xref: /openbmc/linux/kernel/bpf/ringbuf.c (revision 5497b23e)
1 #include <linux/bpf.h>
2 #include <linux/btf.h>
3 #include <linux/err.h>
4 #include <linux/irq_work.h>
5 #include <linux/slab.h>
6 #include <linux/filter.h>
7 #include <linux/mm.h>
8 #include <linux/vmalloc.h>
9 #include <linux/wait.h>
10 #include <linux/poll.h>
11 #include <uapi/linux/btf.h>
12 
13 #define RINGBUF_CREATE_FLAG_MASK (BPF_F_NUMA_NODE)
14 
15 /* non-mmap()'able part of bpf_ringbuf (everything up to consumer page) */
16 #define RINGBUF_PGOFF \
17 	(offsetof(struct bpf_ringbuf, consumer_pos) >> PAGE_SHIFT)
18 /* consumer page and producer page */
19 #define RINGBUF_POS_PAGES 2
20 
21 #define RINGBUF_MAX_RECORD_SZ (UINT_MAX/4)
22 
23 /* Maximum size of ring buffer area is limited by 32-bit page offset within
24  * record header, counted in pages. Reserve 8 bits for extensibility, and take
25  * into account few extra pages for consumer/producer pages and
26  * non-mmap()'able parts. This gives 64GB limit, which seems plenty for single
27  * ring buffer.
28  */
29 #define RINGBUF_MAX_DATA_SZ \
30 	(((1ULL << 24) - RINGBUF_POS_PAGES - RINGBUF_PGOFF) * PAGE_SIZE)
31 
32 struct bpf_ringbuf {
33 	wait_queue_head_t waitq;
34 	struct irq_work work;
35 	u64 mask;
36 	struct page **pages;
37 	int nr_pages;
38 	spinlock_t spinlock ____cacheline_aligned_in_smp;
39 	/* Consumer and producer counters are put into separate pages to allow
40 	 * mapping consumer page as r/w, but restrict producer page to r/o.
41 	 * This protects producer position from being modified by user-space
42 	 * application and ruining in-kernel position tracking.
43 	 */
44 	unsigned long consumer_pos __aligned(PAGE_SIZE);
45 	unsigned long producer_pos __aligned(PAGE_SIZE);
46 	char data[] __aligned(PAGE_SIZE);
47 };
48 
49 struct bpf_ringbuf_map {
50 	struct bpf_map map;
51 	struct bpf_ringbuf *rb;
52 };
53 
54 /* 8-byte ring buffer record header structure */
55 struct bpf_ringbuf_hdr {
56 	u32 len;
57 	u32 pg_off;
58 };
59 
60 static struct bpf_ringbuf *bpf_ringbuf_area_alloc(size_t data_sz, int numa_node)
61 {
62 	const gfp_t flags = GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL |
63 			    __GFP_NOWARN | __GFP_ZERO;
64 	int nr_meta_pages = RINGBUF_PGOFF + RINGBUF_POS_PAGES;
65 	int nr_data_pages = data_sz >> PAGE_SHIFT;
66 	int nr_pages = nr_meta_pages + nr_data_pages;
67 	struct page **pages, *page;
68 	struct bpf_ringbuf *rb;
69 	size_t array_size;
70 	int i;
71 
72 	/* Each data page is mapped twice to allow "virtual"
73 	 * continuous read of samples wrapping around the end of ring
74 	 * buffer area:
75 	 * ------------------------------------------------------
76 	 * | meta pages |  real data pages  |  same data pages  |
77 	 * ------------------------------------------------------
78 	 * |            | 1 2 3 4 5 6 7 8 9 | 1 2 3 4 5 6 7 8 9 |
79 	 * ------------------------------------------------------
80 	 * |            | TA             DA | TA             DA |
81 	 * ------------------------------------------------------
82 	 *                               ^^^^^^^
83 	 *                                  |
84 	 * Here, no need to worry about special handling of wrapped-around
85 	 * data due to double-mapped data pages. This works both in kernel and
86 	 * when mmap()'ed in user-space, simplifying both kernel and
87 	 * user-space implementations significantly.
88 	 */
89 	array_size = (nr_meta_pages + 2 * nr_data_pages) * sizeof(*pages);
90 	pages = bpf_map_area_alloc(array_size, numa_node);
91 	if (!pages)
92 		return NULL;
93 
94 	for (i = 0; i < nr_pages; i++) {
95 		page = alloc_pages_node(numa_node, flags, 0);
96 		if (!page) {
97 			nr_pages = i;
98 			goto err_free_pages;
99 		}
100 		pages[i] = page;
101 		if (i >= nr_meta_pages)
102 			pages[nr_data_pages + i] = page;
103 	}
104 
105 	rb = vmap(pages, nr_meta_pages + 2 * nr_data_pages,
106 		  VM_ALLOC | VM_USERMAP, PAGE_KERNEL);
107 	if (rb) {
108 		rb->pages = pages;
109 		rb->nr_pages = nr_pages;
110 		return rb;
111 	}
112 
113 err_free_pages:
114 	for (i = 0; i < nr_pages; i++)
115 		__free_page(pages[i]);
116 	kvfree(pages);
117 	return NULL;
118 }
119 
120 static void bpf_ringbuf_notify(struct irq_work *work)
121 {
122 	struct bpf_ringbuf *rb = container_of(work, struct bpf_ringbuf, work);
123 
124 	wake_up_all(&rb->waitq);
125 }
126 
127 static struct bpf_ringbuf *bpf_ringbuf_alloc(size_t data_sz, int numa_node)
128 {
129 	struct bpf_ringbuf *rb;
130 
131 	rb = bpf_ringbuf_area_alloc(data_sz, numa_node);
132 	if (!rb)
133 		return NULL;
134 
135 	spin_lock_init(&rb->spinlock);
136 	init_waitqueue_head(&rb->waitq);
137 	init_irq_work(&rb->work, bpf_ringbuf_notify);
138 
139 	rb->mask = data_sz - 1;
140 	rb->consumer_pos = 0;
141 	rb->producer_pos = 0;
142 
143 	return rb;
144 }
145 
146 static struct bpf_map *ringbuf_map_alloc(union bpf_attr *attr)
147 {
148 	struct bpf_ringbuf_map *rb_map;
149 
150 	if (attr->map_flags & ~RINGBUF_CREATE_FLAG_MASK)
151 		return ERR_PTR(-EINVAL);
152 
153 	if (attr->key_size || attr->value_size ||
154 	    !is_power_of_2(attr->max_entries) ||
155 	    !PAGE_ALIGNED(attr->max_entries))
156 		return ERR_PTR(-EINVAL);
157 
158 #ifdef CONFIG_64BIT
159 	/* on 32-bit arch, it's impossible to overflow record's hdr->pgoff */
160 	if (attr->max_entries > RINGBUF_MAX_DATA_SZ)
161 		return ERR_PTR(-E2BIG);
162 #endif
163 
164 	rb_map = kzalloc(sizeof(*rb_map), GFP_USER | __GFP_ACCOUNT);
165 	if (!rb_map)
166 		return ERR_PTR(-ENOMEM);
167 
168 	bpf_map_init_from_attr(&rb_map->map, attr);
169 
170 	rb_map->rb = bpf_ringbuf_alloc(attr->max_entries, rb_map->map.numa_node);
171 	if (!rb_map->rb) {
172 		kfree(rb_map);
173 		return ERR_PTR(-ENOMEM);
174 	}
175 
176 	return &rb_map->map;
177 }
178 
179 static void bpf_ringbuf_free(struct bpf_ringbuf *rb)
180 {
181 	/* copy pages pointer and nr_pages to local variable, as we are going
182 	 * to unmap rb itself with vunmap() below
183 	 */
184 	struct page **pages = rb->pages;
185 	int i, nr_pages = rb->nr_pages;
186 
187 	vunmap(rb);
188 	for (i = 0; i < nr_pages; i++)
189 		__free_page(pages[i]);
190 	kvfree(pages);
191 }
192 
193 static void ringbuf_map_free(struct bpf_map *map)
194 {
195 	struct bpf_ringbuf_map *rb_map;
196 
197 	rb_map = container_of(map, struct bpf_ringbuf_map, map);
198 	bpf_ringbuf_free(rb_map->rb);
199 	kfree(rb_map);
200 }
201 
202 static void *ringbuf_map_lookup_elem(struct bpf_map *map, void *key)
203 {
204 	return ERR_PTR(-ENOTSUPP);
205 }
206 
207 static int ringbuf_map_update_elem(struct bpf_map *map, void *key, void *value,
208 				   u64 flags)
209 {
210 	return -ENOTSUPP;
211 }
212 
213 static int ringbuf_map_delete_elem(struct bpf_map *map, void *key)
214 {
215 	return -ENOTSUPP;
216 }
217 
218 static int ringbuf_map_get_next_key(struct bpf_map *map, void *key,
219 				    void *next_key)
220 {
221 	return -ENOTSUPP;
222 }
223 
224 static size_t bpf_ringbuf_mmap_page_cnt(const struct bpf_ringbuf *rb)
225 {
226 	size_t data_pages = (rb->mask + 1) >> PAGE_SHIFT;
227 
228 	/* consumer page + producer page + 2 x data pages */
229 	return RINGBUF_POS_PAGES + 2 * data_pages;
230 }
231 
232 static int ringbuf_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
233 {
234 	struct bpf_ringbuf_map *rb_map;
235 	size_t mmap_sz;
236 
237 	rb_map = container_of(map, struct bpf_ringbuf_map, map);
238 	mmap_sz = bpf_ringbuf_mmap_page_cnt(rb_map->rb) << PAGE_SHIFT;
239 
240 	if (vma->vm_pgoff * PAGE_SIZE + (vma->vm_end - vma->vm_start) > mmap_sz)
241 		return -EINVAL;
242 
243 	return remap_vmalloc_range(vma, rb_map->rb,
244 				   vma->vm_pgoff + RINGBUF_PGOFF);
245 }
246 
247 static unsigned long ringbuf_avail_data_sz(struct bpf_ringbuf *rb)
248 {
249 	unsigned long cons_pos, prod_pos;
250 
251 	cons_pos = smp_load_acquire(&rb->consumer_pos);
252 	prod_pos = smp_load_acquire(&rb->producer_pos);
253 	return prod_pos - cons_pos;
254 }
255 
256 static __poll_t ringbuf_map_poll(struct bpf_map *map, struct file *filp,
257 				 struct poll_table_struct *pts)
258 {
259 	struct bpf_ringbuf_map *rb_map;
260 
261 	rb_map = container_of(map, struct bpf_ringbuf_map, map);
262 	poll_wait(filp, &rb_map->rb->waitq, pts);
263 
264 	if (ringbuf_avail_data_sz(rb_map->rb))
265 		return EPOLLIN | EPOLLRDNORM;
266 	return 0;
267 }
268 
269 static int ringbuf_map_btf_id;
270 const struct bpf_map_ops ringbuf_map_ops = {
271 	.map_meta_equal = bpf_map_meta_equal,
272 	.map_alloc = ringbuf_map_alloc,
273 	.map_free = ringbuf_map_free,
274 	.map_mmap = ringbuf_map_mmap,
275 	.map_poll = ringbuf_map_poll,
276 	.map_lookup_elem = ringbuf_map_lookup_elem,
277 	.map_update_elem = ringbuf_map_update_elem,
278 	.map_delete_elem = ringbuf_map_delete_elem,
279 	.map_get_next_key = ringbuf_map_get_next_key,
280 	.map_btf_name = "bpf_ringbuf_map",
281 	.map_btf_id = &ringbuf_map_btf_id,
282 };
283 
284 /* Given pointer to ring buffer record metadata and struct bpf_ringbuf itself,
285  * calculate offset from record metadata to ring buffer in pages, rounded
286  * down. This page offset is stored as part of record metadata and allows to
287  * restore struct bpf_ringbuf * from record pointer. This page offset is
288  * stored at offset 4 of record metadata header.
289  */
290 static size_t bpf_ringbuf_rec_pg_off(struct bpf_ringbuf *rb,
291 				     struct bpf_ringbuf_hdr *hdr)
292 {
293 	return ((void *)hdr - (void *)rb) >> PAGE_SHIFT;
294 }
295 
296 /* Given pointer to ring buffer record header, restore pointer to struct
297  * bpf_ringbuf itself by using page offset stored at offset 4
298  */
299 static struct bpf_ringbuf *
300 bpf_ringbuf_restore_from_rec(struct bpf_ringbuf_hdr *hdr)
301 {
302 	unsigned long addr = (unsigned long)(void *)hdr;
303 	unsigned long off = (unsigned long)hdr->pg_off << PAGE_SHIFT;
304 
305 	return (void*)((addr & PAGE_MASK) - off);
306 }
307 
308 static void *__bpf_ringbuf_reserve(struct bpf_ringbuf *rb, u64 size)
309 {
310 	unsigned long cons_pos, prod_pos, new_prod_pos, flags;
311 	u32 len, pg_off;
312 	struct bpf_ringbuf_hdr *hdr;
313 
314 	if (unlikely(size > RINGBUF_MAX_RECORD_SZ))
315 		return NULL;
316 
317 	len = round_up(size + BPF_RINGBUF_HDR_SZ, 8);
318 	cons_pos = smp_load_acquire(&rb->consumer_pos);
319 
320 	if (in_nmi()) {
321 		if (!spin_trylock_irqsave(&rb->spinlock, flags))
322 			return NULL;
323 	} else {
324 		spin_lock_irqsave(&rb->spinlock, flags);
325 	}
326 
327 	prod_pos = rb->producer_pos;
328 	new_prod_pos = prod_pos + len;
329 
330 	/* check for out of ringbuf space by ensuring producer position
331 	 * doesn't advance more than (ringbuf_size - 1) ahead
332 	 */
333 	if (new_prod_pos - cons_pos > rb->mask) {
334 		spin_unlock_irqrestore(&rb->spinlock, flags);
335 		return NULL;
336 	}
337 
338 	hdr = (void *)rb->data + (prod_pos & rb->mask);
339 	pg_off = bpf_ringbuf_rec_pg_off(rb, hdr);
340 	hdr->len = size | BPF_RINGBUF_BUSY_BIT;
341 	hdr->pg_off = pg_off;
342 
343 	/* pairs with consumer's smp_load_acquire() */
344 	smp_store_release(&rb->producer_pos, new_prod_pos);
345 
346 	spin_unlock_irqrestore(&rb->spinlock, flags);
347 
348 	return (void *)hdr + BPF_RINGBUF_HDR_SZ;
349 }
350 
351 BPF_CALL_3(bpf_ringbuf_reserve, struct bpf_map *, map, u64, size, u64, flags)
352 {
353 	struct bpf_ringbuf_map *rb_map;
354 
355 	if (unlikely(flags))
356 		return 0;
357 
358 	rb_map = container_of(map, struct bpf_ringbuf_map, map);
359 	return (unsigned long)__bpf_ringbuf_reserve(rb_map->rb, size);
360 }
361 
362 const struct bpf_func_proto bpf_ringbuf_reserve_proto = {
363 	.func		= bpf_ringbuf_reserve,
364 	.ret_type	= RET_PTR_TO_ALLOC_MEM_OR_NULL,
365 	.arg1_type	= ARG_CONST_MAP_PTR,
366 	.arg2_type	= ARG_CONST_ALLOC_SIZE_OR_ZERO,
367 	.arg3_type	= ARG_ANYTHING,
368 };
369 
370 static void bpf_ringbuf_commit(void *sample, u64 flags, bool discard)
371 {
372 	unsigned long rec_pos, cons_pos;
373 	struct bpf_ringbuf_hdr *hdr;
374 	struct bpf_ringbuf *rb;
375 	u32 new_len;
376 
377 	hdr = sample - BPF_RINGBUF_HDR_SZ;
378 	rb = bpf_ringbuf_restore_from_rec(hdr);
379 	new_len = hdr->len ^ BPF_RINGBUF_BUSY_BIT;
380 	if (discard)
381 		new_len |= BPF_RINGBUF_DISCARD_BIT;
382 
383 	/* update record header with correct final size prefix */
384 	xchg(&hdr->len, new_len);
385 
386 	/* if consumer caught up and is waiting for our record, notify about
387 	 * new data availability
388 	 */
389 	rec_pos = (void *)hdr - (void *)rb->data;
390 	cons_pos = smp_load_acquire(&rb->consumer_pos) & rb->mask;
391 
392 	if (flags & BPF_RB_FORCE_WAKEUP)
393 		irq_work_queue(&rb->work);
394 	else if (cons_pos == rec_pos && !(flags & BPF_RB_NO_WAKEUP))
395 		irq_work_queue(&rb->work);
396 }
397 
398 BPF_CALL_2(bpf_ringbuf_submit, void *, sample, u64, flags)
399 {
400 	bpf_ringbuf_commit(sample, flags, false /* discard */);
401 	return 0;
402 }
403 
404 const struct bpf_func_proto bpf_ringbuf_submit_proto = {
405 	.func		= bpf_ringbuf_submit,
406 	.ret_type	= RET_VOID,
407 	.arg1_type	= ARG_PTR_TO_ALLOC_MEM,
408 	.arg2_type	= ARG_ANYTHING,
409 };
410 
411 BPF_CALL_2(bpf_ringbuf_discard, void *, sample, u64, flags)
412 {
413 	bpf_ringbuf_commit(sample, flags, true /* discard */);
414 	return 0;
415 }
416 
417 const struct bpf_func_proto bpf_ringbuf_discard_proto = {
418 	.func		= bpf_ringbuf_discard,
419 	.ret_type	= RET_VOID,
420 	.arg1_type	= ARG_PTR_TO_ALLOC_MEM,
421 	.arg2_type	= ARG_ANYTHING,
422 };
423 
424 BPF_CALL_4(bpf_ringbuf_output, struct bpf_map *, map, void *, data, u64, size,
425 	   u64, flags)
426 {
427 	struct bpf_ringbuf_map *rb_map;
428 	void *rec;
429 
430 	if (unlikely(flags & ~(BPF_RB_NO_WAKEUP | BPF_RB_FORCE_WAKEUP)))
431 		return -EINVAL;
432 
433 	rb_map = container_of(map, struct bpf_ringbuf_map, map);
434 	rec = __bpf_ringbuf_reserve(rb_map->rb, size);
435 	if (!rec)
436 		return -EAGAIN;
437 
438 	memcpy(rec, data, size);
439 	bpf_ringbuf_commit(rec, flags, false /* discard */);
440 	return 0;
441 }
442 
443 const struct bpf_func_proto bpf_ringbuf_output_proto = {
444 	.func		= bpf_ringbuf_output,
445 	.ret_type	= RET_INTEGER,
446 	.arg1_type	= ARG_CONST_MAP_PTR,
447 	.arg2_type	= ARG_PTR_TO_MEM,
448 	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
449 	.arg4_type	= ARG_ANYTHING,
450 };
451 
452 BPF_CALL_2(bpf_ringbuf_query, struct bpf_map *, map, u64, flags)
453 {
454 	struct bpf_ringbuf *rb;
455 
456 	rb = container_of(map, struct bpf_ringbuf_map, map)->rb;
457 
458 	switch (flags) {
459 	case BPF_RB_AVAIL_DATA:
460 		return ringbuf_avail_data_sz(rb);
461 	case BPF_RB_RING_SIZE:
462 		return rb->mask + 1;
463 	case BPF_RB_CONS_POS:
464 		return smp_load_acquire(&rb->consumer_pos);
465 	case BPF_RB_PROD_POS:
466 		return smp_load_acquire(&rb->producer_pos);
467 	default:
468 		return 0;
469 	}
470 }
471 
472 const struct bpf_func_proto bpf_ringbuf_query_proto = {
473 	.func		= bpf_ringbuf_query,
474 	.ret_type	= RET_INTEGER,
475 	.arg1_type	= ARG_CONST_MAP_PTR,
476 	.arg2_type	= ARG_ANYTHING,
477 };
478