xref: /openbmc/linux/tools/lib/bpf/ringbuf.c (revision 305c8388)
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 <tools/libc_compat.h>
20 
21 #include "libbpf.h"
22 #include "libbpf_internal.h"
23 #include "bpf.h"
24 
25 /* make sure libbpf doesn't use kernel-only integer typedefs */
26 #pragma GCC poison u8 u16 u32 u64 s8 s16 s32 s64
27 
28 struct ring {
29 	ring_buffer_sample_fn sample_cb;
30 	void *ctx;
31 	void *data;
32 	unsigned long *consumer_pos;
33 	unsigned long *producer_pos;
34 	unsigned long mask;
35 	int map_fd;
36 };
37 
38 struct ring_buffer {
39 	struct epoll_event *events;
40 	struct ring *rings;
41 	size_t page_size;
42 	int epoll_fd;
43 	int ring_cnt;
44 };
45 
46 static void ringbuf_unmap_ring(struct ring_buffer *rb, struct ring *r)
47 {
48 	if (r->consumer_pos) {
49 		munmap(r->consumer_pos, rb->page_size);
50 		r->consumer_pos = NULL;
51 	}
52 	if (r->producer_pos) {
53 		munmap(r->producer_pos, rb->page_size + 2 * (r->mask + 1));
54 		r->producer_pos = NULL;
55 	}
56 }
57 
58 /* Add extra RINGBUF maps to this ring buffer manager */
59 int ring_buffer__add(struct ring_buffer *rb, int map_fd,
60 		     ring_buffer_sample_fn sample_cb, void *ctx)
61 {
62 	struct bpf_map_info info;
63 	__u32 len = sizeof(info);
64 	struct epoll_event *e;
65 	struct ring *r;
66 	void *tmp;
67 	int err;
68 
69 	memset(&info, 0, sizeof(info));
70 
71 	err = bpf_obj_get_info_by_fd(map_fd, &info, &len);
72 	if (err) {
73 		err = -errno;
74 		pr_warn("ringbuf: failed to get map info for fd=%d: %d\n",
75 			map_fd, err);
76 		return err;
77 	}
78 
79 	if (info.type != BPF_MAP_TYPE_RINGBUF) {
80 		pr_warn("ringbuf: map fd=%d is not BPF_MAP_TYPE_RINGBUF\n",
81 			map_fd);
82 		return -EINVAL;
83 	}
84 
85 	tmp = reallocarray(rb->rings, rb->ring_cnt + 1, sizeof(*rb->rings));
86 	if (!tmp)
87 		return -ENOMEM;
88 	rb->rings = tmp;
89 
90 	tmp = reallocarray(rb->events, rb->ring_cnt + 1, sizeof(*rb->events));
91 	if (!tmp)
92 		return -ENOMEM;
93 	rb->events = tmp;
94 
95 	r = &rb->rings[rb->ring_cnt];
96 	memset(r, 0, sizeof(*r));
97 
98 	r->map_fd = map_fd;
99 	r->sample_cb = sample_cb;
100 	r->ctx = ctx;
101 	r->mask = info.max_entries - 1;
102 
103 	/* Map writable consumer page */
104 	tmp = mmap(NULL, rb->page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
105 		   map_fd, 0);
106 	if (tmp == MAP_FAILED) {
107 		err = -errno;
108 		pr_warn("ringbuf: failed to mmap consumer page for map fd=%d: %d\n",
109 			map_fd, err);
110 		return err;
111 	}
112 	r->consumer_pos = tmp;
113 
114 	/* Map read-only producer page and data pages. We map twice as big
115 	 * data size to allow simple reading of samples that wrap around the
116 	 * end of a ring buffer. See kernel implementation for details.
117 	 * */
118 	tmp = mmap(NULL, rb->page_size + 2 * info.max_entries, PROT_READ,
119 		   MAP_SHARED, map_fd, rb->page_size);
120 	if (tmp == MAP_FAILED) {
121 		err = -errno;
122 		ringbuf_unmap_ring(rb, r);
123 		pr_warn("ringbuf: failed to mmap data pages for map fd=%d: %d\n",
124 			map_fd, err);
125 		return err;
126 	}
127 	r->producer_pos = tmp;
128 	r->data = tmp + rb->page_size;
129 
130 	e = &rb->events[rb->ring_cnt];
131 	memset(e, 0, sizeof(*e));
132 
133 	e->events = EPOLLIN;
134 	e->data.fd = rb->ring_cnt;
135 	if (epoll_ctl(rb->epoll_fd, EPOLL_CTL_ADD, map_fd, e) < 0) {
136 		err = -errno;
137 		ringbuf_unmap_ring(rb, r);
138 		pr_warn("ringbuf: failed to epoll add map fd=%d: %d\n",
139 			map_fd, err);
140 		return err;
141 	}
142 
143 	rb->ring_cnt++;
144 	return 0;
145 }
146 
147 void ring_buffer__free(struct ring_buffer *rb)
148 {
149 	int i;
150 
151 	if (!rb)
152 		return;
153 
154 	for (i = 0; i < rb->ring_cnt; ++i)
155 		ringbuf_unmap_ring(rb, &rb->rings[i]);
156 	if (rb->epoll_fd >= 0)
157 		close(rb->epoll_fd);
158 
159 	free(rb->events);
160 	free(rb->rings);
161 	free(rb);
162 }
163 
164 struct ring_buffer *
165 ring_buffer__new(int map_fd, ring_buffer_sample_fn sample_cb, void *ctx,
166 		 const struct ring_buffer_opts *opts)
167 {
168 	struct ring_buffer *rb;
169 	int err;
170 
171 	if (!OPTS_VALID(opts, ring_buffer_opts))
172 		return NULL;
173 
174 	rb = calloc(1, sizeof(*rb));
175 	if (!rb)
176 		return NULL;
177 
178 	rb->page_size = getpagesize();
179 
180 	rb->epoll_fd = epoll_create1(EPOLL_CLOEXEC);
181 	if (rb->epoll_fd < 0) {
182 		err = -errno;
183 		pr_warn("ringbuf: failed to create epoll instance: %d\n", err);
184 		goto err_out;
185 	}
186 
187 	err = ring_buffer__add(rb, map_fd, sample_cb, ctx);
188 	if (err)
189 		goto err_out;
190 
191 	return rb;
192 
193 err_out:
194 	ring_buffer__free(rb);
195 	return NULL;
196 }
197 
198 static inline int roundup_len(__u32 len)
199 {
200 	/* clear out top 2 bits (discard and busy, if set) */
201 	len <<= 2;
202 	len >>= 2;
203 	/* add length prefix */
204 	len += BPF_RINGBUF_HDR_SZ;
205 	/* round up to 8 byte alignment */
206 	return (len + 7) / 8 * 8;
207 }
208 
209 static int ringbuf_process_ring(struct ring* r)
210 {
211 	int *len_ptr, len, err, cnt = 0;
212 	unsigned long cons_pos, prod_pos;
213 	bool got_new_data;
214 	void *sample;
215 
216 	cons_pos = smp_load_acquire(r->consumer_pos);
217 	do {
218 		got_new_data = false;
219 		prod_pos = smp_load_acquire(r->producer_pos);
220 		while (cons_pos < prod_pos) {
221 			len_ptr = r->data + (cons_pos & r->mask);
222 			len = smp_load_acquire(len_ptr);
223 
224 			/* sample not committed yet, bail out for now */
225 			if (len & BPF_RINGBUF_BUSY_BIT)
226 				goto done;
227 
228 			got_new_data = true;
229 			cons_pos += roundup_len(len);
230 
231 			if ((len & BPF_RINGBUF_DISCARD_BIT) == 0) {
232 				sample = (void *)len_ptr + BPF_RINGBUF_HDR_SZ;
233 				err = r->sample_cb(r->ctx, sample, len);
234 				if (err) {
235 					/* update consumer pos and bail out */
236 					smp_store_release(r->consumer_pos,
237 							  cons_pos);
238 					return err;
239 				}
240 				cnt++;
241 			}
242 
243 			smp_store_release(r->consumer_pos, cons_pos);
244 		}
245 	} while (got_new_data);
246 done:
247 	return cnt;
248 }
249 
250 /* Consume available ring buffer(s) data without event polling.
251  * Returns number of records consumed across all registered ring buffers, or
252  * negative number if any of the callbacks return error.
253  */
254 int ring_buffer__consume(struct ring_buffer *rb)
255 {
256 	int i, err, res = 0;
257 
258 	for (i = 0; i < rb->ring_cnt; i++) {
259 		struct ring *ring = &rb->rings[i];
260 
261 		err = ringbuf_process_ring(ring);
262 		if (err < 0)
263 			return err;
264 		res += err;
265 	}
266 	return res;
267 }
268 
269 /* Poll for available data and consume records, if any are available.
270  * Returns number of records consumed, or negative number, if any of the
271  * registered callbacks returned error.
272  */
273 int ring_buffer__poll(struct ring_buffer *rb, int timeout_ms)
274 {
275 	int i, cnt, err, res = 0;
276 
277 	cnt = epoll_wait(rb->epoll_fd, rb->events, rb->ring_cnt, timeout_ms);
278 	for (i = 0; i < cnt; i++) {
279 		__u32 ring_id = rb->events[i].data.fd;
280 		struct ring *ring = &rb->rings[ring_id];
281 
282 		err = ringbuf_process_ring(ring);
283 		if (err < 0)
284 			return err;
285 		res += cnt;
286 	}
287 	return cnt < 0 ? -errno : res;
288 }
289