
56 	__u32 len;
57 	__u32 pad;
58 };
59 
60 static void ringbuf_free_ring(struct ring_buffer *rb, struct ring *r)
61 {
62 	if (r->consumer_pos) {
63 		munmap(r->consumer_pos, rb->page_size);
64 		r->consumer_pos = NULL;
65 	}
66 	if (r->producer_pos) {
67 		munmap(r->producer_pos, rb->page_size + 2 * (r->mask + 1));
68 		r->producer_pos = NULL;
69 	}
70 
71 	free(r);
72 }
73 
74 /* Add extra RINGBUF maps to this ring buffer manager */
75 int ring_buffer__add(struct ring_buffer *rb, int map_fd,
76 		     ring_buffer_sample_fn sample_cb, void *ctx)
77 {
78 	struct bpf_map_info info;
79 	__u32 len = sizeof(info);
80 	struct epoll_event *e;
81 	struct ring *r;
82 	__u64 mmap_sz;
83 	void *tmp;
84 	int err;
85 
86 	memset(&info, 0, sizeof(info));
87 
88 	err = bpf_map_get_info_by_fd(map_fd, &info, &len);
89 	if (err) {
90 		err = -errno;
91 		pr_warn("ringbuf: failed to get map info for fd=%d: %d\n",
92 			map_fd, err);
93 		return libbpf_err(err);
94 	}
95 
96 	if (info.type != BPF_MAP_TYPE_RINGBUF) {
97 		pr_warn("ringbuf: map fd=%d is not BPF_MAP_TYPE_RINGBUF\n",
98 			map_fd);
99 		return libbpf_err(-EINVAL);
100 	}
101 
102 	tmp = libbpf_reallocarray(rb->rings, rb->ring_cnt + 1, sizeof(*rb->rings));
103 	if (!tmp)
104 		return libbpf_err(-ENOMEM);
105 	rb->rings = tmp;
106 
107 	tmp = libbpf_reallocarray(rb->events, rb->ring_cnt + 1, sizeof(*rb->events));
108 	if (!tmp)
109 		return libbpf_err(-ENOMEM);
110 	rb->events = tmp;
111 
112 	r = calloc(1, sizeof(*r));
113 	if (!r)
114 		return libbpf_err(-ENOMEM);
115 	rb->rings[rb->ring_cnt] = r;
116 
117 	r->map_fd = map_fd;
118 	r->sample_cb = sample_cb;
119 	r->ctx = ctx;
120 	r->mask = info.max_entries - 1;
121 
122 	/* Map writable consumer page */
123 	tmp = mmap(NULL, rb->page_size, PROT_READ | PROT_WRITE, MAP_SHARED, map_fd, 0);
124 	if (tmp == MAP_FAILED) {
125 		err = -errno;
126 		pr_warn("ringbuf: failed to mmap consumer page for map fd=%d: %d\n",
127 			map_fd, err);
128 		goto err_out;
129 	}
130 	r->consumer_pos = tmp;
131 
132 	/* Map read-only producer page and data pages. We map twice as big
133 	 * data size to allow simple reading of samples that wrap around the
134 	 * end of a ring buffer. See kernel implementation for details.
135 	 */
136 	mmap_sz = rb->page_size + 2 * (__u64)info.max_entries;
137 	if (mmap_sz != (__u64)(size_t)mmap_sz) {
138 		err = -E2BIG;
139 		pr_warn("ringbuf: ring buffer size (%u) is too big\n", info.max_entries);
140 		goto err_out;
141 	}
142 	tmp = mmap(NULL, (size_t)mmap_sz, PROT_READ, MAP_SHARED, map_fd, rb->page_size);
143 	if (tmp == MAP_FAILED) {
144 		err = -errno;
145 		pr_warn("ringbuf: failed to mmap data pages for map fd=%d: %d\n",
146 			map_fd, err);
147 		goto err_out;
148 	}
149 	r->producer_pos = tmp;
150 	r->data = tmp + rb->page_size;
151 
152 	e = &rb->events[rb->ring_cnt];
153 	memset(e, 0, sizeof(*e));
154 
155 	e->events = EPOLLIN;
156 	e->data.fd = rb->ring_cnt;
157 	if (epoll_ctl(rb->epoll_fd, EPOLL_CTL_ADD, map_fd, e) < 0) {
158 		err = -errno;
159 		pr_warn("ringbuf: failed to epoll add map fd=%d: %d\n",
160 			map_fd, err);
161 		goto err_out;
162 	}
163 
164 	rb->ring_cnt++;
165 	return 0;
166 
167 err_out:
168 	ringbuf_free_ring(rb, r);
169 	return libbpf_err(err);
170 }
171 
172 void ring_buffer__free(struct ring_buffer *rb)
173 {
174 	int i;
175 
176 	if (!rb)
177 		return;
178 
179 	for (i = 0; i < rb->ring_cnt; ++i)
180 		ringbuf_free_ring(rb, rb->rings[i]);
181 	if (rb->epoll_fd >= 0)
182 		close(rb->epoll_fd);
183 
184 	free(rb->events);
185 	free(rb->rings);
186 	free(rb);
187 }
188 
189 struct ring_buffer *
190 ring_buffer__new(int map_fd, ring_buffer_sample_fn sample_cb, void *ctx,
191 		 const struct ring_buffer_opts *opts)
192 {
193 	struct ring_buffer *rb;
194 	int err;
195 
196 	if (!OPTS_VALID(opts, ring_buffer_opts))
197 		return errno = EINVAL, NULL;
198 
199 	rb = calloc(1, sizeof(*rb));
200 	if (!rb)
201 		return errno = ENOMEM, NULL;
202 
203 	rb->page_size = getpagesize();
204 
205 	rb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
206 	if (rb->epoll_fd < 0) {
207 		err = -errno;
208 		pr_warn("ringbuf: failed to create epoll instance: %d\n", err);
209 		goto err_out;
210 	}
211 
212 	err = ring_buffer__add(rb, map_fd, sample_cb, ctx);
213 	if (err)
214 		goto err_out;
215 
216 	return rb;
217 
218 err_out:
219 	ring_buffer__free(rb);
220 	return errno = -err, NULL;
221 }
222 
223 static inline int roundup_len(__u32 len)
224 {
225 	/* clear out top 2 bits (discard and busy, if set) */
226 	len <<= 2;
227 	len >>= 2;
228 	/* add length prefix */
229 	len += BPF_RINGBUF_HDR_SZ;
230 	/* round up to 8 byte alignment */
231 	return (len + 7) / 8 * 8;
232 }
233 
234 static int64_t ringbuf_process_ring(struct ring *r, size_t n)
235 {
236 	int *len_ptr, len, err;
237 	/* 64-bit to avoid overflow in case of extreme application behavior */
238 	int64_t cnt = 0;
239 	unsigned long cons_pos, prod_pos;
240 	bool got_new_data;
241 	void *sample;
242 
243 	cons_pos = smp_load_acquire(r->consumer_pos);
244 	do {
245 		got_new_data = false;
246 		prod_pos = smp_load_acquire(r->producer_pos);
247 		while (cons_pos < prod_pos) {
248 			len_ptr = r->data + (cons_pos & r->mask);
249 			len = smp_load_acquire(len_ptr);
250 
251 			/* sample not committed yet, bail out for now */
252 			if (len & BPF_RINGBUF_BUSY_BIT)
253 				goto done;
254 
255 			got_new_data = true;
256 			cons_pos += roundup_len(len);
257 
258 			if ((len & BPF_RINGBUF_DISCARD_BIT) == 0) {
259 				sample = (void *)len_ptr + BPF_RINGBUF_HDR_SZ;
260 				err = r->sample_cb(r->ctx, sample, len);
261 				if (err < 0) {
262 					/* update consumer pos and bail out */
263 					smp_store_release(r->consumer_pos,
264 							  cons_pos);
265 					return err;
266 				}
267 				cnt++;
268 			}
269 
270 			smp_store_release(r->consumer_pos, cons_pos);
271 
272 			if (cnt >= n)
273 				goto done;
274 		}
275 	} while (got_new_data);
276 done:
277 	return cnt;
278 }
279 
280 /* Consume available ring buffer(s) data without event polling, up to n
281  * records.
282  *
283  * Returns number of records consumed across all registered ring buffers (or
284  * n, whichever is less), or negative number if any of the callbacks return
285  * error.
286  */
287 int ring_buffer__consume_n(struct ring_buffer *rb, size_t n)
288 {
289 	int64_t err, res = 0;
290 	int i;
291 
292 	for (i = 0; i < rb->ring_cnt; i++) {
293 		struct ring *ring = rb->rings[i];
294 
295 		err = ringbuf_process_ring(ring, n);
296 		if (err < 0)
297 			return libbpf_err(err);
298 		res += err;
299 		n -= err;
300 
301 		if (n == 0)
302 			break;
303 	}
304 	return res > INT_MAX ? INT_MAX : res;
305 }
306 
307 /* Consume available ring buffer(s) data without event polling.
308  * Returns number of records consumed across all registered ring buffers (or
309  * INT_MAX, whichever is less), or negative number if any of the callbacks
310  * return error.
311  */
312 int ring_buffer__consume(struct ring_buffer *rb)
313 {
314 	int64_t err, res = 0;
315 	int i;
316 
317 	for (i = 0; i < rb->ring_cnt; i++) {
318 		struct ring *ring = rb->rings[i];
319 
320 		err = ringbuf_process_ring(ring, INT_MAX);
321 		if (err < 0)
322 			return libbpf_err(err);
323 		res += err;
324 		if (res > INT_MAX) {
325 			res = INT_MAX;
326 			break;
327 		}
328 	}
329 	return res;
330 }
331 
332 /* Poll for available data and consume records, if any are available.
333  * Returns number of records consumed (or INT_MAX, whichever is less), or
334  * negative number, if any of the registered callbacks returned error.
335  */
336 int ring_buffer__poll(struct ring_buffer *rb, int timeout_ms)
337 {
338 	int i, cnt;
339 	int64_t err, res = 0;
340 
341 	cnt = epoll_wait(rb->epoll_fd, rb->events, rb->ring_cnt, timeout_ms);
342 	if (cnt < 0)
343 		return libbpf_err(-errno);
344 
345 	for (i = 0; i < cnt; i++) {
346 		__u32 ring_id = rb->events[i].data.fd;
347 		struct ring *ring = rb->rings[ring_id];
348 
349 		err = ringbuf_process_ring(ring, INT_MAX);
350 		if (err < 0)
351 			return libbpf_err(err);
352 		res += err;
353 	}
354 	if (res > INT_MAX)
355 		res = INT_MAX;
356 	return res;
357 }
358 
359 /* Get an fd that can be used to sleep until data is available in the ring(s) */
360 int ring_buffer__epoll_fd(const struct ring_buffer *rb)
361 {
362 	return rb->epoll_fd;
363 }
364 
365 struct ring *ring_buffer__ring(struct ring_buffer *rb, unsigned int idx)
366 {
367 	if (idx >= rb->ring_cnt)
368 		return errno = ERANGE, NULL;
369 
370 	return rb->rings[idx];
371 }
372 
373 unsigned long ring__consumer_pos(const struct ring *r)
374 {
375 	/* Synchronizes with smp_store_release() in ringbuf_process_ring(). */
376 	return smp_load_acquire(r->consumer_pos);
377 }
378 
379 unsigned long ring__producer_pos(const struct ring *r)
380 {
381 	/* Synchronizes with smp_store_release() in __bpf_ringbuf_reserve() in
382 	 * the kernel.
383 	 */
384 	return smp_load_acquire(r->producer_pos);
385 }
386 
387 size_t ring__avail_data_size(const struct ring *r)
388 {
389 	unsigned long cons_pos, prod_pos;
390 
391 	cons_pos = ring__consumer_pos(r);
392 	prod_pos = ring__producer_pos(r);
393 	return prod_pos - cons_pos;
394 }
395 
396 size_t ring__size(const struct ring *r)
397 {
398 	return r->mask + 1;
399 }
400 
401 int ring__map_fd(const struct ring *r)
402 {
403 	return r->map_fd;
404 }
405 
406 int ring__consume_n(struct ring *r, size_t n)
407 {
408 	int64_t res;
409 
410 	res = ringbuf_process_ring(r, n);
411 	if (res < 0)
412 		return libbpf_err(res);
413 
414 	return res > INT_MAX ? INT_MAX : res;
415 }
416 
417 int ring__consume(struct ring *r)
418 {
419 	return ring__consume_n(r, INT_MAX);
420 }
421 
422 static void user_ringbuf_unmap_ring(struct user_ring_buffer *rb)
423 {
424 	if (rb->consumer_pos) {
425 		munmap(rb->consumer_pos, rb->page_size);
426 		rb->consumer_pos = NULL;
427 	}
428 	if (rb->producer_pos) {
429 		munmap(rb->producer_pos, rb->page_size + 2 * (rb->mask + 1));
430 		rb->producer_pos = NULL;
431 	}
432 }
433 
434 void user_ring_buffer__free(struct user_ring_buffer *rb)
435 {
436 	if (!rb)
437 		return;
438 
439 	user_ringbuf_unmap_ring(rb);
440 
441 	if (rb->epoll_fd >= 0)
442 		close(rb->epoll_fd);
443 
444 	free(rb);
445 }
446 
447 static int user_ringbuf_map(struct user_ring_buffer *rb, int map_fd)
448 {
449 	struct bpf_map_info info;
450 	__u32 len = sizeof(info);
451 	__u64 mmap_sz;
452 	void *tmp;
453 	struct epoll_event *rb_epoll;
454 	int err;
455 
456 	memset(&info, 0, sizeof(info));
457 
458 	err = bpf_map_get_info_by_fd(map_fd, &info, &len);
459 	if (err) {
460 		err = -errno;
461 		pr_warn("user ringbuf: failed to get map info for fd=%d: %d\n", map_fd, err);
462 		return err;
463 	}
464 
465 	if (info.type != BPF_MAP_TYPE_USER_RINGBUF) {
466 		pr_warn("user ringbuf: map fd=%d is not BPF_MAP_TYPE_USER_RINGBUF\n", map_fd);
467 		return -EINVAL;
468 	}
469 
470 	rb->map_fd = map_fd;
471 	rb->mask = info.max_entries - 1;
472 
473 	/* Map read-only consumer page */
474 	tmp = mmap(NULL, rb->page_size, PROT_READ, MAP_SHARED, map_fd, 0);
475 	if (tmp == MAP_FAILED) {
476 		err = -errno;
477 		pr_warn("user ringbuf: failed to mmap consumer page for map fd=%d: %d\n",
478 			map_fd, err);
479 		return err;
480 	}
481 	rb->consumer_pos = tmp;
482 
483 	/* Map read-write the producer page and data pages. We map the data
484 	 * region as twice the total size of the ring buffer to allow the
485 	 * simple reading and writing of samples that wrap around the end of
486 	 * the buffer.  See the kernel implementation for details.
487 	 */
488 	mmap_sz = rb->page_size + 2 * (__u64)info.max_entries;
489 	if (mmap_sz != (__u64)(size_t)mmap_sz) {
490 		pr_warn("user ringbuf: ring buf size (%u) is too big\n", info.max_entries);
491 		return -E2BIG;
492 	}
493 	tmp = mmap(NULL, (size_t)mmap_sz, PROT_READ | PROT_WRITE, MAP_SHARED,
494 		   map_fd, rb->page_size);
495 	if (tmp == MAP_FAILED) {
496 		err = -errno;
497 		pr_warn("user ringbuf: failed to mmap data pages for map fd=%d: %d\n",
498 			map_fd, err);
499 		return err;
500 	}
501 
502 	rb->producer_pos = tmp;
503 	rb->data = tmp + rb->page_size;
504 
505 	rb_epoll = &rb->event;
506 	rb_epoll->events = EPOLLOUT;
507 	if (epoll_ctl(rb->epoll_fd, EPOLL_CTL_ADD, map_fd, rb_epoll) < 0) {
508 		err = -errno;
509 		pr_warn("user ringbuf: failed to epoll add map fd=%d: %d\n", map_fd, err);
510 		return err;
511 	}
512 
513 	return 0;
514 }
515 
516 struct user_ring_buffer *
517 user_ring_buffer__new(int map_fd, const struct user_ring_buffer_opts *opts)
518 {
519 	struct user_ring_buffer *rb;
520 	int err;
521 
522 	if (!OPTS_VALID(opts, user_ring_buffer_opts))
523 		return errno = EINVAL, NULL;
524 
525 	rb = calloc(1, sizeof(*rb));
526 	if (!rb)
527 		return errno = ENOMEM, NULL;
528 
529 	rb->page_size = getpagesize();
530 
531 	rb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
532 	if (rb->epoll_fd < 0) {
533 		err = -errno;
534 		pr_warn("user ringbuf: failed to create epoll instance: %d\n", err);
535 		goto err_out;
536 	}
537 
538 	err = user_ringbuf_map(rb, map_fd);
539 	if (err)
540 		goto err_out;
541 
542 	return rb;
543 
544 err_out:
545 	user_ring_buffer__free(rb);
546 	return errno = -err, NULL;
547 }
548 
549 static void user_ringbuf_commit(struct user_ring_buffer *rb, void *sample, bool discard)
550 {
551 	__u32 new_len;
552 	struct ringbuf_hdr *hdr;
553 	uintptr_t hdr_offset;
554 
555 	hdr_offset = rb->mask + 1 + (sample - rb->data) - BPF_RINGBUF_HDR_SZ;
556 	hdr = rb->data + (hdr_offset & rb->mask);
557 
558 	new_len = hdr->len & ~BPF_RINGBUF_BUSY_BIT;
559 	if (discard)
560 		new_len |= BPF_RINGBUF_DISCARD_BIT;
561 
562 	/* Synchronizes with smp_load_acquire() in __bpf_user_ringbuf_peek() in
563 	 * the kernel.
564 	 */
565 	__atomic_exchange_n(&hdr->len, new_len, __ATOMIC_ACQ_REL);
566 }
567 
568 void user_ring_buffer__discard(struct user_ring_buffer *rb, void *sample)
569 {
570 	user_ringbuf_commit(rb, sample, true);
571 }
572 
573 void user_ring_buffer__submit(struct user_ring_buffer *rb, void *sample)
574 {
575 	user_ringbuf_commit(rb, sample, false);
576 }
577 
578 void *user_ring_buffer__reserve(struct user_ring_buffer *rb, __u32 size)
579 {
580 	__u32 avail_size, total_size, max_size;
581 	/* 64-bit to avoid overflow in case of extreme application behavior */
582 	__u64 cons_pos, prod_pos;
583 	struct ringbuf_hdr *hdr;
584 
585 	/* The top two bits are used as special flags */
586 	if (size & (BPF_RINGBUF_BUSY_BIT | BPF_RINGBUF_DISCARD_BIT))
587 		return errno = E2BIG, NULL;
588 
589 	/* Synchronizes with smp_store_release() in __bpf_user_ringbuf_peek() in
590 	 * the kernel.
591 	 */
592 	cons_pos = smp_load_acquire(rb->consumer_pos);
593 	/* Synchronizes with smp_store_release() in user_ringbuf_commit() */
594 	prod_pos = smp_load_acquire(rb->producer_pos);
595 
596 	max_size = rb->mask + 1;
597 	avail_size = max_size - (prod_pos - cons_pos);
598 	/* Round up total size to a multiple of 8. */
599 	total_size = (size + BPF_RINGBUF_HDR_SZ + 7) / 8 * 8;
600 
601 	if (total_size > max_size)
602 		return errno = E2BIG, NULL;
603 
604 	if (avail_size < total_size)
605 		return errno = ENOSPC, NULL;
606 
607 	hdr = rb->data + (prod_pos & rb->mask);
608 	hdr->len = size | BPF_RINGBUF_BUSY_BIT;
609 	hdr->pad = 0;
610 
611 	/* Synchronizes with smp_load_acquire() in __bpf_user_ringbuf_peek() in
612 	 * the kernel.
613 	 */
614 	smp_store_release(rb->producer_pos, prod_pos + total_size);
615 
616 	return (void *)rb->data + ((prod_pos + BPF_RINGBUF_HDR_SZ) & rb->mask);
617 }
618 
619 static __u64 ns_elapsed_timespec(const struct timespec *start, const struct timespec *end)
620 {
621 	__u64 start_ns, end_ns, ns_per_s = 1000000000;
622 
623 	start_ns = (__u64)start->tv_sec * ns_per_s + start->tv_nsec;
624 	end_ns = (__u64)end->tv_sec * ns_per_s + end->tv_nsec;
625 
626 	return end_ns - start_ns;
627 }
628 
629 void *user_ring_buffer__reserve_blocking(struct user_ring_buffer *rb, __u32 size, int timeout_ms)
630 {
631 	void *sample;
632 	int err, ms_remaining = timeout_ms;
633 	struct timespec start;
634 
635 	if (timeout_ms < 0 && timeout_ms != -1)
636 		return errno = EINVAL, NULL;
637 
638 	if (timeout_ms != -1) {
639 		err = clock_gettime(CLOCK_MONOTONIC, &start);
640 		if (err)
641 			return NULL;
642 	}
643 
644 	do {
645 		int cnt, ms_elapsed;
646 		struct timespec curr;
647 		__u64 ns_per_ms = 1000000;
648 
649 		sample = user_ring_buffer__reserve(rb, size);
650 		if (sample)
651 			return sample;
652 		else if (errno != ENOSPC)
653 			return NULL;
654 
655 		/* The kernel guarantees at least one event notification
656 		 * delivery whenever at least one sample is drained from the
657 		 * ring buffer in an invocation to bpf_ringbuf_drain(). Other
658 		 * additional events may be delivered at any time, but only one
659 		 * event is guaranteed per bpf_ringbuf_drain() invocation,
660 		 * provided that a sample is drained, and the BPF program did
661 		 * not pass BPF_RB_NO_WAKEUP to bpf_ringbuf_drain(). If
662 		 * BPF_RB_FORCE_WAKEUP is passed to bpf_ringbuf_drain(), a
663 		 * wakeup event will be delivered even if no samples are
664 		 * drained.
665 		 */
666 		cnt = epoll_wait(rb->epoll_fd, &rb->event, 1, ms_remaining);
667 		if (cnt < 0)
668 			return NULL;
669 
670 		if (timeout_ms == -1)
671 			continue;
672 
673 		err = clock_gettime(CLOCK_MONOTONIC, &curr);
674 		if (err)
675 			return NULL;
676 
677 		ms_elapsed = ns_elapsed_timespec(&start, &curr) / ns_per_ms;
678 		ms_remaining = timeout_ms - ms_elapsed;
679 	} while (ms_remaining > 0);
680 
681 	/* Try one more time to reserve a sample after the specified timeout has elapsed. */
682 	return user_ring_buffer__reserve(rb, size);
683 }
684