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 <linux/kmemleak.h>
12#include <uapi/linux/btf.h>
13#include <linux/btf_ids.h>
14
15#define RINGBUF_CREATE_FLAG_MASK (BPF_F_NUMA_NODE)
16
17/* non-mmap()'able part of bpf_ringbuf (everything up to consumer page) */
18#define RINGBUF_PGOFF \
19 (offsetof(struct bpf_ringbuf, consumer_pos) >> PAGE_SHIFT)
20/* consumer page and producer page */
21#define RINGBUF_POS_PAGES 2
22#define RINGBUF_NR_META_PAGES (RINGBUF_PGOFF + RINGBUF_POS_PAGES)
23
24#define RINGBUF_MAX_RECORD_SZ (UINT_MAX/4)
25
26struct bpf_ringbuf {
27 wait_queue_head_t waitq;
28 struct irq_work work;
29 u64 mask;
30 struct page **pages;
31 int nr_pages;
32 spinlock_t spinlock ____cacheline_aligned_in_smp;
33 /* For user-space producer ring buffers, an atomic_t busy bit is used
34 * to synchronize access to the ring buffers in the kernel, rather than
35 * the spinlock that is used for kernel-producer ring buffers. This is
36 * done because the ring buffer must hold a lock across a BPF program's
37 * callback:
38 *
39 * __bpf_user_ringbuf_peek() // lock acquired
40 * -> program callback_fn()
41 * -> __bpf_user_ringbuf_sample_release() // lock released
42 *
43 * It is unsafe and incorrect to hold an IRQ spinlock across what could
44 * be a long execution window, so we instead simply disallow concurrent
45 * access to the ring buffer by kernel consumers, and return -EBUSY from
46 * __bpf_user_ringbuf_peek() if the busy bit is held by another task.
47 */
48 atomic_t busy ____cacheline_aligned_in_smp;
49 /* Consumer and producer counters are put into separate pages to
50 * allow each position to be mapped with different permissions.
51 * This prevents a user-space application from modifying the
52 * position and ruining in-kernel tracking. The permissions of the
53 * pages depend on who is producing samples: user-space or the
54 * kernel.
55 *
56 * Kernel-producer
57 * ---------------
58 * The producer position and data pages are mapped as r/o in
59 * userspace. For this approach, bits in the header of samples are
60 * used to signal to user-space, and to other producers, whether a
61 * sample is currently being written.
62 *
63 * User-space producer
64 * -------------------
65 * Only the page containing the consumer position is mapped r/o in
66 * user-space. User-space producers also use bits of the header to
67 * communicate to the kernel, but the kernel must carefully check and
68 * validate each sample to ensure that they're correctly formatted, and
69 * fully contained within the ring buffer.
70 */
71 unsigned long consumer_pos __aligned(PAGE_SIZE);
72 unsigned long producer_pos __aligned(PAGE_SIZE);
73 char data[] __aligned(PAGE_SIZE);
74};
75
76struct bpf_ringbuf_map {
77 struct bpf_map map;
78 struct bpf_ringbuf *rb;
79};
80
81/* 8-byte ring buffer record header structure */
82struct bpf_ringbuf_hdr {
83 u32 len;
84 u32 pg_off;
85};
86
87static struct bpf_ringbuf *bpf_ringbuf_area_alloc(size_t data_sz, int numa_node)
88{
89 const gfp_t flags = GFP_KERNEL_ACCOUNT | __GFP_RETRY_MAYFAIL |
90 __GFP_NOWARN | __GFP_ZERO;
91 int nr_meta_pages = RINGBUF_NR_META_PAGES;
92 int nr_data_pages = data_sz >> PAGE_SHIFT;
93 int nr_pages = nr_meta_pages + nr_data_pages;
94 struct page **pages, *page;
95 struct bpf_ringbuf *rb;
96 size_t array_size;
97 int i;
98
99 /* Each data page is mapped twice to allow "virtual"
100 * continuous read of samples wrapping around the end of ring
101 * buffer area:
102 * ------------------------------------------------------
103 * | meta pages | real data pages | same data pages |
104 * ------------------------------------------------------
105 * | | 1 2 3 4 5 6 7 8 9 | 1 2 3 4 5 6 7 8 9 |
106 * ------------------------------------------------------
107 * | | TA DA | TA DA |
108 * ------------------------------------------------------
109 * ^^^^^^^
110 * |
111 * Here, no need to worry about special handling of wrapped-around
112 * data due to double-mapped data pages. This works both in kernel and
113 * when mmap()'ed in user-space, simplifying both kernel and
114 * user-space implementations significantly.
115 */
116 array_size = (nr_meta_pages + 2 * nr_data_pages) * sizeof(*pages);
117 pages = bpf_map_area_alloc(size: array_size, numa_node);
118 if (!pages)
119 return NULL;
120
121 for (i = 0; i < nr_pages; i++) {
122 page = alloc_pages_node(nid: numa_node, gfp_mask: flags, order: 0);
123 if (!page) {
124 nr_pages = i;
125 goto err_free_pages;
126 }
127 pages[i] = page;
128 if (i >= nr_meta_pages)
129 pages[nr_data_pages + i] = page;
130 }
131
132 rb = vmap(pages, count: nr_meta_pages + 2 * nr_data_pages,
133 VM_MAP | VM_USERMAP, PAGE_KERNEL);
134 if (rb) {
135 kmemleak_not_leak(ptr: pages);
136 rb->pages = pages;
137 rb->nr_pages = nr_pages;
138 return rb;
139 }
140
141err_free_pages:
142 for (i = 0; i < nr_pages; i++)
143 __free_page(pages[i]);
144 bpf_map_area_free(base: pages);
145 return NULL;
146}
147
148static void bpf_ringbuf_notify(struct irq_work *work)
149{
150 struct bpf_ringbuf *rb = container_of(work, struct bpf_ringbuf, work);
151
152 wake_up_all(&rb->waitq);
153}
154
155/* Maximum size of ring buffer area is limited by 32-bit page offset within
156 * record header, counted in pages. Reserve 8 bits for extensibility, and
157 * take into account few extra pages for consumer/producer pages and
158 * non-mmap()'able parts, the current maximum size would be:
159 *
160 * (((1ULL << 24) - RINGBUF_POS_PAGES - RINGBUF_PGOFF) * PAGE_SIZE)
161 *
162 * This gives 64GB limit, which seems plenty for single ring buffer. Now
163 * considering that the maximum value of data_sz is (4GB - 1), there
164 * will be no overflow, so just note the size limit in the comments.
165 */
166static struct bpf_ringbuf *bpf_ringbuf_alloc(size_t data_sz, int numa_node)
167{
168 struct bpf_ringbuf *rb;
169
170 rb = bpf_ringbuf_area_alloc(data_sz, numa_node);
171 if (!rb)
172 return NULL;
173
174 spin_lock_init(&rb->spinlock);
175 atomic_set(v: &rb->busy, i: 0);
176 init_waitqueue_head(&rb->waitq);
177 init_irq_work(work: &rb->work, func: bpf_ringbuf_notify);
178
179 rb->mask = data_sz - 1;
180 rb->consumer_pos = 0;
181 rb->producer_pos = 0;
182
183 return rb;
184}
185
186static struct bpf_map *ringbuf_map_alloc(union bpf_attr *attr)
187{
188 struct bpf_ringbuf_map *rb_map;
189
190 if (attr->map_flags & ~RINGBUF_CREATE_FLAG_MASK)
191 return ERR_PTR(error: -EINVAL);
192
193 if (attr->key_size || attr->value_size ||
194 !is_power_of_2(n: attr->max_entries) ||
195 !PAGE_ALIGNED(attr->max_entries))
196 return ERR_PTR(error: -EINVAL);
197
198 rb_map = bpf_map_area_alloc(size: sizeof(*rb_map), NUMA_NO_NODE);
199 if (!rb_map)
200 return ERR_PTR(error: -ENOMEM);
201
202 bpf_map_init_from_attr(map: &rb_map->map, attr);
203
204 rb_map->rb = bpf_ringbuf_alloc(data_sz: attr->max_entries, numa_node: rb_map->map.numa_node);
205 if (!rb_map->rb) {
206 bpf_map_area_free(base: rb_map);
207 return ERR_PTR(error: -ENOMEM);
208 }
209
210 return &rb_map->map;
211}
212
213static void bpf_ringbuf_free(struct bpf_ringbuf *rb)
214{
215 /* copy pages pointer and nr_pages to local variable, as we are going
216 * to unmap rb itself with vunmap() below
217 */
218 struct page **pages = rb->pages;
219 int i, nr_pages = rb->nr_pages;
220
221 vunmap(addr: rb);
222 for (i = 0; i < nr_pages; i++)
223 __free_page(pages[i]);
224 bpf_map_area_free(base: pages);
225}
226
227static void ringbuf_map_free(struct bpf_map *map)
228{
229 struct bpf_ringbuf_map *rb_map;
230
231 rb_map = container_of(map, struct bpf_ringbuf_map, map);
232 bpf_ringbuf_free(rb: rb_map->rb);
233 bpf_map_area_free(base: rb_map);
234}
235
236static void *ringbuf_map_lookup_elem(struct bpf_map *map, void *key)
237{
238 return ERR_PTR(error: -ENOTSUPP);
239}
240
241static long ringbuf_map_update_elem(struct bpf_map *map, void *key, void *value,
242 u64 flags)
243{
244 return -ENOTSUPP;
245}
246
247static long ringbuf_map_delete_elem(struct bpf_map *map, void *key)
248{
249 return -ENOTSUPP;
250}
251
252static int ringbuf_map_get_next_key(struct bpf_map *map, void *key,
253 void *next_key)
254{
255 return -ENOTSUPP;
256}
257
258static int ringbuf_map_mmap_kern(struct bpf_map *map, struct vm_area_struct *vma)
259{
260 struct bpf_ringbuf_map *rb_map;
261
262 rb_map = container_of(map, struct bpf_ringbuf_map, map);
263
264 if (vma->vm_flags & VM_WRITE) {
265 /* allow writable mapping for the consumer_pos only */
266 if (vma->vm_pgoff != 0 || vma->vm_end - vma->vm_start != PAGE_SIZE)
267 return -EPERM;
268 } else {
269 vm_flags_clear(vma, VM_MAYWRITE);
270 }
271 /* remap_vmalloc_range() checks size and offset constraints */
272 return remap_vmalloc_range(vma, addr: rb_map->rb,
273 pgoff: vma->vm_pgoff + RINGBUF_PGOFF);
274}
275
276static int ringbuf_map_mmap_user(struct bpf_map *map, struct vm_area_struct *vma)
277{
278 struct bpf_ringbuf_map *rb_map;
279
280 rb_map = container_of(map, struct bpf_ringbuf_map, map);
281
282 if (vma->vm_flags & VM_WRITE) {
283 if (vma->vm_pgoff == 0)
284 /* Disallow writable mappings to the consumer pointer,
285 * and allow writable mappings to both the producer
286 * position, and the ring buffer data itself.
287 */
288 return -EPERM;
289 } else {
290 vm_flags_clear(vma, VM_MAYWRITE);
291 }
292 /* remap_vmalloc_range() checks size and offset constraints */
293 return remap_vmalloc_range(vma, addr: rb_map->rb, pgoff: vma->vm_pgoff + RINGBUF_PGOFF);
294}
295
296static unsigned long ringbuf_avail_data_sz(struct bpf_ringbuf *rb)
297{
298 unsigned long cons_pos, prod_pos;
299
300 cons_pos = smp_load_acquire(&rb->consumer_pos);
301 prod_pos = smp_load_acquire(&rb->producer_pos);
302 return prod_pos - cons_pos;
303}
304
305static u32 ringbuf_total_data_sz(const struct bpf_ringbuf *rb)
306{
307 return rb->mask + 1;
308}
309
310static __poll_t ringbuf_map_poll_kern(struct bpf_map *map, struct file *filp,
311 struct poll_table_struct *pts)
312{
313 struct bpf_ringbuf_map *rb_map;
314
315 rb_map = container_of(map, struct bpf_ringbuf_map, map);
316 poll_wait(filp, wait_address: &rb_map->rb->waitq, p: pts);
317
318 if (ringbuf_avail_data_sz(rb: rb_map->rb))
319 return EPOLLIN | EPOLLRDNORM;
320 return 0;
321}
322
323static __poll_t ringbuf_map_poll_user(struct bpf_map *map, struct file *filp,
324 struct poll_table_struct *pts)
325{
326 struct bpf_ringbuf_map *rb_map;
327
328 rb_map = container_of(map, struct bpf_ringbuf_map, map);
329 poll_wait(filp, wait_address: &rb_map->rb->waitq, p: pts);
330
331 if (ringbuf_avail_data_sz(rb: rb_map->rb) < ringbuf_total_data_sz(rb: rb_map->rb))
332 return EPOLLOUT | EPOLLWRNORM;
333 return 0;
334}
335
336static u64 ringbuf_map_mem_usage(const struct bpf_map *map)
337{
338 struct bpf_ringbuf *rb;
339 int nr_data_pages;
340 int nr_meta_pages;
341 u64 usage = sizeof(struct bpf_ringbuf_map);
342
343 rb = container_of(map, struct bpf_ringbuf_map, map)->rb;
344 usage += (u64)rb->nr_pages << PAGE_SHIFT;
345 nr_meta_pages = RINGBUF_NR_META_PAGES;
346 nr_data_pages = map->max_entries >> PAGE_SHIFT;
347 usage += (nr_meta_pages + 2 * nr_data_pages) * sizeof(struct page *);
348 return usage;
349}
350
351BTF_ID_LIST_SINGLE(ringbuf_map_btf_ids, struct, bpf_ringbuf_map)
352const struct bpf_map_ops ringbuf_map_ops = {
353 .map_meta_equal = bpf_map_meta_equal,
354 .map_alloc = ringbuf_map_alloc,
355 .map_free = ringbuf_map_free,
356 .map_mmap = ringbuf_map_mmap_kern,
357 .map_poll = ringbuf_map_poll_kern,
358 .map_lookup_elem = ringbuf_map_lookup_elem,
359 .map_update_elem = ringbuf_map_update_elem,
360 .map_delete_elem = ringbuf_map_delete_elem,
361 .map_get_next_key = ringbuf_map_get_next_key,
362 .map_mem_usage = ringbuf_map_mem_usage,
363 .map_btf_id = &ringbuf_map_btf_ids[0],
364};
365
366BTF_ID_LIST_SINGLE(user_ringbuf_map_btf_ids, struct, bpf_ringbuf_map)
367const struct bpf_map_ops user_ringbuf_map_ops = {
368 .map_meta_equal = bpf_map_meta_equal,
369 .map_alloc = ringbuf_map_alloc,
370 .map_free = ringbuf_map_free,
371 .map_mmap = ringbuf_map_mmap_user,
372 .map_poll = ringbuf_map_poll_user,
373 .map_lookup_elem = ringbuf_map_lookup_elem,
374 .map_update_elem = ringbuf_map_update_elem,
375 .map_delete_elem = ringbuf_map_delete_elem,
376 .map_get_next_key = ringbuf_map_get_next_key,
377 .map_mem_usage = ringbuf_map_mem_usage,
378 .map_btf_id = &user_ringbuf_map_btf_ids[0],
379};
380
381/* Given pointer to ring buffer record metadata and struct bpf_ringbuf itself,
382 * calculate offset from record metadata to ring buffer in pages, rounded
383 * down. This page offset is stored as part of record metadata and allows to
384 * restore struct bpf_ringbuf * from record pointer. This page offset is
385 * stored at offset 4 of record metadata header.
386 */
387static size_t bpf_ringbuf_rec_pg_off(struct bpf_ringbuf *rb,
388 struct bpf_ringbuf_hdr *hdr)
389{
390 return ((void *)hdr - (void *)rb) >> PAGE_SHIFT;
391}
392
393/* Given pointer to ring buffer record header, restore pointer to struct
394 * bpf_ringbuf itself by using page offset stored at offset 4
395 */
396static struct bpf_ringbuf *
397bpf_ringbuf_restore_from_rec(struct bpf_ringbuf_hdr *hdr)
398{
399 unsigned long addr = (unsigned long)(void *)hdr;
400 unsigned long off = (unsigned long)hdr->pg_off << PAGE_SHIFT;
401
402 return (void*)((addr & PAGE_MASK) - off);
403}
404
405static void *__bpf_ringbuf_reserve(struct bpf_ringbuf *rb, u64 size)
406{
407 unsigned long cons_pos, prod_pos, new_prod_pos, flags;
408 u32 len, pg_off;
409 struct bpf_ringbuf_hdr *hdr;
410
411 if (unlikely(size > RINGBUF_MAX_RECORD_SZ))
412 return NULL;
413
414 len = round_up(size + BPF_RINGBUF_HDR_SZ, 8);
415 if (len > ringbuf_total_data_sz(rb))
416 return NULL;
417
418 cons_pos = smp_load_acquire(&rb->consumer_pos);
419
420 if (in_nmi()) {
421 if (!spin_trylock_irqsave(&rb->spinlock, flags))
422 return NULL;
423 } else {
424 spin_lock_irqsave(&rb->spinlock, flags);
425 }
426
427 prod_pos = rb->producer_pos;
428 new_prod_pos = prod_pos + len;
429
430 /* check for out of ringbuf space by ensuring producer position
431 * doesn't advance more than (ringbuf_size - 1) ahead
432 */
433 if (new_prod_pos - cons_pos > rb->mask) {
434 spin_unlock_irqrestore(lock: &rb->spinlock, flags);
435 return NULL;
436 }
437
438 hdr = (void *)rb->data + (prod_pos & rb->mask);
439 pg_off = bpf_ringbuf_rec_pg_off(rb, hdr);
440 hdr->len = size | BPF_RINGBUF_BUSY_BIT;
441 hdr->pg_off = pg_off;
442
443 /* pairs with consumer's smp_load_acquire() */
444 smp_store_release(&rb->producer_pos, new_prod_pos);
445
446 spin_unlock_irqrestore(lock: &rb->spinlock, flags);
447
448 return (void *)hdr + BPF_RINGBUF_HDR_SZ;
449}
450
451BPF_CALL_3(bpf_ringbuf_reserve, struct bpf_map *, map, u64, size, u64, flags)
452{
453 struct bpf_ringbuf_map *rb_map;
454
455 if (unlikely(flags))
456 return 0;
457
458 rb_map = container_of(map, struct bpf_ringbuf_map, map);
459 return (unsigned long)__bpf_ringbuf_reserve(rb: rb_map->rb, size);
460}
461
462const struct bpf_func_proto bpf_ringbuf_reserve_proto = {
463 .func = bpf_ringbuf_reserve,
464 .ret_type = RET_PTR_TO_RINGBUF_MEM_OR_NULL,
465 .arg1_type = ARG_CONST_MAP_PTR,
466 .arg2_type = ARG_CONST_ALLOC_SIZE_OR_ZERO,
467 .arg3_type = ARG_ANYTHING,
468};
469
470static void bpf_ringbuf_commit(void *sample, u64 flags, bool discard)
471{
472 unsigned long rec_pos, cons_pos;
473 struct bpf_ringbuf_hdr *hdr;
474 struct bpf_ringbuf *rb;
475 u32 new_len;
476
477 hdr = sample - BPF_RINGBUF_HDR_SZ;
478 rb = bpf_ringbuf_restore_from_rec(hdr);
479 new_len = hdr->len ^ BPF_RINGBUF_BUSY_BIT;
480 if (discard)
481 new_len |= BPF_RINGBUF_DISCARD_BIT;
482
483 /* update record header with correct final size prefix */
484 xchg(&hdr->len, new_len);
485
486 /* if consumer caught up and is waiting for our record, notify about
487 * new data availability
488 */
489 rec_pos = (void *)hdr - (void *)rb->data;
490 cons_pos = smp_load_acquire(&rb->consumer_pos) & rb->mask;
491
492 if (flags & BPF_RB_FORCE_WAKEUP)
493 irq_work_queue(work: &rb->work);
494 else if (cons_pos == rec_pos && !(flags & BPF_RB_NO_WAKEUP))
495 irq_work_queue(work: &rb->work);
496}
497
498BPF_CALL_2(bpf_ringbuf_submit, void *, sample, u64, flags)
499{
500 bpf_ringbuf_commit(sample, flags, discard: false /* discard */);
501 return 0;
502}
503
504const struct bpf_func_proto bpf_ringbuf_submit_proto = {
505 .func = bpf_ringbuf_submit,
506 .ret_type = RET_VOID,
507 .arg1_type = ARG_PTR_TO_RINGBUF_MEM | OBJ_RELEASE,
508 .arg2_type = ARG_ANYTHING,
509};
510
511BPF_CALL_2(bpf_ringbuf_discard, void *, sample, u64, flags)
512{
513 bpf_ringbuf_commit(sample, flags, discard: true /* discard */);
514 return 0;
515}
516
517const struct bpf_func_proto bpf_ringbuf_discard_proto = {
518 .func = bpf_ringbuf_discard,
519 .ret_type = RET_VOID,
520 .arg1_type = ARG_PTR_TO_RINGBUF_MEM | OBJ_RELEASE,
521 .arg2_type = ARG_ANYTHING,
522};
523
524BPF_CALL_4(bpf_ringbuf_output, struct bpf_map *, map, void *, data, u64, size,
525 u64, flags)
526{
527 struct bpf_ringbuf_map *rb_map;
528 void *rec;
529
530 if (unlikely(flags & ~(BPF_RB_NO_WAKEUP | BPF_RB_FORCE_WAKEUP)))
531 return -EINVAL;
532
533 rb_map = container_of(map, struct bpf_ringbuf_map, map);
534 rec = __bpf_ringbuf_reserve(rb: rb_map->rb, size);
535 if (!rec)
536 return -EAGAIN;
537
538 memcpy(rec, data, size);
539 bpf_ringbuf_commit(sample: rec, flags, discard: false /* discard */);
540 return 0;
541}
542
543const struct bpf_func_proto bpf_ringbuf_output_proto = {
544 .func = bpf_ringbuf_output,
545 .ret_type = RET_INTEGER,
546 .arg1_type = ARG_CONST_MAP_PTR,
547 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
548 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
549 .arg4_type = ARG_ANYTHING,
550};
551
552BPF_CALL_2(bpf_ringbuf_query, struct bpf_map *, map, u64, flags)
553{
554 struct bpf_ringbuf *rb;
555
556 rb = container_of(map, struct bpf_ringbuf_map, map)->rb;
557
558 switch (flags) {
559 case BPF_RB_AVAIL_DATA:
560 return ringbuf_avail_data_sz(rb);
561 case BPF_RB_RING_SIZE:
562 return ringbuf_total_data_sz(rb);
563 case BPF_RB_CONS_POS:
564 return smp_load_acquire(&rb->consumer_pos);
565 case BPF_RB_PROD_POS:
566 return smp_load_acquire(&rb->producer_pos);
567 default:
568 return 0;
569 }
570}
571
572const struct bpf_func_proto bpf_ringbuf_query_proto = {
573 .func = bpf_ringbuf_query,
574 .ret_type = RET_INTEGER,
575 .arg1_type = ARG_CONST_MAP_PTR,
576 .arg2_type = ARG_ANYTHING,
577};
578
579BPF_CALL_4(bpf_ringbuf_reserve_dynptr, struct bpf_map *, map, u32, size, u64, flags,
580 struct bpf_dynptr_kern *, ptr)
581{
582 struct bpf_ringbuf_map *rb_map;
583 void *sample;
584 int err;
585
586 if (unlikely(flags)) {
587 bpf_dynptr_set_null(ptr);
588 return -EINVAL;
589 }
590
591 err = bpf_dynptr_check_size(size);
592 if (err) {
593 bpf_dynptr_set_null(ptr);
594 return err;
595 }
596
597 rb_map = container_of(map, struct bpf_ringbuf_map, map);
598
599 sample = __bpf_ringbuf_reserve(rb: rb_map->rb, size);
600 if (!sample) {
601 bpf_dynptr_set_null(ptr);
602 return -EINVAL;
603 }
604
605 bpf_dynptr_init(ptr, data: sample, type: BPF_DYNPTR_TYPE_RINGBUF, offset: 0, size);
606
607 return 0;
608}
609
610const struct bpf_func_proto bpf_ringbuf_reserve_dynptr_proto = {
611 .func = bpf_ringbuf_reserve_dynptr,
612 .ret_type = RET_INTEGER,
613 .arg1_type = ARG_CONST_MAP_PTR,
614 .arg2_type = ARG_ANYTHING,
615 .arg3_type = ARG_ANYTHING,
616 .arg4_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_RINGBUF | MEM_UNINIT,
617};
618
619BPF_CALL_2(bpf_ringbuf_submit_dynptr, struct bpf_dynptr_kern *, ptr, u64, flags)
620{
621 if (!ptr->data)
622 return 0;
623
624 bpf_ringbuf_commit(sample: ptr->data, flags, discard: false /* discard */);
625
626 bpf_dynptr_set_null(ptr);
627
628 return 0;
629}
630
631const struct bpf_func_proto bpf_ringbuf_submit_dynptr_proto = {
632 .func = bpf_ringbuf_submit_dynptr,
633 .ret_type = RET_VOID,
634 .arg1_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_RINGBUF | OBJ_RELEASE,
635 .arg2_type = ARG_ANYTHING,
636};
637
638BPF_CALL_2(bpf_ringbuf_discard_dynptr, struct bpf_dynptr_kern *, ptr, u64, flags)
639{
640 if (!ptr->data)
641 return 0;
642
643 bpf_ringbuf_commit(sample: ptr->data, flags, discard: true /* discard */);
644
645 bpf_dynptr_set_null(ptr);
646
647 return 0;
648}
649
650const struct bpf_func_proto bpf_ringbuf_discard_dynptr_proto = {
651 .func = bpf_ringbuf_discard_dynptr,
652 .ret_type = RET_VOID,
653 .arg1_type = ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_RINGBUF | OBJ_RELEASE,
654 .arg2_type = ARG_ANYTHING,
655};
656
657static int __bpf_user_ringbuf_peek(struct bpf_ringbuf *rb, void **sample, u32 *size)
658{
659 int err;
660 u32 hdr_len, sample_len, total_len, flags, *hdr;
661 u64 cons_pos, prod_pos;
662
663 /* Synchronizes with smp_store_release() in user-space producer. */
664 prod_pos = smp_load_acquire(&rb->producer_pos);
665 if (prod_pos % 8)
666 return -EINVAL;
667
668 /* Synchronizes with smp_store_release() in __bpf_user_ringbuf_sample_release() */
669 cons_pos = smp_load_acquire(&rb->consumer_pos);
670 if (cons_pos >= prod_pos)
671 return -ENODATA;
672
673 hdr = (u32 *)((uintptr_t)rb->data + (uintptr_t)(cons_pos & rb->mask));
674 /* Synchronizes with smp_store_release() in user-space producer. */
675 hdr_len = smp_load_acquire(hdr);
676 flags = hdr_len & (BPF_RINGBUF_BUSY_BIT | BPF_RINGBUF_DISCARD_BIT);
677 sample_len = hdr_len & ~flags;
678 total_len = round_up(sample_len + BPF_RINGBUF_HDR_SZ, 8);
679
680 /* The sample must fit within the region advertised by the producer position. */
681 if (total_len > prod_pos - cons_pos)
682 return -EINVAL;
683
684 /* The sample must fit within the data region of the ring buffer. */
685 if (total_len > ringbuf_total_data_sz(rb))
686 return -E2BIG;
687
688 /* The sample must fit into a struct bpf_dynptr. */
689 err = bpf_dynptr_check_size(size: sample_len);
690 if (err)
691 return -E2BIG;
692
693 if (flags & BPF_RINGBUF_DISCARD_BIT) {
694 /* If the discard bit is set, the sample should be skipped.
695 *
696 * Update the consumer pos, and return -EAGAIN so the caller
697 * knows to skip this sample and try to read the next one.
698 */
699 smp_store_release(&rb->consumer_pos, cons_pos + total_len);
700 return -EAGAIN;
701 }
702
703 if (flags & BPF_RINGBUF_BUSY_BIT)
704 return -ENODATA;
705
706 *sample = (void *)((uintptr_t)rb->data +
707 (uintptr_t)((cons_pos + BPF_RINGBUF_HDR_SZ) & rb->mask));
708 *size = sample_len;
709 return 0;
710}
711
712static void __bpf_user_ringbuf_sample_release(struct bpf_ringbuf *rb, size_t size, u64 flags)
713{
714 u64 consumer_pos;
715 u32 rounded_size = round_up(size + BPF_RINGBUF_HDR_SZ, 8);
716
717 /* Using smp_load_acquire() is unnecessary here, as the busy-bit
718 * prevents another task from writing to consumer_pos after it was read
719 * by this task with smp_load_acquire() in __bpf_user_ringbuf_peek().
720 */
721 consumer_pos = rb->consumer_pos;
722 /* Synchronizes with smp_load_acquire() in user-space producer. */
723 smp_store_release(&rb->consumer_pos, consumer_pos + rounded_size);
724}
725
726BPF_CALL_4(bpf_user_ringbuf_drain, struct bpf_map *, map,
727 void *, callback_fn, void *, callback_ctx, u64, flags)
728{
729 struct bpf_ringbuf *rb;
730 long samples, discarded_samples = 0, ret = 0;
731 bpf_callback_t callback = (bpf_callback_t)callback_fn;
732 u64 wakeup_flags = BPF_RB_NO_WAKEUP | BPF_RB_FORCE_WAKEUP;
733 int busy = 0;
734
735 if (unlikely(flags & ~wakeup_flags))
736 return -EINVAL;
737
738 rb = container_of(map, struct bpf_ringbuf_map, map)->rb;
739
740 /* If another consumer is already consuming a sample, wait for them to finish. */
741 if (!atomic_try_cmpxchg(v: &rb->busy, old: &busy, new: 1))
742 return -EBUSY;
743
744 for (samples = 0; samples < BPF_MAX_USER_RINGBUF_SAMPLES && ret == 0; samples++) {
745 int err;
746 u32 size;
747 void *sample;
748 struct bpf_dynptr_kern dynptr;
749
750 err = __bpf_user_ringbuf_peek(rb, sample: &sample, size: &size);
751 if (err) {
752 if (err == -ENODATA) {
753 break;
754 } else if (err == -EAGAIN) {
755 discarded_samples++;
756 continue;
757 } else {
758 ret = err;
759 goto schedule_work_return;
760 }
761 }
762
763 bpf_dynptr_init(ptr: &dynptr, data: sample, type: BPF_DYNPTR_TYPE_LOCAL, offset: 0, size);
764 ret = callback((uintptr_t)&dynptr, (uintptr_t)callback_ctx, 0, 0, 0);
765 __bpf_user_ringbuf_sample_release(rb, size, flags);
766 }
767 ret = samples - discarded_samples;
768
769schedule_work_return:
770 /* Prevent the clearing of the busy-bit from being reordered before the
771 * storing of any rb consumer or producer positions.
772 */
773 atomic_set_release(v: &rb->busy, i: 0);
774
775 if (flags & BPF_RB_FORCE_WAKEUP)
776 irq_work_queue(work: &rb->work);
777 else if (!(flags & BPF_RB_NO_WAKEUP) && samples > 0)
778 irq_work_queue(work: &rb->work);
779 return ret;
780}
781
782const struct bpf_func_proto bpf_user_ringbuf_drain_proto = {
783 .func = bpf_user_ringbuf_drain,
784 .ret_type = RET_INTEGER,
785 .arg1_type = ARG_CONST_MAP_PTR,
786 .arg2_type = ARG_PTR_TO_FUNC,
787 .arg3_type = ARG_PTR_TO_STACK_OR_NULL,
788 .arg4_type = ARG_ANYTHING,
789};
790

source code of linux/kernel/bpf/ringbuf.c